diff options
| author | Thomas Balling Sørensen <tball@tball-laptop.(none)> | 2010-10-05 12:04:08 +0200 |
|---|---|---|
| committer | Thomas Balling Sørensen <tball@tball-laptop.(none)> | 2010-10-05 12:04:08 +0200 |
| commit | 1218430e1200a08cd64b6555d3fd1fd0274ad9e5 (patch) | |
| tree | e060fb27b8388a4bd237ca39fc20f1675c5e367c /src/gallium/drivers/r300 | |
| parent | 63b1525cf0a50e3d31328c3b56355a86056e4c05 (diff) | |
| parent | bf21b7006c63c3dc47045c22d4f372dfe6c7ce67 (diff) | |
Merge branch 'master' into pipe-video
Conflicts:
configs/linux-dri
configure.ac
src/gallium/drivers/nvfx/Makefile
src/gallium/include/pipe/p_defines.h
src/gallium/include/pipe/p_screen.h
src/gallium/include/state_tracker/dri1_api.h
src/gallium/include/state_tracker/drm_api.h
src/gallium/tests/python/samples/tri.py
src/gallium/tests/trivial/Makefile
src/gallium/tests/unit/Makefile
src/gallium/tests/unit/SConscript
src/gallium/tests/unit/u_format_test.c
src/gallium/winsys/nouveau/drm/nouveau_drm_api.c
Diffstat (limited to 'src/gallium/drivers/r300')
51 files changed, 6883 insertions, 3547 deletions
diff --git a/src/gallium/drivers/r300/Makefile b/src/gallium/drivers/r300/Makefile index d3cd6bef96e..66d900ebb5f 100644 --- a/src/gallium/drivers/r300/Makefile +++ b/src/gallium/drivers/r300/Makefile @@ -14,14 +14,17 @@ C_SOURCES = \ r300_hyperz.c \ r300_query.c \ r300_render.c \ + r300_render_stencilref.c \ + r300_render_translate.c \ r300_resource.c \ r300_screen.c \ r300_screen_buffer.c \ r300_state.c \ r300_state_derived.c \ - r300_state_invariant.c \ r300_vs.c \ + r300_vs_draw.c \ r300_texture.c \ + r300_texture_desc.c \ r300_tgsi_to_rc.c \ r300_transfer.c @@ -36,5 +39,6 @@ EXTRA_OBJECTS = \ include ../../Makefile.template +.PHONY: $(COMPILER_ARCHIVE) $(COMPILER_ARCHIVE): $(MAKE) -C $(TOP)/src/mesa/drivers/dri/r300/compiler diff --git a/src/gallium/drivers/r300/SConscript b/src/gallium/drivers/r300/SConscript index 3921085d76a..bf023daaa56 100644 --- a/src/gallium/drivers/r300/SConscript +++ b/src/gallium/drivers/r300/SConscript @@ -24,14 +24,17 @@ r300 = env.ConvenienceLibrary( 'r300_hyperz.c', 'r300_query.c', 'r300_render.c', + 'r300_render_stencilref.c', + 'r300_render_translate.c', 'r300_resource.c', 'r300_screen.c', 'r300_screen_buffer.c', 'r300_state.c', 'r300_state_derived.c', - 'r300_state_invariant.c', 'r300_vs.c', + 'r300_vs_draw.c', 'r300_texture.c', + 'r300_texture_desc.c', 'r300_tgsi_to_rc.c', 'r300_transfer.c', ] + r300compiler) + r300compiler diff --git a/src/gallium/drivers/r300/r300_blit.c b/src/gallium/drivers/r300/r300_blit.c index 928ad300eee..91a374a5838 100644 --- a/src/gallium/drivers/r300/r300_blit.c +++ b/src/gallium/drivers/r300/r300_blit.c @@ -20,14 +20,32 @@ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE * USE OR OTHER DEALINGS IN THE SOFTWARE. */ -#include "r300_blit.h" #include "r300_context.h" +#include "r300_emit.h" +#include "r300_hyperz.h" #include "r300_texture.h" +#include "r300_winsys.h" #include "util/u_format.h" +#include "util/u_pack_color.h" -static void r300_blitter_save_states(struct r300_context* r300) +enum r300_blitter_op /* bitmask */ { + R300_CLEAR = 1, + R300_CLEAR_SURFACE = 2, + R300_COPY = 4 +}; + +static void r300_blitter_begin(struct r300_context* r300, enum r300_blitter_op op) +{ + if (r300->query_current) { + r300->blitter_saved_query = r300->query_current; + r300_stop_query(r300); + } + + /* Yeah we have to save all those states to ensure the blitter operation + * is really transparent. The states will be restored by the blitter once + * copying is done. */ util_blitter_save_blend(r300->blitter, r300->blend_state.state); util_blitter_save_depth_stencil_alpha(r300->blitter, r300->dsa_state.state); util_blitter_save_stencil_ref(r300->blitter, &(r300->stencil_ref)); @@ -35,109 +53,315 @@ static void r300_blitter_save_states(struct r300_context* r300) util_blitter_save_fragment_shader(r300->blitter, r300->fs.state); util_blitter_save_vertex_shader(r300->blitter, r300->vs_state.state); util_blitter_save_viewport(r300->blitter, &r300->viewport); - util_blitter_save_clip(r300->blitter, &r300->clip); + util_blitter_save_clip(r300->blitter, (struct pipe_clip_state*)r300->clip_state.state); util_blitter_save_vertex_elements(r300->blitter, r300->velems); - /* XXX this crashes the driver util_blitter_save_vertex_buffers(r300->blitter, r300->vertex_buffer_count, - r300->vertex_buffer); */ + r300->vertex_buffer); + + if (op & (R300_CLEAR_SURFACE | R300_COPY)) + util_blitter_save_framebuffer(r300->blitter, r300->fb_state.state); + + if (op & R300_COPY) { + struct r300_textures_state* state = + (struct r300_textures_state*)r300->textures_state.state; + + util_blitter_save_fragment_sampler_states( + r300->blitter, state->sampler_state_count, + (void**)state->sampler_states); + + util_blitter_save_fragment_sampler_views( + r300->blitter, state->sampler_view_count, + (struct pipe_sampler_view**)state->sampler_views); + } +} + +static void r300_blitter_end(struct r300_context *r300) +{ + if (r300->blitter_saved_query) { + r300_resume_query(r300, r300->blitter_saved_query); + r300->blitter_saved_query = NULL; + } +} + +static uint32_t r300_depth_clear_cb_value(enum pipe_format format, + const float* rgba) +{ + union util_color uc; + util_pack_color(rgba, format, &uc); + + if (util_format_get_blocksizebits(format) == 32) + return uc.ui; + else + return uc.us | (uc.us << 16); +} + +static boolean r300_cbzb_clear_allowed(struct r300_context *r300, + unsigned clear_buffers) +{ + struct pipe_framebuffer_state *fb = + (struct pipe_framebuffer_state*)r300->fb_state.state; + + /* Only color clear allowed, and only one colorbuffer. */ + if (clear_buffers != PIPE_CLEAR_COLOR || fb->nr_cbufs != 1) + return FALSE; + + return r300_surface(fb->cbufs[0])->cbzb_allowed; +} + +static uint32_t r300_depth_clear_value(enum pipe_format format, + double depth, unsigned stencil) +{ + switch (format) { + case PIPE_FORMAT_Z16_UNORM: + case PIPE_FORMAT_X8Z24_UNORM: + return util_pack_z(format, depth); + + case PIPE_FORMAT_S8_USCALED_Z24_UNORM: + return util_pack_z_stencil(format, depth, stencil); + + default: + assert(0); + return 0; + } } /* Clear currently bound buffers. */ -void r300_clear(struct pipe_context* pipe, - unsigned buffers, - const float* rgba, - double depth, - unsigned stencil) +static void r300_clear(struct pipe_context* pipe, + unsigned buffers, + const float* rgba, + double depth, + unsigned stencil) { - /* XXX Implement fastfill. + /* My notes about fastfill: + * + * 1) Only the zbuffer is cleared. + * + * 2) The zbuffer must be micro-tiled and whole microtiles must be + * written. If microtiling is disabled, it locks up. * - * If fastfill is enabled, a few facts should be considered: + * 3) There is Z Mask RAM which contains a compressed zbuffer and + * it interacts with fastfill. We should figure out how to use it + * to get more performance. + * This is what we know about the Z Mask: * - * 1) Zbuffer must be micro-tiled and whole microtiles must be - * written. + * Each dword of the Z Mask contains compression information + * for 16 4x4 pixel blocks, that is 2 bits for each block. + * On chips with 2 Z pipes, every other dword maps to a different + * pipe. * - * 2) ZB_DEPTHCLEARVALUE is used to clear a zbuffer and Z Mask must be - * equal to 0. + * 4) ZB_DEPTHCLEARVALUE is used to clear the zbuffer and the Z Mask must + * be equal to 0. (clear the Z Mask RAM with zeros) * - * 3) For 16-bit integer buffering, compression causes a hung with one or + * 5) For 16-bit zbuffer, compression causes a hung with one or * two samples and should not be used. * - * 4) Fastfill must not be used if reading of compressed Z data is disabled + * 6) FORCE_COMPRESSED_STENCIL_VALUE should be enabled for stencil clears + * to avoid needless decompression. + * + * 7) Fastfill must not be used if reading of compressed Z data is disabled * and writing of compressed Z data is enabled (RD/WR_COMP_ENABLE), * i.e. it cannot be used to compress the zbuffer. - * (what the hell does that mean and how does it fit in clearing - * the buffers?) + * + * 8) ZB_CB_CLEAR does not interact with fastfill in any way. * * - Marek */ struct r300_context* r300 = r300_context(pipe); - struct pipe_framebuffer_state* fb = + struct pipe_framebuffer_state *fb = (struct pipe_framebuffer_state*)r300->fb_state.state; + struct r300_hyperz_state *hyperz = + (struct r300_hyperz_state*)r300->hyperz_state.state; + struct r300_texture *zstex = + fb->zsbuf ? r300_texture(fb->zsbuf->texture) : NULL; + uint32_t width = fb->width; + uint32_t height = fb->height; + boolean has_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ); + uint32_t hyperz_dcv = hyperz->zb_depthclearvalue; + + /* Enable fast Z clear. + * The zbuffer must be in micro-tiled mode, otherwise it locks up. */ + if ((buffers & PIPE_CLEAR_DEPTHSTENCIL) && has_hyperz) { + hyperz_dcv = hyperz->zb_depthclearvalue = + r300_depth_clear_value(fb->zsbuf->format, depth, stencil); + + r300_mark_fb_state_dirty(r300, R300_CHANGED_ZCLEAR_FLAG); + if (zstex->zmask_mem[fb->zsbuf->level]) { + r300->zmask_clear.dirty = TRUE; + buffers &= ~PIPE_CLEAR_DEPTHSTENCIL; + } + if (zstex->hiz_mem[fb->zsbuf->level]) + r300->hiz_clear.dirty = TRUE; + } + + /* Enable CBZB clear. */ + if (r300_cbzb_clear_allowed(r300, buffers)) { + struct r300_surface *surf = r300_surface(fb->cbufs[0]); - r300_blitter_save_states(r300); + hyperz->zb_depthclearvalue = + r300_depth_clear_cb_value(surf->base.format, rgba); - util_blitter_clear(r300->blitter, - fb->width, - fb->height, - fb->nr_cbufs, - buffers, rgba, depth, stencil); + width = surf->cbzb_width; + height = surf->cbzb_height; + + r300->cbzb_clear = TRUE; + r300_mark_fb_state_dirty(r300, R300_CHANGED_CBZB_FLAG); + } + + /* Clear. */ + if (buffers) { + /* Clear using the blitter. */ + r300_blitter_begin(r300, R300_CLEAR); + util_blitter_clear(r300->blitter, + width, + height, + fb->nr_cbufs, + buffers, rgba, depth, stencil); + r300_blitter_end(r300); + } else if (r300->zmask_clear.dirty) { + /* Just clear zmask and hiz now, this does not use a standard draw + * procedure. */ + unsigned dwords; + + /* Calculate zmask_clear and hiz_clear atom sizes. */ + r300_update_hyperz_state(r300); + dwords = r300->zmask_clear.size + + (r300->hiz_clear.dirty ? r300->hiz_clear.size : 0) + + r300_get_num_cs_end_dwords(r300); + + /* Reserve CS space. */ + if (dwords > (r300->cs->ndw - r300->cs->cdw)) { + r300->context.flush(&r300->context, 0, NULL); + } + + /* Emit clear packets. */ + r300_emit_zmask_clear(r300, r300->zmask_clear.size, + r300->zmask_clear.state); + r300->zmask_clear.dirty = FALSE; + if (r300->hiz_clear.dirty) { + r300_emit_hiz_clear(r300, r300->hiz_clear.size, + r300->hiz_clear.state); + r300->hiz_clear.dirty = FALSE; + } + } else { + assert(0); + } + + /* Disable CBZB clear. */ + if (r300->cbzb_clear) { + r300->cbzb_clear = FALSE; + hyperz->zb_depthclearvalue = hyperz_dcv; + r300_mark_fb_state_dirty(r300, R300_CHANGED_CBZB_FLAG); + } + + /* Enable fastfill and/or hiz. + * + * If we cleared zmask/hiz, it's in use now. The Hyper-Z state update + * looks if zmask/hiz is in use and enables fastfill accordingly. */ + if (zstex && + (zstex->zmask_in_use[fb->zsbuf->level] || + zstex->hiz_in_use[fb->zsbuf->level])) { + r300->hyperz_state.dirty = TRUE; + } } -/* Copy a block of pixels from one surface to another using HW. */ -static void r300_hw_copy(struct pipe_context* pipe, - struct pipe_surface* dst, - unsigned dstx, unsigned dsty, - struct pipe_surface* src, - unsigned srcx, unsigned srcy, - unsigned width, unsigned height) +/* Clear a region of a color surface to a constant value. */ +static void r300_clear_render_target(struct pipe_context *pipe, + struct pipe_surface *dst, + const float *rgba, + unsigned dstx, unsigned dsty, + unsigned width, unsigned height) { - struct r300_context* r300 = r300_context(pipe); - struct r300_textures_state* state = - (struct r300_textures_state*)r300->textures_state.state; + struct r300_context *r300 = r300_context(pipe); - /* Yeah we have to save all those states to ensure this blitter operation - * is really transparent. The states will be restored by the blitter once - * copying is done. */ - r300_blitter_save_states(r300); - util_blitter_save_framebuffer(r300->blitter, r300->fb_state.state); + r300_blitter_begin(r300, R300_CLEAR_SURFACE); + util_blitter_clear_render_target(r300->blitter, dst, rgba, + dstx, dsty, width, height); + r300_blitter_end(r300); +} + +/* Clear a region of a depth stencil surface. */ +static void r300_clear_depth_stencil(struct pipe_context *pipe, + struct pipe_surface *dst, + unsigned clear_flags, + double depth, + unsigned stencil, + unsigned dstx, unsigned dsty, + unsigned width, unsigned height) +{ + struct r300_context *r300 = r300_context(pipe); + + r300_blitter_begin(r300, R300_CLEAR_SURFACE); + util_blitter_clear_depth_stencil(r300->blitter, dst, clear_flags, depth, stencil, + dstx, dsty, width, height); + r300_blitter_end(r300); +} + +/* Flush a depth stencil buffer. */ +void r300_flush_depth_stencil(struct pipe_context *pipe, + struct pipe_resource *dst, + struct pipe_subresource subdst, + unsigned zslice) +{ + struct r300_context *r300 = r300_context(pipe); + struct pipe_surface *dstsurf; + struct r300_texture *tex = r300_texture(dst); - util_blitter_save_fragment_sampler_states( - r300->blitter, state->sampler_state_count, - (void**)state->sampler_states); + if (!tex->zmask_mem[subdst.level]) + return; + if (!tex->zmask_in_use[subdst.level]) + return; - util_blitter_save_fragment_sampler_views( - r300->blitter, state->sampler_view_count, - (struct pipe_sampler_view**)state->sampler_views); + dstsurf = pipe->screen->get_tex_surface(pipe->screen, dst, + subdst.face, subdst.level, zslice, + PIPE_BIND_DEPTH_STENCIL); + r300->z_decomp_rd = TRUE; + r300_blitter_begin(r300, R300_CLEAR_SURFACE); + util_blitter_flush_depth_stencil(r300->blitter, dstsurf); + r300_blitter_end(r300); + r300->z_decomp_rd = FALSE; - /* Do a copy */ - util_blitter_copy(r300->blitter, - dst, dstx, dsty, src, srcx, srcy, width, height, TRUE); + tex->zmask_in_use[subdst.level] = FALSE; } -/* Copy a block of pixels from one surface to another. */ -void r300_surface_copy(struct pipe_context* pipe, - struct pipe_surface* dst, - unsigned dstx, unsigned dsty, - struct pipe_surface* src, - unsigned srcx, unsigned srcy, - unsigned width, unsigned height) +/* Copy a block of pixels from one surface to another using HW. */ +static void r300_hw_copy_region(struct pipe_context* pipe, + struct pipe_resource *dst, + struct pipe_subresource subdst, + unsigned dstx, unsigned dsty, unsigned dstz, + struct pipe_resource *src, + struct pipe_subresource subsrc, + unsigned srcx, unsigned srcy, unsigned srcz, + unsigned width, unsigned height) { - enum pipe_format old_format = dst->texture->format; - enum pipe_format new_format = old_format; + struct r300_context* r300 = r300_context(pipe); - if (dst->texture->format != src->texture->format) { - debug_printf("r300: Implementation error: Format mismatch in %s\n" - " : src: %s dst: %s\n", __FUNCTION__, - util_format_name(src->texture->format), - util_format_name(dst->texture->format)); - debug_assert(0); - } + r300_blitter_begin(r300, R300_COPY); + util_blitter_copy_region(r300->blitter, dst, subdst, dstx, dsty, dstz, + src, subsrc, srcx, srcy, srcz, width, height, + TRUE); + r300_blitter_end(r300); +} +/* Copy a block of pixels from one surface to another. */ +static void r300_resource_copy_region(struct pipe_context *pipe, + struct pipe_resource *dst, + struct pipe_subresource subdst, + unsigned dstx, unsigned dsty, unsigned dstz, + struct pipe_resource *src, + struct pipe_subresource subsrc, + unsigned srcx, unsigned srcy, unsigned srcz, + unsigned width, unsigned height) +{ + enum pipe_format old_format = dst->format; + enum pipe_format new_format = old_format; + boolean is_depth; if (!pipe->screen->is_format_supported(pipe->screen, - old_format, src->texture->target, + old_format, src->target, + src->nr_samples, PIPE_BIND_RENDER_TARGET | - PIPE_BIND_SAMPLER_VIEW, 0)) { + PIPE_BIND_SAMPLER_VIEW, 0) && + util_format_is_plain(old_format)) { switch (util_format_get_blocksize(old_format)) { case 1: new_format = PIPE_FORMAT_I8_UNORM; @@ -154,45 +378,36 @@ void r300_surface_copy(struct pipe_context* pipe, default: debug_printf("r300: surface_copy: Unhandled format: %s. Falling back to software.\n" "r300: surface_copy: Software fallback doesn't work for tiled textures.\n", - util_format_name(old_format)); + util_format_short_name(old_format)); } } + is_depth = util_format_get_component_bits(src->format, UTIL_FORMAT_COLORSPACE_ZS, 0) != 0; + if (is_depth) { + r300_flush_depth_stencil(pipe, src, subsrc, srcz); + } if (old_format != new_format) { - dst->format = new_format; - src->format = new_format; - r300_texture_reinterpret_format(pipe->screen, - dst->texture, new_format); + dst, new_format); r300_texture_reinterpret_format(pipe->screen, - src->texture, new_format); + src, new_format); } - r300_hw_copy(pipe, dst, dstx, dsty, src, srcx, srcy, width, height); + r300_hw_copy_region(pipe, dst, subdst, dstx, dsty, dstz, + src, subsrc, srcx, srcy, srcz, width, height); if (old_format != new_format) { - dst->format = old_format; - src->format = old_format; - r300_texture_reinterpret_format(pipe->screen, - dst->texture, old_format); + dst, old_format); r300_texture_reinterpret_format(pipe->screen, - src->texture, old_format); + src, old_format); } } -/* Fill a region of a surface with a constant value. */ -void r300_surface_fill(struct pipe_context* pipe, - struct pipe_surface* dst, - unsigned dstx, unsigned dsty, - unsigned width, unsigned height, - unsigned value) +void r300_init_blit_functions(struct r300_context *r300) { - struct r300_context* r300 = r300_context(pipe); - - r300_blitter_save_states(r300); - util_blitter_save_framebuffer(r300->blitter, r300->fb_state.state); - - util_blitter_fill(r300->blitter, - dst, dstx, dsty, width, height, value); + r300->context.clear = r300_clear; + r300->context.clear_render_target = r300_clear_render_target; + r300->context.clear_depth_stencil = r300_clear_depth_stencil; + r300->context.resource_copy_region = r300_resource_copy_region; } diff --git a/src/gallium/drivers/r300/r300_blit.h b/src/gallium/drivers/r300/r300_blit.h deleted file mode 100644 index 029e4f98e7d..00000000000 --- a/src/gallium/drivers/r300/r300_blit.h +++ /dev/null @@ -1,48 +0,0 @@ -/* - * Copyright 2008 Marek Olšák <[email protected]> - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * on the rights to use, copy, modify, merge, publish, distribute, sub - * license, and/or sell copies of the Software, and to permit persons to whom - * the Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice (including the next - * paragraph) shall be included in all copies or substantial portions of the - * Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL - * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, - * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR - * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE - * USE OR OTHER DEALINGS IN THE SOFTWARE. */ - -#ifndef R300_BLIT_H -#define R300_BLIT_H - -struct pipe_context; -struct pipe_surface; - -void r300_clear(struct pipe_context* pipe, - unsigned buffers, - const float* rgba, - double depth, - unsigned stencil); - -void r300_surface_copy(struct pipe_context* pipe, - struct pipe_surface* dst, - unsigned dstx, unsigned dsty, - struct pipe_surface* src, - unsigned srcx, unsigned srcy, - unsigned width, unsigned height); - -void r300_surface_fill(struct pipe_context* pipe, - struct pipe_surface* dst, - unsigned dstx, unsigned dsty, - unsigned width, unsigned height, - unsigned value); - -#endif /* R300_BLIT_H */ diff --git a/src/gallium/drivers/r300/r300_cb.h b/src/gallium/drivers/r300/r300_cb.h new file mode 100644 index 00000000000..9d3d4fc1b19 --- /dev/null +++ b/src/gallium/drivers/r300/r300_cb.h @@ -0,0 +1,139 @@ +/* + * Copyright 2008 Corbin Simpson <[email protected]> + * Copyright 2010 Marek Olšák <[email protected]> + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. */ + +/** + * This file contains macros for building command buffers in memory. + * + * Use NEW_CB for buffers with a varying size and it will also allocate + * the buffer. + * Use BEGIN_CB for arrays with a static size. + * + * Example: + * + * uint32_t cb[3]; + * CB_LOCALS; + * + * BEGIN_CB(cb, 3); + * OUT_CB_REG_SEQ(R500_RB3D_CONSTANT_COLOR_AR, 2); + * OUT_CB(blend_color_red_alpha); + * OUT_CB(blend_color_green_blue); + * END_CB; + * + * And later: + * + * CS_LOCALS; + * WRITE_CS_TABLE(cb, 3); + * + * Or using a little slower variant: + * + * CS_LOCALS; + * BEGIN_CS(cb, 3); + * OUT_CS_TABLE(cb, 3); + * END_CS; + */ + +#ifndef R300_CB_H +#define R300_CB_H + +#include "r300_reg.h" + +/* Yes, I know macros are ugly. However, they are much prettier than the code + * that they neatly hide away, and don't have the cost of function setup, so + * we're going to use them. */ + +#ifdef DEBUG +#define CB_DEBUG(x) x +#else +#define CB_DEBUG(x) +#endif + + +/** + * Command buffer setup. + */ + +#define CB_LOCALS \ + CB_DEBUG(int cs_count = 0;) \ + uint32_t *cs_ptr = NULL; \ + CB_DEBUG((void) cs_count;) (void) cs_ptr; + +#define NEW_CB(ptr, size) do { \ + assert(sizeof(*ptr) == sizeof(uint32_t)); \ + cs_ptr = (ptr) = (uint32_t*)malloc((size) * sizeof(uint32_t)); \ + CB_DEBUG(cs_count = size;) \ +} while (0) + +#define BEGIN_CB(ptr, size) do { \ + assert(sizeof(*ptr) == sizeof(uint32_t)); \ + cs_ptr = ptr; \ + CB_DEBUG(cs_count = size;) \ +} while (0) + +#define END_CB do { \ + CB_DEBUG(if (cs_count != 0) \ + debug_printf("r300: Warning: cs_count off by %d at (%s, %s:%i)\n", \ + cs_count, __FUNCTION__, __FILE__, __LINE__);) \ +} while (0) + + +/** + * Storing pure DWORDs. + */ + +#define OUT_CB(value) do { \ + *cs_ptr = (value); \ + cs_ptr++; \ + CB_DEBUG(cs_count--;) \ +} while (0) + +#define OUT_CB_TABLE(values, count) do { \ + memcpy(cs_ptr, values, count * sizeof(uint32_t)); \ + cs_ptr += count; \ + CB_DEBUG(cs_count -= count;) \ +} while (0) + +#define OUT_CB_32F(value) \ + OUT_CB(fui(value)); + +#define OUT_CB_REG(register, value) do { \ + assert(register); \ + OUT_CB(CP_PACKET0(register, 0)); \ + OUT_CB(value); \ +} while (0) + +/* Note: This expects count to be the number of registers, + * not the actual packet0 count! */ +#define OUT_CB_REG_SEQ(register, count) do { \ + assert(register); \ + OUT_CB(CP_PACKET0(register, (count) - 1)); \ +} while (0) + +#define OUT_CB_ONE_REG(register, count) do { \ + assert(register); \ + OUT_CB(CP_PACKET0(register, (count) - 1) | RADEON_ONE_REG_WR); \ +} while (0) + +#define OUT_CB_PKT3(op, count) \ + OUT_CB(CP_PACKET3(op, count)) + +#endif /* R300_CB_H */ diff --git a/src/gallium/drivers/r300/r300_chipset.c b/src/gallium/drivers/r300/r300_chipset.c index e6dca66d4a0..48c24092114 100644 --- a/src/gallium/drivers/r300/r300_chipset.c +++ b/src/gallium/drivers/r300/r300_chipset.c @@ -36,6 +36,7 @@ void r300_parse_chipset(struct r300_capabilities* caps) caps->num_vert_fpus = 2; caps->num_tex_units = 16; caps->has_tcl = debug_get_bool_option("RADEON_NO_TCL", FALSE) ? FALSE : TRUE; + caps->hiz_ram = 0; caps->is_r400 = FALSE; caps->is_r500 = FALSE; caps->high_second_pipe = FALSE; @@ -48,6 +49,8 @@ void r300_parse_chipset(struct r300_capabilities* caps) caps->family = CHIP_FAMILY_R300; caps->high_second_pipe = TRUE; caps->num_vert_fpus = 4; + caps->hiz_ram = R300_HIZ_LIMIT; + caps->zmask_ram = PIPE_ZMASK_SIZE; break; case 0x4145: @@ -60,6 +63,8 @@ void r300_parse_chipset(struct r300_capabilities* caps) caps->family = CHIP_FAMILY_R300; caps->high_second_pipe = TRUE; caps->num_vert_fpus = 4; + caps->hiz_ram = R300_HIZ_LIMIT; + caps->zmask_ram = PIPE_ZMASK_SIZE; break; case 0x4150: @@ -77,6 +82,7 @@ void r300_parse_chipset(struct r300_capabilities* caps) case 0x4E56: caps->family = CHIP_FAMILY_RV350; caps->high_second_pipe = TRUE; + caps->zmask_ram = RV3xx_ZMASK_SIZE; break; case 0x4148: @@ -89,12 +95,16 @@ void r300_parse_chipset(struct r300_capabilities* caps) caps->family = CHIP_FAMILY_R350; caps->high_second_pipe = TRUE; caps->num_vert_fpus = 4; + caps->hiz_ram = R300_HIZ_LIMIT; + caps->zmask_ram = PIPE_ZMASK_SIZE; break; case 0x4E4A: caps->family = CHIP_FAMILY_R360; caps->high_second_pipe = TRUE; caps->num_vert_fpus = 4; + caps->hiz_ram = R300_HIZ_LIMIT; + caps->zmask_ram = PIPE_ZMASK_SIZE; break; case 0x5460: @@ -107,6 +117,7 @@ void r300_parse_chipset(struct r300_capabilities* caps) case 0x5B65: caps->family = CHIP_FAMILY_RV370; caps->high_second_pipe = TRUE; + caps->zmask_ram = RV3xx_ZMASK_SIZE; break; case 0x3150: @@ -117,6 +128,8 @@ void r300_parse_chipset(struct r300_capabilities* caps) case 0x3E54: caps->family = CHIP_FAMILY_RV380; caps->high_second_pipe = TRUE; + caps->hiz_ram = R300_HIZ_LIMIT; + caps->zmask_ram = RV3xx_ZMASK_SIZE; break; case 0x4A48: @@ -132,6 +145,8 @@ void r300_parse_chipset(struct r300_capabilities* caps) caps->family = CHIP_FAMILY_R420; caps->num_vert_fpus = 6; caps->is_r400 = TRUE; + caps->hiz_ram = R300_HIZ_LIMIT; + caps->zmask_ram = PIPE_ZMASK_SIZE; break; case 0x5548: @@ -146,6 +161,8 @@ void r300_parse_chipset(struct r300_capabilities* caps) caps->family = CHIP_FAMILY_R423; caps->num_vert_fpus = 6; caps->is_r400 = TRUE; + caps->hiz_ram = R300_HIZ_LIMIT; + caps->zmask_ram = PIPE_ZMASK_SIZE; break; case 0x554C: @@ -158,6 +175,8 @@ void r300_parse_chipset(struct r300_capabilities* caps) caps->family = CHIP_FAMILY_R430; caps->num_vert_fpus = 6; caps->is_r400 = TRUE; + caps->hiz_ram = R300_HIZ_LIMIT; + caps->zmask_ram = PIPE_ZMASK_SIZE; break; case 0x5D4C: @@ -169,6 +188,8 @@ void r300_parse_chipset(struct r300_capabilities* caps) caps->family = CHIP_FAMILY_R480; caps->num_vert_fpus = 6; caps->is_r400 = TRUE; + caps->hiz_ram = R300_HIZ_LIMIT; + caps->zmask_ram = PIPE_ZMASK_SIZE; break; case 0x4B48: @@ -179,6 +200,8 @@ void r300_parse_chipset(struct r300_capabilities* caps) caps->family = CHIP_FAMILY_R481; caps->num_vert_fpus = 6; caps->is_r400 = TRUE; + caps->hiz_ram = R300_HIZ_LIMIT; + caps->zmask_ram = PIPE_ZMASK_SIZE; break; case 0x5E4C: @@ -196,30 +219,36 @@ void r300_parse_chipset(struct r300_capabilities* caps) caps->family = CHIP_FAMILY_RV410; caps->num_vert_fpus = 6; caps->is_r400 = TRUE; + caps->hiz_ram = R300_HIZ_LIMIT; + caps->zmask_ram = PIPE_ZMASK_SIZE; break; case 0x5954: case 0x5955: caps->family = CHIP_FAMILY_RS480; caps->has_tcl = FALSE; + caps->zmask_ram = RV3xx_ZMASK_SIZE; break; case 0x5974: case 0x5975: caps->family = CHIP_FAMILY_RS482; caps->has_tcl = FALSE; + caps->zmask_ram = RV3xx_ZMASK_SIZE; break; case 0x5A41: case 0x5A42: caps->family = CHIP_FAMILY_RS400; caps->has_tcl = FALSE; + caps->zmask_ram = RV3xx_ZMASK_SIZE; break; case 0x5A61: case 0x5A62: caps->family = CHIP_FAMILY_RC410; caps->has_tcl = FALSE; + caps->zmask_ram = RV3xx_ZMASK_SIZE; break; case 0x791E: @@ -227,6 +256,8 @@ void r300_parse_chipset(struct r300_capabilities* caps) caps->family = CHIP_FAMILY_RS690; caps->has_tcl = FALSE; caps->is_r400 = TRUE; + caps->hiz_ram = R300_HIZ_LIMIT; + caps->zmask_ram = PIPE_ZMASK_SIZE; break; case 0x793F: @@ -235,6 +266,8 @@ void r300_parse_chipset(struct r300_capabilities* caps) caps->family = CHIP_FAMILY_RS600; caps->has_tcl = FALSE; caps->is_r400 = TRUE; + caps->hiz_ram = R300_HIZ_LIMIT; + caps->zmask_ram = PIPE_ZMASK_SIZE; break; case 0x796C: @@ -244,6 +277,8 @@ void r300_parse_chipset(struct r300_capabilities* caps) caps->family = CHIP_FAMILY_RS740; caps->has_tcl = FALSE; caps->is_r400 = TRUE; + caps->hiz_ram = R300_HIZ_LIMIT; + caps->zmask_ram = PIPE_ZMASK_SIZE; break; case 0x7100: @@ -263,6 +298,8 @@ void r300_parse_chipset(struct r300_capabilities* caps) caps->family = CHIP_FAMILY_R520; caps->num_vert_fpus = 8; caps->is_r500 = TRUE; + caps->hiz_ram = R300_HIZ_LIMIT; + caps->zmask_ram = PIPE_ZMASK_SIZE; break; case 0x7140: @@ -306,6 +343,8 @@ void r300_parse_chipset(struct r300_capabilities* caps) caps->family = CHIP_FAMILY_RV515; caps->num_vert_fpus = 2; caps->is_r500 = TRUE; + caps->hiz_ram = R300_HIZ_LIMIT; + caps->zmask_ram = PIPE_ZMASK_SIZE; break; case 0x71C0: @@ -327,6 +366,8 @@ void r300_parse_chipset(struct r300_capabilities* caps) caps->family = CHIP_FAMILY_RV530; caps->num_vert_fpus = 5; caps->is_r500 = TRUE; + /*caps->hiz_ram = RV530_HIZ_LIMIT;*/ + caps->zmask_ram = PIPE_ZMASK_SIZE; break; case 0x7240: @@ -347,12 +388,16 @@ void r300_parse_chipset(struct r300_capabilities* caps) caps->family = CHIP_FAMILY_R580; caps->num_vert_fpus = 8; caps->is_r500 = TRUE; + caps->hiz_ram = RV530_HIZ_LIMIT; + caps->zmask_ram = PIPE_ZMASK_SIZE; break; case 0x7280: caps->family = CHIP_FAMILY_RV570; caps->num_vert_fpus = 8; caps->is_r500 = TRUE; + caps->hiz_ram = RV530_HIZ_LIMIT; + caps->zmask_ram = PIPE_ZMASK_SIZE; break; case 0x7281: @@ -369,6 +414,8 @@ void r300_parse_chipset(struct r300_capabilities* caps) caps->family = CHIP_FAMILY_RV560; caps->num_vert_fpus = 8; caps->is_r500 = TRUE; + caps->hiz_ram = RV530_HIZ_LIMIT; + caps->zmask_ram = PIPE_ZMASK_SIZE; break; default: diff --git a/src/gallium/drivers/r300/r300_chipset.h b/src/gallium/drivers/r300/r300_chipset.h index ab649c38573..e7ca642b4f3 100644 --- a/src/gallium/drivers/r300/r300_chipset.h +++ b/src/gallium/drivers/r300/r300_chipset.h @@ -25,6 +25,14 @@ #include "pipe/p_compiler.h" +/* these are sizes in dwords */ +#define R300_HIZ_LIMIT 10240 +#define RV530_HIZ_LIMIT 15360 + +/* rv3xx have only one pipe */ +#define PIPE_ZMASK_SIZE 4096 +#define RV3xx_ZMASK_SIZE 5120 + /* Structure containing all the possible information about a specific Radeon * in the R3xx, R4xx, and R5xx families. */ struct r300_capabilities { @@ -42,6 +50,10 @@ struct r300_capabilities { unsigned num_tex_units; /* Whether or not TCL is physically present */ boolean has_tcl; + /* Some chipsets do not have HiZ RAM - other have varying amounts . */ + int hiz_ram; + /* some chipsets have zmask ram per pipe some don't */ + int zmask_ram; /* Whether or not this is RV350 or newer, including all r400 and r500 * chipsets. The differences compared to the oldest r300 chips are: * - Blend LTE/GTE thresholds diff --git a/src/gallium/drivers/r300/r300_context.c b/src/gallium/drivers/r300/r300_context.c index 4f721ebb599..624dadd07d7 100644 --- a/src/gallium/drivers/r300/r300_context.c +++ b/src/gallium/drivers/r300/r300_context.c @@ -23,30 +23,86 @@ #include "draw/draw_context.h" #include "util/u_memory.h" +#include "util/u_sampler.h" #include "util/u_simple_list.h" #include "util/u_upload_mgr.h" -#include "r300_blit.h" +#include "r300_cb.h" #include "r300_context.h" #include "r300_emit.h" -#include "r300_flush.h" -#include "r300_query.h" -#include "r300_render.h" +#include "r300_hyperz.h" #include "r300_screen.h" #include "r300_screen_buffer.h" -#include "r300_state_invariant.h" #include "r300_winsys.h" #include <inttypes.h> +static void r300_update_num_contexts(struct r300_screen *r300screen, + int diff) +{ + if (diff > 0) { + p_atomic_inc(&r300screen->num_contexts); + + if (r300screen->num_contexts > 1) + util_mempool_set_thread_safety(&r300screen->pool_buffers, + UTIL_MEMPOOL_MULTITHREADED); + } else { + p_atomic_dec(&r300screen->num_contexts); + + if (r300screen->num_contexts <= 1) + util_mempool_set_thread_safety(&r300screen->pool_buffers, + UTIL_MEMPOOL_SINGLETHREADED); + } +} + +static void r300_release_referenced_objects(struct r300_context *r300) +{ + struct pipe_framebuffer_state *fb = + (struct pipe_framebuffer_state*)r300->fb_state.state; + struct r300_textures_state *textures = + (struct r300_textures_state*)r300->textures_state.state; + struct r300_query *query, *temp; + unsigned i; + + /* Framebuffer state. */ + util_unreference_framebuffer_state(fb); + + /* Textures. */ + for (i = 0; i < textures->sampler_view_count; i++) + pipe_sampler_view_reference( + (struct pipe_sampler_view**)&textures->sampler_views[i], NULL); + + /* The special dummy texture for texkill. */ + if (r300->texkill_sampler) { + pipe_sampler_view_reference( + (struct pipe_sampler_view**)&r300->texkill_sampler, + NULL); + } + + /* The SWTCL VBO. */ + pipe_resource_reference(&r300->vbo, NULL); + + /* Vertex buffers. */ + for (i = 0; i < r300->vertex_buffer_count; i++) { + pipe_resource_reference(&r300->vertex_buffer[i].buffer, NULL); + } + + /* If there are any queries pending or not destroyed, remove them now. */ + foreach_s(query, temp, &r300->query_list) { + remove_from_list(query); + FREE(query); + } +} + static void r300_destroy_context(struct pipe_context* context) { struct r300_context* r300 = r300_context(context); - struct r300_query* query, * temp; struct r300_atom *atom; - util_blitter_destroy(r300->blitter); - draw_destroy(r300->draw); + if (r300->blitter) + util_blitter_destroy(r300->blitter); + if (r300->draw) + draw_destroy(r300->draw); /* Print stats, if enabled. */ if (SCREEN_DBG_ON(r300->screen, DBG_STATS)) { @@ -58,33 +114,53 @@ static void r300_destroy_context(struct pipe_context* context) } } - /* Free the OQ BO. */ - context->screen->resource_destroy(context->screen, r300->oqbo); - - /* If there are any queries pending or not destroyed, remove them now. */ - foreach_s(query, temp, &r300->query_list) { - remove_from_list(query); - FREE(query); + if (r300->upload_vb) + u_upload_destroy(r300->upload_vb); + if (r300->upload_ib) + u_upload_destroy(r300->upload_ib); + + if (r300->tran.translate_cache) + translate_cache_destroy(r300->tran.translate_cache); + + /* XXX: This function assumes r300->query_list was initialized */ + r300_release_referenced_objects(r300); + + if (r300->zmask_mm) + r300_hyperz_destroy_mm(r300); + + if (r300->cs) + r300->rws->cs_destroy(r300->cs); + + /* XXX: No way to tell if this was initialized or not? */ + util_mempool_destroy(&r300->pool_transfers); + + r300_update_num_contexts(r300->screen, -1); + + /* Free the structs allocated in r300_setup_atoms() */ + if (r300->aa_state.state) { + FREE(r300->aa_state.state); + FREE(r300->blend_color_state.state); + FREE(r300->clip_state.state); + FREE(r300->fb_state.state); + FREE(r300->gpu_flush.state); + FREE(r300->hyperz_state.state); + FREE(r300->invariant_state.state); + FREE(r300->rs_block_state.state); + FREE(r300->scissor_state.state); + FREE(r300->textures_state.state); + FREE(r300->vap_invariant_state.state); + FREE(r300->viewport_state.state); + FREE(r300->ztop_state.state); + FREE(r300->fs_constants.state); + FREE(r300->vs_constants.state); + if (!r300->screen->caps.has_tcl) { + FREE(r300->vertex_stream_state.state); + } } - - u_upload_destroy(r300->upload_vb); - u_upload_destroy(r300->upload_ib); - - FREE(r300->blend_color_state.state); - FREE(r300->clip_state.state); - FREE(r300->fb_state.state); - FREE(r300->rs_block_state.state); - FREE(r300->scissor_state.state); - FREE(r300->textures_state.state); - FREE(r300->vap_output_state.state); - FREE(r300->viewport_state.state); - FREE(r300->ztop_state.state); - FREE(r300->fs_constants.state); - FREE(r300->vs_constants.state); FREE(r300); } -static void r300_flush_cb(void *data) +void r300_flush_cb(void *data) { struct r300_context* const cs_context_copy = data; @@ -92,17 +168,24 @@ static void r300_flush_cb(void *data) } #define R300_INIT_ATOM(atomname, atomsize) \ + do { \ r300->atomname.name = #atomname; \ r300->atomname.state = NULL; \ r300->atomname.size = atomsize; \ r300->atomname.emit = r300_emit_##atomname; \ r300->atomname.dirty = FALSE; \ - insert_at_tail(&r300->atom_list, &r300->atomname); + insert_at_tail(&r300->atom_list, &r300->atomname); \ + } while (0) static void r300_setup_atoms(struct r300_context* r300) { + boolean is_rv350 = r300->screen->caps.is_rv350; boolean is_r500 = r300->screen->caps.is_r500; boolean has_tcl = r300->screen->caps.has_tcl; + boolean drm_2_3_0 = r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0); + boolean drm_2_6_0 = r300->rws->get_value(r300->rws, R300_VID_DRM_2_6_0); + boolean has_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ); + boolean has_hiz_ram = r300->screen->caps.hiz_ram > 0; /* Create the actual atom list. * @@ -110,50 +193,206 @@ static void r300_setup_atoms(struct r300_context* r300) * can affect performance and conformance if not handled with care. * * Some atoms never change size, others change every emit - those have - * the size of 0 here. */ + * the size of 0 here. + * + * NOTE: The framebuffer state is split into these atoms: + * - gpu_flush (unpipelined regs) + * - aa_state (unpipelined regs) + * - fb_state (unpipelined regs) + * - hyperz_state (unpipelined regs followed by pipelined ones) + * - fb_state_pipelined (pipelined regs) + * The motivation behind this is to be able to emit a strict + * subset of the regs, and to have reasonable register ordering. */ make_empty_list(&r300->atom_list); - R300_INIT_ATOM(invariant_state, 71); - R300_INIT_ATOM(query_start, 4); + /* SC, GB (unpipelined), RB3D (unpipelined), ZB (unpipelined). */ + R300_INIT_ATOM(gpu_flush, 9); + R300_INIT_ATOM(aa_state, 4); + R300_INIT_ATOM(fb_state, 0); + R300_INIT_ATOM(hyperz_state, is_r500 || (is_rv350 && drm_2_6_0) ? 10 : 8); + /* ZB (unpipelined), SC. */ R300_INIT_ATOM(ztop_state, 2); + /* ZB, FG. */ + R300_INIT_ATOM(dsa_state, is_r500 ? 8 : 6); + /* RB3D. */ R300_INIT_ATOM(blend_state, 8); R300_INIT_ATOM(blend_color_state, is_r500 ? 3 : 2); - R300_INIT_ATOM(clip_state, has_tcl ? 5 + (6 * 4) : 2); - R300_INIT_ATOM(dsa_state, is_r500 ? 8 : 6); - R300_INIT_ATOM(fb_state, 0); - R300_INIT_ATOM(rs_state, 0); + /* SC. */ R300_INIT_ATOM(scissor_state, 3); + /* GB, FG, GA, SU, SC, RB3D. */ + R300_INIT_ATOM(invariant_state, 16 + (is_rv350 ? 4 : 0)); + /* VAP. */ R300_INIT_ATOM(viewport_state, 9); - R300_INIT_ATOM(rs_block_state, 0); - R300_INIT_ATOM(vertex_stream_state, 0); - R300_INIT_ATOM(vap_output_state, 6); R300_INIT_ATOM(pvs_flush, 2); + R300_INIT_ATOM(vap_invariant_state, 9); + R300_INIT_ATOM(vertex_stream_state, 0); R300_INIT_ATOM(vs_state, 0); R300_INIT_ATOM(vs_constants, 0); - R300_INIT_ATOM(texture_cache_inval, 2); - R300_INIT_ATOM(textures_state, 0); + R300_INIT_ATOM(clip_state, has_tcl ? 5 + (6 * 4) : 2); + /* VAP, RS, GA, GB, SU, SC. */ + R300_INIT_ATOM(rs_block_state, 0); + R300_INIT_ATOM(rs_state, 0); + /* SC, US. */ + R300_INIT_ATOM(fb_state_pipelined, 5 + (drm_2_3_0 ? 3 : 0)); + /* US. */ R300_INIT_ATOM(fs, 0); R300_INIT_ATOM(fs_rc_constant_state, 0); R300_INIT_ATOM(fs_constants, 0); + /* TX. */ + R300_INIT_ATOM(texture_cache_inval, 2); + R300_INIT_ATOM(textures_state, 0); + if (has_hyperz) { + /* HiZ Clear */ + if (has_hiz_ram) + R300_INIT_ATOM(hiz_clear, 0); + /* zmask clear */ + R300_INIT_ATOM(zmask_clear, 0); + } + /* ZB (unpipelined), SU. */ + R300_INIT_ATOM(query_start, 4); /* Replace emission functions for r500. */ - if (r300->screen->caps.is_r500) { + if (is_r500) { r300->fs.emit = r500_emit_fs; r300->fs_rc_constant_state.emit = r500_emit_fs_rc_constant_state; r300->fs_constants.emit = r500_emit_fs_constants; } /* Some non-CSO atoms need explicit space to store the state locally. */ + r300->aa_state.state = CALLOC_STRUCT(r300_aa_state); r300->blend_color_state.state = CALLOC_STRUCT(r300_blend_color_state); - r300->clip_state.state = CALLOC_STRUCT(pipe_clip_state); + r300->clip_state.state = CALLOC_STRUCT(r300_clip_state); r300->fb_state.state = CALLOC_STRUCT(pipe_framebuffer_state); + r300->gpu_flush.state = CALLOC_STRUCT(pipe_framebuffer_state); + r300->hyperz_state.state = CALLOC_STRUCT(r300_hyperz_state); + r300->invariant_state.state = CALLOC_STRUCT(r300_invariant_state); r300->rs_block_state.state = CALLOC_STRUCT(r300_rs_block); r300->scissor_state.state = CALLOC_STRUCT(pipe_scissor_state); r300->textures_state.state = CALLOC_STRUCT(r300_textures_state); - r300->vap_output_state.state = CALLOC_STRUCT(r300_vap_output_state); + r300->vap_invariant_state.state = CALLOC_STRUCT(r300_vap_invariant_state); r300->viewport_state.state = CALLOC_STRUCT(r300_viewport_state); r300->ztop_state.state = CALLOC_STRUCT(r300_ztop_state); r300->fs_constants.state = CALLOC_STRUCT(r300_constant_buffer); r300->vs_constants.state = CALLOC_STRUCT(r300_constant_buffer); + if (!r300->screen->caps.has_tcl) { + r300->vertex_stream_state.state = CALLOC_STRUCT(r300_vertex_stream_state); + } + + /* Some non-CSO atoms don't use the state pointer. */ + r300->fb_state_pipelined.allow_null_state = TRUE; + r300->fs_rc_constant_state.allow_null_state = TRUE; + r300->pvs_flush.allow_null_state = TRUE; + r300->query_start.allow_null_state = TRUE; + r300->texture_cache_inval.allow_null_state = TRUE; + + /* Some states must be marked as dirty here to properly set up + * hardware in the first command stream. */ + r300->invariant_state.dirty = TRUE; + r300->pvs_flush.dirty = TRUE; + r300->vap_invariant_state.dirty = TRUE; + r300->texture_cache_inval.dirty = TRUE; + r300->textures_state.dirty = TRUE; +} + +/* Not every state tracker calls every driver function before the first draw + * call and we must initialize the command buffers somehow. */ +static void r300_init_states(struct pipe_context *pipe) +{ + struct r300_context *r300 = r300_context(pipe); + struct pipe_blend_color bc = {{0}}; + struct pipe_clip_state cs = {{{0}}}; + struct pipe_scissor_state ss = {0}; + struct r300_clip_state *clip = + (struct r300_clip_state*)r300->clip_state.state; + struct r300_gpu_flush *gpuflush = + (struct r300_gpu_flush*)r300->gpu_flush.state; + struct r300_vap_invariant_state *vap_invariant = + (struct r300_vap_invariant_state*)r300->vap_invariant_state.state; + struct r300_invariant_state *invariant = + (struct r300_invariant_state*)r300->invariant_state.state; + + CB_LOCALS; + + pipe->set_blend_color(pipe, &bc); + pipe->set_scissor_state(pipe, &ss); + + /* Initialize the clip state. */ + if (r300_context(pipe)->screen->caps.has_tcl) { + pipe->set_clip_state(pipe, &cs); + } else { + BEGIN_CB(clip->cb, 2); + OUT_CB_REG(R300_VAP_CLIP_CNTL, R300_CLIP_DISABLE); + END_CB; + } + + /* Initialize the GPU flush. */ + { + BEGIN_CB(gpuflush->cb_flush_clean, 6); + + /* Flush and free renderbuffer caches. */ + OUT_CB_REG(R300_RB3D_DSTCACHE_CTLSTAT, + R300_RB3D_DSTCACHE_CTLSTAT_DC_FREE_FREE_3D_TAGS | + R300_RB3D_DSTCACHE_CTLSTAT_DC_FLUSH_FLUSH_DIRTY_3D); + OUT_CB_REG(R300_ZB_ZCACHE_CTLSTAT, + R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE | + R300_ZB_ZCACHE_CTLSTAT_ZC_FREE_FREE); + + /* Wait until the GPU is idle. + * This fixes random pixels sometimes appearing probably caused + * by incomplete rendering. */ + OUT_CB_REG(RADEON_WAIT_UNTIL, RADEON_WAIT_3D_IDLECLEAN); + END_CB; + } + + /* Initialize the VAP invariant state. */ + { + BEGIN_CB(vap_invariant->cb, 9); + OUT_CB_REG(VAP_PVS_VTX_TIMEOUT_REG, 0xffff); + OUT_CB_REG_SEQ(R300_VAP_GB_VERT_CLIP_ADJ, 4); + OUT_CB_32F(1.0); + OUT_CB_32F(1.0); + OUT_CB_32F(1.0); + OUT_CB_32F(1.0); + OUT_CB_REG(R300_VAP_PSC_SGN_NORM_CNTL, R300_SGN_NORM_NO_ZERO); + END_CB; + } + + /* Initialize the invariant state. */ + { + BEGIN_CB(invariant->cb, r300->invariant_state.size); + OUT_CB_REG(R300_GB_SELECT, 0); + OUT_CB_REG(R300_FG_FOG_BLEND, 0); + OUT_CB_REG(R300_GA_ROUND_MODE, 1); + OUT_CB_REG(R300_GA_OFFSET, 0); + OUT_CB_REG(R300_SU_TEX_WRAP, 0); + OUT_CB_REG(R300_SU_DEPTH_SCALE, 0x4B7FFFFF); + OUT_CB_REG(R300_SU_DEPTH_OFFSET, 0); + OUT_CB_REG(R300_SC_EDGERULE, 0x2DA49525); + + if (r300->screen->caps.is_rv350) { + OUT_CB_REG(R500_RB3D_DISCARD_SRC_PIXEL_LTE_THRESHOLD, 0x01010101); + OUT_CB_REG(R500_RB3D_DISCARD_SRC_PIXEL_GTE_THRESHOLD, 0xFEFEFEFE); + } + END_CB; + } + + /* Initialize the hyperz state. */ + { + struct r300_hyperz_state *hyperz = + (struct r300_hyperz_state*)r300->hyperz_state.state; + BEGIN_CB(&hyperz->cb_flush_begin, r300->hyperz_state.size); + OUT_CB_REG(R300_ZB_ZCACHE_CTLSTAT, + R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE); + OUT_CB_REG(R300_ZB_BW_CNTL, 0); + OUT_CB_REG(R300_ZB_DEPTHCLEARVALUE, 0); + OUT_CB_REG(R300_SC_HYPERZ, R300_SC_HYPERZ_ADJ_2); + + if (r300->screen->caps.is_r500 || + (r300->screen->caps.is_rv350 && + r300->rws->get_value(r300->rws, R300_VID_DRM_2_6_0))) { + OUT_CB_REG(R300_GB_Z_PEQ_CONFIG, 0); + } + END_CB; + } } struct pipe_context* r300_create_context(struct pipe_screen* screen, @@ -166,6 +405,8 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen, if (!r300) return NULL; + r300_update_num_contexts(r300screen, 1); + r300->rws = rws; r300->screen = r300screen; @@ -175,77 +416,121 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen, r300->context.destroy = r300_destroy_context; - r300->context.clear = r300_clear; - r300->context.surface_copy = r300_surface_copy; - r300->context.surface_fill = r300_surface_fill; - - if (r300screen->caps.has_tcl) { - r300->context.draw_arrays = r300_draw_arrays; - r300->context.draw_elements = r300_draw_elements; - r300->context.draw_range_elements = r300_draw_range_elements; - - if (r300screen->caps.is_r500) { - r300->emit_draw_arrays_immediate = r500_emit_draw_arrays_immediate; - r300->emit_draw_arrays = r500_emit_draw_arrays; - r300->emit_draw_elements = r500_emit_draw_elements; - } else { - r300->emit_draw_arrays_immediate = r300_emit_draw_arrays_immediate; - r300->emit_draw_arrays = r300_emit_draw_arrays; - r300->emit_draw_elements = r300_emit_draw_elements; - } - } else { - r300->context.draw_arrays = r300_swtcl_draw_arrays; - r300->context.draw_elements = r300_draw_elements; - r300->context.draw_range_elements = r300_swtcl_draw_range_elements; + make_empty_list(&r300->query_list); + + util_mempool_create(&r300->pool_transfers, + sizeof(struct pipe_transfer), 64, + UTIL_MEMPOOL_SINGLETHREADED); + + r300->cs = rws->cs_create(rws); + if (r300->cs == NULL) + goto fail; + if (!r300screen->caps.has_tcl) { /* Create a Draw. This is used for SW TCL. */ r300->draw = draw_create(&r300->context); /* Enable our renderer. */ draw_set_rasterize_stage(r300->draw, r300_draw_stage(r300)); - /* Enable Draw's clipping. */ - draw_set_driver_clipping(r300->draw, FALSE); - /* Force Draw to never do viewport transform, since we can do - * transform in hardware, always. */ - draw_set_viewport_state(r300->draw, &r300_viewport_identity); + /* Disable converting points/lines to triangles. */ + draw_wide_line_threshold(r300->draw, 10000000.f); + draw_wide_point_threshold(r300->draw, 10000000.f); } r300_setup_atoms(r300); - /* Open up the OQ BO. */ - r300->oqbo = pipe_buffer_create(screen, - R300_BIND_OQBO, 4096); - make_empty_list(&r300->query_list); - + r300_init_blit_functions(r300); r300_init_flush_functions(r300); r300_init_query_functions(r300); r300_init_state_functions(r300); r300_init_resource_functions(r300); - r300->invariant_state.dirty = TRUE; + r300->blitter = util_blitter_create(&r300->context); + if (r300->blitter == NULL) + goto fail; - rws->set_flush_cb(r300->rws, r300_flush_cb, r300); - r300->dirty_hw++; + /* Render functions must be initialized after blitter. */ + r300_init_render_functions(r300); - r300->blitter = util_blitter_create(&r300->context); + rws->cs_set_flush(r300->cs, r300_flush_cb, r300); + + /* setup hyper-z mm */ + if (r300->rws->get_value(r300->rws, R300_CAN_HYPERZ)) + if (!r300_hyperz_init_mm(r300)) + goto fail; r300->upload_ib = u_upload_create(&r300->context, 32 * 1024, 16, PIPE_BIND_INDEX_BUFFER); if (r300->upload_ib == NULL) - goto no_upload_ib; + goto fail; r300->upload_vb = u_upload_create(&r300->context, 128 * 1024, 16, PIPE_BIND_VERTEX_BUFFER); if (r300->upload_vb == NULL) - goto no_upload_vb; + goto fail; + + r300->tran.translate_cache = translate_cache_create(); + if (r300->tran.translate_cache == NULL) + goto fail; + + r300_init_states(&r300->context); + + /* The KIL opcode needs the first texture unit to be enabled + * on r3xx-r4xx. In order to calm down the CS checker, we bind this + * dummy texture there. */ + if (!r300->screen->caps.is_r500) { + struct pipe_resource *tex; + struct pipe_resource rtempl = {{0}}; + struct pipe_sampler_view vtempl = {{0}}; + + rtempl.target = PIPE_TEXTURE_2D; + rtempl.format = PIPE_FORMAT_I8_UNORM; + rtempl.bind = PIPE_BIND_SAMPLER_VIEW; + rtempl.width0 = 1; + rtempl.height0 = 1; + rtempl.depth0 = 1; + tex = screen->resource_create(screen, &rtempl); + + u_sampler_view_default_template(&vtempl, tex, tex->format); + + r300->texkill_sampler = (struct r300_sampler_view*) + r300->context.create_sampler_view(&r300->context, tex, &vtempl); + + pipe_resource_reference(&tex, NULL); + } return &r300->context; - no_upload_ib: - u_upload_destroy(r300->upload_ib); - no_upload_vb: - FREE(r300); + fail: + r300_destroy_context(&r300->context); return NULL; } + +void r300_finish(struct r300_context *r300) +{ + struct pipe_framebuffer_state *fb; + unsigned i; + + /* This is a preliminary implementation of glFinish. + * + * The ideal implementation should use something like EmitIrqLocked and + * WaitIrq, or better, real fences. + */ + if (r300->fb_state.state) { + fb = r300->fb_state.state; + + for (i = 0; i < fb->nr_cbufs; i++) { + if (fb->cbufs[i]->texture) { + r300->rws->buffer_wait(r300->rws, + r300_texture(fb->cbufs[i]->texture)->buffer); + return; + } + } + if (fb->zsbuf && fb->zsbuf->texture) { + r300->rws->buffer_wait(r300->rws, + r300_texture(fb->zsbuf->texture)->buffer); + } + } +} diff --git a/src/gallium/drivers/r300/r300_context.h b/src/gallium/drivers/r300/r300_context.h index b6e20591eda..b59bc002610 100644 --- a/src/gallium/drivers/r300/r300_context.h +++ b/src/gallium/drivers/r300/r300_context.h @@ -31,14 +31,16 @@ #include "util/u_inlines.h" #include "util/u_transfer.h" +#include "translate/translate_cache.h" + #include "r300_defines.h" #include "r300_screen.h" struct u_upload_mgr; struct r300_context; - struct r300_fragment_shader; struct r300_vertex_shader; +struct r300_stencilref_context; struct r300_atom { /* List pointers. */ @@ -55,70 +57,103 @@ struct r300_atom { unsigned size; /* Whether this atom should be emitted. */ boolean dirty; + /* Whether this atom may be emitted with state == NULL. */ + boolean allow_null_state; +}; + +struct r300_aa_state { + struct r300_surface *dest; + + uint32_t aa_config; + uint32_t aaresolve_ctl; }; struct r300_blend_state { - uint32_t blend_control; /* R300_RB3D_CBLEND: 0x4e04 */ - uint32_t alpha_blend_control; /* R300_RB3D_ABLEND: 0x4e08 */ - uint32_t color_channel_mask; /* R300_RB3D_COLOR_CHANNEL_MASK: 0x4e0c */ - uint32_t rop; /* R300_RB3D_ROPCNTL: 0x4e18 */ - uint32_t dither; /* R300_RB3D_DITHER_CTL: 0x4e50 */ + uint32_t cb[8]; + uint32_t cb_no_readwrite[8]; }; struct r300_blend_color_state { - /* RV515 and earlier */ - uint32_t blend_color; /* R300_RB3D_BLEND_COLOR: 0x4e10 */ - /* R520 and newer */ - uint32_t blend_color_red_alpha; /* R500_RB3D_CONSTANT_COLOR_AR: 0x4ef8 */ - uint32_t blend_color_green_blue; /* R500_RB3D_CONSTANT_COLOR_GB: 0x4efc */ + uint32_t cb[3]; +}; + +struct r300_clip_state { + struct pipe_clip_state clip; + + uint32_t cb[29]; }; struct r300_dsa_state { + struct pipe_depth_stencil_alpha_state dsa; + + /* This is actually a command buffer with named dwords. */ + uint32_t cb_begin; uint32_t alpha_function; /* R300_FG_ALPHA_FUNC: 0x4bd4 */ - uint32_t alpha_reference; /* R500_FG_ALPHA_VALUE: 0x4be0 */ + uint32_t cb_reg_seq; uint32_t z_buffer_control; /* R300_ZB_CNTL: 0x4f00 */ uint32_t z_stencil_control; /* R300_ZB_ZSTENCILCNTL: 0x4f04 */ uint32_t stencil_ref_mask; /* R300_ZB_STENCILREFMASK: 0x4f08 */ + uint32_t cb_reg; uint32_t stencil_ref_bf; /* R500_ZB_STENCILREFMASK_BF: 0x4fd4 */ + /* The second command buffer disables zbuffer reads and writes. */ + uint32_t cb_no_readwrite[8]; + /* Whether a two-sided stencil is enabled. */ boolean two_sided; /* Whether a fallback should be used for a two-sided stencil ref value. */ - boolean stencil_ref_bf_fallback; + boolean two_sided_stencil_ref; }; +struct r300_hyperz_state { + int current_func; /* -1 after a clear before first op */ + int flush; + /* This is actually a command buffer with named dwords. */ + uint32_t cb_flush_begin; + uint32_t zb_zcache_ctlstat; /* R300_ZB_CACHE_CNTL */ + uint32_t cb_begin; + uint32_t zb_bw_cntl; /* R300_ZB_BW_CNTL */ + uint32_t cb_reg1; + uint32_t zb_depthclearvalue; /* R300_ZB_DEPTHCLEARVALUE */ + uint32_t cb_reg2; + uint32_t sc_hyperz; /* R300_SC_HYPERZ */ + uint32_t cb_reg3; + uint32_t gb_z_peq_config; /* R300_GB_Z_PEQ_CONFIG: 0x4028 */ +}; + +struct r300_gpu_flush { + uint32_t cb_flush_clean[6]; +}; + +#define RS_STATE_MAIN_SIZE 23 + struct r300_rs_state { - /* Draw-specific rasterizer state */ + /* Original rasterizer state. */ struct pipe_rasterizer_state rs; + /* Draw-specific rasterizer state. */ + struct pipe_rasterizer_state rs_draw; + + /* Command buffers. */ + uint32_t cb_main[RS_STATE_MAIN_SIZE]; + uint32_t cb_poly_offset_zb16[5]; + uint32_t cb_poly_offset_zb24[5]; + + /* The index to cb_main where the cull_mode register value resides. */ + unsigned cull_mode_index; - uint32_t vap_control_status; /* R300_VAP_CNTL_STATUS: 0x2140 */ - uint32_t antialiasing_config; /* R300_GB_AA_CONFIG: 0x4020 */ - uint32_t point_size; /* R300_GA_POINT_SIZE: 0x421c */ - uint32_t line_control; /* R300_GA_LINE_CNTL: 0x4234 */ - float depth_scale; /* R300_SU_POLY_OFFSET_FRONT_SCALE: 0x42a4 */ - /* R300_SU_POLY_OFFSET_BACK_SCALE: 0x42ac */ - float depth_offset; /* R300_SU_POLY_OFFSET_FRONT_OFFSET: 0x42a8 */ - /* R300_SU_POLY_OFFSET_BACK_OFFSET: 0x42b0 */ - uint32_t polygon_offset_enable; /* R300_SU_POLY_OFFSET_ENABLE: 0x42b4 */ - uint32_t cull_mode; /* R300_SU_CULL_MODE: 0x42b8 */ - uint32_t line_stipple_config; /* R300_GA_LINE_STIPPLE_CONFIG: 0x4328 */ - uint32_t line_stipple_value; /* R300_GA_LINE_STIPPLE_VALUE: 0x4260 */ + /* Whether polygon offset is enabled. */ + boolean polygon_offset_enable; + + /* This is emitted in the draw function. */ uint32_t color_control; /* R300_GA_COLOR_CONTROL: 0x4278 */ - uint32_t polygon_mode; /* R300_GA_POLY_MODE: 0x4288 */ - uint32_t clip_rule; /* R300_SC_CLIP_RULE: 0x43D0 */ - - /* Specifies top of Raster pipe specific enable controls, - * i.e. texture coordinates stuffing for points, lines, triangles */ - uint32_t stuffing_enable; /* R300_GB_ENABLE: 0x4008 */ - - /* Point sprites texture coordinates, 0: lower left, 1: upper right */ - float point_texcoord_left; /* R300_GA_POINT_S0: 0x4200 */ - float point_texcoord_bottom; /* R300_GA_POINT_T0: 0x4204 */ - float point_texcoord_right; /* R300_GA_POINT_S1: 0x4208 */ - float point_texcoord_top; /* R300_GA_POINT_T1: 0x420c */ }; struct r300_rs_block { + uint32_t vap_vtx_state_cntl; /* R300_VAP_VTX_STATE_CNTL: 0x2180 */ + uint32_t vap_vsm_vtx_assm; /* R300_VAP_VSM_VTX_ASSM: 0x2184 */ + uint32_t vap_out_vtx_fmt[2]; /* R300_VAP_OUTPUT_VTX_FMT_[0-1]: 0x2090 */ + uint32_t gb_enable; + uint32_t ip[8]; /* R300_RS_IP_[0-7], R500_RS_IP_[0-7] */ uint32_t count; /* R300_RS_COUNT */ uint32_t inst_count; /* R300_RS_INST_COUNT */ @@ -130,7 +165,6 @@ struct r300_sampler_state { uint32_t filter0; /* R300_TX_FILTER0: 0x4400 */ uint32_t filter1; /* R300_TX_FILTER1: 0x4440 */ - uint32_t border_color; /* R300_TX_BORDER_COLOR: 0x45c0 */ /* Min/max LOD must be clamped to [0, last_level], thus * it's dependent on a currently bound texture */ @@ -147,19 +181,28 @@ struct r300_texture_format_state { struct r300_sampler_view { struct pipe_sampler_view base; + /* Swizzles in the UTIL_FORMAT_SWIZZLE_* representation, + * derived from base. */ + unsigned char swizzle[4]; + /* Copy of r300_texture::texture_format_state with format-specific bits * added. */ struct r300_texture_format_state format; + + /* The texture cache region for this texture. */ + uint32_t texcache_region; }; struct r300_texture_fb_state { - /* Colorbuffer. */ - uint32_t colorpitch[R300_MAX_TEXTURE_LEVELS]; /* R300_RB3D_COLORPITCH[0-3]*/ - uint32_t us_out_fmt; /* R300_US_OUT_FMT[0-3] */ + uint32_t pitch[R300_MAX_TEXTURE_LEVELS]; /* COLORPITCH or DEPTHPITCH. */ + uint32_t format; /* US_OUT_FMT or R300_ZB_FORMAT */ +}; - /* Zbuffer. */ - uint32_t depthpitch[R300_MAX_TEXTURE_LEVELS]; /* R300_RB3D_DEPTHPITCH */ - uint32_t zb_format; /* R300_ZB_FORMAT */ +struct r300_texture_sampler_state { + struct r300_texture_format_state format; + uint32_t filter0; /* R300_TX_FILTER0: 0x4400 */ + uint32_t filter1; /* R300_TX_FILTER1: 0x4440 */ + uint32_t border_color; /* R300_TX_BORDER_COLOR: 0x45c0 */ }; struct r300_textures_state { @@ -173,12 +216,7 @@ struct r300_textures_state { /* This is the merge of the texture and sampler states. */ unsigned count; uint32_t tx_enable; /* R300_TX_ENABLE: 0x4101 */ - struct r300_texture_sampler_state { - struct r300_texture_format_state format; - uint32_t filter0; /* R300_TX_FILTER0: 0x4400 */ - uint32_t filter1; /* R300_TX_FILTER1: 0x4440 */ - uint32_t border_color; /* R300_TX_BORDER_COLOR: 0x45c0 */ - } regs[16]; + struct r300_texture_sampler_state regs[16]; }; struct r300_vertex_stream_state { @@ -190,10 +228,12 @@ struct r300_vertex_stream_state { unsigned count; }; -struct r300_vap_output_state { - uint32_t vap_vtx_state_cntl; /* R300_VAP_VTX_STATE_CNTL: 0x2180 */ - uint32_t vap_vsm_vtx_assm; /* R300_VAP_VSM_VTX_ASSM: 0x2184 */ - uint32_t vap_out_vtx_fmt[2]; /* R300_VAP_OUTPUT_VTX_FMT_[0-1]: 0x2090 */ +struct r300_invariant_state { + uint32_t cb[20]; +}; + +struct r300_vap_invariant_state { + uint32_t cb[9]; }; struct r300_viewport_state { @@ -215,9 +255,9 @@ struct r300_ztop_state { struct r300_constant_buffer { /* Buffer of constants */ - float constants[256][4]; - /* Total number of constants */ - unsigned count; + uint32_t *ptr; + /* Remapping table. */ + unsigned *remap_table; }; /* Query object. @@ -230,38 +270,102 @@ struct r300_constant_buffer { struct r300_query { /* The kind of query. Currently only OQ is supported. */ unsigned type; - /* Whether this query is currently active. Only active queries will - * get emitted into the command stream, and only active queries get - * tallied. */ - boolean active; - /* The current count of this query. Required to be at least 32 bits. */ - unsigned int count; - /* The offset of this query into the query buffer, in bytes. */ - unsigned offset; + /* The number of pipes where query results are stored. */ + unsigned num_pipes; + /* How many results have been written, in dwords. It's incremented + * after end_query and flush. */ + unsigned num_results; /* if we've flushed the query */ boolean flushed; /* if begin has been emitted */ boolean begin_emitted; + + /* The buffer where query results are stored. */ + struct r300_winsys_buffer *buffer; + /* The size of the buffer. */ + unsigned buffer_size; + /* The domain of the buffer. */ + enum r300_buffer_domain domain; + /* Linked list members. */ struct r300_query* prev; struct r300_query* next; }; -struct r300_texture { - /* Parent class */ +/* Fence object. + * + * This is a fake fence. Instead of syncing with the fence, we sync + * with the context, which is inefficient but compliant. + * + * This is not a subclass of pipe_fence_handle because pipe_fence_handle is + * never actually fully defined. So, rather than have it as a member, and do + * subclass-style casting, we treat pipe_fence_handle as an opaque, and just + * trust that our state tracker does not ever mess up fence objects. + */ +struct r300_fence { + struct pipe_reference reference; + struct r300_context *ctx; + boolean signalled; +}; + +struct r300_surface { + struct pipe_surface base; + + /* Winsys buffer backing the texture. */ + struct r300_winsys_buffer *buffer; + + enum r300_buffer_domain domain; + + uint32_t offset; /* COLOROFFSET or DEPTHOFFSET. */ + uint32_t pitch; /* COLORPITCH or DEPTHPITCH. */ + uint32_t format; /* US_OUT_FMT or ZB_FORMAT. */ + + /* Parameters dedicated to the CBZB clear. */ + uint32_t cbzb_width; /* Aligned width. */ + uint32_t cbzb_height; /* Half of the height. */ + uint32_t cbzb_midpoint_offset; /* DEPTHOFFSET. */ + uint32_t cbzb_pitch; /* DEPTHPITCH. */ + uint32_t cbzb_format; /* ZB_FORMAT. */ + + /* Whether the CBZB clear is allowed on the surface. */ + boolean cbzb_allowed; + +}; + +struct r300_texture_desc { + /* Parent class. */ struct u_resource b; + /* Width, height, and depth. + * Most of the time, these are equal to pipe_texture::width0, height0, + * and depth0. However, NPOT 3D textures must have dimensions aligned + * to POT, and this is the only case when these variables differ from + * pipe_texture. */ + unsigned width0, height0, depth0; + + /* Buffer tiling. + * Macrotiling is specified per-level because small mipmaps cannot + * be macrotiled. */ + enum r300_buffer_tiling microtile; + enum r300_buffer_tiling macrotile[R300_MAX_TEXTURE_LEVELS]; + /* Offsets into the buffer. */ - unsigned offset[R300_MAX_TEXTURE_LEVELS]; + unsigned offset_in_bytes[R300_MAX_TEXTURE_LEVELS]; - /* A pitch for each mip-level */ - unsigned pitch[R300_MAX_TEXTURE_LEVELS]; + /* Strides for each mip-level. */ + unsigned stride_in_pixels[R300_MAX_TEXTURE_LEVELS]; + unsigned stride_in_bytes[R300_MAX_TEXTURE_LEVELS]; - /* Size of one zslice or face based on the texture target */ - unsigned layer_size[R300_MAX_TEXTURE_LEVELS]; + /* Size of one zslice or face or 2D image based on the texture target. */ + unsigned layer_size_in_bytes[R300_MAX_TEXTURE_LEVELS]; - /* Whether the mipmap level is macrotiled. */ - enum r300_buffer_tiling mip_macrotile[R300_MAX_TEXTURE_LEVELS]; + /* Total size of this texture, in bytes, + * derived from the texture properties. */ + unsigned size_in_bytes; + + /* Total size of the buffer backing this texture, in bytes. + * It must be >= size. */ + unsigned buffer_size_in_bytes; /** * If non-zero, override the natural texture layout with @@ -271,16 +375,24 @@ struct r300_texture { * * \sa r300_texture_get_stride */ - unsigned stride_override; + unsigned stride_in_bytes_override; - /* Total size of this texture, in bytes. */ - unsigned size; + /* Whether this texture has non-power-of-two dimensions. + * It can be either a regular texture or a rectangle one. */ + boolean is_npot; - /* Whether this texture has non-power-of-two dimensions - * or a user-specified pitch. - * It can be either a regular texture or a rectangle one. - */ - boolean uses_pitch; + /* This flag says that hardware must use the stride for addressing + * instead of the width. */ + boolean uses_stride_addressing; + + /* Whether CBZB fast color clear is allowed on the miplevel. */ + boolean cbzb_allowed[R300_MAX_TEXTURE_LEVELS]; +}; + +struct r300_texture { + struct r300_texture_desc desc; + + enum r300_buffer_domain domain; /* Pipe buffer backing this texture. */ struct r300_winsys_buffer *buffer; @@ -291,72 +403,95 @@ struct r300_texture { /* All bits should be filled in. */ struct r300_texture_fb_state fb_state; - /* Buffer tiling */ - enum r300_buffer_tiling microtile, macrotile; -}; + /* hyper-z memory allocs */ + struct mem_block *hiz_mem[R300_MAX_TEXTURE_LEVELS]; + struct mem_block *zmask_mem[R300_MAX_TEXTURE_LEVELS]; + boolean zmask_in_use[R300_MAX_TEXTURE_LEVELS]; + boolean hiz_in_use[R300_MAX_TEXTURE_LEVELS]; -struct r300_vertex_info { - /* Parent class */ - struct vertex_info vinfo; - - /* R300_VAP_PROG_STREAK_CNTL_[0-7] */ - uint32_t vap_prog_stream_cntl[8]; - /* R300_VAP_PROG_STREAK_CNTL_EXT_[0-7] */ - uint32_t vap_prog_stream_cntl_ext[8]; + /* This is the level tiling flags were last time set for. + * It's used to prevent redundant tiling-flags changes from happening.*/ + unsigned surface_level; }; struct r300_vertex_element_state { unsigned count; struct pipe_vertex_element velem[PIPE_MAX_ATTRIBS]; + /* If (velem[i].src_format != hw_format[i]), the vertex buffer + * referenced by this vertex element cannot be used for rendering and + * its vertex data must be translated to hw_format[i]. */ + enum pipe_format hw_format[PIPE_MAX_ATTRIBS]; + unsigned hw_format_size[PIPE_MAX_ATTRIBS]; + + /* The size of the vertex, in dwords. */ + unsigned vertex_size_dwords; + + /* This might mean two things: + * - src_format != hw_format, as discussed above. + * - src_offset % 4 != 0. */ + boolean incompatible_layout; + struct r300_vertex_stream_state vertex_stream; }; -extern struct pipe_viewport_state r300_viewport_identity; +struct r300_translate_context { + /* Translate cache for incompatible vertex offset/stride/format fallback. */ + struct translate_cache *translate_cache; + + /* The vertex buffer slot containing the translated buffer. */ + unsigned vb_slot; + + /* Saved and new vertex element state. */ + void *saved_velems, *new_velems; +}; struct r300_context { /* Parent class */ struct pipe_context context; - /* Emission of drawing packets. */ - void (*emit_draw_arrays_immediate)( - struct r300_context *r300, - unsigned mode, unsigned start, unsigned count); - - void (*emit_draw_arrays)( - struct r300_context *r300, - unsigned mode, unsigned count); - - void (*emit_draw_elements)( - struct r300_context *r300, struct pipe_resource* indexBuffer, - unsigned indexSize, int indexBias, - unsigned minIndex, unsigned maxIndex, - unsigned mode, unsigned start, unsigned count); - - /* The interface to the windowing system, etc. */ struct r300_winsys_screen *rws; + /* The command stream. */ + struct r300_winsys_cs *cs; /* Screen. */ struct r300_screen *screen; + /* Draw module. Used mostly for SW TCL. */ struct draw_context* draw; + /* Vertex buffer for SW TCL. */ + struct pipe_resource* vbo; + /* Offset and size into the SW TCL VBO. */ + size_t draw_vbo_offset; + size_t draw_vbo_size; + /* Whether the VBO must not be flushed. */ + boolean draw_vbo_locked; + boolean draw_first_emitted; + /* Accelerated blit support. */ struct blitter_context* blitter; + /* Stencil two-sided reference value fallback. */ + struct r300_stencilref_context *stencilref_fallback; + /* For translating vertex buffers having incompatible vertex layout. */ + struct r300_translate_context tran; - /* Vertex buffer for rendering. */ - struct pipe_resource* vbo; - /* Offset into the VBO. */ - size_t vbo_offset; + /* The KIL opcode needs the first texture unit to be enabled + * on r3xx-r4xx. In order to calm down the CS checker, we bind this + * dummy texture there. */ + struct r300_sampler_view *texkill_sampler; - /* Occlusion query buffer. */ - struct pipe_resource* oqbo; - /* Query list. */ + /* The currently active query. */ struct r300_query *query_current; + /* The saved query for blitter operations. */ + struct r300_query *blitter_saved_query; + /* Query list. */ struct r300_query query_list; /* Various CSO state objects. */ /* Beginning of atom list. */ struct r300_atom atom_list; + /* Anti-aliasing (MSAA) state. */ + struct r300_atom aa_state; /* Blend state. */ struct r300_atom blend_state; /* Blend color state. */ @@ -373,11 +508,15 @@ struct r300_context { struct r300_atom fs_constants; /* Framebuffer state. */ struct r300_atom fb_state; + /* Framebuffer state (pipelined regs). */ + struct r300_atom fb_state_pipelined; + /* HyperZ state (various SC/ZB bits). */ + struct r300_atom hyperz_state; /* Occlusion query. */ struct r300_atom query_start; /* Rasterizer state. */ struct r300_atom rs_state; - /* RS block state. */ + /* RS block state + VAP (vertex shader) output mapping state. */ struct r300_atom rs_block_state; /* Scissor state. */ struct r300_atom scissor_state; @@ -385,8 +524,6 @@ struct r300_context { struct r300_atom textures_state; /* Vertex stream formatting state. */ struct r300_atom vertex_stream_state; - /* VAP (vertex shader) output mapping state. */ - struct r300_atom vap_output_state; /* Vertex shader. */ struct r300_atom vs_state; /* Vertex shader constant buffer. */ @@ -397,8 +534,16 @@ struct r300_context { struct r300_atom ztop_state; /* PVS flush. */ struct r300_atom pvs_flush; + /* VAP invariant state. */ + struct r300_atom vap_invariant_state; /* Texture cache invalidate. */ struct r300_atom texture_cache_inval; + /* GPU flush. */ + struct r300_atom gpu_flush; + /* HiZ clear */ + struct r300_atom hiz_clear; + /* zmask clear */ + struct r300_atom zmask_clear; /* Invariant state. This must be emitted to get the engine started. */ struct r300_atom invariant_state; @@ -411,15 +556,17 @@ struct r300_context { struct r300_vertex_element_state *velems; bool any_user_vbs; + struct pipe_index_buffer index_buffer; + /* Vertex info for Draw. */ struct vertex_info vertex_info; struct pipe_stencil_ref stencil_ref; - - struct pipe_clip_state clip; - struct pipe_viewport_state viewport; + /* Stream locations for SWTCL. */ + int stream_loc_notcl[16]; + /* Flag indicating whether or not the HW is dirty. */ uint32_t dirty_hw; /* Whether polygon offset is enabled. */ @@ -428,21 +575,43 @@ struct r300_context { uint32_t zbuffer_bpp; /* Whether rendering is conditional and should be skipped. */ boolean skip_rendering; - /* Whether the two-sided stencil ref value is different for front and - * back faces, and fallback should be used for r3xx-r4xx. */ - boolean stencil_ref_bf_fallback; /* Point sprites texcoord index, 1 bit per texcoord */ int sprite_coord_enable; + /* Whether two-sided color selection is enabled (AKA light_twoside). */ + boolean two_sided_color; + /* Incompatible vertex buffer layout? (misaligned stride or buffer_offset) */ + boolean incompatible_vb_layout; +#define R300_Z_COMPRESS_44 1 +#define RV350_Z_COMPRESS_88 2 + int z_compression; + boolean cbzb_clear; + boolean z_decomp_rd; + + /* two mem block managers for hiz/zmask ram space */ + struct mem_block *hiz_mm; + struct mem_block *zmask_mm; /* upload managers */ struct u_upload_mgr *upload_vb; struct u_upload_mgr *upload_ib; + struct util_mempool pool_transfers; + /* Stat counter. */ uint64_t flush_counter; }; -/* Convenience cast wrapper. */ +/* Convenience cast wrappers. */ +static INLINE struct r300_query* r300_query(struct pipe_query* q) +{ + return (struct r300_query*)q; +} + +static INLINE struct r300_surface* r300_surface(struct pipe_surface* surf) +{ + return (struct r300_surface*)surf; +} + static INLINE struct r300_texture* r300_texture(struct pipe_resource* tex) { return (struct r300_texture*)tex; @@ -461,11 +630,60 @@ static INLINE struct r300_fragment_shader *r300_fs(struct r300_context *r300) struct pipe_context* r300_create_context(struct pipe_screen* screen, void *priv); +void r300_finish(struct r300_context *r300); +void r300_flush_cb(void *data); + /* Context initialization. */ struct draw_stage* r300_draw_stage(struct r300_context* r300); +void r300_init_blit_functions(struct r300_context *r300); +void r300_init_flush_functions(struct r300_context* r300); +void r300_init_query_functions(struct r300_context* r300); +void r300_init_render_functions(struct r300_context *r300); void r300_init_state_functions(struct r300_context* r300); void r300_init_resource_functions(struct r300_context* r300); +/* r300_blit.c */ +void r300_flush_depth_stencil(struct pipe_context *pipe, + struct pipe_resource *dst, + struct pipe_subresource subdst, + unsigned zslice); + +/* r300_query.c */ +void r300_resume_query(struct r300_context *r300, + struct r300_query *query); +void r300_stop_query(struct r300_context *r300); + +/* r300_render_translate.c */ +void r300_begin_vertex_translate(struct r300_context *r300); +void r300_end_vertex_translate(struct r300_context *r300); +void r300_translate_index_buffer(struct r300_context *r300, + struct pipe_resource **index_buffer, + unsigned *index_size, unsigned index_offset, + unsigned *start, unsigned count); + +/* r300_render_stencilref.c */ +void r300_plug_in_stencil_ref_fallback(struct r300_context *r300); + +/* r300_render.c */ +void r300_draw_flush_vbuf(struct r300_context *r300); +boolean r500_index_bias_supported(struct r300_context *r300); +void r500_emit_index_bias(struct r300_context *r300, int index_bias); + +/* r300_state.c */ +enum r300_fb_state_change { + R300_CHANGED_FB_STATE = 0, + R300_CHANGED_CBZB_FLAG, + R300_CHANGED_ZCLEAR_FLAG +}; + +void r300_mark_fb_state_dirty(struct r300_context *r300, + enum r300_fb_state_change change); +void r300_mark_fs_code_dirty(struct r300_context *r300); + +/* r300_debug.c */ +void r500_dump_rs_block(struct r300_rs_block *rs); + + static INLINE boolean CTX_DBG_ON(struct r300_context * ctx, unsigned flags) { return SCREEN_DBG_ON(ctx->screen, flags); diff --git a/src/gallium/drivers/r300/r300_cs.h b/src/gallium/drivers/r300/r300_cs.h index 996a4f491e7..c194d6a1b08 100644 --- a/src/gallium/drivers/r300/r300_cs.h +++ b/src/gallium/drivers/r300/r300_cs.h @@ -20,176 +20,124 @@ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE * USE OR OTHER DEALINGS IN THE SOFTWARE. */ +/** + * This file contains macros for immediate command submission. + */ + #ifndef R300_CS_H #define R300_CS_H -#include "util/u_math.h" - #include "r300_reg.h" +#include "r300_context.h" #include "r300_winsys.h" /* Yes, I know macros are ugly. However, they are much prettier than the code * that they neatly hide away, and don't have the cost of function setup,so * we're going to use them. */ -#define MAX_CS_SIZE 64 * 1024 / 4 - -#define VERY_VERBOSE_CS 1 -#define VERY_VERBOSE_REGISTERS 1 - -/* XXX stolen from radeon_drm.h */ -#define RADEON_GEM_DOMAIN_CPU 0x1 -#define RADEON_GEM_DOMAIN_GTT 0x2 -#define RADEON_GEM_DOMAIN_VRAM 0x4 +#ifdef DEBUG +#define CS_DEBUG(x) x +#else +#define CS_DEBUG(x) +#endif -/* XXX stolen from radeon_reg.h */ -#define RADEON_CP_PACKET0 0x0 - -#define CP_PACKET0(register, count) \ - (RADEON_CP_PACKET0 | ((count) << 16) | ((register) >> 2)) +/** + * Command submission setup. + */ #define CS_LOCALS(context) \ - struct r300_context* const cs_context_copy = (context); \ - struct r300_winsys_screen *cs_winsys = cs_context_copy->rws; \ - int cs_count = 0; (void) cs_count; - -#define CHECK_CS(size) \ - assert(cs_winsys->check_cs(cs_winsys, (size))) + struct r300_winsys_cs *cs_copy = (context)->cs; \ + struct r300_winsys_screen *cs_winsys = (context)->rws; \ + int cs_count = 0; (void) cs_count; (void) cs_winsys; #define BEGIN_CS(size) do { \ - CHECK_CS(size); \ - if (VERY_VERBOSE_CS) { \ - DBG(cs_context_copy, DBG_CS, "r300: BEGIN_CS, count %d, in %s (%s:%d)\n", \ - size, __FUNCTION__, __FILE__, __LINE__); \ - } \ - cs_winsys->begin_cs(cs_winsys, (size), \ - __FILE__, __FUNCTION__, __LINE__); \ - cs_count = size; \ + assert(size <= (cs_copy->ndw - cs_copy->cdw)); \ + CS_DEBUG(cs_count = size;) \ } while (0) -#define OUT_CS(value) do { \ - if (VERY_VERBOSE_CS || VERY_VERBOSE_REGISTERS) { \ - DBG(cs_context_copy, DBG_CS, "r300: writing %08x\n", value); \ - } \ - cs_winsys->write_cs_dword(cs_winsys, (value)); \ - cs_count--; \ +#ifdef DEBUG +#define END_CS do { \ + if (cs_count != 0) \ + debug_printf("r300: Warning: cs_count off by %d at (%s, %s:%i)\n", \ + cs_count, __FUNCTION__, __FILE__, __LINE__); \ + cs_count = 0; \ } while (0) +#else +#define END_CS +#endif + -#define OUT_CS_32F(value) do { \ - if (VERY_VERBOSE_CS || VERY_VERBOSE_REGISTERS) { \ - DBG(cs_context_copy, DBG_CS, "r300: writing %f\n", value); \ - } \ - cs_winsys->write_cs_dword(cs_winsys, fui(value)); \ - cs_count--; \ +/** + * Writing pure DWORDs. + */ + +#define OUT_CS(value) do { \ + cs_copy->ptr[cs_copy->cdw++] = (value); \ + CS_DEBUG(cs_count--;) \ } while (0) +#define OUT_CS_32F(value) \ + OUT_CS(fui(value)) + #define OUT_CS_REG(register, value) do { \ - if (VERY_VERBOSE_REGISTERS) \ - DBG(cs_context_copy, DBG_CS, "r300: writing 0x%08X to register 0x%04X\n", \ - value, register); \ - assert(register); \ - cs_winsys->write_cs_dword(cs_winsys, CP_PACKET0(register, 0)); \ - cs_winsys->write_cs_dword(cs_winsys, value); \ - cs_count -= 2; \ + OUT_CS(CP_PACKET0(register, 0)); \ + OUT_CS(value); \ } while (0) /* Note: This expects count to be the number of registers, * not the actual packet0 count! */ -#define OUT_CS_REG_SEQ(register, count) do { \ - if (VERY_VERBOSE_REGISTERS) \ - DBG(cs_context_copy, DBG_CS, "r300: writing register sequence of %d to 0x%04X\n", \ - count, register); \ - assert(register); \ - cs_winsys->write_cs_dword(cs_winsys, CP_PACKET0((register), ((count) - 1))); \ - cs_count--; \ -} while (0) - -#define OUT_CS_TABLE(values, count) do { \ - if (VERY_VERBOSE_REGISTERS) \ - DBG(cs_context_copy, DBG_CS, "r300: writing table of %d dwords\n", count); \ - cs_winsys->write_cs_table(cs_winsys, values, count); \ - cs_count -= count; \ -} while (0) +#define OUT_CS_REG_SEQ(register, count) \ + OUT_CS(CP_PACKET0((register), ((count) - 1))) -#define OUT_CS_BUF_RELOC(bo, offset, rd, wd, flags) do { \ - DBG(cs_context_copy, DBG_CS, "r300: writing relocation for buffer %p, offset %d, " \ - "domains (%d, %d, %d)\n", \ - bo, offset, rd, wd, flags); \ - assert(bo); \ - cs_winsys->write_cs_dword(cs_winsys, offset); \ - r300_buffer_write_reloc(cs_winsys, r300_buffer(bo), rd, wd, flags); \ - cs_count -= 3; \ -} while (0) +#define OUT_CS_ONE_REG(register, count) \ + OUT_CS(CP_PACKET0((register), ((count) - 1)) | RADEON_ONE_REG_WR) +#define OUT_CS_PKT3(op, count) \ + OUT_CS(CP_PACKET3(op, count)) -#define OUT_CS_TEX_RELOC(tex, offset, rd, wd, flags) do { \ - DBG(cs_context_copy, DBG_CS, "r300: writing relocation for texture %p, offset %d, " \ - "domains (%d, %d, %d)\n", \ - tex, offset, rd, wd, flags); \ - assert(tex); \ - cs_winsys->write_cs_dword(cs_winsys, offset); \ - r300_texture_write_reloc(cs_winsys, tex, rd, wd, flags); \ - cs_count -= 3; \ +#define OUT_CS_TABLE(values, count) do { \ + memcpy(cs_copy->ptr + cs_copy->cdw, values, count * 4); \ + cs_copy->cdw += count; \ + CS_DEBUG(cs_count -= count;) \ } while (0) -#define OUT_CS_BUF_RELOC_NO_OFFSET(bo, rd, wd, flags) do { \ - DBG(cs_context_copy, DBG_CS, "r300: writing relocation for buffer %p, " \ - "domains (%d, %d, %d)\n", \ - bo, rd, wd, flags); \ +/** + * Writing relocations. + */ + +#define OUT_CS_RELOC(bo, offset, rd, wd) do { \ assert(bo); \ - r300_buffer_write_reloc(cs_winsys, r300_buffer(bo), rd, wd, flags); \ - cs_count -= 2; \ + OUT_CS(offset); \ + cs_winsys->cs_write_reloc(cs_copy, bo, rd, wd); \ + CS_DEBUG(cs_count -= 2;) \ } while (0) -#define END_CS do { \ - if (VERY_VERBOSE_CS) { \ - DBG(cs_context_copy, DBG_CS, "r300: END_CS in %s (%s:%d)\n", __FUNCTION__, \ - __FILE__, __LINE__); \ - } \ - if (cs_count != 0) \ - debug_printf("r300: Warning: cs_count off by %d\n", cs_count); \ - cs_winsys->end_cs(cs_winsys, __FILE__, __FUNCTION__, __LINE__); \ +#define OUT_CS_BUF_RELOC(bo, offset, rd, wd) do { \ + assert(bo); \ + OUT_CS_RELOC(r300_buffer(bo)->buf, offset, rd, wd); \ } while (0) -#define FLUSH_CS do { \ - if (VERY_VERBOSE_CS) { \ - DBG(cs_context_copy, DBG_CS, "r300: FLUSH_CS in %s (%s:%d)\n\n", __FUNCTION__, \ - __FILE__, __LINE__); \ - } \ - if (SCREEN_DBG_ON(r300->screen, DBG_STATS)) { \ - r300->flush_counter++; \ - } \ - cs_winsys->flush_cs(cs_winsys); \ +#define OUT_CS_TEX_RELOC(tex, offset, rd, wd) do { \ + assert(tex); \ + OUT_CS_RELOC(tex->buffer, offset, rd, wd); \ } while (0) -#define RADEON_ONE_REG_WR (1 << 15) - -#define OUT_CS_ONE_REG(register, count) do { \ - if (VERY_VERBOSE_REGISTERS) \ - DBG(cs_context_copy, DBG_CS, "r300: writing data sequence of %d to 0x%04X\n", \ - count, register); \ - assert(register); \ - cs_winsys->write_cs_dword(cs_winsys, CP_PACKET0((register), ((count) - 1)) | RADEON_ONE_REG_WR); \ - cs_count--; \ +#define OUT_CS_BUF_RELOC_NO_OFFSET(bo, rd, wd) do { \ + assert(bo); \ + cs_winsys->cs_write_reloc(cs_copy, r300_buffer(bo)->buf, rd, wd); \ + CS_DEBUG(cs_count -= 2;) \ } while (0) -#define CP_PACKET3(op, count) \ - (RADEON_CP_PACKET3 | (op) | ((count) << 16)) -#define OUT_CS_PKT3(op, count) do { \ - cs_winsys->write_cs_dword(cs_winsys, CP_PACKET3(op, count)); \ - cs_count--; \ -} while (0) +/** + * Command buffer emission. + */ -#define OUT_CS_INDEX_RELOC(bo, offset, count, rd, wd, flags) do { \ - DBG(cs_context_copy, DBG_CS, "r300: writing relocation for index buffer %p," \ - "offset %d\n", bo, offset); \ - assert(bo); \ - cs_winsys->write_cs_dword(cs_winsys, offset); \ - cs_winsys->write_cs_dword(cs_winsys, count); \ - cs_winsys->write_cs_reloc(cs_winsys, bo, rd, wd, flags); \ - cs_count -= 4; \ +#define WRITE_CS_TABLE(values, count) do { \ + CS_DEBUG(assert(cs_count == 0);) \ + memcpy(cs_copy->ptr + cs_copy->cdw, (values), (count) * 4); \ + cs_copy->cdw += (count); \ } while (0) #endif /* R300_CS_H */ diff --git a/src/gallium/drivers/r300/r300_debug.c b/src/gallium/drivers/r300/r300_debug.c index aed0ffeeccb..145a7985da3 100644 --- a/src/gallium/drivers/r300/r300_debug.c +++ b/src/gallium/drivers/r300/r300_debug.c @@ -22,76 +22,119 @@ #include "r300_context.h" -#include <stdio.h> +#include "util/u_debug.h" -struct debug_option { - const char * name; - unsigned flag; - const char * description; -}; +#include <stdio.h> -static struct debug_option debug_options[] = { - { "help", DBG_HELP, "Helpful meta-information about the driver" }, - { "fp", DBG_FP, "Fragment program handling (for debugging)" }, - { "vp", DBG_VP, "Vertex program handling (for debugging)" }, - { "cs", DBG_CS, "Command submissions (for debugging)" }, - { "draw", DBG_DRAW, "Draw and emit (for debugging)" }, - { "tex", DBG_TEX, "Textures (for debugging)" }, - { "fall", DBG_FALL, "Fallbacks (for debugging)" }, - { "anisohq", DBG_ANISOHQ, "High quality anisotropic filtering (for benchmarking)" }, - { "notiling", DBG_NO_TILING, "Disable tiling (for benchmarking)" }, - { "noimmd", DBG_NO_IMMD, "Disable immediate mode (for benchmarking)" }, - { "stats", DBG_STATS, "Gather statistics (for lulz)" }, - - { "all", ~0, "Convenience option that enables all debug flags" }, +static const struct debug_named_value debug_options[] = { + { "fp", DBG_FP, "Log fragment program compilation" }, + { "vp", DBG_VP, "Log vertex program compilation" }, + { "draw", DBG_DRAW, "Log draw calls" }, + { "swtcl", DBG_SWTCL, "Log SWTCL-specific info" }, + { "rsblock", DBG_RS_BLOCK, "Log rasterizer registers" }, + { "psc", DBG_PSC, "Log vertex stream registers" }, + { "tex", DBG_TEX, "Log basic info about textures" }, + { "texalloc", DBG_TEXALLOC, "Log texture mipmap tree info" }, + { "fall", DBG_FALL, "Log fallbacks" }, + { "rs", DBG_RS, "Log rasterizer" }, + { "fb", DBG_FB, "Log framebuffer" }, + { "cbzb", DBG_CBZB, "Log fast color clear info" }, + { "stats", DBG_STATS, "Log emission statistics" }, + { "hyperz", DBG_HYPERZ, "Log HyperZ info" }, + { "scissor", DBG_SCISSOR, "Log scissor info" }, + { "fakeocc", DBG_FAKE_OCC, "Use fake occlusion queries" }, + { "anisohq", DBG_ANISOHQ, "Use high quality anisotropic filtering" }, + { "notiling", DBG_NO_TILING, "Disable tiling" }, + { "noimmd", DBG_NO_IMMD, "Disable immediate mode" }, + { "noopt", DBG_NO_OPT, "Disable shader optimizations" }, + { "nocbzb", DBG_NO_CBZB, "Disable fast color clear" }, /* must be last */ - { 0, 0, 0 } + DEBUG_NAMED_VALUE_END }; void r300_init_debug(struct r300_screen * screen) { - const char * options = debug_get_option("RADEON_DEBUG", 0); - boolean printhint = FALSE; - size_t length; - struct debug_option * opt; - - if (options) { - while(*options) { - if (*options == ' ' || *options == ',') { - options++; - continue; - } + screen->debug = debug_get_flags_option("RADEON_DEBUG", debug_options, 0); +} - length = strcspn(options, " ,"); +void r500_dump_rs_block(struct r300_rs_block *rs) +{ + unsigned count, ip, it_count, ic_count, i, j; + unsigned tex_ptr; + unsigned col_ptr, col_fmt; - for(opt = debug_options; opt->name; ++opt) { - if (!strncmp(options, opt->name, length)) { - screen->debug |= opt->flag; - break; - } - } + count = rs->inst_count & 0xf; + count++; - if (!opt->name) { - fprintf(stderr, "Unknown debug option: %s\n", options); - printhint = TRUE; - } + it_count = rs->count & 0x7f; + ic_count = (rs->count >> 7) & 0xf; + + fprintf(stderr, "RS Block: %d texcoords (linear), %d colors (perspective)\n", + it_count, ic_count); + fprintf(stderr, "%d instructions\n", count); + + for (i = 0; i < count; i++) { + if (rs->inst[i] & 0x10) { + ip = rs->inst[i] & 0xf; + fprintf(stderr, "texture: ip %d to psf %d\n", + ip, (rs->inst[i] >> 5) & 0x7f); + + tex_ptr = rs->ip[ip] & 0xffffff; + fprintf(stderr, " : "); - options += length; + j = 3; + do { + if ((tex_ptr & 0x3f) == 63) { + fprintf(stderr, "1.0"); + } else if ((tex_ptr & 0x3f) == 62) { + fprintf(stderr, "0.0"); + } else { + fprintf(stderr, "[%d]", tex_ptr & 0x3f); + } + } while (j-- && fprintf(stderr, "/")); + fprintf(stderr, "\n"); } - if (!screen->debug) - printhint = TRUE; - } + if (rs->inst[i] & 0x10000) { + ip = (rs->inst[i] >> 12) & 0xf; + fprintf(stderr, "color: ip %d to psf %d\n", + ip, (rs->inst[i] >> 18) & 0x7f); - if (printhint || screen->debug & DBG_HELP) { - fprintf(stderr, "You can enable debug output by setting " - "the RADEON_DEBUG environment variable\n" - "to a comma-separated list of debug options. " - "Available options are:\n"); + col_ptr = (rs->ip[ip] >> 24) & 0x7; + col_fmt = (rs->ip[ip] >> 27) & 0xf; + fprintf(stderr, " : offset %d ", col_ptr); - for(opt = debug_options; opt->name; ++opt) { - fprintf(stderr, " %s: %s\n", opt->name, opt->description); + switch (col_fmt) { + case 0: + fprintf(stderr, "(R/G/B/A)"); + break; + case 1: + fprintf(stderr, "(R/G/B/0)"); + break; + case 2: + fprintf(stderr, "(R/G/B/1)"); + break; + case 4: + fprintf(stderr, "(0/0/0/A)"); + break; + case 5: + fprintf(stderr, "(0/0/0/0)"); + break; + case 6: + fprintf(stderr, "(0/0/0/1)"); + break; + case 8: + fprintf(stderr, "(1/1/1/A)"); + break; + case 9: + fprintf(stderr, "(1/1/1/0)"); + break; + case 10: + fprintf(stderr, "(1/1/1/1)"); + break; + } + fprintf(stderr, "\n"); } } } diff --git a/src/gallium/drivers/r300/r300_defines.h b/src/gallium/drivers/r300/r300_defines.h index 98ee3c1edec..896aeef395d 100644 --- a/src/gallium/drivers/r300/r300_defines.h +++ b/src/gallium/drivers/r300/r300_defines.h @@ -30,17 +30,21 @@ #define R300_RESOURCE_FLAG_TRANSFER PIPE_RESOURCE_FLAG_DRV_PRIV -/* XXX: this is just a bandaid on larger problems in - * r300_screen_buffer.h which doesn't seem to be fully ported to - * gallium-resources. - */ -#define R300_BIND_OQBO (1<<21) +#define R300_INVALID_FORMAT 0xffff /* Tiling flags. */ enum r300_buffer_tiling { R300_BUFFER_LINEAR = 0, R300_BUFFER_TILED, - R300_BUFFER_SQUARETILED + R300_BUFFER_SQUARETILED, + + R300_BUFFER_UNKNOWN, + R300_BUFFER_SELECT_LAYOUT = R300_BUFFER_UNKNOWN +}; + +enum r300_buffer_domain { /* bitfield */ + R300_DOMAIN_GTT = 1, + R300_DOMAIN_VRAM = 2 }; #endif diff --git a/src/gallium/drivers/r300/r300_emit.c b/src/gallium/drivers/r300/r300_emit.c index 92b7517b8d6..3a1085d2dc5 100644 --- a/src/gallium/drivers/r300/r300_emit.c +++ b/src/gallium/drivers/r300/r300_emit.c @@ -25,6 +25,7 @@ #include "util/u_format.h" #include "util/u_math.h" +#include "util/u_mm.h" #include "util/u_simple_list.h" #include "r300_context.h" @@ -43,21 +44,11 @@ void r300_emit_blend_state(struct r300_context* r300, (struct pipe_framebuffer_state*)r300->fb_state.state; CS_LOCALS(r300); - BEGIN_CS(size); - OUT_CS_REG(R300_RB3D_ROPCNTL, blend->rop); - OUT_CS_REG_SEQ(R300_RB3D_CBLEND, 3); if (fb->nr_cbufs) { - OUT_CS(blend->blend_control); - OUT_CS(blend->alpha_blend_control); - OUT_CS(blend->color_channel_mask); + WRITE_CS_TABLE(blend->cb, size); } else { - OUT_CS(0); - OUT_CS(0); - OUT_CS(0); - /* XXX also disable fastfill here once it's supported */ + WRITE_CS_TABLE(blend->cb_no_readwrite, size); } - OUT_CS_REG(R300_RB3D_DITHER_CTL, blend->dither); - END_CS; } void r300_emit_blend_color_state(struct r300_context* r300, @@ -66,40 +57,16 @@ void r300_emit_blend_color_state(struct r300_context* r300, struct r300_blend_color_state* bc = (struct r300_blend_color_state*)state; CS_LOCALS(r300); - if (r300->screen->caps.is_r500) { - BEGIN_CS(size); - OUT_CS_REG_SEQ(R500_RB3D_CONSTANT_COLOR_AR, 2); - OUT_CS(bc->blend_color_red_alpha); - OUT_CS(bc->blend_color_green_blue); - END_CS; - } else { - BEGIN_CS(size); - OUT_CS_REG(R300_RB3D_BLEND_COLOR, bc->blend_color); - END_CS; - } + WRITE_CS_TABLE(bc->cb, size); } void r300_emit_clip_state(struct r300_context* r300, unsigned size, void* state) { - struct pipe_clip_state* clip = (struct pipe_clip_state*)state; + struct r300_clip_state* clip = (struct r300_clip_state*)state; CS_LOCALS(r300); - if (r300->screen->caps.has_tcl) { - BEGIN_CS(size); - OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG, - (r300->screen->caps.is_r500 ? - R500_PVS_UCP_START : R300_PVS_UCP_START)); - OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, 6 * 4); - OUT_CS_TABLE(clip->ucp, 6 * 4); - OUT_CS_REG(R300_VAP_CLIP_CNTL, ((1 << clip->nr) - 1) | - R300_PS_UCP_MODE_CLIP_AS_TRIFAN); - END_CS; - } else { - BEGIN_CS(size); - OUT_CS_REG(R300_VAP_CLIP_CNTL, R300_CLIP_DISABLE); - END_CS; - } + WRITE_CS_TABLE(clip->cb, size); } void r300_emit_dsa_state(struct r300_context* r300, unsigned size, void* state) @@ -107,37 +74,22 @@ void r300_emit_dsa_state(struct r300_context* r300, unsigned size, void* state) struct r300_dsa_state* dsa = (struct r300_dsa_state*)state; struct pipe_framebuffer_state* fb = (struct pipe_framebuffer_state*)r300->fb_state.state; - struct pipe_stencil_ref stencil_ref = r300->stencil_ref; CS_LOCALS(r300); - BEGIN_CS(size); - OUT_CS_REG(R300_FG_ALPHA_FUNC, dsa->alpha_function); - OUT_CS_REG_SEQ(R300_ZB_CNTL, 3); - if (fb->zsbuf) { - OUT_CS(dsa->z_buffer_control); - OUT_CS(dsa->z_stencil_control); + WRITE_CS_TABLE(&dsa->cb_begin, size); } else { - OUT_CS(0); - OUT_CS(0); - } - - OUT_CS(dsa->stencil_ref_mask | stencil_ref.ref_value[0]); - - if (r300->screen->caps.is_r500) { - OUT_CS_REG(R500_ZB_STENCILREFMASK_BF, dsa->stencil_ref_bf | stencil_ref.ref_value[1]); + WRITE_CS_TABLE(dsa->cb_no_readwrite, size); } - END_CS; } static const float * get_rc_constant_state( struct r300_context * r300, struct rc_constant * constant) { - struct r300_viewport_state* viewport = r300->viewport_state.state; struct r300_textures_state* texstate = r300->textures_state.state; static float vec[4] = { 0.0, 0.0, 0.0, 1.0 }; - struct pipe_resource *tex; + struct r300_texture *tex; assert(constant->Type == RC_CONSTANT_STATE); @@ -145,21 +97,29 @@ static const float * get_rc_constant_state( /* Factor for converting rectangle coords to * normalized coords. Should only show up on non-r500. */ case RC_STATE_R300_TEXRECT_FACTOR: - tex = texstate->sampler_views[constant->u.State[1]]->base.texture; - vec[0] = 1.0 / tex->width0; - vec[1] = 1.0 / tex->height0; + tex = r300_texture(texstate->sampler_views[constant->u.State[1]]->base.texture); + vec[0] = 1.0 / tex->desc.width0; + vec[1] = 1.0 / tex->desc.height0; + break; + + case RC_STATE_R300_TEXSCALE_FACTOR: + tex = r300_texture(texstate->sampler_views[constant->u.State[1]]->base.texture); + /* Add a small number to the texture size to work around rounding errors in hw. */ + vec[0] = tex->desc.b.b.width0 / (tex->desc.width0 + 0.001f); + vec[1] = tex->desc.b.b.height0 / (tex->desc.height0 + 0.001f); + vec[2] = tex->desc.b.b.depth0 / (tex->desc.depth0 + 0.001f); break; case RC_STATE_R300_VIEWPORT_SCALE: - vec[0] = viewport->xscale; - vec[1] = viewport->yscale; - vec[2] = viewport->zscale; + vec[0] = r300->viewport.scale[0]; + vec[1] = r300->viewport.scale[1]; + vec[2] = r300->viewport.scale[2]; break; case RC_STATE_R300_VIEWPORT_OFFSET: - vec[0] = viewport->xoffset; - vec[1] = viewport->yoffset; - vec[2] = viewport->zoffset; + vec[0] = r300->viewport.translate[0]; + vec[1] = r300->viewport.translate[1]; + vec[2] = r300->viewport.translate[2]; break; default: @@ -176,7 +136,7 @@ static const float * get_rc_constant_state( /* Convert a normal single-precision float into the 7.16 format * used by the R300 fragment shader. */ -static uint32_t pack_float24(float f) +uint32_t pack_float24(float f) { union { float fl; @@ -207,85 +167,20 @@ static uint32_t pack_float24(float f) return float24; } -unsigned r300_get_fs_atom_size(struct r300_context *r300) -{ - struct r300_fragment_shader *fs = r300_fs(r300); - unsigned imm_count = fs->shader->immediates_count; - struct r300_fragment_program_code *code = &fs->shader->code.code.r300; - - return 19 + - code->alu.length * 4 + - (code->tex.length ? (1 + code->tex.length) : 0) + - (imm_count ? imm_count * 5 : 0); -} - void r300_emit_fs(struct r300_context* r300, unsigned size, void *state) { struct r300_fragment_shader *fs = r300_fs(r300); - struct rX00_fragment_program_code* generic_code = &fs->shader->code; - struct r300_fragment_program_code * code = &generic_code->code.r300; - unsigned i; - unsigned imm_count = fs->shader->immediates_count; - unsigned imm_first = fs->shader->externals_count; - unsigned imm_end = generic_code->constants.Count; - struct rc_constant *constants = generic_code->constants.Constants; CS_LOCALS(r300); - BEGIN_CS(size); - OUT_CS_REG(R300_US_CONFIG, code->config); - OUT_CS_REG(R300_US_PIXSIZE, code->pixsize); - OUT_CS_REG(R300_US_CODE_OFFSET, code->code_offset); - - OUT_CS_REG_SEQ(R300_US_CODE_ADDR_0, 4); - OUT_CS_TABLE(code->code_addr, 4); - - OUT_CS_REG_SEQ(R300_US_ALU_RGB_INST_0, code->alu.length); - for (i = 0; i < code->alu.length; i++) - OUT_CS(code->alu.inst[i].rgb_inst); - - OUT_CS_REG_SEQ(R300_US_ALU_RGB_ADDR_0, code->alu.length); - for (i = 0; i < code->alu.length; i++) - OUT_CS(code->alu.inst[i].rgb_addr); - - OUT_CS_REG_SEQ(R300_US_ALU_ALPHA_INST_0, code->alu.length); - for (i = 0; i < code->alu.length; i++) - OUT_CS(code->alu.inst[i].alpha_inst); - - OUT_CS_REG_SEQ(R300_US_ALU_ALPHA_ADDR_0, code->alu.length); - for (i = 0; i < code->alu.length; i++) - OUT_CS(code->alu.inst[i].alpha_addr); - - if (code->tex.length) { - OUT_CS_REG_SEQ(R300_US_TEX_INST_0, code->tex.length); - OUT_CS_TABLE(code->tex.inst, code->tex.length); - } - - /* Emit immediates. */ - if (imm_count) { - for(i = imm_first; i < imm_end; ++i) { - if (constants[i].Type == RC_CONSTANT_IMMEDIATE) { - const float *data = constants[i].u.Immediate; - - OUT_CS_REG_SEQ(R300_PFS_PARAM_0_X + i * 16, 4); - OUT_CS(pack_float24(data[0])); - OUT_CS(pack_float24(data[1])); - OUT_CS(pack_float24(data[2])); - OUT_CS(pack_float24(data[3])); - } - } - } - - OUT_CS_REG(R300_FG_DEPTH_SRC, fs->shader->fg_depth_src); - OUT_CS_REG(R300_US_W_FMT, fs->shader->us_out_w); - END_CS; + WRITE_CS_TABLE(fs->shader->cb_code, fs->shader->cb_code_size); } void r300_emit_fs_constants(struct r300_context* r300, unsigned size, void *state) { struct r300_fragment_shader *fs = r300_fs(r300); - struct rc_constant_list *constants = &fs->shader->code.constants; struct r300_constant_buffer *buf = (struct r300_constant_buffer*)state; - unsigned i, count = fs->shader->externals_count; + unsigned count = fs->shader->externals_count; + unsigned i, j; CS_LOCALS(r300); if (count == 0) @@ -293,15 +188,18 @@ void r300_emit_fs_constants(struct r300_context* r300, unsigned size, void *stat BEGIN_CS(size); OUT_CS_REG_SEQ(R300_PFS_PARAM_0_X, count * 4); - for(i = 0; i < count; ++i) { - const float *data; - assert(constants->Constants[i].Type == RC_CONSTANT_EXTERNAL); - data = buf->constants[i]; - OUT_CS(pack_float24(data[0])); - OUT_CS(pack_float24(data[1])); - OUT_CS(pack_float24(data[2])); - OUT_CS(pack_float24(data[3])); + if (buf->remap_table){ + for (i = 0; i < count; i++) { + float *data = (float*)&buf->ptr[buf->remap_table[i]*4]; + for (j = 0; j < 4; j++) + OUT_CS(pack_float24(data[j])); + } + } else { + for (i = 0; i < count; i++) + for (j = 0; j < 4; j++) + OUT_CS(pack_float24(*(float*)&buf->ptr[i*4+j])); } + END_CS; } @@ -313,6 +211,7 @@ void r300_emit_fs_rc_constant_state(struct r300_context* r300, unsigned size, vo unsigned count = fs->shader->rc_state_count; unsigned first = fs->shader->externals_count; unsigned end = constants->Count; + unsigned j; CS_LOCALS(r300); if (count == 0) @@ -325,84 +224,26 @@ void r300_emit_fs_rc_constant_state(struct r300_context* r300, unsigned size, vo get_rc_constant_state(r300, &constants->Constants[i]); OUT_CS_REG_SEQ(R300_PFS_PARAM_0_X + i * 16, 4); - OUT_CS(pack_float24(data[0])); - OUT_CS(pack_float24(data[1])); - OUT_CS(pack_float24(data[2])); - OUT_CS(pack_float24(data[3])); + for (j = 0; j < 4; j++) + OUT_CS(pack_float24(data[j])); } } END_CS; } -unsigned r500_get_fs_atom_size(struct r300_context *r300) -{ - struct r300_fragment_shader *fs = r300_fs(r300); - unsigned imm_count = fs->shader->immediates_count; - struct r500_fragment_program_code *code = &fs->shader->code.code.r500; - - return 17 + - ((code->inst_end + 1) * 6) + - (imm_count ? imm_count * 7 : 0); -} - void r500_emit_fs(struct r300_context* r300, unsigned size, void *state) { struct r300_fragment_shader *fs = r300_fs(r300); - struct rX00_fragment_program_code* generic_code = &fs->shader->code; - struct r500_fragment_program_code * code = &generic_code->code.r500; - unsigned i; - unsigned imm_count = fs->shader->immediates_count; - unsigned imm_first = fs->shader->externals_count; - unsigned imm_end = generic_code->constants.Count; - struct rc_constant *constants = generic_code->constants.Constants; CS_LOCALS(r300); - BEGIN_CS(size); - OUT_CS_REG(R500_US_CONFIG, R500_ZERO_TIMES_ANYTHING_EQUALS_ZERO); - OUT_CS_REG(R500_US_PIXSIZE, code->max_temp_idx); - OUT_CS_REG(R500_US_CODE_RANGE, - R500_US_CODE_RANGE_ADDR(0) | R500_US_CODE_RANGE_SIZE(code->inst_end)); - OUT_CS_REG(R500_US_CODE_OFFSET, 0); - OUT_CS_REG(R500_US_CODE_ADDR, - R500_US_CODE_START_ADDR(0) | R500_US_CODE_END_ADDR(code->inst_end)); - - OUT_CS_REG(R500_GA_US_VECTOR_INDEX, R500_GA_US_VECTOR_INDEX_TYPE_INSTR); - OUT_CS_ONE_REG(R500_GA_US_VECTOR_DATA, (code->inst_end + 1) * 6); - for (i = 0; i <= code->inst_end; i++) { - OUT_CS(code->inst[i].inst0); - OUT_CS(code->inst[i].inst1); - OUT_CS(code->inst[i].inst2); - OUT_CS(code->inst[i].inst3); - OUT_CS(code->inst[i].inst4); - OUT_CS(code->inst[i].inst5); - } - - /* Emit immediates. */ - if (imm_count) { - for(i = imm_first; i < imm_end; ++i) { - if (constants[i].Type == RC_CONSTANT_IMMEDIATE) { - const float *data = constants[i].u.Immediate; - - OUT_CS_REG(R500_GA_US_VECTOR_INDEX, - R500_GA_US_VECTOR_INDEX_TYPE_CONST | - (i & R500_GA_US_VECTOR_INDEX_MASK)); - OUT_CS_ONE_REG(R500_GA_US_VECTOR_DATA, 4); - OUT_CS_TABLE(data, 4); - } - } - } - - OUT_CS_REG(R300_FG_DEPTH_SRC, fs->shader->fg_depth_src); - OUT_CS_REG(R300_US_W_FMT, fs->shader->us_out_w); - END_CS; + WRITE_CS_TABLE(fs->shader->cb_code, fs->shader->cb_code_size); } void r500_emit_fs_constants(struct r300_context* r300, unsigned size, void *state) { struct r300_fragment_shader *fs = r300_fs(r300); - struct rc_constant_list *constants = &fs->shader->code.constants; struct r300_constant_buffer *buf = (struct r300_constant_buffer*)state; - unsigned i, count = fs->shader->externals_count; + unsigned count = fs->shader->externals_count; CS_LOCALS(r300); if (count == 0) @@ -411,10 +252,14 @@ void r500_emit_fs_constants(struct r300_context* r300, unsigned size, void *stat BEGIN_CS(size); OUT_CS_REG(R500_GA_US_VECTOR_INDEX, R500_GA_US_VECTOR_INDEX_TYPE_CONST); OUT_CS_ONE_REG(R500_GA_US_VECTOR_DATA, count * 4); - for(i = 0; i < count; ++i) { - assert(constants->Constants[i].Type == RC_CONSTANT_EXTERNAL); + if (buf->remap_table){ + for (unsigned i = 0; i < count; i++) { + uint32_t *data = &buf->ptr[buf->remap_table[i]*4]; + OUT_CS_TABLE(data, 4); + } + } else { + OUT_CS_TABLE(buf->ptr, count * 4); } - OUT_CS_TABLE(buf->constants, count * 4); END_CS; } @@ -447,86 +292,236 @@ void r500_emit_fs_rc_constant_state(struct r300_context* r300, unsigned size, vo END_CS; } +void r300_emit_gpu_flush(struct r300_context *r300, unsigned size, void *state) +{ + struct r300_gpu_flush *gpuflush = (struct r300_gpu_flush*)state; + struct pipe_framebuffer_state* fb = + (struct pipe_framebuffer_state*)r300->fb_state.state; + uint32_t height = fb->height; + uint32_t width = fb->width; + CS_LOCALS(r300); + + if (r300->cbzb_clear) { + struct r300_surface *surf = r300_surface(fb->cbufs[0]); + + height = surf->cbzb_height; + width = surf->cbzb_width; + } + + DBG(r300, DBG_SCISSOR, + "r300: Scissor width: %i, height: %i, CBZB clear: %s\n", + width, height, r300->cbzb_clear ? "YES" : "NO"); + + BEGIN_CS(size); + + /* Set up scissors. + * By writing to the SC registers, SC & US assert idle. */ + OUT_CS_REG_SEQ(R300_SC_SCISSORS_TL, 2); + if (r300->screen->caps.is_r500) { + OUT_CS(0); + OUT_CS(((width - 1) << R300_SCISSORS_X_SHIFT) | + ((height - 1) << R300_SCISSORS_Y_SHIFT)); + } else { + OUT_CS((1440 << R300_SCISSORS_X_SHIFT) | + (1440 << R300_SCISSORS_Y_SHIFT)); + OUT_CS(((width + 1440-1) << R300_SCISSORS_X_SHIFT) | + ((height + 1440-1) << R300_SCISSORS_Y_SHIFT)); + } + + /* Flush CB & ZB caches and wait until the 3D engine is idle and clean. */ + OUT_CS_TABLE(gpuflush->cb_flush_clean, 6); + END_CS; +} + +void r300_emit_aa_state(struct r300_context *r300, unsigned size, void *state) +{ + struct r300_aa_state *aa = (struct r300_aa_state*)state; + CS_LOCALS(r300); + + BEGIN_CS(size); + OUT_CS_REG(R300_GB_AA_CONFIG, aa->aa_config); + + if (aa->dest) { + OUT_CS_REG_SEQ(R300_RB3D_AARESOLVE_OFFSET, 1); + OUT_CS_RELOC(aa->dest->buffer, aa->dest->offset, 0, aa->dest->domain); + + OUT_CS_REG_SEQ(R300_RB3D_AARESOLVE_PITCH, 1); + OUT_CS_RELOC(aa->dest->buffer, aa->dest->pitch, 0, aa->dest->domain); + } + + OUT_CS_REG(R300_RB3D_AARESOLVE_CTL, aa->aaresolve_ctl); + END_CS; +} + void r300_emit_fb_state(struct r300_context* r300, unsigned size, void* state) { struct pipe_framebuffer_state* fb = (struct pipe_framebuffer_state*)state; - struct r300_texture* tex; - struct pipe_surface* surf; - int i; + struct r300_surface* surf; + unsigned i; + boolean has_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ); CS_LOCALS(r300); BEGIN_CS(size); - /* Flush and free renderbuffer caches. */ - OUT_CS_REG(R300_RB3D_DSTCACHE_CTLSTAT, - R300_RB3D_DSTCACHE_CTLSTAT_DC_FREE_FREE_3D_TAGS | - R300_RB3D_DSTCACHE_CTLSTAT_DC_FLUSH_FLUSH_DIRTY_3D); - OUT_CS_REG(R300_ZB_ZCACHE_CTLSTAT, - R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE | - R300_ZB_ZCACHE_CTLSTAT_ZC_FREE_FREE); - - /* Set the number of colorbuffers. */ - if (fb->nr_cbufs > 1) { - if (r300->screen->caps.is_r500) { - OUT_CS_REG(R300_RB3D_CCTL, - R300_RB3D_CCTL_NUM_MULTIWRITES(fb->nr_cbufs) | - R300_RB3D_CCTL_INDEPENDENT_COLORFORMAT_ENABLE_ENABLE); - } else { - OUT_CS_REG(R300_RB3D_CCTL, - R300_RB3D_CCTL_NUM_MULTIWRITES(fb->nr_cbufs)); - } + /* NUM_MULTIWRITES replicates COLOR[0] to all colorbuffers, which is not + * what we usually want. */ + if (r300->screen->caps.is_r500) { + OUT_CS_REG(R300_RB3D_CCTL, + R300_RB3D_CCTL_INDEPENDENT_COLORFORMAT_ENABLE_ENABLE); } else { - OUT_CS_REG(R300_RB3D_CCTL, 0x0); + OUT_CS_REG(R300_RB3D_CCTL, 0); } /* Set up colorbuffers. */ for (i = 0; i < fb->nr_cbufs; i++) { - surf = fb->cbufs[i]; - tex = r300_texture(surf->texture); - assert(tex && tex->buffer && "cbuf is marked, but NULL!"); + surf = r300_surface(fb->cbufs[i]); OUT_CS_REG_SEQ(R300_RB3D_COLOROFFSET0 + (4 * i), 1); - OUT_CS_TEX_RELOC(tex, surf->offset, 0, RADEON_GEM_DOMAIN_VRAM, 0); + OUT_CS_RELOC(surf->buffer, surf->offset, 0, surf->domain); OUT_CS_REG_SEQ(R300_RB3D_COLORPITCH0 + (4 * i), 1); - OUT_CS_TEX_RELOC(tex, tex->fb_state.colorpitch[surf->level], - 0, RADEON_GEM_DOMAIN_VRAM, 0); - - OUT_CS_REG(R300_US_OUT_FMT_0 + (4 * i), tex->fb_state.us_out_fmt); - } - for (; i < 4; i++) { - OUT_CS_REG(R300_US_OUT_FMT_0 + (4 * i), R300_US_OUT_FMT_UNUSED); + OUT_CS_RELOC(surf->buffer, surf->pitch, 0, surf->domain); } - /* Set up a zbuffer. */ - if (fb->zsbuf) { - surf = fb->zsbuf; - tex = r300_texture(surf->texture); - assert(tex && tex->buffer && "zsbuf is marked, but NULL!"); + /* Set up the ZB part of the CBZB clear. */ + if (r300->cbzb_clear) { + surf = r300_surface(fb->cbufs[0]); + + OUT_CS_REG(R300_ZB_FORMAT, surf->cbzb_format); OUT_CS_REG_SEQ(R300_ZB_DEPTHOFFSET, 1); - OUT_CS_TEX_RELOC(tex, surf->offset, 0, RADEON_GEM_DOMAIN_VRAM, 0); + OUT_CS_RELOC(surf->buffer, surf->cbzb_midpoint_offset, 0, surf->domain); + + OUT_CS_REG_SEQ(R300_ZB_DEPTHPITCH, 1); + OUT_CS_RELOC(surf->buffer, surf->cbzb_pitch, 0, surf->domain); + + DBG(r300, DBG_CBZB, + "CBZB clearing cbuf %08x %08x\n", surf->cbzb_format, + surf->cbzb_pitch); + } + /* Set up a zbuffer. */ + else if (fb->zsbuf) { + surf = r300_surface(fb->zsbuf); - OUT_CS_REG(R300_ZB_FORMAT, tex->fb_state.zb_format); + OUT_CS_REG(R300_ZB_FORMAT, surf->format); + + OUT_CS_REG_SEQ(R300_ZB_DEPTHOFFSET, 1); + OUT_CS_RELOC(surf->buffer, surf->offset, 0, surf->domain); OUT_CS_REG_SEQ(R300_ZB_DEPTHPITCH, 1); - OUT_CS_TEX_RELOC(tex, tex->fb_state.depthpitch[surf->level], - 0, RADEON_GEM_DOMAIN_VRAM, 0); + OUT_CS_RELOC(surf->buffer, surf->pitch, 0, surf->domain); + + if (has_hyperz) { + uint32_t surf_pitch; + struct r300_texture *tex; + int level = surf->base.level; + tex = r300_texture(surf->base.texture); + + surf_pitch = surf->pitch & R300_DEPTHPITCH_MASK; + /* HiZ RAM. */ + if (r300->screen->caps.hiz_ram) { + if (tex->hiz_mem[level]) { + OUT_CS_REG(R300_ZB_HIZ_OFFSET, tex->hiz_mem[level]->ofs << 2); + OUT_CS_REG(R300_ZB_HIZ_PITCH, surf_pitch); + } else { + OUT_CS_REG(R300_ZB_HIZ_OFFSET, 0); + OUT_CS_REG(R300_ZB_HIZ_PITCH, 0); + } + } + /* Z Mask RAM. (compressed zbuffer) */ + if (tex->zmask_mem[level]) { + OUT_CS_REG(R300_ZB_ZMASK_OFFSET, tex->zmask_mem[level]->ofs << 2); + OUT_CS_REG(R300_ZB_ZMASK_PITCH, surf_pitch); + } else { + OUT_CS_REG(R300_ZB_ZMASK_OFFSET, 0); + OUT_CS_REG(R300_ZB_ZMASK_PITCH, 0); + } + } } - OUT_CS_REG_SEQ(R300_SC_SCISSORS_TL, 2); - if (r300->screen->caps.is_r500) { - OUT_CS(0); - OUT_CS(((fb->width - 1) << R300_SCISSORS_X_SHIFT) | - ((fb->height - 1) << R300_SCISSORS_Y_SHIFT)); - } else { - OUT_CS((1440 << R300_SCISSORS_X_SHIFT) | - (1440 << R300_SCISSORS_Y_SHIFT)); - OUT_CS(((fb->width + 1440-1) << R300_SCISSORS_X_SHIFT) | - ((fb->height + 1440-1) << R300_SCISSORS_Y_SHIFT)); + END_CS; +} + +void r300_emit_hyperz_state(struct r300_context *r300, + unsigned size, void *state) +{ + struct r300_hyperz_state *z = state; + CS_LOCALS(r300); + if (z->flush) + WRITE_CS_TABLE(&z->cb_flush_begin, size); + else + WRITE_CS_TABLE(&z->cb_begin, size - 2); +} + +void r300_emit_hyperz_end(struct r300_context *r300) +{ + struct r300_hyperz_state z = + *(struct r300_hyperz_state*)r300->hyperz_state.state; + + z.flush = 1; + z.zb_bw_cntl = 0; + z.zb_depthclearvalue = 0; + z.sc_hyperz = R300_SC_HYPERZ_ADJ_2; + z.gb_z_peq_config = 0; + + r300_emit_hyperz_state(r300, r300->hyperz_state.size, &z); +} + +void r300_emit_fb_state_pipelined(struct r300_context *r300, + unsigned size, void *state) +{ + struct pipe_framebuffer_state* fb = + (struct pipe_framebuffer_state*)r300->fb_state.state; + unsigned i; + CS_LOCALS(r300); + + BEGIN_CS(size); + + /* Colorbuffer format in the US block. + * (must be written after unpipelined regs) */ + OUT_CS_REG_SEQ(R300_US_OUT_FMT_0, 4); + for (i = 0; i < fb->nr_cbufs; i++) { + OUT_CS(r300_surface(fb->cbufs[i])->format); + } + for (; i < 4; i++) { + OUT_CS(R300_US_OUT_FMT_UNUSED); + } + + /* Multisampling. Depends on framebuffer sample count. + * These are pipelined regs and as such cannot be moved + * to the AA state. */ + if (r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0)) { + unsigned mspos0 = 0x66666666; + unsigned mspos1 = 0x6666666; + + if (fb->nr_cbufs && fb->cbufs[0]->texture->nr_samples > 1) { + /* Subsample placement. These may not be optimal. */ + switch (fb->cbufs[0]->texture->nr_samples) { + case 2: + mspos0 = 0x33996633; + mspos1 = 0x6666663; + break; + case 3: + mspos0 = 0x33936933; + mspos1 = 0x6666663; + break; + case 4: + mspos0 = 0x33939933; + mspos1 = 0x3966663; + break; + case 6: + mspos0 = 0x22a2aa22; + mspos1 = 0x2a65672; + break; + default: + debug_printf("r300: Bad number of multisamples!\n"); + } + } + + OUT_CS_REG_SEQ(R300_GB_MSPOS0, 2); + OUT_CS(mspos0); + OUT_CS(mspos1); } - OUT_CS_REG(R300_GA_POINT_MINMAX, - (MAX2(fb->width, fb->height) * 6) << R300_GA_POINT_MINMAX_MAX_SHIFT); END_CS; } @@ -547,13 +542,14 @@ void r300_emit_query_start(struct r300_context *r300, unsigned size, void*state) OUT_CS_REG(R300_ZB_ZPASS_DATA, 0); END_CS; query->begin_emitted = TRUE; + query->flushed = FALSE; } - -static void r300_emit_query_finish(struct r300_context *r300, - struct r300_query *query) +static void r300_emit_query_end_frag_pipes(struct r300_context *r300, + struct r300_query *query) { struct r300_capabilities* caps = &r300->screen->caps; + struct r300_winsys_buffer *buf = r300->query_current->buffer; CS_LOCALS(r300); assert(caps->num_frag_pipes); @@ -572,28 +568,28 @@ static void r300_emit_query_finish(struct r300_context *r300, /* pipe 3 only */ OUT_CS_REG(R300_SU_REG_DEST, 1 << 3); OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR, 1); - OUT_CS_BUF_RELOC(r300->oqbo, query->offset + (sizeof(uint32_t) * 3), - 0, RADEON_GEM_DOMAIN_GTT, 0); + OUT_CS_RELOC(buf, (query->num_results + 3) * 4, + 0, query->domain); case 3: /* pipe 2 only */ OUT_CS_REG(R300_SU_REG_DEST, 1 << 2); OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR, 1); - OUT_CS_BUF_RELOC(r300->oqbo, query->offset + (sizeof(uint32_t) * 2), - 0, RADEON_GEM_DOMAIN_GTT, 0); + OUT_CS_RELOC(buf, (query->num_results + 2) * 4, + 0, query->domain); case 2: /* pipe 1 only */ /* As mentioned above, accomodate RV380 and older. */ OUT_CS_REG(R300_SU_REG_DEST, 1 << (caps->high_second_pipe ? 3 : 1)); OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR, 1); - OUT_CS_BUF_RELOC(r300->oqbo, query->offset + (sizeof(uint32_t) * 1), - 0, RADEON_GEM_DOMAIN_GTT, 0); + OUT_CS_RELOC(buf, (query->num_results + 1) * 4, + 0, query->domain); case 1: /* pipe 0 only */ OUT_CS_REG(R300_SU_REG_DEST, 1 << 0); OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR, 1); - OUT_CS_BUF_RELOC(r300->oqbo, query->offset + (sizeof(uint32_t) * 0), - 0, RADEON_GEM_DOMAIN_GTT, 0); + OUT_CS_RELOC(buf, (query->num_results + 0) * 4, + 0, query->domain); break; default: fprintf(stderr, "r300: Implementation error: Chipset reports %d" @@ -606,31 +602,33 @@ static void r300_emit_query_finish(struct r300_context *r300, END_CS; } -static void rv530_emit_query_single(struct r300_context *r300, - struct r300_query *query) +static void rv530_emit_query_end_single_z(struct r300_context *r300, + struct r300_query *query) { + struct r300_winsys_buffer *buf = r300->query_current->buffer; CS_LOCALS(r300); BEGIN_CS(8); OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_0); OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR, 1); - OUT_CS_BUF_RELOC(r300->oqbo, query->offset, 0, RADEON_GEM_DOMAIN_GTT, 0); + OUT_CS_RELOC(buf, query->num_results * 4, 0, query->domain); OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL); END_CS; } -static void rv530_emit_query_double(struct r300_context *r300, - struct r300_query *query) +static void rv530_emit_query_end_double_z(struct r300_context *r300, + struct r300_query *query) { + struct r300_winsys_buffer *buf = r300->query_current->buffer; CS_LOCALS(r300); BEGIN_CS(14); OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_0); OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR, 1); - OUT_CS_BUF_RELOC(r300->oqbo, query->offset, 0, RADEON_GEM_DOMAIN_GTT, 0); + OUT_CS_RELOC(buf, (query->num_results + 0) * 4, 0, query->domain); OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_1); OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR, 1); - OUT_CS_BUF_RELOC(r300->oqbo, query->offset + sizeof(uint32_t), 0, RADEON_GEM_DOMAIN_GTT, 0); + OUT_CS_RELOC(buf, (query->num_results + 1) * 4, 0, query->domain); OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL); END_CS; } @@ -648,60 +646,43 @@ void r300_emit_query_end(struct r300_context* r300) if (caps->family == CHIP_FAMILY_RV530) { if (caps->num_z_pipes == 2) - rv530_emit_query_double(r300, query); + rv530_emit_query_end_double_z(r300, query); else - rv530_emit_query_single(r300, query); + rv530_emit_query_end_single_z(r300, query); } else - r300_emit_query_finish(r300, query); + r300_emit_query_end_frag_pipes(r300, query); + + query->begin_emitted = FALSE; + query->num_results += query->num_pipes; + + /* XXX grab all the results and reset the counter. */ + if (query->num_results >= query->buffer_size / 4 - 4) { + query->num_results = (query->buffer_size / 4) / 2; + fprintf(stderr, "r300: Rewinding OQBO...\n"); + } +} + +void r300_emit_invariant_state(struct r300_context *r300, + unsigned size, void *state) +{ + CS_LOCALS(r300); + WRITE_CS_TABLE(state, size); } void r300_emit_rs_state(struct r300_context* r300, unsigned size, void* state) { - struct r300_rs_state* rs = (struct r300_rs_state*)state; - float scale, offset; + struct r300_rs_state* rs = state; CS_LOCALS(r300); BEGIN_CS(size); - OUT_CS_REG(R300_VAP_CNTL_STATUS, rs->vap_control_status); - - OUT_CS_REG(R300_GB_AA_CONFIG, rs->antialiasing_config); - - OUT_CS_REG(R300_GA_POINT_SIZE, rs->point_size); - OUT_CS_REG(R300_GA_LINE_CNTL, rs->line_control); - + OUT_CS_TABLE(rs->cb_main, RS_STATE_MAIN_SIZE); if (rs->polygon_offset_enable) { - scale = rs->depth_scale * 12; - offset = rs->depth_offset; - - switch (r300->zbuffer_bpp) { - case 16: - offset *= 4; - break; - case 24: - offset *= 2; - break; + if (r300->zbuffer_bpp == 16) { + OUT_CS_TABLE(rs->cb_poly_offset_zb16, 5); + } else { + OUT_CS_TABLE(rs->cb_poly_offset_zb24, 5); } - - OUT_CS_REG_SEQ(R300_SU_POLY_OFFSET_FRONT_SCALE, 4); - OUT_CS_32F(scale); - OUT_CS_32F(offset); - OUT_CS_32F(scale); - OUT_CS_32F(offset); } - - OUT_CS_REG_SEQ(R300_SU_POLY_OFFSET_ENABLE, 2); - OUT_CS(rs->polygon_offset_enable); - OUT_CS(rs->cull_mode); - OUT_CS_REG(R300_GA_LINE_STIPPLE_CONFIG, rs->line_stipple_config); - OUT_CS_REG(R300_GA_LINE_STIPPLE_VALUE, rs->line_stipple_value); - OUT_CS_REG(R300_GA_POLY_MODE, rs->polygon_mode); - OUT_CS_REG(R300_SC_CLIP_RULE, rs->clip_rule); - OUT_CS_REG(R300_GB_ENABLE, rs->stuffing_enable); - OUT_CS_REG_SEQ(R300_GA_POINT_S0, 4); - OUT_CS_32F(rs->point_texcoord_left); - OUT_CS_32F(rs->point_texcoord_bottom); - OUT_CS_32F(rs->point_texcoord_right); - OUT_CS_32F(rs->point_texcoord_top); END_CS; } @@ -714,18 +695,37 @@ void r300_emit_rs_block_state(struct r300_context* r300, unsigned count = (rs->inst_count & R300_RS_INST_COUNT_MASK) + 1; CS_LOCALS(r300); - DBG(r300, DBG_DRAW, "r300: RS emit:\n"); + if (DBG_ON(r300, DBG_RS_BLOCK)) { + r500_dump_rs_block(rs); + + fprintf(stderr, "r300: RS emit:\n"); + + for (i = 0; i < count; i++) + fprintf(stderr, " : ip %d: 0x%08x\n", i, rs->ip[i]); + + for (i = 0; i < count; i++) + fprintf(stderr, " : inst %d: 0x%08x\n", i, rs->inst[i]); + + fprintf(stderr, " : count: 0x%08x inst_count: 0x%08x\n", + rs->count, rs->inst_count); + } BEGIN_CS(size); + OUT_CS_REG_SEQ(R300_VAP_VTX_STATE_CNTL, 2); + OUT_CS(rs->vap_vtx_state_cntl); + OUT_CS(rs->vap_vsm_vtx_assm); + OUT_CS_REG_SEQ(R300_VAP_OUTPUT_VTX_FMT_0, 2); + OUT_CS(rs->vap_out_vtx_fmt[0]); + OUT_CS(rs->vap_out_vtx_fmt[1]); + OUT_CS_REG_SEQ(R300_GB_ENABLE, 1); + OUT_CS(rs->gb_enable); + if (r300->screen->caps.is_r500) { OUT_CS_REG_SEQ(R500_RS_IP_0, count); } else { OUT_CS_REG_SEQ(R300_RS_IP_0, count); } OUT_CS_TABLE(rs->ip, count); - for (i = 0; i < count; i++) { - DBG(r300, DBG_DRAW, " : ip %d: 0x%08x\n", i, rs->ip[i]); - } OUT_CS_REG_SEQ(R300_RS_COUNT, 2); OUT_CS(rs->count); @@ -737,13 +737,6 @@ void r300_emit_rs_block_state(struct r300_context* r300, OUT_CS_REG_SEQ(R300_RS_INST_0, count); } OUT_CS_TABLE(rs->inst, count); - for (i = 0; i < count; i++) { - DBG(r300, DBG_DRAW, " : inst %d: 0x%08x\n", i, rs->inst[i]); - } - - DBG(r300, DBG_DRAW, " : count: 0x%08x inst_count: 0x%08x\n", - rs->count, rs->inst_count); - END_CS; } @@ -774,6 +767,7 @@ void r300_emit_textures_state(struct r300_context *r300, { struct r300_textures_state *allstate = (struct r300_textures_state*)state; struct r300_texture_sampler_state *texstate; + struct r300_texture *tex; unsigned i; CS_LOCALS(r300); @@ -783,6 +777,7 @@ void r300_emit_textures_state(struct r300_context *r300, for (i = 0; i < allstate->count; i++) { if ((1 << i) & allstate->tx_enable) { texstate = &allstate->regs[i]; + tex = r300_texture(allstate->sampler_views[i]->base.texture); OUT_CS_REG(R300_TX_FILTER0_0 + (i * 4), texstate->filter0); OUT_CS_REG(R300_TX_FILTER1_0 + (i * 4), texstate->filter1); @@ -794,32 +789,24 @@ void r300_emit_textures_state(struct r300_context *r300, OUT_CS_REG(R300_TX_FORMAT2_0 + (i * 4), texstate->format.format2); OUT_CS_REG_SEQ(R300_TX_OFFSET_0 + (i * 4), 1); - OUT_CS_TEX_RELOC(r300_texture(allstate->sampler_views[i]->base.texture), - texstate->format.tile_config, - RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM, 0, 0); + OUT_CS_TEX_RELOC(tex, texstate->format.tile_config, tex->domain, + 0); } } END_CS; } -void r300_emit_aos(struct r300_context* r300, unsigned offset, boolean indexed) +void r300_emit_aos(struct r300_context* r300, int offset, boolean indexed) { struct pipe_vertex_buffer *vb1, *vb2, *vbuf = r300->vertex_buffer; struct pipe_vertex_element *velem = r300->velems->velem; + struct r300_buffer *buf; int i; + unsigned *hw_format_size = r300->velems->hw_format_size; unsigned size1, size2, aos_count = r300->velems->count; unsigned packet_size = (aos_count * 3 + 1) / 2; CS_LOCALS(r300); - for (i = 0; i < aos_count; i++) { - if ((vbuf[velem[i].vertex_buffer_index].buffer_offset + velem[i].src_offset) % 4 != 0) { - /* XXX We must align the buffer. */ - assert(0); - fprintf(stderr, "r300: Unaligned vertex buffer offsets aren't supported, aborting..\n"); - abort(); - } - } - BEGIN_CS(2 + packet_size + aos_count * 2); OUT_CS_PKT3(R300_PACKET3_3D_LOAD_VBPNTR, packet_size); OUT_CS(aos_count | (!indexed ? R300_VC_FORCE_PREFETCH : 0)); @@ -827,8 +814,8 @@ void r300_emit_aos(struct r300_context* r300, unsigned offset, boolean indexed) for (i = 0; i < aos_count - 1; i += 2) { vb1 = &vbuf[velem[i].vertex_buffer_index]; vb2 = &vbuf[velem[i+1].vertex_buffer_index]; - size1 = util_format_get_blocksize(velem[i].src_format); - size2 = util_format_get_blocksize(velem[i+1].src_format); + size1 = hw_format_size[i]; + size2 = hw_format_size[i+1]; OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(vb1->stride) | R300_VBPNTR_SIZE1(size2) | R300_VBPNTR_STRIDE1(vb2->stride)); @@ -838,24 +825,24 @@ void r300_emit_aos(struct r300_context* r300, unsigned offset, boolean indexed) if (aos_count & 1) { vb1 = &vbuf[velem[i].vertex_buffer_index]; - size1 = util_format_get_blocksize(velem[i].src_format); + size1 = hw_format_size[i]; OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(vb1->stride)); OUT_CS(vb1->buffer_offset + velem[i].src_offset + offset * vb1->stride); } for (i = 0; i < aos_count; i++) { - OUT_CS_BUF_RELOC_NO_OFFSET(vbuf[velem[i].vertex_buffer_index].buffer, - RADEON_GEM_DOMAIN_GTT, 0, 0); + buf = r300_buffer(vbuf[velem[i].vertex_buffer_index].buffer); + OUT_CS_BUF_RELOC_NO_OFFSET(&buf->b.b, buf->domain, 0); } END_CS; } -void r300_emit_vertex_buffer(struct r300_context* r300) +void r300_emit_aos_swtcl(struct r300_context *r300, boolean indexed) { CS_LOCALS(r300); - DBG(r300, DBG_DRAW, "r300: Preparing vertex buffer %p for render, " + DBG(r300, DBG_SWTCL, "r300: Preparing vertex buffer %p for render, " "vertex size %d\n", r300->vbo, r300->vertex_info.size); /* Set the pointer to our vertex buffer. The emitted values are this: @@ -867,11 +854,11 @@ void r300_emit_vertex_buffer(struct r300_context* r300) */ BEGIN_CS(7); OUT_CS_PKT3(R300_PACKET3_3D_LOAD_VBPNTR, 3); - OUT_CS(1); + OUT_CS(1 | (!indexed ? R300_VC_FORCE_PREFETCH : 0)); OUT_CS(r300->vertex_info.size | (r300->vertex_info.size << 8)); - OUT_CS(r300->vbo_offset); - OUT_CS_BUF_RELOC(r300->vbo, 0, RADEON_GEM_DOMAIN_GTT, 0, 0); + OUT_CS(r300->draw_vbo_offset); + OUT_CS_BUF_RELOC(r300->vbo, 0, r300_buffer(r300->vbo)->domain, 0); END_CS; } @@ -883,50 +870,42 @@ void r300_emit_vertex_stream_state(struct r300_context* r300, unsigned i; CS_LOCALS(r300); - DBG(r300, DBG_DRAW, "r300: PSC emit:\n"); + if (DBG_ON(r300, DBG_PSC)) { + fprintf(stderr, "r300: PSC emit:\n"); + + for (i = 0; i < streams->count; i++) { + fprintf(stderr, " : prog_stream_cntl%d: 0x%08x\n", i, + streams->vap_prog_stream_cntl[i]); + } + + for (i = 0; i < streams->count; i++) { + fprintf(stderr, " : prog_stream_cntl_ext%d: 0x%08x\n", i, + streams->vap_prog_stream_cntl_ext[i]); + } + } BEGIN_CS(size); OUT_CS_REG_SEQ(R300_VAP_PROG_STREAM_CNTL_0, streams->count); OUT_CS_TABLE(streams->vap_prog_stream_cntl, streams->count); - for (i = 0; i < streams->count; i++) { - DBG(r300, DBG_DRAW, " : prog_stream_cntl%d: 0x%08x\n", i, - streams->vap_prog_stream_cntl[i]); - } OUT_CS_REG_SEQ(R300_VAP_PROG_STREAM_CNTL_EXT_0, streams->count); OUT_CS_TABLE(streams->vap_prog_stream_cntl_ext, streams->count); - for (i = 0; i < streams->count; i++) { - DBG(r300, DBG_DRAW, " : prog_stream_cntl_ext%d: 0x%08x\n", i, - streams->vap_prog_stream_cntl_ext[i]); - } END_CS; } -void r300_emit_vap_output_state(struct r300_context* r300, - unsigned size, void* state) +void r300_emit_pvs_flush(struct r300_context* r300, unsigned size, void* state) { - struct r300_vap_output_state *vap_out_state = - (struct r300_vap_output_state*)state; CS_LOCALS(r300); - DBG(r300, DBG_DRAW, "r300: VAP emit:\n"); - BEGIN_CS(size); - OUT_CS_REG_SEQ(R300_VAP_VTX_STATE_CNTL, 2); - OUT_CS(vap_out_state->vap_vtx_state_cntl); - OUT_CS(vap_out_state->vap_vsm_vtx_assm); - OUT_CS_REG_SEQ(R300_VAP_OUTPUT_VTX_FMT_0, 2); - OUT_CS(vap_out_state->vap_out_vtx_fmt[0]); - OUT_CS(vap_out_state->vap_out_vtx_fmt[1]); + OUT_CS_REG(R300_VAP_PVS_STATE_FLUSH_REG, 0x0); END_CS; } -void r300_emit_pvs_flush(struct r300_context* r300, unsigned size, void* state) +void r300_emit_vap_invariant_state(struct r300_context *r300, + unsigned size, void *state) { CS_LOCALS(r300); - - BEGIN_CS(size); - OUT_CS_REG(R300_VAP_PVS_STATE_FLUSH_REG, 0x0); - END_CS; + WRITE_CS_TABLE(state, size); } void r300_emit_vs_state(struct r300_context* r300, unsigned size, void* state) @@ -944,7 +923,7 @@ void r300_emit_vs_state(struct r300_context* r300, unsigned size, void* state) unsigned pvs_num_slots = MIN3(vtx_mem_size / input_count, vtx_mem_size / output_count, 10); - unsigned pvs_num_controllers = MIN2(vtx_mem_size / temp_count, 6); + unsigned pvs_num_controllers = MIN2(vtx_mem_size / temp_count, 5); unsigned imm_first = vs->externals_count; unsigned imm_end = vs->code.constants.Count; @@ -953,6 +932,7 @@ void r300_emit_vs_state(struct r300_context* r300, unsigned size, void* state) CS_LOCALS(r300); BEGIN_CS(size); + /* R300_VAP_PVS_CODE_CNTL_0 * R300_VAP_PVS_CONST_CNTL * R300_VAP_PVS_CODE_CNTL_1 @@ -986,6 +966,22 @@ void r300_emit_vs_state(struct r300_context* r300, unsigned size, void* state) OUT_CS_TABLE(data, 4); } } + + /* Emit flow control instructions. */ + if (code->num_fc_ops) { + + OUT_CS_REG(R300_VAP_PVS_FLOW_CNTL_OPC, code->fc_ops); + if (r300screen->caps.is_r500) { + OUT_CS_REG_SEQ(R500_VAP_PVS_FLOW_CNTL_ADDRS_LW_0, code->num_fc_ops * 2); + OUT_CS_TABLE(code->fc_op_addrs.r500, code->num_fc_ops * 2); + } else { + OUT_CS_REG_SEQ(R300_VAP_PVS_FLOW_CNTL_ADDRS_0, code->num_fc_ops); + OUT_CS_TABLE(code->fc_op_addrs.r300, code->num_fc_ops); + } + OUT_CS_REG_SEQ(R300_VAP_PVS_FLOW_CNTL_LOOP_INDEX_0, code->num_fc_ops); + OUT_CS_TABLE(code->fc_loop_index, code->num_fc_ops); + } + END_CS; } @@ -995,6 +991,7 @@ void r300_emit_vs_constants(struct r300_context* r300, unsigned count = ((struct r300_vertex_shader*)r300->vs_state.state)->externals_count; struct r300_constant_buffer *buf = (struct r300_constant_buffer*)state; + unsigned i; CS_LOCALS(r300); if (!count) @@ -1005,7 +1002,14 @@ void r300_emit_vs_constants(struct r300_context* r300, (r300->screen->caps.is_r500 ? R500_PVS_CONST_START : R300_PVS_CONST_START)); OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, count * 4); - OUT_CS_TABLE(buf->constants, count * 4); + if (buf->remap_table){ + for (i = 0; i < count; i++) { + uint32_t *data = &buf->ptr[buf->remap_table[i]*4]; + OUT_CS_TABLE(data, 4); + } + } else { + OUT_CS_TABLE(buf->ptr, count * 4); + } END_CS; } @@ -1015,16 +1019,116 @@ void r300_emit_viewport_state(struct r300_context* r300, struct r300_viewport_state* viewport = (struct r300_viewport_state*)state; CS_LOCALS(r300); - BEGIN_CS(size); - OUT_CS_REG_SEQ(R300_SE_VPORT_XSCALE, 6); - OUT_CS_32F(viewport->xscale); - OUT_CS_32F(viewport->xoffset); - OUT_CS_32F(viewport->yscale); - OUT_CS_32F(viewport->yoffset); - OUT_CS_32F(viewport->zscale); - OUT_CS_32F(viewport->zoffset); - OUT_CS_REG(R300_VAP_VTE_CNTL, viewport->vte_control); - END_CS; + BEGIN_CS(size); + OUT_CS_REG_SEQ(R300_SE_VPORT_XSCALE, 6); + OUT_CS_TABLE(&viewport->xscale, 6); + OUT_CS_REG(R300_VAP_VTE_CNTL, viewport->vte_control); + END_CS; +} + +static void r300_emit_hiz_line_clear(struct r300_context *r300, int start, uint16_t count, uint32_t val) +{ + CS_LOCALS(r300); + BEGIN_CS(4); + OUT_CS_PKT3(R300_PACKET3_3D_CLEAR_HIZ, 2); + OUT_CS(start); + OUT_CS(count); + OUT_CS(val); + END_CS; +} + +static void r300_emit_zmask_line_clear(struct r300_context *r300, int start, uint16_t count, uint32_t val) +{ + CS_LOCALS(r300); + BEGIN_CS(4); + OUT_CS_PKT3(R300_PACKET3_3D_CLEAR_ZMASK, 2); + OUT_CS(start); + OUT_CS(count); + OUT_CS(val); + END_CS; +} + +#define ALIGN_DIVUP(x, y) (((x) + (y) - 1) / (y)) + +void r300_emit_hiz_clear(struct r300_context *r300, unsigned size, void *state) +{ + struct pipe_framebuffer_state *fb = + (struct pipe_framebuffer_state*)r300->fb_state.state; + struct r300_hyperz_state *z = + (struct r300_hyperz_state*)r300->hyperz_state.state; + struct r300_screen* r300screen = r300->screen; + uint32_t stride, offset = 0, height, offset_shift; + struct r300_texture* tex; + int i; + + tex = r300_texture(fb->zsbuf->texture); + + offset = tex->hiz_mem[fb->zsbuf->level]->ofs; + stride = tex->desc.stride_in_pixels[fb->zsbuf->level]; + + /* convert from pixels to 4x4 blocks */ + stride = ALIGN_DIVUP(stride, 4); + + stride = ALIGN_DIVUP(stride, r300screen->caps.num_frag_pipes); + /* there are 4 blocks per dwords */ + stride = ALIGN_DIVUP(stride, 4); + + height = ALIGN_DIVUP(fb->zsbuf->height, 4); + + offset_shift = 2; + offset_shift += (r300screen->caps.num_frag_pipes / 2); + + for (i = 0; i < height; i++) { + offset = i * stride; + offset <<= offset_shift; + r300_emit_hiz_line_clear(r300, offset, stride, 0xffffffff); + } + z->current_func = -1; + + /* Mark the current zbuffer's hiz ram as in use. */ + tex->hiz_in_use[fb->zsbuf->level] = TRUE; +} + +void r300_emit_zmask_clear(struct r300_context *r300, unsigned size, void *state) +{ + struct pipe_framebuffer_state *fb = + (struct pipe_framebuffer_state*)r300->fb_state.state; + struct r300_screen* r300screen = r300->screen; + uint32_t stride, offset = 0; + struct r300_texture* tex; + uint32_t i, height; + int mult, offset_shift; + + tex = r300_texture(fb->zsbuf->texture); + stride = tex->desc.stride_in_pixels[fb->zsbuf->level]; + + offset = tex->zmask_mem[fb->zsbuf->level]->ofs; + + if (r300->z_compression == RV350_Z_COMPRESS_88) + mult = 8; + else + mult = 4; + + height = ALIGN_DIVUP(fb->zsbuf->height, mult); + + offset_shift = 4; + offset_shift += (r300screen->caps.num_frag_pipes / 2); + stride = ALIGN_DIVUP(stride, r300screen->caps.num_frag_pipes); + + /* okay have width in pixels - divide by block width */ + stride = ALIGN_DIVUP(stride, mult); + /* have width in blocks - divide by number of fragment pipes screen width */ + /* 16 blocks per dword */ + stride = ALIGN_DIVUP(stride, 16); + + for (i = 0; i < height; i++) { + offset = i * stride; + offset <<= offset_shift; + r300_emit_zmask_line_clear(r300, offset, stride, 0x0);//0xffffffff); + } + + /* Mark the current zbuffer's zmask as in use. */ + tex->zmask_in_use[fb->zsbuf->level] = TRUE; } void r300_emit_ztop_state(struct r300_context* r300, @@ -1047,9 +1151,9 @@ void r300_emit_texture_cache_inval(struct r300_context* r300, unsigned size, voi END_CS; } -void r300_emit_buffer_validate(struct r300_context *r300, - boolean do_validate_vertex_buffers, - struct pipe_resource *index_buffer) +boolean r300_emit_buffer_validate(struct r300_context *r300, + boolean do_validate_vertex_buffers, + struct pipe_resource *index_buffer) { struct pipe_framebuffer_state* fb = (struct pipe_framebuffer_state*)r300->fb_state.state; @@ -1060,37 +1164,29 @@ void r300_emit_buffer_validate(struct r300_context *r300, struct pipe_vertex_element *velem = r300->velems->velem; struct pipe_resource *pbuf; unsigned i; - boolean invalid = FALSE; /* upload buffers first */ - if (r300->any_user_vbs) { + if (r300->screen->caps.has_tcl && r300->any_user_vbs) { r300_upload_user_buffers(r300); r300->any_user_vbs = false; } /* Clean out BOs. */ - r300->rws->reset_bos(r300->rws); + r300->rws->cs_reset_buffers(r300->cs); -validate: /* Color buffers... */ for (i = 0; i < fb->nr_cbufs; i++) { tex = r300_texture(fb->cbufs[i]->texture); assert(tex && tex->buffer && "cbuf is marked, but NULL!"); - if (!r300_add_texture(r300->rws, tex, - 0, RADEON_GEM_DOMAIN_VRAM)) { - r300->context.flush(&r300->context, 0, NULL); - goto validate; - } + r300->rws->cs_add_buffer(r300->cs, tex->buffer, 0, + r300_surface(fb->cbufs[i])->domain); } /* ...depth buffer... */ if (fb->zsbuf) { tex = r300_texture(fb->zsbuf->texture); assert(tex && tex->buffer && "zsbuf is marked, but NULL!"); - if (!r300_add_texture(r300->rws, tex, - 0, RADEON_GEM_DOMAIN_VRAM)) { - r300->context.flush(&r300->context, 0, NULL); - goto validate; - } + r300->rws->cs_add_buffer(r300->cs, tex->buffer, 0, + r300_surface(fb->zsbuf)->domain); } /* ...textures... */ for (i = 0; i < texstate->count; i++) { @@ -1099,58 +1195,35 @@ validate: } tex = r300_texture(texstate->sampler_views[i]->base.texture); - if (!r300_add_texture(r300->rws, tex, - RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM, 0)) { - r300->context.flush(&r300->context, 0, NULL); - goto validate; - } + r300->rws->cs_add_buffer(r300->cs, tex->buffer, tex->domain, 0); } /* ...occlusion query buffer... */ - if (r300->query_start.dirty) { - if (!r300_add_buffer(r300->rws, r300->oqbo, - 0, RADEON_GEM_DOMAIN_GTT)) { - r300->context.flush(&r300->context, 0, NULL); - goto validate; - } - } + if (r300->query_current) + r300->rws->cs_add_buffer(r300->cs, r300->query_current->buffer, + 0, r300->query_current->domain); /* ...vertex buffer for SWTCL path... */ - if (r300->vbo) { - if (!r300_add_buffer(r300->rws, r300->vbo, - RADEON_GEM_DOMAIN_GTT, 0)) { - r300->context.flush(&r300->context, 0, NULL); - goto validate; - } - } + if (r300->vbo) + r300->rws->cs_add_buffer(r300->cs, r300_buffer(r300->vbo)->buf, + r300_buffer(r300->vbo)->domain, 0); /* ...vertex buffers for HWTCL path... */ if (do_validate_vertex_buffers) { for (i = 0; i < r300->velems->count; i++) { pbuf = vbuf[velem[i].vertex_buffer_index].buffer; - if (!r300_add_buffer(r300->rws, pbuf, - RADEON_GEM_DOMAIN_GTT, 0)) { - r300->context.flush(&r300->context, 0, NULL); - goto validate; - } + r300->rws->cs_add_buffer(r300->cs, r300_buffer(pbuf)->buf, + r300_buffer(pbuf)->domain, 0); } } /* ...and index buffer for HWTCL path. */ - if (index_buffer) { - if (!r300_add_buffer(r300->rws, index_buffer, - RADEON_GEM_DOMAIN_GTT, 0)) { - r300->context.flush(&r300->context, 0, NULL); - goto validate; - } - } - if (!r300->rws->validate(r300->rws)) { - r300->context.flush(&r300->context, 0, NULL); - if (invalid) { - /* Well, hell. */ - fprintf(stderr, "r300: Stuck in validation loop, gonna quit now.\n"); - abort(); - } - invalid = TRUE; - goto validate; + if (index_buffer) + r300->rws->cs_add_buffer(r300->cs, r300_buffer(index_buffer)->buf, + r300_buffer(index_buffer)->domain, 0); + + if (!r300->rws->cs_validate(r300->cs)) { + return FALSE; } + + return TRUE; } unsigned r300_get_num_dirty_dwords(struct r300_context *r300) @@ -1164,18 +1237,28 @@ unsigned r300_get_num_dirty_dwords(struct r300_context *r300) } } - /* emit_query_end is not atomized. */ - dwords += 26; /* let's reserve some more, just in case */ dwords += 32; return dwords; } +unsigned r300_get_num_cs_end_dwords(struct r300_context *r300) +{ + unsigned dwords = 0; + + /* Emitted in flush. */ + dwords += 26; /* emit_query_end */ + dwords += r300->hyperz_state.size + 2; /* emit_hyperz_end + zcache flush */ + if (r500_index_bias_supported(r300)) + dwords += 2; + + return dwords; +} + /* Emit all dirty state. */ void r300_emit_dirty_state(struct r300_context* r300) { - struct r300_screen* r300screen = r300->screen; struct r300_atom* atom; foreach(atom, &r300->atom_list) { @@ -1188,10 +1271,5 @@ void r300_emit_dirty_state(struct r300_context* r300) } } - /* Emit the VBO for SWTCL. */ - if (!r300screen->caps.has_tcl) { - r300_emit_vertex_buffer(r300); - } - r300->dirty_hw++; } diff --git a/src/gallium/drivers/r300/r300_emit.h b/src/gallium/drivers/r300/r300_emit.h index c3eb195d4e7..278dbcb4c7c 100644 --- a/src/gallium/drivers/r300/r300_emit.h +++ b/src/gallium/drivers/r300/r300_emit.h @@ -29,7 +29,9 @@ struct rX00_fragment_program_code; struct r300_vertex_program_code; -void r300_emit_aos(struct r300_context* r300, unsigned offset, boolean indexed); +uint32_t pack_float24(float f); + +void r300_emit_aos(struct r300_context* r300, int offset, boolean indexed); void r300_emit_blend_state(struct r300_context* r300, unsigned size, void* state); @@ -43,7 +45,10 @@ void r300_emit_clip_state(struct r300_context* r300, void r300_emit_dsa_state(struct r300_context* r300, unsigned size, void* state); -unsigned r300_get_fs_atom_size(struct r300_context *r300); +void r300_emit_hyperz_state(struct r300_context *r300, + unsigned size, void *state); + +void r300_emit_hyperz_end(struct r300_context *r300); void r300_emit_fs(struct r300_context* r300, unsigned size, void *state); @@ -51,8 +56,6 @@ void r300_emit_fs_constants(struct r300_context* r300, unsigned size, void *stat void r300_emit_fs_rc_constant_state(struct r300_context* r300, unsigned size, void *state); -unsigned r500_get_fs_atom_size(struct r300_context *r300); - void r500_emit_fs(struct r300_context* r300, unsigned size, void *state); void r500_emit_fs_constants(struct r300_context* r300, unsigned size, void *state); @@ -61,6 +64,13 @@ void r500_emit_fs_rc_constant_state(struct r300_context* r300, unsigned size, vo void r300_emit_fb_state(struct r300_context* r300, unsigned size, void* state); +void r300_emit_fb_state_pipelined(struct r300_context *r300, + unsigned size, void *state); + +void r300_emit_gpu_flush(struct r300_context *r300, unsigned size, void *state); + +void r300_emit_aa_state(struct r300_context *r300, unsigned size, void *state); + void r300_emit_query_start(struct r300_context *r300, unsigned size, void *state); void r300_emit_query_end(struct r300_context* r300); @@ -76,14 +86,14 @@ void r300_emit_scissor_state(struct r300_context* r300, void r300_emit_textures_state(struct r300_context *r300, unsigned size, void *state); -void r300_emit_vertex_buffer(struct r300_context* r300); +void r300_emit_aos_swtcl(struct r300_context *r300, boolean indexed); + +void r300_emit_vap_invariant_state(struct r300_context *r300, + unsigned size, void *state); void r300_emit_vertex_stream_state(struct r300_context* r300, unsigned size, void* state); -void r300_emit_vap_output_state(struct r300_context* r300, - unsigned size, void* state); - void r300_emit_vs_constants(struct r300_context* r300, unsigned size, void *state); @@ -99,13 +109,20 @@ void r300_emit_pvs_flush(struct r300_context* r300, unsigned size, void* state); void r300_emit_texture_cache_inval(struct r300_context* r300, unsigned size, void* state); +void r300_emit_invariant_state(struct r300_context *r300, + unsigned size, void *state); + +void r300_emit_hiz_clear(struct r300_context *r300, unsigned size, void *state); +void r300_emit_zmask_clear(struct r300_context *r300, unsigned size, void *state); + unsigned r300_get_num_dirty_dwords(struct r300_context *r300); +unsigned r300_get_num_cs_end_dwords(struct r300_context *r300); /* Emit all dirty state. */ void r300_emit_dirty_state(struct r300_context* r300); -void r300_emit_buffer_validate(struct r300_context *r300, - boolean do_validate_vertex_buffers, - struct pipe_resource *index_buffer); +boolean r300_emit_buffer_validate(struct r300_context *r300, + boolean do_validate_vertex_buffers, + struct pipe_resource *index_buffer); #endif /* R300_EMIT_H */ diff --git a/src/gallium/drivers/r300/r300_flush.c b/src/gallium/drivers/r300/r300_flush.c index e78c6a3624f..1afd27f0938 100644 --- a/src/gallium/drivers/r300/r300_flush.c +++ b/src/gallium/drivers/r300/r300_flush.c @@ -1,5 +1,6 @@ /* * Copyright 2008 Corbin Simpson <[email protected]> + * Copyright 2010 Marek Olšák <[email protected]> * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), @@ -24,11 +25,11 @@ #include "draw/draw_private.h" #include "util/u_simple_list.h" +#include "util/u_upload_mgr.h" #include "r300_context.h" #include "r300_cs.h" #include "r300_emit.h" -#include "r300_flush.h" static void r300_flush(struct pipe_context* pipe, unsigned flags, @@ -37,26 +38,27 @@ static void r300_flush(struct pipe_context* pipe, struct r300_context *r300 = r300_context(pipe); struct r300_query *query; struct r300_atom *atom; + struct r300_fence **rfence = (struct r300_fence**)fence; - CS_LOCALS(r300); - (void) cs_count; - /* We probably need to flush Draw, but we may have been called from - * within Draw. This feels kludgy, but it might be the best thing. - * - * Of course, the best thing is to kill Draw with fire. :3 */ - if (r300->draw && !r300->draw->flushing) { - draw_flush(r300->draw); - } + u_upload_flush(r300->upload_vb); + u_upload_flush(r300->upload_ib); - r300_emit_query_end(r300); + if (r300->draw && !r300->draw_vbo_locked) + r300_draw_flush_vbuf(r300); if (r300->dirty_hw) { - FLUSH_CS; + r300_emit_hyperz_end(r300); + r300_emit_query_end(r300); + if (r500_index_bias_supported(r300)) + r500_emit_index_bias(r300, 0); + + r300->flush_counter++; + r300->rws->cs_flush(r300->cs); r300->dirty_hw = 0; /* New kitchen sink, baby. */ foreach(atom, &r300->atom_list) { - if (atom->state) { + if (atom->state || atom->allow_null_state) { atom->dirty = TRUE; } } @@ -72,6 +74,13 @@ static void r300_flush(struct pipe_context* pipe, foreach(query, &r300->query_list) { query->flushed = TRUE; } + + /* Create a new fence. */ + if (rfence) { + *rfence = CALLOC_STRUCT(r300_fence); + pipe_reference_init(&(*rfence)->reference, 1); + (*rfence)->ctx = r300; + } } void r300_init_flush_functions(struct r300_context* r300) diff --git a/src/gallium/drivers/r300/r300_flush.h b/src/gallium/drivers/r300/r300_flush.h deleted file mode 100644 index 0e9e6106bb7..00000000000 --- a/src/gallium/drivers/r300/r300_flush.h +++ /dev/null @@ -1,28 +0,0 @@ -/* - * Copyright 2008 Corbin Simpson <[email protected]> - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * on the rights to use, copy, modify, merge, publish, distribute, sub - * license, and/or sell copies of the Software, and to permit persons to whom - * the Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice (including the next - * paragraph) shall be included in all copies or substantial portions of the - * Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL - * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, - * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR - * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE - * USE OR OTHER DEALINGS IN THE SOFTWARE. */ - -#ifndef R300_FLUSH_H -#define R300_FLUSH_H - -void r300_init_flush_functions(struct r300_context* r300); - -#endif /* R300_FLUSH_H */ diff --git a/src/gallium/drivers/r300/r300_fs.c b/src/gallium/drivers/r300/r300_fs.c index 88303f074cd..d9d4a9304df 100644 --- a/src/gallium/drivers/r300/r300_fs.c +++ b/src/gallium/drivers/r300/r300_fs.c @@ -28,7 +28,9 @@ #include "tgsi/tgsi_dump.h" #include "tgsi/tgsi_ureg.h" +#include "r300_cb.h" #include "r300_context.h" +#include "r300_emit.h" #include "r300_screen.h" #include "r300_fs.h" #include "r300_reg.h" @@ -70,6 +72,11 @@ void r300_shader_read_fs_inputs(struct tgsi_shader_info* info, fs_inputs->wpos = i; break; + case TGSI_SEMANTIC_FACE: + assert(index == 0); + fs_inputs->face = i; + break; + default: fprintf(stderr, "r300: FP: Unknown input semantic: %i\n", info->input_semantic_name[i]); @@ -118,6 +125,9 @@ static void allocate_hardware_inputs( allocate(mydata, inputs->color[i], reg++); } } + if (inputs->face != ATTR_UNUSED) { + allocate(mydata, inputs->face, reg++); + } for (i = 0; i < ATTR_GENERIC_COUNT; i++) { if (inputs->generic[i] != ATTR_UNUSED) { allocate(mydata, inputs->generic[i], reg++); @@ -137,18 +147,32 @@ static void get_external_state( { struct r300_textures_state *texstate = r300->textures_state.state; unsigned i; + unsigned char *swizzle; for (i = 0; i < texstate->sampler_state_count; i++) { - struct r300_sampler_state* s = texstate->sampler_states[i]; + struct r300_sampler_state *s = texstate->sampler_states[i]; + struct r300_sampler_view *v = texstate->sampler_views[i]; + struct r300_texture *t; - if (!s) { + if (!s || !v) { continue; } + t = r300_texture(texstate->sampler_views[i]->base.texture); + if (s->state.compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) { - /* XXX Gallium doesn't provide us with any information regarding - * this mode, so we are screwed. I'm setting 0 = LUMINANCE. */ - state->unit[i].depth_texture_mode = 0; + state->unit[i].compare_mode_enabled = 1; + + /* Pass depth texture swizzling to the compiler. */ + if (texstate->sampler_views[i]) { + swizzle = texstate->sampler_views[i]->swizzle; + + state->unit[i].depth_texture_swizzle = + RC_MAKE_SWIZZLE(swizzle[0], swizzle[1], + swizzle[2], swizzle[3]); + } else { + state->unit[i].depth_texture_swizzle = RC_SWIZZLE_XYZW; + } /* Fortunately, no need to translate this. */ state->unit[i].texture_compare_func = s->state.compare_func; @@ -156,35 +180,29 @@ static void get_external_state( state->unit[i].non_normalized_coords = !s->state.normalized_coords; - if (texstate->sampler_views[i]) { - struct r300_texture *t; - t = (struct r300_texture*)texstate->sampler_views[i]->base.texture; - - /* XXX this should probably take into account STR, not just S. */ - if (t->uses_pitch) { - switch (s->state.wrap_s) { - case PIPE_TEX_WRAP_REPEAT: - state->unit[i].wrap_mode = RC_WRAP_REPEAT; - state->unit[i].fake_npot = TRUE; - break; - - case PIPE_TEX_WRAP_MIRROR_REPEAT: - state->unit[i].wrap_mode = RC_WRAP_MIRRORED_REPEAT; - state->unit[i].fake_npot = TRUE; - break; - - case PIPE_TEX_WRAP_MIRROR_CLAMP: - case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE: - case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER: - state->unit[i].wrap_mode = RC_WRAP_MIRRORED_CLAMP; - state->unit[i].fake_npot = TRUE; - break; - - default: - state->unit[i].wrap_mode = RC_WRAP_NONE; - break; - } + /* XXX this should probably take into account STR, not just S. */ + if (t->desc.is_npot) { + switch (s->state.wrap_s) { + case PIPE_TEX_WRAP_REPEAT: + state->unit[i].wrap_mode = RC_WRAP_REPEAT; + break; + + case PIPE_TEX_WRAP_MIRROR_REPEAT: + state->unit[i].wrap_mode = RC_WRAP_MIRRORED_REPEAT; + break; + + case PIPE_TEX_WRAP_MIRROR_CLAMP: + case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE: + case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER: + state->unit[i].wrap_mode = RC_WRAP_MIRRORED_CLAMP; + break; + + default: + state->unit[i].wrap_mode = RC_WRAP_NONE; } + + if (t->desc.b.b.target == PIPE_TEXTURE_3D) + state->unit[i].clamp_and_scale_before_fetch = TRUE; } } } @@ -219,6 +237,128 @@ static void r300_dummy_fragment_shader( ureg_destroy(ureg); } +static void r300_emit_fs_code_to_buffer( + struct r300_context *r300, + struct r300_fragment_shader_code *shader) +{ + struct rX00_fragment_program_code *generic_code = &shader->code; + unsigned imm_count = shader->immediates_count; + unsigned imm_first = shader->externals_count; + unsigned imm_end = generic_code->constants.Count; + struct rc_constant *constants = generic_code->constants.Constants; + unsigned i; + CB_LOCALS; + + if (r300->screen->caps.is_r500) { + struct r500_fragment_program_code *code = &generic_code->code.r500; + + shader->cb_code_size = 19 + + ((code->inst_end + 1) * 6) + + imm_count * 7 + + code->int_constant_count * 2; + + NEW_CB(shader->cb_code, shader->cb_code_size); + OUT_CB_REG(R500_US_CONFIG, R500_ZERO_TIMES_ANYTHING_EQUALS_ZERO); + OUT_CB_REG(R500_US_PIXSIZE, code->max_temp_idx); + OUT_CB_REG(R500_US_FC_CTRL, code->us_fc_ctrl); + for(i = 0; i < code->int_constant_count; i++){ + OUT_CB_REG(R500_US_FC_INT_CONST_0 + (i * 4), + code->int_constants[i]); + } + OUT_CB_REG(R500_US_CODE_RANGE, + R500_US_CODE_RANGE_ADDR(0) | R500_US_CODE_RANGE_SIZE(code->inst_end)); + OUT_CB_REG(R500_US_CODE_OFFSET, 0); + OUT_CB_REG(R500_US_CODE_ADDR, + R500_US_CODE_START_ADDR(0) | R500_US_CODE_END_ADDR(code->inst_end)); + + OUT_CB_REG(R500_GA_US_VECTOR_INDEX, R500_GA_US_VECTOR_INDEX_TYPE_INSTR); + OUT_CB_ONE_REG(R500_GA_US_VECTOR_DATA, (code->inst_end + 1) * 6); + for (i = 0; i <= code->inst_end; i++) { + OUT_CB(code->inst[i].inst0); + OUT_CB(code->inst[i].inst1); + OUT_CB(code->inst[i].inst2); + OUT_CB(code->inst[i].inst3); + OUT_CB(code->inst[i].inst4); + OUT_CB(code->inst[i].inst5); + } + + /* Emit immediates. */ + if (imm_count) { + for(i = imm_first; i < imm_end; ++i) { + if (constants[i].Type == RC_CONSTANT_IMMEDIATE) { + const float *data = constants[i].u.Immediate; + + OUT_CB_REG(R500_GA_US_VECTOR_INDEX, + R500_GA_US_VECTOR_INDEX_TYPE_CONST | + (i & R500_GA_US_VECTOR_INDEX_MASK)); + OUT_CB_ONE_REG(R500_GA_US_VECTOR_DATA, 4); + OUT_CB_TABLE(data, 4); + } + } + } + } else { /* r300 */ + struct r300_fragment_program_code *code = &generic_code->code.r300; + + shader->cb_code_size = 19 + + (r300->screen->caps.is_r400 ? 2 : 0) + + code->alu.length * 4 + + (code->tex.length ? (1 + code->tex.length) : 0) + + imm_count * 5; + + NEW_CB(shader->cb_code, shader->cb_code_size); + + if (r300->screen->caps.is_r400) + OUT_CB_REG(R400_US_CODE_BANK, 0); + + OUT_CB_REG(R300_US_CONFIG, code->config); + OUT_CB_REG(R300_US_PIXSIZE, code->pixsize); + OUT_CB_REG(R300_US_CODE_OFFSET, code->code_offset); + + OUT_CB_REG_SEQ(R300_US_CODE_ADDR_0, 4); + OUT_CB_TABLE(code->code_addr, 4); + + OUT_CB_REG_SEQ(R300_US_ALU_RGB_INST_0, code->alu.length); + for (i = 0; i < code->alu.length; i++) + OUT_CB(code->alu.inst[i].rgb_inst); + + OUT_CB_REG_SEQ(R300_US_ALU_RGB_ADDR_0, code->alu.length); + for (i = 0; i < code->alu.length; i++) + OUT_CB(code->alu.inst[i].rgb_addr); + + OUT_CB_REG_SEQ(R300_US_ALU_ALPHA_INST_0, code->alu.length); + for (i = 0; i < code->alu.length; i++) + OUT_CB(code->alu.inst[i].alpha_inst); + + OUT_CB_REG_SEQ(R300_US_ALU_ALPHA_ADDR_0, code->alu.length); + for (i = 0; i < code->alu.length; i++) + OUT_CB(code->alu.inst[i].alpha_addr); + + if (code->tex.length) { + OUT_CB_REG_SEQ(R300_US_TEX_INST_0, code->tex.length); + OUT_CB_TABLE(code->tex.inst, code->tex.length); + } + + /* Emit immediates. */ + if (imm_count) { + for(i = imm_first; i < imm_end; ++i) { + if (constants[i].Type == RC_CONSTANT_IMMEDIATE) { + const float *data = constants[i].u.Immediate; + + OUT_CB_REG_SEQ(R300_PFS_PARAM_0_X + i * 16, 4); + OUT_CB(pack_float24(data[0])); + OUT_CB(pack_float24(data[1])); + OUT_CB(pack_float24(data[2])); + OUT_CB(pack_float24(data[3])); + } + } + } + } + + OUT_CB_REG(R300_FG_DEPTH_SRC, shader->fg_depth_src); + OUT_CB_REG(R300_US_W_FMT, shader->us_out_w); + END_CB; +} + static void r300_translate_fragment_shader( struct r300_context* r300, struct r300_fragment_shader_code* shader, @@ -226,13 +366,14 @@ static void r300_translate_fragment_shader( { struct r300_fragment_program_compiler compiler; struct tgsi_to_rc ttr; - int wpos; + int wpos, face; unsigned i; tgsi_scan_shader(tokens, &shader->info); r300_shader_read_fs_inputs(&shader->info, &shader->inputs); wpos = shader->inputs.wpos; + face = shader->inputs.face; /* Setup the compiler. */ memset(&compiler, 0, sizeof(compiler)); @@ -241,15 +382,21 @@ static void r300_translate_fragment_shader( compiler.code = &shader->code; compiler.state = shader->compare_state; - compiler.is_r500 = r300->screen->caps.is_r500; - compiler.max_temp_regs = compiler.is_r500 ? 128 : 32; + compiler.Base.is_r500 = r300->screen->caps.is_r500; + compiler.Base.disable_optimizations = DBG_ON(r300, DBG_NO_OPT); + compiler.Base.has_half_swizzles = TRUE; + compiler.Base.has_presub = TRUE; + compiler.Base.max_temp_regs = compiler.Base.is_r500 ? 128 : 32; + compiler.Base.max_constants = compiler.Base.is_r500 ? 256 : 32; + compiler.Base.max_alu_insts = compiler.Base.is_r500 ? 512 : 64; + compiler.Base.remove_unused_constants = TRUE; compiler.AllocateHwInputs = &allocate_hardware_inputs; compiler.UserData = &shader->inputs; find_output_registers(&compiler, shader); if (compiler.Base.Debug) { - debug_printf("r300: Initial fragment program\n"); + DBG(r300, DBG_FP, "r300: Initial fragment program\n"); tgsi_dump(tokens, 0); } @@ -272,17 +419,13 @@ static void r300_translate_fragment_shader( rc_transform_fragment_wpos(&compiler.Base, wpos, wpos, TRUE); } + if (face != ATTR_UNUSED) { + rc_transform_fragment_face(&compiler.Base, face); + } + /* Invoke the compiler */ r3xx_compile_fragment_program(&compiler); - /* Shaders with zero instructions are invalid, - * use the dummy shader instead. */ - if (shader->code.code.r500.inst_end == -1) { - rc_destroy(&compiler.Base); - r300_dummy_fragment_shader(r300, shader); - return; - } - if (compiler.Base.Error) { fprintf(stderr, "r300 FP: Compiler Error:\n%sUsing a dummy shader" " instead.\n", compiler.Base.ErrorMsg); @@ -298,8 +441,21 @@ static void r300_translate_fragment_shader( return; } + /* Shaders with zero instructions are invalid, + * use the dummy shader instead. */ + if (shader->code.code.r500.inst_end == -1) { + rc_destroy(&compiler.Base); + r300_dummy_fragment_shader(r300, shader); + return; + } + /* Initialize numbers of constants for each type. */ - shader->externals_count = ttr.immediate_offset; + shader->externals_count = 0; + for (i = 0; + i < shader->code.constants.Count && + shader->code.constants.Constants[i].Type == RC_CONSTANT_EXTERNAL; i++) { + shader->externals_count = i+1; + } shader->immediates_count = 0; shader->rc_state_count = 0; @@ -327,6 +483,9 @@ static void r300_translate_fragment_shader( /* And, finally... */ rc_destroy(&compiler.Base); + + /* Build the command buffer. */ + r300_emit_fs_code_to_buffer(r300, shader); } boolean r300_pick_fragment_shader(struct r300_context* r300) diff --git a/src/gallium/drivers/r300/r300_fs.h b/src/gallium/drivers/r300/r300_fs.h index 1cc43553595..51bfa88c5ef 100644 --- a/src/gallium/drivers/r300/r300_fs.h +++ b/src/gallium/drivers/r300/r300_fs.h @@ -50,6 +50,9 @@ struct r300_fragment_shader_code { struct r300_fragment_program_external_state compare_state; struct rX00_fragment_program_code code; + unsigned cb_code_size; + uint32_t *cb_code; + struct r300_fragment_shader_code* next; }; diff --git a/src/gallium/drivers/r300/r300_hyperz.c b/src/gallium/drivers/r300/r300_hyperz.c index b41b6b1508d..eb5b0c36f8f 100644 --- a/src/gallium/drivers/r300/r300_hyperz.c +++ b/src/gallium/drivers/r300/r300_hyperz.c @@ -21,43 +21,220 @@ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE * USE OR OTHER DEALINGS IN THE SOFTWARE. */ - -#include "r300_hyperz.h" #include "r300_context.h" +#include "r300_hyperz.h" #include "r300_reg.h" #include "r300_fs.h" +#include "r300_winsys.h" + +#include "util/u_format.h" +#include "util/u_mm.h" + +/* + HiZ rules - taken from various docs + 1. HiZ only works on depth values + 2. Cannot HiZ if stencil fail or zfail is !KEEP + 3. on R300/400, HiZ is disabled if depth test is EQUAL + 4. comparison changes without clears usually mean disabling HiZ +*/ +/*****************************************************************************/ +/* The HyperZ setup */ +/*****************************************************************************/ + +static bool r300_get_sc_hz_max(struct r300_context *r300) +{ + struct r300_dsa_state *dsa_state = r300->dsa_state.state; + int func = dsa_state->z_stencil_control & 0x7; + int ret = R300_SC_HYPERZ_MIN; + + if (func >= 4 && func <= 7) + ret = R300_SC_HYPERZ_MAX; + return ret; +} + +static bool r300_zfunc_same_direction(int func1, int func2) +{ + /* func1 is less/lessthan */ + if (func1 == 1 || func1 == 2) + if (func2 == 3 || func2 == 4 || func2 == 5) + return FALSE; + + if (func2 == 1 || func2 == 2) + if (func1 == 4 || func1 == 5) + return FALSE; + return TRUE; +} + +static int r300_get_hiz_min(struct r300_context *r300) +{ + struct r300_dsa_state *dsa_state = r300->dsa_state.state; + int func = dsa_state->z_stencil_control & 0x7; + int ret = R300_HIZ_MIN; + + if (func == 1 || func == 2) + ret = R300_HIZ_MAX; + return ret; +} + +static boolean r300_dsa_stencil_op_not_keep(struct pipe_stencil_state *s) +{ + if (s->enabled && (s->fail_op != PIPE_STENCIL_OP_KEEP || + s->zfail_op != PIPE_STENCIL_OP_KEEP)) + return TRUE; + return FALSE; +} + +static boolean r300_can_hiz(struct r300_context *r300) +{ + struct r300_dsa_state *dsa_state = r300->dsa_state.state; + struct pipe_depth_stencil_alpha_state *dsa = &dsa_state->dsa; + struct r300_screen* r300screen = r300->screen; + struct r300_hyperz_state *z = r300->hyperz_state.state; + + /* shader writes depth - no HiZ */ + if (r300_fragment_shader_writes_depth(r300_fs(r300))) /* (5) */ + return FALSE; + + if (r300->query_current) + return FALSE; + /* if stencil fail/zfail op is not KEEP */ + if (r300_dsa_stencil_op_not_keep(&dsa->stencil[0]) || + r300_dsa_stencil_op_not_keep(&dsa->stencil[1])) + return FALSE; + + if (dsa->depth.enabled) { + /* if depth func is EQUAL pre-r500 */ + if (dsa->depth.func == PIPE_FUNC_EQUAL && !r300screen->caps.is_r500) + return FALSE; + /* if depth func is NOTEQUAL */ + if (dsa->depth.func == PIPE_FUNC_NOTEQUAL) + return FALSE; + } + /* depth comparison function - if just cleared save and return okay */ + if (z->current_func == -1) { + int func = dsa_state->z_stencil_control & 0x7; + if (func != 0 && func != 7) + z->current_func = dsa_state->z_stencil_control & 0x7; + } else { + /* simple don't change */ + if (!r300_zfunc_same_direction(z->current_func, (dsa_state->z_stencil_control & 0x7))) { + DBG(r300, DBG_HYPERZ, "z func changed direction - disabling hyper-z %d -> %d\n", z->current_func, dsa_state->z_stencil_control); + return FALSE; + } + } + return TRUE; +} + +static void r300_update_hyperz(struct r300_context* r300) +{ + struct r300_hyperz_state *z = + (struct r300_hyperz_state*)r300->hyperz_state.state; + struct pipe_framebuffer_state *fb = + (struct pipe_framebuffer_state*)r300->fb_state.state; + struct r300_texture *zstex = + fb->zsbuf ? r300_texture(fb->zsbuf->texture) : NULL; + boolean zmask_in_use = FALSE; + boolean hiz_in_use = FALSE; + + z->gb_z_peq_config = 0; + z->zb_bw_cntl = 0; + z->sc_hyperz = R300_SC_HYPERZ_ADJ_2; + z->flush = 0; + + if (r300->cbzb_clear) { + z->zb_bw_cntl |= R300_ZB_CB_CLEAR_CACHE_LINE_WRITE_ONLY; + return; + } + + if (!zstex) + return; + + if (!r300->rws->get_value(r300->rws, R300_CAN_HYPERZ)) + return; + + zmask_in_use = zstex->zmask_in_use[fb->zsbuf->level]; + hiz_in_use = zstex->hiz_in_use[fb->zsbuf->level]; + + /* Z fastfill. */ + if (zmask_in_use) { + z->zb_bw_cntl |= R300_FAST_FILL_ENABLE; /* | R300_FORCE_COMPRESSED_STENCIL_VALUE_ENABLE;*/ + } + + /* Zbuffer compression. */ + if (zmask_in_use && r300->z_compression) { + z->zb_bw_cntl |= R300_RD_COMP_ENABLE; + if (r300->z_decomp_rd == false) + z->zb_bw_cntl |= R300_WR_COMP_ENABLE; + } + /* RV350 and up optimizations. */ + /* The section 10.4.9 in the docs is a lie. */ + if (r300->z_compression == RV350_Z_COMPRESS_88) + z->gb_z_peq_config |= R300_GB_Z_PEQ_CONFIG_Z_PEQ_SIZE_8_8; + + if (hiz_in_use) { + bool can_hiz = r300_can_hiz(r300); + if (can_hiz) { + z->zb_bw_cntl |= R300_HIZ_ENABLE; + z->sc_hyperz |= R300_SC_HYPERZ_ENABLE; + z->sc_hyperz |= r300_get_sc_hz_max(r300); + z->zb_bw_cntl |= r300_get_hiz_min(r300); + } + } + + /* R500-specific features and optimizations. */ + if (r300->screen->caps.is_r500) { + z->zb_bw_cntl |= R500_HIZ_FP_EXP_BITS_3; + z->zb_bw_cntl |= + R500_HIZ_EQUAL_REJECT_ENABLE | + R500_PEQ_PACKING_ENABLE | + R500_COVERED_PTR_MASKING_ENABLE; + } +} /*****************************************************************************/ /* The ZTOP state */ /*****************************************************************************/ -static boolean r300_dsa_writes_depth_stencil(struct r300_dsa_state* dsa) +static boolean r300_dsa_writes_stencil( + struct pipe_stencil_state *s) +{ + return s->enabled && s->writemask && + (s->fail_op != PIPE_STENCIL_OP_KEEP || + s->zfail_op != PIPE_STENCIL_OP_KEEP || + s->zpass_op != PIPE_STENCIL_OP_KEEP); +} + +static boolean r300_dsa_writes_depth_stencil( + struct pipe_depth_stencil_alpha_state *dsa) { - /* We are interested only in the cases when a new depth or stencil value - * can be written and changed. */ + /* We are interested only in the cases when a depth or stencil value + * can be changed. */ + + if (dsa->depth.enabled && dsa->depth.writemask && + dsa->depth.func != PIPE_FUNC_NEVER) + return TRUE; - /* We might optionally check for [Z func: never] and inspect the stencil - * state in a similar fashion, but it's not terribly important. */ - return (dsa->z_buffer_control & R300_Z_WRITE_ENABLE) || - (dsa->stencil_ref_mask & R300_STENCILWRITEMASK_MASK) || - ((dsa->z_buffer_control & R500_STENCIL_REFMASK_FRONT_BACK) && - (dsa->stencil_ref_bf & R300_STENCILWRITEMASK_MASK)); + if (r300_dsa_writes_stencil(&dsa->stencil[0]) || + r300_dsa_writes_stencil(&dsa->stencil[1])) + return TRUE; + + return FALSE; } -static boolean r300_dsa_alpha_test_enabled(struct r300_dsa_state* dsa) +static boolean r300_dsa_alpha_test_enabled( + struct pipe_depth_stencil_alpha_state *dsa) { /* We are interested only in the cases when alpha testing can kill * a fragment. */ - uint32_t af = dsa->alpha_function; - return (af & R300_FG_ALPHA_FUNC_ENABLE) && - (af & R300_FG_ALPHA_FUNC_ALWAYS) != R300_FG_ALPHA_FUNC_ALWAYS; + return dsa->alpha.enabled && dsa->alpha.func != PIPE_FUNC_ALWAYS; } static void r300_update_ztop(struct r300_context* r300) { struct r300_ztop_state* ztop_state = (struct r300_ztop_state*)r300->ztop_state.state; + uint32_t old_ztop = ztop_state->z_buffer_top; /* This is important enough that I felt it warranted a comment. * @@ -98,11 +275,139 @@ static void r300_update_ztop(struct r300_context* r300) } else { ztop_state->z_buffer_top = R300_ZTOP_ENABLE; } + if (ztop_state->z_buffer_top != old_ztop) + r300->ztop_state.dirty = TRUE; +} + +#define ALIGN_DIVUP(x, y) (((x) + (y) - 1) / (y)) + +static void r300_update_hiz_clear(struct r300_context *r300) +{ + struct pipe_framebuffer_state *fb = + (struct pipe_framebuffer_state*)r300->fb_state.state; + uint32_t height; + + height = ALIGN_DIVUP(fb->zsbuf->height, 4); + r300->hiz_clear.size = height * 4; +} + +static void r300_update_zmask_clear(struct r300_context *r300) +{ + struct pipe_framebuffer_state *fb = + (struct pipe_framebuffer_state*)r300->fb_state.state; + uint32_t height; + int mult; - r300->ztop_state.dirty = TRUE; + if (r300->z_compression == RV350_Z_COMPRESS_88) + mult = 8; + else + mult = 4; + + height = ALIGN_DIVUP(fb->zsbuf->height, mult); + + r300->zmask_clear.size = height * 4; } void r300_update_hyperz_state(struct r300_context* r300) { r300_update_ztop(r300); + if (r300->hyperz_state.dirty) { + r300_update_hyperz(r300); + } + + if (r300->hiz_clear.dirty) { + r300_update_hiz_clear(r300); + } + if (r300->zmask_clear.dirty) { + r300_update_zmask_clear(r300); + } +} + +void r300_hiz_alloc_block(struct r300_context *r300, struct r300_surface *surf) +{ + struct r300_texture *tex; + uint32_t zsize, ndw; + int level = surf->base.level; + + tex = r300_texture(surf->base.texture); + + if (tex->hiz_mem[level]) + return; + + zsize = tex->desc.layer_size_in_bytes[level]; + zsize /= util_format_get_blocksize(tex->desc.b.b.format); + ndw = ALIGN_DIVUP(zsize, 64); + + tex->hiz_mem[level] = u_mmAllocMem(r300->hiz_mm, ndw, 0, 0); + return; +} + +void r300_zmask_alloc_block(struct r300_context *r300, struct r300_surface *surf, int compress) +{ + int bsize = 256; + uint32_t zsize, ndw; + int level = surf->base.level; + struct r300_texture *tex; + + tex = r300_texture(surf->base.texture); + + /* We currently don't handle decompression for 3D textures and cubemaps + * correctly. */ + if (tex->desc.b.b.target != PIPE_TEXTURE_1D && + tex->desc.b.b.target != PIPE_TEXTURE_2D && + tex->desc.b.b.target != PIPE_TEXTURE_RECT) + return; + + /* Cannot flush zmask of 16-bit zbuffers. */ + if (util_format_get_blocksizebits(tex->desc.b.b.format) == 16) + return; + + if (tex->zmask_mem[level]) + return; + + zsize = tex->desc.layer_size_in_bytes[level]; + zsize /= util_format_get_blocksize(tex->desc.b.b.format); + + /* each zmask dword represents 16 4x4 blocks - which is 256 pixels + or 16 8x8 depending on the gb peq flag = 1024 pixels */ + if (compress == RV350_Z_COMPRESS_88) + bsize = 1024; + + ndw = ALIGN_DIVUP(zsize, bsize); + tex->zmask_mem[level] = u_mmAllocMem(r300->zmask_mm, ndw, 0, 0); + return; +} + +boolean r300_hyperz_init_mm(struct r300_context *r300) +{ + struct r300_screen* r300screen = r300->screen; + int frag_pipes = r300screen->caps.num_frag_pipes; + + r300->zmask_mm = u_mmInit(0, r300screen->caps.zmask_ram * frag_pipes); + if (!r300->zmask_mm) + return FALSE; + + if (r300screen->caps.hiz_ram) { + r300->hiz_mm = u_mmInit(0, r300screen->caps.hiz_ram * frag_pipes); + if (!r300->hiz_mm) { + u_mmDestroy(r300->zmask_mm); + r300->zmask_mm = NULL; + return FALSE; + } + } + + return TRUE; +} + +void r300_hyperz_destroy_mm(struct r300_context *r300) +{ + struct r300_screen* r300screen = r300->screen; + + if (r300screen->caps.hiz_ram) { + u_mmDestroy(r300->hiz_mm); + r300->hiz_mm = NULL; + } + + u_mmDestroy(r300->zmask_mm); + r300->zmask_mm = NULL; } diff --git a/src/gallium/drivers/r300/r300_hyperz.h b/src/gallium/drivers/r300/r300_hyperz.h index 3df5053b896..30a23ec6493 100644 --- a/src/gallium/drivers/r300/r300_hyperz.h +++ b/src/gallium/drivers/r300/r300_hyperz.h @@ -27,4 +27,9 @@ struct r300_context; void r300_update_hyperz_state(struct r300_context* r300); +void r300_hiz_alloc_block(struct r300_context *r300, struct r300_surface *surf); +void r300_zmask_alloc_block(struct r300_context *r300, struct r300_surface *surf, int compress); + +boolean r300_hyperz_init_mm(struct r300_context *r300); +void r300_hyperz_destroy_mm(struct r300_context *r300); #endif diff --git a/src/gallium/drivers/r300/r300_public.h b/src/gallium/drivers/r300/r300_public.h new file mode 100644 index 00000000000..8e7a963c55d --- /dev/null +++ b/src/gallium/drivers/r300/r300_public.h @@ -0,0 +1,9 @@ + +#ifndef R300_PUBLIC_H +#define R300_PUBLIC_H + +struct r300_winsys_screen; + +struct pipe_screen* r300_screen_create(struct r300_winsys_screen *rws); + +#endif diff --git a/src/gallium/drivers/r300/r300_query.c b/src/gallium/drivers/r300/r300_query.c index 5c27796e894..5f34fcb2744 100644 --- a/src/gallium/drivers/r300/r300_query.c +++ b/src/gallium/drivers/r300/r300_query.c @@ -26,7 +26,7 @@ #include "r300_context.h" #include "r300_screen.h" #include "r300_emit.h" -#include "r300_query.h" +#include "r300_winsys.h" #include <stdio.h> @@ -35,31 +35,31 @@ static struct pipe_query *r300_create_query(struct pipe_context *pipe, { struct r300_context *r300 = r300_context(pipe); struct r300_screen *r300screen = r300->screen; - unsigned query_size; - struct r300_query *q, *qptr; + struct r300_query *q; + + if (query_type != PIPE_QUERY_OCCLUSION_COUNTER) { + return NULL; + } q = CALLOC_STRUCT(r300_query); + if (!q) + return NULL; q->type = query_type; - assert(q->type == PIPE_QUERY_OCCLUSION_COUNTER); - - q->active = FALSE; + q->domain = R300_DOMAIN_GTT; + q->buffer_size = 4096; if (r300screen->caps.family == CHIP_FAMILY_RV530) - query_size = r300screen->caps.num_z_pipes * sizeof(uint32_t); + q->num_pipes = r300screen->caps.num_z_pipes; else - query_size = r300screen->caps.num_frag_pipes * sizeof(uint32_t); + q->num_pipes = r300screen->caps.num_frag_pipes; - if (!is_empty_list(&r300->query_list)) { - qptr = last_elem(&r300->query_list); - q->offset = qptr->offset + query_size; - } insert_at_tail(&r300->query_list, q); - /* XXX */ - if (q->offset >= 4096) { - q->offset = 0; - } + /* Open up the occlusion query buffer. */ + q->buffer = r300->rws->buffer_create(r300->rws, q->buffer_size, 4096, + PIPE_BIND_CUSTOM, PIPE_USAGE_STREAM, + q->domain); return (struct pipe_query*)q; } @@ -67,92 +67,87 @@ static struct pipe_query *r300_create_query(struct pipe_context *pipe, static void r300_destroy_query(struct pipe_context* pipe, struct pipe_query* query) { - struct r300_query* q = (struct r300_query*)query; + struct r300_context *r300 = r300_context(pipe); + struct r300_query* q = r300_query(query); + r300->rws->buffer_reference(r300->rws, &q->buffer, NULL); remove_from_list(q); FREE(query); } +void r300_resume_query(struct r300_context *r300, + struct r300_query *query) +{ + r300->query_current = query; + r300->query_start.dirty = TRUE; +} + static void r300_begin_query(struct pipe_context* pipe, struct pipe_query* query) { - uint32_t value = ~0U; struct r300_context* r300 = r300_context(pipe); - struct r300_query* q = (struct r300_query*)query; + struct r300_query* q = r300_query(query); - assert(r300->query_current == NULL); + if (r300->query_current != NULL) { + fprintf(stderr, "r300: begin_query: " + "Some other query has already been started.\n"); + assert(0); + return; + } - pipe_buffer_write(pipe, - r300->oqbo, - q->offset, - sizeof value, - &value); + q->num_results = 0; + r300_resume_query(r300, q); +} - q->flushed = FALSE; - r300->query_current = q; - r300->query_start.dirty = TRUE; +void r300_stop_query(struct r300_context *r300) +{ + r300_emit_query_end(r300); + r300->query_current = NULL; } static void r300_end_query(struct pipe_context* pipe, struct pipe_query* query) { struct r300_context* r300 = r300_context(pipe); - struct r300_query* q = (struct r300_query*)query; + struct r300_query *q = r300_query(query); - r300_emit_query_end(r300); - q->begin_emitted = false; - r300->query_current = NULL; + if (q != r300->query_current) { + fprintf(stderr, "r300: end_query: Got invalid query.\n"); + assert(0); + return; + } + + r300_stop_query(r300); } static boolean r300_get_query_result(struct pipe_context* pipe, struct pipe_query* query, boolean wait, - uint64_t* result) + void* vresult) { struct r300_context* r300 = r300_context(pipe); - struct r300_screen* r300screen = r300->screen; - struct r300_query *q = (struct r300_query*)query; - struct pipe_transfer *transfer; - unsigned flags = PIPE_TRANSFER_READ; - uint32_t* map; - uint32_t temp = 0; - unsigned i, num_results; - - if (q->flushed == FALSE) + struct r300_query *q = r300_query(query); + unsigned flags, i; + uint32_t temp, *map; + uint64_t *result = (uint64_t*)vresult; + + if (!q->flushed) pipe->flush(pipe, 0, NULL); - if (!wait) { - flags |= PIPE_TRANSFER_DONTBLOCK; - } - map = pipe_buffer_map(pipe, r300->oqbo, flags, &transfer); + flags = PIPE_TRANSFER_READ | (!wait ? PIPE_TRANSFER_DONTBLOCK : 0); + + map = r300->rws->buffer_map(r300->rws, q->buffer, r300->cs, flags); if (!map) return FALSE; - map += q->offset / 4; - if (r300screen->caps.family == CHIP_FAMILY_RV530) - num_results = r300screen->caps.num_z_pipes; - else - num_results = r300screen->caps.num_frag_pipes; - - for (i = 0; i < num_results; i++) { - if (*map == ~0U) { - /* Looks like our results aren't ready yet. */ - if (wait) { - fprintf(stderr, "r300: Despite waiting, OQ results haven't " - "come in yet.\n"); - } - temp = ~0U; - break; - } + /* Sum up the results. */ + temp = 0; + for (i = 0; i < q->num_results; i++) { temp += *map; map++; } - pipe_buffer_unmap(pipe, r300->oqbo, transfer); - if (temp == ~0U) { - /* Our results haven't been written yet... */ - return FALSE; - } + r300->rws->buffer_unmap(r300->rws, q->buffer); *result = temp; return TRUE; @@ -163,7 +158,7 @@ static void r300_render_condition(struct pipe_context *pipe, uint mode) { struct r300_context *r300 = r300_context(pipe); - uint64_t result; + uint64_t result = 0; boolean wait; if (query) { @@ -172,19 +167,69 @@ static void r300_render_condition(struct pipe_context *pipe, if (!r300_get_query_result(pipe, query, wait, &result)) { r300->skip_rendering = FALSE; + } else { + r300->skip_rendering = result == 0; } - - r300->skip_rendering = result == 0; } else { r300->skip_rendering = FALSE; } } +/*************************************************************************** + * Fake occlusion queries (for debugging) + ***************************************************************************/ + +static unsigned r300_fake_query; + +static struct pipe_query *r300_fake_create_query(struct pipe_context *pipe, + unsigned query_type) +{ + return (struct pipe_query*)&r300_fake_query; +} + +static void r300_fake_destroy_query(struct pipe_context* pipe, + struct pipe_query* query) +{ +} + +static void r300_fake_begin_query(struct pipe_context* pipe, + struct pipe_query* query) +{ +} + +static void r300_fake_end_query(struct pipe_context* pipe, + struct pipe_query* query) +{ +} + +static boolean r300_fake_get_query_result(struct pipe_context* pipe, + struct pipe_query* query, + boolean wait, void* vresult) +{ + uint64_t *result = (uint64_t*)vresult; + *result = 1000000; + return TRUE; +} + +static void r300_fake_render_condition(struct pipe_context *pipe, + struct pipe_query *query, uint mode) +{ +} + void r300_init_query_functions(struct r300_context* r300) { - r300->context.create_query = r300_create_query; - r300->context.destroy_query = r300_destroy_query; - r300->context.begin_query = r300_begin_query; - r300->context.end_query = r300_end_query; - r300->context.get_query_result = r300_get_query_result; - r300->context.render_condition = r300_render_condition; + if (DBG_ON(r300, DBG_FAKE_OCC)) { + r300->context.create_query = r300_fake_create_query; + r300->context.destroy_query = r300_fake_destroy_query; + r300->context.begin_query = r300_fake_begin_query; + r300->context.end_query = r300_fake_end_query; + r300->context.get_query_result = r300_fake_get_query_result; + r300->context.render_condition = r300_fake_render_condition; + } else { + r300->context.create_query = r300_create_query; + r300->context.destroy_query = r300_destroy_query; + r300->context.begin_query = r300_begin_query; + r300->context.end_query = r300_end_query; + r300->context.get_query_result = r300_get_query_result; + r300->context.render_condition = r300_render_condition; + } } diff --git a/src/gallium/drivers/r300/r300_query.h b/src/gallium/drivers/r300/r300_query.h deleted file mode 100644 index 48876da3123..00000000000 --- a/src/gallium/drivers/r300/r300_query.h +++ /dev/null @@ -1,35 +0,0 @@ -/* - * Copyright 2009 Corbin Simpson <[email protected]> - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * on the rights to use, copy, modify, merge, publish, distribute, sub - * license, and/or sell copies of the Software, and to permit persons to whom - * the Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice (including the next - * paragraph) shall be included in all copies or substantial portions of the - * Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL - * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, - * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR - * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE - * USE OR OTHER DEALINGS IN THE SOFTWARE. */ - -#ifndef R300_QUERY_H -#define R300_QUERY_H - -struct r300_context; - -static INLINE struct r300_query* r300_query(struct pipe_query* q) -{ - return (struct r300_query*)q; -} - -void r300_init_query_functions(struct r300_context* r300); - -#endif /* R300_QUERY_H */ diff --git a/src/gallium/drivers/r300/r300_reg.h b/src/gallium/drivers/r300/r300_reg.h index 675a9317f9c..6bea783f697 100644 --- a/src/gallium/drivers/r300/r300_reg.h +++ b/src/gallium/drivers/r300/r300_reg.h @@ -496,6 +496,12 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. #define R300_VAP_GB_HORZ_CLIP_ADJ 0x2228 #define R300_VAP_GB_HORZ_DISC_ADJ 0x222c +#define R300_VAP_PVS_FLOW_CNTL_ADDRS_0 0x2230 +#define R300_PVS_FC_ACT_ADRS(x) ((x) << 0) +#define R300_PVS_FC_LOOP_CNT_JMP_INST(x) ((x) << 8) +#define R300_PVS_FC_LAST_INST(x) ((x) << 16) +#define R300_PVS_FC_RTN_INST(x) ((x) << 24) + /* gap */ /* Sometimes, END_OF_PKT and 0x2284=0 are the only commands sent between @@ -514,6 +520,10 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. # define R300_2288_R300 0x00750000 /* -- nh */ # define R300_2288_RV350 0x0000FFFF /* -- Vladimir */ +#define R300_VAP_PVS_FLOW_CNTL_LOOP_INDEX_0 0x2290 +#define R300_PVS_FC_LOOP_INIT_VAL(x) ((x) << 0) +#define R300_PVS_FC_LOOP_STEP_VAL(x) ((x) << 8) + /* gap */ /* Addresses are relative to the vertex program instruction area of the @@ -548,6 +558,9 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. #define R300_VAP_PVS_CODE_CNTL_1 0x22D8 # define R300_PVS_LAST_VTX_SRC_INST_SHIFT 0 #define R300_VAP_PVS_FLOW_CNTL_OPC 0x22DC +#define R300_VAP_PVS_FC_OPC_JUMP(x) (1 << (2 * (x))) +#define R300_VAP_PVS_FC_OPC_LOOP(x) (2 << (2 * (x))) +#define R300_VAP_PVS_FC_OPC_JSR(x) (3 << (2 * (x))) /* The entire range from 0x2300 to 0x2AC inclusive seems to be used for * immediate vertices @@ -564,6 +577,14 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. /* write 0 to indicate end of packet? */ #define R300_VAP_VTX_END_OF_PKT 0x24AC +#define R500_VAP_PVS_FLOW_CNTL_ADDRS_LW_0 0x2500 +#define R500_PVS_FC_ACT_ADRS(x) ((x) << 0) +#define R500_PVS_FC_LOOP_CNT_JMP_INST(x) ((x) << 16) + +#define R500_VAP_PVS_FLOW_CNTL_ADDRS_UW_0 0x2504 +#define R500_PVS_FC_LAST_INST(x) ((x) << 0) +#define R500_PVS_FC_RTN_INST(x) ((x) << 16) + /* gap */ /* These are values from r300_reg/r300_reg.h - they are known to be correct @@ -1575,6 +1596,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. # define R500_TX_FORMAT_Y10X10 0x3 # define R500_TX_FORMAT_W10Z10Y10X10 0x4 # define R500_TX_FORMAT_ATI1N 0x5 +# define R500_TX_FORMAT_Y8X24 0x6 # define R300_TX_FORMAT_SIGNED_W (1 << 5) @@ -1585,6 +1607,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. # define R300_TX_FORMAT_3D (1 << 25) # define R300_TX_FORMAT_CUBIC_MAP (2 << 25) +# define R300_TX_FORMAT_TEX_COORD_TYPE_MASK (0x3 << 25) /* alpha modes, convenience mostly */ /* if you have alpha, pick constant appropriate to the @@ -1629,6 +1652,40 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. # define R300_TX_FORMAT_GAMMA (1 << 21) # define R300_TX_FORMAT_YUV_TO_RGB (1 << 22) +# define R300_TX_CACHE(x) ((x) << 27) +# define R300_TX_CACHE_WHOLE 0 +/* reserved */ +# define R300_TX_CACHE_HALF_0 2 +# define R300_TX_CACHE_HALF_1 3 +# define R300_TX_CACHE_FOURTH_0 4 +# define R300_TX_CACHE_FOURTH_1 5 +# define R300_TX_CACHE_FOURTH_2 6 +# define R300_TX_CACHE_FOURTH_3 7 +# define R300_TX_CACHE_EIGHTH_0 8 +# define R300_TX_CACHE_EIGHTH_1 9 +# define R300_TX_CACHE_EIGHTH_2 10 +# define R300_TX_CACHE_EIGHTH_3 11 +# define R300_TX_CACHE_EIGHTH_4 12 +# define R300_TX_CACHE_EIGHTH_5 13 +# define R300_TX_CACHE_EIGHTH_6 14 +# define R300_TX_CACHE_EIGHTH_7 15 +# define R300_TX_CACHE_SIXTEENTH_0 16 +# define R300_TX_CACHE_SIXTEENTH_1 17 +# define R300_TX_CACHE_SIXTEENTH_2 18 +# define R300_TX_CACHE_SIXTEENTH_3 19 +# define R300_TX_CACHE_SIXTEENTH_4 20 +# define R300_TX_CACHE_SIXTEENTH_5 21 +# define R300_TX_CACHE_SIXTEENTH_6 22 +# define R300_TX_CACHE_SIXTEENTH_7 23 +# define R300_TX_CACHE_SIXTEENTH_8 24 +# define R300_TX_CACHE_SIXTEENTH_9 25 +# define R300_TX_CACHE_SIXTEENTH_10 26 +# define R300_TX_CACHE_SIXTEENTH_11 27 +# define R300_TX_CACHE_SIXTEENTH_12 28 +# define R300_TX_CACHE_SIXTEENTH_13 29 +# define R300_TX_CACHE_SIXTEENTH_14 30 +# define R300_TX_CACHE_SIXTEENTH_15 31 + #define R300_TX_FORMAT2_0 0x4500 /* obvious missing in gap */ # define R300_TX_PITCHMASK_SHIFT 0 # define R300_TX_PITCHMASK_MASK (2047 << 0) @@ -2582,7 +2639,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. # define R300_WR_COMP_DISABLE (0 << 4) # define R300_WR_COMP_ENABLE (1 << 4) # define R300_ZB_CB_CLEAR_RMW (0 << 5) -# define R300_ZB_CB_CLEAR_CACHE_LINEAR (1 << 5) +# define R300_ZB_CB_CLEAR_CACHE_LINE_WRITE_ONLY (1 << 5) # define R300_FORCE_COMPRESSED_STENCIL_VALUE_DISABLE (0 << 6) # define R300_FORCE_COMPRESSED_STENCIL_VALUE_ENABLE (1 << 6) @@ -2638,6 +2695,24 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. /* Z Buffer Clear Value */ #define R300_ZB_DEPTHCLEARVALUE 0x4f28 +/* Z Mask RAM is a Z compression buffer. + * Each dword of the Z Mask contains compression info for 16 4x4 pixel blocks, + * that is 2 bits for each block. + * On chips with 2 Z pipes, every other dword maps to a different pipe. + */ + +/* The dword offset into Z mask RAM (bits 18:4) */ +#define R300_ZB_ZMASK_OFFSET 0x4f30 + +/* Z Mask Pitch. */ +#define R300_ZB_ZMASK_PITCH 0x4f34 + +/* Access to Z Mask RAM in a manner similar to HiZ RAM. + * The indices are autoincrementing. */ +#define R300_ZB_ZMASK_WRINDEX 0x4f38 +#define R300_ZB_ZMASK_DWORD 0x4f3c +#define R300_ZB_ZMASK_RDINDEX 0x4f40 + /* Hierarchical Z Memory Offset */ #define R300_ZB_HIZ_OFFSET 0x4f44 @@ -3229,8 +3304,8 @@ enum { # define R500_FC_B_OP0_NONE (0 << 24) # define R500_FC_B_OP0_DECR (1 << 24) # define R500_FC_B_OP0_INCR (2 << 24) -# define R500_FC_B_OP1_DECR (0 << 26) -# define R500_FC_B_OP1_NONE (1 << 26) +# define R500_FC_B_OP1_NONE (0 << 26) +# define R500_FC_B_OP1_DECR (1 << 26) # define R500_FC_B_OP1_INCR (2 << 26) # define R500_FC_IGNORE_UNCOVERED (1 << 28) #define R500_US_FC_INT_CONST_0 0x4c00 @@ -3383,6 +3458,7 @@ enum { # define R300_VBPNTR_SIZE1(x) (((x) >> 2) << 16) # define R300_VBPNTR_STRIDE1(x) (((x) >> 2) << 24) +#define R300_PACKET3_3D_CLEAR_ZMASK 0x00003200 #define R300_PACKET3_INDX_BUFFER 0x00003300 # define R300_INDX_BUFFER_DST_SHIFT 0 # define R300_INDX_BUFFER_SKIP_SHIFT 16 @@ -3436,9 +3512,18 @@ enum { # define RADEON_WAIT_3D_IDLECLEAN (1 << 17) # define RADEON_WAIT_HOST_IDLECLEAN (1 << 18) +#define R200_3D_DRAW_IMMD_2 0xC0003500 + +#define RADEON_CP_PACKET0 0x0 /* XXX stolen from radeon_reg.h */ #define RADEON_CP_PACKET3 0xC0000000 -#define R200_3D_DRAW_IMMD_2 0xC0003500 +#define RADEON_ONE_REG_WR (1 << 15) + +#define CP_PACKET0(register, count) \ + (RADEON_CP_PACKET0 | ((count) << 16) | ((register) >> 2)) + +#define CP_PACKET3(op, count) \ + (RADEON_CP_PACKET3 | (op) | ((count) << 16)) #endif /* _R300_REG_H */ diff --git a/src/gallium/drivers/r300/r300_render.c b/src/gallium/drivers/r300/r300_render.c index 7c3a7902a49..2f00c878f5b 100644 --- a/src/gallium/drivers/r300/r300_render.c +++ b/src/gallium/drivers/r300/r300_render.c @@ -1,5 +1,6 @@ /* * Copyright 2009 Corbin Simpson <[email protected]> + * Copyright 2010 Marek Olšák <[email protected]> * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), @@ -38,9 +39,12 @@ #include "r300_screen_buffer.h" #include "r300_emit.h" #include "r300_reg.h" -#include "r300_render.h" #include "r300_state_derived.h" +#include <limits.h> + +#define IMMD_DWORDS 32 + static uint32_t r300_translate_primitive(unsigned prim) { switch (prim) { @@ -114,16 +118,178 @@ static uint32_t r300_provoking_vertex_fixes(struct r300_context *r300, return color_control; } -/* Check if the requested number of dwords is available in the CS and - * if not, flush. Return TRUE if the flush occured. */ -static boolean r300_reserve_cs_space(struct r300_context *r300, - unsigned dwords) +boolean r500_index_bias_supported(struct r300_context *r300) +{ + return r300->screen->caps.is_r500 && + r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0); +} + +void r500_emit_index_bias(struct r300_context *r300, int index_bias) { - if (!r300->rws->check_cs(r300->rws, dwords)) { + CS_LOCALS(r300); + + BEGIN_CS(2); + OUT_CS_REG(R500_VAP_INDEX_OFFSET, + (index_bias & 0xFFFFFF) | (index_bias < 0 ? 1<<24 : 0)); + END_CS; +} + +/* This function splits the index bias value into two parts: + * - buffer_offset: the value that can be safely added to buffer offsets + * in r300_emit_aos (it must yield a positive offset when added to + * a vertex buffer offset) + * - index_offset: the value that must be manually subtracted from indices + * in an index buffer to achieve negative offsets. */ +static void r300_split_index_bias(struct r300_context *r300, int index_bias, + int *buffer_offset, int *index_offset) +{ + struct pipe_vertex_buffer *vb, *vbufs = r300->vertex_buffer; + struct pipe_vertex_element *velem = r300->velems->velem; + unsigned i, size; + int max_neg_bias; + + if (index_bias < 0) { + /* See how large index bias we may subtract. We must be careful + * here because negative buffer offsets are not allowed + * by the DRM API. */ + max_neg_bias = INT_MAX; + for (i = 0; i < r300->velems->count; i++) { + vb = &vbufs[velem[i].vertex_buffer_index]; + size = (vb->buffer_offset + velem[i].src_offset) / vb->stride; + max_neg_bias = MIN2(max_neg_bias, size); + } + + /* Now set the minimum allowed value. */ + *buffer_offset = MAX2(-max_neg_bias, index_bias); + } else { + /* A positive index bias is OK. */ + *buffer_offset = index_bias; + } + + *index_offset = index_bias - *buffer_offset; +} + +enum r300_prepare_flags { + PREP_FIRST_DRAW = (1 << 0), /* call emit_dirty_state and friends? */ + PREP_VALIDATE_VBOS = (1 << 1), /* validate VBOs? */ + PREP_EMIT_AOS = (1 << 2), /* call emit_aos? */ + PREP_EMIT_AOS_SWTCL = (1 << 3), /* call emit_aos_swtcl? */ + PREP_INDEXED = (1 << 4) /* is this draw_elements? */ +}; + +/** + * Check if the requested number of dwords is available in the CS and + * if not, flush. + * \param r300 The context. + * \param flags See r300_prepare_flags. + * \param cs_dwords The number of dwords to reserve in CS. + * \return TRUE if the CS was flushed + */ +static boolean r300_reserve_cs_dwords(struct r300_context *r300, + enum r300_prepare_flags flags, + unsigned cs_dwords) +{ + boolean flushed = FALSE; + boolean first_draw = flags & PREP_FIRST_DRAW; + boolean emit_aos = flags & PREP_EMIT_AOS; + boolean emit_aos_swtcl = flags & PREP_EMIT_AOS_SWTCL; + boolean hw_index_bias = r500_index_bias_supported(r300); + + /* Add dirty state, index offset, and AOS. */ + if (first_draw) { + cs_dwords += r300_get_num_dirty_dwords(r300); + + if (hw_index_bias) + cs_dwords += 2; /* emit_index_offset */ + + if (emit_aos) + cs_dwords += 55; /* emit_aos */ + + if (emit_aos_swtcl) + cs_dwords += 7; /* emit_aos_swtcl */ + } + + cs_dwords += r300_get_num_cs_end_dwords(r300); + + /* Reserve requested CS space. */ + if (cs_dwords > (r300->cs->ndw - r300->cs->cdw)) { r300->context.flush(&r300->context, 0, NULL); - return TRUE; + flushed = TRUE; } - return FALSE; + + return flushed; +} + +/** + * Validate buffers and emit dirty state. + * \param r300 The context. + * \param flags See r300_prepare_flags. + * \param index_buffer The index buffer to validate. The parameter may be NULL. + * \param aos_offset The offset passed to emit_aos. + * \param index_bias The index bias to emit. + * \return TRUE if rendering should be skipped + */ +static boolean r300_emit_states(struct r300_context *r300, + enum r300_prepare_flags flags, + struct pipe_resource *index_buffer, + int aos_offset, + int index_bias) +{ + boolean first_draw = flags & PREP_FIRST_DRAW; + boolean emit_aos = flags & PREP_EMIT_AOS; + boolean emit_aos_swtcl = flags & PREP_EMIT_AOS_SWTCL; + boolean indexed = flags & PREP_INDEXED; + boolean hw_index_bias = r500_index_bias_supported(r300); + + /* Validate buffers and emit dirty state if needed. */ + if (first_draw) { + if (!r300_emit_buffer_validate(r300, flags & PREP_VALIDATE_VBOS, + index_buffer)) { + fprintf(stderr, "r300: CS space validation failed. " + "(not enough memory?) Skipping rendering.\n"); + return FALSE; + } + + r300_emit_dirty_state(r300); + if (hw_index_bias) { + if (r300->screen->caps.has_tcl) + r500_emit_index_bias(r300, index_bias); + else + r500_emit_index_bias(r300, 0); + } + + if (emit_aos) + r300_emit_aos(r300, aos_offset, indexed); + + if (emit_aos_swtcl) + r300_emit_aos_swtcl(r300, indexed); + } + + return TRUE; +} + +/** + * Check if the requested number of dwords is available in the CS and + * if not, flush. Then validate buffers and emit dirty state. + * \param r300 The context. + * \param flags See r300_prepare_flags. + * \param index_buffer The index buffer to validate. The parameter may be NULL. + * \param cs_dwords The number of dwords to reserve in CS. + * \param aos_offset The offset passed to emit_aos. + * \param index_bias The index bias to emit. + * \return TRUE if rendering should be skipped + */ +static boolean r300_prepare_for_rendering(struct r300_context *r300, + enum r300_prepare_flags flags, + struct pipe_resource *index_buffer, + unsigned cs_dwords, + int aos_offset, + int index_bias) +{ + if (r300_reserve_cs_dwords(r300, flags, cs_dwords)) + flags |= PREP_FIRST_DRAW; + + return r300_emit_states(r300, flags, index_buffer, aos_offset, index_bias); } static boolean immd_is_good_idea(struct r300_context *r300, @@ -135,13 +301,20 @@ static boolean immd_is_good_idea(struct r300_context *r300, unsigned vertex_element_count = r300->velems->count; unsigned i, vbi; - if (count > 10 || DBG_ON(r300, DBG_NO_IMMD)) { + if (DBG_ON(r300, DBG_NO_IMMD)) { + return FALSE; + } + + if (r300->draw) { + return FALSE; + } + + if (count * r300->velems->vertex_size_dwords > IMMD_DWORDS) { return FALSE; } /* We shouldn't map buffers referenced by CS, busy buffers, * and ones placed in VRAM. */ - /* XXX Check for VRAM buffers. */ for (i = 0; i < vertex_element_count; i++) { velem = &r300->velems->velem[i]; vbi = velem->vertex_buffer_index; @@ -149,6 +322,10 @@ static boolean immd_is_good_idea(struct r300_context *r300, if (!checked[vbi]) { vbuf = &r300->vertex_buffer[vbi]; + if (!(r300_buffer(vbuf->buffer)->domain & R300_DOMAIN_GTT)) { + return FALSE; + } + if (r300_buffer_is_referenced(&r300->context, vbuf->buffer, R300_REF_CS | R300_REF_HW)) { @@ -162,85 +339,61 @@ static boolean immd_is_good_idea(struct r300_context *r300, } /***************************************************************************** - * The emission of draw packets for r500. Older GPUs may use these functions * - * after resolving fallback issues (e.g. stencil ref two-sided). * + * The HWTCL draw functions. * ****************************************************************************/ -static boolean r500_emit_index_offset(struct r300_context *r300, int indexBias) -{ - CS_LOCALS(r300); - - if (r300->screen->caps.is_r500 && - r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0)) { - BEGIN_CS(2); - OUT_CS_REG(R500_VAP_INDEX_OFFSET, - (indexBias & 0xFFFFFF) | (indexBias < 0 ? 1<<24 : 0)); - END_CS; - } else { - if (indexBias) - return FALSE; /* Can't do anything :( */ - } - - return TRUE; -} - -void r500_emit_draw_arrays_immediate(struct r300_context *r300, - unsigned mode, - unsigned start, - unsigned count) +static void r300_emit_draw_arrays_immediate(struct r300_context *r300, + unsigned mode, + unsigned start, + unsigned count) { struct pipe_vertex_element* velem; struct pipe_vertex_buffer* vbuf; unsigned vertex_element_count = r300->velems->count; - unsigned i, v, vbi, dw, elem_offset, dwords; + unsigned i, v, vbi; /* Size of the vertex, in dwords. */ - unsigned vertex_size = 0; + unsigned vertex_size = r300->velems->vertex_size_dwords; - /* Offsets of the attribute, in dwords, from the start of the vertex. */ - unsigned offset[PIPE_MAX_ATTRIBS]; + /* The number of dwords for this draw operation. */ + unsigned dwords = 9 + count * vertex_size; /* Size of the vertex element, in dwords. */ unsigned size[PIPE_MAX_ATTRIBS]; /* Stride to the same attrib in the next vertex in the vertex buffer, * in dwords. */ - unsigned stride[PIPE_MAX_ATTRIBS] = {0}; + unsigned stride[PIPE_MAX_ATTRIBS]; /* Mapped vertex buffers. */ - uint32_t* map[PIPE_MAX_ATTRIBS] = {0}; - struct pipe_transfer* transfer[PIPE_MAX_ATTRIBS] = {NULL}; + uint32_t* map[PIPE_MAX_ATTRIBS]; + uint32_t* mapelem[PIPE_MAX_ATTRIBS]; + struct pipe_transfer* transfer[PIPE_MAX_ATTRIBS] = {0}; CS_LOCALS(r300); + if (!r300_prepare_for_rendering(r300, PREP_FIRST_DRAW, NULL, dwords, 0, 0)) + return; + /* Calculate the vertex size, offsets, strides etc. and map the buffers. */ for (i = 0; i < vertex_element_count; i++) { velem = &r300->velems->velem[i]; - offset[i] = velem->src_offset / 4; - size[i] = util_format_get_blocksize(velem->src_format) / 4; - vertex_size += size[i]; + size[i] = r300->velems->hw_format_size[i] / 4; vbi = velem->vertex_buffer_index; + vbuf = &r300->vertex_buffer[vbi]; + stride[i] = vbuf->stride / 4; /* Map the buffer. */ - if (!map[vbi]) { - vbuf = &r300->vertex_buffer[vbi]; + if (!transfer[vbi]) { map[vbi] = (uint32_t*)pipe_buffer_map(&r300->context, vbuf->buffer, PIPE_TRANSFER_READ, &transfer[vbi]); - map[vbi] += vbuf->buffer_offset / 4; - stride[vbi] = vbuf->stride / 4; + map[vbi] += (vbuf->buffer_offset / 4) + stride[i] * start; } + mapelem[i] = map[vbi] + (velem->src_offset / 4); } - dwords = 9 + count * vertex_size; - - r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + 2 + dwords); - r300_emit_buffer_validate(r300, FALSE, NULL); - r300_emit_dirty_state(r300); - - r500_emit_index_offset(r300, 0); - BEGIN_CS(dwords); OUT_CS_REG(R300_GA_COLOR_CONTROL, r300_provoking_vertex_fixes(r300, mode)); @@ -255,13 +408,7 @@ void r500_emit_draw_arrays_immediate(struct r300_context *r300, /* Emit vertices. */ for (v = 0; v < count; v++) { for (i = 0; i < vertex_element_count; i++) { - velem = &r300->velems->velem[i]; - vbi = velem->vertex_buffer_index; - elem_offset = offset[i] + stride[vbi] * (v + start); - - for (dw = 0; dw < size[i]; dw++) { - OUT_CS(map[vbi][elem_offset + dw]); - } + OUT_CS_TABLE(&mapelem[i][stride[i] * v], size[i]); } } END_CS; @@ -270,17 +417,17 @@ void r500_emit_draw_arrays_immediate(struct r300_context *r300, for (i = 0; i < vertex_element_count; i++) { vbi = r300->velems->velem[i].vertex_buffer_index; - if (map[vbi]) { + if (transfer[vbi]) { vbuf = &r300->vertex_buffer[vbi]; pipe_buffer_unmap(&r300->context, vbuf->buffer, transfer[vbi]); - map[vbi] = NULL; + transfer[vbi] = NULL; } } } -void r500_emit_draw_arrays(struct r300_context *r300, - unsigned mode, - unsigned count) +static void r300_emit_draw_arrays(struct r300_context *r300, + unsigned mode, + unsigned count) { boolean alt_num_verts = count > 65535; CS_LOCALS(r300); @@ -291,8 +438,6 @@ void r500_emit_draw_arrays(struct r300_context *r300, return; } - r500_emit_index_offset(r300, 0); - BEGIN_CS(7 + (alt_num_verts ? 2 : 0)); if (alt_num_verts) { OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count); @@ -309,15 +454,14 @@ void r500_emit_draw_arrays(struct r300_context *r300, END_CS; } -void r500_emit_draw_elements(struct r300_context *r300, - struct pipe_resource* indexBuffer, - unsigned indexSize, - int indexBias, - unsigned minIndex, - unsigned maxIndex, - unsigned mode, - unsigned start, - unsigned count) +static void r300_emit_draw_elements(struct r300_context *r300, + struct pipe_resource* indexBuffer, + unsigned indexSize, + unsigned minIndex, + unsigned maxIndex, + unsigned mode, + unsigned start, + unsigned count) { uint32_t count_dwords; uint32_t offset_dwords = indexSize * start / sizeof(uint32_t); @@ -335,12 +479,6 @@ void r500_emit_draw_elements(struct r300_context *r300, DBG(r300, DBG_DRAW, "r300: Indexbuf of %u indices, min %u max %u\n", count, minIndex, maxIndex); - if (!r500_emit_index_offset(r300, indexBias)) { - fprintf(stderr, "r300: Got a non-zero index bias, " - "refusing to render.\n"); - return; - } - BEGIN_CS(13 + (alt_num_verts ? 2 : 0)); if (alt_num_verts) { OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count); @@ -374,189 +512,31 @@ void r500_emit_draw_elements(struct r300_context *r300, (0 << R300_INDX_BUFFER_SKIP_SHIFT)); OUT_CS(offset_dwords << 2); OUT_CS_BUF_RELOC(indexBuffer, count_dwords, - RADEON_GEM_DOMAIN_GTT, 0, 0); - - END_CS; -} - -/***************************************************************************** - * The emission of draw packets for r300 which take care of the two-sided * - * stencil ref fallback and call r500's functions. * - ****************************************************************************/ - -/* Set drawing for front faces. */ -static void r300_begin_stencil_ref_fallback(struct r300_context *r300) -{ - struct r300_rs_state *rs = (struct r300_rs_state*)r300->rs_state.state; - CS_LOCALS(r300); + r300_buffer(indexBuffer)->domain, 0); - BEGIN_CS(2); - OUT_CS_REG(R300_SU_CULL_MODE, rs->cull_mode | R300_CULL_BACK); - END_CS; -} - -/* Set drawing for back faces. */ -static void r300_switch_stencil_ref_side(struct r300_context *r300) -{ - struct r300_rs_state *rs = (struct r300_rs_state*)r300->rs_state.state; - struct r300_dsa_state *dsa = (struct r300_dsa_state*)r300->dsa_state.state; - CS_LOCALS(r300); - - BEGIN_CS(4); - OUT_CS_REG(R300_SU_CULL_MODE, rs->cull_mode | R300_CULL_FRONT); - OUT_CS_REG(R300_ZB_STENCILREFMASK, - dsa->stencil_ref_bf | r300->stencil_ref.ref_value[1]); - END_CS; -} - -/* Restore the original state. */ -static void r300_end_stencil_ref_fallback(struct r300_context *r300) -{ - struct r300_rs_state *rs = (struct r300_rs_state*)r300->rs_state.state; - struct r300_dsa_state *dsa = (struct r300_dsa_state*)r300->dsa_state.state; - CS_LOCALS(r300); - - BEGIN_CS(4); - OUT_CS_REG(R300_SU_CULL_MODE, rs->cull_mode); - OUT_CS_REG(R300_ZB_STENCILREFMASK, - dsa->stencil_ref_mask | r300->stencil_ref.ref_value[0]); END_CS; } -void r300_emit_draw_arrays_immediate(struct r300_context *r300, +/* This is the fast-path drawing & emission for HW TCL. */ +static void r300_draw_range_elements(struct pipe_context* pipe, + struct pipe_resource* indexBuffer, + unsigned indexSize, + int indexBias, + unsigned minIndex, + unsigned maxIndex, unsigned mode, unsigned start, unsigned count) { - if (!r300->stencil_ref_bf_fallback) { - r500_emit_draw_arrays_immediate(r300, mode, start, count); - } else { - r300_begin_stencil_ref_fallback(r300); - r500_emit_draw_arrays_immediate(r300, mode, start, count); - r300_switch_stencil_ref_side(r300); - r500_emit_draw_arrays_immediate(r300, mode, start, count); - r300_end_stencil_ref_fallback(r300); - } -} - -void r300_emit_draw_arrays(struct r300_context *r300, - unsigned mode, - unsigned count) -{ - if (!r300->stencil_ref_bf_fallback) { - r500_emit_draw_arrays(r300, mode, count); - } else { - r300_begin_stencil_ref_fallback(r300); - r500_emit_draw_arrays(r300, mode, count); - r300_switch_stencil_ref_side(r300); - r500_emit_draw_arrays(r300, mode, count); - r300_end_stencil_ref_fallback(r300); - } -} - -void r300_emit_draw_elements(struct r300_context *r300, - struct pipe_resource* indexBuffer, - unsigned indexSize, - int indexBias, - unsigned minIndex, - unsigned maxIndex, - unsigned mode, - unsigned start, - unsigned count) -{ - if (!r300->stencil_ref_bf_fallback) { - r500_emit_draw_elements(r300, indexBuffer, indexSize, indexBias, - minIndex, maxIndex, mode, start, count); - } else { - r300_begin_stencil_ref_fallback(r300); - r500_emit_draw_elements(r300, indexBuffer, indexSize, indexBias, - minIndex, maxIndex, mode, start, count); - r300_switch_stencil_ref_side(r300); - r500_emit_draw_elements(r300, indexBuffer, indexSize, indexBias, - minIndex, maxIndex, mode, start, count); - r300_end_stencil_ref_fallback(r300); - } -} - -static void r300_shorten_ubyte_elts(struct r300_context* r300, - struct pipe_resource** elts, - unsigned start, - unsigned count) -{ - struct pipe_context* context = &r300->context; - struct pipe_screen* screen = r300->context.screen; - struct pipe_resource* new_elts; - unsigned char *in_map; - unsigned short *out_map; - struct pipe_transfer *src_transfer, *dst_transfer; - unsigned i; - - new_elts = pipe_buffer_create(screen, - PIPE_BIND_INDEX_BUFFER, - 2 * count); - - in_map = pipe_buffer_map(context, *elts, PIPE_TRANSFER_READ, &src_transfer); - out_map = pipe_buffer_map(context, new_elts, PIPE_TRANSFER_WRITE, &dst_transfer); - - in_map += start; - - for (i = 0; i < count; i++) { - *out_map = (unsigned short)*in_map; - in_map++; - out_map++; - } - - pipe_buffer_unmap(context, *elts, src_transfer); - pipe_buffer_unmap(context, new_elts, dst_transfer); - - *elts = new_elts; -} - -static void r300_align_ushort_elts(struct r300_context *r300, - struct pipe_resource **elts, - unsigned start, unsigned count) -{ - struct pipe_context* context = &r300->context; - struct pipe_transfer *in_transfer = NULL; - struct pipe_transfer *out_transfer = NULL; - struct pipe_resource* new_elts; - unsigned short *in_map; - unsigned short *out_map; - - new_elts = pipe_buffer_create(context->screen, - PIPE_BIND_INDEX_BUFFER, - 2 * count); - - in_map = pipe_buffer_map(context, *elts, - PIPE_TRANSFER_READ, &in_transfer); - out_map = pipe_buffer_map(context, new_elts, - PIPE_TRANSFER_WRITE, &out_transfer); - - memcpy(out_map, in_map+start, 2 * count); - - pipe_buffer_unmap(context, *elts, in_transfer); - pipe_buffer_unmap(context, new_elts, out_transfer); - - *elts = new_elts; -} - -/* This is the fast-path drawing & emission for HW TCL. */ -void r300_draw_range_elements(struct pipe_context* pipe, - struct pipe_resource* indexBuffer, - unsigned indexSize, - int indexBias, - unsigned minIndex, - unsigned maxIndex, - unsigned mode, - unsigned start, - unsigned count) -{ struct r300_context* r300 = r300_context(pipe); struct pipe_resource* orgIndexBuffer = indexBuffer; boolean alt_num_verts = r300->screen->caps.is_r500 && count > 65536 && r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0); unsigned short_count; + int buffer_offset = 0, index_offset = 0; /* for index bias emulation */ + boolean translate = FALSE; + unsigned new_offset; if (r300->skip_rendering) { return; @@ -566,76 +546,78 @@ void r300_draw_range_elements(struct pipe_context* pipe, return; } - if (indexSize == 1) { - r300_shorten_ubyte_elts(r300, &indexBuffer, start, count); - indexSize = 2; - start = 0; - } else if (indexSize == 2 && start % 2 != 0) { - r300_align_ushort_elts(r300, &indexBuffer, start, count); - start = 0; + /* Index buffer range checking. */ + if ((start + count) * indexSize > indexBuffer->width0) { + fprintf(stderr, "r300: Invalid index buffer range. Skipping rendering.\n"); + return; } + /* Set up fallback for incompatible vertex layout if needed. */ + if (r300->incompatible_vb_layout || r300->velems->incompatible_layout) { + r300_begin_vertex_translate(r300); + translate = TRUE; + } + + if (indexBias && !r500_index_bias_supported(r300)) { + r300_split_index_bias(r300, indexBias, &buffer_offset, &index_offset); + } + + r300_translate_index_buffer(r300, &indexBuffer, &indexSize, index_offset, + &start, count); + r300_update_derived_state(r300); + r300_upload_index_buffer(r300, &indexBuffer, indexSize, start, count, &new_offset); - r300_upload_index_buffer(r300, &indexBuffer, indexSize, start, count); + start = new_offset; - /* 128 dwords for emit_aos and emit_draw_elements */ - r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + 128); - r300_emit_buffer_validate(r300, TRUE, indexBuffer); - r300_emit_dirty_state(r300); - r300_emit_aos(r300, 0, TRUE); + /* 15 dwords for emit_draw_elements. Give up if the function fails. */ + if (!r300_prepare_for_rendering(r300, + PREP_FIRST_DRAW | PREP_VALIDATE_VBOS | PREP_EMIT_AOS | + PREP_INDEXED, indexBuffer, 15, buffer_offset, indexBias)) + goto done; - u_upload_flush(r300->upload_vb); - u_upload_flush(r300->upload_ib); if (alt_num_verts || count <= 65535) { - r300->emit_draw_elements(r300, indexBuffer, indexSize, indexBias, - minIndex, maxIndex, mode, start, count); + r300_emit_draw_elements(r300, indexBuffer, indexSize, + minIndex, maxIndex, mode, start, count); } else { do { short_count = MIN2(count, 65534); - r300->emit_draw_elements(r300, indexBuffer, indexSize, indexBias, + r300_emit_draw_elements(r300, indexBuffer, indexSize, minIndex, maxIndex, mode, start, short_count); start += short_count; count -= short_count; - /* 16 spare dwords are enough for emit_draw_elements. - * Also reserve some space for emit_query_end. */ - if (count && r300_reserve_cs_space(r300, 74)) { - r300_emit_buffer_validate(r300, TRUE, indexBuffer); - r300_emit_dirty_state(r300); - r300_emit_aos(r300, 0, TRUE); + /* 15 dwords for emit_draw_elements */ + if (count) { + if (!r300_prepare_for_rendering(r300, + PREP_VALIDATE_VBOS | PREP_EMIT_AOS | PREP_INDEXED, + indexBuffer, 15, buffer_offset, indexBias)) + goto done; } } while (count); } +done: if (indexBuffer != orgIndexBuffer) { pipe_resource_reference( &indexBuffer, NULL ); } -} - -/* Simple helpers for context setup. Should probably be moved to util. */ -void r300_draw_elements(struct pipe_context* pipe, - struct pipe_resource* indexBuffer, - unsigned indexSize, int indexBias, unsigned mode, - unsigned start, unsigned count) -{ - struct r300_context *r300 = r300_context(pipe); - pipe->draw_range_elements(pipe, indexBuffer, indexSize, indexBias, - 0, r300->vertex_buffer_max_index, - mode, start, count); + if (translate) { + r300_end_vertex_translate(r300); + } } -void r300_draw_arrays(struct pipe_context* pipe, unsigned mode, - unsigned start, unsigned count) +static void r300_draw_arrays(struct pipe_context* pipe, unsigned mode, + unsigned start, unsigned count) { struct r300_context* r300 = r300_context(pipe); boolean alt_num_verts = r300->screen->caps.is_r500 && count > 65536 && r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0); unsigned short_count; + boolean translate = FALSE; if (r300->skip_rendering) { return; @@ -645,135 +627,149 @@ void r300_draw_arrays(struct pipe_context* pipe, unsigned mode, return; } + /* Set up fallback for incompatible vertex layout if needed. */ + if (r300->incompatible_vb_layout || r300->velems->incompatible_layout) { + r300_begin_vertex_translate(r300); + translate = TRUE; + } + r300_update_derived_state(r300); if (immd_is_good_idea(r300, count)) { - r300->emit_draw_arrays_immediate(r300, mode, start, count); + r300_emit_draw_arrays_immediate(r300, mode, start, count); } else { - /* Make sure there are at least 128 spare dwords in the command buffer. - * (most of it being consumed by emit_aos) */ - r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + 128); - r300_emit_buffer_validate(r300, TRUE, NULL); - r300_emit_dirty_state(r300); + /* 9 spare dwords for emit_draw_arrays. Give up if the function fails. */ + if (!r300_prepare_for_rendering(r300, + PREP_FIRST_DRAW | PREP_VALIDATE_VBOS | PREP_EMIT_AOS, + NULL, 9, start, 0)) + goto done; if (alt_num_verts || count <= 65535) { - r300_emit_aos(r300, start, FALSE); - r300->emit_draw_arrays(r300, mode, count); + r300_emit_draw_arrays(r300, mode, count); } else { do { short_count = MIN2(count, 65535); - r300_emit_aos(r300, start, FALSE); - r300->emit_draw_arrays(r300, mode, short_count); + r300_emit_draw_arrays(r300, mode, short_count); start += short_count; count -= short_count; - /* Again, we emit both AOS and draw_arrays so there should be - * at least 128 spare dwords. - * Also reserve some space for emit_query_end. */ - if (count && r300_reserve_cs_space(r300, 186)) { - r300_emit_buffer_validate(r300, TRUE, NULL); - r300_emit_dirty_state(r300); + /* 9 spare dwords for emit_draw_arrays. Give up if the function fails. */ + if (count) { + if (!r300_prepare_for_rendering(r300, + PREP_VALIDATE_VBOS | PREP_EMIT_AOS, NULL, 9, + start, 0)) + goto done; } } while (count); } - u_upload_flush(r300->upload_vb); } -} -/**************************************************************************** - * The rest of this file is for SW TCL rendering only. Please be polite and * - * keep these functions separated so that they are easier to locate. ~C. * - ***************************************************************************/ +done: + if (translate) { + r300_end_vertex_translate(r300); + } +} -/* SW TCL arrays, using Draw. */ -void r300_swtcl_draw_arrays(struct pipe_context* pipe, - unsigned mode, - unsigned start, - unsigned count) +static void r300_draw_vbo(struct pipe_context* pipe, + const struct pipe_draw_info *info) { struct r300_context* r300 = r300_context(pipe); - struct pipe_transfer *vb_transfer[PIPE_MAX_ATTRIBS]; - int i; - - if (r300->skip_rendering) { - return; - } - - if (!u_trim_pipe_prim(mode, &count)) { - return; - } - for (i = 0; i < r300->vertex_buffer_count; i++) { - void* buf = pipe_buffer_map(pipe, - r300->vertex_buffer[i].buffer, - PIPE_TRANSFER_READ, - &vb_transfer[i]); - draw_set_mapped_vertex_buffer(r300->draw, i, buf); - } + if (!r300->velems->count || !r300->vertex_buffer_count) + return; - draw_set_mapped_element_buffer(r300->draw, 0, 0, NULL); + if (info->indexed && r300->index_buffer.buffer) { + unsigned offset; - draw_arrays(r300->draw, mode, start, count); + assert(r300->index_buffer.offset % r300->index_buffer.index_size == 0); + offset = r300->index_buffer.offset / r300->index_buffer.index_size; - for (i = 0; i < r300->vertex_buffer_count; i++) { - pipe_buffer_unmap(pipe, r300->vertex_buffer[i].buffer, - vb_transfer[i]); - draw_set_mapped_vertex_buffer(r300->draw, i, NULL); + r300_draw_range_elements(pipe, + r300->index_buffer.buffer, + r300->index_buffer.index_size, + info->index_bias, + info->min_index, + info->max_index, + info->mode, + info->start + offset, + info->count); + } + else { + r300_draw_arrays(pipe, + info->mode, + info->start, + info->count); } } +/**************************************************************************** + * The rest of this file is for SW TCL rendering only. Please be polite and * + * keep these functions separated so that they are easier to locate. ~C. * + ***************************************************************************/ + /* SW TCL elements, using Draw. */ -void r300_swtcl_draw_range_elements(struct pipe_context* pipe, - struct pipe_resource* indexBuffer, - unsigned indexSize, - int indexBias, - unsigned minIndex, - unsigned maxIndex, - unsigned mode, - unsigned start, - unsigned count) +static void r300_swtcl_draw_vbo(struct pipe_context* pipe, + const struct pipe_draw_info *info) { struct r300_context* r300 = r300_context(pipe); struct pipe_transfer *vb_transfer[PIPE_MAX_ATTRIBS]; - struct pipe_transfer *ib_transfer; + struct pipe_transfer *ib_transfer = NULL; + unsigned count = info->count; int i; - void* indices; + void *indices = NULL; + boolean indexed = info->indexed && r300->index_buffer.buffer; if (r300->skip_rendering) { return; } - if (!u_trim_pipe_prim(mode, &count)) { + if (!u_trim_pipe_prim(info->mode, &count)) { return; } + r300_update_derived_state(r300); + + r300_reserve_cs_dwords(r300, + PREP_FIRST_DRAW | PREP_EMIT_AOS_SWTCL | + (indexed ? PREP_INDEXED : 0), + indexed ? 256 : 6); + for (i = 0; i < r300->vertex_buffer_count; i++) { - void* buf = pipe_buffer_map(pipe, - r300->vertex_buffer[i].buffer, - PIPE_TRANSFER_READ, - &vb_transfer[i]); - draw_set_mapped_vertex_buffer(r300->draw, i, buf); + if (r300->vertex_buffer[i].buffer) { + void *buf = pipe_buffer_map(pipe, + r300->vertex_buffer[i].buffer, + PIPE_TRANSFER_READ, + &vb_transfer[i]); + draw_set_mapped_vertex_buffer(r300->draw, i, buf); + } + } + + if (indexed) { + indices = pipe_buffer_map(pipe, r300->index_buffer.buffer, + PIPE_TRANSFER_READ, &ib_transfer); } - indices = pipe_buffer_map(pipe, indexBuffer, - PIPE_TRANSFER_READ, &ib_transfer); - draw_set_mapped_element_buffer_range(r300->draw, indexSize, indexBias, - minIndex, maxIndex, indices); + draw_set_mapped_index_buffer(r300->draw, indices); - draw_arrays(r300->draw, mode, start, count); + r300->draw_vbo_locked = TRUE; + r300->draw_first_emitted = FALSE; + draw_vbo(r300->draw, info); + draw_flush(r300->draw); + r300->draw_vbo_locked = FALSE; for (i = 0; i < r300->vertex_buffer_count; i++) { - pipe_buffer_unmap(pipe, r300->vertex_buffer[i].buffer, - vb_transfer[i]); - draw_set_mapped_vertex_buffer(r300->draw, i, NULL); + if (r300->vertex_buffer[i].buffer) { + pipe_buffer_unmap(pipe, r300->vertex_buffer[i].buffer, + vb_transfer[i]); + draw_set_mapped_vertex_buffer(r300->draw, i, NULL); + } } - pipe_buffer_unmap(pipe, indexBuffer, - ib_transfer); - draw_set_mapped_element_buffer_range(r300->draw, 0, 0, - start, start + count - 1, - NULL); + if (indexed) { + pipe_buffer_unmap(pipe, r300->index_buffer.buffer, ib_transfer); + draw_set_mapped_index_buffer(r300->draw, NULL); + } } /* Object for rendering using Draw. */ @@ -790,13 +786,10 @@ struct r300_render { unsigned hwprim; /* VBO */ - struct pipe_resource* vbo; - size_t vbo_size; - size_t vbo_offset; size_t vbo_max_used; void * vbo_ptr; - struct pipe_transfer *vbo_transfer; + struct pipe_transfer *vbo_transfer; }; static INLINE struct r300_render* @@ -811,8 +804,6 @@ r300_render_get_vertex_info(struct vbuf_render* render) struct r300_render* r300render = r300_render(render); struct r300_context* r300 = r300render->r300; - r300_update_derived_state(r300); - return &r300->vertex_info; } @@ -825,33 +816,40 @@ static boolean r300_render_allocate_vertices(struct vbuf_render* render, struct pipe_screen* screen = r300->context.screen; size_t size = (size_t)vertex_size * (size_t)count; - if (size + r300render->vbo_offset > r300render->vbo_size) + DBG(r300, DBG_DRAW, "r300: render_allocate_vertices (size: %d)\n", size); + + if (size + r300->draw_vbo_offset > r300->draw_vbo_size) { - pipe_resource_reference(&r300->vbo, NULL); - r300render->vbo = pipe_buffer_create(screen, - PIPE_BIND_VERTEX_BUFFER, - R300_MAX_DRAW_VBO_SIZE); - r300render->vbo_offset = 0; - r300render->vbo_size = R300_MAX_DRAW_VBO_SIZE; + pipe_resource_reference(&r300->vbo, NULL); + r300->vbo = pipe_buffer_create(screen, + PIPE_BIND_VERTEX_BUFFER, + R300_MAX_DRAW_VBO_SIZE); + r300->draw_vbo_offset = 0; + r300->draw_vbo_size = R300_MAX_DRAW_VBO_SIZE; } r300render->vertex_size = vertex_size; - r300->vbo = r300render->vbo; - r300->vbo_offset = r300render->vbo_offset; - return (r300render->vbo) ? TRUE : FALSE; + return (r300->vbo) ? TRUE : FALSE; } static void* r300_render_map_vertices(struct vbuf_render* render) { struct r300_render* r300render = r300_render(render); + struct r300_context* r300 = r300render->r300; + + assert(!r300render->vbo_transfer); + + DBG(r300, DBG_DRAW, "r300: render_map_vertices\n"); r300render->vbo_ptr = pipe_buffer_map(&r300render->r300->context, - r300render->vbo, + r300->vbo, PIPE_TRANSFER_WRITE, &r300render->vbo_transfer); - return ((uint8_t*)r300render->vbo_ptr + r300render->vbo_offset); + assert(r300render->vbo_ptr); + + return ((uint8_t*)r300render->vbo_ptr + r300->draw_vbo_offset); } static void r300_render_unmap_vertices(struct vbuf_render* render, @@ -860,21 +858,27 @@ static void r300_render_unmap_vertices(struct vbuf_render* render, { struct r300_render* r300render = r300_render(render); struct pipe_context* context = &r300render->r300->context; - CS_LOCALS(r300render->r300); - BEGIN_CS(2); - OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, max); - END_CS; + struct r300_context* r300 = r300render->r300; + + assert(r300render->vbo_transfer); + + DBG(r300, DBG_DRAW, "r300: render_unmap_vertices\n"); r300render->vbo_max_used = MAX2(r300render->vbo_max_used, r300render->vertex_size * (max + 1)); - pipe_buffer_unmap(context, r300render->vbo, r300render->vbo_transfer); + pipe_buffer_unmap(context, r300->vbo, r300render->vbo_transfer); + + r300render->vbo_transfer = NULL; } static void r300_render_release_vertices(struct vbuf_render* render) { struct r300_render* r300render = r300_render(render); + struct r300_context* r300 = r300render->r300; - r300render->vbo_offset += r300render->vbo_max_used; + DBG(r300, DBG_DRAW, "r300: render_release_vertices\n"); + + r300->draw_vbo_offset += r300render->vbo_max_used; r300render->vbo_max_used = 0; } @@ -889,92 +893,131 @@ static boolean r300_render_set_primitive(struct vbuf_render* render, return TRUE; } -static void r500_render_draw_arrays(struct vbuf_render* render, +static void r300_render_draw_arrays(struct vbuf_render* render, unsigned start, unsigned count) { struct r300_render* r300render = r300_render(render); struct r300_context* r300 = r300render->r300; + uint8_t* ptr; + unsigned i; + unsigned dwords = 6; CS_LOCALS(r300); + (void) i; (void) ptr; - r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + 2); - r300_emit_buffer_validate(r300, FALSE, NULL); - r300_emit_dirty_state(r300); + DBG(r300, DBG_DRAW, "r300: render_draw_arrays (count: %d)\n", count); - DBG(r300, DBG_DRAW, "r300: Doing vbuf render, count %d\n", count); + if (r300->draw_first_emitted) { + if (!r300_prepare_for_rendering(r300, + PREP_FIRST_DRAW | PREP_EMIT_AOS_SWTCL, + NULL, 6, 0, 0)) + return; + } else { + if (!r300_emit_states(r300, + PREP_FIRST_DRAW | PREP_EMIT_AOS_SWTCL, + NULL, 0, 0)) + return; + } - r500_emit_index_offset(r300, 0); + /* Uncomment to dump all VBOs rendered through this interface. + * Slow and noisy! + ptr = pipe_buffer_map(&r300render->r300->context, + r300render->vbo, PIPE_TRANSFER_READ, + &r300render->vbo_transfer); - BEGIN_CS(2); + for (i = 0; i < count; i++) { + printf("r300: Vertex %d\n", i); + draw_dump_emitted_vertex(&r300->vertex_info, ptr); + ptr += r300->vertex_info.size * 4; + printf("\n"); + } + + pipe_buffer_unmap(&r300render->r300->context, r300render->vbo, + r300render->vbo_transfer); + */ + + BEGIN_CS(dwords); + OUT_CS_REG(R300_GA_COLOR_CONTROL, + r300_provoking_vertex_fixes(r300, r300render->prim)); + OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count - 1); OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0); OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) | r300render->hwprim); END_CS; + + r300->draw_first_emitted = TRUE; } -static void r500_render_draw(struct vbuf_render* render, - const ushort* indices, - uint count) +static void r300_render_draw_elements(struct vbuf_render* render, + const ushort* indices, + uint count) { struct r300_render* r300render = r300_render(render); struct r300_context* r300 = r300render->r300; int i; - unsigned dwords = 2 + (count+1)/2; + unsigned end_cs_dwords; + unsigned max_index = (r300->draw_vbo_size - r300->draw_vbo_offset) / + (r300render->r300->vertex_info.size * 4) - 1; + unsigned short_count; + unsigned free_dwords; CS_LOCALS(r300); + DBG(r300, DBG_DRAW, "r300: render_draw_elements (count: %d)\n", count); - r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + dwords); - r300_emit_buffer_validate(r300, FALSE, NULL); - r300_emit_dirty_state(r300); + if (r300->draw_first_emitted) { + if (!r300_prepare_for_rendering(r300, + PREP_FIRST_DRAW | PREP_EMIT_AOS_SWTCL | PREP_INDEXED, + NULL, 256, 0, 0)) + return; + } else { + if (!r300_emit_states(r300, + PREP_FIRST_DRAW | PREP_EMIT_AOS_SWTCL | PREP_INDEXED, + NULL, 0, 0)) + return; + } - r500_emit_index_offset(r300, 0); + /* Below we manage the CS space manually because there may be more + * indices than it can fit in CS. */ - BEGIN_CS(dwords); - OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, (count+1)/2); - OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) | - r300render->hwprim); - for (i = 0; i < count-1; i += 2) { - OUT_CS(indices[i+1] << 16 | indices[i]); - } - if (count % 2) { - OUT_CS(indices[count-1]); - } - END_CS; -} + end_cs_dwords = r300_get_num_cs_end_dwords(r300); -static void r300_render_draw_arrays(struct vbuf_render* render, - unsigned start, - unsigned count) -{ - struct r300_context* r300 = r300_render(render)->r300; + while (count) { + free_dwords = r300->cs->ndw - r300->cs->cdw; - if (!r300->stencil_ref_bf_fallback) { - r500_render_draw_arrays(render, start, count); - } else { - r300_begin_stencil_ref_fallback(r300); - r500_render_draw_arrays(render, start, count); - r300_switch_stencil_ref_side(r300); - r500_render_draw_arrays(render, start, count); - r300_end_stencil_ref_fallback(r300); - } -} + short_count = MIN2(count, (free_dwords - end_cs_dwords - 6) * 2); -static void r300_render_draw(struct vbuf_render* render, - const ushort* indices, - uint count) -{ - struct r300_context* r300 = r300_render(render)->r300; + BEGIN_CS(6 + (short_count+1)/2); + OUT_CS_REG(R300_GA_COLOR_CONTROL, + r300_provoking_vertex_fixes(r300, r300render->prim)); + OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, max_index); + OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, (short_count+1)/2); + OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (short_count << 16) | + r300render->hwprim); + for (i = 0; i < short_count-1; i += 2) { + OUT_CS(indices[i+1] << 16 | indices[i]); + } + if (short_count % 2) { + OUT_CS(indices[short_count-1]); + } + END_CS; - if (!r300->stencil_ref_bf_fallback) { - r500_render_draw(render, indices, count); - } else { - r300_begin_stencil_ref_fallback(r300); - r500_render_draw(render, indices, count); - r300_switch_stencil_ref_side(r300); - r500_render_draw(render, indices, count); - r300_end_stencil_ref_fallback(r300); + /* OK now subtract the emitted indices and see if we need to emit + * another draw packet. */ + indices += short_count; + count -= short_count; + + if (count) { + if (!r300_prepare_for_rendering(r300, + PREP_EMIT_AOS_SWTCL | PREP_INDEXED, + NULL, 256, 0, 0)) + return; + + end_cs_dwords = r300_get_num_cs_end_dwords(r300); + } } + + r300->draw_first_emitted = TRUE; } static void r300_render_destroy(struct vbuf_render* render) @@ -997,20 +1040,11 @@ static struct vbuf_render* r300_render_create(struct r300_context* r300) r300render->base.map_vertices = r300_render_map_vertices; r300render->base.unmap_vertices = r300_render_unmap_vertices; r300render->base.set_primitive = r300_render_set_primitive; - if (r300->screen->caps.is_r500) { - r300render->base.draw = r500_render_draw; - r300render->base.draw_arrays = r500_render_draw_arrays; - } else { - r300render->base.draw = r300_render_draw; - r300render->base.draw_arrays = r300_render_draw_arrays; - } + r300render->base.draw_elements = r300_render_draw_elements; + r300render->base.draw_arrays = r300_render_draw_arrays; r300render->base.release_vertices = r300_render_release_vertices; r300render->base.destroy = r300_render_destroy; - r300render->vbo = NULL; - r300render->vbo_size = 0; - r300render->vbo_offset = 0; - return &r300render->base; } @@ -1036,3 +1070,154 @@ struct draw_stage* r300_draw_stage(struct r300_context* r300) return stage; } + +void r300_draw_flush_vbuf(struct r300_context *r300) +{ + pipe_resource_reference(&r300->vbo, NULL); + r300->draw_vbo_size = 0; +} + +/**************************************************************************** + * End of SW TCL functions * + ***************************************************************************/ + +/* This functions is used to draw a rectangle for the blitter module. + * + * If we rendered a quad, the pixels on the main diagonal + * would be computed and stored twice, which makes the clear/copy codepaths + * somewhat inefficient. Instead we use a rectangular point sprite. */ +static void r300_blitter_draw_rectangle(struct blitter_context *blitter, + unsigned x1, unsigned y1, + unsigned x2, unsigned y2, + float depth, + enum blitter_attrib_type type, + const float attrib[4]) +{ + struct r300_context *r300 = r300_context(util_blitter_get_pipe(blitter)); + unsigned last_sprite_coord_enable = r300->sprite_coord_enable; + unsigned width = x2 - x1; + unsigned height = y2 - y1; + unsigned vertex_size = + type == UTIL_BLITTER_ATTRIB_COLOR || !r300->draw ? 8 : 4; + unsigned dwords = 13 + vertex_size + + (type == UTIL_BLITTER_ATTRIB_TEXCOORD ? 7 : 0); + const float zeros[4] = {0, 0, 0, 0}; + CS_LOCALS(r300); + + if (type == UTIL_BLITTER_ATTRIB_TEXCOORD) + r300->sprite_coord_enable = 1; + + r300_update_derived_state(r300); + + /* Mark some states we don't care about as non-dirty. */ + r300->clip_state.dirty = FALSE; + r300->viewport_state.dirty = FALSE; + + if (!r300_prepare_for_rendering(r300, PREP_FIRST_DRAW, NULL, dwords, 0, 0)) + goto done; + + DBG(r300, DBG_DRAW, "r300: draw_rectangle\n"); + + BEGIN_CS(dwords); + /* Set up GA. */ + OUT_CS_REG(R300_GA_POINT_SIZE, (height * 6) | ((width * 6) << 16)); + + if (type == UTIL_BLITTER_ATTRIB_TEXCOORD) { + /* Set up the GA to generate texcoords. */ + OUT_CS_REG(R300_GB_ENABLE, R300_GB_POINT_STUFF_ENABLE | + (R300_GB_TEX_STR << R300_GB_TEX0_SOURCE_SHIFT)); + OUT_CS_REG_SEQ(R300_GA_POINT_S0, 4); + OUT_CS_32F(attrib[0]); + OUT_CS_32F(attrib[3]); + OUT_CS_32F(attrib[2]); + OUT_CS_32F(attrib[1]); + } + + /* Set up VAP controls. */ + OUT_CS_REG(R300_VAP_CLIP_CNTL, R300_CLIP_DISABLE); + OUT_CS_REG(R300_VAP_VTE_CNTL, R300_VTX_XY_FMT | R300_VTX_Z_FMT); + OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size); + OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2); + OUT_CS(1); + OUT_CS(0); + + /* Draw. */ + OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, vertex_size); + OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (1 << 16) | + R300_VAP_VF_CNTL__PRIM_POINTS); + + OUT_CS_32F(x1 + width * 0.5f); + OUT_CS_32F(y1 + height * 0.5f); + OUT_CS_32F(depth); + OUT_CS_32F(1); + + if (vertex_size == 8) { + if (!attrib) + attrib = zeros; + OUT_CS_TABLE(attrib, 4); + } + END_CS; + +done: + /* Restore the state. */ + r300->clip_state.dirty = TRUE; + r300->rs_state.dirty = TRUE; + r300->viewport_state.dirty = TRUE; + + r300->sprite_coord_enable = last_sprite_coord_enable; +} + +static void r300_resource_resolve(struct pipe_context* pipe, + struct pipe_resource* dest, + struct pipe_subresource subdest, + struct pipe_resource* src, + struct pipe_subresource subsrc) +{ + struct r300_context* r300 = r300_context(pipe); + struct r300_aa_state *aa = (struct r300_aa_state*)r300->aa_state.state; + struct pipe_surface* srcsurf = src->screen->get_tex_surface(src->screen, + src, subsrc.face, subsrc.level, 0, 0); + float color[] = {0, 0, 0, 0}; + + DBG(r300, DBG_DRAW, "r300: Resolving resource...\n"); + + /* Enable AA resolve. */ + aa->dest = r300_surface( + dest->screen->get_tex_surface(dest->screen, dest, subdest.face, + subdest.level, 0, 0)); + + aa->aaresolve_ctl = + R300_RB3D_AARESOLVE_CTL_AARESOLVE_MODE_RESOLVE | + R300_RB3D_AARESOLVE_CTL_AARESOLVE_ALPHA_AVERAGE; + r300->aa_state.size = 12; + r300->aa_state.dirty = TRUE; + + /* Resolve the surface. */ + r300->context.clear_render_target(pipe, + srcsurf, color, 0, 0, src->width0, src->height0); + + /* Disable AA resolve. */ + aa->aaresolve_ctl = 0; + r300->aa_state.size = 4; + r300->aa_state.dirty = TRUE; + + pipe_surface_reference((struct pipe_surface**)&srcsurf, NULL); + pipe_surface_reference((struct pipe_surface**)&aa->dest, NULL); +} + +void r300_init_render_functions(struct r300_context *r300) +{ + /* Set draw functions based on presence of HW TCL. */ + if (r300->screen->caps.has_tcl) { + r300->context.draw_vbo = r300_draw_vbo; + } else { + r300->context.draw_vbo = r300_swtcl_draw_vbo; + } + + r300->context.resource_resolve = r300_resource_resolve; + r300->blitter->draw_rectangle = r300_blitter_draw_rectangle; + + /* Plug in the two-sided stencil reference value fallback if needed. */ + if (!r300->screen->caps.is_r500) + r300_plug_in_stencil_ref_fallback(r300); +} diff --git a/src/gallium/drivers/r300/r300_render.h b/src/gallium/drivers/r300/r300_render.h deleted file mode 100644 index 4e78914c1ba..00000000000 --- a/src/gallium/drivers/r300/r300_render.h +++ /dev/null @@ -1,97 +0,0 @@ -/* - * Copyright 2009 Corbin Simpson <[email protected]> - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * on the rights to use, copy, modify, merge, publish, distribute, sub - * license, and/or sell copies of the Software, and to permit persons to whom - * the Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice (including the next - * paragraph) shall be included in all copies or substantial portions of the - * Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL - * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, - * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR - * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE - * USE OR OTHER DEALINGS IN THE SOFTWARE. */ - -#ifndef R300_RENDER_H -#define R300_RENDER_H - -void r500_emit_draw_arrays_immediate(struct r300_context *r300, - unsigned mode, - unsigned start, - unsigned count); - -void r500_emit_draw_arrays(struct r300_context *r300, - unsigned mode, - unsigned count); - -void r500_emit_draw_elements(struct r300_context *r300, - struct pipe_resource* indexBuffer, - unsigned indexSize, - int indexBias, - unsigned minIndex, - unsigned maxIndex, - unsigned mode, - unsigned start, - unsigned count); - -void r300_emit_draw_arrays_immediate(struct r300_context *r300, - unsigned mode, - unsigned start, - unsigned count); - -void r300_emit_draw_arrays(struct r300_context *r300, - unsigned mode, - unsigned count); - -void r300_emit_draw_elements(struct r300_context *r300, - struct pipe_resource* indexBuffer, - unsigned indexSize, - int indexBias, - unsigned minIndex, - unsigned maxIndex, - unsigned mode, - unsigned start, - unsigned count); - -void r300_draw_range_elements(struct pipe_context* pipe, - struct pipe_resource* indexBuffer, - unsigned indexSize, - int indexBias, - unsigned minIndex, - unsigned maxIndex, - unsigned mode, - unsigned start, - unsigned count); - -void r300_draw_elements(struct pipe_context* pipe, - struct pipe_resource* indexBuffer, - unsigned indexSize, int indexBias, unsigned mode, - unsigned start, unsigned count); - -void r300_draw_arrays(struct pipe_context* pipe, unsigned mode, - unsigned start, unsigned count); - -void r300_swtcl_draw_arrays(struct pipe_context* pipe, - unsigned mode, - unsigned start, - unsigned count); - -void r300_swtcl_draw_range_elements(struct pipe_context* pipe, - struct pipe_resource* indexBuffer, - unsigned indexSize, - int indexBias, - unsigned minIndex, - unsigned maxIndex, - unsigned mode, - unsigned start, - unsigned count); - -#endif /* R300_RENDER_H */ diff --git a/src/gallium/drivers/r300/r300_render_stencilref.c b/src/gallium/drivers/r300/r300_render_stencilref.c new file mode 100644 index 00000000000..1f035d64a28 --- /dev/null +++ b/src/gallium/drivers/r300/r300_render_stencilref.c @@ -0,0 +1,129 @@ +/* + * Copyright 2010 Marek Olšák <[email protected]> + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. */ + +/** + * The two-sided stencil reference value fallback for r3xx-r4xx chips. + * These chips support two-sided stencil functions but they do not support + * a two-sided reference value. + * + * The functions below split every draw call which uses the two-sided + * reference value into two draw calls -- the first one renders front faces + * and the second renders back faces with the other reference value. + */ + +#include "r300_context.h" +#include "r300_reg.h" + +struct r300_stencilref_context { + void (*draw_vbo)(struct pipe_context *pipe, + const struct pipe_draw_info *info); + + uint32_t rs_cull_mode; + uint32_t zb_stencilrefmask; + ubyte ref_value_front; +}; + +static boolean r300_stencilref_needed(struct r300_context *r300) +{ + struct r300_dsa_state *dsa = (struct r300_dsa_state*)r300->dsa_state.state; + + return dsa->two_sided_stencil_ref || + (dsa->two_sided && + r300->stencil_ref.ref_value[0] != r300->stencil_ref.ref_value[1]); +} + +/* Set drawing for front faces. */ +static void r300_stencilref_begin(struct r300_context *r300) +{ + struct r300_stencilref_context *sr = r300->stencilref_fallback; + struct r300_rs_state *rs = (struct r300_rs_state*)r300->rs_state.state; + struct r300_dsa_state *dsa = (struct r300_dsa_state*)r300->dsa_state.state; + + /* Save state. */ + sr->rs_cull_mode = rs->cb_main[rs->cull_mode_index]; + sr->zb_stencilrefmask = dsa->stencil_ref_mask; + sr->ref_value_front = r300->stencil_ref.ref_value[0]; + + /* We *cull* pixels, therefore no need to mask out the bits. */ + rs->cb_main[rs->cull_mode_index] |= R300_CULL_BACK; + + r300->rs_state.dirty = TRUE; +} + +/* Set drawing for back faces. */ +static void r300_stencilref_switch_side(struct r300_context *r300) +{ + struct r300_stencilref_context *sr = r300->stencilref_fallback; + struct r300_rs_state *rs = (struct r300_rs_state*)r300->rs_state.state; + struct r300_dsa_state *dsa = (struct r300_dsa_state*)r300->dsa_state.state; + + rs->cb_main[rs->cull_mode_index] = sr->rs_cull_mode | R300_CULL_FRONT; + dsa->stencil_ref_mask = dsa->stencil_ref_bf; + r300->stencil_ref.ref_value[0] = r300->stencil_ref.ref_value[1]; + + r300->rs_state.dirty = TRUE; + r300->dsa_state.dirty = TRUE; +} + +/* Restore the original state. */ +static void r300_stencilref_end(struct r300_context *r300) +{ + struct r300_stencilref_context *sr = r300->stencilref_fallback; + struct r300_rs_state *rs = (struct r300_rs_state*)r300->rs_state.state; + struct r300_dsa_state *dsa = (struct r300_dsa_state*)r300->dsa_state.state; + + /* Restore state. */ + rs->cb_main[rs->cull_mode_index] = sr->rs_cull_mode; + dsa->stencil_ref_mask = sr->zb_stencilrefmask; + r300->stencil_ref.ref_value[0] = sr->ref_value_front; + + r300->rs_state.dirty = TRUE; + r300->dsa_state.dirty = TRUE; +} + +static void r300_stencilref_draw_vbo(struct pipe_context *pipe, + const struct pipe_draw_info *info) +{ + struct r300_context *r300 = r300_context(pipe); + struct r300_stencilref_context *sr = r300->stencilref_fallback; + + if (!r300_stencilref_needed(r300)) { + sr->draw_vbo(pipe, info); + } else { + r300_stencilref_begin(r300); + sr->draw_vbo(pipe, info); + r300_stencilref_switch_side(r300); + sr->draw_vbo(pipe, info); + r300_stencilref_end(r300); + } +} + +void r300_plug_in_stencil_ref_fallback(struct r300_context *r300) +{ + r300->stencilref_fallback = CALLOC_STRUCT(r300_stencilref_context); + + /* Save original draw function. */ + r300->stencilref_fallback->draw_vbo = r300->context.draw_vbo; + + /* Override the draw function. */ + r300->context.draw_vbo = r300_stencilref_draw_vbo; +} diff --git a/src/gallium/drivers/r300/r300_render_translate.c b/src/gallium/drivers/r300/r300_render_translate.c new file mode 100644 index 00000000000..9247064508f --- /dev/null +++ b/src/gallium/drivers/r300/r300_render_translate.c @@ -0,0 +1,218 @@ +/* + * Copyright 2010 Marek Olšák <[email protected]> + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. */ + +/** + * The functions below translate vertex and index buffers to the layout + * compatible with the hardware, so that all vertex and index fetches are + * DWORD-aligned and all used vertex and index formats are supported. + * For indices, an optional index offset is added to each index. + */ + +#include "r300_context.h" +#include "translate/translate.h" +#include "util/u_index_modify.h" + +void r300_begin_vertex_translate(struct r300_context *r300) +{ + struct pipe_context *pipe = &r300->context; + struct translate_key key = {0}; + struct translate_element *te; + unsigned tr_elem_index[PIPE_MAX_ATTRIBS] = {0}; + struct translate *tr; + struct r300_vertex_element_state *ve = r300->velems; + boolean vb_translated[PIPE_MAX_ATTRIBS] = {0}; + void *vb_map[PIPE_MAX_ATTRIBS] = {0}, *out_map; + struct pipe_transfer *vb_transfer[PIPE_MAX_ATTRIBS] = {0}, *out_transfer; + struct pipe_resource *out_buffer; + unsigned i, num_verts; + + /* Initialize the translate key, i.e. the recipe how vertices should be + * translated. */ + for (i = 0; i < ve->count; i++) { + struct pipe_vertex_buffer *vb = + &r300->vertex_buffer[ve->velem[i].vertex_buffer_index]; + enum pipe_format output_format = ve->hw_format[i]; + unsigned output_format_size = ve->hw_format_size[i]; + + /* Check for support. */ + if (ve->velem[i].src_format == ve->hw_format[i] && + (vb->buffer_offset + ve->velem[i].src_offset) % 4 == 0 && + vb->stride % 4 == 0) { + continue; + } + + /* Workaround for translate: output floats instead of halfs. */ + switch (output_format) { + case PIPE_FORMAT_R16_FLOAT: + output_format = PIPE_FORMAT_R32_FLOAT; + output_format_size = 4; + break; + case PIPE_FORMAT_R16G16_FLOAT: + output_format = PIPE_FORMAT_R32G32_FLOAT; + output_format_size = 8; + break; + case PIPE_FORMAT_R16G16B16_FLOAT: + output_format = PIPE_FORMAT_R32G32B32_FLOAT; + output_format_size = 12; + break; + case PIPE_FORMAT_R16G16B16A16_FLOAT: + output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; + output_format_size = 16; + break; + default:; + } + + /* Add this vertex element. */ + te = &key.element[key.nr_elements]; + /*te->type; + te->instance_divisor;*/ + te->input_buffer = ve->velem[i].vertex_buffer_index; + te->input_format = ve->velem[i].src_format; + te->input_offset = vb->buffer_offset + ve->velem[i].src_offset; + te->output_format = output_format; + te->output_offset = key.output_stride; + + key.output_stride += output_format_size; + vb_translated[ve->velem[i].vertex_buffer_index] = TRUE; + tr_elem_index[i] = key.nr_elements; + key.nr_elements++; + } + + /* Get a translate object. */ + tr = translate_cache_find(r300->tran.translate_cache, &key); + + /* Map buffers we want to translate. */ + for (i = 0; i < r300->vertex_buffer_count; i++) { + if (vb_translated[i]) { + struct pipe_vertex_buffer *vb = &r300->vertex_buffer[i]; + + vb_map[i] = pipe_buffer_map(pipe, vb->buffer, + PIPE_TRANSFER_READ, &vb_transfer[i]); + + tr->set_buffer(tr, i, vb_map[i], vb->stride, vb->max_index); + } + } + + /* Create and map the output buffer. */ + num_verts = r300->vertex_buffer_max_index + 1; + + out_buffer = pipe_buffer_create(&r300->screen->screen, + PIPE_BIND_VERTEX_BUFFER, + key.output_stride * num_verts); + + out_map = pipe_buffer_map(pipe, out_buffer, PIPE_TRANSFER_WRITE, + &out_transfer); + + /* Translate. */ + tr->run(tr, 0, num_verts, 0, out_map); + + /* Unmap all buffers. */ + for (i = 0; i < r300->vertex_buffer_count; i++) { + if (vb_translated[i]) { + pipe_buffer_unmap(pipe, r300->vertex_buffer[i].buffer, + vb_transfer[i]); + } + } + + pipe_buffer_unmap(pipe, out_buffer, out_transfer); + + /* Setup the new vertex buffer in the first free slot. */ + for (i = 0; i < PIPE_MAX_ATTRIBS; i++) { + struct pipe_vertex_buffer *vb = &r300->vertex_buffer[i]; + + if (!vb->buffer) { + pipe_resource_reference(&vb->buffer, out_buffer); + vb->buffer_offset = 0; + vb->max_index = num_verts - 1; + vb->stride = key.output_stride; + r300->tran.vb_slot = i; + break; + } + } + + /* Save and replace vertex elements. */ + { + struct pipe_vertex_element new_velems[PIPE_MAX_ATTRIBS]; + + r300->tran.saved_velems = r300->velems; + + for (i = 0; i < ve->count; i++) { + if (vb_translated[ve->velem[i].vertex_buffer_index]) { + te = &key.element[tr_elem_index[i]]; + new_velems[i].instance_divisor = ve->velem[i].instance_divisor; + new_velems[i].src_format = te->output_format; + new_velems[i].src_offset = te->output_offset; + new_velems[i].vertex_buffer_index = r300->tran.vb_slot; + } else { + memcpy(&new_velems[i], &ve->velem[i], + sizeof(struct pipe_vertex_element)); + } + } + + r300->tran.new_velems = + pipe->create_vertex_elements_state(pipe, ve->count, new_velems); + pipe->bind_vertex_elements_state(pipe, r300->tran.new_velems); + } + + pipe_resource_reference(&out_buffer, NULL); +} + +void r300_end_vertex_translate(struct r300_context *r300) +{ + struct pipe_context *pipe = &r300->context; + + /* Restore vertex elements. */ + pipe->bind_vertex_elements_state(pipe, r300->tran.saved_velems); + pipe->delete_vertex_elements_state(pipe, r300->tran.new_velems); + + /* Delete the now-unused VBO. */ + pipe_resource_reference(&r300->vertex_buffer[r300->tran.vb_slot].buffer, + NULL); +} + +void r300_translate_index_buffer(struct r300_context *r300, + struct pipe_resource **index_buffer, + unsigned *index_size, unsigned index_offset, + unsigned *start, unsigned count) +{ + switch (*index_size) { + case 1: + util_shorten_ubyte_elts(&r300->context, index_buffer, index_offset, *start, count); + *index_size = 2; + *start = 0; + break; + + case 2: + if (*start % 2 != 0 || index_offset) { + util_rebuild_ushort_elts(&r300->context, index_buffer, index_offset, *start, count); + *start = 0; + } + break; + + case 4: + if (index_offset) { + util_rebuild_uint_elts(&r300->context, index_buffer, index_offset, *start, count); + *start = 0; + } + break; + } +} diff --git a/src/gallium/drivers/r300/r300_resource.c b/src/gallium/drivers/r300/r300_resource.c index 7ed47eaa605..f6f33028dc6 100644 --- a/src/gallium/drivers/r300/r300_resource.c +++ b/src/gallium/drivers/r300/r300_resource.c @@ -23,19 +23,9 @@ * Authors: Dave Airlie */ -#include <stdio.h> - -#include "util/u_inlines.h" -#include "util/u_format.h" -#include "util/u_memory.h" -#include "util/u_upload_mgr.h" -#include "util/u_math.h" - #include "r300_context.h" #include "r300_texture.h" -#include "r300_screen.h" #include "r300_screen_buffer.h" -#include "r300_winsys.h" static struct pipe_resource * r300_resource_create(struct pipe_screen *screen, diff --git a/src/gallium/drivers/r300/r300_screen.c b/src/gallium/drivers/r300/r300_screen.c index 8fc1d5aa00e..7f41ff0e2ec 100644 --- a/src/gallium/drivers/r300/r300_screen.c +++ b/src/gallium/drivers/r300/r300_screen.c @@ -22,12 +22,15 @@ * USE OR OTHER DEALINGS IN THE SOFTWARE. */ #include "util/u_format.h" +#include "util/u_format_s3tc.h" #include "util/u_memory.h" #include "r300_context.h" #include "r300_texture.h" #include "r300_screen_buffer.h" +#include "r300_state_inlines.h" #include "r300_winsys.h" +#include "r300_public.h" /* Return the identifier behind whom the brave coders responsible for this * amalgamation of code, sweat, and duct tape, routinely obscure their names. @@ -75,20 +78,19 @@ static const char* r300_get_name(struct pipe_screen* pscreen) return chip_families[r300screen->caps.family]; } -static int r300_get_param(struct pipe_screen* pscreen, int param) +static int r300_get_param(struct pipe_screen* pscreen, enum pipe_cap param) { struct r300_screen* r300screen = r300_screen(pscreen); + boolean is_r400 = r300screen->caps.is_r400; + boolean is_r500 = r300screen->caps.is_r500; + + /* XXX extended shader capabilities of r400 unimplemented */ + is_r400 = FALSE; switch (param) { - case PIPE_CAP_MAX_TEXTURE_IMAGE_UNITS: - case PIPE_CAP_MAX_COMBINED_SAMPLERS: - return r300screen->caps.num_tex_units; + /* Supported features (boolean caps). */ case PIPE_CAP_NPOT_TEXTURES: - /* XXX enable now to get GL2.1 API, - * figure out later how to emulate this */ - return 1; case PIPE_CAP_TWO_SIDED_STENCIL: - return 1; case PIPE_CAP_GLSL: /* I'll be frank. This is a lie. * @@ -105,53 +107,45 @@ static int r300_get_param(struct pipe_screen* pscreen, int param) * * ~ C. */ - return 1; - case PIPE_CAP_DUAL_SOURCE_BLEND: - return 0; case PIPE_CAP_ANISOTROPIC_FILTER: - return 1; case PIPE_CAP_POINT_SPRITE: - return 1; - case PIPE_CAP_MAX_RENDER_TARGETS: - return 4; case PIPE_CAP_OCCLUSION_QUERY: - return 1; case PIPE_CAP_TEXTURE_SHADOW_MAP: - return 1; - case PIPE_CAP_MAX_TEXTURE_2D_LEVELS: - case PIPE_CAP_MAX_TEXTURE_3D_LEVELS: - case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS: - if (r300screen->caps.is_r500) { - /* 13 == 4096 */ - return 13; - } else { - /* 12 == 2048 */ - return 12; - } case PIPE_CAP_TEXTURE_MIRROR_CLAMP: - return 1; case PIPE_CAP_TEXTURE_MIRROR_REPEAT: - return 1; - case PIPE_CAP_MAX_VERTEX_TEXTURE_UNITS: - return 0; - case PIPE_CAP_TGSI_CONT_SUPPORTED: - return 0; case PIPE_CAP_BLEND_EQUATION_SEPARATE: + case PIPE_CAP_TEXTURE_SWIZZLE: return 1; - case PIPE_CAP_SM3: - if (r300screen->caps.is_r500) { - return 1; - } else { - return 0; - } - case PIPE_CAP_MAX_CONST_BUFFERS: - return 1; - case PIPE_CAP_MAX_CONST_BUFFER_SIZE: - return 256; + + /* Unsupported features (boolean caps). */ + case PIPE_CAP_TIMER_QUERY: + case PIPE_CAP_DUAL_SOURCE_BLEND: case PIPE_CAP_INDEP_BLEND_ENABLE: - return 0; case PIPE_CAP_INDEP_BLEND_FUNC: + case PIPE_CAP_DEPTH_CLAMP: /* XXX implemented, but breaks Regnum Online */ + case PIPE_CAP_DEPTHSTENCIL_CLEAR_SEPARATE: + return 0; + + /* Texturing. */ + case PIPE_CAP_MAX_TEXTURE_IMAGE_UNITS: + case PIPE_CAP_MAX_COMBINED_SAMPLERS: + return r300screen->caps.num_tex_units; + case PIPE_CAP_MAX_VERTEX_TEXTURE_UNITS: return 0; + case PIPE_CAP_MAX_TEXTURE_2D_LEVELS: + case PIPE_CAP_MAX_TEXTURE_3D_LEVELS: + case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS: + /* 13 == 4096, 12 == 2048 */ + return is_r500 ? 13 : 12; + + /* Render targets. */ + case PIPE_CAP_MAX_RENDER_TARGETS: + return 4; + + /* General shader limits and features. */ + case PIPE_CAP_SM3: + return is_r500 ? 1 : 0; + /* Fragment coordinate conventions. */ case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT: case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER: return 1; @@ -165,7 +159,89 @@ static int r300_get_param(struct pipe_screen* pscreen, int param) } } -static float r300_get_paramf(struct pipe_screen* pscreen, int param) +static int r300_get_shader_param(struct pipe_screen *pscreen, unsigned shader, enum pipe_shader_cap param) +{ + struct r300_screen* r300screen = r300_screen(pscreen); + boolean is_r400 = r300screen->caps.is_r400; + boolean is_r500 = r300screen->caps.is_r500; + + /* XXX extended shader capabilities of r400 unimplemented */ + is_r400 = FALSE; + + switch (shader) + { + case PIPE_SHADER_FRAGMENT: + switch (param) + { + case PIPE_SHADER_CAP_MAX_INSTRUCTIONS: + return is_r500 || is_r400 ? 512 : 96; + case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS: + return is_r500 || is_r400 ? 512 : 64; + case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS: + return is_r500 || is_r400 ? 512 : 32; + case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS: + return is_r500 ? 511 : 4; + case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH: + return is_r500 ? 64 : 0; /* Actually unlimited on r500. */ + /* Fragment shader limits. */ + case PIPE_SHADER_CAP_MAX_INPUTS: + /* 2 colors + 8 texcoords are always supported + * (minus fog and wpos). + * + * R500 has the ability to turn 3rd and 4th color into + * additional texcoords but there is no two-sided color + * selection then. However the facing bit can be used instead. */ + return 10; + case PIPE_SHADER_CAP_MAX_CONSTS: + return is_r500 ? 256 : 32; + case PIPE_SHADER_CAP_MAX_CONST_BUFFERS: + return 1; + case PIPE_SHADER_CAP_MAX_TEMPS: + return is_r500 ? 128 : is_r400 ? 64 : 32; + case PIPE_SHADER_CAP_MAX_ADDRS: + return 0; + case PIPE_SHADER_CAP_MAX_PREDS: + return is_r500 ? 1 : 0; + case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED: + return 1; + } + break; + case PIPE_SHADER_VERTEX: + switch (param) + { + case PIPE_SHADER_CAP_MAX_INSTRUCTIONS: + case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS: + return is_r500 ? 1024 : 256; + case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS: + case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS: + return 0; + case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH: + return is_r500 ? 4 : 0; /* For loops; not sure about conditionals. */ + case PIPE_SHADER_CAP_MAX_INPUTS: + return 16; + case PIPE_SHADER_CAP_MAX_CONSTS: + return 256; + case PIPE_SHADER_CAP_MAX_CONST_BUFFERS: + return 1; + case PIPE_SHADER_CAP_MAX_TEMPS: + return 32; + case PIPE_SHADER_CAP_MAX_ADDRS: + return 1; /* XXX guessed */ + case PIPE_SHADER_CAP_MAX_PREDS: + return is_r500 ? 4 : 0; /* XXX guessed. */ + case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED: + return 1; + default: + break; + } + break; + default: + break; + } + return 0; +} + +static float r300_get_paramf(struct pipe_screen* pscreen, enum pipe_cap param) { struct r300_screen* r300screen = r300_screen(pscreen); @@ -197,14 +273,13 @@ static float r300_get_paramf(struct pipe_screen* pscreen, int param) static boolean r300_is_format_supported(struct pipe_screen* screen, enum pipe_format format, enum pipe_texture_target target, + unsigned sample_count, unsigned usage, unsigned geom_flags) { uint32_t retval = 0; boolean is_r500 = r300_screen(screen)->caps.is_r500; boolean is_r400 = r300_screen(screen)->caps.is_r400; - boolean is_z24 = format == PIPE_FORMAT_X8Z24_UNORM || - format == PIPE_FORMAT_S8_USCALED_Z24_UNORM; boolean is_color2101010 = format == PIPE_FORMAT_R10G10B10A2_UNORM || format == PIPE_FORMAT_R10G10B10X2_SNORM || format == PIPE_FORMAT_B10G10R10A2_UNORM || @@ -213,17 +288,35 @@ static boolean r300_is_format_supported(struct pipe_screen* screen, format == PIPE_FORMAT_RGTC1_SNORM; boolean is_ati2n = format == PIPE_FORMAT_RGTC2_UNORM || format == PIPE_FORMAT_RGTC2_SNORM; + boolean is_half_float = format == PIPE_FORMAT_R16_FLOAT || + format == PIPE_FORMAT_R16G16_FLOAT || + format == PIPE_FORMAT_R16G16B16_FLOAT || + format == PIPE_FORMAT_R16G16B16A16_FLOAT; - if (target >= PIPE_MAX_TEXTURE_TYPES) { - fprintf(stderr, "r300: Implementation error: Received bogus texture " - "target %d in %s\n", target, __FUNCTION__); - return FALSE; + /* Check multisampling support. */ + switch (sample_count) { + case 0: + case 1: + break; + case 2: + case 3: + case 4: + case 6: + return FALSE; +#if 0 + if (usage != PIPE_BIND_RENDER_TARGET || + !util_format_is_rgba8_variant( + util_format_description(format))) { + return FALSE; + } +#endif + break; + default: + return FALSE; } /* Check sampler format support. */ if ((usage & PIPE_BIND_SAMPLER_VIEW) && - /* Z24 cannot be sampled from on non-r5xx. */ - (is_r500 || !is_z24) && /* ATI1N is r5xx-only. */ (is_r500 || !is_ati1n) && /* ATI2N is supported on r4xx-r5xx. */ @@ -253,6 +346,20 @@ static boolean r300_is_format_supported(struct pipe_screen* screen, retval |= PIPE_BIND_DEPTH_STENCIL; } + /* Check vertex buffer format support. */ + if (usage & PIPE_BIND_VERTEX_BUFFER && + /* Half float is supported on >= RV350. */ + (is_r400 || is_r500 || !is_half_float) && + r300_translate_vertex_data_type(format) != R300_INVALID_FORMAT) { + retval |= PIPE_BIND_VERTEX_BUFFER; + } + + /* Transfers are always supported. */ + if (usage & PIPE_BIND_TRANSFER_READ) + retval |= PIPE_BIND_TRANSFER_READ; + if (usage & PIPE_BIND_TRANSFER_WRITE) + retval |= PIPE_BIND_TRANSFER_WRITE; + return retval == usage; } @@ -261,6 +368,8 @@ static void r300_destroy_screen(struct pipe_screen* pscreen) struct r300_screen* r300screen = r300_screen(pscreen); struct r300_winsys_screen *rws = r300_winsys_screen(pscreen); + util_mempool_destroy(&r300screen->pool_buffers); + if (rws) rws->destroy(rws); @@ -271,23 +380,36 @@ static void r300_fence_reference(struct pipe_screen *screen, struct pipe_fence_handle **ptr, struct pipe_fence_handle *fence) { + struct r300_fence **oldf = (struct r300_fence**)ptr; + struct r300_fence *newf = (struct r300_fence*)fence; + + if (pipe_reference(&(*oldf)->reference, &newf->reference)) + FREE(*oldf); + + *ptr = fence; } static int r300_fence_signalled(struct pipe_screen *screen, struct pipe_fence_handle *fence, unsigned flags) { - return 0; + struct r300_fence *rfence = (struct r300_fence*)fence; + + return rfence->signalled ? 0 : 1; /* 0 == success */ } static int r300_fence_finish(struct pipe_screen *screen, struct pipe_fence_handle *fence, unsigned flags) { - return 0; + struct r300_fence *rfence = (struct r300_fence*)fence; + + r300_finish(rfence->ctx); + rfence->signalled = TRUE; + return 0; /* 0 == success */ } -struct pipe_screen* r300_create_screen(struct r300_winsys_screen *rws) +struct pipe_screen* r300_screen_create(struct r300_winsys_screen *rws) { struct r300_screen *r300screen = CALLOC_STRUCT(r300_screen); @@ -303,12 +425,17 @@ struct pipe_screen* r300_create_screen(struct r300_winsys_screen *rws) r300_init_debug(r300screen); r300_parse_chipset(&r300screen->caps); + util_mempool_create(&r300screen->pool_buffers, + sizeof(struct r300_buffer), 64, + UTIL_MEMPOOL_SINGLETHREADED); + r300screen->rws = rws; r300screen->screen.winsys = (struct pipe_winsys*)rws; r300screen->screen.destroy = r300_destroy_screen; r300screen->screen.get_name = r300_get_name; r300screen->screen.get_vendor = r300_get_vendor; r300screen->screen.get_param = r300_get_param; + r300screen->screen.get_shader_param = r300_get_shader_param; r300screen->screen.get_paramf = r300_get_paramf; r300screen->screen.is_format_supported = r300_is_format_supported; r300screen->screen.context_create = r300_create_context; @@ -319,11 +446,7 @@ struct pipe_screen* r300_create_screen(struct r300_winsys_screen *rws) r300_init_screen_resource_functions(r300screen); - return &r300screen->screen; -} + util_format_s3tc_init(); -struct r300_winsys_screen * -r300_winsys_screen(struct pipe_screen *screen) -{ - return r300_screen(screen)->rws; + return &r300screen->screen; } diff --git a/src/gallium/drivers/r300/r300_screen.h b/src/gallium/drivers/r300/r300_screen.h index 70ea2fe8d92..dc2bc7e8279 100644 --- a/src/gallium/drivers/r300/r300_screen.h +++ b/src/gallium/drivers/r300/r300_screen.h @@ -28,8 +28,12 @@ #include "r300_chipset.h" +#include "util/u_mempool.h" + #include <stdio.h> +struct r300_winsys_screen; + struct r300_screen { /* Parent class */ struct pipe_screen screen; @@ -39,16 +43,28 @@ struct r300_screen { /* Chipset capabilities */ struct r300_capabilities caps; + /* Memory pools. */ + struct util_mempool pool_buffers; + /** Combination of DBG_xxx flags */ unsigned debug; + + /* The number of created contexts to know whether we have multiple + * contexts or not. */ + int num_contexts; }; -/* Convenience cast wrapper. */ +/* Convenience cast wrappers. */ static INLINE struct r300_screen* r300_screen(struct pipe_screen* screen) { return (struct r300_screen*)screen; } +static INLINE struct r300_winsys_screen * +r300_winsys_screen(struct pipe_screen *screen) { + return r300_screen(screen)->rws; +} + /* Debug functionality. */ /** @@ -61,17 +77,31 @@ static INLINE struct r300_screen* r300_screen(struct pipe_screen* screen) { * those changes. */ /*@{*/ -#define DBG_HELP 0x0000001 -#define DBG_FP 0x0000002 -#define DBG_VP 0x0000004 -#define DBG_CS 0x0000008 -#define DBG_DRAW 0x0000010 -#define DBG_TEX 0x0000020 -#define DBG_FALL 0x0000040 -#define DBG_ANISOHQ 0x0000080 -#define DBG_NO_TILING 0x0000100 -#define DBG_NO_IMMD 0x0000200 -#define DBG_STATS 0x0000400 + +/* Logging. */ +#define DBG_PSC (1 << 0) +#define DBG_FP (1 << 1) +#define DBG_VP (1 << 2) +#define DBG_SWTCL (1 << 3) +#define DBG_DRAW (1 << 4) +#define DBG_TEX (1 << 5) +#define DBG_TEXALLOC (1 << 6) +#define DBG_RS (1 << 7) +#define DBG_FALL (1 << 8) +#define DBG_FB (1 << 9) +#define DBG_RS_BLOCK (1 << 10) +#define DBG_CBZB (1 << 11) +#define DBG_HYPERZ (1 << 12) +#define DBG_SCISSOR (1 << 13) +/* Features. */ +#define DBG_ANISOHQ (1 << 16) +#define DBG_NO_TILING (1 << 17) +#define DBG_NO_IMMD (1 << 18) +#define DBG_FAKE_OCC (1 << 19) +#define DBG_NO_OPT (1 << 20) +#define DBG_NO_CBZB (1 << 21) +/* Statistics. */ +#define DBG_STATS (1 << 24) /*@}*/ static INLINE boolean SCREEN_DBG_ON(struct r300_screen * screen, unsigned flags) diff --git a/src/gallium/drivers/r300/r300_screen_buffer.c b/src/gallium/drivers/r300/r300_screen_buffer.c index 3e2b5afe6f4..37a080ba48b 100644 --- a/src/gallium/drivers/r300/r300_screen_buffer.c +++ b/src/gallium/drivers/r300/r300_screen_buffer.c @@ -43,7 +43,7 @@ unsigned r300_buffer_is_referenced(struct pipe_context *context, if (r300_buffer_is_user_buffer(buf)) return PIPE_UNREFERENCED; - if (r300->rws->is_buffer_referenced(r300->rws, rbuf->buf, domain)) + if (r300->rws->cs_is_buffer_referenced(r300->cs, rbuf->buf, domain)) return PIPE_REFERENCED_FOR_READ | PIPE_REFERENCED_FOR_WRITE; return PIPE_UNREFERENCED; @@ -62,7 +62,8 @@ int r300_upload_index_buffer(struct r300_context *r300, struct pipe_resource **index_buffer, unsigned index_size, unsigned start, - unsigned count) + unsigned count, + unsigned *out_offset) { struct pipe_resource *upload_buffer = NULL; unsigned index_offset = start * index_size; @@ -79,7 +80,10 @@ int r300_upload_index_buffer(struct r300_context *r300, goto done; } *index_buffer = upload_buffer; - } + *out_offset = index_offset / index_size; + } else + *out_offset = start; + done: // if (upload_buffer) // pipe_resource_reference(&upload_buffer, NULL); @@ -93,67 +97,85 @@ int r300_upload_user_buffers(struct r300_context *r300) enum pipe_error ret = PIPE_OK; int i, nr; - nr = r300->vertex_buffer_count; + nr = r300->velems->count; for (i = 0; i < nr; i++) { - if (r300_buffer_is_user_buffer(r300->vertex_buffer[i].buffer)) { - struct pipe_resource *upload_buffer = NULL; - unsigned offset = 0; /*r300->vertex_buffer[i].buffer_offset * 4;*/ - unsigned size = r300->vertex_buffer[i].buffer->width0; - unsigned upload_offset; - ret = u_upload_buffer(r300->upload_vb, - offset, size, - r300->vertex_buffer[i].buffer, - &upload_offset, &upload_buffer); - if (ret) - return ret; - - pipe_resource_reference(&r300->vertex_buffer[i].buffer, NULL); - r300->vertex_buffer[i].buffer = upload_buffer; - r300->vertex_buffer[i].buffer_offset = upload_offset; - } + struct pipe_vertex_buffer *vb = + &r300->vertex_buffer[r300->velems->velem[i].vertex_buffer_index]; + + if (r300_buffer_is_user_buffer(vb->buffer)) { + struct pipe_resource *upload_buffer = NULL; + unsigned offset = 0; /*vb->buffer_offset * 4;*/ + unsigned size = vb->buffer->width0; + unsigned upload_offset; + ret = u_upload_buffer(r300->upload_vb, + offset, size, + vb->buffer, + &upload_offset, &upload_buffer); + if (ret) + return ret; + + pipe_resource_reference(&vb->buffer, NULL); + vb->buffer = upload_buffer; + vb->buffer_offset = upload_offset; + } } return ret; } -static struct r300_winsys_buffer * -r300_winsys_buffer_create(struct r300_screen *r300screen, - unsigned alignment, - unsigned usage, - unsigned size) +static void r300_buffer_destroy(struct pipe_screen *screen, + struct pipe_resource *buf) { + struct r300_screen *r300screen = r300_screen(screen); + struct r300_buffer *rbuf = r300_buffer(buf); struct r300_winsys_screen *rws = r300screen->rws; - struct r300_winsys_buffer *buf; - buf = rws->buffer_create(rws, alignment, usage, size); - return buf; -} + if (rbuf->constant_buffer) + FREE(rbuf->constant_buffer); -static void r300_winsys_buffer_destroy(struct r300_screen *r300screen, - struct r300_buffer *rbuf) -{ - struct r300_winsys_screen *rws = r300screen->rws; + if (rbuf->buf) + rws->buffer_reference(rws, &rbuf->buf, NULL); - if (rbuf->buf) { - rws->buffer_reference(rws, &rbuf->buf, NULL); - rbuf->buf = NULL; - } + util_mempool_free(&r300screen->pool_buffers, rbuf); } -static void r300_buffer_destroy(struct pipe_screen *screen, - struct pipe_resource *buf) +static struct pipe_transfer* +r300_default_get_transfer(struct pipe_context *context, + struct pipe_resource *resource, + struct pipe_subresource sr, + unsigned usage, + const struct pipe_box *box) { - struct r300_screen *r300screen = r300_screen(screen); - struct r300_buffer *rbuf = r300_buffer(buf); + struct r300_context *r300 = r300_context(context); + struct pipe_transfer *transfer = + util_mempool_malloc(&r300->pool_transfers); + + transfer->resource = resource; + transfer->sr = sr; + transfer->usage = usage; + transfer->box = *box; + transfer->stride = 0; + transfer->slice_stride = 0; + transfer->data = NULL; + + /* Note strides are zero, this is ok for buffers, but not for + * textures 2d & higher at least. + */ + return transfer; +} - r300_winsys_buffer_destroy(r300screen, rbuf); - FREE(rbuf); +static void r300_default_transfer_destroy(struct pipe_context *pipe, + struct pipe_transfer *transfer) +{ + struct r300_context *r300 = r300_context(pipe); + util_mempool_free(&r300->pool_transfers, transfer); } static void * r300_buffer_transfer_map( struct pipe_context *pipe, struct pipe_transfer *transfer ) { + struct r300_context *r300 = r300_context(pipe); struct r300_screen *r300screen = r300_screen(pipe->screen); struct r300_winsys_screen *rws = r300screen->rws; struct r300_buffer *rbuf = r300_buffer(transfer->resource); @@ -163,10 +185,8 @@ r300_buffer_transfer_map( struct pipe_context *pipe, if (rbuf->user_buffer) return (uint8_t *) rbuf->user_buffer + transfer->box.x; - - if (rbuf->b.b.bind & PIPE_BIND_CONSTANT_BUFFER) { - goto just_map; - } + if (rbuf->constant_buffer) + return (uint8_t *) rbuf->constant_buffer + transfer->box.x; /* check if the mapping is to a range we already flushed */ if (transfer->usage & PIPE_TRANSFER_DISCARD) { @@ -180,16 +200,18 @@ r300_buffer_transfer_map( struct pipe_context *pipe, rws->buffer_reference(rws, &rbuf->buf, NULL); rbuf->num_ranges = 0; - rbuf->buf = r300_winsys_buffer_create(r300screen, - 16, - rbuf->b.b.bind, /* XXX */ - rbuf->b.b.width0); + rbuf->buf = + r300screen->rws->buffer_create(r300screen->rws, + rbuf->b.b.width0, 16, + rbuf->b.b.bind, + rbuf->b.b.usage, + rbuf->domain); break; } } } -just_map: - map = rws->buffer_map(rws, rbuf->buf, transfer->usage); + + map = rws->buffer_map(rws, rbuf->buf, r300->cs, transfer->usage); if (map == NULL) return NULL; @@ -214,9 +236,8 @@ static void r300_buffer_transfer_flush_region( struct pipe_context *pipe, if (rbuf->user_buffer) return; - - if (rbuf->b.b.bind & PIPE_BIND_CONSTANT_BUFFER) - return; + if (rbuf->constant_buffer) + return; /* mark the range as used */ for(i = 0; i < rbuf->num_ranges; ++i) { @@ -247,14 +268,14 @@ static void r300_buffer_transfer_unmap( struct pipe_context *pipe, struct u_resource_vtbl r300_buffer_vtbl = { u_default_resource_get_handle, /* get_handle */ - r300_buffer_destroy, /* resource_destroy */ - r300_buffer_is_referenced_by_cs, /* is_buffer_referenced */ - u_default_get_transfer, /* get_transfer */ - u_default_transfer_destroy, /* transfer_destroy */ - r300_buffer_transfer_map, /* transfer_map */ + r300_buffer_destroy, /* resource_destroy */ + r300_buffer_is_referenced_by_cs, /* is_buffer_referenced */ + r300_default_get_transfer, /* get_transfer */ + r300_default_transfer_destroy, /* transfer_destroy */ + r300_buffer_transfer_map, /* transfer_map */ r300_buffer_transfer_flush_region, /* transfer_flush_region */ - r300_buffer_transfer_unmap, /* transfer_unmap */ - u_default_transfer_inline_write /* transfer_inline_write */ + r300_buffer_transfer_unmap, /* transfer_unmap */ + u_default_transfer_inline_write /* transfer_inline_write */ }; struct pipe_resource *r300_buffer_create(struct pipe_screen *screen, @@ -264,9 +285,7 @@ struct pipe_resource *r300_buffer_create(struct pipe_screen *screen, struct r300_buffer *rbuf; unsigned alignment = 16; - rbuf = CALLOC_STRUCT(r300_buffer); - if (!rbuf) - goto error1; + rbuf = util_mempool_malloc(&r300screen->pool_buffers); rbuf->magic = R300_BUFFER_MAGIC; @@ -274,23 +293,30 @@ struct pipe_resource *r300_buffer_create(struct pipe_screen *screen, rbuf->b.vtbl = &r300_buffer_vtbl; pipe_reference_init(&rbuf->b.b.reference, 1); rbuf->b.b.screen = screen; + rbuf->domain = R300_DOMAIN_GTT; + rbuf->num_ranges = 0; + rbuf->buf = NULL; + rbuf->constant_buffer = NULL; + rbuf->user_buffer = NULL; + + /* Alloc constant buffers in RAM. */ + if (templ->bind & PIPE_BIND_CONSTANT_BUFFER) { + rbuf->constant_buffer = MALLOC(templ->width0); + return &rbuf->b.b; + } - if (rbuf->b.b.bind & R300_BIND_OQBO) - alignment = 4096; - - rbuf->buf = r300_winsys_buffer_create(r300screen, - alignment, - rbuf->b.b.bind, - rbuf->b.b.width0); + rbuf->buf = + r300screen->rws->buffer_create(r300screen->rws, + rbuf->b.b.width0, alignment, + rbuf->b.b.bind, rbuf->b.b.usage, + rbuf->domain); - if (!rbuf->buf) - goto error2; + if (!rbuf->buf) { + util_mempool_free(&r300screen->pool_buffers, rbuf); + return NULL; + } return &rbuf->b.b; -error2: - FREE(rbuf); -error1: - return NULL; } struct pipe_resource *r300_user_buffer_create(struct pipe_screen *screen, @@ -298,27 +324,28 @@ struct pipe_resource *r300_user_buffer_create(struct pipe_screen *screen, unsigned bytes, unsigned bind) { + struct r300_screen *r300screen = r300_screen(screen); struct r300_buffer *rbuf; - rbuf = CALLOC_STRUCT(r300_buffer); - if (!rbuf) - goto no_rbuf; + rbuf = util_mempool_malloc(&r300screen->pool_buffers); rbuf->magic = R300_BUFFER_MAGIC; pipe_reference_init(&rbuf->b.b.reference, 1); rbuf->b.vtbl = &r300_buffer_vtbl; rbuf->b.b.screen = screen; + rbuf->b.b.target = PIPE_BUFFER; rbuf->b.b.format = PIPE_FORMAT_R8_UNORM; rbuf->b.b.usage = PIPE_USAGE_IMMUTABLE; rbuf->b.b.bind = bind; rbuf->b.b.width0 = bytes; rbuf->b.b.height0 = 1; rbuf->b.b.depth0 = 1; - + rbuf->b.b.flags = 0; + rbuf->domain = R300_DOMAIN_GTT; + rbuf->num_ranges = 0; + rbuf->buf = NULL; + rbuf->constant_buffer = NULL; rbuf->user_buffer = ptr; return &rbuf->b.b; - -no_rbuf: - return NULL; } diff --git a/src/gallium/drivers/r300/r300_screen_buffer.h b/src/gallium/drivers/r300/r300_screen_buffer.h index 57f48229b2e..cafa9f96f20 100644 --- a/src/gallium/drivers/r300/r300_screen_buffer.h +++ b/src/gallium/drivers/r300/r300_screen_buffer.h @@ -52,7 +52,10 @@ struct r300_buffer struct r300_winsys_buffer *buf; + enum r300_buffer_domain domain; + void *user_buffer; + void *constant_buffer; struct r300_buffer_range ranges[R300_BUFFER_MAX_RANGES]; unsigned num_ranges; }; @@ -65,7 +68,7 @@ int r300_upload_index_buffer(struct r300_context *r300, struct pipe_resource **index_buffer, unsigned index_size, unsigned start, - unsigned count); + unsigned count, unsigned *out_offset); struct pipe_resource *r300_buffer_create(struct pipe_screen *screen, const struct pipe_resource *templ); @@ -95,40 +98,4 @@ static INLINE boolean r300_buffer_is_user_buffer(struct pipe_resource *buffer) return r300_buffer(buffer)->user_buffer ? true : false; } -static INLINE boolean r300_add_buffer(struct r300_winsys_screen *rws, - struct pipe_resource *buffer, - int rd, int wr) -{ - struct r300_buffer *buf = r300_buffer(buffer); - - if (!buf->buf) - return true; - - return rws->add_buffer(rws, buf->buf, rd, wr); -} - -static INLINE boolean r300_add_texture(struct r300_winsys_screen *rws, - struct r300_texture *tex, - int rd, int wr) -{ - return rws->add_buffer(rws, tex->buffer, rd, wr); -} - -static INLINE void r300_buffer_write_reloc(struct r300_winsys_screen *rws, - struct r300_buffer *buf, - uint32_t rd, uint32_t wd, uint32_t flags) -{ - if (!buf->buf) - return; - - rws->write_cs_reloc(rws, buf->buf, rd, wd, flags); -} - -static INLINE void r300_texture_write_reloc(struct r300_winsys_screen *rws, - struct r300_texture *texture, - uint32_t rd, uint32_t wd, uint32_t flags) -{ - rws->write_cs_reloc(rws, texture->buffer, rd, wd, flags); -} - #endif diff --git a/src/gallium/drivers/r300/r300_shader_semantics.h b/src/gallium/drivers/r300/r300_shader_semantics.h index cb7a37033f3..4be23e64ce7 100644 --- a/src/gallium/drivers/r300/r300_shader_semantics.h +++ b/src/gallium/drivers/r300/r300_shader_semantics.h @@ -38,6 +38,7 @@ struct r300_shader_semantics { int psize; int color[ATTR_COLOR_COUNT]; int bcolor[ATTR_COLOR_COUNT]; + int face; int generic[ATTR_GENERIC_COUNT]; int fog; int wpos; @@ -50,6 +51,7 @@ static INLINE void r300_shader_semantics_reset( info->pos = ATTR_UNUSED; info->psize = ATTR_UNUSED; + info->face = ATTR_UNUSED; info->fog = ATTR_UNUSED; info->wpos = ATTR_UNUSED; diff --git a/src/gallium/drivers/r300/r300_state.c b/src/gallium/drivers/r300/r300_state.c index edd522288bd..f2479a994c8 100644 --- a/src/gallium/drivers/r300/r300_state.c +++ b/src/gallium/drivers/r300/r300_state.c @@ -23,7 +23,9 @@ #include "draw/draw_context.h" +#include "util/u_framebuffer.h" #include "util/u_math.h" +#include "util/u_mm.h" #include "util/u_memory.h" #include "util/u_pack_color.h" @@ -31,17 +33,18 @@ #include "pipe/p_config.h" +#include "r300_cb.h" #include "r300_context.h" #include "r300_emit.h" #include "r300_reg.h" #include "r300_screen.h" #include "r300_screen_buffer.h" -#include "r300_state.h" #include "r300_state_inlines.h" #include "r300_fs.h" #include "r300_texture.h" #include "r300_vs.h" #include "r300_winsys.h" +#include "r300_hyperz.h" /* r300_state: Functions used to intialize state context by translating * Gallium state objects into semi-native r300 state objects. */ @@ -183,6 +186,12 @@ static void* r300_create_blend_state(struct pipe_context* pipe, { struct r300_screen* r300screen = r300_screen(pipe->screen); struct r300_blend_state* blend = CALLOC_STRUCT(r300_blend_state); + uint32_t blend_control = 0; /* R300_RB3D_CBLEND: 0x4e04 */ + uint32_t alpha_blend_control = 0; /* R300_RB3D_ABLEND: 0x4e08 */ + uint32_t color_channel_mask = 0; /* R300_RB3D_COLOR_CHANNEL_MASK: 0x4e0c */ + uint32_t rop = 0; /* R300_RB3D_ROPCNTL: 0x4e18 */ + uint32_t dither = 0; /* R300_RB3D_DITHER_CTL: 0x4e50 */ + CB_LOCALS; if (state->rt[0].blend_enable) { @@ -196,7 +205,7 @@ static void* r300_create_blend_state(struct pipe_context* pipe, /* despite the name, ALPHA_BLEND_ENABLE has nothing to do with alpha, * this is just the crappy D3D naming */ - blend->blend_control = R300_ALPHA_BLEND_ENABLE | + blend_control = R300_ALPHA_BLEND_ENABLE | r300_translate_blend_function(eqRGB) | ( r300_translate_blend_factor(srcRGB) << R300_SRC_BLEND_SHIFT) | ( r300_translate_blend_factor(dstRGB) << R300_DST_BLEND_SHIFT); @@ -220,7 +229,7 @@ static void* r300_create_blend_state(struct pipe_context* pipe, srcA == PIPE_BLENDFACTOR_INV_DST_ALPHA || srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE) { /* Enable reading from the colorbuffer. */ - blend->blend_control |= R300_READ_ENABLE; + blend_control |= R300_READ_ENABLE; if (r300screen->caps.is_r500) { /* Optimization: Depending on incoming pixels, we can @@ -233,7 +242,7 @@ static void* r300_create_blend_state(struct pipe_context* pipe, (dstA == PIPE_BLENDFACTOR_SRC_COLOR || dstA == PIPE_BLENDFACTOR_SRC_ALPHA || dstA == PIPE_BLENDFACTOR_ZERO)) { - blend->blend_control |= R500_SRC_ALPHA_0_NO_READ; + blend_control |= R500_SRC_ALPHA_0_NO_READ; } /* Disable reading if SRC_ALPHA == 1. */ @@ -242,7 +251,7 @@ static void* r300_create_blend_state(struct pipe_context* pipe, (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR || dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || dstA == PIPE_BLENDFACTOR_ZERO)) { - blend->blend_control |= R500_SRC_ALPHA_1_NO_READ; + blend_control |= R500_SRC_ALPHA_1_NO_READ; } } } @@ -272,31 +281,31 @@ static void* r300_create_blend_state(struct pipe_context* pipe, * pixels. */ if (blend_discard_if_src_alpha_0(srcRGB, srcA, dstRGB, dstA)) { - blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_0; + blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_0; } else if (blend_discard_if_src_alpha_1(srcRGB, srcA, dstRGB, dstA)) { - blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_1; + blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_1; } else if (blend_discard_if_src_color_0(srcRGB, srcA, dstRGB, dstA)) { - blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_0; + blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_0; } else if (blend_discard_if_src_color_1(srcRGB, srcA, dstRGB, dstA)) { - blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_1; + blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_1; } else if (blend_discard_if_src_alpha_color_0(srcRGB, srcA, dstRGB, dstA)) { - blend->blend_control |= + blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_0; } else if (blend_discard_if_src_alpha_color_1(srcRGB, srcA, dstRGB, dstA)) { - blend->blend_control |= + blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_1; } } /* separate alpha */ if (srcA != srcRGB || dstA != dstRGB || eqA != eqRGB) { - blend->blend_control |= R300_SEPARATE_ALPHA_ENABLE; - blend->alpha_blend_control = + blend_control |= R300_SEPARATE_ALPHA_ENABLE; + alpha_blend_control = r300_translate_blend_function(eqA) | (r300_translate_blend_factor(srcA) << R300_SRC_BLEND_SHIFT) | (r300_translate_blend_factor(dstA) << R300_DST_BLEND_SHIFT); @@ -305,21 +314,21 @@ static void* r300_create_blend_state(struct pipe_context* pipe, /* PIPE_LOGICOP_* don't need to be translated, fortunately. */ if (state->logicop_enable) { - blend->rop = R300_RB3D_ROPCNTL_ROP_ENABLE | + rop = R300_RB3D_ROPCNTL_ROP_ENABLE | (state->logicop_func) << R300_RB3D_ROPCNTL_ROP_SHIFT; } /* Color channel masks for all MRTs. */ - blend->color_channel_mask = bgra_cmask(state->rt[0].colormask); + color_channel_mask = bgra_cmask(state->rt[0].colormask); if (r300screen->caps.is_r500 && state->independent_blend_enable) { if (state->rt[1].blend_enable) { - blend->color_channel_mask |= bgra_cmask(state->rt[1].colormask) << 4; + color_channel_mask |= bgra_cmask(state->rt[1].colormask) << 4; } if (state->rt[2].blend_enable) { - blend->color_channel_mask |= bgra_cmask(state->rt[2].colormask) << 8; + color_channel_mask |= bgra_cmask(state->rt[2].colormask) << 8; } if (state->rt[3].blend_enable) { - blend->color_channel_mask |= bgra_cmask(state->rt[3].colormask) << 12; + color_channel_mask |= bgra_cmask(state->rt[3].colormask) << 12; } } @@ -330,11 +339,31 @@ static void* r300_create_blend_state(struct pipe_context* pipe, * This could be revisited if we ever get quality or conformance hints. * if (state->dither) { - blend->dither = R300_RB3D_DITHER_CTL_DITHER_MODE_LUT | + dither = R300_RB3D_DITHER_CTL_DITHER_MODE_LUT | R300_RB3D_DITHER_CTL_ALPHA_DITHER_MODE_LUT; } */ + /* Build a command buffer. */ + BEGIN_CB(blend->cb, 8); + OUT_CB_REG(R300_RB3D_ROPCNTL, rop); + OUT_CB_REG_SEQ(R300_RB3D_CBLEND, 3); + OUT_CB(blend_control); + OUT_CB(alpha_blend_control); + OUT_CB(color_channel_mask); + OUT_CB_REG(R300_RB3D_DITHER_CTL, dither); + END_CB; + + /* The same as above, but with no colorbuffer reads and writes. */ + BEGIN_CB(blend->cb_no_readwrite, 8); + OUT_CB_REG(R300_RB3D_ROPCNTL, rop); + OUT_CB_REG_SEQ(R300_RB3D_CBLEND, 3); + OUT_CB(0); + OUT_CB(0); + OUT_CB(0); + OUT_CB_REG(R300_RB3D_DITHER_CTL, dither); + END_CB; + return (void*)blend; } @@ -368,20 +397,26 @@ static void r300_set_blend_color(struct pipe_context* pipe, struct r300_context* r300 = r300_context(pipe); struct r300_blend_color_state* state = (struct r300_blend_color_state*)r300->blend_color_state.state; - union util_color uc; + CB_LOCALS; - util_pack_color(color->color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc); - state->blend_color = uc.ui; + if (r300->screen->caps.is_r500) { + /* XXX if FP16 blending is enabled, we should use the FP16 format */ + BEGIN_CB(state->cb, 3); + OUT_CB_REG_SEQ(R500_RB3D_CONSTANT_COLOR_AR, 2); + OUT_CB(float_to_fixed10(color->color[0]) | + (float_to_fixed10(color->color[3]) << 16)); + OUT_CB(float_to_fixed10(color->color[2]) | + (float_to_fixed10(color->color[1]) << 16)); + END_CB; + } else { + union util_color uc; + util_pack_color(color->color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc); - /* XXX if FP16 blending is enabled, we should use the FP16 format */ - state->blend_color_red_alpha = - float_to_fixed10(color->color[0]) | - (float_to_fixed10(color->color[3]) << 16); - state->blend_color_green_blue = - float_to_fixed10(color->color[2]) | - (float_to_fixed10(color->color[1]) << 16); + BEGIN_CB(state->cb, 2); + OUT_CB_REG(R300_RB3D_BLEND_COLOR, uc.ui); + END_CB; + } - r300->blend_color_state.size = r300->screen->caps.is_r500 ? 3 : 2; r300->blend_color_state.dirty = TRUE; } @@ -389,21 +424,41 @@ static void r300_set_clip_state(struct pipe_context* pipe, const struct pipe_clip_state* state) { struct r300_context* r300 = r300_context(pipe); + struct r300_clip_state *clip = + (struct r300_clip_state*)r300->clip_state.state; + CB_LOCALS; - r300->clip = *state; + clip->clip = *state; if (r300->screen->caps.has_tcl) { - memcpy(r300->clip_state.state, state, sizeof(struct pipe_clip_state)); - r300->clip_state.size = 29; + r300->clip_state.size = 2 + !!state->nr * 3 + state->nr * 4; + + BEGIN_CB(clip->cb, r300->clip_state.size); + if (state->nr) { + OUT_CB_REG(R300_VAP_PVS_VECTOR_INDX_REG, + (r300->screen->caps.is_r500 ? + R500_PVS_UCP_START : R300_PVS_UCP_START)); + OUT_CB_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, state->nr * 4); + OUT_CB_TABLE(state->ucp, state->nr * 4); + } + OUT_CB_REG(R300_VAP_CLIP_CNTL, ((1 << state->nr) - 1) | + R300_PS_UCP_MODE_CLIP_AS_TRIFAN | + (state->depth_clamp ? R300_CLIP_DISABLE : 0)); + END_CB; + + r300->clip_state.dirty = TRUE; } else { - draw_flush(r300->draw); draw_set_clip_state(r300->draw, state); - r300->clip_state.size = 2; } +} - r300->clip_state.dirty = TRUE; +static void +r300_set_sample_mask(struct pipe_context *pipe, + unsigned sample_mask) +{ } + /* Create a new depth, stencil, and alpha state based on the CSO dsa state. * * This contains the depth buffer, stencil buffer, alpha test, and such. @@ -415,15 +470,18 @@ static void* { struct r300_capabilities *caps = &r300_screen(pipe->screen)->caps; struct r300_dsa_state* dsa = CALLOC_STRUCT(r300_dsa_state); + CB_LOCALS; + + dsa->dsa = *state; + + /* Depth test setup. - separate write mask depth for decomp flush */ + if (state->depth.writemask) { + dsa->z_buffer_control |= R300_Z_WRITE_ENABLE; + } - /* Depth test setup. */ if (state->depth.enabled) { dsa->z_buffer_control |= R300_Z_ENABLE; - if (state->depth.writemask) { - dsa->z_buffer_control |= R300_Z_WRITE_ENABLE; - } - dsa->z_stencil_control |= (r300_translate_depth_stencil_function(state->depth.func) << R300_Z_FUNC_SHIFT); @@ -467,7 +525,7 @@ static void* if (caps->is_r500) { dsa->z_buffer_control |= R500_STENCIL_REFMASK_FRONT_BACK; } else { - dsa->stencil_ref_bf_fallback = + dsa->two_sided_stencil_ref = (state->stencil[0].valuemask != state->stencil[1].valuemask || state->stencil[0].writemask != state->stencil[1].writemask); } @@ -487,21 +545,41 @@ static void* dsa->alpha_function |= R500_FG_ALPHA_FUNC_8BIT; } + BEGIN_CB(&dsa->cb_begin, 8); + OUT_CB_REG(R300_FG_ALPHA_FUNC, dsa->alpha_function); + OUT_CB_REG_SEQ(R300_ZB_CNTL, 3); + OUT_CB(dsa->z_buffer_control); + OUT_CB(dsa->z_stencil_control); + OUT_CB(dsa->stencil_ref_mask); + OUT_CB_REG(R500_ZB_STENCILREFMASK_BF, dsa->stencil_ref_bf); + END_CB; + + BEGIN_CB(dsa->cb_no_readwrite, 8); + OUT_CB_REG(R300_FG_ALPHA_FUNC, dsa->alpha_function); + OUT_CB_REG_SEQ(R300_ZB_CNTL, 3); + OUT_CB(0); + OUT_CB(0); + OUT_CB(0); + OUT_CB_REG(R500_ZB_STENCILREFMASK_BF, 0); + END_CB; + return (void*)dsa; } -static void r300_update_stencil_ref_fallback_status(struct r300_context *r300) +static void r300_dsa_inject_stencilref(struct r300_context *r300) { - struct r300_dsa_state *dsa = (struct r300_dsa_state*)r300->dsa_state.state; + struct r300_dsa_state *dsa = + (struct r300_dsa_state*)r300->dsa_state.state; - if (r300->screen->caps.is_r500) { + if (!dsa) return; - } - r300->stencil_ref_bf_fallback = - dsa->stencil_ref_bf_fallback || - (dsa->two_sided && - r300->stencil_ref.ref_value[0] != r300->stencil_ref.ref_value[1]); + dsa->stencil_ref_mask = + (dsa->stencil_ref_mask & ~R300_STENCILREF_MASK) | + r300->stencil_ref.ref_value[0]; + dsa->stencil_ref_bf = + (dsa->stencil_ref_bf & ~R300_STENCILREF_MASK) | + r300->stencil_ref.ref_value[1]; } /* Bind DSA state. */ @@ -516,7 +594,8 @@ static void r300_bind_dsa_state(struct pipe_context* pipe, UPDATE_STATE(state, r300->dsa_state); - r300_update_stencil_ref_fallback_status(r300); + r300->hyperz_state.dirty = TRUE; /* Will be updated before the emission. */ + r300_dsa_inject_stencilref(r300); } /* Free DSA state. */ @@ -532,72 +611,103 @@ static void r300_set_stencil_ref(struct pipe_context* pipe, struct r300_context* r300 = r300_context(pipe); r300->stencil_ref = *sr; + + r300_dsa_inject_stencilref(r300); r300->dsa_state.dirty = TRUE; +} + +static void r300_tex_set_tiling_flags(struct r300_context *r300, + struct r300_texture *tex, unsigned level) +{ + /* Check if the macrotile flag needs to be changed. + * Skip changing the flags otherwise. */ + if (tex->desc.macrotile[tex->surface_level] != + tex->desc.macrotile[level]) { + /* Tiling determines how DRM treats the buffer data. + * We must flush CS when changing it if the buffer is referenced. */ + if (r300->rws->cs_is_buffer_referenced(r300->cs, + tex->buffer, R300_REF_CS)) + r300->context.flush(&r300->context, 0, NULL); - r300_update_stencil_ref_fallback_status(r300); + r300->rws->buffer_set_tiling(r300->rws, tex->buffer, + tex->desc.microtile, tex->desc.macrotile[level], + tex->desc.stride_in_bytes[0]); + + tex->surface_level = level; + } } /* This switcheroo is needed just because of goddamned MACRO_SWITCH. */ -static void r300_fb_update_tiling_flags(struct r300_context *r300, - const struct pipe_framebuffer_state *old_state, - const struct pipe_framebuffer_state *new_state) +static void r300_fb_set_tiling_flags(struct r300_context *r300, + const struct pipe_framebuffer_state *state) { - struct r300_texture *tex; - unsigned i, j, level; + unsigned i; - /* Reset tiling flags for old surfaces to default values. */ - for (i = 0; i < old_state->nr_cbufs; i++) { - for (j = 0; j < new_state->nr_cbufs; j++) { - if (old_state->cbufs[i]->texture == new_state->cbufs[j]->texture) { - break; - } - } - /* If not binding the surface again... */ - if (j != new_state->nr_cbufs) { - continue; - } + /* Set tiling flags for new surfaces. */ + for (i = 0; i < state->nr_cbufs; i++) { + r300_tex_set_tiling_flags(r300, + r300_texture(state->cbufs[i]->texture), + state->cbufs[i]->level); + } + if (state->zsbuf) { + r300_tex_set_tiling_flags(r300, + r300_texture(state->zsbuf->texture), + state->zsbuf->level); + } +} - tex = r300_texture(old_state->cbufs[i]->texture); +static void r300_print_fb_surf_info(struct pipe_surface *surf, unsigned index, + const char *binding) +{ + struct pipe_resource *tex = surf->texture; + struct r300_texture *rtex = r300_texture(tex); - if (tex) { - r300->rws->buffer_set_tiling(r300->rws, tex->buffer, - tex->pitch[0], - tex->microtile, - tex->macrotile); - } - } - if (old_state->zsbuf && - (!new_state->zsbuf || - old_state->zsbuf->texture != new_state->zsbuf->texture)) { - tex = r300_texture(old_state->zsbuf->texture); + fprintf(stderr, + "r300: %s[%i] Dim: %ix%i, Offset: %i, ZSlice: %i, " + "Face: %i, Level: %i, Format: %s\n" - if (tex) { - r300->rws->buffer_set_tiling(r300->rws, tex->buffer, - tex->pitch[0], - tex->microtile, - tex->macrotile); - } - } + "r300: TEX: Macro: %s, Micro: %s, Pitch: %i, " + "Dim: %ix%ix%i, LastLevel: %i, Format: %s\n", - /* Set tiling flags for new surfaces. */ - for (i = 0; i < new_state->nr_cbufs; i++) { - tex = r300_texture(new_state->cbufs[i]->texture); - level = new_state->cbufs[i]->level; + binding, index, surf->width, surf->height, surf->offset, + surf->zslice, surf->face, surf->level, + util_format_short_name(surf->format), - r300->rws->buffer_set_tiling(r300->rws, tex->buffer, - tex->pitch[level], - tex->microtile, - tex->mip_macrotile[level]); + rtex->desc.macrotile[0] ? "YES" : " NO", + rtex->desc.microtile ? "YES" : " NO", + rtex->desc.stride_in_pixels[0], + tex->width0, tex->height0, tex->depth0, + tex->last_level, util_format_short_name(tex->format)); +} + +void r300_mark_fb_state_dirty(struct r300_context *r300, + enum r300_fb_state_change change) +{ + struct pipe_framebuffer_state *state = r300->fb_state.state; + boolean has_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ); + + /* What is marked as dirty depends on the enum r300_fb_state_change. */ + r300->gpu_flush.dirty = TRUE; + r300->fb_state.dirty = TRUE; + r300->hyperz_state.dirty = TRUE; + + if (change == R300_CHANGED_FB_STATE) { + r300->aa_state.dirty = TRUE; + r300->fb_state_pipelined.dirty = TRUE; } - if (new_state->zsbuf) { - tex = r300_texture(new_state->zsbuf->texture); - level = new_state->zsbuf->level; - r300->rws->buffer_set_tiling(r300->rws, tex->buffer, - tex->pitch[level], - tex->microtile, - tex->mip_macrotile[level]); + /* Now compute the fb_state atom size. */ + r300->fb_state.size = 2 + (8 * state->nr_cbufs); + + if (r300->cbzb_clear) + r300->fb_state.size += 10; + else if (state->zsbuf) { + r300->fb_state.size += 10; + if (has_hyperz) + r300->fb_state.size += r300->screen->caps.hiz_ram ? 8 : 4; } + + /* The size of the rest of atoms stays the same. */ } static void @@ -605,15 +715,12 @@ static void const struct pipe_framebuffer_state* state) { struct r300_context* r300 = r300_context(pipe); + struct r300_aa_state *aa = (struct r300_aa_state*)r300->aa_state.state; struct pipe_framebuffer_state *old_state = r300->fb_state.state; - unsigned max_width, max_height; + boolean has_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ); + unsigned max_width, max_height, i; uint32_t zbuffer_bpp = 0; - - if (state->nr_cbufs > 4) { - fprintf(stderr, "r300: Implementation error: Too many MRTs in %s, " - "refusing to bind framebuffer state!\n", __FUNCTION__); - return; - } + int blocksize; if (r300->screen->caps.is_r500) { max_width = max_height = 4096; @@ -629,12 +736,6 @@ static void return; } - if (r300->draw) { - draw_flush(r300->draw); - } - - r300->fb_state.dirty = TRUE; - /* If nr_cbufs is changed from zero to non-zero or vice versa... */ if (!!old_state->nr_cbufs != !!state->nr_cbufs) { r300->blend_state.dirty = TRUE; @@ -644,27 +745,91 @@ static void r300->dsa_state.dirty = TRUE; } - r300_fb_update_tiling_flags(r300, r300->fb_state.state, state); + /* The tiling flags are dependent on the surface miplevel, unfortunately. */ + r300_fb_set_tiling_flags(r300, state); - memcpy(r300->fb_state.state, state, sizeof(struct pipe_framebuffer_state)); + util_copy_framebuffer_state(r300->fb_state.state, state); - r300->fb_state.size = (10 * state->nr_cbufs) + (2 * (4 - state->nr_cbufs)) + - (state->zsbuf ? 10 : 0) + 11; + r300_mark_fb_state_dirty(r300, R300_CHANGED_FB_STATE); - /* Polygon offset depends on the zbuffer bit depth. */ - if (state->zsbuf && r300->polygon_offset_enabled) { - switch (util_format_get_blocksize(state->zsbuf->texture->format)) { - case 2: - zbuffer_bpp = 16; - break; - case 4: - zbuffer_bpp = 24; - break; + r300->z_compression = false; + + if (state->zsbuf) { + blocksize = util_format_get_blocksize(state->zsbuf->texture->format); + switch (blocksize) { + case 2: + zbuffer_bpp = 16; + break; + case 4: + zbuffer_bpp = 24; + break; + } + if (has_hyperz) { + struct r300_surface *zs_surf = r300_surface(state->zsbuf); + struct r300_texture *tex; + int compress = r300->screen->caps.is_rv350 ? RV350_Z_COMPRESS_88 : R300_Z_COMPRESS_44; + int level = zs_surf->base.level; + + tex = r300_texture(zs_surf->base.texture); + + /* work out whether we can support hiz on this buffer */ + r300_hiz_alloc_block(r300, zs_surf); + + /* work out whether we can support zmask features on this buffer */ + r300_zmask_alloc_block(r300, zs_surf, compress); + + if (tex->zmask_mem[level]) { + /* compression causes hangs on 16-bit */ + if (zbuffer_bpp == 24) + r300->z_compression = compress; + } + DBG(r300, DBG_HYPERZ, + "hyper-z features: hiz: %d @ %08x z-compression: %d z-fastfill: %d @ %08x\n", tex->hiz_mem[level] ? 1 : 0, + tex->hiz_mem[level] ? tex->hiz_mem[level]->ofs : 0xdeadbeef, + r300->z_compression, tex->zmask_mem[level] ? 1 : 0, + tex->zmask_mem[level] ? tex->zmask_mem[level]->ofs : 0xdeadbeef); } + /* Polygon offset depends on the zbuffer bit depth. */ if (r300->zbuffer_bpp != zbuffer_bpp) { r300->zbuffer_bpp = zbuffer_bpp; - r300->rs_state.dirty = TRUE; + + if (r300->polygon_offset_enabled) + r300->rs_state.dirty = TRUE; + } + } + + /* Set up AA config. */ + if (r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0)) { + if (state->nr_cbufs && state->cbufs[0]->texture->nr_samples > 1) { + aa->aa_config = R300_GB_AA_CONFIG_AA_ENABLE; + + switch (state->cbufs[0]->texture->nr_samples) { + case 2: + aa->aa_config |= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_2; + break; + case 3: + aa->aa_config |= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_3; + break; + case 4: + aa->aa_config |= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_4; + break; + case 6: + aa->aa_config |= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_6; + break; + } + } else { + aa->aa_config = 0; + } + } + + if (DBG_ON(r300, DBG_FB)) { + fprintf(stderr, "r300: set_framebuffer_state:\n"); + for (i = 0; i < state->nr_cbufs; i++) { + r300_print_fb_surf_info(state->cbufs[i], i, "CB"); + } + if (state->zsbuf) { + r300_print_fb_surf_info(state->zsbuf, 0, "ZB"); } } } @@ -691,16 +856,18 @@ void r300_mark_fs_code_dirty(struct r300_context *r300) r300->fs.dirty = TRUE; r300->fs_rc_constant_state.dirty = TRUE; r300->fs_constants.dirty = TRUE; + r300->fs.size = fs->shader->cb_code_size; if (r300->screen->caps.is_r500) { - r300->fs.size = r500_get_fs_atom_size(r300); r300->fs_rc_constant_state.size = fs->shader->rc_state_count * 7; r300->fs_constants.size = fs->shader->externals_count * 4 + 3; } else { - r300->fs.size = r300_get_fs_atom_size(r300); r300->fs_rc_constant_state.size = fs->shader->rc_state_count * 5; r300->fs_constants.size = fs->shader->externals_count * 4 + 1; } + + ((struct r300_constant_buffer*)r300->fs_constants.state)->remap_table = + fs->shader->code.constants_remap_table; } /* Bind fragment shader state. */ @@ -719,10 +886,6 @@ static void r300_bind_fs_state(struct pipe_context* pipe, void* shader) r300_mark_fs_code_dirty(r300); r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */ - - if (r300->vs_state.state && r300_vertex_shader_setup_wpos(r300)) { - r300->vap_output_state.dirty = TRUE; - } } /* Delete fragment shader state. */ @@ -735,6 +898,7 @@ static void r300_delete_fs_state(struct pipe_context* pipe, void* shader) tmp = ptr; ptr = ptr->next; rc_constants_destroy(&tmp->code.constants); + FREE(tmp->cb_code); FREE(tmp); } FREE((void*)fs->state.tokens); @@ -758,91 +922,121 @@ static void* r300_create_rs_state(struct pipe_context* pipe, const struct pipe_rasterizer_state* state) { struct r300_rs_state* rs = CALLOC_STRUCT(r300_rs_state); - int i; - - /* Copy rasterizer state for Draw. */ + float psiz; + uint32_t vap_control_status; /* R300_VAP_CNTL_STATUS: 0x2140 */ + uint32_t point_size; /* R300_GA_POINT_SIZE: 0x421c */ + uint32_t point_minmax; /* R300_GA_POINT_MINMAX: 0x4230 */ + uint32_t line_control; /* R300_GA_LINE_CNTL: 0x4234 */ + uint32_t polygon_offset_enable; /* R300_SU_POLY_OFFSET_ENABLE: 0x42b4 */ + uint32_t cull_mode; /* R300_SU_CULL_MODE: 0x42b8 */ + uint32_t line_stipple_config; /* R300_GA_LINE_STIPPLE_CONFIG: 0x4328 */ + uint32_t line_stipple_value; /* R300_GA_LINE_STIPPLE_VALUE: 0x4260 */ + uint32_t polygon_mode; /* R300_GA_POLY_MODE: 0x4288 */ + uint32_t clip_rule; /* R300_SC_CLIP_RULE: 0x43D0 */ + + /* Point sprites texture coordinates, 0: lower left, 1: upper right */ + float point_texcoord_left = 0; /* R300_GA_POINT_S0: 0x4200 */ + float point_texcoord_bottom = 0;/* R300_GA_POINT_T0: 0x4204 */ + float point_texcoord_right = 1; /* R300_GA_POINT_S1: 0x4208 */ + float point_texcoord_top = 0; /* R300_GA_POINT_T1: 0x420c */ + CB_LOCALS; + + /* Copy rasterizer state. */ rs->rs = *state; + rs->rs_draw = *state; + + rs->rs.sprite_coord_enable = state->point_quad_rasterization * + state->sprite_coord_enable; + + /* Override some states for Draw. */ + rs->rs_draw.sprite_coord_enable = 0; /* We can do this in HW. */ #ifdef PIPE_ARCH_LITTLE_ENDIAN - rs->vap_control_status = R300_VC_NO_SWAP; + vap_control_status = R300_VC_NO_SWAP; #else - rs->vap_control_status = R300_VC_32BIT_SWAP; + vap_control_status = R300_VC_32BIT_SWAP; #endif /* If no TCL engine is present, turn off the HW TCL. */ if (!r300_screen(pipe->screen)->caps.has_tcl) { - rs->vap_control_status |= R300_VAP_TCL_BYPASS; + vap_control_status |= R300_VAP_TCL_BYPASS; } - rs->point_size = pack_float_16_6x(state->point_size) | + /* Point size width and height. */ + point_size = + pack_float_16_6x(state->point_size) | (pack_float_16_6x(state->point_size) << R300_POINTSIZE_X_SHIFT); - rs->line_control = pack_float_16_6x(state->line_width) | + /* Point size clamping. */ + if (state->point_size_per_vertex) { + /* Per-vertex point size. + * Clamp to [0, max FB size] */ + psiz = pipe->screen->get_paramf(pipe->screen, + PIPE_CAP_MAX_POINT_WIDTH); + point_minmax = + pack_float_16_6x(psiz) << R300_GA_POINT_MINMAX_MAX_SHIFT; + } else { + /* We cannot disable the point-size vertex output, + * so clamp it. */ + psiz = state->point_size; + point_minmax = + (pack_float_16_6x(psiz) << R300_GA_POINT_MINMAX_MIN_SHIFT) | + (pack_float_16_6x(psiz) << R300_GA_POINT_MINMAX_MAX_SHIFT); + } + + /* Line control. */ + line_control = pack_float_16_6x(state->line_width) | R300_GA_LINE_CNTL_END_TYPE_COMP; /* Enable polygon mode */ - if (state->fill_cw != PIPE_POLYGON_MODE_FILL || - state->fill_ccw != PIPE_POLYGON_MODE_FILL) { - rs->polygon_mode = R300_GA_POLY_MODE_DUAL; + polygon_mode = 0; + if (state->fill_front != PIPE_POLYGON_MODE_FILL || + state->fill_back != PIPE_POLYGON_MODE_FILL) { + polygon_mode = R300_GA_POLY_MODE_DUAL; } - /* Radeons don't think in "CW/CCW", they think in "front/back". */ - if (state->front_winding == PIPE_WINDING_CW) { - rs->cull_mode = R300_FRONT_FACE_CW; - - /* Polygon offset */ - if (state->offset_cw) { - rs->polygon_offset_enable |= R300_FRONT_ENABLE; - } - if (state->offset_ccw) { - rs->polygon_offset_enable |= R300_BACK_ENABLE; - } - - /* Polygon mode */ - if (rs->polygon_mode) { - rs->polygon_mode |= - r300_translate_polygon_mode_front(state->fill_cw); - rs->polygon_mode |= - r300_translate_polygon_mode_back(state->fill_ccw); - } - } else { - rs->cull_mode = R300_FRONT_FACE_CCW; - - /* Polygon offset */ - if (state->offset_ccw) { - rs->polygon_offset_enable |= R300_FRONT_ENABLE; - } - if (state->offset_cw) { - rs->polygon_offset_enable |= R300_BACK_ENABLE; - } + /* Front face */ + if (state->front_ccw) + cull_mode = R300_FRONT_FACE_CCW; + else + cull_mode = R300_FRONT_FACE_CW; - /* Polygon mode */ - if (rs->polygon_mode) { - rs->polygon_mode |= - r300_translate_polygon_mode_front(state->fill_ccw); - rs->polygon_mode |= - r300_translate_polygon_mode_back(state->fill_cw); - } + /* Polygon offset */ + polygon_offset_enable = 0; + if (util_get_offset(state, state->fill_front)) { + polygon_offset_enable |= R300_FRONT_ENABLE; } - if (state->front_winding & state->cull_mode) { - rs->cull_mode |= R300_CULL_FRONT; + if (util_get_offset(state, state->fill_back)) { + polygon_offset_enable |= R300_BACK_ENABLE; } - if (~(state->front_winding) & state->cull_mode) { - rs->cull_mode |= R300_CULL_BACK; + + rs->polygon_offset_enable = polygon_offset_enable != 0; + + /* Polygon mode */ + if (polygon_mode) { + polygon_mode |= + r300_translate_polygon_mode_front(state->fill_front); + polygon_mode |= + r300_translate_polygon_mode_back(state->fill_back); } - if (rs->polygon_offset_enable) { - rs->depth_offset = state->offset_units; - rs->depth_scale = state->offset_scale; + if (state->cull_face & PIPE_FACE_FRONT) { + cull_mode |= R300_CULL_FRONT; + } + if (state->cull_face & PIPE_FACE_BACK) { + cull_mode |= R300_CULL_BACK; } if (state->line_stipple_enable) { - rs->line_stipple_config = + line_stipple_config = R300_GA_LINE_STIPPLE_CONFIG_LINE_RESET_LINE | (fui((float)state->line_stipple_factor) & R300_GA_LINE_STIPPLE_CONFIG_STIPPLE_SCALE_MASK); /* XXX this might need to be scaled up */ - rs->line_stipple_value = state->line_stipple_pattern; + line_stipple_value = state->line_stipple_pattern; + } else { + line_stipple_config = 0; + line_stipple_value = 0; } if (state->flatshade) { @@ -851,35 +1045,68 @@ static void* r300_create_rs_state(struct pipe_context* pipe, rs->color_control = R300_SHADE_MODEL_SMOOTH; } - rs->clip_rule = state->scissor ? 0xAAAA : 0xFFFF; + clip_rule = state->scissor ? 0xAAAA : 0xFFFF; - /* XXX Disable point sprites until we know what's wrong with them. */ - rs->rs.sprite_coord_enable = 0; - - /* Point sprites */ + /* Point sprites coord mode */ if (rs->rs.sprite_coord_enable) { - rs->stuffing_enable = R300_GB_POINT_STUFF_ENABLE; - for (i = 0; i < 8; i++) { - if (state->sprite_coord_enable & (1 << i)) - rs->stuffing_enable |= - R300_GB_TEX_STR << (R300_GB_TEX0_SOURCE_SHIFT + (i*2)); - } - - rs->point_texcoord_left = 0.0f; - rs->point_texcoord_right = 1.0f; - switch (state->sprite_coord_mode) { case PIPE_SPRITE_COORD_UPPER_LEFT: - rs->point_texcoord_top = 0.0f; - rs->point_texcoord_bottom = 1.0f; + point_texcoord_top = 0.0f; + point_texcoord_bottom = 1.0f; break; case PIPE_SPRITE_COORD_LOWER_LEFT: - rs->point_texcoord_top = 1.0f; - rs->point_texcoord_bottom = 0.0f; + point_texcoord_top = 1.0f; + point_texcoord_bottom = 0.0f; break; } } + /* Build the main command buffer. */ + BEGIN_CB(rs->cb_main, RS_STATE_MAIN_SIZE); + OUT_CB_REG(R300_VAP_CNTL_STATUS, vap_control_status); + OUT_CB_REG(R300_GA_POINT_SIZE, point_size); + OUT_CB_REG_SEQ(R300_GA_POINT_MINMAX, 2); + OUT_CB(point_minmax); + OUT_CB(line_control); + OUT_CB_REG_SEQ(R300_SU_POLY_OFFSET_ENABLE, 2); + OUT_CB(polygon_offset_enable); + rs->cull_mode_index = 9; + OUT_CB(cull_mode); + OUT_CB_REG(R300_GA_LINE_STIPPLE_CONFIG, line_stipple_config); + OUT_CB_REG(R300_GA_LINE_STIPPLE_VALUE, line_stipple_value); + OUT_CB_REG(R300_GA_POLY_MODE, polygon_mode); + OUT_CB_REG(R300_SC_CLIP_RULE, clip_rule); + OUT_CB_REG_SEQ(R300_GA_POINT_S0, 4); + OUT_CB_32F(point_texcoord_left); + OUT_CB_32F(point_texcoord_bottom); + OUT_CB_32F(point_texcoord_right); + OUT_CB_32F(point_texcoord_top); + END_CB; + + /* Build the two command buffers for polygon offset setup. */ + if (polygon_offset_enable) { + float scale = state->offset_scale * 12; + float offset = state->offset_units * 4; + + BEGIN_CB(rs->cb_poly_offset_zb16, 5); + OUT_CB_REG_SEQ(R300_SU_POLY_OFFSET_FRONT_SCALE, 4); + OUT_CB_32F(scale); + OUT_CB_32F(offset); + OUT_CB_32F(scale); + OUT_CB_32F(offset); + END_CB; + + offset = state->offset_units * 2; + + BEGIN_CB(rs->cb_poly_offset_zb24, 5); + OUT_CB_REG_SEQ(R300_SU_POLY_OFFSET_FRONT_SCALE, 4); + OUT_CB_32F(scale); + OUT_CB_32F(offset); + OUT_CB_32F(scale); + OUT_CB_32F(offset); + END_CB; + } + return (void*)rs; } @@ -889,24 +1116,27 @@ static void r300_bind_rs_state(struct pipe_context* pipe, void* state) struct r300_context* r300 = r300_context(pipe); struct r300_rs_state* rs = (struct r300_rs_state*)state; int last_sprite_coord_enable = r300->sprite_coord_enable; + boolean last_two_sided_color = r300->two_sided_color; - if (r300->draw) { - draw_flush(r300->draw); - draw_set_rasterizer_state(r300->draw, &rs->rs, state); + if (r300->draw && rs) { + draw_set_rasterizer_state(r300->draw, &rs->rs_draw, state); } if (rs) { - r300->polygon_offset_enabled = rs->rs.offset_cw || rs->rs.offset_ccw; + r300->polygon_offset_enabled = rs->polygon_offset_enable; r300->sprite_coord_enable = rs->rs.sprite_coord_enable; + r300->two_sided_color = rs->rs.light_twoside; } else { r300->polygon_offset_enabled = FALSE; r300->sprite_coord_enable = 0; + r300->two_sided_color = FALSE; } UPDATE_STATE(state, r300->rs_state); - r300->rs_state.size = 26 + (r300->polygon_offset_enabled ? 5 : 0); + r300->rs_state.size = RS_STATE_MAIN_SIZE + (r300->polygon_offset_enabled ? 5 : 0); - if (last_sprite_coord_enable != r300->sprite_coord_enable) { + if (last_sprite_coord_enable != r300->sprite_coord_enable || + last_two_sided_color != r300->two_sided_color) { r300->rs_block_state.dirty = TRUE; } } @@ -925,14 +1155,37 @@ static void* struct r300_sampler_state* sampler = CALLOC_STRUCT(r300_sampler_state); boolean is_r500 = r300->screen->caps.is_r500; int lod_bias; - union util_color uc; sampler->state = *state; + /* r300 doesn't handle CLAMP and MIRROR_CLAMP correctly when either MAG + * or MIN filter is NEAREST. Since texwrap produces same results + * for CLAMP and CLAMP_TO_EDGE, we use them instead. */ + if (sampler->state.min_img_filter == PIPE_TEX_FILTER_NEAREST || + sampler->state.mag_img_filter == PIPE_TEX_FILTER_NEAREST) { + /* Wrap S. */ + if (sampler->state.wrap_s == PIPE_TEX_WRAP_CLAMP) + sampler->state.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE; + else if (sampler->state.wrap_s == PIPE_TEX_WRAP_MIRROR_CLAMP) + sampler->state.wrap_s = PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE; + + /* Wrap T. */ + if (sampler->state.wrap_t == PIPE_TEX_WRAP_CLAMP) + sampler->state.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE; + else if (sampler->state.wrap_t == PIPE_TEX_WRAP_MIRROR_CLAMP) + sampler->state.wrap_t = PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE; + + /* Wrap R. */ + if (sampler->state.wrap_r == PIPE_TEX_WRAP_CLAMP) + sampler->state.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE; + else if (sampler->state.wrap_r == PIPE_TEX_WRAP_MIRROR_CLAMP) + sampler->state.wrap_r = PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE; + } + sampler->filter0 |= - (r300_translate_wrap(state->wrap_s) << R300_TX_WRAP_S_SHIFT) | - (r300_translate_wrap(state->wrap_t) << R300_TX_WRAP_T_SHIFT) | - (r300_translate_wrap(state->wrap_r) << R300_TX_WRAP_R_SHIFT); + (r300_translate_wrap(sampler->state.wrap_s) << R300_TX_WRAP_S_SHIFT) | + (r300_translate_wrap(sampler->state.wrap_t) << R300_TX_WRAP_T_SHIFT) | + (r300_translate_wrap(sampler->state.wrap_r) << R300_TX_WRAP_R_SHIFT); sampler->filter0 |= r300_translate_tex_filters(state->min_img_filter, state->mag_img_filter, @@ -943,12 +1196,12 @@ static void* /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */ /* We must pass these to the merge function to clamp them properly. */ - sampler->min_lod = MAX2((unsigned)state->min_lod, 0); - sampler->max_lod = MAX2((unsigned)ceilf(state->max_lod), 0); + sampler->min_lod = (unsigned)MAX2(state->min_lod, 0); + sampler->max_lod = (unsigned)MAX2(ceilf(state->max_lod), 0); - lod_bias = CLAMP((int)(state->lod_bias * 32), -(1 << 9), (1 << 9) - 1); + lod_bias = CLAMP((int)(state->lod_bias * 32 + 1), -(1 << 9), (1 << 9) - 1); - sampler->filter1 |= lod_bias << R300_LOD_BIAS_SHIFT; + sampler->filter1 |= (lod_bias << R300_LOD_BIAS_SHIFT) & R300_LOD_BIAS_MASK; /* This is very high quality anisotropic filtering for R5xx. * It's good for benchmarking the performance of texturing but @@ -958,9 +1211,6 @@ static void* sampler->filter1 |= r500_anisotropy(state->max_anisotropy); } - util_pack_color(state->border_color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc); - sampler->border_color = uc.ui; - /* R500-specific fixups and optimizations */ if (r300->screen->caps.is_r500) { sampler->filter1 |= R500_BORDER_FIX; @@ -999,6 +1249,31 @@ static void r300_delete_sampler_state(struct pipe_context* pipe, void* state) FREE(state); } +static uint32_t r300_assign_texture_cache_region(unsigned index, unsigned num) +{ + /* This looks like a hack, but I believe it's suppose to work like + * that. To illustrate how this works, let's assume you have 5 textures. + * From docs, 5 and the successive numbers are: + * + * FOURTH_1 = 5 + * FOURTH_2 = 6 + * FOURTH_3 = 7 + * EIGHTH_0 = 8 + * EIGHTH_1 = 9 + * + * First 3 textures will get 3/4 of size of the cache, divived evenly + * between them. The last 1/4 of the cache must be divided between + * the last 2 textures, each will therefore get 1/8 of the cache. + * Why not just to use "5 + texture_index" ? + * + * This simple trick works for all "num" <= 16. + */ + if (num <= 1) + return R300_TX_CACHE(R300_TX_CACHE_WHOLE); + else + return R300_TX_CACHE(num + index); +} + static void r300_set_fragment_sampler_views(struct pipe_context* pipe, unsigned count, struct pipe_sampler_view** views) @@ -1007,7 +1282,7 @@ static void r300_set_fragment_sampler_views(struct pipe_context* pipe, struct r300_textures_state* state = (struct r300_textures_state*)r300->textures_state.state; struct r300_texture *texture; - unsigned i; + unsigned i, real_num_views = 0, view_index = 0; unsigned tex_units = r300->screen->caps.num_tex_units; boolean dirty_tex = FALSE; @@ -1015,6 +1290,12 @@ static void r300_set_fragment_sampler_views(struct pipe_context* pipe, return; } + /* Calculate the real number of views. */ + for (i = 0; i < count; i++) { + if (views[i]) + real_num_views++; + } + for (i = 0; i < count; i++) { if (&state->sampler_views[i]->base != views[i]) { pipe_sampler_view_reference( @@ -1031,9 +1312,13 @@ static void r300_set_fragment_sampler_views(struct pipe_context* pipe, /* Set the texrect factor in the fragment shader. * Needed for RECT and NPOT fallback. */ texture = r300_texture(views[i]->texture); - if (texture->uses_pitch) { + if (texture->desc.is_npot) { r300->fs_rc_constant_state.dirty = TRUE; } + + state->sampler_views[i]->texcache_region = + r300_assign_texture_cache_region(view_index, real_num_views); + view_index++; } } @@ -1061,7 +1346,7 @@ r300_create_sampler_view(struct pipe_context *pipe, { struct r300_sampler_view *view = CALLOC_STRUCT(r300_sampler_view); struct r300_texture *tex = r300_texture(texture); - unsigned char swizzle[4]; + boolean is_r500 = r300_screen(pipe->screen)->caps.is_r500; if (view) { view->base = *templ; @@ -1070,17 +1355,16 @@ r300_create_sampler_view(struct pipe_context *pipe, view->base.texture = NULL; pipe_resource_reference(&view->base.texture, texture); - swizzle[0] = templ->swizzle_r; - swizzle[1] = templ->swizzle_g; - swizzle[2] = templ->swizzle_b; - swizzle[3] = templ->swizzle_a; + view->swizzle[0] = templ->swizzle_r; + view->swizzle[1] = templ->swizzle_g; + view->swizzle[2] = templ->swizzle_b; + view->swizzle[3] = templ->swizzle_a; - /* XXX Enable swizzles when they become supported. Now we get RGBA - * everywhere. And do testing! */ view->format = tex->tx_format; view->format.format1 |= r300_translate_texformat(templ->format, - 0); /*swizzle);*/ - if (r300_screen(pipe->screen)->caps.is_r500) { + view->swizzle, + is_r500); + if (is_r500) { view->format.format2 |= r500_tx_format_msb_bit(templ->format); } } @@ -1116,6 +1400,12 @@ static void r300_set_viewport_state(struct pipe_context* pipe, r300->viewport = *state; + if (r300->draw) { + draw_set_viewport_state(r300->draw, state); + viewport->vte_control = R300_VTX_XY_FMT | R300_VTX_Z_FMT; + return; + } + /* Do the transform in HW. */ viewport->vte_control = R300_VTX_W0_FMT; @@ -1165,48 +1455,60 @@ static void r300_set_vertex_buffers(struct pipe_context* pipe, return; } - /* Check if the stride is aligned to the size of DWORD. */ - for (i = 0; i < count; i++) { - if (buffers[i].buffer) { - if (buffers[i].stride % 4 != 0) { - // XXX Shouldn't we align the buffer? - fprintf(stderr, "r300: set_vertex_buffers: " - "Unaligned buffer stride %i isn't supported.\n", - buffers[i].stride); - abort(); + if (r300->screen->caps.has_tcl) { + /* HW TCL. */ + r300->incompatible_vb_layout = FALSE; + + /* Check if the strides and offsets are aligned to the size of DWORD. */ + for (i = 0; i < count; i++) { + if (buffers[i].buffer) { + if (buffers[i].stride % 4 != 0 || + buffers[i].buffer_offset % 4 != 0) { + r300->incompatible_vb_layout = TRUE; + break; + } } } - } - for (i = 0; i < count; i++) { - /* Why, yes, I AM casting away constness. How did you know? */ - vbo = (struct pipe_vertex_buffer*)&buffers[i]; + for (i = 0; i < count; i++) { + /* Why, yes, I AM casting away constness. How did you know? */ + vbo = (struct pipe_vertex_buffer*)&buffers[i]; - /* Reference our buffer. */ - pipe_resource_reference(&r300->vertex_buffer[i].buffer, vbo->buffer); + /* Skip NULL buffers */ + if (!buffers[i].buffer) { + continue; + } - /* Skip NULL buffers */ - if (!buffers[i].buffer) { - continue; - } + if (r300_buffer_is_user_buffer(vbo->buffer)) { + any_user_buffer = TRUE; + } - if (r300_buffer_is_user_buffer(vbo->buffer)) { - any_user_buffer = TRUE; - } + if (vbo->max_index == ~0) { + /* if no VBO stride then only one vertex value so max index is 1 */ + /* should think about converting to VS constants like svga does */ + if (!vbo->stride) + vbo->max_index = 1; + else + vbo->max_index = + (vbo->buffer->width0 - vbo->buffer_offset) / vbo->stride; + } - if (vbo->max_index == ~0) { - /* if no VBO stride then only one vertex value so max index is 1 */ - /* should think about converting to VS constants like svga does */ - if (!vbo->stride) - vbo->max_index = 1; - else - vbo->max_index = - (vbo->buffer->width0 - vbo->buffer_offset) / vbo->stride; + max_index = MIN2(vbo->max_index, max_index); } - max_index = MIN2(vbo->max_index, max_index); + r300->any_user_vbs = any_user_buffer; + r300->vertex_buffer_max_index = max_index; + + } else { + /* SW TCL. */ + draw_set_vertex_buffers(r300->draw, count, buffers); } + /* Common code. */ + for (i = 0; i < count; i++) { + /* Reference our buffer. */ + pipe_resource_reference(&r300->vertex_buffer[i].buffer, buffers[i].buffer); + } for (; i < r300->vertex_buffer_count; i++) { /* Dereference any old buffers. */ pipe_resource_reference(&r300->vertex_buffer[i].buffer, NULL); @@ -1214,18 +1516,32 @@ static void r300_set_vertex_buffers(struct pipe_context* pipe, memcpy(r300->vertex_buffer, buffers, sizeof(struct pipe_vertex_buffer) * count); - r300->vertex_buffer_count = count; - r300->vertex_buffer_max_index = max_index; - r300->any_user_vbs = any_user_buffer; +} - if (r300->draw) { - draw_flush(r300->draw); - draw_set_vertex_buffers(r300->draw, count, buffers); +static void r300_set_index_buffer(struct pipe_context* pipe, + const struct pipe_index_buffer *ib) +{ + struct r300_context* r300 = r300_context(pipe); + + if (ib) { + pipe_resource_reference(&r300->index_buffer.buffer, ib->buffer); + memcpy(&r300->index_buffer, ib, sizeof(r300->index_buffer)); + } + else { + pipe_resource_reference(&r300->index_buffer.buffer, NULL); + memset(&r300->index_buffer, 0, sizeof(r300->index_buffer)); + } + + if (r300->screen->caps.has_tcl) { + /* TODO make this more like a state */ + } + else { + draw_set_index_buffer(r300->draw, ib); } } -/* Update the PSC tables. */ +/* Initialize the PSC tables. */ static void r300_vertex_psc(struct r300_vertex_element_state *velems) { struct r300_vertex_stream_state *vstream = &velems->vertex_stream; @@ -1243,10 +1559,17 @@ static void r300_vertex_psc(struct r300_vertex_element_state *velems) * so PSC should just route stuff based on the vertex elements, * and not on attrib information. */ for (i = 0; i < velems->count; i++) { - format = velems->velem[i].src_format; + format = velems->hw_format[i]; + + type = r300_translate_vertex_data_type(format); + if (type == R300_INVALID_FORMAT) { + fprintf(stderr, "r300: Bad vertex format %s.\n", + util_format_short_name(format)); + assert(0); + abort(); + } - type = r300_translate_vertex_data_type(format) | - (i << R300_DST_VEC_LOC_SHIFT); + type |= i << R300_DST_VEC_LOC_SHIFT; swizzle = r300_translate_vertex_data_swizzle(format); if (i & 1) { @@ -1268,12 +1591,15 @@ static void r300_vertex_psc(struct r300_vertex_element_state *velems) vstream->count = (i >> 1) + 1; } +#define FORMAT_REPLACE(what, withwhat) \ + case PIPE_FORMAT_##what: *format = PIPE_FORMAT_##withwhat; break + static void* r300_create_vertex_elements_state(struct pipe_context* pipe, unsigned count, const struct pipe_vertex_element* attribs) { struct r300_vertex_element_state *velems; - unsigned i, size; + unsigned i; enum pipe_format *format; assert(count <= PIPE_MAX_ATTRIBS); @@ -1283,88 +1609,69 @@ static void* r300_create_vertex_elements_state(struct pipe_context* pipe, memcpy(velems->velem, attribs, sizeof(struct pipe_vertex_element) * count); if (r300_screen(pipe->screen)->caps.has_tcl) { - r300_vertex_psc(velems); - - /* Check if the format is aligned to the size of DWORD. - * We only care about the blocksizes of the formats since - * swizzles are already set up. */ + /* Set the best hw format in case the original format is not + * supported by hw. */ for (i = 0; i < count; i++) { - format = &velems->velem[i].src_format; + velems->hw_format[i] = velems->velem[i].src_format; + format = &velems->hw_format[i]; - /* Replace some formats with their aligned counterparts, - * this is OK because we check for aligned strides too. */ + /* This is basically the list of unsupported formats. + * For now we don't care about the alignment, that's going to + * be sorted out after the PSC setup. */ switch (*format) { - /* Align to RGBA8. */ - case PIPE_FORMAT_R8_UNORM: - case PIPE_FORMAT_R8G8_UNORM: - case PIPE_FORMAT_R8G8B8_UNORM: - *format = PIPE_FORMAT_R8G8B8A8_UNORM; - continue; - case PIPE_FORMAT_R8_SNORM: - case PIPE_FORMAT_R8G8_SNORM: - case PIPE_FORMAT_R8G8B8_SNORM: - *format = PIPE_FORMAT_R8G8B8A8_SNORM; - continue; - case PIPE_FORMAT_R8_USCALED: - case PIPE_FORMAT_R8G8_USCALED: - case PIPE_FORMAT_R8G8B8_USCALED: - *format = PIPE_FORMAT_R8G8B8A8_USCALED; - continue; - case PIPE_FORMAT_R8_SSCALED: - case PIPE_FORMAT_R8G8_SSCALED: - case PIPE_FORMAT_R8G8B8_SSCALED: - *format = PIPE_FORMAT_R8G8B8A8_SSCALED; - continue; - - /* Align to RG16. */ - case PIPE_FORMAT_R16_UNORM: - *format = PIPE_FORMAT_R16G16_UNORM; - continue; - case PIPE_FORMAT_R16_SNORM: - *format = PIPE_FORMAT_R16G16_SNORM; - continue; - case PIPE_FORMAT_R16_USCALED: - *format = PIPE_FORMAT_R16G16_USCALED; - continue; - case PIPE_FORMAT_R16_SSCALED: - *format = PIPE_FORMAT_R16G16_SSCALED; - continue; - case PIPE_FORMAT_R16_FLOAT: - *format = PIPE_FORMAT_R16G16_FLOAT; - continue; - - /* Align to RGBA16. */ - case PIPE_FORMAT_R16G16B16_UNORM: - *format = PIPE_FORMAT_R16G16B16A16_UNORM; - continue; - case PIPE_FORMAT_R16G16B16_SNORM: - *format = PIPE_FORMAT_R16G16B16A16_SNORM; - continue; - case PIPE_FORMAT_R16G16B16_USCALED: - *format = PIPE_FORMAT_R16G16B16A16_USCALED; - continue; - case PIPE_FORMAT_R16G16B16_SSCALED: - *format = PIPE_FORMAT_R16G16B16A16_SSCALED; - continue; - case PIPE_FORMAT_R16G16B16_FLOAT: - *format = PIPE_FORMAT_R16G16B16A16_FLOAT; - continue; + FORMAT_REPLACE(R64_FLOAT, R32_FLOAT); + FORMAT_REPLACE(R64G64_FLOAT, R32G32_FLOAT); + FORMAT_REPLACE(R64G64B64_FLOAT, R32G32B32_FLOAT); + FORMAT_REPLACE(R64G64B64A64_FLOAT, R32G32B32A32_FLOAT); + + FORMAT_REPLACE(R32_UNORM, R32_FLOAT); + FORMAT_REPLACE(R32G32_UNORM, R32G32_FLOAT); + FORMAT_REPLACE(R32G32B32_UNORM, R32G32B32_FLOAT); + FORMAT_REPLACE(R32G32B32A32_UNORM, R32G32B32A32_FLOAT); + + FORMAT_REPLACE(R32_USCALED, R32_FLOAT); + FORMAT_REPLACE(R32G32_USCALED, R32G32_FLOAT); + FORMAT_REPLACE(R32G32B32_USCALED, R32G32B32_FLOAT); + FORMAT_REPLACE(R32G32B32A32_USCALED,R32G32B32A32_FLOAT); + + FORMAT_REPLACE(R32_SNORM, R32_FLOAT); + FORMAT_REPLACE(R32G32_SNORM, R32G32_FLOAT); + FORMAT_REPLACE(R32G32B32_SNORM, R32G32B32_FLOAT); + FORMAT_REPLACE(R32G32B32A32_SNORM, R32G32B32A32_FLOAT); + + FORMAT_REPLACE(R32_SSCALED, R32_FLOAT); + FORMAT_REPLACE(R32G32_SSCALED, R32G32_FLOAT); + FORMAT_REPLACE(R32G32B32_SSCALED, R32G32B32_FLOAT); + FORMAT_REPLACE(R32G32B32A32_SSCALED,R32G32B32A32_FLOAT); + + FORMAT_REPLACE(R32_FIXED, R32_FLOAT); + FORMAT_REPLACE(R32G32_FIXED, R32G32_FLOAT); + FORMAT_REPLACE(R32G32B32_FIXED, R32G32B32_FLOAT); + FORMAT_REPLACE(R32G32B32A32_FIXED, R32G32B32A32_FLOAT); default:; } - size = util_format_get_blocksize(*format); - - if (size % 4 != 0) { - /* XXX Shouldn't we align the format? */ - fprintf(stderr, "r300_create_vertex_elements_state: " - "Unaligned format %s:%i isn't supported\n", - util_format_name(*format), size); - assert(0); - abort(); - } + velems->incompatible_layout = + velems->incompatible_layout || + velems->velem[i].src_format != velems->hw_format[i] || + velems->velem[i].src_offset % 4 != 0; } + /* Now setup PSC. + * The unused components will be replaced by (..., 0, 1). */ + r300_vertex_psc(velems); + + /* Align the formats to the size of DWORD. + * We only care about the blocksizes of the formats since + * swizzles are already set up. + * Also compute the vertex size. */ + for (i = 0; i < count; i++) { + /* This is OK because we check for aligned strides too. */ + velems->hw_format_size[i] = + align(util_format_get_blocksize(velems->hw_format[i]), 4); + velems->vertex_size_dwords += velems->hw_format_size[i] / 4; + } } } return velems; @@ -1383,8 +1690,8 @@ static void r300_bind_vertex_elements_state(struct pipe_context *pipe, r300->velems = velems; if (r300->draw) { - draw_flush(r300->draw); draw_set_vertex_elements(r300->draw, velems->count, velems->velem); + return; } UPDATE_STATE(&velems->vertex_stream, r300->vertex_stream_state); @@ -1400,7 +1707,6 @@ static void* r300_create_vs_state(struct pipe_context* pipe, const struct pipe_shader_state* shader) { struct r300_context* r300 = r300_context(pipe); - struct r300_vertex_shader* vs = CALLOC_STRUCT(r300_vertex_shader); /* Copy state directly into shader. */ @@ -1408,9 +1714,10 @@ static void* r300_create_vs_state(struct pipe_context* pipe, vs->state.tokens = tgsi_dup_tokens(shader->tokens); if (r300->screen->caps.has_tcl) { - r300_translate_vertex_shader(r300, vs, vs->state.tokens); + r300_init_vs_outputs(vs); + r300_translate_vertex_shader(r300, vs); } else { - vs->draw_vs = draw_create_vertex_shader(r300->draw, shader); + r300_draw_init_vertex_shader(r300->draw, vs); } return vs; @@ -1430,22 +1737,16 @@ static void r300_bind_vs_state(struct pipe_context* pipe, void* shader) } r300->vs_state.state = vs; - // VS output mapping for HWTCL or stream mapping for SWTCL to the RS block - if (r300->fs.state) { - r300_vertex_shader_setup_wpos(r300); - } - memcpy(r300->vap_output_state.state, &vs->vap_out, - sizeof(struct r300_vap_output_state)); - r300->vap_output_state.dirty = TRUE; - /* The majority of the RS block bits is dependent on the vertex shader. */ r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */ if (r300->screen->caps.has_tcl) { + unsigned fc_op_dwords = r300->screen->caps.is_r500 ? 3 : 2; r300->vs_state.dirty = TRUE; r300->vs_state.size = vs->code.length + 9 + - (vs->immediates_count ? vs->immediates_count * 4 + 3 : 0); + (vs->immediates_count ? vs->immediates_count * 4 + 3 : 0) + + (vs->code.num_fc_ops ? vs->code.num_fc_ops * fc_op_dwords + 4 : 0); if (vs->externals_count) { r300->vs_constants.dirty = TRUE; @@ -1454,9 +1755,11 @@ static void r300_bind_vs_state(struct pipe_context* pipe, void* shader) r300->vs_constants.size = 0; } + ((struct r300_constant_buffer*)r300->vs_constants.state)->remap_table = + vs->code.constants_remap_table; + r300->pvs_flush.dirty = TRUE; } else { - draw_flush(r300->draw); draw_bind_vertex_shader(r300->draw, (struct draw_vertex_shader*)vs->draw_vs); } @@ -1469,6 +1772,8 @@ static void r300_delete_vs_state(struct pipe_context* pipe, void* shader) if (r300->screen->caps.has_tcl) { rc_constants_destroy(&vs->code.constants); + if (vs->code.constants_remap_table) + FREE(vs->code.constants_remap_table); } else { draw_delete_vertex_shader(r300->draw, (struct draw_vertex_shader*)vs->draw_vs); @@ -1484,22 +1789,14 @@ static void r300_set_constant_buffer(struct pipe_context *pipe, { struct r300_context* r300 = r300_context(pipe); struct r300_constant_buffer *cbuf; - struct pipe_transfer *tr; - void *mapped; - int max_size = 0; + uint32_t *mapped = r300_buffer(buf)->user_buffer; switch (shader) { case PIPE_SHADER_VERTEX: cbuf = (struct r300_constant_buffer*)r300->vs_constants.state; - max_size = 256; break; case PIPE_SHADER_FRAGMENT: cbuf = (struct r300_constant_buffer*)r300->fs_constants.state; - if (r300->screen->caps.is_r500) { - max_size = 256; - } else { - max_size = 32; - } break; default: assert(0); @@ -1507,26 +1804,16 @@ static void r300_set_constant_buffer(struct pipe_context *pipe, } if (buf == NULL || buf->width0 == 0 || - (mapped = pipe_buffer_map(pipe, buf, PIPE_TRANSFER_READ, &tr)) == NULL) - { - cbuf->count = 0; + (mapped = r300_buffer(buf)->constant_buffer) == NULL) { return; } - assert((buf->width0 % 4 * sizeof(float)) == 0); - - /* Check the size of the constant buffer. */ - /* XXX Subtract immediates and RC_STATE_* variables. */ - if (buf->width0 > (sizeof(float) * 4 * max_size)) { - fprintf(stderr, "r300: Max size of the constant buffer is " - "%i*4 floats.\n", max_size); - abort(); + if (shader == PIPE_SHADER_FRAGMENT || + (shader == PIPE_SHADER_VERTEX && r300->screen->caps.has_tcl)) { + assert((buf->width0 % (4 * sizeof(float))) == 0); + cbuf->ptr = mapped + index*4; } - memcpy(cbuf->constants, mapped, buf->width0); - cbuf->count = buf->width0 / (4 * sizeof(float)); - pipe_buffer_unmap(pipe, buf, tr); - if (shader == PIPE_SHADER_VERTEX) { if (r300->screen->caps.has_tcl) { if (r300->vs_constants.size) { @@ -1535,8 +1822,7 @@ static void r300_set_constant_buffer(struct pipe_context *pipe, r300->pvs_flush.dirty = TRUE; } else if (r300->draw) { draw_set_mapped_constant_buffer(r300->draw, PIPE_SHADER_VERTEX, - 0, cbuf->constants, - buf->width0); + 0, mapped, buf->width0); } } else if (shader == PIPE_SHADER_FRAGMENT) { r300->fs_constants.dirty = TRUE; @@ -1552,6 +1838,7 @@ void r300_init_state_functions(struct r300_context* r300) r300->context.set_blend_color = r300_set_blend_color; r300->context.set_clip_state = r300_set_clip_state; + r300->context.set_sample_mask = r300_set_sample_mask; r300->context.set_constant_buffer = r300_set_constant_buffer; @@ -1587,6 +1874,7 @@ void r300_init_state_functions(struct r300_context* r300) r300->context.set_viewport_state = r300_set_viewport_state; r300->context.set_vertex_buffers = r300_set_vertex_buffers; + r300->context.set_index_buffer = r300_set_index_buffer; r300->context.create_vertex_elements_state = r300_create_vertex_elements_state; r300->context.bind_vertex_elements_state = r300_bind_vertex_elements_state; diff --git a/src/gallium/drivers/r300/r300_state.h b/src/gallium/drivers/r300/r300_state.h deleted file mode 100644 index 1d557506cf3..00000000000 --- a/src/gallium/drivers/r300/r300_state.h +++ /dev/null @@ -1,30 +0,0 @@ -/* - * Copyright 2008 Marek Olšák <[email protected]> - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * on the rights to use, copy, modify, merge, publish, distribute, sub - * license, and/or sell copies of the Software, and to permit persons to whom - * the Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice (including the next - * paragraph) shall be included in all copies or substantial portions of the - * Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL - * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, - * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR - * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE - * USE OR OTHER DEALINGS IN THE SOFTWARE. */ - -#ifndef R300_STATE_H -#define R300_STATE_H - -struct r300_context; - -void r300_mark_fs_code_dirty(struct r300_context *r300); - -#endif /* R300_STATE_H */ diff --git a/src/gallium/drivers/r300/r300_state_derived.c b/src/gallium/drivers/r300/r300_state_derived.c index 10ea579be3b..904736ef06d 100644 --- a/src/gallium/drivers/r300/r300_state_derived.c +++ b/src/gallium/drivers/r300/r300_state_derived.c @@ -31,9 +31,9 @@ #include "r300_hyperz.h" #include "r300_screen.h" #include "r300_shader_semantics.h" -#include "r300_state.h" #include "r300_state_derived.h" #include "r300_state_inlines.h" +#include "r300_texture.h" #include "r300_vs.h" /* r300_state_derived: Various bits of state which are dependent upon @@ -43,6 +43,12 @@ enum r300_rs_swizzle { SWIZ_XYZW = 0, SWIZ_X001, SWIZ_XY01, + SWIZ_0001, +}; + +enum r300_rs_col_write_type { + WRITE_COLOR = 0, + WRITE_FACE }; static void r300_draw_emit_attrib(struct r300_context* r300, @@ -88,14 +94,21 @@ static void r300_draw_emit_all_attribs(struct r300_context* r300) } } - /* XXX Back-face colors. */ + /* Back-face colors. */ + for (i = 0; i < ATTR_COLOR_COUNT; i++) { + if (vs_outputs->bcolor[i] != ATTR_UNUSED) { + r300_draw_emit_attrib(r300, EMIT_4F, INTERP_LINEAR, + vs_outputs->bcolor[i]); + } + } /* Texture coordinates. */ /* Only 8 generic vertex attributes can be used. If there are more, * they won't be rasterized. */ gen_count = 0; for (i = 0; i < ATTR_GENERIC_COUNT && gen_count < 8; i++) { - if (vs_outputs->generic[i] != ATTR_UNUSED) { + if (vs_outputs->generic[i] != ATTR_UNUSED && + !(r300->sprite_coord_enable & (1 << i))) { r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE, vs_outputs->generic[i]); gen_count++; @@ -108,40 +121,54 @@ static void r300_draw_emit_all_attribs(struct r300_context* r300) vs_outputs->fog); gen_count++; } + + /* WPOS. */ + if (r300_fs(r300)->shader->inputs.wpos != ATTR_UNUSED && gen_count < 8) { + DBG(r300, DBG_SWTCL, "draw_emit_attrib: WPOS, index: %i\n", + vs_outputs->wpos); + r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE, + vs_outputs->wpos); + } } /* Update the PSC tables for SW TCL, using Draw. */ static void r300_swtcl_vertex_psc(struct r300_context *r300) { struct r300_vertex_stream_state *vstream = r300->vertex_stream_state.state; - struct r300_vertex_shader* vs = r300->vs_state.state; - struct vertex_info* vinfo = &r300->vertex_info; + struct vertex_info *vinfo = &r300->vertex_info; uint16_t type, swizzle; enum pipe_format format; unsigned i, attrib_count; - int* vs_output_tab = vs->stream_loc_notcl; + int* vs_output_tab = r300->stream_loc_notcl; - /* XXX hax */ memset(vstream, 0, sizeof(struct r300_vertex_stream_state)); /* For each Draw attribute, route it to the fragment shader according * to the vs_output_tab. */ attrib_count = vinfo->num_attribs; - DBG(r300, DBG_DRAW, "r300: attrib count: %d\n", attrib_count); + DBG(r300, DBG_SWTCL, "r300: attrib count: %d\n", attrib_count); for (i = 0; i < attrib_count; i++) { - DBG(r300, DBG_DRAW, "r300: attrib: offset %d, interp %d, size %d," - " vs_output_tab %d\n", vinfo->attrib[i].src_index, - vinfo->attrib[i].interp_mode, vinfo->attrib[i].emit, - vs_output_tab[i]); - - /* Make sure we have a proper destination for our attribute. */ - assert(vs_output_tab[i] != -1); + if (vs_output_tab[i] == -1) { + assert(0); + abort(); + } format = draw_translate_vinfo_format(vinfo->attrib[i].emit); + DBG(r300, DBG_SWTCL, + "r300: swtcl_vertex_psc [%i] <- %s\n", + vs_output_tab[i], util_format_short_name(format)); + /* Obtain the type of data in this attribute. */ - type = r300_translate_vertex_data_type(format) | - vs_output_tab[i] << R300_DST_VEC_LOC_SHIFT; + type = r300_translate_vertex_data_type(format); + if (type == R300_INVALID_FORMAT) { + fprintf(stderr, "r300: Bad vertex format %s.\n", + util_format_short_name(format)); + assert(0); + abort(); + } + + type |= vs_output_tab[i] << R300_DST_VEC_LOC_SHIFT; /* Obtain the swizzle for this attribute. Note that the default * swizzle in the hardware is not XYZW! */ @@ -170,10 +197,10 @@ static void r300_swtcl_vertex_psc(struct r300_context *r300) } static void r300_rs_col(struct r300_rs_block* rs, int id, int ptr, - boolean swizzle_0001) + enum r300_rs_swizzle swiz) { rs->ip[id] |= R300_RS_COL_PTR(ptr); - if (swizzle_0001) { + if (swiz == SWIZ_0001) { rs->ip[id] |= R300_RS_COL_FMT(R300_RS_COL_FMT_0001); } else { rs->ip[id] |= R300_RS_COL_FMT(R300_RS_COL_FMT_RGBA); @@ -181,8 +208,10 @@ static void r300_rs_col(struct r300_rs_block* rs, int id, int ptr, rs->inst[id] |= R300_RS_INST_COL_ID(id); } -static void r300_rs_col_write(struct r300_rs_block* rs, int id, int fp_offset) +static void r300_rs_col_write(struct r300_rs_block* rs, int id, int fp_offset, + enum r300_rs_col_write_type type) { + assert(type == WRITE_COLOR); rs->inst[id] |= R300_RS_INST_COL_CN_WRITE | R300_RS_INST_COL_ADDR(fp_offset); } @@ -191,19 +220,19 @@ static void r300_rs_tex(struct r300_rs_block* rs, int id, int ptr, enum r300_rs_swizzle swiz) { if (swiz == SWIZ_X001) { - rs->ip[id] |= R300_RS_TEX_PTR(ptr*4) | + rs->ip[id] |= R300_RS_TEX_PTR(ptr) | R300_RS_SEL_S(R300_RS_SEL_C0) | R300_RS_SEL_T(R300_RS_SEL_K0) | R300_RS_SEL_R(R300_RS_SEL_K0) | R300_RS_SEL_Q(R300_RS_SEL_K1); } else if (swiz == SWIZ_XY01) { - rs->ip[id] |= R300_RS_TEX_PTR(ptr*4) | + rs->ip[id] |= R300_RS_TEX_PTR(ptr) | R300_RS_SEL_S(R300_RS_SEL_C0) | R300_RS_SEL_T(R300_RS_SEL_C1) | R300_RS_SEL_R(R300_RS_SEL_K0) | R300_RS_SEL_Q(R300_RS_SEL_K1); } else { - rs->ip[id] |= R300_RS_TEX_PTR(ptr*4) | + rs->ip[id] |= R300_RS_TEX_PTR(ptr) | R300_RS_SEL_S(R300_RS_SEL_C0) | R300_RS_SEL_T(R300_RS_SEL_C1) | R300_RS_SEL_R(R300_RS_SEL_C2) | @@ -219,10 +248,10 @@ static void r300_rs_tex_write(struct r300_rs_block* rs, int id, int fp_offset) } static void r500_rs_col(struct r300_rs_block* rs, int id, int ptr, - boolean swizzle_0001) + enum r300_rs_swizzle swiz) { rs->ip[id] |= R500_RS_COL_PTR(ptr); - if (swizzle_0001) { + if (swiz == SWIZ_0001) { rs->ip[id] |= R500_RS_COL_FMT(R300_RS_COL_FMT_0001); } else { rs->ip[id] |= R500_RS_COL_FMT(R300_RS_COL_FMT_RGBA); @@ -230,32 +259,36 @@ static void r500_rs_col(struct r300_rs_block* rs, int id, int ptr, rs->inst[id] |= R500_RS_INST_COL_ID(id); } -static void r500_rs_col_write(struct r300_rs_block* rs, int id, int fp_offset) +static void r500_rs_col_write(struct r300_rs_block* rs, int id, int fp_offset, + enum r300_rs_col_write_type type) { - rs->inst[id] |= R500_RS_INST_COL_CN_WRITE | - R500_RS_INST_COL_ADDR(fp_offset); + if (type == WRITE_FACE) + rs->inst[id] |= R500_RS_INST_COL_CN_WRITE_BACKFACE | + R500_RS_INST_COL_ADDR(fp_offset); + else + rs->inst[id] |= R500_RS_INST_COL_CN_WRITE | + R500_RS_INST_COL_ADDR(fp_offset); + } static void r500_rs_tex(struct r300_rs_block* rs, int id, int ptr, enum r300_rs_swizzle swiz) { - int rs_tex_comp = ptr*4; - if (swiz == SWIZ_X001) { - rs->ip[id] |= R500_RS_SEL_S(rs_tex_comp) | + rs->ip[id] |= R500_RS_SEL_S(ptr) | R500_RS_SEL_T(R500_RS_IP_PTR_K0) | R500_RS_SEL_R(R500_RS_IP_PTR_K0) | R500_RS_SEL_Q(R500_RS_IP_PTR_K1); } else if (swiz == SWIZ_XY01) { - rs->ip[id] |= R500_RS_SEL_S(rs_tex_comp) | - R500_RS_SEL_T(rs_tex_comp + 1) | + rs->ip[id] |= R500_RS_SEL_S(ptr) | + R500_RS_SEL_T(ptr + 1) | R500_RS_SEL_R(R500_RS_IP_PTR_K0) | R500_RS_SEL_Q(R500_RS_IP_PTR_K1); } else { - rs->ip[id] |= R500_RS_SEL_S(rs_tex_comp) | - R500_RS_SEL_T(rs_tex_comp + 1) | - R500_RS_SEL_R(rs_tex_comp + 2) | - R500_RS_SEL_Q(rs_tex_comp + 3); + rs->ip[id] |= R500_RS_SEL_S(ptr) | + R500_RS_SEL_T(ptr + 1) | + R500_RS_SEL_R(ptr + 2) | + R500_RS_SEL_Q(ptr + 3); } rs->inst[id] |= R500_RS_INST_TEX_ID(id); } @@ -268,21 +301,30 @@ static void r500_rs_tex_write(struct r300_rs_block* rs, int id, int fp_offset) /* Set up the RS block. * - * This is the part of the chipset that actually does the rasterization - * of vertices into fragments. This is also the part of the chipset that - * locks up if any part of it is even slightly wrong. */ -static void r300_update_rs_block(struct r300_context* r300, - struct r300_shader_semantics* vs_outputs, - struct r300_shader_semantics* fs_inputs) + * This is the part of the chipset that is responsible for linking vertex + * and fragment shaders and stuffed texture coordinates. + * + * The rasterizer reads data from VAP, which produces vertex shader outputs, + * and GA, which produces stuffed texture coordinates. VAP outputs have + * precedence over GA. All outputs must be rasterized otherwise it locks up. + * If there are more outputs rasterized than is set in VAP/GA, it locks up + * too. The funky part is that this info has been pretty much obtained by trial + * and error. */ +static void r300_update_rs_block(struct r300_context *r300) { - struct r300_rs_block rs = { { 0 } }; - int i, col_count = 0, tex_count = 0, fp_offset = 0, count; - void (*rX00_rs_col)(struct r300_rs_block*, int, int, boolean); - void (*rX00_rs_col_write)(struct r300_rs_block*, int, int); + struct r300_vertex_shader *vs = r300->vs_state.state; + struct r300_shader_semantics *vs_outputs = &vs->outputs; + struct r300_shader_semantics *fs_inputs = &r300_fs(r300)->shader->inputs; + struct r300_rs_block rs = {0}; + int i, col_count = 0, tex_count = 0, fp_offset = 0, count, loc = 0, tex_ptr = 0; + void (*rX00_rs_col)(struct r300_rs_block*, int, int, enum r300_rs_swizzle); + void (*rX00_rs_col_write)(struct r300_rs_block*, int, int, enum r300_rs_col_write_type); void (*rX00_rs_tex)(struct r300_rs_block*, int, int, enum r300_rs_swizzle); void (*rX00_rs_tex_write)(struct r300_rs_block*, int, int); boolean any_bcolor_used = vs_outputs->bcolor[0] != ATTR_UNUSED || vs_outputs->bcolor[1] != ATTR_UNUSED; + int *stream_loc_notcl = r300->stream_loc_notcl; + uint32_t stuffing_enable = 0; if (r300->screen->caps.is_r500) { rX00_rs_col = r500_rs_col; @@ -296,18 +338,44 @@ static void r300_update_rs_block(struct r300_context* r300, rX00_rs_tex_write = r300_rs_tex_write; } - /* Rasterize colors. */ + /* 0x5555 copied from classic, which means: + * Select user color 0 for COLOR0 up to COLOR7. + * What the hell does that mean? */ + rs.vap_vtx_state_cntl = 0x5555; + + /* The position is always present in VAP. */ + rs.vap_vsm_vtx_assm |= R300_INPUT_CNTL_POS; + rs.vap_out_vtx_fmt[0] |= R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT; + stream_loc_notcl[loc++] = 0; + + /* Set up the point size in VAP. */ + if (vs_outputs->psize != ATTR_UNUSED) { + rs.vap_out_vtx_fmt[0] |= R300_VAP_OUTPUT_VTX_FMT_0__PT_SIZE_PRESENT; + stream_loc_notcl[loc++] = 1; + } + + /* Set up and rasterize colors. */ for (i = 0; i < ATTR_COLOR_COUNT; i++) { if (vs_outputs->color[i] != ATTR_UNUSED || any_bcolor_used || vs_outputs->color[1] != ATTR_UNUSED) { - /* Always rasterize if it's written by the VS, - * otherwise it locks up. */ - rX00_rs_col(&rs, col_count, i, FALSE); + /* Set up the color in VAP. */ + rs.vap_vsm_vtx_assm |= R300_INPUT_CNTL_COLOR; + rs.vap_out_vtx_fmt[0] |= + R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << i; + stream_loc_notcl[loc++] = 2 + i; + + /* Rasterize it. */ + rX00_rs_col(&rs, col_count, col_count, SWIZ_XYZW); - /* Write it to the FS input register if it's used by the FS. */ + /* Write it to the FS input register if it's needed by the FS. */ if (fs_inputs->color[i] != ATTR_UNUSED) { - rX00_rs_col_write(&rs, col_count, fp_offset); + rX00_rs_col_write(&rs, col_count, fp_offset, WRITE_COLOR); fp_offset++; + + DBG(r300, DBG_RS, + "r300: Rasterized color %i written to FS.\n", i); + } else { + DBG(r300, DBG_RS, "r300: Rasterized color %i unused.\n", i); } col_count++; } else { @@ -315,92 +383,254 @@ static void r300_update_rs_block(struct r300_context* r300, /* If we try to set it to (0,0,0,1), it will lock up. */ if (fs_inputs->color[i] != ATTR_UNUSED) { fp_offset++; + + DBG(r300, DBG_RS, "r300: FS input color %i unassigned%s.\n", + i); + } + } + } + + /* Set up back-face colors. The rasterizer will do the color selection + * automatically. */ + if (any_bcolor_used) { + if (r300->two_sided_color) { + /* Rasterize as back-face colors. */ + for (i = 0; i < ATTR_COLOR_COUNT; i++) { + rs.vap_vsm_vtx_assm |= R300_INPUT_CNTL_COLOR; + rs.vap_out_vtx_fmt[0] |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << (2+i); + stream_loc_notcl[loc++] = 4 + i; + } + } else { + /* Rasterize two fake texcoords to prevent from the two-sided color + * selection. */ + /* XXX Consider recompiling the vertex shader to save 2 RS units. */ + for (i = 0; i < 2; i++) { + rs.vap_vsm_vtx_assm |= (R300_INPUT_CNTL_TC0 << tex_count); + rs.vap_out_vtx_fmt[1] |= (4 << (3 * tex_count)); + stream_loc_notcl[loc++] = 6 + tex_count; + + /* Rasterize it. */ + rX00_rs_tex(&rs, tex_count, tex_ptr, SWIZ_XYZW); + tex_count++; + tex_ptr += 4; } } } + /* gl_FrontFacing. + * Note that we can use either the two-sided color selection based on + * the front and back vertex shader colors, or gl_FrontFacing, + * but not both! It locks up otherwise. + * + * In Direct3D 9, the two-sided color selection can be used + * with shaders 2.0 only, while gl_FrontFacing can be used + * with shaders 3.0 only. The hardware apparently hasn't been designed + * to support both at the same time. */ + if (r300->screen->caps.is_r500 && fs_inputs->face != ATTR_UNUSED && + !(any_bcolor_used && r300->two_sided_color)) { + rX00_rs_col(&rs, col_count, col_count, SWIZ_XYZW); + rX00_rs_col_write(&rs, col_count, fp_offset, WRITE_FACE); + fp_offset++; + col_count++; + DBG(r300, DBG_RS, "r300: Rasterized FACE written to FS.\n"); + } + /* Rasterize texture coordinates. */ - for (i = 0; i < ATTR_GENERIC_COUNT; i++) { - bool sprite_coord = !!(r300->sprite_coord_enable & (1 << i)); + for (i = 0; i < ATTR_GENERIC_COUNT && tex_count < 8; i++) { + bool sprite_coord = false; + + if (fs_inputs->generic[i] != ATTR_UNUSED) { + sprite_coord = !!(r300->sprite_coord_enable & (1 << i)); + } if (vs_outputs->generic[i] != ATTR_UNUSED || sprite_coord) { - /* Always rasterize if it's written by the VS, - * otherwise it locks up. */ - rX00_rs_tex(&rs, tex_count, tex_count, + if (!sprite_coord) { + /* Set up the texture coordinates in VAP. */ + rs.vap_vsm_vtx_assm |= (R300_INPUT_CNTL_TC0 << tex_count); + rs.vap_out_vtx_fmt[1] |= (4 << (3 * tex_count)); + stream_loc_notcl[loc++] = 6 + tex_count; + } else + stuffing_enable |= + R300_GB_TEX_ST << (R300_GB_TEX0_SOURCE_SHIFT + (tex_count*2)); + + /* Rasterize it. */ + rX00_rs_tex(&rs, tex_count, tex_ptr, sprite_coord ? SWIZ_XY01 : SWIZ_XYZW); - /* Write it to the FS input register if it's used by the FS. */ + /* Write it to the FS input register if it's needed by the FS. */ if (fs_inputs->generic[i] != ATTR_UNUSED) { rX00_rs_tex_write(&rs, tex_count, fp_offset); - if (sprite_coord) - debug_printf("r300: SpriteCoord (generic index %i) is being written to reg %i\n", i, fp_offset); fp_offset++; + + DBG(r300, DBG_RS, + "r300: Rasterized generic %i written to FS%s in texcoord %d.\n", + i, sprite_coord ? " (sprite coord)" : "", tex_count); + } else { + DBG(r300, DBG_RS, + "r300: Rasterized generic %i unused%s.\n", + i, sprite_coord ? " (sprite coord)" : ""); } tex_count++; + tex_ptr += sprite_coord ? 2 : 4; } else { /* Skip the FS input register, leave it uninitialized. */ /* If we try to set it to (0,0,0,1), it will lock up. */ if (fs_inputs->generic[i] != ATTR_UNUSED) { fp_offset++; + + DBG(r300, DBG_RS, "r300: FS input generic %i unassigned%s.\n", + i, sprite_coord ? " (sprite coord)" : ""); + } + } + } + + if (DBG_ON(r300, DBG_RS)) { + for (; i < ATTR_GENERIC_COUNT; i++) { + if (fs_inputs->generic[i] != ATTR_UNUSED) { + DBG(r300, DBG_RS, + "r300: FS input generic %i unassigned.\n", i); } } } /* Rasterize fog coordinates. */ - if (vs_outputs->fog != ATTR_UNUSED) { - /* Always rasterize if it's written by the VS, - * otherwise it locks up. */ - rX00_rs_tex(&rs, tex_count, tex_count, SWIZ_X001); + if (vs_outputs->fog != ATTR_UNUSED && tex_count < 8) { + /* Set up the fog coordinates in VAP. */ + rs.vap_vsm_vtx_assm |= (R300_INPUT_CNTL_TC0 << tex_count); + rs.vap_out_vtx_fmt[1] |= (4 << (3 * tex_count)); + stream_loc_notcl[loc++] = 6 + tex_count; - /* Write it to the FS input register if it's used by the FS. */ + /* Rasterize it. */ + rX00_rs_tex(&rs, tex_count, tex_ptr, SWIZ_X001); + + /* Write it to the FS input register if it's needed by the FS. */ if (fs_inputs->fog != ATTR_UNUSED) { rX00_rs_tex_write(&rs, tex_count, fp_offset); fp_offset++; + + DBG(r300, DBG_RS, "r300: Rasterized fog written to FS.\n"); + } else { + DBG(r300, DBG_RS, "r300: Rasterized fog unused.\n"); } tex_count++; + tex_ptr += 4; } else { /* Skip the FS input register, leave it uninitialized. */ /* If we try to set it to (0,0,0,1), it will lock up. */ if (fs_inputs->fog != ATTR_UNUSED) { fp_offset++; + + DBG(r300, DBG_RS, "r300: FS input fog unassigned.\n"); } } /* Rasterize WPOS. */ - /* If the FS doesn't need it, it's not written by the VS. */ - if (vs_outputs->wpos != ATTR_UNUSED && fs_inputs->wpos != ATTR_UNUSED) { - rX00_rs_tex(&rs, tex_count, tex_count, SWIZ_XYZW); + /* Don't set it in VAP if the FS doesn't need it. */ + if (fs_inputs->wpos != ATTR_UNUSED && tex_count < 8) { + /* Set up the WPOS coordinates in VAP. */ + rs.vap_vsm_vtx_assm |= (R300_INPUT_CNTL_TC0 << tex_count); + rs.vap_out_vtx_fmt[1] |= (4 << (3 * tex_count)); + stream_loc_notcl[loc++] = 6 + tex_count; + + /* Rasterize it. */ + rX00_rs_tex(&rs, tex_count, tex_ptr, SWIZ_XYZW); + + /* Write it to the FS input register. */ rX00_rs_tex_write(&rs, tex_count, fp_offset); + DBG(r300, DBG_RS, "r300: Rasterized WPOS written to FS.\n"); + fp_offset++; tex_count++; + tex_ptr += 4; + } + + /* Invalidate the rest of the no-TCL (GA) stream locations. */ + for (; loc < 16;) { + stream_loc_notcl[loc++] = -1; } /* Rasterize at least one color, or bad things happen. */ if (col_count == 0 && tex_count == 0) { - rX00_rs_col(&rs, 0, 0, TRUE); + rX00_rs_col(&rs, 0, 0, SWIZ_0001); col_count++; + + DBG(r300, DBG_RS, "r300: Rasterized color 0 to prevent lockups.\n"); } - rs.count = (tex_count*4) | (col_count << R300_IC_COUNT_SHIFT) | + DBG(r300, DBG_RS, "r300: --- Rasterizer status ---: colors: %i, " + "generics: %i.\n", col_count, tex_count); + + rs.count = MIN2(tex_ptr, 32) | (col_count << R300_IC_COUNT_SHIFT) | R300_HIRES_EN; count = MAX3(col_count, tex_count, 1); rs.inst_count = count - 1; + /* set the GB enable flags */ + if (r300->sprite_coord_enable) + stuffing_enable |= R300_GB_POINT_STUFF_ENABLE; + + rs.gb_enable = stuffing_enable; + /* Now, after all that, see if we actually need to update the state. */ if (memcmp(r300->rs_block_state.state, &rs, sizeof(struct r300_rs_block))) { memcpy(r300->rs_block_state.state, &rs, sizeof(struct r300_rs_block)); - r300->rs_block_state.size = 5 + count*2; + r300->rs_block_state.size = 13 + count*2; } } -/* Update the shader-dependant states. */ -static void r300_update_derived_shader_state(struct r300_context* r300) +static uint32_t r300_get_border_color(enum pipe_format format, + const float border[4]) { - struct r300_vertex_shader* vs = r300->vs_state.state; + const struct util_format_description *desc; + float border_swizzled[4] = { + border[2], + border[1], + border[0], + border[3] + }; + uint32_t r; + + desc = util_format_description(format); + + /* We don't use util_pack_format because it does not handle the formats + * we want, e.g. R4G4B4A4 is non-existent in Gallium. */ + switch (desc->channel[0].size) { + case 4: + r = ((float_to_ubyte(border_swizzled[0]) & 0xf0) >> 4) | + ((float_to_ubyte(border_swizzled[1]) & 0xf0) << 0) | + ((float_to_ubyte(border_swizzled[2]) & 0xf0) << 4) | + ((float_to_ubyte(border_swizzled[3]) & 0xf0) << 8); + break; + + case 5: + if (desc->channel[1].size == 5) { + r = ((float_to_ubyte(border_swizzled[0]) & 0xf8) >> 3) | + ((float_to_ubyte(border_swizzled[1]) & 0xf8) << 2) | + ((float_to_ubyte(border_swizzled[2]) & 0xf8) << 7) | + ((float_to_ubyte(border_swizzled[3]) & 0x80) << 8); + } else if (desc->channel[1].size == 6) { + r = ((float_to_ubyte(border_swizzled[0]) & 0xf8) >> 3) | + ((float_to_ubyte(border_swizzled[1]) & 0xfc) << 3) | + ((float_to_ubyte(border_swizzled[2]) & 0xf8) << 8); + } else { + assert(0); + r = 0; + } + break; + + default: + /* I think the fat formats (16, 32) are specified + * as the 8-bit ones. I am not sure how compressed formats + * work here. */ + r = ((float_to_ubyte(border_swizzled[0]) & 0xff) << 0) | + ((float_to_ubyte(border_swizzled[1]) & 0xff) << 8) | + ((float_to_ubyte(border_swizzled[2]) & 0xff) << 16) | + ((float_to_ubyte(border_swizzled[3]) & 0xff) << 24); + } - r300_update_rs_block(r300, &vs->outputs, &r300_fs(r300)->shader->inputs); + return r; } static void r300_merge_textures_and_samplers(struct r300_context* r300) @@ -411,10 +641,14 @@ static void r300_merge_textures_and_samplers(struct r300_context* r300) struct r300_sampler_state *sampler; struct r300_sampler_view *view; struct r300_texture *tex; - unsigned min_level, max_level, i, size; + unsigned min_level, max_level, i, j, size; unsigned count = MIN2(state->sampler_view_count, state->sampler_state_count); + /* The KIL opcode fix, see below. */ + if (!count && !r300->screen->caps.is_r500) + count = 1; + state->tx_enable = 0; state->count = 0; size = 2; @@ -431,15 +665,70 @@ static void r300_merge_textures_and_samplers(struct r300_context* r300) texstate->format = view->format; texstate->filter0 = sampler->filter0; texstate->filter1 = sampler->filter1; - texstate->border_color = sampler->border_color; + + /* Set the border color. */ + texstate->border_color = + r300_get_border_color(view->base.format, + sampler->state.border_color); + + /* determine min/max levels */ + max_level = MIN3(sampler->max_lod + view->base.first_level, + tex->desc.b.b.last_level, view->base.last_level); + min_level = MIN2(sampler->min_lod + view->base.first_level, + max_level); + + if (tex->desc.is_npot && min_level > 0) { + /* Even though we do not implement mipmapping for NPOT + * textures, we should at least honor the minimum level + * which is allowed to be displayed. We do this by setting up + * an i-th mipmap level as the zero level. */ + r300_texture_setup_format_state(r300->screen, &tex->desc, + min_level, + &texstate->format); + texstate->format.tile_config |= + tex->desc.offset_in_bytes[min_level] & 0xffffffe0; + assert((tex->desc.offset_in_bytes[min_level] & 0x1f) == 0); + } + + /* Assign a texture cache region. */ + texstate->format.format1 |= view->texcache_region; + + /* Depth textures are kinda special. */ + if (util_format_is_depth_or_stencil(tex->desc.b.b.format)) { + unsigned char depth_swizzle[4]; + + if (!r300->screen->caps.is_r500 && + util_format_get_blocksizebits(tex->desc.b.b.format) == 32) { + /* X24x8 is sampled as Y16X16 on r3xx-r4xx. + * The depth here is at the Y component. */ + for (j = 0; j < 4; j++) + depth_swizzle[j] = UTIL_FORMAT_SWIZZLE_Y; + } else { + for (j = 0; j < 4; j++) + depth_swizzle[j] = UTIL_FORMAT_SWIZZLE_X; + } + + /* If compare mode is disabled, sampler view swizzles + * are stored in the format. + * Otherwise, the swizzles must be applied after the compare + * mode in the fragment shader. */ + if (sampler->state.compare_mode == PIPE_TEX_COMPARE_NONE) { + texstate->format.format1 |= + r300_get_swizzle_combined(depth_swizzle, + view->swizzle); + } else { + texstate->format.format1 |= + r300_get_swizzle_combined(depth_swizzle, 0); + } + } /* to emulate 1D textures through 2D ones correctly */ - if (tex->b.b.target == PIPE_TEXTURE_1D) { + if (tex->desc.b.b.target == PIPE_TEXTURE_1D) { texstate->filter0 &= ~R300_TX_WRAP_T_MASK; texstate->filter0 |= R300_TX_WRAP_T(R300_TX_CLAMP_TO_EDGE); } - if (tex->uses_pitch) { + if (tex->desc.is_npot) { /* NPOT textures don't support mip filter, unfortunately. * This prevents incorrect rendering. */ texstate->filter0 &= ~R300_TX_MIN_FILTER_MIP_MASK; @@ -463,12 +752,7 @@ static void r300_merge_textures_and_samplers(struct r300_context* r300) texstate->filter0 |= R300_TX_WRAP_T(R300_TX_CLAMP_TO_EDGE); } } else { - /* determine min/max levels */ /* the MAX_MIP level is the largest (finest) one */ - max_level = MIN3(sampler->max_lod + view->base.first_level, - tex->b.b.last_level, view->base.last_level); - min_level = MIN2(sampler->min_lod + view->base.first_level, - max_level); texstate->format.format0 |= R300_TX_NUM_LEVELS(max_level); texstate->filter0 |= R300_TX_MAX_MIP_LEVEL(min_level); } @@ -477,6 +761,36 @@ static void r300_merge_textures_and_samplers(struct r300_context* r300) size += 16; state->count = i+1; + } else { + /* For the KIL opcode to work on r3xx-r4xx, the texture unit + * assigned to this opcode (it's always the first one) must be + * enabled. Otherwise the opcode doesn't work. + * + * In order to not depend on the fragment shader, we just make + * the first unit enabled all the time. */ + if (i == 0 && !r300->screen->caps.is_r500) { + pipe_sampler_view_reference( + (struct pipe_sampler_view**)&state->sampler_views[i], + &r300->texkill_sampler->base); + + state->tx_enable |= 1 << i; + + texstate = &state->regs[i]; + + /* Just set some valid state. */ + texstate->format = r300->texkill_sampler->format; + texstate->filter0 = + r300_translate_tex_filters(PIPE_TEX_FILTER_NEAREST, + PIPE_TEX_FILTER_NEAREST, + PIPE_TEX_FILTER_NEAREST, + FALSE); + texstate->filter1 = 0; + texstate->border_color = 0; + + texstate->filter0 |= i << 28; + size += 16; + state->count = i+1; + } } } @@ -491,21 +805,57 @@ static void r300_merge_textures_and_samplers(struct r300_context* r300) } } +/* We can't use compressed zbuffers as samplers. */ +static void r300_flush_depth_textures(struct r300_context *r300) +{ + struct r300_textures_state *state = + (struct r300_textures_state*)r300->textures_state.state; + unsigned i, level; + unsigned count = MIN2(state->sampler_view_count, + state->sampler_state_count); + + if (r300->z_decomp_rd) + return; + + for (i = 0; i < count; i++) + if (state->sampler_views[i] && state->sampler_states[i]) { + struct pipe_resource *tex = state->sampler_views[i]->base.texture; + + if (tex->target == PIPE_TEXTURE_3D || + tex->target == PIPE_TEXTURE_CUBE) + continue; + + /* Ignore non-depth textures. + * Also ignore reinterpreted depth textures, e.g. resource_copy. */ + if (!util_format_is_depth_or_stencil(tex->format)) + continue; + + for (level = 0; level <= tex->last_level; level++) + if (r300_texture(tex)->zmask_in_use[level]) { + /* We don't handle 3D textures and cubemaps yet. */ + r300_flush_depth_stencil(&r300->context, tex, + u_subresource(0, level), 0); + } + } +} + void r300_update_derived_state(struct r300_context* r300) { + r300_flush_depth_textures(r300); + if (r300->textures_state.dirty) { r300_merge_textures_and_samplers(r300); } if (r300->rs_block_state.dirty) { - r300_update_derived_shader_state(r300); - } + r300_update_rs_block(r300); - if (r300->draw) { - memset(&r300->vertex_info, 0, sizeof(struct vertex_info)); - r300_draw_emit_all_attribs(r300); - draw_compute_vertex_size(&r300->vertex_info); - r300_swtcl_vertex_psc(r300); + if (r300->draw) { + memset(&r300->vertex_info, 0, sizeof(struct vertex_info)); + r300_draw_emit_all_attribs(r300); + draw_compute_vertex_size(&r300->vertex_info); + r300_swtcl_vertex_psc(r300); + } } r300_update_hyperz_state(r300); diff --git a/src/gallium/drivers/r300/r300_state_inlines.h b/src/gallium/drivers/r300/r300_state_inlines.h index fcbdb91b67e..7e501221b1f 100644 --- a/src/gallium/drivers/r300/r300_state_inlines.h +++ b/src/gallium/drivers/r300/r300_state_inlines.h @@ -348,41 +348,43 @@ static INLINE uint32_t r300_translate_gb_pipes(int pipe_count) switch (pipe_count) { case 1: return R300_GB_TILE_PIPE_COUNT_RV300; - break; case 2: return R300_GB_TILE_PIPE_COUNT_R300; - break; case 3: return R300_GB_TILE_PIPE_COUNT_R420_3P; - break; case 4: return R300_GB_TILE_PIPE_COUNT_R420; - break; } return 0; } + /* Translate pipe_formats into PSC vertex types. */ static INLINE uint16_t r300_translate_vertex_data_type(enum pipe_format format) { uint32_t result = 0; const struct util_format_description *desc; + unsigned i; desc = util_format_description(format); if (desc->layout != UTIL_FORMAT_LAYOUT_PLAIN) { - fprintf(stderr, "r300: Bad format %s in %s:%d\n", util_format_name(format), - __FUNCTION__, __LINE__); - assert(0); - abort(); + return R300_INVALID_FORMAT; } - switch (desc->channel[0].type) { + /* Find the first non-VOID channel. */ + for (i = 0; i < 4; i++) { + if (desc->channel[i].type != UTIL_FORMAT_TYPE_VOID) { + break; + } + } + + switch (desc->channel[i].type) { /* Half-floats, floats, doubles */ case UTIL_FORMAT_TYPE_FLOAT: - switch (desc->channel[0].size) { + switch (desc->channel[i].size) { case 16: - /* XXX Supported only on RV350 and later. */ + /* Supported only on RV350 and later. */ if (desc->nr_channels > 2) { result = R300_DATA_TYPE_FLT16_4; } else { @@ -393,17 +395,14 @@ r300_translate_vertex_data_type(enum pipe_format format) { result = R300_DATA_TYPE_FLOAT_1 + (desc->nr_channels - 1); break; default: - fprintf(stderr, "r300: Bad format %s in %s:%d\n", - util_format_name(format), __FUNCTION__, __LINE__); - assert(0); - abort(); + return R300_INVALID_FORMAT; } break; /* Unsigned ints */ case UTIL_FORMAT_TYPE_UNSIGNED: /* Signed ints */ case UTIL_FORMAT_TYPE_SIGNED: - switch (desc->channel[0].size) { + switch (desc->channel[i].size) { case 8: result = R300_DATA_TYPE_BYTE; break; @@ -415,25 +414,17 @@ r300_translate_vertex_data_type(enum pipe_format format) { } break; default: - fprintf(stderr, "r300: Bad format %s in %s:%d\n", - util_format_name(format), __FUNCTION__, __LINE__); - fprintf(stderr, "r300: desc->channel[0].size == %d\n", - desc->channel[0].size); - assert(0); - abort(); + return R300_INVALID_FORMAT; } break; default: - fprintf(stderr, "r300: Bad format %s in %s:%d\n", - util_format_name(format), __FUNCTION__, __LINE__); - assert(0); - abort(); + return R300_INVALID_FORMAT; } - if (desc->channel[0].type == UTIL_FORMAT_TYPE_SIGNED) { + if (desc->channel[i].type == UTIL_FORMAT_TYPE_SIGNED) { result |= R300_SIGNED; } - if (desc->channel[0].normalized) { + if (desc->channel[i].normalized) { result |= R300_NORMALIZE; } @@ -449,7 +440,7 @@ r300_translate_vertex_data_swizzle(enum pipe_format format) { if (desc->layout != UTIL_FORMAT_LAYOUT_PLAIN) { fprintf(stderr, "r300: Bad format %s in %s:%d\n", - util_format_name(format), __FUNCTION__, __LINE__); + util_format_short_name(format), __FUNCTION__, __LINE__); return 0; } diff --git a/src/gallium/drivers/r300/r300_state_invariant.c b/src/gallium/drivers/r300/r300_state_invariant.c deleted file mode 100644 index cd9443fa260..00000000000 --- a/src/gallium/drivers/r300/r300_state_invariant.c +++ /dev/null @@ -1,121 +0,0 @@ -/* - * Copyright 2009 Joakim Sindholt <[email protected]> - * Corbin Simpson <[email protected]> - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * on the rights to use, copy, modify, merge, publish, distribute, sub - * license, and/or sell copies of the Software, and to permit persons to whom - * the Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice (including the next - * paragraph) shall be included in all copies or substantial portions of the - * Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL - * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, - * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR - * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE - * USE OR OTHER DEALINGS IN THE SOFTWARE. */ - -#include "r300_context.h" -#include "r300_cs.h" -#include "r300_reg.h" -#include "r300_screen.h" -#include "r300_state_invariant.h" - -struct pipe_viewport_state r300_viewport_identity = { - .scale = {1.0, 1.0, 1.0, 1.0}, - .translate = {0.0, 0.0, 0.0, 0.0}, -}; - -/* Calculate and emit invariant state. This is data that the 3D engine - * will probably want at the beginning of every CS, but it's not currently - * handled by any CSO setup, and in addition it doesn't really change much. - * - * Note that eventually this should be empty, but it's useful for development - * and general unduplication of code. */ -void r300_emit_invariant_state(struct r300_context* r300, - unsigned size, void* state) -{ - CS_LOCALS(r300); - - if (r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0)) { - /* Subpixel multisampling for AA. */ - BEGIN_CS(4); - OUT_CS_REG(R300_GB_MSPOS0, 0x66666666); - OUT_CS_REG(R300_GB_MSPOS1, 0x6666666); - END_CS; - } - - BEGIN_CS(12 + (r300->screen->caps.has_tcl ? 2 : 0)); - - /*** Graphics Backend (GB) ***/ - /* Source of fog depth */ - OUT_CS_REG(R300_GB_SELECT, R300_GB_FOG_SELECT_1_1_W); - - /*** Fog (FG) ***/ - OUT_CS_REG(R300_FG_FOG_BLEND, 0x0); - OUT_CS_REG(R300_FG_FOG_COLOR_R, 0x0); - OUT_CS_REG(R300_FG_FOG_COLOR_G, 0x0); - OUT_CS_REG(R300_FG_FOG_COLOR_B, 0x0); - - /*** VAP ***/ - /* Sign/normalize control */ - OUT_CS_REG(R300_VAP_PSC_SGN_NORM_CNTL, R300_SGN_NORM_NO_ZERO); - /* TCL-only stuff */ - if (r300->screen->caps.has_tcl) { - /* Amount of time to wait for vertex fetches in PVS */ - OUT_CS_REG(VAP_PVS_VTX_TIMEOUT_REG, 0xffff); - } - - END_CS; - - /* XXX unsorted stuff from surface_fill */ - BEGIN_CS(38 + (r300->screen->caps.has_tcl ? 7 : 0) + - (r300->screen->caps.is_rv350 ? 4 : 0)); - - if (r300->screen->caps.has_tcl) { - /*Flushing PVS is required before the VAP_GB registers can be changed*/ - OUT_CS_REG(R300_VAP_PVS_STATE_FLUSH_REG, 0); - OUT_CS_REG_SEQ(R300_VAP_GB_VERT_CLIP_ADJ, 4); - OUT_CS_32F(1.0); - OUT_CS_32F(1.0); - OUT_CS_32F(1.0); - OUT_CS_32F(1.0); - } - /* XXX line tex stuffing */ - OUT_CS_REG_SEQ(R300_GA_LINE_S0, 1); - OUT_CS_32F(0.0); - OUT_CS_REG_SEQ(R300_GA_LINE_S1, 1); - OUT_CS_32F(1.0); - OUT_CS_REG(R300_GA_TRIANGLE_STIPPLE, 0x5 | - (0x5 << R300_GA_TRIANGLE_STIPPLE_Y_SHIFT_SHIFT)); - /* XXX this big chunk should be refactored into rs_state */ - OUT_CS_REG(R300_GA_SOLID_RG, 0x00000000); - OUT_CS_REG(R300_GA_SOLID_BA, 0x00000000); - OUT_CS_REG(R300_GA_ROUND_MODE, 0x00000001); - OUT_CS_REG(R300_GA_OFFSET, 0x00000000); - OUT_CS_REG(R300_GA_FOG_SCALE, 0x3DBF1412); - OUT_CS_REG(R300_GA_FOG_OFFSET, 0x00000000); - OUT_CS_REG(R300_SU_TEX_WRAP, 0x00000000); - OUT_CS_REG(R300_SU_DEPTH_SCALE, 0x4B7FFFFF); - OUT_CS_REG(R300_SU_DEPTH_OFFSET, 0x00000000); - OUT_CS_REG(R300_SC_HYPERZ, 0x0000001C); - OUT_CS_REG(R300_SC_EDGERULE, 0x2DA49525); - OUT_CS_REG(R300_RB3D_AARESOLVE_CTL, 0x00000000); - - if (r300->screen->caps.is_rv350) { - OUT_CS_REG(R500_RB3D_DISCARD_SRC_PIXEL_LTE_THRESHOLD, 0x01010101); - OUT_CS_REG(R500_RB3D_DISCARD_SRC_PIXEL_GTE_THRESHOLD, 0xFEFEFEFE); - } - - OUT_CS_REG(R300_ZB_BW_CNTL, 0x00000000); - OUT_CS_REG(R300_ZB_DEPTHCLEARVALUE, 0x00000000); - OUT_CS_REG(R300_ZB_HIZ_OFFSET, 0x00000000); - OUT_CS_REG(R300_ZB_HIZ_PITCH, 0x00000000); - END_CS; -} diff --git a/src/gallium/drivers/r300/r300_state_invariant.h b/src/gallium/drivers/r300/r300_state_invariant.h deleted file mode 100644 index 83d031c7fe9..00000000000 --- a/src/gallium/drivers/r300/r300_state_invariant.h +++ /dev/null @@ -1,31 +0,0 @@ -/* - * Copyright 2008 Corbin Simpson <[email protected]> - * - * Permission is hereby granted, free of charge, to any person obtaining a - * copy of this software and associated documentation files (the "Software"), - * to deal in the Software without restriction, including without limitation - * on the rights to use, copy, modify, merge, publish, distribute, sub - * license, and/or sell copies of the Software, and to permit persons to whom - * the Software is furnished to do so, subject to the following conditions: - * - * The above copyright notice and this permission notice (including the next - * paragraph) shall be included in all copies or substantial portions of the - * Software. - * - * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR - * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL - * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, - * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR - * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE - * USE OR OTHER DEALINGS IN THE SOFTWARE. */ - -#ifndef R300_STATE_INVARIANT_H -#define R300_STATE_INVARIANT_H - -struct r300_context; - -void r300_emit_invariant_state(struct r300_context* r300, - unsigned size, void* state); - -#endif /* R300_STATE_INVARIANT_H */ diff --git a/src/gallium/drivers/r300/r300_texture.c b/src/gallium/drivers/r300/r300_texture.c index bffda846d70..cee56bccdcd 100644 --- a/src/gallium/drivers/r300/r300_texture.c +++ b/src/gallium/drivers/r300/r300_texture.c @@ -21,44 +21,76 @@ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE * USE OR OTHER DEALINGS IN THE SOFTWARE. */ -#include "pipe/p_screen.h" - -#include "util/u_format.h" -#include "util/u_math.h" -#include "util/u_memory.h" +/* Always include headers in the reverse order!! ~ M. */ +#include "r300_texture.h" #include "r300_context.h" #include "r300_reg.h" -#include "r300_texture.h" +#include "r300_texture_desc.h" #include "r300_transfer.h" #include "r300_screen.h" #include "r300_winsys.h" -/* XXX Enable float textures here. */ -/*#define ENABLE_FLOAT_TEXTURES*/ - -#define TILE_WIDTH 0 -#define TILE_HEIGHT 1 +#include "util/u_format.h" +#include "util/u_format_s3tc.h" +#include "util/u_math.h" +#include "util/u_memory.h" +#include "util/u_mm.h" -static const unsigned microblock_table[5][3][2] = { - /*linear tiled square-tiled */ - {{32, 1}, {8, 4}, {0, 0}}, /* 8 bits per pixel */ - {{16, 1}, {8, 2}, {4, 4}}, /* 16 bits per pixel */ - {{ 8, 1}, {4, 2}, {0, 0}}, /* 32 bits per pixel */ - {{ 4, 1}, {0, 0}, {2, 2}}, /* 64 bits per pixel */ - {{ 2, 1}, {0, 0}, {0, 0}} /* 128 bits per pixel */ -}; +#include "pipe/p_screen.h" -/* Return true for non-compressed and non-YUV formats. */ -static boolean r300_format_is_plain(enum pipe_format format) +unsigned r300_get_swizzle_combined(const unsigned char *swizzle_format, + const unsigned char *swizzle_view) { - const struct util_format_description *desc = util_format_description(format); + unsigned i; + unsigned char swizzle[4]; + unsigned result = 0; + const uint32_t swizzle_shift[4] = { + R300_TX_FORMAT_R_SHIFT, + R300_TX_FORMAT_G_SHIFT, + R300_TX_FORMAT_B_SHIFT, + R300_TX_FORMAT_A_SHIFT + }; + const uint32_t swizzle_bit[4] = { + R300_TX_FORMAT_X, + R300_TX_FORMAT_Y, + R300_TX_FORMAT_Z, + R300_TX_FORMAT_W + }; - if (!format) { - return FALSE; + if (swizzle_view) { + /* Combine two sets of swizzles. */ + for (i = 0; i < 4; i++) { + swizzle[i] = swizzle_view[i] <= UTIL_FORMAT_SWIZZLE_W ? + swizzle_format[swizzle_view[i]] : swizzle_view[i]; + } + } else { + memcpy(swizzle, swizzle_format, 4); } - return desc->layout == UTIL_FORMAT_LAYOUT_PLAIN; + /* Get swizzle. */ + for (i = 0; i < 4; i++) { + switch (swizzle[i]) { + case UTIL_FORMAT_SWIZZLE_Y: + result |= swizzle_bit[1] << swizzle_shift[i]; + break; + case UTIL_FORMAT_SWIZZLE_Z: + result |= swizzle_bit[2] << swizzle_shift[i]; + break; + case UTIL_FORMAT_SWIZZLE_W: + result |= swizzle_bit[3] << swizzle_shift[i]; + break; + case UTIL_FORMAT_SWIZZLE_0: + result |= R300_TX_FORMAT_ZERO << swizzle_shift[i]; + break; + case UTIL_FORMAT_SWIZZLE_1: + result |= R300_TX_FORMAT_ONE << swizzle_shift[i]; + break; + default: /* UTIL_FORMAT_SWIZZLE_X */ + result |= swizzle_bit[0] << swizzle_shift[i]; + } + } + return result; } /* Translate a pipe_format into a useful texture format for sampling. @@ -74,24 +106,13 @@ static boolean r300_format_is_plain(enum pipe_format format) * The FORMAT specifies how the texture sampler will treat the texture, and * makes available X, Y, Z, W, ZERO, and ONE for swizzling. */ uint32_t r300_translate_texformat(enum pipe_format format, - const unsigned char *swizzle) + const unsigned char *swizzle_view, + boolean is_r500) { uint32_t result = 0; const struct util_format_description *desc; unsigned i; boolean uniform = TRUE; - const uint32_t swizzle_shift[4] = { - R300_TX_FORMAT_R_SHIFT, - R300_TX_FORMAT_G_SHIFT, - R300_TX_FORMAT_B_SHIFT, - R300_TX_FORMAT_A_SHIFT - }; - const uint32_t swizzle_bit[4] = { - R300_TX_FORMAT_X, - R300_TX_FORMAT_Y, - R300_TX_FORMAT_Z, - R300_TX_FORMAT_W - }; const uint32_t sign_bit[4] = { R300_TX_FORMAT_SIGNED_X, R300_TX_FORMAT_SIGNED_Y, @@ -103,14 +124,18 @@ uint32_t r300_translate_texformat(enum pipe_format format, /* Colorspace (return non-RGB formats directly). */ switch (desc->colorspace) { - /* Depth stencil formats. */ + /* Depth stencil formats. + * Swizzles are added in r300_merge_textures_and_samplers. */ case UTIL_FORMAT_COLORSPACE_ZS: switch (format) { case PIPE_FORMAT_Z16_UNORM: - return R300_EASY_TX_FORMAT(X, X, X, X, X16); + return R300_TX_FORMAT_X16; case PIPE_FORMAT_X8Z24_UNORM: case PIPE_FORMAT_S8_USCALED_Z24_UNORM: - return R300_EASY_TX_FORMAT(X, X, X, X, W24_FP); + if (is_r500) + return R500_TX_FORMAT_Y8X24; + else + return R300_TX_FORMAT_Y16X16; default: return ~0; /* Unsupported. */ } @@ -144,53 +169,14 @@ uint32_t r300_translate_texformat(enum pipe_format format, } } - /* Add swizzle. */ - if (!swizzle) { - swizzle = desc->swizzle; - } /*else { - if (swizzle[0] != desc->swizzle[0] || - swizzle[1] != desc->swizzle[1] || - swizzle[2] != desc->swizzle[2] || - swizzle[3] != desc->swizzle[3]) - { - const char n[6] = "RGBA01"; - fprintf(stderr, "Got different swizzling! Format: %c%c%c%c, " - "View: %c%c%c%c\n", - n[desc->swizzle[0]], n[desc->swizzle[1]], - n[desc->swizzle[2]], n[desc->swizzle[3]], - n[swizzle[0]], n[swizzle[1]], n[swizzle[2]], - n[swizzle[3]]); - } - }*/ - - for (i = 0; i < 4; i++) { - switch (swizzle[i]) { - case UTIL_FORMAT_SWIZZLE_X: - case UTIL_FORMAT_SWIZZLE_NONE: - result |= swizzle_bit[0] << swizzle_shift[i]; - break; - case UTIL_FORMAT_SWIZZLE_Y: - result |= swizzle_bit[1] << swizzle_shift[i]; - break; - case UTIL_FORMAT_SWIZZLE_Z: - result |= swizzle_bit[2] << swizzle_shift[i]; - break; - case UTIL_FORMAT_SWIZZLE_W: - result |= swizzle_bit[3] << swizzle_shift[i]; - break; - case UTIL_FORMAT_SWIZZLE_0: - result |= R300_TX_FORMAT_ZERO << swizzle_shift[i]; - break; - case UTIL_FORMAT_SWIZZLE_1: - result |= R300_TX_FORMAT_ONE << swizzle_shift[i]; - break; - default: - return ~0; /* Unsupported. */ - } - } + result |= r300_get_swizzle_combined(desc->swizzle, swizzle_view); /* S3TC formats. */ if (desc->layout == UTIL_FORMAT_LAYOUT_S3TC) { + if (!util_format_s3tc_enabled) { + return ~0; /* Unsupported. */ + } + switch (format) { case PIPE_FORMAT_DXT1_RGB: case PIPE_FORMAT_DXT1_RGBA: @@ -274,16 +260,26 @@ uint32_t r300_translate_texformat(enum pipe_format format, return ~0; /* Unsupported/unknown. */ } + /* Find the first non-VOID channel. */ + for (i = 0; i < 4; i++) { + if (desc->channel[i].type != UTIL_FORMAT_TYPE_VOID) { + break; + } + } + + if (i == 4) + return ~0; /* Unsupported/unknown. */ + /* And finally, uniform formats. */ - switch (desc->channel[0].type) { + switch (desc->channel[i].type) { case UTIL_FORMAT_TYPE_UNSIGNED: case UTIL_FORMAT_TYPE_SIGNED: - if (!desc->channel[0].normalized && + if (!desc->channel[i].normalized && desc->colorspace != UTIL_FORMAT_COLORSPACE_SRGB) { return ~0; } - switch (desc->channel[0].size) { + switch (desc->channel[i].size) { case 4: switch (desc->nr_channels) { case 2: @@ -316,9 +312,8 @@ uint32_t r300_translate_texformat(enum pipe_format format, } return ~0; -#if defined(ENABLE_FLOAT_TEXTURES) case UTIL_FORMAT_TYPE_FLOAT: - switch (desc->channel[0].size) { + switch (desc->channel[i].size) { case 16: switch (desc->nr_channels) { case 1: @@ -340,7 +335,6 @@ uint32_t r300_translate_texformat(enum pipe_format format, return R300_TX_FORMAT_32F_32F_32F_32F | result; } } -#endif } return ~0; /* Unsupported/unknown. */ @@ -351,6 +345,8 @@ uint32_t r500_tx_format_msb_bit(enum pipe_format format) switch (format) { case PIPE_FORMAT_RGTC1_UNORM: case PIPE_FORMAT_RGTC1_SNORM: + case PIPE_FORMAT_X8Z24_UNORM: + case PIPE_FORMAT_S8_USCALED_Z24_UNORM: return R500_TXFORMAT_MSB; default: return 0; @@ -373,6 +369,11 @@ static uint32_t r300_translate_colorformat(enum pipe_format format) return R300_COLOR_FORMAT_I8; /* 16-bit buffers. */ + case PIPE_FORMAT_L8A8_UNORM: + case PIPE_FORMAT_R8G8_UNORM: + case PIPE_FORMAT_R8G8_SNORM: + return R300_COLOR_FORMAT_UV88; + case PIPE_FORMAT_B5G6R5_UNORM: return R300_COLOR_FORMAT_RGB565; @@ -405,16 +406,12 @@ static uint32_t r300_translate_colorformat(enum pipe_format format) /* 64-bit buffers. */ case PIPE_FORMAT_R16G16B16A16_UNORM: case PIPE_FORMAT_R16G16B16A16_SNORM: -#if defined(ENABLE_FLOAT_TEXTURES) case PIPE_FORMAT_R16G16B16A16_FLOAT: -#endif return R300_COLOR_FORMAT_ARGB16161616; /* 128-bit buffers. */ -#if defined(ENABLE_FLOAT_TEXTURES) case PIPE_FORMAT_R32G32B32A32_FLOAT: return R300_COLOR_FORMAT_ARGB32323232; -#endif /* YUV buffers. */ case PIPE_FORMAT_UYVY: @@ -461,15 +458,25 @@ static uint32_t r300_translate_out_fmt(enum pipe_format format) desc = util_format_description(format); + /* Find the first non-VOID channel. */ + for (i = 0; i < 4; i++) { + if (desc->channel[i].type != UTIL_FORMAT_TYPE_VOID) { + break; + } + } + + if (i == 4) + return ~0; /* Unsupported/unknown. */ + /* Specifies how the shader output is written to the fog unit. */ - if (desc->channel[0].type == UTIL_FORMAT_TYPE_FLOAT) { - if (desc->channel[0].size == 32) { + if (desc->channel[i].type == UTIL_FORMAT_TYPE_FLOAT) { + if (desc->channel[i].size == 32) { modifier |= R300_US_OUT_FMT_C4_32_FP; } else { modifier |= R300_US_OUT_FMT_C4_16_FP; } } else { - if (desc->channel[0].size == 16) { + if (desc->channel[i].size == 16) { modifier |= R300_US_OUT_FMT_C4_16; } else { /* C4_8 seems to be used for the formats whose pixel size @@ -486,7 +493,7 @@ static uint32_t r300_translate_out_fmt(enum pipe_format format) /* Add swizzles and return. */ switch (format) { - /* 8-bit outputs. + /* 8-bit outputs, one channel. * COLORFORMAT_I8 stores the C2 component. */ case PIPE_FORMAT_A8_UNORM: return modifier | R300_C2_SEL_A; @@ -496,6 +503,14 @@ static uint32_t r300_translate_out_fmt(enum pipe_format format) case PIPE_FORMAT_R8_SNORM: return modifier | R300_C2_SEL_R; + /* 16-bit outputs, two channels. + * COLORFORMAT_UV88 stores C2 and C0. */ + case PIPE_FORMAT_L8A8_UNORM: + return modifier | R300_C0_SEL_A | R300_C2_SEL_R; + case PIPE_FORMAT_R8G8_UNORM: + case PIPE_FORMAT_R8G8_SNORM: + return modifier | R300_C0_SEL_G | R300_C2_SEL_R; + /* BGRA outputs. */ case PIPE_FORMAT_B5G6R5_UNORM: case PIPE_FORMAT_B5G5R5A1_UNORM: @@ -532,7 +547,7 @@ static uint32_t r300_translate_out_fmt(enum pipe_format format) case PIPE_FORMAT_R10SG10SB10SA2U_NORM: case PIPE_FORMAT_R16G16B16A16_UNORM: case PIPE_FORMAT_R16G16B16A16_SNORM: - //case PIPE_FORMAT_R16G16B16A16_FLOAT: /* not in pipe_format */ + case PIPE_FORMAT_R16G16B16A16_FLOAT: case PIPE_FORMAT_R32G32B32A32_FLOAT: return modifier | R300_C0_SEL_R | R300_C1_SEL_G | @@ -556,53 +571,55 @@ boolean r300_is_zs_format_supported(enum pipe_format format) boolean r300_is_sampler_format_supported(enum pipe_format format) { - return r300_translate_texformat(format, 0) != ~0; + return r300_translate_texformat(format, 0, TRUE) != ~0; } -static void r300_texture_setup_immutable_state(struct r300_screen* screen, - struct r300_texture* tex) +void r300_texture_setup_format_state(struct r300_screen *screen, + struct r300_texture_desc *desc, + unsigned level, + struct r300_texture_format_state *out) { - struct r300_texture_format_state* f = &tex->tx_format; - struct pipe_resource *pt = &tex->b.b; + struct pipe_resource *pt = &desc->b.b; boolean is_r500 = screen->caps.is_r500; + /* Mask out all the fields we change. */ + out->format0 = 0; + out->format1 &= ~R300_TX_FORMAT_TEX_COORD_TYPE_MASK; + out->format2 &= R500_TXFORMAT_MSB; + out->tile_config = 0; + /* Set sampler state. */ - f->format0 = R300_TX_WIDTH((pt->width0 - 1) & 0x7ff) | - R300_TX_HEIGHT((pt->height0 - 1) & 0x7ff); + out->format0 = + R300_TX_WIDTH((u_minify(desc->width0, level) - 1) & 0x7ff) | + R300_TX_HEIGHT((u_minify(desc->height0, level) - 1) & 0x7ff) | + R300_TX_DEPTH(util_logbase2(u_minify(desc->depth0, level)) & 0xf); - if (tex->uses_pitch) { + if (desc->uses_stride_addressing) { /* rectangles love this */ - f->format0 |= R300_TX_PITCH_EN; - f->format2 = (tex->pitch[0] - 1) & 0x1fff; - } else { - /* power of two textures (3D, mipmaps, and no pitch) */ - f->format0 |= R300_TX_DEPTH(util_logbase2(pt->depth0) & 0xf); + out->format0 |= R300_TX_PITCH_EN; + out->format2 = (desc->stride_in_pixels[level] - 1) & 0x1fff; } - f->format1 = 0; if (pt->target == PIPE_TEXTURE_CUBE) { - f->format1 |= R300_TX_FORMAT_CUBIC_MAP; + out->format1 |= R300_TX_FORMAT_CUBIC_MAP; } if (pt->target == PIPE_TEXTURE_3D) { - f->format1 |= R300_TX_FORMAT_3D; + out->format1 |= R300_TX_FORMAT_3D; } /* large textures on r500 */ if (is_r500) { - if (pt->width0 > 2048) { - f->format2 |= R500_TXWIDTH_BIT11; + if (desc->width0 > 2048) { + out->format2 |= R500_TXWIDTH_BIT11; } - if (pt->height0 > 2048) { - f->format2 |= R500_TXHEIGHT_BIT11; + if (desc->height0 > 2048) { + out->format2 |= R500_TXHEIGHT_BIT11; } } - f->tile_config = R300_TXO_MACRO_TILE(tex->macrotile) | - R300_TXO_MICRO_TILE(tex->microtile); - - SCREEN_DBG(screen, DBG_TEX, "r300: Set texture state (%dx%d, %d levels)\n", - pt->width0, pt->height0, pt->last_level); + out->tile_config = R300_TXO_MACRO_TILE(desc->macrotile[level]) | + R300_TXO_MICRO_TILE(desc->microtile); } static void r300_texture_setup_fb_state(struct r300_screen* screen, @@ -611,23 +628,23 @@ static void r300_texture_setup_fb_state(struct r300_screen* screen, unsigned i; /* Set framebuffer state. */ - if (util_format_is_depth_or_stencil(tex->b.b.format)) { - for (i = 0; i <= tex->b.b.last_level; i++) { - tex->fb_state.depthpitch[i] = - tex->pitch[i] | - R300_DEPTHMACROTILE(tex->mip_macrotile[i]) | - R300_DEPTHMICROTILE(tex->microtile); + if (util_format_is_depth_or_stencil(tex->desc.b.b.format)) { + for (i = 0; i <= tex->desc.b.b.last_level; i++) { + tex->fb_state.pitch[i] = + tex->desc.stride_in_pixels[i] | + R300_DEPTHMACROTILE(tex->desc.macrotile[i]) | + R300_DEPTHMICROTILE(tex->desc.microtile); } - tex->fb_state.zb_format = r300_translate_zsformat(tex->b.b.format); + tex->fb_state.format = r300_translate_zsformat(tex->desc.b.b.format); } else { - for (i = 0; i <= tex->b.b.last_level; i++) { - tex->fb_state.colorpitch[i] = - tex->pitch[i] | - r300_translate_colorformat(tex->b.b.format) | - R300_COLOR_TILE(tex->mip_macrotile[i]) | - R300_COLOR_MICROTILE(tex->microtile); + for (i = 0; i <= tex->desc.b.b.last_level; i++) { + tex->fb_state.pitch[i] = + tex->desc.stride_in_pixels[i] | + r300_translate_colorformat(tex->desc.b.b.format) | + R300_COLOR_TILE(tex->desc.macrotile[i]) | + R300_COLOR_MICROTILE(tex->desc.microtile); } - tex->fb_state.us_out_fmt = r300_translate_out_fmt(tex->b.b.format); + tex->fb_state.format = r300_translate_out_fmt(tex->desc.b.b.format); } } @@ -637,242 +654,16 @@ void r300_texture_reinterpret_format(struct pipe_screen *screen, { struct r300_screen *r300screen = r300_screen(screen); - SCREEN_DBG(r300screen, DBG_TEX, "r300: Reinterpreting format: %s -> %s\n", - util_format_name(tex->format), util_format_name(new_format)); + SCREEN_DBG(r300screen, DBG_TEX, + "r300: texture_reinterpret_format: %s -> %s\n", + util_format_short_name(tex->format), + util_format_short_name(new_format)); tex->format = new_format; r300_texture_setup_fb_state(r300_screen(screen), r300_texture(tex)); } -unsigned r300_texture_get_offset(struct r300_texture* tex, unsigned level, - unsigned zslice, unsigned face) -{ - unsigned offset = tex->offset[level]; - - switch (tex->b.b.target) { - case PIPE_TEXTURE_3D: - assert(face == 0); - return offset + zslice * tex->layer_size[level]; - - case PIPE_TEXTURE_CUBE: - assert(zslice == 0); - return offset + face * tex->layer_size[level]; - - default: - assert(zslice == 0 && face == 0); - return offset; - } -} - -/** - * Return the width (dim==TILE_WIDTH) or height (dim==TILE_HEIGHT) of one tile - * of the given texture. - */ -static unsigned r300_texture_get_tile_size(struct r300_texture* tex, - int dim, boolean macrotile) -{ - unsigned pixsize, tile_size; - - pixsize = util_format_get_blocksize(tex->b.b.format); - tile_size = microblock_table[util_logbase2(pixsize)][tex->microtile][dim]; - - if (macrotile) { - tile_size *= 8; - } - - assert(tile_size); - return tile_size; -} - -/* Return true if macrotiling should be enabled on the miplevel. */ -static boolean r300_texture_macro_switch(struct r300_texture *tex, - unsigned level, - boolean rv350_mode, - int dim) -{ - unsigned tile, texdim; - - tile = r300_texture_get_tile_size(tex, dim, TRUE); - if (dim == TILE_WIDTH) { - texdim = u_minify(tex->b.b.width0, level); - } else { - texdim = u_minify(tex->b.b.height0, level); - } - - /* See TX_FILTER1_n.MACRO_SWITCH. */ - if (rv350_mode) { - return texdim >= tile; - } else { - return texdim > tile; - } -} - -/** - * Return the stride, in bytes, of the texture images of the given texture - * at the given level. - */ -unsigned r300_texture_get_stride(struct r300_screen* screen, - struct r300_texture* tex, unsigned level) -{ - unsigned tile_width, width; - - if (tex->stride_override) - return tex->stride_override; - - /* Check the level. */ - if (level > tex->b.b.last_level) { - SCREEN_DBG(screen, DBG_TEX, "%s: level (%u) > last_level (%u)\n", - __FUNCTION__, level, tex->b.b.last_level); - return 0; - } - - width = u_minify(tex->b.b.width0, level); - - if (r300_format_is_plain(tex->b.b.format)) { - tile_width = r300_texture_get_tile_size(tex, TILE_WIDTH, - tex->mip_macrotile[level]); - width = align(width, tile_width); - - return util_format_get_stride(tex->b.b.format, width); - } else { - return align(util_format_get_stride(tex->b.b.format, width), 32); - } -} - -static unsigned r300_texture_get_nblocksy(struct r300_texture* tex, - unsigned level) -{ - unsigned height, tile_height; - - height = u_minify(tex->b.b.height0, level); - - if (r300_format_is_plain(tex->b.b.format)) { - tile_height = r300_texture_get_tile_size(tex, TILE_HEIGHT, - tex->mip_macrotile[level]); - height = align(height, tile_height); - - /* This is needed for the kernel checker, unfortunately. */ - height = util_next_power_of_two(height); - } - - return util_format_get_nblocksy(tex->b.b.format, height); -} - -static void r300_texture_3d_fix_mipmapping(struct r300_screen *screen, - struct r300_texture *tex) -{ - /* The kernels <= 2.6.34-rc4 compute the size of mipmapped 3D textures - * incorrectly. This is a workaround to prevent CS from being rejected. */ - - unsigned i, size; - - if (!screen->rws->get_value(screen->rws, R300_VID_DRM_2_3_0) && - tex->b.b.target == PIPE_TEXTURE_3D && - tex->b.b.last_level > 0) { - size = 0; - - for (i = 0; i <= tex->b.b.last_level; i++) { - size += r300_texture_get_stride(screen, tex, i) * - r300_texture_get_nblocksy(tex, i); - } - - size *= tex->b.b.depth0; - tex->size = size; - } -} - -static void r300_setup_miptree(struct r300_screen* screen, - struct r300_texture* tex) -{ - struct pipe_resource* base = &tex->b.b; - unsigned stride, size, layer_size, nblocksy, i; - boolean rv350_mode = screen->caps.is_rv350; - - SCREEN_DBG(screen, DBG_TEX, "r300: Making miptree for texture, format %s\n", - util_format_name(base->format)); - - for (i = 0; i <= base->last_level; i++) { - /* Let's see if this miplevel can be macrotiled. */ - tex->mip_macrotile[i] = - (tex->macrotile == R300_BUFFER_TILED && - r300_texture_macro_switch(tex, i, rv350_mode, TILE_WIDTH) && - r300_texture_macro_switch(tex, i, rv350_mode, TILE_HEIGHT)) ? - R300_BUFFER_TILED : R300_BUFFER_LINEAR; - - stride = r300_texture_get_stride(screen, tex, i); - nblocksy = r300_texture_get_nblocksy(tex, i); - layer_size = stride * nblocksy; - - if (base->target == PIPE_TEXTURE_CUBE) - size = layer_size * 6; - else - size = layer_size * u_minify(base->depth0, i); - - tex->offset[i] = tex->size; - tex->size = tex->offset[i] + size; - tex->layer_size[i] = layer_size; - tex->pitch[i] = stride / util_format_get_blocksize(base->format); - - SCREEN_DBG(screen, DBG_TEX, "r300: Texture miptree: Level %d " - "(%dx%dx%d px, pitch %d bytes) %d bytes total, macrotiled %s\n", - i, u_minify(base->width0, i), u_minify(base->height0, i), - u_minify(base->depth0, i), stride, tex->size, - tex->mip_macrotile[i] ? "TRUE" : "FALSE"); - } -} - -static void r300_setup_flags(struct r300_texture* tex) -{ - tex->uses_pitch = !util_is_power_of_two(tex->b.b.width0) || - !util_is_power_of_two(tex->b.b.height0) || - tex->stride_override; -} - -static void r300_setup_tiling(struct pipe_screen *screen, - struct r300_texture *tex) -{ - struct r300_winsys_screen *rws = (struct r300_winsys_screen *)screen->winsys; - enum pipe_format format = tex->b.b.format; - boolean rv350_mode = r300_screen(screen)->caps.is_rv350; - boolean is_zb = util_format_is_depth_or_stencil(format); - boolean dbg_no_tiling = SCREEN_DBG_ON(r300_screen(screen), DBG_NO_TILING); - - if (!r300_format_is_plain(format)) { - return; - } - - /* If height == 1, disable microtiling except for zbuffer. */ - if (!is_zb && (tex->b.b.height0 == 1 || dbg_no_tiling)) { - return; - } - - /* Set microtiling. */ - switch (util_format_get_blocksize(format)) { - case 1: - case 4: - tex->microtile = R300_BUFFER_TILED; - break; - - case 2: - case 8: - if (rws->get_value(rws, R300_VID_SQUARE_TILING_SUPPORT)) { - tex->microtile = R300_BUFFER_SQUARETILED; - } - break; - } - - if (dbg_no_tiling) { - return; - } - - /* Set macrotiling. */ - if (r300_texture_macro_switch(tex, 0, rv350_mode, TILE_WIDTH) && - r300_texture_macro_switch(tex, 0, rv350_mode, TILE_HEIGHT)) { - tex->macrotile = R300_BUFFER_TILED; - } -} - static unsigned r300_texture_is_referenced(struct pipe_context *context, struct pipe_resource *texture, unsigned face, unsigned level) @@ -880,7 +671,8 @@ static unsigned r300_texture_is_referenced(struct pipe_context *context, struct r300_context *r300 = r300_context(context); struct r300_texture *rtex = (struct r300_texture *)texture; - if (r300->rws->is_buffer_referenced(r300->rws, rtex->buffer, R300_REF_CS)) + if (r300->rws->cs_is_buffer_referenced(r300->cs, + rtex->buffer, R300_REF_CS)) return PIPE_REFERENCED_FOR_READ | PIPE_REFERENCED_FOR_WRITE; return PIPE_UNREFERENCED; @@ -891,8 +683,16 @@ static void r300_texture_destroy(struct pipe_screen *screen, { struct r300_texture* tex = (struct r300_texture*)texture; struct r300_winsys_screen *rws = (struct r300_winsys_screen *)texture->screen->winsys; + int i; rws->buffer_reference(rws, &tex->buffer, NULL); + for (i = 0; i < R300_MAX_TEXTURE_LEVELS; i++) { + if (tex->hiz_mem[i]) + u_mmFreeMem(tex->hiz_mem[i]); + if (tex->zmask_mem[i]) + u_mmFreeMem(tex->zmask_mem[i]); + } + FREE(tex); } @@ -902,20 +702,16 @@ static boolean r300_texture_get_handle(struct pipe_screen* screen, { struct r300_winsys_screen *rws = (struct r300_winsys_screen *)screen->winsys; struct r300_texture* tex = (struct r300_texture*)texture; - unsigned stride; if (!tex) { return FALSE; } - stride = r300_texture_get_stride(r300_screen(screen), tex, 0); - - rws->buffer_get_handle(rws, tex->buffer, stride, whandle); - - return TRUE; + return rws->buffer_get_handle(rws, tex->buffer, + tex->desc.stride_in_bytes[0], whandle); } -struct u_resource_vtbl r300_texture_vtbl = +struct u_resource_vtbl r300_texture_vtbl = { r300_texture_get_handle, /* get_handle */ r300_texture_destroy, /* resource_destroy */ @@ -928,162 +724,208 @@ struct u_resource_vtbl r300_texture_vtbl = u_default_transfer_inline_write /* transfer_inline_write */ }; -/* Create a new texture. */ -struct pipe_resource* r300_texture_create(struct pipe_screen* screen, - const struct pipe_resource* base) +/* The common texture constructor. */ +static struct r300_texture* +r300_texture_create_object(struct r300_screen *rscreen, + const struct pipe_resource *base, + enum r300_buffer_tiling microtile, + enum r300_buffer_tiling macrotile, + unsigned stride_in_bytes_override, + unsigned max_buffer_size, + struct r300_winsys_buffer *buffer) { - struct r300_texture* tex = CALLOC_STRUCT(r300_texture); - struct r300_screen* rscreen = r300_screen(screen); - struct r300_winsys_screen *rws = (struct r300_winsys_screen *)screen->winsys; - + struct r300_winsys_screen *rws = rscreen->rws; + struct r300_texture *tex = CALLOC_STRUCT(r300_texture); if (!tex) { + if (buffer) + rws->buffer_reference(rws, &buffer, NULL); return NULL; } - tex->b.b = *base; - tex->b.vtbl = &r300_texture_vtbl; - pipe_reference_init(&tex->b.b.reference, 1); - tex->b.b.screen = screen; - - r300_setup_flags(tex); - if (!(base->flags & R300_RESOURCE_FLAG_TRANSFER) && - !(base->bind & PIPE_BIND_SCANOUT)) { - r300_setup_tiling(screen, tex); + /* Initialize the descriptor. */ + if (!r300_texture_desc_init(rscreen, &tex->desc, base, + microtile, macrotile, + stride_in_bytes_override, + max_buffer_size)) { + if (buffer) + rws->buffer_reference(rws, &buffer, NULL); + FREE(tex); + return NULL; } - r300_setup_miptree(rscreen, tex); - r300_texture_3d_fix_mipmapping(rscreen, tex); - r300_texture_setup_immutable_state(rscreen, tex); + /* Initialize the hardware state. */ + r300_texture_setup_format_state(rscreen, &tex->desc, 0, &tex->tx_format); r300_texture_setup_fb_state(rscreen, tex); - tex->buffer = rws->buffer_create(rws, 2048, - PIPE_BIND_SAMPLER_VIEW, /* XXX */ - tex->size); - rws->buffer_set_tiling(rws, tex->buffer, - tex->pitch[0], - tex->microtile, - tex->macrotile); + tex->desc.b.vtbl = &r300_texture_vtbl; + pipe_reference_init(&tex->desc.b.b.reference, 1); + tex->domain = base->flags & R300_RESOURCE_FLAG_TRANSFER ? + R300_DOMAIN_GTT : + R300_DOMAIN_VRAM | R300_DOMAIN_GTT; + tex->buffer = buffer; + /* Create the backing buffer if needed. */ if (!tex->buffer) { - FREE(tex); - return NULL; + tex->buffer = rws->buffer_create(rws, tex->desc.size_in_bytes, 2048, + base->bind, base->usage, tex->domain); + + if (!tex->buffer) { + FREE(tex); + return NULL; + } } - return (struct pipe_resource*)tex; + rws->buffer_set_tiling(rws, tex->buffer, + tex->desc.microtile, tex->desc.macrotile[0], + tex->desc.stride_in_bytes[0]); + + return tex; } -/* Not required to implement u_resource_vtbl, consider moving to another file: - */ -struct pipe_surface* r300_get_tex_surface(struct pipe_screen* screen, - struct pipe_resource* texture, - unsigned face, - unsigned level, - unsigned zslice, - unsigned flags) +/* Create a new texture. */ +struct pipe_resource *r300_texture_create(struct pipe_screen *screen, + const struct pipe_resource *base) { - struct r300_texture* tex = r300_texture(texture); - struct pipe_surface* surface = CALLOC_STRUCT(pipe_surface); - unsigned offset; + struct r300_screen *rscreen = r300_screen(screen); + enum r300_buffer_tiling microtile, macrotile; - offset = r300_texture_get_offset(tex, level, zslice, face); - - if (surface) { - pipe_reference_init(&surface->reference, 1); - pipe_resource_reference(&surface->texture, texture); - surface->format = texture->format; - surface->width = u_minify(texture->width0, level); - surface->height = u_minify(texture->height0, level); - surface->offset = offset; - surface->usage = flags; - surface->zslice = zslice; - surface->texture = texture; - surface->face = face; - surface->level = level; + if ((base->flags & R300_RESOURCE_FLAG_TRANSFER) || + (base->bind & PIPE_BIND_SCANOUT)) { + microtile = R300_BUFFER_LINEAR; + macrotile = R300_BUFFER_LINEAR; + } else { + microtile = R300_BUFFER_SELECT_LAYOUT; + macrotile = R300_BUFFER_SELECT_LAYOUT; } - return surface; -} - -/* Not required to implement u_resource_vtbl, consider moving to another file: - */ -void r300_tex_surface_destroy(struct pipe_surface* s) -{ - pipe_resource_reference(&s->texture, NULL); - FREE(s); + return (struct pipe_resource*) + r300_texture_create_object(rscreen, base, microtile, macrotile, + 0, 0, NULL); } -struct pipe_resource* -r300_texture_from_handle(struct pipe_screen* screen, - const struct pipe_resource* base, - struct winsys_handle *whandle) +struct pipe_resource *r300_texture_from_handle(struct pipe_screen *screen, + const struct pipe_resource *base, + struct winsys_handle *whandle) { struct r300_winsys_screen *rws = (struct r300_winsys_screen*)screen->winsys; - struct r300_screen* rscreen = r300_screen(screen); + struct r300_screen *rscreen = r300_screen(screen); struct r300_winsys_buffer *buffer; - struct r300_texture* tex; - unsigned stride; - boolean override_zb_flags; + enum r300_buffer_tiling microtile, macrotile; + unsigned stride, size; /* Support only 2D textures without mipmaps */ - if (base->target != PIPE_TEXTURE_2D || + if ((base->target != PIPE_TEXTURE_2D && + base->target != PIPE_TEXTURE_RECT) || base->depth0 != 1 || base->last_level != 0) { return NULL; } - buffer = rws->buffer_from_handle(rws, screen, whandle, &stride); - if (!buffer) { - return NULL; - } - - tex = CALLOC_STRUCT(r300_texture); - if (!tex) { + buffer = rws->buffer_from_handle(rws, whandle, &stride, &size); + if (!buffer) return NULL; - } - tex->b.b = *base; - tex->b.vtbl = &r300_texture_vtbl; - pipe_reference_init(&tex->b.b.reference, 1); - tex->b.b.screen = screen; + rws->buffer_get_tiling(rws, buffer, µtile, ¯otile); - tex->stride_override = stride; - - /* one ref already taken */ - tex->buffer = buffer; - - rws->buffer_get_tiling(rws, buffer, &tex->microtile, &tex->macrotile); - r300_setup_flags(tex); - - /* Enforce microtiled zbuffer. */ - override_zb_flags = util_format_is_depth_or_stencil(base->format) && - tex->microtile == R300_BUFFER_LINEAR; - - if (override_zb_flags) { + /* Enforce a microtiled zbuffer. */ + if (util_format_is_depth_or_stencil(base->format) && + microtile == R300_BUFFER_LINEAR) { switch (util_format_get_blocksize(base->format)) { case 4: - tex->microtile = R300_BUFFER_TILED; + microtile = R300_BUFFER_TILED; break; case 2: - if (rws->get_value(rws, R300_VID_SQUARE_TILING_SUPPORT)) { - tex->microtile = R300_BUFFER_SQUARETILED; - break; - } - /* Pass through. */ - - default: - override_zb_flags = FALSE; + if (rws->get_value(rws, R300_VID_SQUARE_TILING_SUPPORT)) + microtile = R300_BUFFER_SQUARETILED; + break; } } - r300_setup_miptree(rscreen, tex); - r300_texture_setup_immutable_state(rscreen, tex); - r300_texture_setup_fb_state(rscreen, tex); + return (struct pipe_resource*) + r300_texture_create_object(rscreen, base, microtile, macrotile, + stride, size, buffer); +} - if (override_zb_flags) { - rws->buffer_set_tiling(rws, tex->buffer, - tex->pitch[0], - tex->microtile, - tex->macrotile); +/* Not required to implement u_resource_vtbl, consider moving to another file: + */ +struct pipe_surface* r300_get_tex_surface(struct pipe_screen* screen, + struct pipe_resource* texture, + unsigned face, + unsigned level, + unsigned zslice, + unsigned flags) +{ + struct r300_texture* tex = r300_texture(texture); + struct r300_surface* surface = CALLOC_STRUCT(r300_surface); + + if (surface) { + uint32_t offset, tile_height; + + pipe_reference_init(&surface->base.reference, 1); + pipe_resource_reference(&surface->base.texture, texture); + surface->base.format = texture->format; + surface->base.width = u_minify(texture->width0, level); + surface->base.height = u_minify(texture->height0, level); + surface->base.usage = flags; + surface->base.zslice = zslice; + surface->base.face = face; + surface->base.level = level; + + surface->buffer = tex->buffer; + + /* Prefer VRAM if there are multiple domains to choose from. */ + surface->domain = tex->domain; + if (surface->domain & R300_DOMAIN_VRAM) + surface->domain &= ~R300_DOMAIN_GTT; + + surface->offset = r300_texture_get_offset(&tex->desc, + level, zslice, face); + surface->pitch = tex->fb_state.pitch[level]; + surface->format = tex->fb_state.format; + + /* Parameters for the CBZB clear. */ + surface->cbzb_allowed = tex->desc.cbzb_allowed[level]; + surface->cbzb_width = align(surface->base.width, 64); + + /* Height must be aligned to the size of a tile. */ + tile_height = r300_get_pixel_alignment(tex->desc.b.b.format, + tex->desc.b.b.nr_samples, + tex->desc.microtile, + tex->desc.macrotile[level], + DIM_HEIGHT); + + surface->cbzb_height = align((surface->base.height + 1) / 2, + tile_height); + + /* Offset must be aligned to 2K and must point at the beginning + * of a scanline. */ + offset = surface->offset + + tex->desc.stride_in_bytes[level] * surface->cbzb_height; + surface->cbzb_midpoint_offset = offset & ~2047; + + surface->cbzb_pitch = surface->pitch & 0x1ffffc; + + if (util_format_get_blocksizebits(surface->base.format) == 32) + surface->cbzb_format = R300_DEPTHFORMAT_24BIT_INT_Z_8BIT_STENCIL; + else + surface->cbzb_format = R300_DEPTHFORMAT_16BIT_INT_Z; + + SCREEN_DBG(r300_screen(screen), DBG_CBZB, + "CBZB Allowed: %s, Dim: %ix%i, Misalignment: %i, Micro: %s, Macro: %s\n", + surface->cbzb_allowed ? "YES" : " NO", + surface->cbzb_width, surface->cbzb_height, + offset & 2047, + tex->desc.microtile ? "YES" : " NO", + tex->desc.macrotile[level] ? "YES" : " NO"); } - return (struct pipe_resource*)tex; + + return &surface->base; +} + +/* Not required to implement u_resource_vtbl, consider moving to another file: + */ +void r300_tex_surface_destroy(struct pipe_surface* s) +{ + pipe_resource_reference(&s->texture, NULL); + FREE(s); } diff --git a/src/gallium/drivers/r300/r300_texture.h b/src/gallium/drivers/r300/r300_texture.h index 453d42b188f..c4588a0c90b 100644 --- a/src/gallium/drivers/r300/r300_texture.h +++ b/src/gallium/drivers/r300/r300_texture.h @@ -23,21 +23,26 @@ #ifndef R300_TEXTURE_H #define R300_TEXTURE_H -#include "util/u_format.h" - +#include "pipe/p_compiler.h" +#include "pipe/p_format.h" + +struct pipe_screen; +struct pipe_resource; +struct winsys_handle; +struct r300_texture_format_state; +struct r300_texture_desc; struct r300_texture; +struct r300_screen; + +unsigned r300_get_swizzle_combined(const unsigned char *swizzle_format, + const unsigned char *swizzle_view); uint32_t r300_translate_texformat(enum pipe_format format, - const unsigned char *swizzle); + const unsigned char *swizzle_view, + boolean is_r500); uint32_t r500_tx_format_msb_bit(enum pipe_format format); -unsigned r300_texture_get_stride(struct r300_screen* screen, - struct r300_texture* tex, unsigned level); - -unsigned r300_texture_get_offset(struct r300_texture* tex, unsigned level, - unsigned zslice, unsigned face); - void r300_texture_reinterpret_format(struct pipe_screen *screen, struct pipe_resource *tex, enum pipe_format new_format); @@ -48,6 +53,10 @@ boolean r300_is_zs_format_supported(enum pipe_format format); boolean r300_is_sampler_format_supported(enum pipe_format format); +void r300_texture_setup_format_state(struct r300_screen *screen, + struct r300_texture_desc *desc, + unsigned level, + struct r300_texture_format_state *out); struct pipe_resource* r300_texture_from_handle(struct pipe_screen* screen, diff --git a/src/gallium/drivers/r300/r300_texture_desc.c b/src/gallium/drivers/r300/r300_texture_desc.c new file mode 100644 index 00000000000..543d0fdc15b --- /dev/null +++ b/src/gallium/drivers/r300/r300_texture_desc.c @@ -0,0 +1,495 @@ +/* + * Copyright 2008 Corbin Simpson <[email protected]> + * Copyright 2010 Marek Olšák <[email protected]> + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. */ + +#include "r300_texture_desc.h" + +#include "r300_context.h" +#include "r300_winsys.h" + +#include "util/u_format.h" + +/* Returns the number of pixels that the texture should be aligned to + * in the given dimension. */ +unsigned r300_get_pixel_alignment(enum pipe_format format, + unsigned num_samples, + enum r300_buffer_tiling microtile, + enum r300_buffer_tiling macrotile, + enum r300_dim dim) +{ + static const unsigned table[2][5][3][2] = + { + { + /* Macro: linear linear linear + Micro: linear tiled square-tiled */ + {{ 32, 1}, { 8, 4}, { 0, 0}}, /* 8 bits per pixel */ + {{ 16, 1}, { 8, 2}, { 4, 4}}, /* 16 bits per pixel */ + {{ 8, 1}, { 4, 2}, { 0, 0}}, /* 32 bits per pixel */ + {{ 4, 1}, { 2, 2}, { 0, 0}}, /* 64 bits per pixel */ + {{ 2, 1}, { 0, 0}, { 0, 0}} /* 128 bits per pixel */ + }, + { + /* Macro: tiled tiled tiled + Micro: linear tiled square-tiled */ + {{256, 8}, {64, 32}, { 0, 0}}, /* 8 bits per pixel */ + {{128, 8}, {64, 16}, {32, 32}}, /* 16 bits per pixel */ + {{ 64, 8}, {32, 16}, { 0, 0}}, /* 32 bits per pixel */ + {{ 32, 8}, {16, 16}, { 0, 0}}, /* 64 bits per pixel */ + {{ 16, 8}, { 0, 0}, { 0, 0}} /* 128 bits per pixel */ + } + }; + static const unsigned aa_block[2] = {4, 8}; + unsigned tile = 0; + unsigned pixsize = util_format_get_blocksize(format); + + assert(macrotile <= R300_BUFFER_TILED); + assert(microtile <= R300_BUFFER_SQUARETILED); + assert(pixsize <= 16); + assert(dim <= DIM_HEIGHT); + + if (num_samples > 1) { + /* Multisampled textures have their own alignment scheme. */ + if (pixsize == 4) + tile = aa_block[dim]; + /* XXX FP16 AA. */ + } else { + /* Standard alignment. */ + tile = table[macrotile][util_logbase2(pixsize)][microtile][dim]; + } + + assert(tile); + return tile; +} + +/* Return true if macrotiling should be enabled on the miplevel. */ +static boolean r300_texture_macro_switch(struct r300_texture_desc *desc, + unsigned level, + boolean rv350_mode, + enum r300_dim dim) +{ + unsigned tile, texdim; + + tile = r300_get_pixel_alignment(desc->b.b.format, desc->b.b.nr_samples, + desc->microtile, R300_BUFFER_TILED, dim); + if (dim == DIM_WIDTH) { + texdim = u_minify(desc->width0, level); + } else { + texdim = u_minify(desc->height0, level); + } + + /* See TX_FILTER1_n.MACRO_SWITCH. */ + if (rv350_mode) { + return texdim >= tile; + } else { + return texdim > tile; + } +} + +/** + * Return the stride, in bytes, of the texture image of the given texture + * at the given level. + */ +static unsigned r300_texture_get_stride(struct r300_screen *screen, + struct r300_texture_desc *desc, + unsigned level) +{ + unsigned tile_width, width, stride; + + if (desc->stride_in_bytes_override) + return desc->stride_in_bytes_override; + + /* Check the level. */ + if (level > desc->b.b.last_level) { + SCREEN_DBG(screen, DBG_TEX, "%s: level (%u) > last_level (%u)\n", + __FUNCTION__, level, desc->b.b.last_level); + return 0; + } + + width = u_minify(desc->width0, level); + + if (util_format_is_plain(desc->b.b.format)) { + tile_width = r300_get_pixel_alignment(desc->b.b.format, + desc->b.b.nr_samples, + desc->microtile, + desc->macrotile[level], + DIM_WIDTH); + width = align(width, tile_width); + + stride = util_format_get_stride(desc->b.b.format, width); + + /* Some IGPs need a minimum stride of 64 bytes, hmm... */ + if (!desc->macrotile[level] && + (screen->caps.family == CHIP_FAMILY_RS600 || + screen->caps.family == CHIP_FAMILY_RS690 || + screen->caps.family == CHIP_FAMILY_RS740)) { + unsigned min_stride; + + if (desc->microtile) { + unsigned tile_height = + r300_get_pixel_alignment(desc->b.b.format, + desc->b.b.nr_samples, + desc->microtile, + desc->macrotile[level], + DIM_HEIGHT); + + min_stride = 64 / tile_height; + } else { + min_stride = 64; + } + + return stride < min_stride ? min_stride : stride; + } + + /* The alignment to 32 bytes is sort of implied by the layout... */ + return stride; + } else { + return align(util_format_get_stride(desc->b.b.format, width), 32); + } +} + +static unsigned r300_texture_get_nblocksy(struct r300_texture_desc *desc, + unsigned level, + boolean *out_aligned_for_cbzb) +{ + unsigned height, tile_height; + + height = u_minify(desc->height0, level); + + if (util_format_is_plain(desc->b.b.format)) { + tile_height = r300_get_pixel_alignment(desc->b.b.format, + desc->b.b.nr_samples, + desc->microtile, + desc->macrotile[level], + DIM_HEIGHT); + height = align(height, tile_height); + + /* This is needed for the kernel checker, unfortunately. */ + if ((desc->b.b.target != PIPE_TEXTURE_1D && + desc->b.b.target != PIPE_TEXTURE_2D && + desc->b.b.target != PIPE_TEXTURE_RECT) || + desc->b.b.last_level != 0) { + height = util_next_power_of_two(height); + } + + /* See if the CBZB clear can be used on the buffer, + * taking the texture size into account. */ + if (out_aligned_for_cbzb) { + if (desc->macrotile[level]) { + /* When clearing, the layer (width*height) is horizontally split + * into two, and the upper and lower halves are cleared by the CB + * and ZB units, respectively. Therefore, the number of macrotiles + * in the Y direction must be even. */ + + /* Align the height so that there is an even number of macrotiles. + * Do so for 3 or more macrotiles in the Y direction. */ + if (level == 0 && desc->b.b.last_level == 0 && + (desc->b.b.target == PIPE_TEXTURE_1D || + desc->b.b.target == PIPE_TEXTURE_2D || + desc->b.b.target == PIPE_TEXTURE_RECT) && + height >= tile_height * 3) { + height = align(height, tile_height * 2); + } + + *out_aligned_for_cbzb = height % (tile_height * 2) == 0; + } else { + *out_aligned_for_cbzb = FALSE; + } + } + } + + return util_format_get_nblocksy(desc->b.b.format, height); +} + +static void r300_texture_3d_fix_mipmapping(struct r300_screen *screen, + struct r300_texture_desc *desc) +{ + /* The kernels <= 2.6.34-rc4 compute the size of mipmapped 3D textures + * incorrectly. This is a workaround to prevent CS from being rejected. */ + + unsigned i, size; + + if (!screen->rws->get_value(screen->rws, R300_VID_DRM_2_3_0) && + desc->b.b.target == PIPE_TEXTURE_3D && + desc->b.b.last_level > 0) { + size = 0; + + for (i = 0; i <= desc->b.b.last_level; i++) { + size += desc->stride_in_bytes[i] * + r300_texture_get_nblocksy(desc, i, FALSE); + } + + size *= desc->depth0; + desc->size_in_bytes = size; + } +} + +/* Get a width in pixels from a stride in bytes. */ +static unsigned stride_to_width(enum pipe_format format, + unsigned stride_in_bytes) +{ + return (stride_in_bytes / util_format_get_blocksize(format)) * + util_format_get_blockwidth(format); +} + +static void r300_setup_miptree(struct r300_screen *screen, + struct r300_texture_desc *desc, + boolean align_for_cbzb) +{ + struct pipe_resource *base = &desc->b.b; + unsigned stride, size, layer_size, nblocksy, i; + boolean rv350_mode = screen->caps.family >= CHIP_FAMILY_R350; + boolean aligned_for_cbzb; + + desc->size_in_bytes = 0; + + SCREEN_DBG(screen, DBG_TEXALLOC, + "r300: Making miptree for texture, format %s\n", + util_format_short_name(base->format)); + + for (i = 0; i <= base->last_level; i++) { + /* Let's see if this miplevel can be macrotiled. */ + desc->macrotile[i] = + (desc->macrotile[0] == R300_BUFFER_TILED && + r300_texture_macro_switch(desc, i, rv350_mode, DIM_WIDTH) && + r300_texture_macro_switch(desc, i, rv350_mode, DIM_HEIGHT)) ? + R300_BUFFER_TILED : R300_BUFFER_LINEAR; + + stride = r300_texture_get_stride(screen, desc, i); + + /* Compute the number of blocks in Y, see if the CBZB clear can be + * used on the texture. */ + aligned_for_cbzb = FALSE; + if (align_for_cbzb && desc->cbzb_allowed[i]) + nblocksy = r300_texture_get_nblocksy(desc, i, &aligned_for_cbzb); + else + nblocksy = r300_texture_get_nblocksy(desc, i, NULL); + + layer_size = stride * nblocksy; + + if (base->nr_samples) { + layer_size *= base->nr_samples; + } + + if (base->target == PIPE_TEXTURE_CUBE) + size = layer_size * 6; + else + size = layer_size * u_minify(desc->depth0, i); + + desc->offset_in_bytes[i] = desc->size_in_bytes; + desc->size_in_bytes = desc->offset_in_bytes[i] + size; + desc->layer_size_in_bytes[i] = layer_size; + desc->stride_in_bytes[i] = stride; + desc->stride_in_pixels[i] = stride_to_width(desc->b.b.format, stride); + desc->cbzb_allowed[i] = desc->cbzb_allowed[i] && aligned_for_cbzb; + + SCREEN_DBG(screen, DBG_TEXALLOC, "r300: Texture miptree: Level %d " + "(%dx%dx%d px, pitch %d bytes) %d bytes total, macrotiled %s\n", + i, u_minify(desc->width0, i), u_minify(desc->height0, i), + u_minify(desc->depth0, i), stride, desc->size_in_bytes, + desc->macrotile[i] ? "TRUE" : "FALSE"); + } +} + +static void r300_setup_flags(struct r300_texture_desc *desc) +{ + desc->uses_stride_addressing = + !util_is_power_of_two(desc->b.b.width0) || + (desc->stride_in_bytes_override && + stride_to_width(desc->b.b.format, + desc->stride_in_bytes_override) != desc->b.b.width0); + + desc->is_npot = + desc->uses_stride_addressing || + !util_is_power_of_two(desc->b.b.height0) || + !util_is_power_of_two(desc->b.b.depth0); +} + +static void r300_setup_cbzb_flags(struct r300_screen *rscreen, + struct r300_texture_desc *desc) +{ + unsigned i, bpp; + boolean first_level_valid; + + bpp = util_format_get_blocksizebits(desc->b.b.format); + + /* 1) The texture must be point-sampled, + * 2) The depth must be 16 or 32 bits. + * 3) If the midpoint ZB offset is not aligned to 2048, it returns garbage + * with certain texture sizes. Macrotiling ensures the alignment. */ + first_level_valid = desc->b.b.nr_samples <= 1 && + (bpp == 16 || bpp == 32) && + desc->macrotile[0]; + + if (SCREEN_DBG_ON(rscreen, DBG_NO_CBZB)) + first_level_valid = FALSE; + + for (i = 0; i <= desc->b.b.last_level; i++) + desc->cbzb_allowed[i] = first_level_valid && desc->macrotile[i]; +} + +static void r300_setup_tiling(struct r300_screen *screen, + struct r300_texture_desc *desc) +{ + struct r300_winsys_screen *rws = screen->rws; + enum pipe_format format = desc->b.b.format; + boolean rv350_mode = screen->caps.family >= CHIP_FAMILY_R350; + boolean is_zb = util_format_is_depth_or_stencil(format); + boolean dbg_no_tiling = SCREEN_DBG_ON(screen, DBG_NO_TILING); + + if (!util_format_is_plain(format)) { + return; + } + + /* If height == 1, disable microtiling except for zbuffer. */ + if (!is_zb && (desc->b.b.height0 == 1 || dbg_no_tiling)) { + return; + } + + /* Set microtiling. */ + switch (util_format_get_blocksize(format)) { + case 1: + case 4: + case 8: + desc->microtile = R300_BUFFER_TILED; + break; + + case 2: + if (rws->get_value(rws, R300_VID_SQUARE_TILING_SUPPORT)) { + desc->microtile = R300_BUFFER_SQUARETILED; + } + break; + } + + if (dbg_no_tiling) { + return; + } + + /* Set macrotiling. */ + if (r300_texture_macro_switch(desc, 0, rv350_mode, DIM_WIDTH) && + r300_texture_macro_switch(desc, 0, rv350_mode, DIM_HEIGHT)) { + desc->macrotile[0] = R300_BUFFER_TILED; + } +} + +static void r300_tex_print_info(struct r300_screen *rscreen, + struct r300_texture_desc *desc, + const char *func) +{ + fprintf(stderr, + "r300: %s: Macro: %s, Micro: %s, Pitch: %i, Dim: %ix%ix%i, " + "LastLevel: %i, Size: %i, Format: %s\n", + func, + desc->macrotile[0] ? "YES" : " NO", + desc->microtile ? "YES" : " NO", + desc->stride_in_pixels[0], + desc->b.b.width0, desc->b.b.height0, desc->b.b.depth0, + desc->b.b.last_level, desc->size_in_bytes, + util_format_short_name(desc->b.b.format)); +} + +boolean r300_texture_desc_init(struct r300_screen *rscreen, + struct r300_texture_desc *desc, + const struct pipe_resource *base, + enum r300_buffer_tiling microtile, + enum r300_buffer_tiling macrotile, + unsigned stride_in_bytes_override, + unsigned max_buffer_size) +{ + desc->b.b = *base; + desc->b.b.screen = &rscreen->screen; + desc->stride_in_bytes_override = stride_in_bytes_override; + desc->width0 = base->width0; + desc->height0 = base->height0; + desc->depth0 = base->depth0; + + r300_setup_flags(desc); + + /* Align a 3D NPOT texture to POT. */ + if (base->target == PIPE_TEXTURE_3D && desc->is_npot) { + desc->width0 = util_next_power_of_two(desc->width0); + desc->height0 = util_next_power_of_two(desc->height0); + desc->depth0 = util_next_power_of_two(desc->depth0); + } + + /* Setup tiling. */ + if (microtile == R300_BUFFER_SELECT_LAYOUT || + macrotile == R300_BUFFER_SELECT_LAYOUT) { + r300_setup_tiling(rscreen, desc); + } else { + desc->microtile = microtile; + desc->macrotile[0] = macrotile; + assert(desc->b.b.last_level == 0); + } + + r300_setup_cbzb_flags(rscreen, desc); + + /* Setup the miptree description. */ + r300_setup_miptree(rscreen, desc, TRUE); + /* If the required buffer size is larger the given max size, + * try again without the alignment for the CBZB clear. */ + if (max_buffer_size && desc->size_in_bytes > max_buffer_size) { + r300_setup_miptree(rscreen, desc, FALSE); + } + + r300_texture_3d_fix_mipmapping(rscreen, desc); + + if (max_buffer_size) { + /* Make sure the buffer we got is large enough. */ + if (desc->size_in_bytes > max_buffer_size) { + fprintf(stderr, "r300: texture_desc_init: The buffer is not " + "large enough. Got: %i, Need: %i, Info:\n", + max_buffer_size, desc->size_in_bytes); + r300_tex_print_info(rscreen, desc, "texture_desc_init"); + return FALSE; + } + + desc->buffer_size_in_bytes = max_buffer_size; + } else { + desc->buffer_size_in_bytes = desc->size_in_bytes; + } + + if (SCREEN_DBG_ON(rscreen, DBG_TEX)) + r300_tex_print_info(rscreen, desc, "texture_desc_init"); + + return TRUE; +} + +unsigned r300_texture_get_offset(struct r300_texture_desc *desc, + unsigned level, unsigned zslice, + unsigned face) +{ + unsigned offset = desc->offset_in_bytes[level]; + + switch (desc->b.b.target) { + case PIPE_TEXTURE_3D: + assert(face == 0); + return offset + zslice * desc->layer_size_in_bytes[level]; + + case PIPE_TEXTURE_CUBE: + assert(zslice == 0); + return offset + face * desc->layer_size_in_bytes[level]; + + default: + assert(zslice == 0 && face == 0); + return offset; + } +} diff --git a/src/gallium/drivers/r300/r300_texture_desc.h b/src/gallium/drivers/r300/r300_texture_desc.h new file mode 100644 index 00000000000..3d7fe1fb473 --- /dev/null +++ b/src/gallium/drivers/r300/r300_texture_desc.h @@ -0,0 +1,58 @@ +/* + * Copyright 2008 Corbin Simpson <[email protected]> + * Copyright 2010 Marek Olšák <[email protected]> + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, and/or sell copies of the Software, and to permit persons to whom + * the Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL + * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. */ + +#ifndef R300_TEXTURE_DESC_H +#define R300_TEXTURE_DESC_H + +#include "pipe/p_format.h" +#include "r300_defines.h" + +struct pipe_resource; +struct r300_screen; +struct r300_texture_desc; +struct r300_texture; + +enum r300_dim { + DIM_WIDTH = 0, + DIM_HEIGHT = 1 +}; + +unsigned r300_get_pixel_alignment(enum pipe_format format, + unsigned num_samples, + enum r300_buffer_tiling microtile, + enum r300_buffer_tiling macrotile, + enum r300_dim dim); + +boolean r300_texture_desc_init(struct r300_screen *rscreen, + struct r300_texture_desc *desc, + const struct pipe_resource *base, + enum r300_buffer_tiling microtile, + enum r300_buffer_tiling macrotile, + unsigned stride_in_bytes_override, + unsigned max_buffer_size); + +unsigned r300_texture_get_offset(struct r300_texture_desc *desc, + unsigned level, unsigned zslice, + unsigned face); + +#endif diff --git a/src/gallium/drivers/r300/r300_tgsi_to_rc.c b/src/gallium/drivers/r300/r300_tgsi_to_rc.c index 89f39af9761..a4911b9a2a6 100644 --- a/src/gallium/drivers/r300/r300_tgsi_to_rc.c +++ b/src/gallium/drivers/r300/r300_tgsi_to_rc.c @@ -71,7 +71,7 @@ static unsigned translate_opcode(unsigned opcode) case TGSI_OPCODE_COS: return RC_OPCODE_COS; case TGSI_OPCODE_DDX: return RC_OPCODE_DDX; case TGSI_OPCODE_DDY: return RC_OPCODE_DDY; - /* case TGSI_OPCODE_KILP: return RC_OPCODE_KILP; */ + case TGSI_OPCODE_KILP: return RC_OPCODE_KILP; /* case TGSI_OPCODE_PK2H: return RC_OPCODE_PK2H; */ /* case TGSI_OPCODE_PK2US: return RC_OPCODE_PK2US; */ /* case TGSI_OPCODE_PK4B: return RC_OPCODE_PK4B; */ @@ -97,20 +97,20 @@ static unsigned translate_opcode(unsigned opcode) /* case TGSI_OPCODE_BRA: return RC_OPCODE_BRA; */ /* case TGSI_OPCODE_CAL: return RC_OPCODE_CAL; */ /* case TGSI_OPCODE_RET: return RC_OPCODE_RET; */ - /* case TGSI_OPCODE_SSG: return RC_OPCODE_SSG; */ + case TGSI_OPCODE_SSG: return RC_OPCODE_SSG; case TGSI_OPCODE_CMP: return RC_OPCODE_CMP; case TGSI_OPCODE_SCS: return RC_OPCODE_SCS; case TGSI_OPCODE_TXB: return RC_OPCODE_TXB; /* case TGSI_OPCODE_NRM: return RC_OPCODE_NRM; */ /* case TGSI_OPCODE_DIV: return RC_OPCODE_DIV; */ - /* case TGSI_OPCODE_DP2: return RC_OPCODE_DP2; */ + case TGSI_OPCODE_DP2: return RC_OPCODE_DP2; case TGSI_OPCODE_TXL: return RC_OPCODE_TXL; - /* case TGSI_OPCODE_BRK: return RC_OPCODE_BRK; */ + case TGSI_OPCODE_BRK: return RC_OPCODE_BRK; case TGSI_OPCODE_IF: return RC_OPCODE_IF; - /* case TGSI_OPCODE_LOOP: return RC_OPCODE_LOOP; */ + case TGSI_OPCODE_BGNLOOP: return RC_OPCODE_BGNLOOP; case TGSI_OPCODE_ELSE: return RC_OPCODE_ELSE; case TGSI_OPCODE_ENDIF: return RC_OPCODE_ENDIF; - /* case TGSI_OPCODE_ENDLOOP: return RC_OPCODE_ENDLOOP; */ + case TGSI_OPCODE_ENDLOOP: return RC_OPCODE_ENDLOOP; /* case TGSI_OPCODE_PUSHA: return RC_OPCODE_PUSHA; */ /* case TGSI_OPCODE_POPA: return RC_OPCODE_POPA; */ case TGSI_OPCODE_CEIL: return RC_OPCODE_CEIL; @@ -126,7 +126,7 @@ static unsigned translate_opcode(unsigned opcode) /* case TGSI_OPCODE_SAD: return RC_OPCODE_SAD; */ /* case TGSI_OPCODE_TXF: return RC_OPCODE_TXF; */ /* case TGSI_OPCODE_TXQ: return RC_OPCODE_TXQ; */ - /* case TGSI_OPCODE_CONT: return RC_OPCODE_CONT; */ + case TGSI_OPCODE_CONT: return RC_OPCODE_CONT; /* case TGSI_OPCODE_EMIT: return RC_OPCODE_EMIT; */ /* case TGSI_OPCODE_ENDPRIM: return RC_OPCODE_ENDPRIM; */ /* case TGSI_OPCODE_BGNLOOP2: return RC_OPCODE_BGNLOOP2; */ diff --git a/src/gallium/drivers/r300/r300_transfer.c b/src/gallium/drivers/r300/r300_transfer.c index 0dae9ef98b7..e9333b35ef5 100644 --- a/src/gallium/drivers/r300/r300_transfer.c +++ b/src/gallium/drivers/r300/r300_transfer.c @@ -21,12 +21,9 @@ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE * USE OR OTHER DEALINGS IN THE SOFTWARE. */ -#include "r300_context.h" #include "r300_transfer.h" -#include "r300_texture.h" -#include "r300_screen.h" - -#include "r300_winsys.h" +#include "r300_texture_desc.h" +#include "r300_screen_buffer.h" #include "util/u_memory.h" #include "util/u_format.h" @@ -38,11 +35,8 @@ struct r300_transfer { /* Offset from start of buffer. */ unsigned offset; - /* Detiled texture. */ - struct r300_texture *detiled_texture; - - /* Transfer and format flags. */ - unsigned render_target_usage; + /* Linear texture. */ + struct r300_texture *linear_texture; }; /* Convenience cast wrapper. */ @@ -56,63 +50,38 @@ r300_transfer(struct pipe_transfer* transfer) static void r300_copy_from_tiled_texture(struct pipe_context *ctx, struct r300_transfer *r300transfer) { - struct pipe_screen *screen = ctx->screen; struct pipe_transfer *transfer = (struct pipe_transfer*)r300transfer; struct pipe_resource *tex = transfer->resource; - struct pipe_surface *src, *dst; - - src = screen->get_tex_surface(screen, tex, - transfer->sr.face, - transfer->sr.level, - transfer->box.z, - PIPE_BIND_BLIT_SOURCE); + struct pipe_subresource subdst; - dst = screen->get_tex_surface(screen, &r300transfer->detiled_texture->b.b, - 0, 0, 0, - PIPE_BIND_BLIT_DESTINATION); + subdst.face = 0; + subdst.level = 0; - ctx->surface_copy(ctx, dst, 0, 0, src, - transfer->box.x, transfer->box.y, - transfer->box.width, transfer->box.height); - - pipe_surface_reference(&src, NULL); - pipe_surface_reference(&dst, NULL); + ctx->resource_copy_region(ctx, &r300transfer->linear_texture->desc.b.b, subdst, + 0, 0, 0, + tex, transfer->sr, + transfer->box.x, transfer->box.y, transfer->box.z, + transfer->box.width, transfer->box.height); } /* Copy a detiled texture to a tiled one. */ static void r300_copy_into_tiled_texture(struct pipe_context *ctx, struct r300_transfer *r300transfer) { - struct pipe_screen *screen = ctx->screen; struct pipe_transfer *transfer = (struct pipe_transfer*)r300transfer; struct pipe_resource *tex = transfer->resource; - struct pipe_surface *src, *dst; - - src = screen->get_tex_surface(screen, &r300transfer->detiled_texture->b.b, - 0, 0, 0, - PIPE_BIND_BLIT_SOURCE); + struct pipe_subresource subsrc; - dst = screen->get_tex_surface(screen, tex, - transfer->sr.face, - transfer->sr.level, - transfer->box.z, - PIPE_BIND_BLIT_DESTINATION); + subsrc.face = 0; + subsrc.level = 0; - /* XXX this flush prevents the following DRM error from occuring: - * [drm:radeon_cs_ioctl] *ERROR* Failed to parse relocation ! - * Reproducible with perf/copytex. */ - ctx->flush(ctx, 0, NULL); - - ctx->surface_copy(ctx, dst, - transfer->box.x, transfer->box.y, - src, 0, 0, - transfer->box.width, transfer->box.height); + ctx->resource_copy_region(ctx, tex, transfer->sr, + transfer->box.x, transfer->box.y, transfer->box.z, + &r300transfer->linear_texture->desc.b.b, subsrc, + 0, 0, 0, + transfer->box.width, transfer->box.height); - /* XXX this flush fixes a few piglit tests (e.g. glean/pixelFormats). */ ctx->flush(ctx, 0, NULL); - - pipe_surface_reference(&src, NULL); - pipe_surface_reference(&dst, NULL); } struct pipe_transfer* @@ -122,10 +91,26 @@ r300_texture_get_transfer(struct pipe_context *ctx, unsigned usage, const struct pipe_box *box) { + struct r300_context *r300 = r300_context(ctx); struct r300_texture *tex = r300_texture(texture); - struct r300_screen *r300screen = r300_screen(ctx->screen); struct r300_transfer *trans; struct pipe_resource base; + boolean referenced_cs, referenced_hw, blittable; + + referenced_cs = + r300->rws->cs_is_buffer_referenced(r300->cs, + tex->buffer, R300_REF_CS); + if (referenced_cs) { + referenced_hw = TRUE; + } else { + referenced_hw = + r300->rws->cs_is_buffer_referenced(r300->cs, + tex->buffer, R300_REF_HW); + } + + blittable = ctx->screen->is_format_supported( + ctx->screen, texture->format, texture->target, 0, + PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET, 0); trans = CALLOC_STRUCT(r300_transfer); if (trans) { @@ -136,13 +121,10 @@ r300_texture_get_transfer(struct pipe_context *ctx, trans->transfer.box = *box; /* If the texture is tiled, we must create a temporary detiled texture - * for this transfer. */ - if (tex->microtile || tex->macrotile) { - trans->render_target_usage = - util_format_is_depth_or_stencil(texture->format) ? - PIPE_BIND_DEPTH_STENCIL : - PIPE_BIND_RENDER_TARGET; - + * for this transfer. + * Also make write transfers pipelined. */ + if (tex->desc.microtile || tex->desc.macrotile[sr.level] || + ((referenced_hw & !(usage & PIPE_TRANSFER_READ)) && blittable)) { base.target = PIPE_TEXTURE_2D; base.format = texture->format; base.width0 = box->width; @@ -157,7 +139,7 @@ r300_texture_get_transfer(struct pipe_context *ctx, /* For texture reading, the temporary (detiled) texture is used as * a render target when blitting from a tiled texture. */ if (usage & PIPE_TRANSFER_READ) { - base.bind |= trans->render_target_usage; + base.bind |= PIPE_BIND_RENDER_TARGET; } /* For texture writing, the temporary texture is used as a sampler * when blitting into a tiled texture. */ @@ -166,12 +148,36 @@ r300_texture_get_transfer(struct pipe_context *ctx, } /* Create the temporary texture. */ - trans->detiled_texture = r300_texture( + trans->linear_texture = r300_texture( ctx->screen->resource_create(ctx->screen, &base)); - assert(!trans->detiled_texture->microtile && - !trans->detiled_texture->macrotile); + if (!trans->linear_texture) { + /* Oh crap, the thing can't create the texture. + * Let's flush and try again. */ + ctx->flush(ctx, 0, NULL); + + trans->linear_texture = r300_texture( + ctx->screen->resource_create(ctx->screen, + &base)); + + if (!trans->linear_texture) { + /* For linear textures, it's safe to fallback to + * an unpipelined transfer. */ + if (!tex->desc.microtile && !tex->desc.macrotile[sr.level]) { + goto unpipelined; + } + + /* Otherwise, go to hell. */ + fprintf(stderr, + "r300: Failed to create a transfer object, praise.\n"); + FREE(trans); + return NULL; + } + } + + assert(!trans->linear_texture->desc.microtile && + !trans->linear_texture->desc.macrotile[0]); /* Set the stride. * @@ -181,20 +187,30 @@ r300_texture_get_transfer(struct pipe_context *ctx, * right thing internally. */ trans->transfer.stride = - r300_texture_get_stride(r300screen, trans->detiled_texture, 0); + trans->linear_texture->desc.stride_in_bytes[0]; if (usage & PIPE_TRANSFER_READ) { /* We cannot map a tiled texture directly because the data is * in a different order, therefore we do detiling using a blit. */ r300_copy_from_tiled_texture(ctx, trans); + + /* Always referenced in the blit. */ + ctx->flush(ctx, 0, NULL); } - } else { - trans->transfer.stride = - r300_texture_get_stride(r300screen, tex, sr.level); - trans->offset = r300_texture_get_offset(tex, sr.level, box->z, sr.face); + return &trans->transfer; } + + unpipelined: + /* Unpipelined transfer. */ + trans->transfer.stride = tex->desc.stride_in_bytes[sr.level]; + trans->offset = r300_texture_get_offset(&tex->desc, + sr.level, box->z, sr.face); + + if (referenced_cs) + ctx->flush(ctx, PIPE_FLUSH_RENDER_CACHE, NULL); + return &trans->transfer; } - return &trans->transfer; + return NULL; } void r300_texture_transfer_destroy(struct pipe_context *ctx, @@ -202,13 +218,13 @@ void r300_texture_transfer_destroy(struct pipe_context *ctx, { struct r300_transfer *r300transfer = r300_transfer(trans); - if (r300transfer->detiled_texture) { + if (r300transfer->linear_texture) { if (trans->usage & PIPE_TRANSFER_WRITE) { r300_copy_into_tiled_texture(ctx, r300transfer); } pipe_resource_reference( - (struct pipe_resource**)&r300transfer->detiled_texture, NULL); + (struct pipe_resource**)&r300transfer->linear_texture, NULL); } pipe_resource_reference(&trans->resource, NULL); FREE(trans); @@ -217,21 +233,23 @@ void r300_texture_transfer_destroy(struct pipe_context *ctx, void* r300_texture_transfer_map(struct pipe_context *ctx, struct pipe_transfer *transfer) { + struct r300_context *r300 = r300_context(ctx); struct r300_winsys_screen *rws = (struct r300_winsys_screen *)ctx->winsys; struct r300_transfer *r300transfer = r300_transfer(transfer); struct r300_texture *tex = r300_texture(transfer->resource); char *map; - enum pipe_format format = tex->b.b.format; + enum pipe_format format = tex->desc.b.b.format; - if (r300transfer->detiled_texture) { + if (r300transfer->linear_texture) { /* The detiled texture is of the same size as the region being mapped * (no offset needed). */ return rws->buffer_map(rws, - r300transfer->detiled_texture->buffer, + r300transfer->linear_texture->buffer, + r300->cs, transfer->usage); } else { /* Tiling is disabled. */ - map = rws->buffer_map(rws, tex->buffer, + map = rws->buffer_map(rws, tex->buffer, r300->cs, transfer->usage); if (!map) { @@ -251,10 +269,9 @@ void r300_texture_transfer_unmap(struct pipe_context *ctx, struct r300_transfer *r300transfer = r300_transfer(transfer); struct r300_texture *tex = r300_texture(transfer->resource); - if (r300transfer->detiled_texture) { - rws->buffer_unmap(rws, r300transfer->detiled_texture->buffer); + if (r300transfer->linear_texture) { + rws->buffer_unmap(rws, r300transfer->linear_texture->buffer); } else { rws->buffer_unmap(rws, tex->buffer); } } - diff --git a/src/gallium/drivers/r300/r300_transfer.h b/src/gallium/drivers/r300/r300_transfer.h index d72e54e5ed9..0d32a68d1fa 100644 --- a/src/gallium/drivers/r300/r300_transfer.h +++ b/src/gallium/drivers/r300/r300_transfer.h @@ -24,7 +24,7 @@ #ifndef R300_TRANSFER #define R300_TRANSFER -#include "pipe/p_screen.h" +#include "pipe/p_context.h" struct r300_context; diff --git a/src/gallium/drivers/r300/r300_vs.c b/src/gallium/drivers/r300/r300_vs.c index bfab9c3b014..e2b9af9d018 100644 --- a/src/gallium/drivers/r300/r300_vs.c +++ b/src/gallium/drivers/r300/r300_vs.c @@ -22,7 +22,6 @@ * USE OR OTHER DEALINGS IN THE SOFTWARE. */ #include "r300_vs.h" -#include "r300_fs.h" #include "r300_context.h" #include "r300_screen.h" @@ -94,94 +93,6 @@ static void r300_shader_read_vs_outputs( vs_outputs->wpos = i; } -/* This function sets up: - * - VAP mapping, which maps VS registers to output semantics and - * at the same time it indicates which attributes are enabled and should - * be rasterized. - * - Stream mapping to VS outputs if TCL is not present. */ -static void r300_init_vs_output_mapping(struct r300_vertex_shader* vs) -{ - struct r300_shader_semantics* vs_outputs = &vs->outputs; - struct r300_vap_output_state *vap_out = &vs->vap_out; - int *stream_loc = vs->stream_loc_notcl; - int i, gen_count, tabi = 0; - boolean any_bcolor_used = vs_outputs->bcolor[0] != ATTR_UNUSED || - vs_outputs->bcolor[1] != ATTR_UNUSED; - - vap_out->vap_vtx_state_cntl = 0x5555; /* XXX this is classic Mesa bonghits */ - - /* Position. */ - if (vs_outputs->pos != ATTR_UNUSED) { - vap_out->vap_vsm_vtx_assm |= R300_INPUT_CNTL_POS; - vap_out->vap_out_vtx_fmt[0] |= R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT; - - stream_loc[tabi++] = 0; - } else { - assert(0); - } - - /* Point size. */ - if (vs_outputs->psize != ATTR_UNUSED) { - vap_out->vap_out_vtx_fmt[0] |= R300_VAP_OUTPUT_VTX_FMT_0__PT_SIZE_PRESENT; - - stream_loc[tabi++] = 1; - } - - /* Colors. */ - for (i = 0; i < ATTR_COLOR_COUNT; i++) { - if (vs_outputs->color[i] != ATTR_UNUSED || any_bcolor_used || - vs_outputs->color[1] != ATTR_UNUSED) { - vap_out->vap_vsm_vtx_assm |= R300_INPUT_CNTL_COLOR; - vap_out->vap_out_vtx_fmt[0] |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << i; - - stream_loc[tabi++] = 2 + i; - } - } - - /* Back-face colors. */ - if (any_bcolor_used) { - for (i = 0; i < ATTR_COLOR_COUNT; i++) { - vap_out->vap_vsm_vtx_assm |= R300_INPUT_CNTL_COLOR; - vap_out->vap_out_vtx_fmt[0] |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << (2+i); - - stream_loc[tabi++] = 4 + i; - } - } - - /* Texture coordinates. */ - gen_count = 0; - for (i = 0; i < ATTR_GENERIC_COUNT && gen_count < 8; i++) { - if (vs_outputs->generic[i] != ATTR_UNUSED) { - vap_out->vap_vsm_vtx_assm |= (R300_INPUT_CNTL_TC0 << gen_count); - vap_out->vap_out_vtx_fmt[1] |= (4 << (3 * gen_count)); - - stream_loc[tabi++] = 6 + gen_count; - gen_count++; - } - } - - /* Fog coordinates. */ - if (gen_count < 8 && vs_outputs->fog != ATTR_UNUSED) { - vap_out->vap_vsm_vtx_assm |= (R300_INPUT_CNTL_TC0 << gen_count); - vap_out->vap_out_vtx_fmt[1] |= (4 << (3 * gen_count)); - - stream_loc[tabi++] = 6 + gen_count; - gen_count++; - } - - /* WPOS. */ - if (gen_count < 8) { - vs->wpos_tex_output = gen_count; - stream_loc[tabi++] = 6 + gen_count; - } else { - vs_outputs->wpos = ATTR_UNUSED; - } - - for (; tabi < 16;) { - stream_loc[tabi++] = -1; - } -} - static void set_vertex_inputs_outputs(struct r300_vertex_program_compiler * c) { struct r300_vertex_shader * vs = c->UserData; @@ -246,16 +157,19 @@ static void set_vertex_inputs_outputs(struct r300_vertex_program_compiler * c) } /* WPOS. */ - if (outputs->wpos != ATTR_UNUSED) { - c->code->outputs[outputs->wpos] = reg++; - } + c->code->outputs[outputs->wpos] = reg++; +} + +void r300_init_vs_outputs(struct r300_vertex_shader *vs) +{ + tgsi_scan_shader(vs->state.tokens, &vs->info); + r300_shader_read_vs_outputs(&vs->info, &vs->outputs); } static void r300_dummy_vertex_shader( struct r300_context* r300, struct r300_vertex_shader* shader) { - struct pipe_shader_state state; struct ureg_program *ureg; struct ureg_dst dst; struct ureg_src imm; @@ -269,35 +183,40 @@ static void r300_dummy_vertex_shader( ureg_MOV(ureg, dst, imm); ureg_END(ureg); - state.tokens = ureg_finalize(ureg); + shader->state.tokens = tgsi_dup_tokens(ureg_finalize(ureg)); + ureg_destroy(ureg); shader->dummy = TRUE; - r300_translate_vertex_shader(r300, shader, state.tokens); - - ureg_destroy(ureg); + r300_init_vs_outputs(shader); + r300_translate_vertex_shader(r300, shader); } -void r300_translate_vertex_shader(struct r300_context* r300, - struct r300_vertex_shader* vs, - const struct tgsi_token *tokens) +void r300_translate_vertex_shader(struct r300_context *r300, + struct r300_vertex_shader *vs) { struct r300_vertex_program_compiler compiler; struct tgsi_to_rc ttr; - - tgsi_scan_shader(tokens, &vs->info); - r300_shader_read_vs_outputs(&vs->info, &vs->outputs); - r300_init_vs_output_mapping(vs); + unsigned i; /* Setup the compiler */ + memset(&compiler, 0, sizeof(compiler)); rc_init(&compiler.Base); compiler.Base.Debug = DBG_ON(r300, DBG_VP); compiler.code = &vs->code; compiler.UserData = vs; + compiler.Base.is_r500 = r300->screen->caps.is_r500; + compiler.Base.disable_optimizations = DBG_ON(r300, DBG_NO_OPT); + compiler.Base.has_half_swizzles = FALSE; + compiler.Base.has_presub = FALSE; + compiler.Base.max_temp_regs = 32; + compiler.Base.max_constants = 256; + compiler.Base.max_alu_insts = r300->screen->caps.is_r500 ? 1024 : 256; + compiler.Base.remove_unused_constants = TRUE; if (compiler.Base.Debug) { - debug_printf("r300: Initial vertex program\n"); - tgsi_dump(tokens, 0); + DBG(r300, DBG_VP, "r300: Initial vertex program\n"); + tgsi_dump(vs->state.tokens, 0); } /* Translate TGSI to our internal representation */ @@ -305,23 +224,17 @@ void r300_translate_vertex_shader(struct r300_context* r300, ttr.info = &vs->info; ttr.use_half_swizzles = FALSE; - r300_tgsi_to_rc(&ttr, tokens); - - compiler.RequiredOutputs = - ~(~0 << (vs->info.num_outputs + - (vs->outputs.wpos != ATTR_UNUSED ? 1 : 0))); + r300_tgsi_to_rc(&ttr, vs->state.tokens); + compiler.RequiredOutputs = ~(~0 << (vs->info.num_outputs + 1)); compiler.SetHwInputOutput = &set_vertex_inputs_outputs; /* Insert the WPOS output. */ - if (vs->outputs.wpos != ATTR_UNUSED) { - rc_copy_output(&compiler.Base, 0, vs->outputs.wpos); - } + rc_copy_output(&compiler.Base, 0, vs->outputs.wpos); /* Invoke the compiler */ r3xx_compile_vertex_program(&compiler); if (compiler.Base.Error) { - /* XXX We should fallback using Draw. */ fprintf(stderr, "r300 VP: Compiler error:\n%sUsing a dummy shader" " instead.\n", compiler.Base.ErrorMsg); @@ -337,38 +250,17 @@ void r300_translate_vertex_shader(struct r300_context* r300, } /* Initialize numbers of constants for each type. */ - vs->externals_count = ttr.immediate_offset; + vs->externals_count = 0; + for (i = 0; + i < vs->code.constants.Count && + vs->code.constants.Constants[i].Type == RC_CONSTANT_EXTERNAL; i++) { + vs->externals_count = i+1; + } + for (; i < vs->code.constants.Count; i++) { + assert(vs->code.constants.Constants[i].Type == RC_CONSTANT_IMMEDIATE); + } vs->immediates_count = vs->code.constants.Count - vs->externals_count; /* And, finally... */ rc_destroy(&compiler.Base); } - -boolean r300_vertex_shader_setup_wpos(struct r300_context* r300) -{ - struct r300_vertex_shader* vs = r300->vs_state.state; - struct r300_vap_output_state *vap_out = &vs->vap_out; - int tex_output = vs->wpos_tex_output; - uint32_t tex_fmt = R300_INPUT_CNTL_TC0 << tex_output; - - if (vs->outputs.wpos == ATTR_UNUSED) { - return FALSE; - } - - if (r300_fs(r300)->shader->inputs.wpos != ATTR_UNUSED) { - /* Enable WPOS in VAP. */ - if (!(vap_out->vap_vsm_vtx_assm & tex_fmt)) { - vap_out->vap_vsm_vtx_assm |= tex_fmt; - vap_out->vap_out_vtx_fmt[1] |= (4 << (3 * tex_output)); - return TRUE; - } - } else { - /* Disable WPOS in VAP. */ - if (vap_out->vap_vsm_vtx_assm & tex_fmt) { - vap_out->vap_vsm_vtx_assm &= ~tex_fmt; - vap_out->vap_out_vtx_fmt[1] &= ~(4 << (3 * tex_output)); - return TRUE; - } - } - return FALSE; -} diff --git a/src/gallium/drivers/r300/r300_vs.h b/src/gallium/drivers/r300/r300_vs.h index 56bcc3b70b8..170de6c79db 100644 --- a/src/gallium/drivers/r300/r300_vs.h +++ b/src/gallium/drivers/r300/r300_vs.h @@ -39,7 +39,6 @@ struct r300_vertex_shader { struct tgsi_shader_info info; struct r300_shader_semantics outputs; - struct r300_vap_output_state vap_out; /* Whether the shader was replaced by a dummy one due to a shader * compilation failure. */ @@ -49,12 +48,6 @@ struct r300_vertex_shader { unsigned externals_count; unsigned immediates_count; - /* Stream locations for SWTCL or if TCL is bypassed. */ - int stream_loc_notcl[16]; - - /* Output stream location for WPOS. */ - int wpos_tex_output; - /* HWTCL-specific. */ /* Machine code (if translated) */ struct r300_vertex_program_code code; @@ -63,11 +56,12 @@ struct r300_vertex_shader { void *draw_vs; }; -void r300_translate_vertex_shader(struct r300_context* r300, - struct r300_vertex_shader* vs, - const struct tgsi_token *tokens); +void r300_init_vs_outputs(struct r300_vertex_shader *vs); + +void r300_translate_vertex_shader(struct r300_context *r300, + struct r300_vertex_shader *vs); -/* Return TRUE if VAP (hwfmt) needs to be re-emitted. */ -boolean r300_vertex_shader_setup_wpos(struct r300_context* r300); +void r300_draw_init_vertex_shader(struct draw_context *draw, + struct r300_vertex_shader *vs); #endif /* R300_VS_H */ diff --git a/src/gallium/drivers/r300/r300_vs_draw.c b/src/gallium/drivers/r300/r300_vs_draw.c new file mode 100644 index 00000000000..2939963c355 --- /dev/null +++ b/src/gallium/drivers/r300/r300_vs_draw.c @@ -0,0 +1,358 @@ +/************************************************************************** + * + * Copyright 2009 Marek Olšák <[email protected]> + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/* This file contains the vertex shader tranformations for SW TCL needed + * to overcome the limitations of the r300 rasterizer. + * + * Transformations: + * 1) If the secondary color output is present, the primary color must be + * inserted before it. + * 2) If any back-face color output is present, there must be all 4 color + * outputs and missing ones must be inserted. + * 3) Insert a trailing texcoord output containing a copy of POS, for WPOS. + * + * I know this code is cumbersome, but I don't know of any nicer way + * of transforming TGSI shaders. ~ M. + */ + +#include "r300_vs.h" + +#include <stdio.h> + +#include "tgsi/tgsi_transform.h" +#include "tgsi/tgsi_dump.h" + +#include "draw/draw_context.h" + +struct vs_transform_context { + struct tgsi_transform_context base; + + boolean color_used[2]; + boolean bcolor_used[2]; + boolean temp_used[128]; + + /* Index of the pos output, typically 0. */ + unsigned pos_output; + /* Index of the pos temp where all writes of pos are redirected to. */ + unsigned pos_temp; + /* The index of the last generic output, after which we insert a new + * output for WPOS. */ + int last_generic; + + unsigned num_outputs; + /* Used to shift output decl. indices when inserting new ones. */ + unsigned decl_shift; + /* Used to remap writes to output decls if their indices changed. */ + unsigned out_remap[32]; + + /* First instruction processed? */ + boolean first_instruction; + /* End instruction processed? */ + boolean end_instruction; +}; + +static void emit_temp(struct tgsi_transform_context *ctx, unsigned reg) +{ + struct tgsi_full_declaration decl; + + decl = tgsi_default_full_declaration(); + decl.Declaration.File = TGSI_FILE_TEMPORARY; + decl.Range.First = decl.Range.Last = reg; + ctx->emit_declaration(ctx, &decl); +} + +static void emit_output(struct tgsi_transform_context *ctx, + unsigned name, unsigned index, unsigned interp, + unsigned reg) +{ + struct vs_transform_context *vsctx = (struct vs_transform_context *)ctx; + struct tgsi_full_declaration decl; + + decl = tgsi_default_full_declaration(); + decl.Declaration.File = TGSI_FILE_OUTPUT; + decl.Declaration.Interpolate = interp; + decl.Declaration.Semantic = TRUE; + decl.Semantic.Name = name; + decl.Semantic.Index = index; + decl.Range.First = decl.Range.Last = reg; + ctx->emit_declaration(ctx, &decl); + ++vsctx->num_outputs; +} + +static void insert_output(struct tgsi_transform_context *ctx, + struct tgsi_full_declaration *before, + unsigned name, unsigned index, unsigned interp) +{ + struct vs_transform_context *vsctx = (struct vs_transform_context *)ctx; + unsigned i; + + /* Make a place for the new output. */ + for (i = before->Range.First; i < Elements(vsctx->out_remap); i++) { + ++vsctx->out_remap[i]; + } + + /* Insert the new output. */ + emit_output(ctx, name, index, interp, before->Range.First); + + ++vsctx->decl_shift; +} + +static void insert_trailing_bcolor(struct tgsi_transform_context *ctx, + struct tgsi_full_declaration *before) +{ + struct vs_transform_context *vsctx = (struct vs_transform_context *)ctx; + + /* If BCOLOR0 is used, make sure BCOLOR1 is present too. Otherwise + * the rasterizer doesn't do the color selection correctly. */ + if (vsctx->bcolor_used[0] && !vsctx->bcolor_used[1]) { + if (before) { + insert_output(ctx, before, TGSI_SEMANTIC_BCOLOR, 1, + TGSI_INTERPOLATE_LINEAR); + } else { + emit_output(ctx, TGSI_SEMANTIC_BCOLOR, 1, + TGSI_INTERPOLATE_LINEAR, vsctx->num_outputs); + } + vsctx->bcolor_used[1] = TRUE; + } +} + +static void transform_decl(struct tgsi_transform_context *ctx, + struct tgsi_full_declaration *decl) +{ + struct vs_transform_context *vsctx = (struct vs_transform_context *)ctx; + unsigned i; + + if (decl->Declaration.File == TGSI_FILE_OUTPUT) { + switch (decl->Semantic.Name) { + case TGSI_SEMANTIC_POSITION: + vsctx->pos_output = decl->Range.First; + break; + + case TGSI_SEMANTIC_COLOR: + assert(decl->Semantic.Index < 2); + vsctx->color_used[decl->Semantic.Index] = TRUE; + + /* We must rasterize the first color if the second one is + * used, otherwise the rasterizer doesn't do the color + * selection correctly. Declare it, but don't write to it. */ + if (decl->Semantic.Index == 1 && !vsctx->color_used[0]) { + insert_output(ctx, decl, TGSI_SEMANTIC_COLOR, 0, + TGSI_INTERPOLATE_LINEAR); + vsctx->color_used[0] = TRUE; + } + break; + + case TGSI_SEMANTIC_BCOLOR: + assert(decl->Semantic.Index < 2); + vsctx->bcolor_used[decl->Semantic.Index] = TRUE; + + /* We must rasterize all 4 colors if back-face colors are + * used, otherwise the rasterizer doesn't do the color + * selection correctly. Declare it, but don't write to it. */ + if (!vsctx->color_used[0]) { + insert_output(ctx, decl, TGSI_SEMANTIC_COLOR, 0, + TGSI_INTERPOLATE_LINEAR); + vsctx->color_used[0] = TRUE; + } + if (!vsctx->color_used[1]) { + insert_output(ctx, decl, TGSI_SEMANTIC_COLOR, 1, + TGSI_INTERPOLATE_LINEAR); + vsctx->color_used[1] = TRUE; + } + if (decl->Semantic.Index == 1 && !vsctx->bcolor_used[0]) { + insert_output(ctx, decl, TGSI_SEMANTIC_BCOLOR, 0, + TGSI_INTERPOLATE_LINEAR); + vsctx->bcolor_used[0] = TRUE; + } + /* One more case is handled in insert_trailing_bcolor. */ + break; + + case TGSI_SEMANTIC_GENERIC: + vsctx->last_generic = MAX2(vsctx->last_generic, decl->Semantic.Index); + break; + } + + if (decl->Semantic.Name != TGSI_SEMANTIC_BCOLOR) { + /* Insert it as soon as possible. */ + insert_trailing_bcolor(ctx, decl); + } + + /* Since we're inserting new outputs in between, the following outputs + * should be moved to the right so that they don't overlap with + * the newly added ones. */ + decl->Range.First += vsctx->decl_shift; + decl->Range.Last += vsctx->decl_shift; + + ++vsctx->num_outputs; + } else if (decl->Declaration.File == TGSI_FILE_TEMPORARY) { + for (i = decl->Range.First; i <= decl->Range.Last; i++) { + vsctx->temp_used[i] = TRUE; + } + } + + ctx->emit_declaration(ctx, decl); +} + +static void transform_inst(struct tgsi_transform_context *ctx, + struct tgsi_full_instruction *inst) +{ + struct vs_transform_context *vsctx = (struct vs_transform_context *) ctx; + struct tgsi_full_instruction new_inst; + unsigned i; + + if (!vsctx->first_instruction) { + vsctx->first_instruction = TRUE; + + /* The trailing BCOLOR should be inserted before the code + * if it hasn't already been done so. */ + insert_trailing_bcolor(ctx, NULL); + + /* Insert the generic output for WPOS. */ + emit_output(ctx, TGSI_SEMANTIC_GENERIC, vsctx->last_generic + 1, + TGSI_INTERPOLATE_PERSPECTIVE, vsctx->num_outputs); + + /* Find a free temp for POSITION. */ + for (i = 0; i < Elements(vsctx->temp_used); i++) { + if (!vsctx->temp_used[i]) { + emit_temp(ctx, i); + vsctx->pos_temp = i; + break; + } + } + } + + if (inst->Instruction.Opcode == TGSI_OPCODE_END) { + /* MOV OUT[pos_output], TEMP[pos_temp]; */ + new_inst = tgsi_default_full_instruction(); + new_inst.Instruction.Opcode = TGSI_OPCODE_MOV; + new_inst.Instruction.NumDstRegs = 1; + new_inst.Dst[0].Register.File = TGSI_FILE_OUTPUT; + new_inst.Dst[0].Register.Index = vsctx->pos_output; + new_inst.Dst[0].Register.WriteMask = TGSI_WRITEMASK_XYZW; + new_inst.Instruction.NumSrcRegs = 1; + new_inst.Src[0].Register.File = TGSI_FILE_TEMPORARY; + new_inst.Src[0].Register.Index = vsctx->pos_temp; + ctx->emit_instruction(ctx, &new_inst); + + /* MOV OUT[n-1], TEMP[pos_temp]; */ + new_inst = tgsi_default_full_instruction(); + new_inst.Instruction.Opcode = TGSI_OPCODE_MOV; + new_inst.Instruction.NumDstRegs = 1; + new_inst.Dst[0].Register.File = TGSI_FILE_OUTPUT; + new_inst.Dst[0].Register.Index = vsctx->num_outputs - 1; + new_inst.Dst[0].Register.WriteMask = TGSI_WRITEMASK_XYZW; + new_inst.Instruction.NumSrcRegs = 1; + new_inst.Src[0].Register.File = TGSI_FILE_TEMPORARY; + new_inst.Src[0].Register.Index = vsctx->pos_temp; + ctx->emit_instruction(ctx, &new_inst); + + vsctx->end_instruction = TRUE; + } else { + /* Not an END instruction. */ + /* Fix writes to outputs. */ + for (i = 0; i < inst->Instruction.NumDstRegs; i++) { + struct tgsi_full_dst_register *dst = &inst->Dst[i]; + if (dst->Register.File == TGSI_FILE_OUTPUT) { + if (dst->Register.Index == vsctx->pos_output) { + /* Replace writes to OUT[pos_output] with TEMP[pos_temp]. */ + dst->Register.File = TGSI_FILE_TEMPORARY; + dst->Register.Index = vsctx->pos_temp; + } else { + /* Not a position, good... + * Since we were changing the indices of output decls, + * we must redirect writes into them too. */ + dst->Register.Index = vsctx->out_remap[dst->Register.Index]; + } + } + } + + /* Inserting 2 instructions before the END opcode moves all following + * labels by 2. Subroutines are always after the END opcode so + * they're always moved. */ + if (inst->Instruction.Opcode == TGSI_OPCODE_CAL) { + inst->Label.Label += 2; + } + /* The labels of the following opcodes are moved only after + * the END opcode. */ + if (vsctx->end_instruction && + (inst->Instruction.Opcode == TGSI_OPCODE_IF || + inst->Instruction.Opcode == TGSI_OPCODE_ELSE || + inst->Instruction.Opcode == TGSI_OPCODE_BGNLOOP || + inst->Instruction.Opcode == TGSI_OPCODE_ENDLOOP)) { + inst->Label.Label += 2; + } + } + + ctx->emit_instruction(ctx, inst); +} + +void r300_draw_init_vertex_shader(struct draw_context *draw, + struct r300_vertex_shader *vs) +{ + struct pipe_shader_state new_vs; + struct vs_transform_context transform; + const uint newLen = tgsi_num_tokens(vs->state.tokens) + 100 /* XXX */; + unsigned i; + + new_vs.tokens = tgsi_alloc_tokens(newLen); + if (new_vs.tokens == NULL) + return; + + memset(&transform, 0, sizeof(transform)); + for (i = 0; i < Elements(transform.out_remap); i++) { + transform.out_remap[i] = i; + } + transform.last_generic = -1; + transform.base.transform_instruction = transform_inst; + transform.base.transform_declaration = transform_decl; + + tgsi_transform_shader(vs->state.tokens, + (struct tgsi_token*)new_vs.tokens, + newLen, &transform.base); + +#if 0 + printf("----------------------------------------------\norig shader:\n"); + tgsi_dump(vs->state.tokens, 0); + printf("----------------------------------------------\nnew shader:\n"); + tgsi_dump(new_vs.tokens, 0); + printf("----------------------------------------------\n"); +#endif + + /* Free old tokens. */ + FREE((void*)vs->state.tokens); + + vs->draw_vs = draw_create_vertex_shader(draw, &new_vs); + + /* Instead of duplicating and freeing the tokens, copy the pointer directly. */ + vs->state.tokens = new_vs.tokens; + + /* Init the VS output table for the rasterizer. */ + r300_init_vs_outputs(vs); + + /* Make the last generic be WPOS. */ + vs->outputs.wpos = vs->outputs.generic[transform.last_generic + 1]; + vs->outputs.generic[transform.last_generic + 1] = ATTR_UNUSED; +} diff --git a/src/gallium/drivers/r300/r300_winsys.h b/src/gallium/drivers/r300/r300_winsys.h index 1642981eaa8..4597332399a 100644 --- a/src/gallium/drivers/r300/r300_winsys.h +++ b/src/gallium/drivers/r300/r300_winsys.h @@ -1,5 +1,6 @@ /* * Copyright 2008 Corbin Simpson <[email protected]> + * Copyright 2010 Marek Olšák <[email protected]> * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), @@ -23,23 +24,34 @@ #ifndef R300_WINSYS_H #define R300_WINSYS_H -/* The public interface header for the r300 pipe driver. - * Any winsys hosting this pipe needs to implement r300_winsys and then - * call r300_create_screen to start things. */ +/* The public winsys interface header for the r300 pipe driver. + * Any winsys hosting this pipe needs to implement r300_winsys_screen and then + * call r300_screen_create to start things. */ #include "pipe/p_defines.h" #include "pipe/p_state.h" #include "r300_defines.h" +struct winsys_handle; +struct r300_winsys_screen; + struct r300_winsys_buffer; +struct r300_winsys_cs { + uint32_t *ptr; /* Pointer to the beginning of the CS. */ + unsigned cdw; /* Number of used dwords. */ + unsigned ndw; /* Size of the CS in dwords. */ +}; + enum r300_value_id { R300_VID_PCI_ID, R300_VID_GB_PIPES, R300_VID_Z_PIPES, R300_VID_SQUARE_TILING_SUPPORT, R300_VID_DRM_2_3_0, + R300_VID_DRM_2_6_0, + R300_CAN_HYPERZ, }; enum r300_reference_domain { /* bitfield */ @@ -48,135 +60,251 @@ enum r300_reference_domain { /* bitfield */ }; struct r300_winsys_screen { + /** + * Destroy this winsys. + * + * \param ws The winsys this function is called from. + */ void (*destroy)(struct r300_winsys_screen *ws); - + /** + * Query a system value from a winsys. + * + * \param ws The winsys this function is called from. + * \param vid One of the R300_VID_* enums. + */ + uint32_t (*get_value)(struct r300_winsys_screen *ws, + enum r300_value_id vid); + + /************************************************************************** * Buffer management. Buffer attributes are mostly fixed over its lifetime. * * Remember that gallium gets to choose the interface it needs, and the * window systems must then implement that interface (rather than the * other way around...). + *************************************************************************/ + + /** + * Create a buffer object. * - * usage is a bitmask of R300_WINSYS_BUFFER_USAGE_PIXEL/VERTEX/INDEX/CONSTANT. This - * usage argument is only an optimization hint, not a guarantee, therefore - * proper behavior must be observed in all circumstances. - * - * alignment indicates the client's alignment requirements, eg for - * SSE instructions. + * \param ws The winsys this function is called from. + * \param size The size to allocate. + * \param alignment An alignment of the buffer in memory. + * \param bind A bitmask of the PIPE_BIND_* flags. + * \param usage A bitmask of the PIPE_USAGE_* flags. + * \param domain A bitmask of the R300_DOMAIN_* flags. + * \return The created buffer object. */ struct r300_winsys_buffer *(*buffer_create)(struct r300_winsys_screen *ws, - unsigned alignment, - unsigned usage, - unsigned size); - + unsigned size, + unsigned alignment, + unsigned bind, + unsigned usage, + enum r300_buffer_domain domain); + /** - * Map the entire data store of a buffer object into the client's address. - * flags is bitmask of R300_WINSYS_BUFFER_USAGE_CPU_READ/WRITE flags. + * Reference a buffer object (assign with reference counting). + * + * \param ws The winsys this function is called from. + * \param pdst A destination pointer to set the source buffer to. + * \param src A source buffer object. */ - void *(*buffer_map)( struct r300_winsys_screen *ws, - struct r300_winsys_buffer *buf, - unsigned usage); + void (*buffer_reference)(struct r300_winsys_screen *ws, + struct r300_winsys_buffer **pdst, + struct r300_winsys_buffer *src); - void (*buffer_unmap)( struct r300_winsys_screen *ws, - struct r300_winsys_buffer *buf ); - - void (*buffer_destroy)( struct r300_winsys_buffer *buf ); + /** + * Map the entire data store of a buffer object into the client's address + * space. + * + * \param ws The winsys this function is called from. + * \param buf A winsys buffer object to map. + * \param cs A command stream to flush if the buffer is referenced by it. + * \param usage A bitmask of the PIPE_TRANSFER_* flags. + * \return The pointer at the beginning of the buffer. + */ + void *(*buffer_map)(struct r300_winsys_screen *ws, + struct r300_winsys_buffer *buf, + struct r300_winsys_cs *cs, + enum pipe_transfer_usage usage); + /** + * Unmap a buffer object from the client's address space. + * + * \param ws The winsys this function is called from. + * \param buf A winsys buffer object to unmap. + */ + void (*buffer_unmap)(struct r300_winsys_screen *ws, + struct r300_winsys_buffer *buf); - void (*buffer_reference)(struct r300_winsys_screen *rws, - struct r300_winsys_buffer **pdst, - struct r300_winsys_buffer *src); + /** + * Wait for a buffer object until it is not used by a GPU. This is + * equivalent to a fence placed after the last command using the buffer, + * and synchronizing to the fence. + * + * \param ws The winsys this function is called from. + * \param buf A winsys buffer object to wait for. + */ + void (*buffer_wait)(struct r300_winsys_screen *ws, + struct r300_winsys_buffer *buf); - boolean (*buffer_references)(struct r300_winsys_buffer *a, - struct r300_winsys_buffer *b); + /** + * Return tiling flags describing a memory layout of a buffer object. + * + * \param ws The winsys this function is called from. + * \param buf A winsys buffer object to get the flags from. + * \param macrotile A pointer to the return value of the microtile flag. + * \param microtile A pointer to the return value of the macrotile flag. + * + * \note microtile and macrotile are not bitmasks! + */ + void (*buffer_get_tiling)(struct r300_winsys_screen *ws, + struct r300_winsys_buffer *buf, + enum r300_buffer_tiling *microtile, + enum r300_buffer_tiling *macrotile); - void (*buffer_flush_range)(struct r300_winsys_screen *rws, - struct r300_winsys_buffer *buf, - unsigned offset, - unsigned length); + /** + * Set tiling flags describing a memory layout of a buffer object. + * + * \param ws The winsys this function is called from. + * \param buf A winsys buffer object to set the flags for. + * \param macrotile A macrotile flag. + * \param microtile A microtile flag. + * \param stride A stride of the buffer in bytes, for texturing. + * + * \note microtile and macrotile are not bitmasks! + */ + void (*buffer_set_tiling)(struct r300_winsys_screen *ws, + struct r300_winsys_buffer *buf, + enum r300_buffer_tiling microtile, + enum r300_buffer_tiling macrotile, + unsigned stride); - /* Add a pipe_resource to the list of buffer objects to validate. */ - boolean (*add_buffer)(struct r300_winsys_screen *winsys, - struct r300_winsys_buffer *buf, - uint32_t rd, - uint32_t wd); + /** + * Get a winsys buffer from a winsys handle. The internal structure + * of the handle is platform-specific and only a winsys should access it. + * + * \param ws The winsys this function is called from. + * \param whandle A winsys handle pointer as was received from a state + * tracker. + * \param stride The returned buffer stride in bytes. + * \param size The returned buffer size. + */ + struct r300_winsys_buffer *(*buffer_from_handle)(struct r300_winsys_screen *ws, + struct winsys_handle *whandle, + unsigned *stride, + unsigned *size); + /** + * Get a winsys handle from a winsys buffer. The internal structure + * of the handle is platform-specific and only a winsys should access it. + * + * \param ws The winsys this function is called from. + * \param buf A winsys buffer object to get the handle from. + * \param whandle A winsys handle pointer. + * \param stride A stride of the buffer in bytes, for texturing. + * \return TRUE on success. + */ + boolean (*buffer_get_handle)(struct r300_winsys_screen *ws, + struct r300_winsys_buffer *buf, + unsigned stride, + struct winsys_handle *whandle); - /* Revalidate all currently setup pipe_buffers. - * Returns TRUE if a flush is required. */ - boolean (*validate)(struct r300_winsys_screen* winsys); + /************************************************************************** + * Command submission. + * + * Each pipe context should create its own command stream and submit + * commands independently of other contexts. + *************************************************************************/ - /* Check to see if there's room for commands. */ - boolean (*check_cs)(struct r300_winsys_screen* winsys, int size); + /** + * Create a command stream. + * + * \param ws The winsys this function is called from. + */ + struct r300_winsys_cs *(*cs_create)(struct r300_winsys_screen *ws); - /* Start a command emit. */ - void (*begin_cs)(struct r300_winsys_screen* winsys, - int size, - const char* file, - const char* function, - int line); + /** + * Destroy a command stream. + * + * \param cs A command stream to destroy. + */ + void (*cs_destroy)(struct r300_winsys_cs *cs); - /* Write a dword to the command buffer. */ - void (*write_cs_dword)(struct r300_winsys_screen* winsys, uint32_t dword); + /** + * Add a buffer object to the list of buffers to validate. + * + * \param cs A command stream to add buffer for validation against. + * \param buf A winsys buffer to validate. + * \param rd A read domain containing a bitmask + * of the R300_DOMAIN_* flags. + * \param wd A write domain containing a bitmask + * of the R300_DOMAIN_* flags. + */ + void (*cs_add_buffer)(struct r300_winsys_cs *cs, + struct r300_winsys_buffer *buf, + enum r300_buffer_domain rd, + enum r300_buffer_domain wd); - /* Write a table of dwords to the command buffer. */ - void (*write_cs_table)(struct r300_winsys_screen* winsys, - const void *dwords, unsigned count); + /** + * Revalidate all currently set up winsys buffers. + * Returns TRUE if a flush is required. + * + * \param cs A command stream to validate. + */ + boolean (*cs_validate)(struct r300_winsys_cs *cs); - /* Write a relocated dword to the command buffer. */ - void (*write_cs_reloc)(struct r300_winsys_screen *winsys, + /** + * Write a relocated dword to a command buffer. + * + * \param cs A command stream the relocation is written to. + * \param buf A winsys buffer to write the relocation for. + * \param rd A read domain containing a bitmask of the R300_DOMAIN_* flags. + * \param wd A write domain containing a bitmask of the R300_DOMAIN_* flags. + */ + void (*cs_write_reloc)(struct r300_winsys_cs *cs, struct r300_winsys_buffer *buf, - uint32_t rd, - uint32_t wd, - uint32_t flags); - - /* Finish a command emit. */ - void (*end_cs)(struct r300_winsys_screen* winsys, - const char* file, - const char* function, - int line); - - /* Flush the CS. */ - void (*flush_cs)(struct r300_winsys_screen* winsys); - - /* winsys flush - callback from winsys when flush required */ - void (*set_flush_cb)(struct r300_winsys_screen *winsys, - void (*flush_cb)(void *), void *data); - - void (*reset_bos)(struct r300_winsys_screen *winsys); - - void (*buffer_get_tiling)(struct r300_winsys_screen *winsys, - struct r300_winsys_buffer *buffer, - enum r300_buffer_tiling *microtiled, - enum r300_buffer_tiling *macrotiled); - - void (*buffer_set_tiling)(struct r300_winsys_screen *winsys, - struct r300_winsys_buffer *buffer, - uint32_t pitch, - enum r300_buffer_tiling microtiled, - enum r300_buffer_tiling macrotiled); - - uint32_t (*get_value)(struct r300_winsys_screen *winsys, - enum r300_value_id vid); - - struct r300_winsys_buffer *(*buffer_from_handle)(struct r300_winsys_screen *winsys, - struct pipe_screen *screen, - struct winsys_handle *whandle, - unsigned *stride); - boolean (*buffer_get_handle)(struct r300_winsys_screen *winsys, - struct r300_winsys_buffer *buffer, - unsigned stride, - struct winsys_handle *whandle); - - boolean (*is_buffer_referenced)(struct r300_winsys_screen *winsys, - struct r300_winsys_buffer *buffer, - enum r300_reference_domain domain); -}; + enum r300_buffer_domain rd, + enum r300_buffer_domain wd); + + /** + * Flush a command stream. + * + * \param cs A command stream to flush. + */ + void (*cs_flush)(struct r300_winsys_cs *cs); + + /** + * Set a flush callback which is called from winsys when flush is + * required. + * + * \param cs A command stream to set the callback for. + * \param flush A flush callback function associated with the command stream. + * \param user A user pointer that will be passed to the flush callback. + */ + void (*cs_set_flush)(struct r300_winsys_cs *cs, + void (*flush)(void *), + void *user); -struct r300_winsys_screen * -r300_winsys_screen(struct pipe_screen *screen); + /** + * Reset the list of buffer objects to validate, usually called + * prior to adding buffer objects for validation. + * + * \param cs A command stream to reset buffers for. + */ + void (*cs_reset_buffers)(struct r300_winsys_cs *cs); -/* Creates a new r300 screen. */ -struct pipe_screen* r300_create_screen(struct r300_winsys_screen *rws); + /** + * Return TRUE if a buffer is referenced by a command stream or by hardware + * (i.e. is busy), based on the domain parameter. + * + * \param cs A command stream. + * \param buf A winsys buffer. + * \param domain A bitmask of the R300_REF_* enums. + */ + boolean (*cs_is_buffer_referenced)(struct r300_winsys_cs *cs, + struct r300_winsys_buffer *buf, + enum r300_reference_domain domain); +}; #endif /* R300_WINSYS_H */ |