diff options
Diffstat (limited to 'src/gallium/drivers/radeon')
-rw-r--r-- | src/gallium/drivers/radeon/r600_pipe_common.h | 13 | ||||
-rw-r--r-- | src/gallium/drivers/radeon/r600_texture.c | 547 |
2 files changed, 10 insertions, 550 deletions
diff --git a/src/gallium/drivers/radeon/r600_pipe_common.h b/src/gallium/drivers/radeon/r600_pipe_common.h index 2ece4095c4a..36df0c4068a 100644 --- a/src/gallium/drivers/radeon/r600_pipe_common.h +++ b/src/gallium/drivers/radeon/r600_pipe_common.h @@ -735,6 +735,9 @@ void si_texture_get_fmask_info(struct r600_common_screen *rscreen, struct r600_texture *rtex, unsigned nr_samples, struct r600_fmask_info *out); +void si_texture_get_cmask_info(struct r600_common_screen *rscreen, + struct r600_texture *rtex, + struct r600_cmask_info *out); bool si_init_flushed_depth_texture(struct pipe_context *ctx, struct pipe_resource *texture, struct r600_texture **staging); @@ -757,20 +760,14 @@ struct pipe_surface *si_create_surface_custom(struct pipe_context *pipe, unsigned width0, unsigned height0, unsigned width, unsigned height); unsigned si_translate_colorswap(enum pipe_format format, bool do_endian_swap); +void vi_separate_dcc_try_enable(struct r600_common_context *rctx, + struct r600_texture *tex); void vi_separate_dcc_start_query(struct pipe_context *ctx, struct r600_texture *tex); void vi_separate_dcc_stop_query(struct pipe_context *ctx, struct r600_texture *tex); void vi_separate_dcc_process_and_reset_stats(struct pipe_context *ctx, struct r600_texture *tex); -void vi_dcc_clear_level(struct r600_common_context *rctx, - struct r600_texture *rtex, - unsigned level, unsigned clear_value); -void si_do_fast_color_clear(struct r600_common_context *rctx, - struct pipe_framebuffer_state *fb, - struct r600_atom *fb_state, - unsigned *buffers, ubyte *dirty_cbufs, - const union pipe_color_union *color); bool si_texture_disable_dcc(struct r600_common_context *rctx, struct r600_texture *rtex); void si_init_screen_texture_functions(struct r600_common_screen *rscreen); diff --git a/src/gallium/drivers/radeon/r600_texture.c b/src/gallium/drivers/radeon/r600_texture.c index e3658b465ae..46962ebe01f 100644 --- a/src/gallium/drivers/radeon/r600_texture.c +++ b/src/gallium/drivers/radeon/r600_texture.c @@ -777,9 +777,9 @@ static void r600_texture_allocate_fmask(struct r600_common_screen *rscreen, rtex->size = rtex->fmask.offset + rtex->fmask.size; } -static void si_texture_get_cmask_info(struct r600_common_screen *rscreen, - struct r600_texture *rtex, - struct r600_cmask_info *out) +void si_texture_get_cmask_info(struct r600_common_screen *rscreen, + struct r600_texture *rtex, + struct r600_cmask_info *out) { unsigned pipe_interleave_bytes = rscreen->info.pipe_interleave_bytes; unsigned num_pipes = rscreen->info.num_tile_pipes; @@ -842,37 +842,6 @@ static void r600_texture_allocate_cmask(struct r600_common_screen *rscreen, rtex->cb_color_info |= S_028C70_FAST_CLEAR(1); } -static void r600_texture_alloc_cmask_separate(struct r600_common_screen *rscreen, - struct r600_texture *rtex) -{ - if (rtex->cmask_buffer) - return; - - assert(rtex->cmask.size == 0); - - si_texture_get_cmask_info(rscreen, rtex, &rtex->cmask); - if (!rtex->cmask.size) - return; - - rtex->cmask_buffer = (struct r600_resource *) - si_aligned_buffer_create(&rscreen->b, - R600_RESOURCE_FLAG_UNMAPPABLE, - PIPE_USAGE_DEFAULT, - rtex->cmask.size, - rtex->cmask.alignment); - if (rtex->cmask_buffer == NULL) { - rtex->cmask.size = 0; - return; - } - - /* update colorbuffer state bits */ - rtex->cmask.base_address_reg = rtex->cmask_buffer->gpu_address >> 8; - - rtex->cb_color_info |= S_028C70_FAST_CLEAR(1); - - p_atomic_inc(&rscreen->compressed_colortex_counter); -} - static void r600_texture_get_htile_size(struct r600_common_screen *rscreen, struct r600_texture *rtex) { @@ -1973,71 +1942,6 @@ static void r600_surface_destroy(struct pipe_context *pipe, FREE(surface); } -static void r600_clear_texture(struct pipe_context *pipe, - struct pipe_resource *tex, - unsigned level, - const struct pipe_box *box, - const void *data) -{ - struct pipe_screen *screen = pipe->screen; - struct r600_texture *rtex = (struct r600_texture*)tex; - struct pipe_surface tmpl = {{0}}; - struct pipe_surface *sf; - const struct util_format_description *desc = - util_format_description(tex->format); - - tmpl.format = tex->format; - tmpl.u.tex.first_layer = box->z; - tmpl.u.tex.last_layer = box->z + box->depth - 1; - tmpl.u.tex.level = level; - sf = pipe->create_surface(pipe, tex, &tmpl); - if (!sf) - return; - - if (rtex->is_depth) { - unsigned clear; - float depth; - uint8_t stencil = 0; - - /* Depth is always present. */ - clear = PIPE_CLEAR_DEPTH; - desc->unpack_z_float(&depth, 0, data, 0, 1, 1); - - if (rtex->surface.has_stencil) { - clear |= PIPE_CLEAR_STENCIL; - desc->unpack_s_8uint(&stencil, 0, data, 0, 1, 1); - } - - pipe->clear_depth_stencil(pipe, sf, clear, depth, stencil, - box->x, box->y, - box->width, box->height, false); - } else { - union pipe_color_union color; - - /* pipe_color_union requires the full vec4 representation. */ - if (util_format_is_pure_uint(tex->format)) - desc->unpack_rgba_uint(color.ui, 0, data, 0, 1, 1); - else if (util_format_is_pure_sint(tex->format)) - desc->unpack_rgba_sint(color.i, 0, data, 0, 1, 1); - else - desc->unpack_rgba_float(color.f, 0, data, 0, 1, 1); - - if (screen->is_format_supported(screen, tex->format, - tex->target, 0, - PIPE_BIND_RENDER_TARGET)) { - pipe->clear_render_target(pipe, sf, &color, - box->x, box->y, - box->width, box->height, false); - } else { - /* Software fallback - just for R9G9B9E5_FLOAT */ - util_clear_render_target(pipe, sf, &color, - box->x, box->y, - box->width, box->height); - } - } - pipe_surface_reference(&sf, NULL); -} - unsigned si_translate_colorswap(enum pipe_format format, bool do_endian_swap) { const struct util_format_description *desc = util_format_description(format); @@ -2221,8 +2125,8 @@ static bool vi_should_enable_separate_dcc(struct r600_texture *tex) } /* Called by fast clear. */ -static void vi_separate_dcc_try_enable(struct r600_common_context *rctx, - struct r600_texture *tex) +void vi_separate_dcc_try_enable(struct r600_common_context *rctx, + struct r600_texture *tex) { /* The intent is to use this with shared displayable back buffers, * but it's not strictly limited only to them. @@ -2336,446 +2240,6 @@ void vi_separate_dcc_process_and_reset_stats(struct pipe_context *ctx, } } -/* FAST COLOR CLEAR */ - -static void evergreen_set_clear_color(struct r600_texture *rtex, - enum pipe_format surface_format, - const union pipe_color_union *color) -{ - union util_color uc; - - memset(&uc, 0, sizeof(uc)); - - if (rtex->surface.bpe == 16) { - /* DCC fast clear only: - * CLEAR_WORD0 = R = G = B - * CLEAR_WORD1 = A - */ - assert(color->ui[0] == color->ui[1] && - color->ui[0] == color->ui[2]); - uc.ui[0] = color->ui[0]; - uc.ui[1] = color->ui[3]; - } else if (util_format_is_pure_uint(surface_format)) { - util_format_write_4ui(surface_format, color->ui, 0, &uc, 0, 0, 0, 1, 1); - } else if (util_format_is_pure_sint(surface_format)) { - util_format_write_4i(surface_format, color->i, 0, &uc, 0, 0, 0, 1, 1); - } else { - util_pack_color(color->f, surface_format, &uc); - } - - memcpy(rtex->color_clear_value, &uc, 2 * sizeof(uint32_t)); -} - -static bool vi_get_fast_clear_parameters(enum pipe_format surface_format, - const union pipe_color_union *color, - uint32_t* reset_value, - bool* clear_words_needed) -{ - bool values[4] = {}; - int i; - bool main_value = false; - bool extra_value = false; - int extra_channel; - - /* This is needed to get the correct DCC clear value for luminance formats. - * 1) Get the linear format (because the next step can't handle L8_SRGB). - * 2) Convert luminance to red. (the real hw format for luminance) - */ - surface_format = util_format_linear(surface_format); - surface_format = util_format_luminance_to_red(surface_format); - - const struct util_format_description *desc = util_format_description(surface_format); - - if (desc->block.bits == 128 && - (color->ui[0] != color->ui[1] || - color->ui[0] != color->ui[2])) - return false; - - *clear_words_needed = true; - *reset_value = 0x20202020U; - - /* If we want to clear without needing a fast clear eliminate step, we - * can set each channel to 0 or 1 (or 0/max for integer formats). We - * have two sets of flags, one for the last or first channel(extra) and - * one for the other channels(main). - */ - - if (surface_format == PIPE_FORMAT_R11G11B10_FLOAT || - surface_format == PIPE_FORMAT_B5G6R5_UNORM || - surface_format == PIPE_FORMAT_B5G6R5_SRGB || - util_format_is_alpha(surface_format)) { - extra_channel = -1; - } else if (desc->layout == UTIL_FORMAT_LAYOUT_PLAIN) { - if(si_translate_colorswap(surface_format, false) <= 1) - extra_channel = desc->nr_channels - 1; - else - extra_channel = 0; - } else - return true; - - for (i = 0; i < 4; ++i) { - int index = desc->swizzle[i] - PIPE_SWIZZLE_X; - - if (desc->swizzle[i] < PIPE_SWIZZLE_X || - desc->swizzle[i] > PIPE_SWIZZLE_W) - continue; - - if (desc->channel[i].pure_integer && - desc->channel[i].type == UTIL_FORMAT_TYPE_SIGNED) { - /* Use the maximum value for clamping the clear color. */ - int max = u_bit_consecutive(0, desc->channel[i].size - 1); - - values[i] = color->i[i] != 0; - if (color->i[i] != 0 && MIN2(color->i[i], max) != max) - return true; - } else if (desc->channel[i].pure_integer && - desc->channel[i].type == UTIL_FORMAT_TYPE_UNSIGNED) { - /* Use the maximum value for clamping the clear color. */ - unsigned max = u_bit_consecutive(0, desc->channel[i].size); - - values[i] = color->ui[i] != 0U; - if (color->ui[i] != 0U && MIN2(color->ui[i], max) != max) - return true; - } else { - values[i] = color->f[i] != 0.0F; - if (color->f[i] != 0.0F && color->f[i] != 1.0F) - return true; - } - - if (index == extra_channel) - extra_value = values[i]; - else - main_value = values[i]; - } - - for (int i = 0; i < 4; ++i) - if (values[i] != main_value && - desc->swizzle[i] - PIPE_SWIZZLE_X != extra_channel && - desc->swizzle[i] >= PIPE_SWIZZLE_X && - desc->swizzle[i] <= PIPE_SWIZZLE_W) - return true; - - *clear_words_needed = false; - if (main_value) - *reset_value |= 0x80808080U; - - if (extra_value) - *reset_value |= 0x40404040U; - return true; -} - -void vi_dcc_clear_level(struct r600_common_context *rctx, - struct r600_texture *rtex, - unsigned level, unsigned clear_value) -{ - struct pipe_resource *dcc_buffer; - uint64_t dcc_offset, clear_size; - - assert(vi_dcc_enabled(rtex, level)); - - if (rtex->dcc_separate_buffer) { - dcc_buffer = &rtex->dcc_separate_buffer->b.b; - dcc_offset = 0; - } else { - dcc_buffer = &rtex->resource.b.b; - dcc_offset = rtex->dcc_offset; - } - - if (rctx->chip_class >= GFX9) { - /* Mipmap level clears aren't implemented. */ - assert(rtex->resource.b.b.last_level == 0); - /* MSAA needs a different clear size. */ - assert(rtex->resource.b.b.nr_samples <= 1); - clear_size = rtex->surface.dcc_size; - } else { - unsigned num_layers = util_max_layer(&rtex->resource.b.b, level) + 1; - - /* If this is 0, fast clear isn't possible. (can occur with MSAA) */ - assert(rtex->surface.u.legacy.level[level].dcc_fast_clear_size); - /* Layered MSAA DCC fast clears need to clear dcc_fast_clear_size - * bytes for each layer. This is not currently implemented, and - * therefore MSAA DCC isn't even enabled with multiple layers. - */ - assert(rtex->resource.b.b.nr_samples <= 1 || num_layers == 1); - - dcc_offset += rtex->surface.u.legacy.level[level].dcc_offset; - clear_size = rtex->surface.u.legacy.level[level].dcc_fast_clear_size * - num_layers; - } - - rctx->clear_buffer(&rctx->b, dcc_buffer, dcc_offset, clear_size, - clear_value, R600_COHERENCY_CB_META); -} - -/* Set the same micro tile mode as the destination of the last MSAA resolve. - * This allows hitting the MSAA resolve fast path, which requires that both - * src and dst micro tile modes match. - */ -static void si_set_optimal_micro_tile_mode(struct r600_common_screen *rscreen, - struct r600_texture *rtex) -{ - if (rtex->resource.b.is_shared || - rtex->resource.b.b.nr_samples <= 1 || - rtex->surface.micro_tile_mode == rtex->last_msaa_resolve_target_micro_mode) - return; - - assert(rscreen->chip_class >= GFX9 || - rtex->surface.u.legacy.level[0].mode == RADEON_SURF_MODE_2D); - assert(rtex->resource.b.b.last_level == 0); - - if (rscreen->chip_class >= GFX9) { - /* 4K or larger tiles only. 0 is linear. 1-3 are 256B tiles. */ - assert(rtex->surface.u.gfx9.surf.swizzle_mode >= 4); - - /* If you do swizzle_mode % 4, you'll get: - * 0 = Depth - * 1 = Standard, - * 2 = Displayable - * 3 = Rotated - * - * Depth-sample order isn't allowed: - */ - assert(rtex->surface.u.gfx9.surf.swizzle_mode % 4 != 0); - - switch (rtex->last_msaa_resolve_target_micro_mode) { - case RADEON_MICRO_MODE_DISPLAY: - rtex->surface.u.gfx9.surf.swizzle_mode &= ~0x3; - rtex->surface.u.gfx9.surf.swizzle_mode += 2; /* D */ - break; - case RADEON_MICRO_MODE_THIN: - rtex->surface.u.gfx9.surf.swizzle_mode &= ~0x3; - rtex->surface.u.gfx9.surf.swizzle_mode += 1; /* S */ - break; - case RADEON_MICRO_MODE_ROTATED: - rtex->surface.u.gfx9.surf.swizzle_mode &= ~0x3; - rtex->surface.u.gfx9.surf.swizzle_mode += 3; /* R */ - break; - default: /* depth */ - assert(!"unexpected micro mode"); - return; - } - } else if (rscreen->chip_class >= CIK) { - /* These magic numbers were copied from addrlib. It doesn't use - * any definitions for them either. They are all 2D_TILED_THIN1 - * modes with different bpp and micro tile mode. - */ - switch (rtex->last_msaa_resolve_target_micro_mode) { - case RADEON_MICRO_MODE_DISPLAY: - rtex->surface.u.legacy.tiling_index[0] = 10; - break; - case RADEON_MICRO_MODE_THIN: - rtex->surface.u.legacy.tiling_index[0] = 14; - break; - case RADEON_MICRO_MODE_ROTATED: - rtex->surface.u.legacy.tiling_index[0] = 28; - break; - default: /* depth, thick */ - assert(!"unexpected micro mode"); - return; - } - } else { /* SI */ - switch (rtex->last_msaa_resolve_target_micro_mode) { - case RADEON_MICRO_MODE_DISPLAY: - switch (rtex->surface.bpe) { - case 1: - rtex->surface.u.legacy.tiling_index[0] = 10; - break; - case 2: - rtex->surface.u.legacy.tiling_index[0] = 11; - break; - default: /* 4, 8 */ - rtex->surface.u.legacy.tiling_index[0] = 12; - break; - } - break; - case RADEON_MICRO_MODE_THIN: - switch (rtex->surface.bpe) { - case 1: - rtex->surface.u.legacy.tiling_index[0] = 14; - break; - case 2: - rtex->surface.u.legacy.tiling_index[0] = 15; - break; - case 4: - rtex->surface.u.legacy.tiling_index[0] = 16; - break; - default: /* 8, 16 */ - rtex->surface.u.legacy.tiling_index[0] = 17; - break; - } - break; - default: /* depth, thick */ - assert(!"unexpected micro mode"); - return; - } - } - - rtex->surface.micro_tile_mode = rtex->last_msaa_resolve_target_micro_mode; - - p_atomic_inc(&rscreen->dirty_tex_counter); -} - -void si_do_fast_color_clear(struct r600_common_context *rctx, - struct pipe_framebuffer_state *fb, - struct r600_atom *fb_state, - unsigned *buffers, ubyte *dirty_cbufs, - const union pipe_color_union *color) -{ - int i; - - /* This function is broken in BE, so just disable this path for now */ -#ifdef PIPE_ARCH_BIG_ENDIAN - return; -#endif - - if (rctx->render_cond) - return; - - for (i = 0; i < fb->nr_cbufs; i++) { - struct r600_texture *tex; - unsigned clear_bit = PIPE_CLEAR_COLOR0 << i; - unsigned level = fb->cbufs[i]->u.tex.level; - - if (!fb->cbufs[i]) - continue; - - /* if this colorbuffer is not being cleared */ - if (!(*buffers & clear_bit)) - continue; - - tex = (struct r600_texture *)fb->cbufs[i]->texture; - - /* the clear is allowed if all layers are bound */ - if (fb->cbufs[i]->u.tex.first_layer != 0 || - fb->cbufs[i]->u.tex.last_layer != util_max_layer(&tex->resource.b.b, 0)) { - continue; - } - - /* cannot clear mipmapped textures */ - if (fb->cbufs[i]->texture->last_level != 0) { - continue; - } - - /* only supported on tiled surfaces */ - if (tex->surface.is_linear) { - continue; - } - - /* shared textures can't use fast clear without an explicit flush, - * because there is no way to communicate the clear color among - * all clients - */ - if (tex->resource.b.is_shared && - !(tex->resource.external_usage & PIPE_HANDLE_USAGE_EXPLICIT_FLUSH)) - continue; - - /* fast color clear with 1D tiling doesn't work on old kernels and CIK */ - if (rctx->chip_class == CIK && - tex->surface.u.legacy.level[0].mode == RADEON_SURF_MODE_1D && - rctx->screen->info.drm_major == 2 && - rctx->screen->info.drm_minor < 38) { - continue; - } - - /* Fast clear is the most appropriate place to enable DCC for - * displayable surfaces. - */ - if (rctx->chip_class >= VI && - !(rctx->screen->debug_flags & DBG(NO_DCC_FB))) { - vi_separate_dcc_try_enable(rctx, tex); - - /* RB+ isn't supported with a CMASK clear only on Stoney, - * so all clears are considered to be hypothetically slow - * clears, which is weighed when determining whether to - * enable separate DCC. - */ - if (tex->dcc_gather_statistics && - rctx->family == CHIP_STONEY) - tex->num_slow_clears++; - } - - /* Try to clear DCC first, otherwise try CMASK. */ - if (vi_dcc_enabled(tex, 0)) { - uint32_t reset_value; - bool clear_words_needed, cleared_cmask = false; - - if (rctx->screen->debug_flags & DBG(NO_DCC_CLEAR)) - continue; - - /* This can only occur with MSAA. */ - if (rctx->chip_class == VI && - !tex->surface.u.legacy.level[level].dcc_fast_clear_size) - continue; - - if (!vi_get_fast_clear_parameters(fb->cbufs[i]->format, - color, &reset_value, - &clear_words_needed)) - continue; - - /* DCC fast clear with MSAA should clear CMASK to 0xC. */ - if (tex->resource.b.b.nr_samples >= 2 && tex->cmask.size) { - /* TODO: This doesn't work with MSAA. */ - if (clear_words_needed) - continue; - - rctx->clear_buffer(&rctx->b, &tex->cmask_buffer->b.b, - tex->cmask.offset, tex->cmask.size, - 0xCCCCCCCC, R600_COHERENCY_CB_META); - cleared_cmask = true; - } - - vi_dcc_clear_level(rctx, tex, 0, reset_value); - - if (clear_words_needed || cleared_cmask) { - bool need_compressed_update = !tex->dirty_level_mask; - - tex->dirty_level_mask |= 1 << level; - - if (need_compressed_update) - p_atomic_inc(&rctx->screen->compressed_colortex_counter); - } - tex->separate_dcc_dirty = true; - } else { - /* 128-bit formats are unusupported */ - if (tex->surface.bpe > 8) { - continue; - } - - /* RB+ doesn't work with CMASK fast clear on Stoney. */ - if (rctx->family == CHIP_STONEY) - continue; - - /* ensure CMASK is enabled */ - r600_texture_alloc_cmask_separate(rctx->screen, tex); - if (tex->cmask.size == 0) { - continue; - } - - /* Do the fast clear. */ - rctx->clear_buffer(&rctx->b, &tex->cmask_buffer->b.b, - tex->cmask.offset, tex->cmask.size, 0, - R600_COHERENCY_CB_META); - - bool need_compressed_update = !tex->dirty_level_mask; - - tex->dirty_level_mask |= 1 << level; - - if (need_compressed_update) - p_atomic_inc(&rctx->screen->compressed_colortex_counter); - } - - /* We can change the micro tile mode before a full clear. */ - si_set_optimal_micro_tile_mode(rctx->screen, tex); - - evergreen_set_clear_color(tex, fb->cbufs[i]->format, color); - - if (dirty_cbufs) - *dirty_cbufs |= 1 << i; - rctx->set_atom_dirty(rctx, fb_state, true); - *buffers &= ~clear_bit; - } -} - static struct pipe_memory_object * r600_memobj_from_handle(struct pipe_screen *screen, struct winsys_handle *whandle, @@ -2922,5 +2386,4 @@ void si_init_context_texture_functions(struct r600_common_context *rctx) { rctx->b.create_surface = r600_create_surface; rctx->b.surface_destroy = r600_surface_destroy; - rctx->b.clear_texture = r600_clear_texture; } |