summaryrefslogtreecommitdiffstats
path: root/src/gallium/drivers/r300
diff options
context:
space:
mode:
Diffstat (limited to 'src/gallium/drivers/r300')
-rw-r--r--src/gallium/drivers/r300/Makefile10
-rw-r--r--src/gallium/drivers/r300/SConscript12
-rw-r--r--src/gallium/drivers/r300/r300_blit.c130
-rw-r--r--src/gallium/drivers/r300/r300_blit.h (renamed from src/gallium/drivers/r300/r300_clear.h)26
-rw-r--r--src/gallium/drivers/r300/r300_chipset.c3
-rw-r--r--src/gallium/drivers/r300/r300_chipset.h6
-rw-r--r--src/gallium/drivers/r300/r300_clear.c35
-rw-r--r--src/gallium/drivers/r300/r300_context.c185
-rw-r--r--src/gallium/drivers/r300/r300_context.h108
-rw-r--r--src/gallium/drivers/r300/r300_cs.h20
-rw-r--r--src/gallium/drivers/r300/r300_debug.c8
-rw-r--r--src/gallium/drivers/r300/r300_emit.c675
-rw-r--r--src/gallium/drivers/r300/r300_emit.h32
-rw-r--r--src/gallium/drivers/r300/r300_flush.c27
-rw-r--r--src/gallium/drivers/r300/r300_flush.h7
-rw-r--r--src/gallium/drivers/r300/r300_fs.c207
-rw-r--r--src/gallium/drivers/r300/r300_fs.h42
-rw-r--r--src/gallium/drivers/r300/r300_query.c73
-rw-r--r--src/gallium/drivers/r300/r300_query.h4
-rw-r--r--src/gallium/drivers/r300/r300_reg.h163
-rw-r--r--src/gallium/drivers/r300/r300_render.c432
-rw-r--r--src/gallium/drivers/r300/r300_render.h59
-rw-r--r--src/gallium/drivers/r300/r300_screen.c168
-rw-r--r--src/gallium/drivers/r300/r300_screen.h9
-rw-r--r--src/gallium/drivers/r300/r300_shader_inlines.h47
-rw-r--r--src/gallium/drivers/r300/r300_shader_semantics.h66
-rw-r--r--src/gallium/drivers/r300/r300_state.c592
-rw-r--r--src/gallium/drivers/r300/r300_state_derived.c782
-rw-r--r--src/gallium/drivers/r300/r300_state_derived.h4
-rw-r--r--src/gallium/drivers/r300/r300_state_inlines.h209
-rw-r--r--src/gallium/drivers/r300/r300_state_invariant.c34
-rw-r--r--src/gallium/drivers/r300/r300_state_invariant.h6
-rw-r--r--src/gallium/drivers/r300/r300_surface.c372
-rw-r--r--src/gallium/drivers/r300/r300_surface.h124
-rw-r--r--src/gallium/drivers/r300/r300_texture.c189
-rw-r--r--src/gallium/drivers/r300/r300_texture.h40
-rw-r--r--src/gallium/drivers/r300/r300_tgsi_to_rc.c56
-rw-r--r--src/gallium/drivers/r300/r300_vs.c430
-rw-r--r--src/gallium/drivers/r300/r300_vs.h20
-rw-r--r--src/gallium/drivers/r300/r300_winsys.h61
-rw-r--r--src/gallium/drivers/r300/r3xx_fs.c74
-rw-r--r--src/gallium/drivers/r300/r3xx_fs.h32
-rw-r--r--src/gallium/drivers/r300/r5xx_fs.c125
-rw-r--r--src/gallium/drivers/r300/r5xx_fs.h32
44 files changed, 3452 insertions, 2284 deletions
diff --git a/src/gallium/drivers/r300/Makefile b/src/gallium/drivers/r300/Makefile
index 69e4724790a..afddcb161fa 100644
--- a/src/gallium/drivers/r300/Makefile
+++ b/src/gallium/drivers/r300/Makefile
@@ -4,10 +4,8 @@ include $(TOP)/configs/current
LIBNAME = r300
C_SOURCES = \
- r3xx_fs.c \
- r5xx_fs.c \
+ r300_blit.c \
r300_chipset.c \
- r300_clear.c \
r300_context.c \
r300_debug.c \
r300_emit.c \
@@ -20,12 +18,12 @@ C_SOURCES = \
r300_state_derived.c \
r300_state_invariant.c \
r300_vs.c \
- r300_surface.c \
r300_texture.c \
r300_tgsi_to_rc.c
LIBRARY_INCLUDES = \
- -I$(TOP)/src/mesa/drivers/dri/r300/compiler
+ -I$(TOP)/src/mesa/drivers/dri/r300/compiler \
+ -I$(TOP)/src/gallium/winsys/drm/radeon/core
COMPILER_ARCHIVE = $(TOP)/src/mesa/drivers/dri/r300/compiler/libr300compiler.a
@@ -37,4 +35,4 @@ include ../../Makefile.template
.PHONY : $(COMPILER_ARCHIVE)
$(COMPILER_ARCHIVE):
- cd $(TOP)/src/mesa/drivers/dri/r300/compiler; make
+ $(MAKE) -C $(TOP)/src/mesa/drivers/dri/r300/compiler
diff --git a/src/gallium/drivers/r300/SConscript b/src/gallium/drivers/r300/SConscript
index b4c8ba2015a..183aa17f9b3 100644
--- a/src/gallium/drivers/r300/SConscript
+++ b/src/gallium/drivers/r300/SConscript
@@ -4,15 +4,18 @@ r300compiler = SConscript('#/src/mesa/drivers/dri/r300/compiler/SConscript')
env = env.Clone()
# add the paths for r300compiler
-env.Append(CPPPATH = ['#/src/mesa/drivers/dri/r300/compiler', '#/include', '#/src/mesa'])
+env.Append(CPPPATH = [
+ '#/src/mesa/drivers/dri/r300/compiler',
+ '#/src/gallium/winsys/drm/radeon/core',
+ '#/include',
+ '#/src/mesa',
+])
r300 = env.ConvenienceLibrary(
target = 'r300',
source = [
- 'r3xx_fs.c',
- 'r5xx_fs.c',
+ 'r300_blit.c',
'r300_chipset.c',
- 'r300_clear.c',
'r300_context.c',
'r300_debug.c',
'r300_emit.c',
@@ -25,7 +28,6 @@ r300 = env.ConvenienceLibrary(
'r300_state_derived.c',
'r300_state_invariant.c',
'r300_vs.c',
- 'r300_surface.c',
'r300_texture.c',
'r300_tgsi_to_rc.c',
] + r300compiler) + r300compiler
diff --git a/src/gallium/drivers/r300/r300_blit.c b/src/gallium/drivers/r300/r300_blit.c
new file mode 100644
index 00000000000..ffe066d5369
--- /dev/null
+++ b/src/gallium/drivers/r300/r300_blit.c
@@ -0,0 +1,130 @@
+/*
+ * 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
+ * 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_blit.h"
+#include "r300_context.h"
+
+#include "util/u_rect.h"
+
+static void r300_blitter_save_states(struct r300_context* r300)
+{
+ util_blitter_save_blend(r300->blitter, r300->blend_state);
+ util_blitter_save_depth_stencil_alpha(r300->blitter, r300->dsa_state);
+ util_blitter_save_rasterizer(r300->blitter, r300->rs_state);
+ util_blitter_save_fragment_shader(r300->blitter, r300->fs);
+ util_blitter_save_vertex_shader(r300->blitter, r300->vs);
+}
+
+/* Clear currently bound buffers. */
+void r300_clear(struct pipe_context* pipe,
+ unsigned buffers,
+ const float* rgba,
+ double depth,
+ unsigned stencil)
+{
+ /* XXX Implement fastfill.
+ *
+ * If fastfill is enabled, a few facts should be considered:
+ *
+ * 1) Zbuffer must be micro-tiled and whole microtiles must be
+ * written.
+ *
+ * 2) ZB_DEPTHCLEARVALUE is used to clear a zbuffer and Z Mask must be
+ * equal to 0.
+ *
+ * 3) RB3D_COLOR_CLEAR_VALUE is used to clear a colorbuffer and
+ * RB3D_COLOR_CHANNEL_MASK must be equal to 0.
+ *
+ * 4) ZB_CB_CLEAR can be used to make the ZB units help in clearing
+ * the colorbuffer. The color clear value is supplied through both
+ * RB3D_COLOR_CLEAR_VALUE and ZB_DEPTHCLEARVALUE, and the colorbuffer
+ * must be set in ZB_DEPTHOFFSET and ZB_DEPTHPITCH in addition to
+ * RB3D_COLOROFFSET and RB3D_COLORPITCH. It's obvious that the zbuffer
+ * will not be cleared and multiple render targets cannot be cleared
+ * this way either.
+ *
+ * 5) For 16-bit integer buffering, compression causes a hung with one or
+ * two samples and should not be used.
+ *
+ * 6) 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?)
+ *
+ * - Marek
+ */
+
+ struct r300_context* r300 = r300_context(pipe);
+
+ r300_blitter_save_states(r300);
+
+ util_blitter_clear(r300->blitter,
+ r300->framebuffer_state.width,
+ r300->framebuffer_state.height,
+ r300->framebuffer_state.nr_cbufs,
+ buffers, rgba, depth, stencil);
+}
+
+/* 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)
+{
+ 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->framebuffer_state);
+
+ util_blitter_save_fragment_sampler_states(
+ r300->blitter, r300->sampler_count, (void**)r300->sampler_states);
+
+ util_blitter_save_fragment_sampler_textures(
+ r300->blitter, r300->texture_count,
+ (struct pipe_texture**)r300->textures);
+
+ /* Do a copy */
+ util_blitter_copy(r300->blitter,
+ dst, dstx, dsty, src, srcx, srcy, width, height, TRUE);
+}
+
+/* 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)
+{
+ struct r300_context* r300 = r300_context(pipe);
+
+ r300_blitter_save_states(r300);
+ util_blitter_save_framebuffer(r300->blitter, &r300->framebuffer_state);
+
+ util_blitter_fill(r300->blitter,
+ dst, dstx, dsty, width, height, value);
+}
diff --git a/src/gallium/drivers/r300/r300_clear.h b/src/gallium/drivers/r300/r300_blit.h
index cd5900565e8..029e4f98e7d 100644
--- a/src/gallium/drivers/r300/r300_clear.h
+++ b/src/gallium/drivers/r300/r300_blit.h
@@ -1,5 +1,5 @@
/*
- * Copyright 2008 Corbin Simpson <[email protected]>
+ * 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"),
@@ -20,12 +20,11 @@
* OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
* USE OR OTHER DEALINGS IN THE SOFTWARE. */
-#ifndef R300_CLEAR_H
-#define R300_CLEAR_H
+#ifndef R300_BLIT_H
+#define R300_BLIT_H
-#include "util/u_clear.h"
-
-#include "r300_context.h"
+struct pipe_context;
+struct pipe_surface;
void r300_clear(struct pipe_context* pipe,
unsigned buffers,
@@ -33,4 +32,17 @@ void r300_clear(struct pipe_context* pipe,
double depth,
unsigned stencil);
-#endif /* R300_CLEAR_H */
+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_chipset.c b/src/gallium/drivers/r300/r300_chipset.c
index d138866d33c..51fdb82ff34 100644
--- a/src/gallium/drivers/r300/r300_chipset.c
+++ b/src/gallium/drivers/r300/r300_chipset.c
@@ -21,6 +21,7 @@
* USE OR OTHER DEALINGS IN THE SOFTWARE. */
#include "r300_chipset.h"
+
#include "util/u_debug.h"
/* r300_chipset: A file all to itself for deducing the various properties of
@@ -31,7 +32,7 @@ void r300_parse_chipset(struct r300_capabilities* caps)
{
/* Reasonable defaults */
caps->num_vert_fpus = 4;
- caps->has_tcl = getenv("RADEON_NO_TCL") ? FALSE : TRUE;
+ caps->has_tcl = debug_get_bool_option("RADEON_NO_TCL", FALSE) ? FALSE : TRUE;
caps->is_r500 = FALSE;
caps->high_second_pipe = FALSE;
diff --git a/src/gallium/drivers/r300/r300_chipset.h b/src/gallium/drivers/r300/r300_chipset.h
index 322d4a57e41..0633a8b8a72 100644
--- a/src/gallium/drivers/r300/r300_chipset.h
+++ b/src/gallium/drivers/r300/r300_chipset.h
@@ -33,9 +33,11 @@ struct r300_capabilities {
/* Chipset family */
int family;
/* The number of vertex floating-point units */
- int num_vert_fpus;
+ unsigned num_vert_fpus;
/* The number of fragment pipes */
- int num_frag_pipes;
+ unsigned num_frag_pipes;
+ /* The number of z pipes */
+ unsigned num_z_pipes;
/* Whether or not TCL is physically present */
boolean has_tcl;
/* Whether or not this is an RV515 or newer; R500s have many differences
diff --git a/src/gallium/drivers/r300/r300_clear.c b/src/gallium/drivers/r300/r300_clear.c
deleted file mode 100644
index 8b9cb819ae6..00000000000
--- a/src/gallium/drivers/r300/r300_clear.c
+++ /dev/null
@@ -1,35 +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. */
-
-#include "r300_clear.h"
-
-/* Clears currently bound buffers. */
-void r300_clear(struct pipe_context* pipe,
- unsigned buffers,
- const float* rgba,
- double depth,
- unsigned stencil)
-{
- /* XXX we can and should do one clear if both color and zs are set */
- util_clear(pipe, &r300_context(pipe)->framebuffer_state,
- buffers, rgba, depth, stencil);
-}
diff --git a/src/gallium/drivers/r300/r300_context.c b/src/gallium/drivers/r300/r300_context.c
index 9cc455135db..d5c2d63d393 100644
--- a/src/gallium/drivers/r300/r300_context.c
+++ b/src/gallium/drivers/r300/r300_context.c
@@ -20,178 +20,167 @@
* 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_flush.h"
-#include "r300_state_invariant.h"
-
-static boolean r300_draw_range_elements(struct pipe_context* pipe,
- struct pipe_buffer* indexBuffer,
- unsigned indexSize,
- unsigned minIndex,
- unsigned maxIndex,
- unsigned mode,
- unsigned start,
- unsigned count)
-{
- struct r300_context* r300 = r300_context(pipe);
- int i;
-
- for (i = 0; i < r300->vertex_buffer_count; i++) {
- void* buf = pipe_buffer_map(pipe->screen,
- r300->vertex_buffers[i].buffer,
- PIPE_BUFFER_USAGE_CPU_READ);
- draw_set_mapped_vertex_buffer(r300->draw, i, buf);
- }
-
- if (indexBuffer) {
- void* indices = pipe_buffer_map(pipe->screen, indexBuffer,
- PIPE_BUFFER_USAGE_CPU_READ);
- draw_set_mapped_element_buffer_range(r300->draw, indexSize,
- minIndex, maxIndex, indices);
- } else {
- draw_set_mapped_element_buffer(r300->draw, 0, NULL);
- }
-
- draw_set_mapped_constant_buffer(r300->draw,
- r300->shader_constants[PIPE_SHADER_VERTEX].constants,
- r300->shader_constants[PIPE_SHADER_VERTEX].count *
- (sizeof(float) * 4));
-
- draw_arrays(r300->draw, mode, start, count);
+#include "draw/draw_context.h"
- for (i = 0; i < r300->vertex_buffer_count; i++) {
- pipe_buffer_unmap(pipe->screen, r300->vertex_buffers[i].buffer);
- draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
- }
+#include "tgsi/tgsi_scan.h"
- if (indexBuffer) {
- pipe_buffer_unmap(pipe->screen, indexBuffer);
- draw_set_mapped_element_buffer_range(r300->draw, 0, start,
- start + count - 1, NULL);
- }
+#include "util/u_hash_table.h"
+#include "util/u_memory.h"
+#include "util/u_simple_list.h"
- return TRUE;
-}
+#include "r300_blit.h"
+#include "r300_context.h"
+#include "r300_flush.h"
+#include "r300_query.h"
+#include "r300_render.h"
+#include "r300_screen.h"
+#include "r300_state_derived.h"
+#include "r300_state_invariant.h"
+#include "r300_texture.h"
+#include "r300_winsys.h"
-static boolean r300_draw_elements(struct pipe_context* pipe,
- struct pipe_buffer* indexBuffer,
- unsigned indexSize, unsigned mode,
- unsigned start, unsigned count)
+static enum pipe_error r300_clear_hash_table(void* key, void* value,
+ void* data)
{
- return r300_draw_range_elements(pipe, indexBuffer, indexSize, 0, ~0,
- mode, start, count);
+ FREE(key);
+ FREE(value);
+ return PIPE_OK;
}
-static boolean r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
- unsigned start, unsigned count)
+static void r300_destroy_context(struct pipe_context* context)
{
- return r300_draw_elements(pipe, NULL, 0, mode, start, count);
-}
-
-static void r300_destroy_context(struct pipe_context* context) {
struct r300_context* r300 = r300_context(context);
struct r300_query* query, * temp;
+ util_blitter_destroy(r300->blitter);
+
+ util_hash_table_foreach(r300->shader_hash_table, r300_clear_hash_table,
+ NULL);
+ util_hash_table_destroy(r300->shader_hash_table);
+
draw_destroy(r300->draw);
/* Free the OQ BO. */
context->screen->buffer_destroy(r300->oqbo);
/* If there are any queries pending or not destroyed, remove them now. */
- if (r300->query_list) {
- foreach_s(query, temp, r300->query_list) {
- remove_from_list(query);
- FREE(query);
- }
+ foreach_s(query, temp, &r300->query_list) {
+ remove_from_list(query);
+ FREE(query);
}
FREE(r300->blend_color_state);
FREE(r300->rs_block);
FREE(r300->scissor_state);
+ FREE(r300->vertex_info);
FREE(r300->viewport_state);
FREE(r300);
}
static unsigned int
-r300_is_texture_referenced( struct pipe_context *pipe,
- struct pipe_texture *texture,
- unsigned face, unsigned level)
+r300_is_texture_referenced(struct pipe_context *pipe,
+ struct pipe_texture *texture,
+ unsigned face, unsigned level)
{
- /**
- * FIXME: Optimize.
- */
+ struct pipe_buffer* buf = 0;
- return PIPE_REFERENCED_FOR_READ | PIPE_REFERENCED_FOR_WRITE;
+ r300_get_texture_buffer(texture, &buf, NULL);
+
+ return pipe->is_buffer_referenced(pipe, buf);
}
static unsigned int
-r300_is_buffer_referenced( struct pipe_context *pipe,
- struct pipe_buffer *buf)
+r300_is_buffer_referenced(struct pipe_context *pipe,
+ struct pipe_buffer *buf)
+{
+ /* This only checks to see whether actual hardware buffers are
+ * referenced. Since we use managed BOs and transfers, it's actually not
+ * possible for pipe_buffers to ever reference the actual hardware, so
+ * buffers are never referenced. */
+ return 0;
+}
+
+static void r300_flush_cb(void *data)
{
- /**
- * FIXME: Optimize.
- */
+ struct r300_context* const cs_context_copy = data;
- return PIPE_REFERENCED_FOR_READ | PIPE_REFERENCED_FOR_WRITE;
+ cs_context_copy->context.flush(&cs_context_copy->context, 0, NULL);
}
struct pipe_context* r300_create_context(struct pipe_screen* screen,
- struct r300_winsys* r300_winsys)
+ struct radeon_winsys* radeon_winsys)
{
struct r300_context* r300 = CALLOC_STRUCT(r300_context);
+ struct r300_screen* r300screen = r300_screen(screen);
if (!r300)
return NULL;
- r300->winsys = r300_winsys;
+ r300->winsys = radeon_winsys;
- r300->context.winsys = (struct pipe_winsys*)r300_winsys;
- r300->context.screen = r300_screen(screen);
+ r300->context.winsys = (struct pipe_winsys*)radeon_winsys;
+ r300->context.screen = screen;
r300_init_debug(r300);
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;
- 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->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;
+ } 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;
+
+ /* Create a Draw. This is used for SW TCL. */
+ r300->draw = draw_create();
+ /* 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);
+ }
r300->context.is_texture_referenced = r300_is_texture_referenced;
r300->context.is_buffer_referenced = r300_is_buffer_referenced;
+ r300->shader_hash_table = util_hash_table_create(r300_shader_key_hash,
+ r300_shader_key_compare);
+
r300->blend_color_state = CALLOC_STRUCT(r300_blend_color_state);
r300->rs_block = CALLOC_STRUCT(r300_rs_block);
r300->scissor_state = CALLOC_STRUCT(r300_scissor_state);
+ r300->vertex_info = CALLOC_STRUCT(r300_vertex_info);
r300->viewport_state = CALLOC_STRUCT(r300_viewport_state);
- /* Create a Draw. This is used for vert collation and SW TCL. */
- r300->draw = draw_create();
- /* Enable our renderer. */
- draw_set_rasterize_stage(r300->draw, r300_draw_stage(r300));
- /* Disable Draw's clipping if TCL is present. */
- draw_set_driver_clipping(r300->draw, r300_screen(screen)->caps->has_tcl);
- /* Force Draw to never do viewport transform, since (again) we can do
- * transform in hardware, always. */
- draw_set_viewport_state(r300->draw, &r300_viewport_identity);
-
/* Open up the OQ BO. */
r300->oqbo = screen->buffer_create(screen, 4096,
PIPE_BUFFER_USAGE_VERTEX, 4096);
+ make_empty_list(&r300->query_list);
r300_init_flush_functions(r300);
r300_init_query_functions(r300);
- r300_init_surface_functions(r300);
+ /* r300_init_surface_functions(r300); */
r300_init_state_functions(r300);
r300_emit_invariant_state(r300);
+
+ r300->winsys->set_flush_cb(r300->winsys, r300_flush_cb, r300);
r300->dirty_state = R300_NEW_KITCHEN_SINK;
r300->dirty_hw++;
+ r300->blitter = util_blitter_create(&r300->context);
+
return &r300->context;
}
diff --git a/src/gallium/drivers/r300/r300_context.h b/src/gallium/drivers/r300/r300_context.h
index 52b1c9a6b26..232530b7dc4 100644
--- a/src/gallium/drivers/r300/r300_context.h
+++ b/src/gallium/drivers/r300/r300_context.h
@@ -23,20 +23,12 @@
#ifndef R300_CONTEXT_H
#define R300_CONTEXT_H
-#include "draw/draw_context.h"
#include "draw/draw_vertex.h"
-#include "pipe/p_context.h"
-
-#include "tgsi/tgsi_scan.h"
+#include "util/u_blitter.h"
-#include "util/u_memory.h"
-#include "util/u_simple_list.h"
-
-#include "r300_clear.h"
-#include "r300_query.h"
-#include "r300_screen.h"
-#include "r300_winsys.h"
+#include "pipe/p_context.h"
+#include "pipe/p_inlines.h"
struct r300_fragment_shader;
struct r300_vertex_shader;
@@ -44,6 +36,7 @@ struct r300_vertex_shader;
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 */
};
@@ -62,7 +55,6 @@ struct r300_dsa_state {
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 z_buffer_top; /* R300_ZB_ZTOP: 0x4f14 */
uint32_t stencil_ref_bf; /* R500_ZB_STENCILREFMASK_BF: 0x4fd4 */
};
@@ -88,6 +80,7 @@ struct r300_rs_state {
uint32_t line_stipple_config; /* R300_GA_LINE_STIPPLE_CONFIG: 0x4328 */
uint32_t line_stipple_value; /* R300_GA_LINE_STIPPLE_VALUE: 0x4260 */
uint32_t color_control; /* R300_GA_COLOR_CONTROL: 0x4278 */
+ uint32_t polygon_mode; /* R300_GA_POLY_MODE: 0x4288 */
};
struct r300_rs_block {
@@ -98,14 +91,28 @@ struct r300_rs_block {
};
struct r300_sampler_state {
+ struct pipe_sampler_state 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 */
+ unsigned min_lod, max_lod;
+};
+
+struct r300_scissor_regs {
+ uint32_t top_left; /* R300_SC_SCISSORS_TL: 0x43e0 */
+ uint32_t bottom_right; /* R300_SC_SCISSORS_BR: 0x43e4 */
+
+ /* Whether everything is culled by scissoring. */
+ boolean empty_area;
};
struct r300_scissor_state {
- uint32_t scissor_top_left; /* R300_SC_SCISSORS_TL: 0x43e0 */
- uint32_t scissor_bottom_right; /* R300_SC_SCISSORS_BR: 0x43e4 */
+ struct r300_scissor_regs framebuffer;
+ struct r300_scissor_regs scissor;
};
struct r300_texture_state {
@@ -124,13 +131,17 @@ struct r300_viewport_state {
uint32_t vte_control; /* R300_VAP_VTE_CNTL: 0x20b0 */
};
+struct r300_ztop_state {
+ uint32_t z_buffer_top; /* R300_ZB_ZTOP: 0x4f14 */
+};
+
#define R300_NEW_BLEND 0x00000001
#define R300_NEW_BLEND_COLOR 0x00000002
#define R300_NEW_CLIP 0x00000004
-#define R300_NEW_CONSTANTS 0x00000008
-#define R300_NEW_DSA 0x00000010
-#define R300_NEW_FRAMEBUFFERS 0x00000020
-#define R300_NEW_FRAGMENT_SHADER 0x00000040
+#define R300_NEW_DSA 0x00000008
+#define R300_NEW_FRAMEBUFFERS 0x00000010
+#define R300_NEW_FRAGMENT_SHADER 0x00000020
+#define R300_NEW_FRAGMENT_SHADER_CONSTANTS 0x00000040
#define R300_NEW_RASTERIZER 0x00000080
#define R300_NEW_RS_BLOCK 0x00000100
#define R300_NEW_SAMPLER 0x00000200
@@ -140,8 +151,10 @@ struct r300_viewport_state {
#define R300_ANY_NEW_TEXTURES 0x03fc0000
#define R300_NEW_VERTEX_FORMAT 0x04000000
#define R300_NEW_VERTEX_SHADER 0x08000000
-#define R300_NEW_VIEWPORT 0x10000000
-#define R300_NEW_KITCHEN_SINK 0x1fffffff
+#define R300_NEW_VERTEX_SHADER_CONSTANTS 0x10000000
+#define R300_NEW_VIEWPORT 0x20000000
+#define R300_NEW_QUERY 0x40000000
+#define R300_NEW_KITCHEN_SINK 0x7fffffff
/* The next several objects are not pure Radeon state; they inherit from
* various Gallium classes. */
@@ -172,6 +185,10 @@ struct r300_query {
unsigned int count;
/* The offset of this query into the query buffer, in bytes. */
unsigned offset;
+ /* if we've flushed the query */
+ boolean flushed;
+ /* if begin has been emitted */
+ boolean begin_emitted;
/* Linked list members. */
struct r300_query* prev;
struct r300_query* next;
@@ -184,6 +201,12 @@ struct r300_texture {
/* Offsets into the buffer. */
unsigned offset[PIPE_MAX_TEXTURE_LEVELS];
+ /* A pitch for each mip-level */
+ unsigned pitch[PIPE_MAX_TEXTURE_LEVELS];
+
+ /* Size of one zslice or face based on the texture target */
+ unsigned layer_size[PIPE_MAX_TEXTURE_LEVELS];
+
/**
* If non-zero, override the natural texture layout with
* a custom stride (in bytes).
@@ -197,6 +220,11 @@ struct r300_texture {
/* 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;
+
/* Pipe buffer backing this texture. */
struct pipe_buffer* buffer;
@@ -204,18 +232,14 @@ struct r300_texture {
struct r300_texture_state state;
};
-struct r300_vertex_format {
+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];
- /* Map of vertex attributes into PVS memory for HW TCL,
- * or GA memory for SW TCL. */
- int vs_tab[16];
- /* Map of rasterizer attributes from GB through RS to US. */
- int fs_tab[16];
};
extern struct pipe_viewport_state r300_viewport_identity;
@@ -225,9 +249,11 @@ struct r300_context {
struct pipe_context context;
/* The interface to the windowing system, etc. */
- struct r300_winsys* winsys;
+ struct radeon_winsys* winsys;
/* Draw module. Used mostly for SW TCL. */
struct draw_context* draw;
+ /* Accelerated blit support. */
+ struct blitter_context* blitter;
/* Vertex buffer for rendering. */
struct pipe_buffer* vbo;
@@ -237,7 +263,14 @@ struct r300_context {
/* Occlusion query buffer. */
struct pipe_buffer* oqbo;
/* Query list. */
- struct r300_query* query_list;
+ struct r300_query *query_current;
+ struct r300_query query_list;
+
+ /* Shader hash table. Used to store vertex formatting information, which
+ * depends on the combination of both currently loaded shaders. */
+ struct util_hash_table* shader_hash_table;
+ /* Vertex formatting information. */
+ struct r300_vertex_info* vertex_info;
/* Various CSO state objects. */
/* Blend state. */
@@ -266,15 +299,20 @@ struct r300_context {
/* Texture states. */
struct r300_texture* textures[8];
int texture_count;
- /* Vertex buffers for Gallium. */
- struct pipe_vertex_buffer vertex_buffers[PIPE_MAX_ATTRIBS];
- int vertex_buffer_count;
- /* Vertex information. */
- struct r300_vertex_format vertex_info;
/* Vertex shader. */
struct r300_vertex_shader* vs;
/* Viewport state. */
struct r300_viewport_state* viewport_state;
+ /* ZTOP state. */
+ struct r300_ztop_state ztop_state;
+
+ /* Vertex buffers for Gallium. */
+ struct pipe_vertex_buffer vertex_buffer[PIPE_MAX_ATTRIBS];
+ int vertex_buffer_count;
+ /* Vertex elements for Gallium. */
+ struct pipe_vertex_element vertex_element[PIPE_MAX_ATTRIBS];
+ int vertex_element_count;
+
/* Bitmask of dirty state objects. */
uint32_t dirty_state;
/* Flag indicating whether or not the HW is dirty. */
@@ -312,11 +350,13 @@ void r300_init_surface_functions(struct r300_context* r300);
#define DBG_VP 0x0000004
#define DBG_CS 0x0000008
#define DBG_DRAW 0x0000010
+#define DBG_TEX 0x0000020
+#define DBG_FALL 0x0000040
/*@}*/
static INLINE boolean DBG_ON(struct r300_context * ctx, unsigned flags)
{
- return (ctx->debug & flags) ? true : false;
+ return (ctx->debug & flags) ? TRUE : FALSE;
}
static INLINE void DBG(struct r300_context * ctx, unsigned flags, const char * fmt, ...)
diff --git a/src/gallium/drivers/r300/r300_cs.h b/src/gallium/drivers/r300/r300_cs.h
index 883f0a02dc5..d142fee0502 100644
--- a/src/gallium/drivers/r300/r300_cs.h
+++ b/src/gallium/drivers/r300/r300_cs.h
@@ -26,7 +26,8 @@
#include "util/u_math.h"
#include "r300_reg.h"
-#include "r300_winsys.h"
+
+#include "radeon_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
@@ -34,8 +35,8 @@
#define MAX_CS_SIZE 64 * 1024 / 4
-#define VERY_VERBOSE_CS 0
-#define VERY_VERBOSE_REGISTERS 0
+#define VERY_VERBOSE_CS 1
+#define VERY_VERBOSE_REGISTERS 1
/* XXX stolen from radeon_drm.h */
#define RADEON_GEM_DOMAIN_CPU 0x1
@@ -50,11 +51,11 @@
#define CS_LOCALS(context) \
struct r300_context* const cs_context_copy = (context); \
- struct r300_winsys* cs_winsys = cs_context_copy->winsys; \
+ struct radeon_winsys* cs_winsys = cs_context_copy->winsys; \
int cs_count = 0;
#define CHECK_CS(size) \
- cs_winsys->check_cs(cs_winsys, (size))
+ assert(cs_winsys->check_cs(cs_winsys, (size)))
#define BEGIN_CS(size) do { \
CHECK_CS(size); \
@@ -114,6 +115,15 @@
cs_count -= 3; \
} while (0)
+#define OUT_CS_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); \
+ assert(bo); \
+ cs_winsys->write_cs_reloc(cs_winsys, bo, rd, wd, flags); \
+ 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__, \
diff --git a/src/gallium/drivers/r300/r300_debug.c b/src/gallium/drivers/r300/r300_debug.c
index 85d69c07479..2a6ed54ac9b 100644
--- a/src/gallium/drivers/r300/r300_debug.c
+++ b/src/gallium/drivers/r300/r300_debug.c
@@ -37,6 +37,8 @@ static struct debug_option debug_options[] = {
{ "vp", DBG_VP, "Vertex program handling" },
{ "cs", DBG_CS, "Command submissions" },
{ "draw", DBG_DRAW, "Draw and emit" },
+ { "tex", DBG_TEX, "Textures" },
+ { "fall", DBG_FALL, "Fallbacks" },
{ "all", ~0, "Convenience option that enables all debug flags" },
@@ -47,7 +49,7 @@ static struct debug_option debug_options[] = {
void r300_init_debug(struct r300_context * ctx)
{
const char * options = debug_get_option("RADEON_DEBUG", 0);
- boolean printhint = false;
+ boolean printhint = FALSE;
size_t length;
struct debug_option * opt;
@@ -69,14 +71,14 @@ void r300_init_debug(struct r300_context * ctx)
if (!opt->name) {
debug_printf("Unknown debug option: %s\n", options);
- printhint = true;
+ printhint = TRUE;
}
options += length;
}
if (!ctx->debug)
- printhint = true;
+ printhint = TRUE;
}
if (printhint || ctx->debug & DBG_HELP) {
diff --git a/src/gallium/drivers/r300/r300_emit.c b/src/gallium/drivers/r300/r300_emit.c
index 77ce431cdc3..f8bfa714fef 100644
--- a/src/gallium/drivers/r300/r300_emit.c
+++ b/src/gallium/drivers/r300/r300_emit.c
@@ -1,5 +1,6 @@
/*
* Copyright 2008 Corbin Simpson <[email protected]>
+ * 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"),
@@ -22,20 +23,35 @@
/* r300_emit: Functions for emitting state. */
-#include "r300_emit.h"
+#include "util/u_format.h"
+#include "util/u_math.h"
+#include "r300_context.h"
+#include "r300_cs.h"
+#include "r300_emit.h"
#include "r300_fs.h"
+#include "r300_screen.h"
#include "r300_state_derived.h"
+#include "r300_state_inlines.h"
+#include "r300_texture.h"
#include "r300_vs.h"
void r300_emit_blend_state(struct r300_context* r300,
struct r300_blend_state* blend)
{
CS_LOCALS(r300);
- BEGIN_CS(7);
- OUT_CS_REG_SEQ(R300_RB3D_CBLEND, 2);
- OUT_CS(blend->blend_control);
- OUT_CS(blend->alpha_blend_control);
+ BEGIN_CS(8);
+ OUT_CS_REG_SEQ(R300_RB3D_CBLEND, 3);
+ if (r300->framebuffer_state.nr_cbufs) {
+ OUT_CS(blend->blend_control);
+ OUT_CS(blend->alpha_blend_control);
+ OUT_CS(blend->color_channel_mask);
+ } else {
+ OUT_CS(0);
+ OUT_CS(0);
+ OUT_CS(0);
+ /* XXX also disable fastfill here once it's supported */
+ }
OUT_CS_REG(R300_RB3D_ROPCNTL, blend->rop);
OUT_CS_REG(R300_RB3D_DITHER_CTL, blend->dither);
END_CS;
@@ -96,19 +112,30 @@ void r300_emit_dsa_state(struct r300_context* r300,
struct r300_screen* r300screen = r300_screen(r300->context.screen);
CS_LOCALS(r300);
- BEGIN_CS(r300screen->caps->is_r500 ? 8 : 8);
+ BEGIN_CS(r300screen->caps->is_r500 ? 10 : 8);
OUT_CS_REG(R300_FG_ALPHA_FUNC, dsa->alpha_function);
- /* XXX figure out the r300 counterpart for this */
- if (r300screen->caps->is_r500) {
- /* OUT_CS_REG(R500_FG_ALPHA_VALUE, dsa->alpha_reference); */
- }
+
+ /* not needed since we use the 8bit alpha ref */
+ /*if (r300screen->caps->is_r500) {
+ OUT_CS_REG(R500_FG_ALPHA_VALUE, dsa->alpha_reference);
+ }*/
+
OUT_CS_REG_SEQ(R300_ZB_CNTL, 3);
- OUT_CS(dsa->z_buffer_control);
- OUT_CS(dsa->z_stencil_control);
+
+ if (r300->framebuffer_state.zsbuf) {
+ OUT_CS(dsa->z_buffer_control);
+ OUT_CS(dsa->z_stencil_control);
+ } else {
+ OUT_CS(0);
+ OUT_CS(0);
+ }
+
OUT_CS(dsa->stencil_ref_mask);
- OUT_CS_REG(R300_ZB_ZTOP, dsa->z_buffer_top);
+ OUT_CS_REG(R300_ZB_ZTOP, r300->ztop_state.z_buffer_top);
+
+ /* XXX it seems r3xx doesn't support STENCILREFMASK_BF */
if (r300screen->caps->is_r500) {
- /* OUT_CS_REG(R500_ZB_STENCILREFMASK_BF, dsa->stencil_ref_bf); */
+ OUT_CS_REG(R500_ZB_STENCILREFMASK_BF, dsa->stencil_ref_bf);
}
END_CS;
}
@@ -118,7 +145,9 @@ static const float * get_shader_constant(
struct rc_constant * constant,
struct r300_constant_buffer * externals)
{
- static const float zero[4] = { 0.0, 0.0, 0.0, 0.0 };
+ static float vec[4] = { 0.0, 0.0, 0.0, 1.0 };
+ struct pipe_texture *tex;
+
switch(constant->Type) {
case RC_CONSTANT_EXTERNAL:
return externals->constants[constant->u.External];
@@ -126,11 +155,60 @@ static const float * get_shader_constant(
case RC_CONSTANT_IMMEDIATE:
return constant->u.Immediate;
+ case RC_CONSTANT_STATE:
+ switch (constant->u.State[0]) {
+ /* Factor for converting rectangle coords to
+ * normalized coords. Should only show up on non-r500. */
+ case RC_STATE_R300_TEXRECT_FACTOR:
+ tex = &r300->textures[constant->u.State[1]]->tex;
+ vec[0] = 1.0 / tex->width0;
+ vec[1] = 1.0 / tex->height0;
+ break;
+
+ /* Texture compare-fail value. */
+ /* XXX Since Gallium doesn't support GL_ARB_shadow_ambient,
+ * this is always (0,0,0,0). */
+ case RC_STATE_SHADOW_AMBIENT:
+ vec[3] = 0;
+ break;
+
+ case RC_STATE_R300_VIEWPORT_SCALE:
+ if (r300->rs_state->enable_vte) {
+ vec[0] = r300->viewport_state->xscale;
+ vec[1] = r300->viewport_state->yscale;
+ vec[2] = r300->viewport_state->zscale;
+ } else {
+ vec[0] = 1;
+ vec[1] = 1;
+ vec[2] = 1;
+ }
+ break;
+
+ case RC_STATE_R300_VIEWPORT_OFFSET:
+ if (r300->rs_state->enable_vte) {
+ vec[0] = r300->viewport_state->xoffset;
+ vec[1] = r300->viewport_state->yoffset;
+ vec[2] = r300->viewport_state->zoffset;
+ } else {
+ /* Zeros. */
+ }
+ break;
+
+ default:
+ debug_printf("r300: Implementation error: "
+ "Unknown RC_CONSTANT type %d\n", constant->u.State[0]);
+ }
+ break;
+
default:
- debug_printf("r300: Implementation error: Unhandled constant type %i\n",
- constant->Type);
- return zero;
+ debug_printf("r300: Implementation error: "
+ "Unhandled constant type %d\n", constant->Type);
}
+
+ /* This should either be (0, 0, 0, 1), which should be a relatively safe
+ * RGBA or STRQ value, or it could be one of the RC_CONSTANT_STATE
+ * state factors. */
+ return vec;
}
/* Convert a normal single-precision float into the 7.16 format
@@ -168,18 +246,15 @@ static uint32_t pack_float24(float f)
}
void r300_emit_fragment_program_code(struct r300_context* r300,
- struct rX00_fragment_program_code* generic_code,
- struct r300_constant_buffer* externals)
+ struct rX00_fragment_program_code* generic_code)
{
struct r300_fragment_program_code * code = &generic_code->code.r300;
- struct rc_constant_list * constants = &generic_code->constants;
int i;
CS_LOCALS(r300);
BEGIN_CS(15 +
code->alu.length * 4 +
- (code->tex.length ? (1 + code->tex.length) : 0) +
- (constants->Count ? (1 + constants->Count * 4) : 0));
+ (code->tex.length ? (1 + code->tex.length) : 0));
OUT_CS_REG(R300_US_CONFIG, code->config);
OUT_CS_REG(R300_US_PIXSIZE, code->pixsize);
@@ -211,33 +286,58 @@ void r300_emit_fragment_program_code(struct r300_context* r300,
OUT_CS(code->tex.inst[i]);
}
- if (constants->Count) {
- OUT_CS_ONE_REG(R300_PFS_PARAM_0_X, constants->Count * 4);
- for(i = 0; i < constants->Count; ++i) {
- const float * data = get_shader_constant(r300, &constants->Constants[i], externals);
- OUT_CS(pack_float24(data[0]));
- OUT_CS(pack_float24(data[1]));
- OUT_CS(pack_float24(data[2]));
- OUT_CS(pack_float24(data[3]));
- }
+ END_CS;
+}
+
+void r300_emit_fs_constant_buffer(struct r300_context* r300,
+ struct rc_constant_list* constants)
+{
+ int i;
+ CS_LOCALS(r300);
+
+ if (constants->Count == 0)
+ return;
+
+ BEGIN_CS(constants->Count * 4 + 1);
+ OUT_CS_REG_SEQ(R300_PFS_PARAM_0_X, constants->Count * 4);
+ for(i = 0; i < constants->Count; ++i) {
+ const float * data = get_shader_constant(r300,
+ &constants->Constants[i],
+ &r300->shader_constants[PIPE_SHADER_FRAGMENT]);
+ OUT_CS(pack_float24(data[0]));
+ OUT_CS(pack_float24(data[1]));
+ OUT_CS(pack_float24(data[2]));
+ OUT_CS(pack_float24(data[3]));
}
+ END_CS;
+}
+static void r300_emit_fragment_depth_config(struct r300_context* r300,
+ struct r300_fragment_shader* fs)
+{
+ CS_LOCALS(r300);
+
+ BEGIN_CS(4);
+ if (r300_fragment_shader_writes_depth(fs)) {
+ OUT_CS_REG(R300_FG_DEPTH_SRC, R300_FG_DEPTH_SRC_SHADER);
+ OUT_CS_REG(R300_US_W_FMT, R300_W_FMT_W24 | R300_W_SRC_US);
+ } else {
+ OUT_CS_REG(R300_FG_DEPTH_SRC, R300_FG_DEPTH_SRC_SCAN);
+ OUT_CS_REG(R300_US_W_FMT, R300_W_FMT_W0 | R300_W_SRC_US);
+ }
END_CS;
}
void r500_emit_fragment_program_code(struct r300_context* r300,
- struct rX00_fragment_program_code* generic_code,
- struct r300_constant_buffer* externals)
+ struct rX00_fragment_program_code* generic_code)
{
struct r500_fragment_program_code * code = &generic_code->code.r500;
- struct rc_constant_list * constants = &generic_code->constants;
int i;
CS_LOCALS(r300);
BEGIN_CS(13 +
- ((code->inst_end + 1) * 6) +
- (constants->Count ? (3 + (constants->Count * 4)) : 0));
- OUT_CS_REG(R500_US_CONFIG, 0);
+ ((code->inst_end + 1) * 6));
+ 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));
@@ -256,18 +356,30 @@ void r500_emit_fragment_program_code(struct r300_context* r300,
OUT_CS(code->inst[i].inst5);
}
- if (constants->Count) {
- OUT_CS_REG(R500_GA_US_VECTOR_INDEX, R500_GA_US_VECTOR_INDEX_TYPE_CONST);
- OUT_CS_ONE_REG(R500_GA_US_VECTOR_DATA, constants->Count * 4);
- for (i = 0; i < constants->Count; i++) {
- const float * data = get_shader_constant(r300, &constants->Constants[i], externals);
- OUT_CS_32F(data[0]);
- OUT_CS_32F(data[1]);
- OUT_CS_32F(data[2]);
- OUT_CS_32F(data[3]);
- }
- }
+ END_CS;
+}
+
+void r500_emit_fs_constant_buffer(struct r300_context* r300,
+ struct rc_constant_list* constants)
+{
+ int i;
+ CS_LOCALS(r300);
+
+ if (constants->Count == 0)
+ return;
+ BEGIN_CS(constants->Count * 4 + 3);
+ OUT_CS_REG(R500_GA_US_VECTOR_INDEX, R500_GA_US_VECTOR_INDEX_TYPE_CONST);
+ OUT_CS_ONE_REG(R500_GA_US_VECTOR_DATA, constants->Count * 4);
+ for (i = 0; i < constants->Count; i++) {
+ const float * data = get_shader_constant(r300,
+ &constants->Constants[i],
+ &r300->shader_constants[PIPE_SHADER_FRAGMENT]);
+ OUT_CS_32F(data[0]);
+ OUT_CS_32F(data[1]);
+ OUT_CS_32F(data[2]);
+ OUT_CS_32F(data[3]);
+ }
END_CS;
}
@@ -275,76 +387,98 @@ void r300_emit_fb_state(struct r300_context* r300,
struct pipe_framebuffer_state* fb)
{
struct r300_texture* tex;
- unsigned pixpitch;
+ struct pipe_surface* surf;
int i;
CS_LOCALS(r300);
- BEGIN_CS((10 * fb->nr_cbufs) + (fb->zsbuf ? 10 : 0) + 4);
+ /* Shouldn't fail unless there is a bug in the state tracker. */
+ assert(fb->nr_cbufs <= 4);
+
+ BEGIN_CS((10 * fb->nr_cbufs) + (2 * (4 - fb->nr_cbufs)) +
+ (fb->zsbuf ? 10 : 0) + 6);
+
+ /* 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. */
+ OUT_CS_REG(R300_RB3D_CCTL, R300_RB3D_CCTL_NUM_MULTIWRITES(fb->nr_cbufs));
+
+ /* Set up colorbuffers. */
for (i = 0; i < fb->nr_cbufs; i++) {
- tex = (struct r300_texture*)fb->cbufs[i]->texture;
+ surf = fb->cbufs[i];
+ tex = (struct r300_texture*)surf->texture;
assert(tex && tex->buffer && "cbuf is marked, but NULL!");
- pixpitch = r300_texture_get_stride(tex, 0) / tex->tex.block.size;
OUT_CS_REG_SEQ(R300_RB3D_COLOROFFSET0 + (4 * i), 1);
- OUT_CS_RELOC(tex->buffer, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0);
+ OUT_CS_RELOC(tex->buffer, surf->offset, 0, RADEON_GEM_DOMAIN_VRAM, 0);
OUT_CS_REG_SEQ(R300_RB3D_COLORPITCH0 + (4 * i), 1);
- OUT_CS_RELOC(tex->buffer, pixpitch |
+ OUT_CS_RELOC(tex->buffer, tex->pitch[surf->level] |
r300_translate_colorformat(tex->tex.format), 0,
RADEON_GEM_DOMAIN_VRAM, 0);
OUT_CS_REG(R300_US_OUT_FMT_0 + (4 * i),
- r300_translate_out_fmt(fb->cbufs[i]->format));
+ r300_translate_out_fmt(surf->format));
+ }
+
+ /* Disable unused colorbuffers. */
+ for (; i < 4; i++) {
+ OUT_CS_REG(R300_US_OUT_FMT_0 + (4 * i), R300_US_OUT_FMT_UNUSED);
}
+ /* Set up a zbuffer. */
if (fb->zsbuf) {
- tex = (struct r300_texture*)fb->zsbuf->texture;
+ surf = fb->zsbuf;
+ tex = (struct r300_texture*)surf->texture;
assert(tex && tex->buffer && "zsbuf is marked, but NULL!");
- pixpitch = r300_texture_get_stride(tex, 0) / tex->tex.block.size;
OUT_CS_REG_SEQ(R300_ZB_DEPTHOFFSET, 1);
- OUT_CS_RELOC(tex->buffer, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0);
+ OUT_CS_RELOC(tex->buffer, surf->offset, 0, RADEON_GEM_DOMAIN_VRAM, 0);
OUT_CS_REG(R300_ZB_FORMAT, r300_translate_zsformat(tex->tex.format));
OUT_CS_REG_SEQ(R300_ZB_DEPTHPITCH, 1);
- OUT_CS_RELOC(tex->buffer, pixpitch, 0, RADEON_GEM_DOMAIN_VRAM, 0);
+ OUT_CS_RELOC(tex->buffer, tex->pitch[surf->level], 0,
+ RADEON_GEM_DOMAIN_VRAM, 0);
}
- 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);
END_CS;
}
-void r300_emit_query_begin(struct r300_context* r300,
- struct r300_query* query)
+static void r300_emit_query_start(struct r300_context *r300)
{
+ struct r300_capabilities *caps = r300_screen(r300->context.screen)->caps;
+ struct r300_query *query = r300->query_current;
CS_LOCALS(r300);
- /* XXX This will almost certainly not return good results
- * for overlapping queries. */
- BEGIN_CS(2);
+ if (!query)
+ return;
+
+ BEGIN_CS(4);
+ if (caps->family == CHIP_FAMILY_RV530) {
+ OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL);
+ } else {
+ OUT_CS_REG(R300_SU_REG_DEST, R300_RASTER_PIPE_SELECT_ALL);
+ }
OUT_CS_REG(R300_ZB_ZPASS_DATA, 0);
END_CS;
+ query->begin_emitted = TRUE;
}
-void r300_emit_query_end(struct r300_context* r300,
- struct r300_query* query)
+
+static void r300_emit_query_finish(struct r300_context *r300,
+ struct r300_query *query)
{
struct r300_capabilities* caps = r300_screen(r300->context.screen)->caps;
CS_LOCALS(r300);
- if (!r300->winsys->add_buffer(r300->winsys, r300->oqbo,
- 0, RADEON_GEM_DOMAIN_GTT)) {
- debug_printf("r300: There wasn't room for the OQ buffer!?"
- " Oh noes!\n");
- }
-
assert(caps->num_frag_pipes);
+
BEGIN_CS(6 * caps->num_frag_pipes + 2);
/* I'm not so sure I like this switch, but it's hard to be elegant
* when there's so many special cases...
@@ -381,7 +515,7 @@ void r300_emit_query_end(struct r300_context* r300,
OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR, 1);
OUT_CS_RELOC(r300->oqbo, query->offset + (sizeof(uint32_t) * 0),
0, RADEON_GEM_DOMAIN_GTT, 0);
- break;
+ break;
default:
debug_printf("r300: Implementation error: Chipset reports %d"
" pixel pipes!\n", caps->num_frag_pipes);
@@ -391,14 +525,62 @@ void r300_emit_query_end(struct r300_context* r300,
/* And, finally, reset it to normal... */
OUT_CS_REG(R300_SU_REG_DEST, 0xF);
END_CS;
+}
+
+static void rv530_emit_query_single(struct r300_context *r300,
+ struct r300_query *query)
+{
+ 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_RELOC(r300->oqbo, query->offset, 0, RADEON_GEM_DOMAIN_GTT, 0);
+ 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)
+{
+ 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_RELOC(r300->oqbo, query->offset, 0, RADEON_GEM_DOMAIN_GTT, 0);
+ 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_RELOC(r300->oqbo, query->offset + sizeof(uint32_t), 0, RADEON_GEM_DOMAIN_GTT, 0);
+ OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL);
+ END_CS;
+}
+
+void r300_emit_query_end(struct r300_context* r300)
+{
+ struct r300_capabilities *caps = r300_screen(r300->context.screen)->caps;
+ struct r300_query *query = r300->query_current;
+
+ if (!query)
+ return;
+
+ if (query->begin_emitted == FALSE)
+ return;
+
+ if (caps->family == CHIP_FAMILY_RV530) {
+ if (caps->num_z_pipes == 2)
+ rv530_emit_query_double(r300, query);
+ else
+ rv530_emit_query_single(r300, query);
+ } else
+ r300_emit_query_finish(r300, query);
}
void r300_emit_rs_state(struct r300_context* r300, struct r300_rs_state* rs)
{
CS_LOCALS(r300);
- BEGIN_CS(20);
+ BEGIN_CS(22);
OUT_CS_REG(R300_VAP_CNTL_STATUS, rs->vap_control_status);
OUT_CS_REG(R300_GA_POINT_SIZE, rs->point_size);
OUT_CS_REG_SEQ(R300_GA_POINT_MINMAX, 2);
@@ -414,6 +596,7 @@ void r300_emit_rs_state(struct r300_context* r300, struct r300_rs_state* rs)
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_COLOR_CONTROL, rs->color_control);
+ OUT_CS_REG(R300_GA_POLY_MODE, rs->polygon_mode);
END_CS;
}
@@ -424,6 +607,8 @@ void r300_emit_rs_block_state(struct r300_context* r300,
struct r300_screen* r300screen = r300_screen(r300->context.screen);
CS_LOCALS(r300);
+ DBG(r300, DBG_DRAW, "r300: RS emit:\n");
+
BEGIN_CS(21);
if (r300screen->caps->is_r500) {
OUT_CS_REG_SEQ(R500_RS_IP_0, 8);
@@ -432,7 +617,7 @@ void r300_emit_rs_block_state(struct r300_context* r300,
}
for (i = 0; i < 8; i++) {
OUT_CS(rs->ip[i]);
- /* debug_printf("ip %d: 0x%08x\n", i, rs->ip[i]); */
+ DBG(r300, DBG_DRAW, " : ip %d: 0x%08x\n", i, rs->ip[i]);
}
OUT_CS_REG_SEQ(R300_RS_COUNT, 2);
@@ -446,40 +631,67 @@ void r300_emit_rs_block_state(struct r300_context* r300,
}
for (i = 0; i < 8; i++) {
OUT_CS(rs->inst[i]);
- /* debug_printf("inst %d: 0x%08x\n", i, rs->inst[i]); */
+ DBG(r300, DBG_DRAW, " : inst %d: 0x%08x\n", i, rs->inst[i]);
}
- /* debug_printf("count: 0x%08x inst_count: 0x%08x\n", rs->count,
- * rs->inst_count); */
+ DBG(r300, DBG_DRAW, " : count: 0x%08x inst_count: 0x%08x\n",
+ rs->count, rs->inst_count);
END_CS;
}
-void r300_emit_scissor_state(struct r300_context* r300,
- struct r300_scissor_state* scissor)
+static void r300_emit_scissor_regs(struct r300_context* r300,
+ struct r300_scissor_regs* scissor)
{
CS_LOCALS(r300);
BEGIN_CS(3);
OUT_CS_REG_SEQ(R300_SC_SCISSORS_TL, 2);
- OUT_CS(scissor->scissor_top_left);
- OUT_CS(scissor->scissor_bottom_right);
+ OUT_CS(scissor->top_left);
+ OUT_CS(scissor->bottom_right);
END_CS;
}
+void r300_emit_scissor_state(struct r300_context* r300,
+ struct r300_scissor_state* scissor)
+{
+ if (r300->rs_state->rs.scissor) {
+ r300_emit_scissor_regs(r300, &scissor->scissor);
+ } else {
+ r300_emit_scissor_regs(r300, &scissor->framebuffer);
+ }
+}
+
void r300_emit_texture(struct r300_context* r300,
struct r300_sampler_state* sampler,
struct r300_texture* tex,
unsigned offset)
{
+ uint32_t filter0 = sampler->filter0;
+ uint32_t format0 = tex->state.format0;
+ unsigned min_level, max_level;
CS_LOCALS(r300);
+ /* to emulate 1D textures through 2D ones correctly */
+ if (tex->tex.target == PIPE_TEXTURE_1D) {
+ filter0 &= ~R300_TX_WRAP_T_MASK;
+ filter0 |= R300_TX_WRAP_T(R300_TX_CLAMP_TO_EDGE);
+ }
+
+ /* determine min/max levels */
+ /* the MAX_MIP level is the largest (finest) one */
+ max_level = MIN2(sampler->max_lod, tex->tex.last_level);
+ min_level = MIN2(sampler->min_lod, max_level);
+ format0 |= R300_TX_NUM_LEVELS(max_level);
+ filter0 |= R300_TX_MAX_MIP_LEVEL(min_level);
+
BEGIN_CS(16);
- OUT_CS_REG(R300_TX_FILTER0_0 + (offset * 4), sampler->filter0);
+ OUT_CS_REG(R300_TX_FILTER0_0 + (offset * 4), filter0 |
+ (offset << 28));
OUT_CS_REG(R300_TX_FILTER1_0 + (offset * 4), sampler->filter1);
OUT_CS_REG(R300_TX_BORDER_COLOR_0 + (offset * 4), sampler->border_color);
- OUT_CS_REG(R300_TX_FORMAT0_0 + (offset * 4), tex->state.format0);
+ OUT_CS_REG(R300_TX_FORMAT0_0 + (offset * 4), format0);
OUT_CS_REG(R300_TX_FORMAT1_0 + (offset * 4), tex->state.format1);
OUT_CS_REG(R300_TX_FORMAT2_0 + (offset * 4), tex->state.format2);
OUT_CS_REG_SEQ(R300_TX_OFFSET_0 + (offset * 4), 1);
@@ -488,13 +700,76 @@ void r300_emit_texture(struct r300_context* r300,
END_CS;
}
-void r300_emit_vertex_buffer(struct r300_context* r300)
+static boolean r300_validate_aos(struct r300_context *r300)
+{
+ struct pipe_vertex_buffer *vbuf = r300->vertex_buffer;
+ struct pipe_vertex_element *velem = r300->vertex_element;
+ int i;
+
+ /* Check if formats and strides are aligned to the size of DWORD. */
+ for (i = 0; i < r300->vertex_element_count; i++) {
+ if (vbuf[velem[i].vertex_buffer_index].stride % 4 != 0 ||
+ util_format_get_blocksize(velem[i].src_format) % 4 != 0) {
+ return FALSE;
+ }
+ }
+ return TRUE;
+}
+
+void r300_emit_aos(struct r300_context* r300, unsigned offset)
+{
+ struct pipe_vertex_buffer *vb1, *vb2, *vbuf = r300->vertex_buffer;
+ struct pipe_vertex_element *velem = r300->vertex_element;
+ int i;
+ unsigned size1, size2, aos_count = r300->vertex_element_count;
+ unsigned packet_size = (aos_count * 3 + 1) / 2;
+ CS_LOCALS(r300);
+
+ /* XXX Move this checking to a more approriate place. */
+ if (!r300_validate_aos(r300)) {
+ /* XXX We should fallback using Draw. */
+ assert(0);
+ }
+
+ BEGIN_CS(2 + packet_size + aos_count * 2);
+ OUT_CS_PKT3(R300_PACKET3_3D_LOAD_VBPNTR, packet_size);
+ OUT_CS(aos_count);
+
+ 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);
+
+ OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(vb1->stride) |
+ R300_VBPNTR_SIZE1(size2) | R300_VBPNTR_STRIDE1(vb2->stride));
+ OUT_CS(vb1->buffer_offset + velem[i].src_offset + offset * vb1->stride);
+ OUT_CS(vb2->buffer_offset + velem[i+1].src_offset + offset * vb2->stride);
+ }
+
+ if (aos_count & 1) {
+ vb1 = &vbuf[velem[i].vertex_buffer_index];
+ size1 = util_format_get_blocksize(velem[i].src_format);
+
+ 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_RELOC_NO_OFFSET(vbuf[velem[i].vertex_buffer_index].buffer,
+ RADEON_GEM_DOMAIN_GTT, 0, 0);
+ }
+ END_CS;
+}
+
+#if 0
+void r300_emit_draw_packet(struct r300_context* r300)
{
CS_LOCALS(r300);
DBG(r300, DBG_DRAW, "r300: Preparing vertex buffer %p for render, "
"vertex size %d\n", r300->vbo,
- r300->vertex_info.vinfo.size);
+ r300->vertex_info->vinfo.size);
/* Set the pointer to our vertex buffer. The emitted values are this:
* PACKET3 [3D_LOAD_VBPNTR]
* COUNT [1]
@@ -505,54 +780,66 @@ 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(r300->vertex_info.vinfo.size |
- (r300->vertex_info.vinfo.size << 8));
+ OUT_CS(r300->vertex_info->vinfo.size |
+ (r300->vertex_info->vinfo.size << 8));
OUT_CS(r300->vbo_offset);
OUT_CS_RELOC(r300->vbo, 0, RADEON_GEM_DOMAIN_GTT, 0, 0);
END_CS;
}
+#endif
void r300_emit_vertex_format_state(struct r300_context* r300)
{
int i;
CS_LOCALS(r300);
+ DBG(r300, DBG_DRAW, "r300: VAP/PSC emit:\n");
+
BEGIN_CS(26);
- OUT_CS_REG(R300_VAP_VTX_SIZE, r300->vertex_info.vinfo.size);
+ OUT_CS_REG(R300_VAP_VTX_SIZE, r300->vertex_info->vinfo.size);
OUT_CS_REG_SEQ(R300_VAP_VTX_STATE_CNTL, 2);
- OUT_CS(r300->vertex_info.vinfo.hwfmt[0]);
- OUT_CS(r300->vertex_info.vinfo.hwfmt[1]);
+ OUT_CS(r300->vertex_info->vinfo.hwfmt[0]);
+ OUT_CS(r300->vertex_info->vinfo.hwfmt[1]);
OUT_CS_REG_SEQ(R300_VAP_OUTPUT_VTX_FMT_0, 2);
- OUT_CS(r300->vertex_info.vinfo.hwfmt[2]);
- OUT_CS(r300->vertex_info.vinfo.hwfmt[3]);
- /* for (i = 0; i < 4; i++) {
- * debug_printf("hwfmt%d: 0x%08x\n", i,
- * r300->vertex_info.vinfo.hwfmt[i]);
- * } */
+ OUT_CS(r300->vertex_info->vinfo.hwfmt[2]);
+ OUT_CS(r300->vertex_info->vinfo.hwfmt[3]);
+ for (i = 0; i < 4; i++) {
+ DBG(r300, DBG_DRAW, " : hwfmt%d: 0x%08x\n", i,
+ r300->vertex_info->vinfo.hwfmt[i]);
+ }
OUT_CS_REG_SEQ(R300_VAP_PROG_STREAM_CNTL_0, 8);
for (i = 0; i < 8; i++) {
- OUT_CS(r300->vertex_info.vap_prog_stream_cntl[i]);
- /* debug_printf("prog_stream_cntl%d: 0x%08x\n", i,
- * r300->vertex_info.vap_prog_stream_cntl[i]); */
+ OUT_CS(r300->vertex_info->vap_prog_stream_cntl[i]);
+ DBG(r300, DBG_DRAW, " : prog_stream_cntl%d: 0x%08x\n", i,
+ r300->vertex_info->vap_prog_stream_cntl[i]);
}
OUT_CS_REG_SEQ(R300_VAP_PROG_STREAM_CNTL_EXT_0, 8);
for (i = 0; i < 8; i++) {
- OUT_CS(r300->vertex_info.vap_prog_stream_cntl_ext[i]);
- /* debug_printf("prog_stream_cntl_ext%d: 0x%08x\n", i,
- * r300->vertex_info.vap_prog_stream_cntl_ext[i]); */
+ OUT_CS(r300->vertex_info->vap_prog_stream_cntl_ext[i]);
+ DBG(r300, DBG_DRAW, " : prog_stream_cntl_ext%d: 0x%08x\n", i,
+ r300->vertex_info->vap_prog_stream_cntl_ext[i]);
}
END_CS;
}
+
void r300_emit_vertex_program_code(struct r300_context* r300,
- struct r300_vertex_program_code* code,
- struct r300_constant_buffer* constants)
+ struct r300_vertex_program_code* code)
{
int i;
struct r300_screen* r300screen = r300_screen(r300->context.screen);
unsigned instruction_count = code->length / 4;
+
+ int vtx_mem_size = r300screen->caps->is_r500 ? 128 : 72;
+ int input_count = MAX2(util_bitcount(code->InputsRead), 1);
+ int output_count = MAX2(util_bitcount(code->OutputsWritten), 1);
+ int temp_count = MAX2(code->num_temporaries, 1);
+ int pvs_num_slots = MIN3(vtx_mem_size / input_count,
+ vtx_mem_size / output_count, 10);
+ int pvs_num_controllers = MIN2(vtx_mem_size / temp_count, 6);
+
CS_LOCALS(r300);
if (!r300screen->caps->has_tcl) {
@@ -561,17 +848,11 @@ void r300_emit_vertex_program_code(struct r300_context* r300,
return;
}
- if (code->constants.Count) {
- BEGIN_CS(14 + code->length + (code->constants.Count * 4));
- } else {
- BEGIN_CS(11 + code->length);
- }
-
+ BEGIN_CS(9 + code->length);
/* R300_VAP_PVS_CODE_CNTL_0
* R300_VAP_PVS_CONST_CNTL
* R300_VAP_PVS_CODE_CNTL_1
- * See the r5xx docs for instructions on how to use these.
- * XXX these could be optimized to select better values... */
+ * See the r5xx docs for instructions on how to use these. */
OUT_CS_REG_SEQ(R300_VAP_PVS_CODE_CNTL_0, 3);
OUT_CS(R300_PVS_FIRST_INST(0) |
R300_PVS_XYZW_VALID_INST(instruction_count - 1) |
@@ -584,32 +865,51 @@ void r300_emit_vertex_program_code(struct r300_context* r300,
for (i = 0; i < code->length; i++)
OUT_CS(code->body.d[i]);
- if (code->constants.Count) {
- OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG,
- (r300screen->caps->is_r500 ?
- R500_PVS_CONST_START : R300_PVS_CONST_START));
- OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, code->constants.Count * 4);
- for (i = 0; i < code->constants.Count; i++) {
- const float * data = get_shader_constant(r300, &code->constants.Constants[i], constants);
- OUT_CS_32F(data[0]);
- OUT_CS_32F(data[1]);
- OUT_CS_32F(data[2]);
- OUT_CS_32F(data[3]);
- }
- }
-
- OUT_CS_REG(R300_VAP_CNTL, R300_PVS_NUM_SLOTS(10) |
- R300_PVS_NUM_CNTLRS(5) |
+ OUT_CS_REG(R300_VAP_CNTL, R300_PVS_NUM_SLOTS(pvs_num_slots) |
+ R300_PVS_NUM_CNTLRS(pvs_num_controllers) |
R300_PVS_NUM_FPUS(r300screen->caps->num_vert_fpus) |
- R300_PVS_VF_MAX_VTX_NUM(12));
- OUT_CS_REG(R300_VAP_PVS_STATE_FLUSH_REG, 0x0);
+ R300_PVS_VF_MAX_VTX_NUM(12) |
+ (r300screen->caps->is_r500 ? R500_TCL_STATE_OPTIMIZATION : 0));
END_CS;
}
void r300_emit_vertex_shader(struct r300_context* r300,
struct r300_vertex_shader* vs)
{
- r300_emit_vertex_program_code(r300, &vs->code, &r300->shader_constants[PIPE_SHADER_VERTEX]);
+ r300_emit_vertex_program_code(r300, &vs->code);
+}
+
+void r300_emit_vs_constant_buffer(struct r300_context* r300,
+ struct rc_constant_list* constants)
+{
+ int i;
+ struct r300_screen* r300screen = r300_screen(r300->context.screen);
+ CS_LOCALS(r300);
+
+ if (!r300screen->caps->has_tcl) {
+ debug_printf("r300: Implementation error: emit_vertex_shader called,"
+ " but has_tcl is FALSE!\n");
+ return;
+ }
+
+ if (constants->Count == 0)
+ return;
+
+ BEGIN_CS(constants->Count * 4 + 3);
+ OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG,
+ (r300screen->caps->is_r500 ?
+ R500_PVS_CONST_START : R300_PVS_CONST_START));
+ OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, constants->Count * 4);
+ for (i = 0; i < constants->Count; i++) {
+ const float * data = get_shader_constant(r300,
+ &constants->Constants[i],
+ &r300->shader_constants[PIPE_SHADER_VERTEX]);
+ OUT_CS_32F(data[0]);
+ OUT_CS_32F(data[1]);
+ OUT_CS_32F(data[2]);
+ OUT_CS_32F(data[3]);
+ }
+ END_CS;
}
void r300_emit_viewport_state(struct r300_context* r300,
@@ -634,13 +934,42 @@ void r300_emit_viewport_state(struct r300_context* r300,
END_CS;
}
+void r300_emit_texture_count(struct r300_context* r300)
+{
+ uint32_t tx_enable = 0;
+ int i;
+ CS_LOCALS(r300);
+
+ /* Notice that texture_count and sampler_count are just sizes
+ * of the respective arrays. We still have to check for the individual
+ * elements. */
+ for (i = 0; i < MIN2(r300->sampler_count, r300->texture_count); i++) {
+ if (r300->textures[i]) {
+ tx_enable |= 1 << i;
+ }
+ }
+
+ BEGIN_CS(2);
+ OUT_CS_REG(R300_TX_ENABLE, tx_enable);
+ END_CS;
+
+}
+
void r300_flush_textures(struct r300_context* r300)
{
CS_LOCALS(r300);
- BEGIN_CS(4);
+ BEGIN_CS(2);
OUT_CS_REG(R300_TX_INVALTAGS, 0);
- OUT_CS_REG(R300_TX_ENABLE, (1 << r300->texture_count) - 1);
+ END_CS;
+}
+
+static void r300_flush_pvs(struct r300_context* r300)
+{
+ CS_LOCALS(r300);
+
+ BEGIN_CS(2);
+ OUT_CS_REG(R300_VAP_PVS_STATE_FLUSH_REG, 0x0);
END_CS;
}
@@ -656,9 +985,17 @@ void r300_emit_dirty_state(struct r300_context* r300)
return;
}
- r300_update_derived_state(r300);
+ /* Check size of CS. */
+ /* Make sure we have at least 8*1024 spare dwords. */
+ /* XXX It would be nice to know the number of dwords we really need to
+ * XXX emit. */
+ if (!r300->winsys->check_cs(r300->winsys, 8*1024)) {
+ r300->context.flush(&r300->context, 0, NULL);
+ }
+
+ /* Clean out BOs. */
+ r300->winsys->reset_bos(r300->winsys);
- /* XXX check size */
validate:
/* Color buffers... */
for (i = 0; i < r300->framebuffer_state.nr_cbufs; i++) {
@@ -683,7 +1020,8 @@ validate:
/* ...textures... */
for (i = 0; i < r300->texture_count; i++) {
tex = r300->textures[i];
- assert(tex && tex->buffer && "texture is marked, but NULL!");
+ if (!tex)
+ continue;
if (!r300->winsys->add_buffer(r300->winsys, tex->buffer,
RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM, 0)) {
r300->context.flush(&r300->context, 0, NULL);
@@ -704,7 +1042,7 @@ validate:
goto validate;
}
} else {
- debug_printf("No VBO while emitting dirty state!\n");
+ /* debug_printf("No VBO while emitting dirty state!\n"); */
}
if (!r300->winsys->validate(r300->winsys)) {
r300->context.flush(&r300->context, 0, NULL);
@@ -717,6 +1055,11 @@ validate:
goto validate;
}
+ if (r300->dirty_state & R300_NEW_QUERY) {
+ r300_emit_query_start(r300);
+ r300->dirty_state &= ~R300_NEW_QUERY;
+ }
+
if (r300->dirty_state & R300_NEW_BLEND) {
r300_emit_blend_state(r300, r300->blend_state);
r300->dirty_state &= ~R300_NEW_BLEND;
@@ -738,14 +1081,26 @@ validate:
}
if (r300->dirty_state & R300_NEW_FRAGMENT_SHADER) {
+ r300_emit_fragment_depth_config(r300, r300->fs);
if (r300screen->caps->is_r500) {
- r500_emit_fragment_program_code(r300, &r300->fs->code, &r300->shader_constants[PIPE_SHADER_FRAGMENT]);
+ r500_emit_fragment_program_code(r300, &r300->fs->shader->code);
} else {
- r300_emit_fragment_program_code(r300, &r300->fs->code, &r300->shader_constants[PIPE_SHADER_FRAGMENT]);
+ r300_emit_fragment_program_code(r300, &r300->fs->shader->code);
}
r300->dirty_state &= ~R300_NEW_FRAGMENT_SHADER;
}
+ if (r300->dirty_state & R300_NEW_FRAGMENT_SHADER_CONSTANTS) {
+ if (r300screen->caps->is_r500) {
+ r500_emit_fs_constant_buffer(r300,
+ &r300->fs->shader->code.constants);
+ } else {
+ r300_emit_fs_constant_buffer(r300,
+ &r300->fs->shader->code.constants);
+ }
+ r300->dirty_state &= ~R300_NEW_FRAGMENT_SHADER_CONSTANTS;
+ }
+
if (r300->dirty_state & R300_NEW_FRAMEBUFFERS) {
r300_emit_fb_state(r300, &r300->framebuffer_state);
r300->dirty_state &= ~R300_NEW_FRAMEBUFFERS;
@@ -769,13 +1124,16 @@ validate:
/* Samplers and textures are tracked separately but emitted together. */
if (r300->dirty_state &
(R300_ANY_NEW_SAMPLERS | R300_ANY_NEW_TEXTURES)) {
+ r300_emit_texture_count(r300);
+
for (i = 0; i < MIN2(r300->sampler_count, r300->texture_count); i++) {
- if (r300->dirty_state &
- ((R300_NEW_SAMPLER << i) | (R300_NEW_TEXTURE << i))) {
- r300_emit_texture(r300,
- r300->sampler_states[i],
- r300->textures[i],
- i);
+ if (r300->dirty_state &
+ ((R300_NEW_SAMPLER << i) | (R300_NEW_TEXTURE << i))) {
+ if (r300->textures[i])
+ r300_emit_texture(r300,
+ r300->sampler_states[i],
+ r300->textures[i],
+ i);
r300->dirty_state &=
~((R300_NEW_SAMPLER << i) | (R300_NEW_TEXTURE << i));
dirty_tex++;
@@ -798,17 +1156,26 @@ validate:
r300->dirty_state &= ~R300_NEW_VERTEX_FORMAT;
}
+ if (r300->dirty_state & (R300_NEW_VERTEX_SHADER | R300_NEW_VERTEX_SHADER_CONSTANTS)) {
+ r300_flush_pvs(r300);
+ }
+
if (r300->dirty_state & R300_NEW_VERTEX_SHADER) {
r300_emit_vertex_shader(r300, r300->vs);
r300->dirty_state &= ~R300_NEW_VERTEX_SHADER;
}
+ if (r300->dirty_state & R300_NEW_VERTEX_SHADER_CONSTANTS) {
+ r300_emit_vs_constant_buffer(r300, &r300->vs->code.constants);
+ r300->dirty_state &= ~R300_NEW_VERTEX_SHADER_CONSTANTS;
+ }
+
/* XXX
assert(r300->dirty_state == 0);
*/
/* Finally, emit the VBO. */
- r300_emit_vertex_buffer(r300);
+ /* 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 c4002b8e5d0..3797d3d332a 100644
--- a/src/gallium/drivers/r300/r300_emit.h
+++ b/src/gallium/drivers/r300/r300_emit.h
@@ -23,16 +23,14 @@
#ifndef R300_EMIT_H
#define R300_EMIT_H
-#include "util/u_math.h"
-
#include "r300_context.h"
-#include "r300_cs.h"
-#include "r300_screen.h"
-#include "r300_state_inlines.h"
+#include "radeon_code.h"
struct rX00_fragment_program_code;
struct r300_vertex_program_code;
+void r300_emit_aos(struct r300_context* r300, unsigned offset);
+
void r300_emit_blend_state(struct r300_context* r300,
struct r300_blend_state* blend);
@@ -46,20 +44,24 @@ void r300_emit_dsa_state(struct r300_context* r300,
struct r300_dsa_state* dsa);
void r300_emit_fragment_program_code(struct r300_context* r300,
- struct rX00_fragment_program_code* generic_code,
- struct r300_constant_buffer* externals);
+ struct rX00_fragment_program_code* generic_code);
+
+void r300_emit_fs_constant_buffer(struct r300_context* r300,
+ struct rc_constant_list* constants);
void r500_emit_fragment_program_code(struct r300_context* r300,
- struct rX00_fragment_program_code* generic_code,
- struct r300_constant_buffer* externals);
+ struct rX00_fragment_program_code* generic_code);
+
+void r500_emit_fs_constant_buffer(struct r300_context* r300,
+ struct rc_constant_list* constants);
void r300_emit_fb_state(struct r300_context* r300,
struct pipe_framebuffer_state* fb);
void r300_emit_query_begin(struct r300_context* r300,
struct r300_query* query);
-void r300_emit_query_end(struct r300_context* r300,
- struct r300_query* query);
+
+void r300_emit_query_end(struct r300_context* r300);
void r300_emit_rs_state(struct r300_context* r300, struct r300_rs_state* rs);
@@ -79,8 +81,10 @@ void r300_emit_vertex_buffer(struct r300_context* r300);
void r300_emit_vertex_format_state(struct r300_context* r300);
void r300_emit_vertex_program_code(struct r300_context* r300,
- struct r300_vertex_program_code* code,
- struct r300_constant_buffer* constants);
+ struct r300_vertex_program_code* code);
+
+void r300_emit_vs_constant_buffer(struct r300_context* r300,
+ struct rc_constant_list* constants);
void r300_emit_vertex_shader(struct r300_context* r300,
struct r300_vertex_shader* vs);
@@ -88,6 +92,8 @@ void r300_emit_vertex_shader(struct r300_context* r300,
void r300_emit_viewport_state(struct r300_context* r300,
struct r300_viewport_state* viewport);
+void r300_emit_texture_count(struct r300_context* r300);
+
void r300_flush_textures(struct r300_context* r300);
/* Emit all dirty state. */
diff --git a/src/gallium/drivers/r300/r300_flush.c b/src/gallium/drivers/r300/r300_flush.c
index 0dff1c6f4fb..14a08241fc4 100644
--- a/src/gallium/drivers/r300/r300_flush.c
+++ b/src/gallium/drivers/r300/r300_flush.c
@@ -20,29 +20,48 @@
* OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
* USE OR OTHER DEALINGS IN THE SOFTWARE. */
+#include "draw/draw_context.h"
+#include "draw/draw_private.h"
+
+#include "util/u_simple_list.h"
+
+#include "r300_context.h"
+#include "r300_cs.h"
+#include "r300_emit.h"
#include "r300_flush.h"
+#include "r300_state_invariant.h"
static void r300_flush(struct pipe_context* pipe,
unsigned flags,
struct pipe_fence_handle** fence)
{
- struct r300_context* r300 = r300_context(pipe);
- CS_LOCALS(r300);
+ struct r300_context *r300 = r300_context(pipe);
+ struct r300_query *query;
+ CS_LOCALS(r300);
/* 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. */
- if (!r300->draw->flushing) {
+ * 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);
}
+ r300_emit_query_end(r300);
+
if (r300->dirty_hw) {
FLUSH_CS;
r300_emit_invariant_state(r300);
r300->dirty_state = R300_NEW_KITCHEN_SINK;
r300->dirty_hw = 0;
}
+ /* reset flushed query */
+ foreach(query, &r300->query_list) {
+ query->flushed = TRUE;
+ }
}
+
void r300_init_flush_functions(struct r300_context* r300)
{
r300->context.flush = r300_flush;
diff --git a/src/gallium/drivers/r300/r300_flush.h b/src/gallium/drivers/r300/r300_flush.h
index 9a83d89daab..0e9e6106bb7 100644
--- a/src/gallium/drivers/r300/r300_flush.h
+++ b/src/gallium/drivers/r300/r300_flush.h
@@ -23,13 +23,6 @@
#ifndef R300_FLUSH_H
#define R300_FLUSH_H
-#include "draw/draw_private.h"
-
-#include "pipe/p_context.h"
-
-#include "r300_context.h"
-#include "r300_cs.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 546ad545a53..60ea9c171d5 100644
--- a/src/gallium/drivers/r300/r300_fs.c
+++ b/src/gallium/drivers/r300/r300_fs.c
@@ -1,6 +1,7 @@
/*
* Copyright 2008 Corbin Simpson <[email protected]>
* Joakim Sindholt <[email protected]>
+ * 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"),
@@ -21,12 +22,58 @@
* OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
* USE OR OTHER DEALINGS IN THE SOFTWARE. */
-#include "r300_fs.h"
+#include "util/u_math.h"
+#include "util/u_memory.h"
+
+#include "tgsi/tgsi_dump.h"
+#include "r300_context.h"
+#include "r300_screen.h"
+#include "r300_fs.h"
#include "r300_tgsi_to_rc.h"
+#include "radeon_code.h"
#include "radeon_compiler.h"
+/* Convert info about FS input semantics to r300_shader_semantics. */
+void r300_shader_read_fs_inputs(struct tgsi_shader_info* info,
+ struct r300_shader_semantics* fs_inputs)
+{
+ int i;
+ unsigned index;
+
+ r300_shader_semantics_reset(fs_inputs);
+
+ for (i = 0; i < info->num_inputs; i++) {
+ index = info->input_semantic_index[i];
+
+ switch (info->input_semantic_name[i]) {
+ case TGSI_SEMANTIC_COLOR:
+ assert(index <= ATTR_COLOR_COUNT);
+ fs_inputs->color[index] = i;
+ break;
+
+ case TGSI_SEMANTIC_GENERIC:
+ assert(index <= ATTR_GENERIC_COUNT);
+ fs_inputs->generic[index] = i;
+ break;
+
+ case TGSI_SEMANTIC_FOG:
+ assert(index == 0);
+ fs_inputs->fog = i;
+ break;
+
+ case TGSI_SEMANTIC_POSITION:
+ assert(index == 0);
+ fs_inputs->wpos = i;
+ break;
+
+ default:
+ assert(0);
+ }
+ }
+}
+
static void find_output_registers(struct r300_fragment_program_compiler * compiler,
struct r300_fragment_shader * fs)
{
@@ -54,61 +101,69 @@ static void allocate_hardware_inputs(
void (*allocate)(void * data, unsigned input, unsigned hwreg),
void * mydata)
{
- struct tgsi_shader_info* info = &((struct r300_fragment_shader*)c->UserData)->info;
- int total_colors = 0;
- int colors = 0;
- int total_generic = 0;
- int generic = 0;
- int i;
-
- for (i = 0; i < info->num_inputs; i++) {
- switch (info->input_semantic_name[i]) {
- case TGSI_SEMANTIC_COLOR:
- total_colors++;
- break;
- case TGSI_SEMANTIC_FOG:
- case TGSI_SEMANTIC_GENERIC:
- total_generic++;
- break;
+ struct r300_shader_semantics* inputs =
+ (struct r300_shader_semantics*)c->UserData;
+ int i, reg = 0;
+
+ /* Allocate input registers. */
+ for (i = 0; i < ATTR_COLOR_COUNT; i++) {
+ if (inputs->color[i] != ATTR_UNUSED) {
+ allocate(mydata, inputs->color[i], reg++);
+ }
+ }
+ for (i = 0; i < ATTR_GENERIC_COUNT; i++) {
+ if (inputs->generic[i] != ATTR_UNUSED) {
+ allocate(mydata, inputs->generic[i], reg++);
}
}
+ if (inputs->fog != ATTR_UNUSED) {
+ allocate(mydata, inputs->fog, reg++);
+ }
+ if (inputs->wpos != ATTR_UNUSED) {
+ allocate(mydata, inputs->wpos, reg++);
+ }
+}
- for(i = 0; i < info->num_inputs; i++) {
- switch (info->input_semantic_name[i]) {
- case TGSI_SEMANTIC_COLOR:
- allocate(mydata, i, colors);
- colors++;
- break;
- case TGSI_SEMANTIC_FOG:
- case TGSI_SEMANTIC_GENERIC:
- allocate(mydata, i, total_colors + generic);
- generic++;
- break;
+static void get_compare_state(
+ struct r300_context* r300,
+ struct r300_fragment_program_external_state* state,
+ unsigned shadow_samplers)
+{
+ memset(state, 0, sizeof(*state));
+
+ for (int i = 0; i < r300->sampler_count; i++) {
+ struct r300_sampler_state* s = r300->sampler_states[i];
+
+ if (s && 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;
+
+ /* Fortunately, no need to translate this. */
+ state->unit[i].texture_compare_func = s->state.compare_func;
}
}
}
-void r300_translate_fragment_shader(struct r300_context* r300,
- struct r300_fragment_shader* fs)
+static void r300_translate_fragment_shader(
+ struct r300_context* r300,
+ struct r300_fragment_shader_code* shader)
{
+ struct r300_fragment_shader* fs = r300->fs;
struct r300_fragment_program_compiler compiler;
struct tgsi_to_rc ttr;
+ int wpos = fs->inputs.wpos;
+ /* Setup the compiler. */
memset(&compiler, 0, sizeof(compiler));
rc_init(&compiler.Base);
compiler.Base.Debug = DBG_ON(r300, DBG_FP);
- compiler.code = &fs->code;
+ compiler.code = &shader->code;
+ compiler.state = shader->compare_state;
compiler.is_r500 = r300_screen(r300->context.screen)->caps->is_r500;
compiler.AllocateHwInputs = &allocate_hardware_inputs;
- compiler.UserData = fs;
-
- /* TODO: Program compilation depends on texture compare modes,
- * which are sampler state. Therefore, programs need to be recompiled
- * depending on this state as in the classic Mesa driver.
- *
- * This is not yet handled correctly.
- */
+ compiler.UserData = &fs->inputs;
find_output_registers(&compiler, fs);
@@ -123,20 +178,76 @@ void r300_translate_fragment_shader(struct r300_context* r300,
r300_tgsi_to_rc(&ttr, fs->state.tokens);
+ fs->shadow_samplers = compiler.Base.Program.ShadowSamplers;
+
+ /**
+ * Transform the program to support WPOS.
+ *
+ * Introduce a small fragment at the start of the program that will be
+ * the only code that directly reads the WPOS input.
+ * All other code pieces that reference that input will be rewritten
+ * to read from a newly allocated temporary. */
+ if (wpos != ATTR_UNUSED) {
+ /* Moving the input to some other reg is not really necessary. */
+ rc_transform_fragment_wpos(&compiler.Base, wpos, wpos, TRUE);
+ }
+
/* Invoke the compiler */
r3xx_compile_fragment_program(&compiler);
if (compiler.Base.Error) {
- /* Todo: Fallback to software rendering gracefully? */
- fprintf(stderr, "r300 FP: Compiler error: %s\n", compiler.Base.ErrorMsg);
-
- if (compiler.is_r500) {
- memcpy(compiler.code, &r5xx_passthrough_fragment_shader, sizeof(r5xx_passthrough_fragment_shader));
- } else {
- memcpy(compiler.code, &r3xx_passthrough_fragment_shader, sizeof(r3xx_passthrough_fragment_shader));
- }
+ /* XXX failover maybe? */
+ DBG(r300, DBG_FP, "r300: Error compiling fragment program: %s\n",
+ compiler.Base.ErrorMsg);
+ assert(0);
}
/* And, finally... */
rc_destroy(&compiler.Base);
- fs->translated = TRUE;
+}
+
+boolean r300_pick_fragment_shader(struct r300_context* r300)
+{
+ struct r300_fragment_shader* fs = r300->fs;
+ struct r300_fragment_program_external_state state;
+ struct r300_fragment_shader_code* ptr;
+
+ if (!fs->first) {
+ /* Build the fragment shader for the first time. */
+ fs->first = fs->shader = CALLOC_STRUCT(r300_fragment_shader_code);
+
+ /* BTW shadow samplers will be known after the first translation,
+ * therefore we set ~0, which means it should look at all sampler
+ * states. This choice doesn't have any impact on the correctness. */
+ get_compare_state(r300, &fs->shader->compare_state, ~0);
+ r300_translate_fragment_shader(r300, fs->shader);
+ return TRUE;
+
+ } else if (fs->shadow_samplers) {
+ get_compare_state(r300, &state, fs->shadow_samplers);
+
+ /* Check if the currently-bound shader has been compiled
+ * with the texture-compare state we need. */
+ if (memcmp(&fs->shader->compare_state, &state, sizeof(state)) != 0) {
+ /* Search for the right shader. */
+ ptr = fs->first;
+ while (ptr) {
+ if (memcmp(&ptr->compare_state, &state, sizeof(state)) == 0) {
+ fs->shader = ptr;
+ return TRUE;
+ }
+ ptr = ptr->next;
+ }
+
+ /* Not found, gotta compile a new one. */
+ ptr = CALLOC_STRUCT(r300_fragment_shader_code);
+ ptr->next = fs->first;
+ fs->first = fs->shader = ptr;
+
+ ptr->compare_state = state;
+ r300_translate_fragment_shader(r300, ptr);
+ return TRUE;
+ }
+ }
+
+ return FALSE;
}
diff --git a/src/gallium/drivers/r300/r300_fs.h b/src/gallium/drivers/r300/r300_fs.h
index 967e9f697e9..40ce874353c 100644
--- a/src/gallium/drivers/r300/r300_fs.h
+++ b/src/gallium/drivers/r300/r300_fs.h
@@ -1,6 +1,7 @@
/*
* Copyright 2008 Corbin Simpson <[email protected]>
* Joakim Sindholt <[email protected]>
+ * 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"),
@@ -24,28 +25,47 @@
#ifndef R300_FS_H
#define R300_FS_H
-#include "tgsi/tgsi_dump.h"
+#include "pipe/p_state.h"
+#include "tgsi/tgsi_scan.h"
+#include "radeon_code.h"
+#include "r300_shader_semantics.h"
-#include "r300_context.h"
-#include "r3xx_fs.h"
-#include "r5xx_fs.h"
+struct r300_fragment_shader_code {
+ struct r300_fragment_program_external_state compare_state;
+ struct rX00_fragment_program_code code;
-#include "radeon_code.h"
+ struct r300_fragment_shader_code* next;
+};
struct r300_fragment_shader {
/* Parent class */
struct pipe_shader_state state;
+
struct tgsi_shader_info info;
+ struct r300_shader_semantics inputs;
- /* Has this shader been translated yet? */
- boolean translated;
+ /* Bits 0-15: TRUE if it's a shadow sampler, FALSE otherwise. */
+ unsigned shadow_samplers;
- /* Compiled code */
- struct rX00_fragment_program_code code;
+ /* Currently-bound fragment shader. */
+ struct r300_fragment_shader_code* shader;
+
+ /* List of the same shaders compiled with different texture-compare
+ * states. */
+ struct r300_fragment_shader_code* first;
};
+void r300_shader_read_fs_inputs(struct tgsi_shader_info* info,
+ struct r300_shader_semantics* fs_inputs);
+
+/* Return TRUE if the shader was switched and should be re-emitted. */
+boolean r300_pick_fragment_shader(struct r300_context* r300);
-void r300_translate_fragment_shader(struct r300_context* r300,
- struct r300_fragment_shader* fs);
+static INLINE boolean r300_fragment_shader_writes_depth(struct r300_fragment_shader *fs)
+{
+ if (!fs)
+ return FALSE;
+ return (fs->shader->code.writes_depth) ? TRUE : FALSE;
+}
#endif /* R300_FS_H */
diff --git a/src/gallium/drivers/r300/r300_query.c b/src/gallium/drivers/r300/r300_query.c
index 2880d34877f..ca00b043c51 100644
--- a/src/gallium/drivers/r300/r300_query.c
+++ b/src/gallium/drivers/r300/r300_query.c
@@ -20,17 +20,23 @@
* OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
* USE OR OTHER DEALINGS IN THE SOFTWARE. */
-#include "r300_query.h"
+#include "util/u_memory.h"
+#include "util/u_simple_list.h"
+#include "r300_context.h"
+#include "r300_screen.h"
+#include "r300_cs.h"
#include "r300_emit.h"
+#include "r300_query.h"
+#include "r300_reg.h"
-static struct pipe_query* r300_create_query(struct pipe_context* pipe,
+static struct pipe_query *r300_create_query(struct pipe_context *pipe,
unsigned query_type)
{
- struct r300_context* r300 = r300_context(pipe);
- struct r300_screen* r300screen = r300_screen(r300->context.screen);
- unsigned query_size = r300screen->caps->num_frag_pipes * 4;
- struct r300_query* q, * qptr;
+ struct r300_context *r300 = r300_context(pipe);
+ struct r300_screen *r300screen = r300_screen(r300->context.screen);
+ unsigned query_size;
+ struct r300_query *q, *qptr;
q = CALLOC_STRUCT(r300_query);
@@ -39,13 +45,16 @@ static struct pipe_query* r300_create_query(struct pipe_context* pipe,
q->active = FALSE;
- if (!r300->query_list) {
- r300->query_list = q;
- } else if (!is_empty_list(r300->query_list)) {
- qptr = last_elem(r300->query_list);
+ if (r300screen->caps->family == CHIP_FAMILY_RV530)
+ query_size = r300screen->caps->num_z_pipes * sizeof(uint32_t);
+ else
+ query_size = r300screen->caps->num_frag_pipes * sizeof(uint32_t);
+
+ 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);
}
+ insert_at_tail(&r300->query_list, q);
/* XXX */
if (q->offset >= 4096) {
@@ -71,24 +80,26 @@ static void r300_begin_query(struct pipe_context* pipe,
struct r300_context* r300 = r300_context(pipe);
struct r300_query* q = (struct r300_query*)query;
+ assert(r300->query_current == NULL);
+
map = pipe->screen->buffer_map(pipe->screen, r300->oqbo,
PIPE_BUFFER_USAGE_CPU_WRITE);
map += q->offset / 4;
- *map = ~0;
+ *map = ~0U;
pipe->screen->buffer_unmap(pipe->screen, r300->oqbo);
- r300_emit_dirty_state(r300);
- r300_emit_query_begin(r300, q);
+ q->flushed = FALSE;
+ r300->query_current = q;
+ r300->dirty_state |= R300_NEW_QUERY;
}
static void r300_end_query(struct pipe_context* pipe,
- struct pipe_query* query)
+ struct pipe_query* query)
{
struct r300_context* r300 = r300_context(pipe);
- struct r300_query* q = (struct r300_query*)query;
- r300_emit_dirty_state(r300);
- r300_emit_query_end(r300, q);
+ r300_emit_query_end(r300);
+ r300->query_current = NULL;
}
static boolean r300_get_query_result(struct pipe_context* pipe,
@@ -98,28 +109,36 @@ static boolean r300_get_query_result(struct pipe_context* pipe,
{
struct r300_context* r300 = r300_context(pipe);
struct r300_screen* r300screen = r300_screen(r300->context.screen);
- struct r300_query* q = (struct r300_query*)query;
+ struct r300_query *q = (struct r300_query*)query;
unsigned flags = PIPE_BUFFER_USAGE_CPU_READ;
uint32_t* map;
- uint32_t temp;
- unsigned i;
+ uint32_t temp = 0;
+ unsigned i, num_results;
- if (wait) {
+ if (q->flushed == FALSE)
pipe->flush(pipe, 0, NULL);
- } else {
+ if (!wait) {
flags |= PIPE_BUFFER_USAGE_DONTBLOCK;
}
map = pipe->screen->buffer_map(pipe->screen, r300->oqbo, flags);
+ if (!map)
+ return FALSE;
map += q->offset / 4;
- for (i = 0; i < r300screen->caps->num_frag_pipes; i++) {
- if (*map == ~0) {
+
+ 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) {
debug_printf("r300: Despite waiting, OQ results haven't"
" come in yet.\n");
}
- temp = ~0;
+ temp = ~0U;
break;
}
temp += *map;
@@ -127,7 +146,7 @@ static boolean r300_get_query_result(struct pipe_context* pipe,
}
pipe->screen->buffer_unmap(pipe->screen, r300->oqbo);
- if (temp == ~0) {
+ if (temp == ~0U) {
/* Our results haven't been written yet... */
return FALSE;
}
diff --git a/src/gallium/drivers/r300/r300_query.h b/src/gallium/drivers/r300/r300_query.h
index 4f50e8f8440..48876da3123 100644
--- a/src/gallium/drivers/r300/r300_query.h
+++ b/src/gallium/drivers/r300/r300_query.h
@@ -23,10 +23,6 @@
#ifndef R300_QUERY_H
#define R300_QUERY_H
-#include "r300_context.h"
-#include "r300_cs.h"
-#include "r300_reg.h"
-
struct r300_context;
static INLINE struct r300_query* r300_query(struct pipe_query* q)
diff --git a/src/gallium/drivers/r300/r300_reg.h b/src/gallium/drivers/r300/r300_reg.h
index 3abff5db622..034bfc15cf9 100644
--- a/src/gallium/drivers/r300/r300_reg.h
+++ b/src/gallium/drivers/r300/r300_reg.h
@@ -348,6 +348,27 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
# define R300_WRITE_ENA_W 8
# define R300_SWIZZLE1_SHIFT 16
+# define R300_VAP_SWIZZLE_X001 \
+ ((R300_SWIZZLE_SELECT_X << R300_SWIZZLE_SELECT_X_SHIFT) | \
+ (R300_SWIZZLE_SELECT_FP_ZERO << R300_SWIZZLE_SELECT_Y_SHIFT) | \
+ (R300_SWIZZLE_SELECT_FP_ZERO << R300_SWIZZLE_SELECT_Z_SHIFT) | \
+ (R300_SWIZZLE_SELECT_FP_ONE << R300_SWIZZLE_SELECT_W_SHIFT) | \
+ (0xf << R300_WRITE_ENA_SHIFT))
+
+# define R300_VAP_SWIZZLE_XY01 \
+ ((R300_SWIZZLE_SELECT_X << R300_SWIZZLE_SELECT_X_SHIFT) | \
+ (R300_SWIZZLE_SELECT_Y << R300_SWIZZLE_SELECT_Y_SHIFT) | \
+ (R300_SWIZZLE_SELECT_FP_ZERO << R300_SWIZZLE_SELECT_Z_SHIFT) | \
+ (R300_SWIZZLE_SELECT_FP_ONE << R300_SWIZZLE_SELECT_W_SHIFT) | \
+ (0xf << R300_WRITE_ENA_SHIFT))
+
+# define R300_VAP_SWIZZLE_XYZ1 \
+ ((R300_SWIZZLE_SELECT_X << R300_SWIZZLE_SELECT_X_SHIFT) | \
+ (R300_SWIZZLE_SELECT_Y << R300_SWIZZLE_SELECT_Y_SHIFT) | \
+ (R300_SWIZZLE_SELECT_Z << R300_SWIZZLE_SELECT_Z_SHIFT) | \
+ (R300_SWIZZLE_SELECT_FP_ONE << R300_SWIZZLE_SELECT_W_SHIFT) | \
+ (0xf << R300_WRITE_ENA_SHIFT))
+
# define R300_VAP_SWIZZLE_XYZW \
((R300_SWIZZLE_SELECT_X << R300_SWIZZLE_SELECT_X_SHIFT) | \
(R300_SWIZZLE_SELECT_Y << R300_SWIZZLE_SELECT_Y_SHIFT) | \
@@ -640,20 +661,20 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
# define R300_GB_SUPER_TILE_B (1 << 15)
# define R300_GB_SUBPIXEL_1_12 (0 << 16)
# define R300_GB_SUBPIXEL_1_16 (1 << 16)
-# define GB_TILE_CONFIG_QUADS_PER_RAS_4 (0 << 17)
-# define GB_TILE_CONFIG_QUADS_PER_RAS_8 (1 << 17)
-# define GB_TILE_CONFIG_QUADS_PER_RAS_16 (2 << 17)
-# define GB_TILE_CONFIG_QUADS_PER_RAS_32 (3 << 17)
-# define GB_TILE_CONFIG_BB_SCAN_INTERCEPT (0 << 19)
-# define GB_TILE_CONFIG_BB_SCAN_BOUND_BOX (1 << 19)
-# define GB_TILE_CONFIG_ALT_SCAN_EN_LR (0 << 20)
-# define GB_TILE_CONFIG_ALT_SCAN_EN_LRL (1 << 20)
-# define GB_TILE_CONFIG_ALT_OFFSET (0 << 21)
-# define GB_TILE_CONFIG_SUBPRECISION (0 << 22)
-# define GB_TILE_CONFIG_ALT_TILING_DEF (0 << 23)
-# define GB_TILE_CONFIG_ALT_TILING_3_2 (1 << 23)
-# define GB_TILE_CONFIG_Z_EXTENDED_24_1 (0 << 24)
-# define GB_TILE_CONFIG_Z_EXTENDED_S25_1 (1 << 24)
+# define R300_GB_TILE_CONFIG_QUADS_PER_RAS_4 (0 << 17)
+# define R300_GB_TILE_CONFIG_QUADS_PER_RAS_8 (1 << 17)
+# define R300_GB_TILE_CONFIG_QUADS_PER_RAS_16 (2 << 17)
+# define R300_GB_TILE_CONFIG_QUADS_PER_RAS_32 (3 << 17)
+# define R300_GB_TILE_CONFIG_BB_SCAN_INTERCEPT (0 << 19)
+# define R300_GB_TILE_CONFIG_BB_SCAN_BOUND_BOX (1 << 19)
+# define R300_GB_TILE_CONFIG_ALT_SCAN_EN_LR (0 << 20)
+# define R300_GB_TILE_CONFIG_ALT_SCAN_EN_LRL (1 << 20)
+# define R300_GB_TILE_CONFIG_ALT_OFFSET (0 << 21)
+# define R300_GB_TILE_CONFIG_SUBPRECISION (0 << 22)
+# define R300_GB_TILE_CONFIG_ALT_TILING_DEF (0 << 23)
+# define R300_GB_TILE_CONFIG_ALT_TILING_3_2 (1 << 23)
+# define R300_GB_TILE_CONFIG_Z_EXTENDED_24_1 (0 << 24)
+# define R300_GB_TILE_CONFIG_Z_EXTENDED_S25_1 (1 << 24)
/* Specifies the sizes of the various FIFO`s in the sc/rs/us. This register must be the first one written */
#define R300_GB_FIFO_SIZE 0x4024
@@ -679,9 +700,9 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
# define R300_OFIFO_HIGHWATER_SHIFT 22 /* two bits only */
# define R300_CUBE_FIFO_HIGHWATER_COL_SHIFT 24
-#define GB_Z_PEQ_CONFIG 0x4028
-# define GB_Z_PEQ_CONFIG_Z_PEQ_SIZE_4_4 (0 << 0)
-# define GB_Z_PEQ_CONFIG_Z_PEQ_SIZE_8_8 (1 << 0)
+#define R300_GB_Z_PEQ_CONFIG 0x4028
+# define R300_GB_Z_PEQ_CONFIG_Z_PEQ_SIZE_4_4 (0 << 0)
+# define R300_GB_Z_PEQ_CONFIG_Z_PEQ_SIZE_8_8 (1 << 0)
/* Specifies various polygon specific selects (fog, depth, perspective). */
#define R300_GB_SELECT 0x401c
@@ -704,39 +725,39 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
/* Specifies the graphics pipeline configuration for antialiasing. */
#define R300_GB_AA_CONFIG 0x4020
-# define GB_AA_CONFIG_AA_DISABLE (0 << 0)
-# define GB_AA_CONFIG_AA_ENABLE (1 << 0)
-# define GB_AA_CONFIG_NUM_AA_SUBSAMPLES_2 (0 << 1)
-# define GB_AA_CONFIG_NUM_AA_SUBSAMPLES_3 (1 << 1)
-# define GB_AA_CONFIG_NUM_AA_SUBSAMPLES_4 (2 << 1)
-# define GB_AA_CONFIG_NUM_AA_SUBSAMPLES_6 (3 << 1)
+# define R300_GB_AA_CONFIG_AA_DISABLE (0 << 0)
+# define R300_GB_AA_CONFIG_AA_ENABLE (1 << 0)
+# define R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_2 (0 << 1)
+# define R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_3 (1 << 1)
+# define R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_4 (2 << 1)
+# define R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_6 (3 << 1)
/* Selects which of 4 pipes are active. */
-#define GB_PIPE_SELECT 0x402c
-# define GB_PIPE_SELECT_PIPE0_ID_SHIFT 0
-# define GB_PIPE_SELECT_PIPE1_ID_SHIFT 2
-# define GB_PIPE_SELECT_PIPE2_ID_SHIFT 4
-# define GB_PIPE_SELECT_PIPE3_ID_SHIFT 6
-# define GB_PIPE_SELECT_PIPE_MASK_SHIFT 8
-# define GB_PIPE_SELECT_MAX_PIPE 12
-# define GB_PIPE_SELECT_BAD_PIPES 14
-# define GB_PIPE_SELECT_CONFIG_PIPES 18
+#define R300_GB_PIPE_SELECT 0x402c
+# define R300_GB_PIPE_SELECT_PIPE0_ID_SHIFT 0
+# define R300_GB_PIPE_SELECT_PIPE1_ID_SHIFT 2
+# define R300_GB_PIPE_SELECT_PIPE2_ID_SHIFT 4
+# define R300_GB_PIPE_SELECT_PIPE3_ID_SHIFT 6
+# define R300_GB_PIPE_SELECT_PIPE_MASK_SHIFT 8
+# define R300_GB_PIPE_SELECT_MAX_PIPE 12
+# define R300_GB_PIPE_SELECT_BAD_PIPES 14
+# define R300_GB_PIPE_SELECT_CONFIG_PIPES 18
/* Specifies the sizes of the various FIFO`s in the sc/rs. */
-#define GB_FIFO_SIZE1 0x4070
+#define R300_GB_FIFO_SIZE1 0x4070
/* High water mark for SC input fifo */
-# define GB_FIFO_SIZE1_SC_HIGHWATER_IFIFO_SHIFT 0
-# define GB_FIFO_SIZE1_SC_HIGHWATER_IFIFO_MASK 0x0000003f
+# define R300_GB_FIFO_SIZE1_SC_HIGHWATER_IFIFO_SHIFT 0
+# define R300_GB_FIFO_SIZE1_SC_HIGHWATER_IFIFO_MASK 0x0000003f
/* High water mark for SC input fifo (B) */
-# define GB_FIFO_SIZE1_SC_HIGHWATER_BFIFO_SHIFT 6
-# define GB_FIFO_SIZE1_SC_HIGHWATER_BFIFO_MASK 0x00000fc0
+# define R300_GB_FIFO_SIZE1_SC_HIGHWATER_BFIFO_SHIFT 6
+# define R300_GB_FIFO_SIZE1_SC_HIGHWATER_BFIFO_MASK 0x00000fc0
/* High water mark for RS colors' fifo */
-# define GB_FIFO_SIZE1_SC_HIGHWATER_COL_SHIFT 12
-# define GB_FIFO_SIZE1_SC_HIGHWATER_COL_MASK 0x0003f000
+# define R300_GB_FIFO_SIZE1_SC_HIGHWATER_COL_SHIFT 12
+# define R300_GB_FIFO_SIZE1_SC_HIGHWATER_COL_MASK 0x0003f000
/* High water mark for RS textures' fifo */
-# define GB_FIFO_SIZE1_SC_HIGHWATER_TEX_SHIFT 18
-# define GB_FIFO_SIZE1_SC_HIGHWATER_TEX_MASK 0x00fc0000
+# define R300_GB_FIFO_SIZE1_SC_HIGHWATER_TEX_SHIFT 18
+# define R300_GB_FIFO_SIZE1_SC_HIGHWATER_TEX_MASK 0x00fc0000
/* This table specifies the source location and format for up to 16 texture
* addresses (i[0]:i[15]) and four colors (c[0]:c[3])
@@ -841,10 +862,10 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
# define R300_POINTSIZE_X_MASK 0xffff0000
# define R300_POINTSIZE_MAX (R300_POINTSIZE_Y_MASK / 6)
-/* Blue fill color */
+/* Red fill color */
#define R500_GA_FILL_R 0x4220
-/* Blue fill color */
+/* Green fill color */
#define R500_GA_FILL_G 0x4224
/* Blue fill color */
@@ -1172,6 +1193,13 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
/* SU Depth Offset value */
#define R300_SU_DEPTH_OFFSET 0x42c4
+#define R300_SU_REG_DEST 0x42c8
+# define R300_RASTER_PIPE_SELECT_0 (1 << 0)
+# define R300_RASTER_PIPE_SELECT_1 (1 << 1)
+# define R300_RASTER_PIPE_SELECT_2 (1 << 2)
+# define R300_RASTER_PIPE_SELECT_3 (1 << 3)
+# define R300_RASTER_PIPE_SELECT_ALL 0xf
+
/* BEGIN: Rasterization / Interpolators - many guesses */
@@ -1265,7 +1293,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
# define R500_RS_INST_TEX_ID(x) ((x) << 0)
#define R500_RS_INST_TEX_CN_WRITE (1 << 4)
#define R500_RS_INST_TEX_ADDR_SHIFT 5
-# define R500_RS_INST_TEX_ADDR(x) ((x) << 0)
+# define R500_RS_INST_TEX_ADDR(x) ((x) << 5)
#define R500_RS_INST_COL_ID_SHIFT 12
# define R500_RS_INST_COL_ID(x) ((x) << 12)
#define R500_RS_INST_COL_CN_NO_WRITE (0 << 16)
@@ -1435,6 +1463,8 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
# define R300_TX_MIN_FILTER_MIP_NEAREST (1 << 13)
# define R300_TX_MIN_FILTER_MIP_LINEAR (2 << 13)
# define R300_TX_MIN_FILTER_MIP_MASK (3 << 13)
+# define R300_TX_MAX_MIP_LEVEL_SHIFT 17
+# define R300_TX_MAX_MIP_LEVEL_MASK (0xf << 17)
# define R300_TX_MAX_ANISO_1_TO_1 (0 << 21)
# define R300_TX_MAX_ANISO_2_TO_1 (1 << 21)
# define R300_TX_MAX_ANISO_4_TO_1 (2 << 21)
@@ -1443,6 +1473,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
# define R300_TX_MAX_ANISO_MASK (7 << 21)
# define R300_TX_WRAP_S(x) ((x) << 0)
# define R300_TX_WRAP_T(x) ((x) << 3)
+# define R300_TX_MAX_MIP_LEVEL(x) ((x) << 17)
#define R300_TX_FILTER1_0 0x4440
# define R300_CHROMA_KEY_MODE_DISABLE 0
@@ -1472,8 +1503,6 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
# define R300_TX_HEIGHTMASK_MASK (2047 << 11)
# define R300_TX_DEPTHMASK_SHIFT 22
# define R300_TX_DEPTHMASK_MASK (0xf << 22)
-# define R300_TX_MAX_MIP_LEVEL_SHIFT 26
-# define R300_TX_MAX_MIP_LEVEL_MASK (0xf << 26)
# define R300_TX_SIZE_PROJECTED (1 << 30)
# define R300_TX_PITCH_EN (1 << 31)
# define R300_TX_WIDTH(x) ((x) << 0)
@@ -1856,6 +1885,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
# define R300_RGB_ADDR0(x) ((x) << 0)
# define R300_RGB_ADDR1(x) ((x) << 6)
# define R300_RGB_ADDR2(x) ((x) << 12)
+# define R300_RGB_TARGET(x) ((x) << 29)
#define R300_US_ALU_ALPHA_ADDR_0 0x47C0
# define R300_ALU_SRC0A_SHIFT 0
@@ -1873,9 +1903,10 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
# define R300_ALU_DSTA_REG (1 << 23)
# define R300_ALU_DSTA_OUTPUT (1 << 24)
# define R300_ALU_DSTA_DEPTH (1 << 27)
-# define R300_ALPHA_ADDR0(x) ((x) << 0)
-# define R300_ALPHA_ADDR1(x) ((x) << 6)
-# define R300_ALPHA_ADDR2(x) ((x) << 12)
+# define R300_ALPHA_ADDR0(x) ((x) << 0)
+# define R300_ALPHA_ADDR1(x) ((x) << 6)
+# define R300_ALPHA_ADDR2(x) ((x) << 12)
+# define R300_ALPHA_TARGET(x) ((x) << 25)
#define R300_US_ALU_RGB_INST_0 0x48C0
# define R300_ALU_ARGC_SRC0C_XYZ 0
@@ -2095,6 +2126,10 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
#define R500_FG_ALPHA_VALUE 0x4be0
# define R500_FG_ALPHA_VALUE_MASK 0x0000ffff
+#define RV530_FG_ZBREG_DEST 0x4be8
+# define RV530_FG_ZBREG_DEST_PIPE_SELECT_0 (1 << 0)
+# define RV530_FG_ZBREG_DEST_PIPE_SELECT_1 (1 << 1)
+# define RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL (3 << 0)
/* gap */
/* Fragment program parameters in 7.16 floating point */
@@ -2110,6 +2145,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
/* Unpipelined. */
#define R300_RB3D_CCTL 0x4e00
+# define R300_RB3D_CCTL_NUM_MULTIWRITES(x) (MAX2(((x)-1), 0) << 5)
# define R300_RB3D_CCTL_NUM_MULTIWRITES_1_BUFFER (0 << 5)
# define R300_RB3D_CCTL_NUM_MULTIWRITES_2_BUFFERS (1 << 5)
# define R300_RB3D_CCTL_NUM_MULTIWRITES_3_BUFFERS (2 << 5)
@@ -2150,6 +2186,8 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
# define R300_DISCARD_SRC_PIXELS_SRC_ALPHA_1 (4 << 3)
# define R300_DISCARD_SRC_PIXELS_SRC_COLOR_1 (5 << 3)
# define R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_1 (6 << 3)
+# define R500_SRC_ALPHA_0_NO_READ (1 << 30)
+# define R500_SRC_ALPHA_1_NO_READ (1 << 31)
/* the following are shared between CBLEND and ABLEND */
# define R300_FCN_MASK (3 << 12)
@@ -2384,6 +2422,8 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
# define R300_Z_WRITE_ENABLE (1 << 2)
# define R300_Z_SIGNED_COMPARE (1 << 3)
# define R300_STENCIL_FRONT_BACK (1 << 4)
+# define R500_STENCIL_ZSIGNED_MAGNITUDE (1 << 5)
+# define R500_STENCIL_REFMASK_FRONT_BACK (1 << 6)
#define R300_ZB_ZSTENCILCNTL 0x4f04
/* functions */
@@ -2600,7 +2640,7 @@ enum {
VE_COND_MUX_GTE = 25,
VE_SET_GREATER_THAN = 26,
VE_SET_EQUAL = 27,
- VE_SET_NOT_EQUAL = 28,
+ VE_SET_NOT_EQUAL = 28
};
enum {
@@ -2634,20 +2674,20 @@ enum {
ME_PRED_SET_CLR = 25,
ME_PRED_SET_INV = 26,
ME_PRED_SET_POP = 27,
- ME_PRED_SET_RESTORE = 28,
+ ME_PRED_SET_RESTORE = 28
};
enum {
/* R3XX */
PVS_MACRO_OP_2CLK_MADD = 0,
- PVS_MACRO_OP_2CLK_M2X_ADD = 1,
+ PVS_MACRO_OP_2CLK_M2X_ADD = 1
};
enum {
PVS_SRC_REG_TEMPORARY = 0, /* Intermediate Storage */
PVS_SRC_REG_INPUT = 1, /* Input Vertex Storage */
PVS_SRC_REG_CONSTANT = 2, /* Constant State Storage */
- PVS_SRC_REG_ALT_TEMPORARY = 3, /* Alternate Intermediate Storage */
+ PVS_SRC_REG_ALT_TEMPORARY = 3 /* Alternate Intermediate Storage */
};
enum {
@@ -2656,7 +2696,7 @@ enum {
PVS_DST_REG_OUT = 2, /* Output Memory. Used for all outputs */
PVS_DST_REG_OUT_REPL_X = 3, /* Output Memory & Replicate X to all channels */
PVS_DST_REG_ALT_TEMPORARY = 4, /* Alternate Intermediate Storage */
- PVS_DST_REG_INPUT = 5, /* Output Memory & Replicate X to all channels */
+ PVS_DST_REG_INPUT = 5 /* Output Memory & Replicate X to all channels */
};
enum {
@@ -2665,7 +2705,7 @@ enum {
PVS_SRC_SELECT_Z = 2, /* Select Z Component */
PVS_SRC_SELECT_W = 3, /* Select W Component */
PVS_SRC_SELECT_FORCE_0 = 4, /* Force Component to 0.0 */
- PVS_SRC_SELECT_FORCE_1 = 5, /* Force Component to 1.0 */
+ PVS_SRC_SELECT_FORCE_1 = 5 /* Force Component to 1.0 */
};
/* PVS Opcode & Destination Operand Description */
@@ -2704,7 +2744,7 @@ enum {
PVS_DST_ADDR_SEL_MASK = 0x3,
PVS_DST_ADDR_SEL_SHIFT = 29,
PVS_DST_ADDR_MODE_0_MASK = 0x1,
- PVS_DST_ADDR_MODE_0_SHIFT = 31,
+ PVS_DST_ADDR_MODE_0_SHIFT = 31
};
/* PVS Source Operand Description */
@@ -2739,7 +2779,7 @@ enum {
PVS_SRC_ADDR_SEL_MASK = 0x3,
PVS_SRC_ADDR_SEL_SHIFT = 29,
PVS_SRC_ADDR_MODE_1_MASK = 0x0,
- PVS_SRC_ADDR_MODE_1_SHIFT = 32,
+ PVS_SRC_ADDR_MODE_1_SHIFT = 32
};
/*\}*/
@@ -3256,6 +3296,11 @@ enum {
*/
#define R300_PACKET3_3D_LOAD_VBPNTR 0x00002F00
+# define R300_VBPNTR_SIZE0(x) ((x) >> 2)
+# define R300_VBPNTR_STRIDE0(x) (((x) >> 2) << 8)
+# define R300_VBPNTR_SIZE1(x) (((x) >> 2) << 16)
+# define R300_VBPNTR_STRIDE1(x) (((x) >> 2) << 24)
+
#define R300_PACKET3_INDX_BUFFER 0x00003300
# define R300_INDX_BUFFER_DST_SHIFT 0
# define R300_INDX_BUFFER_SKIP_SHIFT 16
@@ -3313,10 +3358,6 @@ enum {
#define R200_3D_DRAW_IMMD_2 0xC0003500
-/* XXX Oh look, stuff not brought over from docs yet */
-
-#define R300_SU_REG_DEST 0x42C8
-
#endif /* _R300_REG_H */
/* *INDENT-ON* */
diff --git a/src/gallium/drivers/r300/r300_render.c b/src/gallium/drivers/r300/r300_render.c
index ca44e0f6615..a4ac9ad9a7b 100644
--- a/src/gallium/drivers/r300/r300_render.c
+++ b/src/gallium/drivers/r300/r300_render.c
@@ -20,19 +20,389 @@
* OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
* USE OR OTHER DEALINGS IN THE SOFTWARE. */
-#include "draw/draw_pipe.h"
+/* r300_render: Vertex and index buffer primitive emission. Contains both
+ * HW TCL fastpath rendering, and SW TCL Draw-assisted rendering. */
+
+#include "draw/draw_context.h"
#include "draw/draw_vbuf.h"
+
+#include "pipe/p_inlines.h"
+
#include "util/u_memory.h"
+#include "util/u_prim.h"
#include "r300_cs.h"
#include "r300_context.h"
#include "r300_emit.h"
#include "r300_reg.h"
+#include "r300_render.h"
#include "r300_state_derived.h"
/* r300_render: Vertex and index buffer primitive emission. */
#define R300_MAX_VBO_SIZE (1024 * 1024)
+uint32_t r300_translate_primitive(unsigned prim)
+{
+ switch (prim) {
+ case PIPE_PRIM_POINTS:
+ return R300_VAP_VF_CNTL__PRIM_POINTS;
+ case PIPE_PRIM_LINES:
+ return R300_VAP_VF_CNTL__PRIM_LINES;
+ case PIPE_PRIM_LINE_LOOP:
+ return R300_VAP_VF_CNTL__PRIM_LINE_LOOP;
+ case PIPE_PRIM_LINE_STRIP:
+ return R300_VAP_VF_CNTL__PRIM_LINE_STRIP;
+ case PIPE_PRIM_TRIANGLES:
+ return R300_VAP_VF_CNTL__PRIM_TRIANGLES;
+ case PIPE_PRIM_TRIANGLE_STRIP:
+ return R300_VAP_VF_CNTL__PRIM_TRIANGLE_STRIP;
+ case PIPE_PRIM_TRIANGLE_FAN:
+ return R300_VAP_VF_CNTL__PRIM_TRIANGLE_FAN;
+ case PIPE_PRIM_QUADS:
+ return R300_VAP_VF_CNTL__PRIM_QUADS;
+ case PIPE_PRIM_QUAD_STRIP:
+ return R300_VAP_VF_CNTL__PRIM_QUAD_STRIP;
+ case PIPE_PRIM_POLYGON:
+ return R300_VAP_VF_CNTL__PRIM_POLYGON;
+ default:
+ return 0;
+ }
+}
+
+static boolean r300_nothing_to_draw(struct r300_context *r300)
+{
+ return r300->rs_state->rs.scissor &&
+ r300->scissor_state->scissor.empty_area;
+}
+
+static uint32_t r300_provoking_vertex_fixes(struct r300_context *r300,
+ unsigned mode)
+{
+ uint32_t color_control = r300->rs_state->color_control;
+
+ /* By default (see r300_state.c:r300_create_rs_state) color_control is
+ * initialized to provoking the first vertex.
+ *
+ * Triangle fans must be reduced to the second vertex, not the first, in
+ * Gallium flatshade-first mode, as per the GL spec.
+ * (http://www.opengl.org/registry/specs/ARB/provoking_vertex.txt)
+ *
+ * Quads never provoke correctly in flatshade-first mode. The first
+ * vertex is never considered as provoking, so only the second, third,
+ * and fourth vertices can be selected, and both "third" and "last" modes
+ * select the fourth vertex. This is probably due to D3D lacking quads.
+ *
+ * Similarly, polygons reduce to the first, not the last, vertex, when in
+ * "last" mode, and all other modes start from the second vertex.
+ *
+ * ~ C.
+ */
+
+ if (r300->rs_state->rs.flatshade_first) {
+ switch (mode) {
+ case PIPE_PRIM_TRIANGLE_FAN:
+ color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_SECOND;
+ break;
+ case PIPE_PRIM_QUADS:
+ case PIPE_PRIM_QUAD_STRIP:
+ case PIPE_PRIM_POLYGON:
+ color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST;
+ break;
+ default:
+ color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_FIRST;
+ break;
+ }
+ } else {
+ color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST;
+ }
+
+ return color_control;
+}
+
+static void r300_emit_draw_arrays(struct r300_context *r300,
+ unsigned mode,
+ unsigned count)
+{
+ CS_LOCALS(r300);
+
+ BEGIN_CS(8);
+ OUT_CS_REG(R300_GA_COLOR_CONTROL,
+ r300_provoking_vertex_fixes(r300, mode));
+ OUT_CS_REG(R300_VAP_VF_MIN_VTX_INDX, 0);
+ 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) |
+ r300_translate_primitive(mode));
+ END_CS;
+}
+
+static void r300_emit_draw_elements(struct r300_context *r300,
+ struct pipe_buffer* 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);
+ CS_LOCALS(r300);
+
+ /* XXX most of these are stupid */
+ assert(indexSize == 4 || indexSize == 2);
+ assert((start * indexSize) % 4 == 0);
+ assert(offset_dwords == 0);
+
+ BEGIN_CS(14);
+ OUT_CS_REG(R300_GA_COLOR_CONTROL,
+ r300_provoking_vertex_fixes(r300, mode));
+ OUT_CS_REG(R300_VAP_VF_MIN_VTX_INDX, minIndex);
+ OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, maxIndex);
+ OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0);
+ if (indexSize == 4) {
+ count_dwords = count + start;
+ OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
+ R300_VAP_VF_CNTL__INDEX_SIZE_32bit |
+ r300_translate_primitive(mode));
+ } else {
+ count_dwords = (count + start + 1) / 2;
+ OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
+ r300_translate_primitive(mode));
+ }
+
+ /* INDX_BUFFER is a truly special packet3.
+ * Unlike most other packet3, where the offset is after the count,
+ * the order is reversed, so the relocation ends up carrying the
+ * size of the indexbuf instead of the offset.
+ *
+ * XXX Fix offset
+ */
+ OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER, 2);
+ OUT_CS(R300_INDX_BUFFER_ONE_REG_WR | (R300_VAP_PORT_IDX0 >> 2) |
+ (0 << R300_INDX_BUFFER_SKIP_SHIFT));
+ OUT_CS(offset_dwords);
+ OUT_CS_RELOC(indexBuffer, count_dwords,
+ RADEON_GEM_DOMAIN_GTT, 0, 0);
+
+ END_CS;
+}
+
+
+static boolean r300_setup_vertex_buffers(struct r300_context *r300)
+{
+ struct pipe_vertex_buffer *vbuf = r300->vertex_buffer;
+ struct pipe_vertex_element *velem = r300->vertex_element;
+
+validate:
+ for (int i = 0; i < r300->vertex_element_count; i++) {
+ if (!r300->winsys->add_buffer(r300->winsys,
+ vbuf[velem[i].vertex_buffer_index].buffer,
+ RADEON_GEM_DOMAIN_GTT, 0)) {
+ r300->context.flush(&r300->context, 0, NULL);
+ goto validate;
+ }
+ }
+
+ if (!r300->winsys->validate(r300->winsys)) {
+ r300->context.flush(&r300->context, 0, NULL);
+ return r300->winsys->validate(r300->winsys);
+ }
+
+ return TRUE;
+}
+
+/* This is the fast-path drawing & emission for HW TCL. */
+void r300_draw_range_elements(struct pipe_context* pipe,
+ struct pipe_buffer* indexBuffer,
+ unsigned indexSize,
+ unsigned minIndex,
+ unsigned maxIndex,
+ unsigned mode,
+ unsigned start,
+ unsigned count)
+{
+ struct r300_context* r300 = r300_context(pipe);
+
+ if (!u_trim_pipe_prim(mode, &count)) {
+ return;
+ }
+
+ if (count > 65535) {
+ /* XXX: use aux/indices functions to split this into smaller
+ * primitives.
+ */
+ return;
+ }
+
+ if (r300_nothing_to_draw(r300)) {
+ return;
+ }
+
+ r300_update_derived_state(r300);
+
+ if (!r300_setup_vertex_buffers(r300)) {
+ return;
+ }
+
+ if (!r300->winsys->add_buffer(r300->winsys, indexBuffer,
+ RADEON_GEM_DOMAIN_GTT, 0)) {
+ return;
+ }
+
+ if (!r300->winsys->validate(r300->winsys)) {
+ return;
+ }
+
+ r300_emit_dirty_state(r300);
+
+ r300_emit_aos(r300, 0);
+
+ r300_emit_draw_elements(r300, indexBuffer, indexSize, minIndex, maxIndex,
+ mode, start, count);
+}
+
+/* Simple helpers for context setup. Should probably be moved to util. */
+void r300_draw_elements(struct pipe_context* pipe,
+ struct pipe_buffer* indexBuffer,
+ unsigned indexSize, unsigned mode,
+ unsigned start, unsigned count)
+{
+ pipe->draw_range_elements(pipe, indexBuffer, indexSize, 0, ~0,
+ mode, start, count);
+}
+
+void r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
+ unsigned start, unsigned count)
+{
+ struct r300_context* r300 = r300_context(pipe);
+
+ if (!u_trim_pipe_prim(mode, &count)) {
+ return;
+ }
+
+ if (count > 65535) {
+ /* XXX: driver needs to handle this -- use the functions in
+ * aux/indices to split this into several smaller primitives.
+ */
+ return;
+ }
+
+ if (r300_nothing_to_draw(r300)) {
+ return;
+ }
+
+ r300_update_derived_state(r300);
+
+ if (!r300_setup_vertex_buffers(r300)) {
+ return;
+ }
+
+ r300_emit_dirty_state(r300);
+
+ r300_emit_aos(r300, start);
+
+ r300_emit_draw_arrays(r300, mode, 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 arrays, using Draw. */
+void r300_swtcl_draw_arrays(struct pipe_context* pipe,
+ unsigned mode,
+ unsigned start,
+ unsigned count)
+{
+ struct r300_context* r300 = r300_context(pipe);
+ int i;
+
+ if (!u_trim_pipe_prim(mode, &count)) {
+ return;
+ }
+
+ if (r300_nothing_to_draw(r300)) {
+ return;
+ }
+
+ for (i = 0; i < r300->vertex_buffer_count; i++) {
+ void* buf = pipe_buffer_map(pipe->screen,
+ r300->vertex_buffer[i].buffer,
+ PIPE_BUFFER_USAGE_CPU_READ);
+ draw_set_mapped_vertex_buffer(r300->draw, i, buf);
+ }
+
+ draw_set_mapped_element_buffer(r300->draw, 0, NULL);
+
+ draw_set_mapped_constant_buffer(r300->draw,
+ PIPE_SHADER_VERTEX,
+ r300->shader_constants[PIPE_SHADER_VERTEX].constants,
+ r300->shader_constants[PIPE_SHADER_VERTEX].count *
+ (sizeof(float) * 4));
+
+ draw_arrays(r300->draw, mode, start, count);
+
+ for (i = 0; i < r300->vertex_buffer_count; i++) {
+ pipe_buffer_unmap(pipe->screen, r300->vertex_buffer[i].buffer);
+ draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
+ }
+}
+
+/* SW TCL elements, using Draw. */
+void r300_swtcl_draw_range_elements(struct pipe_context* pipe,
+ struct pipe_buffer* indexBuffer,
+ unsigned indexSize,
+ unsigned minIndex,
+ unsigned maxIndex,
+ unsigned mode,
+ unsigned start,
+ unsigned count)
+{
+ struct r300_context* r300 = r300_context(pipe);
+ int i;
+ void* indices;
+
+ if (!u_trim_pipe_prim(mode, &count)) {
+ return;
+ }
+
+ if (r300_nothing_to_draw(r300)) {
+ return;
+ }
+
+ for (i = 0; i < r300->vertex_buffer_count; i++) {
+ void* buf = pipe_buffer_map(pipe->screen,
+ r300->vertex_buffer[i].buffer,
+ PIPE_BUFFER_USAGE_CPU_READ);
+ draw_set_mapped_vertex_buffer(r300->draw, i, buf);
+ }
+
+ indices = pipe_buffer_map(pipe->screen, indexBuffer,
+ PIPE_BUFFER_USAGE_CPU_READ);
+ draw_set_mapped_element_buffer_range(r300->draw, indexSize,
+ minIndex, maxIndex, indices);
+
+ draw_set_mapped_constant_buffer(r300->draw,
+ PIPE_SHADER_VERTEX,
+ r300->shader_constants[PIPE_SHADER_VERTEX].constants,
+ r300->shader_constants[PIPE_SHADER_VERTEX].count *
+ (sizeof(float) * 4));
+
+ draw_arrays(r300->draw, mode, start, count);
+
+ for (i = 0; i < r300->vertex_buffer_count; i++) {
+ pipe_buffer_unmap(pipe->screen, r300->vertex_buffer[i].buffer);
+ draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
+ }
+
+ pipe_buffer_unmap(pipe->screen, indexBuffer);
+ draw_set_mapped_element_buffer_range(r300->draw, 0, start,
+ start + count - 1, NULL);
+}
+
+/* Object for rendering using Draw. */
struct r300_render {
/* Parent class */
struct vbuf_render base;
@@ -67,7 +437,7 @@ r300_render_get_vertex_info(struct vbuf_render* render)
r300_update_derived_state(r300);
- return &r300->vertex_info.vinfo;
+ return &r300->vertex_info->vinfo;
}
static boolean r300_render_allocate_vertices(struct vbuf_render* render,
@@ -105,7 +475,7 @@ static void* r300_render_map_vertices(struct vbuf_render* render)
r300render->vbo_ptr = pipe_buffer_map(screen, r300render->vbo,
PIPE_BUFFER_USAGE_CPU_WRITE);
- return (r300render->vbo_ptr + r300render->vbo_offset);
+ return ((uint8_t*)r300render->vbo_ptr + r300render->vbo_offset);
}
static void r300_render_unmap_vertices(struct vbuf_render* render,
@@ -127,7 +497,6 @@ static void r300_render_unmap_vertices(struct vbuf_render* render,
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;
r300render->vbo_max_used = 0;
@@ -137,56 +506,13 @@ static boolean r300_render_set_primitive(struct vbuf_render* render,
unsigned prim)
{
struct r300_render* r300render = r300_render(render);
- r300render->prim = prim;
- switch (prim) {
- case PIPE_PRIM_POINTS:
- r300render->hwprim = R300_VAP_VF_CNTL__PRIM_POINTS;
- break;
- case PIPE_PRIM_LINES:
- r300render->hwprim = R300_VAP_VF_CNTL__PRIM_LINES;
- break;
- case PIPE_PRIM_LINE_LOOP:
- r300render->hwprim = R300_VAP_VF_CNTL__PRIM_LINE_LOOP;
- break;
- case PIPE_PRIM_LINE_STRIP:
- r300render->hwprim = R300_VAP_VF_CNTL__PRIM_LINE_STRIP;
- break;
- case PIPE_PRIM_TRIANGLES:
- r300render->hwprim = R300_VAP_VF_CNTL__PRIM_TRIANGLES;
- break;
- case PIPE_PRIM_TRIANGLE_STRIP:
- r300render->hwprim = R300_VAP_VF_CNTL__PRIM_TRIANGLE_STRIP;
- break;
- case PIPE_PRIM_TRIANGLE_FAN:
- r300render->hwprim = R300_VAP_VF_CNTL__PRIM_TRIANGLE_FAN;
- break;
- case PIPE_PRIM_QUADS:
- r300render->hwprim = R300_VAP_VF_CNTL__PRIM_QUADS;
- break;
- case PIPE_PRIM_QUAD_STRIP:
- r300render->hwprim = R300_VAP_VF_CNTL__PRIM_QUAD_STRIP;
- break;
- case PIPE_PRIM_POLYGON:
- r300render->hwprim = R300_VAP_VF_CNTL__PRIM_POLYGON;
- break;
- default:
- return FALSE;
- break;
- }
+ r300render->prim = prim;
+ r300render->hwprim = r300_translate_primitive(prim);
return TRUE;
}
-static void r300_prepare_render(struct r300_render* render, unsigned count)
-{
- struct r300_context* r300 = render->r300;
-
- CS_LOCALS(r300);
-
- r300_emit_dirty_state(r300);
-}
-
static void r300_render_draw_arrays(struct vbuf_render* render,
unsigned start,
unsigned count)
@@ -196,7 +522,7 @@ static void r300_render_draw_arrays(struct vbuf_render* render,
CS_LOCALS(r300);
- r300_prepare_render(r300render, count);
+ r300_emit_dirty_state(r300);
DBG(r300, DBG_DRAW, "r300: Doing vbuf render, count %d\n", count);
@@ -213,15 +539,11 @@ static void r300_render_draw(struct vbuf_render* render,
{
struct r300_render* r300render = r300_render(render);
struct r300_context* r300 = r300render->r300;
- struct pipe_screen* screen = r300->context.screen;
- struct pipe_buffer* index_buffer;
- void* index_map;
int i;
- uint32_t index;
CS_LOCALS(r300);
- r300_prepare_render(r300render, count);
+ r300_emit_dirty_state(r300);
BEGIN_CS(2 + (count+1)/2);
OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, (count+1)/2);
diff --git a/src/gallium/drivers/r300/r300_render.h b/src/gallium/drivers/r300/r300_render.h
new file mode 100644
index 00000000000..27b5e6a9630
--- /dev/null
+++ b/src/gallium/drivers/r300/r300_render.h
@@ -0,0 +1,59 @@
+/*
+ * 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
+
+uint32_t r300_translate_primitive(unsigned prim);
+
+void r300_draw_range_elements(struct pipe_context* pipe,
+ struct pipe_buffer* indexBuffer,
+ unsigned indexSize,
+ unsigned minIndex,
+ unsigned maxIndex,
+ unsigned mode,
+ unsigned start,
+ unsigned count);
+
+void r300_draw_elements(struct pipe_context* pipe,
+ struct pipe_buffer* indexBuffer,
+ unsigned indexSize, 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_buffer* indexBuffer,
+ unsigned indexSize,
+ unsigned minIndex,
+ unsigned maxIndex,
+ unsigned mode,
+ unsigned start,
+ unsigned count);
+
+#endif /* R300_RENDER_H */
diff --git a/src/gallium/drivers/r300/r300_screen.c b/src/gallium/drivers/r300/r300_screen.c
index 81d01b1320c..287664b1d20 100644
--- a/src/gallium/drivers/r300/r300_screen.c
+++ b/src/gallium/drivers/r300/r300_screen.c
@@ -20,7 +20,16 @@
* OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
* USE OR OTHER DEALINGS IN THE SOFTWARE. */
+#include "pipe/p_inlines.h"
+#include "util/u_format.h"
+#include "util/u_memory.h"
+#include "util/u_simple_screen.h"
+
+#include "r300_context.h"
#include "r300_screen.h"
+#include "r300_texture.h"
+
+#include "radeon_winsys.h"
/* Return the identifier behind whom the brave coders responsible for this
* amalgamation of code, sweat, and duct tape, routinely obscure their names.
@@ -73,14 +82,14 @@ static int r300_get_param(struct pipe_screen* pscreen, int param)
struct r300_screen* r300screen = r300_screen(pscreen);
switch (param) {
- /* XXX cases marked "IN THEORY" are possible on the hardware,
- * but haven't been implemented yet. */
case PIPE_CAP_MAX_TEXTURE_IMAGE_UNITS:
+ case PIPE_CAP_MAX_COMBINED_SAMPLERS:
/* XXX I'm told this goes up to 16 */
return 8;
case PIPE_CAP_NPOT_TEXTURES:
- /* IN THEORY */
- return 0;
+ /* XXX enable now to get GL2.1 API,
+ * figure out later how to emulate this */
+ return 1;
case PIPE_CAP_TWO_SIDED_STENCIL:
if (r300screen->caps->is_r500) {
return 1;
@@ -88,16 +97,26 @@ static int r300_get_param(struct pipe_screen* pscreen, int param)
return 0;
}
case PIPE_CAP_GLSL:
- if (r300screen->caps->is_r500) {
- return 1;
- } else {
- return 0;
- }
+ /* I'll be frank. This is a lie.
+ *
+ * We don't truly support GLSL on any of this driver's chipsets.
+ * To be fair, no chipset supports the full GLSL specification
+ * to the best of our knowledge, but some of the less esoteric
+ * features are still missing here.
+ *
+ * Rather than cripple ourselves intentionally, I'm going to set
+ * this flag, and as Gallium's interface continues to change, I
+ * hope that this single monolithic GLSL enable can slowly get
+ * split down into many different pieces and the state tracker
+ * will handle fallbacks transparently, like it should.
+ *
+ * ~ C.
+ */
+ return 1;
case PIPE_CAP_ANISOTROPIC_FILTER:
return 1;
case PIPE_CAP_POINT_SPRITE:
- /* IN THEORY */
- return 0;
+ return 1;
case PIPE_CAP_MAX_RENDER_TARGETS:
return 4;
case PIPE_CAP_OCCLUSION_QUERY:
@@ -105,32 +124,13 @@ static int r300_get_param(struct pipe_screen* pscreen, int param)
case PIPE_CAP_TEXTURE_SHADOW_MAP:
return 1;
case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
- if (r300screen->caps->is_r500) {
- /* 13 == 4096x4096 */
- return 13;
- } else {
- /* 12 == 2048x2048 */
- return 12;
- }
case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
- /* So, technically, the limit is the same as above, but some math
- * shows why this is silly. Assuming RGBA, 4cpp, we can see that
- * 4096*4096*4096 = 64.0 GiB exactly, so it's not exactly
- * practical. However, if at some point a game really wants this,
- * then we can remove or raise this limit. */
- if (r300screen->caps->is_r500) {
- /* 9 == 256x256x256 */
- return 9;
- } else {
- /* 8 == 128*128*128 */
- return 8;
- }
case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
if (r300screen->caps->is_r500) {
- /* 13 == 4096x4096 */
+ /* 13 == 4096 */
return 13;
} else {
- /* 12 == 2048x2048 */
+ /* 12 == 2048 */
return 12;
}
case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
@@ -138,13 +138,17 @@ static int r300_get_param(struct pipe_screen* pscreen, int param)
case PIPE_CAP_TEXTURE_MIRROR_REPEAT:
return 1;
case PIPE_CAP_MAX_VERTEX_TEXTURE_UNITS:
- /* XXX guessing (what a terrible guess) */
- return 2;
+ return 0;
case PIPE_CAP_TGSI_CONT_SUPPORTED:
- /* XXX */
return 0;
case PIPE_CAP_BLEND_EQUATION_SEPARATE:
return 1;
+ case PIPE_CAP_SM3:
+ if (r300screen->caps->is_r500) {
+ return 1;
+ } else {
+ return 0;
+ }
default:
debug_printf("r300: Implementation error: Bad param %d\n",
param);
@@ -179,19 +183,22 @@ static float r300_get_paramf(struct pipe_screen* pscreen, int param)
}
}
-static boolean check_tex_2d_format(enum pipe_format format, uint32_t usage,
- boolean is_r500)
+static boolean check_tex_format(enum pipe_format format, uint32_t usage,
+ boolean is_r500)
{
+ uint32_t retval = 0;
+
switch (format) {
/* Supported formats. */
/* Colorbuffer */
case PIPE_FORMAT_A4R4G4B4_UNORM:
case PIPE_FORMAT_R5G6B5_UNORM:
case PIPE_FORMAT_A1R5G5B5_UNORM:
- return usage &
+ retval = usage &
(PIPE_TEXTURE_USAGE_RENDER_TARGET |
PIPE_TEXTURE_USAGE_DISPLAY_TARGET |
PIPE_TEXTURE_USAGE_PRIMARY);
+ break;
/* Texture */
case PIPE_FORMAT_A8R8G8B8_SRGB:
@@ -201,7 +208,10 @@ static boolean check_tex_2d_format(enum pipe_format format, uint32_t usage,
case PIPE_FORMAT_DXT3_RGBA:
case PIPE_FORMAT_DXT5_RGBA:
case PIPE_FORMAT_YCBCR:
- return usage & PIPE_TEXTURE_USAGE_SAMPLER;
+ case PIPE_FORMAT_L8_UNORM:
+ case PIPE_FORMAT_A8L8_UNORM:
+ retval = usage & PIPE_TEXTURE_USAGE_SAMPLER;
+ break;
/* Colorbuffer or texture */
case PIPE_FORMAT_A8R8G8B8_UNORM:
@@ -209,25 +219,32 @@ static boolean check_tex_2d_format(enum pipe_format format, uint32_t usage,
case PIPE_FORMAT_R8G8B8A8_UNORM:
case PIPE_FORMAT_R8G8B8X8_UNORM:
case PIPE_FORMAT_I8_UNORM:
- return usage &
+ retval = usage &
(PIPE_TEXTURE_USAGE_RENDER_TARGET |
PIPE_TEXTURE_USAGE_DISPLAY_TARGET |
PIPE_TEXTURE_USAGE_PRIMARY |
PIPE_TEXTURE_USAGE_SAMPLER);
+ break;
- /* Z buffer */
+ /* Z buffer or texture */
case PIPE_FORMAT_Z16_UNORM:
- return usage & PIPE_TEXTURE_USAGE_DEPTH_STENCIL;
+ retval = usage &
+ (PIPE_TEXTURE_USAGE_DEPTH_STENCIL |
+ PIPE_TEXTURE_USAGE_SAMPLER);
+ break;
+ /* 24bit Z buffer can only be used as a texture on R500. */
+ case PIPE_FORMAT_Z24X8_UNORM:
/* Z buffer with stencil or texture */
case PIPE_FORMAT_Z24S8_UNORM:
- return usage &
+ retval = usage &
(PIPE_TEXTURE_USAGE_DEPTH_STENCIL |
- PIPE_TEXTURE_USAGE_SAMPLER);
+ (is_r500 ? PIPE_TEXTURE_USAGE_SAMPLER : 0));
+ break;
/* Definitely unsupported formats. */
/* Non-usable Z buffer/stencil formats. */
- case PIPE_FORMAT_Z24X8_UNORM:
+ case PIPE_FORMAT_Z32_UNORM:
case PIPE_FORMAT_S8Z24_UNORM:
case PIPE_FORMAT_X8Z24_UNORM:
debug_printf("r300: Note: Got unsupported format: %s in %s\n",
@@ -237,7 +254,6 @@ static boolean check_tex_2d_format(enum pipe_format format, uint32_t usage,
/* XXX These don't even exist
case PIPE_FORMAT_A32R32G32B32:
case PIPE_FORMAT_A16R16G16B16: */
- /* XXX Insert YUV422 packed VYUY and YVYU here */
/* XXX What the deuce is UV88? (r3xx accel page 14)
debug_printf("r300: Warning: Got unimplemented format: %s in %s\n",
pf_name(format), __FUNCTION__);
@@ -261,10 +277,15 @@ static boolean check_tex_2d_format(enum pipe_format format, uint32_t usage,
break;
}
- return FALSE;
+ /* If usage was a mask that contained multiple bits, and not all of them
+ * are supported, this will catch that and return FALSE.
+ * e.g. usage = 2 | 4; retval = 4; (retval >= usage) == FALSE
+ *
+ * This also returns FALSE for any unknown formats.
+ */
+ return (retval >= usage);
}
-/* XXX moar targets */
static boolean r300_is_format_supported(struct pipe_screen* pscreen,
enum pipe_format format,
enum pipe_texture_target target,
@@ -272,15 +293,13 @@ static boolean r300_is_format_supported(struct pipe_screen* pscreen,
unsigned geom_flags)
{
switch (target) {
+ case PIPE_TEXTURE_1D: /* handle 1D textures as 2D ones */
case PIPE_TEXTURE_2D:
- return check_tex_2d_format(format, tex_usage,
- r300_screen(pscreen)->caps->is_r500);
- case PIPE_TEXTURE_1D:
case PIPE_TEXTURE_3D:
case PIPE_TEXTURE_CUBE:
- debug_printf("r300: Implementation error: Unsupported format "
- "target: %d\n", target);
- break;
+ return check_tex_format(format, tex_usage,
+ r300_screen(pscreen)->caps->is_r500);
+
default:
debug_printf("r300: Fatal: This is not a format target: %d\n",
target);
@@ -300,36 +319,25 @@ r300_get_tex_transfer(struct pipe_screen *screen,
{
struct r300_texture *tex = (struct r300_texture *)texture;
struct r300_transfer *trans;
- unsigned offset; /* in bytes */
+ unsigned offset;
- /* XXX Add support for these things */
- if (texture->target == PIPE_TEXTURE_CUBE) {
- debug_printf("PIPE_TEXTURE_CUBE is not yet supported.\n");
- /* offset = tex->image_offset[level][face]; */
- }
- else if (texture->target == PIPE_TEXTURE_3D) {
- debug_printf("PIPE_TEXTURE_3D is not yet supported.\n");
- /* offset = tex->image_offset[level][zslice]; */
- }
- else {
- offset = tex->offset[level];
- assert(face == 0);
- assert(zslice == 0);
- }
+ offset = r300_texture_get_offset(tex, level, zslice, face); /* in bytes */
trans = CALLOC_STRUCT(r300_transfer);
if (trans) {
pipe_texture_reference(&trans->transfer.texture, texture);
- trans->transfer.format = texture->format;
trans->transfer.x = x;
trans->transfer.y = y;
trans->transfer.width = w;
trans->transfer.height = h;
- trans->transfer.block = texture->block;
- trans->transfer.nblocksx = texture->nblocksx[level];
- trans->transfer.nblocksy = texture->nblocksy[level];
trans->transfer.stride = r300_texture_get_stride(tex, level);
trans->transfer.usage = usage;
+
+ /* XXX not sure whether it's required to set these two,
+ the driver doesn't use them */
+ trans->transfer.zslice = zslice;
+ trans->transfer.face = face;
+
trans->offset = offset;
}
return &trans->transfer;
@@ -347,6 +355,7 @@ static void* r300_transfer_map(struct pipe_screen* screen,
{
struct r300_texture* tex = (struct r300_texture*)transfer->texture;
char* map;
+ enum pipe_format format = tex->tex.format;
map = pipe_buffer_map(screen, tex->buffer,
pipe_transfer_buffer_flags(transfer));
@@ -356,8 +365,8 @@ static void* r300_transfer_map(struct pipe_screen* screen,
}
return map + r300_transfer(transfer)->offset +
- transfer->y / transfer->block.height * transfer->stride +
- transfer->x / transfer->block.width * transfer->block.size;
+ transfer->y / util_format_get_blockheight(format) * transfer->stride +
+ transfer->x / util_format_get_blockwidth(format) * util_format_get_blocksize(format);
}
static void r300_transfer_unmap(struct pipe_screen* screen,
@@ -375,7 +384,7 @@ static void r300_destroy_screen(struct pipe_screen* pscreen)
FREE(r300screen);
}
-struct pipe_screen* r300_create_screen(struct r300_winsys* r300_winsys)
+struct pipe_screen* r300_create_screen(struct radeon_winsys* radeon_winsys)
{
struct r300_screen* r300screen = CALLOC_STRUCT(r300_screen);
struct r300_capabilities* caps = CALLOC_STRUCT(r300_capabilities);
@@ -383,13 +392,14 @@ struct pipe_screen* r300_create_screen(struct r300_winsys* r300_winsys)
if (!r300screen || !caps)
return NULL;
- caps->pci_id = r300_winsys->pci_id;
- caps->num_frag_pipes = r300_winsys->gb_pipes;
+ caps->pci_id = radeon_winsys->pci_id;
+ caps->num_frag_pipes = radeon_winsys->gb_pipes;
+ caps->num_z_pipes = radeon_winsys->z_pipes;
r300_parse_chipset(caps);
r300screen->caps = caps;
- r300screen->screen.winsys = (struct pipe_winsys*)r300_winsys;
+ r300screen->screen.winsys = (struct pipe_winsys*)radeon_winsys;
r300screen->screen.destroy = r300_destroy_screen;
r300screen->screen.get_name = r300_get_name;
r300screen->screen.get_vendor = r300_get_vendor;
diff --git a/src/gallium/drivers/r300/r300_screen.h b/src/gallium/drivers/r300/r300_screen.h
index 2a0e41fbc3b..2217988addd 100644
--- a/src/gallium/drivers/r300/r300_screen.h
+++ b/src/gallium/drivers/r300/r300_screen.h
@@ -23,14 +23,11 @@
#ifndef R300_SCREEN_H
#define R300_SCREEN_H
-#include "pipe/p_inlines.h"
#include "pipe/p_screen.h"
-#include "util/u_memory.h"
-#include "util/u_simple_screen.h"
#include "r300_chipset.h"
-#include "r300_texture.h"
-#include "r300_winsys.h"
+
+struct radeon_winsys;
struct r300_screen {
/* Parent class */
@@ -61,6 +58,6 @@ r300_transfer(struct pipe_transfer* transfer)
}
/* Creates a new r300 screen. */
-struct pipe_screen* r300_create_screen(struct r300_winsys* r300_winsys);
+struct pipe_screen* r300_create_screen(struct radeon_winsys* radeon_winsys);
#endif /* R300_SCREEN_H */
diff --git a/src/gallium/drivers/r300/r300_shader_inlines.h b/src/gallium/drivers/r300/r300_shader_inlines.h
deleted file mode 100644
index a04f45b03e2..00000000000
--- a/src/gallium/drivers/r300/r300_shader_inlines.h
+++ /dev/null
@@ -1,47 +0,0 @@
-/*
- * Copyright 2009 Corbin Simpson <[email protected]>
- * Joakim Sindholt <[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_SHADER_INLINES_H
-#define R300_SHADER_INLINES_H
-
-/* TGSI constants. TGSI is like XML: If it can't solve your problems, you're
- * not using enough of it. */
-static const struct tgsi_full_src_register r300_constant_zero = {
- .SrcRegister.Extended = TRUE,
- .SrcRegister.File = TGSI_FILE_NULL,
- .SrcRegisterExtSwz.ExtSwizzleX = TGSI_EXTSWIZZLE_ZERO,
- .SrcRegisterExtSwz.ExtSwizzleY = TGSI_EXTSWIZZLE_ZERO,
- .SrcRegisterExtSwz.ExtSwizzleZ = TGSI_EXTSWIZZLE_ZERO,
- .SrcRegisterExtSwz.ExtSwizzleW = TGSI_EXTSWIZZLE_ZERO,
-};
-
-static const struct tgsi_full_src_register r300_constant_one = {
- .SrcRegister.Extended = TRUE,
- .SrcRegister.File = TGSI_FILE_NULL,
- .SrcRegisterExtSwz.ExtSwizzleX = TGSI_EXTSWIZZLE_ONE,
- .SrcRegisterExtSwz.ExtSwizzleY = TGSI_EXTSWIZZLE_ONE,
- .SrcRegisterExtSwz.ExtSwizzleZ = TGSI_EXTSWIZZLE_ONE,
- .SrcRegisterExtSwz.ExtSwizzleW = TGSI_EXTSWIZZLE_ONE,
-};
-
-#endif /* R300_SHADER_INLINES_H */
diff --git a/src/gallium/drivers/r300/r300_shader_semantics.h b/src/gallium/drivers/r300/r300_shader_semantics.h
new file mode 100644
index 00000000000..6796841b29b
--- /dev/null
+++ b/src/gallium/drivers/r300/r300_shader_semantics.h
@@ -0,0 +1,66 @@
+/*
+ * 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
+ * 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_SHADER_SEMANTICS_H
+#define R300_SHADER_SEMANTICS_H
+
+#define ATTR_UNUSED (-1)
+#define ATTR_COLOR_COUNT 2
+#define ATTR_GENERIC_COUNT 16
+
+/* This structure contains information about what attributes are written by VS
+ * or read by FS. (but not both) It's much easier to work with than
+ * tgsi_shader_info.
+ *
+ * The variables contain indices to tgsi_shader_info semantics and those
+ * indices are nothing else than input/output register numbers. */
+struct r300_shader_semantics {
+ int pos;
+ int psize;
+ int color[ATTR_COLOR_COUNT];
+ int bcolor[ATTR_COLOR_COUNT];
+ int generic[ATTR_GENERIC_COUNT];
+ int fog;
+ int wpos;
+};
+
+static INLINE void r300_shader_semantics_reset(
+ struct r300_shader_semantics* info)
+{
+ int i;
+
+ info->pos = ATTR_UNUSED;
+ info->psize = ATTR_UNUSED;
+ info->fog = ATTR_UNUSED;
+ info->wpos = ATTR_UNUSED;
+
+ for (i = 0; i < ATTR_COLOR_COUNT; i++) {
+ info->color[i] = ATTR_UNUSED;
+ info->bcolor[i] = ATTR_UNUSED;
+ }
+
+ for (i = 0; i < ATTR_GENERIC_COUNT; i++) {
+ info->generic[i] = ATTR_UNUSED;
+ }
+}
+
+#endif
diff --git a/src/gallium/drivers/r300/r300_state.c b/src/gallium/drivers/r300/r300_state.c
index 88cb9af6fb7..a145a7f18a5 100644
--- a/src/gallium/drivers/r300/r300_state.c
+++ b/src/gallium/drivers/r300/r300_state.c
@@ -1,5 +1,6 @@
/*
* Copyright 2008 Corbin Simpson <[email protected]>
+ * 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"),
@@ -20,8 +21,10 @@
* OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
* USE OR OTHER DEALINGS IN THE SOFTWARE. */
-#include "util/u_debug.h"
+#include "draw/draw_context.h"
+
#include "util/u_math.h"
+#include "util/u_memory.h"
#include "util/u_pack_color.h"
#include "tgsi/tgsi_parse.h"
@@ -31,6 +34,7 @@
#include "r300_context.h"
#include "r300_reg.h"
+#include "r300_screen.h"
#include "r300_state_inlines.h"
#include "r300_fs.h"
#include "r300_vs.h"
@@ -38,6 +42,120 @@
/* r300_state: Functions used to intialize state context by translating
* Gallium state objects into semi-native r300 state objects. */
+static boolean blend_discard_if_src_alpha_0(unsigned srcRGB, unsigned srcA,
+ unsigned dstRGB, unsigned dstA)
+{
+ /* If the blend equation is ADD or REVERSE_SUBTRACT,
+ * SRC_ALPHA == 0, and the following state is set, the colorbuffer
+ * will not be changed.
+ * Notice that the dst factors are the src factors inverted. */
+ return (srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
+ srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE ||
+ srcRGB == PIPE_BLENDFACTOR_ZERO) &&
+ (srcA == PIPE_BLENDFACTOR_SRC_COLOR ||
+ srcA == PIPE_BLENDFACTOR_SRC_ALPHA ||
+ srcA == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE ||
+ srcA == PIPE_BLENDFACTOR_ZERO) &&
+ (dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
+ dstRGB == PIPE_BLENDFACTOR_ONE) &&
+ (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
+ dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
+ dstA == PIPE_BLENDFACTOR_ONE);
+}
+
+static boolean blend_discard_if_src_alpha_1(unsigned srcRGB, unsigned srcA,
+ unsigned dstRGB, unsigned dstA)
+{
+ /* If the blend equation is ADD or REVERSE_SUBTRACT,
+ * SRC_ALPHA == 1, and the following state is set, the colorbuffer
+ * will not be changed.
+ * Notice that the dst factors are the src factors inverted. */
+ return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
+ srcRGB == PIPE_BLENDFACTOR_ZERO) &&
+ (srcA == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
+ srcA == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
+ srcA == PIPE_BLENDFACTOR_ZERO) &&
+ (dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
+ dstRGB == PIPE_BLENDFACTOR_ONE) &&
+ (dstA == PIPE_BLENDFACTOR_SRC_COLOR ||
+ dstA == PIPE_BLENDFACTOR_SRC_ALPHA ||
+ dstA == PIPE_BLENDFACTOR_ONE);
+}
+
+static boolean blend_discard_if_src_color_0(unsigned srcRGB, unsigned srcA,
+ unsigned dstRGB, unsigned dstA)
+{
+ /* If the blend equation is ADD or REVERSE_SUBTRACT,
+ * SRC_COLOR == (0,0,0), and the following state is set, the colorbuffer
+ * will not be changed.
+ * Notice that the dst factors are the src factors inverted. */
+ return (srcRGB == PIPE_BLENDFACTOR_SRC_COLOR ||
+ srcRGB == PIPE_BLENDFACTOR_ZERO) &&
+ (srcA == PIPE_BLENDFACTOR_ZERO) &&
+ (dstRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
+ dstRGB == PIPE_BLENDFACTOR_ONE) &&
+ (dstA == PIPE_BLENDFACTOR_ONE);
+}
+
+static boolean blend_discard_if_src_color_1(unsigned srcRGB, unsigned srcA,
+ unsigned dstRGB, unsigned dstA)
+{
+ /* If the blend equation is ADD or REVERSE_SUBTRACT,
+ * SRC_COLOR == (1,1,1), and the following state is set, the colorbuffer
+ * will not be changed.
+ * Notice that the dst factors are the src factors inverted. */
+ return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
+ srcRGB == PIPE_BLENDFACTOR_ZERO) &&
+ (srcA == PIPE_BLENDFACTOR_ZERO) &&
+ (dstRGB == PIPE_BLENDFACTOR_SRC_COLOR ||
+ dstRGB == PIPE_BLENDFACTOR_ONE) &&
+ (dstA == PIPE_BLENDFACTOR_ONE);
+}
+
+static boolean blend_discard_if_src_alpha_color_0(unsigned srcRGB, unsigned srcA,
+ unsigned dstRGB, unsigned dstA)
+{
+ /* If the blend equation is ADD or REVERSE_SUBTRACT,
+ * SRC_ALPHA_COLOR == (0,0,0,0), and the following state is set,
+ * the colorbuffer will not be changed.
+ * Notice that the dst factors are the src factors inverted. */
+ return (srcRGB == PIPE_BLENDFACTOR_SRC_COLOR ||
+ srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
+ srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE ||
+ srcRGB == PIPE_BLENDFACTOR_ZERO) &&
+ (srcA == PIPE_BLENDFACTOR_SRC_COLOR ||
+ srcA == PIPE_BLENDFACTOR_SRC_ALPHA ||
+ srcA == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE ||
+ srcA == PIPE_BLENDFACTOR_ZERO) &&
+ (dstRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
+ dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
+ dstRGB == PIPE_BLENDFACTOR_ONE) &&
+ (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
+ dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
+ dstA == PIPE_BLENDFACTOR_ONE);
+}
+
+static boolean blend_discard_if_src_alpha_color_1(unsigned srcRGB, unsigned srcA,
+ unsigned dstRGB, unsigned dstA)
+{
+ /* If the blend equation is ADD or REVERSE_SUBTRACT,
+ * SRC_ALPHA_COLOR == (1,1,1,1), and the following state is set,
+ * the colorbuffer will not be changed.
+ * Notice that the dst factors are the src factors inverted. */
+ return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
+ srcRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
+ srcRGB == PIPE_BLENDFACTOR_ZERO) &&
+ (srcA == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
+ srcA == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
+ srcA == PIPE_BLENDFACTOR_ZERO) &&
+ (dstRGB == PIPE_BLENDFACTOR_SRC_COLOR ||
+ dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
+ dstRGB == PIPE_BLENDFACTOR_ONE) &&
+ (dstA == PIPE_BLENDFACTOR_SRC_COLOR ||
+ dstA == PIPE_BLENDFACTOR_SRC_ALPHA ||
+ dstA == PIPE_BLENDFACTOR_ONE);
+}
+
/* Create a new blend state based on the CSO blend state.
*
* This encompasses alpha blending, logic/raster ops, and blend dithering. */
@@ -46,23 +164,123 @@ static void* r300_create_blend_state(struct pipe_context* pipe,
{
struct r300_blend_state* blend = CALLOC_STRUCT(r300_blend_state);
- if (state->blend_enable) {
- /* XXX for now, always do separate alpha...
- * is it faster to do it with one reg? */
+ if (state->blend_enable)
+ {
+ unsigned eqRGB = state->rgb_func;
+ unsigned srcRGB = state->rgb_src_factor;
+ unsigned dstRGB = state->rgb_dst_factor;
+
+ unsigned eqA = state->alpha_func;
+ unsigned srcA = state->alpha_src_factor;
+ unsigned dstA = state->alpha_dst_factor;
+
+ /* 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 |
- R300_SEPARATE_ALPHA_ENABLE |
- R300_READ_ENABLE |
- r300_translate_blend_function(state->rgb_func) |
- (r300_translate_blend_factor(state->rgb_src_factor) <<
- R300_SRC_BLEND_SHIFT) |
- (r300_translate_blend_factor(state->rgb_dst_factor) <<
- R300_DST_BLEND_SHIFT);
- blend->alpha_blend_control =
- r300_translate_blend_function(state->alpha_func) |
- (r300_translate_blend_factor(state->alpha_src_factor) <<
- R300_SRC_BLEND_SHIFT) |
- (r300_translate_blend_factor(state->alpha_dst_factor) <<
- R300_DST_BLEND_SHIFT);
+ r300_translate_blend_function(eqRGB) |
+ ( r300_translate_blend_factor(srcRGB) << R300_SRC_BLEND_SHIFT) |
+ ( r300_translate_blend_factor(dstRGB) << R300_DST_BLEND_SHIFT);
+
+ /* Optimization: some operations do not require the destination color.
+ *
+ * When SRC_ALPHA_SATURATE is used, colorbuffer reads must be enabled,
+ * otherwise blending gives incorrect results. It seems to be
+ * a hardware bug. */
+ if (eqRGB == PIPE_BLEND_MIN || eqA == PIPE_BLEND_MIN ||
+ eqRGB == PIPE_BLEND_MAX || eqA == PIPE_BLEND_MAX ||
+ dstRGB != PIPE_BLENDFACTOR_ZERO ||
+ dstA != PIPE_BLENDFACTOR_ZERO ||
+ srcRGB == PIPE_BLENDFACTOR_DST_COLOR ||
+ srcRGB == PIPE_BLENDFACTOR_DST_ALPHA ||
+ srcRGB == PIPE_BLENDFACTOR_INV_DST_COLOR ||
+ srcRGB == PIPE_BLENDFACTOR_INV_DST_ALPHA ||
+ srcA == PIPE_BLENDFACTOR_DST_COLOR ||
+ srcA == PIPE_BLENDFACTOR_DST_ALPHA ||
+ srcA == PIPE_BLENDFACTOR_INV_DST_COLOR ||
+ srcA == PIPE_BLENDFACTOR_INV_DST_ALPHA ||
+ srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE) {
+ /* Enable reading from the colorbuffer. */
+ blend->blend_control |= R300_READ_ENABLE;
+
+ if (r300_screen(r300_context(pipe)->context.screen)->caps->is_r500) {
+ /* Optimization: Depending on incoming pixels, we can
+ * conditionally disable the reading in hardware... */
+ if (eqRGB != PIPE_BLEND_MIN && eqA != PIPE_BLEND_MIN &&
+ eqRGB != PIPE_BLEND_MAX && eqA != PIPE_BLEND_MAX) {
+ /* Disable reading if SRC_ALPHA == 0. */
+ if ((dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
+ dstRGB == PIPE_BLENDFACTOR_ZERO) &&
+ (dstA == PIPE_BLENDFACTOR_SRC_COLOR ||
+ dstA == PIPE_BLENDFACTOR_SRC_ALPHA ||
+ dstA == PIPE_BLENDFACTOR_ZERO)) {
+ blend->blend_control |= R500_SRC_ALPHA_0_NO_READ;
+ }
+
+ /* Disable reading if SRC_ALPHA == 1. */
+ if ((dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
+ dstRGB == PIPE_BLENDFACTOR_ZERO) &&
+ (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;
+ }
+ }
+ }
+ }
+
+ /* Optimization: discard pixels which don't change the colorbuffer.
+ *
+ * The code below is non-trivial and some math is involved.
+ *
+ * Discarding pixels must be disabled when FP16 AA is enabled.
+ * This is a hardware bug. Also, this implementation wouldn't work
+ * with FP blending enabled and equation clamping disabled.
+ *
+ * Equations other than ADD are rarely used and therefore won't be
+ * optimized. */
+ if ((eqRGB == PIPE_BLEND_ADD || eqRGB == PIPE_BLEND_REVERSE_SUBTRACT) &&
+ (eqA == PIPE_BLEND_ADD || eqA == PIPE_BLEND_REVERSE_SUBTRACT)) {
+ /* ADD: X+Y
+ * REVERSE_SUBTRACT: Y-X
+ *
+ * The idea is:
+ * If X = src*srcFactor = 0 and Y = dst*dstFactor = 1,
+ * then CB will not be changed.
+ *
+ * Given the srcFactor and dstFactor variables, we can derive
+ * what src and dst should be equal to and discard appropriate
+ * pixels.
+ */
+ if (blend_discard_if_src_alpha_0(srcRGB, srcA, dstRGB, dstA)) {
+ blend->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;
+ } else if (blend_discard_if_src_color_0(srcRGB, srcA,
+ dstRGB, dstA)) {
+ blend->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;
+ } else if (blend_discard_if_src_alpha_color_0(srcRGB, srcA,
+ dstRGB, dstA)) {
+ blend->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 |=
+ 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 =
+ r300_translate_blend_function(eqA) |
+ (r300_translate_blend_factor(srcA) << R300_SRC_BLEND_SHIFT) |
+ (r300_translate_blend_factor(dstA) << R300_DST_BLEND_SHIFT);
+ }
}
/* PIPE_LOGICOP_* don't need to be translated, fortunately. */
@@ -71,6 +289,20 @@ static void* r300_create_blend_state(struct pipe_context* pipe,
(state->logicop_func) << R300_RB3D_ROPCNTL_ROP_SHIFT;
}
+ /* Color Channel Mask */
+ if (state->colormask & PIPE_MASK_R) {
+ blend->color_channel_mask |= RB3D_COLOR_CHANNEL_MASK_RED_MASK0;
+ }
+ if (state->colormask & PIPE_MASK_G) {
+ blend->color_channel_mask |= RB3D_COLOR_CHANNEL_MASK_GREEN_MASK0;
+ }
+ if (state->colormask & PIPE_MASK_B) {
+ blend->color_channel_mask |= RB3D_COLOR_CHANNEL_MASK_BLUE_MASK0;
+ }
+ if (state->colormask & PIPE_MASK_A) {
+ blend->color_channel_mask |= RB3D_COLOR_CHANNEL_MASK_ALPHA_MASK0;
+ }
+
if (state->dither) {
blend->dither = R300_RB3D_DITHER_CTL_DITHER_MODE_LUT |
R300_RB3D_DITHER_CTL_ALPHA_DITHER_MODE_LUT;
@@ -96,25 +328,30 @@ static void r300_delete_blend_state(struct pipe_context* pipe,
FREE(state);
}
+/* Convert float to 10bit integer */
+static unsigned float_to_fixed10(float f)
+{
+ return CLAMP((unsigned)(f * 1023.9f), 0, 1023);
+}
+
/* Set blend color.
* Setup both R300 and R500 registers, figure out later which one to write. */
static void r300_set_blend_color(struct pipe_context* pipe,
const struct pipe_blend_color* color)
{
struct r300_context* r300 = r300_context(pipe);
- ubyte ur, ug, ub, ua;
-
- ur = float_to_ubyte(color->color[0]);
- ug = float_to_ubyte(color->color[1]);
- ub = float_to_ubyte(color->color[2]);
- ua = float_to_ubyte(color->color[3]);
+ union util_color uc;
- util_pack_color(color->color, PIPE_FORMAT_A8R8G8B8_UNORM,
- &r300->blend_color_state->blend_color);
+ util_pack_color(color->color, PIPE_FORMAT_A8R8G8B8_UNORM, &uc);
+ r300->blend_color_state->blend_color = uc.ui;
- /* XXX this is wrong */
- r300->blend_color_state->blend_color_red_alpha = ur | (ua << 16);
- r300->blend_color_state->blend_color_green_blue = ub | (ug << 16);
+ /* XXX if FP16 blending is enabled, we should use the FP16 format */
+ r300->blend_color_state->blend_color_red_alpha =
+ float_to_fixed10(color->color[0]) |
+ (float_to_fixed10(color->color[3]) << 16);
+ r300->blend_color_state->blend_color_green_blue =
+ float_to_fixed10(color->color[2]) |
+ (float_to_fixed10(color->color[1]) << 16);
r300->dirty_state |= R300_NEW_BLEND_COLOR;
}
@@ -133,31 +370,6 @@ static void r300_set_clip_state(struct pipe_context* pipe,
}
}
-static void
- r300_set_constant_buffer(struct pipe_context* pipe,
- uint shader, uint index,
- const struct pipe_constant_buffer* buffer)
-{
- struct r300_context* r300 = r300_context(pipe);
-
- /* This entire chunk of code seems ever-so-slightly baked.
- * It's as if I've got pipe_buffer* matryoshkas... */
- if (buffer && buffer->buffer && buffer->buffer->size) {
- void* map = pipe->winsys->buffer_map(pipe->winsys, buffer->buffer,
- PIPE_BUFFER_USAGE_CPU_READ);
- memcpy(r300->shader_constants[shader].constants,
- map, buffer->buffer->size);
- pipe->winsys->buffer_unmap(pipe->winsys, buffer->buffer);
-
- r300->shader_constants[shader].count =
- buffer->buffer->size / (sizeof(float) * 4);
- } else {
- r300->shader_constants[shader].count = 0;
- }
-
- r300->dirty_state |= R300_NEW_CONSTANTS;
-}
-
/* 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.
@@ -167,6 +379,8 @@ static void*
r300_create_dsa_state(struct pipe_context* pipe,
const struct pipe_depth_stencil_alpha_state* state)
{
+ struct r300_capabilities *caps =
+ r300_screen(r300_context(pipe)->context.screen)->caps;
struct r300_dsa_state* dsa = CALLOC_STRUCT(r300_dsa_state);
/* Depth test setup. */
@@ -211,9 +425,16 @@ static void*
(r300_translate_stencil_op(state->stencil[1].zfail_op) <<
R300_S_BACK_ZFAIL_OP_SHIFT);
- dsa->stencil_ref_bf = (state->stencil[1].ref_value) |
- (state->stencil[1].valuemask << R300_STENCILMASK_SHIFT) |
- (state->stencil[1].writemask << R300_STENCILWRITEMASK_SHIFT);
+ /* XXX it seems r3xx doesn't support STENCILREFMASK_BF */
+ if (caps->is_r500)
+ {
+ dsa->z_buffer_control |= R500_STENCIL_REFMASK_FRONT_BACK;
+ dsa->stencil_ref_bf = (state->stencil[1].ref_value) |
+ (state->stencil[1].valuemask <<
+ R300_STENCILMASK_SHIFT) |
+ (state->stencil[1].writemask <<
+ R300_STENCILWRITEMASK_SHIFT);
+ }
}
}
@@ -222,11 +443,13 @@ static void*
dsa->alpha_function =
r300_translate_alpha_function(state->alpha.func) |
R300_FG_ALPHA_FUNC_ENABLE;
- dsa->alpha_reference = CLAMP(state->alpha.ref_value * 1023.0f,
- 0, 1023);
- } else {
- /* XXX need to fix this to be dynamically set
- dsa->z_buffer_top = R300_ZTOP_ENABLE; */
+
+ /* XXX figure out why emitting 10bit alpha ref causes CS to dump */
+ /* always use 8bit alpha ref */
+ dsa->alpha_function |= float_to_ubyte(state->alpha.ref_value);
+
+ if (caps->is_r500)
+ dsa->alpha_function |= R500_FG_ALPHA_FUNC_8BIT;
}
return (void*)dsa;
@@ -249,11 +472,29 @@ static void r300_delete_dsa_state(struct pipe_context* pipe,
FREE(state);
}
-static void r300_set_edgeflags(struct pipe_context* pipe,
- const unsigned* bitfield)
+static void r300_set_scissor_regs(const struct pipe_scissor_state* state,
+ struct r300_scissor_regs *scissor,
+ boolean is_r500)
{
- /* XXX you know it's bad when i915 has this blank too */
- /* XXX and even worse, I have no idea WTF the bitfield is */
+ if (is_r500) {
+ scissor->top_left =
+ (state->minx << R300_SCISSORS_X_SHIFT) |
+ (state->miny << R300_SCISSORS_Y_SHIFT);
+ scissor->bottom_right =
+ ((state->maxx - 1) << R300_SCISSORS_X_SHIFT) |
+ ((state->maxy - 1) << R300_SCISSORS_Y_SHIFT);
+ } else {
+ /* Offset of 1440 in non-R500 chipsets. */
+ scissor->top_left =
+ ((state->minx + 1440) << R300_SCISSORS_X_SHIFT) |
+ ((state->miny + 1440) << R300_SCISSORS_Y_SHIFT);
+ scissor->bottom_right =
+ (((state->maxx - 1) + 1440) << R300_SCISSORS_X_SHIFT) |
+ (((state->maxy - 1) + 1440) << R300_SCISSORS_Y_SHIFT);
+ }
+
+ scissor->empty_area = state->minx >= state->maxx ||
+ state->miny >= state->maxy;
}
static void
@@ -261,12 +502,27 @@ static void
const struct pipe_framebuffer_state* state)
{
struct r300_context* r300 = r300_context(pipe);
+ struct pipe_scissor_state scissor;
- draw_flush(r300->draw);
+ if (r300->draw) {
+ draw_flush(r300->draw);
+ }
r300->framebuffer_state = *state;
+ scissor.minx = scissor.miny = 0;
+ scissor.maxx = state->width;
+ scissor.maxy = state->height;
+ r300_set_scissor_regs(&scissor, &r300->scissor_state->framebuffer,
+ r300_screen(r300->context.screen)->caps->is_r500);
+
+ /* Don't rely on the order of states being set for the first time. */
+ if (!r300->rs_state || !r300->rs_state->rs.scissor) {
+ r300->dirty_state |= R300_NEW_SCISSOR;
+ }
r300->dirty_state |= R300_NEW_FRAMEBUFFERS;
+ r300->dirty_state |= R300_NEW_BLEND;
+ r300->dirty_state |= R300_NEW_DSA;
}
/* Create fragment shader state. */
@@ -282,6 +538,7 @@ static void* r300_create_fs_state(struct pipe_context* pipe,
fs->state.tokens = tgsi_dup_tokens(shader->tokens);
tgsi_scan_shader(shader->tokens, &fs->info);
+ r300_shader_read_fs_inputs(&fs->info, &fs->inputs);
return (void*)fs;
}
@@ -295,21 +552,31 @@ static void r300_bind_fs_state(struct pipe_context* pipe, void* shader)
if (fs == NULL) {
r300->fs = NULL;
return;
- } else if (!fs->translated) {
- r300_translate_fragment_shader(r300, fs);
}
r300->fs = fs;
+ r300_pick_fragment_shader(r300);
- r300->dirty_state |= R300_NEW_FRAGMENT_SHADER;
+ if (r300->vs && r300_vertex_shader_setup_wpos(r300)) {
+ r300->dirty_state |= R300_NEW_VERTEX_FORMAT;
+ }
+
+ r300->dirty_state |= R300_NEW_FRAGMENT_SHADER | R300_NEW_FRAGMENT_SHADER_CONSTANTS;
}
/* Delete fragment shader state. */
static void r300_delete_fs_state(struct pipe_context* pipe, void* shader)
{
struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader;
- rc_constants_destroy(&fs->code.constants);
- FREE(fs->state.tokens);
+ struct r300_fragment_shader_code *tmp, *ptr = fs->first;
+
+ while (ptr) {
+ tmp = ptr;
+ ptr = ptr->next;
+ rc_constants_destroy(&tmp->code.constants);
+ FREE(tmp);
+ }
+ FREE((void*)fs->state.tokens);
FREE(shader);
}
@@ -347,8 +614,6 @@ static void* r300_create_rs_state(struct pipe_context* pipe,
if (state->bypass_vs_clip_and_viewport ||
!r300_screen(pipe->screen)->caps->has_tcl) {
rs->vap_control_status |= R300_VAP_TCL_BYPASS;
- } else {
- rs->rs.bypass_vs_clip_and_viewport = TRUE;
}
rs->point_size = pack_float_16_6x(state->point_size) |
@@ -363,25 +628,52 @@ static void* r300_create_rs_state(struct pipe_context* pipe,
rs->line_control = pack_float_16_6x(state->line_width) |
R300_GA_LINE_CNTL_END_TYPE_COMP;
+ /* XXX I think there is something wrong with the polygon mode,
+ * XXX re-test when r300g is in a better shape */
+
+ /* 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;
+ }
+
/* 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;
}
+
+ /* 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);
+ }
}
if (state->front_winding & state->cull_mode) {
rs->cull_mode |= R300_CULL_FRONT;
@@ -412,10 +704,6 @@ static void* r300_create_rs_state(struct pipe_context* pipe,
rs->color_control = R300_SHADE_MODEL_SMOOTH;
}
- if (!state->flatshade_first) {
- rs->color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST;
- }
-
return (void*)rs;
}
@@ -425,14 +713,20 @@ 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;
- draw_flush(r300->draw);
- draw_set_rasterizer_state(r300->draw, &rs->rs);
+ if (r300->draw) {
+ draw_flush(r300->draw);
+ draw_set_rasterizer_state(r300->draw, &rs->rs);
+ }
r300->rs_state = rs;
+ /* XXX Clean these up when we move to atom emits */
r300->dirty_state |= R300_NEW_RASTERIZER;
r300->dirty_state |= R300_NEW_RS_BLOCK;
r300->dirty_state |= R300_NEW_SCISSOR;
r300->dirty_state |= R300_NEW_VIEWPORT;
+ if (r300->fs && r300->fs->inputs.wpos != ATTR_UNUSED) {
+ r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS;
+ }
}
/* Free rasterizer state. */
@@ -448,6 +742,9 @@ static void*
struct r300_context* r300 = r300_context(pipe);
struct r300_sampler_state* sampler = CALLOC_STRUCT(r300_sampler_state);
int lod_bias;
+ union util_color uc;
+
+ sampler->state = *state;
sampler->filter0 |=
(r300_translate_wrap(state->wrap_s) << R300_TX_WRAP_S_SHIFT) |
@@ -456,7 +753,13 @@ static void*
sampler->filter0 |= r300_translate_tex_filters(state->min_img_filter,
state->mag_img_filter,
- state->min_mip_filter);
+ state->min_mip_filter,
+ state->max_anisotropy > 1.0);
+
+ /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */
+ /* We must pass these to the emit function to clamp them properly. */
+ sampler->min_lod = MAX2((unsigned)state->min_lod, 0);
+ sampler->max_lod = MAX2((unsigned)ceilf(state->max_lod), 0);
lod_bias = CLAMP((int)(state->lod_bias * 32), -(1 << 9), (1 << 9) - 1);
@@ -464,8 +767,8 @@ static void*
sampler->filter1 |= r300_anisotropy(state->max_anisotropy);
- util_pack_color(state->border_color, PIPE_FORMAT_A8R8G8B8_UNORM,
- &sampler->border_color);
+ util_pack_color(state->border_color, PIPE_FORMAT_A8R8G8B8_UNORM, &uc);
+ sampler->border_color = uc.ui;
/* R500-specific fixups and optimizations */
if (r300_screen(r300->context.screen)->caps->is_r500) {
@@ -494,6 +797,20 @@ static void r300_bind_sampler_states(struct pipe_context* pipe,
}
r300->sampler_count = count;
+
+ /* Pick a fragment shader based on the texture compare state. */
+ if (r300->fs && (r300->dirty_state & R300_ANY_NEW_SAMPLERS)) {
+ if (r300_pick_fragment_shader(r300)) {
+ r300->dirty_state |= R300_NEW_FRAGMENT_SHADER |
+ R300_NEW_FRAGMENT_SHADER_CONSTANTS;
+ }
+ }
+}
+
+static void r300_lacks_vertex_textures(struct pipe_context* pipe,
+ unsigned count,
+ void** states)
+{
}
static void r300_delete_sampler_state(struct pipe_context* pipe, void* state)
@@ -506,18 +823,26 @@ static void r300_set_sampler_textures(struct pipe_context* pipe,
struct pipe_texture** texture)
{
struct r300_context* r300 = r300_context(pipe);
+ boolean is_r500 = r300_screen(r300->context.screen)->caps->is_r500;
int i;
/* XXX magic num */
if (count > 8) {
return;
}
-
+
for (i = 0; i < count; i++) {
if (r300->textures[i] != (struct r300_texture*)texture[i]) {
pipe_texture_reference((struct pipe_texture**)&r300->textures[i],
texture[i]);
r300->dirty_state |= (R300_NEW_TEXTURE << i);
+
+ /* R300-specific - set the texrect factor in a fragment shader */
+ if (!is_r500 && r300->textures[i]->is_npot) {
+ /* XXX It would be nice to re-emit just 1 constant,
+ * XXX not all of them */
+ r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS;
+ }
}
}
@@ -537,24 +862,13 @@ static void r300_set_scissor_state(struct pipe_context* pipe,
{
struct r300_context* r300 = r300_context(pipe);
- if (r300_screen(r300->context.screen)->caps->is_r500) {
- r300->scissor_state->scissor_top_left =
- (state->minx << R300_SCISSORS_X_SHIFT) |
- (state->miny << R300_SCISSORS_Y_SHIFT);
- r300->scissor_state->scissor_bottom_right =
- ((state->maxx - 1) << R300_SCISSORS_X_SHIFT) |
- ((state->maxy - 1) << R300_SCISSORS_Y_SHIFT);
- } else {
- /* Offset of 1440 in non-R500 chipsets. */
- r300->scissor_state->scissor_top_left =
- ((state->minx + 1440) << R300_SCISSORS_X_SHIFT) |
- ((state->miny + 1440) << R300_SCISSORS_Y_SHIFT);
- r300->scissor_state->scissor_bottom_right =
- (((state->maxx - 1) + 1440) << R300_SCISSORS_X_SHIFT) |
- (((state->maxy - 1) + 1440) << R300_SCISSORS_Y_SHIFT);
- }
+ r300_set_scissor_regs(state, &r300->scissor_state->scissor,
+ r300_screen(r300->context.screen)->caps->is_r500);
- r300->dirty_state |= R300_NEW_SCISSOR;
+ /* Don't rely on the order of states being set for the first time. */
+ if (!r300->rs_state || r300->rs_state->rs.scissor) {
+ r300->dirty_state |= R300_NEW_SCISSOR;
+ }
}
static void r300_set_viewport_state(struct pipe_context* pipe,
@@ -566,17 +880,14 @@ static void r300_set_viewport_state(struct pipe_context* pipe,
r300->viewport_state->vte_control = R300_VTX_W0_FMT;
if (state->scale[0] != 1.0f) {
- assert(state->scale[0] != 0.0f);
r300->viewport_state->xscale = state->scale[0];
r300->viewport_state->vte_control |= R300_VPORT_X_SCALE_ENA;
}
if (state->scale[1] != 1.0f) {
- assert(state->scale[1] != 0.0f);
r300->viewport_state->yscale = state->scale[1];
r300->viewport_state->vte_control |= R300_VPORT_Y_SCALE_ENA;
}
if (state->scale[2] != 1.0f) {
- assert(state->scale[2] != 0.0f);
r300->viewport_state->zscale = state->scale[2];
r300->viewport_state->vte_control |= R300_VPORT_Z_SCALE_ENA;
}
@@ -594,6 +905,9 @@ static void r300_set_viewport_state(struct pipe_context* pipe,
}
r300->dirty_state |= R300_NEW_VIEWPORT;
+ if (r300->fs && r300->fs->inputs.wpos != ATTR_UNUSED) {
+ r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS;
+ }
}
static void r300_set_vertex_buffers(struct pipe_context* pipe,
@@ -602,13 +916,16 @@ static void r300_set_vertex_buffers(struct pipe_context* pipe,
{
struct r300_context* r300 = r300_context(pipe);
- memcpy(r300->vertex_buffers, buffers,
+ memcpy(r300->vertex_buffer, buffers,
sizeof(struct pipe_vertex_buffer) * count);
-
r300->vertex_buffer_count = count;
- draw_flush(r300->draw);
- draw_set_vertex_buffers(r300->draw, count, buffers);
+ if (r300->draw) {
+ draw_flush(r300->draw);
+ draw_set_vertex_buffers(r300->draw, count, buffers);
+ }
+
+ r300->dirty_state |= R300_NEW_VERTEX_FORMAT;
}
static void r300_set_vertex_elements(struct pipe_context* pipe,
@@ -617,8 +934,15 @@ static void r300_set_vertex_elements(struct pipe_context* pipe,
{
struct r300_context* r300 = r300_context(pipe);
- draw_flush(r300->draw);
- draw_set_vertex_elements(r300->draw, count, elements);
+ memcpy(r300->vertex_element,
+ elements,
+ sizeof(struct pipe_vertex_element) * count);
+ r300->vertex_element_count = count;
+
+ if (r300->draw) {
+ draw_flush(r300->draw);
+ draw_set_vertex_elements(r300->draw, count, elements);
+ }
}
static void* r300_create_vs_state(struct pipe_context* pipe,
@@ -634,9 +958,6 @@ static void* r300_create_vs_state(struct pipe_context* pipe,
tgsi_scan_shader(shader->tokens, &vs->info);
- /* Appease Draw. */
- vs->draw = draw_create_vertex_shader(r300->draw, shader);
-
return (void*)vs;
} else {
return draw_create_vertex_shader(r300->draw, shader);
@@ -647,8 +968,6 @@ static void r300_bind_vs_state(struct pipe_context* pipe, void* shader)
{
struct r300_context* r300 = r300_context(pipe);
- draw_flush(r300->draw);
-
if (r300_screen(pipe->screen)->caps->has_tcl) {
struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader;
@@ -659,10 +978,16 @@ static void r300_bind_vs_state(struct pipe_context* pipe, void* shader)
r300_translate_vertex_shader(r300, vs);
}
- draw_bind_vertex_shader(r300->draw, vs->draw);
r300->vs = vs;
- r300->dirty_state |= R300_NEW_VERTEX_SHADER;
+ if (r300->fs) {
+ r300_vertex_shader_setup_wpos(r300);
+ }
+
+ r300->dirty_state |=
+ R300_NEW_VERTEX_SHADER | R300_NEW_VERTEX_SHADER_CONSTANTS |
+ R300_NEW_VERTEX_FORMAT;
} else {
+ draw_flush(r300->draw);
draw_bind_vertex_shader(r300->draw,
(struct draw_vertex_shader*)shader);
}
@@ -676,8 +1001,7 @@ static void r300_delete_vs_state(struct pipe_context* pipe, void* shader)
struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader;
rc_constants_destroy(&vs->code.constants);
- draw_delete_vertex_shader(r300->draw, vs->draw);
- FREE(vs->state.tokens);
+ FREE((void*)vs->state.tokens);
FREE(shader);
} else {
draw_delete_vertex_shader(r300->draw,
@@ -685,6 +1009,31 @@ static void r300_delete_vs_state(struct pipe_context* pipe, void* shader)
}
}
+static void r300_set_constant_buffer(struct pipe_context *pipe,
+ uint shader, uint index,
+ const struct pipe_constant_buffer *buf)
+{
+ struct r300_context* r300 = r300_context(pipe);
+ void *mapped;
+
+ if (buf == NULL || buf->buffer->size == 0 ||
+ (mapped = pipe_buffer_map(pipe->screen, buf->buffer, PIPE_BUFFER_USAGE_CPU_READ)) == NULL)
+ {
+ r300->shader_constants[shader].count = 0;
+ return;
+ }
+
+ assert((buf->buffer->size % 4 * sizeof(float)) == 0);
+ memcpy(r300->shader_constants[shader].constants, mapped, buf->buffer->size);
+ r300->shader_constants[shader].count = buf->buffer->size / (4 * sizeof(float));
+ pipe_buffer_unmap(pipe->screen, buf->buffer);
+
+ if (shader == PIPE_SHADER_VERTEX)
+ r300->dirty_state |= R300_NEW_VERTEX_SHADER_CONSTANTS;
+ else if (shader == PIPE_SHADER_FRAGMENT)
+ r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS;
+}
+
void r300_init_state_functions(struct r300_context* r300)
{
r300->context.create_blend_state = r300_create_blend_state;
@@ -701,8 +1050,6 @@ void r300_init_state_functions(struct r300_context* r300)
r300->context.bind_depth_stencil_alpha_state = r300_bind_dsa_state;
r300->context.delete_depth_stencil_alpha_state = r300_delete_dsa_state;
- r300->context.set_edgeflags = r300_set_edgeflags;
-
r300->context.set_framebuffer_state = r300_set_framebuffer_state;
r300->context.create_fs_state = r300_create_fs_state;
@@ -716,10 +1063,11 @@ void r300_init_state_functions(struct r300_context* r300)
r300->context.delete_rasterizer_state = r300_delete_rs_state;
r300->context.create_sampler_state = r300_create_sampler_state;
- r300->context.bind_sampler_states = r300_bind_sampler_states;
+ r300->context.bind_fragment_sampler_states = r300_bind_sampler_states;
+ r300->context.bind_vertex_sampler_states = r300_lacks_vertex_textures;
r300->context.delete_sampler_state = r300_delete_sampler_state;
- r300->context.set_sampler_textures = r300_set_sampler_textures;
+ r300->context.set_fragment_sampler_textures = r300_set_sampler_textures;
r300->context.set_scissor_state = r300_set_scissor_state;
diff --git a/src/gallium/drivers/r300/r300_state_derived.c b/src/gallium/drivers/r300/r300_state_derived.c
index 02b7ab91076..22660a52d9a 100644
--- a/src/gallium/drivers/r300/r300_state_derived.c
+++ b/src/gallium/drivers/r300/r300_state_derived.c
@@ -1,5 +1,6 @@
/*
* Copyright 2008 Corbin Simpson <[email protected]>
+ * 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"),
@@ -20,439 +21,540 @@
* OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
* USE OR OTHER DEALINGS IN THE SOFTWARE. */
-#include "r300_state_derived.h"
+#include "draw/draw_context.h"
+
+#include "util/u_math.h"
+#include "util/u_memory.h"
+#include "r300_context.h"
#include "r300_fs.h"
+#include "r300_screen.h"
+#include "r300_shader_semantics.h"
+#include "r300_state_derived.h"
#include "r300_state_inlines.h"
#include "r300_vs.h"
/* r300_state_derived: Various bits of state which are dependent upon
* currently bound CSO data. */
-/* Set up the vs_tab and routes. */
-static void r300_vs_tab_routes(struct r300_context* r300,
- struct r300_vertex_format* vformat)
+struct r300_shader_key {
+ struct r300_vertex_shader* vs;
+ struct r300_fragment_shader* fs;
+};
+
+struct r300_shader_derived_value {
+ struct r300_vertex_format* vformat;
+ struct r300_rs_block* rs_block;
+};
+
+unsigned r300_shader_key_hash(void* key) {
+ struct r300_shader_key* shader_key = (struct r300_shader_key*)key;
+ unsigned vs = (intptr_t)shader_key->vs;
+ unsigned fs = (intptr_t)shader_key->fs;
+
+ return (vs << 16) | (fs & 0xffff);
+}
+
+int r300_shader_key_compare(void* key1, void* key2) {
+ struct r300_shader_key* shader_key1 = (struct r300_shader_key*)key1;
+ struct r300_shader_key* shader_key2 = (struct r300_shader_key*)key2;
+
+ return (shader_key1->vs == shader_key2->vs) &&
+ (shader_key1->fs == shader_key2->fs);
+}
+
+static void r300_draw_emit_attrib(struct r300_context* r300,
+ enum attrib_emit emit,
+ enum interp_mode interp,
+ int index)
{
- struct r300_screen* r300screen = r300_screen(r300->context.screen);
- struct vertex_info* vinfo = &vformat->vinfo;
- int* tab = vformat->vs_tab;
- boolean pos = FALSE, psize = FALSE, fog = FALSE;
- int i, texs = 0, cols = 0;
- struct tgsi_shader_info* info;
+ struct tgsi_shader_info* info = &r300->vs->info;
+ int output;
- if (r300screen->caps->has_tcl) {
- /* Use vertex shader to determine required routes. */
- info = &r300->vs->info;
+ output = draw_find_shader_output(r300->draw,
+ info->output_semantic_name[index],
+ info->output_semantic_index[index]);
+ draw_emit_vertex_attr(&r300->vertex_info->vinfo, emit, interp, output);
+}
+
+static void r300_draw_emit_all_attribs(struct r300_context* r300)
+{
+ struct r300_shader_semantics* vs_outputs = &r300->vs->outputs;
+ int i, gen_count;
+
+ /* Position. */
+ if (vs_outputs->pos != ATTR_UNUSED) {
+ r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE,
+ vs_outputs->pos);
} else {
- /* Use fragment shader to determine required routes. */
- info = &r300->fs->info;
+ assert(0);
}
- assert(info->num_inputs <= 16);
-
- if (!r300screen->caps->has_tcl || !r300->rs_state->enable_vte)
- {
- for (i = 0; i < info->num_inputs; i++) {
- switch (r300->vs->code.inputs[i]) {
- case TGSI_SEMANTIC_POSITION:
- pos = TRUE;
- tab[i] = 0;
- break;
- case TGSI_SEMANTIC_COLOR:
- tab[i] = 2 + cols;
- cols++;
- break;
- case TGSI_SEMANTIC_PSIZE:
- assert(psize == FALSE);
- psize = TRUE;
- tab[i] = 15;
- break;
- case TGSI_SEMANTIC_FOG:
- assert(fog == FALSE);
- fog = TRUE;
- /* Fall through */
- case TGSI_SEMANTIC_GENERIC:
- tab[i] = 6 + texs;
- texs++;
- break;
- default:
- debug_printf("r300: Unknown vertex input %d\n",
- info->input_semantic_name[i]);
- break;
- }
- }
+ /* Point size. */
+ if (vs_outputs->psize != ATTR_UNUSED) {
+ r300_draw_emit_attrib(r300, EMIT_1F_PSIZE, INTERP_POS,
+ vs_outputs->psize);
}
- else
- {
- /* Just copy vert attribs over as-is. */
- for (i = 0; i < info->num_inputs; i++) {
- tab[i] = i;
- }
- for (i = 0; i < info->num_outputs; i++) {
- switch (info->output_semantic_name[i]) {
- case TGSI_SEMANTIC_POSITION:
- pos = TRUE;
- break;
- case TGSI_SEMANTIC_COLOR:
- cols++;
- break;
- case TGSI_SEMANTIC_PSIZE:
- psize = TRUE;
- break;
- case TGSI_SEMANTIC_FOG:
- fog = TRUE;
- /* Fall through */
- case TGSI_SEMANTIC_GENERIC:
- texs++;
- break;
- default:
- debug_printf("r300: Unknown vertex output %d\n",
- info->output_semantic_name[i]);
- break;
- }
+ /* Colors. */
+ for (i = 0; i < ATTR_COLOR_COUNT; i++) {
+ if (vs_outputs->color[i] != ATTR_UNUSED) {
+ r300_draw_emit_attrib(r300, EMIT_4F, INTERP_LINEAR,
+ vs_outputs->color[i]);
}
}
- /* XXX magic */
- assert(texs <= 8);
+ /* XXX Back-face colors. */
- /* Do the actual vertex_info setup.
- *
- * vertex_info has four uints of hardware-specific data in it.
- * vinfo.hwfmt[0] is R300_VAP_VTX_STATE_CNTL
- * vinfo.hwfmt[1] is R300_VAP_VSM_VTX_ASSM
- * vinfo.hwfmt[2] is R300_VAP_OUTPUT_VTX_FMT_0
- * vinfo.hwfmt[3] is R300_VAP_OUTPUT_VTX_FMT_1 */
-
- vinfo->hwfmt[0] = 0x5555; /* XXX this is classic Mesa bonghits */
-
- /* We need to add vertex position attribute only for SW TCL case,
- * for HW TCL case it could be generated by vertex shader */
- if (!pos && !r300screen->caps->has_tcl) {
- debug_printf("r300: Forcing vertex position attribute emit...\n");
- /* Make room for the position attribute
- * at the beginning of the tab. */
- for (i = 15; i > 0; i--) {
- tab[i] = tab[i-1];
+ /* Texture coordinates. */
+ gen_count = 0;
+ for (i = 0; i < ATTR_GENERIC_COUNT; i++) {
+ if (vs_outputs->generic[i] != ATTR_UNUSED) {
+ r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE,
+ vs_outputs->generic[i]);
+ gen_count++;
}
- tab[0] = 0;
- }
- draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE,
- draw_find_vs_output(r300->draw, TGSI_SEMANTIC_POSITION, 0));
- vinfo->hwfmt[1] |= R300_INPUT_CNTL_POS;
- vinfo->hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT;
-
- if (psize) {
- draw_emit_vertex_attr(vinfo, EMIT_1F_PSIZE, INTERP_POS,
- draw_find_vs_output(r300->draw, TGSI_SEMANTIC_PSIZE, 0));
- vinfo->hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__PT_SIZE_PRESENT;
}
- for (i = 0; i < cols; i++) {
- draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_LINEAR,
- draw_find_vs_output(r300->draw, TGSI_SEMANTIC_COLOR, i));
- vinfo->hwfmt[1] |= R300_INPUT_CNTL_COLOR;
- vinfo->hwfmt[2] |= (R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << i);
+ /* Fog coordinates. */
+ if (vs_outputs->fog != ATTR_UNUSED) {
+ r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE,
+ vs_outputs->fog);
+ gen_count++;
}
- /* Init i right here, increment it if fog is enabled.
- * This gets around a double-increment problem. */
- i = 0;
+ /* XXX magic */
+ assert(gen_count <= 8);
+}
- if (fog) {
- i++;
- draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE,
- draw_find_vs_output(r300->draw, TGSI_SEMANTIC_FOG, 0));
- vinfo->hwfmt[1] |= (R300_INPUT_CNTL_TC0 << i);
- vinfo->hwfmt[3] |= (4 << (3 * i));
+/* Update the PSC tables. */
+static void r300_vertex_psc(struct r300_context* r300)
+{
+ struct r300_vertex_info *vformat = r300->vertex_info;
+ uint16_t type, swizzle;
+ enum pipe_format format;
+ unsigned i;
+ int identity[16] = {0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15};
+ int* stream_tab;
+
+ /* If TCL is bypassed, map vertex streams to equivalent VS output
+ * locations. */
+ if (r300->rs_state->enable_vte) {
+ stream_tab = identity;
+ } else {
+ stream_tab = r300->vs->stream_loc_notcl;
}
- for (; i < texs; i++) {
- draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE,
- draw_find_vs_output(r300->draw, TGSI_SEMANTIC_GENERIC, i));
- vinfo->hwfmt[1] |= (R300_INPUT_CNTL_TC0 << i);
- vinfo->hwfmt[3] |= (4 << (3 * i));
- }
+ /* Vertex shaders have no semantics on their inputs,
+ * so PSC should just route stuff based on the vertex elements,
+ * and not on attrib information. */
+ DBG(r300, DBG_DRAW, "r300: vs expects %d attribs, routing %d elements"
+ " in psc\n",
+ r300->vs->info.num_inputs,
+ r300->vertex_element_count);
- /* Handle the case where the vertex shader will be generating some of
- * the attribs based on its inputs. */
- if (r300screen->caps->has_tcl &&
- info->num_inputs < info->num_outputs) {
- vinfo->num_attribs = info->num_inputs;
+ for (i = 0; i < r300->vertex_element_count; i++) {
+ format = r300->vertex_element[i].src_format;
+
+ type = r300_translate_vertex_data_type(format) |
+ (stream_tab[i] << R300_DST_VEC_LOC_SHIFT);
+ swizzle = r300_translate_vertex_data_swizzle(format);
+
+ if (i & 1) {
+ vformat->vap_prog_stream_cntl[i >> 1] |= type << 16;
+ vformat->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 16;
+ } else {
+ vformat->vap_prog_stream_cntl[i >> 1] |= type;
+ vformat->vap_prog_stream_cntl_ext[i >> 1] |= swizzle;
+ }
}
- draw_compute_vertex_size(vinfo);
+ assert(i <= 15);
+
+ /* Set the last vector in the PSC. */
+ if (i) {
+ i -= 1;
+ }
+ vformat->vap_prog_stream_cntl[i >> 1] |=
+ (R300_LAST_VEC << (i & 1 ? 16 : 0));
}
-/* Update the PSC tables. */
-static void r300_vertex_psc(struct r300_context* r300,
- struct r300_vertex_format* vformat)
+/* Update the PSC tables for SW TCL, using Draw. */
+static void r300_swtcl_vertex_psc(struct r300_context* r300)
{
- struct r300_screen* r300screen = r300_screen(r300->context.screen);
+ struct r300_vertex_info *vformat = r300->vertex_info;
struct vertex_info* vinfo = &vformat->vinfo;
- int* tab = vformat->vs_tab;
- uint32_t temp;
- int i, attrib_count;
-
- /* Vertex shaders have no semantics on their inputs,
- * so PSC should just route stuff based on their info,
- * and not on attrib information. */
- if (r300screen->caps->has_tcl) {
- attrib_count = r300->vs->info.num_inputs;
- DBG(r300, DBG_DRAW, "r300: routing %d attribs in psc for vs\n",
- attrib_count);
- } else {
- attrib_count = vinfo->num_attribs;
- DBG(r300, DBG_DRAW, "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,"
- " tab %d\n", vinfo->attrib[i].src_index,
- vinfo->attrib[i].interp_mode, vinfo->attrib[i].emit,
- tab[i]);
- }
+ uint16_t type, swizzle;
+ enum pipe_format format;
+ unsigned i, attrib_count;
+ int* vs_output_tab = r300->vs->stream_loc_notcl;
+
+ /* 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);
+ 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]);
}
for (i = 0; i < attrib_count; i++) {
- /* Make sure we have a proper destination for our attribute */
- assert(tab[i] != -1);
+ /* Make sure we have a proper destination for our attribute. */
+ assert(vs_output_tab[i] != -1);
- /* Add the attribute to the PSC table. */
- temp = r300screen->caps->has_tcl ?
- R300_DATA_TYPE_FLOAT_4 :
- translate_vertex_data_type(vinfo->attrib[i].emit);
- temp |= tab[i] << R300_DST_VEC_LOC_SHIFT;
+ format = draw_translate_vinfo_format(vinfo->attrib[i].emit);
- if (i & 1) {
- vformat->vap_prog_stream_cntl[i >> 1] &= 0x0000ffff;
- vformat->vap_prog_stream_cntl[i >> 1] |= temp << 16;
+ /* Obtain the type of data in this attribute. */
+ type = r300_translate_vertex_data_type(format) |
+ vs_output_tab[i] << R300_DST_VEC_LOC_SHIFT;
- vformat->vap_prog_stream_cntl_ext[i >> 1] |=
- (R300_VAP_SWIZZLE_XYZW << 16);
- } else {
- vformat->vap_prog_stream_cntl[i >> 1] &= 0xffff0000;
- vformat->vap_prog_stream_cntl[i >> 1] |= temp << 0;
+ /* Obtain the swizzle for this attribute. Note that the default
+ * swizzle in the hardware is not XYZW! */
+ swizzle = r300_translate_vertex_data_swizzle(format);
- vformat->vap_prog_stream_cntl_ext[i >> 1] |=
- (R300_VAP_SWIZZLE_XYZW << 0);
+ /* Add the attribute to the PSC table. */
+ if (i & 1) {
+ vformat->vap_prog_stream_cntl[i >> 1] |= type << 16;
+ vformat->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 16;
+ } else {
+ vformat->vap_prog_stream_cntl[i >> 1] |= type;
+ vformat->vap_prog_stream_cntl_ext[i >> 1] |= swizzle;
}
}
/* Set the last vector in the PSC. */
- i--;
+ if (i) {
+ i -= 1;
+ }
vformat->vap_prog_stream_cntl[i >> 1] |=
(R300_LAST_VEC << (i & 1 ? 16 : 0));
}
-/* Update the vertex format. */
-static void r300_update_vertex_format(struct r300_context* r300)
+static void r300_rs_col(struct r300_rs_block* rs, int id, int ptr,
+ boolean swizzle_0001)
{
- struct r300_vertex_format vformat;
- int i;
-
- memset(&vformat, 0, sizeof(struct r300_vertex_format));
- for (i = 0; i < 16; i++) {
- vformat.vs_tab[i] = -1;
- vformat.fs_tab[i] = -1;
+ rs->ip[id] |= R300_RS_COL_PTR(ptr);
+ if (swizzle_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);
}
+ rs->inst[id] |= R300_RS_INST_COL_ID(id);
+}
- r300_vs_tab_routes(r300, &vformat);
-
- r300_vertex_psc(r300, &vformat);
+static void r300_rs_col_write(struct r300_rs_block* rs, int id, int fp_offset)
+{
+ rs->inst[id] |= R300_RS_INST_COL_CN_WRITE |
+ R300_RS_INST_COL_ADDR(fp_offset);
+}
- if (memcmp(&r300->vertex_info, &vformat,
- sizeof(struct r300_vertex_format))) {
- memcpy(&r300->vertex_info, &vformat,
- sizeof(struct r300_vertex_format));
- r300->dirty_state |= R300_NEW_VERTEX_FORMAT;
+static void r300_rs_tex(struct r300_rs_block* rs, int id, int ptr,
+ boolean swizzle_X001)
+{
+ if (swizzle_X001) {
+ rs->ip[id] |= R300_RS_TEX_PTR(ptr*4) |
+ 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 {
+ rs->ip[id] |= R300_RS_TEX_PTR(ptr*4) |
+ R300_RS_SEL_S(R300_RS_SEL_C0) |
+ R300_RS_SEL_T(R300_RS_SEL_C1) |
+ R300_RS_SEL_R(R300_RS_SEL_C2) |
+ R300_RS_SEL_Q(R300_RS_SEL_C3);
}
+ rs->inst[id] |= R300_RS_INST_TEX_ID(id);
}
-/* Set up the mappings from GB to US, for RS block. */
-static void r300_update_fs_tab(struct r300_context* r300)
+static void r300_rs_tex_write(struct r300_rs_block* rs, int id, int fp_offset)
{
- struct r300_vertex_format* vformat = &r300->vertex_info;
- struct tgsi_shader_info* info = &r300->fs->info;
- int i, cols = 0, texs = 0, cols_emitted = 0;
- int* tab = vformat->fs_tab;
+ rs->inst[id] |= R300_RS_INST_TEX_CN_WRITE |
+ R300_RS_INST_TEX_ADDR(fp_offset);
+}
- for (i = 0; i < 16; i++) {
- tab[i] = -1;
+static void r500_rs_col(struct r300_rs_block* rs, int id, int ptr,
+ boolean swizzle_0001)
+{
+ rs->ip[id] |= R500_RS_COL_PTR(ptr);
+ if (swizzle_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);
}
+ rs->inst[id] |= R500_RS_INST_COL_ID(id);
+}
- assert(info->num_inputs <= 16);
- for (i = 0; i < info->num_inputs; i++) {
- switch (info->input_semantic_name[i]) {
- case TGSI_SEMANTIC_COLOR:
- tab[i] = INTERP_LINEAR;
- cols++;
- break;
- case TGSI_SEMANTIC_POSITION:
- case TGSI_SEMANTIC_PSIZE:
- debug_printf("r300: Implementation error: Can't use "
- "pos attribs in fragshader yet!\n");
- /* Pass through for now */
- case TGSI_SEMANTIC_FOG:
- case TGSI_SEMANTIC_GENERIC:
- tab[i] = INTERP_PERSPECTIVE;
- break;
- default:
- debug_printf("r300: Unknown vertex input %d\n",
- info->input_semantic_name[i]);
- break;
- }
- }
+static void r500_rs_col_write(struct r300_rs_block* rs, int id, int fp_offset)
+{
+ rs->inst[id] |= R500_RS_INST_COL_CN_WRITE |
+ R500_RS_INST_COL_ADDR(fp_offset);
+}
- /* Now that we know where everything is... */
- DBG(r300, DBG_DRAW, "r300: fp input count: %d\n", info->num_inputs);
- for (i = 0; i < info->num_inputs; i++) {
- switch (tab[i]) {
- case INTERP_LINEAR:
- DBG(r300, DBG_DRAW, "r300: attrib: "
- "stack offset %d, color, tab %d\n",
- i, cols_emitted);
- tab[i] = cols_emitted;
- cols_emitted++;
- break;
- case INTERP_PERSPECTIVE:
- DBG(r300, DBG_DRAW, "r300: attrib: "
- "stack offset %d, texcoord, tab %d\n",
- i, cols + texs);
- tab[i] = cols + texs;
- texs++;
- break;
- case -1:
- debug_printf("r300: Implementation error: Bad fp interp!\n");
- default:
- break;
- }
+static void r500_rs_tex(struct r300_rs_block* rs, int id, int ptr,
+ boolean swizzle_X001)
+{
+ int rs_tex_comp = ptr*4;
+
+ if (swizzle_X001) {
+ rs->ip[id] |= R500_RS_SEL_S(rs_tex_comp) |
+ 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 {
+ 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->inst[id] |= R500_RS_INST_TEX_ID(id);
+}
+static void r500_rs_tex_write(struct r300_rs_block* rs, int id, int fp_offset)
+{
+ rs->inst[id] |= R500_RS_INST_TEX_CN_WRITE |
+ R500_RS_INST_TEX_ADDR(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)
+/* 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)
{
struct r300_rs_block* rs = r300->rs_block;
- struct tgsi_shader_info* info = &r300->fs->info;
- int* tab = r300->vertex_info.fs_tab;
- int col_count = 0, fp_offset = 0, i, tex_count = 0;
- int rs_tex_comp = 0;
- memset(rs, 0, sizeof(struct r300_rs_block));
+ int i, col_count = 0, tex_count = 0, fp_offset = 0;
+ void (*rX00_rs_col)(struct r300_rs_block*, int, int, boolean);
+ void (*rX00_rs_col_write)(struct r300_rs_block*, int, int);
+ void (*rX00_rs_tex)(struct r300_rs_block*, int, int, boolean);
+ 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;
if (r300_screen(r300->context.screen)->caps->is_r500) {
- for (i = 0; i < info->num_inputs; i++) {
- assert(tab[i] != -1);
- switch (info->input_semantic_name[i]) {
- case TGSI_SEMANTIC_COLOR:
- rs->ip[col_count] |=
- R500_RS_COL_PTR(col_count) |
- R500_RS_COL_FMT(R300_RS_COL_FMT_RGBA);
- col_count++;
- break;
- case TGSI_SEMANTIC_GENERIC:
- rs->ip[tex_count] |=
- 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);
- tex_count++;
- rs_tex_comp += 4;
- break;
- default:
- break;
- }
- }
+ rX00_rs_col = r500_rs_col;
+ rX00_rs_col_write = r500_rs_col_write;
+ rX00_rs_tex = r500_rs_tex;
+ rX00_rs_tex_write = r500_rs_tex_write;
+ } else {
+ rX00_rs_col = r300_rs_col;
+ rX00_rs_col_write = r300_rs_col_write;
+ rX00_rs_tex = r300_rs_tex;
+ rX00_rs_tex_write = r300_rs_tex_write;
+ }
- /* Rasterize at least one color, or bad things happen. */
- if ((col_count == 0) && (tex_count == 0)) {
- rs->ip[0] |= R500_RS_COL_FMT(R300_RS_COL_FMT_0001);
+ /* Rasterize colors. */
+ for (i = 0; i < ATTR_COLOR_COUNT; i++) {
+ if (vs_outputs->color[i] != ATTR_UNUSED || any_bcolor_used) {
+ /* Always rasterize if it's written by the VS,
+ * otherwise it locks up. */
+ rX00_rs_col(rs, col_count, i, FALSE);
+
+ /* Write it to the FS input register if it's used by the FS. */
+ if (fs_inputs->color[i] != ATTR_UNUSED) {
+ rX00_rs_col_write(rs, col_count, fp_offset);
+ fp_offset++;
+ }
col_count++;
+ } 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->color[i] != ATTR_UNUSED) {
+ fp_offset++;
+ }
}
+ }
- for (i = 0; i < tex_count; i++) {
- rs->inst[i] |= R500_RS_INST_TEX_ID(i) |
- R500_RS_INST_TEX_CN_WRITE | R500_RS_INST_TEX_ADDR(fp_offset);
- fp_offset++;
+ /* Rasterize texture coordinates. */
+ for (i = 0; i < ATTR_GENERIC_COUNT; i++) {
+ if (vs_outputs->generic[i] != ATTR_UNUSED) {
+ /* Always rasterize if it's written by the VS,
+ * otherwise it locks up. */
+ rX00_rs_tex(rs, tex_count, tex_count, FALSE);
+
+ /* Write it to the FS input register if it's used by the FS. */
+ if (fs_inputs->generic[i] != ATTR_UNUSED) {
+ rX00_rs_tex_write(rs, tex_count, fp_offset);
+ fp_offset++;
+ }
+ tex_count++;
+ } 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++;
+ }
}
+ }
- for (i = 0; i < col_count; i++) {
- rs->inst[i] |= R500_RS_INST_COL_ID(i) |
- R500_RS_INST_COL_CN_WRITE | R500_RS_INST_COL_ADDR(fp_offset);
+ /* 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, TRUE);
+
+ /* Write it to the FS input register if it's used by the FS. */
+ if (fs_inputs->fog != ATTR_UNUSED) {
+ rX00_rs_tex_write(rs, tex_count, fp_offset);
fp_offset++;
}
+ tex_count++;
} else {
- for (i = 0; i < info->num_inputs; i++) {
- assert(tab[i] != -1);
- switch (info->input_semantic_name[i]) {
- case TGSI_SEMANTIC_COLOR:
- rs->ip[col_count] |=
- R300_RS_COL_PTR(col_count) |
- R300_RS_COL_FMT(R300_RS_COL_FMT_RGBA);
- col_count++;
- break;
- case TGSI_SEMANTIC_GENERIC:
- rs->ip[tex_count] |=
- R300_RS_TEX_PTR(rs_tex_comp) |
- R300_RS_SEL_S(R300_RS_SEL_C0) |
- R300_RS_SEL_T(R300_RS_SEL_C1) |
- R300_RS_SEL_R(R300_RS_SEL_C2) |
- R300_RS_SEL_Q(R300_RS_SEL_C3);
- tex_count++;
- rs_tex_comp+=4;
- break;
- default:
- break;
- }
+ /* 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++;
}
+ }
- if (col_count == 0) {
- rs->ip[0] |= R300_RS_COL_FMT(R300_RS_COL_FMT_0001);
- }
+ /* Rasterize WPOS. */
+ /* If the FS doesn't need it, it's not written by the VS. */
+ if (fs_inputs->wpos != ATTR_UNUSED) {
+ rX00_rs_tex(rs, tex_count, tex_count, FALSE);
+ rX00_rs_tex_write(rs, tex_count, fp_offset);
- if (tex_count == 0) {
- rs->ip[0] |=
- R300_RS_SEL_S(R300_RS_SEL_K0) |
- R300_RS_SEL_T(R300_RS_SEL_K0) |
- R300_RS_SEL_R(R300_RS_SEL_K0) |
- R300_RS_SEL_Q(R300_RS_SEL_K1);
- }
+ fp_offset++;
+ tex_count++;
+ }
- /* Rasterize at least one color, or bad things happen. */
- if ((col_count == 0) && (tex_count == 0)) {
- col_count++;
- }
+ /* Rasterize at least one color, or bad things happen. */
+ if (col_count == 0 && tex_count == 0) {
+ rX00_rs_col(rs, 0, 0, TRUE);
+ col_count++;
+ }
- for (i = 0; i < tex_count; i++) {
- rs->inst[i] |= R300_RS_INST_TEX_ID(i) |
- R300_RS_INST_TEX_CN_WRITE | R300_RS_INST_TEX_ADDR(fp_offset);
- fp_offset++;
- }
+ rs->count = (tex_count*4) | (col_count << R300_IC_COUNT_SHIFT) |
+ R300_HIRES_EN;
- for (i = 0; i < col_count; i++) {
- rs->inst[i] |= R300_RS_INST_COL_ID(i) |
- R300_RS_INST_COL_CN_WRITE | R300_RS_INST_COL_ADDR(fp_offset);
- fp_offset++;
- }
+ rs->inst_count = MAX3(col_count - 1, tex_count - 1, 0);
+}
+
+/* Update the vertex format. */
+static void r300_update_derived_shader_state(struct r300_context* r300)
+{
+ struct r300_screen* r300screen = r300_screen(r300->context.screen);
+
+ /*
+ struct r300_shader_key* key;
+ struct r300_shader_derived_value* value;
+ key = CALLOC_STRUCT(r300_shader_key);
+ key->vs = r300->vs;
+ key->fs = r300->fs;
+
+ value = (struct r300_shader_derived_value*)
+ util_hash_table_get(r300->shader_hash_table, (void*)key);
+ if (value) {
+ //vformat = value->vformat;
+ rs_block = value->rs_block;
+
+ FREE(key);
+ } else {
+ rs_block = CALLOC_STRUCT(r300_rs_block);
+ value = CALLOC_STRUCT(r300_shader_derived_value);
+
+ r300_update_rs_block(r300, rs_block);
+
+ //value->vformat = vformat;
+ value->rs_block = rs_block;
+ util_hash_table_set(r300->shader_hash_table,
+ (void*)key, (void*)value);
+ } */
+
+ /* Reset structures */
+ memset(r300->rs_block, 0, sizeof(struct r300_rs_block));
+ memset(r300->vertex_info, 0, sizeof(struct r300_vertex_info));
+ memcpy(r300->vertex_info->vinfo.hwfmt, r300->vs->hwfmt, sizeof(uint)*4);
+
+ r300_update_rs_block(r300, &r300->vs->outputs, &r300->fs->inputs);
+
+ if (r300screen->caps->has_tcl) {
+ r300_vertex_psc(r300);
+ } else {
+ r300_draw_emit_all_attribs(r300);
+ draw_compute_vertex_size(&r300->vertex_info->vinfo);
+ r300_swtcl_vertex_psc(r300);
}
- rs->count = (rs_tex_comp) | (col_count << R300_IC_COUNT_SHIFT) |
- R300_HIRES_EN;
+ r300->dirty_state |= R300_NEW_RS_BLOCK;
+}
+
+static boolean r300_dsa_writes_depth_stencil(struct r300_dsa_state* dsa)
+{
+ /* We are interested only in the cases when a new depth or stencil value
+ * can be written and changed. */
+
+ /* 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));
+}
+
+static boolean r300_dsa_alpha_test_enabled(struct r300_dsa_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;
+}
- rs->inst_count = MAX2(MAX2(col_count - 1, tex_count - 1), 0);
+static void r300_update_ztop(struct r300_context* r300)
+{
+ r300->ztop_state.z_buffer_top = R300_ZTOP_ENABLE;
+
+ /* This is important enough that I felt it warranted a comment.
+ *
+ * According to the docs, these are the conditions where ZTOP must be
+ * disabled:
+ * 1) Alpha testing enabled
+ * 2) Texture kill instructions in fragment shader
+ * 3) Chroma key culling enabled
+ * 4) W-buffering enabled
+ *
+ * The docs claim that for the first three cases, if no ZS writes happen,
+ * then ZTOP can be used.
+ *
+ * (3) will never apply since we do not support chroma-keyed operations.
+ * (4) will need to be re-examined (and this comment updated) if/when
+ * Hyper-Z becomes supported.
+ *
+ * Additionally, the following conditions require disabled ZTOP:
+ * 5) Depth writes in fragment shader
+ * 6) Outstanding occlusion queries
+ *
+ * ~C.
+ */
+
+ /* ZS writes */
+ if (r300_dsa_writes_depth_stencil(r300->dsa_state) &&
+ (r300_dsa_alpha_test_enabled(r300->dsa_state) || /* (1) */
+ r300->fs->info.uses_kill)) { /* (2) */
+ r300->ztop_state.z_buffer_top = R300_ZTOP_DISABLE;
+ } else if (r300_fragment_shader_writes_depth(r300->fs)) { /* (5) */
+ r300->ztop_state.z_buffer_top = R300_ZTOP_DISABLE;
+ } else if (r300->query_current) { /* (6) */
+ r300->ztop_state.z_buffer_top = R300_ZTOP_DISABLE;
+ }
}
void r300_update_derived_state(struct r300_context* r300)
{
if (r300->dirty_state &
- (R300_NEW_FRAGMENT_SHADER | R300_NEW_VERTEX_SHADER)) {
- r300_update_vertex_format(r300);
+ (R300_NEW_FRAGMENT_SHADER | R300_NEW_VERTEX_SHADER |
+ R300_NEW_VERTEX_FORMAT)) {
+ r300_update_derived_shader_state(r300);
}
- if (r300->dirty_state & R300_NEW_VERTEX_FORMAT) {
- r300_update_fs_tab(r300);
- r300_update_rs_block(r300);
+ if (r300->dirty_state &
+ (R300_NEW_DSA | R300_NEW_FRAGMENT_SHADER | R300_NEW_QUERY)) {
+ r300_update_ztop(r300);
}
}
diff --git a/src/gallium/drivers/r300/r300_state_derived.h b/src/gallium/drivers/r300/r300_state_derived.h
index 71a4a47b003..05ad535e2de 100644
--- a/src/gallium/drivers/r300/r300_state_derived.h
+++ b/src/gallium/drivers/r300/r300_state_derived.h
@@ -25,6 +25,10 @@
struct r300_context;
+unsigned r300_shader_key_hash(void* key);
+
+int r300_shader_key_compare(void* key1, void* key2);
+
void r300_update_derived_state(struct r300_context* r300);
#endif /* R300_STATE_DERIVED_H */
diff --git a/src/gallium/drivers/r300/r300_state_inlines.h b/src/gallium/drivers/r300/r300_state_inlines.h
index 88eb66b79e7..35be00e1b01 100644
--- a/src/gallium/drivers/r300/r300_state_inlines.h
+++ b/src/gallium/drivers/r300/r300_state_inlines.h
@@ -24,8 +24,12 @@
#ifndef R300_STATE_INLINES_H
#define R300_STATE_INLINES_H
+#include "draw/draw_vertex.h"
+
#include "pipe/p_format.h"
+#include "util/u_format.h"
+
#include "r300_reg.h"
/* Some maths. These should probably find their way to u_math, if needed. */
@@ -51,6 +55,7 @@ static INLINE uint32_t r300_translate_blend_function(int blend_func)
return R300_COMB_FCN_MAX;
default:
debug_printf("r300: Unknown blend function %d\n", blend_func);
+ assert(0);
break;
}
return 0;
@@ -98,6 +103,7 @@ static INLINE uint32_t r300_translate_blend_factor(int blend_fact)
case PIPE_BLENDFACTOR_INV_SRC1_ALPHA: */
default:
debug_printf("r300: Unknown blend factor %d\n", blend_fact);
+ assert(0);
break;
}
return 0;
@@ -127,6 +133,7 @@ static INLINE uint32_t r300_translate_depth_stencil_function(int zs_func)
default:
debug_printf("r300: Unknown depth/stencil function %d\n",
zs_func);
+ assert(0);
break;
}
return 0;
@@ -153,6 +160,7 @@ static INLINE uint32_t r300_translate_stencil_op(int s_op)
return R300_ZS_INVERT;
default:
debug_printf("r300: Unknown stencil op %d", s_op);
+ assert(0);
break;
}
return 0;
@@ -179,11 +187,48 @@ static INLINE uint32_t r300_translate_alpha_function(int alpha_func)
return R300_FG_ALPHA_FUNC_ALWAYS;
default:
debug_printf("r300: Unknown alpha function %d", alpha_func);
+ assert(0);
break;
}
return 0;
}
+static INLINE uint32_t
+r300_translate_polygon_mode_front(unsigned mode) {
+ switch (mode)
+ {
+ case PIPE_POLYGON_MODE_FILL:
+ return R300_GA_POLY_MODE_FRONT_PTYPE_TRI;
+ case PIPE_POLYGON_MODE_LINE:
+ return R300_GA_POLY_MODE_FRONT_PTYPE_LINE;
+ case PIPE_POLYGON_MODE_POINT:
+ return R300_GA_POLY_MODE_FRONT_PTYPE_POINT;
+
+ default:
+ debug_printf("r300: Bad polygon mode %i in %s\n", mode,
+ __FUNCTION__);
+ return R300_GA_POLY_MODE_FRONT_PTYPE_TRI;
+ }
+}
+
+static INLINE uint32_t
+r300_translate_polygon_mode_back(unsigned mode) {
+ switch (mode)
+ {
+ case PIPE_POLYGON_MODE_FILL:
+ return R300_GA_POLY_MODE_BACK_PTYPE_TRI;
+ case PIPE_POLYGON_MODE_LINE:
+ return R300_GA_POLY_MODE_BACK_PTYPE_LINE;
+ case PIPE_POLYGON_MODE_POINT:
+ return R300_GA_POLY_MODE_BACK_PTYPE_POINT;
+
+ default:
+ debug_printf("r300: Bad polygon mode %i in %s\n", mode,
+ __FUNCTION__);
+ return R300_GA_POLY_MODE_BACK_PTYPE_TRI;
+ }
+}
+
/* Texture sampler state. */
static INLINE uint32_t r300_translate_wrap(int wrap)
@@ -207,40 +252,42 @@ static INLINE uint32_t r300_translate_wrap(int wrap)
return R300_TX_CLAMP_TO_EDGE | R300_TX_MIRRORED;
default:
debug_printf("r300: Unknown texture wrap %d", wrap);
+ assert(0);
return 0;
}
}
-static INLINE uint32_t r300_translate_tex_filters(int min, int mag, int mip)
+static INLINE uint32_t r300_translate_tex_filters(int min, int mag, int mip,
+ int is_anisotropic)
{
uint32_t retval = 0;
- switch (min) {
+ if (is_anisotropic)
+ retval |= R300_TX_MIN_FILTER_ANISO | R300_TX_MAG_FILTER_ANISO;
+ else {
+ switch (min) {
case PIPE_TEX_FILTER_NEAREST:
retval |= R300_TX_MIN_FILTER_NEAREST;
break;
case PIPE_TEX_FILTER_LINEAR:
retval |= R300_TX_MIN_FILTER_LINEAR;
break;
- case PIPE_TEX_FILTER_ANISO:
- retval |= R300_TX_MIN_FILTER_ANISO;
- break;
default:
debug_printf("r300: Unknown texture filter %d\n", min);
+ assert(0);
break;
- }
- switch (mag) {
+ }
+ switch (mag) {
case PIPE_TEX_FILTER_NEAREST:
retval |= R300_TX_MAG_FILTER_NEAREST;
break;
case PIPE_TEX_FILTER_LINEAR:
retval |= R300_TX_MAG_FILTER_LINEAR;
break;
- case PIPE_TEX_FILTER_ANISO:
- retval |= R300_TX_MAG_FILTER_ANISO;
- break;
default:
debug_printf("r300: Unknown texture filter %d\n", mag);
+ assert(0);
break;
+ }
}
switch (mip) {
case PIPE_TEX_MIPFILTER_NONE:
@@ -254,6 +301,7 @@ static INLINE uint32_t r300_translate_tex_filters(int min, int mag, int mip)
break;
default:
debug_printf("r300: Unknown texture filter %d\n", mip);
+ assert(0);
break;
}
@@ -277,6 +325,8 @@ static INLINE uint32_t r300_anisotropy(float max_aniso)
/* Buffer formats. */
+/* Colorbuffer formats. This is the unswizzled format of the RB3D block's
+ * output. For the swizzling of the targets, check the shader's format. */
static INLINE uint32_t r300_translate_colorformat(enum pipe_format format)
{
switch (format) {
@@ -295,7 +345,6 @@ static INLINE uint32_t r300_translate_colorformat(enum pipe_format format)
case PIPE_FORMAT_X8R8G8B8_UNORM:
case PIPE_FORMAT_R8G8B8A8_UNORM:
case PIPE_FORMAT_R8G8B8X8_UNORM:
- case PIPE_FORMAT_Z24S8_UNORM:
return R300_COLOR_FORMAT_ARGB8888;
/* XXX Not in pipe_format
case PIPE_FORMAT_A32R32G32B32:
@@ -312,17 +361,21 @@ static INLINE uint32_t r300_translate_colorformat(enum pipe_format format)
debug_printf("r300: Implementation error: "
"Got unsupported color format %s in %s\n",
pf_name(format), __FUNCTION__);
+ assert(0);
break;
}
return 0;
}
+/* Depthbuffer and stencilbuffer. Thankfully, we only support two flavors. */
static INLINE uint32_t r300_translate_zsformat(enum pipe_format format)
{
switch (format) {
/* 16-bit depth, no stencil */
case PIPE_FORMAT_Z16_UNORM:
return R300_DEPTHFORMAT_16BIT_INT_Z;
+ /* 24-bit depth, ignored stencil */
+ case PIPE_FORMAT_Z24X8_UNORM:
/* 24-bit depth, 8-bit stencil */
case PIPE_FORMAT_Z24S8_UNORM:
return R300_DEPTHFORMAT_24BIT_INT_Z_8BIT_STENCIL;
@@ -330,18 +383,22 @@ static INLINE uint32_t r300_translate_zsformat(enum pipe_format format)
debug_printf("r300: Implementation error: "
"Got unsupported ZS format %s in %s\n",
pf_name(format), __FUNCTION__);
+ assert(0);
break;
}
return 0;
}
-/* Translate pipe_format into US_OUT_FMT.
+/* Shader output formats. This is essentially the swizzle from the shader
+ * to the RB3D block.
+ *
* Note that formats are stored from C3 to C0. */
static INLINE uint32_t r300_translate_out_fmt(enum pipe_format format)
{
switch (format) {
case PIPE_FORMAT_A8R8G8B8_UNORM:
case PIPE_FORMAT_X8R8G8B8_UNORM:
+ /* XXX */
case PIPE_FORMAT_Z24S8_UNORM:
return R300_US_OUT_FMT_C4_8 |
R300_C0_SEL_B | R300_C1_SEL_G |
@@ -355,6 +412,7 @@ static INLINE uint32_t r300_translate_out_fmt(enum pipe_format format)
debug_printf("r300: Implementation error: "
"Got unsupported output format %s in %s\n",
pf_name(format), __FUNCTION__);
+ assert(0);
return R300_US_OUT_FMT_UNUSED;
}
return 0;
@@ -381,32 +439,119 @@ static INLINE uint32_t r300_translate_gb_pipes(int pipe_count)
return 0;
}
-static INLINE uint32_t translate_vertex_data_type(int type) {
- switch (type) {
- case EMIT_1F:
- case EMIT_1F_PSIZE:
- return R300_DATA_TYPE_FLOAT_1;
- break;
- case EMIT_2F:
- return R300_DATA_TYPE_FLOAT_2;
- break;
- case EMIT_3F:
- return R300_DATA_TYPE_FLOAT_3;
- break;
- case EMIT_4F:
- return R300_DATA_TYPE_FLOAT_4;
+/* Utility function to count the number of components in RGBAZS formats.
+ * XXX should go to util or p_format.h */
+static INLINE unsigned pf_component_count(enum pipe_format format) {
+ unsigned count = 0;
+
+ if (util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_RGB, 0)) {
+ count++;
+ }
+ if (util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_RGB, 1)) {
+ count++;
+ }
+ if (util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_RGB, 2)) {
+ count++;
+ }
+ if (util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_RGB, 3)) {
+ count++;
+ }
+ if (util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_ZS, 0)) {
+ count++;
+ }
+ if (util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_ZS, 1)) {
+ count++;
+ }
+
+ return count;
+}
+
+/* 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 components = pf_component_count(format);
+
+ desc = util_format_description(format);
+
+ if (desc->layout != UTIL_FORMAT_LAYOUT_ARITH &&
+ desc->layout != UTIL_FORMAT_LAYOUT_ARRAY) {
+ debug_printf("r300: Bad format %s in %s:%d\n", pf_name(format),
+ __FUNCTION__, __LINE__);
+ assert(0);
+ }
+
+ switch (desc->channel[0].type) {
+ /* Half-floats, floats, doubles */
+ case UTIL_FORMAT_TYPE_FLOAT:
+ switch (util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_RGB, 0)) {
+ case 32:
+ result = R300_DATA_TYPE_FLOAT_1 + (components - 1);
+ break;
+ default:
+ debug_printf("r300: Bad format %s in %s:%d\n",
+ pf_name(format), __FUNCTION__, __LINE__);
+ assert(0);
+ }
break;
- case EMIT_4UB:
- return R300_DATA_TYPE_BYTE;
+ /* Unsigned ints */
+ case UTIL_FORMAT_TYPE_UNSIGNED:
+ /* Signed ints */
+ case UTIL_FORMAT_TYPE_SIGNED:
+ switch (util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_RGB, 0)) {
+ case 8:
+ result = R300_DATA_TYPE_BYTE;
+ break;
+ case 16:
+ if (components > 2) {
+ result = R300_DATA_TYPE_SHORT_4;
+ } else {
+ result = R300_DATA_TYPE_SHORT_2;
+ }
+ break;
+ default:
+ debug_printf("r300: Bad format %s in %s:%d\n",
+ pf_name(format), __FUNCTION__, __LINE__);
+ debug_printf("r300: util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_RGB, 0) == %d\n",
+ util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_RGB, 0));
+ assert(0);
+ }
break;
default:
- debug_printf("r300: Implementation error: "
- "Bad vertex data type!\n");
+ debug_printf("r300: Bad format %s in %s:%d\n",
+ pf_name(format), __FUNCTION__, __LINE__);
assert(0);
- break;
}
- return 0;
+ if (desc->channel[0].type == UTIL_FORMAT_TYPE_SIGNED) {
+ result |= R300_SIGNED;
+ }
+ if (desc->channel[0].normalized) {
+ result |= R300_NORMALIZE;
+ }
+
+ return result;
+}
+
+static INLINE uint16_t
+r300_translate_vertex_data_swizzle(enum pipe_format format) {
+ const struct util_format_description *desc = util_format_description(format);
+
+ assert(format);
+
+ if (desc->layout != UTIL_FORMAT_LAYOUT_ARITH &&
+ desc->layout != UTIL_FORMAT_LAYOUT_ARRAY) {
+ debug_printf("r300: Bad format %s in %s:%d\n",
+ pf_name(format), __FUNCTION__, __LINE__);
+ return 0;
+ }
+
+ return ((desc->swizzle[0] << R300_SWIZZLE_SELECT_X_SHIFT) |
+ (desc->swizzle[1] << R300_SWIZZLE_SELECT_Y_SHIFT) |
+ (desc->swizzle[2] << R300_SWIZZLE_SELECT_Z_SHIFT) |
+ (desc->swizzle[3] << R300_SWIZZLE_SELECT_W_SHIFT) |
+ (0xf << R300_WRITE_ENA_SHIFT));
}
#endif /* R300_STATE_INLINES_H */
diff --git a/src/gallium/drivers/r300/r300_state_invariant.c b/src/gallium/drivers/r300/r300_state_invariant.c
index 3865730d635..f25f3ca217d 100644
--- a/src/gallium/drivers/r300/r300_state_invariant.c
+++ b/src/gallium/drivers/r300/r300_state_invariant.c
@@ -21,9 +21,12 @@
* 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},
@@ -40,7 +43,7 @@ void r300_emit_invariant_state(struct r300_context* r300)
struct r300_capabilities* caps = r300_screen(r300->context.screen)->caps;
CS_LOCALS(r300);
- BEGIN_CS(24 + (caps->has_tcl ? 2: 0));
+ BEGIN_CS(16 + (caps->has_tcl ? 2: 0));
/*** Graphics Backend (GB) ***/
/* Various GB enables */
@@ -63,13 +66,8 @@ void r300_emit_invariant_state(struct r300_context* r300)
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);
- OUT_CS_REG(R300_FG_DEPTH_SRC, 0x0);
- OUT_CS_REG(R300_US_W_FMT, 0x0);
/*** VAP ***/
- /* Max and min vertex index clamp. */
- OUT_CS_REG(R300_VAP_VF_MIN_VTX_INDX, 0x0);
- OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, 0xffffff);
/* Sign/normalize control */
OUT_CS_REG(R300_VAP_PSC_SGN_NORM_CNTL, R300_SGN_NORM_NO_ZERO);
/* TCL-only stuff */
@@ -81,15 +79,11 @@ void r300_emit_invariant_state(struct r300_context* r300)
END_CS;
/* XXX unsorted stuff from surface_fill */
- BEGIN_CS(64 + (caps->has_tcl ? 5 : 0) + (caps->is_r500 ? 4 : 0));
- /* Flush PVS. */
- OUT_CS_REG(R300_VAP_PVS_STATE_FLUSH_REG, 0x0);
+ BEGIN_CS(44 + (caps->has_tcl ? 7 : 0) + (caps->is_r500 ? 4 : 0));
- OUT_CS_REG(R300_SE_VTE_CNTL, R300_VPORT_X_SCALE_ENA |
- R300_VPORT_X_OFFSET_ENA | R300_VPORT_Y_SCALE_ENA |
- R300_VPORT_Y_OFFSET_ENA | R300_VPORT_Z_SCALE_ENA |
- R300_VPORT_Z_OFFSET_ENA | R300_VTX_W0_FMT);
if (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);
@@ -111,7 +105,6 @@ void r300_emit_invariant_state(struct r300_context* r300)
/* 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_POLY_MODE, 0x00000000);
OUT_CS_REG(R300_GA_ROUND_MODE, 0x00000001);
OUT_CS_REG(R300_GA_OFFSET, 0x00000000);
OUT_CS_REG(R300_GA_FOG_SCALE, 0x3DBF1412);
@@ -121,22 +114,15 @@ void r300_emit_invariant_state(struct r300_context* r300)
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_CCTL, 0x00000000);
- OUT_CS_REG(RB3D_COLOR_CHANNEL_MASK, 0x0000000F);
OUT_CS_REG(R300_RB3D_AARESOLVE_CTL, 0x00000000);
if (caps->is_r500) {
- OUT_CS_REG(R500_RB3D_DISCARD_SRC_PIXEL_LTE_THRESHOLD, 0x00000000);
- OUT_CS_REG(R500_RB3D_DISCARD_SRC_PIXEL_GTE_THRESHOLD, 0xFFFFFFFF);
+ 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_FORMAT, 0x00000002);
- OUT_CS_REG(R300_ZB_ZCACHE_CTLSTAT, 0x00000003);
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);
- OUT_CS_REG(R300_VAP_VTX_STATE_CNTL, 0x1);
- OUT_CS_REG(R300_VAP_VSM_VTX_ASSM, 0x405);
- OUT_CS_REG(R300_SE_VTE_CNTL, 0x0000043F);
/* XXX */
OUT_CS_REG(R300_SC_CLIP_RULE, 0xaaaa);
diff --git a/src/gallium/drivers/r300/r300_state_invariant.h b/src/gallium/drivers/r300/r300_state_invariant.h
index 5bea6779fe5..05cff0d6dfe 100644
--- a/src/gallium/drivers/r300/r300_state_invariant.h
+++ b/src/gallium/drivers/r300/r300_state_invariant.h
@@ -23,11 +23,7 @@
#ifndef R300_STATE_INVARIANT_H
#define R300_STATE_INVARIANT_H
-#include "r300_chipset.h"
-#include "r300_context.h"
-#include "r300_cs.h"
-#include "r300_reg.h"
-#include "r300_state_inlines.h"
+struct r300_context;
void r300_emit_invariant_state(struct r300_context* r300);
diff --git a/src/gallium/drivers/r300/r300_surface.c b/src/gallium/drivers/r300/r300_surface.c
deleted file mode 100644
index cc6288cb519..00000000000
--- a/src/gallium/drivers/r300/r300_surface.c
+++ /dev/null
@@ -1,372 +0,0 @@
-/*
- * Copyright 2008 Corbin Simpson <[email protected]>
- * Joakim Sindholt <[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_surface.h"
-
-static void r300_surface_setup(struct r300_context* r300,
- struct r300_texture* dest,
- unsigned x, unsigned y,
- unsigned w, unsigned h)
-{
- struct r300_capabilities* caps = r300_screen(r300->context.screen)->caps;
- unsigned pixpitch = r300_texture_get_stride(dest, 0) / dest->tex.block.size;
- CS_LOCALS(r300);
-
- r300_emit_blend_state(r300, &blend_clear_state);
- r300_emit_blend_color_state(r300, &blend_color_clear_state);
- r300_emit_dsa_state(r300, &dsa_clear_state);
- r300_emit_rs_state(r300, &rs_clear_state);
-
- BEGIN_CS(26);
-
- /* Viewport setup */
- OUT_CS_REG_SEQ(R300_SE_VPORT_XSCALE, 6);
- OUT_CS_32F((float)w);
- OUT_CS_32F((float)x);
- OUT_CS_32F((float)h);
- OUT_CS_32F((float)y);
- OUT_CS_32F(1.0);
- OUT_CS_32F(0.0);
-
- OUT_CS_REG(R300_VAP_VTE_CNTL, R300_VPORT_X_SCALE_ENA |
- R300_VPORT_X_OFFSET_ENA |
- R300_VPORT_Y_SCALE_ENA |
- R300_VPORT_Y_OFFSET_ENA |
- R300_VTX_XY_FMT | R300_VTX_Z_FMT);
-
- /* Pixel scissors. */
- OUT_CS_REG_SEQ(R300_SC_SCISSORS_TL, 2);
- if (caps->is_r500) {
- OUT_CS((x << R300_SCISSORS_X_SHIFT) | (y << R300_SCISSORS_Y_SHIFT));
- OUT_CS(((w - 1) << R300_SCISSORS_X_SHIFT) | ((h - 1) << R300_SCISSORS_Y_SHIFT));
- } else {
- /* Non-R500 chipsets have an offset of 1440 in their scissors. */
- OUT_CS(((x + 1440) << R300_SCISSORS_X_SHIFT) |
- ((y + 1440) << R300_SCISSORS_Y_SHIFT));
- OUT_CS((((w - 1) + 1440) << R300_SCISSORS_X_SHIFT) |
- (((h - 1) + 1440) << R300_SCISSORS_Y_SHIFT));
- }
-
- /* Flush colorbuffer and blend caches. */
- OUT_CS_REG(R300_RB3D_DSTCACHE_CTLSTAT,
- R300_RB3D_DSTCACHE_CTLSTAT_DC_FLUSH_FLUSH_DIRTY_3D |
- R300_RB3D_DSTCACHE_CTLSTAT_DC_FINISH_SIGNAL);
- OUT_CS_REG(R300_ZB_ZCACHE_CTLSTAT,
- R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE |
- R300_ZB_ZCACHE_CTLSTAT_ZC_FREE_FREE);
-
- /* Setup colorbuffer. */
- OUT_CS_REG_SEQ(R300_RB3D_COLOROFFSET0, 1);
- OUT_CS_RELOC(dest->buffer, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0);
- OUT_CS_REG_SEQ(R300_RB3D_COLORPITCH0, 1);
- OUT_CS_RELOC(dest->buffer, pixpitch |
- r300_translate_colorformat(dest->tex.format), 0,
- RADEON_GEM_DOMAIN_VRAM, 0);
- OUT_CS_REG(RB3D_COLOR_CHANNEL_MASK, 0xf);
-
- END_CS;
-}
-
-/* Provides pipe_context's "surface_fill". Commonly used for clearing
- * buffers. */
-static void r300_surface_fill(struct pipe_context* pipe,
- struct pipe_surface* dest,
- unsigned x, unsigned y,
- unsigned w, unsigned h,
- unsigned color)
-{
- int i;
- float r, g, b, a, depth;
- struct r300_context* r300 = r300_context(pipe);
- struct r300_capabilities* caps = r300_screen(pipe->screen)->caps;
- struct r300_texture* tex = (struct r300_texture*)dest->texture;
- unsigned pixpitch = r300_texture_get_stride(tex, 0) / tex->tex.block.size;
- boolean invalid = FALSE;
- CS_LOCALS(r300);
-
- a = (float)((color >> 24) & 0xff) / 255.0f;
- r = (float)((color >> 16) & 0xff) / 255.0f;
- g = (float)((color >> 8) & 0xff) / 255.0f;
- b = (float)((color >> 0) & 0xff) / 255.0f;
- debug_printf("r300: Filling surface %p at (%d,%d),"
- " dimensions %dx%d (pixel pitch %d), color 0x%x\n",
- dest, x, y, w, h, pixpitch, color);
-
- /* Fallback? */
- if (FALSE) {
-fallback:
- debug_printf("r300: Falling back on surface clear...");
- util_surface_fill(pipe, dest, x, y, w, h, color);
- return;
- }
-
- /* Make sure our target BO is okay. */
-validate:
- if (!r300->winsys->add_buffer(r300->winsys, tex->buffer,
- 0, RADEON_GEM_DOMAIN_VRAM)) {
- r300->context.flush(&r300->context, 0, NULL);
- goto validate;
- }
- if (!r300->winsys->validate(r300->winsys)) {
- r300->context.flush(&r300->context, 0, NULL);
- if (invalid) {
- debug_printf("r300: Stuck in validation loop, gonna fallback.");
- goto fallback;
- }
- invalid = TRUE;
- goto validate;
- }
-
- r300_surface_setup(r300, tex, x, y, w, h);
-
- /* Vertex shader setup */
- if (caps->has_tcl) {
- r300_emit_vertex_program_code(r300, &r300_passthrough_vertex_shader, 0);
- } else {
- BEGIN_CS(4);
- OUT_CS_REG(R300_VAP_CNTL_STATUS,
-#ifdef PIPE_ARCH_BIG_ENDIAN
- R300_VC_32BIT_SWAP |
-#endif
- R300_VAP_TCL_BYPASS);
- OUT_CS_REG(R300_VAP_CNTL, R300_PVS_NUM_SLOTS(5) |
- R300_PVS_NUM_CNTLRS(5) |
- R300_PVS_NUM_FPUS(caps->num_vert_fpus) |
- R300_PVS_VF_MAX_VTX_NUM(12));
- END_CS;
- }
-
- /* Fragment shader setup */
- if (caps->is_r500) {
- r500_emit_fragment_program_code(r300, &r5xx_passthrough_fragment_shader, 0);
- r300_emit_rs_block_state(r300, &r5xx_rs_block_clear_state);
- } else {
- r300_emit_fragment_program_code(r300, &r3xx_passthrough_fragment_shader, 0);
- r300_emit_rs_block_state(r300, &r3xx_rs_block_clear_state);
- }
-
- BEGIN_CS(26);
-
- /* VAP stream control, mapping from input memory to PVS/RS memory */
- if (caps->has_tcl) {
- OUT_CS_REG(R300_VAP_PROG_STREAM_CNTL_0,
- (R300_DATA_TYPE_FLOAT_4 << R300_DATA_TYPE_0_SHIFT) |
- ((R300_LAST_VEC | (1 << R300_DST_VEC_LOC_SHIFT) |
- R300_DATA_TYPE_FLOAT_4) << R300_DATA_TYPE_1_SHIFT));
- } else {
- OUT_CS_REG(R300_VAP_PROG_STREAM_CNTL_0,
- (R300_DATA_TYPE_FLOAT_4 << R300_DATA_TYPE_0_SHIFT) |
- ((R300_LAST_VEC | (2 << R300_DST_VEC_LOC_SHIFT) |
- R300_DATA_TYPE_FLOAT_4) << R300_DATA_TYPE_1_SHIFT));
- }
- OUT_CS_REG(R300_VAP_PROG_STREAM_CNTL_EXT_0,
- (R300_VAP_SWIZZLE_XYZW << R300_SWIZZLE0_SHIFT) |
- (R300_VAP_SWIZZLE_XYZW << R300_SWIZZLE1_SHIFT));
-
- /* VAP format controls */
- OUT_CS_REG(R300_VAP_OUTPUT_VTX_FMT_0,
- R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT |
- R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT);
- OUT_CS_REG(R300_VAP_OUTPUT_VTX_FMT_1, 0x0);
-
- /* Disable textures */
- OUT_CS_REG(R300_TX_ENABLE, 0x0);
-
- /* The size of the point we're about to draw, in sixths of pixels */
- OUT_CS_REG(R300_GA_POINT_SIZE,
- ((h * 6) & R300_POINTSIZE_Y_MASK) |
- ((w * 6) << R300_POINTSIZE_X_SHIFT));
-
- /* Vertex size. */
- OUT_CS_REG(R300_VAP_VTX_SIZE, 0x8);
-
- /* Packet3 with our point vertex */
- OUT_CS_PKT3(R200_3D_DRAW_IMMD_2, 8);
- OUT_CS(R300_PRIM_TYPE_POINT | R300_PRIM_WALK_RING |
- (1 << R300_PRIM_NUM_VERTICES_SHIFT));
- /* Position */
- OUT_CS_32F(0.5);
- OUT_CS_32F(0.5);
- OUT_CS_32F(1.0);
- OUT_CS_32F(1.0);
- /* Color */
- OUT_CS_32F(r);
- OUT_CS_32F(g);
- OUT_CS_32F(b);
- OUT_CS_32F(a);
-
- OUT_CS_REG(R300_RB3D_DSTCACHE_CTLSTAT, 0xA);
-
- END_CS;
-
- r300->dirty_hw++;
-}
-
-static void r300_surface_copy(struct pipe_context* pipe,
- struct pipe_surface* dest,
- unsigned destx, unsigned desty,
- struct pipe_surface* src,
- unsigned srcx, unsigned srcy,
- unsigned w, unsigned h)
-{
- struct r300_context* r300 = r300_context(pipe);
- struct r300_capabilities* caps = r300_screen(pipe->screen)->caps;
- struct r300_texture* srctex = (struct r300_texture*)src->texture;
- struct r300_texture* desttex = (struct r300_texture*)dest->texture;
- unsigned pixpitch = r300_texture_get_stride(srctex, 0) / srctex->tex.block.size;
- boolean invalid = FALSE;
- float fsrcx = srcx, fsrcy = srcy, fdestx = destx, fdesty = desty;
- CS_LOCALS(r300);
-
- debug_printf("r300: Copying surface %p at (%d,%d) to %p at (%d, %d),"
- " dimensions %dx%d (pixel pitch %d)\n",
- src, srcx, srcy, dest, destx, desty, w, h, pixpitch);
-
- if ((srctex->buffer == desttex->buffer) &&
- ((destx < srcx + w) || (srcx < destx + w)) &&
- ((desty < srcy + h) || (srcy < desty + h))) {
-fallback:
- debug_printf("r300: Falling back on surface_copy\n");
- util_surface_copy(pipe, FALSE, dest, destx, desty, src,
- srcx, srcy, w, h);
- }
-
- /* Add our target BOs to the list. */
-validate:
- if (!r300->winsys->add_buffer(r300->winsys, srctex->buffer,
- RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM, 0)) {
- r300->context.flush(&r300->context, 0, NULL);
- goto validate;
- }
- if (!r300->winsys->add_buffer(r300->winsys, desttex->buffer,
- 0, RADEON_GEM_DOMAIN_VRAM)) {
- r300->context.flush(&r300->context, 0, NULL);
- goto validate;
- }
- if (!r300->winsys->validate(r300->winsys)) {
- r300->context.flush(&r300->context, 0, NULL);
- if (invalid) {
- debug_printf("r300: Stuck in validation loop, gonna fallback.");
- goto fallback;
- }
- invalid = TRUE;
- goto validate;
- }
-
- r300_surface_setup(r300, desttex, destx, desty, w, h);
-
- /* Setup the texture. */
- r300_emit_texture(r300, &r300_sampler_copy_state, srctex, 0);
-
- /* Flush and enable. */
- r300_flush_textures(r300);
-
- /* Vertex shader setup */
- if (caps->has_tcl) {
- r300_emit_vertex_program_code(r300, &r300_passthrough_vertex_shader, 0);
- } else {
- BEGIN_CS(4);
- OUT_CS_REG(R300_VAP_CNTL_STATUS,
-#ifdef PIPE_ARCH_BIG_ENDIAN
- R300_VC_32BIT_SWAP |
-#endif
- R300_VAP_TCL_BYPASS);
- OUT_CS_REG(R300_VAP_CNTL, R300_PVS_NUM_SLOTS(5) |
- R300_PVS_NUM_CNTLRS(5) |
- R300_PVS_NUM_FPUS(caps->num_vert_fpus) |
- R300_PVS_VF_MAX_VTX_NUM(12));
- END_CS;
- }
-
- /* Fragment shader setup */
- if (caps->is_r500) {
- r500_emit_fragment_program_code(r300, &r5xx_texture_fragment_shader, 0);
- r300_emit_rs_block_state(r300, &r5xx_rs_block_copy_state);
- } else {
- r300_emit_fragment_program_code(r300, &r3xx_texture_fragment_shader, 0);
- r300_emit_rs_block_state(r300, &r3xx_rs_block_copy_state);
- }
-
- BEGIN_CS(30);
- /* VAP stream control, mapping from input memory to PVS/RS memory */
- if (caps->has_tcl) {
- OUT_CS_REG(R300_VAP_PROG_STREAM_CNTL_0,
- (R300_DATA_TYPE_FLOAT_2 << R300_DATA_TYPE_0_SHIFT) |
- ((R300_LAST_VEC | (1 << R300_DST_VEC_LOC_SHIFT) |
- R300_DATA_TYPE_FLOAT_2) << R300_DATA_TYPE_1_SHIFT));
- } else {
- OUT_CS_REG(R300_VAP_PROG_STREAM_CNTL_0,
- (R300_DATA_TYPE_FLOAT_2 << R300_DATA_TYPE_0_SHIFT) |
- ((R300_LAST_VEC | (6 << R300_DST_VEC_LOC_SHIFT) |
- R300_DATA_TYPE_FLOAT_2) << R300_DATA_TYPE_1_SHIFT));
- }
- OUT_CS_REG(R300_VAP_PROG_STREAM_CNTL_EXT_0,
- (R300_VAP_SWIZZLE_XYZW << R300_SWIZZLE0_SHIFT) |
- (R300_VAP_SWIZZLE_XYZW << R300_SWIZZLE1_SHIFT));
-
- /* VAP format controls */
- OUT_CS_REG(R300_VAP_OUTPUT_VTX_FMT_0,
- R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT);
- /* Two components of texture 0 */
- OUT_CS_REG(R300_VAP_OUTPUT_VTX_FMT_1, 0x2);
-
- /* Vertex size. */
- OUT_CS_REG(R300_VAP_VTX_SIZE, 0x4);
-
- /* Packet3 with our texcoords */
- OUT_CS_PKT3(R200_3D_DRAW_IMMD_2, 16);
- OUT_CS(R300_PRIM_TYPE_QUADS | R300_PRIM_WALK_RING |
- (4 << R300_PRIM_NUM_VERTICES_SHIFT));
- /* (x , y ) */
- OUT_CS_32F(fdestx / dest->width);
- OUT_CS_32F(fdesty / dest->height);
- OUT_CS_32F(fsrcx / src->width);
- OUT_CS_32F(fsrcy / src->height);
- /* (x , y + h) */
- OUT_CS_32F(fdestx / dest->width);
- OUT_CS_32F((fdesty + h) / dest->height);
- OUT_CS_32F(fsrcx / src->width);
- OUT_CS_32F((fsrcy + h) / src->height);
- /* (x + w, y + h) */
- OUT_CS_32F((fdestx + w) / dest->width);
- OUT_CS_32F((fdesty + h) / dest->height);
- OUT_CS_32F((fsrcx + w) / src->width);
- OUT_CS_32F((fsrcy + h) / src->height);
- /* (x + w, y ) */
- OUT_CS_32F((fdestx + w) / dest->width);
- OUT_CS_32F(fdesty / dest->height);
- OUT_CS_32F((fsrcx + w) / src->width);
- OUT_CS_32F(fsrcy / src->height);
-
- OUT_CS_REG(R300_RB3D_DSTCACHE_CTLSTAT, 0xA);
-
- END_CS;
-
- r300->dirty_hw++;
-}
-
-void r300_init_surface_functions(struct r300_context* r300)
-{
- r300->context.surface_fill = r300_surface_fill;
- r300->context.surface_copy = r300_surface_copy;
-}
diff --git a/src/gallium/drivers/r300/r300_surface.h b/src/gallium/drivers/r300/r300_surface.h
deleted file mode 100644
index f9e98b2ec9c..00000000000
--- a/src/gallium/drivers/r300/r300_surface.h
+++ /dev/null
@@ -1,124 +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_SURFACE_H
-#define R300_SURFACE_H
-
-#include "pipe/p_context.h"
-#include "pipe/p_screen.h"
-
-#include "util/u_rect.h"
-
-#include "r300_context.h"
-#include "r300_cs.h"
-#include "r300_emit.h"
-#include "r300_fs.h"
-#include "r300_vs.h"
-#include "r300_state_inlines.h"
-
-static struct r300_blend_state blend_clear_state = {
- .blend_control = 0x0,
- .alpha_blend_control = 0x0,
- .rop = 0x0,
- .dither = 0x0,
-};
-
-static struct r300_blend_color_state blend_color_clear_state = {
- .blend_color = 0x0,
- .blend_color_red_alpha = 0x0,
- .blend_color_green_blue = 0x0,
-};
-
-static struct r300_dsa_state dsa_clear_state = {
- .alpha_function = 0x0,
- .alpha_reference = 0x0,
- .z_buffer_control = 0x0,
- .z_stencil_control = 0x0,
- .stencil_ref_mask = R300_STENCILWRITEMASK_MASK,
- .z_buffer_top = R300_ZTOP_ENABLE,
- .stencil_ref_bf = 0x0,
-};
-
-static struct r300_rs_state rs_clear_state = {
- .point_minmax = 0x36000006,
- .line_control = 0x00030006,
- .depth_scale_front = 0x0,
- .depth_offset_front = 0x0,
- .depth_scale_back = 0x0,
- .depth_offset_back = 0x0,
- .polygon_offset_enable = 0x0,
- .cull_mode = 0x0,
- .line_stipple_config = 0x3BAAAAAB,
- .line_stipple_value = 0x0,
- .color_control = R300_SHADE_MODEL_FLAT,
-};
-
-static struct r300_rs_block r3xx_rs_block_clear_state = {
- .ip[0] = R500_RS_SEL_S(R300_RS_SEL_C0) |
- R500_RS_SEL_T(R300_RS_SEL_C0) |
- R500_RS_SEL_R(R300_RS_SEL_C0) |
- R500_RS_SEL_Q(R300_RS_SEL_K1),
- .inst[0] = R300_RS_INST_COL_CN_WRITE,
- .count = R300_IT_COUNT(0) | R300_IC_COUNT(1) | R300_HIRES_EN,
- .inst_count = 0,
-};
-
-static struct r300_rs_block r5xx_rs_block_clear_state = {
- .ip[0] = R500_RS_SEL_S(R500_RS_IP_PTR_K0) |
- 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),
- .inst[0] = R500_RS_INST_COL_CN_WRITE,
- .count = R300_IT_COUNT(0) | R300_IC_COUNT(1) | R300_HIRES_EN,
- .inst_count = 0,
-};
-
-/* The following state is used for surface_copy only. */
-
-static struct r300_rs_block r3xx_rs_block_copy_state = {
- .ip[0] = R500_RS_SEL_S(R300_RS_SEL_K0) |
- R500_RS_SEL_T(R300_RS_SEL_K0) |
- R500_RS_SEL_R(R300_RS_SEL_K0) |
- R500_RS_SEL_Q(R300_RS_SEL_K1),
- .inst[0] = R300_RS_INST_COL_CN_WRITE,
- .count = R300_IT_COUNT(2) | R300_IC_COUNT(0) | R300_HIRES_EN,
- .inst_count = R300_RS_TX_OFFSET(0),
-};
-
-static struct r300_rs_block r5xx_rs_block_copy_state = {
- .ip[0] = R500_RS_SEL_S(0) |
- R500_RS_SEL_T(1) |
- R500_RS_SEL_R(R500_RS_IP_PTR_K0) |
- R500_RS_SEL_Q(R500_RS_IP_PTR_K1),
- .inst[0] = R500_RS_INST_TEX_CN_WRITE,
- .count = R300_IT_COUNT(2) | R300_IC_COUNT(0) | R300_HIRES_EN,
- .inst_count = R300_RS_TX_OFFSET(0),
-};
-
-static struct r300_sampler_state r300_sampler_copy_state = {
- .filter0 = R300_TX_WRAP_S(R300_TX_CLAMP) |
- R300_TX_WRAP_T(R300_TX_CLAMP) |
- R300_TX_MAG_FILTER_NEAREST |
- R300_TX_MIN_FILTER_NEAREST,
-};
-
-#endif /* R300_SURFACE_H */
diff --git a/src/gallium/drivers/r300/r300_texture.c b/src/gallium/drivers/r300/r300_texture.c
index ce60ded7caf..9a96206a4dc 100644
--- a/src/gallium/drivers/r300/r300_texture.c
+++ b/src/gallium/drivers/r300/r300_texture.c
@@ -20,41 +20,75 @@
* 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"
+
+#include "r300_context.h"
#include "r300_texture.h"
+#include "r300_screen.h"
-static void r300_setup_texture_state(struct r300_texture* tex)
+static void r300_setup_texture_state(struct r300_texture* tex, boolean is_r500)
{
struct r300_texture_state* state = &tex->state;
struct pipe_texture *pt = &tex->tex;
- state->format0 = R300_TX_WIDTH((pt->width[0] - 1) & 0x7ff) |
- R300_TX_HEIGHT((pt->height[0] - 1) & 0x7ff) |
- R300_TX_DEPTH(util_logbase2(pt->depth[0]) & 0xf) |
- R300_TX_NUM_LEVELS(pt->last_level) |
- R300_TX_PITCH_EN;
+ state->format0 = R300_TX_WIDTH((pt->width0 - 1) & 0x7ff) |
+ R300_TX_HEIGHT((pt->height0 - 1) & 0x7ff);
+
+ if (tex->is_npot) {
+ /* rectangles love this */
+ state->format0 |= R300_TX_PITCH_EN;
+ state->format2 = (tex->pitch[0] - 1) & 0x1fff;
+ } else {
+ /* power of two textures (3D, mipmaps, and no pitch) */
+ state->format0 |= R300_TX_DEPTH(util_logbase2(pt->depth0) & 0xf);
+ }
- /* XXX */
state->format1 = r300_translate_texformat(pt->format);
if (pt->target == PIPE_TEXTURE_CUBE) {
- state->format1 |= R300_TX_FORMAT_CUBIC_MAP;
+ state->format1 |= R300_TX_FORMAT_CUBIC_MAP;
}
if (pt->target == PIPE_TEXTURE_3D) {
- state->format1 |= R300_TX_FORMAT_3D;
+ state->format1 |= R300_TX_FORMAT_3D;
}
- state->format2 = (r300_texture_get_stride(tex, 0) / pt->block.size) - 1;
-
- /* Assume (somewhat foolishly) that oversized textures will
- * not be permitted by the state tracker. */
- if (pt->width[0] > 2048) {
- state->format2 |= R500_TXWIDTH_BIT11;
- }
- if (pt->height[0] > 2048) {
- state->format2 |= R500_TXHEIGHT_BIT11;
+ /* large textures on r500 */
+ if (is_r500)
+ {
+ if (pt->width0 > 2048) {
+ state->format2 |= R500_TXWIDTH_BIT11;
+ }
+ if (pt->height0 > 2048) {
+ state->format2 |= R500_TXHEIGHT_BIT11;
+ }
}
+ assert(is_r500 || (pt->width0 <= 2048 && pt->height0 <= 2048));
debug_printf("r300: Set texture state (%dx%d, %d levels)\n",
- pt->width[0], pt->height[0], pt->last_level);
+ pt->width0, pt->height0, pt->last_level);
+}
+
+unsigned r300_texture_get_offset(struct r300_texture* tex, unsigned level,
+ unsigned zslice, unsigned face)
+{
+ unsigned offset = tex->offset[level];
+
+ switch (tex->tex.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;
+ }
}
/**
@@ -67,47 +101,49 @@ unsigned r300_texture_get_stride(struct r300_texture* tex, unsigned level)
return tex->stride_override;
if (level > tex->tex.last_level) {
- debug_printf("%s: level (%u) > last_level (%u)\n", __FUNCTION__, level, tex->tex.last_level);
+ debug_printf("%s: level (%u) > last_level (%u)\n", __FUNCTION__,
+ level, tex->tex.last_level);
return 0;
}
- return align(pf_get_stride(&tex->tex.block, tex->tex.width[level]), 32);
+ return align(util_format_get_stride(tex->tex.format, u_minify(tex->tex.width0, level)), 32);
}
static void r300_setup_miptree(struct r300_texture* tex)
{
struct pipe_texture* base = &tex->tex;
- int stride, size;
+ int stride, size, layer_size;
int i;
for (i = 0; i <= base->last_level; i++) {
- if (i > 0) {
- base->width[i] = minify(base->width[i-1]);
- base->height[i] = minify(base->height[i-1]);
- base->depth[i] = minify(base->depth[i-1]);
- }
+ unsigned nblocksy = util_format_get_nblocksy(base->format, u_minify(base->height0, i));
- base->nblocksx[i] = pf_get_nblocksx(&base->block, base->width[i]);
- base->nblocksy[i] = pf_get_nblocksy(&base->block, base->height[i]);
-
- /* Radeons enjoy things in multiples of 64.
- *
- * XXX
- * POT, uncompressed, unmippmapped textures can be aligned to 32,
- * instead of 64. */
stride = r300_texture_get_stride(tex, i);
- size = stride * base->nblocksy[i] * base->depth[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] = align(tex->size, 32);
tex->size = tex->offset[i] + size;
+ tex->layer_size[i] = layer_size;
+ tex->pitch[i] = stride / util_format_get_blocksize(base->format);
debug_printf("r300: Texture miptree: Level %d "
"(%dx%dx%d px, pitch %d bytes)\n",
- i, base->width[i], base->height[i], base->depth[i],
- stride);
+ i, u_minify(base->width0, i), u_minify(base->height0, i),
+ u_minify(base->depth0, i), stride);
}
}
+static void r300_setup_flags(struct r300_texture* tex)
+{
+ tex->is_npot = !util_is_power_of_two(tex->tex.width0) ||
+ !util_is_power_of_two(tex->tex.height0);
+}
+
/* Create a new texture. */
static struct pipe_texture*
r300_texture_create(struct pipe_screen* screen,
@@ -123,9 +159,9 @@ static struct pipe_texture*
pipe_reference_init(&tex->tex.reference, 1);
tex->tex.screen = screen;
+ r300_setup_flags(tex);
r300_setup_miptree(tex);
-
- r300_setup_texture_state(tex);
+ r300_setup_texture_state(tex, r300_screen(screen)->caps->is_r500);
tex->buffer = screen->buffer_create(screen, 1024,
PIPE_BUFFER_USAGE_PIXEL,
@@ -159,17 +195,20 @@ static struct pipe_surface* r300_get_tex_surface(struct pipe_screen* screen,
struct pipe_surface* surface = CALLOC_STRUCT(pipe_surface);
unsigned offset;
- /* XXX this is certainly dependent on tex target */
- offset = tex->offset[level];
+ offset = r300_texture_get_offset(tex, level, zslice, face);
if (surface) {
pipe_reference_init(&surface->reference, 1);
pipe_texture_reference(&surface->texture, texture);
surface->format = texture->format;
- surface->width = texture->width[level];
- surface->height = texture->height[level];
+ 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;
}
return surface;
@@ -189,10 +228,10 @@ static struct pipe_texture*
{
struct r300_texture* tex;
- /* XXX we should start doing mips now... */
+ /* Support only 2D textures without mipmaps */
if (base->target != PIPE_TEXTURE_2D ||
- base->last_level != 0 ||
- base->depth[0] != 1) {
+ base->depth0 != 1 ||
+ base->last_level != 0) {
return NULL;
}
@@ -206,15 +245,64 @@ static struct pipe_texture*
tex->tex.screen = screen;
tex->stride_override = *stride;
+ tex->pitch[0] = *stride / util_format_get_blocksize(base->format);
- /* XXX */
- r300_setup_texture_state(tex);
+ r300_setup_flags(tex);
+ r300_setup_texture_state(tex, r300_screen(screen)->caps->is_r500);
pipe_buffer_reference(&tex->buffer, buffer);
return (struct pipe_texture*)tex;
}
+static struct pipe_video_surface *
+r300_video_surface_create(struct pipe_screen *screen,
+ enum pipe_video_chroma_format chroma_format,
+ unsigned width, unsigned height)
+{
+ struct r300_video_surface *r300_vsfc;
+ struct pipe_texture template;
+
+ assert(screen);
+ assert(width && height);
+
+ r300_vsfc = CALLOC_STRUCT(r300_video_surface);
+ if (!r300_vsfc)
+ return NULL;
+
+ pipe_reference_init(&r300_vsfc->base.reference, 1);
+ r300_vsfc->base.screen = screen;
+ r300_vsfc->base.chroma_format = chroma_format;
+ r300_vsfc->base.width = width;
+ r300_vsfc->base.height = height;
+
+ memset(&template, 0, sizeof(struct pipe_texture));
+ template.target = PIPE_TEXTURE_2D;
+ template.format = PIPE_FORMAT_X8R8G8B8_UNORM;
+ template.last_level = 0;
+ template.width0 = util_next_power_of_two(width);
+ template.height0 = util_next_power_of_two(height);
+ template.depth0 = 1;
+ template.tex_usage = PIPE_TEXTURE_USAGE_SAMPLER |
+ PIPE_TEXTURE_USAGE_RENDER_TARGET;
+
+ r300_vsfc->tex = screen->texture_create(screen, &template);
+ if (!r300_vsfc->tex)
+ {
+ FREE(r300_vsfc);
+ return NULL;
+ }
+
+ return &r300_vsfc->base;
+}
+
+static void r300_video_surface_destroy(struct pipe_video_surface *vsfc)
+{
+ struct r300_video_surface *r300_vsfc = r300_video_surface(vsfc);
+ pipe_texture_reference(&r300_vsfc->tex, NULL);
+ FREE(r300_vsfc);
+}
+
void r300_init_screen_texture_functions(struct pipe_screen* screen)
{
screen->texture_create = r300_texture_create;
@@ -222,6 +310,9 @@ void r300_init_screen_texture_functions(struct pipe_screen* screen)
screen->get_tex_surface = r300_get_tex_surface;
screen->tex_surface_destroy = r300_tex_surface_destroy;
screen->texture_blanket = r300_texture_blanket;
+
+ screen->video_surface_create = r300_video_surface_create;
+ screen->video_surface_destroy= r300_video_surface_destroy;
}
boolean r300_get_texture_buffer(struct pipe_texture* texture,
diff --git a/src/gallium/drivers/r300/r300_texture.h b/src/gallium/drivers/r300/r300_texture.h
index bd87790bc34..55ceb1a5136 100644
--- a/src/gallium/drivers/r300/r300_texture.h
+++ b/src/gallium/drivers/r300/r300_texture.h
@@ -23,11 +23,8 @@
#ifndef R300_TEXTURE_H
#define R300_TEXTURE_H
-#include "pipe/p_screen.h"
+#include "pipe/p_video_state.h"
-#include "util/u_math.h"
-
-#include "r300_context.h"
#include "r300_reg.h"
struct r300_texture;
@@ -36,6 +33,9 @@ void r300_init_screen_texture_functions(struct pipe_screen* screen);
unsigned r300_texture_get_stride(struct r300_texture* tex, unsigned level);
+unsigned r300_texture_get_offset(struct r300_texture* tex, unsigned level,
+ unsigned zslice, unsigned face);
+
/* Note the signature of R300_EASY_TX_FORMAT(A, R, G, B, FORMAT)... */
static INLINE uint32_t r300_translate_texformat(enum pipe_format format)
{
@@ -43,6 +43,19 @@ static INLINE uint32_t r300_translate_texformat(enum pipe_format format)
/* X8 */
case PIPE_FORMAT_I8_UNORM:
return R300_EASY_TX_FORMAT(X, X, X, X, X8);
+ case PIPE_FORMAT_L8_UNORM:
+ return R300_EASY_TX_FORMAT(X, X, X, ONE, X8);
+ /* X16 */
+ case PIPE_FORMAT_R16_UNORM:
+ return R300_EASY_TX_FORMAT(X, X, X, X, X16);
+ case PIPE_FORMAT_R16_SNORM:
+ return R300_EASY_TX_FORMAT(X, X, X, X, X16) |
+ R300_TX_FORMAT_SIGNED;
+ case PIPE_FORMAT_Z16_UNORM:
+ return R300_EASY_TX_FORMAT(X, X, X, X, X16);
+ /* Y8X8 */
+ case PIPE_FORMAT_A8L8_UNORM:
+ return R300_EASY_TX_FORMAT(X, X, X, Y, Y8X8);
/* W8Z8Y8X8 */
case PIPE_FORMAT_A8R8G8B8_UNORM:
return R300_EASY_TX_FORMAT(X, Y, Z, W, W8Z8Y8X8);
@@ -75,12 +88,9 @@ static INLINE uint32_t r300_translate_texformat(enum pipe_format format)
R300_TX_FORMAT_YUV_TO_RGB;
/* W24_FP */
case PIPE_FORMAT_Z24S8_UNORM:
+ case PIPE_FORMAT_Z24X8_UNORM:
return R300_EASY_TX_FORMAT(X, X, X, X, W24_FP);
- /* Z5_Y6_X5 */
- case PIPE_FORMAT_R16_SNORM:
- return R300_EASY_TX_FORMAT(X, X, X, X, Z5Y6X5);
- case PIPE_FORMAT_Z16_UNORM:
- return R300_EASY_TX_FORMAT(X, X, X, X, X16);
+
default:
debug_printf("r300: Implementation error: "
"Got unsupported texture format %s in %s\n",
@@ -91,6 +101,18 @@ static INLINE uint32_t r300_translate_texformat(enum pipe_format format)
return 0;
}
+struct r300_video_surface
+{
+ struct pipe_video_surface base;
+ struct pipe_texture *tex;
+};
+
+static INLINE struct r300_video_surface *
+r300_video_surface(struct pipe_video_surface *pvs)
+{
+ return (struct r300_video_surface *)pvs;
+}
+
#ifndef R300_WINSYS_H
boolean r300_get_texture_buffer(struct pipe_texture* texture,
diff --git a/src/gallium/drivers/r300/r300_tgsi_to_rc.c b/src/gallium/drivers/r300/r300_tgsi_to_rc.c
index 74d4fb50876..a792c2cf989 100644
--- a/src/gallium/drivers/r300/r300_tgsi_to_rc.c
+++ b/src/gallium/drivers/r300/r300_tgsi_to_rc.c
@@ -120,7 +120,7 @@ static unsigned translate_opcode(unsigned opcode)
/* case TGSI_OPCODE_NOT: return RC_OPCODE_NOT; */
/* case TGSI_OPCODE_TRUNC: return RC_OPCODE_TRUNC; */
/* case TGSI_OPCODE_SHL: return RC_OPCODE_SHL; */
- /* case TGSI_OPCODE_SHR: return RC_OPCODE_SHR; */
+ /* case TGSI_OPCODE_ISHR: return RC_OPCODE_SHR; */
/* case TGSI_OPCODE_AND: return RC_OPCODE_AND; */
/* case TGSI_OPCODE_OR: return RC_OPCODE_OR; */
/* case TGSI_OPCODE_MOD: return RC_OPCODE_MOD; */
@@ -135,10 +135,6 @@ static unsigned translate_opcode(unsigned opcode)
/* case TGSI_OPCODE_BGNSUB: return RC_OPCODE_BGNSUB; */
/* case TGSI_OPCODE_ENDLOOP2: return RC_OPCODE_ENDLOOP2; */
/* case TGSI_OPCODE_ENDSUB: return RC_OPCODE_ENDSUB; */
- /* case TGSI_OPCODE_NOISE1: return RC_OPCODE_NOISE1; */
- /* case TGSI_OPCODE_NOISE2: return RC_OPCODE_NOISE2; */
- /* case TGSI_OPCODE_NOISE3: return RC_OPCODE_NOISE3; */
- /* case TGSI_OPCODE_NOISE4: return RC_OPCODE_NOISE4; */
case TGSI_OPCODE_NOP: return RC_OPCODE_NOP;
/* gap */
/* case TGSI_OPCODE_NRM4: return RC_OPCODE_NRM4; */
@@ -146,7 +142,6 @@ static unsigned translate_opcode(unsigned opcode)
/* case TGSI_OPCODE_IFC: return RC_OPCODE_IFC; */
/* case TGSI_OPCODE_BREAKC: return RC_OPCODE_BREAKC; */
case TGSI_OPCODE_KIL: return RC_OPCODE_KIL;
- case TGSI_OPCODE_SWZ: return RC_OPCODE_SWZ;
}
fprintf(stderr, "Unknown opcode: %i\n", opcode);
@@ -195,10 +190,10 @@ static void transform_dstreg(
struct rc_dst_register * dst,
struct tgsi_full_dst_register * src)
{
- dst->File = translate_register_file(src->DstRegister.File);
- dst->Index = translate_register_index(ttr, src->DstRegister.File, src->DstRegister.Index);
- dst->WriteMask = src->DstRegister.WriteMask;
- dst->RelAddr = src->DstRegister.Indirect;
+ dst->File = translate_register_file(src->Register.File);
+ dst->Index = translate_register_index(ttr, src->Register.File, src->Register.Index);
+ dst->WriteMask = src->Register.WriteMask;
+ dst->RelAddr = src->Register.Indirect;
}
static void transform_srcreg(
@@ -206,23 +201,19 @@ static void transform_srcreg(
struct rc_src_register * dst,
struct tgsi_full_src_register * src)
{
- dst->File = translate_register_file(src->SrcRegister.File);
- dst->Index = translate_register_index(ttr, src->SrcRegister.File, src->SrcRegister.Index);
- dst->RelAddr = src->SrcRegister.Indirect;
- dst->Swizzle = tgsi_util_get_full_src_register_extswizzle(src, 0);
- dst->Swizzle |= tgsi_util_get_full_src_register_extswizzle(src, 1) << 3;
- dst->Swizzle |= tgsi_util_get_full_src_register_extswizzle(src, 2) << 6;
- dst->Swizzle |= tgsi_util_get_full_src_register_extswizzle(src, 3) << 9;
- dst->Abs = src->SrcRegisterExtMod.Absolute;
- dst->Negate =
- src->SrcRegisterExtSwz.NegateX |
- (src->SrcRegisterExtSwz.NegateY << 1) |
- (src->SrcRegisterExtSwz.NegateZ << 2) |
- (src->SrcRegisterExtSwz.NegateW << 3);
- dst->Negate ^= src->SrcRegister.Negate ? RC_MASK_XYZW : 0;
+ dst->File = translate_register_file(src->Register.File);
+ dst->Index = translate_register_index(ttr, src->Register.File, src->Register.Index);
+ dst->RelAddr = src->Register.Indirect;
+ dst->Swizzle = tgsi_util_get_full_src_register_swizzle(src, 0);
+ dst->Swizzle |= tgsi_util_get_full_src_register_swizzle(src, 1) << 3;
+ dst->Swizzle |= tgsi_util_get_full_src_register_swizzle(src, 2) << 6;
+ dst->Swizzle |= tgsi_util_get_full_src_register_swizzle(src, 3) << 9;
+ dst->Abs = src->Register.Absolute;
+ dst->Negate = src->Register.Negate ? RC_MASK_XYZW : 0;
}
-static void transform_texture(struct rc_instruction * dst, struct tgsi_instruction_ext_texture src)
+static void transform_texture(struct rc_instruction * dst, struct tgsi_instruction_texture src,
+ uint32_t *shadowSamplers)
{
switch(src.Texture) {
case TGSI_TEXTURE_1D:
@@ -243,14 +234,17 @@ static void transform_texture(struct rc_instruction * dst, struct tgsi_instructi
case TGSI_TEXTURE_SHADOW1D:
dst->U.I.TexSrcTarget = RC_TEXTURE_1D;
dst->U.I.TexShadow = 1;
+ *shadowSamplers |= 1 << dst->U.I.TexSrcUnit;
break;
case TGSI_TEXTURE_SHADOW2D:
dst->U.I.TexSrcTarget = RC_TEXTURE_2D;
dst->U.I.TexShadow = 1;
+ *shadowSamplers |= 1 << dst->U.I.TexSrcUnit;
break;
case TGSI_TEXTURE_SHADOWRECT:
dst->U.I.TexSrcTarget = RC_TEXTURE_RECT;
dst->U.I.TexShadow = 1;
+ *shadowSamplers |= 1 << dst->U.I.TexSrcUnit;
break;
}
}
@@ -268,17 +262,19 @@ static void transform_instruction(struct tgsi_to_rc * ttr, struct tgsi_full_inst
dst->U.I.SaturateMode = translate_saturate(src->Instruction.Saturate);
if (src->Instruction.NumDstRegs)
- transform_dstreg(ttr, &dst->U.I.DstReg, &src->FullDstRegisters[0]);
+ transform_dstreg(ttr, &dst->U.I.DstReg, &src->Dst[0]);
for(i = 0; i < src->Instruction.NumSrcRegs; ++i) {
- if (src->FullSrcRegisters[i].SrcRegister.File == TGSI_FILE_SAMPLER)
- dst->U.I.TexSrcUnit = src->FullSrcRegisters[i].SrcRegister.Index;
+ if (src->Src[i].Register.File == TGSI_FILE_SAMPLER)
+ dst->U.I.TexSrcUnit = src->Src[i].Register.Index;
else
- transform_srcreg(ttr, &dst->U.I.SrcReg[i], &src->FullSrcRegisters[i]);
+ transform_srcreg(ttr, &dst->U.I.SrcReg[i], &src->Src[i]);
}
/* Texturing. */
- transform_texture(dst, src->InstructionExtTexture);
+ if (src->Instruction.Texture)
+ transform_texture(dst, src->Texture,
+ &ttr->compiler->Program.ShadowSamplers);
}
static void handle_immediate(struct tgsi_to_rc * ttr, struct tgsi_full_immediate * imm)
diff --git a/src/gallium/drivers/r300/r300_vs.c b/src/gallium/drivers/r300/r300_vs.c
index 8460cfaf518..68aef70872e 100644
--- a/src/gallium/drivers/r300/r300_vs.c
+++ b/src/gallium/drivers/r300/r300_vs.c
@@ -1,5 +1,6 @@
/*
* Copyright 2009 Corbin Simpson <[email protected]>
+ * 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"),
@@ -21,91 +22,296 @@
* USE OR OTHER DEALINGS IN THE SOFTWARE. */
#include "r300_vs.h"
+#include "r300_fs.h"
#include "r300_context.h"
+#include "r300_screen.h"
#include "r300_tgsi_to_rc.h"
+#include "r300_reg.h"
#include "tgsi/tgsi_dump.h"
#include "tgsi/tgsi_parse.h"
#include "radeon_compiler.h"
+#include "util/u_math.h"
-static void set_vertex_inputs_outputs(struct r300_vertex_program_compiler * c)
+/* Convert info about VS output semantics into r300_shader_semantics. */
+static void r300_shader_read_vs_outputs(
+ struct tgsi_shader_info* info,
+ struct r300_shader_semantics* vs_outputs)
{
- struct r300_vertex_shader * vs = c->UserData;
- struct tgsi_shader_info* info = &vs->info;
- struct tgsi_parse_context parser;
- struct tgsi_full_declaration * decl;
- boolean pointsize = false;
- int out_colors = 0;
- int colors = 0;
- int out_generic = 0;
- int generic = 0;
int i;
+ unsigned index;
- /* Fill in the input mapping */
- for (i = 0; i < info->num_inputs; i++)
- c->code->inputs[i] = i;
+ r300_shader_semantics_reset(vs_outputs);
- /* Fill in the output mapping */
for (i = 0; i < info->num_outputs; i++) {
+ index = info->output_semantic_index[i];
+
switch (info->output_semantic_name[i]) {
+ case TGSI_SEMANTIC_POSITION:
+ assert(index == 0);
+ vs_outputs->pos = i;
+ break;
+
case TGSI_SEMANTIC_PSIZE:
- pointsize = true;
+ assert(index == 0);
+ vs_outputs->psize = i;
break;
+
case TGSI_SEMANTIC_COLOR:
- out_colors++;
+ assert(index <= ATTR_COLOR_COUNT);
+ vs_outputs->color[index] = i;
break;
- case TGSI_SEMANTIC_FOG:
+
+ case TGSI_SEMANTIC_BCOLOR:
+ assert(index <= ATTR_COLOR_COUNT);
+ vs_outputs->bcolor[index] = i;
+ break;
+
case TGSI_SEMANTIC_GENERIC:
- out_generic++;
+ assert(index <= ATTR_GENERIC_COUNT);
+ vs_outputs->generic[index] = i;
+ break;
+
+ case TGSI_SEMANTIC_FOG:
+ assert(index == 0);
+ vs_outputs->fog = i;
break;
+
+ case TGSI_SEMANTIC_EDGEFLAG:
+ assert(index == 0);
+ fprintf(stderr, "r300 VP: cannot handle edgeflag output\n");
+ assert(0);
+ break;
+ default:
+ assert(0);
}
}
+}
- tgsi_parse_init(&parser, vs->state.tokens);
+static void r300_shader_vap_output_fmt(struct r300_vertex_shader* vs)
+{
+ struct r300_shader_semantics* vs_outputs = &vs->outputs;
+ uint32_t* hwfmt = vs->hwfmt;
+ int i, gen_count;
+ boolean any_bcolor_used = vs_outputs->bcolor[0] != ATTR_UNUSED ||
+ vs_outputs->bcolor[1] != ATTR_UNUSED;
+
+ /* Do the actual vertex_info setup.
+ *
+ * vertex_info has four uints of hardware-specific data in it.
+ * vinfo.hwfmt[0] is R300_VAP_VTX_STATE_CNTL
+ * vinfo.hwfmt[1] is R300_VAP_VSM_VTX_ASSM
+ * vinfo.hwfmt[2] is R300_VAP_OUTPUT_VTX_FMT_0
+ * vinfo.hwfmt[3] is R300_VAP_OUTPUT_VTX_FMT_1 */
+
+ hwfmt[0] = 0x5555; /* XXX this is classic Mesa bonghits */
+
+ /* Position. */
+ if (vs_outputs->pos != ATTR_UNUSED) {
+ hwfmt[1] |= R300_INPUT_CNTL_POS;
+ hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT;
+ } else {
+ assert(0);
+ }
- while (!tgsi_parse_end_of_tokens(&parser)) {
- tgsi_parse_token(&parser);
+ /* Point size. */
+ if (vs_outputs->psize != ATTR_UNUSED) {
+ hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__PT_SIZE_PRESENT;
+ }
- if (parser.FullToken.Token.Type != TGSI_TOKEN_TYPE_DECLARATION)
- continue;
+ /* Colors. */
+ for (i = 0; i < ATTR_COLOR_COUNT; i++) {
+ if (vs_outputs->color[i] != ATTR_UNUSED || any_bcolor_used) {
+ hwfmt[1] |= R300_INPUT_CNTL_COLOR;
+ hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << i;
+ }
+ }
- decl = &parser.FullToken.FullDeclaration;
+ /* Back-face colors. */
+ if (any_bcolor_used) {
+ for (i = 0; i < ATTR_COLOR_COUNT; i++) {
+ hwfmt[1] |= R300_INPUT_CNTL_COLOR;
+ hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << (2+i);
+ }
+ }
- if (decl->Declaration.File != TGSI_FILE_OUTPUT)
- continue;
+ /* Texture coordinates. */
+ gen_count = 0;
+ for (i = 0; i < ATTR_GENERIC_COUNT; i++) {
+ if (vs_outputs->generic[i] != ATTR_UNUSED) {
+ hwfmt[1] |= (R300_INPUT_CNTL_TC0 << gen_count);
+ hwfmt[3] |= (4 << (3 * gen_count));
+ gen_count++;
+ }
+ }
- switch (decl->Semantic.SemanticName) {
- case TGSI_SEMANTIC_POSITION:
- c->code->outputs[decl->DeclarationRange.First] = 0;
- break;
- case TGSI_SEMANTIC_PSIZE:
- c->code->outputs[decl->DeclarationRange.First] = 1;
- break;
- case TGSI_SEMANTIC_COLOR:
- c->code->outputs[decl->DeclarationRange.First] = 1 +
- (pointsize ? 1 : 0) +
- colors++;
- break;
- case TGSI_SEMANTIC_FOG:
- case TGSI_SEMANTIC_GENERIC:
- c->code->outputs[decl->DeclarationRange.First] = 1 +
- (pointsize ? 1 : 0) +
- out_colors +
- generic++;
- break;
- default:
- debug_printf("r300: vs: Bad semantic declaration %d\n",
- decl->Semantic.SemanticName);
- break;
+ /* Fog coordinates. */
+ if (vs_outputs->fog != ATTR_UNUSED) {
+ hwfmt[1] |= (R300_INPUT_CNTL_TC0 << gen_count);
+ hwfmt[3] |= (4 << (3 * gen_count));
+ gen_count++;
+ }
+
+ /* XXX magic */
+ assert(gen_count <= 8);
+
+ /* WPOS. */
+ vs->wpos_tex_output = gen_count;
+}
+
+/* Sets up stream mapping to equivalent VS outputs if TCL is bypassed
+ * or isn't present. */
+static void r300_stream_locations_notcl(
+ struct r300_shader_semantics* vs_outputs,
+ int* stream_loc)
+{
+ int i, tabi = 0, gen_count;
+ boolean any_bcolor_used = vs_outputs->bcolor[0] != ATTR_UNUSED ||
+ vs_outputs->bcolor[1] != ATTR_UNUSED;
+
+ /* Position. */
+ stream_loc[tabi++] = 0;
+
+ /* Point size. */
+ if (vs_outputs->psize != ATTR_UNUSED) {
+ stream_loc[tabi++] = 1;
+ }
+
+ /* Colors. */
+ for (i = 0; i < ATTR_COLOR_COUNT; i++) {
+ if (vs_outputs->color[i] != ATTR_UNUSED || any_bcolor_used) {
+ stream_loc[tabi++] = 2 + i;
+ }
+ }
+
+ /* Back-face colors. */
+ if (any_bcolor_used) {
+ for (i = 0; i < ATTR_COLOR_COUNT; i++) {
+ stream_loc[tabi++] = 4 + i;
}
}
- tgsi_parse_free(&parser);
+ /* Texture coordinates. */
+ gen_count = 0;
+ for (i = 0; i < ATTR_GENERIC_COUNT; i++) {
+ if (vs_outputs->generic[i] != ATTR_UNUSED) {
+ assert(tabi < 16);
+ stream_loc[tabi++] = 6 + gen_count;
+ gen_count++;
+ }
+ }
+
+ /* Fog coordinates. */
+ if (vs_outputs->fog != ATTR_UNUSED) {
+ assert(tabi < 16);
+ stream_loc[tabi++] = 6 + gen_count;
+ gen_count++;
+ }
+
+ /* WPOS. */
+ if (vs_outputs->wpos != ATTR_UNUSED) {
+ assert(tabi < 16);
+ stream_loc[tabi++] = 6 + gen_count;
+ gen_count++;
+ }
+
+ 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;
+ struct r300_shader_semantics* outputs = &vs->outputs;
+ struct tgsi_shader_info* info = &vs->info;
+ int i, reg = 0;
+ boolean any_bcolor_used = outputs->bcolor[0] != ATTR_UNUSED ||
+ outputs->bcolor[1] != ATTR_UNUSED;
+
+ /* Fill in the input mapping */
+ for (i = 0; i < info->num_inputs; i++)
+ c->code->inputs[i] = i;
+
+ /* Position. */
+ if (outputs->pos != ATTR_UNUSED) {
+ c->code->outputs[outputs->pos] = reg++;
+ } else {
+ assert(0);
+ }
+
+ /* Point size. */
+ if (outputs->psize != ATTR_UNUSED) {
+ c->code->outputs[outputs->psize] = reg++;
+ }
+
+ /* If we're writing back facing colors we need to send
+ * four colors to make front/back face colors selection work.
+ * If the vertex program doesn't write all 4 colors, lets
+ * pretend it does by skipping output index reg so the colors
+ * get written into appropriate output vectors.
+ */
+
+ /* Colors. */
+ for (i = 0; i < ATTR_COLOR_COUNT; i++) {
+ if (outputs->color[i] != ATTR_UNUSED) {
+ c->code->outputs[outputs->color[i]] = reg++;
+ } else if (any_bcolor_used) {
+ reg++;
+ }
+ }
+
+ /* Back-face colors. */
+ for (i = 0; i < ATTR_COLOR_COUNT; i++) {
+ if (outputs->bcolor[i] != ATTR_UNUSED) {
+ c->code->outputs[outputs->bcolor[i]] = reg++;
+ } else if (any_bcolor_used) {
+ reg++;
+ }
+ }
+
+ /* Texture coordinates. */
+ for (i = 0; i < ATTR_GENERIC_COUNT; i++) {
+ if (outputs->generic[i] != ATTR_UNUSED) {
+ c->code->outputs[outputs->generic[i]] = reg++;
+ }
+ }
+
+ /* Fog coordinates. */
+ if (outputs->fog != ATTR_UNUSED) {
+ c->code->outputs[outputs->fog] = reg++;
+ }
+
+ /* WPOS. */
+ if (outputs->wpos != ATTR_UNUSED) {
+ c->code->outputs[outputs->wpos] = reg++;
+ }
+}
+
+static void r300_insert_wpos(struct r300_vertex_program_compiler* c,
+ struct r300_shader_semantics* outputs)
+{
+ int i, lastOutput = 0;
+
+ /* Find the max output index. */
+ lastOutput = MAX2(lastOutput, outputs->psize);
+ for (i = 0; i < ATTR_COLOR_COUNT; i++) {
+ lastOutput = MAX2(lastOutput, outputs->color[i]);
+ lastOutput = MAX2(lastOutput, outputs->bcolor[i]);
+ }
+ for (i = 0; i < ATTR_GENERIC_COUNT; i++) {
+ lastOutput = MAX2(lastOutput, outputs->generic[i]);
+ }
+ lastOutput = MAX2(lastOutput, outputs->fog);
+
+ /* Set WPOS after the last output. */
+ lastOutput++;
+ rc_copy_output(&c->Base, 0, lastOutput); /* out[lastOutput] = out[0]; */
+ outputs->wpos = lastOutput;
+}
void r300_translate_vertex_shader(struct r300_context* r300,
struct r300_vertex_shader* vs)
@@ -113,6 +319,9 @@ void r300_translate_vertex_shader(struct r300_context* r300,
struct r300_vertex_program_compiler compiler;
struct tgsi_to_rc ttr;
+ /* Initialize. */
+ r300_shader_read_vs_outputs(&vs->info, &vs->outputs);
+
/* Setup the compiler */
rc_init(&compiler.Base);
@@ -131,13 +340,19 @@ void r300_translate_vertex_shader(struct r300_context* r300,
r300_tgsi_to_rc(&ttr, vs->state.tokens);
- compiler.RequiredOutputs = ~(~0 << vs->info.num_outputs);
+ compiler.RequiredOutputs = ~(~0 << (vs->info.num_outputs+1));
compiler.SetHwInputOutput = &set_vertex_inputs_outputs;
+ /* Insert the WPOS output. */
+ r300_insert_wpos(&compiler, &vs->outputs);
+
+ r300_shader_vap_output_fmt(vs);
+ r300_stream_locations_notcl(&vs->outputs, vs->stream_loc_notcl);
+
/* Invoke the compiler */
r3xx_compile_vertex_program(&compiler);
if (compiler.Base.Error) {
- /* Todo: Fail gracefully */
+ /* XXX We should fallback using Draw. */
fprintf(stderr, "r300 VP: Compiler error\n");
abort();
}
@@ -147,88 +362,29 @@ void r300_translate_vertex_shader(struct r300_context* r300,
vs->translated = TRUE;
}
-
-/* XXX get these to r300_reg */
-#define R300_PVS_DST_OPCODE(x) ((x) << 0)
-# define R300_VE_DOT_PRODUCT 1
-# define R300_VE_MULTIPLY 2
-# define R300_VE_ADD 3
-# define R300_VE_MAXIMUM 7
-# define R300_VE_SET_LESS_THAN 10
-#define R300_PVS_DST_MATH_INST (1 << 6)
-# define R300_ME_RECIP_DX 6
-#define R300_PVS_DST_MACRO_INST (1 << 7)
-# define R300_PVS_MACRO_OP_2CLK_MADD 0
-#define R300_PVS_DST_REG_TYPE(x) ((x) << 8)
-# define R300_PVS_DST_REG_TEMPORARY 0
-# define R300_PVS_DST_REG_A0 1
-# define R300_PVS_DST_REG_OUT 2
-# define R300_PVS_DST_REG_OUT_REPL_X 3
-# define R300_PVS_DST_REG_ALT_TEMPORARY 4
-# define R300_PVS_DST_REG_INPUT 5
-#define R300_PVS_DST_OFFSET(x) ((x) << 13)
-#define R300_PVS_DST_WE(x) ((x) << 20)
-#define R300_PVS_DST_WE_XYZW (0xf << 20)
-
-#define R300_PVS_SRC_REG_TYPE(x) ((x) << 0)
-# define R300_PVS_SRC_REG_TEMPORARY 0
-# define R300_PVS_SRC_REG_INPUT 1
-# define R300_PVS_SRC_REG_CONSTANT 2
-# define R300_PVS_SRC_REG_ALT_TEMPORARY 3
-#define R300_PVS_SRC_OFFSET(x) ((x) << 5)
-#define R300_PVS_SRC_SWIZZLE(x) ((x) << 13)
-# define R300_PVS_SRC_SELECT_X 0
-# define R300_PVS_SRC_SELECT_Y 1
-# define R300_PVS_SRC_SELECT_Z 2
-# define R300_PVS_SRC_SELECT_W 3
-# define R300_PVS_SRC_SELECT_FORCE_0 4
-# define R300_PVS_SRC_SELECT_FORCE_1 5
-# define R300_PVS_SRC_SWIZZLE_XYZW \
- ((R300_PVS_SRC_SELECT_X | (R300_PVS_SRC_SELECT_Y << 3) | \
- (R300_PVS_SRC_SELECT_Z << 6) | (R300_PVS_SRC_SELECT_W << 9)) << 13)
-# define R300_PVS_SRC_SWIZZLE_ZERO \
- ((R300_PVS_SRC_SELECT_FORCE_0 | (R300_PVS_SRC_SELECT_FORCE_0 << 3) | \
- (R300_PVS_SRC_SELECT_FORCE_0 << 6) | \
- (R300_PVS_SRC_SELECT_FORCE_0 << 9)) << 13)
-# define R300_PVS_SRC_SWIZZLE_ONE \
- ((R300_PVS_SRC_SELECT_FORCE_1 | (R300_PVS_SRC_SELECT_FORCE_1 << 3) | \
- (R300_PVS_SRC_SELECT_FORCE_1 << 6) | \
- (R300_PVS_SRC_SELECT_FORCE_1 << 9)) << 13)
-#define R300_PVS_MODIFIER_X (1 << 25)
-#define R300_PVS_MODIFIER_Y (1 << 26)
-#define R300_PVS_MODIFIER_Z (1 << 27)
-#define R300_PVS_MODIFIER_W (1 << 28)
-#define R300_PVS_NEGATE_XYZW \
- (R300_PVS_MODIFIER_X | R300_PVS_MODIFIER_Y | \
- R300_PVS_MODIFIER_Z | R300_PVS_MODIFIER_W)
-
-struct r300_vertex_program_code r300_passthrough_vertex_shader = {
- .length = 8, /* two instructions */
-
- /* MOV out[0], in[0] */
- .body.d[0] = R300_PVS_DST_OPCODE(R300_VE_ADD) |
- R300_PVS_DST_REG_TYPE(R300_PVS_DST_REG_OUT) |
- R300_PVS_DST_OFFSET(0) | R300_PVS_DST_WE_XYZW,
- .body.d[1] = R300_PVS_SRC_REG_TYPE(R300_PVS_SRC_REG_INPUT) |
- R300_PVS_SRC_OFFSET(0) | R300_PVS_SRC_SWIZZLE_XYZW,
- .body.d[2] = R300_PVS_SRC_SWIZZLE_ZERO,
- .body.d[3] = 0x0,
-
- /* MOV out[1], in[1] */
- .body.d[4] = R300_PVS_DST_OPCODE(R300_VE_ADD) |
- R300_PVS_DST_REG_TYPE(R300_PVS_DST_REG_OUT) |
- R300_PVS_DST_OFFSET(1) | R300_PVS_DST_WE_XYZW,
- .body.d[5] = R300_PVS_SRC_REG_TYPE(R300_PVS_SRC_REG_INPUT) |
- R300_PVS_SRC_OFFSET(1) | R300_PVS_SRC_SWIZZLE_XYZW,
- .body.d[6] = R300_PVS_SRC_SWIZZLE_ZERO,
- .body.d[7] = 0x0,
-
- .inputs[0] = 0,
- .inputs[1] = 1,
- .outputs[0] = 0,
- .outputs[1] = 1,
-
- .InputsRead = 3,
- .OutputsWritten = 3
-};
-
+boolean r300_vertex_shader_setup_wpos(struct r300_context* r300)
+{
+ struct r300_vertex_shader* vs = r300->vs;
+ int tex_output = r300->vs->wpos_tex_output;
+ uint32_t tex_fmt = R300_INPUT_CNTL_TC0 << tex_output;
+ uint32_t* hwfmt = vs->hwfmt;
+
+ if (r300->fs->inputs.wpos != ATTR_UNUSED) {
+ /* Enable WPOS in VAP. */
+ if (!(hwfmt[1] & tex_fmt)) {
+ hwfmt[1] |= tex_fmt;
+ hwfmt[3] |= (4 << (3 * tex_output));
+
+ assert(tex_output < 8);
+ return TRUE;
+ }
+ } else {
+ /* Disable WPOS in VAP. */
+ if (hwfmt[1] & tex_fmt) {
+ hwfmt[1] &= ~tex_fmt;
+ hwfmt[3] &= ~(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 2a4ce315e32..18cfeee3cd4 100644
--- a/src/gallium/drivers/r300/r300_vs.h
+++ b/src/gallium/drivers/r300/r300_vs.h
@@ -1,5 +1,6 @@
/*
* Copyright 2009 Corbin Simpson <[email protected]>
+ * 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"),
@@ -25,18 +26,25 @@
#include "pipe/p_state.h"
#include "tgsi/tgsi_scan.h"
-
#include "radeon_code.h"
+#include "r300_shader_semantics.h"
+
struct r300_context;
struct r300_vertex_shader {
/* Parent class */
struct pipe_shader_state state;
+
struct tgsi_shader_info info;
+ struct r300_shader_semantics outputs;
+ uint hwfmt[4];
+
+ /* Stream locations for SWTCL or if TCL is bypassed. */
+ int stream_loc_notcl[16];
- /* Fallback shader, because Draw has issues */
- struct draw_vertex_shader* draw;
+ /* Output stream location for WPOS. */
+ int wpos_tex_output;
/* Has this shader been translated yet? */
boolean translated;
@@ -45,10 +53,10 @@ struct r300_vertex_shader {
struct r300_vertex_program_code code;
};
-
-extern struct r300_vertex_program_code r300_passthrough_vertex_shader;
-
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);
+
#endif /* R300_VS_H */
diff --git a/src/gallium/drivers/r300/r300_winsys.h b/src/gallium/drivers/r300/r300_winsys.h
index f18ad75a47d..1ae6de70fee 100644
--- a/src/gallium/drivers/r300/r300_winsys.h
+++ b/src/gallium/drivers/r300/r300_winsys.h
@@ -35,67 +35,10 @@ extern "C" {
#include "pipe/p_state.h"
#include "pipe/internal/p_winsys_screen.h"
-struct r300_winsys {
- /* Parent class */
- struct pipe_winsys base;
-
- /* Opaque Radeon-specific winsys object. */
- void* radeon_winsys;
-
- /* PCI ID */
- uint32_t pci_id;
-
- /* GB pipe count */
- uint32_t gb_pipes;
-
- /* GART size. */
- uint32_t gart_size;
-
- /* VRAM size. */
- uint32_t vram_size;
-
- /* Add a pipe_buffer to the list of buffer objects to validate. */
- boolean (*add_buffer)(struct r300_winsys* winsys,
- struct pipe_buffer* pbuffer,
- uint32_t rd,
- uint32_t wd);
-
- /* Revalidate all currently setup pipe_buffers.
- * Returns TRUE if a flush is required. */
- boolean (*validate)(struct r300_winsys* winsys);
-
- /* Check to see if there's room for commands. */
- boolean (*check_cs)(struct r300_winsys* winsys, int size);
-
- /* Start a command emit. */
- void (*begin_cs)(struct r300_winsys* winsys,
- int size,
- const char* file,
- const char* function,
- int line);
-
- /* Write a dword to the command buffer. */
- void (*write_cs_dword)(struct r300_winsys* winsys, uint32_t dword);
-
- /* Write a relocated dword to the command buffer. */
- void (*write_cs_reloc)(struct r300_winsys* winsys,
- struct pipe_buffer* bo,
- uint32_t rd,
- uint32_t wd,
- uint32_t flags);
-
- /* Finish a command emit. */
- void (*end_cs)(struct r300_winsys* winsys,
- const char* file,
- const char* function,
- int line);
-
- /* Flush the CS. */
- void (*flush_cs)(struct r300_winsys* winsys);
-};
+#include "radeon_winsys.h"
struct pipe_context* r300_create_context(struct pipe_screen* screen,
- struct r300_winsys* r300_winsys);
+ struct radeon_winsys* radeon_winsys);
boolean r300_get_texture_buffer(struct pipe_texture* texture,
struct pipe_buffer** buffer,
diff --git a/src/gallium/drivers/r300/r3xx_fs.c b/src/gallium/drivers/r300/r3xx_fs.c
deleted file mode 100644
index c1c1194d58e..00000000000
--- a/src/gallium/drivers/r300/r3xx_fs.c
+++ /dev/null
@@ -1,74 +0,0 @@
-/*
- * Copyright 2008 Corbin Simpson <[email protected]>
- * Joakim Sindholt <[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 "r3xx_fs.h"
-
-#include "r300_reg.h"
-
-struct rX00_fragment_program_code r3xx_passthrough_fragment_shader = {
- .code.r300.alu.length = 1,
- .code.r300.tex.length = 0,
-
- .code.r300.config = 0,
- .code.r300.pixsize = 0,
- .code.r300.code_offset = 0,
- .code.r300.code_addr[3] = R300_RGBA_OUT,
-
- .code.r300.alu.inst[0].rgb_inst = R300_RGB_SWIZA(R300_ALU_ARGC_SRC0C_XYZ) |
- R300_RGB_SWIZB(R300_ALU_ARGC_SRC0C_XYZ) |
- R300_RGB_SWIZC(R300_ALU_ARGC_ZERO) |
- R300_ALU_OUTC_CMP,
- .code.r300.alu.inst[0].rgb_addr = R300_RGB_ADDR0(0) | R300_RGB_ADDR1(0) |
- R300_RGB_ADDR2(0) | R300_ALU_DSTC_OUTPUT_XYZ,
- .code.r300.alu.inst[0].alpha_inst = R300_ALPHA_SWIZA(R300_ALU_ARGA_SRC0A) |
- R300_ALPHA_SWIZB(R300_ALU_ARGA_SRC0A) |
- R300_ALPHA_SWIZC(R300_ALU_ARGA_ZERO) |
- R300_ALU_OUTA_CMP,
- .code.r300.alu.inst[0].alpha_addr = R300_ALPHA_ADDR0(0) |
- R300_ALPHA_ADDR1(0) | R300_ALPHA_ADDR2(0) | R300_ALU_DSTA_OUTPUT,
-};
-
-struct rX00_fragment_program_code r3xx_texture_fragment_shader = {
- .code.r300.alu.length = 1,
- .code.r300.tex.length = 1,
-
- .code.r300.config = R300_PFS_CNTL_FIRST_NODE_HAS_TEX,
- .code.r300.pixsize = 0,
- .code.r300.code_offset = 0,
- .code.r300.code_addr[3] = R300_RGBA_OUT,
-
- .code.r300.tex.inst[0] = R300_TEX_OP_LD << R300_TEX_INST_SHIFT,
-
- .code.r300.alu.inst[0].rgb_inst = R300_RGB_SWIZA(R300_ALU_ARGC_SRC0C_XYZ) |
- R300_RGB_SWIZB(R300_ALU_ARGC_SRC0C_XYZ) |
- R300_RGB_SWIZC(R300_ALU_ARGC_ZERO) |
- R300_ALU_OUTC_CMP,
- .code.r300.alu.inst[0].rgb_addr = R300_RGB_ADDR0(0) | R300_RGB_ADDR1(0) |
- R300_RGB_ADDR2(0) | R300_ALU_DSTC_OUTPUT_XYZ,
- .code.r300.alu.inst[0].alpha_inst = R300_ALPHA_SWIZA(R300_ALU_ARGA_SRC0A) |
- R300_ALPHA_SWIZB(R300_ALU_ARGA_SRC0A) |
- R300_ALPHA_SWIZC(R300_ALU_ARGA_ZERO) |
- R300_ALU_OUTA_CMP,
- .code.r300.alu.inst[0].alpha_addr = R300_ALPHA_ADDR0(0) |
- R300_ALPHA_ADDR1(0) | R300_ALPHA_ADDR2(0) | R300_ALU_DSTA_OUTPUT,
-};
diff --git a/src/gallium/drivers/r300/r3xx_fs.h b/src/gallium/drivers/r300/r3xx_fs.h
deleted file mode 100644
index 51cd245724d..00000000000
--- a/src/gallium/drivers/r300/r3xx_fs.h
+++ /dev/null
@@ -1,32 +0,0 @@
-/*
- * Copyright 2008 Corbin Simpson <[email protected]>
- * Joakim Sindholt <[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 R3XX_FS_H
-#define R3XX_FS_H
-
-#include "radeon_code.h"
-
-struct rX00_fragment_program_code r3xx_passthrough_fragment_shader;
-struct rX00_fragment_program_code r3xx_texture_fragment_shader;
-
-#endif /* R3XX_FS_H */
diff --git a/src/gallium/drivers/r300/r5xx_fs.c b/src/gallium/drivers/r300/r5xx_fs.c
deleted file mode 100644
index f072deab0d9..00000000000
--- a/src/gallium/drivers/r300/r5xx_fs.c
+++ /dev/null
@@ -1,125 +0,0 @@
-/*
- * Copyright 2008 Corbin Simpson <[email protected]>
- * Joakim Sindholt <[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 "r5xx_fs.h"
-
-#include "r300_reg.h"
-
-/* XXX this all should find its way back to r300_reg */
-/* Swizzle tools */
-#define R500_SWIZZLE_ZERO 4
-#define R500_SWIZZLE_HALF 5
-#define R500_SWIZZLE_ONE 6
-#define R500_SWIZ_RGB_ZERO ((4 << 0) | (4 << 3) | (4 << 6))
-#define R500_SWIZ_RGB_ONE ((6 << 0) | (6 << 3) | (6 << 6))
-#define R500_SWIZ_RGB_RGB ((0 << 0) | (1 << 3) | (2 << 6))
-#define R500_SWIZ_MOD_NEG 1
-#define R500_SWIZ_MOD_ABS 2
-#define R500_SWIZ_MOD_NEG_ABS 3
-/* Swizzles for inst2 */
-#define R500_SWIZ_TEX_STRQ(x) ((x) << 8)
-#define R500_SWIZ_TEX_RGBA(x) ((x) << 24)
-/* Swizzles for inst3 */
-#define R500_SWIZ_RGB_A(x) ((x) << 2)
-#define R500_SWIZ_RGB_B(x) ((x) << 15)
-/* Swizzles for inst4 */
-#define R500_SWIZ_ALPHA_A(x) ((x) << 14)
-#define R500_SWIZ_ALPHA_B(x) ((x) << 21)
-/* Swizzle for inst5 */
-#define R500_SWIZ_RGBA_C(x) ((x) << 14)
-#define R500_SWIZ_ALPHA_C(x) ((x) << 27)
-/* Writemasks */
-#define R500_TEX_WMASK(x) ((x) << 11)
-#define R500_ALU_WMASK(x) ((x) << 11)
-#define R500_ALU_OMASK(x) ((x) << 15)
-#define R500_W_OMASK (1 << 31)
-
-struct rX00_fragment_program_code r5xx_passthrough_fragment_shader = {
- .code.r500.max_temp_idx = 0,
- .code.r500.inst_end = 0,
-
- .code.r500.inst[0].inst0 = R500_INST_TYPE_OUT |
- R500_INST_TEX_SEM_WAIT | R500_INST_LAST |
- R500_INST_RGB_OMASK_RGB | R500_INST_ALPHA_OMASK |
- R500_INST_RGB_CLAMP | R500_INST_ALPHA_CLAMP,
- .code.r500.inst[0].inst1 =
- R500_RGB_ADDR0(0) | R500_RGB_ADDR1(0) | R500_RGB_ADDR1_CONST |
- R500_RGB_ADDR2(0) | R500_RGB_ADDR2_CONST,
- .code.r500.inst[0].inst2 =
- R500_ALPHA_ADDR0(0) | R500_ALPHA_ADDR1(0) | R500_ALPHA_ADDR1_CONST |
- R500_ALPHA_ADDR2(0) | R500_ALPHA_ADDR2_CONST,
- .code.r500.inst[0].inst3 =
- R500_ALU_RGB_SEL_A_SRC0 | R500_ALU_RGB_R_SWIZ_A_R |
- R500_ALU_RGB_G_SWIZ_A_G | R500_ALU_RGB_B_SWIZ_A_B |
- R500_ALU_RGB_SEL_B_SRC0 | R500_ALU_RGB_R_SWIZ_B_R |
- R500_ALU_RGB_B_SWIZ_B_G | R500_ALU_RGB_G_SWIZ_B_B,
- .code.r500.inst[0].inst4 =
- R500_ALPHA_OP_CMP | R500_ALPHA_SWIZ_A_A | R500_ALPHA_SWIZ_B_A,
- .code.r500.inst[0].inst5 =
- R500_ALU_RGBA_OP_CMP | R500_ALU_RGBA_R_SWIZ_0 |
- R500_ALU_RGBA_G_SWIZ_0 | R500_ALU_RGBA_B_SWIZ_0 |
- R500_ALU_RGBA_A_SWIZ_0,
-};
-
-struct rX00_fragment_program_code r5xx_texture_fragment_shader = {
- .code.r500.max_temp_idx = 0,
- .code.r500.inst_end = 1,
-
- .code.r500.inst[0].inst0 = R500_INST_TYPE_TEX |
- R500_INST_TEX_SEM_WAIT |
- R500_INST_RGB_WMASK_RGB | R500_INST_ALPHA_WMASK |
- R500_INST_RGB_CLAMP | R500_INST_ALPHA_CLAMP,
- .code.r500.inst[0].inst1 = R500_TEX_ID(0) | R500_TEX_INST_LD |
- R500_TEX_SEM_ACQUIRE | R500_TEX_IGNORE_UNCOVERED,
- .code.r500.inst[0].inst2 = R500_TEX_SRC_ADDR(0) |
- R500_TEX_SRC_S_SWIZ_R | R500_TEX_SRC_T_SWIZ_G |
- R500_TEX_SRC_R_SWIZ_B | R500_TEX_SRC_Q_SWIZ_A |
- R500_TEX_DST_ADDR(0) |
- R500_TEX_DST_R_SWIZ_R | R500_TEX_DST_G_SWIZ_G |
- R500_TEX_DST_B_SWIZ_B | R500_TEX_DST_A_SWIZ_A,
- .code.r500.inst[0].inst3 = 0x0,
- .code.r500.inst[0].inst4 = 0x0,
- .code.r500.inst[0].inst5 = 0x0,
-
- .code.r500.inst[1].inst0 = R500_INST_TYPE_OUT |
- R500_INST_TEX_SEM_WAIT | R500_INST_LAST |
- R500_INST_RGB_OMASK_RGB | R500_INST_ALPHA_OMASK |
- R500_INST_RGB_CLAMP | R500_INST_ALPHA_CLAMP,
- .code.r500.inst[1].inst1 =
- R500_RGB_ADDR0(0) | R500_RGB_ADDR1(0) | R500_RGB_ADDR1_CONST |
- R500_RGB_ADDR2(0) | R500_RGB_ADDR2_CONST,
- .code.r500.inst[1].inst2 =
- R500_ALPHA_ADDR0(0) | R500_ALPHA_ADDR1(0) | R500_ALPHA_ADDR1_CONST |
- R500_ALPHA_ADDR2(0) | R500_ALPHA_ADDR2_CONST,
- .code.r500.inst[1].inst3 =
- R500_ALU_RGB_SEL_A_SRC0 | R500_ALU_RGB_R_SWIZ_A_R |
- R500_ALU_RGB_G_SWIZ_A_G | R500_ALU_RGB_B_SWIZ_A_B |
- R500_ALU_RGB_SEL_B_SRC0 | R500_ALU_RGB_R_SWIZ_B_R |
- R500_ALU_RGB_B_SWIZ_B_G | R500_ALU_RGB_G_SWIZ_B_B,
- .code.r500.inst[1].inst4 =
- R500_ALPHA_OP_CMP | R500_ALPHA_SWIZ_A_A | R500_ALPHA_SWIZ_B_A,
- .code.r500.inst[1].inst5 =
- R500_ALU_RGBA_OP_CMP | R500_ALU_RGBA_R_SWIZ_0 |
- R500_ALU_RGBA_G_SWIZ_0 | R500_ALU_RGBA_B_SWIZ_0 |
- R500_ALU_RGBA_A_SWIZ_0,
-};
diff --git a/src/gallium/drivers/r300/r5xx_fs.h b/src/gallium/drivers/r300/r5xx_fs.h
deleted file mode 100644
index a4addde32b2..00000000000
--- a/src/gallium/drivers/r300/r5xx_fs.h
+++ /dev/null
@@ -1,32 +0,0 @@
-/*
- * Copyright 2008 Corbin Simpson <[email protected]>
- * Joakim Sindholt <[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 R5XX_FS_H
-#define R5XX_FS_H
-
-#include "radeon_code.h"
-
-struct rX00_fragment_program_code r5xx_passthrough_fragment_shader;
-struct rX00_fragment_program_code r5xx_texture_fragment_shader;
-
-#endif /* R5XX_FS_H */