summaryrefslogtreecommitdiff
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/Makefile3
-rw-r--r--src/gallium/drivers/r300/r300_context.c40
-rw-r--r--src/gallium/drivers/r300/r300_context.h23
-rw-r--r--src/gallium/drivers/r300/r300_cs.h7
-rw-r--r--src/gallium/drivers/r300/r300_emit.c117
-rw-r--r--src/gallium/drivers/r300/r300_emit.h2
-rw-r--r--src/gallium/drivers/r300/r300_fs.c85
-rw-r--r--src/gallium/drivers/r300/r300_fs.h6
-rw-r--r--src/gallium/drivers/r300/r300_reg.h89
-rw-r--r--src/gallium/drivers/r300/r300_render.c27
-rw-r--r--src/gallium/drivers/r300/r300_screen.c22
-rw-r--r--src/gallium/drivers/r300/r300_screen.h4
-rw-r--r--src/gallium/drivers/r300/r300_shader_semantics.h64
-rw-r--r--src/gallium/drivers/r300/r300_state.c93
-rw-r--r--src/gallium/drivers/r300/r300_state_derived.c647
-rw-r--r--src/gallium/drivers/r300/r300_state_invariant.c4
-rw-r--r--src/gallium/drivers/r300/r300_texture.c55
-rw-r--r--src/gallium/drivers/r300/r300_tgsi_to_rc.c31
-rw-r--r--src/gallium/drivers/r300/r300_vbo.c47
-rw-r--r--src/gallium/drivers/r300/r300_vs.c250
-rw-r--r--src/gallium/drivers/r300/r300_vs.h14
-rw-r--r--src/gallium/drivers/r300/r300_winsys.h70
22 files changed, 966 insertions, 734 deletions
diff --git a/src/gallium/drivers/r300/Makefile b/src/gallium/drivers/r300/Makefile
index 121b65063f..9c9fc6f64b 100644
--- a/src/gallium/drivers/r300/Makefile
+++ b/src/gallium/drivers/r300/Makefile
@@ -23,7 +23,8 @@ C_SOURCES = \
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
diff --git a/src/gallium/drivers/r300/r300_context.c b/src/gallium/drivers/r300/r300_context.c
index ae23329b83..5b337f03ac 100644
--- a/src/gallium/drivers/r300/r300_context.c
+++ b/src/gallium/drivers/r300/r300_context.c
@@ -36,6 +36,7 @@
#include "r300_screen.h"
#include "r300_state_derived.h"
#include "r300_state_invariant.h"
+#include "r300_texture.h"
#include "r300_winsys.h"
static enum pipe_error r300_clear_hash_table(void* key, void* value,
@@ -69,6 +70,7 @@ static void r300_destroy_context(struct pipe_context* context)
FREE(r300->blend_color_state);
FREE(r300->rs_block);
FREE(r300->scissor_state);
+ FREE(r300->vertex_info);
FREE(r300->viewport_state);
FREE(r300);
}
@@ -104,7 +106,7 @@ static void r300_flush_cb(void *data)
}
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);
@@ -112,9 +114,9 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen,
if (!r300)
return NULL;
- r300->winsys = r300_winsys;
+ r300->winsys = radeon_winsys;
- r300->context.winsys = (struct pipe_winsys*)r300_winsys;
+ r300->context.winsys = (struct pipe_winsys*)radeon_winsys;
r300->context.screen = screen;
r300_init_debug(r300);
@@ -123,15 +125,24 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen,
r300->context.clear = r300_clear;
- if (r300screen->caps->has_tcl)
- {
+ 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
- {
- assert(0);
+ } 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;
@@ -143,18 +154,9 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen,
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);
diff --git a/src/gallium/drivers/r300/r300_context.h b/src/gallium/drivers/r300/r300_context.h
index f954ba7f9a..0be190392a 100644
--- a/src/gallium/drivers/r300/r300_context.h
+++ b/src/gallium/drivers/r300/r300_context.h
@@ -92,11 +92,23 @@ struct r300_sampler_state {
uint32_t filter0; /* R300_TX_FILTER0: 0x4400 */
uint32_t filter1; /* R300_TX_FILTER1: 0x4440 */
uint32_t border_color; /* R300_TX_BORDER_COLOR: 0x45c0 */
+
+ /* Min/max LOD must be clamped to [0, last_level], thus
+ * it's dependent on a currently bound texture */
+ 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 {
@@ -219,11 +231,6 @@ struct r300_texture {
struct r300_vertex_info {
/* Parent class */
struct vertex_info vinfo;
- /* 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];
/* R300_VAP_PROG_STREAK_CNTL_[0-7] */
uint32_t vap_prog_stream_cntl[8];
@@ -238,7 +245,7 @@ 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;
diff --git a/src/gallium/drivers/r300/r300_cs.h b/src/gallium/drivers/r300/r300_cs.h
index 86ba91db52..9fcf3ab538 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
@@ -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); \
diff --git a/src/gallium/drivers/r300/r300_emit.c b/src/gallium/drivers/r300/r300_emit.c
index eeb97a2d37..dbf316a9b5 100644
--- a/src/gallium/drivers/r300/r300_emit.c
+++ b/src/gallium/drivers/r300/r300_emit.c
@@ -129,7 +129,9 @@ 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];
@@ -137,11 +139,31 @@ 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;
+
+ 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
@@ -360,8 +382,6 @@ static void r300_emit_query_start(struct r300_context *r300)
if (!query)
return;
- /* XXX This will almost certainly not return good results
- * for overlapping queries. */
BEGIN_CS(4);
if (caps->family == CHIP_FAMILY_RV530) {
OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL);
@@ -543,24 +563,36 @@ void r300_emit_rs_block_state(struct r300_context* r300,
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 */
@@ -569,13 +601,20 @@ void r300_emit_texture(struct r300_context* r300,
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), 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);
@@ -600,10 +639,10 @@ void r300_emit_aos(struct r300_context* r300, unsigned offset)
for (i = 0; i < aos_count - 1; i += 2) {
int buf_num1 = velem[i].vertex_buffer_index;
int buf_num2 = velem[i+1].vertex_buffer_index;
- assert(vbuf[buf_num1].stride % 4 == 0 && pf_get_size(velem[i].src_format) % 4 == 0);
- assert(vbuf[buf_num2].stride % 4 == 0 && pf_get_size(velem[i+1].src_format) % 4 == 0);
- OUT_CS((pf_get_size(velem[i].src_format) >> 2) | (vbuf[buf_num1].stride << 6) |
- (pf_get_size(velem[i+1].src_format) << 14) | (vbuf[buf_num2].stride << 22));
+ assert(vbuf[buf_num1].stride % 4 == 0 && pf_get_blocksize(velem[i].src_format) % 4 == 0);
+ assert(vbuf[buf_num2].stride % 4 == 0 && pf_get_blocksize(velem[i+1].src_format) % 4 == 0);
+ OUT_CS((pf_get_blocksize(velem[i].src_format) >> 2) | (vbuf[buf_num1].stride << 6) |
+ (pf_get_blocksize(velem[i+1].src_format) << 14) | (vbuf[buf_num2].stride << 22));
OUT_CS(vbuf[buf_num1].buffer_offset + velem[i].src_offset +
offset * vbuf[buf_num1].stride);
OUT_CS(vbuf[buf_num2].buffer_offset + velem[i+1].src_offset +
@@ -611,8 +650,8 @@ void r300_emit_aos(struct r300_context* r300, unsigned offset)
}
if (aos_count & 1) {
int buf_num = velem[i].vertex_buffer_index;
- assert(vbuf[buf_num].stride % 4 == 0 && pf_get_size(velem[i].src_format) % 4 == 0);
- OUT_CS((pf_get_size(velem[i].src_format) >> 2) | (vbuf[buf_num].stride << 6));
+ assert(vbuf[buf_num].stride % 4 == 0 && pf_get_blocksize(velem[i].src_format) % 4 == 0);
+ OUT_CS((pf_get_blocksize(velem[i].src_format) >> 2) | (vbuf[buf_num].stride << 6));
OUT_CS(vbuf[buf_num].buffer_offset + velem[i].src_offset +
offset * vbuf[buf_num].stride);
}
@@ -690,12 +729,22 @@ void r300_emit_vertex_format_state(struct r300_context* r300)
END_CS;
}
+
void r300_emit_vertex_program_code(struct r300_context* r300,
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) {
@@ -708,8 +757,7 @@ void r300_emit_vertex_program_code(struct r300_context* r300,
/* 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) |
@@ -722,10 +770,11 @@ void r300_emit_vertex_program_code(struct r300_context* r300,
for (i = 0; i < code->length; i++)
OUT_CS(code->body.d[i]);
- 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));
+ R300_PVS_VF_MAX_VTX_NUM(12) |
+ (r300screen->caps->is_r500 ? R500_TCL_STATE_OPTIMIZATION : 0));
END_CS;
}
@@ -790,13 +839,22 @@ void r300_emit_viewport_state(struct r300_context* r300,
END_CS;
}
+void r300_emit_texture_count(struct r300_context* r300)
+{
+ CS_LOCALS(r300);
+
+ BEGIN_CS(2);
+ OUT_CS_REG(R300_TX_ENABLE, (1 << r300->texture_count) - 1);
+ 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;
}
@@ -821,10 +879,17 @@ void r300_emit_dirty_state(struct r300_context* r300)
return;
}
+ /* 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++) {
@@ -950,6 +1015,8 @@ 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))) {
diff --git a/src/gallium/drivers/r300/r300_emit.h b/src/gallium/drivers/r300/r300_emit.h
index 7c83c5166d..3797d3d332 100644
--- a/src/gallium/drivers/r300/r300_emit.h
+++ b/src/gallium/drivers/r300/r300_emit.h
@@ -92,6 +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_fs.c b/src/gallium/drivers/r300/r300_fs.c
index 29ddc84c41..79b01bb4dc 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 <MostAwesomeDude@gmail.com>
* Joakim Sindholt <opensource@zhasha.com>
+ * Copyright 2009 Marek Olšák <maraeo@gmail.com>
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
@@ -31,6 +32,41 @@
#include "radeon_code.h"
#include "radeon_compiler.h"
+/* Convert info about FS input semantics to r300_shader_semantics. */
+static 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;
+
+ default:
+ assert(0);
+ }
+ }
+}
+
+
static void find_output_registers(struct r300_fragment_program_compiler * compiler,
struct r300_fragment_shader * fs)
{
@@ -58,38 +94,24 @@ 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_fragment_shader*)c->UserData)->inputs;
+ 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 < 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;
+ 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++);
+ }
}
void r300_translate_fragment_shader(struct r300_context* r300,
@@ -98,6 +120,10 @@ void r300_translate_fragment_shader(struct r300_context* r300,
struct r300_fragment_program_compiler compiler;
struct tgsi_to_rc ttr;
+ /* Initialize. */
+ r300_shader_read_fs_inputs(&fs->info, &fs->inputs);
+
+ /* Setup the compiler. */
memset(&compiler, 0, sizeof(compiler));
rc_init(&compiler.Base);
compiler.Base.Debug = DBG_ON(r300, DBG_FP);
@@ -107,7 +133,7 @@ void r300_translate_fragment_shader(struct r300_context* r300,
compiler.AllocateHwInputs = &allocate_hardware_inputs;
compiler.UserData = fs;
- /* TODO: Program compilation depends on texture compare modes,
+ /* XXX: 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.
*
@@ -133,6 +159,7 @@ void r300_translate_fragment_shader(struct r300_context* r300,
/* XXX failover maybe? */
DBG(r300, DBG_FP, "r300: Error compiling fragment program: %s\n",
compiler.Base.ErrorMsg);
+ assert(0);
}
/* And, finally... */
diff --git a/src/gallium/drivers/r300/r300_fs.h b/src/gallium/drivers/r300/r300_fs.h
index e831c30301..630e2d0c8a 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 <MostAwesomeDude@gmail.com>
* Joakim Sindholt <opensource@zhasha.com>
+ * Copyright 2009 Marek Olšák <maraeo@gmail.com>
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
@@ -25,15 +26,16 @@
#define R300_FS_H
#include "pipe/p_state.h"
-
#include "tgsi/tgsi_scan.h"
-
#include "radeon_code.h"
+#include "r300_shader_semantics.h"
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;
diff --git a/src/gallium/drivers/r300/r300_reg.h b/src/gallium/drivers/r300/r300_reg.h
index 8ca785cb58..85b1ea568a 100644
--- a/src/gallium/drivers/r300/r300_reg.h
+++ b/src/gallium/drivers/r300/r300_reg.h
@@ -661,20 +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
@@ -700,9 +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
@@ -725,39 +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])
@@ -1293,7 +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)
@@ -1463,6 +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)
@@ -1471,6 +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
@@ -1500,8 +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)
diff --git a/src/gallium/drivers/r300/r300_render.c b/src/gallium/drivers/r300/r300_render.c
index 62e1456ed3..35b335df6a 100644
--- a/src/gallium/drivers/r300/r300_render.c
+++ b/src/gallium/drivers/r300/r300_render.c
@@ -70,6 +70,12 @@ uint32_t r300_translate_primitive(unsigned prim)
}
}
+static boolean r300_nothing_to_draw(struct r300_context *r300)
+{
+ return r300->rs_state->rs.scissor &&
+ r300->scissor_state->scissor.empty_area;
+}
+
static void r300_emit_draw_arrays(struct r300_context *r300,
unsigned mode,
unsigned count)
@@ -173,18 +179,21 @@ boolean r300_draw_range_elements(struct pipe_context* pipe,
return FALSE;
}
+
if (count > 65535) {
return FALSE;
}
+ if (r300_nothing_to_draw(r300)) {
+ return TRUE;
+ }
+
r300_update_derived_state(r300);
if (!r300_setup_vertex_buffers(r300)) {
return FALSE;
}
- setup_vertex_attributes(r300);
-
setup_index_buffer(r300, indexBuffer, indexSize);
r300_emit_dirty_state(r300);
@@ -220,14 +229,16 @@ boolean r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
return FALSE;
}
+ if (r300_nothing_to_draw(r300)) {
+ return TRUE;
+ }
+
r300_update_derived_state(r300);
if (!r300_setup_vertex_buffers(r300)) {
return FALSE;
}
- setup_vertex_attributes(r300);
-
r300_emit_dirty_state(r300);
r300_emit_aos(r300, start);
@@ -255,6 +266,10 @@ boolean r300_swtcl_draw_arrays(struct pipe_context* pipe,
return FALSE;
}
+ if (r300_nothing_to_draw(r300)) {
+ return TRUE;
+ }
+
for (i = 0; i < r300->vertex_buffer_count; i++) {
void* buf = pipe_buffer_map(pipe->screen,
r300->vertex_buffer[i].buffer,
@@ -296,6 +311,10 @@ boolean r300_swtcl_draw_range_elements(struct pipe_context* pipe,
return FALSE;
}
+ if (r300_nothing_to_draw(r300)) {
+ return TRUE;
+ }
+
for (i = 0; i < r300->vertex_buffer_count; i++) {
void* buf = pipe_buffer_map(pipe->screen,
r300->vertex_buffer[i].buffer,
diff --git a/src/gallium/drivers/r300/r300_screen.c b/src/gallium/drivers/r300/r300_screen.c
index 390b63007e..c0d9797020 100644
--- a/src/gallium/drivers/r300/r300_screen.c
+++ b/src/gallium/drivers/r300/r300_screen.c
@@ -27,7 +27,8 @@
#include "r300_context.h"
#include "r300_screen.h"
#include "r300_texture.h"
-#include "r300_winsys.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.
@@ -311,14 +312,10 @@ r300_get_tex_transfer(struct pipe_screen *screen,
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;
@@ -344,6 +341,7 @@ 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));
@@ -353,8 +351,8 @@ 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 / pf_get_blockheight(format) * transfer->stride +
+ transfer->x / pf_get_blockwidth(format) * pf_get_blocksize(format);
}
static void r300_transfer_unmap(struct pipe_screen* screen,
@@ -372,7 +370,7 @@ 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);
@@ -380,14 +378,14 @@ 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->num_z_pipes = r300_winsys->z_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 41df31f670..2217988add 100644
--- a/src/gallium/drivers/r300/r300_screen.h
+++ b/src/gallium/drivers/r300/r300_screen.h
@@ -27,6 +27,8 @@
#include "r300_chipset.h"
+struct radeon_winsys;
+
struct r300_screen {
/* Parent class */
struct pipe_screen screen;
@@ -56,6 +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_semantics.h b/src/gallium/drivers/r300/r300_shader_semantics.h
new file mode 100644
index 0000000000..85184e2cfd
--- /dev/null
+++ b/src/gallium/drivers/r300/r300_shader_semantics.h
@@ -0,0 +1,64 @@
+/*
+ * Copyright 2009 Marek Olšák <maraeo@gmail.com>
+ *
+ * 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;
+};
+
+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;
+
+ 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 d1eced61db..8ef0b3b268 100644
--- a/src/gallium/drivers/r300/r300_state.c
+++ b/src/gallium/drivers/r300/r300_state.c
@@ -289,11 +289,37 @@ static void r300_set_edgeflags(struct pipe_context* pipe,
/* XXX and even worse, I have no idea WTF the bitfield is */
}
+static void r300_set_scissor_regs(const struct pipe_scissor_state* state,
+ struct r300_scissor_regs *scissor,
+ boolean is_r500)
+{
+ 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
r300_set_framebuffer_state(struct pipe_context* pipe,
const struct pipe_framebuffer_state* state)
{
struct r300_context* r300 = r300_context(pipe);
+ struct pipe_scissor_state scissor;
if (r300->draw) {
draw_flush(r300->draw);
@@ -301,6 +327,16 @@ static void
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;
}
@@ -523,6 +559,11 @@ static void*
state->mag_img_filter,
state->min_mip_filter);
+ /* 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);
sampler->filter1 |= lod_bias << R300_LOD_BIAS_SHIFT;
@@ -561,6 +602,12 @@ static void r300_bind_sampler_states(struct pipe_context* pipe,
r300->sampler_count = count;
}
+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)
{
FREE(state);
@@ -571,6 +618,7 @@ 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 */
@@ -585,6 +633,13 @@ static void r300_set_sampler_textures(struct pipe_context* pipe,
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;
+ }
}
}
@@ -604,24 +659,13 @@ 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,
@@ -674,6 +718,8 @@ static void r300_set_vertex_buffers(struct pipe_context* pipe,
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,
@@ -706,9 +752,6 @@ 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);
@@ -719,8 +762,6 @@ 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;
@@ -731,10 +772,10 @@ 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 | R300_NEW_VERTEX_SHADER_CONSTANTS;
} else {
+ draw_flush(r300->draw);
draw_bind_vertex_shader(r300->draw,
(struct draw_vertex_shader*)shader);
}
@@ -748,7 +789,6 @@ 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((void*)vs->state.tokens);
FREE(shader);
} else {
@@ -813,10 +853,11 @@ 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 7166694edf..6af49888b9 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 <MostAwesomeDude@gmail.com>
+ * Copyright 2009 Marek Olšák <maraeo@gmail.com>
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
@@ -28,6 +29,7 @@
#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"
@@ -47,8 +49,8 @@ struct r300_shader_derived_value {
unsigned r300_shader_key_hash(void* key) {
struct r300_shader_key* shader_key = (struct r300_shader_key*)key;
- unsigned vs = (unsigned)shader_key->vs;
- unsigned fs = (unsigned)shader_key->fs;
+ unsigned vs = (intptr_t)shader_key->vs;
+ unsigned fs = (intptr_t)shader_key->fs;
return (vs << 16) | (fs & 0xffff);
}
@@ -61,209 +63,143 @@ int r300_shader_key_compare(void* key1, void* key2) {
(shader_key1->fs == shader_key2->fs);
}
-/* Set up the vs_tab and routes. */
-static void r300_vs_tab_routes(struct r300_context* r300,
- struct r300_vertex_info* vformat)
+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_vs_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;
}
- /* Position. */
- if (r300->draw) {
- draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE,
- draw_find_vs_output(r300->draw, TGSI_SEMANTIC_POSITION, 0));
+ /* Fog coordinates. */
+ if (vs_outputs->fog != ATTR_UNUSED) {
+ r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE,
+ vs_outputs->fog);
+ gen_count++;
}
- vinfo->hwfmt[1] |= R300_INPUT_CNTL_POS;
- vinfo->hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT;
- /* Point size. */
- if (psize) {
- if (r300->draw) {
- 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;
- }
+ /* XXX magic */
+ assert(gen_count <= 8);
+}
- /* Colors. */
- for (i = 0; i < cols; i++) {
- if (r300->draw) {
- 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);
- }
+/* 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;
- /* Init i right here, increment it if fog is enabled.
- * This gets around a double-increment problem. */
- i = 0;
+ /* 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);
- /* Fog. This is a special-cased texcoord. */
- if (fog) {
- i++;
- if (r300->draw) {
- 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));
- }
+ for (i = 0; i < r300->vertex_element_count; i++) {
+ format = r300->vertex_element[i].src_format;
+
+ type = r300_translate_vertex_data_type(format) |
+ (i << R300_DST_VEC_LOC_SHIFT);
+ swizzle = r300_translate_vertex_data_swizzle(format);
- /* Texcoords. */
- for (; i < texs; i++) {
- if (r300->draw) {
- draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE,
- draw_find_vs_output(r300->draw, TGSI_SEMANTIC_GENERIC, i));
+ if (i % 2) {
+ 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;
}
- vinfo->hwfmt[1] |= (R300_INPUT_CNTL_TC0 << i);
- vinfo->hwfmt[3] |= (4 << (3 * i));
}
- 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_info* 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;
uint16_t type, swizzle;
enum pipe_format format;
unsigned i, attrib_count;
+ int* vs_output_tab = r300->vs->output_stream_loc_swtcl;
- /* 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]);
- }
+ /* 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);
+ assert(vs_output_tab[i] != -1);
format = draw_translate_vinfo_format(vinfo->attrib[i].emit);
/* Obtain the type of data in this attribute. */
type = r300_translate_vertex_data_type(format) |
- tab[i] << R300_DST_VEC_LOC_SHIFT;
+ vs_output_tab[i] << R300_DST_VEC_LOC_SHIFT;
/* Obtain the swizzle for this attribute. Note that the default
* swizzle in the hardware is not XYZW! */
@@ -272,12 +208,10 @@ static void r300_vertex_psc(struct r300_context* r300,
/* 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 << 0;
-
- vformat->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 0;
+ vformat->vap_prog_stream_cntl[i >> 1] |= type;
+ vformat->vap_prog_stream_cntl_ext[i >> 1] |= swizzle;
}
}
@@ -289,185 +223,200 @@ static void r300_vertex_psc(struct r300_context* r300,
(R300_LAST_VEC << (i & 1 ? 16 : 0));
}
-/* Set up the mappings from GB to US, for RS block. */
-static void r300_update_fs_tab(struct r300_context* r300,
- struct r300_vertex_info* vformat)
+static void r300_rs_col(struct r300_rs_block* rs, int id, int ptr,
+ boolean swizzle_0001)
+{
+ 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);
+}
+
+static void r300_rs_col_write(struct r300_rs_block* rs, int id, int fp_offset)
{
- 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_COL_CN_WRITE |
+ R300_RS_INST_COL_ADDR(fp_offset);
+}
- for (i = 0; i < 16; i++) {
- tab[i] = -1;
+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);
+}
- 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 r300_rs_tex_write(struct r300_rs_block* rs, int id, int fp_offset)
+{
+ rs->inst[id] |= R300_RS_INST_TEX_CN_WRITE |
+ R300_RS_INST_TEX_ADDR(fp_offset);
+}
+
+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);
+}
- /* 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_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);
+}
+
+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. */
+/* 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_rs_block* rs)
+ struct r300_shader_semantics* vs_outputs,
+ struct r300_shader_semantics* fs_inputs)
{
- struct tgsi_shader_info* info = &r300->fs->info;
- int col_count = 0, fp_offset = 0, i, tex_count = 0;
- int rs_tex_comp = 0;
+ struct r300_rs_block* rs = 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);
if (r300_screen(r300->context.screen)->caps->is_r500) {
- for (i = 0; i < info->num_inputs; i++) {
- 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) {
+ /* 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++;
- }
-
- 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++;
- }
-
- 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);
- fp_offset++;
- }
- } else {
- for (i = 0; i < info->num_inputs; i++) {
- 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;
+ } 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++;
}
}
+ }
- if (col_count == 0) {
- rs->ip[0] |= R300_RS_COL_FMT(R300_RS_COL_FMT_0001);
- }
-
- 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);
+ /* 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++;
+ }
}
+ }
- /* Rasterize at least one color, or bad things happen. */
- if ((col_count == 0) && (tex_count == 0)) {
- col_count++;
- }
+ /* 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);
- 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);
+ /* 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++;
}
-
- 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);
+ 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->fog != ATTR_UNUSED) {
fp_offset++;
}
}
- rs->count = (rs_tex_comp) | (col_count << R300_IC_COUNT_SHIFT) |
+ /* 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++;
+ }
+
+ rs->count = (tex_count*4) | (col_count << R300_IC_COUNT_SHIFT) |
R300_HIRES_EN;
- rs->inst_count = MAX2(MAX2(col_count - 1, tex_count - 1), 0);
+ 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_vertex_info* vformat;
- struct r300_rs_block* rs_block;
- int i;
/*
struct r300_shader_key* key;
@@ -495,27 +444,45 @@ static void r300_update_derived_shader_state(struct r300_context* r300)
(void*)key, (void*)value);
} */
- /* XXX This will be refactored ASAP. */
- vformat = CALLOC_STRUCT(r300_vertex_info);
- rs_block = CALLOC_STRUCT(r300_rs_block);
+ /* 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);
- for (i = 0; i < 16; i++) {
- vformat->vs_tab[i] = -1;
- vformat->fs_tab[i] = -1;
+ 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);
}
- r300_vs_tab_routes(r300, vformat);
- r300_vertex_psc(r300, vformat);
- r300_update_fs_tab(r300, vformat);
+ r300->dirty_state |= R300_NEW_RS_BLOCK;
+}
- r300_update_rs_block(r300, 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));
+}
- FREE(r300->vertex_info);
- FREE(r300->rs_block);
+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;
- r300->vertex_info = vformat;
- r300->rs_block = rs_block;
- r300->dirty_state |= (R300_NEW_VERTEX_FORMAT | R300_NEW_RS_BLOCK);
+ return (af & R300_FG_ALPHA_FUNC_ENABLE) &&
+ (af & R300_FG_ALPHA_FUNC_ALWAYS) != R300_FG_ALPHA_FUNC_ALWAYS;
}
static void r300_update_ztop(struct r300_context* r300)
@@ -534,28 +501,34 @@ static void r300_update_ztop(struct r300_context* r300)
* 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:
- * ~) Depth writes in fragment shader
- * ~) Outstanding occlusion queries
+ * 5) Depth writes in fragment shader
+ * 6) Outstanding occlusion queries
*
* ~C.
*/
- if (r300->dsa_state->alpha_function) {
- r300->ztop_state.z_buffer_top = R300_ZTOP_DISABLE;
- } else if (r300->fs->info.uses_kill) {
+
+ /* 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)) {
+ } else if (r300_fragment_shader_writes_depth(r300->fs)) { /* (5) */
r300->ztop_state.z_buffer_top = R300_ZTOP_DISABLE;
- } else if (r300->query_current) {
+ } else if (r300->query_current) { /* (6) */
r300->ztop_state.z_buffer_top = R300_ZTOP_DISABLE;
}
}
void r300_update_derived_state(struct r300_context* r300)
{
- /* XXX */
- if (TRUE || r300->dirty_state &
- (R300_NEW_FRAGMENT_SHADER | R300_NEW_VERTEX_SHADER)) {
+ if (r300->dirty_state &
+ (R300_NEW_FRAGMENT_SHADER | R300_NEW_VERTEX_SHADER |
+ R300_NEW_VERTEX_FORMAT)) {
r300_update_derived_shader_state(r300);
}
diff --git a/src/gallium/drivers/r300/r300_state_invariant.c b/src/gallium/drivers/r300/r300_state_invariant.c
index c07e6ae676..46d1cb39b5 100644
--- a/src/gallium/drivers/r300/r300_state_invariant.c
+++ b/src/gallium/drivers/r300/r300_state_invariant.c
@@ -84,7 +84,7 @@ void r300_emit_invariant_state(struct r300_context* r300)
END_CS;
/* XXX unsorted stuff from surface_fill */
- BEGIN_CS(60 + (caps->has_tcl ? 5 : 0) + (caps->is_r500 ? 4 : 0));
+ BEGIN_CS(56 + (caps->has_tcl ? 5 : 0) + (caps->is_r500 ? 4 : 0));
/* Flush PVS. */
OUT_CS_REG(R300_VAP_PVS_STATE_FLUSH_REG, 0x0);
@@ -135,8 +135,6 @@ void r300_emit_invariant_state(struct r300_context* r300)
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 */
diff --git a/src/gallium/drivers/r300/r300_texture.c b/src/gallium/drivers/r300/r300_texture.c
index aea25cf71d..63fc6a235a 100644
--- a/src/gallium/drivers/r300/r300_texture.c
+++ b/src/gallium/drivers/r300/r300_texture.c
@@ -34,8 +34,8 @@ 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);
+ state->format0 = R300_TX_WIDTH((pt->width0 - 1) & 0x7ff) |
+ R300_TX_HEIGHT((pt->height0 - 1) & 0x7ff);
if (tex->is_npot) {
/* rectangles love this */
@@ -43,8 +43,7 @@ static void r300_setup_texture_state(struct r300_texture* tex, boolean is_r500)
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->depth[0]) & 0xf) |
- R300_TX_NUM_LEVELS(pt->last_level & 0xf);
+ state->format0 |= R300_TX_DEPTH(util_logbase2(pt->depth0) & 0xf);
}
state->format1 = r300_translate_texformat(pt->format);
@@ -58,17 +57,17 @@ static void r300_setup_texture_state(struct r300_texture* tex, boolean is_r500)
/* large textures on r500 */
if (is_r500)
{
- if (pt->width[0] > 2048) {
+ if (pt->width0 > 2048) {
state->format2 |= R500_TXWIDTH_BIT11;
}
- if (pt->height[0] > 2048) {
+ if (pt->height0 > 2048) {
state->format2 |= R500_TXHEIGHT_BIT11;
}
}
- assert(is_r500 || (pt->width[0] <= 2048 && pt->height[0] <= 2048));
+ 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,
@@ -106,7 +105,7 @@ unsigned r300_texture_get_stride(struct r300_texture* tex, unsigned level)
return 0;
}
- return align(pf_get_stride(&tex->tex.block, tex->tex.width[level]), 32);
+ return align(pf_get_stride(tex->tex.format, u_minify(tex->tex.width0, level)), 32);
}
static void r300_setup_miptree(struct r300_texture* tex)
@@ -116,39 +115,32 @@ static void r300_setup_miptree(struct r300_texture* tex)
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]);
- }
-
- base->nblocksx[i] = pf_get_nblocksx(&base->block, base->width[i]);
- base->nblocksy[i] = pf_get_nblocksy(&base->block, base->height[i]);
+ unsigned nblocksy = pf_get_nblocksy(base->format, u_minify(base->height0, i));
stride = r300_texture_get_stride(tex, i);
- layer_size = stride * base->nblocksy[i];
+ layer_size = stride * nblocksy;
if (base->target == PIPE_TEXTURE_CUBE)
size = layer_size * 6;
else
- size = layer_size * base->depth[i];
+ 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 / base->block.size;
+ tex->pitch[i] = stride / pf_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.width[0]) ||
- !util_is_power_of_two(tex->tex.height[0]);
+ tex->is_npot = !util_is_power_of_two(tex->tex.width0) ||
+ !util_is_power_of_two(tex->tex.height0);
}
/* Create a new texture. */
@@ -208,8 +200,8 @@ static struct pipe_surface* r300_get_tex_surface(struct pipe_screen* screen,
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;
@@ -237,7 +229,7 @@ static struct pipe_texture*
/* Support only 2D textures without mipmaps */
if (base->target != PIPE_TEXTURE_2D ||
- base->depth[0] != 1 ||
+ base->depth0 != 1 ||
base->last_level != 0) {
return NULL;
}
@@ -252,7 +244,7 @@ static struct pipe_texture*
tex->tex.screen = screen;
tex->stride_override = *stride;
- tex->pitch[0] = *stride / base->block.size;
+ tex->pitch[0] = *stride / pf_get_blocksize(base->format);
r300_setup_flags(tex);
r300_setup_texture_state(tex, r300_screen(screen)->caps->is_r500);
@@ -287,10 +279,9 @@ r300_video_surface_create(struct pipe_screen *screen,
template.target = PIPE_TEXTURE_2D;
template.format = PIPE_FORMAT_X8R8G8B8_UNORM;
template.last_level = 0;
- template.width[0] = util_next_power_of_two(width);
- template.height[0] = util_next_power_of_two(height);
- template.depth[0] = 1;
- pf_get_block(template.format, &template.block);
+ 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;
diff --git a/src/gallium/drivers/r300/r300_tgsi_to_rc.c b/src/gallium/drivers/r300/r300_tgsi_to_rc.c
index 589f1984ee..9fb2de2403 100644
--- a/src/gallium/drivers/r300/r300_tgsi_to_rc.c
+++ b/src/gallium/drivers/r300/r300_tgsi_to_rc.c
@@ -190,10 +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(
@@ -201,18 +201,18 @@ 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->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->SrcRegisterExtMod.Absolute;
- dst->Negate = src->SrcRegister.Negate ? RC_MASK_XYZW : 0;
+ 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)
{
switch(src.Texture) {
case TGSI_TEXTURE_1D:
@@ -258,17 +258,18 @@ 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);
}
static void handle_immediate(struct tgsi_to_rc * ttr, struct tgsi_full_immediate * imm)
diff --git a/src/gallium/drivers/r300/r300_vbo.c b/src/gallium/drivers/r300/r300_vbo.c
index a6a159667a..d8610dadfa 100644
--- a/src/gallium/drivers/r300/r300_vbo.c
+++ b/src/gallium/drivers/r300/r300_vbo.c
@@ -32,53 +32,8 @@
#include "r300_context.h"
#include "r300_state_inlines.h"
#include "r300_reg.h"
-#include "r300_winsys.h"
-static INLINE void setup_vertex_attribute(struct r300_vertex_info *vinfo,
- struct pipe_vertex_element *vert_elem,
- unsigned attr_num)
-{
- uint16_t hw_fmt1, hw_fmt2;
-
- hw_fmt1 = r300_translate_vertex_data_type(vert_elem->src_format) |
- (attr_num << R300_DST_VEC_LOC_SHIFT);
- hw_fmt2 = r300_translate_vertex_data_swizzle(vert_elem->src_format);
-
- if (attr_num % 2 == 0)
- {
- vinfo->vap_prog_stream_cntl[attr_num >> 1] = hw_fmt1;
- vinfo->vap_prog_stream_cntl_ext[attr_num >> 1] = hw_fmt2;
- }
- else
- {
- vinfo->vap_prog_stream_cntl[attr_num >> 1] |= hw_fmt1 << 16;
- vinfo->vap_prog_stream_cntl_ext[attr_num >> 1] |= hw_fmt2 << 16;
- }
-}
-
-static void finish_vertex_attribs_setup(struct r300_vertex_info *vinfo,
- unsigned attribs_num)
-{
- uint32_t last_vec_bit = (attribs_num % 2 == 0) ?
- (R300_LAST_VEC << 16) : R300_LAST_VEC;
-
- assert(attribs_num > 0 && attribs_num <= 16);
- vinfo->vap_prog_stream_cntl[(attribs_num - 1) >> 1] |= last_vec_bit;
-}
-
-void setup_vertex_attributes(struct r300_context *r300)
-{
- struct pipe_vertex_element *vert_elem;
- int i;
-
- for (i = 0; i < r300->vertex_element_count; i++) {
- vert_elem = &r300->vertex_element[i];
- setup_vertex_attribute(r300->vertex_info, vert_elem, i);
- }
-
- finish_vertex_attribs_setup(r300->vertex_info,
- r300->vertex_element_count);
-}
+#include "radeon_winsys.h"
static INLINE int get_buffer_offset(struct r300_context *r300,
unsigned int buf_nr,
diff --git a/src/gallium/drivers/r300/r300_vs.c b/src/gallium/drivers/r300/r300_vs.c
index 74ef416dc1..31248346bc 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 <MostAwesomeDude@gmail.com>
+ * Copyright 2009 Marek Olšák <maraeo@gmail.com>
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
@@ -23,89 +24,226 @@
#include "r300_vs.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"
-
-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;
+
+ default:
+ assert(0);
}
}
+}
- tgsi_parse_init(&parser, vs->state.tokens);
+static void r300_shader_vap_output_fmt(
+ struct r300_shader_semantics* vs_outputs,
+ uint* hwfmt)
+{
+ int i, gen_count;
- while (!tgsi_parse_end_of_tokens(&parser)) {
- tgsi_parse_token(&parser);
+ /* 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 */
- if (parser.FullToken.Token.Type != TGSI_TOKEN_TYPE_DECLARATION)
- continue;
+ hwfmt[0] = 0x5555; /* XXX this is classic Mesa bonghits */
- decl = &parser.FullToken.FullDeclaration;
+ /* 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);
+ }
- if (decl->Declaration.File != TGSI_FILE_OUTPUT)
- continue;
+ /* Point size. */
+ if (vs_outputs->psize != ATTR_UNUSED) {
+ hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__PT_SIZE_PRESENT;
+ }
- 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;
+ /* Colors. */
+ for (i = 0; i < ATTR_COLOR_COUNT; i++) {
+ if (vs_outputs->color[i] != ATTR_UNUSED) {
+ hwfmt[1] |= R300_INPUT_CNTL_COLOR;
+ hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << i;
+ }
+ }
+
+ /* XXX Back-face colors. */
+
+ /* 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++;
+ }
+ }
+
+ /* 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);
+}
+
+/* Set VS output stream locations for SWTCL. */
+static void r300_stream_locations_swtcl(
+ struct r300_shader_semantics* vs_outputs,
+ int* output_stream_loc)
+{
+ int i, tabi = 0, gen_count;
+
+ /* XXX Check whether the numbers (0, 1, 2+i, etc.) are correct.
+ * These should go to VAP_PROG_STREAM_CNTL/DST_VEC_LOC. */
+
+ /* Position. */
+ output_stream_loc[tabi++] = 0;
+
+ /* Point size. */
+ if (vs_outputs->psize != ATTR_UNUSED) {
+ output_stream_loc[tabi++] = 1;
+ }
+
+ /* Colors. */
+ for (i = 0; i < ATTR_COLOR_COUNT; i++) {
+ if (vs_outputs->color[i] != ATTR_UNUSED) {
+ output_stream_loc[tabi++] = 2 + i;
+ }
+ }
+
+ /* Back-face colors. */
+ for (i = 0; i < ATTR_COLOR_COUNT; i++) {
+ if (vs_outputs->bcolor[i] != ATTR_UNUSED) {
+ output_stream_loc[tabi++] = 4 + i;
+ }
+ }
+
+ /* Texture coordinates. */
+ gen_count = 0;
+ for (i = 0; i < ATTR_GENERIC_COUNT; i++) {
+ if (vs_outputs->bcolor[i] != ATTR_UNUSED) {
+ assert(tabi < 16);
+ output_stream_loc[tabi++] = 6 + gen_count;
+ gen_count++;
}
}
- tgsi_parse_free(&parser);
+ /* Fog coordinates. */
+ if (vs_outputs->fog != ATTR_UNUSED) {
+ assert(tabi < 16);
+ output_stream_loc[tabi++] = 6 + gen_count;
+ gen_count++;
+ }
+
+ /* XXX magic */
+ assert(gen_count <= 8);
+
+ for (; tabi < 16;) {
+ output_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;
+
+ /* 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++;
+ }
+
+ /* Colors. */
+ for (i = 0; i < ATTR_COLOR_COUNT; i++) {
+ if (outputs->color[i] != ATTR_UNUSED) {
+ c->code->outputs[outputs->color[i]] = reg++;
+ }
+ }
+
+ /* XXX Back-face colors. */
+
+ /* 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++;
+ }
+}
void r300_translate_vertex_shader(struct r300_context* r300,
struct r300_vertex_shader* vs)
@@ -113,6 +251,14 @@ 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);
+ r300_shader_vap_output_fmt(&vs->outputs, vs->hwfmt);
+
+ if (!r300_screen(r300->context.screen)->caps->has_tcl) {
+ r300_stream_locations_swtcl(&vs->outputs, vs->output_stream_loc_swtcl);
+ }
+
/* Setup the compiler */
rc_init(&compiler.Base);
@@ -137,7 +283,7 @@ void r300_translate_vertex_shader(struct r300_context* r300,
/* Invoke the compiler */
r3xx_compile_vertex_program(&compiler);
if (compiler.Base.Error) {
- /* Todo: Fail gracefully */
+ /* XXX Fail gracefully */
fprintf(stderr, "r300 VP: Compiler error\n");
abort();
}
diff --git a/src/gallium/drivers/r300/r300_vs.h b/src/gallium/drivers/r300/r300_vs.h
index 2a4ce315e3..283dd5a9e8 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 <MostAwesomeDude@gmail.com>
+ * Copyright 2009 Marek Olšák <maraeo@gmail.com>
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
@@ -25,18 +26,20 @@
#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;
- /* Fallback shader, because Draw has issues */
- struct draw_vertex_shader* draw;
+ struct tgsi_shader_info info;
+ struct r300_shader_semantics outputs;
+ int output_stream_loc_swtcl[16];
+ uint hwfmt[4];
/* Has this shader been translated yet? */
boolean translated;
@@ -45,9 +48,6 @@ 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);
diff --git a/src/gallium/drivers/r300/r300_winsys.h b/src/gallium/drivers/r300/r300_winsys.h
index 864a6146b2..1ae6de70fe 100644
--- a/src/gallium/drivers/r300/r300_winsys.h
+++ b/src/gallium/drivers/r300/r300_winsys.h
@@ -35,76 +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;
-
- /* Z pipe count (rv530 only) */
- uint32_t z_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);
-
- /* winsys flush - callback from winsys when flush required */
- void (*set_flush_cb)(struct r300_winsys *winsys,
- void (*flush_cb)(void *), void *data);
-
- void (*reset_bos)(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,