AlexIndustrial/mesa
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

r600g: plug in optimizing backend

Browse Source 
Optimization is enabled with "R600_DEBUG=sb".

Signed-off-by: Vadim Girlin <vadimgirlin@gmail.com>
This commit is contained in:
Vadim Girlin
2013-04-30 20:53:15 +04:00
parent 2cd7691793
commit ad1df471d0
8 changed files with 155 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/gallium/drivers/r600/Makefile.am
+9 -1
View File
@@ -10,8 +10,16 @@ AM_CFLAGS = \
$(RADEON_CFLAGS) \
$(VISIBILITY_CFLAGS)
AM_CXXFLAGS = \
-I$(top_srcdir)/src/gallium/drivers \
-I$(top_srcdir)/include \
-I$(top_srcdir)/src/gallium/include \
-I$(top_srcdir)/src/gallium/auxiliary \
$(DEFINES)
libr600_la_SOURCES = \
$(C_SOURCES)
$(C_SOURCES) \
$(CXX_SOURCES)
libr600_la_LIBADD = ../radeon/libradeon.la
src/gallium/drivers/r600/Makefile.sources
+28
View File
@@ -19,4 +19,32 @@ C_SOURCES = \
compute_memory_pool.c \
r600_uvd.c
CXX_SOURCES = \
sb/sb_bc_builder.cpp \
sb/sb_bc_decoder.cpp \
sb/sb_bc_dump.cpp \
sb/sb_bc_finalize.cpp \
sb/sb_bc_parser.cpp \
sb/sb_context.cpp \
sb/sb_core.cpp \
sb/sb_dce_cleanup.cpp \
sb/sb_def_use.cpp \
sb/sb_dump.cpp \
sb/sb_expr.cpp \
sb/sb_gcm.cpp \
sb/sb_gvn.cpp \
sb/sb_if_conversion.cpp \
sb/sb_ir.cpp \
sb/sb_liveness.cpp \
sb/sb_pass.cpp \
sb/sb_peephole.cpp \
sb/sb_psi_ops.cpp \
sb/sb_ra_checker.cpp \
sb/sb_ra_coalesce.cpp \
sb/sb_ra_init.cpp \
sb/sb_sched.cpp \
sb/sb_shader.cpp \
sb/sb_ssa_builder.cpp \
sb/sb_valtable.cpp
LLVM_C_SOURCES = r600_llvm.c
src/gallium/drivers/r600/r600_asm.c
+11
View File
@@ -32,6 +32,8 @@
#include "util/u_memory.h"
#include "pipe/p_shader_tokens.h"
#include "sb/sb_public.h"
#define NUM_OF_CYCLES 3
#define NUM_OF_COMPONENTS 4
@@ -126,6 +128,10 @@ void r600_bytecode_init(struct r600_bytecode *bc,
enum radeon_family family,
enum r600_msaa_texture_mode msaa_texture_mode)
{
static unsigned next_shader_id = 0;
bc->debug_id = ++next_shader_id;
if ((chip_class == R600) &&
(family != CHIP_RV670 && family != CHIP_RS780 && family != CHIP_RS880)) {
bc->ar_handling = AR_HANDLE_RV6XX;
@@ -2381,8 +2387,13 @@ void *r600_create_vertex_fetch_shader(struct pipe_context *ctx,
fprintf(stderr, "\n");
}
#if 0
r600_bytecode_disasm(&bc);
fprintf(stderr, "______________________________________________________________\n");
#else
r600_sb_bytecode_process(rctx, &bc, NULL, 1 /*dump*/, 0 /*optimize*/);
#endif
}
fs_size = bc.ndw*4;
src/gallium/drivers/r600/r600_asm.h
+1
View File
@@ -214,6 +214,7 @@ struct r600_bytecode {
unsigned ar_chan;
unsigned ar_handling;
unsigned r6xx_nop_after_rel_dst;
unsigned debug_id;
struct r600_isa* isa;
};
src/gallium/drivers/r600/r600_pipe.c
+11
View File
@@ -26,6 +26,8 @@
#include "evergreen_compute.h"
#include "r600d.h"
#include "sb/sb_public.h"
#include <errno.h>
#include "pipe/p_shader_tokens.h"
#include "util/u_blitter.h"
@@ -64,6 +66,13 @@ static const struct debug_named_value debug_options[] = {
/* GL uses the word INVALIDATE, gallium uses the word DISCARD */
{ "noinvalrange", DBG_NO_DISCARD_RANGE, "Disable handling of INVALIDATE_RANGE map flags" },
/* shader backend */
{ "sb", DBG_SB, "Enable optimization of graphics shaders" },
{ "sbcl", DBG_SB_CS, "Enable optimization of compute shaders" },
{ "sbdry", DBG_SB_DRY_RUN, "Don't use optimized bytecode (just print the dumps)" },
{ "sbstat", DBG_SB_STAT, "Print optimization statistics for shaders" },
{ "sbdump", DBG_SB_DUMP, "Print IR dumps after some optimization passes" },
DEBUG_NAMED_VALUE_END /* must be last */
};
@@ -305,6 +314,8 @@ static void r600_destroy_context(struct pipe_context *context)
r600_isa_destroy(rctx->isa);
r600_sb_context_destroy(rctx->sb_context);
pipe_resource_reference((struct pipe_resource**)&rctx->dummy_cmask, NULL);
pipe_resource_reference((struct pipe_resource**)&rctx->dummy_fmask, NULL);
src/gallium/drivers/r600/r600_pipe.h
+7
View File
@@ -257,6 +257,12 @@ typedef boolean (*r600g_dma_blit_t)(struct pipe_context *ctx,
#define DBG_NO_CP_DMA (1 << 18)
#define DBG_NO_ASYNC_DMA (1 << 19)
#define DBG_NO_DISCARD_RANGE (1 << 20)
/* shader backend */
#define DBG_SB (1 << 21)
#define DBG_SB_CS (1 << 22)
#define DBG_SB_DRY_RUN (1 << 23)
#define DBG_SB_STAT (1 << 24)
#define DBG_SB_DUMP (1 << 25)
struct r600_tiling_info {
unsigned num_channels;
@@ -640,6 +646,7 @@ struct r600_context {
unsigned current_render_cond_mode;
boolean predicate_drawing;
void *sb_context;
struct r600_isa *isa;
};
src/gallium/drivers/r600/r600_shader.c
+77 -2
View File
@@ -27,6 +27,8 @@
#include "r600_shader.h"
#include "r600d.h"
#include "sb/sb_public.h"
#include "pipe/p_shader_tokens.h"
#include "tgsi/tgsi_info.h"
#include "tgsi/tgsi_parse.h"
@@ -62,6 +64,26 @@ static int r600_shader_from_tgsi(struct r600_screen *rscreen,
struct r600_pipe_shader *pipeshader,
struct r600_shader_key key);
static void r600_add_gpr_array(struct r600_shader *ps, int start_gpr,
int size, unsigned comp_mask) {
if (!size)
return;
if (ps->num_arrays == ps->max_arrays) {
ps->max_arrays += 64;
ps->arrays = realloc(ps->arrays, ps->max_arrays *
sizeof(struct r600_shader_array));
}
int n = ps->num_arrays;
++ps->num_arrays;
ps->arrays[n].comp_mask = comp_mask;
ps->arrays[n].gpr_start = start_gpr;
ps->arrays[n].gpr_count = size;
}
static unsigned tgsi_get_processor_type(const struct tgsi_token *tokens)
{
struct tgsi_parse_context parse;
@@ -118,6 +140,7 @@ int r600_pipe_shader_create(struct pipe_context *ctx,
int r, i;
uint32_t *ptr;
bool dump = r600_can_dump_shader(rctx->screen, tgsi_get_processor_type(sel->tokens));
unsigned use_sb = rctx->screen->debug_flags & DBG_SB;
shader->shader.bc.isa = rctx->isa;
@@ -139,12 +162,22 @@ int r600_pipe_shader_create(struct pipe_context *ctx,
R600_ERR("building bytecode failed !\n");
return r;
}
#if 0
if (dump) {
fprintf(stderr, "--------------------------------------------------------------\n");
r600_bytecode_disasm(&shader->shader.bc);
fprintf(stderr, "______________________________________________________________\n");
}
#else
if (dump || use_sb) {
r = r600_sb_bytecode_process(rctx, &shader->shader.bc, &shader->shader, dump, use_sb);
if (r) {
R600_ERR("r600_sb_bytecode_process failed !\n");
return r;
}
}
#endif
/* Store the shader in a buffer. */
if (shader->bo == NULL) {
@@ -273,6 +306,7 @@ int r600_compute_shader_create(struct pipe_context * ctx,
struct r600_shader_ctx shader_ctx;
boolean use_kill = false;
bool dump = (r600_ctx->screen->debug_flags & DBG_CS) != 0;
unsigned use_sb = r600_ctx->screen->debug_flags & DBG_SB_CS;
shader_ctx.bc = bytecode;
r600_bytecode_init(shader_ctx.bc, r600_ctx->chip_class, r600_ctx->family,
@@ -286,9 +320,18 @@ int r600_compute_shader_create(struct pipe_context * ctx,
cm_bytecode_add_cf_end(shader_ctx.bc);
}
r600_bytecode_build(shader_ctx.bc);
#if 0
if (dump) {
r600_bytecode_disasm(shader_ctx.bc);
}
#else
if (dump || use_sb) {
if (r600_sb_bytecode_process(r600_ctx, shader_ctx.bc, NULL, dump, use_sb))
R600_ERR("r600_sb_bytecode_process failed!\n");
}
#endif
free(bytes);
return 1;
}
@@ -956,8 +999,18 @@ static int tgsi_declaration(struct r600_shader_ctx *ctx)
}
}
break;
case TGSI_FILE_CONSTANT:
case TGSI_FILE_TEMPORARY:
if (ctx->info.indirect_files & (1 << TGSI_FILE_TEMPORARY)) {
if (d->Array.ArrayID) {
r600_add_gpr_array(ctx->shader,
ctx->file_offset[TGSI_FILE_TEMPORARY] +
d->Range.First,
d->Range.Last - d->Range.First + 1, 0b1111);
}
}
break;
case TGSI_FILE_CONSTANT:
case TGSI_FILE_SAMPLER:
case TGSI_FILE_ADDRESS:
break;
@@ -1248,6 +1301,7 @@ static int process_twoside_color_inputs(struct r600_shader_ctx *ctx)
return 0;
}
static int r600_shader_from_tgsi(struct r600_screen *rscreen,
struct r600_pipe_shader *pipeshader,
struct r600_shader_key key)
@@ -1267,6 +1321,7 @@ static int r600_shader_from_tgsi(struct r600_screen *rscreen,
bool use_llvm = false;
unsigned char * inst_bytes = NULL;
unsigned inst_byte_count = 0;
bool indirect_gprs;
#ifdef R600_USE_LLVM
use_llvm = !(rscreen->debug_flags & DBG_NO_LLVM);
@@ -1279,6 +1334,8 @@ static int r600_shader_from_tgsi(struct r600_screen *rscreen,
rscreen->msaa_texture_support);
ctx.tokens = tokens;
tgsi_scan_shader(tokens, &ctx.info);
shader->indirect_files = ctx.info.indirect_files;
indirect_gprs = ctx.info.indirect_files & ~(1 << TGSI_FILE_CONSTANT);
tgsi_parse_init(&ctx.parse, tokens);
ctx.type = ctx.parse.FullHeader.Processor.Processor;
shader->processor_type = ctx.type;
@@ -1356,6 +1413,24 @@ static int r600_shader_from_tgsi(struct r600_screen *rscreen,
ctx.info.file_max[TGSI_FILE_TEMPORARY] + 1;
ctx.temp_reg = ctx.bc->ar_reg + 1;
if (indirect_gprs) {
shader->max_arrays = 0;
shader->num_arrays = 0;
if (ctx.info.indirect_files & (1 << TGSI_FILE_INPUT)) {
r600_add_gpr_array(shader, ctx.file_offset[TGSI_FILE_INPUT],
ctx.file_offset[TGSI_FILE_OUTPUT] -
ctx.file_offset[TGSI_FILE_INPUT],
0b1111);
}
if (ctx.info.indirect_files & (1 << TGSI_FILE_OUTPUT)) {
r600_add_gpr_array(shader, ctx.file_offset[TGSI_FILE_OUTPUT],
ctx.file_offset[TGSI_FILE_TEMPORARY] -
ctx.file_offset[TGSI_FILE_OUTPUT],
0b1111);
}
}
ctx.nliterals = 0;
ctx.literals = NULL;
shader->fs_write_all = FALSE;
src/gallium/drivers/r600/r600_shader.h
+11
View File
@@ -63,6 +63,11 @@ struct r600_shader {
boolean vs_out_point_size;
boolean has_txq_cube_array_z_comp;
boolean uses_tex_buffers;
unsigned indirect_files;
unsigned max_arrays;
unsigned num_arrays;
struct r600_shader_array * arrays;
};
struct r600_shader_key {
@@ -71,6 +76,12 @@ struct r600_shader_key {
unsigned nr_cbufs:4;
};
struct r600_shader_array {
unsigned gpr_start;
unsigned gpr_count;
unsigned comp_mask;
};
struct r600_pipe_shader {
struct r600_pipe_shader_selector *selector;
struct r600_pipe_shader *next_variant;