Mesa (master): radeonsi: use ac_shader_config
GitLab Mirror
gitlab-mirror at kemper.freedesktop.org
Thu Jun 13 00:40:13 UTC 2019
Module: Mesa
Branch: master
Commit: 77b05cc42df29472a7852b90575a19e8991815cd
URL: http://cgit.freedesktop.org/mesa/mesa/commit/?id=77b05cc42df29472a7852b90575a19e8991815cd
Author: Nicolai Hähnle <nicolai.haehnle at amd.com>
Date: Wed May 9 16:38:33 2018 +0200
radeonsi: use ac_shader_config
Reviewed-by: Marek Olšák <marek.olsak at amd.com>
---
src/amd/common/ac_binary.c | 2 +
src/gallium/drivers/radeonsi/si_compute.c | 14 ++--
src/gallium/drivers/radeonsi/si_shader.c | 112 ++++--------------------------
src/gallium/drivers/radeonsi/si_shader.h | 25 ++-----
4 files changed, 27 insertions(+), 126 deletions(-)
diff --git a/src/amd/common/ac_binary.c b/src/amd/common/ac_binary.c
index ee8d3836177..c047666b656 100644
--- a/src/amd/common/ac_binary.c
+++ b/src/amd/common/ac_binary.c
@@ -225,12 +225,14 @@ void ac_parse_shader_binary_config(const char *data, size_t nbytes,
conf->num_sgprs = MAX2(conf->num_sgprs, (G_00B028_SGPRS(value) + 1) * 8);
conf->num_vgprs = MAX2(conf->num_vgprs, (G_00B028_VGPRS(value) + 1) * 4);
conf->float_mode = G_00B028_FLOAT_MODE(value);
+ conf->rsrc1 = value;
break;
case R_00B02C_SPI_SHADER_PGM_RSRC2_PS:
conf->lds_size = MAX2(conf->lds_size, G_00B02C_EXTRA_LDS_SIZE(value));
break;
case R_00B84C_COMPUTE_PGM_RSRC2:
conf->lds_size = MAX2(conf->lds_size, G_00B84C_LDS_SIZE(value));
+ conf->rsrc2 = value;
break;
case R_0286CC_SPI_PS_INPUT_ENA:
conf->spi_ps_input_ena = value;
diff --git a/src/gallium/drivers/radeonsi/si_compute.c b/src/gallium/drivers/radeonsi/si_compute.c
index bb75132a1a2..a08ed7d65f5 100644
--- a/src/gallium/drivers/radeonsi/si_compute.c
+++ b/src/gallium/drivers/radeonsi/si_compute.c
@@ -66,7 +66,7 @@ static const amd_kernel_code_t *si_compute_get_code_object(
}
static void code_object_to_config(const amd_kernel_code_t *code_object,
- struct si_shader_config *out_config) {
+ struct ac_shader_config *out_config) {
uint32_t rsrc1 = code_object->compute_pgm_resource_registers;
uint32_t rsrc2 = code_object->compute_pgm_resource_registers >> 32;
@@ -249,8 +249,8 @@ static void *si_create_compute_state(
return NULL;
}
} else {
- si_shader_binary_read_config(&program->shader.binary,
- &program->shader.config, 0);
+ ac_shader_binary_read_config(&program->shader.binary,
+ &program->shader.config, 0, false);
}
si_shader_dump(sctx->screen, &program->shader, &sctx->debug,
PIPE_SHADER_COMPUTE, stderr, true);
@@ -366,7 +366,7 @@ void si_emit_initial_compute_regs(struct si_context *sctx, struct radeon_cmdbuf
static bool si_setup_compute_scratch_buffer(struct si_context *sctx,
struct si_shader *shader,
- struct si_shader_config *config)
+ struct ac_shader_config *config)
{
uint64_t scratch_bo_size, scratch_needed;
scratch_bo_size = 0;
@@ -409,8 +409,8 @@ static bool si_switch_compute_shader(struct si_context *sctx,
unsigned offset)
{
struct radeon_cmdbuf *cs = sctx->gfx_cs;
- struct si_shader_config inline_config = {0};
- struct si_shader_config *config;
+ struct ac_shader_config inline_config = {0};
+ struct ac_shader_config *config;
uint64_t shader_va;
if (sctx->cs_shader_state.emitted_program == program &&
@@ -426,7 +426,7 @@ static bool si_switch_compute_shader(struct si_context *sctx,
if (code_object) {
code_object_to_config(code_object, config);
} else {
- si_shader_binary_read_config(&shader->binary, config, offset);
+ ac_shader_binary_read_config(&shader->binary, config, offset, false);
}
lds_blocks = config->lds_size;
diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c
index 8c4f4e75653..d532d664307 100644
--- a/src/gallium/drivers/radeonsi/si_shader.c
+++ b/src/gallium/drivers/radeonsi/si_shader.c
@@ -5045,90 +5045,6 @@ static void si_llvm_emit_polygon_stipple(struct si_shader_context *ctx,
ac_build_kill_if_false(&ctx->ac, bit);
}
-void si_shader_binary_read_config(struct ac_shader_binary *binary,
- struct si_shader_config *conf,
- unsigned symbol_offset)
-{
- unsigned i;
- const unsigned char *config =
- ac_shader_binary_config_start(binary, symbol_offset);
- bool really_needs_scratch = false;
-
- /* LLVM adds SGPR spills to the scratch size.
- * Find out if we really need the scratch buffer.
- */
- for (i = 0; i < binary->reloc_count; i++) {
- const struct ac_shader_reloc *reloc = &binary->relocs[i];
-
- if (!strcmp(scratch_rsrc_dword0_symbol, reloc->name) ||
- !strcmp(scratch_rsrc_dword1_symbol, reloc->name)) {
- really_needs_scratch = true;
- break;
- }
- }
-
- /* XXX: We may be able to emit some of these values directly rather than
- * extracting fields to be emitted later.
- */
-
- for (i = 0; i < binary->config_size_per_symbol; i+= 8) {
- unsigned reg = util_le32_to_cpu(*(uint32_t*)(config + i));
- unsigned value = util_le32_to_cpu(*(uint32_t*)(config + i + 4));
- switch (reg) {
- case R_00B028_SPI_SHADER_PGM_RSRC1_PS:
- case R_00B128_SPI_SHADER_PGM_RSRC1_VS:
- case R_00B228_SPI_SHADER_PGM_RSRC1_GS:
- case R_00B428_SPI_SHADER_PGM_RSRC1_HS:
- case R_00B848_COMPUTE_PGM_RSRC1:
- conf->num_sgprs = MAX2(conf->num_sgprs, (G_00B028_SGPRS(value) + 1) * 8);
- conf->num_vgprs = MAX2(conf->num_vgprs, (G_00B028_VGPRS(value) + 1) * 4);
- conf->float_mode = G_00B028_FLOAT_MODE(value);
- conf->rsrc1 = value;
- break;
- case R_00B02C_SPI_SHADER_PGM_RSRC2_PS:
- conf->lds_size = MAX2(conf->lds_size, G_00B02C_EXTRA_LDS_SIZE(value));
- break;
- case R_00B84C_COMPUTE_PGM_RSRC2:
- conf->lds_size = MAX2(conf->lds_size, G_00B84C_LDS_SIZE(value));
- conf->rsrc2 = value;
- break;
- case R_0286CC_SPI_PS_INPUT_ENA:
- conf->spi_ps_input_ena = value;
- break;
- case R_0286D0_SPI_PS_INPUT_ADDR:
- conf->spi_ps_input_addr = value;
- break;
- case R_0286E8_SPI_TMPRING_SIZE:
- case R_00B860_COMPUTE_TMPRING_SIZE:
- /* WAVESIZE is in units of 256 dwords. */
- if (really_needs_scratch)
- conf->scratch_bytes_per_wave =
- G_00B860_WAVESIZE(value) * 256 * 4;
- break;
- case 0x4: /* SPILLED_SGPRS */
- conf->spilled_sgprs = value;
- break;
- case 0x8: /* SPILLED_VGPRS */
- conf->spilled_vgprs = value;
- break;
- default:
- {
- static bool printed;
-
- if (!printed) {
- fprintf(stderr, "Warning: LLVM emitted unknown "
- "config register: 0x%x\n", reg);
- printed = true;
- }
- }
- break;
- }
- }
-
- if (!conf->spi_ps_input_addr)
- conf->spi_ps_input_addr = conf->spi_ps_input_ena;
-}
-
void si_shader_apply_scratch_relocs(struct si_shader *shader,
uint64_t scratch_va)
{
@@ -5296,7 +5212,7 @@ static void si_shader_dump_disassembly(const struct ac_shader_binary *binary,
static void si_calculate_max_simd_waves(struct si_shader *shader)
{
struct si_screen *sscreen = shader->selector->screen;
- struct si_shader_config *conf = &shader->config;
+ struct ac_shader_config *conf = &shader->config;
unsigned num_inputs = shader->selector->info.num_inputs;
unsigned lds_increment = sscreen->info.chip_class >= GFX7 ? 512 : 256;
unsigned lds_per_wave = 0;
@@ -5345,13 +5261,13 @@ static void si_calculate_max_simd_waves(struct si_shader *shader)
if (lds_per_wave)
max_simd_waves = MIN2(max_simd_waves, 16384 / lds_per_wave);
- conf->max_simd_waves = max_simd_waves;
+ shader->info.max_simd_waves = max_simd_waves;
}
void si_shader_dump_stats_for_shader_db(const struct si_shader *shader,
struct pipe_debug_callback *debug)
{
- const struct si_shader_config *conf = &shader->config;
+ const struct ac_shader_config *conf = &shader->config;
pipe_debug_message(debug, SHADER_INFO,
"Shader Stats: SGPRS: %d VGPRS: %d Code Size: %d "
@@ -5360,8 +5276,8 @@ void si_shader_dump_stats_for_shader_db(const struct si_shader *shader,
conf->num_sgprs, conf->num_vgprs,
si_get_shader_binary_size(shader),
conf->lds_size, conf->scratch_bytes_per_wave,
- conf->max_simd_waves, conf->spilled_sgprs,
- conf->spilled_vgprs, conf->private_mem_vgprs);
+ shader->info.max_simd_waves, conf->spilled_sgprs,
+ conf->spilled_vgprs, shader->info.private_mem_vgprs);
}
static void si_shader_dump_stats(struct si_screen *sscreen,
@@ -5370,7 +5286,7 @@ static void si_shader_dump_stats(struct si_screen *sscreen,
FILE *file,
bool check_debug_option)
{
- const struct si_shader_config *conf = &shader->config;
+ const struct ac_shader_config *conf = &shader->config;
if (!check_debug_option ||
si_can_dump_shader(sscreen, processor)) {
@@ -5394,10 +5310,10 @@ static void si_shader_dump_stats(struct si_screen *sscreen,
"********************\n\n\n",
conf->num_sgprs, conf->num_vgprs,
conf->spilled_sgprs, conf->spilled_vgprs,
- conf->private_mem_vgprs,
+ shader->info.private_mem_vgprs,
si_get_shader_binary_size(shader),
conf->lds_size, conf->scratch_bytes_per_wave,
- conf->max_simd_waves);
+ shader->info.max_simd_waves);
}
}
@@ -5484,7 +5400,7 @@ void si_shader_dump(struct si_screen *sscreen, const struct si_shader *shader,
static int si_compile_llvm(struct si_screen *sscreen,
struct ac_shader_binary *binary,
- struct si_shader_config *conf,
+ struct ac_shader_config *conf,
struct ac_llvm_compiler *compiler,
LLVMModuleRef mod,
struct pipe_debug_callback *debug,
@@ -5518,7 +5434,7 @@ static int si_compile_llvm(struct si_screen *sscreen,
return r;
}
- si_shader_binary_read_config(binary, conf, 0);
+ ac_shader_binary_read_config(binary, conf, 0, false);
/* Enable 64-bit and 16-bit denormals, because there is no performance
* cost.
@@ -6893,7 +6809,7 @@ int si_compile_tgsi_shader(struct si_screen *sscreen,
if ((debug && debug->debug_message) ||
si_can_dump_shader(sscreen, ctx.type)) {
- ctx.shader->config.private_mem_vgprs =
+ ctx.shader->info.private_mem_vgprs =
ac_count_scratch_private_memory(ctx.main_fn);
}
@@ -8066,9 +7982,9 @@ int si_shader_create(struct si_screen *sscreen, struct ac_llvm_compiler *compile
shader->config.spilled_vgprs =
MAX2(shader->config.spilled_vgprs,
shader->previous_stage->config.spilled_vgprs);
- shader->config.private_mem_vgprs =
- MAX2(shader->config.private_mem_vgprs,
- shader->previous_stage->config.private_mem_vgprs);
+ shader->info.private_mem_vgprs =
+ MAX2(shader->info.private_mem_vgprs,
+ shader->previous_stage->info.private_mem_vgprs);
shader->config.scratch_bytes_per_wave =
MAX2(shader->config.scratch_bytes_per_wave,
shader->previous_stage->config.scratch_bytes_per_wave);
diff --git a/src/gallium/drivers/radeonsi/si_shader.h b/src/gallium/drivers/radeonsi/si_shader.h
index bc9299bda66..b63045f662d 100644
--- a/src/gallium/drivers/radeonsi/si_shader.h
+++ b/src/gallium/drivers/radeonsi/si_shader.h
@@ -574,22 +574,6 @@ struct si_shader_key {
/* Restore the pack alignment to default. */
#pragma pack(pop)
-struct si_shader_config {
- unsigned num_sgprs;
- unsigned num_vgprs;
- unsigned spilled_sgprs;
- unsigned spilled_vgprs;
- unsigned private_mem_vgprs;
- unsigned lds_size;
- unsigned max_simd_waves;
- unsigned spi_ps_input_ena;
- unsigned spi_ps_input_addr;
- unsigned float_mode;
- unsigned scratch_bytes_per_wave;
- unsigned rsrc1;
- unsigned rsrc2;
-};
-
/* GCN-specific shader info. */
struct si_shader_info {
ubyte vs_output_param_offset[SI_MAX_VS_OUTPUTS];
@@ -600,6 +584,8 @@ struct si_shader_info {
bool uses_instanceid;
ubyte nr_pos_exports;
ubyte nr_param_exports;
+ unsigned private_mem_vgprs;
+ unsigned max_simd_waves;
};
struct si_shader {
@@ -627,7 +613,7 @@ struct si_shader {
/* The following data is all that's needed for binary shaders. */
struct ac_shader_binary binary;
- struct si_shader_config config;
+ struct ac_shader_config config;
struct si_shader_info info;
/* Shader key + LLVM IR + disassembly + statistics.
@@ -684,7 +670,7 @@ struct si_shader_part {
struct si_shader_part *next;
union si_shader_part_key key;
struct ac_shader_binary binary;
- struct si_shader_config config;
+ struct ac_shader_config config;
};
/* si_shader.c */
@@ -714,9 +700,6 @@ void si_multiwave_lds_size_workaround(struct si_screen *sscreen,
unsigned *lds_size);
void si_shader_apply_scratch_relocs(struct si_shader *shader,
uint64_t scratch_va);
-void si_shader_binary_read_config(struct ac_shader_binary *binary,
- struct si_shader_config *conf,
- unsigned symbol_offset);
const char *si_get_shader_name(const struct si_shader *shader, unsigned processor);
/* si_shader_nir.c */
More information about the mesa-commit
mailing list