diff options
author | Marek Olšák <[email protected]> | 2018-01-10 23:09:58 +0100 |
---|---|---|
committer | Marek Olšák <[email protected]> | 2018-01-31 03:21:20 +0100 |
commit | c02c9ee550d137fbea3ed105131d621d6af5813b (patch) | |
tree | 2e36f13080af74ac91fe11dd23bd830a1a00ea06 /src/gallium/drivers/radeonsi | |
parent | a7311cd7eee309adb8ebb8b556dfa2fe15c5cdde (diff) |
radeonsi: move max_simd_waves computation into a separate function
Reviewed-by: Nicolai Hähnle <[email protected]>
Diffstat (limited to 'src/gallium/drivers/radeonsi')
-rw-r--r-- | src/gallium/drivers/radeonsi/si_shader.c | 34 | ||||
-rw-r--r-- | src/gallium/drivers/radeonsi/si_shader.h | 1 |
2 files changed, 23 insertions, 12 deletions
diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c index 76215b318a0..7304d7e00db 100644 --- a/src/gallium/drivers/radeonsi/si_shader.c +++ b/src/gallium/drivers/radeonsi/si_shader.c @@ -5362,16 +5362,11 @@ static void si_shader_dump_disassembly(const struct ac_shader_binary *binary, } } -static void si_shader_dump_stats(struct si_screen *sscreen, - const struct si_shader *shader, - struct pipe_debug_callback *debug, - unsigned processor, - FILE *file, - bool check_debug_option) +static void si_calculate_max_simd_waves(struct si_shader *shader) { - const struct si_shader_config *conf = &shader->config; - unsigned num_inputs = shader->selector ? shader->selector->info.num_inputs : 0; - unsigned code_size = si_get_shader_binary_size(shader); + struct si_screen *sscreen = shader->selector->screen; + struct si_shader_config *conf = &shader->config; + unsigned num_inputs = shader->selector->info.num_inputs; unsigned lds_increment = sscreen->info.chip_class >= CIK ? 512 : 256; unsigned lds_per_wave = 0; unsigned max_simd_waves; @@ -5388,7 +5383,7 @@ static void si_shader_dump_stats(struct si_screen *sscreen, } /* Compute LDS usage for PS. */ - switch (processor) { + switch (shader->selector->type) { case PIPE_SHADER_FRAGMENT: /* The minimum usage per wave is (num_inputs * 48). The maximum * usage is (num_inputs * 48 * 16). @@ -5429,6 +5424,19 @@ static void si_shader_dump_stats(struct si_screen *sscreen, if (lds_per_wave) max_simd_waves = MIN2(max_simd_waves, 16384 / lds_per_wave); + conf->max_simd_waves = max_simd_waves; +} + +static void si_shader_dump_stats(struct si_screen *sscreen, + const struct si_shader *shader, + struct pipe_debug_callback *debug, + unsigned processor, + FILE *file, + bool check_debug_option) +{ + const struct si_shader_config *conf = &shader->config; + unsigned code_size = si_get_shader_binary_size(shader); + if (!check_debug_option || si_can_dump_shader(sscreen, processor)) { if (processor == PIPE_SHADER_FRAGMENT) { @@ -5453,7 +5461,7 @@ static void si_shader_dump_stats(struct si_screen *sscreen, conf->spilled_sgprs, conf->spilled_vgprs, conf->private_mem_vgprs, code_size, conf->lds_size, conf->scratch_bytes_per_wave, - max_simd_waves); + conf->max_simd_waves); } pipe_debug_message(debug, SHADER_INFO, @@ -5462,7 +5470,7 @@ static void si_shader_dump_stats(struct si_screen *sscreen, "Spilled VGPRs: %d PrivMem VGPRs: %d", conf->num_sgprs, conf->num_vgprs, code_size, conf->lds_size, conf->scratch_bytes_per_wave, - max_simd_waves, conf->spilled_sgprs, + conf->max_simd_waves, conf->spilled_sgprs, conf->spilled_vgprs, conf->private_mem_vgprs); } @@ -6971,6 +6979,7 @@ int si_compile_tgsi_shader(struct si_screen *sscreen, shader->info.num_input_vgprs += 1; } + si_calculate_max_simd_waves(shader); return 0; } @@ -8044,6 +8053,7 @@ int si_shader_create(struct si_screen *sscreen, LLVMTargetMachineRef tm, shader->config.num_vgprs = MAX2(shader->config.num_vgprs, shader->epilog->config.num_vgprs); } + si_calculate_max_simd_waves(shader); } si_fix_resource_usage(sscreen, shader); diff --git a/src/gallium/drivers/radeonsi/si_shader.h b/src/gallium/drivers/radeonsi/si_shader.h index c449aa96844..6ed1646817f 100644 --- a/src/gallium/drivers/radeonsi/si_shader.h +++ b/src/gallium/drivers/radeonsi/si_shader.h @@ -557,6 +557,7 @@ struct si_shader_config { 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; |