| /xsrc/external/mit/MesaLib/dist/src/amd/compiler/ |
| aco_scheduler.cpp | 34 #define SMEM_WINDOW_SIZE (350 - ctx.num_waves * 35)
35 #define VMEM_WINDOW_SIZE (1024 - ctx.num_waves * 64)
37 #define SMEM_MAX_MOVES (64 - ctx.num_waves * 4)
38 #define VMEM_MAX_MOVES (256 - ctx.num_waves * 16)
39 /* creating clauses decreases def-use distances, so make it less aggressive the lower num_waves is */
40 #define VMEM_CLAUSE_MAX_GRAB_DIST (ctx.num_waves * 2)
125 int16_t num_waves; member in struct:aco::sched_ctx
676 /* only move VMEM instructions below descriptor loads. be more aggressive at higher num_waves
678 if (candidate->isVMEM() && (cursor.insert_idx - cursor.source_idx > (ctx.num_waves * 4) ||
780 ctx.last_SMEM_stall = 10 - ctx.num_waves - k
[all...] |
| aco_live_var_analysis.cpp | 366 program->num_waves = 0;
369 program->num_waves = program->dev.physical_sgprs / get_sgpr_alloc(program, new_demand.sgpr);
372 program->num_waves =
373 std::min<uint16_t>(program->num_waves, program->dev.physical_vgprs / vgpr_demand);
408 program->num_waves = std::min<uint16_t>(program->num_waves, program->max_waves);
409 program->max_reg_demand.vgpr = get_addr_vgpr_from_waves(program, program->num_waves);
410 program->max_reg_demand.sgpr = get_addr_sgpr_from_waves(program, program->num_waves);
|
| aco_statistics.cpp | 533 double parallelism = program->num_waves;
553 fprintf(stderr, "num_waves: %u\n", program->num_waves);
|
| aco_spill.cpp | 1883 if (program->num_waves > 0)
1928 assert(program->num_waves > 0);
|
| aco_validate.cpp | 863 uint16_t sgpr_limit = get_addr_sgpr_from_waves(program, program->num_waves);
|
| aco_ir.h | 2041 uint16_t num_waves = 0;
|
| /xsrc/external/mit/MesaLib.old/dist/src/amd/vulkan/ |
| radv_debug.c | 375 unsigned num_waves, FILE *f)
387 for (i = 0; i < num_waves; i++) {
392 if (i == num_waves)
397 num_waves -= i;
419 while (num_waves && start_addr + inst->offset == waves->pc) {
436 num_waves--;
449 unsigned num_waves = ac_get_wave_info(waves); local
452 "\n\n", num_waves);
459 stage, waves, num_waves, f);
465 for (i = 0; i < num_waves; i++)
[all...] |
| /xsrc/external/mit/MesaLib/dist/src/gallium/drivers/radeonsi/ |
| si_debug.c | 900 unsigned num_waves, FILE *f)
912 for (i = 0; i < num_waves; i++) {
916 if (i == num_waves)
921 num_waves -= i;
963 while (num_waves && inst->addr == waves->pc) {
977 num_waves--;
990 unsigned num_waves = ac_get_wave_info(sctx->chip_class, waves); local
992 fprintf(f, COLOR_CYAN "The number of active waves = %u" COLOR_RESET "\n\n", num_waves);
994 si_print_annotated_shader(sctx->shader.vs.current, waves, num_waves, f);
995 si_print_annotated_shader(sctx->shader.tcs.current, waves, num_waves, f)
[all...] |
| gfx10_shader_ngg.c | 687 LLVMValueRef num_waves = get_tgsize(ctx); local
692 * 8 waves: valid_mask = ~0ull >> (64 - num_waves * 8)
693 * 4 waves: valid_mask = ~0 >> (32 - num_waves * 8)
695 LLVMValueRef num_waves8 = LLVMBuildShl(builder, num_waves, LLVMConstInt(ctx->ac.i32, 3, 0), "");
|
| /xsrc/external/mit/MesaLib.old/dist/src/gallium/drivers/radeonsi/ |
| si_debug.c | 916 unsigned num_waves,
927 for (i = 0; i < num_waves; i++) {
931 if (i == num_waves)
936 num_waves -= i;
976 while (num_waves && inst->addr == waves->pc) {
993 num_waves--;
1004 unsigned num_waves = ac_get_wave_info(waves); local
1007 "\n\n", num_waves);
1009 si_print_annotated_shader(sctx->vs_shader.current, waves, num_waves, f);
1010 si_print_annotated_shader(sctx->tcs_shader.current, waves, num_waves, f)
[all...] |
| /xsrc/external/mit/MesaLib/dist/src/amd/vulkan/ |
| radv_debug.c | 301 struct ac_wave_info *waves, unsigned num_waves, FILE *f)
313 for (i = 0; i < num_waves; i++) {
318 if (i == num_waves)
323 num_waves -= i;
344 while (num_waves && start_addr + inst->offset == waves->pc) {
358 num_waves--;
372 unsigned num_waves = ac_get_wave_info(chip_class, waves); local
374 fprintf(f, COLOR_CYAN "The number of active waves = %u" COLOR_RESET "\n\n", num_waves);
381 radv_dump_annotated_shader(pipeline->shaders[stage], stage, waves, num_waves, f);
387 for (i = 0; i < num_waves; i++)
[all...] |
| /xsrc/external/mit/MesaLib/dist/src/amd/common/ |
| ac_debug.c | 807 unsigned num_waves = 0;
824 assert(num_waves < AC_MAX_WAVES_PER_CHIP);
825 w = &waves[num_waves];
833 num_waves++;
837 qsort(waves, num_waves, sizeof(struct ac_wave_info), compare_wave);
840 return num_waves;
|
| ac_nir_lower_ngg.c | 170 /* Broadcast the packed data we read from LDS (to the first 16 lanes, but we only care up to num_waves). */
183 /* Broadcast the packed data we read from LDS (to the first 16 lanes, but we only care up to num_waves). */
275 nir_ssa_def *num_waves = nir_build_load_num_subgroups(b); local
279 nir_ssa_def *wg_num_repacked_invocations = nir_build_read_invocation(b, sum, num_waves);
|
| /xsrc/external/mit/MesaLib.old/dist/src/amd/common/ |
| ac_debug.c | 764 unsigned num_waves = 0; local
780 assert(num_waves < AC_MAX_WAVES_PER_CHIP);
781 w = &waves[num_waves];
790 num_waves++;
794 qsort(waves, num_waves, sizeof(struct ac_wave_info), compare_wave);
797 return num_waves;
|
| /xsrc/external/mit/MesaLib/dist/src/gallium/drivers/r600/ |
| evergreen_compute.c | 604 unsigned num_waves; local
624 /* num_waves = ceil((tg_size.x * tg_size.y, tg_size.z) / (16 * num_pipes)) */
625 num_waves = (info->block[0] * info->block[1] * info->block[2] +
631 num_pipes, num_waves, lds_size);
657 lds_size | (num_waves << 14));
|
| evergreen_state.c | 4544 unsigned num_waves; local
4605 num_waves = ceilf((float)(*num_patches * num_tcs_output_cp) / (float)wave_divisor);
4607 rctx->lds_alloc = (lds_size | (num_waves << 14));
|
| /xsrc/external/mit/MesaLib.old/dist/src/gallium/drivers/r600/ |
| evergreen_compute.c | 583 unsigned num_waves; local
602 /* num_waves = ceil((tg_size.x * tg_size.y, tg_size.z) / (16 * num_pipes)) */
603 num_waves = (info->block[0] * info->block[1] * info->block[2] +
609 num_pipes, num_waves, lds_size);
635 lds_size | (num_waves << 14));
|
| evergreen_state.c | 4511 unsigned num_waves; local
4572 num_waves = ceilf((float)(*num_patches * num_tcs_output_cp) / (float)wave_divisor);
4574 rctx->lds_alloc = (lds_size | (num_waves << 14));
|
