| /third_party/mesa3d/src/amd/compiler/ |
| D | 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)
40 #define VMEM_CLAUSE_MAX_GRAB_DIST (ctx.num_waves * 2)
125 int16_t num_waves; member
681 (cursor.insert_idx - cursor.source_idx > (ctx.num_waves * 4) || in schedule_SMEM()
783 ctx.last_SMEM_stall = 10 - ctx.num_waves - k; in schedule_SMEM()
1065 if (program->num_waves <= 5 * wave_fac) in schedule_program()
1066 ctx.num_waves = program->num_waves; in schedule_program()
[all …]
|
| D | aco_live_var_analysis.cpp | 416 program->num_waves = 0; in update_vgpr_sgpr_demand()
419 program->num_waves = program->dev.physical_sgprs / get_sgpr_alloc(program, new_demand.sgpr); in update_vgpr_sgpr_demand()
422 program->num_waves = in update_vgpr_sgpr_demand()
423 std::min<uint16_t>(program->num_waves, program->dev.physical_vgprs / vgpr_demand); in update_vgpr_sgpr_demand()
425 program->num_waves = std::min(program->num_waves, max_waves); in update_vgpr_sgpr_demand()
428 program->num_waves = max_suitable_waves(program, program->num_waves); in update_vgpr_sgpr_demand()
429 program->max_reg_demand.vgpr = get_addr_vgpr_from_waves(program, program->num_waves); in update_vgpr_sgpr_demand()
430 program->max_reg_demand.sgpr = get_addr_sgpr_from_waves(program, program->num_waves); in update_vgpr_sgpr_demand()
|
| D | aco_statistics.cpp | 534 double parallelism = program->num_waves; in collect_preasm_stats()
554 fprintf(stderr, "num_waves: %u\n", program->num_waves); in collect_preasm_stats()
|
| D | aco_spill.cpp | 1926 if (program->num_waves > 0) in spill()
1974 assert(program->num_waves > 0); in spill()
|
| D | aco_validate.cpp | 999 uint16_t sgpr_limit = get_addr_sgpr_from_waves(program, program->num_waves); in validate_ra()
|
| D | aco_ir.h | 2103 uint16_t num_waves = 0; variable
|
| /third_party/mesa3d/src/gallium/drivers/radeonsi/ |
| D | si_debug.c | 887 unsigned num_waves, FILE *f) in si_print_annotated_shader() argument
899 for (i = 0; i < num_waves; i++) { in si_print_annotated_shader()
903 if (i == num_waves) in si_print_annotated_shader()
908 num_waves -= i; in si_print_annotated_shader()
945 while (num_waves && inst->addr == waves->pc) { in si_print_annotated_shader()
959 num_waves--; in si_print_annotated_shader()
972 unsigned num_waves = ac_get_wave_info(sctx->gfx_level, waves); in si_dump_annotated_shaders() local
974 fprintf(f, COLOR_CYAN "The number of active waves = %u" COLOR_RESET "\n\n", num_waves); in si_dump_annotated_shaders()
976 si_print_annotated_shader(sctx->shader.vs.current, waves, num_waves, f); in si_dump_annotated_shaders()
977 si_print_annotated_shader(sctx->shader.tcs.current, waves, num_waves, f); in si_dump_annotated_shaders()
[all …]
|
| D | gfx10_shader_ngg.c | 778 LLVMValueRef num_waves = get_tgsize(ctx); in load_vertex_counts() local
786 LLVMValueRef num_waves8 = LLVMBuildShl(builder, num_waves, LLVMConstInt(ctx->ac.i32, 3, 0), ""); in load_vertex_counts()
|
| /third_party/mesa3d/src/amd/vulkan/ |
| D | radv_debug.c | 302 struct ac_wave_info *waves, unsigned num_waves, FILE *f) in radv_dump_annotated_shader() argument
314 for (i = 0; i < num_waves; i++) { in radv_dump_annotated_shader()
319 if (i == num_waves) in radv_dump_annotated_shader()
324 num_waves -= i; in radv_dump_annotated_shader()
345 while (num_waves && start_addr + inst->offset == waves->pc) { in radv_dump_annotated_shader()
359 num_waves--; in radv_dump_annotated_shader()
373 unsigned num_waves = ac_get_wave_info(gfx_level, waves); in radv_dump_annotated_shaders() local
375 fprintf(f, COLOR_CYAN "The number of active waves = %u" COLOR_RESET "\n\n", num_waves); in radv_dump_annotated_shaders()
382 radv_dump_annotated_shader(pipeline->shaders[stage], stage, waves, num_waves, f); in radv_dump_annotated_shaders()
388 for (i = 0; i < num_waves; i++) { in radv_dump_annotated_shaders()
|
| /third_party/mesa3d/src/amd/common/ |
| D | ac_debug.c | 822 unsigned num_waves = 0; in ac_get_wave_info()
839 assert(num_waves < AC_MAX_WAVES_PER_CHIP); in ac_get_wave_info()
840 w = &waves[num_waves]; in ac_get_wave_info()
848 num_waves++; in ac_get_wave_info()
852 qsort(waves, num_waves, sizeof(struct ac_wave_info), compare_wave); in ac_get_wave_info()
855 return num_waves; in ac_get_wave_info()
|
| D | ac_nir_lower_ngg.c | 348 nir_ssa_def *num_waves = nir_load_num_subgroups(b); in repack_invocations_in_workgroup() local
352 nir_ssa_def *wg_num_repacked_invocations = nir_read_invocation(b, sum, num_waves); in repack_invocations_in_workgroup()
|
| /third_party/mesa3d/src/gallium/drivers/r600/ |
| D | evergreen_compute.c | 610 unsigned num_waves; in evergreen_emit_dispatch() local
626 num_waves = (info->block[0] * info->block[1] * info->block[2] + in evergreen_emit_dispatch()
632 num_pipes, num_waves, lds_size); in evergreen_emit_dispatch()
658 lds_size | (num_waves << 14)); in evergreen_emit_dispatch()
|
| D | evergreen_state.c | 4539 unsigned num_waves; in evergreen_setup_tess_constants() local
4600 num_waves = ceilf((float)(*num_patches * num_tcs_output_cp) / (float)wave_divisor); in evergreen_setup_tess_constants()
4602 rctx->lds_alloc = (lds_size | (num_waves << 14)); in evergreen_setup_tess_constants()
|
| /third_party/mesa3d/docs/relnotes/ |
| D | 20.3.1.rst | 160 - aco: Use program->num_waves as maximum in scheduler.
|
| D | 20.2.5.rst | 164 - aco: Use program->num_waves as maximum in scheduler.
|
| D | 22.2.0.rst | 1400 - aco: remove 'max_waves' and use 'num_waves' to adjust for LDS and workgroup size
1401 - aco: split num_waves adjustment into separate function
1402 - aco: adjust num_waves for LDS before scheduling
|
| D | 19.3.0.rst | 2859 - aco: take LDS into account when calculating num_waves
|
| D | 21.0.0.rst | 3160 - aco: Use program->num_waves as maximum in scheduler.
|
| D | 21.1.0.rst | 4704 - aco: fix num_waves on GFX10+
|