diff options
author | Bas Nieuwenhuizen <[email protected]> | 2018-01-15 00:41:59 +0100 |
---|---|---|
committer | Bas Nieuwenhuizen <[email protected]> | 2018-01-30 22:01:26 +0100 |
commit | 694c34314b5ff707d4e6eedcb1cb7f46f4e09c33 (patch) | |
tree | f179b8ced07a808550ca4cf4dd0382bce59da7ae | |
parent | 0bea0851aa65d62d67f935a901505f1f00699e84 (diff) |
radv: Split out the ia_multi_vgt_param precomputation.
Also moved everything in a struct and then return the struct from
the helper function, so it is clear in the caller what part of the
pipeline gets modified.
Reviewed-by: Dave Airlie <[email protected]>
Reviewed-by: Samuel Pitoiset <[email protected]>
-rw-r--r-- | src/amd/vulkan/radv_pipeline.c | 169 | ||||
-rw-r--r-- | src/amd/vulkan/radv_private.h | 16 | ||||
-rw-r--r-- | src/amd/vulkan/si_cmd_buffer.c | 12 |
3 files changed, 106 insertions, 91 deletions
diff --git a/src/amd/vulkan/radv_pipeline.c b/src/amd/vulkan/radv_pipeline.c index 8ca3497b12c..7b7ced5cd79 100644 --- a/src/amd/vulkan/radv_pipeline.c +++ b/src/amd/vulkan/radv_pipeline.c @@ -2921,6 +2921,95 @@ radv_pipeline_generate_pm4(struct radv_pipeline *pipeline, assert(pipeline->cs.cdw <= pipeline->cs.max_dw); } +static struct radv_ia_multi_vgt_param_helpers +radv_compute_ia_multi_vgt_param_helpers(struct radv_pipeline *pipeline) +{ + struct radv_ia_multi_vgt_param_helpers ia_multi_vgt_param = {0}; + const struct radv_device *device = pipeline->device; + + if (radv_pipeline_has_tess(pipeline)) + ia_multi_vgt_param.primgroup_size = pipeline->graphics.tess.num_patches; + else if (radv_pipeline_has_gs(pipeline)) + ia_multi_vgt_param.primgroup_size = 64; + else + ia_multi_vgt_param.primgroup_size = 128; /* recommended without a GS */ + + ia_multi_vgt_param.partial_es_wave = false; + if (pipeline->device->has_distributed_tess) { + if (radv_pipeline_has_gs(pipeline)) { + if (device->physical_device->rad_info.chip_class <= VI) + ia_multi_vgt_param.partial_es_wave = true; + } + } + /* GS requirement. */ + if (SI_GS_PER_ES / ia_multi_vgt_param.primgroup_size >= pipeline->device->gs_table_depth - 3) + ia_multi_vgt_param.partial_es_wave = true; + + ia_multi_vgt_param.wd_switch_on_eop = false; + if (device->physical_device->rad_info.chip_class >= CIK) { + unsigned prim = pipeline->graphics.prim; + /* WD_SWITCH_ON_EOP has no effect on GPUs with less than + * 4 shader engines. Set 1 to pass the assertion below. + * The other cases are hardware requirements. */ + if (device->physical_device->rad_info.max_se < 4 || + prim == V_008958_DI_PT_POLYGON || + prim == V_008958_DI_PT_LINELOOP || + prim == V_008958_DI_PT_TRIFAN || + prim == V_008958_DI_PT_TRISTRIP_ADJ || + (pipeline->graphics.prim_restart_enable && + (device->physical_device->rad_info.family < CHIP_POLARIS10 || + (prim != V_008958_DI_PT_POINTLIST && + prim != V_008958_DI_PT_LINESTRIP && + prim != V_008958_DI_PT_TRISTRIP)))) + ia_multi_vgt_param.wd_switch_on_eop = true; + } + + ia_multi_vgt_param.ia_switch_on_eoi = false; + if (pipeline->shaders[MESA_SHADER_FRAGMENT]->info.fs.prim_id_input) + ia_multi_vgt_param.ia_switch_on_eoi = true; + if (radv_pipeline_has_gs(pipeline) && + pipeline->shaders[MESA_SHADER_GEOMETRY]->info.info.uses_prim_id) + ia_multi_vgt_param.ia_switch_on_eoi = true; + if (radv_pipeline_has_tess(pipeline)) { + /* SWITCH_ON_EOI must be set if PrimID is used. */ + if (pipeline->shaders[MESA_SHADER_TESS_CTRL]->info.info.uses_prim_id || + radv_get_tess_eval_shader(pipeline)->info.info.uses_prim_id) + ia_multi_vgt_param.ia_switch_on_eoi = true; + } + + ia_multi_vgt_param.partial_vs_wave = false; + if (radv_pipeline_has_tess(pipeline)) { + /* Bug with tessellation and GS on Bonaire and older 2 SE chips. */ + if ((device->physical_device->rad_info.family == CHIP_TAHITI || + device->physical_device->rad_info.family == CHIP_PITCAIRN || + device->physical_device->rad_info.family == CHIP_BONAIRE) && + radv_pipeline_has_gs(pipeline)) + ia_multi_vgt_param.partial_vs_wave = true; + /* Needed for 028B6C_DISTRIBUTION_MODE != 0 */ + if (device->has_distributed_tess) { + if (radv_pipeline_has_gs(pipeline)) { + if (device->physical_device->rad_info.family == CHIP_TONGA || + device->physical_device->rad_info.family == CHIP_FIJI || + device->physical_device->rad_info.family == CHIP_POLARIS10 || + device->physical_device->rad_info.family == CHIP_POLARIS11 || + device->physical_device->rad_info.family == CHIP_POLARIS12) + ia_multi_vgt_param.partial_vs_wave = true; + } else { + ia_multi_vgt_param.partial_vs_wave = true; + } + } + } + + ia_multi_vgt_param.base = + S_028AA8_PRIMGROUP_SIZE(ia_multi_vgt_param.primgroup_size - 1) | + /* The following field was moved to VGT_SHADER_STAGES_EN in GFX9. */ + S_028AA8_MAX_PRIMGRP_IN_WAVE(device->physical_device->rad_info.chip_class == VI ? 2 : 0) | + S_030960_EN_INST_OPT_BASIC(device->physical_device->rad_info.chip_class >= GFX9) | + S_030960_EN_INST_OPT_ADV(device->physical_device->rad_info.chip_class >= GFX9); + + return ia_multi_vgt_param; +} + static VkResult radv_pipeline_init(struct radv_pipeline *pipeline, struct radv_device *device, @@ -3041,85 +3130,7 @@ radv_pipeline_init(struct radv_pipeline *pipeline, calculate_tess_state(pipeline, pCreateInfo); } - if (radv_pipeline_has_tess(pipeline)) - pipeline->graphics.primgroup_size = pipeline->graphics.tess.num_patches; - else if (radv_pipeline_has_gs(pipeline)) - pipeline->graphics.primgroup_size = 64; - else - pipeline->graphics.primgroup_size = 128; /* recommended without a GS */ - - pipeline->graphics.partial_es_wave = false; - if (pipeline->device->has_distributed_tess) { - if (radv_pipeline_has_gs(pipeline)) { - if (device->physical_device->rad_info.chip_class <= VI) - pipeline->graphics.partial_es_wave = true; - } - } - /* GS requirement. */ - if (SI_GS_PER_ES / pipeline->graphics.primgroup_size >= pipeline->device->gs_table_depth - 3) - pipeline->graphics.partial_es_wave = true; - - pipeline->graphics.wd_switch_on_eop = false; - if (device->physical_device->rad_info.chip_class >= CIK) { - unsigned prim = pipeline->graphics.prim; - /* WD_SWITCH_ON_EOP has no effect on GPUs with less than - * 4 shader engines. Set 1 to pass the assertion below. - * The other cases are hardware requirements. */ - if (device->physical_device->rad_info.max_se < 4 || - prim == V_008958_DI_PT_POLYGON || - prim == V_008958_DI_PT_LINELOOP || - prim == V_008958_DI_PT_TRIFAN || - prim == V_008958_DI_PT_TRISTRIP_ADJ || - (pipeline->graphics.prim_restart_enable && - (device->physical_device->rad_info.family < CHIP_POLARIS10 || - (prim != V_008958_DI_PT_POINTLIST && - prim != V_008958_DI_PT_LINESTRIP && - prim != V_008958_DI_PT_TRISTRIP)))) - pipeline->graphics.wd_switch_on_eop = true; - } - - pipeline->graphics.ia_switch_on_eoi = false; - if (pipeline->shaders[MESA_SHADER_FRAGMENT]->info.fs.prim_id_input) - pipeline->graphics.ia_switch_on_eoi = true; - if (radv_pipeline_has_gs(pipeline) && - pipeline->shaders[MESA_SHADER_GEOMETRY]->info.info.uses_prim_id) - pipeline->graphics.ia_switch_on_eoi = true; - if (radv_pipeline_has_tess(pipeline)) { - /* SWITCH_ON_EOI must be set if PrimID is used. */ - if (pipeline->shaders[MESA_SHADER_TESS_CTRL]->info.info.uses_prim_id || - radv_get_tess_eval_shader(pipeline)->info.info.uses_prim_id) - pipeline->graphics.ia_switch_on_eoi = true; - } - - pipeline->graphics.partial_vs_wave = false; - if (radv_pipeline_has_tess(pipeline)) { - /* Bug with tessellation and GS on Bonaire and older 2 SE chips. */ - if ((device->physical_device->rad_info.family == CHIP_TAHITI || - device->physical_device->rad_info.family == CHIP_PITCAIRN || - device->physical_device->rad_info.family == CHIP_BONAIRE) && - radv_pipeline_has_gs(pipeline)) - pipeline->graphics.partial_vs_wave = true; - /* Needed for 028B6C_DISTRIBUTION_MODE != 0 */ - if (device->has_distributed_tess) { - if (radv_pipeline_has_gs(pipeline)) { - if (device->physical_device->rad_info.family == CHIP_TONGA || - device->physical_device->rad_info.family == CHIP_FIJI || - device->physical_device->rad_info.family == CHIP_POLARIS10 || - device->physical_device->rad_info.family == CHIP_POLARIS11 || - device->physical_device->rad_info.family == CHIP_POLARIS12) - pipeline->graphics.partial_vs_wave = true; - } else { - pipeline->graphics.partial_vs_wave = true; - } - } - } - - pipeline->graphics.base_ia_multi_vgt_param = - S_028AA8_PRIMGROUP_SIZE(pipeline->graphics.primgroup_size - 1) | - /* The following field was moved to VGT_SHADER_STAGES_EN in GFX9. */ - S_028AA8_MAX_PRIMGRP_IN_WAVE(device->physical_device->rad_info.chip_class == VI ? 2 : 0) | - S_030960_EN_INST_OPT_BASIC(device->physical_device->rad_info.chip_class >= GFX9) | - S_030960_EN_INST_OPT_ADV(device->physical_device->rad_info.chip_class >= GFX9); + pipeline->graphics.ia_multi_vgt_param = radv_compute_ia_multi_vgt_param_helpers(pipeline); const VkPipelineVertexInputStateCreateInfo *vi_info = pCreateInfo->pVertexInputState; diff --git a/src/amd/vulkan/radv_private.h b/src/amd/vulkan/radv_private.h index 26c0c1074bd..dd757fe41f0 100644 --- a/src/amd/vulkan/radv_private.h +++ b/src/amd/vulkan/radv_private.h @@ -1189,6 +1189,15 @@ struct radv_vertex_elements_info { uint32_t count; }; +struct radv_ia_multi_vgt_param_helpers { + uint32_t base; + bool partial_es_wave; + uint8_t primgroup_size; + bool wd_switch_on_eop; + bool ia_switch_on_eoi; + bool partial_vs_wave; +}; + #define SI_GS_PER_ES 128 struct radv_pipeline { @@ -1223,16 +1232,11 @@ struct radv_pipeline { uint32_t vgt_gs_mode; bool vgt_primitiveid_en; bool prim_restart_enable; - bool partial_es_wave; - uint8_t primgroup_size; unsigned esgs_ring_size; unsigned gsvs_ring_size; uint32_t vgt_shader_stages_en; uint32_t vtx_base_sgpr; - uint32_t base_ia_multi_vgt_param; - bool wd_switch_on_eop; - bool ia_switch_on_eoi; - bool partial_vs_wave; + struct radv_ia_multi_vgt_param_helpers ia_multi_vgt_param; uint8_t vtx_emit_num; struct radv_prim_vertex_count prim_vertex_count; bool can_use_guardband; diff --git a/src/amd/vulkan/si_cmd_buffer.c b/src/amd/vulkan/si_cmd_buffer.c index f5f4eefcd28..d9c78bf1170 100644 --- a/src/amd/vulkan/si_cmd_buffer.c +++ b/src/amd/vulkan/si_cmd_buffer.c @@ -741,21 +741,21 @@ si_get_ia_multi_vgt_param(struct radv_cmd_buffer *cmd_buffer, bool ia_switch_on_eop = false; bool ia_switch_on_eoi = false; bool partial_vs_wave = false; - bool partial_es_wave = cmd_buffer->state.pipeline->graphics.partial_es_wave; + bool partial_es_wave = cmd_buffer->state.pipeline->graphics.ia_multi_vgt_param.partial_es_wave; bool multi_instances_smaller_than_primgroup; multi_instances_smaller_than_primgroup = indirect_draw; if (!multi_instances_smaller_than_primgroup && instanced_draw) { uint32_t num_prims = radv_prims_for_vertices(&cmd_buffer->state.pipeline->graphics.prim_vertex_count, draw_vertex_count); - if (num_prims < cmd_buffer->state.pipeline->graphics.primgroup_size) + if (num_prims < cmd_buffer->state.pipeline->graphics.ia_multi_vgt_param.primgroup_size) multi_instances_smaller_than_primgroup = true; } - ia_switch_on_eoi = cmd_buffer->state.pipeline->graphics.ia_switch_on_eoi; - partial_vs_wave = cmd_buffer->state.pipeline->graphics.partial_vs_wave; + ia_switch_on_eoi = cmd_buffer->state.pipeline->graphics.ia_multi_vgt_param.ia_switch_on_eoi; + partial_vs_wave = cmd_buffer->state.pipeline->graphics.ia_multi_vgt_param.partial_vs_wave; if (chip_class >= CIK) { - wd_switch_on_eop = cmd_buffer->state.pipeline->graphics.wd_switch_on_eop; + wd_switch_on_eop = cmd_buffer->state.pipeline->graphics.ia_multi_vgt_param.wd_switch_on_eop; /* Hawaii hangs if instancing is enabled and WD_SWITCH_ON_EOP is 0. * We don't know that for indirect drawing, so treat it as @@ -815,7 +815,7 @@ si_get_ia_multi_vgt_param(struct radv_cmd_buffer *cmd_buffer, } } - return cmd_buffer->state.pipeline->graphics.base_ia_multi_vgt_param | + return cmd_buffer->state.pipeline->graphics.ia_multi_vgt_param.base | S_028AA8_SWITCH_ON_EOP(ia_switch_on_eop) | S_028AA8_SWITCH_ON_EOI(ia_switch_on_eoi) | S_028AA8_PARTIAL_VS_WAVE_ON(partial_vs_wave) | |