#include "util/debug.h"
#include "ac_exp_param.h"
#include "ac_shader_util.h"
-#include "main/menums.h"
struct radv_blend_state {
uint32_t blend_enable_4bit;
if(pipeline->cs.buf)
free(pipeline->cs.buf);
- vk_free2(&device->alloc, allocator, pipeline);
+
+ vk_object_base_finish(&pipeline->base);
+ vk_free2(&device->vk.alloc, allocator, pipeline);
}
void radv_DestroyPipeline(
}
if (vkms->sampleShadingEnable) {
- ps_iter_samples = ceil(vkms->minSampleShading * num_samples);
+ ps_iter_samples = ceilf(vkms->minSampleShading * num_samples);
ps_iter_samples = util_next_power_of_two(ps_iter_samples);
}
return ps_iter_samples;
}
}
+static void
+radv_set_linked_driver_locations(struct radv_pipeline *pipeline, nir_shader **shaders,
+ struct radv_shader_info infos[MESA_SHADER_STAGES])
+{
+ bool has_tess = shaders[MESA_SHADER_TESS_CTRL];
+ bool has_gs = shaders[MESA_SHADER_GEOMETRY];
+
+ if (!has_tess && !has_gs)
+ return;
+
+ unsigned vs_info_idx = MESA_SHADER_VERTEX;
+ unsigned tes_info_idx = MESA_SHADER_TESS_EVAL;
+
+ if (pipeline->device->physical_device->rad_info.chip_class >= GFX9) {
+ /* These are merged into the next stage */
+ vs_info_idx = has_tess ? MESA_SHADER_TESS_CTRL : MESA_SHADER_GEOMETRY;
+ tes_info_idx = has_gs ? MESA_SHADER_GEOMETRY : MESA_SHADER_TESS_EVAL;
+ }
+
+ if (has_tess) {
+ nir_linked_io_var_info vs2tcs =
+ nir_assign_linked_io_var_locations(shaders[MESA_SHADER_VERTEX], shaders[MESA_SHADER_TESS_CTRL]);
+ nir_linked_io_var_info tcs2tes =
+ nir_assign_linked_io_var_locations(shaders[MESA_SHADER_TESS_CTRL], shaders[MESA_SHADER_TESS_EVAL]);
+
+ infos[vs_info_idx].vs.num_linked_outputs = vs2tcs.num_linked_io_vars;
+ infos[MESA_SHADER_TESS_CTRL].tcs.num_linked_inputs = vs2tcs.num_linked_io_vars;
+ infos[MESA_SHADER_TESS_CTRL].tcs.num_linked_outputs = tcs2tes.num_linked_io_vars;
+ infos[MESA_SHADER_TESS_CTRL].tcs.num_linked_patch_outputs = tcs2tes.num_linked_patch_io_vars;
+ infos[tes_info_idx].tes.num_linked_inputs = tcs2tes.num_linked_io_vars;
+ infos[tes_info_idx].tes.num_linked_patch_inputs = tcs2tes.num_linked_patch_io_vars;
+
+ if (has_gs) {
+ nir_linked_io_var_info tes2gs =
+ nir_assign_linked_io_var_locations(shaders[MESA_SHADER_TESS_EVAL], shaders[MESA_SHADER_GEOMETRY]);
+
+ infos[tes_info_idx].tes.num_linked_outputs = tes2gs.num_linked_io_vars;
+ infos[MESA_SHADER_GEOMETRY].gs.num_linked_inputs = tes2gs.num_linked_io_vars;
+ }
+ } else if (has_gs) {
+ nir_linked_io_var_info vs2gs =
+ nir_assign_linked_io_var_locations(shaders[MESA_SHADER_VERTEX], shaders[MESA_SHADER_GEOMETRY]);
+
+ infos[vs_info_idx].vs.num_linked_outputs = vs2gs.num_linked_io_vars;
+ infos[MESA_SHADER_GEOMETRY].gs.num_linked_inputs = vs2gs.num_linked_io_vars;
+ }
+}
+
static uint32_t
radv_get_attrib_stride(const VkPipelineVertexInputStateCreateInfo *input_state,
uint32_t attrib_binding)
keys[MESA_SHADER_TESS_EVAL].vs_common_out.as_ngg = false;
}
+ if (!device->physical_device->use_ngg_gs) {
+ if (nir[MESA_SHADER_GEOMETRY]) {
+ if (nir[MESA_SHADER_TESS_CTRL])
+ keys[MESA_SHADER_TESS_EVAL].vs_common_out.as_ngg = false;
+ else
+ keys[MESA_SHADER_VERTEX].vs_common_out.as_ngg = false;
+ }
+ }
+
gl_shader_stage last_xfb_stage = MESA_SHADER_VERTEX;
for (int i = MESA_SHADER_VERTEX; i <= MESA_SHADER_GEOMETRY; i++) {
return device->physical_device->ge_wave_size;
}
+static uint8_t
+radv_get_ballot_bit_size(struct radv_device *device,
+ const VkPipelineShaderStageCreateInfo *pStage,
+ gl_shader_stage stage,
+ const struct radv_shader_variant_key *key)
+{
+ if (stage == MESA_SHADER_COMPUTE && key->cs.subgroup_size)
+ return key->cs.subgroup_size;
+ return 64;
+}
+
static void
radv_fill_shader_info(struct radv_pipeline *pipeline,
const VkPipelineShaderStageCreateInfo **pStages,
radv_nir_shader_info_pass(nir[MESA_SHADER_FRAGMENT],
pipeline->layout,
&keys[MESA_SHADER_FRAGMENT],
- &infos[MESA_SHADER_FRAGMENT]);
+ &infos[MESA_SHADER_FRAGMENT],
+ pipeline->device->physical_device->use_aco);
/* TODO: These are no longer used as keys we should refactor this */
keys[MESA_SHADER_VERTEX].vs_common_out.export_prim_id =
infos[MESA_SHADER_FRAGMENT].ps.layer_input;
keys[MESA_SHADER_VERTEX].vs_common_out.export_clip_dists =
!!infos[MESA_SHADER_FRAGMENT].ps.num_input_clips_culls;
+ keys[MESA_SHADER_VERTEX].vs_common_out.export_viewport_index =
+ infos[MESA_SHADER_FRAGMENT].ps.viewport_index_input;
keys[MESA_SHADER_TESS_EVAL].vs_common_out.export_prim_id =
infos[MESA_SHADER_FRAGMENT].ps.prim_id_input;
keys[MESA_SHADER_TESS_EVAL].vs_common_out.export_layer_id =
infos[MESA_SHADER_FRAGMENT].ps.layer_input;
keys[MESA_SHADER_TESS_EVAL].vs_common_out.export_clip_dists =
!!infos[MESA_SHADER_FRAGMENT].ps.num_input_clips_culls;
+ keys[MESA_SHADER_TESS_EVAL].vs_common_out.export_viewport_index =
+ infos[MESA_SHADER_FRAGMENT].ps.viewport_index_input;
/* NGG passthrough mode can't be enabled for vertex shaders
* that export the primitive ID.
filled_stages |= (1 << MESA_SHADER_FRAGMENT);
}
+ if (nir[MESA_SHADER_TESS_CTRL]) {
+ infos[MESA_SHADER_TESS_CTRL].tcs.tes_inputs_read =
+ nir[MESA_SHADER_TESS_EVAL]->info.inputs_read;
+ infos[MESA_SHADER_TESS_CTRL].tcs.tes_patch_inputs_read =
+ nir[MESA_SHADER_TESS_EVAL]->info.patch_inputs_read;
+ }
+
if (pipeline->device->physical_device->rad_info.chip_class >= GFX9 &&
nir[MESA_SHADER_TESS_CTRL]) {
struct nir_shader *combined_nir[] = {nir[MESA_SHADER_VERTEX], nir[MESA_SHADER_TESS_CTRL]};
for (int i = 0; i < 2; i++) {
radv_nir_shader_info_pass(combined_nir[i],
pipeline->layout, &key,
- &infos[MESA_SHADER_TESS_CTRL]);
+ &infos[MESA_SHADER_TESS_CTRL],
+ pipeline->device->physical_device->use_aco);
}
keys[MESA_SHADER_TESS_EVAL].tes.num_patches =
radv_nir_shader_info_pass(combined_nir[i],
pipeline->layout,
&keys[pre_stage],
- &infos[MESA_SHADER_GEOMETRY]);
+ &infos[MESA_SHADER_GEOMETRY],
+ pipeline->device->physical_device->use_aco);
}
filled_stages |= (1 << pre_stage);
radv_nir_shader_info_init(&infos[i]);
radv_nir_shader_info_pass(nir[i], pipeline->layout,
- &keys[i], &infos[i]);
+ &keys[i], &infos[i], pipeline->device->physical_device->use_aco);
}
for (int i = 0; i < MESA_SHADER_STAGES; i++) {
- if (nir[i])
+ if (nir[i]) {
infos[i].wave_size =
radv_get_wave_size(pipeline->device, pStages[i],
i, &keys[i]);
+ infos[i].ballot_bit_size =
+ radv_get_ballot_bit_size(pipeline->device,
+ pStages[i], i,
+ &keys[i]);
+ }
}
}
struct radv_shader_info infos[MESA_SHADER_STAGES] = {0};
unsigned char hash[20], gs_copy_hash[20];
bool keep_executable_info = (flags & VK_PIPELINE_CREATE_CAPTURE_INTERNAL_REPRESENTATIONS_BIT_KHR) || device->keep_shader_info;
+ bool keep_statistic_info = (flags & VK_PIPELINE_CREATE_CAPTURE_STATISTICS_BIT_KHR) || device->keep_shader_info;
radv_start_feedback(pipeline_feedback);
gs_copy_hash[0] ^= 1;
bool found_in_application_cache = true;
- if (modules[MESA_SHADER_GEOMETRY] && !keep_executable_info) {
+ if (modules[MESA_SHADER_GEOMETRY] && !keep_executable_info && !keep_statistic_info) {
struct radv_shader_variant *variants[MESA_SHADER_STAGES] = {0};
radv_create_shader_variants_from_pipeline_cache(device, cache, gs_copy_hash, variants,
&found_in_application_cache);
pipeline->gs_copy_shader = variants[MESA_SHADER_GEOMETRY];
}
- if (!keep_executable_info &&
+ if (!keep_executable_info && !keep_statistic_info &&
radv_create_shader_variants_from_pipeline_cache(device, cache, hash, pipeline->shaders,
&found_in_application_cache) &&
(!modules[MESA_SHADER_GEOMETRY] || pipeline->gs_copy_shader)) {
for (unsigned i = 0; i < MESA_SHADER_STAGES; ++i) {
const VkPipelineShaderStageCreateInfo *stage = pStages[i];
- unsigned subgroup_size = 64;
+ unsigned subgroup_size = 64, ballot_bit_size = 64;
if (!modules[i])
continue;
radv_start_feedback(stage_feedbacks[i]);
if (key->compute_subgroup_size) {
- /* Only GFX10+ and compute shaders currently support
- * requiring a specific subgroup size.
- */
- assert(device->physical_device->rad_info.chip_class >= GFX10 &&
- i == MESA_SHADER_COMPUTE);
+ /* Only compute shaders currently support requiring a
+ * specific subgroup size.
+ */
+ assert(i == MESA_SHADER_COMPUTE);
subgroup_size = key->compute_subgroup_size;
+ ballot_bit_size = key->compute_subgroup_size;
}
nir[i] = radv_shader_compile_to_nir(device, modules[i],
stage ? stage->pName : "main", i,
stage ? stage->pSpecializationInfo : NULL,
flags, pipeline->layout,
- subgroup_size);
+ subgroup_size, ballot_bit_size);
/* We don't want to alter meta shaders IR directly so clone it
* first.
if (!(flags & VK_PIPELINE_CREATE_DISABLE_OPTIMIZATION_BIT))
radv_link_shaders(pipeline, nir);
+ radv_set_linked_driver_locations(pipeline, nir, infos);
+
for (int i = 0; i < MESA_SHADER_STAGES; ++i) {
if (nir[i]) {
+ /* do this again since information such as outputs_read can be out-of-date */
+ nir_shader_gather_info(nir[i], nir_shader_get_entrypoint(nir[i]));
+
if (device->physical_device->use_aco) {
NIR_PASS_V(nir[i], nir_lower_non_uniform_access,
nir_lower_non_uniform_ubo_access |
radv_nir_shader_info_pass(nir[MESA_SHADER_GEOMETRY],
pipeline->layout, &key,
- &info);
+ &info, pipeline->device->physical_device->use_aco);
info.wave_size = 64; /* Wave32 not supported. */
+ info.ballot_bit_size = 64;
pipeline->gs_copy_shader = radv_create_gs_copy_shader(
device, nir[MESA_SHADER_GEOMETRY], &info,
- &gs_copy_binary, keep_executable_info,
+ &gs_copy_binary, keep_executable_info, keep_statistic_info,
keys[MESA_SHADER_GEOMETRY].has_multiview_view_index);
}
- if (!keep_executable_info && pipeline->gs_copy_shader) {
+ if (!keep_executable_info && !keep_statistic_info && pipeline->gs_copy_shader) {
struct radv_shader_binary *binaries[MESA_SHADER_STAGES] = {NULL};
struct radv_shader_variant *variants[MESA_SHADER_STAGES] = {0};
radv_shader_variant_compile(device, modules[MESA_SHADER_FRAGMENT], &nir[MESA_SHADER_FRAGMENT], 1,
pipeline->layout, keys + MESA_SHADER_FRAGMENT,
infos + MESA_SHADER_FRAGMENT,
- keep_executable_info,
+ keep_executable_info, keep_statistic_info,
&binaries[MESA_SHADER_FRAGMENT]);
radv_stop_feedback(stage_feedbacks[MESA_SHADER_FRAGMENT], false);
pipeline->shaders[MESA_SHADER_TESS_CTRL] = radv_shader_variant_compile(device, modules[MESA_SHADER_TESS_CTRL], combined_nir, 2,
pipeline->layout,
&key, &infos[MESA_SHADER_TESS_CTRL], keep_executable_info,
- &binaries[MESA_SHADER_TESS_CTRL]);
+ keep_statistic_info, &binaries[MESA_SHADER_TESS_CTRL]);
radv_stop_feedback(stage_feedbacks[MESA_SHADER_TESS_CTRL], false);
}
pipeline->shaders[MESA_SHADER_GEOMETRY] = radv_shader_variant_compile(device, modules[MESA_SHADER_GEOMETRY], combined_nir, 2,
pipeline->layout,
&keys[pre_stage], &infos[MESA_SHADER_GEOMETRY], keep_executable_info,
- &binaries[MESA_SHADER_GEOMETRY]);
+ keep_statistic_info, &binaries[MESA_SHADER_GEOMETRY]);
radv_stop_feedback(stage_feedbacks[MESA_SHADER_GEOMETRY], false);
}
pipeline->shaders[i] = radv_shader_variant_compile(device, modules[i], &nir[i], 1,
pipeline->layout,
- keys + i, infos + i,keep_executable_info,
- &binaries[i]);
+ keys + i, infos + i, keep_executable_info,
+ keep_statistic_info, &binaries[i]);
radv_stop_feedback(stage_feedbacks[i], false);
}
}
- if (!keep_executable_info) {
+ if (!keep_executable_info && !keep_statistic_info) {
radv_pipeline_cache_insert_shaders(device, cache, hash, pipeline->shaders,
binaries);
}
db_render_control |= S_028000_DEPTH_CLEAR_ENABLE(extra->db_depth_clear);
db_render_control |= S_028000_STENCIL_CLEAR_ENABLE(extra->db_stencil_clear);
- db_render_control |= S_028000_RESUMMARIZE_ENABLE(extra->db_resummarize);
- db_render_control |= S_028000_DEPTH_COMPRESS_DISABLE(extra->db_flush_depth_inplace);
- db_render_control |= S_028000_STENCIL_COMPRESS_DISABLE(extra->db_flush_stencil_inplace);
+ db_render_control |= S_028000_RESUMMARIZE_ENABLE(extra->resummarize_enable);
+ db_render_control |= S_028000_DEPTH_COMPRESS_DISABLE(extra->depth_compress_disable);
+ db_render_control |= S_028000_STENCIL_COMPRESS_DISABLE(extra->stencil_compress_disable);
db_render_override2 |= S_028010_DISABLE_ZMASK_EXPCLEAR_OPTIMIZATION(extra->db_depth_disable_expclear);
db_render_override2 |= S_028010_DISABLE_SMEM_EXPCLEAR_OPTIMIZATION(extra->db_stencil_disable_expclear);
}
*
* Requirement: GE_CNTL.VERT_GRP_SIZE = VGT_GS_ONCHIP_CNTL.ES_VERTS_PER_SUBGRP - 5
*/
- if ((pipeline->device->physical_device->rad_info.family == CHIP_NAVI10 ||
- pipeline->device->physical_device->rad_info.family == CHIP_NAVI12 ||
- pipeline->device->physical_device->rad_info.family == CHIP_NAVI14) &&
+ if (pipeline->device->physical_device->rad_info.chip_class == GFX10 &&
!radv_pipeline_has_tess(pipeline) &&
ngg_state->hw_max_esverts != 256) {
ge_cntl &= C_03096C_VERT_GRP_SIZE;
++ps_offset;
}
+ if (ps->info.ps.viewport_index_input) {
+ unsigned vs_offset = outinfo->vs_output_param_offset[VARYING_SLOT_VIEWPORT];
+ if (vs_offset != AC_EXP_PARAM_UNDEFINED)
+ ps_input_cntl[ps_offset] = offset_to_ps_input(vs_offset, true, false, false);
+ else
+ ps_input_cntl[ps_offset] = offset_to_ps_input(AC_EXP_PARAM_DEFAULT_VAL_0000, true, false, false);
+ ++ps_offset;
+ }
+
if (ps->info.ps.has_pcoord) {
unsigned val;
val = S_028644_PT_SPRITE_TEX(1) | S_028644_OFFSET(0x20);
struct radv_pipeline *pipeline;
VkResult result;
- pipeline = vk_zalloc2(&device->alloc, pAllocator, sizeof(*pipeline), 8,
+ pipeline = vk_zalloc2(&device->vk.alloc, pAllocator, sizeof(*pipeline), 8,
VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
if (pipeline == NULL)
return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
+ vk_object_base_init(&device->vk, &pipeline->base,
+ VK_OBJECT_TYPE_PIPELINE);
+
result = radv_pipeline_init(pipeline, device, cache,
pCreateInfo, extra);
if (result != VK_SUCCESS) {
struct radv_pipeline *pipeline;
VkResult result;
- pipeline = vk_zalloc2(&device->alloc, pAllocator, sizeof(*pipeline), 8,
+ pipeline = vk_zalloc2(&device->vk.alloc, pAllocator, sizeof(*pipeline), 8,
VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
if (pipeline == NULL)
return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
+ vk_object_base_init(&device->vk, &pipeline->base,
+ VK_OBJECT_TYPE_PIPELINE);
+
pipeline->device = device;
pipeline->layout = radv_pipeline_layout_from_handle(pCreateInfo->layout);
assert(pipeline->layout);
}
++s;
+ if (shader->statistics) {
+ for (unsigned i = 0; i < shader->statistics->count; i++) {
+ struct radv_compiler_statistic_info *info = &shader->statistics->infos[i];
+ uint32_t value = shader->statistics->values[i];
+ if (s < end) {
+ desc_copy(s->name, info->name);
+ desc_copy(s->description, info->desc);
+ s->format = VK_PIPELINE_EXECUTABLE_STATISTIC_FORMAT_UINT64_KHR;
+ s->value.u64 = value;
+ }
+ ++s;
+ }
+ }
+
if (!pStatistics)
*pStatisticCount = s - pStatistics;
else if (s > end) {