if (scratch_bytes_per_wave && max_waves < min_waves) {
/* Not really true at this moment, but will be true on first
* execution. Avoid having hanging shaders. */
- return VK_ERROR_OUT_OF_DEVICE_MEMORY;
+ return vk_error(VK_ERROR_OUT_OF_DEVICE_MEMORY);
}
pipeline->scratch_bytes_per_wave = scratch_bytes_per_wave;
pipeline->max_waves = max_waves;
radv_pipeline_init_dynamic_state(struct radv_pipeline *pipeline,
const VkGraphicsPipelineCreateInfo *pCreateInfo)
{
- radv_cmd_dirty_mask_t states = RADV_CMD_DIRTY_DYNAMIC_ALL;
+ uint32_t states = RADV_CMD_DIRTY_DYNAMIC_ALL;
RADV_FROM_HANDLE(radv_render_pass, pass, pCreateInfo->renderPass);
struct radv_subpass *subpass = &pass->subpasses[pCreateInfo->subpass];
}
}
- pipeline->dynamic_state_mask = states;
+ pipeline->dynamic_state.mask = states;
}
-static struct ac_shader_variant_key
-radv_compute_vs_key(const VkGraphicsPipelineCreateInfo *pCreateInfo, bool as_es, bool as_ls)
-{
- struct ac_shader_variant_key key;
- const VkPipelineVertexInputStateCreateInfo *input_state =
- pCreateInfo->pVertexInputState;
-
- memset(&key, 0, sizeof(key));
- key.vs.instance_rate_inputs = 0;
- key.vs.as_es = as_es;
- key.vs.as_ls = as_ls;
-
- for (unsigned i = 0; i < input_state->vertexAttributeDescriptionCount; ++i) {
- unsigned binding;
- binding = input_state->pVertexAttributeDescriptions[i].binding;
- if (input_state->pVertexBindingDescriptions[binding].inputRate)
- key.vs.instance_rate_inputs |= 1u << input_state->pVertexAttributeDescriptions[i].location;
- }
- return key;
-}
-
-
static void calculate_gfx9_gs_info(const VkGraphicsPipelineCreateInfo *pCreateInfo,
struct radv_pipeline *pipeline)
{
if (pipeline->device->physical_device->rad_info.chip_class <= VI)
pipeline->graphics.esgs_ring_size = CLAMP(esgs_ring_size, min_esgs_ring_size, max_size);
+ pipeline->graphics.gs.vgt_esgs_ring_itemsize = es_info->esgs_itemsize / 4;
pipeline->graphics.gsvs_ring_size = MIN2(gsvs_ring_size, max_size);
}
}
}
-static void calculate_pa_cl_vs_out_cntl(struct radv_pipeline *pipeline)
+static void calculate_vs_outinfo(struct radv_pipeline *pipeline)
{
struct ac_vs_output_info *outinfo = get_vs_output_info(pipeline);
bool misc_vec_ena = outinfo->writes_pointsize ||
outinfo->writes_layer ||
outinfo->writes_viewport_index;
- pipeline->graphics.pa_cl_vs_out_cntl =
+ pipeline->graphics.vs.pa_cl_vs_out_cntl =
S_02881C_USE_VTX_POINT_SIZE(outinfo->writes_pointsize) |
S_02881C_USE_VTX_RENDER_TARGET_INDX(outinfo->writes_layer) |
S_02881C_USE_VTX_VIEWPORT_INDX(outinfo->writes_viewport_index) |
cull_dist_mask << 8 |
clip_dist_mask;
+ pipeline->graphics.vs.spi_shader_pos_format =
+ S_02870C_POS0_EXPORT_FORMAT(V_02870C_SPI_SHADER_4COMP) |
+ S_02870C_POS1_EXPORT_FORMAT(outinfo->pos_exports > 1 ?
+ V_02870C_SPI_SHADER_4COMP :
+ V_02870C_SPI_SHADER_NONE) |
+ S_02870C_POS2_EXPORT_FORMAT(outinfo->pos_exports > 2 ?
+ V_02870C_SPI_SHADER_4COMP :
+ V_02870C_SPI_SHADER_NONE) |
+ S_02870C_POS3_EXPORT_FORMAT(outinfo->pos_exports > 3 ?
+ V_02870C_SPI_SHADER_4COMP :
+ V_02870C_SPI_SHADER_NONE);
+
+ pipeline->graphics.vs.spi_vs_out_config = S_0286C4_VS_EXPORT_COUNT(MAX2(1, outinfo->param_exports) - 1);
+ /* only emitted on pre-VI */
+ pipeline->graphics.vs.vgt_reuse_off = S_028AB4_REUSE_OFF(outinfo->writes_viewport_index);
}
static uint32_t offset_to_ps_input(uint32_t offset, bool flat_shade)
}
}
+
+static struct radv_pipeline_key
+radv_generate_graphics_pipeline_key(struct radv_pipeline *pipeline,
+ const VkGraphicsPipelineCreateInfo *pCreateInfo,
+ bool has_view_index)
+{
+ const VkPipelineVertexInputStateCreateInfo *input_state =
+ pCreateInfo->pVertexInputState;
+ struct radv_pipeline_key key;
+ memset(&key, 0, sizeof(key));
+
+ key.has_multiview_view_index = has_view_index;
+
+ for (unsigned i = 0; i < input_state->vertexAttributeDescriptionCount; ++i) {
+ unsigned binding;
+ binding = input_state->pVertexAttributeDescriptions[i].binding;
+ if (input_state->pVertexBindingDescriptions[binding].inputRate)
+ key.instance_rate_inputs |= 1u << input_state->pVertexAttributeDescriptions[i].location;
+ }
+
+ if (pCreateInfo->pTessellationState)
+ key.tess_input_vertices = pCreateInfo->pTessellationState->patchControlPoints;
+
+
+ if (pCreateInfo->pMultisampleState &&
+ pCreateInfo->pMultisampleState->rasterizationSamples > 1)
+ key.multisample = true;
+
+ key.col_format = pipeline->graphics.blend.spi_shader_col_format;
+ if (pipeline->device->physical_device->rad_info.chip_class < VI)
+ radv_pipeline_compute_get_int_clamp(pCreateInfo, &key.is_int8, &key.is_int10);
+
+ return key;
+}
+
+static void
+radv_fill_shader_keys(struct ac_shader_variant_key *keys,
+ const struct radv_pipeline_key *key,
+ nir_shader **nir)
+{
+ keys[MESA_SHADER_VERTEX].vs.instance_rate_inputs = key->instance_rate_inputs;
+
+ if (nir[MESA_SHADER_TESS_CTRL]) {
+ keys[MESA_SHADER_VERTEX].vs.as_ls = true;
+ keys[MESA_SHADER_TESS_CTRL].tcs.input_vertices = key->tess_input_vertices;
+ keys[MESA_SHADER_TESS_CTRL].tcs.primitive_mode = nir[MESA_SHADER_TESS_EVAL]->info.tess.primitive_mode;
+
+ keys[MESA_SHADER_TESS_CTRL].tcs.tes_reads_tess_factors = !!(nir[MESA_SHADER_TESS_EVAL]->info.inputs_read & (VARYING_BIT_TESS_LEVEL_INNER | VARYING_BIT_TESS_LEVEL_OUTER));
+ }
+
+ if (nir[MESA_SHADER_GEOMETRY]) {
+ if (nir[MESA_SHADER_TESS_CTRL])
+ keys[MESA_SHADER_TESS_EVAL].tes.as_es = true;
+ else
+ keys[MESA_SHADER_VERTEX].vs.as_es = true;
+ }
+
+ for(int i = 0; i < MESA_SHADER_STAGES; ++i)
+ keys[i].has_multiview_view_index = key->has_multiview_view_index;
+
+ keys[MESA_SHADER_FRAGMENT].fs.multisample = key->multisample;
+ keys[MESA_SHADER_FRAGMENT].fs.col_format = key->col_format;
+ keys[MESA_SHADER_FRAGMENT].fs.is_int8 = key->is_int8;
+ keys[MESA_SHADER_FRAGMENT].fs.is_int10 = key->is_int10;
+}
+
static
void radv_create_shaders(struct radv_pipeline *pipeline,
struct radv_device *device,
struct radv_pipeline_cache *cache,
- struct ac_shader_variant_key *keys,
+ struct radv_pipeline_key key,
const VkPipelineShaderStageCreateInfo **pStages)
{
struct radv_shader_module fs_m = {0};
nir_shader *nir[MESA_SHADER_STAGES] = {0};
void *codes[MESA_SHADER_STAGES] = {0};
unsigned code_sizes[MESA_SHADER_STAGES] = {0};
+ struct ac_shader_variant_key keys[MESA_SHADER_STAGES] = {{{{0}}}};
unsigned char hash[20], gs_copy_hash[20];
for (unsigned i = 0; i < MESA_SHADER_STAGES; ++i) {
}
}
- radv_hash_shaders(hash, pStages, pipeline->layout, keys, get_hash_flags(device));
+ radv_hash_shaders(hash, pStages, pipeline->layout, &key, get_hash_flags(device));
memcpy(gs_copy_hash, hash, 20);
gs_copy_hash[0] ^= 1;
modules[MESA_SHADER_FRAGMENT] = &fs_m;
}
+ /* Determine first and last stage. */
+ unsigned first = MESA_SHADER_STAGES;
+ unsigned last = 0;
+ for (unsigned i = 0; i < MESA_SHADER_STAGES; i++) {
+ if (!pStages[i])
+ continue;
+ if (first == MESA_SHADER_STAGES)
+ first = i;
+ last = i;
+ }
+
for (unsigned i = 0; i < MESA_SHADER_STAGES; ++i) {
const VkPipelineShaderStageCreateInfo *stage = pStages[i];
stage ? stage->pName : "main", i,
stage ? stage->pSpecializationInfo : NULL);
pipeline->active_stages |= mesa_to_vk_shader_stage(i);
+
/* We don't want to alter meta shaders IR directly so clone it
* first.
*/
nir[i] = nir_shader_clone(NULL, nir[i]);
}
+ if (first != last) {
+ nir_variable_mode mask = 0;
+
+ if (i != first)
+ mask = mask | nir_var_shader_in;
+
+ if (i != last)
+ mask = mask | nir_var_shader_out;
+
+ nir_lower_io_to_scalar_early(nir[i], mask);
+ radv_optimize_nir(nir[i]);
+ }
}
if (nir[MESA_SHADER_TESS_CTRL]) {
- /* TODO: This is no longer used as a key we should refactor this */
- if (keys)
- keys[MESA_SHADER_TESS_CTRL].tcs.primitive_mode = nir[MESA_SHADER_TESS_EVAL]->info.tess.primitive_mode;
-
- keys[MESA_SHADER_TESS_CTRL].tcs.tes_reads_tess_factors = !!(nir[MESA_SHADER_TESS_EVAL]->info.inputs_read & (VARYING_BIT_TESS_LEVEL_INNER | VARYING_BIT_TESS_LEVEL_OUTER));
nir_lower_tes_patch_vertices(nir[MESA_SHADER_TESS_EVAL], nir[MESA_SHADER_TESS_CTRL]->info.tess.tcs_vertices_out);
}
radv_link_shaders(pipeline, nir);
+ for (int i = 0; i < MESA_SHADER_STAGES; ++i) {
+ if (!(device->instance->debug_flags & RADV_DEBUG_DUMP_SHADERS))
+ continue;
+
+ if (modules[i])
+ nir_print_shader(nir[i], stderr);
+ }
+
+ radv_fill_shader_keys(keys, &key, nir);
+
if (nir[MESA_SHADER_FRAGMENT]) {
if (!pipeline->shaders[MESA_SHADER_FRAGMENT]) {
pipeline->shaders[MESA_SHADER_FRAGMENT] =
radv_shader_variant_create(device, modules[MESA_SHADER_FRAGMENT], &nir[MESA_SHADER_FRAGMENT], 1,
- pipeline->layout, keys ? keys + MESA_SHADER_FRAGMENT : 0,
+ pipeline->layout, keys + MESA_SHADER_FRAGMENT,
&codes[MESA_SHADER_FRAGMENT], &code_sizes[MESA_SHADER_FRAGMENT]);
}
/* TODO: These are no longer used as keys we should refactor this */
- if (keys) {
- keys[MESA_SHADER_VERTEX].vs.export_prim_id =
- pipeline->shaders[MESA_SHADER_FRAGMENT]->info.fs.prim_id_input;
- keys[MESA_SHADER_TESS_EVAL].tes.export_prim_id =
- pipeline->shaders[MESA_SHADER_FRAGMENT]->info.fs.prim_id_input;
- }
+ keys[MESA_SHADER_VERTEX].vs.export_prim_id =
+ pipeline->shaders[MESA_SHADER_FRAGMENT]->info.fs.prim_id_input;
+ keys[MESA_SHADER_TESS_EVAL].tes.export_prim_id =
+ pipeline->shaders[MESA_SHADER_FRAGMENT]->info.fs.prim_id_input;
}
if (device->physical_device->rad_info.chip_class >= GFX9 && modules[MESA_SHADER_TESS_CTRL]) {
if(modules[i] && !pipeline->shaders[i]) {
pipeline->shaders[i] = radv_shader_variant_create(device, modules[i], &nir[i], 1,
pipeline->layout,
- keys ? keys + i : 0, &codes[i],
+ keys + i, &codes[i],
&code_sizes[i]);
}
}
for (int i = 0; i < MESA_SHADER_STAGES; ++i) {
free(codes[i]);
- if (modules[i] && !pipeline->device->trace_bo)
+ if (modules[i] && !pipeline->device->keep_shader_info)
ralloc_free(nir[i]);
}
ralloc_free(fs_m.nir);
}
+static uint32_t
+radv_pipeline_stage_to_user_data_0(struct radv_pipeline *pipeline,
+ gl_shader_stage stage, enum chip_class chip_class)
+{
+ bool has_gs = radv_pipeline_has_gs(pipeline);
+ bool has_tess = radv_pipeline_has_tess(pipeline);
+ switch (stage) {
+ case MESA_SHADER_FRAGMENT:
+ return R_00B030_SPI_SHADER_USER_DATA_PS_0;
+ case MESA_SHADER_VERTEX:
+ if (chip_class >= GFX9) {
+ return has_tess ? R_00B430_SPI_SHADER_USER_DATA_LS_0 :
+ has_gs ? R_00B330_SPI_SHADER_USER_DATA_ES_0 :
+ R_00B130_SPI_SHADER_USER_DATA_VS_0;
+ }
+ if (has_tess)
+ return R_00B530_SPI_SHADER_USER_DATA_LS_0;
+ else
+ return has_gs ? R_00B330_SPI_SHADER_USER_DATA_ES_0 : R_00B130_SPI_SHADER_USER_DATA_VS_0;
+ case MESA_SHADER_GEOMETRY:
+ return chip_class >= GFX9 ? R_00B330_SPI_SHADER_USER_DATA_ES_0 :
+ R_00B230_SPI_SHADER_USER_DATA_GS_0;
+ case MESA_SHADER_COMPUTE:
+ return R_00B900_COMPUTE_USER_DATA_0;
+ case MESA_SHADER_TESS_CTRL:
+ return chip_class >= GFX9 ? R_00B430_SPI_SHADER_USER_DATA_LS_0 :
+ R_00B430_SPI_SHADER_USER_DATA_HS_0;
+ case MESA_SHADER_TESS_EVAL:
+ if (chip_class >= GFX9) {
+ return has_gs ? R_00B330_SPI_SHADER_USER_DATA_ES_0 :
+ R_00B130_SPI_SHADER_USER_DATA_VS_0;
+ }
+ if (has_gs)
+ return R_00B330_SPI_SHADER_USER_DATA_ES_0;
+ else
+ return R_00B130_SPI_SHADER_USER_DATA_VS_0;
+ default:
+ unreachable("unknown shader");
+ }
+}
+
+
static VkResult
radv_pipeline_init(struct radv_pipeline *pipeline,
struct radv_device *device,
pStages[stage] = &pCreateInfo->pStages[i];
}
- struct ac_shader_variant_key keys[MESA_SHADER_STAGES];
- memset(keys, 0, sizeof(keys));
-
- if (pStages[MESA_SHADER_VERTEX]) {
- bool as_es = false;
- bool as_ls = false;
- if (pStages[MESA_SHADER_TESS_CTRL])
- as_ls = true;
- else if (pStages[MESA_SHADER_GEOMETRY])
- as_es = true;
-
- keys[MESA_SHADER_VERTEX] = radv_compute_vs_key(pCreateInfo, as_es, as_ls);
- keys[MESA_SHADER_VERTEX].has_multiview_view_index = has_view_index;
- }
-
- if (pStages[MESA_SHADER_TESS_EVAL]) {
- keys[MESA_SHADER_TESS_EVAL].has_multiview_view_index = has_view_index;
- if (pStages[MESA_SHADER_GEOMETRY])
- keys[MESA_SHADER_TESS_EVAL].tes.as_es = true;
- }
-
- if (pCreateInfo->pTessellationState)
- keys[MESA_SHADER_TESS_CTRL].tcs.input_vertices = pCreateInfo->pTessellationState->patchControlPoints;
-
- if (pStages[MESA_SHADER_GEOMETRY]) {
- keys[MESA_SHADER_GEOMETRY] = radv_compute_vs_key(pCreateInfo, false, false);
- keys[MESA_SHADER_GEOMETRY].has_multiview_view_index = has_view_index;
- }
-
- if (pCreateInfo->pMultisampleState &&
- pCreateInfo->pMultisampleState->rasterizationSamples > 1)
- keys[MESA_SHADER_FRAGMENT].fs.multisample = true;
-
- keys[MESA_SHADER_FRAGMENT].fs.col_format = pipeline->graphics.blend.spi_shader_col_format;
- if (pipeline->device->physical_device->rad_info.chip_class < VI)
- radv_pipeline_compute_get_int_clamp(pCreateInfo, &keys[MESA_SHADER_FRAGMENT].fs.is_int8, &keys[MESA_SHADER_FRAGMENT].fs.is_int10);
-
- radv_create_shaders(pipeline, device, cache, keys, pStages);
+ radv_create_shaders(pipeline, device, cache,
+ radv_generate_graphics_pipeline_key(pipeline, pCreateInfo, has_view_index),
+ pStages);
radv_pipeline_init_depth_stencil_state(pipeline, pCreateInfo, extra);
radv_pipeline_init_raster_state(pipeline, pCreateInfo);
V_028710_SPI_SHADER_ZERO;
calculate_vgt_gs_mode(pipeline);
- calculate_pa_cl_vs_out_cntl(pipeline);
+ calculate_vs_outinfo(pipeline);
calculate_ps_inputs(pipeline);
for (unsigned i = 0; i < MESA_SHADER_STAGES; i++) {
pipeline->binding_stride[desc->binding] = desc->stride;
}
+ for (uint32_t i = 0; i < MESA_SHADER_STAGES; i++)
+ pipeline->user_data_0[i] = radv_pipeline_stage_to_user_data_0(pipeline, i, device->physical_device->rad_info.chip_class);
+
struct ac_userdata_info *loc = radv_lookup_user_sgpr(pipeline, MESA_SHADER_VERTEX,
AC_UD_VS_BASE_VERTEX_START_INSTANCE);
if (loc->sgpr_idx != -1) {
- pipeline->graphics.vtx_base_sgpr = radv_shader_stage_to_user_data_0(MESA_SHADER_VERTEX, device->physical_device->rad_info.chip_class, radv_pipeline_has_gs(pipeline), radv_pipeline_has_tess(pipeline));
+ pipeline->graphics.vtx_base_sgpr = pipeline->user_data_0[MESA_SHADER_VERTEX];
pipeline->graphics.vtx_base_sgpr += loc->sgpr_idx * 4;
if (radv_get_vertex_shader(pipeline)->info.info.vs.needs_draw_id)
pipeline->graphics.vtx_emit_num = 3;
struct radv_pipeline *pipeline;
VkResult result;
- pipeline = vk_alloc2(&device->alloc, pAllocator, sizeof(*pipeline), 8,
- VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
+ pipeline = vk_zalloc2(&device->alloc, pAllocator, sizeof(*pipeline), 8,
+ VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
if (pipeline == NULL)
return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
- memset(pipeline, 0, sizeof(*pipeline));
result = radv_pipeline_init(pipeline, device, cache,
pCreateInfo, extra, pAllocator);
if (result != VK_SUCCESS) {
struct radv_pipeline *pipeline;
VkResult result;
- pipeline = vk_alloc2(&device->alloc, pAllocator, sizeof(*pipeline), 8,
- VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
+ pipeline = vk_zalloc2(&device->alloc, pAllocator, sizeof(*pipeline), 8,
+ VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
if (pipeline == NULL)
return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
- memset(pipeline, 0, sizeof(*pipeline));
pipeline->device = device;
pipeline->layout = radv_pipeline_layout_from_handle(pCreateInfo->layout);
pStages[MESA_SHADER_COMPUTE] = &pCreateInfo->stage;
- radv_create_shaders(pipeline, device, cache, NULL, pStages);
-
+ radv_create_shaders(pipeline, device, cache, (struct radv_pipeline_key) {0}, pStages);
+ pipeline->user_data_0[MESA_SHADER_COMPUTE] = radv_pipeline_stage_to_user_data_0(pipeline, MESA_SHADER_COMPUTE, device->physical_device->rad_info.chip_class);
pipeline->need_indirect_descriptor_sets |= pipeline->shaders[MESA_SHADER_COMPUTE]->info.need_indirect_descriptor_sets;
result = radv_pipeline_scratch_init(device, pipeline);
if (result != VK_SUCCESS) {
projects / mesa.git / blobdiff