X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Famd%2Fvulkan%2Fradv_descriptor_set.c;h=3d56f8c2176fbed6434e5c268aba2169526c6ddb;hb=b0460bbf1c7d09122c5e23546a23a8550375e03a;hp=ec7fd3d8cc80c48651a0558383689beadd4872c0;hpb=9da104593386f6e8ddec8f0d9d288aceb8908fe1;p=mesa.git diff --git a/src/amd/vulkan/radv_descriptor_set.c b/src/amd/vulkan/radv_descriptor_set.c index ec7fd3d8cc8..3d56f8c2176 100644 --- a/src/amd/vulkan/radv_descriptor_set.c +++ b/src/amd/vulkan/radv_descriptor_set.c @@ -31,6 +31,20 @@ #include "radv_private.h" #include "sid.h" + +static bool has_equal_immutable_samplers(const VkSampler *samplers, uint32_t count) +{ + if (!samplers) + return false; + for(uint32_t i = 1; i < count; ++i) { + if (memcmp(radv_sampler_from_handle(samplers[0])->state, + radv_sampler_from_handle(samplers[i])->state, 16)) { + return false; + } + } + return true; +} + VkResult radv_CreateDescriptorSetLayout( VkDevice _device, const VkDescriptorSetLayoutCreateInfo* pCreateInfo, @@ -66,6 +80,8 @@ VkResult radv_CreateDescriptorSetLayout( set_layout->binding_count = max_binding + 1; set_layout->shader_stages = 0; + set_layout->dynamic_shader_stages = 0; + set_layout->has_immutable_samplers = false; set_layout->size = 0; memset(set_layout->binding, 0, size - sizeof(struct radv_descriptor_set_layout)); @@ -121,7 +137,6 @@ VkResult radv_CreateDescriptorSetLayout( } set_layout->size = align(set_layout->size, alignment); - assert(binding->descriptorCount > 0); set_layout->binding[b].type = binding->descriptorType; set_layout->binding[b].array_size = binding->descriptorCount; set_layout->binding[b].offset = set_layout->size; @@ -130,14 +145,13 @@ VkResult radv_CreateDescriptorSetLayout( if (binding->pImmutableSamplers) { set_layout->binding[b].immutable_samplers_offset = samplers_offset; - set_layout->binding[b].immutable_samplers_equal = true; + set_layout->binding[b].immutable_samplers_equal = + has_equal_immutable_samplers(binding->pImmutableSamplers, binding->descriptorCount); + set_layout->has_immutable_samplers = true; for (uint32_t i = 0; i < binding->descriptorCount; i++) memcpy(samplers + 4 * i, &radv_sampler_from_handle(binding->pImmutableSamplers[i])->state, 16); - for (uint32_t i = 1; i < binding->descriptorCount; i++) - if (memcmp(samplers + 4 * i, samplers, 16) != 0) - set_layout->binding[b].immutable_samplers_equal = false; /* Don't reserve space for the samplers if they're not accessed. */ if (set_layout->binding[b].immutable_samplers_equal) { @@ -179,6 +193,69 @@ void radv_DestroyDescriptorSetLayout( vk_free2(&device->alloc, pAllocator, set_layout); } +void radv_GetDescriptorSetLayoutSupport(VkDevice device, + const VkDescriptorSetLayoutCreateInfo* pCreateInfo, + VkDescriptorSetLayoutSupport* pSupport) +{ + bool supported = true; + uint64_t size = 0; + for (uint32_t i = 0; i < pCreateInfo->bindingCount; i++) { + const VkDescriptorSetLayoutBinding *binding = &pCreateInfo->pBindings[i]; + + if (binding->descriptorCount == 0) + continue; + + uint64_t descriptor_size = 0; + uint64_t descriptor_alignment = 1; + switch (binding->descriptorType) { + case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC: + case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC: + break; + case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER: + case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER: + case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER: + case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER: + descriptor_size = 16; + descriptor_alignment = 16; + break; + case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE: + case VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE: + case VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT: + descriptor_size = 64; + descriptor_alignment = 32; + break; + case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER: + if (!has_equal_immutable_samplers(binding->pImmutableSamplers, binding->descriptorCount)) { + descriptor_size = 64; + } else { + descriptor_size = 96; + } + descriptor_alignment = 32; + break; + case VK_DESCRIPTOR_TYPE_SAMPLER: + if (!has_equal_immutable_samplers(binding->pImmutableSamplers, binding->descriptorCount)) { + descriptor_size = 16; + descriptor_alignment = 16; + } + break; + default: + unreachable("unknown descriptor type\n"); + break; + } + + if (size && !align_u64(size, descriptor_alignment)) { + supported = false; + } + size = align_u64(size, descriptor_alignment); + if (descriptor_size && (UINT64_MAX - size) / descriptor_size < binding->descriptorCount) { + supported = false; + } + size += binding->descriptorCount * descriptor_size; + } + + pSupport->supported = supported; +} + /* * Pipeline layouts. These have nothing to do with the pipeline. They are * just muttiple descriptor set layouts pasted together @@ -225,6 +302,7 @@ VkResult radv_CreatePipelineLayout( layout->dynamic_offset_count = dynamic_offset_count; layout->push_constant_size = 0; + for (unsigned i = 0; i < pCreateInfo->pushConstantRangeCount; ++i) { const VkPushConstantRange *range = pCreateInfo->pPushConstantRanges + i; layout->push_constant_size = MAX2(layout->push_constant_size, @@ -292,58 +370,69 @@ radv_descriptor_set_create(struct radv_device *device, uint32_t layout_size = align_u32(layout->size, 32); set->size = layout->size; + if (!pool->host_memory_base && pool->entry_count == pool->max_entry_count) { + vk_free2(&device->alloc, NULL, set); + return vk_error(VK_ERROR_OUT_OF_POOL_MEMORY_KHR); + } + /* try to allocate linearly first, so that we don't spend * time looking for gaps if the app only allocates & * resets via the pool. */ if (pool->current_offset + layout_size <= pool->size) { set->bo = pool->bo; set->mapped_ptr = (uint32_t*)(pool->mapped_ptr + pool->current_offset); - set->va = device->ws->buffer_get_va(set->bo) + pool->current_offset; + set->va = radv_buffer_get_va(set->bo) + pool->current_offset; + if (!pool->host_memory_base) { + pool->entries[pool->entry_count].offset = pool->current_offset; + pool->entries[pool->entry_count].size = layout_size; + pool->entries[pool->entry_count].set = set; + pool->entry_count++; + } pool->current_offset += layout_size; - list_addtail(&set->vram_list, &pool->vram_list); } else if (!pool->host_memory_base) { uint64_t offset = 0; - struct list_head *prev = &pool->vram_list; - struct radv_descriptor_set *cur; + int index; - assert(!pool->host_memory_base); - LIST_FOR_EACH_ENTRY(cur, &pool->vram_list, vram_list) { - uint64_t start = (uint8_t*)cur->mapped_ptr - pool->mapped_ptr; - if (start - offset >= layout_size) + for (index = 0; index < pool->entry_count; ++index) { + if (pool->entries[index].offset - offset >= layout_size) break; - - offset = start + cur->size; - prev = &cur->vram_list; + offset = pool->entries[index].offset + pool->entries[index].size; } if (pool->size - offset < layout_size) { - vk_free2(&device->alloc, NULL, set->dynamic_descriptors); vk_free2(&device->alloc, NULL, set); return vk_error(VK_ERROR_OUT_OF_POOL_MEMORY_KHR); } set->bo = pool->bo; set->mapped_ptr = (uint32_t*)(pool->mapped_ptr + offset); - set->va = device->ws->buffer_get_va(set->bo) + offset; - list_add(&set->vram_list, prev); + set->va = radv_buffer_get_va(set->bo) + offset; + memmove(&pool->entries[index + 1], &pool->entries[index], + sizeof(pool->entries[0]) * (pool->entry_count - index)); + pool->entries[index].offset = offset; + pool->entries[index].size = layout_size; + pool->entries[index].set = set; + pool->entry_count++; } else return vk_error(VK_ERROR_OUT_OF_POOL_MEMORY_KHR); } - for (unsigned i = 0; i < layout->binding_count; ++i) { - if (!layout->binding[i].immutable_samplers_offset || - layout->binding[i].immutable_samplers_equal) - continue; + if (layout->has_immutable_samplers) { + for (unsigned i = 0; i < layout->binding_count; ++i) { + if (!layout->binding[i].immutable_samplers_offset || + layout->binding[i].immutable_samplers_equal) + continue; - unsigned offset = layout->binding[i].offset / 4; - if (layout->binding[i].type == VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER) - offset += 16; + unsigned offset = layout->binding[i].offset / 4; + if (layout->binding[i].type == VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER) + offset += 16; - const uint32_t *samplers = (const uint32_t*)((const char*)layout + layout->binding[i].immutable_samplers_offset); - for (unsigned j = 0; j < layout->binding[i].array_size; ++j) { - memcpy(set->mapped_ptr + offset, samplers + 4 * j, 16); - offset += layout->binding[i].size / 4; - } + const uint32_t *samplers = (const uint32_t*)((const char*)layout + layout->binding[i].immutable_samplers_offset); + for (unsigned j = 0; j < layout->binding[i].array_size; ++j) { + memcpy(set->mapped_ptr + offset, samplers + 4 * j, 16); + offset += layout->binding[i].size / 4; + } + } } *out_set = set; return VK_SUCCESS; @@ -357,8 +446,17 @@ radv_descriptor_set_destroy(struct radv_device *device, { assert(!pool->host_memory_base); - if (free_bo && set->size) - list_del(&set->vram_list); + if (free_bo && set->size && !pool->host_memory_base) { + uint32_t offset = (uint8_t*)set->mapped_ptr - pool->mapped_ptr; + for (int i = 0; i < pool->entry_count; ++i) { + if (pool->entries[i].offset == offset) { + memmove(&pool->entries[i], &pool->entries[i+1], + sizeof(pool->entries[i]) * (pool->entry_count - i - 1)); + --pool->entry_count; + break; + } + } + } vk_free2(&device->alloc, NULL, set); } @@ -410,6 +508,8 @@ VkResult radv_CreateDescriptorPool( host_size += sizeof(struct radeon_winsys_bo*) * bo_count; host_size += sizeof(struct radv_descriptor_range) * range_count; size += host_size; + } else { + size += sizeof(struct radv_descriptor_pool_entry) * pCreateInfo->maxSets; } pool = vk_alloc2(&device->alloc, pAllocator, size, 8, @@ -426,13 +526,15 @@ VkResult radv_CreateDescriptorPool( } if (bo_size) { - pool->bo = device->ws->buffer_create(device->ws, bo_size, - 32, RADEON_DOMAIN_VRAM, 0); + pool->bo = device->ws->buffer_create(device->ws, bo_size, 32, + RADEON_DOMAIN_VRAM, + RADEON_FLAG_NO_INTERPROCESS_SHARING | + RADEON_FLAG_READ_ONLY); pool->mapped_ptr = (uint8_t*)device->ws->buffer_map(pool->bo); } pool->size = bo_size; + pool->max_entry_count = pCreateInfo->maxSets; - list_inithead(&pool->vram_list); *pDescriptorPool = radv_descriptor_pool_to_handle(pool); return VK_SUCCESS; } @@ -449,9 +551,8 @@ void radv_DestroyDescriptorPool( return; if (!pool->host_memory_base) { - list_for_each_entry_safe(struct radv_descriptor_set, set, - &pool->vram_list, vram_list) { - radv_descriptor_set_destroy(device, pool, set, false); + for(int i = 0; i < pool->entry_count; ++i) { + radv_descriptor_set_destroy(device, pool, pool->entries[i].set, false); } } @@ -469,14 +570,12 @@ VkResult radv_ResetDescriptorPool( RADV_FROM_HANDLE(radv_descriptor_pool, pool, descriptorPool); if (!pool->host_memory_base) { - list_for_each_entry_safe(struct radv_descriptor_set, set, - &pool->vram_list, vram_list) { - radv_descriptor_set_destroy(device, pool, set, false); + for(int i = 0; i < pool->entry_count; ++i) { + radv_descriptor_set_destroy(device, pool, pool->entries[i].set, false); } + pool->entry_count = 0; } - list_inithead(&pool->vram_list); - pool->current_offset = 0; pool->host_memory_ptr = pool->host_memory_base; @@ -493,7 +592,7 @@ VkResult radv_AllocateDescriptorSets( VkResult result = VK_SUCCESS; uint32_t i; - struct radv_descriptor_set *set; + struct radv_descriptor_set *set = NULL; /* allocate a set of buffers for each shader to contain descriptors */ for (i = 0; i < pAllocateInfo->descriptorSetCount; i++) { @@ -544,7 +643,7 @@ static void write_texel_buffer_descriptor(struct radv_device *device, memcpy(dst, buffer_view->state, 4 * 4); if (cmd_buffer) - device->ws->cs_add_buffer(cmd_buffer->cs, buffer_view->bo, 7); + radv_cs_add_buffer(device->ws, cmd_buffer->cs, buffer_view->bo, 7); else *buffer_list = buffer_view->bo; } @@ -556,7 +655,7 @@ static void write_buffer_descriptor(struct radv_device *device, const VkDescriptorBufferInfo *buffer_info) { RADV_FROM_HANDLE(radv_buffer, buffer, buffer_info->buffer); - uint64_t va = device->ws->buffer_get_va(buffer->bo); + uint64_t va = radv_buffer_get_va(buffer->bo); uint32_t range = buffer_info->range; if (buffer_info->range == VK_WHOLE_SIZE) @@ -574,7 +673,7 @@ static void write_buffer_descriptor(struct radv_device *device, S_008F0C_DATA_FORMAT(V_008F0C_BUF_DATA_FORMAT_32); if (cmd_buffer) - device->ws->cs_add_buffer(cmd_buffer->cs, buffer->bo, 7); + radv_cs_add_buffer(device->ws, cmd_buffer->cs, buffer->bo, 7); else *buffer_list = buffer->bo; } @@ -585,7 +684,7 @@ static void write_dynamic_buffer_descriptor(struct radv_device *device, const VkDescriptorBufferInfo *buffer_info) { RADV_FROM_HANDLE(radv_buffer, buffer, buffer_info->buffer); - uint64_t va = device->ws->buffer_get_va(buffer->bo); + uint64_t va = radv_buffer_get_va(buffer->bo); unsigned size = buffer_info->range; if (buffer_info->range == VK_WHOLE_SIZE) @@ -603,14 +702,22 @@ write_image_descriptor(struct radv_device *device, struct radv_cmd_buffer *cmd_buffer, unsigned *dst, struct radeon_winsys_bo **buffer_list, + VkDescriptorType descriptor_type, const VkDescriptorImageInfo *image_info) { RADV_FROM_HANDLE(radv_image_view, iview, image_info->imageView); - memcpy(dst, iview->descriptor, 8 * 4); - memcpy(dst + 8, iview->fmask_descriptor, 8 * 4); + uint32_t *descriptor; + + if (descriptor_type == VK_DESCRIPTOR_TYPE_STORAGE_IMAGE) { + descriptor = iview->storage_descriptor; + } else { + descriptor = iview->descriptor; + } + + memcpy(dst, descriptor, 16 * 4); if (cmd_buffer) - device->ws->cs_add_buffer(cmd_buffer->cs, iview->bo, 7); + radv_cs_add_buffer(device->ws, cmd_buffer->cs, iview->bo, 7); else *buffer_list = iview->bo; } @@ -620,12 +727,13 @@ write_combined_image_sampler_descriptor(struct radv_device *device, struct radv_cmd_buffer *cmd_buffer, unsigned *dst, struct radeon_winsys_bo **buffer_list, + VkDescriptorType descriptor_type, const VkDescriptorImageInfo *image_info, bool has_sampler) { RADV_FROM_HANDLE(radv_sampler, sampler, image_info->sampler); - write_image_descriptor(device, cmd_buffer, dst, buffer_list, image_info); + write_image_descriptor(device, cmd_buffer, dst, buffer_list, descriptor_type, image_info); /* copy over sampler state */ if (has_sampler) memcpy(dst + 16, sampler->state, 16); @@ -696,10 +804,12 @@ void radv_update_descriptor_sets( case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE: case VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT: write_image_descriptor(device, cmd_buffer, ptr, buffer_list, + writeset->descriptorType, writeset->pImageInfo + j); break; case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER: write_combined_image_sampler_descriptor(device, cmd_buffer, ptr, buffer_list, + writeset->descriptorType, writeset->pImageInfo + j, !binding_layout->immutable_samplers_offset); if (copy_immutable_samplers) { @@ -725,8 +835,59 @@ void radv_update_descriptor_sets( } } - if (descriptorCopyCount) - radv_finishme("copy descriptors"); + + for (i = 0; i < descriptorCopyCount; i++) { + const VkCopyDescriptorSet *copyset = &pDescriptorCopies[i]; + RADV_FROM_HANDLE(radv_descriptor_set, src_set, + copyset->srcSet); + RADV_FROM_HANDLE(radv_descriptor_set, dst_set, + copyset->dstSet); + const struct radv_descriptor_set_binding_layout *src_binding_layout = + src_set->layout->binding + copyset->srcBinding; + const struct radv_descriptor_set_binding_layout *dst_binding_layout = + dst_set->layout->binding + copyset->dstBinding; + uint32_t *src_ptr = src_set->mapped_ptr; + uint32_t *dst_ptr = dst_set->mapped_ptr; + struct radeon_winsys_bo **src_buffer_list = src_set->descriptors; + struct radeon_winsys_bo **dst_buffer_list = dst_set->descriptors; + + src_ptr += src_binding_layout->offset / 4; + dst_ptr += dst_binding_layout->offset / 4; + + src_ptr += src_binding_layout->size * copyset->srcArrayElement / 4; + dst_ptr += dst_binding_layout->size * copyset->dstArrayElement / 4; + + src_buffer_list += src_binding_layout->buffer_offset; + src_buffer_list += copyset->srcArrayElement; + + dst_buffer_list += dst_binding_layout->buffer_offset; + dst_buffer_list += copyset->dstArrayElement; + + for (j = 0; j < copyset->descriptorCount; ++j) { + switch (src_binding_layout->type) { + case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC: + case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC: { + unsigned src_idx = copyset->srcArrayElement + j; + unsigned dst_idx = copyset->dstArrayElement + j; + struct radv_descriptor_range *src_range, *dst_range; + src_idx += src_binding_layout->dynamic_offset_offset; + dst_idx += dst_binding_layout->dynamic_offset_offset; + + src_range = src_set->dynamic_descriptors + src_idx; + dst_range = dst_set->dynamic_descriptors + dst_idx; + *dst_range = *src_range; + break; + } + default: + memcpy(dst_ptr, src_ptr, src_binding_layout->size); + } + src_ptr += src_binding_layout->size / 4; + dst_ptr += dst_binding_layout->size / 4; + dst_buffer_list[j] = src_buffer_list[j]; + ++src_buffer_list; + ++dst_buffer_list; + } + } } void radv_UpdateDescriptorSets( @@ -742,10 +903,10 @@ void radv_UpdateDescriptorSets( descriptorCopyCount, pDescriptorCopies); } -VkResult radv_CreateDescriptorUpdateTemplateKHR(VkDevice _device, - const VkDescriptorUpdateTemplateCreateInfoKHR *pCreateInfo, - const VkAllocationCallbacks *pAllocator, - VkDescriptorUpdateTemplateKHR *pDescriptorUpdateTemplate) +VkResult radv_CreateDescriptorUpdateTemplate(VkDevice _device, + const VkDescriptorUpdateTemplateCreateInfoKHR *pCreateInfo, + const VkAllocationCallbacks *pAllocator, + VkDescriptorUpdateTemplateKHR *pDescriptorUpdateTemplate) { RADV_FROM_HANDLE(radv_device, device, _device); RADV_FROM_HANDLE(radv_descriptor_set_layout, set_layout, pCreateInfo->descriptorSetLayout); @@ -760,6 +921,7 @@ VkResult radv_CreateDescriptorUpdateTemplateKHR(VkDevice _device, return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY); templ->entry_count = entry_count; + templ->bind_point = pCreateInfo->pipelineBindPoint; for (i = 0; i < entry_count; i++) { const VkDescriptorUpdateTemplateEntryKHR *entry = &pCreateInfo->pDescriptorUpdateEntries[i]; @@ -814,9 +976,9 @@ VkResult radv_CreateDescriptorUpdateTemplateKHR(VkDevice _device, return VK_SUCCESS; } -void radv_DestroyDescriptorUpdateTemplateKHR(VkDevice _device, - VkDescriptorUpdateTemplateKHR descriptorUpdateTemplate, - const VkAllocationCallbacks *pAllocator) +void radv_DestroyDescriptorUpdateTemplate(VkDevice _device, + VkDescriptorUpdateTemplateKHR descriptorUpdateTemplate, + const VkAllocationCallbacks *pAllocator) { RADV_FROM_HANDLE(radv_device, device, _device); RADV_FROM_HANDLE(radv_descriptor_update_template, templ, descriptorUpdateTemplate); @@ -866,10 +1028,12 @@ void radv_update_descriptor_set_with_template(struct radv_device *device, case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE: case VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT: write_image_descriptor(device, cmd_buffer, pDst, buffer_list, + templ->entry[i].descriptor_type, (struct VkDescriptorImageInfo *) pSrc); break; case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER: write_combined_image_sampler_descriptor(device, cmd_buffer, pDst, buffer_list, + templ->entry[i].descriptor_type, (struct VkDescriptorImageInfo *) pSrc, templ->entry[i].has_sampler); if (templ->entry[i].immutable_samplers) @@ -893,13 +1057,31 @@ void radv_update_descriptor_set_with_template(struct radv_device *device, } } -void radv_UpdateDescriptorSetWithTemplateKHR(VkDevice _device, - VkDescriptorSet descriptorSet, - VkDescriptorUpdateTemplateKHR descriptorUpdateTemplate, - const void *pData) +void radv_UpdateDescriptorSetWithTemplate(VkDevice _device, + VkDescriptorSet descriptorSet, + VkDescriptorUpdateTemplateKHR descriptorUpdateTemplate, + const void *pData) { RADV_FROM_HANDLE(radv_device, device, _device); RADV_FROM_HANDLE(radv_descriptor_set, set, descriptorSet); radv_update_descriptor_set_with_template(device, NULL, set, descriptorUpdateTemplate, pData); } + + +VkResult radv_CreateSamplerYcbcrConversion(VkDevice device, + const VkSamplerYcbcrConversionCreateInfo* pCreateInfo, + const VkAllocationCallbacks* pAllocator, + VkSamplerYcbcrConversion* pYcbcrConversion) +{ + *pYcbcrConversion = VK_NULL_HANDLE; + return VK_SUCCESS; +} + + +void radv_DestroySamplerYcbcrConversion(VkDevice device, + VkSamplerYcbcrConversion ycbcrConversion, + const VkAllocationCallbacks* pAllocator) +{ + /* Do nothing. */ +}