summaryrefslogtreecommitdiffstats
path: root/src/video_core/renderer_vulkan
diff options
context:
space:
mode:
Diffstat (limited to 'src/video_core/renderer_vulkan')
-rw-r--r--src/video_core/renderer_vulkan/maxwell_to_vk.cpp7
-rw-r--r--src/video_core/renderer_vulkan/vk_blit_screen.cpp1
-rw-r--r--src/video_core/renderer_vulkan/vk_buffer_cache.cpp3
-rw-r--r--src/video_core/renderer_vulkan/vk_compute_pass.cpp320
-rw-r--r--src/video_core/renderer_vulkan/vk_compute_pass.h50
-rw-r--r--src/video_core/renderer_vulkan/vk_fence_manager.h2
-rw-r--r--src/video_core/renderer_vulkan/vk_query_cache.cpp1596
-rw-r--r--src/video_core/renderer_vulkan/vk_query_cache.h106
-rw-r--r--src/video_core/renderer_vulkan/vk_rasterizer.cpp135
-rw-r--r--src/video_core/renderer_vulkan/vk_rasterizer.h14
-rw-r--r--src/video_core/renderer_vulkan/vk_scheduler.cpp9
-rw-r--r--src/video_core/renderer_vulkan/vk_scheduler.h12
-rw-r--r--src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp33
-rw-r--r--src/video_core/renderer_vulkan/vk_staging_buffer_pool.h5
-rw-r--r--src/video_core/renderer_vulkan/vk_texture_cache.cpp128
-rw-r--r--src/video_core/renderer_vulkan/vk_texture_cache.h7
16 files changed, 2119 insertions, 309 deletions
diff --git a/src/video_core/renderer_vulkan/maxwell_to_vk.cpp b/src/video_core/renderer_vulkan/maxwell_to_vk.cpp
index 35bf80ea3..a08f2f67f 100644
--- a/src/video_core/renderer_vulkan/maxwell_to_vk.cpp
+++ b/src/video_core/renderer_vulkan/maxwell_to_vk.cpp
@@ -185,7 +185,7 @@ struct FormatTuple {
{VK_FORMAT_BC2_SRGB_BLOCK}, // BC2_SRGB
{VK_FORMAT_BC3_SRGB_BLOCK}, // BC3_SRGB
{VK_FORMAT_BC7_SRGB_BLOCK}, // BC7_SRGB
- {VK_FORMAT_R4G4B4A4_UNORM_PACK16}, // A4B4G4R4_UNORM
+ {VK_FORMAT_A4B4G4R4_UNORM_PACK16_EXT}, // A4B4G4R4_UNORM
{VK_FORMAT_R4G4_UNORM_PACK8}, // G4R4_UNORM
{VK_FORMAT_ASTC_4x4_SRGB_BLOCK}, // ASTC_2D_4X4_SRGB
{VK_FORMAT_ASTC_8x8_SRGB_BLOCK}, // ASTC_2D_8X8_SRGB
@@ -214,8 +214,9 @@ struct FormatTuple {
{VK_FORMAT_E5B9G9R9_UFLOAT_PACK32}, // E5B9G9R9_FLOAT
// Depth formats
- {VK_FORMAT_D32_SFLOAT, Attachable}, // D32_FLOAT
- {VK_FORMAT_D16_UNORM, Attachable}, // D16_UNORM
+ {VK_FORMAT_D32_SFLOAT, Attachable}, // D32_FLOAT
+ {VK_FORMAT_D16_UNORM, Attachable}, // D16_UNORM
+ {VK_FORMAT_X8_D24_UNORM_PACK32, Attachable}, // X8_D24_UNORM
// Stencil formats
{VK_FORMAT_S8_UINT, Attachable}, // S8_UINT
diff --git a/src/video_core/renderer_vulkan/vk_blit_screen.cpp b/src/video_core/renderer_vulkan/vk_blit_screen.cpp
index 31928bb94..52fc142d1 100644
--- a/src/video_core/renderer_vulkan/vk_blit_screen.cpp
+++ b/src/video_core/renderer_vulkan/vk_blit_screen.cpp
@@ -96,6 +96,7 @@ std::size_t GetSizeInBytes(const Tegra::FramebufferConfig& framebuffer) {
VkFormat GetFormat(const Tegra::FramebufferConfig& framebuffer) {
switch (framebuffer.pixel_format) {
case Service::android::PixelFormat::Rgba8888:
+ case Service::android::PixelFormat::Rgbx8888:
return VK_FORMAT_A8B8G8R8_UNORM_PACK32;
case Service::android::PixelFormat::Rgb565:
return VK_FORMAT_R5G6B5_UNORM_PACK16;
diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp
index e15865d16..d8148e89a 100644
--- a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp
@@ -61,6 +61,9 @@ vk::Buffer CreateBuffer(const Device& device, const MemoryAllocator& memory_allo
if (device.IsExtTransformFeedbackSupported()) {
flags |= VK_BUFFER_USAGE_TRANSFORM_FEEDBACK_BUFFER_BIT_EXT;
}
+ if (device.IsExtConditionalRendering()) {
+ flags |= VK_BUFFER_USAGE_CONDITIONAL_RENDERING_BIT_EXT;
+ }
const VkBufferCreateInfo buffer_ci = {
.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
.pNext = nullptr,
diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.cpp b/src/video_core/renderer_vulkan/vk_compute_pass.cpp
index 54ee030ce..617f92910 100644
--- a/src/video_core/renderer_vulkan/vk_compute_pass.cpp
+++ b/src/video_core/renderer_vulkan/vk_compute_pass.cpp
@@ -3,6 +3,7 @@
#include <array>
#include <memory>
+#include <numeric>
#include <optional>
#include <utility>
@@ -11,7 +12,13 @@
#include "common/assert.h"
#include "common/common_types.h"
#include "common/div_ceil.h"
+#include "common/vector_math.h"
#include "video_core/host_shaders/astc_decoder_comp_spv.h"
+#include "video_core/host_shaders/convert_msaa_to_non_msaa_comp_spv.h"
+#include "video_core/host_shaders/convert_non_msaa_to_msaa_comp_spv.h"
+#include "video_core/host_shaders/queries_prefix_scan_sum_comp_spv.h"
+#include "video_core/host_shaders/queries_prefix_scan_sum_nosubgroups_comp_spv.h"
+#include "video_core/host_shaders/resolve_conditional_render_comp_spv.h"
#include "video_core/host_shaders/vulkan_quad_indexed_comp_spv.h"
#include "video_core/host_shaders/vulkan_uint8_comp_spv.h"
#include "video_core/renderer_vulkan/vk_compute_pass.h"
@@ -57,6 +64,30 @@ constexpr std::array<VkDescriptorSetLayoutBinding, 2> INPUT_OUTPUT_DESCRIPTOR_SE
},
}};
+constexpr std::array<VkDescriptorSetLayoutBinding, 3> QUERIES_SCAN_DESCRIPTOR_SET_BINDINGS{{
+ {
+ .binding = 0,
+ .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+ .descriptorCount = 1,
+ .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
+ .pImmutableSamplers = nullptr,
+ },
+ {
+ .binding = 1,
+ .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+ .descriptorCount = 1,
+ .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
+ .pImmutableSamplers = nullptr,
+ },
+ {
+ .binding = 2,
+ .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+ .descriptorCount = 1,
+ .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
+ .pImmutableSamplers = nullptr,
+ },
+}};
+
constexpr DescriptorBankInfo INPUT_OUTPUT_BANK_INFO{
.uniform_buffers = 0,
.storage_buffers = 2,
@@ -67,6 +98,16 @@ constexpr DescriptorBankInfo INPUT_OUTPUT_BANK_INFO{
.score = 2,
};
+constexpr DescriptorBankInfo QUERIES_SCAN_BANK_INFO{
+ .uniform_buffers = 0,
+ .storage_buffers = 3,
+ .texture_buffers = 0,
+ .image_buffers = 0,
+ .textures = 0,
+ .images = 0,
+ .score = 3,
+};
+
constexpr std::array<VkDescriptorSetLayoutBinding, ASTC_NUM_BINDINGS> ASTC_DESCRIPTOR_SET_BINDINGS{{
{
.binding = ASTC_BINDING_INPUT_BUFFER,
@@ -94,6 +135,33 @@ constexpr DescriptorBankInfo ASTC_BANK_INFO{
.score = 2,
};
+constexpr std::array<VkDescriptorSetLayoutBinding, ASTC_NUM_BINDINGS> MSAA_DESCRIPTOR_SET_BINDINGS{{
+ {
+ .binding = 0,
+ .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
+ .descriptorCount = 1,
+ .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
+ .pImmutableSamplers = nullptr,
+ },
+ {
+ .binding = 1,
+ .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
+ .descriptorCount = 1,
+ .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
+ .pImmutableSamplers = nullptr,
+ },
+}};
+
+constexpr DescriptorBankInfo MSAA_BANK_INFO{
+ .uniform_buffers = 0,
+ .storage_buffers = 0,
+ .texture_buffers = 0,
+ .image_buffers = 0,
+ .textures = 0,
+ .images = 2,
+ .score = 2,
+};
+
constexpr VkDescriptorUpdateTemplateEntry INPUT_OUTPUT_DESCRIPTOR_UPDATE_TEMPLATE{
.dstBinding = 0,
.dstArrayElement = 0,
@@ -103,6 +171,24 @@ constexpr VkDescriptorUpdateTemplateEntry INPUT_OUTPUT_DESCRIPTOR_UPDATE_TEMPLAT
.stride = sizeof(DescriptorUpdateEntry),
};
+constexpr VkDescriptorUpdateTemplateEntry QUERIES_SCAN_DESCRIPTOR_UPDATE_TEMPLATE{
+ .dstBinding = 0,
+ .dstArrayElement = 0,
+ .descriptorCount = 3,
+ .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+ .offset = 0,
+ .stride = sizeof(DescriptorUpdateEntry),
+};
+
+constexpr VkDescriptorUpdateTemplateEntry MSAA_DESCRIPTOR_UPDATE_TEMPLATE{
+ .dstBinding = 0,
+ .dstArrayElement = 0,
+ .descriptorCount = 2,
+ .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
+ .offset = 0,
+ .stride = sizeof(DescriptorUpdateEntry),
+};
+
constexpr std::array<VkDescriptorUpdateTemplateEntry, ASTC_NUM_BINDINGS>
ASTC_PASS_DESCRIPTOR_UPDATE_TEMPLATE_ENTRY{{
{
@@ -131,13 +217,21 @@ struct AstcPushConstants {
u32 block_height;
u32 block_height_mask;
};
+
+struct QueriesPrefixScanPushConstants {
+ u32 min_accumulation_base;
+ u32 max_accumulation_base;
+ u32 accumulation_limit;
+ u32 buffer_offset;
+};
} // Anonymous namespace
ComputePass::ComputePass(const Device& device_, DescriptorPool& descriptor_pool,
vk::Span<VkDescriptorSetLayoutBinding> bindings,
vk::Span<VkDescriptorUpdateTemplateEntry> templates,
const DescriptorBankInfo& bank_info,
- vk::Span<VkPushConstantRange> push_constants, std::span<const u32> code)
+ vk::Span<VkPushConstantRange> push_constants, std::span<const u32> code,
+ std::optional<u32> optional_subgroup_size)
: device{device_} {
descriptor_set_layout = device.GetLogical().CreateDescriptorSetLayout({
.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO,
@@ -170,6 +264,9 @@ ComputePass::ComputePass(const Device& device_, DescriptorPool& descriptor_pool,
});
descriptor_allocator = descriptor_pool.Allocator(*descriptor_set_layout, bank_info);
}
+ if (code.empty()) {
+ return;
+ }
module = device.GetLogical().CreateShaderModule({
.sType = VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO,
.pNext = nullptr,
@@ -178,13 +275,19 @@ ComputePass::ComputePass(const Device& device_, DescriptorPool& descriptor_pool,
.pCode = code.data(),
});
device.SaveShader(code);
+ const VkPipelineShaderStageRequiredSubgroupSizeCreateInfoEXT subgroup_size_ci{
+ .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_REQUIRED_SUBGROUP_SIZE_CREATE_INFO_EXT,
+ .pNext = nullptr,
+ .requiredSubgroupSize = optional_subgroup_size ? *optional_subgroup_size : 32U,
+ };
+ bool use_setup_size = device.IsExtSubgroupSizeControlSupported() && optional_subgroup_size;
pipeline = device.GetLogical().CreateComputePipeline({
.sType = VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.stage{
.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
- .pNext = nullptr,
+ .pNext = use_setup_size ? &subgroup_size_ci : nullptr,
.flags = 0,
.stage = VK_SHADER_STAGE_COMPUTE_BIT,
.module = *module,
@@ -302,6 +405,123 @@ std::pair<VkBuffer, VkDeviceSize> QuadIndexedPass::Assemble(
return {staging.buffer, staging.offset};
}
+ConditionalRenderingResolvePass::ConditionalRenderingResolvePass(
+ const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_,
+ ComputePassDescriptorQueue& compute_pass_descriptor_queue_)
+ : ComputePass(device_, descriptor_pool_, INPUT_OUTPUT_DESCRIPTOR_SET_BINDINGS,
+ INPUT_OUTPUT_DESCRIPTOR_UPDATE_TEMPLATE, INPUT_OUTPUT_BANK_INFO, nullptr,
+ RESOLVE_CONDITIONAL_RENDER_COMP_SPV),
+ scheduler{scheduler_}, compute_pass_descriptor_queue{compute_pass_descriptor_queue_} {}
+
+void ConditionalRenderingResolvePass::Resolve(VkBuffer dst_buffer, VkBuffer src_buffer,
+ u32 src_offset, bool compare_to_zero) {
+ const size_t compare_size = compare_to_zero ? 8 : 24;
+
+ compute_pass_descriptor_queue.Acquire();
+ compute_pass_descriptor_queue.AddBuffer(src_buffer, src_offset, compare_size);
+ compute_pass_descriptor_queue.AddBuffer(dst_buffer, 0, sizeof(u32));
+ const void* const descriptor_data{compute_pass_descriptor_queue.UpdateData()};
+
+ scheduler.RequestOutsideRenderPassOperationContext();
+ scheduler.Record([this, descriptor_data](vk::CommandBuffer cmdbuf) {
+ static constexpr VkMemoryBarrier read_barrier{
+ .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
+ .pNext = nullptr,
+ .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT | VK_ACCESS_SHADER_WRITE_BIT,
+ .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT,
+ };
+ static constexpr VkMemoryBarrier write_barrier{
+ .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
+ .pNext = nullptr,
+ .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT,
+ .dstAccessMask = VK_ACCESS_CONDITIONAL_RENDERING_READ_BIT_EXT,
+ };
+ const VkDescriptorSet set = descriptor_allocator.Commit();
+ device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data);
+
+ cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
+ VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, read_barrier);
+ cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, *pipeline);
+ cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {});
+ cmdbuf.Dispatch(1, 1, 1);
+ cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
+ VK_PIPELINE_STAGE_CONDITIONAL_RENDERING_BIT_EXT, 0, write_barrier);
+ });
+}
+
+QueriesPrefixScanPass::QueriesPrefixScanPass(
+ const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_,
+ ComputePassDescriptorQueue& compute_pass_descriptor_queue_)
+ : ComputePass(
+ device_, descriptor_pool_, QUERIES_SCAN_DESCRIPTOR_SET_BINDINGS,
+ QUERIES_SCAN_DESCRIPTOR_UPDATE_TEMPLATE, QUERIES_SCAN_BANK_INFO,
+ COMPUTE_PUSH_CONSTANT_RANGE<sizeof(QueriesPrefixScanPushConstants)>,
+ device_.IsSubgroupFeatureSupported(VK_SUBGROUP_FEATURE_BASIC_BIT) &&
+ device_.IsSubgroupFeatureSupported(VK_SUBGROUP_FEATURE_ARITHMETIC_BIT) &&
+ device_.IsSubgroupFeatureSupported(VK_SUBGROUP_FEATURE_SHUFFLE_BIT) &&
+ device_.IsSubgroupFeatureSupported(VK_SUBGROUP_FEATURE_SHUFFLE_RELATIVE_BIT)
+ ? std::span<const u32>(QUERIES_PREFIX_SCAN_SUM_COMP_SPV)
+ : std::span<const u32>(QUERIES_PREFIX_SCAN_SUM_NOSUBGROUPS_COMP_SPV)),
+ scheduler{scheduler_}, compute_pass_descriptor_queue{compute_pass_descriptor_queue_} {}
+
+void QueriesPrefixScanPass::Run(VkBuffer accumulation_buffer, VkBuffer dst_buffer,
+ VkBuffer src_buffer, size_t number_of_sums,
+ size_t min_accumulation_limit, size_t max_accumulation_limit) {
+ size_t current_runs = number_of_sums;
+ size_t offset = 0;
+ while (current_runs != 0) {
+ static constexpr size_t DISPATCH_SIZE = 2048U;
+ size_t runs_to_do = std::min<size_t>(current_runs, DISPATCH_SIZE);
+ current_runs -= runs_to_do;
+ compute_pass_descriptor_queue.Acquire();
+ compute_pass_descriptor_queue.AddBuffer(src_buffer, 0, number_of_sums * sizeof(u64));
+ compute_pass_descriptor_queue.AddBuffer(dst_buffer, 0, number_of_sums * sizeof(u64));
+ compute_pass_descriptor_queue.AddBuffer(accumulation_buffer, 0, sizeof(u64));
+ const void* const descriptor_data{compute_pass_descriptor_queue.UpdateData()};
+ size_t used_offset = offset;
+ offset += runs_to_do;
+
+ scheduler.RequestOutsideRenderPassOperationContext();
+ scheduler.Record([this, descriptor_data, min_accumulation_limit, max_accumulation_limit,
+ runs_to_do, used_offset](vk::CommandBuffer cmdbuf) {
+ static constexpr VkMemoryBarrier read_barrier{
+ .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
+ .pNext = nullptr,
+ .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
+ .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT,
+ };
+ static constexpr VkMemoryBarrier write_barrier{
+ .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
+ .pNext = nullptr,
+ .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT,
+ .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_TRANSFER_READ_BIT |
+ VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT |
+ VK_ACCESS_INDIRECT_COMMAND_READ_BIT | VK_ACCESS_INDEX_READ_BIT |
+ VK_ACCESS_UNIFORM_READ_BIT |
+ VK_ACCESS_CONDITIONAL_RENDERING_READ_BIT_EXT,
+ };
+ const QueriesPrefixScanPushConstants uniforms{
+ .min_accumulation_base = static_cast<u32>(min_accumulation_limit),
+ .max_accumulation_base = static_cast<u32>(max_accumulation_limit),
+ .accumulation_limit = static_cast<u32>(runs_to_do - 1),
+ .buffer_offset = static_cast<u32>(used_offset),
+ };
+ const VkDescriptorSet set = descriptor_allocator.Commit();
+ device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data);
+
+ cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
+ VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, read_barrier);
+ cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, *pipeline);
+ cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {});
+ cmdbuf.PushConstants(*layout, VK_SHADER_STAGE_COMPUTE_BIT, uniforms);
+ cmdbuf.Dispatch(1, 1, 1);
+ cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
+ VK_PIPELINE_STAGE_CONDITIONAL_RENDERING_BIT_EXT, 0,
+ write_barrier);
+ });
+ }
+}
+
ASTCDecoderPass::ASTCDecoderPass(const Device& device_, Scheduler& scheduler_,
DescriptorPool& descriptor_pool_,
StagingBufferPool& staging_buffer_pool_,
@@ -413,4 +633,100 @@ void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map,
scheduler.Finish();
}
+MSAACopyPass::MSAACopyPass(const Device& device_, Scheduler& scheduler_,
+ DescriptorPool& descriptor_pool_,
+ StagingBufferPool& staging_buffer_pool_,
+ ComputePassDescriptorQueue& compute_pass_descriptor_queue_)
+ : ComputePass(device_, descriptor_pool_, MSAA_DESCRIPTOR_SET_BINDINGS,
+ MSAA_DESCRIPTOR_UPDATE_TEMPLATE, MSAA_BANK_INFO, {},
+ CONVERT_NON_MSAA_TO_MSAA_COMP_SPV),
+ scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_},
+ compute_pass_descriptor_queue{compute_pass_descriptor_queue_} {
+ const auto make_msaa_pipeline = [this](size_t i, std::span<const u32> code) {
+ modules[i] = device.GetLogical().CreateShaderModule({
+ .sType = VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO,
+ .pNext = nullptr,
+ .flags = 0,
+ .codeSize = static_cast<u32>(code.size_bytes()),
+ .pCode = code.data(),
+ });
+ pipelines[i] = device.GetLogical().CreateComputePipeline({
+ .sType = VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO,
+ .pNext = nullptr,
+ .flags = 0,
+ .stage{
+ .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
+ .pNext = nullptr,
+ .flags = 0,
+ .stage = VK_SHADER_STAGE_COMPUTE_BIT,
+ .module = *modules[i],
+ .pName = "main",
+ .pSpecializationInfo = nullptr,
+ },
+ .layout = *layout,
+ .basePipelineHandle = nullptr,
+ .basePipelineIndex = 0,
+ });
+ };
+ make_msaa_pipeline(0, CONVERT_NON_MSAA_TO_MSAA_COMP_SPV);
+ make_msaa_pipeline(1, CONVERT_MSAA_TO_NON_MSAA_COMP_SPV);
+}
+
+MSAACopyPass::~MSAACopyPass() = default;
+
+void MSAACopyPass::CopyImage(Image& dst_image, Image& src_image,
+ std::span<const VideoCommon::ImageCopy> copies,
+ bool msaa_to_non_msaa) {
+ const VkPipeline msaa_pipeline = *pipelines[msaa_to_non_msaa ? 1 : 0];
+ scheduler.RequestOutsideRenderPassOperationContext();
+ for (const VideoCommon::ImageCopy& copy : copies) {
+ ASSERT(copy.src_subresource.base_layer == 0);
+ ASSERT(copy.src_subresource.num_layers == 1);
+ ASSERT(copy.dst_subresource.base_layer == 0);
+ ASSERT(copy.dst_subresource.num_layers == 1);
+
+ compute_pass_descriptor_queue.Acquire();
+ compute_pass_descriptor_queue.AddImage(
+ src_image.StorageImageView(copy.src_subresource.base_level));
+ compute_pass_descriptor_queue.AddImage(
+ dst_image.StorageImageView(copy.dst_subresource.base_level));
+ const void* const descriptor_data{compute_pass_descriptor_queue.UpdateData()};
+
+ const Common::Vec3<u32> num_dispatches = {
+ Common::DivCeil(copy.extent.width, 8U),
+ Common::DivCeil(copy.extent.height, 8U),
+ copy.extent.depth,
+ };
+
+ scheduler.Record([this, dst = dst_image.Handle(), msaa_pipeline, num_dispatches,
+ descriptor_data](vk::CommandBuffer cmdbuf) {
+ const VkDescriptorSet set = descriptor_allocator.Commit();
+ device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data);
+ cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, msaa_pipeline);
+ cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *layout, 0, set, {});
+ cmdbuf.Dispatch(num_dispatches.x, num_dispatches.y, num_dispatches.z);
+ const VkImageMemoryBarrier write_barrier{
+ .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+ .pNext = nullptr,
+ .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT,
+ .dstAccessMask = VK_ACCESS_SHADER_READ_BIT,
+ .oldLayout = VK_IMAGE_LAYOUT_GENERAL,
+ .newLayout = VK_IMAGE_LAYOUT_GENERAL,
+ .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+ .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+ .image = dst,
+ .subresourceRange{
+ .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
+ .baseMipLevel = 0,
+ .levelCount = VK_REMAINING_MIP_LEVELS,
+ .baseArrayLayer = 0,
+ .layerCount = VK_REMAINING_ARRAY_LAYERS,
+ },
+ };
+ cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
+ VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, 0, write_barrier);
+ });
+ }
+}
+
} // namespace Vulkan
diff --git a/src/video_core/renderer_vulkan/vk_compute_pass.h b/src/video_core/renderer_vulkan/vk_compute_pass.h
index dd3927376..7b8f938c1 100644
--- a/src/video_core/renderer_vulkan/vk_compute_pass.h
+++ b/src/video_core/renderer_vulkan/vk_compute_pass.h
@@ -3,6 +3,7 @@
#pragma once
+#include <optional>
#include <span>
#include <utility>
@@ -10,6 +11,7 @@
#include "video_core/engines/maxwell_3d.h"
#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
#include "video_core/renderer_vulkan/vk_update_descriptor.h"
+#include "video_core/texture_cache/types.h"
#include "video_core/vulkan_common/vulkan_memory_allocator.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
@@ -31,7 +33,8 @@ public:
vk::Span<VkDescriptorSetLayoutBinding> bindings,
vk::Span<VkDescriptorUpdateTemplateEntry> templates,
const DescriptorBankInfo& bank_info,
- vk::Span<VkPushConstantRange> push_constants, std::span<const u32> code);
+ vk::Span<VkPushConstantRange> push_constants, std::span<const u32> code,
+ std::optional<u32> optional_subgroup_size = std::nullopt);
~ComputePass();
protected:
@@ -82,6 +85,33 @@ private:
ComputePassDescriptorQueue& compute_pass_descriptor_queue;
};
+class ConditionalRenderingResolvePass final : public ComputePass {
+public:
+ explicit ConditionalRenderingResolvePass(
+ const Device& device_, Scheduler& scheduler_, DescriptorPool& descriptor_pool_,
+ ComputePassDescriptorQueue& compute_pass_descriptor_queue_);
+
+ void Resolve(VkBuffer dst_buffer, VkBuffer src_buffer, u32 src_offset, bool compare_to_zero);
+
+private:
+ Scheduler& scheduler;
+ ComputePassDescriptorQueue& compute_pass_descriptor_queue;
+};
+
+class QueriesPrefixScanPass final : public ComputePass {
+public:
+ explicit QueriesPrefixScanPass(const Device& device_, Scheduler& scheduler_,
+ DescriptorPool& descriptor_pool_,
+ ComputePassDescriptorQueue& compute_pass_descriptor_queue_);
+
+ void Run(VkBuffer accumulation_buffer, VkBuffer dst_buffer, VkBuffer src_buffer,
+ size_t number_of_sums, size_t min_accumulation_limit, size_t max_accumulation_limit);
+
+private:
+ Scheduler& scheduler;
+ ComputePassDescriptorQueue& compute_pass_descriptor_queue;
+};
+
class ASTCDecoderPass final : public ComputePass {
public:
explicit ASTCDecoderPass(const Device& device_, Scheduler& scheduler_,
@@ -101,4 +131,22 @@ private:
MemoryAllocator& memory_allocator;
};
+class MSAACopyPass final : public ComputePass {
+public:
+ explicit MSAACopyPass(const Device& device_, Scheduler& scheduler_,
+ DescriptorPool& descriptor_pool_, StagingBufferPool& staging_buffer_pool_,
+ ComputePassDescriptorQueue& compute_pass_descriptor_queue_);
+ ~MSAACopyPass();
+
+ void CopyImage(Image& dst_image, Image& src_image,
+ std::span<const VideoCommon::ImageCopy> copies, bool msaa_to_non_msaa);
+
+private:
+ Scheduler& scheduler;
+ StagingBufferPool& staging_buffer_pool;
+ ComputePassDescriptorQueue& compute_pass_descriptor_queue;
+ std::array<vk::ShaderModule, 2> modules;
+ std::array<vk::Pipeline, 2> pipelines;
+};
+
} // namespace Vulkan
diff --git a/src/video_core/renderer_vulkan/vk_fence_manager.h b/src/video_core/renderer_vulkan/vk_fence_manager.h
index 145359d4e..336573574 100644
--- a/src/video_core/renderer_vulkan/vk_fence_manager.h
+++ b/src/video_core/renderer_vulkan/vk_fence_manager.h
@@ -7,6 +7,7 @@
#include "video_core/fence_manager.h"
#include "video_core/renderer_vulkan/vk_buffer_cache.h"
+#include "video_core/renderer_vulkan/vk_query_cache.h"
#include "video_core/renderer_vulkan/vk_texture_cache.h"
namespace Core {
@@ -20,7 +21,6 @@ class RasterizerInterface;
namespace Vulkan {
class Device;
-class QueryCache;
class Scheduler;
class InnerFence : public VideoCommon::FenceBase {
diff --git a/src/video_core/renderer_vulkan/vk_query_cache.cpp b/src/video_core/renderer_vulkan/vk_query_cache.cpp
index 29e0b797b..2edaafa7e 100644
--- a/src/video_core/renderer_vulkan/vk_query_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_query_cache.cpp
@@ -1,139 +1,1555 @@
-// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
-// SPDX-License-Identifier: GPL-2.0-or-later
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
-#include <algorithm>
#include <cstddef>
+#include <limits>
+#include <map>
+#include <memory>
+#include <span>
+#include <type_traits>
+#include <unordered_map>
#include <utility>
#include <vector>
+#include "common/bit_util.h"
+#include "common/common_types.h"
+#include "core/memory.h"
+#include "video_core/engines/draw_manager.h"
+#include "video_core/query_cache/query_cache.h"
+#include "video_core/renderer_vulkan/vk_buffer_cache.h"
+#include "video_core/renderer_vulkan/vk_compute_pass.h"
#include "video_core/renderer_vulkan/vk_query_cache.h"
#include "video_core/renderer_vulkan/vk_resource_pool.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
+#include "video_core/renderer_vulkan/vk_staging_buffer_pool.h"
+#include "video_core/renderer_vulkan/vk_update_descriptor.h"
#include "video_core/vulkan_common/vulkan_device.h"
+#include "video_core/vulkan_common/vulkan_memory_allocator.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
namespace Vulkan {
-using VideoCore::QueryType;
+using Tegra::Engines::Maxwell3D;
+using VideoCommon::QueryType;
namespace {
+class SamplesQueryBank : public VideoCommon::BankBase {
+public:
+ static constexpr size_t BANK_SIZE = 256;
+ static constexpr size_t QUERY_SIZE = 8;
+ explicit SamplesQueryBank(const Device& device_, size_t index_)
+ : BankBase(BANK_SIZE), device{device_}, index{index_} {
+ const auto& dev = device.GetLogical();
+ query_pool = dev.CreateQueryPool({
+ .sType = VK_STRUCTURE_TYPE_QUERY_POOL_CREATE_INFO,
+ .pNext = nullptr,
+ .flags = 0,
+ .queryType = VK_QUERY_TYPE_OCCLUSION,
+ .queryCount = BANK_SIZE,
+ .pipelineStatistics = 0,
+ });
+ Reset();
+ }
-constexpr std::array QUERY_TARGETS = {VK_QUERY_TYPE_OCCLUSION};
+ ~SamplesQueryBank() = default;
-constexpr VkQueryType GetTarget(QueryType type) {
- return QUERY_TARGETS[static_cast<std::size_t>(type)];
-}
+ void Reset() override {
+ ASSERT(references == 0);
+ VideoCommon::BankBase::Reset();
+ const auto& dev = device.GetLogical();
+ dev.ResetQueryPool(*query_pool, 0, BANK_SIZE);
+ host_results.fill(0ULL);
+ next_bank = 0;
+ }
+
+ void Sync(size_t start, size_t size) {
+ const auto& dev = device.GetLogical();
+ const VkResult query_result = dev.GetQueryResults(
+ *query_pool, static_cast<u32>(start), static_cast<u32>(size), sizeof(u64) * size,
+ &host_results[start], sizeof(u64), VK_QUERY_RESULT_64_BIT | VK_QUERY_RESULT_WAIT_BIT);
+ switch (query_result) {
+ case VK_SUCCESS:
+ return;
+ case VK_ERROR_DEVICE_LOST:
+ device.ReportLoss();
+ [[fallthrough]];
+ default:
+ throw vk::Exception(query_result);
+ }
+ }
+
+ VkQueryPool GetInnerPool() {
+ return *query_pool;
+ }
+
+ size_t GetIndex() const {
+ return index;
+ }
+
+ const std::array<u64, BANK_SIZE>& GetResults() const {
+ return host_results;
+ }
+
+ size_t next_bank;
+
+private:
+ const Device& device;
+ const size_t index;
+ vk::QueryPool query_pool;
+ std::array<u64, BANK_SIZE> host_results;
+};
+
+using BaseStreamer = VideoCommon::SimpleStreamer<VideoCommon::HostQueryBase>;
+
+struct HostSyncValues {
+ VAddr address;
+ size_t size;
+ size_t offset;
+
+ static constexpr bool GeneratesBaseBuffer = false;
+};
+
+class SamplesStreamer : public BaseStreamer {
+public:
+ explicit SamplesStreamer(size_t id_, QueryCacheRuntime& runtime_,
+ VideoCore::RasterizerInterface* rasterizer_, const Device& device_,
+ Scheduler& scheduler_, const MemoryAllocator& memory_allocator_,
+ ComputePassDescriptorQueue& compute_pass_descriptor_queue,
+ DescriptorPool& descriptor_pool)
+ : BaseStreamer(id_), runtime{runtime_}, rasterizer{rasterizer_}, device{device_},
+ scheduler{scheduler_}, memory_allocator{memory_allocator_} {
+ current_bank = nullptr;
+ current_query = nullptr;
+ ammend_value = 0;
+ acumulation_value = 0;
+ queries_prefix_scan_pass = std::make_unique<QueriesPrefixScanPass>(
+ device, scheduler, descriptor_pool, compute_pass_descriptor_queue);
+
+ const VkBufferCreateInfo buffer_ci = {
+ .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
+ .pNext = nullptr,
+ .flags = 0,
+ .size = 8,
+ .usage = VK_BUFFER_USAGE_TRANSFER_DST_BIT | VK_BUFFER_USAGE_TRANSFER_SRC_BIT |
+ VK_BUFFER_USAGE_STORAGE_BUFFER_BIT,
+ .sharingMode = VK_SHARING_MODE_EXCLUSIVE,
+ .queueFamilyIndexCount = 0,
+ .pQueueFamilyIndices = nullptr,
+ };
+ accumulation_buffer = memory_allocator.CreateBuffer(buffer_ci, MemoryUsage::DeviceLocal);
+ scheduler.RequestOutsideRenderPassOperationContext();
+ scheduler.Record([buffer = *accumulation_buffer](vk::CommandBuffer cmdbuf) {
+ cmdbuf.FillBuffer(buffer, 0, 8, 0);
+ });
+ }
+
+ ~SamplesStreamer() = default;
+
+ void StartCounter() override {
+ if (has_started) {
+ return;
+ }
+ ReserveHostQuery();
+ scheduler.Record([query_pool = current_query_pool,
+ query_index = current_bank_slot](vk::CommandBuffer cmdbuf) {
+ const bool use_precise = Settings::IsGPULevelHigh();
+ cmdbuf.BeginQuery(query_pool, static_cast<u32>(query_index),
+ use_precise ? VK_QUERY_CONTROL_PRECISE_BIT : 0);
+ });
+ has_started = true;
+ }
+
+ void PauseCounter() override {
+ if (!has_started) {
+ return;
+ }
+ scheduler.Record([query_pool = current_query_pool,
+ query_index = current_bank_slot](vk::CommandBuffer cmdbuf) {
+ cmdbuf.EndQuery(query_pool, static_cast<u32>(query_index));
+ });
+ has_started = false;
+ }
+
+ void ResetCounter() override {
+ if (has_started) {
+ PauseCounter();
+ }
+ AbandonCurrentQuery();
+ std::function<void()> func([this, counts = pending_flush_queries.size()] {
+ ammend_value = 0;
+ acumulation_value = 0;
+ });
+ rasterizer->SyncOperation(std::move(func));
+ accumulation_since_last_sync = false;
+ first_accumulation_checkpoint = std::min(first_accumulation_checkpoint, num_slots_used);
+ last_accumulation_checkpoint = std::max(last_accumulation_checkpoint, num_slots_used);
+ }
+
+ void CloseCounter() override {
+ PauseCounter();
+ }
-} // Anonymous namespace
+ bool HasPendingSync() const override {
+ return !pending_sync.empty();
+ }
+
+ void SyncWrites() override {
+ if (sync_values_stash.empty()) {
+ return;
+ }
-QueryPool::QueryPool(const Device& device_, Scheduler& scheduler, QueryType type_)
- : ResourcePool{scheduler.GetMasterSemaphore(), GROW_STEP}, device{device_}, type{type_} {}
+ for (size_t i = 0; i < sync_values_stash.size(); i++) {
+ runtime.template SyncValues<HostSyncValues>(sync_values_stash[i],
+ *buffers[resolve_buffers[i]]);
+ }
+
+ sync_values_stash.clear();
+ }
-QueryPool::~QueryPool() = default;
+ void PresyncWrites() override {
+ if (pending_sync.empty()) {
+ return;
+ }
+ PauseCounter();
+ sync_values_stash.clear();
+ sync_values_stash.emplace_back();
+ std::vector<HostSyncValues>* sync_values = &sync_values_stash.back();
+ sync_values->reserve(num_slots_used);
+ std::unordered_map<size_t, std::pair<size_t, size_t>> offsets;
+ resolve_buffers.clear();
+ size_t resolve_buffer_index = ObtainBuffer<true>(num_slots_used);
+ resolve_buffers.push_back(resolve_buffer_index);
+ size_t base_offset = 0;
-std::pair<VkQueryPool, u32> QueryPool::Commit() {
- std::size_t index;
- do {
- index = CommitResource();
- } while (usage[index]);
- usage[index] = true;
+ ApplyBanksWideOp<true>(pending_sync, [&](SamplesQueryBank* bank, size_t start,
+ size_t amount) {
+ size_t bank_id = bank->GetIndex();
+ auto& resolve_buffer = buffers[resolve_buffer_index];
+ VkQueryPool query_pool = bank->GetInnerPool();
+ scheduler.RequestOutsideRenderPassOperationContext();
+ scheduler.Record([start, amount, base_offset, query_pool,
+ buffer = *resolve_buffer](vk::CommandBuffer cmdbuf) {
+ const VkBufferMemoryBarrier copy_query_pool_barrier{
+ .sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER,
+ .pNext = nullptr,
+ .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
+ .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT,
+ .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+ .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+ .buffer = buffer,
+ .offset = base_offset,
+ .size = amount * SamplesQueryBank::QUERY_SIZE,
+ };
+
+ cmdbuf.CopyQueryPoolResults(
+ query_pool, static_cast<u32>(start), static_cast<u32>(amount), buffer,
+ static_cast<u32>(base_offset), SamplesQueryBank::QUERY_SIZE,
+ VK_QUERY_RESULT_WAIT_BIT | VK_QUERY_RESULT_64_BIT);
+ cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT,
+ VK_PIPELINE_STAGE_TRANSFER_BIT, 0, copy_query_pool_barrier);
+ });
+ offsets[bank_id] = {start, base_offset};
+ base_offset += amount * SamplesQueryBank::QUERY_SIZE;
+ });
+
+ // Convert queries
+ bool has_multi_queries = false;
+ for (auto q : pending_sync) {
+ auto* query = GetQuery(q);
+ size_t sync_value_slot = 0;
+ if (True(query->flags & VideoCommon::QueryFlagBits::IsRewritten)) {
+ continue;
+ }
+ if (True(query->flags & VideoCommon::QueryFlagBits::IsInvalidated)) {
+ continue;
+ }
+ if (accumulation_since_last_sync || query->size_slots > 1) {
+ if (!has_multi_queries) {
+ has_multi_queries = true;
+ sync_values_stash.emplace_back();
+ }
+ sync_value_slot = 1;
+ }
+ query->flags |= VideoCommon::QueryFlagBits::IsHostSynced;
+ auto loc_data = offsets[query->start_bank_id];
+ sync_values_stash[sync_value_slot].emplace_back(HostSyncValues{
+ .address = query->guest_address,
+ .size = SamplesQueryBank::QUERY_SIZE,
+ .offset =
+ loc_data.second + (query->start_slot - loc_data.first + query->size_slots - 1) *
+ SamplesQueryBank::QUERY_SIZE,
+ });
+ }
+
+ if (has_multi_queries) {
+ size_t intermediary_buffer_index = ObtainBuffer<false>(num_slots_used);
+ resolve_buffers.push_back(intermediary_buffer_index);
+ queries_prefix_scan_pass->Run(*accumulation_buffer, *buffers[intermediary_buffer_index],
+ *buffers[resolve_buffer_index], num_slots_used,
+ std::min(first_accumulation_checkpoint, num_slots_used),
+ last_accumulation_checkpoint);
+
+ } else {
+ scheduler.RequestOutsideRenderPassOperationContext();
+ scheduler.Record([buffer = *accumulation_buffer](vk::CommandBuffer cmdbuf) {
+ cmdbuf.FillBuffer(buffer, 0, 8, 0);
+ });
+ }
+
+ ReplicateCurrentQueryIfNeeded();
+ std::function<void()> func([this] { ammend_value = acumulation_value; });
+ rasterizer->SyncOperation(std::move(func));
+ AbandonCurrentQuery();
+ num_slots_used = 0;
+ first_accumulation_checkpoint = std::numeric_limits<size_t>::max();
+ last_accumulation_checkpoint = 0;
+ accumulation_since_last_sync = has_multi_queries;
+ pending_sync.clear();
+ }
- return {*pools[index / GROW_STEP], static_cast<u32>(index % GROW_STEP)};
+ size_t WriteCounter(VAddr address, bool has_timestamp, u32 value,
+ [[maybe_unused]] std::optional<u32> subreport) override {
+ PauseCounter();
+ auto index = BuildQuery();
+ auto* new_query = GetQuery(index);
+ new_query->guest_address = address;
+ new_query->value = 0;
+ new_query->flags &= ~VideoCommon::QueryFlagBits::IsOrphan;
+ if (has_timestamp) {
+ new_query->flags |= VideoCommon::QueryFlagBits::HasTimestamp;
+ }
+ if (!current_query) {
+ new_query->flags |= VideoCommon::QueryFlagBits::IsFinalValueSynced;
+ return index;
+ }
+ new_query->start_bank_id = current_query->start_bank_id;
+ new_query->size_banks = current_query->size_banks;
+ new_query->start_slot = current_query->start_slot;
+ new_query->size_slots = current_query->size_slots;
+ ApplyBankOp(new_query, [](SamplesQueryBank* bank, size_t start, size_t amount) {
+ bank->AddReference(amount);
+ });
+ pending_sync.push_back(index);
+ pending_flush_queries.push_back(index);
+ return index;
+ }
+
+ bool HasUnsyncedQueries() const override {
+ return !pending_flush_queries.empty();
+ }
+
+ void PushUnsyncedQueries() override {
+ PauseCounter();
+ current_bank->Close();
+ {
+ std::scoped_lock lk(flush_guard);
+ pending_flush_sets.emplace_back(std::move(pending_flush_queries));
+ }
+ }
+
+ void PopUnsyncedQueries() override {
+ std::vector<size_t> current_flush_queries;
+ {
+ std::scoped_lock lk(flush_guard);
+ current_flush_queries = std::move(pending_flush_sets.front());
+ pending_flush_sets.pop_front();
+ }
+ ApplyBanksWideOp<false>(
+ current_flush_queries,
+ [](SamplesQueryBank* bank, size_t start, size_t amount) { bank->Sync(start, amount); });
+ for (auto q : current_flush_queries) {
+ auto* query = GetQuery(q);
+ u64 total = 0;
+ ApplyBankOp(query, [&total](SamplesQueryBank* bank, size_t start, size_t amount) {
+ const auto& results = bank->GetResults();
+ for (size_t i = 0; i < amount; i++) {
+ total += results[start + i];
+ }
+ });
+ query->value = total;
+ query->flags |= VideoCommon::QueryFlagBits::IsFinalValueSynced;
+ }
+ }
+
+private:
+ template <typename Func>
+ void ApplyBankOp(VideoCommon::HostQueryBase* query, Func&& func) {
+ size_t size_slots = query->size_slots;
+ if (size_slots == 0) {
+ return;
+ }
+ size_t bank_id = query->start_bank_id;
+ size_t banks_set = query->size_banks;
+ size_t start_slot = query->start_slot;
+ for (size_t i = 0; i < banks_set; i++) {
+ auto& the_bank = bank_pool.GetBank(bank_id);
+ size_t amount = std::min(the_bank.Size() - start_slot, size_slots);
+ func(&the_bank, start_slot, amount);
+ bank_id = the_bank.next_bank - 1;
+ start_slot = 0;
+ size_slots -= amount;
+ }
+ }
+
+ template <bool is_ordered, typename Func>
+ void ApplyBanksWideOp(std::vector<size_t>& queries, Func&& func) {
+ std::conditional_t<is_ordered, std::map<size_t, std::pair<size_t, size_t>>,
+ std::unordered_map<size_t, std::pair<size_t, size_t>>>
+ indexer;
+ for (auto q : queries) {
+ auto* query = GetQuery(q);
+ ApplyBankOp(query, [&indexer](SamplesQueryBank* bank, size_t start, size_t amount) {
+ auto id_ = bank->GetIndex();
+ auto pair = indexer.try_emplace(id_, std::numeric_limits<size_t>::max(),
+ std::numeric_limits<size_t>::min());
+ auto& current_pair = pair.first->second;
+ current_pair.first = std::min(current_pair.first, start);
+ current_pair.second = std::max(current_pair.second, amount + start);
+ });
+ }
+ for (auto& cont : indexer) {
+ func(&bank_pool.GetBank(cont.first), cont.second.first,
+ cont.second.second - cont.second.first);
+ }
+ }
+
+ void ReserveBank() {
+ current_bank_id =
+ bank_pool.ReserveBank([this](std::deque<SamplesQueryBank>& queue, size_t index) {
+ queue.emplace_back(device, index);
+ });
+ if (current_bank) {
+ current_bank->next_bank = current_bank_id + 1;
+ }
+ current_bank = &bank_pool.GetBank(current_bank_id);
+ current_query_pool = current_bank->GetInnerPool();
+ }
+
+ size_t ReserveBankSlot() {
+ if (!current_bank || current_bank->IsClosed()) {
+ ReserveBank();
+ }
+ auto [built, index] = current_bank->Reserve();
+ current_bank_slot = index;
+ return index;
+ }
+
+ void ReserveHostQuery() {
+ size_t new_slot = ReserveBankSlot();
+ current_bank->AddReference(1);
+ num_slots_used++;
+ if (current_query) {
+ size_t bank_id = current_query->start_bank_id;
+ size_t banks_set = current_query->size_banks - 1;
+ bool found = bank_id == current_bank_id;
+ while (!found && banks_set > 0) {
+ SamplesQueryBank& some_bank = bank_pool.GetBank(bank_id);
+ bank_id = some_bank.next_bank - 1;
+ found = bank_id == current_bank_id;
+ banks_set--;
+ }
+ if (!found) {
+ current_query->size_banks++;
+ }
+ current_query->size_slots++;
+ } else {
+ current_query_id = BuildQuery();
+ current_query = GetQuery(current_query_id);
+ current_query->start_bank_id = static_cast<u32>(current_bank_id);
+ current_query->size_banks = 1;
+ current_query->start_slot = new_slot;
+ current_query->size_slots = 1;
+ }
+ }
+
+ void Free(size_t query_id) override {
+ std::scoped_lock lk(guard);
+ auto* query = GetQuery(query_id);
+ ApplyBankOp(query, [](SamplesQueryBank* bank, size_t start, size_t amount) {
+ bank->CloseReference(amount);
+ });
+ ReleaseQuery(query_id);
+ }
+
+ void AbandonCurrentQuery() {
+ if (!current_query) {
+ return;
+ }
+ Free(current_query_id);
+ current_query = nullptr;
+ current_query_id = 0;
+ }
+
+ void ReplicateCurrentQueryIfNeeded() {
+ if (pending_sync.empty()) {
+ return;
+ }
+ if (!current_query) {
+ return;
+ }
+ auto index = BuildQuery();
+ auto* new_query = GetQuery(index);
+ new_query->guest_address = 0;
+ new_query->value = 0;
+ new_query->flags &= ~VideoCommon::QueryFlagBits::IsOrphan;
+ new_query->start_bank_id = current_query->start_bank_id;
+ new_query->size_banks = current_query->size_banks;
+ new_query->start_slot = current_query->start_slot;
+ new_query->size_slots = current_query->size_slots;
+ ApplyBankOp(new_query, [](SamplesQueryBank* bank, size_t start, size_t amount) {
+ bank->AddReference(amount);
+ });
+ pending_flush_queries.push_back(index);
+ std::function<void()> func([this, index] {
+ auto* query = GetQuery(index);
+ query->value += GetAmmendValue();
+ SetAccumulationValue(query->value);
+ Free(index);
+ });
+ rasterizer->SyncOperation(std::move(func));
+ }
+
+ template <bool is_resolve>
+ size_t ObtainBuffer(size_t num_needed) {
+ const size_t log_2 = std::max<size_t>(11U, Common::Log2Ceil64(num_needed));
+ if constexpr (is_resolve) {
+ if (resolve_table[log_2] != 0) {
+ return resolve_table[log_2] - 1;
+ }
+ } else {
+ if (intermediary_table[log_2] != 0) {
+ return intermediary_table[log_2] - 1;
+ }
+ }
+ const VkBufferCreateInfo buffer_ci = {
+ .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
+ .pNext = nullptr,
+ .flags = 0,
+ .size = SamplesQueryBank::QUERY_SIZE * (1ULL << log_2),
+ .usage = VK_BUFFER_USAGE_TRANSFER_DST_BIT | VK_BUFFER_USAGE_TRANSFER_SRC_BIT |
+ VK_BUFFER_USAGE_STORAGE_BUFFER_BIT,
+ .sharingMode = VK_SHARING_MODE_EXCLUSIVE,
+ .queueFamilyIndexCount = 0,
+ .pQueueFamilyIndices = nullptr,
+ };
+ buffers.emplace_back(memory_allocator.CreateBuffer(buffer_ci, MemoryUsage::DeviceLocal));
+ if constexpr (is_resolve) {
+ resolve_table[log_2] = buffers.size();
+ } else {
+ intermediary_table[log_2] = buffers.size();
+ }
+ return buffers.size() - 1;
+ }
+
+ QueryCacheRuntime& runtime;
+ VideoCore::RasterizerInterface* rasterizer;
+ const Device& device;
+ Scheduler& scheduler;
+ const MemoryAllocator& memory_allocator;
+ VideoCommon::BankPool<SamplesQueryBank> bank_pool;
+ std::deque<vk::Buffer> buffers;
+ std::array<size_t, 32> resolve_table{};
+ std::array<size_t, 32> intermediary_table{};
+ vk::Buffer accumulation_buffer;
+ std::deque<std::vector<HostSyncValues>> sync_values_stash;
+ std::vector<size_t> resolve_buffers;
+
+ // syncing queue
+ std::vector<size_t> pending_sync;
+
+ // flush levels
+ std::vector<size_t> pending_flush_queries;
+ std::deque<std::vector<size_t>> pending_flush_sets;
+
+ // State Machine
+ size_t current_bank_slot;
+ size_t current_bank_id;
+ SamplesQueryBank* current_bank;
+ VkQueryPool current_query_pool;
+ size_t current_query_id;
+ size_t num_slots_used{};
+ size_t first_accumulation_checkpoint{};
+ size_t last_accumulation_checkpoint{};
+ bool accumulation_since_last_sync{};
+ VideoCommon::HostQueryBase* current_query;
+ bool has_started{};
+ std::mutex flush_guard;
+
+ std::unique_ptr<QueriesPrefixScanPass> queries_prefix_scan_pass;
+};
+
+// Transform feedback queries
+class TFBQueryBank : public VideoCommon::BankBase {
+public:
+ static constexpr size_t BANK_SIZE = 1024;
+ static constexpr size_t QUERY_SIZE = 4;
+ explicit TFBQueryBank(Scheduler& scheduler_, const MemoryAllocator& memory_allocator,
+ size_t index_)
+ : BankBase(BANK_SIZE), scheduler{scheduler_}, index{index_} {
+ const VkBufferCreateInfo buffer_ci = {
+ .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
+ .pNext = nullptr,
+ .flags = 0,
+ .size = QUERY_SIZE * BANK_SIZE,
+ .usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT,
+ .sharingMode = VK_SHARING_MODE_EXCLUSIVE,
+ .queueFamilyIndexCount = 0,
+ .pQueueFamilyIndices = nullptr,
+ };
+ buffer = memory_allocator.CreateBuffer(buffer_ci, MemoryUsage::DeviceLocal);
+ }
+
+ ~TFBQueryBank() = default;
+
+ void Reset() override {
+ ASSERT(references == 0);
+ VideoCommon::BankBase::Reset();
+ }
+
+ void Sync(StagingBufferRef& stagging_buffer, size_t extra_offset, size_t start, size_t size) {
+ scheduler.RequestOutsideRenderPassOperationContext();
+ scheduler.Record([this, dst_buffer = stagging_buffer.buffer, extra_offset, start,
+ size](vk::CommandBuffer cmdbuf) {
+ std::array<VkBufferCopy, 1> copy{VkBufferCopy{
+ .srcOffset = start * QUERY_SIZE,
+ .dstOffset = extra_offset,
+ .size = size * QUERY_SIZE,
+ }};
+ cmdbuf.CopyBuffer(*buffer, dst_buffer, copy);
+ });
+ }
+
+ size_t GetIndex() const {
+ return index;
+ }
+
+ VkBuffer GetBuffer() const {
+ return *buffer;
+ }
+
+private:
+ Scheduler& scheduler;
+ const size_t index;
+ vk::Buffer buffer;
+};
+
+class PrimitivesSucceededStreamer;
+
+class TFBCounterStreamer : public BaseStreamer {
+public:
+ explicit TFBCounterStreamer(size_t id_, QueryCacheRuntime& runtime_, const Device& device_,
+ Scheduler& scheduler_, const MemoryAllocator& memory_allocator_,
+ StagingBufferPool& staging_pool_)
+ : BaseStreamer(id_), runtime{runtime_}, device{device_}, scheduler{scheduler_},
+ memory_allocator{memory_allocator_}, staging_pool{staging_pool_} {
+ buffers_count = 0;
+ current_bank = nullptr;
+ counter_buffers.fill(VK_NULL_HANDLE);
+ offsets.fill(0);
+ last_queries.fill(0);
+ last_queries_stride.fill(1);
+ const VkBufferCreateInfo buffer_ci = {
+ .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
+ .pNext = nullptr,
+ .flags = 0,
+ .size = TFBQueryBank::QUERY_SIZE * NUM_STREAMS,
+ .usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT |
+ VK_BUFFER_USAGE_TRANSFORM_FEEDBACK_COUNTER_BUFFER_BIT_EXT,
+ .sharingMode = VK_SHARING_MODE_EXCLUSIVE,
+ .queueFamilyIndexCount = 0,
+ .pQueueFamilyIndices = nullptr,
+ };
+
+ counters_buffer = memory_allocator.CreateBuffer(buffer_ci, MemoryUsage::DeviceLocal);
+ for (auto& c : counter_buffers) {
+ c = *counters_buffer;
+ }
+ size_t base_offset = 0;
+ for (auto& o : offsets) {
+ o = base_offset;
+ base_offset += TFBQueryBank::QUERY_SIZE;
+ }
+ }
+
+ ~TFBCounterStreamer() = default;
+
+ void StartCounter() override {
+ FlushBeginTFB();
+ has_started = true;
+ }
+
+ void PauseCounter() override {
+ CloseCounter();
+ }
+
+ void ResetCounter() override {
+ CloseCounter();
+ }
+
+ void CloseCounter() override {
+ if (has_flushed_end_pending) {
+ FlushEndTFB();
+ }
+ runtime.View3DRegs([this](Maxwell3D& maxwell3d) {
+ if (maxwell3d.regs.transform_feedback_enabled == 0) {
+ streams_mask = 0;
+ has_started = false;
+ }
+ });
+ }
+
+ bool HasPendingSync() const override {
+ return !pending_sync.empty();
+ }
+
+ void SyncWrites() override {
+ CloseCounter();
+ std::unordered_map<size_t, std::vector<HostSyncValues>> sync_values_stash;
+ for (auto q : pending_sync) {
+ auto* query = GetQuery(q);
+ if (True(query->flags & VideoCommon::QueryFlagBits::IsRewritten)) {
+ continue;
+ }
+ if (True(query->flags & VideoCommon::QueryFlagBits::IsInvalidated)) {
+ continue;
+ }
+ query->flags |= VideoCommon::QueryFlagBits::IsHostSynced;
+ sync_values_stash.try_emplace(query->start_bank_id);
+ sync_values_stash[query->start_bank_id].emplace_back(HostSyncValues{
+ .address = query->guest_address,
+ .size = TFBQueryBank::QUERY_SIZE,
+ .offset = query->start_slot * TFBQueryBank::QUERY_SIZE,
+ });
+ }
+ for (auto& p : sync_values_stash) {
+ auto& bank = bank_pool.GetBank(p.first);
+ runtime.template SyncValues<HostSyncValues>(p.second, bank.GetBuffer());
+ }
+ pending_sync.clear();
+ }
+
+ size_t WriteCounter(VAddr address, bool has_timestamp, u32 value,
+ std::optional<u32> subreport_) override {
+ auto index = BuildQuery();
+ auto* new_query = GetQuery(index);
+ new_query->guest_address = address;
+ new_query->value = 0;
+ new_query->flags &= ~VideoCommon::QueryFlagBits::IsOrphan;
+ if (has_timestamp) {
+ new_query->flags |= VideoCommon::QueryFlagBits::HasTimestamp;
+ }
+ if (!subreport_) {
+ new_query->flags |= VideoCommon::QueryFlagBits::IsFinalValueSynced;
+ return index;
+ }
+ const size_t subreport = static_cast<size_t>(*subreport_);
+ last_queries[subreport] = address;
+ if ((streams_mask & (1ULL << subreport)) == 0) {
+ new_query->flags |= VideoCommon::QueryFlagBits::IsFinalValueSynced;
+ return index;
+ }
+ CloseCounter();
+ auto [bank_slot, data_slot] = ProduceCounterBuffer(subreport);
+ new_query->start_bank_id = static_cast<u32>(bank_slot);
+ new_query->size_banks = 1;
+ new_query->start_slot = static_cast<u32>(data_slot);
+ new_query->size_slots = 1;
+ pending_sync.push_back(index);
+ pending_flush_queries.push_back(index);
+ return index;
+ }
+
+ std::optional<std::pair<VAddr, size_t>> GetLastQueryStream(size_t stream) {
+ if (last_queries[stream] != 0) {
+ std::pair<VAddr, size_t> result(last_queries[stream], last_queries_stride[stream]);
+ return result;
+ }
+ return std::nullopt;
+ }
+
+ Maxwell3D::Regs::PrimitiveTopology GetOutputTopology() const {
+ return out_topology;
+ }
+
+ bool HasUnsyncedQueries() const override {
+ return !pending_flush_queries.empty();
+ }
+
+ void PushUnsyncedQueries() override {
+ CloseCounter();
+ auto staging_ref = staging_pool.Request(
+ pending_flush_queries.size() * TFBQueryBank::QUERY_SIZE, MemoryUsage::Download, true);
+ size_t offset_base = staging_ref.offset;
+ for (auto q : pending_flush_queries) {
+ auto* query = GetQuery(q);
+ auto& bank = bank_pool.GetBank(query->start_bank_id);
+ bank.Sync(staging_ref, offset_base, query->start_slot, 1);
+ offset_base += TFBQueryBank::QUERY_SIZE;
+ bank.CloseReference();
+ }
+ static constexpr VkMemoryBarrier WRITE_BARRIER{
+ .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
+ .pNext = nullptr,
+ .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
+ .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT,
+ };
+ scheduler.RequestOutsideRenderPassOperationContext();
+ scheduler.Record([](vk::CommandBuffer cmdbuf) {
+ cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT,
+ VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, WRITE_BARRIER);
+ });
+
+ std::scoped_lock lk(flush_guard);
+ for (auto& str : free_queue) {
+ staging_pool.FreeDeferred(str);
+ }
+ free_queue.clear();
+ download_buffers.emplace_back(staging_ref);
+ pending_flush_sets.emplace_back(std::move(pending_flush_queries));
+ }
+
+ void PopUnsyncedQueries() override {
+ StagingBufferRef staging_ref;
+ std::vector<size_t> flushed_queries;
+ {
+ std::scoped_lock lk(flush_guard);
+ staging_ref = download_buffers.front();
+ flushed_queries = std::move(pending_flush_sets.front());
+ download_buffers.pop_front();
+ pending_flush_sets.pop_front();
+ }
+
+ size_t offset_base = staging_ref.offset;
+ for (auto q : flushed_queries) {
+ auto* query = GetQuery(q);
+ u32 result = 0;
+ std::memcpy(&result, staging_ref.mapped_span.data() + offset_base, sizeof(u32));
+ query->value = static_cast<u64>(result);
+ query->flags |= VideoCommon::QueryFlagBits::IsFinalValueSynced;
+ offset_base += TFBQueryBank::QUERY_SIZE;
+ }
+
+ {
+ std::scoped_lock lk(flush_guard);
+ free_queue.emplace_back(staging_ref);
+ }
+ }
+
+private:
+ void FlushBeginTFB() {
+ if (has_flushed_end_pending) [[unlikely]] {
+ return;
+ }
+ has_flushed_end_pending = true;
+ if (!has_started || buffers_count == 0) {
+ scheduler.Record([](vk::CommandBuffer cmdbuf) {
+ cmdbuf.BeginTransformFeedbackEXT(0, 0, nullptr, nullptr);
+ });
+ UpdateBuffers();
+ return;
+ }
+ scheduler.Record([this, total = static_cast<u32>(buffers_count)](vk::CommandBuffer cmdbuf) {
+ cmdbuf.BeginTransformFeedbackEXT(0, total, counter_buffers.data(), offsets.data());
+ });
+ UpdateBuffers();
+ }
+
+ void FlushEndTFB() {
+ if (!has_flushed_end_pending) [[unlikely]] {
+ UNREACHABLE();
+ return;
+ }
+ has_flushed_end_pending = false;
+
+ if (buffers_count == 0) {
+ scheduler.Record([](vk::CommandBuffer cmdbuf) {
+ cmdbuf.EndTransformFeedbackEXT(0, 0, nullptr, nullptr);
+ });
+ } else {
+ scheduler.Record([this,
+ total = static_cast<u32>(buffers_count)](vk::CommandBuffer cmdbuf) {
+ cmdbuf.EndTransformFeedbackEXT(0, total, counter_buffers.data(), offsets.data());
+ });
+ }
+ }
+
+ void UpdateBuffers() {
+ last_queries.fill(0);
+ last_queries_stride.fill(1);
+ runtime.View3DRegs([this](Maxwell3D& maxwell3d) {
+ buffers_count = 0;
+ out_topology = maxwell3d.draw_manager->GetDrawState().topology;
+ for (size_t i = 0; i < Maxwell3D::Regs::NumTransformFeedbackBuffers; i++) {
+ const auto& tf = maxwell3d.regs.transform_feedback;
+ if (tf.buffers[i].enable == 0) {
+ continue;
+ }
+ const size_t stream = tf.controls[i].stream;
+ last_queries_stride[stream] = tf.controls[i].stride;
+ streams_mask |= 1ULL << stream;
+ buffers_count = std::max<size_t>(buffers_count, stream + 1);
+ }
+ });
+ }
+
+ std::pair<size_t, size_t> ProduceCounterBuffer(size_t stream) {
+ if (current_bank == nullptr || current_bank->IsClosed()) {
+ current_bank_id =
+ bank_pool.ReserveBank([this](std::deque<TFBQueryBank>& queue, size_t index) {
+ queue.emplace_back(scheduler, memory_allocator, index);
+ });
+ current_bank = &bank_pool.GetBank(current_bank_id);
+ }
+ auto [dont_care, other] = current_bank->Reserve();
+ const size_t slot = other; // workaround to compile bug.
+ current_bank->AddReference();
+
+ static constexpr VkMemoryBarrier READ_BARRIER{
+ .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
+ .pNext = nullptr,
+ .srcAccessMask = VK_ACCESS_TRANSFORM_FEEDBACK_COUNTER_WRITE_BIT_EXT,
+ .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT,
+ };
+ static constexpr VkMemoryBarrier WRITE_BARRIER{
+ .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
+ .pNext = nullptr,
+ .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
+ .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT,
+ };
+ scheduler.RequestOutsideRenderPassOperationContext();
+ scheduler.Record([dst_buffer = current_bank->GetBuffer(),
+ src_buffer = counter_buffers[stream], src_offset = offsets[stream],
+ slot](vk::CommandBuffer cmdbuf) {
+ cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFORM_FEEDBACK_BIT_EXT,
+ VK_PIPELINE_STAGE_TRANSFER_BIT, 0, READ_BARRIER);
+ std::array<VkBufferCopy, 1> copy{VkBufferCopy{
+ .srcOffset = src_offset,
+ .dstOffset = slot * TFBQueryBank::QUERY_SIZE,
+ .size = TFBQueryBank::QUERY_SIZE,
+ }};
+ cmdbuf.CopyBuffer(src_buffer, dst_buffer, copy);
+ cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
+ 0, WRITE_BARRIER);
+ });
+ return {current_bank_id, slot};
+ }
+
+ friend class PrimitivesSucceededStreamer;
+
+ static constexpr size_t NUM_STREAMS = 4;
+
+ QueryCacheRuntime& runtime;
+ const Device& device;
+ Scheduler& scheduler;
+ const MemoryAllocator& memory_allocator;
+ StagingBufferPool& staging_pool;
+ VideoCommon::BankPool<TFBQueryBank> bank_pool;
+ size_t current_bank_id;
+ TFBQueryBank* current_bank;
+ vk::Buffer counters_buffer;
+
+ // syncing queue
+ std::vector<size_t> pending_sync;
+
+ // flush levels
+ std::vector<size_t> pending_flush_queries;
+ std::deque<StagingBufferRef> download_buffers;
+ std::deque<std::vector<size_t>> pending_flush_sets;
+ std::vector<StagingBufferRef> free_queue;
+ std::mutex flush_guard;
+
+ // state machine
+ bool has_started{};
+ bool has_flushed_end_pending{};
+ size_t buffers_count{};
+ std::array<VkBuffer, NUM_STREAMS> counter_buffers{};
+ std::array<VkDeviceSize, NUM_STREAMS> offsets{};
+ std::array<VAddr, NUM_STREAMS> last_queries;
+ std::array<size_t, NUM_STREAMS> last_queries_stride;
+ Maxwell3D::Regs::PrimitiveTopology out_topology;
+ u64 streams_mask;
+};
+
+class PrimitivesQueryBase : public VideoCommon::QueryBase {
+public:
+ // Default constructor
+ PrimitivesQueryBase()
+ : VideoCommon::QueryBase(0, VideoCommon::QueryFlagBits::IsHostManaged, 0) {}
+
+ // Parameterized constructor
+ PrimitivesQueryBase(bool has_timestamp, VAddr address)
+ : VideoCommon::QueryBase(address, VideoCommon::QueryFlagBits::IsHostManaged, 0) {
+ if (has_timestamp) {
+ flags |= VideoCommon::QueryFlagBits::HasTimestamp;
+ }
+ }
+
+ u64 stride{};
+ VAddr dependant_address{};
+ Maxwell3D::Regs::PrimitiveTopology topology{Maxwell3D::Regs::PrimitiveTopology::Points};
+ size_t dependant_index{};
+ bool dependant_manage{};
+};
+
+class PrimitivesSucceededStreamer : public VideoCommon::SimpleStreamer<PrimitivesQueryBase> {
+public:
+ explicit PrimitivesSucceededStreamer(size_t id_, QueryCacheRuntime& runtime_,
+ TFBCounterStreamer& tfb_streamer_,
+ Core::Memory::Memory& cpu_memory_)
+ : VideoCommon::SimpleStreamer<PrimitivesQueryBase>(id_), runtime{runtime_},
+ tfb_streamer{tfb_streamer_}, cpu_memory{cpu_memory_} {
+ MakeDependent(&tfb_streamer);
+ }
+
+ ~PrimitivesSucceededStreamer() = default;
+
+ size_t WriteCounter(VAddr address, bool has_timestamp, u32 value,
+ std::optional<u32> subreport_) override {
+ auto index = BuildQuery();
+ auto* new_query = GetQuery(index);
+ new_query->guest_address = address;
+ new_query->value = 0;
+ if (has_timestamp) {
+ new_query->flags |= VideoCommon::QueryFlagBits::HasTimestamp;
+ }
+ if (!subreport_) {
+ new_query->flags |= VideoCommon::QueryFlagBits::IsFinalValueSynced;
+ return index;
+ }
+ const size_t subreport = static_cast<size_t>(*subreport_);
+ auto dependant_address_opt = tfb_streamer.GetLastQueryStream(subreport);
+ bool must_manage_dependance = false;
+ new_query->topology = tfb_streamer.GetOutputTopology();
+ if (dependant_address_opt) {
+ auto [dep_address, stride] = *dependant_address_opt;
+ new_query->dependant_address = dep_address;
+ new_query->stride = stride;
+ } else {
+ new_query->dependant_index =
+ tfb_streamer.WriteCounter(address, has_timestamp, value, subreport_);
+ auto* dependant_query = tfb_streamer.GetQuery(new_query->dependant_index);
+ dependant_query->flags |= VideoCommon::QueryFlagBits::IsInvalidated;
+ must_manage_dependance = true;
+ if (True(dependant_query->flags & VideoCommon::QueryFlagBits::IsFinalValueSynced)) {
+ new_query->value = 0;
+ new_query->flags |= VideoCommon::QueryFlagBits::IsFinalValueSynced;
+ if (must_manage_dependance) {
+ tfb_streamer.Free(new_query->dependant_index);
+ }
+ return index;
+ }
+ new_query->stride = 1;
+ runtime.View3DRegs([new_query, subreport](Maxwell3D& maxwell3d) {
+ for (size_t i = 0; i < Maxwell3D::Regs::NumTransformFeedbackBuffers; i++) {
+ const auto& tf = maxwell3d.regs.transform_feedback;
+ if (tf.buffers[i].enable == 0) {
+ continue;
+ }
+ if (tf.controls[i].stream != subreport) {
+ continue;
+ }
+ new_query->stride = tf.controls[i].stride;
+ break;
+ }
+ });
+ }
+
+ new_query->dependant_manage = must_manage_dependance;
+ pending_flush_queries.push_back(index);
+ return index;
+ }
+
+ bool HasUnsyncedQueries() const override {
+ return !pending_flush_queries.empty();
+ }
+
+ void PushUnsyncedQueries() override {
+ std::scoped_lock lk(flush_guard);
+ pending_flush_sets.emplace_back(std::move(pending_flush_queries));
+ pending_flush_queries.clear();
+ }
+
+ void PopUnsyncedQueries() override {
+ std::vector<size_t> flushed_queries;
+ {
+ std::scoped_lock lk(flush_guard);
+ flushed_queries = std::move(pending_flush_sets.front());
+ pending_flush_sets.pop_front();
+ }
+
+ for (auto q : flushed_queries) {
+ auto* query = GetQuery(q);
+ if (True(query->flags & VideoCommon::QueryFlagBits::IsFinalValueSynced)) {
+ continue;
+ }
+
+ query->flags |= VideoCommon::QueryFlagBits::IsFinalValueSynced;
+ u64 num_vertices = 0;
+ if (query->dependant_manage) {
+ auto* dependant_query = tfb_streamer.GetQuery(query->dependant_index);
+ num_vertices = dependant_query->value / query->stride;
+ tfb_streamer.Free(query->dependant_index);
+ } else {
+ u8* pointer = cpu_memory.GetPointer(query->dependant_address);
+ u32 result;
+ std::memcpy(&result, pointer, sizeof(u32));
+ num_vertices = static_cast<u64>(result) / query->stride;
+ }
+ query->value = [&]() -> u64 {
+ switch (query->topology) {
+ case Maxwell3D::Regs::PrimitiveTopology::Points:
+ return num_vertices;
+ case Maxwell3D::Regs::PrimitiveTopology::Lines:
+ return num_vertices / 2;
+ case Maxwell3D::Regs::PrimitiveTopology::LineLoop:
+ return (num_vertices / 2) + 1;
+ case Maxwell3D::Regs::PrimitiveTopology::LineStrip:
+ return num_vertices - 1;
+ case Maxwell3D::Regs::PrimitiveTopology::Patches:
+ case Maxwell3D::Regs::PrimitiveTopology::Triangles:
+ case Maxwell3D::Regs::PrimitiveTopology::TrianglesAdjacency:
+ return num_vertices / 3;
+ case Maxwell3D::Regs::PrimitiveTopology::TriangleFan:
+ case Maxwell3D::Regs::PrimitiveTopology::TriangleStrip:
+ case Maxwell3D::Regs::PrimitiveTopology::TriangleStripAdjacency:
+ return num_vertices - 2;
+ case Maxwell3D::Regs::PrimitiveTopology::Quads:
+ return num_vertices / 4;
+ case Maxwell3D::Regs::PrimitiveTopology::Polygon:
+ return 1U;
+ default:
+ return num_vertices;
+ }
+ }();
+ }
+ }
+
+private:
+ QueryCacheRuntime& runtime;
+ TFBCounterStreamer& tfb_streamer;
+ Core::Memory::Memory& cpu_memory;
+
+ // syncing queue
+ std::vector<size_t> pending_sync;
+
+ // flush levels
+ std::vector<size_t> pending_flush_queries;
+ std::deque<std::vector<size_t>> pending_flush_sets;
+ std::mutex flush_guard;
+};
+
+} // namespace
+
+struct QueryCacheRuntimeImpl {
+ QueryCacheRuntimeImpl(QueryCacheRuntime& runtime, VideoCore::RasterizerInterface* rasterizer_,
+ Core::Memory::Memory& cpu_memory_, Vulkan::BufferCache& buffer_cache_,
+ const Device& device_, const MemoryAllocator& memory_allocator_,
+ Scheduler& scheduler_, StagingBufferPool& staging_pool_,
+ ComputePassDescriptorQueue& compute_pass_descriptor_queue,
+ DescriptorPool& descriptor_pool)
+ : rasterizer{rasterizer_}, cpu_memory{cpu_memory_},
+ buffer_cache{buffer_cache_}, device{device_},
+ memory_allocator{memory_allocator_}, scheduler{scheduler_}, staging_pool{staging_pool_},
+ guest_streamer(0, runtime),
+ sample_streamer(static_cast<size_t>(QueryType::ZPassPixelCount64), runtime, rasterizer,
+ device, scheduler, memory_allocator, compute_pass_descriptor_queue,
+ descriptor_pool),
+ tfb_streamer(static_cast<size_t>(QueryType::StreamingByteCount), runtime, device,
+ scheduler, memory_allocator, staging_pool),
+ primitives_succeeded_streamer(
+ static_cast<size_t>(QueryType::StreamingPrimitivesSucceeded), runtime, tfb_streamer,
+ cpu_memory_),
+ primitives_needed_minus_suceeded_streamer(
+ static_cast<size_t>(QueryType::StreamingPrimitivesNeededMinusSucceeded), runtime, 0u),
+ hcr_setup{}, hcr_is_set{}, is_hcr_running{}, maxwell3d{} {
+
+ hcr_setup.sType = VK_STRUCTURE_TYPE_CONDITIONAL_RENDERING_BEGIN_INFO_EXT;
+ hcr_setup.pNext = nullptr;
+ hcr_setup.flags = 0;
+
+ conditional_resolve_pass = std::make_unique<ConditionalRenderingResolvePass>(
+ device, scheduler, descriptor_pool, compute_pass_descriptor_queue);
+
+ const VkBufferCreateInfo buffer_ci = {
+ .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
+ .pNext = nullptr,
+ .flags = 0,
+ .size = sizeof(u32),
+ .usage = VK_BUFFER_USAGE_TRANSFER_DST_BIT | VK_BUFFER_USAGE_STORAGE_BUFFER_BIT |
+ VK_BUFFER_USAGE_CONDITIONAL_RENDERING_BIT_EXT,
+ .sharingMode = VK_SHARING_MODE_EXCLUSIVE,
+ .queueFamilyIndexCount = 0,
+ .pQueueFamilyIndices = nullptr,
+ };
+ hcr_resolve_buffer = memory_allocator.CreateBuffer(buffer_ci, MemoryUsage::DeviceLocal);
+ }
+
+ VideoCore::RasterizerInterface* rasterizer;
+ Core::Memory::Memory& cpu_memory;
+ Vulkan::BufferCache& buffer_cache;
+
+ const Device& device;
+ const MemoryAllocator& memory_allocator;
+ Scheduler& scheduler;
+ StagingBufferPool& staging_pool;
+
+ // Streamers
+ VideoCommon::GuestStreamer<QueryCacheParams> guest_streamer;
+ SamplesStreamer sample_streamer;
+ TFBCounterStreamer tfb_streamer;
+ PrimitivesSucceededStreamer primitives_succeeded_streamer;
+ VideoCommon::StubStreamer<QueryCacheParams> primitives_needed_minus_suceeded_streamer;
+
+ std::vector<std::pair<VAddr, VAddr>> little_cache;
+ std::vector<std::pair<VkBuffer, VkDeviceSize>> buffers_to_upload_to;
+ std::vector<size_t> redirect_cache;
+ std::vector<std::vector<VkBufferCopy>> copies_setup;
+
+ // Host conditional rendering data
+ std::unique_ptr<ConditionalRenderingResolvePass> conditional_resolve_pass;
+ vk::Buffer hcr_resolve_buffer;
+ VkConditionalRenderingBeginInfoEXT hcr_setup;
+ VkBuffer hcr_buffer;
+ size_t hcr_offset;
+ bool hcr_is_set;
+ bool is_hcr_running;
+
+ // maxwell3d
+ Maxwell3D* maxwell3d;
+};
+
+QueryCacheRuntime::QueryCacheRuntime(VideoCore::RasterizerInterface* rasterizer,
+ Core::Memory::Memory& cpu_memory_,
+ Vulkan::BufferCache& buffer_cache_, const Device& device_,
+ const MemoryAllocator& memory_allocator_,
+ Scheduler& scheduler_, StagingBufferPool& staging_pool_,
+ ComputePassDescriptorQueue& compute_pass_descriptor_queue,
+ DescriptorPool& descriptor_pool) {
+ impl = std::make_unique<QueryCacheRuntimeImpl>(
+ *this, rasterizer, cpu_memory_, buffer_cache_, device_, memory_allocator_, scheduler_,
+ staging_pool_, compute_pass_descriptor_queue, descriptor_pool);
}
-void QueryPool::Allocate(std::size_t begin, std::size_t end) {
- usage.resize(end);
+void QueryCacheRuntime::Bind3DEngine(Maxwell3D* maxwell3d) {
+ impl->maxwell3d = maxwell3d;
+}
- pools.push_back(device.GetLogical().CreateQueryPool({
- .sType = VK_STRUCTURE_TYPE_QUERY_POOL_CREATE_INFO,
- .pNext = nullptr,
- .flags = 0,
- .queryType = GetTarget(type),
- .queryCount = static_cast<u32>(end - begin),
- .pipelineStatistics = 0,
- }));
+template <typename Func>
+void QueryCacheRuntime::View3DRegs(Func&& func) {
+ if (impl->maxwell3d) {
+ func(*impl->maxwell3d);
+ }
+}
+
+void QueryCacheRuntime::EndHostConditionalRendering() {
+ PauseHostConditionalRendering();
+ impl->hcr_is_set = false;
+ impl->is_hcr_running = false;
+ impl->hcr_buffer = nullptr;
+ impl->hcr_offset = 0;
+}
+
+void QueryCacheRuntime::PauseHostConditionalRendering() {
+ if (!impl->hcr_is_set) {
+ return;
+ }
+ if (impl->is_hcr_running) {
+ impl->scheduler.Record(
+ [](vk::CommandBuffer cmdbuf) { cmdbuf.EndConditionalRenderingEXT(); });
+ }
+ impl->is_hcr_running = false;
}
-void QueryPool::Reserve(std::pair<VkQueryPool, u32> query) {
- const auto it =
- std::find_if(pools.begin(), pools.end(), [query_pool = query.first](vk::QueryPool& pool) {
- return query_pool == *pool;
+void QueryCacheRuntime::ResumeHostConditionalRendering() {
+ if (!impl->hcr_is_set) {
+ return;
+ }
+ if (!impl->is_hcr_running) {
+ impl->scheduler.Record([hcr_setup = impl->hcr_setup](vk::CommandBuffer cmdbuf) {
+ cmdbuf.BeginConditionalRenderingEXT(hcr_setup);
});
+ }
+ impl->is_hcr_running = true;
+}
- if (it != std::end(pools)) {
- const std::ptrdiff_t pool_index = std::distance(std::begin(pools), it);
- usage[pool_index * GROW_STEP + static_cast<std::ptrdiff_t>(query.second)] = false;
+void QueryCacheRuntime::HostConditionalRenderingCompareValueImpl(VideoCommon::LookupData object,
+ bool is_equal) {
+ {
+ std::scoped_lock lk(impl->buffer_cache.mutex);
+ static constexpr auto sync_info = VideoCommon::ObtainBufferSynchronize::FullSynchronize;
+ const auto post_op = VideoCommon::ObtainBufferOperation::DoNothing;
+ const auto [buffer, offset] =
+ impl->buffer_cache.ObtainCPUBuffer(object.address, 8, sync_info, post_op);
+ impl->hcr_buffer = buffer->Handle();
+ impl->hcr_offset = offset;
+ }
+ if (impl->hcr_is_set) {
+ if (impl->hcr_setup.buffer == impl->hcr_buffer &&
+ impl->hcr_setup.offset == impl->hcr_offset) {
+ ResumeHostConditionalRendering();
+ return;
+ }
+ PauseHostConditionalRendering();
}
+ impl->hcr_setup.buffer = impl->hcr_buffer;
+ impl->hcr_setup.offset = impl->hcr_offset;
+ impl->hcr_setup.flags = is_equal ? VK_CONDITIONAL_RENDERING_INVERTED_BIT_EXT : 0;
+ impl->hcr_is_set = true;
+ impl->is_hcr_running = false;
+ ResumeHostConditionalRendering();
}
-QueryCache::QueryCache(VideoCore::RasterizerInterface& rasterizer_,
- Core::Memory::Memory& cpu_memory_, const Device& device_,
- Scheduler& scheduler_)
- : QueryCacheBase{rasterizer_, cpu_memory_}, device{device_}, scheduler{scheduler_},
- query_pools{
- QueryPool{device_, scheduler_, QueryType::SamplesPassed},
- } {}
-
-QueryCache::~QueryCache() {
- // TODO(Rodrigo): This is a hack to destroy all HostCounter instances before the base class
- // destructor is called. The query cache should be redesigned to have a proper ownership model
- // instead of using shared pointers.
- for (size_t query_type = 0; query_type < VideoCore::NumQueryTypes; ++query_type) {
- auto& stream = Stream(static_cast<QueryType>(query_type));
- stream.Update(false);
- stream.Reset();
+void QueryCacheRuntime::HostConditionalRenderingCompareBCImpl(VAddr address, bool is_equal) {
+ VkBuffer to_resolve;
+ u32 to_resolve_offset;
+ {
+ std::scoped_lock lk(impl->buffer_cache.mutex);
+ static constexpr auto sync_info = VideoCommon::ObtainBufferSynchronize::NoSynchronize;
+ const auto post_op = VideoCommon::ObtainBufferOperation::DoNothing;
+ const auto [buffer, offset] =
+ impl->buffer_cache.ObtainCPUBuffer(address, 24, sync_info, post_op);
+ to_resolve = buffer->Handle();
+ to_resolve_offset = static_cast<u32>(offset);
}
+ if (impl->is_hcr_running) {
+ PauseHostConditionalRendering();
+ }
+ impl->conditional_resolve_pass->Resolve(*impl->hcr_resolve_buffer, to_resolve,
+ to_resolve_offset, false);
+ impl->hcr_setup.buffer = *impl->hcr_resolve_buffer;
+ impl->hcr_setup.offset = 0;
+ impl->hcr_setup.flags = is_equal ? 0 : VK_CONDITIONAL_RENDERING_INVERTED_BIT_EXT;
+ impl->hcr_is_set = true;
+ impl->is_hcr_running = false;
+ ResumeHostConditionalRendering();
}
-std::pair<VkQueryPool, u32> QueryCache::AllocateQuery(QueryType type) {
- return query_pools[static_cast<std::size_t>(type)].Commit();
+bool QueryCacheRuntime::HostConditionalRenderingCompareValue(VideoCommon::LookupData object_1,
+ [[maybe_unused]] bool qc_dirty) {
+ if (!impl->device.IsExtConditionalRendering()) {
+ return false;
+ }
+ HostConditionalRenderingCompareValueImpl(object_1, false);
+ return true;
}
-void QueryCache::Reserve(QueryType type, std::pair<VkQueryPool, u32> query) {
- query_pools[static_cast<std::size_t>(type)].Reserve(query);
+bool QueryCacheRuntime::HostConditionalRenderingCompareValues(VideoCommon::LookupData object_1,
+ VideoCommon::LookupData object_2,
+ bool qc_dirty, bool equal_check) {
+ if (!impl->device.IsExtConditionalRendering()) {
+ return false;
+ }
+
+ const auto check_in_bc = [&](VAddr address) {
+ return impl->buffer_cache.IsRegionGpuModified(address, 8);
+ };
+ const auto check_value = [&](VAddr address) {
+ u8* ptr = impl->cpu_memory.GetPointer(address);
+ u64 value{};
+ std::memcpy(&value, ptr, sizeof(value));
+ return value == 0;
+ };
+ std::array<VideoCommon::LookupData*, 2> objects{&object_1, &object_2};
+ std::array<bool, 2> is_in_bc{};
+ std::array<bool, 2> is_in_qc{};
+ std::array<bool, 2> is_in_ac{};
+ std::array<bool, 2> is_null{};
+ {
+ std::scoped_lock lk(impl->buffer_cache.mutex);
+ for (size_t i = 0; i < 2; i++) {
+ is_in_qc[i] = objects[i]->found_query != nullptr;
+ is_in_bc[i] = !is_in_qc[i] && check_in_bc(objects[i]->address);
+ is_in_ac[i] = is_in_qc[i] || is_in_bc[i];
+ }
+ }
+
+ if (!is_in_ac[0] && !is_in_ac[1]) {
+ EndHostConditionalRendering();
+ return false;
+ }
+
+ if (!qc_dirty && !is_in_bc[0] && !is_in_bc[1]) {
+ EndHostConditionalRendering();
+ return false;
+ }
+
+ const bool is_gpu_high = Settings::IsGPULevelHigh();
+ if (!is_gpu_high && impl->device.GetDriverID() == VK_DRIVER_ID_INTEL_PROPRIETARY_WINDOWS) {
+ return true;
+ }
+
+ for (size_t i = 0; i < 2; i++) {
+ is_null[i] = !is_in_ac[i] && check_value(objects[i]->address);
+ }
+
+ for (size_t i = 0; i < 2; i++) {
+ if (is_null[i]) {
+ size_t j = (i + 1) % 2;
+ HostConditionalRenderingCompareValueImpl(*objects[j], equal_check);
+ return true;
+ }
+ }
+
+ if (!is_gpu_high) {
+ return true;
+ }
+
+ if (!is_in_bc[0] && !is_in_bc[1]) {
+ // Both queries are in query cache, it's best to just flush.
+ return true;
+ }
+ HostConditionalRenderingCompareBCImpl(object_1.address, equal_check);
+ return true;
}
-HostCounter::HostCounter(QueryCache& cache_, std::shared_ptr<HostCounter> dependency_,
- QueryType type_)
- : HostCounterBase{std::move(dependency_)}, cache{cache_}, type{type_},
- query{cache_.AllocateQuery(type_)}, tick{cache_.GetScheduler().CurrentTick()} {
- const vk::Device* logical = &cache.GetDevice().GetLogical();
- cache.GetScheduler().Record([logical, query_ = query](vk::CommandBuffer cmdbuf) {
- const bool use_precise = Settings::IsGPULevelHigh();
- logical->ResetQueryPool(query_.first, query_.second, 1);
- cmdbuf.BeginQuery(query_.first, query_.second,
- use_precise ? VK_QUERY_CONTROL_PRECISE_BIT : 0);
- });
+QueryCacheRuntime::~QueryCacheRuntime() = default;
+
+VideoCommon::StreamerInterface* QueryCacheRuntime::GetStreamerInterface(QueryType query_type) {
+ switch (query_type) {
+ case QueryType::Payload:
+ return &impl->guest_streamer;
+ case QueryType::ZPassPixelCount64:
+ return &impl->sample_streamer;
+ case QueryType::StreamingByteCount:
+ return &impl->tfb_streamer;
+ case QueryType::StreamingPrimitivesNeeded:
+ case QueryType::VtgPrimitivesOut:
+ case QueryType::StreamingPrimitivesSucceeded:
+ return &impl->primitives_succeeded_streamer;
+ case QueryType::StreamingPrimitivesNeededMinusSucceeded:
+ return &impl->primitives_needed_minus_suceeded_streamer;
+ default:
+ return nullptr;
+ }
}
-HostCounter::~HostCounter() {
- cache.Reserve(type, query);
+void QueryCacheRuntime::Barriers(bool is_prebarrier) {
+ static constexpr VkMemoryBarrier READ_BARRIER{
+ .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
+ .pNext = nullptr,
+ .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT,
+ .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT,
+ };
+ static constexpr VkMemoryBarrier WRITE_BARRIER{
+ .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
+ .pNext = nullptr,
+ .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
+ .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT,
+ };
+ if (is_prebarrier) {
+ impl->scheduler.Record([](vk::CommandBuffer cmdbuf) {
+ cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
+ VK_PIPELINE_STAGE_TRANSFER_BIT, 0, READ_BARRIER);
+ });
+ } else {
+ impl->scheduler.Record([](vk::CommandBuffer cmdbuf) {
+ cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT,
+ VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, WRITE_BARRIER);
+ });
+ }
}
-void HostCounter::EndQuery() {
- cache.GetScheduler().Record([query_ = query](vk::CommandBuffer cmdbuf) {
- cmdbuf.EndQuery(query_.first, query_.second);
+template <typename SyncValuesType>
+void QueryCacheRuntime::SyncValues(std::span<SyncValuesType> values, VkBuffer base_src_buffer) {
+ if (values.size() == 0) {
+ return;
+ }
+ impl->redirect_cache.clear();
+ impl->little_cache.clear();
+ size_t total_size = 0;
+ for (auto& sync_val : values) {
+ total_size += sync_val.size;
+ bool found = false;
+ VAddr base = Common::AlignDown(sync_val.address, Core::Memory::YUZU_PAGESIZE);
+ VAddr base_end = base + Core::Memory::YUZU_PAGESIZE;
+ for (size_t i = 0; i < impl->little_cache.size(); i++) {
+ const auto set_found = [&] {
+ impl->redirect_cache.push_back(i);
+ found = true;
+ };
+ auto& loc = impl->little_cache[i];
+ if (base < loc.second && loc.first < base_end) {
+ set_found();
+ break;
+ }
+ if (loc.first == base_end) {
+ loc.first = base;
+ set_found();
+ break;
+ }
+ if (loc.second == base) {
+ loc.second = base_end;
+ set_found();
+ break;
+ }
+ }
+ if (!found) {
+ impl->redirect_cache.push_back(impl->little_cache.size());
+ impl->little_cache.emplace_back(base, base_end);
+ }
+ }
+
+ // Vulkan part.
+ std::scoped_lock lk(impl->buffer_cache.mutex);
+ impl->buffer_cache.BufferOperations([&] {
+ impl->buffers_to_upload_to.clear();
+ for (auto& pair : impl->little_cache) {
+ static constexpr auto sync_info = VideoCommon::ObtainBufferSynchronize::FullSynchronize;
+ const auto post_op = VideoCommon::ObtainBufferOperation::DoNothing;
+ const auto [buffer, offset] = impl->buffer_cache.ObtainCPUBuffer(
+ pair.first, static_cast<u32>(pair.second - pair.first), sync_info, post_op);
+ impl->buffers_to_upload_to.emplace_back(buffer->Handle(), offset);
+ }
});
-}
-u64 HostCounter::BlockingQuery(bool async) const {
- if (!async) {
- cache.GetScheduler().Wait(tick);
- }
- u64 data;
- const VkResult query_result = cache.GetDevice().GetLogical().GetQueryResults(
- query.first, query.second, 1, sizeof(data), &data, sizeof(data),
- VK_QUERY_RESULT_64_BIT | VK_QUERY_RESULT_WAIT_BIT);
-
- switch (query_result) {
- case VK_SUCCESS:
- return data;
- case VK_ERROR_DEVICE_LOST:
- cache.GetDevice().ReportLoss();
- [[fallthrough]];
- default:
- throw vk::Exception(query_result);
+ VkBuffer src_buffer;
+ [[maybe_unused]] StagingBufferRef ref;
+ impl->copies_setup.clear();
+ impl->copies_setup.resize(impl->little_cache.size());
+ if constexpr (SyncValuesType::GeneratesBaseBuffer) {
+ ref = impl->staging_pool.Request(total_size, MemoryUsage::Upload);
+ size_t current_offset = ref.offset;
+ size_t accumulated_size = 0;
+ for (size_t i = 0; i < values.size(); i++) {
+ size_t which_copy = impl->redirect_cache[i];
+ impl->copies_setup[which_copy].emplace_back(VkBufferCopy{
+ .srcOffset = current_offset + accumulated_size,
+ .dstOffset = impl->buffers_to_upload_to[which_copy].second + values[i].address -
+ impl->little_cache[which_copy].first,
+ .size = values[i].size,
+ });
+ std::memcpy(ref.mapped_span.data() + accumulated_size, &values[i].value,
+ values[i].size);
+ accumulated_size += values[i].size;
+ }
+ src_buffer = ref.buffer;
+ } else {
+ for (size_t i = 0; i < values.size(); i++) {
+ size_t which_copy = impl->redirect_cache[i];
+ impl->copies_setup[which_copy].emplace_back(VkBufferCopy{
+ .srcOffset = values[i].offset,
+ .dstOffset = impl->buffers_to_upload_to[which_copy].second + values[i].address -
+ impl->little_cache[which_copy].first,
+ .size = values[i].size,
+ });
+ }
+ src_buffer = base_src_buffer;
}
+
+ impl->scheduler.RequestOutsideRenderPassOperationContext();
+ impl->scheduler.Record([src_buffer, dst_buffers = std::move(impl->buffers_to_upload_to),
+ vk_copies = std::move(impl->copies_setup)](vk::CommandBuffer cmdbuf) {
+ size_t size = dst_buffers.size();
+ for (size_t i = 0; i < size; i++) {
+ cmdbuf.CopyBuffer(src_buffer, dst_buffers[i].first, vk_copies[i]);
+ }
+ });
}
} // namespace Vulkan
+
+namespace VideoCommon {
+
+template class QueryCacheBase<Vulkan::QueryCacheParams>;
+
+} // namespace VideoCommon
diff --git a/src/video_core/renderer_vulkan/vk_query_cache.h b/src/video_core/renderer_vulkan/vk_query_cache.h
index c1b9552eb..e9a1ea169 100644
--- a/src/video_core/renderer_vulkan/vk_query_cache.h
+++ b/src/video_core/renderer_vulkan/vk_query_cache.h
@@ -1,101 +1,75 @@
-// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
-// SPDX-License-Identifier: GPL-2.0-or-later
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
#pragma once
-#include <cstddef>
#include <memory>
-#include <utility>
-#include <vector>
-#include "common/common_types.h"
-#include "video_core/query_cache.h"
-#include "video_core/renderer_vulkan/vk_resource_pool.h"
-#include "video_core/vulkan_common/vulkan_wrapper.h"
+#include "video_core/query_cache/query_cache_base.h"
+#include "video_core/renderer_vulkan/vk_buffer_cache.h"
namespace VideoCore {
class RasterizerInterface;
}
+namespace VideoCommon {
+class StreamerInterface;
+}
+
namespace Vulkan {
-class CachedQuery;
class Device;
-class HostCounter;
-class QueryCache;
class Scheduler;
+class StagingBufferPool;
-using CounterStream = VideoCommon::CounterStreamBase<QueryCache, HostCounter>;
+struct QueryCacheRuntimeImpl;
-class QueryPool final : public ResourcePool {
+class QueryCacheRuntime {
public:
- explicit QueryPool(const Device& device, Scheduler& scheduler, VideoCore::QueryType type);
- ~QueryPool() override;
+ explicit QueryCacheRuntime(VideoCore::RasterizerInterface* rasterizer,
+ Core::Memory::Memory& cpu_memory_,
+ Vulkan::BufferCache& buffer_cache_, const Device& device_,
+ const MemoryAllocator& memory_allocator_, Scheduler& scheduler_,
+ StagingBufferPool& staging_pool_,
+ ComputePassDescriptorQueue& compute_pass_descriptor_queue,
+ DescriptorPool& descriptor_pool);
+ ~QueryCacheRuntime();
- std::pair<VkQueryPool, u32> Commit();
+ template <typename SyncValuesType>
+ void SyncValues(std::span<SyncValuesType> values, VkBuffer base_src_buffer = nullptr);
- void Reserve(std::pair<VkQueryPool, u32> query);
+ void Barriers(bool is_prebarrier);
-protected:
- void Allocate(std::size_t begin, std::size_t end) override;
+ void EndHostConditionalRendering();
-private:
- static constexpr std::size_t GROW_STEP = 512;
+ void PauseHostConditionalRendering();
- const Device& device;
- const VideoCore::QueryType type;
+ void ResumeHostConditionalRendering();
- std::vector<vk::QueryPool> pools;
- std::vector<bool> usage;
-};
+ bool HostConditionalRenderingCompareValue(VideoCommon::LookupData object_1, bool qc_dirty);
-class QueryCache final
- : public VideoCommon::QueryCacheBase<QueryCache, CachedQuery, CounterStream, HostCounter> {
-public:
- explicit QueryCache(VideoCore::RasterizerInterface& rasterizer_,
- Core::Memory::Memory& cpu_memory_, const Device& device_,
- Scheduler& scheduler_);
- ~QueryCache();
-
- std::pair<VkQueryPool, u32> AllocateQuery(VideoCore::QueryType type);
+ bool HostConditionalRenderingCompareValues(VideoCommon::LookupData object_1,
+ VideoCommon::LookupData object_2, bool qc_dirty,
+ bool equal_check);
- void Reserve(VideoCore::QueryType type, std::pair<VkQueryPool, u32> query);
+ VideoCommon::StreamerInterface* GetStreamerInterface(VideoCommon::QueryType query_type);
- const Device& GetDevice() const noexcept {
- return device;
- }
+ void Bind3DEngine(Tegra::Engines::Maxwell3D* maxwell3d);
- Scheduler& GetScheduler() const noexcept {
- return scheduler;
- }
+ template <typename Func>
+ void View3DRegs(Func&& func);
private:
- const Device& device;
- Scheduler& scheduler;
- std::array<QueryPool, VideoCore::NumQueryTypes> query_pools;
+ void HostConditionalRenderingCompareValueImpl(VideoCommon::LookupData object, bool is_equal);
+ void HostConditionalRenderingCompareBCImpl(VAddr address, bool is_equal);
+ friend struct QueryCacheRuntimeImpl;
+ std::unique_ptr<QueryCacheRuntimeImpl> impl;
};
-class HostCounter final : public VideoCommon::HostCounterBase<QueryCache, HostCounter> {
-public:
- explicit HostCounter(QueryCache& cache_, std::shared_ptr<HostCounter> dependency_,
- VideoCore::QueryType type_);
- ~HostCounter();
-
- void EndQuery();
-
-private:
- u64 BlockingQuery(bool async = false) const override;
-
- QueryCache& cache;
- const VideoCore::QueryType type;
- const std::pair<VkQueryPool, u32> query;
- const u64 tick;
+struct QueryCacheParams {
+ using RuntimeType = typename Vulkan::QueryCacheRuntime;
};
-class CachedQuery : public VideoCommon::CachedQueryBase<HostCounter> {
-public:
- explicit CachedQuery(QueryCache&, VideoCore::QueryType, VAddr cpu_addr_, u8* host_ptr_)
- : CachedQueryBase{cpu_addr_, host_ptr_} {}
-};
+using QueryCache = VideoCommon::QueryCacheBase<QueryCacheParams>;
} // namespace Vulkan
diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp
index 2fbfc14cc..83f2b6045 100644
--- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp
+++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp
@@ -24,6 +24,7 @@
#include "video_core/renderer_vulkan/vk_compute_pipeline.h"
#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
#include "video_core/renderer_vulkan/vk_pipeline_cache.h"
+#include "video_core/renderer_vulkan/vk_query_cache.h"
#include "video_core/renderer_vulkan/vk_rasterizer.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/vk_staging_buffer_pool.h"
@@ -170,9 +171,11 @@ RasterizerVulkan::RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra
buffer_cache_runtime(device, memory_allocator, scheduler, staging_pool,
guest_descriptor_queue, compute_pass_descriptor_queue, descriptor_pool),
buffer_cache(*this, cpu_memory_, buffer_cache_runtime),
+ query_cache_runtime(this, cpu_memory_, buffer_cache, device, memory_allocator, scheduler,
+ staging_pool, compute_pass_descriptor_queue, descriptor_pool),
+ query_cache(gpu, *this, cpu_memory_, query_cache_runtime),
pipeline_cache(*this, device, scheduler, descriptor_pool, guest_descriptor_queue,
render_pass_cache, buffer_cache, texture_cache, gpu.ShaderNotify()),
- query_cache{*this, cpu_memory_, device, scheduler},
accelerate_dma(buffer_cache, texture_cache, scheduler),
fence_manager(*this, gpu, texture_cache, buffer_cache, query_cache, device, scheduler),
wfi_event(device.GetLogical().CreateEvent()) {
@@ -189,14 +192,7 @@ void RasterizerVulkan::PrepareDraw(bool is_indexed, Func&& draw_func) {
FlushWork();
gpu_memory->FlushCaching();
-#if ANDROID
- if (Settings::IsGPULevelHigh()) {
- // This is problematic on Android, disable on GPU Normal.
- query_cache.UpdateCounters();
- }
-#else
- query_cache.UpdateCounters();
-#endif
+ query_cache.NotifySegment(true);
GraphicsPipeline* const pipeline{pipeline_cache.CurrentGraphicsPipeline()};
if (!pipeline) {
@@ -207,13 +203,12 @@ void RasterizerVulkan::PrepareDraw(bool is_indexed, Func&& draw_func) {
pipeline->SetEngine(maxwell3d, gpu_memory);
pipeline->Configure(is_indexed);
- BeginTransformFeedback();
-
UpdateDynamicStates();
+ HandleTransformFeedback();
+ query_cache.CounterEnable(VideoCommon::QueryType::ZPassPixelCount64,
+ maxwell3d->regs.zpass_pixel_count_enable);
draw_func();
-
- EndTransformFeedback();
}
void RasterizerVulkan::Draw(bool is_indexed, u32 instance_count) {
@@ -241,6 +236,14 @@ void RasterizerVulkan::DrawIndirect() {
const auto indirect_buffer = buffer_cache.GetDrawIndirectBuffer();
const auto& buffer = indirect_buffer.first;
const auto& offset = indirect_buffer.second;
+ if (params.is_byte_count) {
+ scheduler.Record([buffer_obj = buffer->Handle(), offset,
+ stride = params.stride](vk::CommandBuffer cmdbuf) {
+ cmdbuf.DrawIndirectByteCountEXT(1, 0, buffer_obj, offset, 0,
+ static_cast<u32>(stride));
+ });
+ return;
+ }
if (params.include_count) {
const auto count = buffer_cache.GetDrawIndirectCount();
const auto& draw_buffer = count.first;
@@ -280,20 +283,15 @@ void RasterizerVulkan::DrawTexture() {
SCOPE_EXIT({ gpu.TickWork(); });
FlushWork();
-#if ANDROID
- if (Settings::IsGPULevelHigh()) {
- // This is problematic on Android, disable on GPU Normal.
- query_cache.UpdateCounters();
- }
-#else
- query_cache.UpdateCounters();
-#endif
+ query_cache.NotifySegment(true);
texture_cache.SynchronizeGraphicsDescriptors();
texture_cache.UpdateRenderTargets(false);
UpdateDynamicStates();
+ query_cache.CounterEnable(VideoCommon::QueryType::ZPassPixelCount64,
+ maxwell3d->regs.zpass_pixel_count_enable);
const auto& draw_texture_state = maxwell3d->draw_manager->GetDrawTextureState();
const auto& sampler = texture_cache.GetGraphicsSampler(draw_texture_state.src_sampler);
const auto& texture = texture_cache.GetImageView(draw_texture_state.src_texture);
@@ -316,14 +314,9 @@ void RasterizerVulkan::Clear(u32 layer_count) {
FlushWork();
gpu_memory->FlushCaching();
-#if ANDROID
- if (Settings::IsGPULevelHigh()) {
- // This is problematic on Android, disable on GPU Normal.
- query_cache.UpdateCounters();
- }
-#else
- query_cache.UpdateCounters();
-#endif
+ query_cache.NotifySegment(true);
+ query_cache.CounterEnable(VideoCommon::QueryType::ZPassPixelCount64,
+ maxwell3d->regs.zpass_pixel_count_enable);
auto& regs = maxwell3d->regs;
const bool use_color = regs.clear_surface.R || regs.clear_surface.G || regs.clear_surface.B ||
@@ -483,13 +476,13 @@ void RasterizerVulkan::DispatchCompute() {
scheduler.Record([dim](vk::CommandBuffer cmdbuf) { cmdbuf.Dispatch(dim[0], dim[1], dim[2]); });
}
-void RasterizerVulkan::ResetCounter(VideoCore::QueryType type) {
- query_cache.ResetCounter(type);
+void RasterizerVulkan::ResetCounter(VideoCommon::QueryType type) {
+ query_cache.CounterReset(type);
}
-void RasterizerVulkan::Query(GPUVAddr gpu_addr, VideoCore::QueryType type,
- std::optional<u64> timestamp) {
- query_cache.Query(gpu_addr, type, timestamp);
+void RasterizerVulkan::Query(GPUVAddr gpu_addr, VideoCommon::QueryType type,
+ VideoCommon::QueryPropertiesFlags flags, u32 payload, u32 subreport) {
+ query_cache.CounterReport(gpu_addr, type, flags, payload, subreport);
}
void RasterizerVulkan::BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr gpu_addr,
@@ -670,8 +663,8 @@ void RasterizerVulkan::SignalReference() {
fence_manager.SignalReference();
}
-void RasterizerVulkan::ReleaseFences() {
- fence_manager.WaitPendingFences();
+void RasterizerVulkan::ReleaseFences(bool force) {
+ fence_manager.WaitPendingFences(force);
}
void RasterizerVulkan::FlushAndInvalidateRegion(VAddr addr, u64 size,
@@ -695,6 +688,8 @@ void RasterizerVulkan::WaitForIdle() {
flags |= VK_PIPELINE_STAGE_TRANSFORM_FEEDBACK_BIT_EXT;
}
+ query_cache.NotifyWFI();
+
scheduler.RequestOutsideRenderPassOperationContext();
scheduler.Record([event = *wfi_event, flags](vk::CommandBuffer cmdbuf) {
cmdbuf.SetEvent(event, flags);
@@ -738,19 +733,7 @@ void RasterizerVulkan::TickFrame() {
bool RasterizerVulkan::AccelerateConditionalRendering() {
gpu_memory->FlushCaching();
- if (Settings::IsGPULevelHigh()) {
- // TODO(Blinkhawk): Reimplement Host conditional rendering.
- return false;
- }
- // Medium / Low Hack: stub any checks on queries written into the buffer cache.
- const GPUVAddr condition_address{maxwell3d->regs.render_enable.Address()};
- Maxwell::ReportSemaphore::Compare cmp;
- if (gpu_memory->IsMemoryDirty(condition_address, sizeof(cmp),
- VideoCommon::CacheType::BufferCache |
- VideoCommon::CacheType::QueryCache)) {
- return true;
- }
- return false;
+ return query_cache.AccelerateHostConditionalRendering();
}
bool RasterizerVulkan::AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Surface& src,
@@ -796,6 +779,7 @@ bool RasterizerVulkan::AccelerateDisplay(const Tegra::FramebufferConfig& config,
if (!image_view) {
return false;
}
+ query_cache.NotifySegment(false);
screen_info.image = image_view->ImageHandle();
screen_info.image_view = image_view->Handle(Shader::TextureType::Color2D);
screen_info.width = image_view->size.width;
@@ -934,31 +918,18 @@ void RasterizerVulkan::UpdateDynamicStates() {
}
}
-void RasterizerVulkan::BeginTransformFeedback() {
+void RasterizerVulkan::HandleTransformFeedback() {
const auto& regs = maxwell3d->regs;
- if (regs.transform_feedback_enabled == 0) {
- return;
- }
if (!device.IsExtTransformFeedbackSupported()) {
LOG_ERROR(Render_Vulkan, "Transform feedbacks used but not supported");
return;
}
- UNIMPLEMENTED_IF(regs.IsShaderConfigEnabled(Maxwell::ShaderType::TessellationInit) ||
- regs.IsShaderConfigEnabled(Maxwell::ShaderType::Tessellation));
- scheduler.Record(
- [](vk::CommandBuffer cmdbuf) { cmdbuf.BeginTransformFeedbackEXT(0, 0, nullptr, nullptr); });
-}
-
-void RasterizerVulkan::EndTransformFeedback() {
- const auto& regs = maxwell3d->regs;
- if (regs.transform_feedback_enabled == 0) {
- return;
- }
- if (!device.IsExtTransformFeedbackSupported()) {
- return;
+ query_cache.CounterEnable(VideoCommon::QueryType::StreamingByteCount,
+ regs.transform_feedback_enabled);
+ if (regs.transform_feedback_enabled != 0) {
+ UNIMPLEMENTED_IF(regs.IsShaderConfigEnabled(Maxwell::ShaderType::TessellationInit) ||
+ regs.IsShaderConfigEnabled(Maxwell::ShaderType::Tessellation));
}
- scheduler.Record(
- [](vk::CommandBuffer cmdbuf) { cmdbuf.EndTransformFeedbackEXT(0, 0, nullptr, nullptr); });
}
void RasterizerVulkan::UpdateViewportsState(Tegra::Engines::Maxwell3D::Regs& regs) {
@@ -1044,15 +1015,37 @@ void RasterizerVulkan::UpdateDepthBias(Tegra::Engines::Maxwell3D::Regs& regs) {
regs.zeta.format == Tegra::DepthFormat::X8Z24_UNORM ||
regs.zeta.format == Tegra::DepthFormat::S8Z24_UNORM ||
regs.zeta.format == Tegra::DepthFormat::V8Z24_UNORM;
- if (is_d24 && !device.SupportsD24DepthBuffer()) {
+ bool force_unorm = ([&] {
+ if (!is_d24 || device.SupportsD24DepthBuffer()) {
+ return false;
+ }
+ if (device.IsExtDepthBiasControlSupported()) {
+ return true;
+ }
+ if (!Settings::values.renderer_amdvlk_depth_bias_workaround) {
+ return false;
+ }
// the base formulas can be obtained from here:
// https://docs.microsoft.com/en-us/windows/win32/direct3d11/d3d10-graphics-programming-guide-output-merger-stage-depth-bias
const double rescale_factor =
static_cast<double>(1ULL << (32 - 24)) / (static_cast<double>(0x1.ep+127));
units = static_cast<float>(static_cast<double>(units) * rescale_factor);
- }
+ return false;
+ })();
scheduler.Record([constant = units, clamp = regs.depth_bias_clamp,
- factor = regs.slope_scale_depth_bias](vk::CommandBuffer cmdbuf) {
+ factor = regs.slope_scale_depth_bias, force_unorm,
+ precise = device.HasExactDepthBiasControl()](vk::CommandBuffer cmdbuf) {
+ if (force_unorm) {
+ VkDepthBiasRepresentationInfoEXT info{
+ .sType = VK_STRUCTURE_TYPE_DEPTH_BIAS_REPRESENTATION_INFO_EXT,
+ .pNext = nullptr,
+ .depthBiasRepresentation =
+ VK_DEPTH_BIAS_REPRESENTATION_LEAST_REPRESENTABLE_VALUE_FORCE_UNORM_EXT,
+ .depthBiasExact = precise ? VK_TRUE : VK_FALSE,
+ };
+ cmdbuf.SetDepthBias(constant, clamp, factor, &info);
+ return;
+ }
cmdbuf.SetDepthBias(constant, clamp, factor);
});
}
diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.h b/src/video_core/renderer_vulkan/vk_rasterizer.h
index b31982485..ad069556c 100644
--- a/src/video_core/renderer_vulkan/vk_rasterizer.h
+++ b/src/video_core/renderer_vulkan/vk_rasterizer.h
@@ -84,8 +84,9 @@ public:
void DrawTexture() override;
void Clear(u32 layer_count) override;
void DispatchCompute() override;
- void ResetCounter(VideoCore::QueryType type) override;
- void Query(GPUVAddr gpu_addr, VideoCore::QueryType type, std::optional<u64> timestamp) override;
+ void ResetCounter(VideoCommon::QueryType type) override;
+ void Query(GPUVAddr gpu_addr, VideoCommon::QueryType type,
+ VideoCommon::QueryPropertiesFlags flags, u32 payload, u32 subreport) override;
void BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr gpu_addr, u32 size) override;
void DisableGraphicsUniformBuffer(size_t stage, u32 index) override;
void FlushAll() override;
@@ -106,7 +107,7 @@ public:
void SyncOperation(std::function<void()>&& func) override;
void SignalSyncPoint(u32 value) override;
void SignalReference() override;
- void ReleaseFences() override;
+ void ReleaseFences(bool force = true) override;
void FlushAndInvalidateRegion(
VAddr addr, u64 size, VideoCommon::CacheType which = VideoCommon::CacheType::All) override;
void WaitForIdle() override;
@@ -146,9 +147,7 @@ private:
void UpdateDynamicStates();
- void BeginTransformFeedback();
-
- void EndTransformFeedback();
+ void HandleTransformFeedback();
void UpdateViewportsState(Tegra::Engines::Maxwell3D::Regs& regs);
void UpdateScissorsState(Tegra::Engines::Maxwell3D::Regs& regs);
@@ -195,8 +194,9 @@ private:
TextureCache texture_cache;
BufferCacheRuntime buffer_cache_runtime;
BufferCache buffer_cache;
- PipelineCache pipeline_cache;
+ QueryCacheRuntime query_cache_runtime;
QueryCache query_cache;
+ PipelineCache pipeline_cache;
AccelerateDMA accelerate_dma;
FenceManager fence_manager;
diff --git a/src/video_core/renderer_vulkan/vk_scheduler.cpp b/src/video_core/renderer_vulkan/vk_scheduler.cpp
index 89fd31b4f..3be7837f4 100644
--- a/src/video_core/renderer_vulkan/vk_scheduler.cpp
+++ b/src/video_core/renderer_vulkan/vk_scheduler.cpp
@@ -243,10 +243,10 @@ void Scheduler::AllocateNewContext() {
#if ANDROID
if (Settings::IsGPULevelHigh()) {
// This is problematic on Android, disable on GPU Normal.
- query_cache->UpdateCounters();
+ query_cache->NotifySegment(true);
}
#else
- query_cache->UpdateCounters();
+ query_cache->NotifySegment(true);
#endif
}
}
@@ -261,11 +261,12 @@ void Scheduler::EndPendingOperations() {
#if ANDROID
if (Settings::IsGPULevelHigh()) {
// This is problematic on Android, disable on GPU Normal.
- query_cache->DisableStreams();
+ // query_cache->DisableStreams();
}
#else
- query_cache->DisableStreams();
+ // query_cache->DisableStreams();
#endif
+ query_cache->NotifySegment(false);
EndRenderPass();
}
diff --git a/src/video_core/renderer_vulkan/vk_scheduler.h b/src/video_core/renderer_vulkan/vk_scheduler.h
index 475c682eb..da03803aa 100644
--- a/src/video_core/renderer_vulkan/vk_scheduler.h
+++ b/src/video_core/renderer_vulkan/vk_scheduler.h
@@ -17,6 +17,11 @@
#include "video_core/renderer_vulkan/vk_master_semaphore.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
+namespace VideoCommon {
+template <typename Trait>
+class QueryCacheBase;
+}
+
namespace Vulkan {
class CommandPool;
@@ -24,7 +29,8 @@ class Device;
class Framebuffer;
class GraphicsPipeline;
class StateTracker;
-class QueryCache;
+
+struct QueryCacheParams;
/// The scheduler abstracts command buffer and fence management with an interface that's able to do
/// OpenGL-like operations on Vulkan command buffers.
@@ -63,7 +69,7 @@ public:
void InvalidateState();
/// Assigns the query cache.
- void SetQueryCache(QueryCache& query_cache_) {
+ void SetQueryCache(VideoCommon::QueryCacheBase<QueryCacheParams>& query_cache_) {
query_cache = &query_cache_;
}
@@ -219,7 +225,7 @@ private:
std::unique_ptr<MasterSemaphore> master_semaphore;
std::unique_ptr<CommandPool> command_pool;
- QueryCache* query_cache = nullptr;
+ VideoCommon::QueryCacheBase<QueryCacheParams>* query_cache = nullptr;
vk::CommandBuffer current_cmdbuf;
diff --git a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp
index ce92f66ab..b278614e6 100644
--- a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp
+++ b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp
@@ -24,25 +24,38 @@ using namespace Common::Literals;
// Maximum potential alignment of a Vulkan buffer
constexpr VkDeviceSize MAX_ALIGNMENT = 256;
-// Maximum size to put elements in the stream buffer
-constexpr VkDeviceSize MAX_STREAM_BUFFER_REQUEST_SIZE = 8_MiB;
// Stream buffer size in bytes
-constexpr VkDeviceSize STREAM_BUFFER_SIZE = 128_MiB;
-constexpr VkDeviceSize REGION_SIZE = STREAM_BUFFER_SIZE / StagingBufferPool::NUM_SYNCS;
+constexpr VkDeviceSize MAX_STREAM_BUFFER_SIZE = 128_MiB;
-size_t Region(size_t iterator) noexcept {
- return iterator / REGION_SIZE;
+size_t GetStreamBufferSize(const Device& device) {
+ VkDeviceSize size{0};
+ if (device.HasDebuggingToolAttached()) {
+ ForEachDeviceLocalHostVisibleHeap(device, [&size](size_t index, VkMemoryHeap& heap) {
+ size = std::max(size, heap.size);
+ });
+ // If rebar is not supported, cut the max heap size to 40%. This will allow 2 captures to be
+ // loaded at the same time in RenderDoc. If rebar is supported, this shouldn't be an issue
+ // as the heap will be much larger.
+ if (size <= 256_MiB) {
+ size = size * 40 / 100;
+ }
+ } else {
+ size = MAX_STREAM_BUFFER_SIZE;
+ }
+ return std::min(Common::AlignUp(size, MAX_ALIGNMENT), MAX_STREAM_BUFFER_SIZE);
}
} // Anonymous namespace
StagingBufferPool::StagingBufferPool(const Device& device_, MemoryAllocator& memory_allocator_,
Scheduler& scheduler_)
- : device{device_}, memory_allocator{memory_allocator_}, scheduler{scheduler_} {
+ : device{device_}, memory_allocator{memory_allocator_}, scheduler{scheduler_},
+ stream_buffer_size{GetStreamBufferSize(device)}, region_size{stream_buffer_size /
+ StagingBufferPool::NUM_SYNCS} {
VkBufferCreateInfo stream_ci = {
.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
- .size = STREAM_BUFFER_SIZE,
+ .size = stream_buffer_size,
.usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT |
VK_BUFFER_USAGE_INDEX_BUFFER_BIT | VK_BUFFER_USAGE_STORAGE_BUFFER_BIT,
.sharingMode = VK_SHARING_MODE_EXCLUSIVE,
@@ -63,7 +76,7 @@ StagingBufferPool::StagingBufferPool(const Device& device_, MemoryAllocator& mem
StagingBufferPool::~StagingBufferPool() = default;
StagingBufferRef StagingBufferPool::Request(size_t size, MemoryUsage usage, bool deferred) {
- if (!deferred && usage == MemoryUsage::Upload && size <= MAX_STREAM_BUFFER_REQUEST_SIZE) {
+ if (!deferred && usage == MemoryUsage::Upload && size <= region_size) {
return GetStreamBuffer(size);
}
return GetStagingBuffer(size, usage, deferred);
@@ -101,7 +114,7 @@ StagingBufferRef StagingBufferPool::GetStreamBuffer(size_t size) {
used_iterator = iterator;
free_iterator = std::max(free_iterator, iterator + size);
- if (iterator + size >= STREAM_BUFFER_SIZE) {
+ if (iterator + size >= stream_buffer_size) {
std::fill(sync_ticks.begin() + Region(used_iterator), sync_ticks.begin() + NUM_SYNCS,
current_tick);
used_iterator = 0;
diff --git a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.h b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.h
index 5f69f08b1..d3deb9072 100644
--- a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.h
+++ b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.h
@@ -90,6 +90,9 @@ private:
void ReleaseCache(MemoryUsage usage);
void ReleaseLevel(StagingBuffersCache& cache, size_t log2);
+ size_t Region(size_t iter) const noexcept {
+ return iter / region_size;
+ }
const Device& device;
MemoryAllocator& memory_allocator;
@@ -97,6 +100,8 @@ private:
vk::Buffer stream_buffer;
std::span<u8> stream_pointer;
+ VkDeviceSize stream_buffer_size;
+ VkDeviceSize region_size;
size_t iterator = 0;
size_t used_iterator = 0;
diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp
index b3e17c332..e266c1dbe 100644
--- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp
@@ -120,19 +120,9 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) {
return usage;
}
-/// Returns the preferred format for a VkImage
-[[nodiscard]] PixelFormat StorageFormat(PixelFormat format) {
- switch (format) {
- case PixelFormat::A8B8G8R8_SRGB:
- return PixelFormat::A8B8G8R8_UNORM;
- default:
- return format;
- }
-}
-
[[nodiscard]] VkImageCreateInfo MakeImageCreateInfo(const Device& device, const ImageInfo& info) {
- const PixelFormat format = StorageFormat(info.format);
- const auto format_info = MaxwellToVK::SurfaceFormat(device, FormatType::Optimal, false, format);
+ const auto format_info =
+ MaxwellToVK::SurfaceFormat(device, FormatType::Optimal, false, info.format);
VkImageCreateFlags flags{};
if (info.type == ImageType::e2D && info.resources.layers >= 6 &&
info.size.width == info.size.height && !device.HasBrokenCubeImageCompability()) {
@@ -157,7 +147,7 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) {
.arrayLayers = static_cast<u32>(info.resources.layers),
.samples = ConvertSampleCount(info.num_samples),
.tiling = VK_IMAGE_TILING_OPTIMAL,
- .usage = ImageUsageFlags(format_info, format),
+ .usage = ImageUsageFlags(format_info, info.format),
.sharingMode = VK_SHARING_MODE_EXCLUSIVE,
.queueFamilyIndexCount = 0,
.pQueueFamilyIndices = nullptr,
@@ -186,6 +176,36 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) {
return allocator.CreateImage(image_ci);
}
+[[nodiscard]] vk::ImageView MakeStorageView(const vk::Device& device, u32 level, VkImage image,
+ VkFormat format) {
+ static constexpr VkImageViewUsageCreateInfo storage_image_view_usage_create_info{
+ .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_USAGE_CREATE_INFO,
+ .pNext = nullptr,
+ .usage = VK_IMAGE_USAGE_STORAGE_BIT,
+ };
+ return device.CreateImageView(VkImageViewCreateInfo{
+ .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
+ .pNext = &storage_image_view_usage_create_info,
+ .flags = 0,
+ .image = image,
+ .viewType = VK_IMAGE_VIEW_TYPE_2D_ARRAY,
+ .format = format,
+ .components{
+ .r = VK_COMPONENT_SWIZZLE_IDENTITY,
+ .g = VK_COMPONENT_SWIZZLE_IDENTITY,
+ .b = VK_COMPONENT_SWIZZLE_IDENTITY,
+ .a = VK_COMPONENT_SWIZZLE_IDENTITY,
+ },
+ .subresourceRange{
+ .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
+ .baseMipLevel = level,
+ .levelCount = 1,
+ .baseArrayLayer = 0,
+ .layerCount = VK_REMAINING_ARRAY_LAYERS,
+ },
+ });
+}
+
[[nodiscard]] VkImageAspectFlags ImageAspectMask(PixelFormat format) {
switch (VideoCore::Surface::GetFormatType(format)) {
case VideoCore::Surface::SurfaceType::ColorTexture:
@@ -218,6 +238,7 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) {
return any_r ? VK_IMAGE_ASPECT_STENCIL_BIT : VK_IMAGE_ASPECT_DEPTH_BIT;
case PixelFormat::D16_UNORM:
case PixelFormat::D32_FLOAT:
+ case PixelFormat::X8_D24_UNORM:
return VK_IMAGE_ASPECT_DEPTH_BIT;
case PixelFormat::S8_UINT:
return VK_IMAGE_ASPECT_STENCIL_BIT;
@@ -600,7 +621,7 @@ void CopyBufferToImage(vk::CommandBuffer cmdbuf, VkBuffer src_buffer, VkImage im
}
void TryTransformSwizzleIfNeeded(PixelFormat format, std::array<SwizzleSource, 4>& swizzle,
- bool emulate_bgr565) {
+ bool emulate_bgr565, bool emulate_a4b4g4r4) {
switch (format) {
case PixelFormat::A1B5G5R5_UNORM:
std::ranges::transform(swizzle, swizzle.begin(), SwapBlueRed);
@@ -616,6 +637,11 @@ void TryTransformSwizzleIfNeeded(PixelFormat format, std::array<SwizzleSource, 4
case PixelFormat::G4R4_UNORM:
std::ranges::transform(swizzle, swizzle.begin(), SwapGreenRed);
break;
+ case PixelFormat::A4B4G4R4_UNORM:
+ if (emulate_a4b4g4r4) {
+ std::ranges::reverse(swizzle);
+ }
+ break;
default:
break;
}
@@ -822,6 +848,10 @@ TextureCacheRuntime::TextureCacheRuntime(const Device& device_, Scheduler& sched
astc_decoder_pass.emplace(device, scheduler, descriptor_pool, staging_buffer_pool,
compute_pass_descriptor_queue, memory_allocator);
}
+ if (device.IsStorageImageMultisampleSupported()) {
+ msaa_copy_pass = std::make_unique<MSAACopyPass>(
+ device, scheduler, descriptor_pool, staging_buffer_pool, compute_pass_descriptor_queue);
+ }
if (!device.IsKhrImageFormatListSupported()) {
return;
}
@@ -1044,15 +1074,27 @@ void TextureCacheRuntime::BlitImage(Framebuffer* dst_framebuffer, ImageView& dst
dst_region, src_region, filter, operation);
return;
}
+ ASSERT(src.format == dst.format);
if (aspect_mask == (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) {
- if (!device.IsBlitDepthStencilSupported()) {
+ const auto format = src.format;
+ const auto can_blit_depth_stencil = [this, format] {
+ switch (format) {
+ case VideoCore::Surface::PixelFormat::D24_UNORM_S8_UINT:
+ case VideoCore::Surface::PixelFormat::S8_UINT_D24_UNORM:
+ return device.IsBlitDepth24Stencil8Supported();
+ case VideoCore::Surface::PixelFormat::D32_FLOAT_S8_UINT:
+ return device.IsBlitDepth32Stencil8Supported();
+ default:
+ UNREACHABLE();
+ }
+ }();
+ if (!can_blit_depth_stencil) {
UNIMPLEMENTED_IF(is_src_msaa || is_dst_msaa);
blit_image_helper.BlitDepthStencil(dst_framebuffer, src.DepthView(), src.StencilView(),
dst_region, src_region, filter, operation);
return;
}
}
- ASSERT(src.format == dst.format);
ASSERT(!(is_dst_msaa && !is_src_msaa));
ASSERT(operation == Fermi2D::Operation::SrcCopy);
@@ -1278,7 +1320,11 @@ void TextureCacheRuntime::CopyImage(Image& dst, Image& src,
void TextureCacheRuntime::CopyImageMSAA(Image& dst, Image& src,
std::span<const VideoCommon::ImageCopy> copies) {
- UNIMPLEMENTED_MSG("Copying images with different samples is not implemented in Vulkan.");
+ const bool msaa_to_non_msaa = src.info.num_samples > 1 && dst.info.num_samples == 1;
+ if (msaa_copy_pass) {
+ return msaa_copy_pass->CopyImage(dst, src, copies, msaa_to_non_msaa);
+ }
+ UNIMPLEMENTED_MSG("Copying images with different samples is not supported.");
}
u64 TextureCacheRuntime::GetDeviceLocalMemory() const {
@@ -1326,39 +1372,15 @@ Image::Image(TextureCacheRuntime& runtime_, const ImageInfo& info_, GPUVAddr gpu
if (runtime->device.HasDebuggingToolAttached()) {
original_image.SetObjectNameEXT(VideoCommon::Name(*this).c_str());
}
- static constexpr VkImageViewUsageCreateInfo storage_image_view_usage_create_info{
- .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_USAGE_CREATE_INFO,
- .pNext = nullptr,
- .usage = VK_IMAGE_USAGE_STORAGE_BIT,
- };
current_image = *original_image;
+ storage_image_views.resize(info.resources.levels);
if (IsPixelFormatASTC(info.format) && !runtime->device.IsOptimalAstcSupported() &&
Settings::values.astc_recompression.GetValue() ==
Settings::AstcRecompression::Uncompressed) {
const auto& device = runtime->device.GetLogical();
- storage_image_views.reserve(info.resources.levels);
for (s32 level = 0; level < info.resources.levels; ++level) {
- storage_image_views.push_back(device.CreateImageView(VkImageViewCreateInfo{
- .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
- .pNext = &storage_image_view_usage_create_info,
- .flags = 0,
- .image = *original_image,
- .viewType = VK_IMAGE_VIEW_TYPE_2D_ARRAY,
- .format = VK_FORMAT_A8B8G8R8_UNORM_PACK32,
- .components{
- .r = VK_COMPONENT_SWIZZLE_IDENTITY,
- .g = VK_COMPONENT_SWIZZLE_IDENTITY,
- .b = VK_COMPONENT_SWIZZLE_IDENTITY,
- .a = VK_COMPONENT_SWIZZLE_IDENTITY,
- },
- .subresourceRange{
- .aspectMask = aspect_mask,
- .baseMipLevel = static_cast<u32>(level),
- .levelCount = 1,
- .baseArrayLayer = 0,
- .layerCount = VK_REMAINING_ARRAY_LAYERS,
- },
- }));
+ storage_image_views[level] =
+ MakeStorageView(device, level, *original_image, VK_FORMAT_A8B8G8R8_UNORM_PACK32);
}
}
}
@@ -1489,6 +1511,17 @@ void Image::DownloadMemory(const StagingBufferRef& map, std::span<const BufferIm
DownloadMemory(buffers, offsets, copies);
}
+VkImageView Image::StorageImageView(s32 level) noexcept {
+ auto& view = storage_image_views[level];
+ if (!view) {
+ const auto format_info =
+ MaxwellToVK::SurfaceFormat(runtime->device, FormatType::Optimal, true, info.format);
+ view =
+ MakeStorageView(runtime->device.GetLogical(), level, current_image, format_info.format);
+ }
+ return *view;
+}
+
bool Image::IsRescaled() const noexcept {
return True(flags & ImageFlagBits::Rescaled);
}
@@ -1626,8 +1659,8 @@ bool Image::NeedsScaleHelper() const {
return true;
}
static constexpr auto OPTIMAL_FORMAT = FormatType::Optimal;
- const PixelFormat format = StorageFormat(info.format);
- const auto vk_format = MaxwellToVK::SurfaceFormat(device, OPTIMAL_FORMAT, false, format).format;
+ const auto vk_format =
+ MaxwellToVK::SurfaceFormat(device, OPTIMAL_FORMAT, false, info.format).format;
const auto blit_usage = VK_FORMAT_FEATURE_BLIT_SRC_BIT | VK_FORMAT_FEATURE_BLIT_DST_BIT;
const bool needs_blit_helper = !device.IsFormatSupported(vk_format, blit_usage, OPTIMAL_FORMAT);
return needs_blit_helper;
@@ -1649,7 +1682,8 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI
};
if (!info.IsRenderTarget()) {
swizzle = info.Swizzle();
- TryTransformSwizzleIfNeeded(format, swizzle, device->MustEmulateBGR565());
+ TryTransformSwizzleIfNeeded(format, swizzle, device->MustEmulateBGR565(),
+ !device->IsExt4444FormatsSupported());
if ((aspect_mask & (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) != 0) {
std::ranges::transform(swizzle, swizzle.begin(), ConvertGreenRed);
}
diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.h b/src/video_core/renderer_vulkan/vk_texture_cache.h
index 565ce19a9..d6c5a15cc 100644
--- a/src/video_core/renderer_vulkan/vk_texture_cache.h
+++ b/src/video_core/renderer_vulkan/vk_texture_cache.h
@@ -117,6 +117,7 @@ public:
BlitImageHelper& blit_image_helper;
RenderPassCache& render_pass_cache;
std::optional<ASTCDecoderPass> astc_decoder_pass;
+ std::unique_ptr<MSAACopyPass> msaa_copy_pass;
const Settings::ResolutionScalingInfo& resolution;
std::array<std::vector<VkFormat>, VideoCore::Surface::MaxPixelFormat> view_formats;
@@ -161,15 +162,13 @@ public:
return aspect_mask;
}
- [[nodiscard]] VkImageView StorageImageView(s32 level) const noexcept {
- return *storage_image_views[level];
- }
-
/// Returns true when the image is already initialized and mark it as initialized
[[nodiscard]] bool ExchangeInitialization() noexcept {
return std::exchange(initialized, true);
}
+ VkImageView StorageImageView(s32 level) noexcept;
+
bool IsRescaled() const noexcept;
bool ScaleUp(bool ignore = false);
generated by cgit v1.2.3 (git 2.25.1) at 2024-06-29 07:05:45 +0200