rooty/dxvk
1
0
Fork 0
mirror of https://github.com/doitsujin/dxvk.git synced 2025-04-12 03:02:33 +02:00
Code Issues Projects Releases Wiki Activity
dxvk/src/dxvk/dxvk_context.cpp
Philip Rebohle a8f9fdb21d [dxvk] Rearrange descriptor sets 
This allows us not to unnecessarily dirty the FS UBO set when
changing tetxure bindings, leading to a cleaner separation.
2023-01-15 15:36:05 +01:00

6290 lines
229 KiB
C++
Raw Blame History

#include <cstring>
#include <vector>
#include <utility>
#include "dxvk_device.h"
#include "dxvk_context.h"
namespace dxvk {
DxvkContext::DxvkContext(const Rc<DxvkDevice>& device, DxvkContextType type)
: m_device (device),
m_type (type),
m_common (&device->m_objects),
m_sdmaAcquires(DxvkCmdBuffer::SdmaBuffer),
m_sdmaBarriers(DxvkCmdBuffer::SdmaBuffer),
m_initBarriers(DxvkCmdBuffer::InitBuffer),
m_execAcquires(DxvkCmdBuffer::ExecBuffer),
m_execBarriers(DxvkCmdBuffer::ExecBuffer),
m_queryManager(m_common->queryPool()),
m_staging (device, StagingBufferSize) {
// Init framebuffer info with default render pass in case
// the app does not explicitly bind any render targets
m_state.om.framebufferInfo = makeFramebufferInfo(m_state.om.renderTargets);
m_descriptorManager = new DxvkDescriptorManager(device.ptr(), type);
// Default destination barriers for graphics pipelines
m_globalRoGraphicsBarrier.stages = m_device->getShaderPipelineStages()
| VK_PIPELINE_STAGE_TRANSFER_BIT
| VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT
| VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT
| VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT;
m_globalRoGraphicsBarrier.access = 0;
if (m_device->features().extTransformFeedback.transformFeedback)
m_globalRoGraphicsBarrier.stages |= VK_PIPELINE_STAGE_TRANSFORM_FEEDBACK_BIT_EXT;
m_globalRwGraphicsBarrier = m_globalRoGraphicsBarrier;
m_globalRwGraphicsBarrier.stages |= VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT
| VK_PIPELINE_STAGE_VERTEX_INPUT_BIT;
m_globalRwGraphicsBarrier.access |= VK_ACCESS_INDIRECT_COMMAND_READ_BIT
| VK_ACCESS_INDEX_READ_BIT
| VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT
| VK_ACCESS_UNIFORM_READ_BIT
| VK_ACCESS_SHADER_READ_BIT
| VK_ACCESS_COLOR_ATTACHMENT_READ_BIT
| VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT
| VK_ACCESS_TRANSFER_READ_BIT;
if (m_device->features().extTransformFeedback.transformFeedback)
m_globalRwGraphicsBarrier.access |= VK_ACCESS_TRANSFORM_FEEDBACK_COUNTER_READ_BIT_EXT;
// Store the lifetime tracking bit as a context feature so
// that we don't have to scan device features at draw time
if (m_device->mustTrackPipelineLifetime())
m_features.set(DxvkContextFeature::TrackGraphicsPipeline);
// Variable multisample rate is needed to efficiently support
// rendering without bound render targets, otherwise we may
// have to interrupt the current render pass whenever the
// requested rasterizer sample count changes
if (m_device->features().core.features.variableMultisampleRate)
m_features.set(DxvkContextFeature::VariableMultisampleRate);
}
DxvkContext::~DxvkContext() {
}
void DxvkContext::beginRecording(const Rc<DxvkCommandList>& cmdList) {
m_cmd = cmdList;
m_cmd->init();
if (m_descriptorPool == nullptr)
m_descriptorPool = m_descriptorManager->getDescriptorPool();
this->beginCurrentCommands();
}
Rc<DxvkCommandList> DxvkContext::endRecording() {
this->endCurrentCommands();
if (m_descriptorPool->shouldSubmit(false)) {
m_cmd->trackDescriptorPool(m_descriptorPool, m_descriptorManager);
m_descriptorPool = m_descriptorManager->getDescriptorPool();
}
m_cmd->finalize();
return std::exchange(m_cmd, nullptr);
}
void DxvkContext::endFrame() {
if (m_descriptorPool->shouldSubmit(true)) {
m_cmd->trackDescriptorPool(m_descriptorPool, m_descriptorManager);
m_descriptorPool = m_descriptorManager->getDescriptorPool();
}
}
void DxvkContext::flushCommandList() {
m_device->submitCommandList(
this->endRecording());
this->beginRecording(
m_device->createCommandList());
}
void DxvkContext::beginQuery(const Rc<DxvkGpuQuery>& query) {
m_queryManager.enableQuery(m_cmd, query);
}
void DxvkContext::endQuery(const Rc<DxvkGpuQuery>& query) {
m_queryManager.disableQuery(m_cmd, query);
}
void DxvkContext::blitImage(
const Rc<DxvkImage>& dstImage,
const VkComponentMapping& dstMapping,
const Rc<DxvkImage>& srcImage,
const VkComponentMapping& srcMapping,
const VkImageBlit& region,
VkFilter filter) {
this->spillRenderPass(true);
this->prepareImage(dstImage, vk::makeSubresourceRange(region.dstSubresource));
this->prepareImage(srcImage, vk::makeSubresourceRange(region.srcSubresource));
auto mapping = util::resolveSrcComponentMapping(dstMapping, srcMapping);
bool canUseFb = (srcImage->info().usage & VK_IMAGE_USAGE_SAMPLED_BIT)
&& (dstImage->info().usage & VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT)
&& ((dstImage->info().flags & VK_IMAGE_CREATE_2D_ARRAY_COMPATIBLE_BIT)
|| (dstImage->info().type != VK_IMAGE_TYPE_3D));
bool useFb = dstImage->info().sampleCount != VK_SAMPLE_COUNT_1_BIT
|| !util::isIdentityMapping(mapping);
if (!useFb) {
this->blitImageHw(
dstImage, srcImage,
region, filter);
} else if (canUseFb) {
this->blitImageFb(
dstImage, srcImage,
region, mapping, filter);
} else {
Logger::err("DxvkContext: Unsupported blit operation");
}
}
void DxvkContext::changeImageLayout(
const Rc<DxvkImage>& image,
VkImageLayout layout) {
if (image->info().layout != layout) {
this->spillRenderPass(true);
VkImageSubresourceRange subresources = image->getAvailableSubresources();
this->prepareImage(image, subresources);
if (m_execBarriers.isImageDirty(image, subresources, DxvkAccess::Write))
m_execBarriers.recordCommands(m_cmd);
m_execBarriers.accessImage(image, subresources,
image->info().layout,
image->info().stages, 0,
layout,
image->info().stages,
image->info().access);
image->setLayout(layout);
m_cmd->trackResource<DxvkAccess::Write>(image);
}
}
void DxvkContext::clearBuffer(
const Rc<DxvkBuffer>& buffer,
VkDeviceSize offset,
VkDeviceSize length,
uint32_t value) {
bool replaceBuffer = this->tryInvalidateDeviceLocalBuffer(buffer, length);
auto bufferSlice = buffer->getSliceHandle(offset, align(length, sizeof(uint32_t)));
if (!replaceBuffer) {
this->spillRenderPass(true);
if (m_execBarriers.isBufferDirty(bufferSlice, DxvkAccess::Write))
m_execBarriers.recordCommands(m_cmd);
}
DxvkCmdBuffer cmdBuffer = replaceBuffer
? DxvkCmdBuffer::InitBuffer
: DxvkCmdBuffer::ExecBuffer;
if (length > sizeof(value)) {
m_cmd->cmdFillBuffer(cmdBuffer,
bufferSlice.handle,
bufferSlice.offset,
bufferSlice.length,
value);
} else {
m_cmd->cmdUpdateBuffer(cmdBuffer,
bufferSlice.handle,
bufferSlice.offset,
bufferSlice.length,
&value);
}
auto& barriers = replaceBuffer
? m_initBarriers
: m_execBarriers;
barriers.accessBuffer(bufferSlice,
VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_ACCESS_TRANSFER_WRITE_BIT,
buffer->info().stages,
buffer->info().access);
m_cmd->trackResource<DxvkAccess::Write>(buffer);
}
void DxvkContext::clearBufferView(
const Rc<DxvkBufferView>& bufferView,
VkDeviceSize offset,
VkDeviceSize length,
VkClearColorValue value) {
this->spillRenderPass(true);
this->invalidateState();
// The view range might have been invalidated, so
// we need to make sure the handle is up to date
bufferView->updateView();
auto bufferSlice = bufferView->getSliceHandle();
if (m_execBarriers.isBufferDirty(bufferSlice, DxvkAccess::Write))
m_execBarriers.recordCommands(m_cmd);
// Query pipeline objects to use for this clear operation
DxvkMetaClearPipeline pipeInfo = m_common->metaClear().getClearBufferPipeline(
lookupFormatInfo(bufferView->info().format)->flags);
// Create a descriptor set pointing to the view
VkBufferView viewObject = bufferView->handle();
VkDescriptorSet descriptorSet = m_descriptorPool->alloc(pipeInfo.dsetLayout);
VkWriteDescriptorSet descriptorWrite = { VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET };
descriptorWrite.dstSet = descriptorSet;
descriptorWrite.dstBinding = 0;
descriptorWrite.dstArrayElement = 0;
descriptorWrite.descriptorCount = 1;
descriptorWrite.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER;
descriptorWrite.pTexelBufferView = &viewObject;
m_cmd->updateDescriptorSets(1, &descriptorWrite);
// Prepare shader arguments
DxvkMetaClearArgs pushArgs = { };
pushArgs.clearValue = value;
pushArgs.offset = VkOffset3D { int32_t(offset), 0, 0 };
pushArgs.extent = VkExtent3D { uint32_t(length), 1, 1 };
VkExtent3D workgroups = util::computeBlockCount(
pushArgs.extent, pipeInfo.workgroupSize);
m_cmd->cmdBindPipeline(
VK_PIPELINE_BIND_POINT_COMPUTE,
pipeInfo.pipeline);
m_cmd->cmdBindDescriptorSet(
VK_PIPELINE_BIND_POINT_COMPUTE,
pipeInfo.pipeLayout, descriptorSet,
0, nullptr);
m_cmd->cmdPushConstants(
pipeInfo.pipeLayout,
VK_SHADER_STAGE_COMPUTE_BIT,
0, sizeof(pushArgs), &pushArgs);
m_cmd->cmdDispatch(
workgroups.width,
workgroups.height,
workgroups.depth);
m_execBarriers.accessBuffer(bufferSlice,
VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
VK_ACCESS_SHADER_WRITE_BIT,
bufferView->bufferInfo().stages,
bufferView->bufferInfo().access);
m_cmd->trackResource<DxvkAccess::None>(bufferView);
m_cmd->trackResource<DxvkAccess::Write>(bufferView->buffer());
}
void DxvkContext::clearRenderTarget(
const Rc<DxvkImageView>& imageView,
VkImageAspectFlags clearAspects,
VkClearValue clearValue) {
// Make sure the color components are ordered correctly
if (clearAspects & VK_IMAGE_ASPECT_COLOR_BIT) {
clearValue.color = util::swizzleClearColor(clearValue.color,
util::invertComponentMapping(imageView->info().swizzle));
}
// Check whether the render target view is an attachment
// of the current framebuffer and is included entirely.
// If not, we need to create a temporary framebuffer.
int32_t attachmentIndex = -1;
if (m_state.om.framebufferInfo.isFullSize(imageView))
attachmentIndex = m_state.om.framebufferInfo.findAttachment(imageView);
if (attachmentIndex < 0) {
// Suspend works here because we'll end up with one of these scenarios:
// 1) The render pass gets ended for good, in which case we emit barriers
// 2) The clear gets folded into render pass ops, so the layout is correct
// 3) The clear gets executed separately, in which case updateFramebuffer
// will indirectly emit barriers for the given render target.
// If there is overlap, we need to explicitly transition affected attachments.
this->spillRenderPass(true);
this->prepareImage(imageView->image(), imageView->subresources(), false);
} else if (!m_state.om.framebufferInfo.isWritable(attachmentIndex, clearAspects)) {
// We cannot inline clears if the clear aspects are not writable
this->spillRenderPass(true);
}
if (m_flags.test(DxvkContextFlag::GpRenderPassBound)) {
uint32_t colorIndex = std::max(0, m_state.om.framebufferInfo.getColorAttachmentIndex(attachmentIndex));
VkClearAttachment clearInfo;
clearInfo.aspectMask = clearAspects;
clearInfo.colorAttachment = colorIndex;
clearInfo.clearValue = clearValue;
VkClearRect clearRect;
clearRect.rect.offset.x = 0;
clearRect.rect.offset.y = 0;
clearRect.rect.extent.width = imageView->mipLevelExtent(0).width;
clearRect.rect.extent.height = imageView->mipLevelExtent(0).height;
clearRect.baseArrayLayer = 0;
clearRect.layerCount = imageView->info().numLayers;
m_cmd->cmdClearAttachments(1, &clearInfo, 1, &clearRect);
} else
this->deferClear(imageView, clearAspects, clearValue);
}
void DxvkContext::clearImageView(
const Rc<DxvkImageView>& imageView,
VkOffset3D offset,
VkExtent3D extent,
VkImageAspectFlags aspect,
VkClearValue value) {
const VkImageUsageFlags viewUsage = imageView->info().usage;
if (aspect & VK_IMAGE_ASPECT_COLOR_BIT) {
value.color = util::swizzleClearColor(value.color,
util::invertComponentMapping(imageView->info().swizzle));
}
if (viewUsage & (VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT | VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT))
this->clearImageViewFb(imageView, offset, extent, aspect, value);
else if (viewUsage & VK_IMAGE_USAGE_STORAGE_BIT)
this->clearImageViewCs(imageView, offset, extent, value);
}
void DxvkContext::copyBuffer(
const Rc<DxvkBuffer>& dstBuffer,
VkDeviceSize dstOffset,
const Rc<DxvkBuffer>& srcBuffer,
VkDeviceSize srcOffset,
VkDeviceSize numBytes) {
// When overwriting small buffers, we can allocate a new slice in order to
// avoid suspending the current render pass or inserting barriers. The source
// buffer must be read-only since otherwise we cannot schedule the copy early.
bool srcIsReadOnly = DxvkBarrierSet::getAccessTypes(srcBuffer->info().access) == DxvkAccess::Read;
bool replaceBuffer = srcIsReadOnly && this->tryInvalidateDeviceLocalBuffer(dstBuffer, numBytes);
auto srcSlice = srcBuffer->getSliceHandle(srcOffset, numBytes);
auto dstSlice = dstBuffer->getSliceHandle(dstOffset, numBytes);
if (!replaceBuffer) {
this->spillRenderPass(true);
if (m_execBarriers.isBufferDirty(srcSlice, DxvkAccess::Read)
|| m_execBarriers.isBufferDirty(dstSlice, DxvkAccess::Write))
m_execBarriers.recordCommands(m_cmd);
}
DxvkCmdBuffer cmdBuffer = replaceBuffer
? DxvkCmdBuffer::InitBuffer
: DxvkCmdBuffer::ExecBuffer;
VkBufferCopy2 copyRegion = { VK_STRUCTURE_TYPE_BUFFER_COPY_2 };
copyRegion.srcOffset = srcSlice.offset;
copyRegion.dstOffset = dstSlice.offset;
copyRegion.size = dstSlice.length;
VkCopyBufferInfo2 copyInfo = { VK_STRUCTURE_TYPE_COPY_BUFFER_INFO_2 };
copyInfo.srcBuffer = srcSlice.handle;
copyInfo.dstBuffer = dstSlice.handle;
copyInfo.regionCount = 1;
copyInfo.pRegions = &copyRegion;
m_cmd->cmdCopyBuffer(cmdBuffer, &copyInfo);
auto& barriers = replaceBuffer
? m_initBarriers
: m_execBarriers;
barriers.accessBuffer(srcSlice,
VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_ACCESS_TRANSFER_READ_BIT,
srcBuffer->info().stages,
srcBuffer->info().access);
barriers.accessBuffer(dstSlice,
VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_ACCESS_TRANSFER_WRITE_BIT,
dstBuffer->info().stages,
dstBuffer->info().access);
m_cmd->trackResource<DxvkAccess::Write>(dstBuffer);
m_cmd->trackResource<DxvkAccess::Read>(srcBuffer);
}
void DxvkContext::copyBufferRegion(
const Rc<DxvkBuffer>& dstBuffer,
VkDeviceSize dstOffset,
VkDeviceSize srcOffset,
VkDeviceSize numBytes) {
VkDeviceSize loOvl = std::max(dstOffset, srcOffset);
VkDeviceSize hiOvl = std::min(dstOffset, srcOffset) + numBytes;
if (hiOvl > loOvl) {
DxvkBufferCreateInfo bufInfo;
bufInfo.size = numBytes;
bufInfo.usage = VK_BUFFER_USAGE_TRANSFER_DST_BIT
| VK_BUFFER_USAGE_TRANSFER_SRC_BIT;
bufInfo.stages = VK_PIPELINE_STAGE_TRANSFER_BIT;
bufInfo.access = VK_ACCESS_TRANSFER_WRITE_BIT
| VK_ACCESS_TRANSFER_READ_BIT;
auto tmpBuffer = m_device->createBuffer(
bufInfo, VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT);
VkDeviceSize tmpOffset = 0;
this->copyBuffer(tmpBuffer, tmpOffset, dstBuffer, srcOffset, numBytes);
this->copyBuffer(dstBuffer, dstOffset, tmpBuffer, tmpOffset, numBytes);
} else {
this->copyBuffer(dstBuffer, dstOffset, dstBuffer, srcOffset, numBytes);
}
}
void DxvkContext::copyBufferToImage(
const Rc<DxvkImage>& dstImage,
VkImageSubresourceLayers dstSubresource,
VkOffset3D dstOffset,
VkExtent3D dstExtent,
const Rc<DxvkBuffer>& srcBuffer,
VkDeviceSize srcOffset,
VkDeviceSize rowAlignment,
VkDeviceSize sliceAlignment) {
this->spillRenderPass(true);
this->prepareImage(dstImage, vk::makeSubresourceRange(dstSubresource));
auto srcSlice = srcBuffer->getSliceHandle(srcOffset, 0);
// We may copy to only one aspect at a time, but pipeline
// barriers need to have all available aspect bits set
auto dstFormatInfo = dstImage->formatInfo();
auto dstSubresourceRange = vk::makeSubresourceRange(dstSubresource);
dstSubresourceRange.aspectMask = dstFormatInfo->aspectMask;
if (m_execBarriers.isImageDirty(dstImage, dstSubresourceRange, DxvkAccess::Write)
|| m_execBarriers.isBufferDirty(srcSlice, DxvkAccess::Read))
m_execBarriers.recordCommands(m_cmd);
// Initialize the image if the entire subresource is covered
VkImageLayout dstImageLayoutInitial = dstImage->info().layout;
VkImageLayout dstImageLayoutTransfer = dstImage->pickLayout(VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL);
if (dstImage->isFullSubresource(dstSubresource, dstExtent))
dstImageLayoutInitial = VK_IMAGE_LAYOUT_UNDEFINED;
if (dstImageLayoutTransfer != dstImageLayoutInitial) {
m_execAcquires.accessImage(
dstImage, dstSubresourceRange,
dstImageLayoutInitial,
VK_PIPELINE_STAGE_TRANSFER_BIT, 0,
dstImageLayoutTransfer,
VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_ACCESS_TRANSFER_WRITE_BIT);
}
m_execAcquires.recordCommands(m_cmd);
this->copyImageBufferData<true>(DxvkCmdBuffer::ExecBuffer, dstImage, dstSubresource,
dstOffset, dstExtent, dstImageLayoutTransfer, srcSlice, rowAlignment, sliceAlignment);
m_execBarriers.accessImage(
dstImage, dstSubresourceRange,
dstImageLayoutTransfer,
VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_ACCESS_TRANSFER_WRITE_BIT,
dstImage->info().layout,
dstImage->info().stages,
dstImage->info().access);
m_execBarriers.accessBuffer(srcSlice,
VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_ACCESS_TRANSFER_READ_BIT,
srcBuffer->info().stages,
srcBuffer->info().access);
m_cmd->trackResource<DxvkAccess::Write>(dstImage);
m_cmd->trackResource<DxvkAccess::Read>(srcBuffer);
}
void DxvkContext::copyImage(
const Rc<DxvkImage>& dstImage,
VkImageSubresourceLayers dstSubresource,
VkOffset3D dstOffset,
const Rc<DxvkImage>& srcImage,
VkImageSubresourceLayers srcSubresource,
VkOffset3D srcOffset,
VkExtent3D extent) {
this->spillRenderPass(true);
if (this->copyImageClear(dstImage, dstSubresource, dstOffset, extent, srcImage, srcSubresource))
return;
this->prepareImage(dstImage, vk::makeSubresourceRange(dstSubresource));
this->prepareImage(srcImage, vk::makeSubresourceRange(srcSubresource));
bool useFb = dstSubresource.aspectMask != srcSubresource.aspectMask;
if (m_device->perfHints().preferFbDepthStencilCopy) {
useFb |= (dstSubresource.aspectMask == (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT))
&& (dstImage->info().usage & VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT)
&& (srcImage->info().usage & VK_IMAGE_USAGE_SAMPLED_BIT);
}
if (!useFb) {
this->copyImageHw(
dstImage, dstSubresource, dstOffset,
srcImage, srcSubresource, srcOffset,
extent);
} else {
this->copyImageFb(
dstImage, dstSubresource, dstOffset,
srcImage, srcSubresource, srcOffset,
extent);
}
}
void DxvkContext::copyImageRegion(
const Rc<DxvkImage>& dstImage,
VkImageSubresourceLayers dstSubresource,
VkOffset3D dstOffset,
VkOffset3D srcOffset,
VkExtent3D extent) {
VkOffset3D loOvl = {
std::max(dstOffset.x, srcOffset.x),
std::max(dstOffset.y, srcOffset.y),
std::max(dstOffset.z, srcOffset.z) };
VkOffset3D hiOvl = {
std::min(dstOffset.x, srcOffset.x) + int32_t(extent.width),
std::min(dstOffset.y, srcOffset.y) + int32_t(extent.height),
std::min(dstOffset.z, srcOffset.z) + int32_t(extent.depth) };
bool overlap = hiOvl.x > loOvl.x
&& hiOvl.y > loOvl.y
&& hiOvl.z > loOvl.z;
if (overlap) {
DxvkImageCreateInfo imgInfo;
imgInfo.type = dstImage->info().type;
imgInfo.format = dstImage->info().format;
imgInfo.flags = 0;
imgInfo.sampleCount = dstImage->info().sampleCount;
imgInfo.extent = extent;
imgInfo.numLayers = dstSubresource.layerCount;
imgInfo.mipLevels = 1;
imgInfo.usage = VK_IMAGE_USAGE_TRANSFER_DST_BIT
| VK_IMAGE_USAGE_TRANSFER_SRC_BIT;
imgInfo.stages = VK_PIPELINE_STAGE_TRANSFER_BIT;
imgInfo.access = VK_ACCESS_TRANSFER_WRITE_BIT
| VK_ACCESS_TRANSFER_READ_BIT;
imgInfo.tiling = dstImage->info().tiling;
imgInfo.layout = VK_IMAGE_LAYOUT_GENERAL;
auto tmpImage = m_device->createImage(
imgInfo, VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT);
VkImageSubresourceLayers tmpSubresource;
tmpSubresource.aspectMask = dstSubresource.aspectMask;
tmpSubresource.mipLevel = 0;
tmpSubresource.baseArrayLayer = 0;
tmpSubresource.layerCount = dstSubresource.layerCount;
VkOffset3D tmpOffset = { 0, 0, 0 };
this->copyImage(
tmpImage, tmpSubresource, tmpOffset,
dstImage, dstSubresource, srcOffset,
extent);
this->copyImage(
dstImage, dstSubresource, dstOffset,
tmpImage, tmpSubresource, tmpOffset,
extent);
} else {
this->copyImage(
dstImage, dstSubresource, dstOffset,
dstImage, dstSubresource, srcOffset,
extent);
}
}
void DxvkContext::copyImageToBuffer(
const Rc<DxvkBuffer>& dstBuffer,
VkDeviceSize dstOffset,
VkDeviceSize rowAlignment,
VkDeviceSize sliceAlignment,
const Rc<DxvkImage>& srcImage,
VkImageSubresourceLayers srcSubresource,
VkOffset3D srcOffset,
VkExtent3D srcExtent) {
this->spillRenderPass(true);
this->prepareImage(srcImage, vk::makeSubresourceRange(srcSubresource));
auto dstSlice = dstBuffer->getSliceHandle(dstOffset, 0);
// We may copy to only one aspect of a depth-stencil image,
// but pipeline barriers need to have all aspect bits set
auto srcFormatInfo = srcImage->formatInfo();
auto srcSubresourceRange = vk::makeSubresourceRange(srcSubresource);
srcSubresourceRange.aspectMask = srcFormatInfo->aspectMask;
if (m_execBarriers.isImageDirty(srcImage, srcSubresourceRange, DxvkAccess::Write)
|| m_execBarriers.isBufferDirty(dstSlice, DxvkAccess::Write))
m_execBarriers.recordCommands(m_cmd);
// Select a suitable image layout for the transfer op
VkImageLayout srcImageLayoutTransfer = srcImage->pickLayout(VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL);
m_execAcquires.accessImage(
srcImage, srcSubresourceRange,
srcImage->info().layout,
VK_PIPELINE_STAGE_TRANSFER_BIT, 0,
srcImageLayoutTransfer,
VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_ACCESS_TRANSFER_READ_BIT);
m_execAcquires.recordCommands(m_cmd);
this->copyImageBufferData<false>(DxvkCmdBuffer::ExecBuffer, srcImage, srcSubresource,
srcOffset, srcExtent, srcImageLayoutTransfer, dstSlice, rowAlignment, sliceAlignment);
m_execBarriers.accessImage(
srcImage, srcSubresourceRange,
srcImageLayoutTransfer,
VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_ACCESS_TRANSFER_READ_BIT,
srcImage->info().layout,
srcImage->info().stages,
srcImage->info().access);
m_execBarriers.accessBuffer(dstSlice,
VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_ACCESS_TRANSFER_WRITE_BIT,
dstBuffer->info().stages,
dstBuffer->info().access);
m_cmd->trackResource<DxvkAccess::Write>(dstBuffer);
m_cmd->trackResource<DxvkAccess::Read>(srcImage);
}
void DxvkContext::copyDepthStencilImageToPackedBuffer(
const Rc<DxvkBuffer>& dstBuffer,
VkDeviceSize dstBufferOffset,
VkOffset2D dstOffset,
VkExtent2D dstExtent,
const Rc<DxvkImage>& srcImage,
VkImageSubresourceLayers srcSubresource,
VkOffset2D srcOffset,
VkExtent2D srcExtent,
VkFormat format) {
this->spillRenderPass(true);
this->prepareImage(srcImage, vk::makeSubresourceRange(srcSubresource));
this->invalidateState();
// Retrieve compute pipeline for the given format
auto pipeInfo = m_common->metaPack().getPackPipeline(format);
if (!pipeInfo.pipeHandle)
return;
// Create one depth view and one stencil view
DxvkImageViewCreateInfo dViewInfo;
dViewInfo.type = VK_IMAGE_VIEW_TYPE_2D_ARRAY;
dViewInfo.format = srcImage->info().format;
dViewInfo.usage = VK_IMAGE_USAGE_SAMPLED_BIT;
dViewInfo.aspect = VK_IMAGE_ASPECT_DEPTH_BIT;
dViewInfo.minLevel = srcSubresource.mipLevel;
dViewInfo.numLevels = 1;
dViewInfo.minLayer = srcSubresource.baseArrayLayer;
dViewInfo.numLayers = srcSubresource.layerCount;
DxvkImageViewCreateInfo sViewInfo = dViewInfo;
sViewInfo.aspect = VK_IMAGE_ASPECT_STENCIL_BIT;
Rc<DxvkImageView> dView = m_device->createImageView(srcImage, dViewInfo);
Rc<DxvkImageView> sView = m_device->createImageView(srcImage, sViewInfo);
// Create a descriptor set for the pack operation
VkImageLayout layout = srcImage->pickLayout(VK_IMAGE_LAYOUT_DEPTH_STENCIL_READ_ONLY_OPTIMAL);
DxvkMetaPackDescriptors descriptors;
descriptors.dstBuffer = dstBuffer->getDescriptor(dstBufferOffset, VK_WHOLE_SIZE).buffer;
descriptors.srcDepth = dView->getDescriptor(VK_IMAGE_VIEW_TYPE_2D_ARRAY, layout).image;
descriptors.srcStencil = sView->getDescriptor(VK_IMAGE_VIEW_TYPE_2D_ARRAY, layout).image;
VkDescriptorSet dset = m_descriptorPool->alloc(pipeInfo.dsetLayout);
m_cmd->updateDescriptorSetWithTemplate(dset, pipeInfo.dsetTemplate, &descriptors);
// Since this is a meta operation, the image may be
// in a different layout and we have to transition it
auto subresourceRange = vk::makeSubresourceRange(srcSubresource);
if (m_execBarriers.isImageDirty(srcImage, subresourceRange, DxvkAccess::Write))
m_execBarriers.recordCommands(m_cmd);
if (srcImage->info().layout != layout) {
m_execAcquires.accessImage(
srcImage, subresourceRange,
srcImage->info().layout,
VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0,
layout,
VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
VK_ACCESS_SHADER_READ_BIT);
m_execAcquires.recordCommands(m_cmd);
}
// Execute the actual pack operation
DxvkMetaPackArgs args;
args.srcOffset = srcOffset;
args.srcExtent = srcExtent;
args.dstOffset = dstOffset;
args.dstExtent = dstExtent;
m_cmd->cmdBindPipeline(
VK_PIPELINE_BIND_POINT_COMPUTE,
pipeInfo.pipeHandle);
m_cmd->cmdBindDescriptorSet(
VK_PIPELINE_BIND_POINT_COMPUTE,
pipeInfo.pipeLayout, dset,
0, nullptr);
m_cmd->cmdPushConstants(
pipeInfo.pipeLayout,
VK_SHADER_STAGE_COMPUTE_BIT,
0, sizeof(args), &args);
m_cmd->cmdDispatch(
(srcExtent.width + 7) / 8,
(srcExtent.height + 7) / 8,
srcSubresource.layerCount);
m_execBarriers.accessImage(
srcImage, subresourceRange, layout,
VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
VK_ACCESS_SHADER_READ_BIT,
srcImage->info().layout,
srcImage->info().stages,
srcImage->info().access);
m_execBarriers.accessBuffer(
dstBuffer->getSliceHandle(),
VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
VK_ACCESS_SHADER_WRITE_BIT,
dstBuffer->info().stages,
dstBuffer->info().access);
m_cmd->trackResource<DxvkAccess::None>(dView);
m_cmd->trackResource<DxvkAccess::None>(sView);
m_cmd->trackResource<DxvkAccess::Write>(dstBuffer);
m_cmd->trackResource<DxvkAccess::Read>(srcImage);
}
void DxvkContext::copyPackedBufferImage(
const Rc<DxvkBuffer>& dstBuffer,
VkDeviceSize dstBufferOffset,
VkOffset3D dstOffset,
VkExtent3D dstSize,
const Rc<DxvkBuffer>& srcBuffer,
VkDeviceSize srcBufferOffset,
VkOffset3D srcOffset,
VkExtent3D srcSize,
VkExtent3D extent,
VkDeviceSize elementSize) {
this->spillRenderPass(true);
this->invalidateState();
auto dstBufferSlice = dstBuffer->getSliceHandle(dstBufferOffset, elementSize * util::flattenImageExtent(dstSize));
auto srcBufferSlice = srcBuffer->getSliceHandle(srcBufferOffset, elementSize * util::flattenImageExtent(srcSize));
if (m_execBarriers.isBufferDirty(dstBufferSlice, DxvkAccess::Write)
|| m_execBarriers.isBufferDirty(srcBufferSlice, DxvkAccess::Read))
m_execBarriers.recordCommands(m_cmd);
// We'll use texel buffer views with an appropriately
// sized integer format to perform the copy
VkFormat format = VK_FORMAT_UNDEFINED;
switch (elementSize) {
case 1: format = VK_FORMAT_R8_UINT; break;
case 2: format = VK_FORMAT_R16_UINT; break;
case 4: format = VK_FORMAT_R32_UINT; break;
case 8: format = VK_FORMAT_R32G32_UINT; break;
case 12: format = VK_FORMAT_R32G32B32_UINT; break;
case 16: format = VK_FORMAT_R32G32B32A32_UINT; break;
}
if (!format) {
Logger::err(str::format("DxvkContext: copyPackedBufferImage: Unsupported element size ", elementSize));
return;
}
DxvkBufferViewCreateInfo viewInfo;
viewInfo.format = format;
viewInfo.rangeOffset = dstBufferOffset;
viewInfo.rangeLength = dstBufferSlice.length;
Rc<DxvkBufferView> dstView = m_device->createBufferView(dstBuffer, viewInfo);
viewInfo.rangeOffset = srcBufferOffset;
viewInfo.rangeLength = srcBufferSlice.length;
Rc<DxvkBufferView> srcView;
if (srcBuffer == dstBuffer
&& srcBufferSlice.offset < dstBufferSlice.offset + dstBufferSlice.length
&& srcBufferSlice.offset + srcBufferSlice.length > dstBufferSlice.offset) {
// Create temporary copy in case of overlapping regions
DxvkBufferCreateInfo bufferInfo;
bufferInfo.size = srcBufferSlice.length;
bufferInfo.usage = VK_BUFFER_USAGE_TRANSFER_DST_BIT
| VK_BUFFER_USAGE_UNIFORM_TEXEL_BUFFER_BIT;
bufferInfo.stages = VK_PIPELINE_STAGE_TRANSFER_BIT
| VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT;
bufferInfo.access = VK_ACCESS_TRANSFER_WRITE_BIT
| VK_ACCESS_SHADER_READ_BIT;
Rc<DxvkBuffer> tmpBuffer = m_device->createBuffer(bufferInfo, VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT);
auto tmpBufferSlice = tmpBuffer->getSliceHandle();
VkBufferCopy2 copyRegion = { VK_STRUCTURE_TYPE_BUFFER_COPY_2 };
copyRegion.srcOffset = srcBufferSlice.offset;
copyRegion.dstOffset = tmpBufferSlice.offset;
copyRegion.size = tmpBufferSlice.length;
VkCopyBufferInfo2 copyInfo = { VK_STRUCTURE_TYPE_COPY_BUFFER_INFO_2 };
copyInfo.srcBuffer = srcBufferSlice.handle;
copyInfo.dstBuffer = tmpBufferSlice.handle;
copyInfo.regionCount = 1;
copyInfo.pRegions = &copyRegion;
m_cmd->cmdCopyBuffer(DxvkCmdBuffer::ExecBuffer, &copyInfo);
emitMemoryBarrier(
VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_ACCESS_TRANSFER_WRITE_BIT,
VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
VK_ACCESS_SHADER_READ_BIT);
viewInfo.rangeOffset = 0;
srcView = m_device->createBufferView(tmpBuffer, viewInfo);
m_cmd->trackResource<DxvkAccess::Write>(tmpBuffer);
} else {
srcView = m_device->createBufferView(srcBuffer, viewInfo);
}
auto pipeInfo = m_common->metaCopy().getCopyBufferImagePipeline();
VkDescriptorSet descriptorSet = m_descriptorPool->alloc(pipeInfo.dsetLayout);
std::array<VkWriteDescriptorSet, 2> descriptorWrites;
std::array<std::pair<VkDescriptorType, VkBufferView>, 2> descriptorInfos = {{
{ VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER, dstView->handle() },
{ VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER, srcView->handle() },
}};
for (uint32_t i = 0; i < descriptorWrites.size(); i++) {
auto write = &descriptorWrites[i];
auto info = &descriptorInfos[i];
write->sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET;
write->pNext = nullptr;
write->dstSet = descriptorSet;
write->dstBinding = i;
write->dstArrayElement = 0;
write->descriptorCount = 1;
write->descriptorType = info->first;
write->pImageInfo = nullptr;
write->pBufferInfo = nullptr;
write->pTexelBufferView = &info->second;
}
m_cmd->updateDescriptorSets(descriptorWrites.size(), descriptorWrites.data());
DxvkCopyBufferImageArgs args = { };
args.dstOffset = dstOffset;
args.srcOffset = srcOffset;
args.extent = extent;
args.dstSize = { dstSize.width, dstSize.height };
args.srcSize = { srcSize.width, srcSize.height };
m_cmd->cmdBindPipeline(
VK_PIPELINE_BIND_POINT_COMPUTE,
pipeInfo.pipeHandle);
m_cmd->cmdBindDescriptorSet(
VK_PIPELINE_BIND_POINT_COMPUTE,
pipeInfo.pipeLayout, descriptorSet,
0, nullptr);
m_cmd->cmdPushConstants(
pipeInfo.pipeLayout,
VK_SHADER_STAGE_COMPUTE_BIT,
0, sizeof(args), &args);
m_cmd->cmdDispatch(
(extent.width + 7) / 8,
(extent.height + 7) / 8,
extent.depth);
m_execBarriers.accessBuffer(
dstView->getSliceHandle(),
VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
VK_ACCESS_SHADER_WRITE_BIT,
dstBuffer->info().stages,
dstBuffer->info().access);
m_execBarriers.accessBuffer(
srcView->getSliceHandle(),
VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
VK_ACCESS_SHADER_READ_BIT,
srcBuffer->info().stages,
srcBuffer->info().access);
// Track all involved resources
m_cmd->trackResource<DxvkAccess::Write>(dstBuffer);
m_cmd->trackResource<DxvkAccess::Read>(srcBuffer);
m_cmd->trackResource<DxvkAccess::None>(dstView);
m_cmd->trackResource<DxvkAccess::None>(srcView);
}
void DxvkContext::copyPackedBufferToDepthStencilImage(
const Rc<DxvkImage>& dstImage,
VkImageSubresourceLayers dstSubresource,
VkOffset2D dstOffset,
VkExtent2D dstExtent,
const Rc<DxvkBuffer>& srcBuffer,
VkDeviceSize srcBufferOffset,
VkOffset2D srcOffset,
VkExtent2D srcExtent,
VkFormat format) {
this->spillRenderPass(true);
this->invalidateState();
this->prepareImage(dstImage, vk::makeSubresourceRange(dstSubresource));
if (m_execBarriers.isBufferDirty(srcBuffer->getSliceHandle(), DxvkAccess::Read)
|| m_execBarriers.isImageDirty(dstImage, vk::makeSubresourceRange(dstSubresource), DxvkAccess::Write))
m_execBarriers.recordCommands(m_cmd);
// Retrieve compute pipeline for the given format
auto pipeInfo = m_common->metaPack().getUnpackPipeline(dstImage->info().format, format);
if (!pipeInfo.pipeHandle) {
Logger::err(str::format(
"DxvkContext: copyPackedBufferToDepthStencilImage: Unhandled formats"
"\n dstFormat = ", dstImage->info().format,
"\n srcFormat = ", format));
return;
}
// Pick depth and stencil data formats
VkFormat dataFormatD = VK_FORMAT_UNDEFINED;
VkFormat dataFormatS = VK_FORMAT_UNDEFINED;
const std::array<std::tuple<VkFormat, VkFormat, VkFormat>, 2> formats = {{
{ VK_FORMAT_D24_UNORM_S8_UINT, VK_FORMAT_R32_UINT, VK_FORMAT_R8_UINT },
{ VK_FORMAT_D32_SFLOAT_S8_UINT, VK_FORMAT_R32_SFLOAT, VK_FORMAT_R8_UINT },
}};
for (const auto& e : formats) {
if (std::get<0>(e) == dstImage->info().format) {
dataFormatD = std::get<1>(e);
dataFormatS = std::get<2>(e);
}
}
// Create temporary buffer for depth/stencil data
VkDeviceSize pixelCount = dstExtent.width * dstExtent.height * dstSubresource.layerCount;
VkDeviceSize dataSizeD = align(pixelCount * lookupFormatInfo(dataFormatD)->elementSize, 256);
VkDeviceSize dataSizeS = align(pixelCount * lookupFormatInfo(dataFormatS)->elementSize, 256);
DxvkBufferCreateInfo tmpBufferInfo;
tmpBufferInfo.size = dataSizeD + dataSizeS;
tmpBufferInfo.usage = VK_BUFFER_USAGE_STORAGE_TEXEL_BUFFER_BIT
| VK_BUFFER_USAGE_TRANSFER_SRC_BIT;
tmpBufferInfo.stages = VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT
| VK_PIPELINE_STAGE_TRANSFER_BIT;
tmpBufferInfo.access = VK_ACCESS_SHADER_WRITE_BIT
| VK_ACCESS_TRANSFER_READ_BIT;
auto tmpBuffer = m_device->createBuffer(tmpBufferInfo, VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT);
// Create formatted buffer views
DxvkBufferViewCreateInfo tmpViewInfoD;
tmpViewInfoD.format = dataFormatD;
tmpViewInfoD.rangeOffset = 0;
tmpViewInfoD.rangeLength = dataSizeD;
DxvkBufferViewCreateInfo tmpViewInfoS;
tmpViewInfoS.format = dataFormatS;
tmpViewInfoS.rangeOffset = dataSizeD;
tmpViewInfoS.rangeLength = dataSizeS;
auto tmpBufferViewD = m_device->createBufferView(tmpBuffer, tmpViewInfoD);
auto tmpBufferViewS = m_device->createBufferView(tmpBuffer, tmpViewInfoS);
// Create descriptor set for the unpack operation
DxvkMetaUnpackDescriptors descriptors;
descriptors.dstDepth = tmpBufferViewD->handle();
descriptors.dstStencil = tmpBufferViewS->handle();
descriptors.srcBuffer = srcBuffer->getDescriptor(srcBufferOffset, VK_WHOLE_SIZE).buffer;
VkDescriptorSet dset = m_descriptorPool->alloc(pipeInfo.dsetLayout);
m_cmd->updateDescriptorSetWithTemplate(dset, pipeInfo.dsetTemplate, &descriptors);
// Unpack the source buffer to temporary buffers
DxvkMetaPackArgs args;
args.srcOffset = srcOffset;
args.srcExtent = srcExtent;
args.dstOffset = VkOffset2D { 0, 0 };
args.dstExtent = dstExtent;
m_cmd->cmdBindPipeline(
VK_PIPELINE_BIND_POINT_COMPUTE,
pipeInfo.pipeHandle);
m_cmd->cmdBindDescriptorSet(
VK_PIPELINE_BIND_POINT_COMPUTE,
pipeInfo.pipeLayout, dset,
0, nullptr);
m_cmd->cmdPushConstants(
pipeInfo.pipeLayout,
VK_SHADER_STAGE_COMPUTE_BIT,
0, sizeof(args), &args);
m_cmd->cmdDispatch(
(dstExtent.width + 63) / 64,
dstExtent.height,
dstSubresource.layerCount);
m_execBarriers.accessBuffer(
tmpBuffer->getSliceHandle(),
VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
VK_ACCESS_SHADER_WRITE_BIT,
VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_ACCESS_TRANSFER_READ_BIT);
m_execBarriers.accessBuffer(
srcBuffer->getSliceHandle(),
VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
VK_ACCESS_SHADER_READ_BIT,
srcBuffer->info().stages,
srcBuffer->info().access);
// Prepare image for the data transfer operation
VkOffset3D dstOffset3D = { dstOffset.x, dstOffset.y, 0 };
VkExtent3D dstExtent3D = { dstExtent.width, dstExtent.height, 1 };
VkImageLayout initialImageLayout = dstImage->info().layout;
if (dstImage->isFullSubresource(dstSubresource, dstExtent3D))
initialImageLayout = VK_IMAGE_LAYOUT_UNDEFINED;
m_execBarriers.accessImage(
dstImage, vk::makeSubresourceRange(dstSubresource),
initialImageLayout,
dstImage->info().stages,
dstImage->info().access,
dstImage->pickLayout(VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL),
VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_ACCESS_TRANSFER_WRITE_BIT);
m_execBarriers.recordCommands(m_cmd);
// Copy temporary buffer data to depth-stencil image
VkImageSubresourceLayers dstSubresourceD = dstSubresource;
dstSubresourceD.aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT;
VkImageSubresourceLayers dstSubresourceS = dstSubresource;
dstSubresourceS.aspectMask = VK_IMAGE_ASPECT_STENCIL_BIT;
std::array<VkBufferImageCopy2, 2> copyRegions;
copyRegions[0] = { VK_STRUCTURE_TYPE_BUFFER_IMAGE_COPY_2 };
copyRegions[0].bufferOffset = tmpBufferViewD->info().rangeOffset;
copyRegions[0].imageSubresource = dstSubresourceD;
copyRegions[0].imageOffset = dstOffset3D;
copyRegions[0].imageExtent = dstExtent3D;
copyRegions[1] = { VK_STRUCTURE_TYPE_BUFFER_IMAGE_COPY_2 };
copyRegions[1].bufferOffset = tmpBufferViewS->info().rangeOffset;
copyRegions[1].imageSubresource = dstSubresourceS;
copyRegions[1].imageOffset = dstOffset3D;
copyRegions[1].imageExtent = dstExtent3D;
VkCopyBufferToImageInfo2 copyInfo = { VK_STRUCTURE_TYPE_COPY_BUFFER_TO_IMAGE_INFO_2 };
copyInfo.srcBuffer = tmpBuffer->getSliceHandle().handle;
copyInfo.dstImage = dstImage->handle();
copyInfo.dstImageLayout = dstImage->pickLayout(VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL);
copyInfo.regionCount = copyRegions.size();
copyInfo.pRegions = copyRegions.data();
m_cmd->cmdCopyBufferToImage(DxvkCmdBuffer::ExecBuffer, &copyInfo);
m_execBarriers.accessImage(
dstImage, vk::makeSubresourceRange(dstSubresource),
dstImage->pickLayout(VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL),
VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_ACCESS_TRANSFER_WRITE_BIT,
dstImage->info().layout,
dstImage->info().stages,
dstImage->info().access);
// Track all involved resources
m_cmd->trackResource<DxvkAccess::Write>(dstImage);
m_cmd->trackResource<DxvkAccess::Read>(srcBuffer);
m_cmd->trackResource<DxvkAccess::None>(tmpBufferViewD);
m_cmd->trackResource<DxvkAccess::None>(tmpBufferViewS);
}
void DxvkContext::copySparsePagesToBuffer(
const Rc<DxvkBuffer>& dstBuffer,
VkDeviceSize dstOffset,
const Rc<DxvkPagedResource>& srcResource,
uint32_t pageCount,
const uint32_t* pages) {
this->copySparsePages<true>(
srcResource, pageCount, pages,
dstBuffer, dstOffset);
}
void DxvkContext::copySparsePagesFromBuffer(
const Rc<DxvkPagedResource>& dstResource,
uint32_t pageCount,
const uint32_t* pages,
const Rc<DxvkBuffer>& srcBuffer,
VkDeviceSize srcOffset) {
this->copySparsePages<false>(
dstResource, pageCount, pages,
srcBuffer, srcOffset);
}
void DxvkContext::discardBuffer(
const Rc<DxvkBuffer>& buffer) {
if ((buffer->memFlags() & VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT)
|| (buffer->info().flags & VK_BUFFER_CREATE_SPARSE_BINDING_BIT))
return;
if (m_execBarriers.isBufferDirty(buffer->getSliceHandle(), DxvkAccess::Write))
this->invalidateBuffer(buffer, buffer->allocSlice());
}
void DxvkContext::discardImageView(
const Rc<DxvkImageView>& imageView,
VkImageAspectFlags discardAspects) {
VkImageUsageFlags viewUsage = imageView->info().usage;
// Ignore non-render target views since there's likely no good use case for
// discarding those. Also, force reinitialization even if the image is bound
// as a render target, which may have niche use cases for depth buffers.
if (viewUsage & (VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT | VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT)) {
this->spillRenderPass(true);
this->deferDiscard(imageView, discardAspects);
}
}
void DxvkContext::dispatch(
uint32_t x,
uint32_t y,
uint32_t z) {
if (this->commitComputeState()) {
this->commitComputeBarriers<false>();
this->commitComputeBarriers<true>();
m_queryManager.beginQueries(m_cmd,
VK_QUERY_TYPE_PIPELINE_STATISTICS);
m_cmd->cmdDispatch(x, y, z);
m_queryManager.endQueries(m_cmd,
VK_QUERY_TYPE_PIPELINE_STATISTICS);
}
m_cmd->addStatCtr(DxvkStatCounter::CmdDispatchCalls, 1);
}
void DxvkContext::dispatchIndirect(
VkDeviceSize offset) {
auto bufferSlice = m_state.id.argBuffer.getSliceHandle(
offset, sizeof(VkDispatchIndirectCommand));
if (m_execBarriers.isBufferDirty(bufferSlice, DxvkAccess::Read))
m_execBarriers.recordCommands(m_cmd);
if (this->commitComputeState()) {
this->commitComputeBarriers<false>();
this->commitComputeBarriers<true>();
m_queryManager.beginQueries(m_cmd,
VK_QUERY_TYPE_PIPELINE_STATISTICS);
m_cmd->cmdDispatchIndirect(
bufferSlice.handle,
bufferSlice.offset);
m_queryManager.endQueries(m_cmd,
VK_QUERY_TYPE_PIPELINE_STATISTICS);
m_execBarriers.accessBuffer(bufferSlice,
VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT,
VK_ACCESS_INDIRECT_COMMAND_READ_BIT,
m_state.id.argBuffer.bufferInfo().stages,
m_state.id.argBuffer.bufferInfo().access);
this->trackDrawBuffer();
}
m_cmd->addStatCtr(DxvkStatCounter::CmdDispatchCalls, 1);
}
void DxvkContext::draw(
uint32_t vertexCount,
uint32_t instanceCount,
uint32_t firstVertex,
uint32_t firstInstance) {
if (this->commitGraphicsState<false, false>()) {
m_cmd->cmdDraw(
vertexCount, instanceCount,
firstVertex, firstInstance);
}
m_cmd->addStatCtr(DxvkStatCounter::CmdDrawCalls, 1);
}
void DxvkContext::drawIndirect(
VkDeviceSize offset,
uint32_t count,
uint32_t stride) {
if (this->commitGraphicsState<false, true>()) {
auto descriptor = m_state.id.argBuffer.getDescriptor();
m_cmd->cmdDrawIndirect(
descriptor.buffer.buffer,
descriptor.buffer.offset + offset,
count, stride);
}
m_cmd->addStatCtr(DxvkStatCounter::CmdDrawCalls, 1);
}
void DxvkContext::drawIndirectCount(
VkDeviceSize offset,
VkDeviceSize countOffset,
uint32_t maxCount,
uint32_t stride) {
if (this->commitGraphicsState<false, true>()) {
auto argDescriptor = m_state.id.argBuffer.getDescriptor();
auto cntDescriptor = m_state.id.cntBuffer.getDescriptor();
m_cmd->cmdDrawIndirectCount(
argDescriptor.buffer.buffer,
argDescriptor.buffer.offset + offset,
cntDescriptor.buffer.buffer,
cntDescriptor.buffer.offset + countOffset,
maxCount, stride);
}
m_cmd->addStatCtr(DxvkStatCounter::CmdDrawCalls, 1);
}
void DxvkContext::drawIndexed(
uint32_t indexCount,
uint32_t instanceCount,
uint32_t firstIndex,
uint32_t vertexOffset,
uint32_t firstInstance) {
if (this->commitGraphicsState<true, false>()) {
m_cmd->cmdDrawIndexed(
indexCount, instanceCount,
firstIndex, vertexOffset,
firstInstance);
}
m_cmd->addStatCtr(DxvkStatCounter::CmdDrawCalls, 1);
}
void DxvkContext::drawIndexedIndirect(
VkDeviceSize offset,
uint32_t count,
uint32_t stride) {
if (this->commitGraphicsState<true, true>()) {
auto descriptor = m_state.id.argBuffer.getDescriptor();
m_cmd->cmdDrawIndexedIndirect(
descriptor.buffer.buffer,
descriptor.buffer.offset + offset,
count, stride);
}
m_cmd->addStatCtr(DxvkStatCounter::CmdDrawCalls, 1);
}
void DxvkContext::drawIndexedIndirectCount(
VkDeviceSize offset,
VkDeviceSize countOffset,
uint32_t maxCount,
uint32_t stride) {
if (this->commitGraphicsState<true, true>()) {
auto argDescriptor = m_state.id.argBuffer.getDescriptor();
auto cntDescriptor = m_state.id.cntBuffer.getDescriptor();
m_cmd->cmdDrawIndexedIndirectCount(
argDescriptor.buffer.buffer,
argDescriptor.buffer.offset + offset,
cntDescriptor.buffer.buffer,
cntDescriptor.buffer.offset + countOffset,
maxCount, stride);
}
m_cmd->addStatCtr(DxvkStatCounter::CmdDrawCalls, 1);
}
void DxvkContext::drawIndirectXfb(
const DxvkBufferSlice& counterBuffer,
uint32_t counterDivisor,
uint32_t counterBias) {
if (this->commitGraphicsState<false, false>()) {
auto physSlice = counterBuffer.getSliceHandle();
m_cmd->cmdDrawIndirectVertexCount(1, 0,
physSlice.handle,
physSlice.offset,
counterBias,
counterDivisor);
}
m_cmd->addStatCtr(DxvkStatCounter::CmdDrawCalls, 1);
}
void DxvkContext::initBuffer(
const Rc<DxvkBuffer>& buffer) {
auto slice = buffer->getSliceHandle();
m_cmd->cmdFillBuffer(DxvkCmdBuffer::InitBuffer,
slice.handle, slice.offset,
dxvk::align(slice.length, 4), 0);
m_initBarriers.accessBuffer(slice,
VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_ACCESS_TRANSFER_WRITE_BIT,
buffer->info().stages,
buffer->info().access);
m_cmd->trackResource<DxvkAccess::Write>(buffer);
}
void DxvkContext::initImage(
const Rc<DxvkImage>& image,
const VkImageSubresourceRange& subresources,
VkImageLayout initialLayout) {
if (initialLayout == VK_IMAGE_LAYOUT_PREINITIALIZED) {
m_initBarriers.accessImage(image, subresources,
initialLayout,
VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, 0,
image->info().layout,
image->info().stages,
image->info().access);
m_cmd->trackResource<DxvkAccess::None>(image);
} else {
VkImageLayout clearLayout = image->pickLayout(VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL);
m_initBarriers.accessImage(image, subresources,
initialLayout,
VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, 0,
clearLayout,
VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_ACCESS_TRANSFER_WRITE_BIT);
auto formatInfo = image->formatInfo();
if (formatInfo->flags.any(DxvkFormatFlag::BlockCompressed, DxvkFormatFlag::MultiPlane)) {
for (auto aspects = formatInfo->aspectMask; aspects; ) {
auto aspect = vk::getNextAspect(aspects);
auto extent = image->mipLevelExtent(subresources.baseMipLevel);
auto elementSize = formatInfo->elementSize;
if (formatInfo->flags.test(DxvkFormatFlag::MultiPlane)) {
auto plane = &formatInfo->planes[vk::getPlaneIndex(aspect)];
extent.width /= plane->blockSize.width;
extent.height /= plane->blockSize.height;
elementSize = plane->elementSize;
}
// Allocate enough staging buffer memory to fit one
// single subresource, then dispatch multiple copies
VkExtent3D blockCount = util::computeBlockCount(extent, formatInfo->blockSize);
VkDeviceSize dataSize = util::flattenImageExtent(blockCount) * elementSize;
auto zeroBuffer = createZeroBuffer(dataSize);
auto zeroHandle = zeroBuffer->getSliceHandle();
for (uint32_t level = 0; level < subresources.levelCount; level++) {
VkOffset3D offset = VkOffset3D { 0, 0, 0 };
VkExtent3D extent = image->mipLevelExtent(subresources.baseMipLevel + level);
if (formatInfo->flags.test(DxvkFormatFlag::MultiPlane)) {
auto plane = &formatInfo->planes[vk::getPlaneIndex(aspect)];
extent.width /= plane->blockSize.width;
extent.height /= plane->blockSize.height;
}
for (uint32_t layer = 0; layer < subresources.layerCount; layer++) {
VkBufferImageCopy2 copyRegion = { VK_STRUCTURE_TYPE_BUFFER_IMAGE_COPY_2 };
copyRegion.bufferOffset = zeroHandle.offset;
copyRegion.imageSubresource = vk::makeSubresourceLayers(
vk::pickSubresource(subresources, level, layer));
copyRegion.imageSubresource.aspectMask = aspect;
copyRegion.imageOffset = offset;
copyRegion.imageExtent = extent;
VkCopyBufferToImageInfo2 copyInfo = { VK_STRUCTURE_TYPE_COPY_BUFFER_TO_IMAGE_INFO_2 };
copyInfo.srcBuffer = zeroHandle.handle;
copyInfo.dstImage = image->handle();
copyInfo.dstImageLayout = clearLayout;
copyInfo.regionCount = 1;
copyInfo.pRegions = &copyRegion;
m_cmd->cmdCopyBufferToImage(DxvkCmdBuffer::ExecBuffer, &copyInfo);
}
}
m_cmd->trackResource<DxvkAccess::Read>(zeroBuffer);
}
} else {
if (subresources.aspectMask & (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) {
VkClearDepthStencilValue value = { };
m_cmd->cmdClearDepthStencilImage(image->handle(),
clearLayout, &value, 1, &subresources);
} else {
VkClearColorValue value = { };
m_cmd->cmdClearColorImage(image->handle(),
clearLayout, &value, 1, &subresources);
}
}
m_execBarriers.accessImage(image, subresources,
clearLayout,
VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_ACCESS_TRANSFER_WRITE_BIT,
image->info().layout,
image->info().stages,
image->info().access);
m_cmd->trackResource<DxvkAccess::Write>(image);
}
}
void DxvkContext::initSparseImage(
const Rc<DxvkImage>& image) {
auto vk = m_device->vkd();
// Query sparse memory requirements
uint32_t reqCount = 0;
vk->vkGetImageSparseMemoryRequirements(vk->device(), image->handle(), &reqCount, nullptr);
std::vector<VkSparseImageMemoryRequirements> req(reqCount);
vk->vkGetImageSparseMemoryRequirements(vk->device(), image->handle(), &reqCount, req.data());
// Bind metadata aspects. Since the image was just created,
// we do not need to interrupt our command list for that.
VkDeviceMemory imageMemory = image->memory().memory();
VkDeviceSize imageOffset = image->memory().offset();
for (const auto& r : req) {
if (!(r.formatProperties.aspectMask & VK_IMAGE_ASPECT_METADATA_BIT))
continue;
uint32_t layerCount = (r.formatProperties.flags & VK_SPARSE_IMAGE_FORMAT_SINGLE_MIPTAIL_BIT)
? 1u : image->info().numLayers;
for (uint32_t i = 0; i < layerCount; i++) {
DxvkSparseImageOpaqueBindKey key;
key.image = image->handle();
key.offset = r.imageMipTailOffset + i * r.imageMipTailStride;
key.size = r.imageMipTailSize;
key.flags = VK_SPARSE_MEMORY_BIND_METADATA_BIT;
DxvkSparsePageHandle page;
page.memory = imageMemory;
page.offset = imageOffset;
page.length = r.imageMipTailSize;
m_cmd->bindImageOpaqueMemory(key, page);
imageOffset += r.imageMipTailSize;
}
}
// Perform initial layout transition
m_initBarriers.accessImage(image,
image->getAvailableSubresources(),
VK_IMAGE_LAYOUT_UNDEFINED,
VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, 0,
image->info().layout,
image->info().stages,
image->info().access);
m_cmd->trackResource<DxvkAccess::Write>(image);
}
void DxvkContext::emitGraphicsBarrier(
VkPipelineStageFlags srcStages,
VkAccessFlags srcAccess,
VkPipelineStageFlags dstStages,
VkAccessFlags dstAccess) {
// Emit barrier early so we can fold this into
// the spill render pass barrier if possible
if (srcStages | dstStages) {
m_execBarriers.accessMemory(
srcStages, srcAccess,
dstStages, dstAccess);
}
this->spillRenderPass(true);
// Flush barriers if there was no active render pass.
// This is necessary because there are no resources
// associated with the barrier to allow tracking.
if (srcStages | dstStages)
m_execBarriers.recordCommands(m_cmd);
}
void DxvkContext::generateMipmaps(
const Rc<DxvkImageView>& imageView,
VkFilter filter) {
if (imageView->info().numLevels <= 1)
return;
this->spillRenderPass(false);
this->invalidateState();
// Create image views, etc.
Rc<DxvkMetaMipGenRenderPass> mipGenerator = new DxvkMetaMipGenRenderPass(m_device->vkd(), imageView);
if (m_execBarriers.isImageDirty(imageView->image(), imageView->imageSubresources(), DxvkAccess::Write))
m_execBarriers.recordCommands(m_cmd);
VkImageLayout dstLayout = imageView->pickLayout(VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL);
VkImageLayout srcLayout = imageView->pickLayout(VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL);
// If necessary, transition first mip level to the read-only layout
if (imageView->imageInfo().layout != srcLayout) {
m_execAcquires.accessImage(imageView->image(),
mipGenerator->getTopSubresource(),
imageView->imageInfo().layout,
imageView->imageInfo().stages, 0,
srcLayout,
VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT,
VK_ACCESS_SHADER_READ_BIT);
}
// If necessary, initialize all levels that are written to
if (imageView->imageInfo().layout != dstLayout) {
m_execAcquires.accessImage(imageView->image(),
mipGenerator->getAllTargetSubresources(),
VK_IMAGE_LAYOUT_UNDEFINED,
imageView->imageInfo().stages, 0,
dstLayout,
VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT,
VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT);
}
m_execAcquires.recordCommands(m_cmd);
// Common descriptor set properties that we use to
// bind the source image view to the fragment shader
VkDescriptorImageInfo descriptorImage = { };
descriptorImage.sampler = m_common->metaBlit().getSampler(filter);
descriptorImage.imageLayout = srcLayout;
VkWriteDescriptorSet descriptorWrite = { VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET };
descriptorWrite.dstBinding = 0;
descriptorWrite.dstArrayElement = 0;
descriptorWrite.descriptorCount = 1;
descriptorWrite.descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER;
descriptorWrite.pImageInfo = &descriptorImage;
// Common render pass info
VkRenderingAttachmentInfo attachmentInfo = { VK_STRUCTURE_TYPE_RENDERING_ATTACHMENT_INFO };
attachmentInfo.imageLayout = dstLayout;
attachmentInfo.loadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
attachmentInfo.storeOp = VK_ATTACHMENT_STORE_OP_STORE;
VkRenderingInfo renderingInfo = { VK_STRUCTURE_TYPE_RENDERING_INFO };
renderingInfo.colorAttachmentCount = 1;
renderingInfo.pColorAttachments = &attachmentInfo;
// Retrieve a compatible pipeline to use for rendering
DxvkMetaBlitPipeline pipeInfo = m_common->metaBlit().getPipeline(
mipGenerator->getSrcViewType(), imageView->info().format, VK_SAMPLE_COUNT_1_BIT);
for (uint32_t i = 0; i < mipGenerator->getPassCount(); i++) {
// Width, height and layer count for the current pass
VkExtent3D passExtent = mipGenerator->computePassExtent(i);
// Create descriptor set with the current source view
descriptorImage.imageView = mipGenerator->getSrcView(i);
descriptorWrite.dstSet = m_descriptorPool->alloc(pipeInfo.dsetLayout);
m_cmd->updateDescriptorSets(1, &descriptorWrite);
// Set up viewport and scissor rect
VkViewport viewport;
viewport.x = 0.0f;
viewport.y = 0.0f;
viewport.width = float(passExtent.width);
viewport.height = float(passExtent.height);
viewport.minDepth = 0.0f;
viewport.maxDepth = 1.0f;
VkRect2D scissor;
scissor.offset = { 0, 0 };
scissor.extent = { passExtent.width, passExtent.height };
// Set up rendering info
attachmentInfo.imageView = mipGenerator->getDstView(i);
renderingInfo.renderArea = scissor;
renderingInfo.layerCount = passExtent.depth;
// Set up push constants
DxvkMetaBlitPushConstants pushConstants = { };
pushConstants.srcCoord0 = { 0.0f, 0.0f, 0.0f };
pushConstants.srcCoord1 = { 1.0f, 1.0f, 1.0f };
pushConstants.layerCount = passExtent.depth;
if (i) {
m_execAcquires.accessImage(imageView->image(),
mipGenerator->getSourceSubresource(i),
dstLayout,
VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT,
VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT,
srcLayout,
VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT,
VK_ACCESS_SHADER_READ_BIT);
m_execAcquires.recordCommands(m_cmd);
}
m_cmd->cmdBeginRendering(&renderingInfo);
m_cmd->cmdBindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeInfo.pipeHandle);
m_cmd->cmdBindDescriptorSet(VK_PIPELINE_BIND_POINT_GRAPHICS,
pipeInfo.pipeLayout, descriptorWrite.dstSet, 0, nullptr);
m_cmd->cmdSetViewport(1, &viewport);
m_cmd->cmdSetScissor(1, &scissor);
m_cmd->cmdPushConstants(
pipeInfo.pipeLayout,
VK_SHADER_STAGE_FRAGMENT_BIT,
0, sizeof(pushConstants),
&pushConstants);
m_cmd->cmdDraw(3, passExtent.depth, 0, 0);
m_cmd->cmdEndRendering();
}
// Issue barriers to ensure we can safely access all mip
// levels of the image in all ways the image can be used
if (srcLayout == dstLayout) {
m_execBarriers.accessImage(imageView->image(),
imageView->imageSubresources(),
srcLayout,
VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT,
VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT |
VK_ACCESS_SHADER_READ_BIT,
imageView->imageInfo().layout,
imageView->imageInfo().stages,
imageView->imageInfo().access);
} else {
m_execBarriers.accessImage(imageView->image(),
mipGenerator->getAllSourceSubresources(),
srcLayout,
VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT,
VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT |
VK_ACCESS_SHADER_READ_BIT,
imageView->imageInfo().layout,
imageView->imageInfo().stages,
imageView->imageInfo().access);
m_execBarriers.accessImage(imageView->image(),
mipGenerator->getBottomSubresource(),
dstLayout,
VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT,
VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT,
imageView->imageInfo().layout,
imageView->imageInfo().stages,
imageView->imageInfo().access);
}
m_cmd->trackResource<DxvkAccess::None>(mipGenerator);
m_cmd->trackResource<DxvkAccess::Write>(imageView->image());
}
void DxvkContext::invalidateBuffer(
const Rc<DxvkBuffer>& buffer,
const DxvkBufferSliceHandle& slice) {
// Allocate new backing resource
DxvkBufferSliceHandle prevSlice = buffer->rename(slice);
m_cmd->freeBufferSlice(buffer, prevSlice);
// We also need to update all bindings that the buffer
// may be bound to either directly or through views.
VkBufferUsageFlags usage = buffer->info().usage &
~(VK_BUFFER_USAGE_TRANSFER_DST_BIT |
VK_BUFFER_USAGE_TRANSFER_SRC_BIT);
// Fast early-out for plain uniform buffers, very common
if (likely(usage == VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT)) {
m_descriptorState.dirtyBuffers(buffer->getShaderStages());
return;
}
if (usage & (VK_BUFFER_USAGE_STORAGE_BUFFER_BIT | VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT))
m_descriptorState.dirtyBuffers(buffer->getShaderStages());
if (usage & (VK_BUFFER_USAGE_STORAGE_BUFFER_BIT | VK_BUFFER_USAGE_UNIFORM_TEXEL_BUFFER_BIT | VK_BUFFER_USAGE_STORAGE_TEXEL_BUFFER_BIT))
m_descriptorState.dirtyViews(buffer->getShaderStages());
if (usage & VK_BUFFER_USAGE_INDEX_BUFFER_BIT)
m_flags.set(DxvkContextFlag::GpDirtyIndexBuffer);
if (usage & VK_BUFFER_USAGE_VERTEX_BUFFER_BIT)
m_flags.set(DxvkContextFlag::GpDirtyVertexBuffers);
if (usage & VK_BUFFER_USAGE_INDIRECT_BUFFER_BIT)
m_flags.set(DxvkContextFlag::DirtyDrawBuffer);
if (usage & VK_BUFFER_USAGE_TRANSFORM_FEEDBACK_BUFFER_BIT_EXT)
m_flags.set(DxvkContextFlag::GpDirtyXfbBuffers);
}
void DxvkContext::resolveImage(
const Rc<DxvkImage>& dstImage,
const Rc<DxvkImage>& srcImage,
const VkImageResolve& region,
VkFormat format) {
this->spillRenderPass(true);
this->prepareImage(dstImage, vk::makeSubresourceRange(region.dstSubresource));
this->prepareImage(srcImage, vk::makeSubresourceRange(region.srcSubresource));
if (format == VK_FORMAT_UNDEFINED)
format = srcImage->info().format;
bool useFb = srcImage->info().format != format
|| dstImage->info().format != format;
if (m_device->perfHints().preferFbResolve) {
useFb |= (dstImage->info().usage & VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT)
&& (srcImage->info().usage & VK_IMAGE_USAGE_SAMPLED_BIT);
}
if (!useFb) {
this->resolveImageHw(
dstImage, srcImage, region);
} else {
this->resolveImageFb(
dstImage, srcImage, region, format,
VK_RESOLVE_MODE_NONE,
VK_RESOLVE_MODE_NONE);
}
}
void DxvkContext::resolveDepthStencilImage(
const Rc<DxvkImage>& dstImage,
const Rc<DxvkImage>& srcImage,
const VkImageResolve& region,
VkResolveModeFlagBits depthMode,
VkResolveModeFlagBits stencilMode) {
this->spillRenderPass(true);
this->prepareImage(dstImage, vk::makeSubresourceRange(region.dstSubresource));
this->prepareImage(srcImage, vk::makeSubresourceRange(region.srcSubresource));
// Technically legal, but no-op
if (!depthMode && !stencilMode)
return;
// Subsequent functions expect stencil mode to be None
// if either of the images have no stencil aspect
if (!(region.dstSubresource.aspectMask
& region.srcSubresource.aspectMask
& VK_IMAGE_ASPECT_STENCIL_BIT))
stencilMode = VK_RESOLVE_MODE_NONE;
// We can only use the depth-stencil resolve path if we are resolving
// a full subresource and both images have the same format.
bool useFb = !dstImage->isFullSubresource(region.dstSubresource, region.extent)
|| !srcImage->isFullSubresource(region.srcSubresource, region.extent)
|| dstImage->info().format != srcImage->info().format;
if (!useFb) {
// Additionally, the given mode combination must be supported.
const auto& properties = m_device->properties().vk12;
useFb |= (properties.supportedDepthResolveModes & depthMode) != depthMode
|| (properties.supportedStencilResolveModes & stencilMode) != stencilMode;
if (depthMode != stencilMode) {
useFb |= (!depthMode || !stencilMode)
? !properties.independentResolveNone
: !properties.independentResolve;
}
}
if (useFb) {
this->resolveImageFb(
dstImage, srcImage, region, VK_FORMAT_UNDEFINED,
depthMode, stencilMode);
} else {
this->resolveImageDs(
dstImage, srcImage, region,
depthMode, stencilMode);
}
}
void DxvkContext::transformImage(
const Rc<DxvkImage>& dstImage,
const VkImageSubresourceRange& dstSubresources,
VkImageLayout srcLayout,
VkImageLayout dstLayout) {
this->spillRenderPass(false);
if (srcLayout != dstLayout) {
m_execBarriers.recordCommands(m_cmd);
m_execBarriers.accessImage(
dstImage, dstSubresources,
srcLayout,
dstImage->info().stages,
dstImage->info().access,
dstLayout,
dstImage->info().stages,
dstImage->info().access);
m_cmd->trackResource<DxvkAccess::Write>(dstImage);
}
}
void DxvkContext::performClear(
const Rc<DxvkImageView>& imageView,
int32_t attachmentIndex,
VkImageAspectFlags discardAspects,
VkImageAspectFlags clearAspects,
VkClearValue clearValue) {
DxvkColorAttachmentOps colorOp;
colorOp.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD;
colorOp.loadLayout = imageView->imageInfo().layout;
colorOp.storeLayout = imageView->imageInfo().layout;
DxvkDepthAttachmentOps depthOp;
depthOp.loadOpD = VK_ATTACHMENT_LOAD_OP_LOAD;
depthOp.loadOpS = VK_ATTACHMENT_LOAD_OP_LOAD;
depthOp.loadLayout = imageView->imageInfo().layout;
depthOp.storeLayout = imageView->imageInfo().layout;
if (clearAspects & VK_IMAGE_ASPECT_COLOR_BIT)
colorOp.loadOp = VK_ATTACHMENT_LOAD_OP_CLEAR;
else if (discardAspects & VK_IMAGE_ASPECT_COLOR_BIT)
colorOp.loadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
if (clearAspects & VK_IMAGE_ASPECT_DEPTH_BIT)
depthOp.loadOpD = VK_ATTACHMENT_LOAD_OP_CLEAR;
else if (discardAspects & VK_IMAGE_ASPECT_DEPTH_BIT)
depthOp.loadOpD = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
if (clearAspects & VK_IMAGE_ASPECT_STENCIL_BIT)
depthOp.loadOpS = VK_ATTACHMENT_LOAD_OP_CLEAR;
else if (discardAspects & VK_IMAGE_ASPECT_DEPTH_BIT)
depthOp.loadOpS = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
if (attachmentIndex >= 0 && !m_state.om.framebufferInfo.isWritable(attachmentIndex, clearAspects | discardAspects)) {
// Do not fold the clear/discard into the render pass if any of the affected aspects
// isn't writable. We can only hit this particular path when starting a render pass,
// so we can safely manipulate load layouts here.
int32_t colorIndex = m_state.om.framebufferInfo.getColorAttachmentIndex(attachmentIndex);
VkImageLayout renderLayout = m_state.om.framebufferInfo.getAttachment(attachmentIndex).layout;
if (colorIndex < 0) {
depthOp.loadLayout = m_state.om.renderPassOps.depthOps.loadLayout;
depthOp.storeLayout = renderLayout;
m_state.om.renderPassOps.depthOps.loadLayout = renderLayout;
} else {
colorOp.loadLayout = m_state.om.renderPassOps.colorOps[colorIndex].loadLayout;
colorOp.storeLayout = renderLayout;
m_state.om.renderPassOps.colorOps[colorIndex].loadLayout = renderLayout;
}
attachmentIndex = -1;
}
bool is3D = imageView->imageInfo().type == VK_IMAGE_TYPE_3D;
if ((clearAspects | discardAspects) == imageView->info().aspect && !is3D) {
colorOp.loadLayout = VK_IMAGE_LAYOUT_UNDEFINED;
depthOp.loadLayout = VK_IMAGE_LAYOUT_UNDEFINED;
}
if (attachmentIndex < 0) {
if (m_execBarriers.isImageDirty(imageView->image(), imageView->imageSubresources(), DxvkAccess::Write))
m_execBarriers.recordCommands(m_cmd);
// Set up a temporary render pass to execute the clear
VkImageLayout imageLayout = ((clearAspects | discardAspects) & VK_IMAGE_ASPECT_COLOR_BIT)
? imageView->pickLayout(VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL)
: imageView->pickLayout(VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL);
VkRenderingAttachmentInfo attachmentInfo = { VK_STRUCTURE_TYPE_RENDERING_ATTACHMENT_INFO };
attachmentInfo.imageView = imageView->handle();
attachmentInfo.imageLayout = imageLayout;
attachmentInfo.storeOp = VK_ATTACHMENT_STORE_OP_STORE;
attachmentInfo.clearValue = clearValue;
VkRenderingAttachmentInfo stencilInfo = attachmentInfo;
VkExtent3D extent = imageView->mipLevelExtent(0);
VkRenderingInfo renderingInfo = { VK_STRUCTURE_TYPE_RENDERING_INFO };
renderingInfo.renderArea.extent = { extent.width, extent.height };
renderingInfo.layerCount = imageView->info().numLayers;
VkImageLayout loadLayout;
VkImageLayout storeLayout;
VkPipelineStageFlags clearStages = 0;
VkAccessFlags clearAccess = 0;
if ((clearAspects | discardAspects) & VK_IMAGE_ASPECT_COLOR_BIT) {
clearStages |= VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT;
clearAccess |= VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT;
attachmentInfo.loadOp = colorOp.loadOp;
renderingInfo.colorAttachmentCount = 1;
renderingInfo.pColorAttachments = &attachmentInfo;
loadLayout = colorOp.loadLayout;
storeLayout = colorOp.storeLayout;
} else {
clearStages |= VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT
| VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT;
clearAccess |= VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT;
if (imageView->info().aspect & VK_IMAGE_ASPECT_DEPTH_BIT) {
renderingInfo.pDepthAttachment = &attachmentInfo;
attachmentInfo.loadOp = depthOp.loadOpD;
}
if (imageView->info().aspect & VK_IMAGE_ASPECT_STENCIL_BIT) {
renderingInfo.pStencilAttachment = &stencilInfo;
stencilInfo.loadOp = depthOp.loadOpS;
}
loadLayout = depthOp.loadLayout;
storeLayout = depthOp.storeLayout;
}
if (loadLayout != imageLayout) {
m_execAcquires.accessImage(
imageView->image(),
imageView->imageSubresources(),
loadLayout, clearStages, 0,
imageLayout, clearStages, clearAccess);
m_execAcquires.recordCommands(m_cmd);
}
m_cmd->cmdBeginRendering(&renderingInfo);
m_cmd->cmdEndRendering();
m_execBarriers.accessImage(
imageView->image(),
imageView->imageSubresources(),
imageLayout, clearStages, clearAccess,
storeLayout,
imageView->imageInfo().stages,
imageView->imageInfo().access);
m_cmd->trackResource<DxvkAccess::None>(imageView);
m_cmd->trackResource<DxvkAccess::Write>(imageView->image());
} else {
// Perform the operation when starting the next render pass
if ((clearAspects | discardAspects) & VK_IMAGE_ASPECT_COLOR_BIT) {
uint32_t colorIndex = m_state.om.framebufferInfo.getColorAttachmentIndex(attachmentIndex);
m_state.om.renderPassOps.colorOps[colorIndex].loadOp = colorOp.loadOp;
if (m_state.om.renderPassOps.colorOps[colorIndex].loadOp != VK_ATTACHMENT_LOAD_OP_LOAD && !is3D)
m_state.om.renderPassOps.colorOps[colorIndex].loadLayout = VK_IMAGE_LAYOUT_UNDEFINED;
m_state.om.renderPassOps.colorOps[colorIndex].clearValue = clearValue.color;
}
if ((clearAspects | discardAspects) & VK_IMAGE_ASPECT_DEPTH_BIT) {
m_state.om.renderPassOps.depthOps.loadOpD = depthOp.loadOpD;
m_state.om.renderPassOps.depthOps.clearValue.depth = clearValue.depthStencil.depth;
}
if ((clearAspects | discardAspects) & VK_IMAGE_ASPECT_STENCIL_BIT) {
m_state.om.renderPassOps.depthOps.loadOpS = depthOp.loadOpS;
m_state.om.renderPassOps.depthOps.clearValue.stencil = clearValue.depthStencil.stencil;
}
if ((clearAspects | discardAspects) & (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) {
if (m_state.om.renderPassOps.depthOps.loadOpD != VK_ATTACHMENT_LOAD_OP_LOAD
&& m_state.om.renderPassOps.depthOps.loadOpS != VK_ATTACHMENT_LOAD_OP_LOAD)
m_state.om.renderPassOps.depthOps.loadLayout = VK_IMAGE_LAYOUT_UNDEFINED;
}
}
}
void DxvkContext::deferClear(
const Rc<DxvkImageView>& imageView,
VkImageAspectFlags clearAspects,
VkClearValue clearValue) {
for (auto& entry : m_deferredClears) {
if (entry.imageView->matchesView(imageView)) {
entry.imageView = imageView;
entry.discardAspects &= ~clearAspects;
entry.clearAspects |= clearAspects;
if (clearAspects & VK_IMAGE_ASPECT_COLOR_BIT)
entry.clearValue.color = clearValue.color;
if (clearAspects & VK_IMAGE_ASPECT_DEPTH_BIT)
entry.clearValue.depthStencil.depth = clearValue.depthStencil.depth;
if (clearAspects & VK_IMAGE_ASPECT_STENCIL_BIT)
entry.clearValue.depthStencil.stencil = clearValue.depthStencil.stencil;
return;
} else if (entry.imageView->checkSubresourceOverlap(imageView)) {
this->spillRenderPass(false);
break;
}
}
m_deferredClears.push_back({ imageView, 0, clearAspects, clearValue });
}
void DxvkContext::deferDiscard(
const Rc<DxvkImageView>& imageView,
VkImageAspectFlags discardAspects) {
for (auto& entry : m_deferredClears) {
if (entry.imageView->matchesView(imageView)) {
entry.imageView = imageView;
entry.discardAspects |= discardAspects;
entry.clearAspects &= ~discardAspects;
return;
} else if (entry.imageView->checkSubresourceOverlap(imageView)) {
this->spillRenderPass(false);
break;
}
}
m_deferredClears.push_back({ imageView, discardAspects });
}
void DxvkContext::flushClears(
bool useRenderPass) {
for (const auto& clear : m_deferredClears) {
int32_t attachmentIndex = -1;
if (useRenderPass && m_state.om.framebufferInfo.isFullSize(clear.imageView))
attachmentIndex = m_state.om.framebufferInfo.findAttachment(clear.imageView);
this->performClear(clear.imageView, attachmentIndex,
clear.discardAspects, clear.clearAspects, clear.clearValue);
}
m_deferredClears.clear();
}
void DxvkContext::flushSharedImages() {
for (auto i = m_deferredClears.begin(); i != m_deferredClears.end(); ) {
if (i->imageView->imageInfo().shared) {
this->performClear(i->imageView, -1, i->discardAspects, i->clearAspects, i->clearValue);
i = m_deferredClears.erase(i);
} else {
i++;
}
}
this->transitionRenderTargetLayouts(true);
}
void DxvkContext::updateBuffer(
const Rc<DxvkBuffer>& buffer,
VkDeviceSize offset,
VkDeviceSize size,
const void* data) {
bool replaceBuffer = this->tryInvalidateDeviceLocalBuffer(buffer, size);
auto bufferSlice = buffer->getSliceHandle(offset, size);
if (!replaceBuffer) {
this->spillRenderPass(true);
if (m_execBarriers.isBufferDirty(bufferSlice, DxvkAccess::Write))
m_execBarriers.recordCommands(m_cmd);
}
DxvkCmdBuffer cmdBuffer = replaceBuffer
? DxvkCmdBuffer::InitBuffer
: DxvkCmdBuffer::ExecBuffer;
m_cmd->cmdUpdateBuffer(cmdBuffer,
bufferSlice.handle,
bufferSlice.offset,
bufferSlice.length,
data);
auto& barriers = replaceBuffer
? m_initBarriers
: m_execBarriers;
barriers.accessBuffer(bufferSlice,
VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_ACCESS_TRANSFER_WRITE_BIT,
buffer->info().stages,
buffer->info().access);
m_cmd->trackResource<DxvkAccess::Write>(buffer);
}
void DxvkContext::updateDepthStencilImage(
const Rc<DxvkImage>& image,
const VkImageSubresourceLayers& subresources,
VkOffset2D imageOffset,
VkExtent2D imageExtent,
const void* data,
VkDeviceSize pitchPerRow,
VkDeviceSize pitchPerLayer,
VkFormat format) {
auto formatInfo = lookupFormatInfo(format);
VkExtent3D extent3D;
extent3D.width = imageExtent.width;
extent3D.height = imageExtent.height;
extent3D.depth = subresources.layerCount;
VkDeviceSize pixelCount = extent3D.width * extent3D.height * extent3D.depth;
DxvkBufferCreateInfo tmpBufferInfo;
tmpBufferInfo.size = pixelCount * formatInfo->elementSize;
tmpBufferInfo.usage = VK_BUFFER_USAGE_STORAGE_BUFFER_BIT;
tmpBufferInfo.stages = VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT;
tmpBufferInfo.access = VK_ACCESS_SHADER_READ_BIT;
auto tmpBuffer = m_device->createBuffer(tmpBufferInfo,
VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
VK_MEMORY_PROPERTY_HOST_COHERENT_BIT);
util::packImageData(tmpBuffer->mapPtr(0), data,
extent3D, formatInfo->elementSize,
pitchPerRow, pitchPerLayer);
copyPackedBufferToDepthStencilImage(
image, subresources, imageOffset, imageExtent,
tmpBuffer, 0, VkOffset2D { 0, 0 }, imageExtent,
format);
}
void DxvkContext::uploadBuffer(
const Rc<DxvkBuffer>& buffer,
const void* data) {
auto bufferSlice = buffer->getSliceHandle();
auto stagingSlice = m_staging.alloc(CACHE_LINE_SIZE, bufferSlice.length);
auto stagingHandle = stagingSlice.getSliceHandle();
std::memcpy(stagingHandle.mapPtr, data, bufferSlice.length);
VkBufferCopy2 copyRegion = { VK_STRUCTURE_TYPE_BUFFER_COPY_2 };
copyRegion.srcOffset = stagingHandle.offset;
copyRegion.dstOffset = bufferSlice.offset;
copyRegion.size = bufferSlice.length;
VkCopyBufferInfo2 copyInfo = { VK_STRUCTURE_TYPE_COPY_BUFFER_INFO_2 };
copyInfo.srcBuffer = stagingHandle.handle;
copyInfo.dstBuffer = bufferSlice.handle;
copyInfo.regionCount = 1;
copyInfo.pRegions = &copyRegion;
m_cmd->cmdCopyBuffer(DxvkCmdBuffer::SdmaBuffer, &copyInfo);
m_sdmaBarriers.releaseBuffer(
m_initBarriers, bufferSlice,
m_device->queues().transfer.queueFamily,
VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_ACCESS_TRANSFER_WRITE_BIT,
m_device->queues().graphics.queueFamily,
buffer->info().stages,
buffer->info().access);
m_cmd->trackResource<DxvkAccess::Read>(stagingSlice.buffer());
m_cmd->trackResource<DxvkAccess::Write>(buffer);
}
void DxvkContext::uploadImage(
const Rc<DxvkImage>& image,
const VkImageSubresourceLayers& subresources,
const void* data,
VkDeviceSize pitchPerRow,
VkDeviceSize pitchPerLayer) {
VkOffset3D imageOffset = { 0, 0, 0 };
VkExtent3D imageExtent = image->mipLevelExtent(subresources.mipLevel);
DxvkCmdBuffer cmdBuffer = DxvkCmdBuffer::SdmaBuffer;
DxvkBarrierSet* barriers = &m_sdmaAcquires;
if (subresources.aspectMask & (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) {
cmdBuffer = DxvkCmdBuffer::InitBuffer;
barriers = &m_initBarriers;
}
// Discard previous subresource contents
barriers->accessImage(image,
vk::makeSubresourceRange(subresources),
VK_IMAGE_LAYOUT_UNDEFINED,
VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, 0,
image->pickLayout(VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL),
VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_ACCESS_TRANSFER_WRITE_BIT);
barriers->recordCommands(m_cmd);
this->copyImageHostData(cmdBuffer,
image, subresources, imageOffset, imageExtent,
data, pitchPerRow, pitchPerLayer);
// Transfer ownership to graphics queue
if (cmdBuffer == DxvkCmdBuffer::SdmaBuffer) {
m_sdmaBarriers.releaseImage(m_initBarriers,
image, vk::makeSubresourceRange(subresources),
m_device->queues().transfer.queueFamily,
image->pickLayout(VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL),
VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_ACCESS_TRANSFER_WRITE_BIT,
m_device->queues().graphics.queueFamily,
image->info().layout,
image->info().stages,
image->info().access);
} else {
barriers->accessImage(image,
vk::makeSubresourceRange(subresources),
image->pickLayout(VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL),
VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_ACCESS_TRANSFER_WRITE_BIT,
image->info().layout,
image->info().stages,
image->info().access);
}
m_cmd->trackResource<DxvkAccess::Write>(image);
}
void DxvkContext::setViewports(
uint32_t viewportCount,
const VkViewport* viewports,
const VkRect2D* scissorRects) {
for (uint32_t i = 0; i < viewportCount; i++) {
m_state.vp.viewports[i] = viewports[i];
m_state.vp.scissorRects[i] = scissorRects[i];
// Vulkan viewports are not allowed to have a width or
// height of zero, so we fall back to a dummy viewport
// and instead set an empty scissor rect, which is legal.
if (viewports[i].width == 0.0f || viewports[i].height == 0.0f) {
m_state.vp.viewports[i] = VkViewport {
0.0f, 0.0f, 1.0f, 1.0f, 0.0f, 1.0f };
m_state.vp.scissorRects[i] = VkRect2D {
VkOffset2D { 0, 0 },
VkExtent2D { 0, 0 } };
}
}
m_state.vp.viewportCount = viewportCount;
m_flags.set(DxvkContextFlag::GpDirtyViewport);
}
void DxvkContext::setBlendConstants(
DxvkBlendConstants blendConstants) {
if (m_state.dyn.blendConstants != blendConstants) {
m_state.dyn.blendConstants = blendConstants;
m_flags.set(DxvkContextFlag::GpDirtyBlendConstants);
}
}
void DxvkContext::setDepthBias(
DxvkDepthBias depthBias) {
if (m_state.dyn.depthBias != depthBias) {
m_state.dyn.depthBias = depthBias;
m_flags.set(DxvkContextFlag::GpDirtyDepthBias);
}
}
void DxvkContext::setDepthBounds(
DxvkDepthBounds depthBounds) {
if (m_state.dyn.depthBounds != depthBounds) {
m_state.dyn.depthBounds = depthBounds;
m_flags.set(DxvkContextFlag::GpDirtyDepthBounds);
}
if (m_state.gp.state.ds.enableDepthBoundsTest() != depthBounds.enableDepthBounds) {
m_state.gp.state.ds.setEnableDepthBoundsTest(depthBounds.enableDepthBounds);
m_flags.set(DxvkContextFlag::GpDirtyPipelineState);
}
}
void DxvkContext::setStencilReference(
uint32_t reference) {
if (m_state.dyn.stencilReference != reference) {
m_state.dyn.stencilReference = reference;
m_flags.set(DxvkContextFlag::GpDirtyStencilRef);
}
}
void DxvkContext::setInputAssemblyState(const DxvkInputAssemblyState& ia) {
m_state.gp.state.ia = DxvkIaInfo(
ia.primitiveTopology,
ia.primitiveRestart,
ia.patchVertexCount);
m_flags.set(DxvkContextFlag::GpDirtyPipelineState);
}
void DxvkContext::setInputLayout(
uint32_t attributeCount,
const DxvkVertexAttribute* attributes,
uint32_t bindingCount,
const DxvkVertexBinding* bindings) {
m_flags.set(
DxvkContextFlag::GpDirtyPipelineState,
DxvkContextFlag::GpDirtyVertexBuffers);
for (uint32_t i = 0; i < bindingCount; i++) {
m_state.gp.state.ilBindings[i] = DxvkIlBinding(
bindings[i].binding, 0, bindings[i].inputRate,
bindings[i].fetchRate);
m_state.vi.vertexExtents[i] = bindings[i].extent;
}
for (uint32_t i = bindingCount; i < m_state.gp.state.il.bindingCount(); i++) {
m_state.gp.state.ilBindings[i] = DxvkIlBinding();
m_state.vi.vertexExtents[i] = 0;
}
for (uint32_t i = 0; i < attributeCount; i++) {
m_state.gp.state.ilAttributes[i] = DxvkIlAttribute(
attributes[i].location, attributes[i].binding,
attributes[i].format, attributes[i].offset);
}
for (uint32_t i = attributeCount; i < m_state.gp.state.il.attributeCount(); i++)
m_state.gp.state.ilAttributes[i] = DxvkIlAttribute();
m_state.gp.state.il = DxvkIlInfo(attributeCount, bindingCount);
}
void DxvkContext::setRasterizerState(const DxvkRasterizerState& rs) {
if (m_state.dyn.cullMode != rs.cullMode || m_state.dyn.frontFace != rs.frontFace) {
m_state.dyn.cullMode = rs.cullMode;
m_state.dyn.frontFace = rs.frontFace;
m_flags.set(DxvkContextFlag::GpDirtyRasterizerState);
}
if (unlikely(rs.sampleCount != m_state.gp.state.rs.sampleCount())) {
if (!m_state.gp.state.ms.sampleCount())
m_flags.set(DxvkContextFlag::GpDirtyMultisampleState);
if (!m_features.test(DxvkContextFeature::VariableMultisampleRate))
m_flags.set(DxvkContextFlag::GpDirtyFramebuffer);
}
DxvkRsInfo rsInfo(
rs.depthClipEnable,
rs.depthBiasEnable,
rs.polygonMode,
rs.sampleCount,
rs.conservativeMode,
rs.flatShading);
if (!m_state.gp.state.rs.eq(rsInfo)) {
m_flags.set(DxvkContextFlag::GpDirtyPipelineState);
// Since depth bias enable is only dynamic for base pipelines,
// it is applied as part of the dynamic depth-stencil state
if (m_state.gp.state.rs.depthBiasEnable() != rs.depthBiasEnable)
m_flags.set(DxvkContextFlag::GpDirtyDepthStencilState);
m_state.gp.state.rs = rsInfo;
}
}
void DxvkContext::setMultisampleState(const DxvkMultisampleState& ms) {
m_state.gp.state.ms = DxvkMsInfo(
m_state.gp.state.ms.sampleCount(),
ms.sampleMask,
ms.enableAlphaToCoverage);
m_flags.set(
DxvkContextFlag::GpDirtyPipelineState,
DxvkContextFlag::GpDirtyMultisampleState);
}
void DxvkContext::setDepthStencilState(const DxvkDepthStencilState& ds) {
m_state.gp.state.ds = DxvkDsInfo(
ds.enableDepthTest,
ds.enableDepthWrite,
m_state.gp.state.ds.enableDepthBoundsTest(),
ds.enableStencilTest,
ds.depthCompareOp);
m_state.gp.state.dsFront = DxvkDsStencilOp(ds.stencilOpFront);
m_state.gp.state.dsBack = DxvkDsStencilOp(ds.stencilOpBack);
m_flags.set(
DxvkContextFlag::GpDirtyPipelineState,
DxvkContextFlag::GpDirtyDepthStencilState);
}
void DxvkContext::setLogicOpState(const DxvkLogicOpState& lo) {
m_state.gp.state.om = DxvkOmInfo(
lo.enableLogicOp,
lo.logicOp,
m_state.gp.state.om.feedbackLoop());
m_flags.set(DxvkContextFlag::GpDirtyPipelineState);
}
void DxvkContext::setBlendMode(
uint32_t attachment,
const DxvkBlendMode& blendMode) {
m_state.gp.state.omBlend[attachment] = DxvkOmAttachmentBlend(
blendMode.enableBlending,
blendMode.colorSrcFactor,
blendMode.colorDstFactor,
blendMode.colorBlendOp,
blendMode.alphaSrcFactor,
blendMode.alphaDstFactor,
blendMode.alphaBlendOp,
blendMode.writeMask);
m_flags.set(DxvkContextFlag::GpDirtyPipelineState);
}
void DxvkContext::setBarrierControl(DxvkBarrierControlFlags control) {
m_barrierControl = control;
}
void DxvkContext::updatePageTable(
const DxvkSparseBindInfo& bindInfo,
DxvkSparseBindFlags flags) {
// Split command buffers here so that we execute
// the sparse binding operation at the right time
if (!flags.test(DxvkSparseBindFlag::SkipSynchronization))
this->splitCommands();
DxvkSparsePageAllocator* srcAllocator = bindInfo.srcAllocator.ptr();
DxvkSparsePageTable* dstPageTable = bindInfo.dstResource->getSparsePageTable();
DxvkSparsePageTable* srcPageTable = nullptr;
if (bindInfo.srcResource != nullptr)
srcPageTable = bindInfo.srcResource->getSparsePageTable();
// In order to support copies properly, we need to buffer the new
// mappings first before we apply them to the destination resource.
size_t bindCount = bindInfo.binds.size();
std::vector<DxvkSparseMapping> mappings(bindCount);
for (size_t i = 0; i < bindCount; i++) {
DxvkSparseBind bind = bindInfo.binds[i];
switch (bind.mode) {
case DxvkSparseBindMode::Null:
// The mapping array is already default-initialized
// so we don't actually need to do anything here
break;
case DxvkSparseBindMode::Bind:
mappings[i] = srcAllocator->acquirePage(bind.srcPage);
break;
case DxvkSparseBindMode::Copy:
mappings[i] = srcPageTable->getMapping(bind.srcPage);
break;
}
}
// Process the actual page table updates here and resolve
// our internal structures to Vulkan resource and memory
// handles. The rest will be done at submission time.
for (size_t i = 0; i < bindCount; i++) {
DxvkSparseBind bind = bindInfo.binds[i];
DxvkSparseMapping mapping = std::move(mappings[i]);
DxvkSparsePageInfo pageInfo = dstPageTable->getPageInfo(bind.dstPage);
switch (pageInfo.type) {
case DxvkSparsePageType::None:
break;
case DxvkSparsePageType::Buffer: {
DxvkSparseBufferBindKey key;
key.buffer = dstPageTable->getBufferHandle();
key.offset = pageInfo.buffer.offset;
key.size = pageInfo.buffer.length;
m_cmd->bindBufferMemory(key, mapping.getHandle());
} break;
case DxvkSparsePageType::Image: {
DxvkSparseImageBindKey key;
key.image = dstPageTable->getImageHandle();
key.subresource = pageInfo.image.subresource;
key.offset = pageInfo.image.offset;
key.extent = pageInfo.image.extent;
m_cmd->bindImageMemory(key, mapping.getHandle());
} break;
case DxvkSparsePageType::ImageMipTail: {
DxvkSparseImageOpaqueBindKey key;
key.image = dstPageTable->getImageHandle();
key.offset = pageInfo.mipTail.resourceOffset;
key.size = pageInfo.mipTail.resourceLength;
key.flags = 0;
m_cmd->bindImageOpaqueMemory(key, mapping.getHandle());
} break;
}
// Update the page table mapping for tracking purposes
if (pageInfo.type != DxvkSparsePageType::None)
dstPageTable->updateMapping(m_cmd.ptr(), bind.dstPage, std::move(mapping));
}
m_cmd->trackResource<DxvkAccess::Write>(bindInfo.dstResource);
}
void DxvkContext::signalGpuEvent(const Rc<DxvkGpuEvent>& event) {
this->spillRenderPass(true);
DxvkGpuEventHandle handle = m_common->eventPool().allocEvent();
// Supported client APIs can't access device memory in a defined manner
// without triggering a queue submission first, so we really only need
// to wait for prior commands, especially queries, to complete.
VkMemoryBarrier2 barrier = { VK_STRUCTURE_TYPE_MEMORY_BARRIER_2 };
barrier.srcStageMask = VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT;
VkDependencyInfo depInfo = { VK_STRUCTURE_TYPE_DEPENDENCY_INFO };
depInfo.memoryBarrierCount = 1;
depInfo.pMemoryBarriers = &barrier;
m_cmd->cmdSetEvent(handle.event, &depInfo);
m_cmd->trackGpuEvent(event->reset(handle));
m_cmd->trackResource<DxvkAccess::None>(event);
}
void DxvkContext::launchCuKernelNVX(
const VkCuLaunchInfoNVX& nvxLaunchInfo,
const std::vector<std::pair<Rc<DxvkBuffer>, DxvkAccessFlags>>& buffers,
const std::vector<std::pair<Rc<DxvkImage>, DxvkAccessFlags>>& images) {
// The resources in the std::vectors above are called-out
// explicitly in the API for barrier and tracking purposes
// since they're being used bindlessly.
this->spillRenderPass(true);
VkPipelineStageFlags srcStages = 0;
VkAccessFlags srcAccess = 0;
for (auto& r : buffers) {
srcStages |= r.first->info().stages;
srcAccess |= r.first->info().access;
}
for (auto& r : images) {
srcStages |= r.first->info().stages;
srcAccess |= r.first->info().access;
this->prepareImage(r.first, r.first->getAvailableSubresources());
}
m_execBarriers.accessMemory(srcStages, srcAccess,
VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT);
m_execBarriers.recordCommands(m_cmd);
m_cmd->cmdLaunchCuKernel(nvxLaunchInfo);
for (auto& r : buffers) {
VkAccessFlags accessFlags = (r.second.test(DxvkAccess::Read) * VK_ACCESS_SHADER_READ_BIT)
| (r.second.test(DxvkAccess::Write) * VK_ACCESS_SHADER_WRITE_BIT);
DxvkBufferSliceHandle bufferSlice = r.first->getSliceHandle();
m_execBarriers.accessBuffer(bufferSlice,
VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
accessFlags,
r.first->info().stages,
r.first->info().access);
}
for (auto& r : images) {
VkAccessFlags accessFlags = (r.second.test(DxvkAccess::Read) * VK_ACCESS_SHADER_READ_BIT)
| (r.second.test(DxvkAccess::Write) * VK_ACCESS_SHADER_WRITE_BIT);
m_execBarriers.accessImage(r.first,
r.first->getAvailableSubresources(),
r.first->info().layout,
VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
accessFlags,
r.first->info().layout,
r.first->info().stages,
r.first->info().access);
}
for (auto& r : images) {
if (r.second.test(DxvkAccess::Read)) m_cmd->trackResource<DxvkAccess::Read>(r.first);
if (r.second.test(DxvkAccess::Write)) m_cmd->trackResource<DxvkAccess::Write>(r.first);
}
for (auto& r : buffers) {
if (r.second.test(DxvkAccess::Read)) m_cmd->trackResource<DxvkAccess::Read>(r.first);
if (r.second.test(DxvkAccess::Write)) m_cmd->trackResource<DxvkAccess::Write>(r.first);
}
}
void DxvkContext::writeTimestamp(const Rc<DxvkGpuQuery>& query) {
m_queryManager.writeTimestamp(m_cmd, query);
}
void DxvkContext::signal(const Rc<sync::Signal>& signal, uint64_t value) {
m_cmd->queueSignal(signal, value);
}
void DxvkContext::waitFence(const Rc<DxvkFence>& fence, uint64_t value) {
m_cmd->waitFence(fence, value);
}
void DxvkContext::signalFence(const Rc<DxvkFence>& fence, uint64_t value) {
m_cmd->signalFence(fence, value);
}
void DxvkContext::beginDebugLabel(VkDebugUtilsLabelEXT *label) {
if (!m_device->instance()->extensions().extDebugUtils)
return;
m_cmd->cmdBeginDebugUtilsLabel(label);
}
void DxvkContext::endDebugLabel() {
if (!m_device->instance()->extensions().extDebugUtils)
return;
m_cmd->cmdEndDebugUtilsLabel();
}
void DxvkContext::insertDebugLabel(VkDebugUtilsLabelEXT *label) {
if (!m_device->instance()->extensions().extDebugUtils)
return;
m_cmd->cmdInsertDebugUtilsLabel(label);
}
void DxvkContext::blitImageFb(
const Rc<DxvkImage>& dstImage,
const Rc<DxvkImage>& srcImage,
const VkImageBlit& region,
const VkComponentMapping& mapping,
VkFilter filter) {
this->invalidateState();
auto dstSubresourceRange = vk::makeSubresourceRange(region.dstSubresource);
auto srcSubresourceRange = vk::makeSubresourceRange(region.srcSubresource);
if (m_execBarriers.isImageDirty(dstImage, dstSubresourceRange, DxvkAccess::Write)
|| m_execBarriers.isImageDirty(srcImage, srcSubresourceRange, DxvkAccess::Write))
m_execBarriers.recordCommands(m_cmd);
bool srcIsDepthStencil = region.srcSubresource.aspectMask & (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT);
VkImageLayout srcLayout = srcImage->pickLayout(srcIsDepthStencil
? VK_IMAGE_LAYOUT_DEPTH_STENCIL_READ_ONLY_OPTIMAL
: VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL);
VkImageLayout dstLayout = dstImage->pickLayout(
VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL);
if (dstImage->info().layout != dstLayout) {
m_execAcquires.accessImage(
dstImage, dstSubresourceRange,
dstImage->info().layout,
dstImage->info().stages, 0,
dstLayout,
VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT,
VK_ACCESS_COLOR_ATTACHMENT_READ_BIT |
VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT);
}
if (srcImage->info().layout != srcLayout) {
m_execAcquires.accessImage(
srcImage, srcSubresourceRange,
srcImage->info().layout,
srcImage->info().stages, 0,
srcLayout,
VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT,
VK_ACCESS_SHADER_READ_BIT);
}
m_execAcquires.recordCommands(m_cmd);
// Sort out image offsets so that dstOffset[0] points
// to the top-left corner of the target area
VkOffset3D srcOffsets[2] = { region.srcOffsets[0], region.srcOffsets[1] };
VkOffset3D dstOffsets[2] = { region.dstOffsets[0], region.dstOffsets[1] };
if (dstOffsets[0].x > dstOffsets[1].x) {
std::swap(dstOffsets[0].x, dstOffsets[1].x);
std::swap(srcOffsets[0].x, srcOffsets[1].x);
}
if (dstOffsets[0].y > dstOffsets[1].y) {
std::swap(dstOffsets[0].y, dstOffsets[1].y);
std::swap(srcOffsets[0].y, srcOffsets[1].y);
}
if (dstOffsets[0].z > dstOffsets[1].z) {
std::swap(dstOffsets[0].z, dstOffsets[1].z);
std::swap(srcOffsets[0].z, srcOffsets[1].z);
}
VkExtent3D dstExtent = {
uint32_t(dstOffsets[1].x - dstOffsets[0].x),
uint32_t(dstOffsets[1].y - dstOffsets[0].y),
uint32_t(dstOffsets[1].z - dstOffsets[0].z) };
// Begin render pass
Rc<DxvkMetaBlitRenderPass> pass = new DxvkMetaBlitRenderPass(
m_device, dstImage, srcImage, region, mapping);
VkExtent3D imageExtent = dstImage->mipLevelExtent(region.dstSubresource.mipLevel);
VkRenderingAttachmentInfo attachmentInfo = { VK_STRUCTURE_TYPE_RENDERING_ATTACHMENT_INFO };
attachmentInfo.imageView = pass->getDstView();
attachmentInfo.imageLayout = dstLayout;
attachmentInfo.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD;
attachmentInfo.storeOp = VK_ATTACHMENT_STORE_OP_STORE;
VkRenderingInfo renderingInfo = { VK_STRUCTURE_TYPE_RENDERING_INFO };
renderingInfo.renderArea = VkRect2D {
VkOffset2D { 0, 0 },
VkExtent2D { imageExtent.width, imageExtent.height } };
renderingInfo.layerCount = pass->framebufferLayerCount();
renderingInfo.colorAttachmentCount = 1;
renderingInfo.pColorAttachments = &attachmentInfo;
m_cmd->cmdBeginRendering(&renderingInfo);
// Bind pipeline
DxvkMetaBlitPipeline pipeInfo = m_common->metaBlit().getPipeline(
pass->viewType(), dstImage->info().format, dstImage->info().sampleCount);
m_cmd->cmdBindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeInfo.pipeHandle);
// Set up viewport
VkViewport viewport;
viewport.x = float(dstOffsets[0].x);
viewport.y = float(dstOffsets[0].y);
viewport.width = float(dstExtent.width);
viewport.height = float(dstExtent.height);
viewport.minDepth = 0.0f;
viewport.maxDepth = 1.0f;
VkRect2D scissor;
scissor.offset = { dstOffsets[0].x, dstOffsets[0].y };
scissor.extent = { dstExtent.width, dstExtent.height };
m_cmd->cmdSetViewport(1, &viewport);
m_cmd->cmdSetScissor(1, &scissor);
// Bind source image view
VkDescriptorImageInfo descriptorImage;
descriptorImage.sampler = m_common->metaBlit().getSampler(filter);
descriptorImage.imageView = pass->getSrcView();
descriptorImage.imageLayout = srcLayout;
VkWriteDescriptorSet descriptorWrite = { VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET };
descriptorWrite.dstSet = m_descriptorPool->alloc(pipeInfo.dsetLayout);
descriptorWrite.dstBinding = 0;
descriptorWrite.dstArrayElement = 0;
descriptorWrite.descriptorCount = 1;
descriptorWrite.descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER;
descriptorWrite.pImageInfo = &descriptorImage;
m_cmd->updateDescriptorSets(1, &descriptorWrite);
m_cmd->cmdBindDescriptorSet(VK_PIPELINE_BIND_POINT_GRAPHICS,
pipeInfo.pipeLayout, descriptorWrite.dstSet, 0, nullptr);
// Compute shader parameters for the operation
VkExtent3D srcExtent = srcImage->mipLevelExtent(region.srcSubresource.mipLevel);
DxvkMetaBlitPushConstants pushConstants = { };
pushConstants.srcCoord0 = {
float(srcOffsets[0].x) / float(srcExtent.width),
float(srcOffsets[0].y) / float(srcExtent.height),
float(srcOffsets[0].z) / float(srcExtent.depth) };
pushConstants.srcCoord1 = {
float(srcOffsets[1].x) / float(srcExtent.width),
float(srcOffsets[1].y) / float(srcExtent.height),
float(srcOffsets[1].z) / float(srcExtent.depth) };
pushConstants.layerCount = pass->framebufferLayerCount();
m_cmd->cmdPushConstants(
pipeInfo.pipeLayout,
VK_SHADER_STAGE_FRAGMENT_BIT,
0, sizeof(pushConstants),
&pushConstants);
m_cmd->cmdDraw(3, pushConstants.layerCount, 0, 0);
m_cmd->cmdEndRendering();
// Add barriers and track image objects
m_execBarriers.accessImage(dstImage,
vk::makeSubresourceRange(region.dstSubresource),
dstLayout,
VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT,
VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT,
dstImage->info().layout,
dstImage->info().stages,
dstImage->info().access);
m_execBarriers.accessImage(srcImage,
vk::makeSubresourceRange(region.srcSubresource),
srcLayout,
VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT,
VK_ACCESS_SHADER_READ_BIT,
srcImage->info().layout,
srcImage->info().stages,
srcImage->info().access);
m_cmd->trackResource<DxvkAccess::Write>(dstImage);
m_cmd->trackResource<DxvkAccess::Read>(srcImage);
m_cmd->trackResource<DxvkAccess::None>(pass);
}
void DxvkContext::blitImageHw(
const Rc<DxvkImage>& dstImage,
const Rc<DxvkImage>& srcImage,
const VkImageBlit& region,
VkFilter filter) {
auto dstSubresourceRange = vk::makeSubresourceRange(region.dstSubresource);
auto srcSubresourceRange = vk::makeSubresourceRange(region.srcSubresource);
if (m_execBarriers.isImageDirty(dstImage, dstSubresourceRange, DxvkAccess::Write)
|| m_execBarriers.isImageDirty(srcImage, srcSubresourceRange, DxvkAccess::Write))
m_execBarriers.recordCommands(m_cmd);
// Prepare the two images for transfer ops if necessary
auto dstLayout = dstImage->pickLayout(VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL);
auto srcLayout = srcImage->pickLayout(VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL);
if (dstImage->info().layout != dstLayout) {
m_execAcquires.accessImage(
dstImage, dstSubresourceRange,
dstImage->info().layout,
VK_PIPELINE_STAGE_TRANSFER_BIT, 0,
dstLayout,
VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_ACCESS_TRANSFER_WRITE_BIT);
}
if (srcImage->info().layout != srcLayout) {
m_execAcquires.accessImage(
srcImage, srcSubresourceRange,
srcImage->info().layout,
VK_PIPELINE_STAGE_TRANSFER_BIT, 0,
srcLayout,
VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_ACCESS_TRANSFER_READ_BIT);
}
m_execAcquires.recordCommands(m_cmd);
// Perform the blit operation
VkImageBlit2 blitRegion = { VK_STRUCTURE_TYPE_IMAGE_BLIT_2 };
blitRegion.srcSubresource = region.srcSubresource;
blitRegion.dstSubresource = region.dstSubresource;
for (uint32_t i = 0; i < 2; i++) {
blitRegion.srcOffsets[i] = region.srcOffsets[i];
blitRegion.dstOffsets[i] = region.dstOffsets[i];
}
VkBlitImageInfo2 blitInfo = { VK_STRUCTURE_TYPE_BLIT_IMAGE_INFO_2 };
blitInfo.srcImage = srcImage->handle();
blitInfo.srcImageLayout = srcLayout;
blitInfo.dstImage = dstImage->handle();
blitInfo.dstImageLayout = dstLayout;
blitInfo.regionCount = 1;
blitInfo.pRegions = &blitRegion;
blitInfo.filter = filter;
m_cmd->cmdBlitImage(&blitInfo);
m_execBarriers.accessImage(
dstImage, dstSubresourceRange, dstLayout,
VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_ACCESS_TRANSFER_WRITE_BIT,
dstImage->info().layout,
dstImage->info().stages,
dstImage->info().access);
m_execBarriers.accessImage(
srcImage, srcSubresourceRange, srcLayout,
VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_ACCESS_TRANSFER_READ_BIT,
srcImage->info().layout,
srcImage->info().stages,
srcImage->info().access);
m_cmd->trackResource<DxvkAccess::Write>(dstImage);
m_cmd->trackResource<DxvkAccess::Read>(srcImage);
}
template<bool ToImage>
void DxvkContext::copyImageBufferData(
DxvkCmdBuffer cmd,
const Rc<DxvkImage>& image,
const VkImageSubresourceLayers& imageSubresource,
VkOffset3D imageOffset,
VkExtent3D imageExtent,
VkImageLayout imageLayout,
const DxvkBufferSliceHandle& bufferSlice,
VkDeviceSize bufferRowAlignment,
VkDeviceSize bufferSliceAlignment) {
auto formatInfo = image->formatInfo();
auto layers = imageSubresource.layerCount;
VkDeviceSize bufferOffset = bufferSlice.offset;
// Do one copy region per layer in case the buffer memory layout is weird
if (bufferSliceAlignment || formatInfo->flags.test(DxvkFormatFlag::MultiPlane))
layers = 1;
for (uint32_t i = 0; i < imageSubresource.layerCount; i += layers) {
auto aspectOffset = bufferOffset;
for (auto aspects = imageSubresource.aspectMask; aspects; ) {
auto aspect = vk::getNextAspect(aspects);
auto elementSize = formatInfo->elementSize;
VkBufferImageCopy2 copyRegion = { VK_STRUCTURE_TYPE_BUFFER_IMAGE_COPY_2 };
copyRegion.imageSubresource.aspectMask = aspect;
copyRegion.imageSubresource.baseArrayLayer = imageSubresource.baseArrayLayer + i;
copyRegion.imageSubresource.layerCount = layers;
copyRegion.imageSubresource.mipLevel = imageSubresource.mipLevel;
copyRegion.imageOffset = imageOffset;
copyRegion.imageExtent = imageExtent;
if (formatInfo->flags.test(DxvkFormatFlag::MultiPlane)) {
auto plane = &formatInfo->planes[vk::getPlaneIndex(aspect)];
copyRegion.imageOffset.x /= plane->blockSize.width;
copyRegion.imageOffset.y /= plane->blockSize.height;
copyRegion.imageExtent.width /= plane->blockSize.width;
copyRegion.imageExtent.height /= plane->blockSize.height;
elementSize = plane->elementSize;
}
// Vulkan can't really express row pitch in the same way that client APIs
// may expect, so we'll need to do some heroics here and hope that it works
VkExtent3D blockCount = util::computeBlockCount(copyRegion.imageExtent, formatInfo->blockSize);
VkDeviceSize rowPitch = blockCount.width * elementSize;
if (bufferRowAlignment > elementSize)
rowPitch = bufferRowAlignment >= rowPitch ? bufferRowAlignment : align(rowPitch, bufferRowAlignment);
VkDeviceSize slicePitch = blockCount.height * rowPitch;
if (image->info().type == VK_IMAGE_TYPE_3D && bufferSliceAlignment > elementSize)
slicePitch = bufferSliceAlignment >= slicePitch ? bufferSliceAlignment : align(slicePitch, bufferSliceAlignment);
copyRegion.bufferOffset = aspectOffset;
copyRegion.bufferRowLength = formatInfo->blockSize.width * rowPitch / elementSize;
copyRegion.bufferImageHeight = formatInfo->blockSize.height * slicePitch / rowPitch;
// Perform the actual copy
if constexpr (ToImage) {
VkCopyBufferToImageInfo2 copyInfo = { VK_STRUCTURE_TYPE_COPY_BUFFER_TO_IMAGE_INFO_2 };
copyInfo.srcBuffer = bufferSlice.handle;
copyInfo.dstImage = image->handle();
copyInfo.dstImageLayout = imageLayout;
copyInfo.regionCount = 1;
copyInfo.pRegions = &copyRegion;
m_cmd->cmdCopyBufferToImage(cmd, &copyInfo);
} else {
VkCopyImageToBufferInfo2 copyInfo = { VK_STRUCTURE_TYPE_COPY_IMAGE_TO_BUFFER_INFO_2 };
copyInfo.srcImage = image->handle();
copyInfo.srcImageLayout = imageLayout;
copyInfo.dstBuffer = bufferSlice.handle;
copyInfo.regionCount = 1;
copyInfo.pRegions = &copyRegion;
m_cmd->cmdCopyImageToBuffer(cmd, &copyInfo);
}
aspectOffset += blockCount.depth * slicePitch;
}
// Advance to next layer. This is non-trivial for multi-plane formats
// since plane data for each layer is expected to be packed.
VkDeviceSize layerPitch = aspectOffset - bufferOffset;
if (bufferSliceAlignment)
layerPitch = bufferSliceAlignment >= layerPitch ? bufferSliceAlignment : align(layerPitch, bufferSliceAlignment);
bufferOffset += layerPitch;
}
}
void DxvkContext::copyImageHostData(
DxvkCmdBuffer cmd,
const Rc<DxvkImage>& image,
const VkImageSubresourceLayers& imageSubresource,
VkOffset3D imageOffset,
VkExtent3D imageExtent,
const void* hostData,
VkDeviceSize rowPitch,
VkDeviceSize slicePitch) {
auto formatInfo = image->formatInfo();
auto srcData = reinterpret_cast<const char*>(hostData);
for (uint32_t i = 0; i < imageSubresource.layerCount; i++) {
auto layerData = srcData + i * slicePitch;
for (auto aspects = imageSubresource.aspectMask; aspects; ) {
auto aspect = vk::getNextAspect(aspects);
auto extent = imageExtent;
VkDeviceSize elementSize = formatInfo->elementSize;
if (formatInfo->flags.test(DxvkFormatFlag::MultiPlane)) {
auto plane = &formatInfo->planes[vk::getPlaneIndex(aspect)];
extent.width /= plane->blockSize.width;
extent.height /= plane->blockSize.height;
elementSize = plane->elementSize;
}
auto blockCount = util::computeBlockCount(extent, formatInfo->blockSize);
auto stagingSlice = m_staging.alloc(CACHE_LINE_SIZE, elementSize * util::flattenImageExtent(blockCount));
auto stagingHandle = stagingSlice.getSliceHandle();
util::packImageData(stagingHandle.mapPtr, layerData,
blockCount, elementSize, rowPitch, slicePitch);
auto subresource = imageSubresource;
subresource.aspectMask = aspect;
this->copyImageBufferData<true>(cmd,
image, subresource, imageOffset, imageExtent,
image->pickLayout(VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL),
stagingHandle, 0, 0);
layerData += blockCount.height * rowPitch;
m_cmd->trackResource<DxvkAccess::Read>(stagingSlice.buffer());
}
}
}
void DxvkContext::clearImageViewFb(
const Rc<DxvkImageView>& imageView,
VkOffset3D offset,
VkExtent3D extent,
VkImageAspectFlags aspect,
VkClearValue value) {
this->updateFramebuffer();
VkPipelineStageFlags clearStages = 0;
VkAccessFlags clearAccess = 0;
VkImageLayout clearLayout = VK_IMAGE_LAYOUT_UNDEFINED;
// Find out if the render target view is currently bound,
// so that we can avoid spilling the render pass if it is.
int32_t attachmentIndex = -1;
if (m_state.om.framebufferInfo.isFullSize(imageView))
attachmentIndex = m_state.om.framebufferInfo.findAttachment(imageView);
if (attachmentIndex >= 0 && !m_state.om.framebufferInfo.isWritable(attachmentIndex, aspect))
attachmentIndex = -1;
if (attachmentIndex < 0) {
this->spillRenderPass(false);
if (m_execBarriers.isImageDirty(imageView->image(), imageView->imageSubresources(), DxvkAccess::Write))
m_execBarriers.recordCommands(m_cmd);
clearLayout = (imageView->info().aspect & VK_IMAGE_ASPECT_COLOR_BIT)
? imageView->pickLayout(VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL)
: imageView->pickLayout(VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL);
VkExtent3D extent = imageView->mipLevelExtent(0);
VkRenderingAttachmentInfo attachmentInfo = { VK_STRUCTURE_TYPE_RENDERING_ATTACHMENT_INFO };
attachmentInfo.imageView = imageView->handle();
attachmentInfo.imageLayout = clearLayout;
attachmentInfo.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD;
attachmentInfo.storeOp = VK_ATTACHMENT_STORE_OP_STORE;
VkRenderingInfo renderingInfo = { VK_STRUCTURE_TYPE_RENDERING_INFO };
renderingInfo.renderArea.extent = { extent.width, extent.height };
renderingInfo.layerCount = imageView->info().numLayers;
if (imageView->info().aspect & VK_IMAGE_ASPECT_COLOR_BIT) {
clearStages |= VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT;
clearAccess |= VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT
| VK_ACCESS_COLOR_ATTACHMENT_READ_BIT;
renderingInfo.colorAttachmentCount = 1;
renderingInfo.pColorAttachments = &attachmentInfo;
} else {
clearStages |= VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT
| VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT;
clearAccess |= VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT
| VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT;
if (imageView->info().aspect & VK_IMAGE_ASPECT_DEPTH_BIT)
renderingInfo.pDepthAttachment = &attachmentInfo;
if (imageView->info().aspect & VK_IMAGE_ASPECT_STENCIL_BIT)
renderingInfo.pStencilAttachment = &attachmentInfo;
}
if (clearLayout != imageView->imageInfo().layout) {
m_execAcquires.accessImage(
imageView->image(),
imageView->imageSubresources(),
imageView->imageInfo().layout, clearStages, 0,
clearLayout, clearStages, clearAccess);
m_execAcquires.recordCommands(m_cmd);
}
// We cannot leverage render pass clears
// because we clear only part of the view
m_cmd->cmdBeginRendering(&renderingInfo);
} else {
// Make sure the render pass is active so
// that we can actually perform the clear
this->startRenderPass();
}
// Perform the actual clear operation
VkClearAttachment clearInfo;
clearInfo.aspectMask = aspect;
clearInfo.colorAttachment = 0;
clearInfo.clearValue = value;
if ((aspect & VK_IMAGE_ASPECT_COLOR_BIT) && (attachmentIndex >= 0))
clearInfo.colorAttachment = m_state.om.framebufferInfo.getColorAttachmentIndex(attachmentIndex);
VkClearRect clearRect;
clearRect.rect.offset.x = offset.x;
clearRect.rect.offset.y = offset.y;
clearRect.rect.extent.width = extent.width;
clearRect.rect.extent.height = extent.height;
clearRect.baseArrayLayer = 0;
clearRect.layerCount = imageView->info().numLayers;
m_cmd->cmdClearAttachments(1, &clearInfo, 1, &clearRect);
// Unbind temporary framebuffer
if (attachmentIndex < 0) {
m_cmd->cmdEndRendering();
m_execBarriers.accessImage(
imageView->image(),
imageView->imageSubresources(),
clearLayout, clearStages, clearAccess,
imageView->imageInfo().layout,
imageView->imageInfo().stages,
imageView->imageInfo().access);
m_cmd->trackResource<DxvkAccess::None>(imageView);
m_cmd->trackResource<DxvkAccess::Write>(imageView->image());
}
}
void DxvkContext::clearImageViewCs(
const Rc<DxvkImageView>& imageView,
VkOffset3D offset,
VkExtent3D extent,
VkClearValue value) {
this->spillRenderPass(false);
this->invalidateState();
if (m_execBarriers.isImageDirty(
imageView->image(),
imageView->imageSubresources(),
DxvkAccess::Write))
m_execBarriers.recordCommands(m_cmd);
// Query pipeline objects to use for this clear operation
DxvkMetaClearPipeline pipeInfo = m_common->metaClear().getClearImagePipeline(
imageView->type(), lookupFormatInfo(imageView->info().format)->flags);
// Create a descriptor set pointing to the view
VkDescriptorSet descriptorSet = m_descriptorPool->alloc(pipeInfo.dsetLayout);
VkDescriptorImageInfo viewInfo;
viewInfo.sampler = VK_NULL_HANDLE;
viewInfo.imageView = imageView->handle();
viewInfo.imageLayout = imageView->imageInfo().layout;
VkWriteDescriptorSet descriptorWrite = { VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET };
descriptorWrite.dstSet = descriptorSet;
descriptorWrite.dstBinding = 0;
descriptorWrite.dstArrayElement = 0;
descriptorWrite.descriptorCount = 1;
descriptorWrite.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE;
descriptorWrite.pImageInfo = &viewInfo;
m_cmd->updateDescriptorSets(1, &descriptorWrite);
// Prepare shader arguments
DxvkMetaClearArgs pushArgs = { };
pushArgs.clearValue = value.color;
pushArgs.offset = offset;
pushArgs.extent = extent;
VkExtent3D workgroups = util::computeBlockCount(
pushArgs.extent, pipeInfo.workgroupSize);
if (imageView->type() == VK_IMAGE_VIEW_TYPE_1D_ARRAY)
workgroups.height = imageView->subresources().layerCount;
else if (imageView->type() == VK_IMAGE_VIEW_TYPE_2D_ARRAY)
workgroups.depth = imageView->subresources().layerCount;
m_cmd->cmdBindPipeline(
VK_PIPELINE_BIND_POINT_COMPUTE,
pipeInfo.pipeline);
m_cmd->cmdBindDescriptorSet(
VK_PIPELINE_BIND_POINT_COMPUTE,
pipeInfo.pipeLayout, descriptorSet,
0, nullptr);
m_cmd->cmdPushConstants(
pipeInfo.pipeLayout,
VK_SHADER_STAGE_COMPUTE_BIT,
0, sizeof(pushArgs), &pushArgs);
m_cmd->cmdDispatch(
workgroups.width,
workgroups.height,
workgroups.depth);
m_execBarriers.accessImage(
imageView->image(),
imageView->imageSubresources(),
imageView->imageInfo().layout,
VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
VK_ACCESS_SHADER_WRITE_BIT,
imageView->imageInfo().layout,
imageView->imageInfo().stages,
imageView->imageInfo().access);
m_cmd->trackResource<DxvkAccess::None>(imageView);
m_cmd->trackResource<DxvkAccess::Write>(imageView->image());
}
void DxvkContext::copyImageHw(
const Rc<DxvkImage>& dstImage,
VkImageSubresourceLayers dstSubresource,
VkOffset3D dstOffset,
const Rc<DxvkImage>& srcImage,
VkImageSubresourceLayers srcSubresource,
VkOffset3D srcOffset,
VkExtent3D extent) {
auto dstSubresourceRange = vk::makeSubresourceRange(dstSubresource);
auto srcSubresourceRange = vk::makeSubresourceRange(srcSubresource);
auto dstFormatInfo = dstImage->formatInfo();
if (m_execBarriers.isImageDirty(dstImage, dstSubresourceRange, DxvkAccess::Write)
|| m_execBarriers.isImageDirty(srcImage, srcSubresourceRange, DxvkAccess::Write))
m_execBarriers.recordCommands(m_cmd);
VkImageLayout dstImageLayout = dstImage->pickLayout(VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL);
VkImageLayout srcImageLayout = srcImage->pickLayout(VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL);
VkImageLayout dstInitImageLayout = dstImage->info().layout;
if (dstImage->isFullSubresource(dstSubresource, extent))
dstInitImageLayout = VK_IMAGE_LAYOUT_UNDEFINED;
if (dstImageLayout != dstInitImageLayout) {
m_execAcquires.accessImage(
dstImage, dstSubresourceRange,
dstInitImageLayout,
VK_PIPELINE_STAGE_TRANSFER_BIT, 0,
dstImageLayout,
VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_ACCESS_TRANSFER_WRITE_BIT);
}
if (srcImageLayout != srcImage->info().layout) {
m_execAcquires.accessImage(
srcImage, srcSubresourceRange,
srcImage->info().layout,
VK_PIPELINE_STAGE_TRANSFER_BIT, 0,
srcImageLayout,
VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_ACCESS_TRANSFER_READ_BIT);
}
m_execAcquires.recordCommands(m_cmd);
for (auto aspects = dstSubresource.aspectMask; aspects; ) {
auto aspect = vk::getNextAspect(aspects);
VkImageCopy2 copyRegion = { VK_STRUCTURE_TYPE_IMAGE_COPY_2 };
copyRegion.srcSubresource = srcSubresource;
copyRegion.srcSubresource.aspectMask = aspect;
copyRegion.srcOffset = srcOffset;
copyRegion.dstSubresource = dstSubresource;
copyRegion.dstSubresource.aspectMask = aspect;
copyRegion.dstOffset = dstOffset;
copyRegion.extent = extent;
if (dstFormatInfo->flags.test(DxvkFormatFlag::MultiPlane)) {
auto plane = &dstFormatInfo->planes[vk::getPlaneIndex(aspect)];
copyRegion.srcOffset.x /= plane->blockSize.width;
copyRegion.srcOffset.y /= plane->blockSize.height;
copyRegion.dstOffset.x /= plane->blockSize.width;
copyRegion.dstOffset.y /= plane->blockSize.height;
copyRegion.extent.width /= plane->blockSize.width;
copyRegion.extent.height /= plane->blockSize.height;
}
VkCopyImageInfo2 copyInfo = { VK_STRUCTURE_TYPE_COPY_IMAGE_INFO_2 };
copyInfo.srcImage = srcImage->handle();
copyInfo.srcImageLayout = srcImageLayout;
copyInfo.dstImage = dstImage->handle();
copyInfo.dstImageLayout = dstImageLayout;
copyInfo.regionCount = 1;
copyInfo.pRegions = &copyRegion;
m_cmd->cmdCopyImage(DxvkCmdBuffer::ExecBuffer, &copyInfo);
}
m_execBarriers.accessImage(
dstImage, dstSubresourceRange,
dstImageLayout,
VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_ACCESS_TRANSFER_WRITE_BIT,
dstImage->info().layout,
dstImage->info().stages,
dstImage->info().access);
m_execBarriers.accessImage(
srcImage, srcSubresourceRange,
srcImageLayout,
VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_ACCESS_TRANSFER_READ_BIT,
srcImage->info().layout,
srcImage->info().stages,
srcImage->info().access);
m_cmd->trackResource<DxvkAccess::Write>(dstImage);
m_cmd->trackResource<DxvkAccess::Read>(srcImage);
}
void DxvkContext::copyImageFb(
const Rc<DxvkImage>& dstImage,
VkImageSubresourceLayers dstSubresource,
VkOffset3D dstOffset,
const Rc<DxvkImage>& srcImage,
VkImageSubresourceLayers srcSubresource,
VkOffset3D srcOffset,
VkExtent3D extent) {
VkFormat viewFormat = m_common->metaCopy().getCopyDestinationFormat(
dstSubresource.aspectMask,
srcSubresource.aspectMask,
srcImage->info().format);
if (!viewFormat) {
Logger::err("DxvkContext: copyImageFb: Unsupported format");
return;
}
// Usually we should be able to draw directly to the destination image,
// but in some cases this might not be possible, e.g. if when copying
// from something like D32_SFLOAT to RGBA8_UNORM. In those situations,
// create a temporary image to draw to, and then copy to the actual
// destination image using a regular Vulkan transfer function.
bool useDirectCopy = (dstImage->info().usage & (VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT | VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT))
&& (dstImage->isViewCompatible(viewFormat));
if (useDirectCopy) {
this->copyImageFbDirect(
dstImage, dstSubresource, dstOffset, viewFormat,
srcImage, srcSubresource, srcOffset, extent);
} else {
DxvkImageCreateInfo imageInfo = dstImage->info();
imageInfo.format = viewFormat;
imageInfo.flags = 0;
imageInfo.usage = VK_IMAGE_USAGE_TRANSFER_SRC_BIT;
imageInfo.extent = extent;
imageInfo.numLayers = dstSubresource.layerCount;
imageInfo.mipLevels = 1;
imageInfo.tiling = VK_IMAGE_TILING_OPTIMAL;
imageInfo.layout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL;
imageInfo.stages = VK_PIPELINE_STAGE_TRANSFER_BIT;
imageInfo.access = VK_ACCESS_TRANSFER_READ_BIT;
imageInfo.viewFormatCount = 0;
if (dstImage->formatInfo()->aspectMask & VK_IMAGE_ASPECT_COLOR_BIT) {
imageInfo.usage |= VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT;
imageInfo.stages |= VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT;
imageInfo.access |= VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT;
} else {
imageInfo.usage |= VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT;
imageInfo.stages |= VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT
| VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT;
imageInfo.access |= VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT;
}
Rc<DxvkImage> tmpImage = m_device->createImage(imageInfo,
VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT);
VkImageSubresourceLayers tmpSubresource = dstSubresource;
tmpSubresource.mipLevel = 0;
tmpSubresource.baseArrayLayer = 0;
VkOffset3D tmpOffset = { 0, 0, 0 };
this->copyImageFbDirect(
tmpImage, tmpSubresource, tmpOffset, viewFormat,
srcImage, srcSubresource, srcOffset, extent);
this->copyImageHw(
dstImage, dstSubresource, dstOffset,
tmpImage, tmpSubresource, tmpOffset, extent);
}
}
void DxvkContext::copyImageFbDirect(
const Rc<DxvkImage>& dstImage,
VkImageSubresourceLayers dstSubresource,
VkOffset3D dstOffset,
VkFormat dstFormat,
const Rc<DxvkImage>& srcImage,
VkImageSubresourceLayers srcSubresource,
VkOffset3D srcOffset,
VkExtent3D extent) {
this->invalidateState();
auto dstSubresourceRange = vk::makeSubresourceRange(dstSubresource);
auto srcSubresourceRange = vk::makeSubresourceRange(srcSubresource);
if (m_execBarriers.isImageDirty(dstImage, dstSubresourceRange, DxvkAccess::Write)
|| m_execBarriers.isImageDirty(srcImage, srcSubresourceRange, DxvkAccess::Write))
m_execBarriers.recordCommands(m_cmd);
// Flag used to determine whether we can do an UNDEFINED transition
bool doDiscard = dstImage->isFullSubresource(dstSubresource, extent);
// This function can process both color and depth-stencil images, so
// some things change a lot depending on the destination image type
VkPipelineStageFlags dstStages;
VkAccessFlags dstAccess;
VkImageLayout dstLayout;
if (dstSubresource.aspectMask & VK_IMAGE_ASPECT_COLOR_BIT) {
dstLayout = dstImage->pickLayout(VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL);
dstStages = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT;
dstAccess = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT;
if (!doDiscard)
dstAccess |= VK_ACCESS_COLOR_ATTACHMENT_READ_BIT;
} else {
dstLayout = dstImage->pickLayout(VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL);
dstStages = VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT
| VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT;
dstAccess = VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT;
if (!doDiscard)
dstAccess |= VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT;
}
// Might have to transition source image as well
VkImageLayout srcLayout = (srcSubresource.aspectMask & VK_IMAGE_ASPECT_COLOR_BIT)
? srcImage->pickLayout(VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL)
: srcImage->pickLayout(VK_IMAGE_LAYOUT_DEPTH_STENCIL_READ_ONLY_OPTIMAL);
if (srcImage->info().layout != srcLayout) {
m_execAcquires.accessImage(
srcImage, srcSubresourceRange,
srcImage->info().layout,
srcImage->info().stages, 0,
srcLayout,
VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT,
VK_ACCESS_SHADER_READ_BIT);
}
if (dstImage->info().layout != dstLayout) {
m_execAcquires.accessImage(
dstImage, dstSubresourceRange,
doDiscard ? VK_IMAGE_LAYOUT_UNDEFINED
: dstImage->info().layout,
dstImage->info().stages, 0,
dstLayout, dstStages, dstAccess);
}
m_execAcquires.recordCommands(m_cmd);
// Create source and destination image views
VkFormat srcFormat = srcImage->info().format;
Rc<DxvkMetaCopyViews> views = new DxvkMetaCopyViews(m_device->vkd(),
dstImage, dstSubresource, dstFormat,
srcImage, srcSubresource, srcFormat);
// Create pipeline for the copy operation
DxvkMetaCopyPipeline pipeInfo = m_common->metaCopy().getPipeline(
views->getSrcViewType(), dstFormat, dstImage->info().sampleCount);
// Create and initialize descriptor set
VkDescriptorSet descriptorSet = m_descriptorPool->alloc(pipeInfo.dsetLayout);
std::array<VkDescriptorImageInfo, 2> descriptorImages = {{
{ VK_NULL_HANDLE, views->getSrcView(), srcLayout },
{ VK_NULL_HANDLE, views->getSrcStencilView(), srcLayout },
}};
std::array<VkWriteDescriptorSet, 2> descriptorWrites;
for (uint32_t i = 0; i < descriptorWrites.size(); i++) {
descriptorWrites[i] = { VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET };
descriptorWrites[i].dstSet = descriptorSet;
descriptorWrites[i].dstBinding = i;
descriptorWrites[i].descriptorCount = 1;
descriptorWrites[i].descriptorType = VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE;
descriptorWrites[i].pImageInfo = &descriptorImages[i];
}
m_cmd->updateDescriptorSets(
descriptorWrites.size(),
descriptorWrites.data());
// Set up render state
VkViewport viewport;
viewport.x = float(dstOffset.x);
viewport.y = float(dstOffset.y);
viewport.width = float(extent.width);
viewport.height = float(extent.height);
viewport.minDepth = 0.0f;
viewport.maxDepth = 1.0f;
VkRect2D scissor;
scissor.offset = { dstOffset.x, dstOffset.y };
scissor.extent = { extent.width, extent.height };
VkExtent3D mipExtent = dstImage->mipLevelExtent(dstSubresource.mipLevel);
VkRenderingAttachmentInfo attachmentInfo = { VK_STRUCTURE_TYPE_RENDERING_ATTACHMENT_INFO };
attachmentInfo.imageView = views->getDstView();
attachmentInfo.imageLayout = dstLayout;
attachmentInfo.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD;
attachmentInfo.storeOp = VK_ATTACHMENT_STORE_OP_STORE;
if (doDiscard)
attachmentInfo.loadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
VkRenderingInfo renderingInfo = { VK_STRUCTURE_TYPE_RENDERING_INFO };
renderingInfo.renderArea.offset = VkOffset2D { 0, 0 };
renderingInfo.renderArea.extent = VkExtent2D { mipExtent.width, mipExtent.height };
renderingInfo.layerCount = dstSubresource.layerCount;
VkImageAspectFlags dstAspects = dstImage->formatInfo()->aspectMask;
if (dstAspects & VK_IMAGE_ASPECT_COLOR_BIT) {
renderingInfo.colorAttachmentCount = 1;
renderingInfo.pColorAttachments = &attachmentInfo;
} else {
if (dstAspects & VK_IMAGE_ASPECT_DEPTH_BIT)
renderingInfo.pDepthAttachment = &attachmentInfo;
if (dstAspects & VK_IMAGE_ASPECT_STENCIL_BIT)
renderingInfo.pStencilAttachment = &attachmentInfo;
}
// Perform the actual copy operation
m_cmd->cmdBeginRendering(&renderingInfo);
m_cmd->cmdBindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeInfo.pipeHandle);
m_cmd->cmdBindDescriptorSet(VK_PIPELINE_BIND_POINT_GRAPHICS,
pipeInfo.pipeLayout, descriptorSet, 0, nullptr);
m_cmd->cmdSetViewport(1, &viewport);
m_cmd->cmdSetScissor(1, &scissor);
VkOffset2D srcCoordOffset = {
srcOffset.x - dstOffset.x,
srcOffset.y - dstOffset.y };
m_cmd->cmdPushConstants(pipeInfo.pipeLayout,
VK_SHADER_STAGE_FRAGMENT_BIT,
0, sizeof(srcCoordOffset),
&srcCoordOffset);
m_cmd->cmdDraw(3, dstSubresource.layerCount, 0, 0);
m_cmd->cmdEndRendering();
m_execBarriers.accessImage(
srcImage, srcSubresourceRange, srcLayout,
VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT,
VK_ACCESS_SHADER_READ_BIT,
srcImage->info().layout,
srcImage->info().stages,
srcImage->info().access);
m_execBarriers.accessImage(
dstImage, dstSubresourceRange,
dstLayout, dstStages, dstAccess,
dstImage->info().layout,
dstImage->info().stages,
dstImage->info().access);
m_cmd->trackResource<DxvkAccess::Write>(dstImage);
m_cmd->trackResource<DxvkAccess::Read>(srcImage);
m_cmd->trackResource<DxvkAccess::None>(views);
}
bool DxvkContext::copyImageClear(
const Rc<DxvkImage>& dstImage,
VkImageSubresourceLayers dstSubresource,
VkOffset3D dstOffset,
VkExtent3D dstExtent,
const Rc<DxvkImage>& srcImage,
VkImageSubresourceLayers srcSubresource) {
// If the source image has a pending deferred clear, we can
// implement the copy by clearing the destination image to
// the same clear value.
const VkImageUsageFlags attachmentUsage
= VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT
| VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT;
if (!(dstImage->info().usage & attachmentUsage)
|| !(srcImage->info().usage & attachmentUsage))
return false;
// Ignore 3D images since those are complicated to handle
if (dstImage->info().type == VK_IMAGE_TYPE_3D
|| srcImage->info().type == VK_IMAGE_TYPE_3D)
return false;
// Find a pending clear that overlaps with the source image
const DxvkDeferredClear* clear = nullptr;
for (const auto& entry : m_deferredClears) {
// Entries in the deferred clear array cannot overlap, so
// if we find an entry covering all source subresources,
// it's the only one in the list that does.
if ((entry.imageView->image() == srcImage) && ((srcSubresource.aspectMask & entry.clearAspects) == srcSubresource.aspectMask)
&& (vk::checkSubresourceRangeSuperset(entry.imageView->subresources(), vk::makeSubresourceRange(srcSubresource)))) {
clear = &entry;
break;
}
}
if (!clear)
return false;
// Create a view for the destination image with the general
// properties ofthe source image view used for the clear
DxvkImageViewCreateInfo viewInfo = clear->imageView->info();
viewInfo.type = dstImage->info().type == VK_IMAGE_TYPE_1D
? VK_IMAGE_VIEW_TYPE_1D_ARRAY
: VK_IMAGE_VIEW_TYPE_2D_ARRAY;
viewInfo.minLevel = dstSubresource.mipLevel;
viewInfo.numLevels = 1;
viewInfo.minLayer = dstSubresource.baseArrayLayer;
viewInfo.numLayers = dstSubresource.layerCount;
// That is, if the formats are actually compatible
// so that we can safely use the same clear value
if (!dstImage->isViewCompatible(viewInfo.format))
return false;
// Ignore mismatched size for now, needs more testing since we'd
// need to prepare the image first and then call clearImageViewFb
if (dstImage->mipLevelExtent(dstSubresource.mipLevel) != dstExtent)
return false;
auto view = m_device->createImageView(dstImage, viewInfo);
this->deferClear(view, srcSubresource.aspectMask, clear->clearValue);
return true;
}
template<bool ToBuffer>
void DxvkContext::copySparsePages(
const Rc<DxvkPagedResource>& sparse,
uint32_t pageCount,
const uint32_t* pages,
const Rc<DxvkBuffer>& buffer,
VkDeviceSize offset) {
auto pageTable = sparse->getSparsePageTable();
auto bufferHandle = buffer->getSliceHandle(offset, SparseMemoryPageSize * pageCount);
if (m_execBarriers.isBufferDirty(bufferHandle,
ToBuffer ? DxvkAccess::Write : DxvkAccess::Read))
m_execBarriers.recordCommands(m_cmd);
if (pageTable->getBufferHandle()) {
this->copySparseBufferPages<ToBuffer>(
static_cast<DxvkBuffer*>(sparse.ptr()),
pageCount, pages, buffer, offset);
} else {
this->copySparseImagePages<ToBuffer>(
static_cast<DxvkImage*>(sparse.ptr()),
pageCount, pages, buffer, offset);
}
}
template<bool ToBuffer>
void DxvkContext::copySparseBufferPages(
const Rc<DxvkBuffer>& sparse,
uint32_t pageCount,
const uint32_t* pages,
const Rc<DxvkBuffer>& buffer,
VkDeviceSize offset) {
std::vector<VkBufferCopy2> regions;
regions.reserve(pageCount);
auto pageTable = sparse->getSparsePageTable();
auto sparseHandle = sparse->getSliceHandle();
auto bufferHandle = buffer->getSliceHandle(offset, SparseMemoryPageSize * pageCount);
if (m_execBarriers.isBufferDirty(sparseHandle,
ToBuffer ? DxvkAccess::Read : DxvkAccess::Write))
m_execBarriers.recordCommands(m_cmd);
for (uint32_t i = 0; i < pageCount; i++) {
auto pageInfo = pageTable->getPageInfo(pages[i]);
if (pageInfo.type == DxvkSparsePageType::Buffer) {
VkDeviceSize sparseOffset = pageInfo.buffer.offset;
VkDeviceSize bufferOffset = bufferHandle.offset + SparseMemoryPageSize * i;
VkBufferCopy2 copy = { VK_STRUCTURE_TYPE_BUFFER_COPY_2 };
copy.srcOffset = ToBuffer ? sparseOffset : bufferOffset;
copy.dstOffset = ToBuffer ? bufferOffset : sparseOffset;
copy.size = pageInfo.buffer.length;
regions.push_back(copy);
}
}
VkCopyBufferInfo2 info = { VK_STRUCTURE_TYPE_COPY_BUFFER_INFO_2 };
info.srcBuffer = ToBuffer ? sparseHandle.handle : bufferHandle.handle;
info.dstBuffer = ToBuffer ? bufferHandle.handle : sparseHandle.handle;
info.regionCount = uint32_t(regions.size());
info.pRegions = regions.data();
if (info.regionCount)
m_cmd->cmdCopyBuffer(DxvkCmdBuffer::ExecBuffer, &info);
m_execBarriers.accessBuffer(sparseHandle,
VK_PIPELINE_STAGE_TRANSFER_BIT,
ToBuffer ? VK_ACCESS_TRANSFER_READ_BIT : VK_ACCESS_TRANSFER_WRITE_BIT,
sparse->info().stages,
sparse->info().access);
m_execBarriers.accessBuffer(bufferHandle,
VK_PIPELINE_STAGE_TRANSFER_BIT,
ToBuffer ? VK_ACCESS_TRANSFER_WRITE_BIT : VK_ACCESS_TRANSFER_READ_BIT,
buffer->info().stages,
buffer->info().access);
m_cmd->trackResource<ToBuffer ? DxvkAccess::Read : DxvkAccess::Write>(sparse);
m_cmd->trackResource<ToBuffer ? DxvkAccess::Write : DxvkAccess::Read>(buffer);
}
template<bool ToBuffer>
void DxvkContext::copySparseImagePages(
const Rc<DxvkImage>& sparse,
uint32_t pageCount,
const uint32_t* pages,
const Rc<DxvkBuffer>& buffer,
VkDeviceSize offset) {
std::vector<VkBufferImageCopy2> regions;
regions.reserve(pageCount);
auto pageTable = sparse->getSparsePageTable();
auto pageExtent = pageTable->getProperties().pageRegionExtent;
auto bufferHandle = buffer->getSliceHandle(offset, SparseMemoryPageSize * pageCount);
auto sparseSubresources = sparse->getAvailableSubresources();
if (m_execBarriers.isImageDirty(sparse, sparseSubresources, DxvkAccess::Write))
m_execBarriers.recordCommands(m_cmd);
VkImageLayout transferLayout = sparse->pickLayout(ToBuffer
? VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL
: VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL);
VkAccessFlags transferAccess = ToBuffer
? VK_ACCESS_TRANSFER_READ_BIT
: VK_ACCESS_TRANSFER_WRITE_BIT;
if (sparse->info().layout != transferLayout) {
m_execAcquires.accessImage(sparse, sparseSubresources,
sparse->info().layout,
sparse->info().stages, 0,
transferLayout,
VK_PIPELINE_STAGE_TRANSFER_BIT,
transferAccess);
m_execAcquires.recordCommands(m_cmd);
}
for (uint32_t i = 0; i < pageCount; i++) {
auto pageInfo = pageTable->getPageInfo(pages[i]);
if (pageInfo.type == DxvkSparsePageType::Image) {
VkBufferImageCopy2 copy = { VK_STRUCTURE_TYPE_BUFFER_IMAGE_COPY_2 };
copy.bufferOffset = bufferHandle.offset + SparseMemoryPageSize * i;
copy.bufferRowLength = pageExtent.width;
copy.bufferImageHeight = pageExtent.height;
copy.imageSubresource = vk::makeSubresourceLayers(pageInfo.image.subresource);
copy.imageOffset = pageInfo.image.offset;
copy.imageExtent = pageInfo.image.extent;
regions.push_back(copy);
}
}
if (ToBuffer) {
VkCopyImageToBufferInfo2 info = { VK_STRUCTURE_TYPE_COPY_IMAGE_TO_BUFFER_INFO_2 };
info.srcImage = sparse->handle();
info.srcImageLayout = transferLayout;
info.dstBuffer = bufferHandle.handle;
info.regionCount = regions.size();
info.pRegions = regions.data();
if (info.regionCount)
m_cmd->cmdCopyImageToBuffer(DxvkCmdBuffer::ExecBuffer, &info);
} else {
VkCopyBufferToImageInfo2 info = { VK_STRUCTURE_TYPE_COPY_BUFFER_TO_IMAGE_INFO_2 };
info.srcBuffer = bufferHandle.handle;
info.dstImage = sparse->handle();
info.dstImageLayout = transferLayout;
info.regionCount = regions.size();
info.pRegions = regions.data();
if (info.regionCount)
m_cmd->cmdCopyBufferToImage(DxvkCmdBuffer::ExecBuffer, &info);
}
m_execBarriers.accessImage(sparse, sparseSubresources,
transferLayout,
VK_PIPELINE_STAGE_TRANSFER_BIT,
transferAccess,
sparse->info().layout,
sparse->info().stages,
sparse->info().access);
m_execBarriers.accessBuffer(bufferHandle,
VK_PIPELINE_STAGE_TRANSFER_BIT,
ToBuffer ? VK_ACCESS_TRANSFER_WRITE_BIT : VK_ACCESS_TRANSFER_READ_BIT,
buffer->info().stages,
buffer->info().access);
m_cmd->trackResource<ToBuffer ? DxvkAccess::Read : DxvkAccess::Write>(sparse);
m_cmd->trackResource<ToBuffer ? DxvkAccess::Write : DxvkAccess::Read>(buffer);
}
void DxvkContext::resolveImageHw(
const Rc<DxvkImage>& dstImage,
const Rc<DxvkImage>& srcImage,
const VkImageResolve& region) {
auto dstSubresourceRange = vk::makeSubresourceRange(region.dstSubresource);
auto srcSubresourceRange = vk::makeSubresourceRange(region.srcSubresource);
if (m_execBarriers.isImageDirty(dstImage, dstSubresourceRange, DxvkAccess::Write)
|| m_execBarriers.isImageDirty(srcImage, srcSubresourceRange, DxvkAccess::Write))
m_execBarriers.recordCommands(m_cmd);
// We only support resolving to the entire image
// area, so we might as well discard its contents
VkImageLayout dstLayout = dstImage->pickLayout(VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL);
VkImageLayout srcLayout = srcImage->pickLayout(VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL);
VkImageLayout initialLayout = dstImage->info().layout;
if (dstImage->isFullSubresource(region.dstSubresource, region.extent))
initialLayout = VK_IMAGE_LAYOUT_UNDEFINED;
if (dstLayout != initialLayout) {
m_execAcquires.accessImage(
dstImage, dstSubresourceRange, initialLayout,
VK_PIPELINE_STAGE_TRANSFER_BIT, 0,
dstLayout,
VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_ACCESS_TRANSFER_WRITE_BIT);
}
if (srcLayout != srcImage->info().layout) {
m_execAcquires.accessImage(
srcImage, srcSubresourceRange,
srcImage->info().layout,
VK_PIPELINE_STAGE_TRANSFER_BIT, 0,
srcLayout,
VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_ACCESS_TRANSFER_READ_BIT);
}
m_execAcquires.recordCommands(m_cmd);
VkImageResolve2 resolveRegion = { VK_STRUCTURE_TYPE_IMAGE_RESOLVE_2 };
resolveRegion.srcSubresource = region.srcSubresource;
resolveRegion.srcOffset = region.srcOffset;
resolveRegion.dstSubresource = region.dstSubresource;
resolveRegion.dstOffset = region.dstOffset;
resolveRegion.extent = region.extent;
VkResolveImageInfo2 resolveInfo = { VK_STRUCTURE_TYPE_RESOLVE_IMAGE_INFO_2 };
resolveInfo.srcImage = srcImage->handle();
resolveInfo.srcImageLayout = srcLayout;
resolveInfo.dstImage = dstImage->handle();
resolveInfo.dstImageLayout = dstLayout;
resolveInfo.regionCount = 1;
resolveInfo.pRegions = &resolveRegion;
m_cmd->cmdResolveImage(&resolveInfo);
m_execBarriers.accessImage(
dstImage, dstSubresourceRange, dstLayout,
VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_ACCESS_TRANSFER_WRITE_BIT,
dstImage->info().layout,
dstImage->info().stages,
dstImage->info().access);
m_execBarriers.accessImage(
srcImage, srcSubresourceRange, srcLayout,
VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_ACCESS_TRANSFER_READ_BIT,
srcImage->info().layout,
srcImage->info().stages,
srcImage->info().access);
m_cmd->trackResource<DxvkAccess::Write>(dstImage);
m_cmd->trackResource<DxvkAccess::Read>(srcImage);
}
void DxvkContext::resolveImageDs(
const Rc<DxvkImage>& dstImage,
const Rc<DxvkImage>& srcImage,
const VkImageResolve& region,
VkResolveModeFlagBits depthMode,
VkResolveModeFlagBits stencilMode) {
auto dstSubresourceRange = vk::makeSubresourceRange(region.dstSubresource);
auto srcSubresourceRange = vk::makeSubresourceRange(region.srcSubresource);
if (m_execBarriers.isImageDirty(dstImage, dstSubresourceRange, DxvkAccess::Write)
|| m_execBarriers.isImageDirty(srcImage, srcSubresourceRange, DxvkAccess::Write))
m_execBarriers.recordCommands(m_cmd);
// Transition both images to usable layouts if necessary. For the source image we
// can be fairly leniet since writable layouts are allowed for resolve attachments.
VkImageLayout dstLayout = dstImage->pickLayout(VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL);
VkImageLayout srcLayout = srcImage->info().layout;
if (srcLayout != VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL
&& srcLayout != VK_IMAGE_LAYOUT_DEPTH_STENCIL_READ_ONLY_OPTIMAL)
srcLayout = srcImage->pickLayout(VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL);
if (srcImage->info().layout != srcLayout) {
m_execAcquires.accessImage(srcImage, srcSubresourceRange,
srcImage->info().layout,
srcImage->info().stages, 0,
srcLayout,
VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT,
VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT);
}
if (dstImage->info().layout != dstLayout) {
m_execAcquires.accessImage(dstImage, dstSubresourceRange,
VK_IMAGE_LAYOUT_UNDEFINED, dstImage->info().stages, 0,
dstLayout,
VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT,
VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT);
}
m_execAcquires.recordCommands(m_cmd);
// Create a pair of views for the attachment resolve
Rc<DxvkMetaResolveViews> views = new DxvkMetaResolveViews(m_device->vkd(),
dstImage, region.dstSubresource, srcImage, region.srcSubresource,
dstImage->info().format);
VkRenderingAttachmentInfo depthAttachment = { VK_STRUCTURE_TYPE_RENDERING_ATTACHMENT_INFO };
depthAttachment.imageView = views->getSrcView();
depthAttachment.imageLayout = srcLayout;
depthAttachment.resolveMode = depthMode;
depthAttachment.resolveImageView = views->getDstView();
depthAttachment.resolveImageLayout = dstLayout;
depthAttachment.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD;
depthAttachment.storeOp = VK_ATTACHMENT_STORE_OP_STORE;
VkRenderingAttachmentInfo stencilAttachment = depthAttachment;
stencilAttachment.resolveMode = stencilMode;
VkExtent3D extent = dstImage->mipLevelExtent(region.dstSubresource.mipLevel);
VkRenderingInfo renderingInfo = { VK_STRUCTURE_TYPE_RENDERING_INFO };
renderingInfo.renderArea.offset = VkOffset2D { 0, 0 };
renderingInfo.renderArea.extent = VkExtent2D { extent.width, extent.height };
renderingInfo.layerCount = region.dstSubresource.layerCount;
if (dstImage->formatInfo()->aspectMask & VK_IMAGE_ASPECT_DEPTH_BIT)
renderingInfo.pDepthAttachment = &depthAttachment;
if (dstImage->formatInfo()->aspectMask & VK_IMAGE_ASPECT_STENCIL_BIT)
renderingInfo.pStencilAttachment = &stencilAttachment;
m_cmd->cmdBeginRendering(&renderingInfo);
m_cmd->cmdEndRendering();
// Add barriers for the resolve operation
m_execBarriers.accessImage(srcImage, srcSubresourceRange,
srcLayout,
VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT,
VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT,
srcImage->info().layout,
srcImage->info().stages,
srcImage->info().access);
m_execBarriers.accessImage(dstImage, dstSubresourceRange,
dstLayout,
VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT,
VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT,
dstImage->info().layout,
dstImage->info().stages,
dstImage->info().access);
m_cmd->trackResource<DxvkAccess::Write>(dstImage);
m_cmd->trackResource<DxvkAccess::Read>(srcImage);
m_cmd->trackResource<DxvkAccess::None>(views);
}
void DxvkContext::resolveImageFb(
const Rc<DxvkImage>& dstImage,
const Rc<DxvkImage>& srcImage,
const VkImageResolve& region,
VkFormat format,
VkResolveModeFlagBits depthMode,
VkResolveModeFlagBits stencilMode) {
this->invalidateState();
auto dstSubresourceRange = vk::makeSubresourceRange(region.dstSubresource);
auto srcSubresourceRange = vk::makeSubresourceRange(region.srcSubresource);
if (m_execBarriers.isImageDirty(dstImage, dstSubresourceRange, DxvkAccess::Write)
|| m_execBarriers.isImageDirty(srcImage, srcSubresourceRange, DxvkAccess::Write))
m_execBarriers.recordCommands(m_cmd);
// Discard the destination image if we're fully writing it,
// and transition the image layout if necessary
bool doDiscard = dstImage->isFullSubresource(region.dstSubresource, region.extent);
if (region.dstSubresource.aspectMask & VK_IMAGE_ASPECT_DEPTH_BIT)
doDiscard &= depthMode != VK_RESOLVE_MODE_NONE;
if (region.dstSubresource.aspectMask & VK_IMAGE_ASPECT_STENCIL_BIT)
doDiscard &= stencilMode != VK_RESOLVE_MODE_NONE;
VkPipelineStageFlags dstStages;
VkImageLayout dstLayout;
VkAccessFlags dstAccess;
if (region.dstSubresource.aspectMask & VK_IMAGE_ASPECT_COLOR_BIT) {
dstLayout = dstImage->pickLayout(VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL);
dstStages = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT;
dstAccess = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT;
if (!doDiscard)
dstAccess |= VK_ACCESS_COLOR_ATTACHMENT_READ_BIT;
} else {
dstLayout = dstImage->pickLayout(VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL);
dstStages = VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT
| VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT;
dstAccess = VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT;
if (!doDiscard)
dstAccess |= VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT;
}
if (dstImage->info().layout != dstLayout) {
m_execAcquires.accessImage(
dstImage, dstSubresourceRange,
doDiscard ? VK_IMAGE_LAYOUT_UNDEFINED
: dstImage->info().layout,
dstImage->info().stages, 0,
dstLayout, dstStages, dstAccess);
}
// Check source image layout, and try to avoid transitions if we can
VkImageLayout srcLayout = srcImage->info().layout;
if (srcLayout != VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL
&& srcLayout != VK_IMAGE_LAYOUT_DEPTH_STENCIL_READ_ONLY_OPTIMAL) {
srcLayout = (region.srcSubresource.aspectMask & VK_IMAGE_ASPECT_COLOR_BIT & VK_IMAGE_ASPECT_COLOR_BIT)
? srcImage->pickLayout(VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL)
: srcImage->pickLayout(VK_IMAGE_LAYOUT_DEPTH_STENCIL_READ_ONLY_OPTIMAL);
}
if (srcImage->info().layout != srcLayout) {
m_execAcquires.accessImage(
srcImage, srcSubresourceRange,
srcImage->info().layout,
srcImage->info().stages, 0,
srcLayout,
VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT,
VK_ACCESS_SHADER_READ_BIT);
}
m_execAcquires.recordCommands(m_cmd);
// Create a framebuffer and pipeline for the resolve op
VkFormat dstFormat = format ? format : dstImage->info().format;
VkFormat srcFormat = format ? format : srcImage->info().format;
VkExtent3D passExtent = dstImage->mipLevelExtent(region.dstSubresource.mipLevel);
Rc<DxvkMetaCopyViews> views = new DxvkMetaCopyViews(m_device->vkd(),
dstImage, region.dstSubresource, dstFormat,
srcImage, region.srcSubresource, srcFormat);
DxvkMetaResolvePipeline pipeInfo = m_common->metaResolve().getPipeline(
dstFormat, srcImage->info().sampleCount, depthMode, stencilMode);
// Create and initialize descriptor set
VkDescriptorSet descriptorSet = m_descriptorPool->alloc(pipeInfo.dsetLayout);
std::array<VkDescriptorImageInfo, 2> descriptorImages = {{
{ VK_NULL_HANDLE, views->getSrcView(), srcLayout },
{ VK_NULL_HANDLE, views->getSrcStencilView(), srcLayout },
}};
std::array<VkWriteDescriptorSet, 2> descriptorWrites;
for (uint32_t i = 0; i < descriptorWrites.size(); i++) {
descriptorWrites[i] = { VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET };
descriptorWrites[i].dstSet = descriptorSet;
descriptorWrites[i].dstBinding = i;
descriptorWrites[i].descriptorCount = 1;
descriptorWrites[i].descriptorType = VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE;
descriptorWrites[i].pImageInfo = &descriptorImages[i];
}
m_cmd->updateDescriptorSets(
descriptorWrites.size(),
descriptorWrites.data());
// Set up render state
VkViewport viewport;
viewport.x = float(region.dstOffset.x);
viewport.y = float(region.dstOffset.y);
viewport.width = float(region.extent.width);
viewport.height = float(region.extent.height);
viewport.minDepth = 0.0f;
viewport.maxDepth = 1.0f;
VkRect2D scissor;
scissor.offset = { region.dstOffset.x, region.dstOffset.y };
scissor.extent = { region.extent.width, region.extent.height };
VkRenderingAttachmentInfo attachmentInfo = { VK_STRUCTURE_TYPE_RENDERING_ATTACHMENT_INFO };
attachmentInfo.imageView = views->getDstView();
attachmentInfo.imageLayout = dstLayout;
attachmentInfo.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD;
attachmentInfo.storeOp = VK_ATTACHMENT_STORE_OP_STORE;
if (doDiscard)
attachmentInfo.loadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
VkRenderingInfo renderingInfo = { VK_STRUCTURE_TYPE_RENDERING_INFO };
renderingInfo.renderArea.offset = VkOffset2D { 0, 0 };
renderingInfo.renderArea.extent = VkExtent2D { passExtent.width, passExtent.height };
renderingInfo.layerCount = region.dstSubresource.layerCount;
VkImageAspectFlags dstAspects = dstImage->formatInfo()->aspectMask;
if (dstAspects & VK_IMAGE_ASPECT_COLOR_BIT) {
renderingInfo.colorAttachmentCount = 1;
renderingInfo.pColorAttachments = &attachmentInfo;
} else {
if (dstAspects & VK_IMAGE_ASPECT_DEPTH_BIT)
renderingInfo.pDepthAttachment = &attachmentInfo;
if (dstAspects & VK_IMAGE_ASPECT_STENCIL_BIT)
renderingInfo.pStencilAttachment = &attachmentInfo;
}
// Perform the actual resolve operation
VkOffset2D srcOffset = {
region.srcOffset.x - region.dstOffset.x,
region.srcOffset.y - region.dstOffset.y };
m_cmd->cmdBeginRendering(&renderingInfo);
m_cmd->cmdBindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeInfo.pipeHandle);
m_cmd->cmdBindDescriptorSet(VK_PIPELINE_BIND_POINT_GRAPHICS,
pipeInfo.pipeLayout, descriptorSet, 0, nullptr);
m_cmd->cmdSetViewport(1, &viewport);
m_cmd->cmdSetScissor(1, &scissor);
m_cmd->cmdPushConstants(pipeInfo.pipeLayout,
VK_SHADER_STAGE_FRAGMENT_BIT,
0, sizeof(srcOffset), &srcOffset);
m_cmd->cmdDraw(3, region.dstSubresource.layerCount, 0, 0);
m_cmd->cmdEndRendering();
m_execBarriers.accessImage(
srcImage, srcSubresourceRange, srcLayout,
VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, 0,
srcImage->info().layout,
srcImage->info().stages,
srcImage->info().access);
m_execBarriers.accessImage(
dstImage, dstSubresourceRange,
dstLayout, dstStages, dstAccess,
dstImage->info().layout,
dstImage->info().stages,
dstImage->info().access);
m_cmd->trackResource<DxvkAccess::Write>(dstImage);
m_cmd->trackResource<DxvkAccess::Read>(srcImage);
m_cmd->trackResource<DxvkAccess::None>(views);
}
void DxvkContext::startRenderPass() {
if (!m_flags.test(DxvkContextFlag::GpRenderPassBound)) {
this->applyRenderTargetLoadLayouts();
this->flushClears(true);
// Make sure all graphics state gets reapplied on the next draw
m_descriptorState.dirtyStages(VK_SHADER_STAGE_ALL_GRAPHICS);
m_flags.set(
DxvkContextFlag::GpRenderPassBound,
DxvkContextFlag::GpDirtyPipeline,
DxvkContextFlag::GpDirtyPipelineState,
DxvkContextFlag::GpDirtyVertexBuffers,
DxvkContextFlag::GpDirtyIndexBuffer,
DxvkContextFlag::GpDirtyXfbBuffers,
DxvkContextFlag::GpDirtyBlendConstants,
DxvkContextFlag::GpDirtyStencilRef,
DxvkContextFlag::GpDirtyMultisampleState,
DxvkContextFlag::GpDirtyRasterizerState,
DxvkContextFlag::GpDirtyViewport,
DxvkContextFlag::GpDirtyDepthBias,
DxvkContextFlag::GpDirtyDepthBounds,
DxvkContextFlag::GpDirtyDepthStencilState,
DxvkContextFlag::DirtyPushConstants);
m_flags.clr(
DxvkContextFlag::GpRenderPassSuspended,
DxvkContextFlag::GpIndependentSets);
this->renderPassBindFramebuffer(
m_state.om.framebufferInfo,
m_state.om.renderPassOps);
// Track the final layout of each render target
this->applyRenderTargetStoreLayouts();
// Don't discard image contents if we have
// to spill the current render pass
this->resetRenderPassOps(
m_state.om.renderTargets,
m_state.om.renderPassOps);
// Begin occlusion queries
m_queryManager.beginQueries(m_cmd, VK_QUERY_TYPE_OCCLUSION);
m_queryManager.beginQueries(m_cmd, VK_QUERY_TYPE_PIPELINE_STATISTICS);
}
}
void DxvkContext::spillRenderPass(bool suspend) {
if (m_flags.test(DxvkContextFlag::GpRenderPassBound)) {
m_flags.clr(DxvkContextFlag::GpRenderPassBound);
this->pauseTransformFeedback();
m_queryManager.endQueries(m_cmd, VK_QUERY_TYPE_OCCLUSION);
m_queryManager.endQueries(m_cmd, VK_QUERY_TYPE_PIPELINE_STATISTICS);
this->renderPassUnbindFramebuffer();
if (suspend)
m_flags.set(DxvkContextFlag::GpRenderPassSuspended);
else
this->transitionRenderTargetLayouts(false);
m_execBarriers.recordCommands(m_cmd);
} else if (!suspend) {
// We may end a previously suspended render pass
if (m_flags.test(DxvkContextFlag::GpRenderPassSuspended)) {
m_flags.clr(DxvkContextFlag::GpRenderPassSuspended);
this->transitionRenderTargetLayouts(false);
m_execBarriers.recordCommands(m_cmd);
}
// Execute deferred clears if necessary
this->flushClears(false);
}
}
void DxvkContext::renderPassEmitInitBarriers(
const DxvkFramebufferInfo& framebufferInfo,
const DxvkRenderPassOps& ops) {
// If any of the involved images are dirty, emit all pending barriers now.
// Otherwise, skip this step so that we can more efficiently batch barriers.
for (uint32_t i = 0; i < framebufferInfo.numAttachments(); i++) {
const auto& attachment = framebufferInfo.getAttachment(i);
if (m_execBarriers.isImageDirty(
attachment.view->image(),
attachment.view->imageSubresources(),
DxvkAccess::Write)) {
m_execBarriers.recordCommands(m_cmd);
break;
}
}
// Transition all images to the render layout as necessary
const auto& depthAttachment = framebufferInfo.getDepthTarget();
if (depthAttachment.layout != ops.depthOps.loadLayout
&& depthAttachment.view != nullptr) {
VkImageAspectFlags depthAspects = depthAttachment.view->info().aspect;
VkPipelineStageFlags depthStages =
VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT;
VkAccessFlags depthAccess = 0;
if (((depthAspects & VK_IMAGE_ASPECT_DEPTH_BIT) && ops.depthOps.loadOpD == VK_ATTACHMENT_LOAD_OP_LOAD)
|| ((depthAspects & VK_IMAGE_ASPECT_STENCIL_BIT) && ops.depthOps.loadOpS == VK_ATTACHMENT_LOAD_OP_LOAD))
depthAccess |= VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT;
if (((depthAspects & VK_IMAGE_ASPECT_DEPTH_BIT) && ops.depthOps.loadOpD != VK_ATTACHMENT_LOAD_OP_LOAD)
|| ((depthAspects & VK_IMAGE_ASPECT_STENCIL_BIT) && ops.depthOps.loadOpS != VK_ATTACHMENT_LOAD_OP_LOAD)
|| (depthAttachment.layout != VK_IMAGE_LAYOUT_DEPTH_STENCIL_READ_ONLY_OPTIMAL))
depthAccess |= VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT;
if (depthAttachment.layout != VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL) {
depthStages |= m_device->getShaderPipelineStages();
depthAccess |= VK_ACCESS_SHADER_READ_BIT;
}
m_execBarriers.accessImage(
depthAttachment.view->image(),
depthAttachment.view->imageSubresources(),
ops.depthOps.loadLayout,
depthStages, 0,
depthAttachment.layout,
depthStages, depthAccess);
}
for (uint32_t i = 0; i < MaxNumRenderTargets; i++) {
const auto& colorAttachment = framebufferInfo.getColorTarget(i);
if (colorAttachment.layout != ops.colorOps[i].loadLayout
&& colorAttachment.view != nullptr) {
VkAccessFlags colorAccess = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT;
if (ops.colorOps[i].loadOp == VK_ATTACHMENT_LOAD_OP_LOAD)
colorAccess |= VK_ACCESS_COLOR_ATTACHMENT_READ_BIT;
m_execBarriers.accessImage(
colorAttachment.view->image(),
colorAttachment.view->imageSubresources(),
ops.colorOps[i].loadLayout,
VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT, 0,
colorAttachment.layout,
VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT,
colorAccess);
}
}
// Unconditionally emit barriers here. We need to do this
// even if there are no layout transitions, since we don't
// track resource usage during render passes.
m_execBarriers.recordCommands(m_cmd);
}
void DxvkContext::renderPassEmitPostBarriers(
const DxvkFramebufferInfo& framebufferInfo,
const DxvkRenderPassOps& ops) {
const auto& depthAttachment = framebufferInfo.getDepthTarget();
if (depthAttachment.view != nullptr) {
if (depthAttachment.layout != ops.depthOps.storeLayout) {
m_execBarriers.accessImage(
depthAttachment.view->image(),
depthAttachment.view->imageSubresources(),
depthAttachment.layout,
VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT,
VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT |
VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT,
ops.depthOps.storeLayout,
depthAttachment.view->imageInfo().stages,
depthAttachment.view->imageInfo().access);
} else {
VkAccessFlags srcAccess = VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT;
if (depthAttachment.layout != VK_IMAGE_LAYOUT_DEPTH_STENCIL_READ_ONLY_OPTIMAL)
srcAccess |= VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT;
m_execBarriers.accessMemory(
VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT,
srcAccess,
depthAttachment.view->imageInfo().stages,
depthAttachment.view->imageInfo().access);
}
}
for (uint32_t i = 0; i < MaxNumRenderTargets; i++) {
const auto& colorAttachment = framebufferInfo.getColorTarget(i);
if (colorAttachment.view != nullptr) {
if (colorAttachment.layout != ops.colorOps[i].storeLayout) {
m_execBarriers.accessImage(
colorAttachment.view->image(),
colorAttachment.view->imageSubresources(),
colorAttachment.layout,
VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT,
VK_ACCESS_COLOR_ATTACHMENT_READ_BIT |
VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT,
ops.colorOps[i].storeLayout,
colorAttachment.view->imageInfo().stages,
colorAttachment.view->imageInfo().access);
} else {
m_execBarriers.accessMemory(
VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT,
VK_ACCESS_COLOR_ATTACHMENT_READ_BIT |
VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT,
colorAttachment.view->imageInfo().stages,
colorAttachment.view->imageInfo().access);
}
}
}
// Do not flush barriers here. This is intended since
// we pre-record them when binding the framebuffer.
}
void DxvkContext::renderPassBindFramebuffer(
const DxvkFramebufferInfo& framebufferInfo,
const DxvkRenderPassOps& ops) {
const DxvkFramebufferSize fbSize = framebufferInfo.size();
this->renderPassEmitInitBarriers(framebufferInfo, ops);
this->renderPassEmitPostBarriers(framebufferInfo, ops);
uint32_t colorInfoCount = 0;
std::array<VkRenderingAttachmentInfo, MaxNumRenderTargets> colorInfos;
for (uint32_t i = 0; i < MaxNumRenderTargets; i++) {
const auto& colorTarget = framebufferInfo.getColorTarget(i);
colorInfos[i] = { VK_STRUCTURE_TYPE_RENDERING_ATTACHMENT_INFO };
if (colorTarget.view != nullptr) {
colorInfos[i].imageView = colorTarget.view->handle();
colorInfos[i].imageLayout = colorTarget.layout;
colorInfos[i].loadOp = ops.colorOps[i].loadOp;
colorInfos[i].storeOp = VK_ATTACHMENT_STORE_OP_STORE;
if (ops.colorOps[i].loadOp == VK_ATTACHMENT_LOAD_OP_CLEAR)
colorInfos[i].clearValue.color = ops.colorOps[i].clearValue;
colorInfoCount = i + 1;
}
}
VkRenderingAttachmentInfo depthInfo = { VK_STRUCTURE_TYPE_RENDERING_ATTACHMENT_INFO };
VkImageAspectFlags depthStencilAspects = 0;
if (framebufferInfo.getDepthTarget().view != nullptr) {
const auto& depthTarget = framebufferInfo.getDepthTarget();
depthStencilAspects = depthTarget.view->info().aspect;
depthInfo.imageView = depthTarget.view->handle();
depthInfo.imageLayout = depthTarget.layout;
depthInfo.loadOp = ops.depthOps.loadOpD;
depthInfo.storeOp = VK_ATTACHMENT_STORE_OP_STORE;
if (ops.depthOps.loadOpD == VK_ATTACHMENT_LOAD_OP_CLEAR)
depthInfo.clearValue.depthStencil.depth = ops.depthOps.clearValue.depth;
}
VkRenderingAttachmentInfo stencilInfo = depthInfo;
if (framebufferInfo.getDepthTarget().view != nullptr) {
stencilInfo.loadOp = ops.depthOps.loadOpS;
stencilInfo.storeOp = VK_ATTACHMENT_STORE_OP_STORE;
if (ops.depthOps.loadOpS == VK_ATTACHMENT_LOAD_OP_CLEAR)
stencilInfo.clearValue.depthStencil.stencil = ops.depthOps.clearValue.stencil;
}
VkRenderingInfo renderingInfo = { VK_STRUCTURE_TYPE_RENDERING_INFO };
renderingInfo.renderArea.offset = VkOffset2D { 0, 0 };
renderingInfo.renderArea.extent = VkExtent2D { fbSize.width, fbSize.height };
renderingInfo.layerCount = fbSize.layers;
if (colorInfoCount) {
renderingInfo.colorAttachmentCount = colorInfoCount;
renderingInfo.pColorAttachments = colorInfos.data();
}
if (depthStencilAspects & VK_IMAGE_ASPECT_DEPTH_BIT)
renderingInfo.pDepthAttachment = &depthInfo;
if (depthStencilAspects & VK_IMAGE_ASPECT_STENCIL_BIT)
renderingInfo.pStencilAttachment = &stencilInfo;
m_cmd->cmdBeginRendering(&renderingInfo);
for (uint32_t i = 0; i < framebufferInfo.numAttachments(); i++) {
m_cmd->trackResource<DxvkAccess::None> (framebufferInfo.getAttachment(i).view);
m_cmd->trackResource<DxvkAccess::Write>(framebufferInfo.getAttachment(i).view->image());
}
m_cmd->addStatCtr(DxvkStatCounter::CmdRenderPassCount, 1);
}
void DxvkContext::renderPassUnbindFramebuffer() {
m_cmd->cmdEndRendering();
// If there are pending layout transitions, execute them immediately
// since the backend expects images to be in the store layout after
// a render pass instance. This is expected to be rare.
if (m_execBarriers.hasResourceBarriers())
m_execBarriers.recordCommands(m_cmd);
}
void DxvkContext::resetRenderPassOps(
const DxvkRenderTargets& renderTargets,
DxvkRenderPassOps& renderPassOps) {
if (renderTargets.depth.view != nullptr) {
renderPassOps.depthOps = DxvkDepthAttachmentOps {
VK_ATTACHMENT_LOAD_OP_LOAD, VK_ATTACHMENT_LOAD_OP_LOAD,
renderTargets.depth.layout, renderTargets.depth.layout };
} else {
renderPassOps.depthOps = DxvkDepthAttachmentOps { };
}
for (uint32_t i = 0; i < MaxNumRenderTargets; i++) {
if (renderTargets.color[i].view != nullptr) {
renderPassOps.colorOps[i] = DxvkColorAttachmentOps {
VK_ATTACHMENT_LOAD_OP_LOAD,
renderTargets.color[i].layout,
renderTargets.color[i].layout };
} else {
renderPassOps.colorOps[i] = DxvkColorAttachmentOps { };
}
}
}
void DxvkContext::startTransformFeedback() {
if (!m_flags.test(DxvkContextFlag::GpXfbActive)) {
m_flags.set(DxvkContextFlag::GpXfbActive);
VkBuffer ctrBuffers[MaxNumXfbBuffers];
VkDeviceSize ctrOffsets[MaxNumXfbBuffers];
for (uint32_t i = 0; i < MaxNumXfbBuffers; i++) {
auto physSlice = m_state.xfb.counters[i].getSliceHandle();
ctrBuffers[i] = physSlice.handle;
ctrOffsets[i] = physSlice.offset;
if (physSlice.handle != VK_NULL_HANDLE)
m_cmd->trackResource<DxvkAccess::Read>(m_state.xfb.counters[i].buffer());
}
m_cmd->cmdBeginTransformFeedback(
0, MaxNumXfbBuffers, ctrBuffers, ctrOffsets);
m_queryManager.beginQueries(m_cmd,
VK_QUERY_TYPE_TRANSFORM_FEEDBACK_STREAM_EXT);
}
}
void DxvkContext::pauseTransformFeedback() {
if (m_flags.test(DxvkContextFlag::GpXfbActive)) {
m_flags.clr(DxvkContextFlag::GpXfbActive);
VkBuffer ctrBuffers[MaxNumXfbBuffers];
VkDeviceSize ctrOffsets[MaxNumXfbBuffers];
for (uint32_t i = 0; i < MaxNumXfbBuffers; i++) {
auto physSlice = m_state.xfb.counters[i].getSliceHandle();
ctrBuffers[i] = physSlice.handle;
ctrOffsets[i] = physSlice.offset;
if (physSlice.handle != VK_NULL_HANDLE)
m_cmd->trackResource<DxvkAccess::Write>(m_state.xfb.counters[i].buffer());
}
m_queryManager.endQueries(m_cmd,
VK_QUERY_TYPE_TRANSFORM_FEEDBACK_STREAM_EXT);
m_cmd->cmdEndTransformFeedback(
0, MaxNumXfbBuffers, ctrBuffers, ctrOffsets);
}
}
void DxvkContext::unbindComputePipeline() {
m_flags.set(DxvkContextFlag::CpDirtyPipelineState);
m_state.cp.pipeline = nullptr;
}
bool DxvkContext::updateComputePipelineState() {
if (unlikely(m_state.gp.pipeline != nullptr))
this->unbindGraphicsPipeline();
// Look up pipeline object based on the bound compute shader
auto newPipeline = lookupComputePipeline(m_state.cp.shaders);
m_state.cp.pipeline = newPipeline;
if (unlikely(!newPipeline))
return false;
if (unlikely(newPipeline->getSpecConstantMask() != m_state.cp.constants.mask))
this->resetSpecConstants<VK_PIPELINE_BIND_POINT_COMPUTE>(newPipeline->getSpecConstantMask());
if (m_flags.test(DxvkContextFlag::CpDirtySpecConstants))
this->updateSpecConstants<VK_PIPELINE_BIND_POINT_COMPUTE>();
// Look up Vulkan pipeline handle for the given compute state
auto pipelineHandle = newPipeline->getPipelineHandle(m_state.cp.state);
if (unlikely(!pipelineHandle))
return false;
m_cmd->cmdBindPipeline(
VK_PIPELINE_BIND_POINT_COMPUTE,
pipelineHandle);
// Mark compute resources and push constants as dirty
m_descriptorState.dirtyStages(VK_SHADER_STAGE_COMPUTE_BIT);
if (newPipeline->getBindings()->layout().getPushConstantRange().size)
m_flags.set(DxvkContextFlag::DirtyPushConstants);
m_flags.clr(DxvkContextFlag::CpDirtyPipelineState);
return true;
}
void DxvkContext::unbindGraphicsPipeline() {
m_flags.set(
DxvkContextFlag::GpDirtyPipeline,
DxvkContextFlag::GpDirtyPipelineState,
DxvkContextFlag::GpDirtyVertexBuffers,
DxvkContextFlag::GpDirtyIndexBuffer,
DxvkContextFlag::GpDirtyXfbBuffers,
DxvkContextFlag::GpDirtyBlendConstants,
DxvkContextFlag::GpDirtyStencilRef,
DxvkContextFlag::GpDirtyMultisampleState,
DxvkContextFlag::GpDirtyRasterizerState,
DxvkContextFlag::GpDirtyViewport,
DxvkContextFlag::GpDirtyDepthBias,
DxvkContextFlag::GpDirtyDepthBounds,
DxvkContextFlag::GpDirtyDepthStencilState);
m_state.gp.pipeline = nullptr;
}
bool DxvkContext::updateGraphicsPipeline() {
if (unlikely(m_state.cp.pipeline != nullptr))
this->unbindComputePipeline();
auto newPipeline = lookupGraphicsPipeline(m_state.gp.shaders);
m_state.gp.pipeline = newPipeline;
if (unlikely(!newPipeline)) {
m_state.gp.flags = DxvkGraphicsPipelineFlags();
return false;
}
if (m_features.test(DxvkContextFeature::TrackGraphicsPipeline))
m_cmd->trackGraphicsPipeline(newPipeline);
if (unlikely(newPipeline->getSpecConstantMask() != m_state.gp.constants.mask))
this->resetSpecConstants<VK_PIPELINE_BIND_POINT_GRAPHICS>(newPipeline->getSpecConstantMask());
if (m_state.gp.flags != newPipeline->flags()) {
m_state.gp.flags = newPipeline->flags();
// Force-update vertex/index buffers for hazard checks
m_flags.set(DxvkContextFlag::GpDirtyIndexBuffer,
DxvkContextFlag::GpDirtyVertexBuffers,
DxvkContextFlag::GpDirtyXfbBuffers,
DxvkContextFlag::DirtyDrawBuffer);
// This is necessary because we'll only do hazard
// tracking if the active pipeline has side effects
if (!m_barrierControl.test(DxvkBarrierControl::IgnoreGraphicsBarriers))
this->spillRenderPass(true);
}
m_descriptorState.dirtyStages(VK_SHADER_STAGE_ALL_GRAPHICS);
if (newPipeline->getBindings()->layout().getPushConstantRange().size)
m_flags.set(DxvkContextFlag::DirtyPushConstants);
m_flags.clr(DxvkContextFlag::GpDirtyPipeline);
return true;
}
bool DxvkContext::updateGraphicsPipelineState(DxvkGlobalPipelineBarrier srcBarrier) {
bool oldIndependentSets = m_flags.test(DxvkContextFlag::GpIndependentSets);
// Check which dynamic states need to be active. States that
// are not dynamic will be invalidated in the command buffer.
m_flags.clr(DxvkContextFlag::GpDynamicBlendConstants,
DxvkContextFlag::GpDynamicDepthStencilState,
DxvkContextFlag::GpDynamicDepthBias,
DxvkContextFlag::GpDynamicDepthBounds,
DxvkContextFlag::GpDynamicStencilRef,
DxvkContextFlag::GpDynamicMultisampleState,
DxvkContextFlag::GpDynamicRasterizerState,
DxvkContextFlag::GpIndependentSets);
m_flags.set(m_state.gp.state.useDynamicBlendConstants()
? DxvkContextFlag::GpDynamicBlendConstants
: DxvkContextFlag::GpDirtyBlendConstants);
m_flags.set((!m_state.gp.flags.test(DxvkGraphicsPipelineFlag::HasRasterizerDiscard))
? DxvkContextFlag::GpDynamicRasterizerState
: DxvkContextFlag::GpDirtyRasterizerState);
// Retrieve and bind actual Vulkan pipeline handle
auto pipelineInfo = m_state.gp.pipeline->getPipelineHandle(m_state.gp.state);
if (unlikely(!pipelineInfo.first))
return false;
m_cmd->cmdBindPipeline(
VK_PIPELINE_BIND_POINT_GRAPHICS,
pipelineInfo.first);
// For pipelines created from graphics pipeline libraries, we need to
// apply a bunch of dynamic state that is otherwise static or unused
if (pipelineInfo.second == DxvkGraphicsPipelineType::BasePipeline) {
m_flags.set(
DxvkContextFlag::GpDynamicDepthStencilState,
DxvkContextFlag::GpDynamicDepthBias,
DxvkContextFlag::GpDynamicStencilRef,
DxvkContextFlag::GpIndependentSets);
if (m_device->features().core.features.depthBounds)
m_flags.set(DxvkContextFlag::GpDynamicDepthBounds);
if (m_device->features().extExtendedDynamicState3.extendedDynamicState3RasterizationSamples
&& m_device->features().extExtendedDynamicState3.extendedDynamicState3SampleMask)
m_flags.set(DxvkContextFlag::GpDynamicMultisampleState);
} else {
m_flags.set(m_state.gp.state.useDynamicDepthBias()
? DxvkContextFlag::GpDynamicDepthBias
: DxvkContextFlag::GpDirtyDepthBias);
m_flags.set(m_state.gp.state.useDynamicDepthBounds()
? DxvkContextFlag::GpDynamicDepthBounds
: DxvkContextFlag::GpDirtyDepthBounds);
m_flags.set(m_state.gp.state.useDynamicStencilRef()
? DxvkContextFlag::GpDynamicStencilRef
: DxvkContextFlag::GpDirtyStencilRef);
m_flags.set(
DxvkContextFlag::GpDirtyDepthStencilState,
DxvkContextFlag::GpDirtyMultisampleState);
}
// If necessary, dirty descriptor sets due to layout incompatibilities
bool newIndependentSets = m_flags.test(DxvkContextFlag::GpIndependentSets);
if (newIndependentSets != oldIndependentSets)
m_descriptorState.dirtyStages(VK_SHADER_STAGE_ALL_GRAPHICS);
// Emit barrier based on pipeline properties, in order to avoid
// accidental write-after-read hazards after the render pass.
DxvkGlobalPipelineBarrier pipelineBarrier = m_state.gp.pipeline->getGlobalBarrier(m_state.gp.state);
srcBarrier.stages |= pipelineBarrier.stages;
srcBarrier.access |= pipelineBarrier.access;
if (srcBarrier.stages) {
DxvkAccessFlags access = DxvkBarrierSet::getAccessTypes(srcBarrier.access);
DxvkGlobalPipelineBarrier dstBarrier = access.test(DxvkAccess::Write)
? m_globalRwGraphicsBarrier
: m_globalRoGraphicsBarrier;
m_execBarriers.accessMemory(
srcBarrier.stages, srcBarrier.access,
dstBarrier.stages, dstBarrier.access);
}
m_flags.clr(DxvkContextFlag::GpDirtyPipelineState);
return true;
}
template<VkPipelineBindPoint BindPoint>
void DxvkContext::resetSpecConstants(
uint32_t newMask) {
auto& scInfo = BindPoint == VK_PIPELINE_BIND_POINT_GRAPHICS ? m_state.gp.state.sc : m_state.cp.state.sc;
auto& scState = BindPoint == VK_PIPELINE_BIND_POINT_GRAPHICS ? m_state.gp.constants : m_state.cp.constants;
// Set all constants to 0 that were used by the previous pipeline
// but are not used by the old one. Any stale data could otherwise
// lead to unnecessary pipeline variants being created.
for (auto i : bit::BitMask(scState.mask & ~newMask))
scInfo.specConstants[i] = 0;
scState.mask = newMask;
auto flag = BindPoint == VK_PIPELINE_BIND_POINT_GRAPHICS
? DxvkContextFlag::GpDirtySpecConstants
: DxvkContextFlag::CpDirtySpecConstants;
if (newMask)
m_flags.set(flag);
else
m_flags.clr(flag);
}
template<VkPipelineBindPoint BindPoint>
void DxvkContext::updateSpecConstants() {
auto& scInfo = BindPoint == VK_PIPELINE_BIND_POINT_GRAPHICS ? m_state.gp.state.sc : m_state.cp.state.sc;
auto& scState = BindPoint == VK_PIPELINE_BIND_POINT_GRAPHICS ? m_state.gp.constants : m_state.cp.constants;
for (auto i : bit::BitMask(scState.mask))
scInfo.specConstants[i] = scState.data[i];
if (BindPoint == VK_PIPELINE_BIND_POINT_GRAPHICS) {
m_flags.clr(DxvkContextFlag::GpDirtySpecConstants);
m_flags.set(DxvkContextFlag::GpDirtyPipelineState);
} else {
m_flags.clr(DxvkContextFlag::CpDirtySpecConstants);
m_flags.set(DxvkContextFlag::CpDirtyPipelineState);
}
}
void DxvkContext::invalidateState() {
this->unbindComputePipeline();
this->unbindGraphicsPipeline();
}
template<VkPipelineBindPoint BindPoint>
void DxvkContext::updateResourceBindings(const DxvkBindingLayoutObjects* layout) {
const auto& bindings = layout->layout();
// Ensure that the arrays we write descriptor info to are big enough
if (unlikely(layout->getBindingCount() > m_descriptors.size()))
this->resizeDescriptorArrays(layout->getBindingCount());
// On 32-bit wine, vkUpdateDescriptorSets has significant overhead due
// to struct conversion, so we should use descriptor update templates.
// For 64-bit applications, using templates is slower on some drivers.
constexpr bool useDescriptorTemplates = env::is32BitHostPlatform();
bool independentSets = BindPoint == VK_PIPELINE_BIND_POINT_GRAPHICS
&& m_flags.test(DxvkContextFlag::GpIndependentSets);
uint32_t layoutSetMask = layout->getSetMask();
uint32_t dirtySetMask = BindPoint == VK_PIPELINE_BIND_POINT_GRAPHICS
? m_descriptorState.getDirtyGraphicsSets()
: m_descriptorState.getDirtyComputeSets();
dirtySetMask &= layoutSetMask;
std::array<VkDescriptorSet, DxvkDescriptorSets::SetCount> sets;
m_descriptorPool->alloc(layout, dirtySetMask, sets.data());
uint32_t descriptorCount = 0;
for (auto setIndex : bit::BitMask(dirtySetMask)) {
uint32_t bindingCount = bindings.getBindingCount(setIndex);
VkDescriptorSet set = sets[setIndex];
for (uint32_t j = 0; j < bindingCount; j++) {
const auto& binding = bindings.getBinding(setIndex, j);
if (!useDescriptorTemplates) {
auto& descriptorWrite = m_descriptorWrites[descriptorCount];
descriptorWrite.dstSet = set;
descriptorWrite.dstBinding = j;
descriptorWrite.descriptorType = binding.descriptorType;
}
auto& descriptorInfo = m_descriptors[descriptorCount++];
switch (binding.descriptorType) {
case VK_DESCRIPTOR_TYPE_SAMPLER: {
const auto& res = m_rc[binding.resourceBinding];
if (res.sampler != nullptr) {
descriptorInfo.image.sampler = res.sampler->handle();
descriptorInfo.image.imageView = VK_NULL_HANDLE;
descriptorInfo.image.imageLayout = VK_IMAGE_LAYOUT_UNDEFINED;
if (m_rcTracked.set(binding.resourceBinding))
m_cmd->trackResource<DxvkAccess::None>(res.sampler);
} else {
descriptorInfo.image.sampler = m_common->dummyResources().samplerHandle();
descriptorInfo.image.imageView = VK_NULL_HANDLE;
descriptorInfo.image.imageLayout = VK_IMAGE_LAYOUT_UNDEFINED;
}
} break;
case VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE: {
const auto& res = m_rc[binding.resourceBinding];
if (res.imageView != nullptr && res.imageView->handle(binding.viewType) != VK_NULL_HANDLE) {
descriptorInfo.image.sampler = VK_NULL_HANDLE;
descriptorInfo.image.imageView = res.imageView->handle(binding.viewType);
descriptorInfo.image.imageLayout = res.imageView->imageInfo().layout;
if (m_rcTracked.set(binding.resourceBinding)) {
m_cmd->trackResource<DxvkAccess::None>(res.imageView);
m_cmd->trackResource<DxvkAccess::Read>(res.imageView->image());
}
} else {
descriptorInfo.image.sampler = VK_NULL_HANDLE;
descriptorInfo.image.imageView = VK_NULL_HANDLE;
descriptorInfo.image.imageLayout = VK_IMAGE_LAYOUT_UNDEFINED;
}
} break;
case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE: {
const auto& res = m_rc[binding.resourceBinding];
if (res.imageView != nullptr && res.imageView->handle(binding.viewType) != VK_NULL_HANDLE) {
descriptorInfo.image.sampler = VK_NULL_HANDLE;
descriptorInfo.image.imageView = res.imageView->handle(binding.viewType);
descriptorInfo.image.imageLayout = res.imageView->imageInfo().layout;
if (m_rcTracked.set(binding.resourceBinding)) {
m_cmd->trackResource<DxvkAccess::None>(res.imageView);
m_cmd->trackResource<DxvkAccess::Write>(res.imageView->image());
}
} else {
descriptorInfo.image.sampler = VK_NULL_HANDLE;
descriptorInfo.image.imageView = VK_NULL_HANDLE;
descriptorInfo.image.imageLayout = VK_IMAGE_LAYOUT_UNDEFINED;
}
} break;
case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER: {
const auto& res = m_rc[binding.resourceBinding];
if (res.sampler != nullptr && res.imageView != nullptr
&& res.imageView->handle(binding.viewType) != VK_NULL_HANDLE) {
descriptorInfo.image.sampler = res.sampler->handle();
descriptorInfo.image.imageView = res.imageView->handle(binding.viewType);
descriptorInfo.image.imageLayout = res.imageView->imageInfo().layout;
if (m_rcTracked.set(binding.resourceBinding)) {
m_cmd->trackResource<DxvkAccess::None>(res.sampler);
m_cmd->trackResource<DxvkAccess::None>(res.imageView);
m_cmd->trackResource<DxvkAccess::Read>(res.imageView->image());
}
} else {
descriptorInfo.image.sampler = m_common->dummyResources().samplerHandle();
descriptorInfo.image.imageView = VK_NULL_HANDLE;
descriptorInfo.image.imageLayout = VK_IMAGE_LAYOUT_UNDEFINED;
}
} break;
case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER: {
const auto& res = m_rc[binding.resourceBinding];
if (res.bufferView != nullptr) {
res.bufferView->updateView();
descriptorInfo.texelBuffer = res.bufferView->handle();
if (m_rcTracked.set(binding.resourceBinding)) {
m_cmd->trackResource<DxvkAccess::None>(res.bufferView);
m_cmd->trackResource<DxvkAccess::Read>(res.bufferView->buffer());
}
} else {
descriptorInfo.texelBuffer = VK_NULL_HANDLE;
}
} break;
case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER: {
const auto& res = m_rc[binding.resourceBinding];
if (res.bufferView != nullptr) {
res.bufferView->updateView();
descriptorInfo.texelBuffer = res.bufferView->handle();
if (m_rcTracked.set(binding.resourceBinding)) {
m_cmd->trackResource<DxvkAccess::None>(res.bufferView);
m_cmd->trackResource<DxvkAccess::Write>(res.bufferView->buffer());
}
} else {
descriptorInfo.texelBuffer = VK_NULL_HANDLE;
}
} break;
case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER: {
const auto& res = m_rc[binding.resourceBinding];
if (res.bufferSlice.length()) {
descriptorInfo = res.bufferSlice.getDescriptor();
if (m_rcTracked.set(binding.resourceBinding))
m_cmd->trackResource<DxvkAccess::Read>(res.bufferSlice.buffer());
} else {
descriptorInfo.buffer.buffer = VK_NULL_HANDLE;
descriptorInfo.buffer.offset = 0;
descriptorInfo.buffer.range = VK_WHOLE_SIZE;
}
} break;
case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER: {
const auto& res = m_rc[binding.resourceBinding];
if (res.bufferSlice.length()) {
descriptorInfo = res.bufferSlice.getDescriptor();
if (m_rcTracked.set(binding.resourceBinding))
m_cmd->trackResource<DxvkAccess::Write>(res.bufferSlice.buffer());
} else {
descriptorInfo.buffer.buffer = VK_NULL_HANDLE;
descriptorInfo.buffer.offset = 0;
descriptorInfo.buffer.range = VK_WHOLE_SIZE;
}
} break;
default:
break;
}
}
if (useDescriptorTemplates) {
m_cmd->updateDescriptorSetWithTemplate(set,
layout->getSetUpdateTemplate(setIndex),
&m_descriptors[0]);
descriptorCount = 0;
}
// If the next set is not dirty, update and bind all previously
// updated sets in one go in order to reduce api call overhead.
if (!(((dirtySetMask >> 1) >> setIndex) & 1u)) {
if (!useDescriptorTemplates) {
m_cmd->updateDescriptorSets(descriptorCount,
m_descriptorWrites.data());
descriptorCount = 0;
}
// Find first dirty set in the mask and clear bits
// for all sets that we're going to update here.
uint32_t firstSet = bit::tzcnt(dirtySetMask);
dirtySetMask &= (~1u) << setIndex;
m_cmd->cmdBindDescriptorSets(BindPoint,
layout->getPipelineLayout(independentSets),
firstSet, setIndex - firstSet + 1, &sets[firstSet],
0, nullptr);
}
}
}
void DxvkContext::updateComputeShaderResources() {
this->updateResourceBindings<VK_PIPELINE_BIND_POINT_COMPUTE>(m_state.cp.pipeline->getBindings());
m_descriptorState.clearStages(VK_SHADER_STAGE_COMPUTE_BIT);
}
void DxvkContext::updateGraphicsShaderResources() {
this->updateResourceBindings<VK_PIPELINE_BIND_POINT_GRAPHICS>(m_state.gp.pipeline->getBindings());
m_descriptorState.clearStages(VK_SHADER_STAGE_ALL_GRAPHICS);
}
DxvkFramebufferInfo DxvkContext::makeFramebufferInfo(
const DxvkRenderTargets& renderTargets) {
return DxvkFramebufferInfo(renderTargets, m_device->getDefaultFramebufferSize());
}
void DxvkContext::updateFramebuffer() {
if (m_flags.test(DxvkContextFlag::GpDirtyFramebuffer)) {
m_flags.clr(DxvkContextFlag::GpDirtyFramebuffer);
this->spillRenderPass(true);
DxvkFramebufferInfo fbInfo = makeFramebufferInfo(m_state.om.renderTargets);
this->updateRenderTargetLayouts(fbInfo, m_state.om.framebufferInfo);
// Update relevant graphics pipeline state
m_state.gp.state.ms.setSampleCount(fbInfo.getSampleCount());
m_state.gp.state.rt = fbInfo.getRtInfo();
m_state.om.framebufferInfo = fbInfo;
for (uint32_t i = 0; i < MaxNumRenderTargets; i++) {
const Rc<DxvkImageView>& attachment = fbInfo.getColorTarget(i).view;
VkComponentMapping mapping = attachment != nullptr
? util::invertComponentMapping(attachment->info().swizzle)
: VkComponentMapping();
m_state.gp.state.omSwizzle[i] = DxvkOmAttachmentSwizzle(mapping);
}
m_flags.set(DxvkContextFlag::GpDirtyPipelineState);
}
}
void DxvkContext::applyRenderTargetLoadLayouts() {
for (uint32_t i = 0; i < MaxNumRenderTargets; i++)
m_state.om.renderPassOps.colorOps[i].loadLayout = m_rtLayouts.color[i];
m_state.om.renderPassOps.depthOps.loadLayout = m_rtLayouts.depth;
}
void DxvkContext::applyRenderTargetStoreLayouts() {
for (uint32_t i = 0; i < MaxNumRenderTargets; i++)
m_rtLayouts.color[i] = m_state.om.renderPassOps.colorOps[i].storeLayout;
m_rtLayouts.depth = m_state.om.renderPassOps.depthOps.storeLayout;
}
void DxvkContext::transitionRenderTargetLayouts(
bool sharedOnly) {
for (uint32_t i = 0; i < MaxNumRenderTargets; i++) {
const DxvkAttachment& color = m_state.om.framebufferInfo.getColorTarget(i);
if (color.view != nullptr && (!sharedOnly || color.view->imageInfo().shared)) {
this->transitionColorAttachment(color, m_rtLayouts.color[i]);
m_rtLayouts.color[i] = color.view->imageInfo().layout;
}
}
const DxvkAttachment& depth = m_state.om.framebufferInfo.getDepthTarget();
if (depth.view != nullptr && (!sharedOnly || depth.view->imageInfo().shared)) {
this->transitionDepthAttachment(depth, m_rtLayouts.depth);
m_rtLayouts.depth = depth.view->imageInfo().layout;
}
}
void DxvkContext::transitionColorAttachment(
const DxvkAttachment& attachment,
VkImageLayout oldLayout) {
if (oldLayout != attachment.view->imageInfo().layout) {
m_execBarriers.accessImage(
attachment.view->image(),
attachment.view->imageSubresources(), oldLayout,
VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT,
VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT,
attachment.view->imageInfo().layout,
attachment.view->imageInfo().stages,
attachment.view->imageInfo().access);
m_cmd->trackResource<DxvkAccess::Write>(attachment.view->image());
}
}
void DxvkContext::transitionDepthAttachment(
const DxvkAttachment& attachment,
VkImageLayout oldLayout) {
if (oldLayout != attachment.view->imageInfo().layout) {
m_execBarriers.accessImage(
attachment.view->image(),
attachment.view->imageSubresources(), oldLayout,
VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT,
oldLayout != VK_IMAGE_LAYOUT_DEPTH_STENCIL_READ_ONLY_OPTIMAL
? VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT : 0,
attachment.view->imageInfo().layout,
attachment.view->imageInfo().stages,
attachment.view->imageInfo().access);
m_cmd->trackResource<DxvkAccess::Write>(attachment.view->image());
}
}
void DxvkContext::updateRenderTargetLayouts(
const DxvkFramebufferInfo& newFb,
const DxvkFramebufferInfo& oldFb) {
DxvkRenderTargetLayouts layouts = { };
for (uint32_t i = 0; i < MaxNumRenderTargets; i++) {
if (newFb.getColorTarget(i).view != nullptr)
layouts.color[i] = newFb.getColorTarget(i).view->imageInfo().layout;
}
if (newFb.getDepthTarget().view != nullptr)
layouts.depth = newFb.getDepthTarget().view->imageInfo().layout;
// Check whether any of the previous attachments have been moved
// around or been rebound with a different view. This may help
// reduce the number of image layout transitions between passes.
for (uint32_t i = 0; i < MaxNumRenderTargets; i++) {
const DxvkAttachment& oldAttachment = oldFb.getColorTarget(i);
if (oldAttachment.view != nullptr) {
bool found = false;
for (uint32_t j = 0; j < MaxNumRenderTargets && !found; j++) {
const DxvkAttachment& newAttachment = newFb.getColorTarget(j);
found = newAttachment.view == oldAttachment.view || (newAttachment.view != nullptr
&& newAttachment.view->image() == oldAttachment.view->image()
&& newAttachment.view->subresources() == oldAttachment.view->subresources());
if (found)
layouts.color[j] = m_rtLayouts.color[i];
}
if (!found && m_flags.test(DxvkContextFlag::GpRenderPassSuspended))
this->transitionColorAttachment(oldAttachment, m_rtLayouts.color[i]);
}
}
const DxvkAttachment& oldAttachment = oldFb.getDepthTarget();
if (oldAttachment.view != nullptr) {
const DxvkAttachment& newAttachment = newFb.getDepthTarget();
bool found = newAttachment.view == oldAttachment.view || (newAttachment.view != nullptr
&& newAttachment.view->image() == oldAttachment.view->image()
&& newAttachment.view->subresources() == oldAttachment.view->subresources());
if (found)
layouts.depth = m_rtLayouts.depth;
else if (m_flags.test(DxvkContextFlag::GpRenderPassSuspended))
this->transitionDepthAttachment(oldAttachment, m_rtLayouts.depth);
}
m_rtLayouts = layouts;
}
void DxvkContext::prepareImage(
const Rc<DxvkImage>& image,
const VkImageSubresourceRange& subresources,
bool flushClears) {
// Images that can't be used as attachments are always in their
// default layout, so we don't have to do anything in this case
if (!(image->info().usage & (VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT | VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT)))
return;
// Flush clears if there are any since they may affect the image
if (!m_deferredClears.empty() && flushClears)
this->spillRenderPass(false);
// All images are in their default layout for suspended passes
if (!m_flags.test(DxvkContextFlag::GpRenderPassSuspended))
return;
// 3D images require special care because they only have one
// layer, but views may address individual 2D slices as layers
bool is3D = image->info().type == VK_IMAGE_TYPE_3D;
// Transition any attachment with overlapping subresources
if (image->info().usage & VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT) {
for (uint32_t i = 0; i < MaxNumRenderTargets; i++) {
const DxvkAttachment& attachment = m_state.om.framebufferInfo.getColorTarget(i);
if (attachment.view != nullptr && attachment.view->image() == image
&& (is3D || vk::checkSubresourceRangeOverlap(attachment.view->subresources(), subresources))) {
this->transitionColorAttachment(attachment, m_rtLayouts.color[i]);
m_rtLayouts.color[i] = image->info().layout;
}
}
} else {
const DxvkAttachment& attachment = m_state.om.framebufferInfo.getDepthTarget();
if (attachment.view != nullptr && attachment.view->image() == image
&& (is3D || vk::checkSubresourceRangeOverlap(attachment.view->subresources(), subresources))) {
this->transitionDepthAttachment(attachment, m_rtLayouts.depth);
m_rtLayouts.depth = image->info().layout;
}
}
}
bool DxvkContext::updateIndexBufferBinding() {
if (unlikely(!m_state.vi.indexBuffer.length()))
return false;
m_flags.clr(DxvkContextFlag::GpDirtyIndexBuffer);
auto bufferInfo = m_state.vi.indexBuffer.getDescriptor();
m_cmd->cmdBindIndexBuffer(
bufferInfo.buffer.buffer,
bufferInfo.buffer.offset,
m_state.vi.indexType);
if (m_vbTracked.set(MaxNumVertexBindings))
m_cmd->trackResource<DxvkAccess::Read>(m_state.vi.indexBuffer.buffer());
return true;
}
void DxvkContext::updateVertexBufferBindings() {
m_flags.clr(DxvkContextFlag::GpDirtyVertexBuffers);
if (unlikely(!m_state.gp.state.il.bindingCount()))
return;
std::array<VkBuffer, MaxNumVertexBindings> buffers;
std::array<VkDeviceSize, MaxNumVertexBindings> offsets;
std::array<VkDeviceSize, MaxNumVertexBindings> lengths;
std::array<VkDeviceSize, MaxNumVertexBindings> strides;
bool oldDynamicStrides = m_flags.test(DxvkContextFlag::GpDynamicVertexStrides);
bool newDynamicStrides = true;
// Set buffer handles and offsets for active bindings
for (uint32_t i = 0; i < m_state.gp.state.il.bindingCount(); i++) {
uint32_t binding = m_state.gp.state.ilBindings[i].binding();
if (likely(m_state.vi.vertexBuffers[binding].length())) {
auto vbo = m_state.vi.vertexBuffers[binding].getDescriptor();
buffers[i] = vbo.buffer.buffer;
offsets[i] = vbo.buffer.offset;
lengths[i] = vbo.buffer.range;
strides[i] = m_state.vi.vertexStrides[binding];
if (strides[i]) {
// Dynamic strides are only allowed if the stride is not smaller
// than highest attribute offset + format size for given binding
newDynamicStrides &= strides[i] >= m_state.vi.vertexExtents[i];
}
if (m_vbTracked.set(binding))
m_cmd->trackResource<DxvkAccess::Read>(m_state.vi.vertexBuffers[binding].buffer());
} else {
buffers[i] = VK_NULL_HANDLE;
offsets[i] = 0;
lengths[i] = 0;
strides[i] = 0;
}
}
// If vertex strides are static or if we are switching between static or
// dynamic strides, we'll have to apply them to the pipeline state and
// also sort out our state flags
if (unlikely(!oldDynamicStrides) || unlikely(!newDynamicStrides)) {
m_flags.clr(DxvkContextFlag::GpDynamicVertexStrides);
for (uint32_t i = 0; i < m_state.gp.state.il.bindingCount(); i++) {
uint32_t stride = newDynamicStrides ? 0 : strides[i];
if (m_state.gp.state.ilBindings[i].stride() != stride) {
m_state.gp.state.ilBindings[i].setStride(stride);
m_flags.set(DxvkContextFlag::GpDirtyPipelineState);
}
}
if (newDynamicStrides)
m_flags.set(DxvkContextFlag::GpDynamicVertexStrides);
}
// Vertex bindigs get remapped when compiling the
// pipeline, so this actually does the right thing
m_cmd->cmdBindVertexBuffers(0, m_state.gp.state.il.bindingCount(),
buffers.data(), offsets.data(), lengths.data(),
newDynamicStrides ? strides.data() : nullptr);
}
void DxvkContext::updateTransformFeedbackBuffers() {
const auto& gsInfo = m_state.gp.shaders.gs->info();
VkBuffer xfbBuffers[MaxNumXfbBuffers];
VkDeviceSize xfbOffsets[MaxNumXfbBuffers];
VkDeviceSize xfbLengths[MaxNumXfbBuffers];
for (size_t i = 0; i < MaxNumXfbBuffers; i++) {
auto physSlice = m_state.xfb.buffers[i].getSliceHandle();
xfbBuffers[i] = physSlice.handle;
xfbOffsets[i] = physSlice.offset;
xfbLengths[i] = physSlice.length;
if (physSlice.handle == VK_NULL_HANDLE)
xfbBuffers[i] = m_common->dummyResources().bufferHandle();
if (physSlice.handle != VK_NULL_HANDLE) {
const Rc<DxvkBuffer>& buffer = m_state.xfb.buffers[i].buffer();
buffer->setXfbVertexStride(gsInfo.xfbStrides[i]);
m_cmd->trackResource<DxvkAccess::Write>(buffer);
}
}
m_cmd->cmdBindTransformFeedbackBuffers(
0, MaxNumXfbBuffers,
xfbBuffers, xfbOffsets, xfbLengths);
}
void DxvkContext::updateTransformFeedbackState() {
if (m_flags.test(DxvkContextFlag::GpDirtyXfbBuffers)) {
m_flags.clr(DxvkContextFlag::GpDirtyXfbBuffers);
this->pauseTransformFeedback();
this->updateTransformFeedbackBuffers();
}
this->startTransformFeedback();
}
void DxvkContext::updateDynamicState() {
if (unlikely(m_flags.test(DxvkContextFlag::GpDirtyViewport))) {
m_flags.clr(DxvkContextFlag::GpDirtyViewport);
m_cmd->cmdSetViewport(m_state.vp.viewportCount, m_state.vp.viewports.data());
m_cmd->cmdSetScissor(m_state.vp.viewportCount, m_state.vp.scissorRects.data());
}
if (unlikely(m_flags.all(DxvkContextFlag::GpDirtyDepthStencilState,
DxvkContextFlag::GpDynamicDepthStencilState))) {
m_flags.clr(DxvkContextFlag::GpDirtyDepthStencilState);
// Make sure to not enable writes to aspects that cannot be
// written in the current depth-stencil attachment layout.
// This mirrors what we do for monolithic pipelines.
VkImageAspectFlags dsReadOnlyAspects = m_state.gp.state.rt.getDepthStencilReadOnlyAspects();
bool enableDepthWrites = !(dsReadOnlyAspects & VK_IMAGE_ASPECT_DEPTH_BIT);
bool enableStencilWrites = !(dsReadOnlyAspects & VK_IMAGE_ASPECT_STENCIL_BIT);
m_cmd->cmdSetDepthState(
m_state.gp.state.ds.enableDepthTest(),
m_state.gp.state.ds.enableDepthWrite() && enableDepthWrites,
m_state.gp.state.ds.depthCompareOp());
if (m_device->features().core.features.depthBounds) {
m_cmd->cmdSetDepthBoundsState(
m_state.gp.state.ds.enableDepthBoundsTest());
m_flags.set(DxvkContextFlag::GpDynamicDepthBounds);
}
m_cmd->cmdSetStencilState(
m_state.gp.state.ds.enableStencilTest(),
m_state.gp.state.dsFront.state(enableStencilWrites),
m_state.gp.state.dsBack.state(enableStencilWrites));
m_cmd->cmdSetDepthBiasState(
m_state.gp.state.rs.depthBiasEnable());
if (m_device->features().extExtendedDynamicState3.extendedDynamicState3DepthClipEnable) {
m_cmd->cmdSetDepthClipState(
m_state.gp.state.rs.depthClipEnable());
}
}
if (unlikely(m_flags.all(DxvkContextFlag::GpDirtyMultisampleState,
DxvkContextFlag::GpDynamicMultisampleState))) {
m_flags.clr(DxvkContextFlag::GpDirtyMultisampleState);
// Infer actual sample count from both the multisample state
// and rasterizer state, just like during pipeline creation
VkSampleCountFlagBits sampleCount = VkSampleCountFlagBits(m_state.gp.state.ms.sampleCount());
if (!sampleCount) {
sampleCount = m_state.gp.state.rs.sampleCount()
? VkSampleCountFlagBits(m_state.gp.state.rs.sampleCount())
: VK_SAMPLE_COUNT_1_BIT;
}
VkSampleMask sampleMask = m_state.gp.state.ms.sampleMask() & ((1u << sampleCount) - 1u);
m_cmd->cmdSetMultisampleState(sampleCount, sampleMask);
if (m_device->features().extExtendedDynamicState3.extendedDynamicState3AlphaToCoverageEnable)
m_cmd->cmdSetAlphaToCoverageState(m_state.gp.state.ms.enableAlphaToCoverage());
}
if (unlikely(m_flags.all(DxvkContextFlag::GpDirtyBlendConstants,
DxvkContextFlag::GpDynamicBlendConstants))) {
m_flags.clr(DxvkContextFlag::GpDirtyBlendConstants);
m_cmd->cmdSetBlendConstants(&m_state.dyn.blendConstants.r);
}
if (m_flags.all(DxvkContextFlag::GpDirtyRasterizerState,
DxvkContextFlag::GpDynamicRasterizerState)) {
m_flags.clr(DxvkContextFlag::GpDirtyRasterizerState);
m_cmd->cmdSetRasterizerState(
m_state.dyn.cullMode,
m_state.dyn.frontFace);
}
if (m_flags.all(DxvkContextFlag::GpDirtyStencilRef,
DxvkContextFlag::GpDynamicStencilRef)) {
m_flags.clr(DxvkContextFlag::GpDirtyStencilRef);
m_cmd->cmdSetStencilReference(
VK_STENCIL_FRONT_AND_BACK,
m_state.dyn.stencilReference);
}
if (m_flags.all(DxvkContextFlag::GpDirtyDepthBias,
DxvkContextFlag::GpDynamicDepthBias)) {
m_flags.clr(DxvkContextFlag::GpDirtyDepthBias);
m_cmd->cmdSetDepthBias(
m_state.dyn.depthBias.depthBiasConstant,
m_state.dyn.depthBias.depthBiasClamp,
m_state.dyn.depthBias.depthBiasSlope);
}
if (m_flags.all(DxvkContextFlag::GpDirtyDepthBounds,
DxvkContextFlag::GpDynamicDepthBounds)) {
m_flags.clr(DxvkContextFlag::GpDirtyDepthBounds);
m_cmd->cmdSetDepthBounds(
m_state.dyn.depthBounds.minDepthBounds,
m_state.dyn.depthBounds.maxDepthBounds);
}
}
template<VkPipelineBindPoint BindPoint>
void DxvkContext::updatePushConstants() {
m_flags.clr(DxvkContextFlag::DirtyPushConstants);
auto bindings = BindPoint == VK_PIPELINE_BIND_POINT_GRAPHICS
? m_state.gp.pipeline->getBindings()
: m_state.cp.pipeline->getBindings();
VkPushConstantRange pushConstRange = bindings->layout().getPushConstantRange();
if (!pushConstRange.size)
return;
// Push constants should be compatible between complete and
// independent layouts, so always ask for the complete one
m_cmd->cmdPushConstants(
bindings->getPipelineLayout(false),
pushConstRange.stageFlags,
pushConstRange.offset,
pushConstRange.size,
&m_state.pc.data[pushConstRange.offset]);
}
bool DxvkContext::commitComputeState() {
this->spillRenderPass(false);
if (m_flags.any(
DxvkContextFlag::CpDirtyPipelineState,
DxvkContextFlag::CpDirtySpecConstants)) {
if (unlikely(!this->updateComputePipelineState()))
return false;
}
if (m_descriptorState.hasDirtyComputeSets())
this->updateComputeShaderResources();
if (m_flags.test(DxvkContextFlag::DirtyPushConstants))
this->updatePushConstants<VK_PIPELINE_BIND_POINT_COMPUTE>();
return true;
}
template<bool Indexed, bool Indirect>
bool DxvkContext::commitGraphicsState() {
if (m_flags.test(DxvkContextFlag::GpDirtyPipeline)) {
if (unlikely(!this->updateGraphicsPipeline()))
return false;
}
if (m_flags.test(DxvkContextFlag::GpDirtyFramebuffer))
this->updateFramebuffer();
if (!m_flags.test(DxvkContextFlag::GpRenderPassBound))
this->startRenderPass();
if (m_state.gp.flags.any(
DxvkGraphicsPipelineFlag::HasStorageDescriptors,
DxvkGraphicsPipelineFlag::HasTransformFeedback)) {
this->commitGraphicsBarriers<Indexed, Indirect, false>();
// This can only happen if the render pass was active before,
// so we'll never strat the render pass twice in one draw
if (!m_flags.test(DxvkContextFlag::GpRenderPassBound))
this->startRenderPass();
this->commitGraphicsBarriers<Indexed, Indirect, true>();
}
if (m_flags.test(DxvkContextFlag::GpDirtyIndexBuffer) && Indexed) {
if (unlikely(!this->updateIndexBufferBinding()))
return false;
}
if (m_flags.test(DxvkContextFlag::GpDirtyVertexBuffers))
this->updateVertexBufferBindings();
if (m_flags.test(DxvkContextFlag::GpDirtySpecConstants))
this->updateSpecConstants<VK_PIPELINE_BIND_POINT_GRAPHICS>();
if (m_flags.test(DxvkContextFlag::GpDirtyPipelineState)) {
DxvkGlobalPipelineBarrier barrier = { };
if (Indexed) {
barrier.stages |= VK_PIPELINE_STAGE_VERTEX_INPUT_BIT;
barrier.access |= VK_ACCESS_INDEX_READ_BIT;
}
if (Indirect) {
barrier.stages |= VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT;
barrier.access |= VK_ACCESS_INDIRECT_COMMAND_READ_BIT;
}
if (unlikely(!this->updateGraphicsPipelineState(barrier)))
return false;
}
if (m_descriptorState.hasDirtyGraphicsSets())
this->updateGraphicsShaderResources();
if (m_state.gp.flags.test(DxvkGraphicsPipelineFlag::HasTransformFeedback))
this->updateTransformFeedbackState();
this->updateDynamicState();
if (m_flags.test(DxvkContextFlag::DirtyPushConstants))
this->updatePushConstants<VK_PIPELINE_BIND_POINT_GRAPHICS>();
if (m_flags.test(DxvkContextFlag::DirtyDrawBuffer) && Indirect)
this->trackDrawBuffer();
return true;
}
template<bool DoEmit>
void DxvkContext::commitComputeBarriers() {
const auto& layout = m_state.cp.pipeline->getBindings()->layout();
// Exit early if we're only checking for hazards and
// if the barrier set is empty, to avoid some overhead.
if (!DoEmit && !m_execBarriers.hasResourceBarriers())
return;
for (uint32_t i = 0; i < DxvkDescriptorSets::CsSetCount; i++) {
uint32_t bindingCount = layout.getBindingCount(i);
for (uint32_t j = 0; j < bindingCount; j++) {
const DxvkBindingInfo& binding = layout.getBinding(i, j);
const DxvkShaderResourceSlot& slot = m_rc[binding.resourceBinding];
bool requiresBarrier = false;
switch (binding.descriptorType) {
case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER:
case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER:
if (likely(slot.bufferSlice.length())) {
requiresBarrier = this->checkBufferBarrier<DoEmit>(slot.bufferSlice,
VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, binding.access);
}
break;
case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER:
case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER:
if (likely(slot.bufferView != nullptr)) {
requiresBarrier = this->checkBufferViewBarrier<DoEmit>(slot.bufferView,
VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, binding.access);
}
break;
case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE:
case VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE:
case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
if (likely(slot.imageView != nullptr)) {
requiresBarrier = this->checkImageViewBarrier<DoEmit>(slot.imageView,
VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, binding.access);
}
break;
default:
/* nothing to do */;
}
if (requiresBarrier) {
m_execBarriers.recordCommands(m_cmd);
return;
}
}
}
}
template<bool Indexed, bool Indirect, bool DoEmit>
void DxvkContext::commitGraphicsBarriers() {
if (m_barrierControl.test(DxvkBarrierControl::IgnoreGraphicsBarriers))
return;
constexpr auto storageBufferAccess = VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_TRANSFORM_FEEDBACK_WRITE_BIT_EXT;
constexpr auto storageImageAccess = VK_ACCESS_SHADER_WRITE_BIT;
bool requiresBarrier = false;
// Check the draw buffer for indirect draw calls
if (m_flags.test(DxvkContextFlag::DirtyDrawBuffer) && Indirect) {
std::array<DxvkBufferSlice*, 2> slices = {{
&m_state.id.argBuffer,
&m_state.id.cntBuffer,
}};
for (uint32_t i = 0; i < slices.size() && !requiresBarrier; i++) {
if ((slices[i]->length())
&& (slices[i]->bufferInfo().access & storageBufferAccess)) {
requiresBarrier = this->checkBufferBarrier<DoEmit>(*slices[i],
VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT,
VK_ACCESS_INDIRECT_COMMAND_READ_BIT);
}
}
}
// Read-only stage, so we only have to check this if
// the bindngs have actually changed between draws
if (m_flags.test(DxvkContextFlag::GpDirtyIndexBuffer) && !requiresBarrier && Indexed) {
const auto& indexBufferSlice = m_state.vi.indexBuffer;
if ((indexBufferSlice.length())
&& (indexBufferSlice.bufferInfo().access & storageBufferAccess)) {
requiresBarrier = this->checkBufferBarrier<DoEmit>(indexBufferSlice,
VK_PIPELINE_STAGE_VERTEX_INPUT_BIT,
VK_ACCESS_INDEX_READ_BIT);
}
}
// Same here, also ignore unused vertex bindings
if (m_flags.test(DxvkContextFlag::GpDirtyVertexBuffers)) {
uint32_t bindingCount = m_state.gp.state.il.bindingCount();
for (uint32_t i = 0; i < bindingCount && !requiresBarrier; i++) {
uint32_t binding = m_state.gp.state.ilBindings[i].binding();
const auto& vertexBufferSlice = m_state.vi.vertexBuffers[binding];
if ((vertexBufferSlice.length())
&& (vertexBufferSlice.bufferInfo().access & storageBufferAccess)) {
requiresBarrier = this->checkBufferBarrier<DoEmit>(vertexBufferSlice,
VK_PIPELINE_STAGE_VERTEX_INPUT_BIT,
VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT);
}
}
}
// Transform feedback buffer writes won't overlap, so we
// also only need to check those when they are rebound
if (m_flags.test(DxvkContextFlag::GpDirtyXfbBuffers)
&& m_state.gp.flags.test(DxvkGraphicsPipelineFlag::HasTransformFeedback)) {
for (uint32_t i = 0; i < MaxNumXfbBuffers && !requiresBarrier; i++) {
const auto& xfbBufferSlice = m_state.xfb.buffers[i];
const auto& xfbCounterSlice = m_state.xfb.counters[i];
if (xfbBufferSlice.length()) {
requiresBarrier = this->checkBufferBarrier<DoEmit>(xfbBufferSlice,
VK_PIPELINE_STAGE_TRANSFORM_FEEDBACK_BIT_EXT,
VK_ACCESS_TRANSFORM_FEEDBACK_WRITE_BIT_EXT);
if (xfbCounterSlice.length()) {
requiresBarrier |= this->checkBufferBarrier<DoEmit>(xfbCounterSlice,
VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT |
VK_PIPELINE_STAGE_TRANSFORM_FEEDBACK_BIT_EXT,
VK_ACCESS_TRANSFORM_FEEDBACK_COUNTER_READ_BIT_EXT |
VK_ACCESS_TRANSFORM_FEEDBACK_COUNTER_WRITE_BIT_EXT);
}
}
}
}
// Check shader resources on every draw to handle WAW hazards
auto layout = m_state.gp.pipeline->getBindings()->layout();
for (uint32_t i = 0; i < DxvkDescriptorSets::SetCount && !requiresBarrier; i++) {
uint32_t bindingCount = layout.getBindingCount(i);
for (uint32_t j = 0; j < bindingCount && !requiresBarrier; j++) {
const DxvkBindingInfo& binding = layout.getBinding(i, j);
const DxvkShaderResourceSlot& slot = m_rc[binding.resourceBinding];
switch (binding.descriptorType) {
case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER:
case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER:
if ((slot.bufferSlice.length())
&& (slot.bufferSlice.bufferInfo().access & storageBufferAccess)) {
requiresBarrier = this->checkBufferBarrier<DoEmit>(slot.bufferSlice,
util::pipelineStages(binding.stage), binding.access);
}
break;
case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER:
case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER:
if ((slot.bufferView != nullptr)
&& (slot.bufferView->bufferInfo().access & storageBufferAccess)) {
slot.bufferView->updateView();
requiresBarrier = this->checkBufferViewBarrier<DoEmit>(slot.bufferView,
util::pipelineStages(binding.stage), binding.access);
}
break;
case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE:
case VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE:
case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
if ((slot.imageView != nullptr)
&& (slot.imageView->imageInfo().access & storageImageAccess)) {
requiresBarrier = this->checkImageViewBarrier<DoEmit>(slot.imageView,
util::pipelineStages(binding.stage), binding.access);
}
break;
default:
/* nothing to do */;
}
}
}
// External subpass dependencies serve as full memory
// and execution barriers, so we can use this to allow
// inter-stage synchronization.
if (requiresBarrier)
this->spillRenderPass(true);
}
template<bool DoEmit>
bool DxvkContext::checkBufferBarrier(
const DxvkBufferSlice& bufferSlice,
VkPipelineStageFlags stages,
VkAccessFlags access) {
if constexpr (DoEmit) {
m_execBarriers.accessBuffer(
bufferSlice.getSliceHandle(),
stages, access,
bufferSlice.bufferInfo().stages,
bufferSlice.bufferInfo().access);
return false;
} else {
DxvkAccessFlags dstAccess = DxvkBarrierSet::getAccessTypes(access);
bool dirty = m_execBarriers.isBufferDirty(
bufferSlice.getSliceHandle(), dstAccess);
if (!dirty || dstAccess.test(DxvkAccess::Read) || !this->canIgnoreWawHazards(stages))
return dirty;
DxvkAccessFlags srcAccess = m_execBarriers.getBufferAccess(bufferSlice.getSliceHandle());
return srcAccess.test(DxvkAccess::Read);
}
}
template<bool DoEmit>
bool DxvkContext::checkBufferViewBarrier(
const Rc<DxvkBufferView>& bufferView,
VkPipelineStageFlags stages,
VkAccessFlags access) {
if constexpr (DoEmit) {
m_execBarriers.accessBuffer(
bufferView->getSliceHandle(),
stages, access,
bufferView->bufferInfo().stages,
bufferView->bufferInfo().access);
return false;
} else {
DxvkAccessFlags dstAccess = DxvkBarrierSet::getAccessTypes(access);
bool dirty = m_execBarriers.isBufferDirty(
bufferView->getSliceHandle(), dstAccess);
if (!dirty || dstAccess.test(DxvkAccess::Read) || !this->canIgnoreWawHazards(stages))
return dirty;
DxvkAccessFlags srcAccess = m_execBarriers.getBufferAccess(bufferView->getSliceHandle());
return srcAccess.test(DxvkAccess::Read);
}
}
template<bool DoEmit>
bool DxvkContext::checkImageViewBarrier(
const Rc<DxvkImageView>& imageView,
VkPipelineStageFlags stages,
VkAccessFlags access) {
if constexpr (DoEmit) {
m_execBarriers.accessImage(
imageView->image(),
imageView->imageSubresources(),
imageView->imageInfo().layout,
stages, access,
imageView->imageInfo().layout,
imageView->imageInfo().stages,
imageView->imageInfo().access);
return false;
} else {
DxvkAccessFlags dstAccess = DxvkBarrierSet::getAccessTypes(access);
bool dirty = m_execBarriers.isImageDirty(
imageView->image(),
imageView->imageSubresources(),
dstAccess);
if (!dirty || dstAccess.test(DxvkAccess::Read) || !this->canIgnoreWawHazards(stages))
return dirty;
DxvkAccessFlags srcAccess = m_execBarriers.getImageAccess(
imageView->image(), imageView->imageSubresources());
return srcAccess.test(DxvkAccess::Read);
}
}
bool DxvkContext::canIgnoreWawHazards(VkPipelineStageFlags stages) {
if (!m_barrierControl.test(DxvkBarrierControl::IgnoreWriteAfterWrite))
return false;
if (stages & VK_SHADER_STAGE_COMPUTE_BIT) {
VkPipelineStageFlags stageMask = VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT | VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT;
return !(m_execBarriers.getSrcStages() & ~stageMask);
}
return true;
}
void DxvkContext::emitMemoryBarrier(
VkPipelineStageFlags srcStages,
VkAccessFlags srcAccess,
VkPipelineStageFlags dstStages,
VkAccessFlags dstAccess) {
VkMemoryBarrier2 barrier = { VK_STRUCTURE_TYPE_MEMORY_BARRIER_2 };
barrier.srcStageMask = srcStages;
barrier.srcAccessMask = srcAccess;
barrier.dstStageMask = dstStages;
barrier.dstAccessMask = dstAccess;
VkDependencyInfo depInfo = { VK_STRUCTURE_TYPE_DEPENDENCY_INFO };
depInfo.memoryBarrierCount = 1;
depInfo.pMemoryBarriers = &barrier;
m_cmd->cmdPipelineBarrier(DxvkCmdBuffer::ExecBuffer, &depInfo);
m_cmd->addStatCtr(DxvkStatCounter::CmdBarrierCount, 1);
}
void DxvkContext::trackDrawBuffer() {
if (m_flags.test(DxvkContextFlag::DirtyDrawBuffer)) {
m_flags.clr(DxvkContextFlag::DirtyDrawBuffer);
if (m_state.id.argBuffer.length())
m_cmd->trackResource<DxvkAccess::Read>(m_state.id.argBuffer.buffer());
if (m_state.id.cntBuffer.length())
m_cmd->trackResource<DxvkAccess::Read>(m_state.id.cntBuffer.buffer());
}
}
bool DxvkContext::tryInvalidateDeviceLocalBuffer(
const Rc<DxvkBuffer>& buffer,
VkDeviceSize copySize) {
// We can only discard if the full buffer gets written, and we will only discard
// small buffers in order to not waste significant amounts of memory.
if (copySize != buffer->info().size || copySize > 0x40000)
return false;
// Don't discard host-visible buffers since that may interfere with the frontend
if (buffer->memFlags() & VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT)
return false;
// Don't discard sparse buffers
if (buffer->info().flags & VK_BUFFER_CREATE_SPARSE_BINDING_BIT)
return false;
// Suspend the current render pass if transform feedback is active prior to
// invalidating the buffer, since otherwise we may invalidate a bound buffer.
if ((buffer->info().usage & VK_BUFFER_USAGE_TRANSFORM_FEEDBACK_COUNTER_BUFFER_BIT_EXT)
&& (m_flags.test(DxvkContextFlag::GpXfbActive)))
this->spillRenderPass(true);
this->invalidateBuffer(buffer, buffer->allocSlice());
return true;
}
DxvkGraphicsPipeline* DxvkContext::lookupGraphicsPipeline(
const DxvkGraphicsPipelineShaders& shaders) {
auto idx = shaders.hash() % m_gpLookupCache.size();
if (unlikely(!m_gpLookupCache[idx] || !shaders.eq(m_gpLookupCache[idx]->shaders())))
m_gpLookupCache[idx] = m_common->pipelineManager().createGraphicsPipeline(shaders);
return m_gpLookupCache[idx];
}
DxvkComputePipeline* DxvkContext::lookupComputePipeline(
const DxvkComputePipelineShaders& shaders) {
auto idx = shaders.hash() % m_cpLookupCache.size();
if (unlikely(!m_cpLookupCache[idx] || !shaders.eq(m_cpLookupCache[idx]->shaders())))
m_cpLookupCache[idx] = m_common->pipelineManager().createComputePipeline(shaders);
return m_cpLookupCache[idx];
}
Rc<DxvkBuffer> DxvkContext::createZeroBuffer(
VkDeviceSize size) {
if (m_zeroBuffer != nullptr && m_zeroBuffer->info().size >= size)
return m_zeroBuffer;
DxvkBufferCreateInfo bufInfo;
bufInfo.size = align<VkDeviceSize>(size, 1 << 20);
bufInfo.usage = VK_BUFFER_USAGE_TRANSFER_DST_BIT
| VK_BUFFER_USAGE_TRANSFER_SRC_BIT;
bufInfo.stages = VK_PIPELINE_STAGE_TRANSFER_BIT;
bufInfo.access = VK_ACCESS_TRANSFER_WRITE_BIT
| VK_ACCESS_TRANSFER_READ_BIT;
m_zeroBuffer = m_device->createBuffer(bufInfo,
VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT);
this->initBuffer(m_zeroBuffer);
return m_zeroBuffer;
}
void DxvkContext::resizeDescriptorArrays(
uint32_t bindingCount) {
m_descriptors.resize(bindingCount);
m_descriptorWrites.resize(bindingCount);
for (uint32_t i = 0; i < bindingCount; i++) {
m_descriptorWrites[i] = { VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET };
m_descriptorWrites[i].descriptorCount = 1;
m_descriptorWrites[i].descriptorType = VK_DESCRIPTOR_TYPE_MAX_ENUM;
m_descriptorWrites[i].pImageInfo = &m_descriptors[i].image;
m_descriptorWrites[i].pBufferInfo = &m_descriptors[i].buffer;
m_descriptorWrites[i].pTexelBufferView = &m_descriptors[i].texelBuffer;
}
}
void DxvkContext::beginCurrentCommands() {
// Mark all resources as untracked
m_vbTracked.clear();
m_rcTracked.clear();
// The current state of the internal command buffer is
// undefined, so we have to bind and set up everything
// before any draw or dispatch command is recorded.
m_flags.clr(
DxvkContextFlag::GpRenderPassBound,
DxvkContextFlag::GpXfbActive,
DxvkContextFlag::GpIndependentSets);
m_flags.set(
DxvkContextFlag::GpDirtyFramebuffer,
DxvkContextFlag::GpDirtyPipeline,
DxvkContextFlag::GpDirtyPipelineState,
DxvkContextFlag::GpDirtyVertexBuffers,
DxvkContextFlag::GpDirtyIndexBuffer,
DxvkContextFlag::GpDirtyXfbBuffers,
DxvkContextFlag::GpDirtyBlendConstants,
DxvkContextFlag::GpDirtyStencilRef,
DxvkContextFlag::GpDirtyMultisampleState,
DxvkContextFlag::GpDirtyRasterizerState,
DxvkContextFlag::GpDirtyViewport,
DxvkContextFlag::GpDirtyDepthBias,
DxvkContextFlag::GpDirtyDepthBounds,
DxvkContextFlag::GpDirtyDepthStencilState,
DxvkContextFlag::CpDirtyPipelineState,
DxvkContextFlag::DirtyDrawBuffer);
m_descriptorState.dirtyStages(
VK_SHADER_STAGE_ALL_GRAPHICS |
VK_SHADER_STAGE_COMPUTE_BIT);
m_state.gp.pipeline = nullptr;
m_state.cp.pipeline = nullptr;
}
void DxvkContext::endCurrentCommands() {
this->spillRenderPass(true);
this->flushSharedImages();
m_sdmaBarriers.finalize(m_cmd);
m_initBarriers.finalize(m_cmd);
m_execBarriers.finalize(m_cmd);
}
void DxvkContext::splitCommands() {
// This behaves the same as a pair of endRecording and
// beginRecording calls, except that we keep the same
// command list object for subsequent commands.
this->endCurrentCommands();
m_cmd->next();
this->beginCurrentCommands();
}
}
Reference in New Issue View Git Blame Copy Permalink