/* * Copyright 2015 Google Inc. * * Use of this source code is governed by a BSD-style license that can be * found in the LICENSE file. */ #include "src/gpu/vk/GrVkGpu.h" #include "include/gpu/GrBackendSemaphore.h" #include "include/gpu/GrBackendSurface.h" #include "include/gpu/GrContextOptions.h" #include "include/gpu/GrDirectContext.h" #include "include/private/SkTo.h" #include "src/core/SkCompressedDataUtils.h" #include "src/core/SkConvertPixels.h" #include "src/core/SkMipmap.h" #include "src/gpu/GrBackendUtils.h" #include "src/gpu/GrDataUtils.h" #include "src/gpu/GrDirectContextPriv.h" #include "src/gpu/GrGeometryProcessor.h" #include "src/gpu/GrGpuResourceCacheAccess.h" #include "src/gpu/GrNativeRect.h" #include "src/gpu/GrPipeline.h" #include "src/gpu/GrRenderTarget.h" #include "src/gpu/GrSurfaceDrawContext.h" #include "src/gpu/GrTexture.h" #include "src/gpu/GrThreadSafePipelineBuilder.h" #include "src/gpu/SkGpuDevice.h" #include "src/gpu/SkGr.h" #include "src/gpu/vk/GrVkAMDMemoryAllocator.h" #include "src/gpu/vk/GrVkBuffer.h" #include "src/gpu/vk/GrVkCommandBuffer.h" #include "src/gpu/vk/GrVkCommandPool.h" #include "src/gpu/vk/GrVkFramebuffer.h" #include "src/gpu/vk/GrVkImage.h" #include "src/gpu/vk/GrVkInterface.h" #include "src/gpu/vk/GrVkMemory.h" #include "src/gpu/vk/GrVkOpsRenderPass.h" #include "src/gpu/vk/GrVkPipeline.h" #include "src/gpu/vk/GrVkPipelineState.h" #include "src/gpu/vk/GrVkRenderPass.h" #include "src/gpu/vk/GrVkResourceProvider.h" #include "src/gpu/vk/GrVkSemaphore.h" #include "src/gpu/vk/GrVkTexture.h" #include "src/gpu/vk/GrVkTextureRenderTarget.h" #include "src/image/SkImage_Gpu.h" #include "src/image/SkSurface_Gpu.h" #include "include/gpu/vk/GrVkExtensions.h" #include "include/gpu/vk/GrVkTypes.h" #include #define VK_CALL(X) GR_VK_CALL(this->vkInterface(), X) #define VK_CALL_RET(RET, X) GR_VK_CALL_RESULT(this, RET, X) sk_sp GrVkGpu::Make(const GrVkBackendContext& backendContext, const GrContextOptions& options, GrDirectContext* direct) { if (backendContext.fInstance == VK_NULL_HANDLE || backendContext.fPhysicalDevice == VK_NULL_HANDLE || backendContext.fDevice == VK_NULL_HANDLE || backendContext.fQueue == VK_NULL_HANDLE) { return nullptr; } if (!backendContext.fGetProc) { return nullptr; } PFN_vkEnumerateInstanceVersion localEnumerateInstanceVersion = reinterpret_cast( backendContext.fGetProc("vkEnumerateInstanceVersion", VK_NULL_HANDLE, VK_NULL_HANDLE)); uint32_t instanceVersion = 0; if (!localEnumerateInstanceVersion) { instanceVersion = VK_MAKE_VERSION(1, 0, 0); } else { VkResult err = localEnumerateInstanceVersion(&instanceVersion); if (err) { SkDebugf("Failed to enumerate instance version. Err: %d\n", err); return nullptr; } } PFN_vkGetPhysicalDeviceProperties localGetPhysicalDeviceProperties = reinterpret_cast( backendContext.fGetProc("vkGetPhysicalDeviceProperties", backendContext.fInstance, VK_NULL_HANDLE)); if (!localGetPhysicalDeviceProperties) { return nullptr; } VkPhysicalDeviceProperties physDeviceProperties; localGetPhysicalDeviceProperties(backendContext.fPhysicalDevice, &physDeviceProperties); uint32_t physDevVersion = physDeviceProperties.apiVersion; uint32_t apiVersion = backendContext.fMaxAPIVersion ? backendContext.fMaxAPIVersion : instanceVersion; instanceVersion = std::min(instanceVersion, apiVersion); physDevVersion = std::min(physDevVersion, apiVersion); sk_sp interface; if (backendContext.fVkExtensions) { interface.reset(new GrVkInterface(backendContext.fGetProc, backendContext.fInstance, backendContext.fDevice, instanceVersion, physDevVersion, backendContext.fVkExtensions)); if (!interface->validate(instanceVersion, physDevVersion, backendContext.fVkExtensions)) { return nullptr; } } else { GrVkExtensions extensions; // The only extension flag that may effect the vulkan backend is the swapchain extension. We // need to know if this is enabled to know if we can transition to a present layout when // flushing a surface. if (backendContext.fExtensions & kKHR_swapchain_GrVkExtensionFlag) { const char* swapChainExtName = VK_KHR_SWAPCHAIN_EXTENSION_NAME; extensions.init(backendContext.fGetProc, backendContext.fInstance, backendContext.fPhysicalDevice, 0, nullptr, 1, &swapChainExtName); } interface.reset(new GrVkInterface(backendContext.fGetProc, backendContext.fInstance, backendContext.fDevice, instanceVersion, physDevVersion, &extensions)); if (!interface->validate(instanceVersion, physDevVersion, &extensions)) { return nullptr; } } sk_sp caps; if (backendContext.fDeviceFeatures2) { caps.reset(new GrVkCaps(options, interface.get(), backendContext.fPhysicalDevice, *backendContext.fDeviceFeatures2, instanceVersion, physDevVersion, *backendContext.fVkExtensions, backendContext.fProtectedContext)); } else if (backendContext.fDeviceFeatures) { VkPhysicalDeviceFeatures2 features2; features2.pNext = nullptr; features2.features = *backendContext.fDeviceFeatures; caps.reset(new GrVkCaps(options, interface.get(), backendContext.fPhysicalDevice, features2, instanceVersion, physDevVersion, *backendContext.fVkExtensions, backendContext.fProtectedContext)); } else { VkPhysicalDeviceFeatures2 features; memset(&features, 0, sizeof(VkPhysicalDeviceFeatures2)); features.pNext = nullptr; if (backendContext.fFeatures & kGeometryShader_GrVkFeatureFlag) { features.features.geometryShader = true; } if (backendContext.fFeatures & kDualSrcBlend_GrVkFeatureFlag) { features.features.dualSrcBlend = true; } if (backendContext.fFeatures & kSampleRateShading_GrVkFeatureFlag) { features.features.sampleRateShading = true; } GrVkExtensions extensions; // The only extension flag that may effect the vulkan backend is the swapchain extension. We // need to know if this is enabled to know if we can transition to a present layout when // flushing a surface. if (backendContext.fExtensions & kKHR_swapchain_GrVkExtensionFlag) { const char* swapChainExtName = VK_KHR_SWAPCHAIN_EXTENSION_NAME; extensions.init(backendContext.fGetProc, backendContext.fInstance, backendContext.fPhysicalDevice, 0, nullptr, 1, &swapChainExtName); } caps.reset(new GrVkCaps(options, interface.get(), backendContext.fPhysicalDevice, features, instanceVersion, physDevVersion, extensions, backendContext.fProtectedContext)); } if (!caps) { return nullptr; } sk_sp memoryAllocator = backendContext.fMemoryAllocator; if (!memoryAllocator) { // We were not given a memory allocator at creation memoryAllocator = GrVkAMDMemoryAllocator::Make(backendContext.fInstance, backendContext.fPhysicalDevice, backendContext.fDevice, physDevVersion, backendContext.fVkExtensions, interface, caps.get()); } if (!memoryAllocator) { SkDEBUGFAIL("No supplied vulkan memory allocator and unable to create one internally."); return nullptr; } sk_sp vkGpu(new GrVkGpu(direct, backendContext, std::move(caps), interface, instanceVersion, physDevVersion, std::move(memoryAllocator))); if (backendContext.fProtectedContext == GrProtected::kYes && !vkGpu->vkCaps().supportsProtectedMemory()) { return nullptr; } return std::move(vkGpu); } //////////////////////////////////////////////////////////////////////////////// GrVkGpu::GrVkGpu(GrDirectContext* direct, const GrVkBackendContext& backendContext, sk_sp caps, sk_sp interface, uint32_t instanceVersion, uint32_t physicalDeviceVersion, sk_sp memoryAllocator) : INHERITED(direct) , fInterface(std::move(interface)) , fMemoryAllocator(std::move(memoryAllocator)) , fVkCaps(std::move(caps)) , fPhysicalDevice(backendContext.fPhysicalDevice) , fDevice(backendContext.fDevice) , fQueue(backendContext.fQueue) , fQueueIndex(backendContext.fGraphicsQueueIndex) , fResourceProvider(this) , fStagingBufferManager(this) , fDisconnected(false) , fProtectedContext(backendContext.fProtectedContext) { SkASSERT(!backendContext.fOwnsInstanceAndDevice); SkASSERT(fMemoryAllocator); this->initCapsAndCompiler(fVkCaps); VK_CALL(GetPhysicalDeviceProperties(backendContext.fPhysicalDevice, &fPhysDevProps)); VK_CALL(GetPhysicalDeviceMemoryProperties(backendContext.fPhysicalDevice, &fPhysDevMemProps)); fResourceProvider.init(); fMainCmdPool = fResourceProvider.findOrCreateCommandPool(); if (fMainCmdPool) { fMainCmdBuffer = fMainCmdPool->getPrimaryCommandBuffer(); SkASSERT(this->currentCommandBuffer()); this->currentCommandBuffer()->begin(this); } } void GrVkGpu::destroyResources() { if (fMainCmdPool) { fMainCmdPool->getPrimaryCommandBuffer()->end(this, /*abandoningBuffer=*/true); fMainCmdPool->close(); } // wait for all commands to finish this->finishOutstandingGpuWork(); if (fMainCmdPool) { fMainCmdPool->unref(); fMainCmdPool = nullptr; } for (int i = 0; i < fSemaphoresToWaitOn.count(); ++i) { fSemaphoresToWaitOn[i]->unref(); } fSemaphoresToWaitOn.reset(); for (int i = 0; i < fSemaphoresToSignal.count(); ++i) { fSemaphoresToSignal[i]->unref(); } fSemaphoresToSignal.reset(); fStagingBufferManager.reset(); fMSAALoadManager.destroyResources(this); // must call this just before we destroy the command pool and VkDevice fResourceProvider.destroyResources(); } GrVkGpu::~GrVkGpu() { if (!fDisconnected) { this->destroyResources(); } // We don't delete the memory allocator until the very end of the GrVkGpu lifetime so that // clients can continue to delete backend textures even after a context has been abandoned. fMemoryAllocator.reset(); } void GrVkGpu::disconnect(DisconnectType type) { INHERITED::disconnect(type); if (!fDisconnected) { this->destroyResources(); fSemaphoresToWaitOn.reset(); fSemaphoresToSignal.reset(); fMainCmdBuffer = nullptr; fDisconnected = true; } } GrThreadSafePipelineBuilder* GrVkGpu::pipelineBuilder() { return fResourceProvider.pipelineStateCache(); } sk_sp GrVkGpu::refPipelineBuilder() { return fResourceProvider.refPipelineStateCache(); } /////////////////////////////////////////////////////////////////////////////// GrOpsRenderPass* GrVkGpu::onGetOpsRenderPass( GrRenderTarget* rt, bool useMSAASurface, GrAttachment* stencil, GrSurfaceOrigin origin, const SkIRect& bounds, const GrOpsRenderPass::LoadAndStoreInfo& colorInfo, const GrOpsRenderPass::StencilLoadAndStoreInfo& stencilInfo, const SkTArray& sampledProxies, GrXferBarrierFlags renderPassXferBarriers) { if (!fCachedOpsRenderPass) { fCachedOpsRenderPass = std::make_unique(this); } // For the given render target and requested render pass features we need to find a compatible // framebuffer to use for the render pass. Technically it is the underlying VkRenderPass that // is compatible, but that is part of the framebuffer that we get here. GrVkRenderTarget* vkRT = static_cast(rt); SkASSERT(!useMSAASurface || (rt->numSamples() > 1 || (this->vkCaps().preferDiscardableMSAAAttachment() && vkRT->resolveAttachment() && vkRT->resolveAttachment()->supportsInputAttachmentUsage()))); // Covert the GrXferBarrierFlags into render pass self dependency flags GrVkRenderPass::SelfDependencyFlags selfDepFlags = GrVkRenderPass::SelfDependencyFlags::kNone; if (renderPassXferBarriers & GrXferBarrierFlags::kBlend) { selfDepFlags |= GrVkRenderPass::SelfDependencyFlags::kForNonCoherentAdvBlend; } if (renderPassXferBarriers & GrXferBarrierFlags::kTexture) { selfDepFlags |= GrVkRenderPass::SelfDependencyFlags::kForInputAttachment; } // Figure out if we need a resolve attachment for this render pass. A resolve attachment is // needed if we are using msaa to draw with a discardable msaa attachment. If we are in this // case we also need to update the color load/store ops since we don't want to ever load or // store the msaa color attachment, but may need to for the resolve attachment. GrOpsRenderPass::LoadAndStoreInfo localColorInfo = colorInfo; bool withResolve = false; GrVkRenderPass::LoadFromResolve loadFromResolve = GrVkRenderPass::LoadFromResolve::kNo; GrOpsRenderPass::LoadAndStoreInfo resolveInfo{GrLoadOp::kLoad, GrStoreOp::kStore, {}}; if (useMSAASurface && this->vkCaps().preferDiscardableMSAAAttachment() && vkRT->resolveAttachment() && vkRT->resolveAttachment()->supportsInputAttachmentUsage()) { withResolve = true; localColorInfo.fStoreOp = GrStoreOp::kDiscard; if (colorInfo.fLoadOp == GrLoadOp::kLoad) { loadFromResolve = GrVkRenderPass::LoadFromResolve::kLoad; localColorInfo.fLoadOp = GrLoadOp::kDiscard; } else { resolveInfo.fLoadOp = GrLoadOp::kDiscard; } } // Get the framebuffer to use for the render pass sk_sp framebuffer; if (vkRT->wrapsSecondaryCommandBuffer()) { framebuffer = vkRT->externalFramebuffer(); } else { auto fb = vkRT->getFramebuffer(withResolve, SkToBool(stencil), selfDepFlags, loadFromResolve); framebuffer = sk_ref_sp(fb); } if (!framebuffer) { return nullptr; } if (!fCachedOpsRenderPass->set(rt, std::move(framebuffer), origin, bounds, localColorInfo, stencilInfo, resolveInfo, selfDepFlags, loadFromResolve, sampledProxies)) { return nullptr; } return fCachedOpsRenderPass.get(); } bool GrVkGpu::submitCommandBuffer(SyncQueue sync) { TRACE_EVENT0("skia.gpu", TRACE_FUNC); if (!this->currentCommandBuffer()) { return false; } SkASSERT(!fCachedOpsRenderPass || !fCachedOpsRenderPass->isActive()); if (!this->currentCommandBuffer()->hasWork() && kForce_SyncQueue != sync && !fSemaphoresToSignal.count() && !fSemaphoresToWaitOn.count()) { // We may have added finished procs during the flush call. Since there is no actual work // we are not submitting the command buffer and may never come back around to submit it. // Thus we call all current finished procs manually, since the work has technically // finished. this->currentCommandBuffer()->callFinishedProcs(); SkASSERT(fDrawables.empty()); fResourceProvider.checkCommandBuffers(); return true; } fMainCmdBuffer->end(this); SkASSERT(fMainCmdPool); fMainCmdPool->close(); bool didSubmit = fMainCmdBuffer->submitToQueue(this, fQueue, fSemaphoresToSignal, fSemaphoresToWaitOn); if (didSubmit && sync == kForce_SyncQueue) { fMainCmdBuffer->forceSync(this); } // We must delete any drawables that had to wait until submit to destroy. fDrawables.reset(); // If we didn't submit the command buffer then we did not wait on any semaphores. We will // continue to hold onto these semaphores and wait on them during the next command buffer // submission. if (didSubmit) { for (int i = 0; i < fSemaphoresToWaitOn.count(); ++i) { fSemaphoresToWaitOn[i]->unref(); } fSemaphoresToWaitOn.reset(); } // Even if we did not submit the command buffer, we drop all the signal semaphores since we will // not try to recover the work that wasn't submitted and instead just drop it all. The client // will be notified that the semaphores were not submit so that they will not try to wait on // them. for (int i = 0; i < fSemaphoresToSignal.count(); ++i) { fSemaphoresToSignal[i]->unref(); } fSemaphoresToSignal.reset(); // Release old command pool and create a new one fMainCmdPool->unref(); fMainCmdPool = fResourceProvider.findOrCreateCommandPool(); if (fMainCmdPool) { fMainCmdBuffer = fMainCmdPool->getPrimaryCommandBuffer(); SkASSERT(fMainCmdBuffer); fMainCmdBuffer->begin(this); } else { fMainCmdBuffer = nullptr; } // We must wait to call checkCommandBuffers until after we get a new command buffer. The // checkCommandBuffers may trigger a releaseProc which may cause us to insert a barrier for a // released GrVkImage. That barrier needs to be put into a new command buffer and not the old // one that was just submitted. fResourceProvider.checkCommandBuffers(); return didSubmit; } /////////////////////////////////////////////////////////////////////////////// sk_sp GrVkGpu::onCreateBuffer(size_t size, GrGpuBufferType type, GrAccessPattern accessPattern, const void* data) { #ifdef SK_DEBUG switch (type) { case GrGpuBufferType::kVertex: case GrGpuBufferType::kIndex: case GrGpuBufferType::kDrawIndirect: SkASSERT(accessPattern == kDynamic_GrAccessPattern || accessPattern == kStatic_GrAccessPattern); break; case GrGpuBufferType::kXferCpuToGpu: SkASSERT(accessPattern == kDynamic_GrAccessPattern); break; case GrGpuBufferType::kXferGpuToCpu: SkASSERT(accessPattern == kDynamic_GrAccessPattern || accessPattern == kStream_GrAccessPattern); break; case GrGpuBufferType::kUniform: SkASSERT(accessPattern == kDynamic_GrAccessPattern); break; } #endif sk_sp buff = GrVkBuffer::Make(this, size, type, accessPattern); if (data && buff) { buff->updateData(data, size); } return buff; } bool GrVkGpu::onWritePixels(GrSurface* surface, int left, int top, int width, int height, GrColorType surfaceColorType, GrColorType srcColorType, const GrMipLevel texels[], int mipLevelCount, bool prepForTexSampling) { GrVkTexture* texture = static_cast(surface->asTexture()); if (!texture) { return false; } GrVkAttachment* texAttachment = texture->textureAttachment(); // Make sure we have at least the base level if (!mipLevelCount || !texels[0].fPixels) { return false; } SkASSERT(!GrVkFormatIsCompressed(texAttachment->imageFormat())); bool success = false; bool linearTiling = texAttachment->isLinearTiled(); if (linearTiling) { if (mipLevelCount > 1) { SkDebugf("Can't upload mipmap data to linear tiled texture"); return false; } if (VK_IMAGE_LAYOUT_PREINITIALIZED != texAttachment->currentLayout()) { // Need to change the layout to general in order to perform a host write texAttachment->setImageLayout(this, VK_IMAGE_LAYOUT_GENERAL, VK_ACCESS_HOST_WRITE_BIT, VK_PIPELINE_STAGE_HOST_BIT, false); if (!this->submitCommandBuffer(kForce_SyncQueue)) { return false; } } success = this->uploadTexDataLinear(texAttachment, left, top, width, height, srcColorType, texels[0].fPixels, texels[0].fRowBytes); } else { SkASSERT(mipLevelCount <= (int)texAttachment->mipLevels()); success = this->uploadTexDataOptimal(texAttachment, left, top, width, height, srcColorType, texels, mipLevelCount); if (1 == mipLevelCount) { texture->markMipmapsDirty(); } } if (prepForTexSampling) { texAttachment->setImageLayout(this, VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL, VK_ACCESS_SHADER_READ_BIT, VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, false); } return success; } bool GrVkGpu::onTransferPixelsTo(GrTexture* texture, int left, int top, int width, int height, GrColorType surfaceColorType, GrColorType bufferColorType, sk_sp transferBuffer, size_t bufferOffset, size_t rowBytes) { if (!this->currentCommandBuffer()) { return false; } size_t bpp = GrColorTypeBytesPerPixel(bufferColorType); if (GrBackendFormatBytesPerPixel(texture->backendFormat()) != bpp) { return false; } // Vulkan only supports offsets that are both 4-byte aligned and aligned to a pixel. if ((bufferOffset & 0x3) || (bufferOffset % bpp)) { return false; } GrVkTexture* tex = static_cast(texture); if (!tex) { return false; } GrVkAttachment* vkTex = tex->textureAttachment(); VkFormat format = vkTex->imageFormat(); // Can't transfer compressed data SkASSERT(!GrVkFormatIsCompressed(format)); if (!transferBuffer) { return false; } if (bufferColorType != this->vkCaps().transferColorType(format, surfaceColorType)) { return false; } SkASSERT(GrVkFormatBytesPerBlock(format) == GrColorTypeBytesPerPixel(bufferColorType)); SkDEBUGCODE( SkIRect subRect = SkIRect::MakeXYWH(left, top, width, height); SkIRect bounds = SkIRect::MakeWH(texture->width(), texture->height()); SkASSERT(bounds.contains(subRect)); ) // Set up copy region VkBufferImageCopy region; memset(®ion, 0, sizeof(VkBufferImageCopy)); region.bufferOffset = bufferOffset; region.bufferRowLength = (uint32_t)(rowBytes/bpp); region.bufferImageHeight = 0; region.imageSubresource = { VK_IMAGE_ASPECT_COLOR_BIT, 0, 0, 1 }; region.imageOffset = { left, top, 0 }; region.imageExtent = { (uint32_t)width, (uint32_t)height, 1 }; // Change layout of our target so it can be copied to vkTex->setImageLayout(this, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, VK_ACCESS_TRANSFER_WRITE_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, false); const GrVkBuffer* vkBuffer = static_cast(transferBuffer.get()); // Copy the buffer to the image. this->currentCommandBuffer()->copyBufferToImage(this, vkBuffer->vkBuffer(), vkTex, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 1, ®ion); this->currentCommandBuffer()->addGrBuffer(std::move(transferBuffer)); tex->markMipmapsDirty(); return true; } bool GrVkGpu::onTransferPixelsFrom(GrSurface* surface, int left, int top, int width, int height, GrColorType surfaceColorType, GrColorType bufferColorType, sk_sp transferBuffer, size_t offset) { if (!this->currentCommandBuffer()) { return false; } SkASSERT(surface); SkASSERT(transferBuffer); if (fProtectedContext == GrProtected::kYes) { return false; } GrVkImage* srcImage; if (GrVkRenderTarget* rt = static_cast(surface->asRenderTarget())) { // Reading from render targets that wrap a secondary command buffer is not allowed since // it would require us to know the VkImage, which we don't have, as well as need us to // stop and start the VkRenderPass which we don't have access to. if (rt->wrapsSecondaryCommandBuffer()) { return false; } if (!rt->nonMSAAAttachment()) { return false; } srcImage = rt->nonMSAAAttachment(); } else { SkASSERT(surface->asTexture()); srcImage = static_cast(surface->asTexture())->textureAttachment(); } VkFormat format = srcImage->imageFormat(); if (bufferColorType != this->vkCaps().transferColorType(format, surfaceColorType)) { return false; } SkASSERT(GrVkFormatBytesPerBlock(format) == GrColorTypeBytesPerPixel(bufferColorType)); // Set up copy region VkBufferImageCopy region; memset(®ion, 0, sizeof(VkBufferImageCopy)); region.bufferOffset = offset; region.bufferRowLength = width; region.bufferImageHeight = 0; region.imageSubresource = { VK_IMAGE_ASPECT_COLOR_BIT, 0, 0, 1 }; region.imageOffset = { left, top, 0 }; region.imageExtent = { (uint32_t)width, (uint32_t)height, 1 }; srcImage->setImageLayout(this, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, VK_ACCESS_TRANSFER_READ_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, false); this->currentCommandBuffer()->copyImageToBuffer(this, srcImage, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, transferBuffer, 1, ®ion); GrVkBuffer* vkBuffer = static_cast(transferBuffer.get()); // Make sure the copy to buffer has finished. vkBuffer->addMemoryBarrier(VK_ACCESS_TRANSFER_WRITE_BIT, VK_ACCESS_HOST_READ_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_HOST_BIT, false); return true; } void GrVkGpu::resolveImage(GrSurface* dst, GrVkRenderTarget* src, const SkIRect& srcRect, const SkIPoint& dstPoint) { if (!this->currentCommandBuffer()) { return; } SkASSERT(dst); SkASSERT(src && src->colorAttachment() && src->colorAttachment()->numSamples() > 1); VkImageResolve resolveInfo; resolveInfo.srcSubresource = {VK_IMAGE_ASPECT_COLOR_BIT, 0, 0, 1}; resolveInfo.srcOffset = {srcRect.fLeft, srcRect.fTop, 0}; resolveInfo.dstSubresource = {VK_IMAGE_ASPECT_COLOR_BIT, 0, 0, 1}; resolveInfo.dstOffset = {dstPoint.fX, dstPoint.fY, 0}; resolveInfo.extent = {(uint32_t)srcRect.width(), (uint32_t)srcRect.height(), 1}; GrVkImage* dstImage; GrRenderTarget* dstRT = dst->asRenderTarget(); GrTexture* dstTex = dst->asTexture(); if (dstTex) { dstImage = static_cast(dstTex)->textureAttachment(); } else { SkASSERT(dst->asRenderTarget()); dstImage = static_cast(dstRT)->nonMSAAAttachment(); } SkASSERT(dstImage); dstImage->setImageLayout(this, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, VK_ACCESS_TRANSFER_WRITE_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, false); src->colorAttachment()->setImageLayout(this, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, VK_ACCESS_TRANSFER_READ_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, false); this->currentCommandBuffer()->addGrSurface(sk_ref_sp(src->colorAttachment())); this->currentCommandBuffer()->addGrSurface(sk_ref_sp(dst)); this->currentCommandBuffer()->resolveImage(this, *src->colorAttachment(), *dstImage, 1, &resolveInfo); } void GrVkGpu::onResolveRenderTarget(GrRenderTarget* target, const SkIRect& resolveRect) { SkASSERT(target->numSamples() > 1); GrVkRenderTarget* rt = static_cast(target); SkASSERT(rt->colorAttachmentView() && rt->resolveAttachmentView()); if (this->vkCaps().preferDiscardableMSAAAttachment() && rt->resolveAttachment() && rt->resolveAttachment()->supportsInputAttachmentUsage()) { // We would have resolved the RT during the render pass; return; } this->resolveImage(target, rt, resolveRect, SkIPoint::Make(resolveRect.x(), resolveRect.y())); } bool GrVkGpu::uploadTexDataLinear(GrVkAttachment* texAttachment, int left, int top, int width, int height, GrColorType dataColorType, const void* data, size_t rowBytes) { SkASSERT(data); SkASSERT(texAttachment->isLinearTiled()); SkDEBUGCODE( SkIRect subRect = SkIRect::MakeXYWH(left, top, width, height); SkIRect bounds = SkIRect::MakeWH(texAttachment->width(), texAttachment->height()); SkASSERT(bounds.contains(subRect)); ) size_t bpp = GrColorTypeBytesPerPixel(dataColorType); size_t trimRowBytes = width * bpp; SkASSERT(VK_IMAGE_LAYOUT_PREINITIALIZED == texAttachment->currentLayout() || VK_IMAGE_LAYOUT_GENERAL == texAttachment->currentLayout()); const VkImageSubresource subres = { VK_IMAGE_ASPECT_COLOR_BIT, 0, // mipLevel 0, // arraySlice }; VkSubresourceLayout layout; const GrVkInterface* interface = this->vkInterface(); GR_VK_CALL(interface, GetImageSubresourceLayout(fDevice, texAttachment->image(), &subres, &layout)); const GrVkAlloc& alloc = texAttachment->alloc(); if (VK_NULL_HANDLE == alloc.fMemory) { return false; } VkDeviceSize offset = top * layout.rowPitch + left * bpp; VkDeviceSize size = height*layout.rowPitch; SkASSERT(size + offset <= alloc.fSize); void* mapPtr = GrVkMemory::MapAlloc(this, alloc); if (!mapPtr) { return false; } mapPtr = reinterpret_cast(mapPtr) + offset; SkRectMemcpy(mapPtr, static_cast(layout.rowPitch), data, rowBytes, trimRowBytes, height); GrVkMemory::FlushMappedAlloc(this, alloc, offset, size); GrVkMemory::UnmapAlloc(this, alloc); return true; } // This fills in the 'regions' vector in preparation for copying a buffer to an image. // 'individualMipOffsets' is filled in as a side-effect. static size_t fill_in_compressed_regions(GrStagingBufferManager* stagingBufferManager, SkTArray* regions, SkTArray* individualMipOffsets, GrStagingBufferManager::Slice* slice, SkImage::CompressionType compression, VkFormat vkFormat, SkISize dimensions, GrMipmapped mipmapped) { SkASSERT(compression != SkImage::CompressionType::kNone); int numMipLevels = 1; if (mipmapped == GrMipmapped::kYes) { numMipLevels = SkMipmap::ComputeLevelCount(dimensions.width(), dimensions.height()) + 1; } regions->reserve_back(numMipLevels); individualMipOffsets->reserve_back(numMipLevels); size_t bytesPerBlock = GrVkFormatBytesPerBlock(vkFormat); size_t bufferSize = SkCompressedDataSize(compression, dimensions, individualMipOffsets, mipmapped == GrMipmapped::kYes); SkASSERT(individualMipOffsets->count() == numMipLevels); // Get a staging buffer slice to hold our mip data. // Vulkan requires offsets in the buffer to be aligned to multiple of the texel size and 4 size_t alignment = bytesPerBlock; switch (alignment & 0b11) { case 0: break; // alignment is already a multiple of 4. case 2: alignment *= 2; break; // alignment is a multiple of 2 but not 4. default: alignment *= 4; break; // alignment is not a multiple of 2. } *slice = stagingBufferManager->allocateStagingBufferSlice(bufferSize, alignment); if (!slice->fBuffer) { return 0; } for (int i = 0; i < numMipLevels; ++i) { VkBufferImageCopy& region = regions->push_back(); memset(®ion, 0, sizeof(VkBufferImageCopy)); region.bufferOffset = slice->fOffset + (*individualMipOffsets)[i]; SkISize revisedDimensions = GrCompressedDimensions(compression, dimensions); region.bufferRowLength = revisedDimensions.width(); region.bufferImageHeight = revisedDimensions.height(); region.imageSubresource = {VK_IMAGE_ASPECT_COLOR_BIT, SkToU32(i), 0, 1}; region.imageOffset = {0, 0, 0}; region.imageExtent = {SkToU32(dimensions.width()), SkToU32(dimensions.height()), 1}; dimensions = {std::max(1, dimensions.width() /2), std::max(1, dimensions.height()/2)}; } return bufferSize; } bool GrVkGpu::uploadTexDataOptimal(GrVkAttachment* texAttachment, int left, int top, int width, int height, GrColorType dataColorType, const GrMipLevel texels[], int mipLevelCount) { if (!this->currentCommandBuffer()) { return false; } SkASSERT(!texAttachment->isLinearTiled()); // The assumption is either that we have no mipmaps, or that our rect is the entire texture SkASSERT(1 == mipLevelCount || (0 == left && 0 == top && width == texAttachment->width() && height == texAttachment->height())); // We assume that if the texture has mip levels, we either upload to all the levels or just the // first. SkASSERT(1 == mipLevelCount || mipLevelCount == (int)texAttachment->mipLevels()); if (width == 0 || height == 0) { return false; } SkASSERT(this->vkCaps().surfaceSupportsWritePixels(texAttachment)); SkASSERT(this->vkCaps().isVkFormatTexturable(texAttachment->imageFormat())); size_t bpp = GrColorTypeBytesPerPixel(dataColorType); // texels is const. // But we may need to adjust the fPixels ptr based on the copyRect, or fRowBytes. // Because of this we need to make a non-const shallow copy of texels. SkAutoTArray texelsShallowCopy(mipLevelCount); std::copy_n(texels, mipLevelCount, texelsShallowCopy.get()); SkTArray individualMipOffsets; size_t combinedBufferSize; if (mipLevelCount > 1) { combinedBufferSize = GrComputeTightCombinedBufferSize(bpp, {width, height}, &individualMipOffsets, mipLevelCount); } else { SkASSERT(texelsShallowCopy[0].fPixels && texelsShallowCopy[0].fRowBytes); combinedBufferSize = width*height*bpp; individualMipOffsets.push_back(0); } SkASSERT(combinedBufferSize); // Get a staging buffer slice to hold our mip data. // Vulkan requires offsets in the buffer to be aligned to multiple of the texel size and 4 size_t alignment = bpp; switch (alignment & 0b11) { case 0: break; // alignment is already a multiple of 4. case 2: alignment *= 2; break; // alignment is a multiple of 2 but not 4. default: alignment *= 4; break; // alignment is not a multiple of 2. } GrStagingBufferManager::Slice slice = fStagingBufferManager.allocateStagingBufferSlice(combinedBufferSize, alignment); if (!slice.fBuffer) { return false; } int uploadLeft = left; int uploadTop = top; char* buffer = (char*) slice.fOffsetMapPtr; SkTArray regions(mipLevelCount); int currentWidth = width; int currentHeight = height; int layerHeight = texAttachment->height(); for (int currentMipLevel = 0; currentMipLevel < mipLevelCount; currentMipLevel++) { if (texelsShallowCopy[currentMipLevel].fPixels) { SkASSERT(1 == mipLevelCount || currentHeight == layerHeight); const size_t trimRowBytes = currentWidth * bpp; const size_t rowBytes = texelsShallowCopy[currentMipLevel].fRowBytes; // copy data into the buffer, skipping the trailing bytes char* dst = buffer + individualMipOffsets[currentMipLevel]; const char* src = (const char*)texelsShallowCopy[currentMipLevel].fPixels; SkRectMemcpy(dst, trimRowBytes, src, rowBytes, trimRowBytes, currentHeight); VkBufferImageCopy& region = regions.push_back(); memset(®ion, 0, sizeof(VkBufferImageCopy)); region.bufferOffset = slice.fOffset + individualMipOffsets[currentMipLevel]; region.bufferRowLength = currentWidth; region.bufferImageHeight = currentHeight; region.imageSubresource = {VK_IMAGE_ASPECT_COLOR_BIT, SkToU32(currentMipLevel), 0, 1}; region.imageOffset = {uploadLeft, uploadTop, 0}; region.imageExtent = {(uint32_t)currentWidth, (uint32_t)currentHeight, 1}; } currentWidth = std::max(1, currentWidth/2); currentHeight = std::max(1, currentHeight/2); layerHeight = currentHeight; } // Change layout of our target so it can be copied to texAttachment->setImageLayout(this, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, VK_ACCESS_TRANSFER_WRITE_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, false); // Copy the buffer to the image. This call takes the raw VkBuffer instead of a GrGpuBuffer // because we don't need the command buffer to ref the buffer here. The reason being is that // the buffer is coming from the staging manager and the staging manager will make sure the // command buffer has a ref on the buffer. This avoids having to add and remove a ref for ever // upload in the frame. GrVkBuffer* vkBuffer = static_cast(slice.fBuffer); this->currentCommandBuffer()->copyBufferToImage(this, vkBuffer->vkBuffer(), texAttachment, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, regions.count(), regions.begin()); return true; } // It's probably possible to roll this into uploadTexDataOptimal, // but for now it's easier to maintain as a separate entity. bool GrVkGpu::uploadTexDataCompressed(GrVkAttachment* uploadTexture, SkImage::CompressionType compression, VkFormat vkFormat, SkISize dimensions, GrMipmapped mipMapped, const void* data, size_t dataSize) { if (!this->currentCommandBuffer()) { return false; } SkASSERT(data); SkASSERT(!uploadTexture->isLinearTiled()); // For now the assumption is that our rect is the entire texture. // Compressed textures are read-only so this should be a reasonable assumption. SkASSERT(dimensions.fWidth == uploadTexture->width() && dimensions.fHeight == uploadTexture->height()); if (dimensions.fWidth == 0 || dimensions.fHeight == 0) { return false; } SkASSERT(uploadTexture->imageFormat() == vkFormat); SkASSERT(this->vkCaps().isVkFormatTexturable(vkFormat)); GrStagingBufferManager::Slice slice; SkTArray regions; SkTArray individualMipOffsets; SkDEBUGCODE(size_t combinedBufferSize =) fill_in_compressed_regions(&fStagingBufferManager, ®ions, &individualMipOffsets, &slice, compression, vkFormat, dimensions, mipMapped); if (!slice.fBuffer) { return false; } SkASSERT(dataSize == combinedBufferSize); { char* buffer = (char*)slice.fOffsetMapPtr; memcpy(buffer, data, dataSize); } // Change layout of our target so it can be copied to uploadTexture->setImageLayout(this, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, VK_ACCESS_TRANSFER_WRITE_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, false); // Copy the buffer to the image. This call takes the raw VkBuffer instead of a GrGpuBuffer // because we don't need the command buffer to ref the buffer here. The reason being is that // the buffer is coming from the staging manager and the staging manager will make sure the // command buffer has a ref on the buffer. This avoids having to add and remove a ref for ever // upload in the frame. GrVkBuffer* vkBuffer = static_cast(slice.fBuffer); this->currentCommandBuffer()->copyBufferToImage(this, vkBuffer->vkBuffer(), uploadTexture, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, regions.count(), regions.begin()); return true; } //////////////////////////////////////////////////////////////////////////////// // TODO: make this take a GrMipmapped sk_sp GrVkGpu::onCreateTexture(SkISize dimensions, const GrBackendFormat& format, GrRenderable renderable, int renderTargetSampleCnt, SkBudgeted budgeted, GrProtected isProtected, int mipLevelCount, uint32_t levelClearMask) { VkFormat pixelFormat; SkAssertResult(format.asVkFormat(&pixelFormat)); SkASSERT(!GrVkFormatIsCompressed(pixelFormat)); SkASSERT(mipLevelCount > 0); GrMipmapStatus mipmapStatus = mipLevelCount > 1 ? GrMipmapStatus::kDirty : GrMipmapStatus::kNotAllocated; sk_sp tex; if (renderable == GrRenderable::kYes) { tex = GrVkTextureRenderTarget::MakeNewTextureRenderTarget( this, budgeted, dimensions, pixelFormat, mipLevelCount, renderTargetSampleCnt, mipmapStatus, isProtected); } else { tex = GrVkTexture::MakeNewTexture(this, budgeted, dimensions, pixelFormat, mipLevelCount, isProtected, mipmapStatus); } if (!tex) { return nullptr; } if (levelClearMask) { if (!this->currentCommandBuffer()) { return nullptr; } SkSTArray<1, VkImageSubresourceRange> ranges; bool inRange = false; GrVkImage* texImage = tex->textureAttachment(); for (uint32_t i = 0; i < texImage->mipLevels(); ++i) { if (levelClearMask & (1U << i)) { if (inRange) { ranges.back().levelCount++; } else { auto& range = ranges.push_back(); range.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT; range.baseArrayLayer = 0; range.baseMipLevel = i; range.layerCount = 1; range.levelCount = 1; inRange = true; } } else if (inRange) { inRange = false; } } SkASSERT(!ranges.empty()); static constexpr VkClearColorValue kZeroClearColor = {}; texImage->setImageLayout(this, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, VK_ACCESS_TRANSFER_WRITE_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, false); this->currentCommandBuffer()->clearColorImage(this, texImage, &kZeroClearColor, ranges.count(), ranges.begin()); } return std::move(tex); } sk_sp GrVkGpu::onCreateCompressedTexture(SkISize dimensions, const GrBackendFormat& format, SkBudgeted budgeted, GrMipmapped mipMapped, GrProtected isProtected, const void* data, size_t dataSize) { VkFormat pixelFormat; SkAssertResult(format.asVkFormat(&pixelFormat)); SkASSERT(GrVkFormatIsCompressed(pixelFormat)); int numMipLevels = 1; if (mipMapped == GrMipmapped::kYes) { numMipLevels = SkMipmap::ComputeLevelCount(dimensions.width(), dimensions.height())+1; } GrMipmapStatus mipmapStatus = (mipMapped == GrMipmapped::kYes) ? GrMipmapStatus::kValid : GrMipmapStatus::kNotAllocated; auto tex = GrVkTexture::MakeNewTexture(this, budgeted, dimensions, pixelFormat, numMipLevels, isProtected, mipmapStatus); if (!tex) { return nullptr; } SkImage::CompressionType compression = GrBackendFormatToCompressionType(format); if (!this->uploadTexDataCompressed(tex->textureAttachment(), compression, pixelFormat, dimensions, mipMapped, data, dataSize)) { return nullptr; } return std::move(tex); } //////////////////////////////////////////////////////////////////////////////// void GrVkGpu::copyBuffer(sk_sp srcBuffer, sk_sp dstBuffer, VkDeviceSize srcOffset, VkDeviceSize dstOffset, VkDeviceSize size) { if (!this->currentCommandBuffer()) { return; } VkBufferCopy copyRegion; copyRegion.srcOffset = srcOffset; copyRegion.dstOffset = dstOffset; copyRegion.size = size; this->currentCommandBuffer()->copyBuffer(this, std::move(srcBuffer), std::move(dstBuffer), 1, ©Region); } bool GrVkGpu::updateBuffer(sk_sp buffer, const void* src, VkDeviceSize offset, VkDeviceSize size) { if (!this->currentCommandBuffer()) { return false; } // Update the buffer this->currentCommandBuffer()->updateBuffer(this, std::move(buffer), offset, size, src); return true; } //////////////////////////////////////////////////////////////////////////////// static bool check_image_info(const GrVkCaps& caps, const GrVkImageInfo& info, bool needsAllocation, uint32_t graphicsQueueIndex) { if (VK_NULL_HANDLE == info.fImage) { return false; } if (VK_NULL_HANDLE == info.fAlloc.fMemory && needsAllocation) { return false; } if (info.fImageLayout == VK_IMAGE_LAYOUT_PRESENT_SRC_KHR && !caps.supportsSwapchain()) { return false; } if (info.fCurrentQueueFamily != VK_QUEUE_FAMILY_IGNORED && info.fCurrentQueueFamily != VK_QUEUE_FAMILY_EXTERNAL && info.fCurrentQueueFamily != VK_QUEUE_FAMILY_FOREIGN_EXT) { if (info.fSharingMode == VK_SHARING_MODE_EXCLUSIVE) { if (info.fCurrentQueueFamily != graphicsQueueIndex) { return false; } } else { return false; } } if (info.fYcbcrConversionInfo.isValid()) { if (!caps.supportsYcbcrConversion()) { return false; } if (info.fYcbcrConversionInfo.fExternalFormat != 0) { return true; } } // We currently require everything to be made with transfer bits set if (!SkToBool(info.fImageUsageFlags & VK_IMAGE_USAGE_TRANSFER_SRC_BIT) || !SkToBool(info.fImageUsageFlags & VK_IMAGE_USAGE_TRANSFER_DST_BIT)) { return false; } return true; } static bool check_tex_image_info(const GrVkCaps& caps, const GrVkImageInfo& info) { // We don't support directly importing multisampled textures for sampling from shaders. if (info.fSampleCount != 1) { return false; } if (info.fYcbcrConversionInfo.isValid() && info.fYcbcrConversionInfo.fExternalFormat != 0) { return true; } if (info.fImageTiling == VK_IMAGE_TILING_OPTIMAL) { if (!caps.isVkFormatTexturable(info.fFormat)) { return false; } } else { SkASSERT(info.fImageTiling == VK_IMAGE_TILING_LINEAR); if (!caps.isVkFormatTexturableLinearly(info.fFormat)) { return false; } } // We currently require all textures to be made with sample support if (!SkToBool(info.fImageUsageFlags & VK_IMAGE_USAGE_SAMPLED_BIT)) { return false; } return true; } static bool check_rt_image_info(const GrVkCaps& caps, const GrVkImageInfo& info, bool resolveOnly) { if (!caps.isFormatRenderable(info.fFormat, info.fSampleCount)) { return false; } if (!resolveOnly && !SkToBool(info.fImageUsageFlags & VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT)) { return false; } return true; } sk_sp GrVkGpu::onWrapBackendTexture(const GrBackendTexture& backendTex, GrWrapOwnership ownership, GrWrapCacheable cacheable, GrIOType ioType) { GrVkImageInfo imageInfo; if (!backendTex.getVkImageInfo(&imageInfo)) { return nullptr; } if (!check_image_info(this->vkCaps(), imageInfo, kAdopt_GrWrapOwnership == ownership, this->queueIndex())) { return nullptr; } if (!check_tex_image_info(this->vkCaps(), imageInfo)) { return nullptr; } if (backendTex.isProtected() && (fProtectedContext == GrProtected::kNo)) { return nullptr; } sk_sp mutableState = backendTex.getMutableState(); SkASSERT(mutableState); return GrVkTexture::MakeWrappedTexture(this, backendTex.dimensions(), ownership, cacheable, ioType, imageInfo, std::move(mutableState)); } sk_sp GrVkGpu::onWrapCompressedBackendTexture(const GrBackendTexture& beTex, GrWrapOwnership ownership, GrWrapCacheable cacheable) { return this->onWrapBackendTexture(beTex, ownership, cacheable, kRead_GrIOType); } sk_sp GrVkGpu::onWrapRenderableBackendTexture(const GrBackendTexture& backendTex, int sampleCnt, GrWrapOwnership ownership, GrWrapCacheable cacheable) { GrVkImageInfo imageInfo; if (!backendTex.getVkImageInfo(&imageInfo)) { return nullptr; } if (!check_image_info(this->vkCaps(), imageInfo, kAdopt_GrWrapOwnership == ownership, this->queueIndex())) { return nullptr; } if (!check_tex_image_info(this->vkCaps(), imageInfo)) { return nullptr; } // If sampleCnt is > 1 we will create an intermediate MSAA VkImage and then resolve into // the wrapped VkImage. bool resolveOnly = sampleCnt > 1; if (!check_rt_image_info(this->vkCaps(), imageInfo, resolveOnly)) { return nullptr; } if (backendTex.isProtected() && (fProtectedContext == GrProtected::kNo)) { return nullptr; } sampleCnt = this->vkCaps().getRenderTargetSampleCount(sampleCnt, imageInfo.fFormat); sk_sp mutableState = backendTex.getMutableState(); SkASSERT(mutableState); return GrVkTextureRenderTarget::MakeWrappedTextureRenderTarget(this, backendTex.dimensions(), sampleCnt, ownership, cacheable, imageInfo, std::move(mutableState)); } sk_sp GrVkGpu::onWrapBackendRenderTarget(const GrBackendRenderTarget& backendRT) { GrVkImageInfo info; if (!backendRT.getVkImageInfo(&info)) { return nullptr; } if (!check_image_info(this->vkCaps(), info, false, this->queueIndex())) { return nullptr; } // We will always render directly to this VkImage. static bool kResolveOnly = false; if (!check_rt_image_info(this->vkCaps(), info, kResolveOnly)) { return nullptr; } if (backendRT.isProtected() && (fProtectedContext == GrProtected::kNo)) { return nullptr; } sk_sp mutableState = backendRT.getMutableState(); SkASSERT(mutableState); sk_sp tgt = GrVkRenderTarget::MakeWrappedRenderTarget( this, backendRT.dimensions(), backendRT.sampleCnt(), info, std::move(mutableState)); // We don't allow the client to supply a premade stencil buffer. We always create one if needed. SkASSERT(!backendRT.stencilBits()); if (tgt) { SkASSERT(tgt->canAttemptStencilAttachment(tgt->numSamples() > 1)); } return std::move(tgt); } sk_sp GrVkGpu::onWrapVulkanSecondaryCBAsRenderTarget( const SkImageInfo& imageInfo, const GrVkDrawableInfo& vkInfo) { int maxSize = this->caps()->maxTextureSize(); if (imageInfo.width() > maxSize || imageInfo.height() > maxSize) { return nullptr; } GrBackendFormat backendFormat = GrBackendFormat::MakeVk(vkInfo.fFormat); if (!backendFormat.isValid()) { return nullptr; } int sampleCnt = this->vkCaps().getRenderTargetSampleCount(1, vkInfo.fFormat); if (!sampleCnt) { return nullptr; } return GrVkRenderTarget::MakeSecondaryCBRenderTarget(this, imageInfo.dimensions(), vkInfo); } bool GrVkGpu::loadMSAAFromResolve(GrVkCommandBuffer* commandBuffer, const GrVkRenderPass& renderPass, GrAttachment* dst, GrVkAttachment* src, const SkIRect& srcRect) { return fMSAALoadManager.loadMSAAFromResolve(this, commandBuffer, renderPass, dst, src, srcRect); } bool GrVkGpu::onRegenerateMipMapLevels(GrTexture* tex) { if (!this->currentCommandBuffer()) { return false; } auto* vkTex = static_cast(tex)->textureAttachment(); // don't do anything for linearly tiled textures (can't have mipmaps) if (vkTex->isLinearTiled()) { SkDebugf("Trying to create mipmap for linear tiled texture"); return false; } SkASSERT(tex->textureType() == GrTextureType::k2D); // determine if we can blit to and from this format const GrVkCaps& caps = this->vkCaps(); if (!caps.formatCanBeDstofBlit(vkTex->imageFormat(), false) || !caps.formatCanBeSrcofBlit(vkTex->imageFormat(), false) || !caps.mipmapSupport()) { return false; } int width = tex->width(); int height = tex->height(); VkImageBlit blitRegion; memset(&blitRegion, 0, sizeof(VkImageBlit)); // SkMipmap doesn't include the base level in the level count so we have to add 1 uint32_t levelCount = SkMipmap::ComputeLevelCount(tex->width(), tex->height()) + 1; SkASSERT(levelCount == vkTex->mipLevels()); // change layout of the layers so we can write to them. vkTex->setImageLayout(this, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, VK_ACCESS_TRANSFER_WRITE_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, false); // setup memory barrier SkASSERT(GrVkFormatIsSupported(vkTex->imageFormat())); VkImageMemoryBarrier imageMemoryBarrier = { VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, // sType nullptr, // pNext VK_ACCESS_TRANSFER_WRITE_BIT, // srcAccessMask VK_ACCESS_TRANSFER_READ_BIT, // dstAccessMask VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, // oldLayout VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, // newLayout VK_QUEUE_FAMILY_IGNORED, // srcQueueFamilyIndex VK_QUEUE_FAMILY_IGNORED, // dstQueueFamilyIndex vkTex->image(), // image {VK_IMAGE_ASPECT_COLOR_BIT, 0, 1, 0, 1} // subresourceRange }; // Blit the miplevels uint32_t mipLevel = 1; while (mipLevel < levelCount) { int prevWidth = width; int prevHeight = height; width = std::max(1, width / 2); height = std::max(1, height / 2); imageMemoryBarrier.subresourceRange.baseMipLevel = mipLevel - 1; this->addImageMemoryBarrier(vkTex->resource(), VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, false, &imageMemoryBarrier); blitRegion.srcSubresource = { VK_IMAGE_ASPECT_COLOR_BIT, mipLevel - 1, 0, 1 }; blitRegion.srcOffsets[0] = { 0, 0, 0 }; blitRegion.srcOffsets[1] = { prevWidth, prevHeight, 1 }; blitRegion.dstSubresource = { VK_IMAGE_ASPECT_COLOR_BIT, mipLevel, 0, 1 }; blitRegion.dstOffsets[0] = { 0, 0, 0 }; blitRegion.dstOffsets[1] = { width, height, 1 }; this->currentCommandBuffer()->blitImage(this, vkTex->resource(), vkTex->image(), VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, vkTex->resource(), vkTex->image(), VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 1, &blitRegion, VK_FILTER_LINEAR); ++mipLevel; } if (levelCount > 1) { // This barrier logically is not needed, but it changes the final level to the same layout // as all the others, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL. This makes tracking of the // layouts and future layout changes easier. The alternative here would be to track layout // and memory accesses per layer which doesn't seem work it. imageMemoryBarrier.subresourceRange.baseMipLevel = mipLevel - 1; this->addImageMemoryBarrier(vkTex->resource(), VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, false, &imageMemoryBarrier); vkTex->updateImageLayout(VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL); } return true; } //////////////////////////////////////////////////////////////////////////////// sk_sp GrVkGpu::makeStencilAttachment(const GrBackendFormat& /*colorFormat*/, SkISize dimensions, int numStencilSamples) { VkFormat sFmt = this->vkCaps().preferredStencilFormat(); fStats.incStencilAttachmentCreates(); return GrVkAttachment::MakeStencil(this, dimensions, numStencilSamples, sFmt); } sk_sp GrVkGpu::makeMSAAAttachment(SkISize dimensions, const GrBackendFormat& format, int numSamples, GrProtected isProtected) { VkFormat pixelFormat; SkAssertResult(format.asVkFormat(&pixelFormat)); SkASSERT(!GrVkFormatIsCompressed(pixelFormat)); SkASSERT(this->vkCaps().isFormatRenderable(pixelFormat, numSamples)); fStats.incMSAAAttachmentCreates(); return GrVkAttachment::MakeMSAA(this, dimensions, numSamples, pixelFormat, isProtected); } //////////////////////////////////////////////////////////////////////////////// bool copy_src_data(char* mapPtr, VkFormat vkFormat, const SkTArray& individualMipOffsets, const GrPixmap srcData[], int numMipLevels) { SkASSERT(srcData && numMipLevels); SkASSERT(!GrVkFormatIsCompressed(vkFormat)); SkASSERT(individualMipOffsets.count() == numMipLevels); SkASSERT(mapPtr); size_t bytesPerPixel = GrVkFormatBytesPerBlock(vkFormat); for (int level = 0; level < numMipLevels; ++level) { const size_t trimRB = srcData[level].info().width() * bytesPerPixel; SkRectMemcpy(mapPtr + individualMipOffsets[level], trimRB, srcData[level].addr(), srcData[level].rowBytes(), trimRB, srcData[level].height()); } return true; } bool GrVkGpu::createVkImageForBackendSurface(VkFormat vkFormat, SkISize dimensions, int sampleCnt, GrTexturable texturable, GrRenderable renderable, GrMipmapped mipMapped, GrVkImageInfo* info, GrProtected isProtected) { SkASSERT(texturable == GrTexturable::kYes || renderable == GrRenderable::kYes); if (fProtectedContext != isProtected) { return false; } if (texturable == GrTexturable::kYes && !fVkCaps->isVkFormatTexturable(vkFormat)) { return false; } // MSAA images are only currently used by createTestingOnlyBackendRenderTarget. if (sampleCnt > 1 && (texturable == GrTexturable::kYes || renderable == GrRenderable::kNo)) { return false; } if (renderable == GrRenderable::kYes) { sampleCnt = fVkCaps->getRenderTargetSampleCount(sampleCnt, vkFormat); if (!sampleCnt) { return false; } } int numMipLevels = 1; if (mipMapped == GrMipmapped::kYes) { numMipLevels = SkMipmap::ComputeLevelCount(dimensions.width(), dimensions.height()) + 1; } VkImageUsageFlags usageFlags = VK_IMAGE_USAGE_TRANSFER_SRC_BIT | VK_IMAGE_USAGE_TRANSFER_DST_BIT; if (texturable == GrTexturable::kYes) { usageFlags |= VK_IMAGE_USAGE_SAMPLED_BIT; } if (renderable == GrRenderable::kYes) { usageFlags |= VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT; // We always make our render targets support being used as input attachments usageFlags |= VK_IMAGE_USAGE_INPUT_ATTACHMENT_BIT; } GrVkImage::ImageDesc imageDesc; imageDesc.fImageType = VK_IMAGE_TYPE_2D; imageDesc.fFormat = vkFormat; imageDesc.fWidth = dimensions.width(); imageDesc.fHeight = dimensions.height(); imageDesc.fLevels = numMipLevels; imageDesc.fSamples = sampleCnt; imageDesc.fImageTiling = VK_IMAGE_TILING_OPTIMAL; imageDesc.fUsageFlags = usageFlags; imageDesc.fMemProps = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT; imageDesc.fIsProtected = fProtectedContext; if (!GrVkImage::InitImageInfo(this, imageDesc, info)) { SkDebugf("Failed to init image info\n"); return false; } return true; } bool GrVkGpu::onClearBackendTexture(const GrBackendTexture& backendTexture, sk_sp finishedCallback, std::array color) { GrVkImageInfo info; SkAssertResult(backendTexture.getVkImageInfo(&info)); sk_sp mutableState = backendTexture.getMutableState(); SkASSERT(mutableState); sk_sp texture = GrVkTexture::MakeWrappedTexture(this, backendTexture.dimensions(), kBorrow_GrWrapOwnership, GrWrapCacheable::kNo, kRW_GrIOType, info, std::move(mutableState)); if (!texture) { return false; } GrVkAttachment* texAttachment = texture->textureAttachment(); GrVkPrimaryCommandBuffer* cmdBuffer = this->currentCommandBuffer(); if (!cmdBuffer) { return false; } texAttachment->setImageLayout(this, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, VK_ACCESS_TRANSFER_WRITE_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, false); // CmdClearColorImage doesn't work for compressed formats SkASSERT(!GrVkFormatIsCompressed(info.fFormat)); VkClearColorValue vkColor; // If we ever support SINT or UINT formats this needs to be updated to use the int32 and // uint32 union members in those cases. vkColor.float32[0] = color[0]; vkColor.float32[1] = color[1]; vkColor.float32[2] = color[2]; vkColor.float32[3] = color[3]; VkImageSubresourceRange range; range.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT; range.baseArrayLayer = 0; range.baseMipLevel = 0; range.layerCount = 1; range.levelCount = info.fLevelCount; cmdBuffer->clearColorImage(this, texAttachment, &vkColor, 1, &range); // Change image layout to shader read since if we use this texture as a borrowed // texture within Ganesh we require that its layout be set to that texAttachment->setImageLayout(this, VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL, VK_ACCESS_SHADER_READ_BIT, VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, false); if (finishedCallback) { this->addFinishedCallback(std::move(finishedCallback)); } return true; } GrBackendTexture GrVkGpu::onCreateBackendTexture(SkISize dimensions, const GrBackendFormat& format, GrRenderable renderable, GrMipmapped mipMapped, GrProtected isProtected) { const GrVkCaps& caps = this->vkCaps(); if (fProtectedContext != isProtected) { return {}; } VkFormat vkFormat; if (!format.asVkFormat(&vkFormat)) { return {}; } // TODO: move the texturability check up to GrGpu::createBackendTexture and just assert here if (!caps.isVkFormatTexturable(vkFormat)) { return {}; } if (GrVkFormatNeedsYcbcrSampler(vkFormat)) { return {}; } GrVkImageInfo info; if (!this->createVkImageForBackendSurface(vkFormat, dimensions, 1, GrTexturable::kYes, renderable, mipMapped, &info, isProtected)) { return {}; } return GrBackendTexture(dimensions.width(), dimensions.height(), info); } GrBackendTexture GrVkGpu::onCreateCompressedBackendTexture( SkISize dimensions, const GrBackendFormat& format, GrMipmapped mipMapped, GrProtected isProtected) { return this->onCreateBackendTexture(dimensions, format, GrRenderable::kNo, mipMapped, isProtected); } bool GrVkGpu::onUpdateCompressedBackendTexture(const GrBackendTexture& backendTexture, sk_sp finishedCallback, const void* data, size_t size) { GrVkImageInfo info; SkAssertResult(backendTexture.getVkImageInfo(&info)); sk_sp mutableState = backendTexture.getMutableState(); SkASSERT(mutableState); sk_sp texture = GrVkTexture::MakeWrappedTexture(this, backendTexture.dimensions(), kBorrow_GrWrapOwnership, GrWrapCacheable::kNo, kRW_GrIOType, info, std::move(mutableState)); if (!texture) { return false; } GrVkPrimaryCommandBuffer* cmdBuffer = this->currentCommandBuffer(); if (!cmdBuffer) { return false; } GrVkAttachment* attachment = texture->textureAttachment(); attachment->setImageLayout(this, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, VK_ACCESS_TRANSFER_WRITE_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, false); SkImage::CompressionType compression = GrBackendFormatToCompressionType(backendTexture.getBackendFormat()); SkTArray regions; SkTArray individualMipOffsets; GrStagingBufferManager::Slice slice; fill_in_compressed_regions(&fStagingBufferManager, ®ions, &individualMipOffsets, &slice, compression, info.fFormat, backendTexture.dimensions(), backendTexture.fMipmapped); if (!slice.fBuffer) { return false; } memcpy(slice.fOffsetMapPtr, data, size); cmdBuffer->addGrSurface(texture); // Copy the buffer to the image. This call takes the raw VkBuffer instead of a GrGpuBuffer // because we don't need the command buffer to ref the buffer here. The reason being is that // the buffer is coming from the staging manager and the staging manager will make sure the // command buffer has a ref on the buffer. This avoids having to add and remove a ref for // every upload in the frame. cmdBuffer->copyBufferToImage(this, static_cast(slice.fBuffer)->vkBuffer(), attachment, attachment->currentLayout(), regions.count(), regions.begin()); // Change image layout to shader read since if we use this texture as a borrowed // texture within Ganesh we require that its layout be set to that attachment->setImageLayout(this, VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL, VK_ACCESS_SHADER_READ_BIT, VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, false); if (finishedCallback) { this->addFinishedCallback(std::move(finishedCallback)); } return true; } void set_layout_and_queue_from_mutable_state(GrVkGpu* gpu, GrVkImage* image, const GrVkSharedImageInfo& newInfo) { // Even though internally we use this helper for getting src access flags and stages they // can also be used for general dst flags since we don't know exactly what the client // plans on using the image for. VkImageLayout newLayout = newInfo.getImageLayout(); if (newLayout == VK_IMAGE_LAYOUT_UNDEFINED) { newLayout = image->currentLayout(); } VkPipelineStageFlags dstStage = GrVkImage::LayoutToPipelineSrcStageFlags(newLayout); VkAccessFlags dstAccess = GrVkImage::LayoutToSrcAccessMask(newLayout); uint32_t currentQueueFamilyIndex = image->currentQueueFamilyIndex(); uint32_t newQueueFamilyIndex = newInfo.getQueueFamilyIndex(); auto isSpecialQueue = [](uint32_t queueFamilyIndex) { return queueFamilyIndex == VK_QUEUE_FAMILY_EXTERNAL || queueFamilyIndex == VK_QUEUE_FAMILY_FOREIGN_EXT; }; if (isSpecialQueue(currentQueueFamilyIndex) && isSpecialQueue(newQueueFamilyIndex)) { // It is illegal to have both the new and old queue be special queue families (i.e. external // or foreign). return; } image->setImageLayoutAndQueueIndex(gpu, newLayout, dstAccess, dstStage, false, newQueueFamilyIndex); } bool GrVkGpu::setBackendSurfaceState(GrVkImageInfo info, sk_sp currentState, SkISize dimensions, const GrVkSharedImageInfo& newInfo, GrBackendSurfaceMutableState* previousState, sk_sp finishedCallback) { sk_sp texture = GrVkAttachment::MakeWrapped( this, dimensions, info, std::move(currentState), GrVkAttachment::UsageFlags::kColorAttachment, kBorrow_GrWrapOwnership, GrWrapCacheable::kNo, /*forSecondaryCB=*/false); SkASSERT(texture); if (!texture) { return false; } if (previousState) { previousState->setVulkanState(texture->currentLayout(), texture->currentQueueFamilyIndex()); } set_layout_and_queue_from_mutable_state(this, texture.get(), newInfo); if (finishedCallback) { this->addFinishedCallback(std::move(finishedCallback)); } return true; } bool GrVkGpu::setBackendTextureState(const GrBackendTexture& backendTeture, const GrBackendSurfaceMutableState& newState, GrBackendSurfaceMutableState* previousState, sk_sp finishedCallback) { GrVkImageInfo info; SkAssertResult(backendTeture.getVkImageInfo(&info)); sk_sp currentState = backendTeture.getMutableState(); SkASSERT(currentState); SkASSERT(newState.isValid() && newState.fBackend == GrBackend::kVulkan); return this->setBackendSurfaceState(info, std::move(currentState), backendTeture.dimensions(), newState.fVkState, previousState, std::move(finishedCallback)); } bool GrVkGpu::setBackendRenderTargetState(const GrBackendRenderTarget& backendRenderTarget, const GrBackendSurfaceMutableState& newState, GrBackendSurfaceMutableState* previousState, sk_sp finishedCallback) { GrVkImageInfo info; SkAssertResult(backendRenderTarget.getVkImageInfo(&info)); sk_sp currentState = backendRenderTarget.getMutableState(); SkASSERT(currentState); SkASSERT(newState.fBackend == GrBackend::kVulkan); return this->setBackendSurfaceState(info, std::move(currentState), backendRenderTarget.dimensions(), newState.fVkState, previousState, std::move(finishedCallback)); } void GrVkGpu::xferBarrier(GrRenderTarget* rt, GrXferBarrierType barrierType) { GrVkRenderTarget* vkRT = static_cast(rt); VkPipelineStageFlags dstStage; VkAccessFlags dstAccess; if (barrierType == kBlend_GrXferBarrierType) { dstStage = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT; dstAccess = VK_ACCESS_COLOR_ATTACHMENT_READ_NONCOHERENT_BIT_EXT; } else { SkASSERT(barrierType == kTexture_GrXferBarrierType); dstStage = VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT; dstAccess = VK_ACCESS_INPUT_ATTACHMENT_READ_BIT; } GrVkAttachment* colorAttachment = vkRT->colorAttachment(); VkImageMemoryBarrier barrier; barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER; barrier.pNext = nullptr; barrier.srcAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT; barrier.dstAccessMask = dstAccess; barrier.oldLayout = colorAttachment->currentLayout(); barrier.newLayout = barrier.oldLayout; barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; barrier.image = colorAttachment->image(); barrier.subresourceRange = {VK_IMAGE_ASPECT_COLOR_BIT, 0, colorAttachment->mipLevels(), 0, 1}; this->addImageMemoryBarrier(colorAttachment->resource(), VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT, dstStage, true, &barrier); } void GrVkGpu::deleteBackendTexture(const GrBackendTexture& tex) { SkASSERT(GrBackendApi::kVulkan == tex.fBackend); GrVkImageInfo info; if (tex.getVkImageInfo(&info)) { GrVkImage::DestroyImageInfo(this, const_cast(&info)); } } bool GrVkGpu::compile(const GrProgramDesc& desc, const GrProgramInfo& programInfo) { GrVkRenderPass::AttachmentsDescriptor attachmentsDescriptor; GrVkRenderPass::AttachmentFlags attachmentFlags; GrVkRenderTarget::ReconstructAttachmentsDescriptor(this->vkCaps(), programInfo, &attachmentsDescriptor, &attachmentFlags); GrVkRenderPass::SelfDependencyFlags selfDepFlags = GrVkRenderPass::SelfDependencyFlags::kNone; if (programInfo.renderPassBarriers() & GrXferBarrierFlags::kBlend) { selfDepFlags |= GrVkRenderPass::SelfDependencyFlags::kForNonCoherentAdvBlend; } if (programInfo.renderPassBarriers() & GrXferBarrierFlags::kTexture) { selfDepFlags |= GrVkRenderPass::SelfDependencyFlags::kForInputAttachment; } GrVkRenderPass::LoadFromResolve loadFromResolve = GrVkRenderPass::LoadFromResolve::kNo; if (programInfo.targetSupportsVkResolveLoad() && programInfo.colorLoadOp() == GrLoadOp::kLoad && this->vkCaps().preferDiscardableMSAAAttachment()) { loadFromResolve = GrVkRenderPass::LoadFromResolve::kLoad; } sk_sp renderPass(this->resourceProvider().findCompatibleRenderPass( &attachmentsDescriptor, attachmentFlags, selfDepFlags, loadFromResolve)); if (!renderPass) { return false; } GrThreadSafePipelineBuilder::Stats::ProgramCacheResult stat; auto pipelineState = this->resourceProvider().findOrCreateCompatiblePipelineState( desc, programInfo, renderPass->vkRenderPass(), &stat); if (!pipelineState) { return false; } return stat != GrThreadSafePipelineBuilder::Stats::ProgramCacheResult::kHit; } #if GR_TEST_UTILS bool GrVkGpu::isTestingOnlyBackendTexture(const GrBackendTexture& tex) const { SkASSERT(GrBackendApi::kVulkan == tex.fBackend); GrVkImageInfo backend; if (!tex.getVkImageInfo(&backend)) { return false; } if (backend.fImage && backend.fAlloc.fMemory) { VkMemoryRequirements req; memset(&req, 0, sizeof(req)); GR_VK_CALL(this->vkInterface(), GetImageMemoryRequirements(fDevice, backend.fImage, &req)); // TODO: find a better check // This will probably fail with a different driver return (req.size > 0) && (req.size <= 8192 * 8192); } return false; } GrBackendRenderTarget GrVkGpu::createTestingOnlyBackendRenderTarget(SkISize dimensions, GrColorType ct, int sampleCnt, GrProtected isProtected) { if (dimensions.width() > this->caps()->maxRenderTargetSize() || dimensions.height() > this->caps()->maxRenderTargetSize()) { return {}; } VkFormat vkFormat = this->vkCaps().getFormatFromColorType(ct); GrVkImageInfo info; if (!this->createVkImageForBackendSurface(vkFormat, dimensions, sampleCnt, GrTexturable::kNo, GrRenderable::kYes, GrMipmapped::kNo, &info, isProtected)) { return {}; } return GrBackendRenderTarget(dimensions.width(), dimensions.height(), 0, info); } void GrVkGpu::deleteTestingOnlyBackendRenderTarget(const GrBackendRenderTarget& rt) { SkASSERT(GrBackendApi::kVulkan == rt.fBackend); GrVkImageInfo info; if (rt.getVkImageInfo(&info)) { // something in the command buffer may still be using this, so force submit SkAssertResult(this->submitCommandBuffer(kForce_SyncQueue)); GrVkImage::DestroyImageInfo(this, const_cast(&info)); } } #endif //////////////////////////////////////////////////////////////////////////////// void GrVkGpu::addBufferMemoryBarrier(const GrManagedResource* resource, VkPipelineStageFlags srcStageMask, VkPipelineStageFlags dstStageMask, bool byRegion, VkBufferMemoryBarrier* barrier) const { if (!this->currentCommandBuffer()) { return; } SkASSERT(resource); this->currentCommandBuffer()->pipelineBarrier(this, resource, srcStageMask, dstStageMask, byRegion, GrVkCommandBuffer::kBufferMemory_BarrierType, barrier); } void GrVkGpu::addBufferMemoryBarrier(VkPipelineStageFlags srcStageMask, VkPipelineStageFlags dstStageMask, bool byRegion, VkBufferMemoryBarrier* barrier) const { if (!this->currentCommandBuffer()) { return; } // We don't pass in a resource here to the command buffer. The command buffer only is using it // to hold a ref, but every place where we add a buffer memory barrier we are doing some other // command with the buffer on the command buffer. Thus those other commands will already cause // the command buffer to be holding a ref to the buffer. this->currentCommandBuffer()->pipelineBarrier(this, /*resource=*/nullptr, srcStageMask, dstStageMask, byRegion, GrVkCommandBuffer::kBufferMemory_BarrierType, barrier); } void GrVkGpu::addImageMemoryBarrier(const GrManagedResource* resource, VkPipelineStageFlags srcStageMask, VkPipelineStageFlags dstStageMask, bool byRegion, VkImageMemoryBarrier* barrier) const { // If we are in the middle of destroying or abandoning the context we may hit a release proc // that triggers the destruction of a GrVkImage. This could cause us to try and transfer the // VkImage back to the original queue. In this state we don't submit anymore work and we may not // have a current command buffer. Thus we won't do the queue transfer. if (!this->currentCommandBuffer()) { return; } SkASSERT(resource); this->currentCommandBuffer()->pipelineBarrier(this, resource, srcStageMask, dstStageMask, byRegion, GrVkCommandBuffer::kImageMemory_BarrierType, barrier); } void GrVkGpu::prepareSurfacesForBackendAccessAndStateUpdates( SkSpan proxies, SkSurface::BackendSurfaceAccess access, const GrBackendSurfaceMutableState* newState) { // Submit the current command buffer to the Queue. Whether we inserted semaphores or not does // not effect what we do here. if (!proxies.empty() && (access == SkSurface::BackendSurfaceAccess::kPresent || newState)) { // We currently don't support passing in new surface state for multiple proxies here. The // only time we have multiple proxies is if we are flushing a yuv SkImage which won't have // state updates anyways. Additionally if we have a newState than we must not have any // BackendSurfaceAccess. SkASSERT(!newState || proxies.size() == 1); SkASSERT(!newState || access == SkSurface::BackendSurfaceAccess::kNoAccess); GrVkImage* image; for (GrSurfaceProxy* proxy : proxies) { SkASSERT(proxy->isInstantiated()); if (GrTexture* tex = proxy->peekTexture()) { image = static_cast(tex)->textureAttachment(); } else { GrRenderTarget* rt = proxy->peekRenderTarget(); SkASSERT(rt); GrVkRenderTarget* vkRT = static_cast(rt); image = vkRT->externalAttachment(); } if (newState) { const GrVkSharedImageInfo& newInfo = newState->fVkState; set_layout_and_queue_from_mutable_state(this, image, newInfo); } else { SkASSERT(access == SkSurface::BackendSurfaceAccess::kPresent); image->prepareForPresent(this); } } } } void GrVkGpu::addFinishedProc(GrGpuFinishedProc finishedProc, GrGpuFinishedContext finishedContext) { SkASSERT(finishedProc); this->addFinishedCallback(GrRefCntedCallback::Make(finishedProc, finishedContext)); } void GrVkGpu::addFinishedCallback(sk_sp finishedCallback) { SkASSERT(finishedCallback); fResourceProvider.addFinishedProcToActiveCommandBuffers(std::move(finishedCallback)); } void GrVkGpu::takeOwnershipOfBuffer(sk_sp buffer) { this->currentCommandBuffer()->addGrBuffer(std::move(buffer)); } bool GrVkGpu::onSubmitToGpu(bool syncCpu) { if (syncCpu) { return this->submitCommandBuffer(kForce_SyncQueue); } else { return this->submitCommandBuffer(kSkip_SyncQueue); } } void GrVkGpu::finishOutstandingGpuWork() { VK_CALL(QueueWaitIdle(fQueue)); if (this->vkCaps().mustSyncCommandBuffersWithQueue()) { fResourceProvider.forceSyncAllCommandBuffers(); } } void GrVkGpu::onReportSubmitHistograms() { #if SK_HISTOGRAMS_ENABLED uint64_t allocatedMemory = fMemoryAllocator->totalAllocatedMemory(); uint64_t usedMemory = fMemoryAllocator->totalUsedMemory(); SkASSERT(usedMemory <= allocatedMemory); if (allocatedMemory > 0) { SK_HISTOGRAM_PERCENTAGE("VulkanMemoryAllocator.PercentUsed", (usedMemory * 100) / allocatedMemory); } // allocatedMemory is in bytes and need to be reported it in kilobytes. SK_HISTOGRAM_MEMORY_KB // supports samples up to around 500MB which should support the amounts of memory we allocate. SK_HISTOGRAM_MEMORY_KB("VulkanMemoryAllocator.AmountAllocated", allocatedMemory >> 10); #endif } void GrVkGpu::copySurfaceAsCopyImage(GrSurface* dst, GrSurface* src, GrVkImage* dstImage, GrVkImage* srcImage, const SkIRect& srcRect, const SkIPoint& dstPoint) { if (!this->currentCommandBuffer()) { return; } #ifdef SK_DEBUG int dstSampleCnt = dstImage->vkImageInfo().fSampleCount; int srcSampleCnt = srcImage->vkImageInfo().fSampleCount; bool dstHasYcbcr = dstImage->ycbcrConversionInfo().isValid(); bool srcHasYcbcr = srcImage->ycbcrConversionInfo().isValid(); VkFormat dstFormat = dstImage->imageFormat(); VkFormat srcFormat; SkAssertResult(dst->backendFormat().asVkFormat(&srcFormat)); SkASSERT(this->vkCaps().canCopyImage(dstFormat, dstSampleCnt, dstHasYcbcr, srcFormat, srcSampleCnt, srcHasYcbcr)); #endif if (src->isProtected() && !dst->isProtected()) { SkDebugf("Can't copy from protected memory to non-protected"); return; } // These flags are for flushing/invalidating caches and for the dst image it doesn't matter if // the cache is flushed since it is only being written to. dstImage->setImageLayout(this, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, VK_ACCESS_TRANSFER_WRITE_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, false); srcImage->setImageLayout(this, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, VK_ACCESS_TRANSFER_READ_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, false); VkImageCopy copyRegion; memset(©Region, 0, sizeof(VkImageCopy)); copyRegion.srcSubresource = { VK_IMAGE_ASPECT_COLOR_BIT, 0, 0, 1 }; copyRegion.srcOffset = { srcRect.fLeft, srcRect.fTop, 0 }; copyRegion.dstSubresource = { VK_IMAGE_ASPECT_COLOR_BIT, 0, 0, 1 }; copyRegion.dstOffset = { dstPoint.fX, dstPoint.fY, 0 }; copyRegion.extent = { (uint32_t)srcRect.width(), (uint32_t)srcRect.height(), 1 }; this->currentCommandBuffer()->addGrSurface(sk_ref_sp(src)); this->currentCommandBuffer()->addGrSurface(sk_ref_sp(dst)); this->currentCommandBuffer()->copyImage(this, srcImage, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, dstImage, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 1, ©Region); SkIRect dstRect = SkIRect::MakeXYWH(dstPoint.fX, dstPoint.fY, srcRect.width(), srcRect.height()); // The rect is already in device space so we pass in kTopLeft so no flip is done. this->didWriteToSurface(dst, kTopLeft_GrSurfaceOrigin, &dstRect); } void GrVkGpu::copySurfaceAsBlit(GrSurface* dst, GrSurface* src, GrVkImage* dstImage, GrVkImage* srcImage, const SkIRect& srcRect, const SkIPoint& dstPoint) { if (!this->currentCommandBuffer()) { return; } #ifdef SK_DEBUG int dstSampleCnt = dstImage->vkImageInfo().fSampleCount; int srcSampleCnt = srcImage->vkImageInfo().fSampleCount; bool dstHasYcbcr = dstImage->ycbcrConversionInfo().isValid(); bool srcHasYcbcr = srcImage->ycbcrConversionInfo().isValid(); VkFormat dstFormat = dstImage->imageFormat(); VkFormat srcFormat; SkAssertResult(dst->backendFormat().asVkFormat(&srcFormat)); SkASSERT(this->vkCaps().canCopyAsBlit(dstFormat, dstSampleCnt, dstImage->isLinearTiled(), dstHasYcbcr, srcFormat, srcSampleCnt, srcImage->isLinearTiled(), srcHasYcbcr)); #endif if (src->isProtected() && !dst->isProtected()) { SkDebugf("Can't copy from protected memory to non-protected"); return; } dstImage->setImageLayout(this, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, VK_ACCESS_TRANSFER_WRITE_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, false); srcImage->setImageLayout(this, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, VK_ACCESS_TRANSFER_READ_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, false); // Flip rect if necessary SkIRect dstRect = SkIRect::MakeXYWH(dstPoint.fX, dstPoint.fY, srcRect.width(), srcRect.height()); VkImageBlit blitRegion; memset(&blitRegion, 0, sizeof(VkImageBlit)); blitRegion.srcSubresource = { VK_IMAGE_ASPECT_COLOR_BIT, 0, 0, 1 }; blitRegion.srcOffsets[0] = { srcRect.fLeft, srcRect.fTop, 0 }; blitRegion.srcOffsets[1] = { srcRect.fRight, srcRect.fBottom, 1 }; blitRegion.dstSubresource = { VK_IMAGE_ASPECT_COLOR_BIT, 0, 0, 1 }; blitRegion.dstOffsets[0] = { dstRect.fLeft, dstRect.fTop, 0 }; blitRegion.dstOffsets[1] = { dstRect.fRight, dstRect.fBottom, 1 }; this->currentCommandBuffer()->addGrSurface(sk_ref_sp(src)); this->currentCommandBuffer()->addGrSurface(sk_ref_sp(dst)); this->currentCommandBuffer()->blitImage(this, *srcImage, *dstImage, 1, &blitRegion, VK_FILTER_NEAREST); // We never scale so any filter works here // The rect is already in device space so we pass in kTopLeft so no flip is done. this->didWriteToSurface(dst, kTopLeft_GrSurfaceOrigin, &dstRect); } void GrVkGpu::copySurfaceAsResolve(GrSurface* dst, GrSurface* src, const SkIRect& srcRect, const SkIPoint& dstPoint) { if (src->isProtected() && !dst->isProtected()) { SkDebugf("Can't copy from protected memory to non-protected"); return; } GrVkRenderTarget* srcRT = static_cast(src->asRenderTarget()); this->resolveImage(dst, srcRT, srcRect, dstPoint); SkIRect dstRect = SkIRect::MakeXYWH(dstPoint.fX, dstPoint.fY, srcRect.width(), srcRect.height()); // The rect is already in device space so we pass in kTopLeft so no flip is done. this->didWriteToSurface(dst, kTopLeft_GrSurfaceOrigin, &dstRect); } bool GrVkGpu::onCopySurface(GrSurface* dst, GrSurface* src, const SkIRect& srcRect, const SkIPoint& dstPoint) { #ifdef SK_DEBUG if (GrVkRenderTarget* srcRT = static_cast(src->asRenderTarget())) { SkASSERT(!srcRT->wrapsSecondaryCommandBuffer()); } if (GrVkRenderTarget* dstRT = static_cast(dst->asRenderTarget())) { SkASSERT(!dstRT->wrapsSecondaryCommandBuffer()); } #endif if (src->isProtected() && !dst->isProtected()) { SkDebugf("Can't copy from protected memory to non-protected"); return false; } bool useDiscardableMSAA = this->vkCaps().preferDiscardableMSAAAttachment(); GrVkImage* dstImage; GrVkImage* srcImage; GrRenderTarget* dstRT = dst->asRenderTarget(); if (dstRT) { GrVkRenderTarget* vkRT = static_cast(dstRT); if (vkRT->wrapsSecondaryCommandBuffer()) { return false; } if (useDiscardableMSAA && vkRT->resolveAttachment() && vkRT->resolveAttachment()->supportsInputAttachmentUsage()) { dstImage = vkRT->resolveAttachment(); } else { dstImage = vkRT->colorAttachment(); } } else if (dst->asTexture()) { dstImage = static_cast(dst->asTexture())->textureAttachment(); } else { // The surface in a GrAttachment already dstImage = static_cast(dst); } GrRenderTarget* srcRT = src->asRenderTarget(); if (srcRT) { GrVkRenderTarget* vkRT = static_cast(srcRT); if (useDiscardableMSAA && vkRT->resolveAttachment() && vkRT->resolveAttachment()->supportsInputAttachmentUsage()) { srcImage = vkRT->resolveAttachment(); } else { srcImage = vkRT->colorAttachment(); } } else if (src->asTexture()) { SkASSERT(src->asTexture()); srcImage = static_cast(src->asTexture())->textureAttachment(); } else { // The surface in a GrAttachment already srcImage = static_cast(src); } VkFormat dstFormat = dstImage->imageFormat(); VkFormat srcFormat = srcImage->imageFormat(); int dstSampleCnt = dstImage->vkImageInfo().fSampleCount; int srcSampleCnt = srcImage->vkImageInfo().fSampleCount; bool dstHasYcbcr = dstImage->ycbcrConversionInfo().isValid(); bool srcHasYcbcr = srcImage->ycbcrConversionInfo().isValid(); if (this->vkCaps().canCopyAsResolve(dstFormat, dstSampleCnt, dstHasYcbcr, srcFormat, srcSampleCnt, srcHasYcbcr)) { this->copySurfaceAsResolve(dst, src, srcRect, dstPoint); return true; } if (this->vkCaps().canCopyImage(dstFormat, dstSampleCnt, dstHasYcbcr, srcFormat, srcSampleCnt, srcHasYcbcr)) { this->copySurfaceAsCopyImage(dst, src, dstImage, srcImage, srcRect, dstPoint); return true; } if (this->vkCaps().canCopyAsBlit(dstFormat, dstSampleCnt, dstImage->isLinearTiled(), dstHasYcbcr, srcFormat, srcSampleCnt, srcImage->isLinearTiled(), srcHasYcbcr)) { this->copySurfaceAsBlit(dst, src, dstImage, srcImage, srcRect, dstPoint); return true; } return false; } bool GrVkGpu::onReadPixels(GrSurface* surface, int left, int top, int width, int height, GrColorType surfaceColorType, GrColorType dstColorType, void* buffer, size_t rowBytes) { if (surface->isProtected()) { return false; } if (!this->currentCommandBuffer()) { return false; } GrVkImage* image = nullptr; GrVkRenderTarget* rt = static_cast(surface->asRenderTarget()); if (rt) { // Reading from render targets that wrap a secondary command buffer is not allowed since // it would require us to know the VkImage, which we don't have, as well as need us to // stop and start the VkRenderPass which we don't have access to. if (rt->wrapsSecondaryCommandBuffer()) { return false; } image = rt->nonMSAAAttachment(); } else { image = static_cast(surface->asTexture())->textureAttachment(); } if (!image) { return false; } if (dstColorType == GrColorType::kUnknown || dstColorType != this->vkCaps().transferColorType(image->imageFormat(), surfaceColorType)) { return false; } // Change layout of our target so it can be used as copy image->setImageLayout(this, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, VK_ACCESS_TRANSFER_READ_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, false); size_t bpp = GrColorTypeBytesPerPixel(dstColorType); if (GrVkFormatBytesPerBlock(image->imageFormat()) != bpp) { return false; } size_t tightRowBytes = bpp * width; VkBufferImageCopy region; memset(®ion, 0, sizeof(VkBufferImageCopy)); VkOffset3D offset = { left, top, 0 }; region.imageOffset = offset; region.imageExtent = { (uint32_t)width, (uint32_t)height, 1 }; size_t transBufferRowBytes = bpp * region.imageExtent.width; size_t imageRows = region.imageExtent.height; GrResourceProvider* resourceProvider = this->getContext()->priv().resourceProvider(); sk_sp transferBuffer = resourceProvider->createBuffer( transBufferRowBytes * imageRows, GrGpuBufferType::kXferGpuToCpu, kDynamic_GrAccessPattern); if (!transferBuffer) { return false; } GrVkBuffer* vkBuffer = static_cast(transferBuffer.get()); // Copy the image to a buffer so we can map it to cpu memory region.bufferOffset = 0; region.bufferRowLength = 0; // Forces RowLength to be width. We handle the rowBytes below. region.bufferImageHeight = 0; // Forces height to be tightly packed. Only useful for 3d images. region.imageSubresource = { VK_IMAGE_ASPECT_COLOR_BIT, 0, 0, 1 }; this->currentCommandBuffer()->copyImageToBuffer(this, image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, transferBuffer, 1, ®ion); // make sure the copy to buffer has finished vkBuffer->addMemoryBarrier(VK_ACCESS_TRANSFER_WRITE_BIT, VK_ACCESS_HOST_READ_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_HOST_BIT, false); // We need to submit the current command buffer to the Queue and make sure it finishes before // we can copy the data out of the buffer. if (!this->submitCommandBuffer(kForce_SyncQueue)) { return false; } void* mappedMemory = transferBuffer->map(); SkRectMemcpy(buffer, rowBytes, mappedMemory, transBufferRowBytes, tightRowBytes, height); transferBuffer->unmap(); return true; } bool GrVkGpu::beginRenderPass(const GrVkRenderPass* renderPass, sk_sp framebuffer, const VkClearValue* colorClear, const GrSurface* target, const SkIRect& renderPassBounds, bool forSecondaryCB) { if (!this->currentCommandBuffer()) { return false; } SkASSERT (!framebuffer->isExternal()); #ifdef SK_DEBUG uint32_t index; bool result = renderPass->colorAttachmentIndex(&index); SkASSERT(result && 0 == index); result = renderPass->stencilAttachmentIndex(&index); if (result) { SkASSERT(1 == index); } #endif VkClearValue clears[3]; int stencilIndex = renderPass->hasResolveAttachment() ? 2 : 1; clears[0].color = colorClear->color; clears[stencilIndex].depthStencil.depth = 0.0f; clears[stencilIndex].depthStencil.stencil = 0; return this->currentCommandBuffer()->beginRenderPass( this, renderPass, std::move(framebuffer), clears, target, renderPassBounds, forSecondaryCB); } void GrVkGpu::endRenderPass(GrRenderTarget* target, GrSurfaceOrigin origin, const SkIRect& bounds) { // We had a command buffer when we started the render pass, we should have one now as well. SkASSERT(this->currentCommandBuffer()); this->currentCommandBuffer()->endRenderPass(this); this->didWriteToSurface(target, origin, &bounds); } bool GrVkGpu::checkVkResult(VkResult result) { switch (result) { case VK_SUCCESS: return true; case VK_ERROR_DEVICE_LOST: fDeviceIsLost = true; return false; case VK_ERROR_OUT_OF_DEVICE_MEMORY: case VK_ERROR_OUT_OF_HOST_MEMORY: this->setOOMed(); return false; default: return false; } } void GrVkGpu::submitSecondaryCommandBuffer(std::unique_ptr buffer) { if (!this->currentCommandBuffer()) { return; } this->currentCommandBuffer()->executeCommands(this, std::move(buffer)); } void GrVkGpu::submit(GrOpsRenderPass* renderPass) { SkASSERT(fCachedOpsRenderPass.get() == renderPass); fCachedOpsRenderPass->submit(); fCachedOpsRenderPass->reset(); } GrFence SK_WARN_UNUSED_RESULT GrVkGpu::insertFence() { VkFenceCreateInfo createInfo; memset(&createInfo, 0, sizeof(VkFenceCreateInfo)); createInfo.sType = VK_STRUCTURE_TYPE_FENCE_CREATE_INFO; createInfo.pNext = nullptr; createInfo.flags = 0; VkFence fence = VK_NULL_HANDLE; VkResult result; VK_CALL_RET(result, CreateFence(this->device(), &createInfo, nullptr, &fence)); if (result != VK_SUCCESS) { return 0; } VK_CALL_RET(result, QueueSubmit(this->queue(), 0, nullptr, fence)); if (result != VK_SUCCESS) { VK_CALL(DestroyFence(this->device(), fence, nullptr)); return 0; } static_assert(sizeof(GrFence) >= sizeof(VkFence)); return (GrFence)fence; } bool GrVkGpu::waitFence(GrFence fence) { SkASSERT(VK_NULL_HANDLE != (VkFence)fence); VkResult result; VK_CALL_RET(result, WaitForFences(this->device(), 1, (VkFence*)&fence, VK_TRUE, 0)); return (VK_SUCCESS == result); } void GrVkGpu::deleteFence(GrFence fence) const { VK_CALL(DestroyFence(this->device(), (VkFence)fence, nullptr)); } std::unique_ptr SK_WARN_UNUSED_RESULT GrVkGpu::makeSemaphore(bool isOwned) { return GrVkSemaphore::Make(this, isOwned); } std::unique_ptr GrVkGpu::wrapBackendSemaphore( const GrBackendSemaphore& semaphore, GrResourceProvider::SemaphoreWrapType wrapType, GrWrapOwnership ownership) { return GrVkSemaphore::MakeWrapped(this, semaphore.vkSemaphore(), wrapType, ownership); } void GrVkGpu::insertSemaphore(GrSemaphore* semaphore) { SkASSERT(semaphore); GrVkSemaphore* vkSem = static_cast(semaphore); GrVkSemaphore::Resource* resource = vkSem->getResource(); if (resource->shouldSignal()) { resource->ref(); fSemaphoresToSignal.push_back(resource); } } void GrVkGpu::waitSemaphore(GrSemaphore* semaphore) { SkASSERT(semaphore); GrVkSemaphore* vkSem = static_cast(semaphore); GrVkSemaphore::Resource* resource = vkSem->getResource(); if (resource->shouldWait()) { resource->ref(); fSemaphoresToWaitOn.push_back(resource); } } std::unique_ptr GrVkGpu::prepareTextureForCrossContextUsage(GrTexture* texture) { SkASSERT(texture); GrVkAttachment* vkTexture = static_cast(texture)->textureAttachment(); vkTexture->setImageLayout(this, VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL, VK_ACCESS_SHADER_READ_BIT, VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, false); // TODO: should we have a way to notify the caller that this has failed? Currently if the submit // fails (caused by DEVICE_LOST) this will just cause us to fail the next use of the gpu. // Eventually we will abandon the whole GPU if this fails. this->submitToGpu(false); // The image layout change serves as a barrier, so no semaphore is needed. // If we ever decide we need to return a semaphore here, we need to make sure GrVkSemaphore is // thread safe so that only the first thread that tries to use the semaphore actually submits // it. This additionally would also require thread safety in command buffer submissions to // queues in general. return nullptr; } void GrVkGpu::addDrawable(std::unique_ptr drawable) { fDrawables.emplace_back(std::move(drawable)); } void GrVkGpu::storeVkPipelineCacheData() { if (this->getContext()->priv().getPersistentCache()) { this->resourceProvider().storePipelineCacheData(); } }