blob: a4c92f9b51d9404658882c8e7511d978a2d2a114 [file] [log] [blame]
//
// Copyright 2018 The ANGLE Project Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
//
// UtilsVk.cpp:
// Implements the UtilsVk class.
//
#include "libANGLE/renderer/vulkan/UtilsVk.h"
#include "common/spirv/spirv_instruction_builder_autogen.h"
#include "libANGLE/renderer/glslang_wrapper_utils.h"
#include "libANGLE/renderer/vulkan/ContextVk.h"
#include "libANGLE/renderer/vulkan/DisplayVk.h"
#include "libANGLE/renderer/vulkan/FramebufferVk.h"
#include "libANGLE/renderer/vulkan/RenderTargetVk.h"
#include "libANGLE/renderer/vulkan/RendererVk.h"
#include "libANGLE/renderer/vulkan/vk_utils.h"
namespace rx
{
namespace ConvertVertex_comp = vk::InternalShader::ConvertVertex_comp;
namespace ImageClear_frag = vk::InternalShader::ImageClear_frag;
namespace ImageCopy_frag = vk::InternalShader::ImageCopy_frag;
namespace BlitResolve_frag = vk::InternalShader::BlitResolve_frag;
namespace BlitResolveStencilNoExport_comp = vk::InternalShader::BlitResolveStencilNoExport_comp;
namespace OverlayCull_comp = vk::InternalShader::OverlayCull_comp;
namespace OverlayDraw_comp = vk::InternalShader::OverlayDraw_comp;
namespace ConvertIndexIndirectLineLoop_comp = vk::InternalShader::ConvertIndexIndirectLineLoop_comp;
namespace GenerateMipmap_comp = vk::InternalShader::GenerateMipmap_comp;
namespace spirv = angle::spirv;
namespace
{
constexpr uint32_t kConvertIndexDestinationBinding = 0;
constexpr uint32_t kConvertVertexDestinationBinding = 0;
constexpr uint32_t kConvertVertexSourceBinding = 1;
constexpr uint32_t kImageCopySourceBinding = 0;
constexpr uint32_t kBlitResolveColorOrDepthBinding = 0;
constexpr uint32_t kBlitResolveStencilBinding = 1;
constexpr uint32_t kBlitResolveSamplerBinding = 2;
constexpr uint32_t kBlitResolveStencilNoExportDestBinding = 0;
constexpr uint32_t kBlitResolveStencilNoExportSrcBinding = 1;
constexpr uint32_t kBlitResolveStencilNoExportSamplerBinding = 2;
constexpr uint32_t kOverlayCullCulledWidgetsBinding = 0;
constexpr uint32_t kOverlayCullWidgetCoordsBinding = 1;
constexpr uint32_t kOverlayDrawOutputBinding = 0;
constexpr uint32_t kOverlayDrawTextWidgetsBinding = 1;
constexpr uint32_t kOverlayDrawGraphWidgetsBinding = 2;
constexpr uint32_t kOverlayDrawCulledWidgetsBinding = 3;
constexpr uint32_t kOverlayDrawFontBinding = 4;
constexpr uint32_t kGenerateMipmapDestinationBinding = 0;
constexpr uint32_t kGenerateMipmapSourceBinding = 1;
bool ValidateFloatOneAsUint()
{
union
{
uint32_t asUint;
float asFloat;
} one;
one.asUint = gl::Float32One;
return one.asFloat == 1.0f;
}
uint32_t GetConvertVertexFlags(const UtilsVk::ConvertVertexParameters &params)
{
bool srcIsSint = params.srcFormat->isSint();
bool srcIsUint = params.srcFormat->isUint();
bool srcIsSnorm = params.srcFormat->isSnorm();
bool srcIsUnorm = params.srcFormat->isUnorm();
bool srcIsFixed = params.srcFormat->isFixed;
bool srcIsFloat = params.srcFormat->isFloat();
bool srcIsHalfFloat = params.srcFormat->isVertexTypeHalfFloat();
bool dstIsSint = params.dstFormat->isSint();
bool dstIsUint = params.dstFormat->isUint();
bool dstIsSnorm = params.dstFormat->isSnorm();
bool dstIsUnorm = params.dstFormat->isUnorm();
bool dstIsFloat = params.dstFormat->isFloat();
bool dstIsHalfFloat = params.dstFormat->isVertexTypeHalfFloat();
// Assert on the types to make sure the shader supports its. These are based on
// ConvertVertex_comp::Conversion values.
ASSERT(!dstIsSint || srcIsSint); // If destination is sint, src must be sint too
ASSERT(!dstIsUint || srcIsUint); // If destination is uint, src must be uint too
ASSERT(!srcIsFixed || dstIsFloat); // If source is fixed, dst must be float
// One of each bool set must be true
ASSERT(srcIsSint || srcIsUint || srcIsSnorm || srcIsUnorm || srcIsFixed || srcIsFloat);
ASSERT(dstIsSint || dstIsUint || dstIsSnorm || dstIsUnorm || dstIsFloat || dstIsHalfFloat);
// We currently don't have any big-endian devices in the list of supported platforms. The
// shader is capable of supporting big-endian architectures, but the relevant flag (IsBigEndian)
// is not added to the build configuration file (to reduce binary size). If necessary, add
// IsBigEndian to ConvertVertex.comp.json and select the appropriate flag based on the
// endian-ness test here.
ASSERT(IsLittleEndian());
uint32_t flags = 0;
if (srcIsHalfFloat && dstIsHalfFloat)
{
// Note that HalfFloat conversion uses the same shader as Uint.
flags = ConvertVertex_comp::kUintToUint;
}
else if ((srcIsSnorm && dstIsSnorm) || (srcIsUnorm && dstIsUnorm))
{
// Do snorm->snorm and unorm->unorm copies using the uint->uint shader. Currently only
// supported for same-width formats, so it's only used when adding channels.
ASSERT(params.srcFormat->redBits == params.dstFormat->redBits);
flags = ConvertVertex_comp::kUintToUint;
}
else if (srcIsSint && dstIsSint)
{
flags = ConvertVertex_comp::kSintToSint;
}
else if (srcIsUint && dstIsUint)
{
flags = ConvertVertex_comp::kUintToUint;
}
else if (srcIsSint)
{
flags = ConvertVertex_comp::kSintToFloat;
}
else if (srcIsUint)
{
flags = ConvertVertex_comp::kUintToFloat;
}
else if (srcIsSnorm)
{
flags = ConvertVertex_comp::kSnormToFloat;
}
else if (srcIsUnorm)
{
flags = ConvertVertex_comp::kUnormToFloat;
}
else if (srcIsFixed)
{
flags = ConvertVertex_comp::kFixedToFloat;
}
else if (srcIsFloat)
{
flags = ConvertVertex_comp::kFloatToFloat;
}
else
{
UNREACHABLE();
}
return flags;
}
uint32_t GetImageClearFlags(const angle::Format &format, uint32_t attachmentIndex, bool clearDepth)
{
constexpr uint32_t kAttachmentFlagStep =
ImageClear_frag::kAttachment1 - ImageClear_frag::kAttachment0;
static_assert(gl::IMPLEMENTATION_MAX_DRAW_BUFFERS == 8,
"ImageClear shader assumes maximum 8 draw buffers");
static_assert(
ImageClear_frag::kAttachment0 + 7 * kAttachmentFlagStep == ImageClear_frag::kAttachment7,
"ImageClear AttachmentN flag calculation needs correction");
uint32_t flags = ImageClear_frag::kAttachment0 + attachmentIndex * kAttachmentFlagStep;
if (format.isSint())
{
flags |= ImageClear_frag::kIsSint;
}
else if (format.isUint())
{
flags |= ImageClear_frag::kIsUint;
}
else
{
flags |= ImageClear_frag::kIsFloat;
}
if (clearDepth)
{
flags |= ImageClear_frag::kClearDepth;
}
return flags;
}
uint32_t GetFormatFlags(const angle::Format &format,
uint32_t intFlag,
uint32_t uintFlag,
uint32_t floatFlag)
{
if (format.isSint())
{
return intFlag;
}
if (format.isUint())
{
return uintFlag;
}
return floatFlag;
}
uint32_t GetImageCopyFlags(const angle::Format &srcIntendedFormat,
const angle::Format &dstIntendedFormat)
{
uint32_t flags = 0;
flags |= GetFormatFlags(srcIntendedFormat, ImageCopy_frag::kSrcIsSint,
ImageCopy_frag::kSrcIsUint, ImageCopy_frag::kSrcIsFloat);
flags |= GetFormatFlags(dstIntendedFormat, ImageCopy_frag::kDestIsSint,
ImageCopy_frag::kDestIsUint, ImageCopy_frag::kDestIsFloat);
return flags;
}
uint32_t GetBlitResolveFlags(bool blitColor,
bool blitDepth,
bool blitStencil,
const angle::Format &intendedFormat)
{
if (blitColor)
{
return GetFormatFlags(intendedFormat, BlitResolve_frag::kBlitColorInt,
BlitResolve_frag::kBlitColorUint, BlitResolve_frag::kBlitColorFloat);
}
if (blitDepth)
{
if (blitStencil)
{
return BlitResolve_frag::kBlitDepthStencil;
}
else
{
return BlitResolve_frag::kBlitDepth;
}
}
else
{
return BlitResolve_frag::kBlitStencil;
}
}
uint32_t GetConvertIndexIndirectLineLoopFlag(uint32_t indicesBitsWidth)
{
switch (indicesBitsWidth)
{
case 8:
return ConvertIndexIndirectLineLoop_comp::kIs8Bits;
case 16:
return ConvertIndexIndirectLineLoop_comp::kIs16Bits;
case 32:
return ConvertIndexIndirectLineLoop_comp::kIs32Bits;
default:
UNREACHABLE();
return 0;
}
}
uint32_t GetGenerateMipmapFlags(ContextVk *contextVk, const angle::Format &actualFormat)
{
uint32_t flags = 0;
// Note: If bits-per-component is 8 or 16 and float16 is supported in the shader, use that for
// faster math.
const bool hasShaderFloat16 =
contextVk->getRenderer()->getFeatures().supportsShaderFloat16.enabled;
if (actualFormat.redBits <= 8)
{
flags = hasShaderFloat16 ? GenerateMipmap_comp::kIsRGBA8_UseHalf
: GenerateMipmap_comp::kIsRGBA8;
}
else if (actualFormat.redBits <= 16)
{
flags = hasShaderFloat16 ? GenerateMipmap_comp::kIsRGBA16_UseHalf
: GenerateMipmap_comp::kIsRGBA16;
}
else
{
flags = GenerateMipmap_comp::kIsRGBA32F;
}
flags |= UtilsVk::GetGenerateMipmapMaxLevels(contextVk) == UtilsVk::kGenerateMipmapMaxLevels
? GenerateMipmap_comp::kDestSize6
: GenerateMipmap_comp::kDestSize4;
return flags;
}
enum UnresolveColorAttachmentType
{
kUnresolveTypeUnused = 0,
kUnresolveTypeFloat = 1,
kUnresolveTypeSint = 2,
kUnresolveTypeUint = 3,
};
uint32_t GetUnresolveFlags(uint32_t colorAttachmentCount,
const gl::DrawBuffersArray<vk::ImageHelper *> &colorSrc,
bool unresolveDepth,
bool unresolveStencil,
gl::DrawBuffersArray<UnresolveColorAttachmentType> *attachmentTypesOut)
{
uint32_t flags = 0;
for (uint32_t attachmentIndex = 0; attachmentIndex < colorAttachmentCount; ++attachmentIndex)
{
const angle::Format &format = colorSrc[attachmentIndex]->getIntendedFormat();
UnresolveColorAttachmentType type = kUnresolveTypeFloat;
if (format.isSint())
{
type = kUnresolveTypeSint;
}
else if (format.isUint())
{
type = kUnresolveTypeUint;
}
(*attachmentTypesOut)[attachmentIndex] = type;
// |flags| is comprised of |colorAttachmentCount| values from
// |UnresolveColorAttachmentType|, each taking up 2 bits.
flags |= type << (2 * attachmentIndex);
}
// Additionally, two bits are used for depth and stencil unresolve.
constexpr uint32_t kDepthUnresolveFlagBit = 2 * gl::IMPLEMENTATION_MAX_DRAW_BUFFERS;
constexpr uint32_t kStencilUnresolveFlagBit = kDepthUnresolveFlagBit + 1;
if (unresolveDepth)
{
flags |= 1 << kDepthUnresolveFlagBit;
}
if (unresolveStencil)
{
flags |= 1 << kStencilUnresolveFlagBit;
}
return flags;
}
uint32_t GetFormatDefaultChannelMask(const angle::Format &intendedImageFormat,
const angle::Format &actualImageFormat)
{
uint32_t mask = 0;
// Red can never be introduced due to format emulation (except for luma which is handled
// especially)
ASSERT(((intendedImageFormat.redBits > 0) == (actualImageFormat.redBits > 0)) ||
intendedImageFormat.isLUMA());
mask |= intendedImageFormat.greenBits == 0 && actualImageFormat.greenBits > 0 ? 2 : 0;
mask |= intendedImageFormat.blueBits == 0 && actualImageFormat.blueBits > 0 ? 4 : 0;
mask |= intendedImageFormat.alphaBits == 0 && actualImageFormat.alphaBits > 0 ? 8 : 0;
return mask;
}
// Calculate the transformation offset for blit/resolve. See BlitResolve.frag for details on how
// these values are derived.
void CalculateBlitOffset(const UtilsVk::BlitResolveParameters &params, float offset[2])
{
int srcOffsetFactorX = params.flipX ? -1 : 1;
int srcOffsetFactorY = params.flipY ? -1 : 1;
offset[0] = params.dstOffset[0] * params.stretch[0] - params.srcOffset[0] * srcOffsetFactorX;
offset[1] = params.dstOffset[1] * params.stretch[1] - params.srcOffset[1] * srcOffsetFactorY;
}
void CalculateResolveOffset(const UtilsVk::BlitResolveParameters &params, int32_t offset[2])
{
int srcOffsetFactorX = params.flipX ? -1 : 1;
int srcOffsetFactorY = params.flipY ? -1 : 1;
// There's no stretching in resolve.
offset[0] = params.dstOffset[0] - params.srcOffset[0] * srcOffsetFactorX;
offset[1] = params.dstOffset[1] - params.srcOffset[1] * srcOffsetFactorY;
}
// Sets the appropriate settings in the pipeline for the shader to output stencil. Requires the
// shader stencil export extension.
void SetStencilForShaderExport(ContextVk *contextVk, vk::GraphicsPipelineDesc *desc)
{
ASSERT(contextVk->getRenderer()->getFeatures().supportsShaderStencilExport.enabled);
const uint8_t completeMask = 0xFF;
const uint8_t unusedReference = 0x00;
desc->setStencilTestEnabled(true);
desc->setStencilFrontFuncs(unusedReference, VK_COMPARE_OP_ALWAYS, completeMask);
desc->setStencilBackFuncs(unusedReference, VK_COMPARE_OP_ALWAYS, completeMask);
desc->setStencilFrontOps(VK_STENCIL_OP_REPLACE, VK_STENCIL_OP_REPLACE, VK_STENCIL_OP_REPLACE);
desc->setStencilBackOps(VK_STENCIL_OP_REPLACE, VK_STENCIL_OP_REPLACE, VK_STENCIL_OP_REPLACE);
desc->setStencilFrontWriteMask(completeMask);
desc->setStencilBackWriteMask(completeMask);
}
namespace unresolve
{
// The unresolve shader looks like the following, based on the number and types of unresolve
// attachments.
//
// #version 450 core
// #extension GL_ARB_shader_stencil_export : require
//
// layout(location = 0) out vec4 colorOut0;
// layout(location = 1) out ivec4 colorOut1;
// layout(location = 2) out uvec4 colorOut2;
// layout(input_attachment_index = 0, set = 0, binding = 0) uniform subpassInput colorIn0;
// layout(input_attachment_index = 1, set = 0, binding = 1) uniform isubpassInput colorIn1;
// layout(input_attachment_index = 2, set = 0, binding = 2) uniform usubpassInput colorIn2;
// layout(input_attachment_index = 3, set = 0, binding = 3) uniform subpassInput depthIn;
// layout(input_attachment_index = 3, set = 0, binding = 4) uniform usubpassInput stencilIn;
//
// void main()
// {
// colorOut0 = subpassLoad(colorIn0);
// colorOut1 = subpassLoad(colorIn1);
// colorOut2 = subpassLoad(colorIn2);
// gl_FragDepth = subpassLoad(depthIn).x;
// gl_FragStencilRefARB = int(subpassLoad(stencilIn).x);
// }
//
// This shader compiles to the following SPIR-V:
//
// OpCapability Shader \
// OpCapability InputAttachment \
// OpCapability StencilExportEXT \ Preamble. Mostly fixed, except
// OpExtension "SPV_EXT_shader_stencil_export" \ OpEntryPoint should enumerate
// %1 = OpExtInstImport "GLSL.std.450" \ out variables, stencil export
// OpMemoryModel Logical GLSL450 / is conditional to stencil
// OpEntryPoint Fragment %4 "main" %26 %27 %28 %29 %30 / unresolve, and depth replacing
// OpExecutionMode %4 OriginUpperLeft / conditional to depth unresolve.
// OpExecutionMode %4 DepthReplacing /
// OpSource GLSL 450 /
//
// OpName %4 "main" \
// OpName %26 "colorOut0" \
// OpName %27 "colorOut1" \
// OpName %28 "colorOut2" \
// OpName %29 "gl_FragDepth" \ Debug information. Not generated here.
// OpName %30 "gl_FragStencilRefARB" /
// OpName %31 "colorIn0" /
// OpName %32 "colorIn1" /
// OpName %33 "colorIn2" /
// OpName %34 "depthIn" /
// OpName %35 "stencilIn" /
//
// OpDecorate %26 Location 0 \
// OpDecorate %27 Location 1 \ Location decoration of out variables.
// OpDecorate %28 Location 2 /
//
// OpDecorate %29 BuiltIn FragDepth \ Builtin outputs, conditional to depth
// OpDecorate %30 BuiltIn FragStencilRefEXT / and stencil unresolve.
//
// OpDecorate %31 DescriptorSet 0 \
// OpDecorate %31 Binding 0 \
// OpDecorate %31 InputAttachmentIndex 0 \
// OpDecorate %32 DescriptorSet 0 \
// OpDecorate %32 Binding 1 \
// OpDecorate %32 InputAttachmentIndex 1 \
// OpDecorate %33 DescriptorSet 0 \ set, binding and input_attachment
// OpDecorate %33 Binding 2 \ decorations of the subpassInput
// OpDecorate %33 InputAttachmentIndex 2 / variables.
// OpDecorate %34 DescriptorSet 0 /
// OpDecorate %34 Binding 3 /
// OpDecorate %34 InputAttachmentIndex 3 /
// OpDecorate %35 DescriptorSet 0 /
// OpDecorate %35 Binding 4 /
// OpDecorate %35 InputAttachmentIndex 3 /
//
// %2 = OpTypeVoid \ Type of main(). Fixed.
// %3 = OpTypeFunction %2 /
//
// %6 = OpTypeFloat 32 \
// %7 = OpTypeVector %6 4 \
// %8 = OpTypePointer Output %7 \ Type declaration for "out vec4"
// %9 = OpTypeImage %6 SubpassData 0 0 0 2 Unknown / and "subpassInput". Fixed.
// %10 = OpTypePointer UniformConstant %9 /
//
// %11 = OpTypeInt 32 1 \
// %12 = OpTypeVector %11 4 \
// %13 = OpTypePointer Output %12 \ Type declaration for "out ivec4"
// %14 = OpTypeImage %11 SubpassData 0 0 0 2 Unknown / and "isubpassInput". Fixed.
// %15 = OpTypePointer UniformConstant %14 /
//
// %16 = OpTypeInt 32 0 \
// %17 = OpTypeVector %16 4 \
// %18 = OpTypePointer Output %17 \ Type declaration for "out uvec4"
// %19 = OpTypeImage %16 SubpassData 0 0 0 2 Unknown / and "usubpassInput". Fixed.
// %20 = OpTypePointer UniformConstant %19 /
//
// %21 = OpTypePointer Output %6 \ Type declaraions for depth and stencil. Fixed.
// %22 = OpTypePointer Output %11 /
//
// %23 = OpConstant %11 0 \
// %24 = OpTypeVector %11 2 \ ivec2(0) for OpImageRead. subpassLoad
// %25 = OpConstantComposite %22 %21 %21 / doesn't require coordinates. Fixed.
//
// %26 = OpVariable %8 Output \
// %27 = OpVariable %13 Output \
// %28 = OpVariable %18 Output \
// %29 = OpVariable %21 Output \
// %30 = OpVariable %22 Output \ Actual "out" and "*subpassInput"
// %31 = OpVariable %10 UniformConstant / variable declarations.
// %32 = OpVariable %15 UniformConstant /
// %33 = OpVariable %20 UniformConstant /
// %34 = OpVariable %10 UniformConstant /
// %35 = OpVariable %20 UniformConstant /
//
// %4 = OpFunction %2 None %3 \ Top of main(). Fixed.
// %5 = OpLabel /
//
// %36 = OpLoad %9 %31 \
// %37 = OpImageRead %7 %36 %23 \ colorOut0 = subpassLoad(colorIn0);
// OpStore %26 %37 /
//
// %38 = OpLoad %14 %32 \
// %39 = OpImageRead %12 %38 %23 \ colorOut1 = subpassLoad(colorIn1);
// OpStore %27 %39 /
//
// %40 = OpLoad %19 %33 \
// %41 = OpImageRead %17 %40 %23 \ colorOut2 = subpassLoad(colorIn2);
// OpStore %28 %41 /
//
// %42 = OpLoad %9 %34 \
// %43 = OpImageRead %7 %42 %23 \ gl_FragDepth = subpassLoad(depthIn).x;
// %44 = OpCompositeExtract %6 %43 0 /
// OpStore %29 %44 /
//
// %45 = OpLoad %19 %35 \
// %46 = OpImageRead %17 %45 %23 \
// %47 = OpCompositeExtract %16 %46 0 \ gl_FragStencilRefARB = int(subpassLoad(stencilIn).x);
// %48 = OpBitcast %11 %47 /
// OpStore %30 %48 /
//
// OpReturn \ Bottom of main(). Fixed.
// OpFunctionEnd /
//
// What makes the generation of this shader manageable is that the majority of it is constant
// between the different variations of the shader. The rest are repeating patterns with different
// ids or indices.
enum
{
// main() ids
kIdExtInstImport = 1,
kIdVoid,
kIdMainType,
kIdMain,
kIdMainLabel,
// Types for "out vec4" and "subpassInput"
kIdFloatType,
kIdFloat4Type,
kIdFloat4OutType,
kIdFloatSubpassImageType,
kIdFloatSubpassInputType,
// Types for "out ivec4" and "isubpassInput"
kIdSIntType,
kIdSInt4Type,
kIdSInt4OutType,
kIdSIntSubpassImageType,
kIdSIntSubpassInputType,
// Types for "out uvec4" and "usubpassInput"
kIdUIntType,
kIdUInt4Type,
kIdUInt4OutType,
kIdUIntSubpassImageType,
kIdUIntSubpassInputType,
// Types for gl_FragDepth && gl_FragStencilRefARB
kIdFloatOutType,
kIdSIntOutType,
// ivec2(0) constant
kIdSIntZero,
kIdSInt2Type,
kIdSInt2Zero,
// Output variable ids
kIdColor0Out,
kIdDepthOut = kIdColor0Out + gl::IMPLEMENTATION_MAX_DRAW_BUFFERS,
kIdStencilOut,
// Input variable ids
kIdColor0In,
kIdDepthIn = kIdColor0In + gl::IMPLEMENTATION_MAX_DRAW_BUFFERS,
kIdStencilIn,
// Ids for temp variables
kIdColor0Load,
// 2 temp ids per color unresolve
kIdDepthLoad = kIdColor0Load + gl::IMPLEMENTATION_MAX_DRAW_BUFFERS * 2,
// 3 temp ids for depth unresolve
kIdStencilLoad = kIdDepthLoad + 3,
// Total number of ids used
// 4 temp ids for stencil unresolve
kIdCount = kIdStencilLoad + 4,
};
void InsertPreamble(uint32_t colorAttachmentCount,
bool unresolveDepth,
bool unresolveStencil,
angle::spirv::Blob *blobOut)
{
spirv::WriteCapability(blobOut, spv::CapabilityShader);
spirv::WriteCapability(blobOut, spv::CapabilityInputAttachment);
if (unresolveStencil)
{
spirv::WriteCapability(blobOut, spv::CapabilityStencilExportEXT);
spirv::WriteExtension(blobOut, "SPV_EXT_shader_stencil_export");
}
// OpExtInstImport is actually not needed by this shader. We don't use any instructions from
// GLSL.std.450.
spirv::WriteMemoryModel(blobOut, spv::AddressingModelLogical, spv::MemoryModelGLSL450);
// Create the list of entry point ids, including only the out variables.
spirv::IdRefList entryPointIds;
for (uint32_t colorIndex = 0; colorIndex < colorAttachmentCount; ++colorIndex)
{
entryPointIds.push_back(spirv::IdRef(kIdColor0Out + colorIndex));
}
if (unresolveDepth)
{
entryPointIds.push_back(spirv::IdRef(kIdDepthOut));
}
if (unresolveStencil)
{
entryPointIds.push_back(spirv::IdRef(kIdStencilOut));
}
spirv::WriteEntryPoint(blobOut, spv::ExecutionModelFragment, spirv::IdRef(kIdMain), "main",
entryPointIds);
spirv::WriteExecutionMode(blobOut, spirv::IdRef(kIdMain), spv::ExecutionModeOriginUpperLeft,
{});
if (unresolveDepth)
{
spirv::WriteExecutionMode(blobOut, spirv::IdRef(kIdMain), spv::ExecutionModeDepthReplacing,
{});
}
spirv::WriteSource(blobOut, spv::SourceLanguageGLSL, spirv::LiteralInteger(450), nullptr,
nullptr);
}
void InsertInputDecorations(spirv::IdRef id,
uint32_t attachmentIndex,
uint32_t binding,
angle::spirv::Blob *blobOut)
{
spirv::WriteDecorate(blobOut, id, spv::DecorationDescriptorSet,
{spirv::LiteralInteger(ToUnderlying(DescriptorSetIndex::Internal))});
spirv::WriteDecorate(blobOut, id, spv::DecorationBinding, {spirv::LiteralInteger(binding)});
spirv::WriteDecorate(blobOut, id, spv::DecorationInputAttachmentIndex,
{spirv::LiteralInteger(attachmentIndex)});
}
void InsertColorDecorations(uint32_t colorIndex, angle::spirv::Blob *blobOut)
{
// Decorate the output color attachment with Location
spirv::WriteDecorate(blobOut, spirv::IdRef(kIdColor0Out + colorIndex), spv::DecorationLocation,
{spirv::LiteralInteger(colorIndex)});
// Decorate the subpasss input color attachment with Set/Binding/InputAttachmentIndex.
InsertInputDecorations(spirv::IdRef(kIdColor0In + colorIndex), colorIndex, colorIndex, blobOut);
}
void InsertDepthStencilDecorations(uint32_t depthStencilInputIndex,
uint32_t depthStencilBindingIndex,
bool unresolveDepth,
bool unresolveStencil,
angle::spirv::Blob *blobOut)
{
if (unresolveDepth)
{
// Decorate the output depth attachment with Location
spirv::WriteDecorate(blobOut, spirv::IdRef(kIdDepthOut), spv::DecorationBuiltIn,
{spirv::LiteralInteger(spv::BuiltInFragDepth)});
// Decorate the subpasss input depth attachment with Set/Binding/InputAttachmentIndex.
InsertInputDecorations(spirv::IdRef(kIdDepthIn), depthStencilInputIndex,
depthStencilBindingIndex, blobOut);
// Advance the binding. Note that the depth/stencil attachment has the same input
// attachment index (it's the same attachment in the subpass), but different bindings (one
// aspect per image view).
++depthStencilBindingIndex;
}
if (unresolveStencil)
{
// Decorate the output stencil attachment with Location
spirv::WriteDecorate(blobOut, spirv::IdRef(kIdStencilOut), spv::DecorationBuiltIn,
{spirv::LiteralInteger(spv::BuiltInFragStencilRefEXT)});
// Decorate the subpasss input stencil attachment with Set/Binding/InputAttachmentIndex.
InsertInputDecorations(spirv::IdRef(kIdStencilIn), depthStencilInputIndex,
depthStencilBindingIndex, blobOut);
}
}
void InsertDerivativeTypes(spirv::IdRef baseId,
spirv::IdRef vec4Id,
spirv::IdRef vec4OutId,
spirv::IdRef imageTypeId,
spirv::IdRef inputTypeId,
angle::spirv::Blob *blobOut)
{
spirv::WriteTypeVector(blobOut, vec4Id, baseId, spirv::LiteralInteger(4));
spirv::WriteTypePointer(blobOut, vec4OutId, spv::StorageClassOutput, vec4Id);
spirv::WriteTypeImage(blobOut, imageTypeId, baseId, spv::DimSubpassData,
// Unused with subpass inputs
spirv::LiteralInteger(0),
// Not arrayed
spirv::LiteralInteger(0),
// Not multisampled
spirv::LiteralInteger(0),
// Used without a sampler
spirv::LiteralInteger(2), spv::ImageFormatUnknown, nullptr);
spirv::WriteTypePointer(blobOut, inputTypeId, spv::StorageClassUniformConstant, imageTypeId);
}
void InsertCommonTypes(angle::spirv::Blob *blobOut)
{
// Types to support main().
spirv::WriteTypeVoid(blobOut, spirv::IdRef(kIdVoid));
spirv::WriteTypeFunction(blobOut, spirv::IdRef(kIdMainType), spirv::IdRef(kIdVoid), {});
// Float types
spirv::WriteTypeFloat(blobOut, spirv::IdRef(kIdFloatType), spirv::LiteralInteger(32));
InsertDerivativeTypes(spirv::IdRef(kIdFloatType), spirv::IdRef(kIdFloat4Type),
spirv::IdRef(kIdFloat4OutType), spirv::IdRef(kIdFloatSubpassImageType),
spirv::IdRef(kIdFloatSubpassInputType), blobOut);
// Int types
spirv::WriteTypeInt(blobOut, spirv::IdRef(kIdSIntType), spirv::LiteralInteger(32),
spirv::LiteralInteger(1));
InsertDerivativeTypes(spirv::IdRef(kIdSIntType), spirv::IdRef(kIdSInt4Type),
spirv::IdRef(kIdSInt4OutType), spirv::IdRef(kIdSIntSubpassImageType),
spirv::IdRef(kIdSIntSubpassInputType), blobOut);
// Unsigned int types
spirv::WriteTypeInt(blobOut, spirv::IdRef(kIdUIntType), spirv::LiteralInteger(32),
spirv::LiteralInteger(0));
InsertDerivativeTypes(spirv::IdRef(kIdUIntType), spirv::IdRef(kIdUInt4Type),
spirv::IdRef(kIdUInt4OutType), spirv::IdRef(kIdUIntSubpassImageType),
spirv::IdRef(kIdUIntSubpassInputType), blobOut);
// Types to support depth/stencil
spirv::WriteTypePointer(blobOut, spirv::IdRef(kIdFloatOutType), spv::StorageClassOutput,
spirv::IdRef(kIdFloatType));
spirv::WriteTypePointer(blobOut, spirv::IdRef(kIdSIntOutType), spv::StorageClassOutput,
spirv::IdRef(kIdSIntType));
// Constants used to load from subpass inputs
spirv::WriteConstant(blobOut, spirv::IdRef(kIdSIntType), spirv::IdRef(kIdSIntZero),
spirv::LiteralInteger(0));
spirv::WriteTypeVector(blobOut, spirv::IdRef(kIdSInt2Type), spirv::IdRef(kIdSIntType),
spirv::LiteralInteger(2));
spirv::WriteConstantComposite(blobOut, spirv::IdRef(kIdSInt2Type), spirv::IdRef(kIdSInt2Zero),
{spirv::IdRef(kIdSIntZero), spirv::IdRef(kIdSIntZero)});
}
void InsertVariableDecl(spirv::IdRef outType,
spirv::IdRef outId,
spirv::IdRef inType,
spirv::IdRef inId,
angle::spirv::Blob *blobOut)
{
// Declare both the output and subpass input variables.
spirv::WriteVariable(blobOut, outType, outId, spv::StorageClassOutput, nullptr);
spirv::WriteVariable(blobOut, inType, inId, spv::StorageClassUniformConstant, nullptr);
}
void InsertColorVariableDecl(uint32_t colorIndex,
UnresolveColorAttachmentType type,
angle::spirv::Blob *blobOut)
{
// Find the correct types for color variable declarations.
spirv::IdRef outType(kIdFloat4OutType);
spirv::IdRef outId(kIdColor0Out + colorIndex);
spirv::IdRef inType(kIdFloatSubpassInputType);
spirv::IdRef inId(kIdColor0In + colorIndex);
switch (type)
{
case kUnresolveTypeSint:
outType = spirv::IdRef(kIdSInt4OutType);
inType = spirv::IdRef(kIdSIntSubpassInputType);
break;
case kUnresolveTypeUint:
outType = spirv::IdRef(kIdUInt4OutType);
inType = spirv::IdRef(kIdUIntSubpassInputType);
break;
default:
break;
}
InsertVariableDecl(outType, outId, inType, inId, blobOut);
}
void InsertDepthStencilVariableDecl(bool unresolveDepth,
bool unresolveStencil,
angle::spirv::Blob *blobOut)
{
if (unresolveDepth)
{
InsertVariableDecl(spirv::IdRef(kIdFloatOutType), spirv::IdRef(kIdDepthOut),
spirv::IdRef(kIdFloatSubpassInputType), spirv::IdRef(kIdDepthIn),
blobOut);
}
if (unresolveStencil)
{
InsertVariableDecl(spirv::IdRef(kIdSIntOutType), spirv::IdRef(kIdStencilOut),
spirv::IdRef(kIdUIntSubpassInputType), spirv::IdRef(kIdStencilIn),
blobOut);
}
}
void InsertTopOfMain(angle::spirv::Blob *blobOut)
{
spirv::WriteFunction(blobOut, spirv::IdRef(kIdVoid), spirv::IdRef(kIdMain),
spv::FunctionControlMaskNone, spirv::IdRef(kIdMainType));
spirv::WriteLabel(blobOut, spirv::IdRef(kIdMainLabel));
}
void InsertColorUnresolveLoadStore(uint32_t colorIndex,
UnresolveColorAttachmentType type,
angle::spirv::Blob *blobOut)
{
spirv::IdRef loadResult(kIdColor0Load + colorIndex * 2);
spirv::IdRef imageReadResult(loadResult + 1);
// Find the correct types for load/store.
spirv::IdRef loadType(kIdFloatSubpassImageType);
spirv::IdRef readType(kIdFloat4Type);
spirv::IdRef inId(kIdColor0In + colorIndex);
spirv::IdRef outId(kIdColor0Out + colorIndex);
switch (type)
{
case kUnresolveTypeSint:
loadType = spirv::IdRef(kIdSIntSubpassImageType);
readType = spirv::IdRef(kIdSInt4Type);
break;
case kUnresolveTypeUint:
loadType = spirv::IdRef(kIdUIntSubpassImageType);
readType = spirv::IdRef(kIdUInt4Type);
break;
default:
break;
}
// Load the subpass input image, read from it, and store in output.
spirv::WriteLoad(blobOut, loadType, loadResult, inId, nullptr);
spirv::WriteImageRead(blobOut, readType, imageReadResult, loadResult,
spirv::IdRef(kIdSInt2Zero), nullptr, {});
spirv::WriteStore(blobOut, outId, imageReadResult, nullptr);
}
void InsertDepthStencilUnresolveLoadStore(bool unresolveDepth,
bool unresolveStencil,
angle::spirv::Blob *blobOut)
{
if (unresolveDepth)
{
spirv::IdRef loadResult(kIdDepthLoad);
spirv::IdRef imageReadResult(loadResult + 1);
spirv::IdRef extractResult(imageReadResult + 1);
spirv::IdRef loadType(kIdFloatSubpassImageType);
spirv::IdRef readType(kIdFloat4Type);
spirv::IdRef inId(kIdDepthIn);
spirv::IdRef outId(kIdDepthOut);
// Load the subpass input image, read from it, select .x, and store in output.
spirv::WriteLoad(blobOut, loadType, loadResult, inId, nullptr);
spirv::WriteImageRead(blobOut, readType, imageReadResult, loadResult,
spirv::IdRef(kIdSInt2Zero), nullptr, {});
spirv::WriteCompositeExtract(blobOut, spirv::IdRef(kIdFloatType), extractResult,
imageReadResult, {spirv::LiteralInteger(0)});
spirv::WriteStore(blobOut, outId, extractResult, nullptr);
}
if (unresolveStencil)
{
spirv::IdRef loadResult(kIdStencilLoad);
spirv::IdRef imageReadResult(loadResult + 1);
spirv::IdRef extractResult(imageReadResult + 1);
spirv::IdRef bitcastResult(extractResult + 1);
spirv::IdRef loadType(kIdUIntSubpassImageType);
spirv::IdRef readType(kIdUInt4Type);
spirv::IdRef inId(kIdStencilIn);
spirv::IdRef outId(kIdStencilOut);
// Load the subpass input image, read from it, select .x, and store in output. There's a
// bitcast involved since the stencil subpass input has unsigned type, while
// gl_FragStencilRefARB is signed!
spirv::WriteLoad(blobOut, loadType, loadResult, inId, nullptr);
spirv::WriteImageRead(blobOut, readType, imageReadResult, loadResult,
spirv::IdRef(kIdSInt2Zero), nullptr, {});
spirv::WriteCompositeExtract(blobOut, spirv::IdRef(kIdUIntType), extractResult,
imageReadResult, {spirv::LiteralInteger(0)});
spirv::WriteBitcast(blobOut, spirv::IdRef(kIdSIntType), bitcastResult, extractResult);
spirv::WriteStore(blobOut, outId, bitcastResult, nullptr);
}
}
void InsertBottomOfMain(angle::spirv::Blob *blobOut)
{
spirv::WriteReturn(blobOut);
spirv::WriteFunctionEnd(blobOut);
}
angle::spirv::Blob MakeFragShader(
uint32_t colorAttachmentCount,
gl::DrawBuffersArray<UnresolveColorAttachmentType> &colorAttachmentTypes,
bool unresolveDepth,
bool unresolveStencil)
{
angle::spirv::Blob code;
// Reserve a sensible amount of memory. A single-attachment shader is 169 words.
code.reserve(169);
// Header
spirv::WriteSpirvHeader(&code, kIdCount);
// The preamble
InsertPreamble(colorAttachmentCount, unresolveDepth, unresolveStencil, &code);
// Color attachment decorations
for (uint32_t colorIndex = 0; colorIndex < colorAttachmentCount; ++colorIndex)
{
InsertColorDecorations(colorIndex, &code);
}
const uint32_t depthStencilInputIndex = colorAttachmentCount;
uint32_t depthStencilBindingIndex = colorAttachmentCount;
InsertDepthStencilDecorations(depthStencilInputIndex, depthStencilBindingIndex, unresolveDepth,
unresolveStencil, &code);
// Common types
InsertCommonTypes(&code);
// Attachment declarations
for (uint32_t colorIndex = 0; colorIndex < colorAttachmentCount; ++colorIndex)
{
InsertColorVariableDecl(colorIndex, colorAttachmentTypes[colorIndex], &code);
}
InsertDepthStencilVariableDecl(unresolveDepth, unresolveStencil, &code);
// Top of main
InsertTopOfMain(&code);
// Load and store for each attachment
for (uint32_t colorIndex = 0; colorIndex < colorAttachmentCount; ++colorIndex)
{
InsertColorUnresolveLoadStore(colorIndex, colorAttachmentTypes[colorIndex], &code);
}
InsertDepthStencilUnresolveLoadStore(unresolveDepth, unresolveStencil, &code);
// Bottom of main
InsertBottomOfMain(&code);
return code;
}
} // namespace unresolve
angle::Result GetUnresolveFrag(
vk::Context *context,
uint32_t colorAttachmentCount,
gl::DrawBuffersArray<UnresolveColorAttachmentType> &colorAttachmentTypes,
bool unresolveDepth,
bool unresolveStencil,
vk::RefCounted<vk::ShaderAndSerial> *shader)
{
if (shader->get().valid())
{
return angle::Result::Continue;
}
angle::spirv::Blob shaderCode = unresolve::MakeFragShader(
colorAttachmentCount, colorAttachmentTypes, unresolveDepth, unresolveStencil);
ASSERT(spirv::Validate(shaderCode));
// Create shader lazily. Access will need to be locked for multi-threading.
return vk::InitShaderAndSerial(context, &shader->get(), shaderCode.data(),
shaderCode.size() * 4);
}
} // namespace
UtilsVk::ConvertVertexShaderParams::ConvertVertexShaderParams() = default;
UtilsVk::ImageCopyShaderParams::ImageCopyShaderParams() = default;
uint32_t UtilsVk::GetGenerateMipmapMaxLevels(ContextVk *contextVk)
{
RendererVk *renderer = contextVk->getRenderer();
uint32_t maxPerStageDescriptorStorageImages =
renderer->getPhysicalDeviceProperties().limits.maxPerStageDescriptorStorageImages;
// Vulkan requires that there be support for at least 4 storage images per stage.
constexpr uint32_t kMinimumStorageImagesLimit = 4;
ASSERT(maxPerStageDescriptorStorageImages >= kMinimumStorageImagesLimit);
// If fewer than max-levels are supported, use 4 levels (which is the minimum required number
// of storage image bindings).
return maxPerStageDescriptorStorageImages < kGenerateMipmapMaxLevels
? kMinimumStorageImagesLimit
: kGenerateMipmapMaxLevels;
}
UtilsVk::UtilsVk() : mPerfCounters{}, mCumulativePerfCounters{} {}
UtilsVk::~UtilsVk() = default;
void UtilsVk::destroy(RendererVk *renderer)
{
VkDevice device = renderer->getDevice();
outputCumulativePerfCounters();
for (Function f : angle::AllEnums<Function>())
{
for (auto &descriptorSetLayout : mDescriptorSetLayouts[f])
{
descriptorSetLayout.reset();
}
mPipelineLayouts[f].reset();
mDescriptorPools[f].destroy(device);
}
for (vk::ShaderProgramHelper &program : mConvertIndexPrograms)
{
program.destroy(renderer);
}
for (vk::ShaderProgramHelper &program : mConvertIndirectLineLoopPrograms)
{
program.destroy(renderer);
}
for (vk::ShaderProgramHelper &program : mConvertIndexIndirectLineLoopPrograms)
{
program.destroy(renderer);
}
for (vk::ShaderProgramHelper &program : mConvertVertexPrograms)
{
program.destroy(renderer);
}
mImageClearProgramVSOnly.destroy(renderer);
for (vk::ShaderProgramHelper &program : mImageClearPrograms)
{
program.destroy(renderer);
}
for (vk::ShaderProgramHelper &program : mImageCopyPrograms)
{
program.destroy(renderer);
}
for (vk::ShaderProgramHelper &program : mBlitResolvePrograms)
{
program.destroy(renderer);
}
for (vk::ShaderProgramHelper &program : mBlitResolveStencilNoExportPrograms)
{
program.destroy(renderer);
}
for (vk::ShaderProgramHelper &program : mOverlayCullPrograms)
{
program.destroy(renderer);
}
for (vk::ShaderProgramHelper &program : mOverlayDrawPrograms)
{
program.destroy(renderer);
}
for (vk::ShaderProgramHelper &program : mGenerateMipmapPrograms)
{
program.destroy(renderer);
}
for (auto &programIter : mUnresolvePrograms)
{
vk::ShaderProgramHelper &program = programIter.second;
program.destroy(renderer);
}
mUnresolvePrograms.clear();
for (auto &shaderIter : mUnresolveFragShaders)
{
vk::RefCounted<vk::ShaderAndSerial> &shader = shaderIter.second;
shader.get().destroy(device);
}
mUnresolveFragShaders.clear();
mPointSampler.destroy(device);
mLinearSampler.destroy(device);
}
angle::Result UtilsVk::ensureResourcesInitialized(ContextVk *contextVk,
Function function,
VkDescriptorPoolSize *setSizes,
size_t setSizesCount,
size_t pushConstantsSize)
{
vk::DescriptorSetLayoutDesc descriptorSetDesc;
bool isCompute = function >= Function::ComputeStartIndex;
const VkShaderStageFlags descStages =
isCompute ? VK_SHADER_STAGE_COMPUTE_BIT : VK_SHADER_STAGE_FRAGMENT_BIT;
uint32_t currentBinding = 0;
for (size_t i = 0; i < setSizesCount; ++i)
{
descriptorSetDesc.update(currentBinding, setSizes[i].type, setSizes[i].descriptorCount,
descStages, nullptr);
++currentBinding;
}
ANGLE_TRY(contextVk->getDescriptorSetLayoutCache().getDescriptorSetLayout(
contextVk, descriptorSetDesc,
&mDescriptorSetLayouts[function][DescriptorSetIndex::Internal]));
vk::DescriptorSetLayoutBindingVector bindingVector;
std::vector<VkSampler> immutableSamplers;
descriptorSetDesc.unpackBindings(&bindingVector, &immutableSamplers);
std::vector<VkDescriptorPoolSize> descriptorPoolSizes;
for (const VkDescriptorSetLayoutBinding &binding : bindingVector)
{
if (binding.descriptorCount > 0)
{
VkDescriptorPoolSize poolSize = {};
poolSize.type = binding.descriptorType;
poolSize.descriptorCount = binding.descriptorCount;
descriptorPoolSizes.emplace_back(poolSize);
}
}
if (!descriptorPoolSizes.empty())
{
ANGLE_TRY(mDescriptorPools[function].init(
contextVk, descriptorPoolSizes.data(), descriptorPoolSizes.size(),
mDescriptorSetLayouts[function][DescriptorSetIndex::Internal].get().getHandle()));
}
gl::ShaderType pushConstantsShaderStage =
isCompute ? gl::ShaderType::Compute : gl::ShaderType::Fragment;
// Corresponding pipeline layouts:
vk::PipelineLayoutDesc pipelineLayoutDesc;
pipelineLayoutDesc.updateDescriptorSetLayout(DescriptorSetIndex::Internal, descriptorSetDesc);
if (pushConstantsSize)
{
pipelineLayoutDesc.updatePushConstantRange(pushConstantsShaderStage, 0,
static_cast<uint32_t>(pushConstantsSize));
}
ANGLE_TRY(contextVk->getPipelineLayoutCache().getPipelineLayout(contextVk, pipelineLayoutDesc,
mDescriptorSetLayouts[function],
&mPipelineLayouts[function]));
return angle::Result::Continue;
}
angle::Result UtilsVk::ensureConvertIndexResourcesInitialized(ContextVk *contextVk)
{
if (mPipelineLayouts[Function::ConvertIndexBuffer].valid())
{
return angle::Result::Continue;
}
VkDescriptorPoolSize setSizes[2] = {
{VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, 1},
{VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, 1},
};
return ensureResourcesInitialized(contextVk, Function::ConvertIndexBuffer, setSizes,
ArraySize(setSizes), sizeof(ConvertIndexShaderParams));
}
angle::Result UtilsVk::ensureConvertIndexIndirectResourcesInitialized(ContextVk *contextVk)
{
if (mPipelineLayouts[Function::ConvertIndexIndirectBuffer].valid())
{
return angle::Result::Continue;
}
VkDescriptorPoolSize setSizes[4] = {
{VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, 1}, // dst index buffer
{VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, 1}, // source index buffer
{VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, 1}, // src indirect buffer
{VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, 1}, // dst indirect buffer
};
return ensureResourcesInitialized(contextVk, Function::ConvertIndexIndirectBuffer, setSizes,
ArraySize(setSizes),
sizeof(ConvertIndexIndirectShaderParams));
}
angle::Result UtilsVk::ensureConvertIndexIndirectLineLoopResourcesInitialized(ContextVk *contextVk)
{
if (mPipelineLayouts[Function::ConvertIndexIndirectLineLoopBuffer].valid())
{
return angle::Result::Continue;
}
VkDescriptorPoolSize setSizes[4] = {
{VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, 1}, // cmd buffer
{VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, 1}, // dst cmd buffer
{VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, 1}, // source index buffer
{VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, 1}, // dst index buffer
};
return ensureResourcesInitialized(contextVk, Function::ConvertIndexIndirectLineLoopBuffer,
setSizes, ArraySize(setSizes),
sizeof(ConvertIndexIndirectLineLoopShaderParams));
}
angle::Result UtilsVk::ensureConvertIndirectLineLoopResourcesInitialized(ContextVk *contextVk)
{
if (mPipelineLayouts[Function::ConvertIndirectLineLoopBuffer].valid())
{
return angle::Result::Continue;
}
VkDescriptorPoolSize setSizes[3] = {
{VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, 1}, // cmd buffer
{VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, 1}, // dst cmd buffer
{VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, 1}, // dst index buffer
};
return ensureResourcesInitialized(contextVk, Function::ConvertIndirectLineLoopBuffer, setSizes,
ArraySize(setSizes),
sizeof(ConvertIndirectLineLoopShaderParams));
}
angle::Result UtilsVk::ensureConvertVertexResourcesInitialized(ContextVk *contextVk)
{
if (mPipelineLayouts[Function::ConvertVertexBuffer].valid())
{
return angle::Result::Continue;
}
VkDescriptorPoolSize setSizes[2] = {
{VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, 1},
{VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, 1},
};
return ensureResourcesInitialized(contextVk, Function::ConvertVertexBuffer, setSizes,
ArraySize(setSizes), sizeof(ConvertVertexShaderParams));
}
angle::Result UtilsVk::ensureImageClearResourcesInitialized(ContextVk *contextVk)
{
if (mPipelineLayouts[Function::ImageClear].valid())
{
return angle::Result::Continue;
}
// The shader does not use any descriptor sets.
return ensureResourcesInitialized(contextVk, Function::ImageClear, nullptr, 0,
sizeof(ImageClearShaderParams));
}
angle::Result UtilsVk::ensureImageCopyResourcesInitialized(ContextVk *contextVk)
{
if (mPipelineLayouts[Function::ImageCopy].valid())
{
return angle::Result::Continue;
}
VkDescriptorPoolSize setSizes[1] = {
{VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE, 1},
};
return ensureResourcesInitialized(contextVk, Function::ImageCopy, setSizes, ArraySize(setSizes),
sizeof(ImageCopyShaderParams));
}
angle::Result UtilsVk::ensureBlitResolveResourcesInitialized(ContextVk *contextVk)
{
if (!mPipelineLayouts[Function::BlitResolve].valid())
{
VkDescriptorPoolSize setSizes[3] = {
{VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE, 1},
{VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE, 1},
{VK_DESCRIPTOR_TYPE_SAMPLER, 1},
};
ANGLE_TRY(ensureResourcesInitialized(contextVk, Function::BlitResolve, setSizes,
ArraySize(setSizes), sizeof(BlitResolveShaderParams)));
}
return ensureSamplersInitialized(contextVk);
}
angle::Result UtilsVk::ensureBlitResolveStencilNoExportResourcesInitialized(ContextVk *contextVk)
{
if (!mPipelineLayouts[Function::BlitResolveStencilNoExport].valid())
{
VkDescriptorPoolSize setSizes[3] = {
{VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, 1},
{VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE, 1},
{VK_DESCRIPTOR_TYPE_SAMPLER, 1},
};
ANGLE_TRY(ensureResourcesInitialized(contextVk, Function::BlitResolveStencilNoExport,
setSizes, ArraySize(setSizes),
sizeof(BlitResolveStencilNoExportShaderParams)));
}
return ensureSamplersInitialized(contextVk);
}
angle::Result UtilsVk::ensureOverlayCullResourcesInitialized(ContextVk *contextVk)
{
if (mPipelineLayouts[Function::OverlayCull].valid())
{
return angle::Result::Continue;
}
VkDescriptorPoolSize setSizes[2] = {
{VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, 1},
{VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, 1},
};
return ensureResourcesInitialized(contextVk, Function::OverlayCull, setSizes,
ArraySize(setSizes), 0);
}
angle::Result UtilsVk::ensureOverlayDrawResourcesInitialized(ContextVk *contextVk)
{
if (!mPipelineLayouts[Function::OverlayDraw].valid())
{
VkDescriptorPoolSize setSizes[5] = {
{VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, 1}, {VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, 1},
{VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, 1}, {VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE, 1},
{VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE, 1},
};
ANGLE_TRY(ensureResourcesInitialized(contextVk, Function::OverlayDraw, setSizes,
ArraySize(setSizes), sizeof(OverlayDrawShaderParams)));
}
return ensureSamplersInitialized(contextVk);
}
angle::Result UtilsVk::ensureGenerateMipmapResourcesInitialized(ContextVk *contextVk)
{
if (mPipelineLayouts[Function::GenerateMipmap].valid())
{
return angle::Result::Continue;
}
VkDescriptorPoolSize setSizes[2] = {
{VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, GetGenerateMipmapMaxLevels(contextVk)},
{VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, 1},
};
return ensureResourcesInitialized(contextVk, Function::GenerateMipmap, setSizes,
ArraySize(setSizes), sizeof(GenerateMipmapShaderParams));
}
angle::Result UtilsVk::ensureUnresolveResourcesInitialized(ContextVk *contextVk,
Function function,
uint32_t attachmentCount)
{
ASSERT(static_cast<uint32_t>(function) -
static_cast<uint32_t>(Function::Unresolve1Attachment) ==
attachmentCount - 1);
if (mPipelineLayouts[function].valid())
{
return angle::Result::Continue;
}
vk::FramebufferAttachmentArray<VkDescriptorPoolSize> setSizes;
std::fill(setSizes.begin(), setSizes.end(),
VkDescriptorPoolSize{VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT, 1});
return ensureResourcesInitialized(contextVk, function, setSizes.data(), attachmentCount, 0);
}
angle::Result UtilsVk::ensureSamplersInitialized(ContextVk *contextVk)
{
VkSamplerCreateInfo samplerInfo = {};
samplerInfo.sType = VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO;
samplerInfo.flags = 0;
samplerInfo.magFilter = VK_FILTER_NEAREST;
samplerInfo.minFilter = VK_FILTER_NEAREST;
samplerInfo.mipmapMode = VK_SAMPLER_MIPMAP_MODE_NEAREST;
samplerInfo.addressModeU = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE;
samplerInfo.addressModeV = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE;
samplerInfo.addressModeW = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE;
samplerInfo.mipLodBias = 0.0f;
samplerInfo.anisotropyEnable = VK_FALSE;
samplerInfo.maxAnisotropy = 1;
samplerInfo.compareEnable = VK_FALSE;
samplerInfo.compareOp = VK_COMPARE_OP_ALWAYS;
samplerInfo.minLod = 0;
samplerInfo.maxLod = 0;
samplerInfo.borderColor = VK_BORDER_COLOR_INT_TRANSPARENT_BLACK;
samplerInfo.unnormalizedCoordinates = VK_FALSE;
if (!mPointSampler.valid())
{
ANGLE_VK_TRY(contextVk, mPointSampler.init(contextVk->getDevice(), samplerInfo));
}
samplerInfo.magFilter = VK_FILTER_LINEAR;
samplerInfo.minFilter = VK_FILTER_LINEAR;
if (!mLinearSampler.valid())
{
ANGLE_VK_TRY(contextVk, mLinearSampler.init(contextVk->getDevice(), samplerInfo));
}
return angle::Result::Continue;
}
angle::Result UtilsVk::setupComputeProgram(ContextVk *contextVk,
Function function,
vk::RefCounted<vk::ShaderAndSerial> *csShader,
vk::ShaderProgramHelper *program,
const VkDescriptorSet descriptorSet,
const void *pushConstants,
size_t pushConstantsSize,
vk::OutsideRenderPassCommandBuffer *commandBuffer)
{
ASSERT(function >= Function::ComputeStartIndex);
const vk::BindingPointer<vk::PipelineLayout> &pipelineLayout = mPipelineLayouts[function];
vk::PipelineHelper *pipeline;
program->setShader(gl::ShaderType::Compute, csShader);
ANGLE_TRY(program->getComputePipeline(contextVk, pipelineLayout.get(), &pipeline));
pipeline->retain(&contextVk->getResourceUseList());
commandBuffer->bindComputePipeline(pipeline->getPipeline());
contextVk->invalidateComputePipelineBinding();
if (descriptorSet != VK_NULL_HANDLE)
{
commandBuffer->bindDescriptorSets(pipelineLayout.get(), VK_PIPELINE_BIND_POINT_COMPUTE,
DescriptorSetIndex::Internal, 1, &descriptorSet, 0,
nullptr);
contextVk->invalidateComputeDescriptorSet(DescriptorSetIndex::Internal);
}
if (pushConstants)
{
commandBuffer->pushConstants(pipelineLayout.get(), VK_SHADER_STAGE_COMPUTE_BIT, 0,
static_cast<uint32_t>(pushConstantsSize), pushConstants);
}
return angle::Result::Continue;
}
angle::Result UtilsVk::setupGraphicsProgram(ContextVk *contextVk,
Function function,
vk::RefCounted<vk::ShaderAndSerial> *vsShader,
vk::RefCounted<vk::ShaderAndSerial> *fsShader,
vk::ShaderProgramHelper *program,
const vk::GraphicsPipelineDesc *pipelineDesc,
const VkDescriptorSet descriptorSet,
const void *pushConstants,
size_t pushConstantsSize,
vk::RenderPassCommandBuffer *commandBuffer)
{
RendererVk *renderer = contextVk->getRenderer();
ASSERT(function < Function::ComputeStartIndex);
const vk::BindingPointer<vk::PipelineLayout> &pipelineLayout = mPipelineLayouts[function];
program->setShader(gl::ShaderType::Vertex, vsShader);
if (fsShader)
{
program->setShader(gl::ShaderType::Fragment, fsShader);
}
// This value is not used but is passed to getGraphicsPipeline to avoid a nullptr check.
const vk::GraphicsPipelineDesc *descPtr;
vk::PipelineHelper *helper;
vk::PipelineCache *pipelineCache = nullptr;
ANGLE_TRY(renderer->getPipelineCache(&pipelineCache));
ANGLE_TRY(program->getGraphicsPipeline(contextVk, &contextVk->getRenderPassCache(),
*pipelineCache, pipelineLayout.get(), *pipelineDesc,
gl::AttributesMask(), gl::ComponentTypeMask(),
gl::DrawBufferMask(), &descPtr, &helper));
helper->retain(&contextVk->getResourceUseList());
commandBuffer->bindGraphicsPipeline(helper->getPipeline());
contextVk->invalidateGraphicsPipelineBinding();
if (descriptorSet != VK_NULL_HANDLE)
{
commandBuffer->bindDescriptorSets(pipelineLayout.get(), VK_PIPELINE_BIND_POINT_GRAPHICS,
DescriptorSetIndex::Internal, 1, &descriptorSet, 0,
nullptr);
contextVk->invalidateGraphicsDescriptorSet(DescriptorSetIndex::Internal);
}
if (pushConstants)
{
commandBuffer->pushConstants(pipelineLayout.get(), VK_SHADER_STAGE_FRAGMENT_BIT, 0,
static_cast<uint32_t>(pushConstantsSize), pushConstants);
}
return angle::Result::Continue;
}
angle::Result UtilsVk::convertIndexBuffer(ContextVk *contextVk,
vk::BufferHelper *dst,
vk::BufferHelper *src,
const ConvertIndexParameters &params)
{
ANGLE_TRY(ensureConvertIndexResourcesInitialized(contextVk));
vk::CommandBufferAccess access;
access.onBufferComputeShaderRead(src);
access.onBufferComputeShaderWrite(dst);
vk::OutsideRenderPassCommandBuffer *commandBuffer;
ANGLE_TRY(contextVk->getOutsideRenderPassCommandBuffer(access, &commandBuffer));
VkDescriptorSet descriptorSet;
vk::RefCountedDescriptorPoolBinding descriptorPoolBinding;
ANGLE_TRY(allocateDescriptorSet(contextVk, Function::ConvertIndexBuffer, &descriptorPoolBinding,
&descriptorSet));
std::array<VkDescriptorBufferInfo, 2> buffers = {{
{dst->getBuffer().getHandle(), dst->getOffset(), dst->getSize()},
{src->getBuffer().getHandle(), src->getOffset(), src->getSize()},
}};
VkWriteDescriptorSet writeInfo = {};
writeInfo.sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET;
writeInfo.dstSet = descriptorSet;
writeInfo.dstBinding = kConvertIndexDestinationBinding;
writeInfo.descriptorCount = 2;
writeInfo.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER;
writeInfo.pBufferInfo = buffers.data();
vkUpdateDescriptorSets(contextVk->getDevice(), 1, &writeInfo, 0, nullptr);
ConvertIndexShaderParams shaderParams = {params.srcOffset, params.dstOffset >> 2,
params.maxIndex, 0};
uint32_t flags = 0;
if (contextVk->getState().isPrimitiveRestartEnabled())
{
flags |= vk::InternalShader::ConvertIndex_comp::kIsPrimitiveRestartEnabled;
}
vk::RefCounted<vk::ShaderAndSerial> *shader = nullptr;
ANGLE_TRY(contextVk->getShaderLibrary().getConvertIndex_comp(contextVk, flags, &shader));
ANGLE_TRY(setupComputeProgram(contextVk, Function::ConvertIndexBuffer, shader,
&mConvertIndexPrograms[flags], descriptorSet, &shaderParams,
sizeof(ConvertIndexShaderParams), commandBuffer));
constexpr uint32_t kInvocationsPerGroup = 64;
constexpr uint32_t kInvocationsPerIndex = 2;
const uint32_t kIndexCount = params.maxIndex;
const uint32_t kGroupCount =
UnsignedCeilDivide(kIndexCount * kInvocationsPerIndex, kInvocationsPerGroup);
commandBuffer->dispatch(kGroupCount, 1, 1);
descriptorPoolBinding.reset();
return angle::Result::Continue;
}
angle::Result UtilsVk::convertIndexIndirectBuffer(ContextVk *contextVk,
vk::BufferHelper *srcIndirectBuf,
vk::BufferHelper *srcIndexBuf,
vk::BufferHelper *dstIndirectBuf,
vk::BufferHelper *dstIndexBuf,
const ConvertIndexIndirectParameters &params)
{
ANGLE_TRY(ensureConvertIndexIndirectResourcesInitialized(contextVk));
vk::CommandBufferAccess access;
access.onBufferComputeShaderRead(srcIndirectBuf);
access.onBufferComputeShaderRead(srcIndexBuf);
access.onBufferComputeShaderWrite(dstIndirectBuf);
access.onBufferComputeShaderWrite(dstIndexBuf);
vk::OutsideRenderPassCommandBuffer *commandBuffer;
ANGLE_TRY(contextVk->getOutsideRenderPassCommandBuffer(access, &commandBuffer));
VkDescriptorSet descriptorSet;
vk::RefCountedDescriptorPoolBinding descriptorPoolBinding;
ANGLE_TRY(allocateDescriptorSet(contextVk, Function::ConvertIndexIndirectBuffer,
&descriptorPoolBinding, &descriptorSet));
std::array<VkDescriptorBufferInfo, 4> buffers = {{
{dstIndexBuf->getBuffer().getHandle(), dstIndexBuf->getOffset(), dstIndexBuf->getSize()},
{srcIndexBuf->getBuffer().getHandle(), srcIndexBuf->getOffset(), srcIndexBuf->getSize()},
{srcIndirectBuf->getBuffer().getHandle(), srcIndirectBuf->getOffset(),
srcIndirectBuf->getSize()},
{dstIndirectBuf->getBuffer().getHandle(), dstIndirectBuf->getOffset(),
dstIndirectBuf->getSize()},
}};
VkWriteDescriptorSet writeInfo = {};
writeInfo.sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET;
writeInfo.dstSet = descriptorSet;
writeInfo.dstBinding = kConvertIndexDestinationBinding;
writeInfo.descriptorCount = 4;
writeInfo.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER;
writeInfo.pBufferInfo = buffers.data();
vkUpdateDescriptorSets(contextVk->getDevice(), 1, &writeInfo, 0, nullptr);
ConvertIndexIndirectShaderParams shaderParams = {
params.srcIndirectBufOffset >> 2, params.srcIndexBufOffset, params.dstIndexBufOffset >> 2,
params.maxIndex, params.dstIndirectBufOffset >> 2};
uint32_t flags = vk::InternalShader::ConvertIndex_comp::kIsIndirect;
if (contextVk->getState().isPrimitiveRestartEnabled())
{
flags |= vk::InternalShader::ConvertIndex_comp::kIsPrimitiveRestartEnabled;
}
vk::RefCounted<vk::ShaderAndSerial> *shader = nullptr;
ANGLE_TRY(contextVk->getShaderLibrary().getConvertIndex_comp(contextVk, flags, &shader));
ANGLE_TRY(setupComputeProgram(contextVk, Function::ConvertIndexIndirectBuffer, shader,
&mConvertIndexPrograms[flags], descriptorSet, &shaderParams,
sizeof(ConvertIndexIndirectShaderParams), commandBuffer));
constexpr uint32_t kInvocationsPerGroup = 64;
constexpr uint32_t kInvocationsPerIndex = 2;
const uint32_t kIndexCount = params.maxIndex;
const uint32_t kGroupCount =
UnsignedCeilDivide(kIndexCount * kInvocationsPerIndex, kInvocationsPerGroup);
commandBuffer->dispatch(kGroupCount, 1, 1);
descriptorPoolBinding.reset();
return angle::Result::Continue;
}
angle::Result UtilsVk::convertLineLoopIndexIndirectBuffer(
ContextVk *contextVk,
vk::BufferHelper *srcIndirectBuffer,
vk::BufferHelper *dstIndirectBuffer,
vk::BufferHelper *dstIndexBuffer,
vk::BufferHelper *srcIndexBuffer,
const ConvertLineLoopIndexIndirectParameters &params)
{
ANGLE_TRY(ensureConvertIndexIndirectLineLoopResourcesInitialized(contextVk));
vk::CommandBufferAccess access;
access.onBufferComputeShaderRead(srcIndirectBuffer);
access.onBufferComputeShaderRead(srcIndexBuffer);
access.onBufferComputeShaderWrite(dstIndirectBuffer);
access.onBufferComputeShaderWrite(dstIndexBuffer);
vk::OutsideRenderPassCommandBuffer *commandBuffer;
ANGLE_TRY(contextVk->getOutsideRenderPassCommandBuffer(access, &commandBuffer));
VkDescriptorSet descriptorSet;
vk::RefCountedDescriptorPoolBinding descriptorPoolBinding;
ANGLE_TRY(allocateDescriptorSet(contextVk, Function::ConvertIndexIndirectLineLoopBuffer,
&descriptorPoolBinding, &descriptorSet));
std::array<VkDescriptorBufferInfo, 4> buffers = {{
{dstIndexBuffer->getBuffer().getHandle(), dstIndexBuffer->getOffset(),
dstIndexBuffer->getSize()},
{srcIndexBuffer->getBuffer().getHandle(), srcIndexBuffer->getOffset(),
srcIndexBuffer->getSize()},
{srcIndirectBuffer->getBuffer().getHandle(), srcIndirectBuffer->getOffset(),
srcIndirectBuffer->getSize()},
{dstIndirectBuffer->getBuffer().getHandle(), dstIndirectBuffer->getOffset(),
dstIndirectBuffer->getSize()},
}};
VkWriteDescriptorSet writeInfo = {};
writeInfo.sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET;
writeInfo.dstSet = descriptorSet;
writeInfo.dstBinding = kConvertIndexDestinationBinding;
writeInfo.descriptorCount = 4;
writeInfo.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER;
writeInfo.pBufferInfo = buffers.data();
vkUpdateDescriptorSets(contextVk->getDevice(), 1, &writeInfo, 0, nullptr);
ConvertIndexIndirectLineLoopShaderParams shaderParams = {
params.indirectBufferOffset >> 2, params.dstIndirectBufferOffset >> 2,
params.srcIndexBufferOffset, params.dstIndexBufferOffset >> 2,
contextVk->getState().isPrimitiveRestartEnabled()};
uint32_t flags = GetConvertIndexIndirectLineLoopFlag(params.indicesBitsWidth);
vk::RefCounted<vk::ShaderAndSerial> *shader = nullptr;
ANGLE_TRY(contextVk->getShaderLibrary().getConvertIndexIndirectLineLoop_comp(contextVk, flags,
&shader));
ANGLE_TRY(setupComputeProgram(contextVk, Function::ConvertIndexIndirectLineLoopBuffer, shader,
&mConvertIndexIndirectLineLoopPrograms[flags], descriptorSet,
&shaderParams, sizeof(ConvertIndexIndirectLineLoopShaderParams),
commandBuffer));
commandBuffer->dispatch(1, 1, 1);
descriptorPoolBinding.reset();
return angle::Result::Continue;
}
angle::Result UtilsVk::convertLineLoopArrayIndirectBuffer(
ContextVk *contextVk,
vk::BufferHelper *srcIndirectBuffer,
vk::BufferHelper *dstIndirectBuffer,
vk::BufferHelper *dstIndexBuffer,
const ConvertLineLoopArrayIndirectParameters &params)
{
ANGLE_TRY(ensureConvertIndirectLineLoopResourcesInitialized(contextVk));
vk::CommandBufferAccess access;
access.onBufferComputeShaderRead(srcIndirectBuffer);
access.onBufferComputeShaderWrite(dstIndirectBuffer);
access.onBufferComputeShaderWrite(dstIndexBuffer);
vk::OutsideRenderPassCommandBuffer *commandBuffer;
ANGLE_TRY(contextVk->getOutsideRenderPassCommandBuffer(access, &commandBuffer));
VkDescriptorSet descriptorSet;
vk::RefCountedDescriptorPoolBinding descriptorPoolBinding;
ANGLE_TRY(allocateDescriptorSet(contextVk, Function::ConvertIndirectLineLoopBuffer,
&descriptorPoolBinding, &descriptorSet));
std::array<VkDescriptorBufferInfo, 3> buffers = {{
{srcIndirectBuffer->getBuffer().getHandle(), srcIndirectBuffer->getOffset(),
srcIndirectBuffer->getSize()},
{dstIndirectBuffer->getBuffer().getHandle(), dstIndirectBuffer->getOffset(),
dstIndirectBuffer->getSize()},
{dstIndexBuffer->getBuffer().getHandle(), dstIndexBuffer->getOffset(),
dstIndexBuffer->getSize()},
}};
VkWriteDescriptorSet writeInfo = {};
writeInfo.sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET;
writeInfo.dstSet = descriptorSet;
writeInfo.dstBinding = kConvertIndexDestinationBinding;
writeInfo.descriptorCount = 3;
writeInfo.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER;
writeInfo.pBufferInfo = buffers.data();
vkUpdateDescriptorSets(contextVk->getDevice(), 1, &writeInfo, 0, nullptr);
ConvertIndirectLineLoopShaderParams shaderParams = {params.indirectBufferOffset >> 2,
params.dstIndirectBufferOffset >> 2,
params.dstIndexBufferOffset >> 2};
uint32_t flags = 0;
vk::RefCounted<vk::ShaderAndSerial> *shader = nullptr;
ANGLE_TRY(
contextVk->getShaderLibrary().getConvertIndirectLineLoop_comp(contextVk, flags, &shader));
ANGLE_TRY(setupComputeProgram(contextVk, Function::ConvertIndirectLineLoopBuffer, shader,
&mConvertIndirectLineLoopPrograms[flags], descriptorSet,
&shaderParams, sizeof(ConvertIndirectLineLoopShaderParams),
commandBuffer));
commandBuffer->dispatch(1, 1, 1);
descriptorPoolBinding.reset();
return angle::Result::Continue;
}
angle::Result UtilsVk::convertVertexBuffer(ContextVk *contextVk,
vk::BufferHelper *dst,
vk::BufferHelper *src,
const ConvertVertexParameters &params)
{
vk::CommandBufferAccess access;
access.onBufferComputeShaderRead(src);
access.onBufferComputeShaderWrite(dst);
vk::OutsideRenderPassCommandBuffer *commandBuffer;
ANGLE_TRY(contextVk->getOutsideRenderPassCommandBuffer(access, &commandBuffer));
ConvertVertexShaderParams shaderParams;
shaderParams.Ns = params.srcFormat->channelCount;
shaderParams.Bs = params.srcFormat->pixelBytes / params.srcFormat->channelCount;
shaderParams.Ss = static_cast<uint32_t>(params.srcStride);
shaderParams.Nd = params.dstFormat->channelCount;
shaderParams.Bd = params.dstFormat->pixelBytes / params.dstFormat->channelCount;
shaderParams.Sd = shaderParams.Nd * shaderParams.Bd;
// The component size is expected to either be 1, 2 or 4 bytes.
ASSERT(4 % shaderParams.Bs == 0);
ASSERT(4 % shaderParams.Bd == 0);
shaderParams.Es = 4 / shaderParams.Bs;
shaderParams.Ed = 4 / shaderParams.Bd;
// Total number of output components is simply the number of vertices by number of components in
// each.
shaderParams.componentCount = static_cast<uint32_t>(params.vertexCount * shaderParams.Nd);
// Total number of 4-byte outputs is the number of components divided by how many components can
// fit in a 4-byte value. Note that this value is also the invocation size of the shader.
shaderParams.outputCount = UnsignedCeilDivide(shaderParams.componentCount, shaderParams.Ed);
shaderParams.srcOffset = static_cast<uint32_t>(params.srcOffset);
shaderParams.dstOffset = static_cast<uint32_t>(params.dstOffset);
bool isSrcA2BGR10 =
params.srcFormat->vertexAttribType == gl::VertexAttribType::UnsignedInt2101010 ||
params.srcFormat->vertexAttribType == gl::VertexAttribType::Int2101010;
bool isSrcRGB10A2 =
params.srcFormat->vertexAttribType == gl::VertexAttribType::UnsignedInt1010102 ||
params.srcFormat->vertexAttribType == gl::VertexAttribType::Int1010102;
shaderParams.isSrcHDR = isSrcA2BGR10 || isSrcRGB10A2;
shaderParams.isSrcA2BGR10 = isSrcA2BGR10;
uint32_t flags = GetConvertVertexFlags(params);
// See GLES3.0 section 2.9.1 Transferring Array Elements
const uint32_t srcValueBits = shaderParams.isSrcHDR ? 2 : shaderParams.Bs * 8;
const uint32_t srcValueMask =
srcValueBits == 32 ? 0xFFFFFFFFu : angle::BitMask<uint32_t>(srcValueBits);
switch (flags)
{
case ConvertVertex_comp::kSintToSint:
case ConvertVertex_comp::kSintToFloat:
case ConvertVertex_comp::kUintToFloat:
// For integers, alpha should take a value of 1.
shaderParams.srcEmulatedAlpha = 1;
break;
case ConvertVertex_comp::kUintToUint:
// For integers, alpha should take a value of 1. However, uint->uint is also used to
// add channels to RGB snorm, unorm and half formats.
if (params.dstFormat->isSnorm())
{
// See case ConvertVertex_comp::kSnormToFloat below.
shaderParams.srcEmulatedAlpha = srcValueMask >> 1;
}
else if (params.dstFormat->isUnorm())
{
// See case ConvertVertex_comp::kUnormToFloat below.
shaderParams.srcEmulatedAlpha = srcValueMask;
}
else if (params.dstFormat->isVertexTypeHalfFloat())
{
shaderParams.srcEmulatedAlpha = gl::Float16One;
}
else
{
shaderParams.srcEmulatedAlpha = 1;
}
break;
case ConvertVertex_comp::kSnormToFloat:
// The largest signed number with as many bits as the alpha channel of the source is
// 0b011...1 which is srcValueMask >> 1
shaderParams.srcEmulatedAlpha = srcValueMask >> 1;
break;
case ConvertVertex_comp::kUnormToFloat:
// The largest unsigned number with as many bits as the alpha channel of the source is
// 0b11...1 which is srcValueMask
shaderParams.srcEmulatedAlpha = srcValueMask;
break;
case ConvertVertex_comp::kFixedToFloat:
// 1.0 in fixed point is 0x10000
shaderParams.srcEmulatedAlpha = 0x10000;
break;
case ConvertVertex_comp::kFloatToFloat:
ASSERT(ValidateFloatOneAsUint());
shaderParams.srcEmulatedAlpha = gl::Float32One;
break;
default:
UNREACHABLE();
}
return convertVertexBufferImpl(contextVk, dst, src, flags, commandBuffer, shaderParams);
}
angle::Result UtilsVk::convertVertexBufferImpl(ContextVk *contextVk,
vk::BufferHelper *dst,
vk::BufferHelper *src,
uint32_t flags,
vk::OutsideRenderPassCommandBuffer *commandBuffer,
const ConvertVertexShaderParams &shaderParams)
{
ANGLE_TRY(ensureConvertVertexResourcesInitialized(contextVk));
VkDescriptorSet descriptorSet;
vk::RefCountedDescriptorPoolBinding descriptorPoolBinding;
ANGLE_TRY(allocateDescriptorSet(contextVk, Function::ConvertVertexBuffer,
&descriptorPoolBinding, &descriptorSet));
VkWriteDescriptorSet writeInfo = {};
VkDescriptorBufferInfo buffers[2] = {
{dst->getBuffer().getHandle(), dst->getOffset(), dst->getSize()},
{src->getBuffer().getHandle(), src->getOffset(), src->getSize()},
};
static_assert(kConvertVertexDestinationBinding + 1 == kConvertVertexSourceBinding,
"Update write info");
writeInfo.sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET;
writeInfo.dstSet = descriptorSet;
writeInfo.dstBinding = kConvertVertexDestinationBinding;
writeInfo.descriptorCount = 2;
writeInfo.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER;
writeInfo.pBufferInfo = buffers;
vkUpdateDescriptorSets(contextVk->getDevice(), 1, &writeInfo, 0, nullptr);
vk::RefCounted<vk::ShaderAndSerial> *shader = nullptr;
ANGLE_TRY(contextVk->getShaderLibrary().getConvertVertex_comp(contextVk, flags, &shader));
ANGLE_TRY(setupComputeProgram(contextVk, Function::ConvertVertexBuffer, shader,
&mConvertVertexPrograms[flags], descriptorSet, &shaderParams,
sizeof(shaderParams), commandBuffer));
commandBuffer->dispatch(UnsignedCeilDivide(shaderParams.outputCount, 64), 1, 1);
descriptorPoolBinding.reset();
return angle::Result::Continue;
}
angle::Result UtilsVk::startRenderPass(ContextVk *contextVk,
vk::ImageHelper *image,
const vk::ImageView *imageView,
const vk::RenderPassDesc &renderPassDesc,
const gl::Rectangle &renderArea,
vk::RenderPassCommandBuffer **commandBufferOut)
{
vk::RenderPass *compatibleRenderPass = nullptr;
ANGLE_TRY(contextVk->getCompatibleRenderPass(renderPassDesc, &compatibleRenderPass));
VkFramebufferCreateInfo framebufferInfo = {};
// Minimize the framebuffer coverage to only cover up to the render area.
framebufferInfo.sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO;
framebufferInfo.flags = 0;
framebufferInfo.renderPass = compatibleRenderPass->getHandle();
framebufferInfo.attachmentCount = 1;
framebufferInfo.pAttachments = imageView->ptr();
framebufferInfo.width = renderArea.x + renderArea.width;
framebufferInfo.height = renderArea.y + renderArea.height;
framebufferInfo.layers = 1;
vk::Framebuffer framebuffer;
ANGLE_VK_TRY(contextVk, framebuffer.init(contextVk->getDevice(), framebufferInfo));
vk::AttachmentOpsArray renderPassAttachmentOps;
vk::PackedClearValuesArray clearValues;
clearValues.store(vk::kAttachmentIndexZero, VK_IMAGE_ASPECT_COLOR_BIT, {});
renderPassAttachmentOps.initWithLoadStore(vk::kAttachmentIndexZero,
vk::ImageLayout::ColorAttachment,
vk::ImageLayout::ColorAttachment);
ANGLE_TRY(contextVk->beginNewRenderPass(
framebuffer, renderArea, renderPassDesc, renderPassAttachmentOps,
vk::PackedAttachmentCount(1), vk::kAttachmentIndexInvalid, clearValues, commandBufferOut));
contextVk->addGarbage(&framebuffer);
return angle::Result::Continue;
}
angle::Result UtilsVk::clearFramebuffer(ContextVk *contextVk,
FramebufferVk *framebuffer,
const ClearFramebufferParameters &params)
{
ANGLE_TRY(ensureImageClearResourcesInitialized(contextVk));
const gl::Rectangle &scissoredRenderArea = params.clearArea;
vk::Framebuffer *currentFramebuffer = nullptr;
vk::RenderPassCommandBuffer *commandBuffer;
// Start a new render pass if not already started
ANGLE_TRY(framebuffer->getFramebuffer(contextVk, &currentFramebuffer, nullptr));
if (contextVk->hasStartedRenderPassWithFramebuffer(currentFramebuffer))
{
commandBuffer = &contextVk->getStartedRenderPassCommands().getCommandBuffer();
}
else
{
ANGLE_TRY(contextVk->startRenderPass(scissoredRenderArea, &commandBuffer, nullptr));
}
if (params.clearStencil || params.clearDepth)
{
vk::RenderPassCommandBufferHelper *renderpassCommands;
renderpassCommands = &contextVk->getStartedRenderPassCommands();
// Because clear is not affected by depth/stencil test, we have to explicitly mark
// depth/stencil write here.
if (params.clearDepth)
{
renderpassCommands->onDepthAccess(vk::ResourceAccess::Write);
}
if (params.clearStencil)
{
renderpassCommands->onStencilAccess(vk::ResourceAccess::Write);
}
// We may have changed depth stencil access mode, so update read only depth stencil mode
// here.
framebuffer->updateRenderPassReadOnlyDepthMode(contextVk, renderpassCommands);
}
ImageClearShaderParams shaderParams;
shaderParams.clearValue = params.colorClearValue;
shaderParams.clearDepth = params.depthStencilClearValue.depth;
vk::GraphicsPipelineDesc pipelineDesc;
pipelineDesc.initDefaults(contextVk);
pipelineDesc.setCullMode(VK_CULL_MODE_NONE);
pipelineDesc.setColorWriteMasks(0, gl::DrawBufferMask(), gl::DrawBufferMask());
pipelineDesc.setSingleColorWriteMask(params.colorAttachmentIndexGL, params.colorMaskFlags);
pipelineDesc.setRasterizationSamples(framebuffer->getSamples());
pipelineDesc.setRenderPassDesc(framebuffer->getRenderPassDesc());
// Note: depth test is disabled by default so this should be unnecessary, but works around an
// Intel bug on windows. http://anglebug.com/3348
pipelineDesc.setDepthWriteEnabled(false);
// Clears can be done on a currently open render pass, so make sure the correct subpass index is
// used.
pipelineDesc.setSubpass(contextVk->getCurrentSubpassIndex());
// Clear depth by enabling depth clamping and setting the viewport depth range to the clear
// value if possible. Otherwise use the shader to export depth.
const bool supportsDepthClamp =
contextVk->getRenderer()->getPhysicalDeviceFeatures().depthClamp == VK_TRUE;
if (params.clearDepth)
{
pipelineDesc.setDepthTestEnabled(true);
pipelineDesc.setDepthWriteEnabled(true);
pipelineDesc.setDepthFunc(VK_COMPARE_OP_ALWAYS);
if (supportsDepthClamp)
{
// Note: this path requires the depthClamp Vulkan feature.
pipelineDesc.setDepthClampEnabled(true);
}
}
// Clear stencil by enabling stencil write with the right mask.
if (params.clearStencil)
{
const uint8_t compareMask = 0xFF;
const uint8_t clearStencilValue =
static_cast<uint8_t>(params.depthStencilClearValue.stencil);
pipelineDesc.setStencilTestEnabled(true);
pipelineDesc.setStencilFrontFuncs(clearStencilValue, VK_COMPARE_OP_ALWAYS, compareMask);
pipelineDesc.setStencilBackFuncs(clearStencilValue, VK_COMPARE_OP_ALWAYS, compareMask);
pipelineDesc.setStencilFrontOps(VK_STENCIL_OP_REPLACE, VK_STENCIL_OP_REPLACE,
VK_STENCIL_OP_REPLACE);
pipelineDesc.setStencilBackOps(VK_STENCIL_OP_REPLACE, VK_STENCIL_OP_REPLACE,
VK_STENCIL_OP_REPLACE);
pipelineDesc.setStencilFrontWriteMask(params.stencilMask);
pipelineDesc.setStencilBackWriteMask(params.stencilMask);
}
VkViewport viewport;
gl::Rectangle completeRenderArea = framebuffer->getRotatedCompleteRenderArea(contextVk);
bool invertViewport = contextVk->isViewportFlipEnabledForDrawFBO();
bool clipSpaceOriginUpperLeft =
contextVk->getState().getClipSpaceOrigin() == gl::ClipSpaceOrigin::UpperLeft;
// Set depth range to clear value. If clearing depth, the vertex shader depth output is clamped
// to this value, thus clearing the depth buffer to the desired clear value.
const float clearDepthValue = params.depthStencilClearValue.depth;
gl_vk::GetViewport(completeRenderArea, clearDepthValue, clearDepthValue, invertViewport,
clipSpaceOriginUpperLeft, completeRenderArea.height, &viewport);
commandBuffer->setViewport(0, 1, &viewport);
const VkRect2D scissor = gl_vk::GetRect(params.clearArea);
commandBuffer->setScissor(0, 1, &scissor);
contextVk->invalidateViewportAndScissor();
vk::ShaderLibrary &shaderLibrary = contextVk->getShaderLibrary();
vk::RefCounted<vk::ShaderAndSerial> *vertexShader = nullptr;
vk::RefCounted<vk::ShaderAndSerial> *fragmentShader = nullptr;
vk::ShaderProgramHelper *imageClearProgram = &mImageClearProgramVSOnly;
ANGLE_TRY(shaderLibrary.getFullScreenTri_vert(contextVk, 0, &vertexShader));
if (params.clearColor)
{
const uint32_t flags =
GetImageClearFlags(*params.colorFormat, params.colorAttachmentIndexGL,
params.clearDepth && !supportsDepthClamp);
ANGLE_TRY(shaderLibrary.getImageClear_frag(contextVk, flags, &fragmentShader));
imageClearProgram = &mImageClearPrograms[flags];
}
// Make sure transform feedback is paused. Needs to be done before binding the pipeline as
// that's not allowed in Vulkan.
const bool isTransformFeedbackActiveUnpaused =
contextVk->getStartedRenderPassCommands().isTransformFeedbackActiveUnpaused();
contextVk->pauseTransformFeedbackIfActiveUnpaused();
ANGLE_TRY(setupGraphicsProgram(contextVk, Function::ImageClear, vertexShader, fragmentShader,
imageClearProgram, &pipelineDesc, VK_NULL_HANDLE, &shaderParams,
sizeof(shaderParams), commandBuffer));
// Make sure this draw call doesn't count towards occlusion query results.
contextVk->pauseRenderPassQueriesIfActive();
commandBuffer->draw(3, 0);
ANGLE_TRY(contextVk->resumeRenderPassQueriesIfActive());
// If transform feedback was active, we can't pause and resume it in the same render pass
// because we can't insert a memory barrier for the counter buffers. In that case, break the
// render pass.
if (isTransformFeedbackActiveUnpaused)
{
ANGLE_TRY(contextVk->flushCommandsAndEndRenderPass(
RenderPassClosureReason::XfbResumeAfterDrawBasedClear));
}
return angle::Result::Continue;
}
angle::Result UtilsVk::clearImage(ContextVk *contextVk,
vk::ImageHelper *dst,
const ClearImageParameters &params)
{
ANGLE_TRY(ensureImageClearResourcesInitialized(contextVk));
const angle::Format &dstActualFormat = dst->getActualFormat();
// Currently, this function is only used to clear emulated channels of color images.
ASSERT(!dstActualFormat.hasDepthOrStencilBits());
// TODO: currently this function is only implemented for images that are drawable. If needed,
// for images that are not drawable, the following algorithm can be used.
//
// - Copy image to temp buffer
// - Use convertVertexBufferImpl to overwrite the alpha channel
// - Copy the result back to the image
//
// Note that the following check is not enough; if the image is AHB-imported, then the draw path
// cannot be taken if AHARDWAREBUFFER_USAGE_GPU_FRAMEBUFFER hasn't been specified, even if the
// format is renderable.
//
// http://anglebug.com/6151
if (!vk::FormatHasNecessaryFeature(contextVk->getRenderer(), dstActualFormat.id,
dst->getTilingMode(),
VK_FORMAT_FEATURE_COLOR_ATTACHMENT_BIT))
{
UNIMPLEMENTED();
return angle::Result::Continue;
}
vk::DeviceScoped<vk::ImageView> destView(contextVk->getDevice());
const gl::TextureType destViewType = vk::Get2DTextureType(1, dst->getSamples());
ANGLE_TRY(dst->initLayerImageView(contextVk, destViewType, VK_IMAGE_ASPECT_COLOR_BIT,
gl::SwizzleState(), &destView.get(), params.dstMip, 1,
params.dstLayer, 1, gl::SrgbWriteControlMode::Default));
const gl::Rectangle &renderArea = params.clearArea;
ImageClearShaderParams shaderParams;
shaderParams.clearValue = params.colorClearValue;
shaderParams.clearDepth = 0;
vk::RenderPassDesc renderPassDesc;
renderPassDesc.setSamples(dst->getSamples());
renderPassDesc.packColorAttachment(0, dstActualFormat.id);
vk::GraphicsPipelineDesc pipelineDesc;
pipelineDesc.initDefaults(contextVk);
pipelineDesc.setCullMode(VK_CULL_MODE_NONE);
pipelineDesc.setSingleColorWriteMask(0, params.colorMaskFlags);
pipelineDesc.setRasterizationSamples(dst->getSamples());
pipelineDesc.setRenderPassDesc(renderPassDesc);
vk::RenderPassCommandBuffer *commandBuffer;
ANGLE_TRY(startRenderPass(contextVk, dst, &destView.get(), renderPassDesc, renderArea,
&commandBuffer));
VkViewport viewport;
gl_vk::GetViewport(renderArea, 0.0f, 1.0f, false, false, dst->getExtents().height, &viewport);
commandBuffer->setViewport(0, 1, &viewport);
VkRect2D scissor = gl_vk::GetRect(renderArea);
commandBuffer->setScissor(0, 1, &scissor);
contextVk->invalidateViewportAndScissor();
contextVk->onImageRenderPassWrite(dst->toGLLevel(params.dstMip), params.dstLayer, 1,
VK_IMAGE_ASPECT_COLOR_BIT, vk::ImageLayout::ColorAttachment,
dst);
const uint32_t flags = GetImageClearFlags(dstActualFormat, 0, false);
vk::ShaderLibrary &shaderLibrary = contextVk->getShaderLibrary();
vk::RefCounted<vk::ShaderAndSerial> *vertexShader = nullptr;
vk::RefCounted<vk::ShaderAndSerial> *fragmentShader = nullptr;
ANGLE_TRY(shaderLibrary.getFullScreenTri_vert(contextVk, 0, &vertexShader));
ANGLE_TRY(shaderLibrary.getImageClear_frag(contextVk, flags, &fragmentShader));
ANGLE_TRY(setupGraphicsProgram(contextVk, Function::ImageClear, vertexShader, fragmentShader,
&mImageClearPrograms[flags], &pipelineDesc, VK_NULL_HANDLE,
&shaderParams, sizeof(shaderParams), commandBuffer));
// Note: this utility creates its own framebuffer, thus bypassing ContextVk::startRenderPass.
// As such, occlusion queries are not enabled.
commandBuffer->draw(3, 0);
vk::ImageView destViewObject = destView.release();
contextVk->addGarbage(&destViewObject);
// Close the render pass for this temporary framebuffer.
return contextVk->flushCommandsAndEndRenderPass(
RenderPassClosureReason::TemporaryForImageClear);
}
angle::Result UtilsVk::colorBlitResolve(ContextVk *contextVk,
FramebufferVk *framebuffer,
vk::ImageHelper *src,
const vk::ImageView *srcView,
const BlitResolveParameters &params)
{
// The views passed to this function are already retained, so a render pass cannot be already
// open. Otherwise, this function closes the render pass, which may incur a vkQueueSubmit and
// then the views are used in a new command buffer without having been retained for it.
// http://crbug.com/1272266#c22
//
// Note that depth/stencil views for blit are not derived from a ResourceVk object and are
// retained differently.
ASSERT(!contextVk->hasStartedRenderPass());
return blitResolveImpl(contextVk, framebuffer, src, srcView, nullptr, nullptr, params);
}
angle::Result UtilsVk::depthStencilBlitResolve(ContextVk *contextVk,
FramebufferVk *framebuffer,
vk::ImageHelper *src,
const vk::ImageView *srcDepthView,
const vk::ImageView *srcStencilView,
const BlitResolveParameters &params)
{
return blitResolveImpl(contextVk, framebuffer, src, nullptr, srcDepthView, srcStencilView,
params);
}
angle::Result UtilsVk::blitResolveImpl(ContextVk *contextVk,
FramebufferVk *framebuffer,
vk::ImageHelper *src,
const vk::ImageView *srcColorView,
const vk::ImageView *srcDepthView,
const vk::ImageView *srcStencilView,
const BlitResolveParameters &params)
{
// Possible ways to resolve color are:
//
// - vkCmdResolveImage: This is by far the easiest method, but lacks the ability to flip
// images during resolve.
// - Manual resolve: A shader can read all samples from input, average them and output.
// - Using subpass resolve attachment: A shader can transform the sample colors from source to
// destination coordinates and the subpass resolve would finish the job.
//
// The first method is unable to handle flipping, so it's not generally applicable. The last
// method would have been great were we able to modify the last render pass that rendered into
// source, but still wouldn't be able to handle flipping. The second method is implemented in
// this function for complete control.
// Possible ways to resolve depth/stencil are:
//
// - Manual resolve: A shader can read a samples from input and choose that for output.
// - Using subpass resolve attachment through VkSubpassDescriptionDepthStencilResolveKHR: This
// requires an extension that's not very well supported.
//
// The first method is implemented in this function.
// Possible ways to blit color, depth or stencil are:
//
// - vkCmdBlitImage: This function works if the source and destination formats have the blit
// feature.
// - Manual blit: A shader can sample from the source image and write it to the destination.
//
// The first method has a serious shortcoming. GLES allows blit parameters to exceed the
// source or destination boundaries. The actual blit is clipped to these limits, but the
// scaling applied is determined solely by the input areas. Vulkan requires the blit parameters
// to be within the source and destination bounds. This makes it hard to keep the scaling
// constant.
//
// The second method is implemented in this function, which shares code with the resolve method.
ANGLE_TRY(ensureBlitResolveResourcesInitialized(contextVk));
bool isResolve = src->getSamples() > 1;
BlitResolveShaderParams shaderParams;
// Note: adjustments made for pre-rotatation in FramebufferVk::blit() affect these
// Calculate*Offset() functions.
if (isResolve)
{
CalculateResolveOffset(params, shaderParams.offset.resolve);
}
else
{
CalculateBlitOffset(params, shaderParams.offset.blit);
}
shaderParams.stretch[0] = params.stretch[0];
shaderParams.stretch[1] = params.stretch[1];
shaderParams.invSrcExtent[0] = 1.0f / params.srcExtents[0];
shaderParams.invSrcExtent[1] = 1.0f / params.srcExtents[1];
shaderParams.srcLayer = params.srcLayer;
shaderParams.samples = src->getSamples();
shaderParams.invSamples = 1.0f / shaderParams.samples;
shaderParams.outputMask =
static_cast<uint32_t>(framebuffer->getState().getEnabledDrawBuffers().to_ulong());
shaderParams.flipX = params.flipX;
shaderParams.flipY = params.flipY;
shaderParams.rotateXY = 0;
// Potentially make adjustments for pre-rotation. Depending on the angle some of the
// shaderParams need to be adjusted.
switch (params.rotation)
{
case SurfaceRotation::Identity:
break;
case SurfaceRotation::Rotated90Degrees:
shaderParams.rotateXY = 1;
break;
case SurfaceRotation::Rotated180Degrees:
if (isResolve)
{
shaderParams.offset.resolve[0] += params.rotatedOffsetFactor[0];
shaderParams.offset.resolve[1] += params.rotatedOffsetFactor[1];
}
else
{
shaderParams.offset.blit[0] += params.rotatedOffsetFactor[0];
shaderParams.offset.blit[1] += params.rotatedOffsetFactor[1];
}
break;
case SurfaceRotation::Rotated270Degrees:
if (isResolve)
{
shaderParams.offset.resolve[0] += params.rotatedOffsetFactor[0];
shaderParams.offset.resolve[1] += params.rotatedOffsetFactor[1];
}
else
{
shaderParams.offset.blit[0] += params.rotatedOffsetFactor[0];
shaderParams.offset.blit[1] += params.rotatedOffsetFactor[1];
}
shaderParams.rotateXY = 1;
break;
default:
UNREACHABLE();
break;
}
bool blitColor = srcColorView != nullptr;
bool blitDepth = srcDepthView != nullptr;
bool blitStencil = srcStencilView != nullptr;
// Either color is blitted/resolved or depth/stencil, but not both.
ASSERT(blitColor != (blitDepth || blitStencil));
// Linear sampling is only valid with color blitting.
ASSERT((blitColor && !isResolve) || !params.linear);
uint32_t flags =
GetBlitResolveFlags(blitColor, blitDepth, blitStencil, src->getIntendedFormat());
flags |= src->getLayerCount() > 1 ? BlitResolve_frag::kSrcIsArray : 0;
flags |= isResolve ? BlitResolve_frag::kIsResolve : 0;
VkDescriptorSet descriptorSet;
vk::RefCountedDescriptorPoolBinding descriptorPoolBinding;
ANGLE_TRY(allocateDescriptorSet(contextVk, Function::BlitResolve, &descriptorPoolBinding,
&descriptorSet));
constexpr VkColorComponentFlags kAllColorComponents =
VK_COLOR_COMPONENT_R_BIT | VK_COLOR_COMPONENT_G_BIT | VK_COLOR_COMPONENT_B_BIT |
VK_COLOR_COMPONENT_A_BIT;
vk::GraphicsPipelineDesc pipelineDesc;
pipelineDesc.initDefaults(contextVk);
if (blitColor)
{
pipelineDesc.setColorWriteMasks(
gl::BlendStateExt::ColorMaskStorage::GetReplicatedValue(
kAllColorComponents, gl::BlendStateExt::ColorMaskStorage::GetMask(
framebuffer->getRenderPassDesc().colorAttachmentRange())),
framebuffer->getEmulatedAlphaAttachmentMask(), ~gl::DrawBufferMask());
}
else
{
pipelineDesc.setColorWriteMasks(0, gl::DrawBufferMask(), gl::DrawBufferMask());
}
pipelineDesc.setCullMode(VK_CULL_MODE_NONE);
pipelineDesc.setRenderPassDesc(framebuffer->getRenderPassDesc());
pipelineDesc.setDepthTestEnabled(blitDepth);
pipelineDesc.setDepthWriteEnabled(blitDepth);
pipelineDesc.setDepthFunc(VK_COMPARE_OP_ALWAYS);
if (blitStencil)
{
SetStencilForShaderExport(contextVk, &pipelineDesc);
}
vk::RenderPassCommandBuffer *commandBuffer;
ANGLE_TRY(framebuffer->startNewRenderPass(contextVk, params.blitArea, &commandBuffer, nullptr));
VkViewport viewport;
gl::Rectangle completeRenderArea = framebuffer->getRotatedCompleteRenderArea(contextVk);
gl_vk::GetViewport(completeRenderArea, 0.0f, 1.0f, false, false, completeRenderArea.height,
&viewport);
commandBuffer->setViewport(0, 1, &viewport);
VkRect2D scissor = gl_vk::GetRect(params.blitArea);
commandBuffer->setScissor(0, 1, &scissor);
contextVk->invalidateViewportAndScissor();
contextVk->onImageRenderPassRead(src->getAspectFlags(), vk::ImageLayout::FragmentShaderReadOnly,
src);
vk::RenderPassCommandBufferHelper *renderPassCommands =
&contextVk->getStartedRenderPassCommands();
if (blitDepth)
{
// Explicitly mark a depth write because we are modifying the depth buffer.
renderPassCommands->onDepthAccess(vk::ResourceAccess::Write);
}
if (blitStencil)
{
// Explicitly mark a stencil write because we are modifying the stencil buffer.
renderPassCommands->onStencilAccess(vk::ResourceAccess::Write);
}
if (blitDepth || blitStencil)
{
// Because we may have changed the depth stencil access mode, update read only depth mode
// now.
framebuffer->updateRenderPassReadOnlyDepthMode(contextVk, renderPassCommands);
}
VkDescriptorImageInfo imageInfos[2] = {};
if (blitColor)
{
imageInfos[0].imageView = srcColorView->getHandle();
imageInfos[0].imageLayout = src->getCurrentLayout();
}
if (blitDepth)
{
imageInfos[0].imageView = srcDepthView->getHandle();
imageInfos[0].imageLayout = src->getCurrentLayout();
}
if (blitStencil)
{
imageInfos[1].imageView = srcStencilView->getHandle();
imageInfos[1].imageLayout = src->getCurrentLayout();
}
VkDescriptorImageInfo samplerInfo = {};
samplerInfo.sampler = params.linear ? mLinearSampler.getHandle() : mPointSampler.getHandle();
VkWriteDescriptorSet writeInfos[3] = {};
writeInfos[0].sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET;
writeInfos[0].dstSet = descriptorSet;
writeInfos[0].dstBinding = kBlitResolveColorOrDepthBinding;
writeInfos[0].descriptorCount = 1;
writeInfos[0].descriptorType = VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE;
writeInfos[0].pImageInfo = &imageInfos[0];
writeInfos[1] = writeInfos[0];
writeInfos[1].dstBinding = kBlitResolveStencilBinding;
writeInfos[1].pImageInfo = &imageInfos[1];
writeInfos[2].sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET;
writeInfos[2].dstSet = descriptorSet;
writeInfos[2].dstBinding = kBlitResolveSamplerBinding;
writeInfos[2].descriptorCount = 1;
writeInfos[2].descriptorType = VK_DESCRIPTOR_TYPE_SAMPLER;
writeInfos[2].pImageInfo = &samplerInfo;
// If resolving color, there's one write info; index 0
// If resolving depth, write info index 0 must be written
// If resolving stencil, write info index 1 must also be written
//
// Note again that resolving color and depth/stencil are mutually exclusive here.
uint32_t writeInfoOffset = blitDepth || blitColor ? 0 : 1;
uint32_t writeInfoCount = blitColor + blitDepth + blitStencil;
vkUpdateDescriptorSets(contextVk->getDevice(), writeInfoCount, writeInfos + writeInfoOffset, 0,
nullptr);
vkUpdateDescriptorSets(contextVk->getDevice(), 1, &writeInfos[2], 0, nullptr);
vk::ShaderLibrary &shaderLibrary = contextVk->getShaderLibrary();
vk::RefCounted<vk::ShaderAndSerial> *vertexShader = nullptr;
vk::RefCounted<vk::ShaderAndSerial> *fragmentShader = nullptr;
ANGLE_TRY(shaderLibrary.getFullScreenTri_vert(contextVk, 0, &vertexShader));
ANGLE_TRY(shaderLibrary.getBlitResolve_frag(contextVk, flags, &fragmentShader));
ANGLE_TRY(setupGraphicsProgram(contextVk, Function::BlitResolve, vertexShader, fragmentShader,
&mBlitResolvePrograms[flags], &pipelineDesc, descriptorSet,
&shaderParams, sizeof(shaderParams), commandBuffer));
// Note: this utility starts the render pass directly, thus bypassing
// ContextVk::startRenderPass. As such, occlusion queries are not enabled.
commandBuffer->draw(3, 0);
descriptorPoolBinding.reset();
return angle::Result::Continue;
}
angle::Result UtilsVk::stencilBlitResolveNoShaderExport(ContextVk *contextVk,
FramebufferVk *framebuffer,
vk::ImageHelper *src,
const vk::ImageView *srcStencilView,
const BlitResolveParameters &params)
{
// When VK_EXT_shader_stencil_export is not available, stencil is blitted/resolved into a
// temporary buffer which is then copied into the stencil aspect of the image.
ANGLE_TRY(ensureBlitResolveStencilNoExportResourcesInitialized(contextVk));
bool isResolve = src->getSamples() > 1;
VkDescriptorSet descriptorSet;
vk::RefCountedDescriptorPoolBinding descriptorPoolBinding;
ANGLE_TRY(allocateDescriptorSet(contextVk, Function::BlitResolveStencilNoExport,
&descriptorPoolBinding, &descriptorSet));
// Create a temporary buffer to blit/resolve stencil into.
vk::RendererScoped<vk::BufferHelper> blitBuffer(contextVk->getRenderer());
uint32_t bufferRowLengthInUints = UnsignedCeilDivide(params.blitArea.width, sizeof(uint32_t));
VkDeviceSize bufferSize = bufferRowLengthInUints * sizeof(uint32_t) * params.blitArea.height;
VkBufferCreateInfo blitBufferInfo = {};
blitBufferInfo.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO;
blitBufferInfo.flags = 0;
blitBufferInfo.size = bufferSize;
blitBufferInfo.usage = VK_BUFFER_USAGE_STORAGE_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_SRC_BIT;
blitBufferInfo.sharingMode = VK_SHARING_MODE_EXCLUSIVE;
blitBufferInfo.queueFamilyIndexCount = 0;
blitBufferInfo.pQueueFamilyIndices = nullptr;
ANGLE_TRY(
blitBuffer.get().init(contextVk, blitBufferInfo, VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT));
blitBuffer.get().retainReadWrite(&contextVk->getResourceUseList());
BlitResolveStencilNoExportShaderParams shaderParams;
// Note: adjustments made for pre-rotatation in FramebufferVk::blit() affect these
// Calculate*Offset() functions.
if (isResolve)
{
CalculateResolveOffset(params, shaderParams.offset.resolve);
}
else
{
CalculateBlitOffset(params, shaderParams.offset.blit);
}
shaderParams.stretch[0] = params.stretch[0];
shaderParams.stretch[1] = params.stretch[1];
shaderParams.invSrcExtent[0] = 1.0f / params.srcExtents[0];
shaderParams.invSrcExtent[1] = 1.0f / params.srcExtents[1];
shaderParams.srcLayer = params.srcLayer;
shaderParams.srcWidth = params.srcExtents[0];
shaderParams.dstPitch = bufferRowLengthInUints;
shaderParams.blitArea[0] = params.blitArea.x;
shaderParams.blitArea[1] = params.blitArea.y;
shaderParams.blitArea[2] = params.blitArea.width;
shaderParams.blitArea[3] = params.blitArea.height;
shaderParams.flipX = params.flipX;
shaderParams.flipY = params.flipY;
shaderParams.rotateXY = 0;
// Potentially make adjustments for pre-rotatation. Depending on the angle some of the
// shaderParams need to be adjusted.
switch (params.rotation)
{
case SurfaceRotation::Identity:
break;
case SurfaceRotation::Rotated90Degrees:
shaderParams.rotateXY = 1;
break;
case SurfaceRotation::Rotated180Degrees:
if (isResolve)
{
// Align the offset with minus 1, or the sample position near the edge will be
// wrong.
shaderParams.offset.resolve[0] += params.rotatedOffsetFactor[0] - 1;
shaderParams.offset.resolve[1] += params.rotatedOffsetFactor[1];
}
else
{
shaderParams.offset.blit[0] += params.rotatedOffsetFactor[0] - 1;
shaderParams.offset.blit[1] += params.rotatedOffsetFactor[1];
}
break;
case SurfaceRotation::Rotated270Degrees:
if (isResolve)
{
shaderParams.offset.resolve[0] += params.rotatedOffsetFactor[0] - 1;
shaderParams.offset.resolve[1] += params.rotatedOffsetFactor[1] - 1;
}
else
{
shaderParams.offset.blit[0] += params.rotatedOffsetFactor[0] - 1;
shaderParams.offset.blit[1] += params.rotatedOffsetFactor[1] - 1;
}
shaderParams.rotateXY = 1;
break;
default:
UNREACHABLE();
break;
}
// Linear sampling is only valid with color blitting.
ASSERT(!params.linear);
uint32_t flags = src->getLayerCount() > 1 ? BlitResolveStencilNoExport_comp::kSrcIsArray : 0;
flags |= isResolve ? BlitResolve_frag::kIsResolve : 0;
RenderTargetVk *depthStencilRenderTarget = framebuffer->getDepthStencilRenderTarget();
ASSERT(depthStencilRenderTarget != nullptr);
vk::ImageHelper *depthStencilImage = &depthStencilRenderTarget->getImageForWrite();
// Change layouts prior to computation.
vk::CommandBufferAccess access;
access.onImageComputeShaderRead(src->getAspectFlags(), src);
access.onImageTransferWrite(depthStencilRenderTarget->getLevelIndex(), 1,
depthStencilRenderTarget->getLayerIndex(), 1,
depthStencilImage->getAspectFlags(), depthStencilImage);
vk::OutsideRenderPassCommandBuffer *commandBuffer;
ANGLE_TRY(contextVk->getOutsideRenderPassCommandBuffer(access, &commandBuffer));
// Blit/resolve stencil into the buffer.
VkDescriptorImageInfo imageInfo = {};
imageInfo.imageView = srcStencilView->getHandle();
imageInfo.imageLayout = src->getCurrentLayout();
VkDescriptorBufferInfo bufferInfo = {};
bufferInfo.buffer = blitBuffer.get().getBuffer().getHandle();
bufferInfo.offset = blitBuffer.get().getOffset();
bufferInfo.range = blitBuffer.get().getSize();
VkDescriptorImageInfo samplerInfo = {};
samplerInfo.sampler = params.linear ? mLinearSampler.getHandle() : mPointSampler.getHandle();
VkWriteDescriptorSet writeInfos[3] = {};
writeInfos[0].sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET;
writeInfos[0].dstSet = descriptorSet;
writeInfos[0].dstBinding = kBlitResolveStencilNoExportDestBinding;
writeInfos[0].descriptorCount = 1;
writeInfos[0].descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER;
writeInfos[0].pBufferInfo = &bufferInfo;
writeInfos[1].sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET;
writeInfos[1].dstSet = descriptorSet;
writeInfos[1].dstBinding = kBlitResolveStencilNoExportSrcBinding;
writeInfos[1].descriptorCount = 1;
writeInfos[1].descriptorType = VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE;
writeInfos[1].pImageInfo = &imageInfo;
writeInfos[2].sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET;
writeInfos[2].dstSet = descriptorSet;
writeInfos[2].dstBinding = kBlitResolveStencilNoExportSamplerBinding;
writeInfos[2].descriptorCount = 1;
writeInfos[2].descriptorType = VK_DESCRIPTOR_TYPE_SAMPLER;
writeInfos[2].pImageInfo = &samplerInfo;
vkUpdateDescriptorSets(contextVk->getDevice(), 3, writeInfos, 0, nullptr);
vk::RefCounted<vk::ShaderAndSerial> *shader = nullptr;
ANGLE_TRY(contextVk->getShaderLibrary().getBlitResolveStencilNoExport_comp(contextVk, flags,
&shader));
ANGLE_TRY(setupComputeProgram(contextVk, Function::BlitResolveStencilNoExport, shader,
&mBlitResolveStencilNoExportPrograms[flags], descriptorSet,
&shaderParams, sizeof(shaderParams), commandBuffer));
commandBuffer->dispatch(UnsignedCeilDivide(bufferRowLengthInUints, 8),
UnsignedCeilDivide(params.blitArea.height, 8), 1);
descriptorPoolBinding.reset();
// Add a barrier prior to copy.
VkMemoryBarrier memoryBarrier = {};
memoryBarrier.sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER;
memoryBarrier.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT;
memoryBarrier.dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT;
commandBuffer->memoryBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
VK_PIPELINE_STAGE_TRANSFER_BIT, &memoryBarrier);
// Copy the resulting buffer into dst.
VkBufferImageCopy region = {};
region.bufferOffset = 0;
region.bufferRowLength = bufferRowLengthInUints * sizeof(uint32_t);
region.bufferImageHeight = params.blitArea.height;
region.imageSubresource.aspectMask = VK_IMAGE_ASPECT_STENCIL_BIT;
region.imageSubresource.mipLevel =
depthStencilImage->toVkLevel(depthStencilRenderTarget->getLevelIndex()).get();
region.imageSubresource.baseArrayLayer = depthStencilRenderTarget->getLayerIndex();
region.imageSubresource.layerCount = 1;
region.imageOffset.x = params.blitArea.x;
region.imageOffset.y = params.blitArea.y;
region.imageOffset.z = 0;
region.imageExtent.width = params.blitArea.width;
region.imageExtent.height = params.blitArea.height;
region.imageExtent.depth = 1;
commandBuffer->copyBufferToImage(blitBuffer.get().getBuffer().getHandle(),
depthStencilImage->getImage(),
depthStencilImage->getCurrentLayout(), 1, &region);
return angle::Result::Continue;
}
angle::Result UtilsVk::copyImage(ContextVk *contextVk,
vk::ImageHelper *dst,
const vk::ImageView *destView,
vk::ImageHelper *src,
const vk::ImageView *srcView,
const CopyImageParameters &params)
{
// The views passed to this function are already retained, so a render pass cannot be already
// open. Otherwise, this function closes the render pass, which may incur a vkQueueSubmit and
// then the views are used in a new command buffer without having been retained for it.
// http://crbug.com/1272266#c22
ASSERT(!contextVk->hasStartedRenderPass());
ANGLE_TRY(ensureImageCopyResourcesInitialized(contextVk));
const angle::Format &srcIntendedFormat = src->getIntendedFormat();
const angle::Format &dstIntendedFormat = dst->getIntendedFormat();
ImageCopyShaderParams shaderParams;
shaderParams.flipX = 0;
shaderParams.flipY = params.srcFlipY || params.dstFlipY;
shaderParams.premultiplyAlpha = params.srcPremultiplyAlpha;
shaderParams.unmultiplyAlpha = params.srcUnmultiplyAlpha;
shaderParams.dstHasLuminance = dstIntendedFormat.luminanceBits > 0;
shaderParams.dstIsAlpha = dstIntendedFormat.isLUMA() && dstIntendedFormat.alphaBits > 0;
shaderParams.dstDefaultChannelsMask =
GetFormatDefaultChannelMask(dst->getIntendedFormat(), dst->getActualFormat());
shaderParams.srcMip = params.srcMip;
shaderParams.srcLayer = params.srcLayer;
shaderParams.srcOffset[0] = params.srcOffset[0];
shaderParams.srcOffset[1] = params.srcOffset[1];
shaderParams.dstOffset[0] = params.dstOffset[0];
shaderParams.dstOffset[1] = params.dstOffset[1];
shaderParams.rotateXY = 0;
shaderParams.srcIsSRGB = params.srcColorEncoding == GL_SRGB;
shaderParams.dstIsSRGB = params.dstColorEncoding == GL_SRGB;
// If both src and dst are sRGB, and there is no alpha multiplication/division necessary, then
// the shader can work with sRGB data and pretend they are linear.
if (shaderParams.srcIsSRGB && shaderParams.dstIsSRGB && !shaderParams.premultiplyAlpha &&
!shaderParams.unmultiplyAlpha)
{
shaderParams.srcIsSRGB = false;
shaderParams.dstIsSRGB = false;
}
ASSERT(!(params.srcFlipY && params.dstFlipY));
if (params.srcFlipY)
{
// If viewport is flipped, the shader expects srcOffset[1] to have the
// last row's index instead of the first's.
shaderParams.srcOffset[1] = params.srcHeight - params.srcOffset[1] - 1;
}
else if (params.dstFlipY)
{
// If image is flipped during copy, the shader uses the same code path as above,
// with srcOffset being set to the last row's index instead of the first's.
shaderParams.srcOffset[1] = params.srcOffset[1] + params.srcExtents[1] - 1;
}
switch (params.srcRotation)
{
case SurfaceRotation::Identity:
break;
case SurfaceRotation::Rotated90Degrees:
shaderParams.rotateXY = 1;
break;
case SurfaceRotation::Rotated180Degrees:
shaderParams.flipX = true;
ASSERT(shaderParams.flipY);
shaderParams.flipY = false;
shaderParams.srcOffset[0] += params.srcExtents[0];
shaderParams.srcOffset[1] -= params.srcExtents[1];
break;
case SurfaceRotation::Rotated270Degrees:
shaderParams.flipX = true;
ASSERT(!shaderParams.flipY);
shaderParams.flipY = true;
shaderParams.srcOffset[0] += params.srcExtents[0];
shaderParams.srcOffset[1] += params.srcExtents[1];
shaderParams.rotateXY = 1;
break;
default:
UNREACHABLE();
break;
}
uint32_t flags = GetImageCopyFlags(srcIntendedFormat, dstIntendedFormat);
if (src->getType() == VK_IMAGE_TYPE_3D)
{
flags |= ImageCopy_frag::kSrcIs3D;
}
else if (src->getLayerCount() > 1)
{
flags |= ImageCopy_frag::kSrcIs2DArray;
}
else
{
flags |= ImageCopy_frag::kSrcIs2D;
}
VkDescriptorSet descriptorSet;
vk::RefCountedDescriptorPoolBinding descriptorPoolBinding;
ANGLE_TRY(allocateDescriptorSet(contextVk, Function::ImageCopy, &descriptorPoolBinding,
&descriptorSet));
vk::RenderPassDesc renderPassDesc;
renderPassDesc.setSamples(dst->getSamples());
renderPassDesc.packColorAttachment(0, dst->getActualFormatID());
// Copy from multisampled image is not supported.
ASSERT(src->getSamples() == 1);
vk::GraphicsPipelineDesc pipelineDesc;
pipelineDesc.initDefaults(contextVk);
pipelineDesc.setCullMode(VK_CULL_MODE_NONE);
pipelineDesc.setRenderPassDesc(renderPassDesc);
pipelineDesc.setRasterizationSamples(dst->getSamples());
gl::Rectangle renderArea;
renderArea.x = params.dstOffset[0];
renderArea.y = params.dstOffset[1];
renderArea.width = params.srcExtents[0];
renderArea.height = params.srcExtents[1];
if ((params.srcRotation == SurfaceRotation::Rotated90Degrees) ||
(params.srcRotation == SurfaceRotation::Rotated270Degrees))
{
// The surface is rotated 90/270 degrees. This changes the aspect ratio of the surface.
std::swap(renderArea.x, renderArea.y);
std::swap(renderArea.width, renderArea.height);
}
vk::RenderPassCommandBuffer *commandBuffer;
ANGLE_TRY(
startRenderPass(contextVk, dst, destView, renderPassDesc, renderArea, &commandBuffer));
VkViewport viewport;
gl_vk::GetViewport(renderArea, 0.0f, 1.0f, false, false, dst->getExtents().height, &viewport);
commandBuffer->setViewport(0, 1, &viewport);
VkRect2D scissor = gl_vk::GetRect(renderArea);
commandBuffer->setScissor(0, 1, &scissor);
contextVk->invalidateViewportAndScissor();
// Change source layout inside render pass.
contextVk->onImageRenderPassRead(VK_IMAGE_ASPECT_COLOR_BIT,
vk::ImageLayout::FragmentShaderReadOnly, src);
contextVk->onImageRenderPassWrite(params.dstMip, params.dstLayer, 1, VK_IMAGE_ASPECT_COLOR_BIT,
vk::ImageLayout::ColorAttachment, dst);
VkDescriptorImageInfo imageInfo = {};
imageInfo.imageView = srcView->getHandle();
imageInfo.imageLayout = src->getCurrentLayout();
VkWriteDescriptorSet writeInfo = {};
writeInfo.sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET;
writeInfo.dstSet = descriptorSet;
writeInfo.dstBinding = kImageCopySourceBinding;
writeInfo.descriptorCount = 1;
writeInfo.descriptorType = VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE;
writeInfo.pImageInfo = &imageInfo;
vkUpdateDescriptorSets(contextVk->getDevice(), 1, &writeInfo, 0, nullptr);
vk::ShaderLibrary &shaderLibrary = contextVk->getShaderLibrary();
vk::RefCounted<vk::ShaderAndSerial> *vertexShader = nullptr;
vk::RefCounted<vk::ShaderAndSerial> *fragmentShader = nullptr;
ANGLE_TRY(shaderLibrary.getFullScreenTri_vert(contextVk, 0, &vertexShader));
ANGLE_TRY(shaderLibrary.getImageCopy_frag(contextVk, flags, &fragmentShader));
ANGLE_TRY(setupGraphicsProgram(contextVk, Function::ImageCopy, vertexShader, fragmentShader,
&mImageCopyPrograms[flags], &pipelineDesc, descriptorSet,
&shaderParams, sizeof(shaderParams), commandBuffer));
// Note: this utility creates its own framebuffer, thus bypassing ContextVk::startRenderPass.
// As such, occlusion queries are not enabled.
commandBuffer->draw(3, 0);
descriptorPoolBinding.reset();
// Close the render pass for this temporary framebuffer.
return contextVk->flushCommandsAndEndRenderPass(RenderPassClosureReason::TemporaryForImageCopy);
}
angle::Result UtilsVk::copyImageBits(ContextVk *contextVk,
vk::ImageHelper *dst,
vk::ImageHelper *src,
const CopyImageBitsParameters &params)
{
// This function is used to copy the bit representation of an image to another, and is used to
// support EXT_copy_image when a format is emulated. Currently, only RGB->RGBA emulation is
// possible, and so this function is tailored to this specific kind of emulation.
//
// The copy can be done with various degrees of efficiency:
//
// - If the UINT reinterpretation format for src supports SAMPLED usage, texels can be read
// directly from that. Otherwise vkCmdCopyImageToBuffer can be used and data then read from
// the buffer.
// - If the UINT reinterpretation format for dst supports STORAGE usage, texels can be written
// directly to that. Otherwise conversion can be done to a buffer and then
// vkCmdCopyBufferToImage used.
//
// This requires four different shaders. For simplicity, this function unconditionally copies
// src to a temp buffer, transforms to another temp buffer and copies to the dst. No known
// applications use EXT_copy_image on RGB formats, so no further optimization is currently
// necessary.
//
// The conversion between buffers can be done with ConvertVertex.comp in UintToUint mode, so no
// new shader is necessary. The srcEmulatedAlpha parameter is used to make sure the destination
// alpha value is correct, if dst is RGBA.
// This path should only be necessary for when RGBA is used as fallback for RGB. No other
// format which can be used with EXT_copy_image has a fallback.
ASSERT(src->getIntendedFormat().blueBits > 0 && src->getIntendedFormat().alphaBits == 0);
ASSERT(dst->getIntendedFormat().blueBits > 0 && dst->getIntendedFormat().alphaBits == 0);
const angle::Format &srcImageFormat = src->getActualFormat();
const angle::Format &dstImageFormat = dst->getActualFormat();
// Create temporary buffers.
vk::RendererScoped<vk::BufferHelper> srcBuffer(contextVk->getRenderer());
vk::RendererScoped<vk::BufferHelper> dstBuffer(contextVk->getRenderer());
const uint32_t srcPixelBytes = srcImageFormat.pixelBytes;
const uint32_t dstPixelBytes = dstImageFormat.pixelBytes;
const uint32_t totalPixelCount =
params.copyExtents[0] * params.copyExtents[1] * params.copyExtents[2];
// Note that buffer sizes are rounded up a multiple of uint size, as that the granularity in
// which the compute shader accesses these buffers.
const VkDeviceSize srcBufferSize =
roundUpPow2<uint32_t>(srcPixelBytes * totalPixelCount, sizeof(uint32_t));
const VkDeviceSize dstBufferSize =
roundUpPow2<uint32_t>(dstPixelBytes * totalPixelCount, sizeof(uint32_t));
VkBufferCreateInfo bufferInfo = {};
bufferInfo.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO;
bufferInfo.flags = 0;
bufferInfo.size = srcBufferSize;
bufferInfo.usage = VK_BUFFER_USAGE_STORAGE_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT;
bufferInfo.sharingMode = VK_SHARING_MODE_EXCLUSIVE;
bufferInfo.queueFamilyIndexCount = 0;
bufferInfo.pQueueFamilyIndices = nullptr;
ANGLE_TRY(srcBuffer.get().init(contextVk, bufferInfo, VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT));
bufferInfo.size = dstBufferSize;
bufferInfo.usage = VK_BUFFER_USAGE_STORAGE_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_SRC_BIT;
ANGLE_TRY(dstBuffer.get().init(contextVk, bufferInfo, VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT));
srcBuffer.get().retainReadOnly(&contextVk->getResourceUseList());
dstBuffer.get().retainReadWrite(&contextVk->getResourceUseList());
bool isSrc3D = src->getType() == VK_IMAGE_TYPE_3D;
bool isDst3D = dst->getType() == VK_IMAGE_TYPE_3D;
// Change layouts prior to computation.
vk::CommandBufferAccess access;
access.onImageTransferRead(src->getAspectFlags(), src);
access.onImageTransferWrite(params.dstLevel, 1, isDst3D ? 0 : params.dstOffset[2],
isDst3D ? 1 : params.copyExtents[2], VK_IMAGE_ASPECT_COLOR_BIT,
dst);
vk::OutsideRenderPassCommandBuffer *commandBuffer;
ANGLE_TRY(contextVk->getOutsideRenderPassCommandBuffer(access, &commandBuffer));
// Copy src into buffer, completely packed.
VkBufferImageCopy srcRegion = {};
srcRegion.imageSubresource.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
srcRegion.imageSubresource.mipLevel = src->toVkLevel(params.srcLevel).get();
srcRegion.imageSubresource.baseArrayLayer = isSrc3D ? 0 : params.srcOffset[2];
srcRegion.imageSubresource.layerCount = isSrc3D ? 1 : params.copyExtents[2];
srcRegion.imageOffset.x = params.srcOffset[0];
srcRegion.imageOffset.y = params.srcOffset[1];
srcRegion.imageOffset.z = isSrc3D ? params.srcOffset[2] : 0;
srcRegion.imageExtent.width = params.copyExtents[0];
srcRegion.imageExtent.height = params.copyExtents[1];
srcRegion.imageExtent.depth = isSrc3D ? params.copyExtents[2] : 1;
commandBuffer->copyImageToBuffer(src->getImage(), src->getCurrentLayout(),
srcBuffer.get().getBuffer().getHandle(), 1, &srcRegion);
// Add a barrier prior to dispatch call.
VkMemoryBarrier memoryBarrier = {};
memoryBarrier.sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER;
memoryBarrier.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
memoryBarrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT;
commandBuffer->memoryBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, &memoryBarrier);
// Set up ConvertVertex shader to convert between the formats. Only the following three cases
// are possible:
//
// - RGB -> RGBA: Ns = 3, Ss = src.pixelBytes,
// Nd = 4, Sd = dst.pixelBytes, use srcEmulatedAlpha
//
// - RGBA -> RGBA: Ns = 3, Ss = src.pixelBytes,
// Nd = 4, Sd = dst.pixelBytes, use srcEmulatedAlpha
//
// - RGBA -> RGB: Ns = 3, Ss = src.pixelBytes,
// Nd = 3, Sd = dst.pixelBytes
//
// The trick here is with RGBA -> RGBA, where Ns is specified as 3, so that the emulated alpha
// from source is not taken (as uint), but rather one is provided such that the destination
// alpha would contain the correct emulated alpha.
//
ConvertVertexShaderParams shaderParams;
shaderParams.Ns = 3;
shaderParams.Bs = srcImageFormat.pixelBytes / srcImageFormat.channelCount;
shaderParams.Ss = srcImageFormat.pixelBytes;
shaderParams.Nd = dstImageFormat.channelCount;
shaderParams.Bd = dstImageFormat.pixelBytes / dstImageFormat.channelCount;
shaderParams.Sd = shaderParams.Nd * shaderParams.Bd;
// The component size is expected to either be 1, 2 or 4 bytes.
ASSERT(4 % shaderParams.Bs == 0);
ASSERT(4 % shaderParams.Bd == 0);
shaderParams.Es = 4 / shaderParams.Bs;
shaderParams.Ed = 4 / shaderParams.Bd;
// Total number of output components is simply the number of pixels by number of components in
// each.
shaderParams.componentCount = totalPixelCount * shaderParams.Nd;
// Total number of 4-byte outputs is the number of components divided by how many components can
// fit in a 4-byte value. Note that this value is also the invocation size of the shader.
shaderParams.outputCount = UnsignedCeilDivide(shaderParams.componentCount, shaderParams.Ed);
shaderParams.srcOffset = 0;
shaderParams.dstOffset = 0;
shaderParams.isSrcHDR = 0;
shaderParams.isSrcA2BGR10 = 0;
// Due to the requirements of EXT_copy_image, the channel size of src and dst must be
// identical. Usage of srcEmulatedAlpha relies on this as it's used to output an alpha value in
// dst through the source.
ASSERT(shaderParams.Bs == shaderParams.Bd);
// The following RGB formats are allowed in EXT_copy_image:
//
// - RGB32F, RGB32UI, RGB32I
// - RGB16F, RGB16UI, RGB16I
// - RGB8, RGB8_SNORM, SRGB8, RGB8UI, RGB8I
//
// The value of emulated alpha is:
//
// - 1 for all RGB*I and RGB*UI formats
// - bit representation of 1.0f for RGB32F
// - bit representation of half-float 1.0f for RGB16F
// - 0xFF for RGB8 and SRGB8
// - 0x7F for RGB8_SNORM
if (dstImageFormat.isInt())
{
shaderParams.srcEmulatedAlpha = 1;
}
else if (dstImageFormat.isUnorm())
{
ASSERT(shaderParams.Bd == 1);
shaderParams.srcEmulatedAlpha = 0xFF;
}
else if (dstImageFormat.isSnorm())
{
ASSERT(shaderParams.Bd == 1);
shaderParams.srcEmulatedAlpha = 0x7F;
}
else if (shaderParams.Bd == 2)
{
ASSERT(dstImageFormat.isFloat());
shaderParams.srcEmulatedAlpha = gl::Float16One;
}
else if (shaderParams.Bd == 4)
{
ASSERT(dstImageFormat.isFloat());
ASSERT(ValidateFloatOneAsUint());
shaderParams.srcEmulatedAlpha = gl::Float32One;
}
else
{
UNREACHABLE();
}
// Use UintToUint conversion to preserve the bit pattern during transfer.
const uint32_t flags = ConvertVertex_comp::kUintToUint;
ANGLE_TRY(convertVertexBufferImpl(contextVk, &dstBuffer.get(), &srcBuffer.get(), flags,
commandBuffer, shaderParams));
// Add a barrier prior to copy.
memoryBarrier.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT;
memoryBarrier.dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT;
commandBuffer->memoryBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
VK_PIPELINE_STAGE_TRANSFER_BIT, &memoryBarrier);
// Copy buffer into dst. It's completely packed.
VkBufferImageCopy dstRegion = {};
dstRegion.imageSubresource.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
dstRegion.imageSubresource.mipLevel = dst->toVkLevel(params.dstLevel).get();
dstRegion.imageSubresource.baseArrayLayer = isDst3D ? 0 : params.dstOffset[2];
dstRegion.imageSubresource.layerCount = isDst3D ? 1 : params.copyExtents[2];
dstRegion.imageOffset.x = params.dstOffset[0];
dstRegion.imageOffset.y = params.dstOffset[1];
dstRegion.imageOffset.z = isDst3D ? params.dstOffset[2] : 0;
dstRegion.imageExtent.width = params.copyExtents[0];
dstRegion.imageExtent.height = params.copyExtents[1];
dstRegion.imageExtent.depth = isDst3D ? params.copyExtents[2] : 1;
commandBuffer->copyBufferToImage(dstBuffer.get().getBuffer().getHandle(), dst->getImage(),
dst->getCurrentLayout(), 1, &dstRegion);
return angle::Result::Continue;
}
angle::Result UtilsVk::generateMipmap(ContextVk *contextVk,
vk::ImageHelper *src,
const vk::ImageView *srcLevelZeroView,
vk::ImageHelper *dst,
const GenerateMipmapDestLevelViews &destLevelViews,
const vk::Sampler &sampler,
const GenerateMipmapParameters &params)
{
ANGLE_TRY(ensureGenerateMipmapResourcesInitialized(contextVk));
const gl::Extents &srcExtents = src->getLevelExtents(vk::LevelIndex(params.srcLevel));
ASSERT(srcExtents.depth == 1);
// Each workgroup processes a 64x64 tile of the image.
constexpr uint32_t kPixelWorkgroupRatio = 64;
const uint32_t workGroupX = UnsignedCeilDivide(srcExtents.width, kPixelWorkgroupRatio);
const uint32_t workGroupY = UnsignedCeilDivide(srcExtents.height, kPixelWorkgroupRatio);
GenerateMipmapShaderParams shaderParams;
shaderParams.invSrcExtent[0] = 1.0f / srcExtents.width;
shaderParams.invSrcExtent[1] = 1.0f / srcExtents.height;
shaderParams.levelCount = params.dstLevelCount;
uint32_t flags = GetGenerateMipmapFlags(contextVk, src->getActualFormat());
VkDescriptorSet descriptorSet;
vk::RefCountedDescriptorPoolBinding descriptorPoolBinding;
ANGLE_TRY(allocateDescriptorSet(contextVk, Function::GenerateMipmap, &descriptorPoolBinding,
&descriptorSet));
VkDescriptorImageInfo destImageInfos[kGenerateMipmapMaxLevels] = {};
for (uint32_t level = 0; level < kGenerateMipmapMaxLevels; ++level)
{
destImageInfos[level].imageView = destLevelViews[level]->getHandle();
destImageInfos[level].imageLayout = dst->getCurrentLayout();
}
VkDescriptorImageInfo srcImageInfo = {};
srcImageInfo.imageView = srcLevelZeroView->getHandle();
srcImageInfo.imageLayout = src->getCurrentLayout();
srcImageInfo.sampler = sampler.getHandle();
VkWriteDescriptorSet writeInfos[2] = {};
writeInfos[0].sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET;
writeInfos[0].dstSet = descriptorSet;
writeInfos[0].dstBinding = kGenerateMipmapDestinationBinding;
writeInfos[0].descriptorCount = GetGenerateMipmapMaxLevels(contextVk);
writeInfos[0].descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE;
writeInfos[0].pImageInfo = destImageInfos;
writeInfos[1].sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET;
writeInfos[1].dstSet = descriptorSet;
writeInfos[1].dstBinding = kGenerateMipmapSourceBinding;
writeInfos[1].descriptorCount = 1;
writeInfos[1].descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER;
writeInfos[1].pImageInfo = &srcImageInfo;
vkUpdateDescriptorSets(contextVk->getDevice(), 2, writeInfos, 0, nullptr);
vk::RefCounted<vk::ShaderAndSerial> *shader = nullptr;
ANGLE_TRY(contextVk->getShaderLibrary().getGenerateMipmap_comp(contextVk, flags, &shader));
// Note: onImageRead/onImageWrite is expected to be called by the caller. This avoids inserting
// barriers between calls for each layer of the image.
vk::OutsideRenderPassCommandBuffer *commandBuffer;
ANGLE_TRY(contextVk->getOutsideRenderPassCommandBuffer({}, &commandBuffer));
ANGLE_TRY(setupComputeProgram(contextVk, Function::GenerateMipmap, shader,
&mGenerateMipmapPrograms[flags], descriptorSet, &shaderParams,
sizeof(shaderParams), commandBuffer));
commandBuffer->dispatch(workGroupX, workGroupY, 1);
descriptorPoolBinding.reset();
return angle::Result::Continue;
}
angle::Result UtilsVk::unresolve(ContextVk *contextVk,
const FramebufferVk *framebuffer,
const UnresolveParameters &params)
{
// Get attachment count and pointers to resolve images and views.
gl::DrawBuffersArray<vk::ImageHelper *> colorSrc = {};
gl::DrawBuffersArray<const vk::ImageView *> colorSrcView = {};
vk::DeviceScoped<vk::ImageView> depthView(contextVk->getDevice());
vk::DeviceScoped<vk::ImageView> stencilView(contextVk->getDevice());
const vk::ImageView *depthSrcView = nullptr;
const vk::ImageView *stencilSrcView = nullptr;
// The subpass that initializes the multisampled-render-to-texture attachments packs the
// attachments that need to be unresolved, so the attachment indices of this subpass are not the
// same. See InitializeUnresolveSubpass for details.
vk::PackedAttachmentIndex colorIndexVk(0);
for (size_t colorIndexGL : params.unresolveColorMask)
{
RenderTargetVk *colorRenderTarget = framebuffer->getColorDrawRenderTarget(colorIndexGL);
ASSERT(colorRenderTarget->hasResolveAttachment());
ASSERT(colorRenderTarget->isImageTransient());
colorSrc[colorIndexVk.get()] = &colorRenderTarget->getResolveImageForRenderPass();
ANGLE_TRY(
colorRenderTarget->getResolveImageView(contextVk, &colorSrcView[colorIndexVk.get()]));
++colorIndexVk;
}
if (params.unresolveDepth || params.unresolveStencil)
{
RenderTargetVk *depthStencilRenderTarget = framebuffer->getDepthStencilRenderTarget();
ASSERT(depthStencilRenderTarget->hasResolveAttachment());
ASSERT(depthStencilRenderTarget->isImageTransient());
vk::ImageHelper *depthStencilSrc =
&depthStencilRenderTarget->getResolveImageForRenderPass();
// The resolved depth/stencil image is necessarily single-sampled.
ASSERT(depthStencilSrc->getSamples() == 1);
gl::TextureType textureType = vk::Get2DTextureType(depthStencilSrc->getLayerCount(), 1);
const vk::LevelIndex levelIndex =
depthStencilSrc->toVkLevel(depthStencilRenderTarget->getLevelIndex());
const uint32_t layerIndex = depthStencilRenderTarget->getLayerIndex();
if (params.unresolveDepth)
{
ANGLE_TRY(depthStencilSrc->initLayerImageView(
contextVk, textureType, VK_IMAGE_ASPECT_DEPTH_BIT, gl::SwizzleState(),
&depthView.get(), levelIndex, 1, layerIndex, 1, gl::SrgbWriteControlMode::Default));
depthSrcView = &depthView.get();
}
if (params.unresolveStencil)
{
ANGLE_TRY(depthStencilSrc->initLayerImageView(
contextVk, textureType, VK_IMAGE_ASPECT_STENCIL_BIT, gl::SwizzleState(),
&stencilView.get(), levelIndex, 1, layerIndex, 1,
gl::SrgbWriteControlMode::Default));
stencilSrcView = &stencilView.get();
}
}
const uint32_t colorAttachmentCount = colorIndexVk.get();
const uint32_t depthStencilBindingCount =
(params.unresolveDepth ? 1 : 0) + (params.unresolveStencil ? 1 : 0);
const uint32_t totalBindingCount = colorAttachmentCount + depthStencilBindingCount;
ASSERT(totalBindingCount > 0);
const Function function = static_cast<Function>(
static_cast<uint32_t>(Function::Unresolve1Attachment) + totalBindingCount - 1);
ANGLE_TRY(ensureUnresolveResourcesInitialized(contextVk, function, totalBindingCount));
vk::GraphicsPipelineDesc pipelineDesc;
pipelineDesc.initDefaults(contextVk);
pipelineDesc.setCullMode(VK_CULL_MODE_NONE);
pipelineDesc.setRasterizationSamples(framebuffer->getSamples());
pipelineDesc.setRenderPassDesc(framebuffer->getRenderPassDesc());
pipelineDesc.setDepthTestEnabled(params.unresolveDepth);
pipelineDesc.setDepthWriteEnabled(params.unresolveDepth);
pipelineDesc.setDepthFunc(VK_COMPARE_OP_ALWAYS);
if (params.unresolveStencil)
{
SetStencilForShaderExport(contextVk, &pipelineDesc);
}
vk::RenderPassCommandBuffer *commandBuffer =
&contextVk->getStartedRenderPassCommands().getCommandBuffer();
VkViewport viewport;
gl::Rectangle completeRenderArea = framebuffer->getRotatedCompleteRenderArea(contextVk);
bool invertViewport = contextVk->isViewportFlipEnabledForDrawFBO();
bool clipSpaceOriginUpperLeft =
contextVk->getState().getClipSpaceOrigin() == gl::ClipSpaceOrigin::UpperLeft;
gl_vk::GetViewport(completeRenderArea, 0.0f, 1.0f, invertViewport, clipSpaceOriginUpperLeft,
completeRenderArea.height, &viewport);
commandBuffer->setViewport(0, 1, &viewport);
VkRect2D scissor = gl_vk::GetRect(completeRenderArea);
commandBuffer->setScissor(0, 1, &scissor);
contextVk->invalidateViewportAndScissor();
VkDescriptorSet descriptorSet;
vk::RefCountedDescriptorPoolBinding descriptorPoolBinding;
ANGLE_TRY(allocateDescriptorSet(contextVk, function, &descriptorPoolBinding, &descriptorSet));
vk::FramebufferAttachmentArray<VkDescriptorImageInfo> inputImageInfo = {};
for (uint32_t attachmentIndex = 0; attachmentIndex < colorAttachmentCount; ++attachmentIndex)
{
inputImageInfo[attachmentIndex].imageView = colorSrcView[attachmentIndex]->getHandle();
inputImageInfo[attachmentIndex].imageLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL;
}
uint32_t depthStencilBindingIndex = colorAttachmentCount;
if (params.unresolveDepth)
{
inputImageInfo[depthStencilBindingIndex].imageView = depthSrcView->getHandle();
inputImageInfo[depthStencilBindingIndex].imageLayout =
VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL;
++depthStencilBindingIndex;
}
if (params.unresolveStencil)
{
inputImageInfo[depthStencilBindingIndex].imageView = stencilSrcView->getHandle();
inputImageInfo[depthStencilBindingIndex].imageLayout =
VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL;
}
VkWriteDescriptorSet writeInfo = {};
writeInfo.sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET;
writeInfo.dstSet = descriptorSet;
writeInfo.dstBinding = 0;
writeInfo.descriptorCount = totalBindingCount;
writeInfo.descriptorType = VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT;
writeInfo.pImageInfo = inputImageInfo.data();
vkUpdateDescriptorSets(contextVk->getDevice(), 1, &writeInfo, 0, nullptr);
gl::DrawBuffersArray<UnresolveColorAttachmentType> colorAttachmentTypes;
uint32_t flags = GetUnresolveFlags(colorAttachmentCount, colorSrc, params.unresolveDepth,
params.unresolveStencil, &colorAttachmentTypes);
vk::ShaderLibrary &shaderLibrary = contextVk->getShaderLibrary();
vk::RefCounted<vk::ShaderAndSerial> *vertexShader = nullptr;
vk::RefCounted<vk::ShaderAndSerial> *fragmentShader = &mUnresolveFragShaders[flags];
ANGLE_TRY(shaderLibrary.getFullScreenTri_vert(contextVk, 0, &vertexShader));
ANGLE_TRY(GetUnresolveFrag(contextVk, colorAttachmentCount, colorAttachmentTypes,
params.unresolveDepth, params.unresolveStencil, fragmentShader));
ANGLE_TRY(setupGraphicsProgram(contextVk, function, vertexShader, fragmentShader,
&mUnresolvePrograms[flags], &pipelineDesc, descriptorSet,
nullptr, 0, commandBuffer));
// This draw call is made before ContextVk gets a chance to start the occlusion query. As such,
// occlusion queries are not enabled.
commandBuffer->draw(3, 0);
// Release temporary views
vk::ImageView depthViewObject = depthView.release();
vk::ImageView stencilViewObject = stencilView.release();
contextVk->addGarbage(&depthViewObject);
contextVk->addGarbage(&stencilViewObject);
return angle::Result::Continue;
}
angle::Result UtilsVk::cullOverlayWidgets(ContextVk *contextVk,
vk::BufferHelper *enabledWidgetsBuffer,
vk::ImageHelper *dst,
const vk::ImageView *destView,
const OverlayCullParameters &params)
{
ANGLE_TRY(ensureOverlayCullResourcesInitialized(contextVk));
ASSERT(params.subgroupSize[0] == 8 &&
(params.subgroupSize[1] == 8 || params.subgroupSize[1] == 4));
uint32_t flags =
params.subgroupSize[1] == 8 ? OverlayCull_comp::kIs8x8 : OverlayCull_comp::kIs8x4;
if (params.supportsSubgroupBallot)
{
flags |= OverlayCull_comp::kSupportsBallot;
}
else if (params.supportsSubgroupBallot)
{
flags |= OverlayCull_comp::kSupportsArithmetic;
}
else
{
flags |= OverlayCull_comp::kSupportsNone;
}
VkDescriptorSet descriptorSet;
vk::RefCountedDescriptorPoolBinding descriptorPoolBinding;
ANGLE_TRY(allocateDescriptorSet(contextVk, Function::OverlayCull, &descriptorPoolBinding,
&descriptorSet));
ASSERT(dst->getLevelCount() == 1 && dst->getLayerCount() == 1 &&
dst->getFirstAllocatedLevel() == gl::LevelIndex(0));
vk::CommandBufferAccess access;
access.onBufferComputeShaderRead(enabledWidgetsBuffer);
access.onImageComputeShaderWrite(gl::LevelIndex(0), 1, 0, 1, VK_IMAGE_ASPECT_COLOR_BIT, dst);
vk::OutsideRenderPassCommandBuffer *commandBuffer;
ANGLE_TRY(contextVk->getOutsideRenderPassCommandBuffer(access, &commandBuffer));
VkDescriptorImageInfo imageInfo = {};
imageInfo.imageView = destView->getHandle();
imageInfo.imageLayout = dst->getCurrentLayout();
VkDescriptorBufferInfo bufferInfo = {};
bufferInfo.buffer = enabledWidgetsBuffer->getBuffer().getHandle();
bufferInfo.offset = enabledWidgetsBuffer->getOffset();
bufferInfo.range = enabledWidgetsBuffer->getSize();
VkWriteDescriptorSet writeInfos[2] = {};
writeInfos[0].sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET;
writeInfos[0].dstSet = descriptorSet;
writeInfos[0].dstBinding = kOverlayCullCulledWidgetsBinding;
writeInfos[0].descriptorCount = 1;
writeInfos[0].descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE;
writeInfos[0].pImageInfo = &imageInfo;
writeInfos[1].sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET;
writeInfos[1].dstSet = descriptorSet;
writeInfos[1].dstBinding = kOverlayCullWidgetCoordsBinding;
writeInfos[1].descriptorCount = 1;
writeInfos[1].descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
writeInfos[1].pBufferInfo = &bufferInfo;
vkUpdateDescriptorSets(contextVk->getDevice(), 2, writeInfos, 0, nullptr);
vk::RefCounted<vk::ShaderAndSerial> *shader = nullptr;
ANGLE_TRY(contextVk->getShaderLibrary().getOverlayCull_comp(contextVk, flags, &shader));
ANGLE_TRY(setupComputeProgram(contextVk, Function::OverlayCull, shader,
&mOverlayCullPrograms[flags], descriptorSet, nullptr, 0,
commandBuffer));
commandBuffer->dispatch(dst->getExtents().width, dst->getExtents().height, 1);
descriptorPoolBinding.reset();
return angle::Result::Continue;
}
angle::Result UtilsVk::drawOverlay(ContextVk *contextVk,
vk::BufferHelper *textWidgetsBuffer,
vk::BufferHelper *graphWidgetsBuffer,
vk::ImageHelper *font,
const vk::ImageView *fontView,
vk::ImageHelper *culledWidgets,
const vk::ImageView *culledWidgetsView,
vk::ImageHelper *dst,
const vk::ImageView *destView,
const OverlayDrawParameters &params)
{
ANGLE_TRY(ensureOverlayDrawResourcesInitialized(contextVk));
OverlayDrawShaderParams shaderParams;
shaderParams.outputSize[0] = dst->getExtents().width;
shaderParams.outputSize[1] = dst->getExtents().height;
shaderParams.rotateXY = params.rotateXY;
ASSERT(params.subgroupSize[0] == 8 &&
(params.subgroupSize[1] == 8 || params.subgroupSize[1] == 4));
uint32_t flags =
params.subgroupSize[1] == 8 ? OverlayDraw_comp::kIs8x8 : OverlayDraw_comp::kIs8x4;
VkDescriptorSet descriptorSet;
vk::RefCountedDescriptorPoolBinding descriptorPoolBinding;
ANGLE_TRY(allocateDescriptorSet(contextVk, Function::OverlayDraw, &descriptorPoolBinding,
&descriptorSet));
ASSERT(dst->getLevelCount() == 1 && dst->getLayerCount() == 1 &&
dst->getFirstAllocatedLevel() == gl::LevelIndex(0));
vk::CommandBufferAccess access;
access.onImageComputeShaderWrite(gl::LevelIndex(0), 1, 0, 1, VK_IMAGE_ASPECT_COLOR_BIT, dst);
access.onImageComputeShaderRead(VK_IMAGE_ASPECT_COLOR_BIT, culledWidgets);
access.onImageComputeShaderRead(VK_IMAGE_ASPECT_COLOR_BIT, font);
access.onBufferComputeShaderRead(textWidgetsBuffer);
access.onBufferComputeShaderRead(graphWidgetsBuffer);
vk::OutsideRenderPassCommandBuffer *commandBuffer;
ANGLE_TRY(contextVk->getOutsideRenderPassCommandBuffer(access, &commandBuffer));
VkDescriptorImageInfo imageInfos[3] = {};
imageInfos[0].imageView = destView->getHandle();
imageInfos[0].imageLayout = dst->getCurrentLayout();
imageInfos[1].imageView = culledWidgetsView->getHandle();
imageInfos[1].imageLayout = culledWidgets->getCurrentLayout();
imageInfos[2].imageView = fontView->getHandle();
imageInfos[2].imageLayout = font->getCurrentLayout();
VkDescriptorBufferInfo bufferInfos[2] = {};
bufferInfos[0].buffer = textWidgetsBuffer->getBuffer().getHandle();
bufferInfos[0].offset = textWidgetsBuffer->getOffset();
bufferInfos[0].range = textWidgetsBuffer->getSize();
bufferInfos[1].buffer = graphWidgetsBuffer->getBuffer().getHandle();
bufferInfos[1].offset = graphWidgetsBuffer->getOffset();
bufferInfos[1].range = graphWidgetsBuffer->getSize();
VkWriteDescriptorSet writeInfos[5] = {};
writeInfos[0].sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET;
writeInfos[0].dstSet = descriptorSet;
writeInfos[0].dstBinding = kOverlayDrawOutputBinding;
writeInfos[0].descriptorCount = 1;
writeInfos[0].descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE;
writeInfos[0].pImageInfo = &imageInfos[0];
writeInfos[1].sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET;
writeInfos[1].dstSet = descriptorSet;
writeInfos[1].dstBinding = kOverlayDrawCulledWidgetsBinding;
writeInfos[1].descriptorCount = 1;
writeInfos[1].descriptorType = VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE;
writeInfos[1].pImageInfo = &imageInfos[1];
writeInfos[2].sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET;
writeInfos[2].dstSet = descriptorSet;
writeInfos[2].dstBinding = kOverlayDrawFontBinding;
writeInfos[2].descriptorCount = 1;
writeInfos[2].descriptorType = VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE;
writeInfos[2].pImageInfo = &imageInfos[2];
writeInfos[3].sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET;
writeInfos[3].dstSet = descriptorSet;
writeInfos[3].dstBinding = kOverlayDrawTextWidgetsBinding;
writeInfos[3].descriptorCount = 1;
writeInfos[3].descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
writeInfos[3].pBufferInfo = &bufferInfos[0];
writeInfos[4] = writeInfos[3];
writeInfos[4].dstBinding = kOverlayDrawGraphWidgetsBinding;
writeInfos[4].pBufferInfo = &bufferInfos[1];
vkUpdateDescriptorSets(contextVk->getDevice(), 5, writeInfos, 0, nullptr);
vk::RefCounted<vk::ShaderAndSerial> *shader = nullptr;
ANGLE_TRY(contextVk->getShaderLibrary().getOverlayDraw_comp(contextVk, flags, &shader));
ANGLE_TRY(setupComputeProgram(contextVk, Function::OverlayDraw, shader,
&mOverlayDrawPrograms[flags], descriptorSet, &shaderParams,
sizeof(shaderParams), commandBuffer));
// Every pixel of culledWidgets corresponds to one workgroup, so we can use that as dispatch
// size.
const VkExtent3D &extents = culledWidgets->getExtents();
commandBuffer->dispatch(extents.width, extents.height, 1);
descriptorPoolBinding.reset();
return angle::Result::Continue;
}
angle::Result UtilsVk::allocateDescriptorSet(ContextVk *contextVk,
Function function,
vk::RefCountedDescriptorPoolBinding *bindingOut,
VkDescriptorSet *descriptorSetOut)
{
ANGLE_TRY(mDescriptorPools[function].allocateSets(
contextVk, mDescriptorSetLayouts[function][DescriptorSetIndex::Internal].get().ptr(), 1,
bindingOut, descriptorSetOut));
mPerfCounters.descriptorSetsAllocated++;
return angle::Result::Continue;
}
UtilsVk::ClearFramebufferParameters::ClearFramebufferParameters()
: clearColor(false),
clearDepth(false),
clearStencil(false),
stencilMask(0),
colorMaskFlags(0),
colorAttachmentIndexGL(0),
colorFormat(nullptr),
colorClearValue{},
depthStencilClearValue{}
{}
// Requires that trace is enabled to see the output, which is supported with is_debug=true
void UtilsVk::outputCumulativePerfCounters()
{
if (!vk::kOutputCumulativePerfCounters)
{
return;
}
INFO() << "Utils Descriptor Set Allocations: "
<< mCumulativePerfCounters.descriptorSetsAllocated;
}
InternalShaderPerfCounters UtilsVk::getAndResetObjectPerfCounters()
{
mCumulativePerfCounters.descriptorSetsAllocated += mPerfCounters.descriptorSetsAllocated;
InternalShaderPerfCounters counters = mPerfCounters;
mPerfCounters.descriptorSetsAllocated = 0;
return counters;
}
} // namespace rx