swung0x48/MobileGL
1
0
Fork 0
mirror of https://github.com/MobileGL-Dev/MobileGL synced 2026-09-11 21:58:31 +09:00
Code Issues Packages Projects Releases Wiki Activity

[Fix] (MG_Backend, MG_Impl, ShaderTranspiler, MG_Test): support iterationRP custom images and storage format reinterpretation

Browse Source
This commit is contained in:
swung0x48
2026-07-20 02:35:30 -04:00
parent 8bc4808b1a
commit e5388c0e7e
33 changed files with 2382 additions and 88 deletions
Download Patch File Download Diff File Expand all files Collapse all files
MobileGL/MG_Util/BackendLoaders/OpenGL/Loader.cpp
+18
View File
@@ -865,6 +865,9 @@ namespace MobileGL::MG_Util::BackendLoader {
GLint maxUniformBlockSize = 16384;
GLint maxImageUnits = 8;
GLint maxCombinedImageUniforms = 8;
GLint maxVertexImageUniforms = 0;
GLint maxGeometryImageUniforms = 0;
GLint maxFragmentImageUniforms = 8;
GLint maxComputeImageUniforms = 8;
GLint maxDrawBuffers = 8;
GLint maxColorAttachments = 8;
@@ -904,7 +907,16 @@ namespace MobileGL::MG_Util::BackendLoader {
glesFuncs.glGetIntegerv(GL_MAX_UNIFORM_BLOCK_SIZE, &maxUniformBlockSize);
glesFuncs.glGetIntegerv(GL_MAX_IMAGE_UNITS, &maxImageUnits);
glesFuncs.glGetIntegerv(GL_MAX_COMBINED_IMAGE_UNIFORMS, &maxCombinedImageUniforms);
glesFuncs.glGetIntegerv(GL_MAX_VERTEX_IMAGE_UNIFORMS, &maxVertexImageUniforms);
glesFuncs.glGetIntegerv(GL_MAX_FRAGMENT_IMAGE_UNIFORMS, &maxFragmentImageUniforms);
glesFuncs.glGetIntegerv(GL_MAX_COMPUTE_IMAGE_UNIFORMS, &maxComputeImageUniforms);
// Geometry shaders and their image-uniform query are core only in ES 3.2. DirectGLES
// emits ESSL 3.10 on an ES 3.1 context, so reporting zero there is both legal and an
// accurate description of what the backend compiler can consume.
if (caps.GLESVersion.Major > 3 ||
(caps.GLESVersion.Major == 3 && caps.GLESVersion.Minor >= 2)) {
glesFuncs.glGetIntegerv(GL_MAX_GEOMETRY_IMAGE_UNIFORMS, &maxGeometryImageUniforms);
}
glesFuncs.glGetIntegerv(GL_MAX_DRAW_BUFFERS, &maxDrawBuffers);
glesFuncs.glGetIntegerv(GL_MAX_COLOR_ATTACHMENTS, &maxColorAttachments);
glesFuncs.glGetIntegerv(GL_MAX_CLIP_DISTANCES, &maxClipDistances);
@@ -955,6 +967,9 @@ namespace MobileGL::MG_Util::BackendLoader {
caps.MaxUniformBlockSize = maxUniformBlockSize;
caps.MaxImageUnits = maxImageUnits;
caps.MaxCombinedImageUniforms = maxCombinedImageUniforms;
caps.MaxVertexImageUniforms = maxVertexImageUniforms;
caps.MaxGeometryImageUniforms = maxGeometryImageUniforms;
caps.MaxFragmentImageUniforms = maxFragmentImageUniforms;
caps.MaxComputeImageUniforms = maxComputeImageUniforms;
caps.MaxDrawBuffers = maxDrawBuffers;
caps.MaxColorAttachments = maxColorAttachments;
@@ -1000,6 +1015,9 @@ namespace MobileGL::MG_Util::BackendLoader {
MGLOG_I(" GL_MAX_UNIFORM_BLOCK_SIZE: %d", caps.MaxUniformBlockSize);
MGLOG_I(" GL_MAX_IMAGE_UNITS: %d", caps.MaxImageUnits);
MGLOG_I(" GL_MAX_COMBINED_IMAGE_UNIFORMS: %d", caps.MaxCombinedImageUniforms);
MGLOG_I(" GL_MAX_VERTEX_IMAGE_UNIFORMS: %d", caps.MaxVertexImageUniforms);
MGLOG_I(" GL_MAX_GEOMETRY_IMAGE_UNIFORMS: %d", caps.MaxGeometryImageUniforms);
MGLOG_I(" GL_MAX_FRAGMENT_IMAGE_UNIFORMS: %d", caps.MaxFragmentImageUniforms);
MGLOG_I(" GL_MAX_COMPUTE_IMAGE_UNIFORMS: %d", caps.MaxComputeImageUniforms);
MGLOG_I(" GL_MAX_DRAW_BUFFERS: %d", caps.MaxDrawBuffers);
MGLOG_I(" GL_MAX_COLOR_ATTACHMENTS: %d", caps.MaxColorAttachments);
MobileGL/MG_Util/BackendLoaders/OpenGL/Loader.h
+3
View File
@@ -1102,6 +1102,9 @@ namespace MobileGL {
Int MaxUniformBlockSize = 16384;
Int MaxImageUnits = 8;
Int MaxCombinedImageUniforms = 8;
Int MaxVertexImageUniforms = 0;
Int MaxGeometryImageUniforms = 0;
Int MaxFragmentImageUniforms = 8;
Int MaxComputeImageUniforms = 8;
Int MaxDrawBuffers = 8;
Int MaxColorAttachments = 8;
MobileGL/MG_Util/BackendLoaders/Vulkan/Loader.cpp
+9
View File
@@ -174,6 +174,10 @@ namespace MobileGL::MG_Util::BackendLoader {
VkPhysicalDeviceFeatures supportedFeatures{};
vkGetPhysicalDeviceFeatures(physicalDevice, &supportedFeatures);
caps.SupportsWideLines = supportedFeatures.wideLines == VK_TRUE;
caps.SupportsVertexPipelineStoresAndAtomics =
supportedFeatures.vertexPipelineStoresAndAtomics == VK_TRUE;
caps.SupportsFragmentStoresAndAtomics = supportedFeatures.fragmentStoresAndAtomics == VK_TRUE;
caps.SupportsGeometryShader = supportedFeatures.geometryShader == VK_TRUE;
caps.MaxShaderStorageBlockSize = static_cast<SizeT>(p.limits.maxStorageBufferRange);
const Bool supportsShaderSubgroup = vk.vkGetPhysicalDeviceProperties2 &&
HasUsableShaderSubgroupSupport(subgroupProps);
@@ -256,6 +260,11 @@ namespace MobileGL::MG_Util::BackendLoader {
caps.ViewportBoundsRangeMax = properties.limits.viewportBoundsRange[1];
caps.ViewportSubpixelBits = static_cast<Int>(properties.limits.viewportSubPixelBits);
caps.SupportsWideLines = false;
// This helper only receives properties, not VkPhysicalDeviceFeatures. Leave optional
// stage writes disabled rather than inferring them from descriptor limits alone.
caps.SupportsVertexPipelineStoresAndAtomics = false;
caps.SupportsFragmentStoresAndAtomics = false;
caps.SupportsGeometryShader = false;
caps.MaxShaderStorageBlockSize = static_cast<SizeT>(properties.limits.maxStorageBufferRange);
caps.SupportsShaderSubgroup = false;
caps.SubgroupSize = 0;
MobileGL/MG_Util/BackendLoaders/Vulkan/Loader.h
+6
View File
@@ -67,6 +67,12 @@ namespace MobileGL {
Float ViewportBoundsRangeMax = 0.0f;
Int ViewportSubpixelBits = 0;
Bool SupportsWideLines = false;
// Storage-image descriptors are limited per stage by
// maxPerStageDescriptorStorageImages, but writes/atomics outside compute additionally
// require these core Vulkan features to be enabled on the logical device.
Bool SupportsVertexPipelineStoresAndAtomics = false;
Bool SupportsFragmentStoresAndAtomics = false;
Bool SupportsGeometryShader = false;
SizeT MaxShaderStorageBlockSize = 128 * 1024 * 1024;
Bool SupportsShaderSubgroup = false;
Uint32 SubgroupSize = 0;
MobileGL/MG_Util/ShaderTranspiler/ShaderCompiler.cpp
+155 -3
View File
@@ -6,6 +6,10 @@
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
#define SPV_ENABLE_UTILITY_CODE
#include "glslang/SPIRV/spirv.hpp11"
#undef SPV_ENABLE_UTILITY_CODE
#include "ShaderCompiler.h"
#include "SpirvPasses/EliminateFloatEqualsZeroPass.h"
@@ -19,6 +23,7 @@
#include "spirv-tools/optimizer.hpp"
#include "ShaderSourceProcessor.h"
#include <MG_Backend/BackendObjects.h>
#include <MG_Util/Converters/GLToStr/GLEnumConverter.h>
#include <MG_Util/Converters/GLToGlslang/ProgramEnumConverter.h>
#include <cstdlib>
@@ -26,8 +31,8 @@
namespace MobileGL {
namespace MG_Util {
namespace ShaderTranspiler {
TBuiltInResource& GetTBuiltInResourceInstance() {
static TBuiltInResource Resources{};
TBuiltInResource BuildTBuiltInResource() {
TBuiltInResource Resources{};
Resources.maxLights = 32;
Resources.maxClipPlanes = 6;
Resources.maxTextureUnits = 32;
@@ -122,6 +127,22 @@ namespace MobileGL {
Resources.maxTaskWorkGroupSizeZ_NV = 1;
Resources.maxMeshViewCountNV = 4;
// Resource checking must describe the same backend contract exposed through
// glGetIntegerv. Keeping this copy local also avoids racing on a process-global
// TBuiltInResource when Iris compiles shaders concurrently.
const MG_Backend::DynamicBackendParameters fallbackParameters{};
const auto& activeBackend = MG_Backend::pActiveBackendObject;
const auto& dynamicParameters =
activeBackend ? activeBackend->GetDynamicParameters() : fallbackParameters;
Resources.maxImageUnits = dynamicParameters.MaxImageUnits;
Resources.maxCombinedImageUnitsAndFragmentOutputs =
dynamicParameters.MaxImageUnits + dynamicParameters.MaxDrawBuffers;
Resources.maxVertexImageUniforms = dynamicParameters.MaxVertexImageUniforms;
Resources.maxGeometryImageUniforms = dynamicParameters.MaxGeometryImageUniforms;
Resources.maxFragmentImageUniforms = dynamicParameters.MaxFragmentImageUniforms;
Resources.maxComputeImageUniforms = dynamicParameters.MaxComputeImageUniforms;
Resources.maxCombinedImageUniforms = dynamicParameters.MaxCombinedImageUniforms;
Resources.limits.nonInductiveForLoops = true;
Resources.limits.whileLoops = true;
Resources.limits.doWhileLoops = true;
@@ -167,7 +188,8 @@ namespace MobileGL {
tshader->setAutoMapLocations(true);
tshader->setAutoMapBindings(true);
tshader->setGlobalUniformBlockName(GLOBAL_UBO_NAME);
if (!tshader->parse(&GetTBuiltInResourceInstance(), 460, ECoreProfile,
auto resources = BuildTBuiltInResource();
if (!tshader->parse(&resources, 460, ECoreProfile,
/*forceDefaultVersionAndProfile: */ false,
/*forwardCompatible: */ true, EShMsgDefault)) {
ResultInfo r;
@@ -392,6 +414,136 @@ namespace MobileGL {
return true;
}
bool ShaderCompiler::UseUnformattedFloatStorageImagesForVulkan(
const Vector<Uint32>& inputBinary, Vector<uint32_t>& outputBinary) {
constexpr SizeT kSpirvHeaderWordCount = 5;
outputBinary.clear();
if (inputBinary.size() < kSpirvHeaderWordCount || inputBinary[0] != spv::MagicNumber) {
return false;
}
Vector<Uint32> floatTypeIds;
Vector<Uint32> resultTypeById(inputBinary[3], 0);
Vector<Uint32> pointerPointeeTypeById(inputBinary[3], 0);
Bool hasReadWithoutFormatCapability = false;
Bool hasWriteWithoutFormatCapability = false;
SizeT capabilityInsertOffset = kSpirvHeaderWordCount;
for (SizeT offset = kSpirvHeaderWordCount; offset < inputBinary.size();) {
const Uint32 instructionWord = inputBinary[offset];
const Uint32 wordCount = instructionWord >> 16u;
const auto opcode = static_cast<spv::Op>(instructionWord & 0xffffu);
if (wordCount == 0 || offset + wordCount > inputBinary.size()) {
return false;
}
if (opcode == spv::Op::OpCapability && wordCount >= 2) {
capabilityInsertOffset = offset + wordCount;
const auto capability = static_cast<spv::Capability>(inputBinary[offset + 1]);
hasReadWithoutFormatCapability |=
capability == spv::Capability::StorageImageReadWithoutFormat;
hasWriteWithoutFormatCapability |=
capability == spv::Capability::StorageImageWriteWithoutFormat;
} else if (opcode == spv::Op::OpTypeFloat && wordCount >= 3) {
floatTypeIds.push_back(inputBinary[offset + 1]);
} else if (opcode == spv::Op::OpTypePointer && wordCount >= 4) {
const Uint32 pointerTypeId = inputBinary[offset + 1];
if (pointerTypeId >= pointerPointeeTypeById.size()) {
return false;
}
pointerPointeeTypeById[pointerTypeId] = inputBinary[offset + 3];
}
bool hasResult = false;
bool hasResultType = false;
spv::HasResultAndType(opcode, &hasResult, &hasResultType);
if (hasResult && hasResultType && wordCount >= 3) {
const Uint32 resultTypeId = inputBinary[offset + 1];
const Uint32 resultId = inputBinary[offset + 2];
if (resultId >= resultTypeById.size()) {
return false;
}
resultTypeById[resultId] = resultTypeId;
}
offset += wordCount;
}
// OpImageTexelPointer is the bridge to image atomic instructions. Vulkan requires
// those image types to retain an atomic-compatible declared format, so exclude only
// the exact image types used by an atomic path rather than disabling formatless
// access for unrelated float images in the same module.
Vector<Uint32> atomicImageTypeIds;
for (SizeT offset = kSpirvHeaderWordCount; offset < inputBinary.size();) {
const Uint32 instructionWord = inputBinary[offset];
const Uint32 wordCount = instructionWord >> 16u;
const auto opcode = static_cast<spv::Op>(instructionWord & 0xffffu);
if (opcode == spv::Op::OpImageTexelPointer && wordCount >= 6) {
const Uint32 imageId = inputBinary[offset + 3];
if (imageId >= resultTypeById.size()) {
return false;
}
Uint32 imageTypeId = resultTypeById[imageId];
if (imageTypeId < pointerPointeeTypeById.size() &&
pointerPointeeTypeById[imageTypeId] != 0) {
imageTypeId = pointerPointeeTypeById[imageTypeId];
}
if (imageTypeId != 0 &&
std::find(atomicImageTypeIds.begin(), atomicImageTypeIds.end(), imageTypeId) ==
atomicImageTypeIds.end()) {
atomicImageTypeIds.push_back(imageTypeId);
}
}
offset += wordCount;
}
outputBinary = inputBinary;
Bool hasFloatStorageImage = false;
for (SizeT offset = kSpirvHeaderWordCount; offset < outputBinary.size();) {
const Uint32 instructionWord = outputBinary[offset];
const Uint32 wordCount = instructionWord >> 16u;
const auto opcode = static_cast<spv::Op>(instructionWord & 0xffffu);
// OpTypeImage operands are: result id, sampled type, dim, depth, arrayed,
// multisampled, sampled, image format, and an optional access qualifier.
if (opcode == spv::Op::OpTypeImage && wordCount >= 9) {
const Uint32 imageTypeId = outputBinary[offset + 1];
const Uint32 sampledTypeId = outputBinary[offset + 2];
const Uint32 sampled = outputBinary[offset + 7];
const Bool hasFloatSampledType =
std::find(floatTypeIds.begin(), floatTypeIds.end(), sampledTypeId) != floatTypeIds.end();
const Bool usedByAtomic =
std::find(atomicImageTypeIds.begin(), atomicImageTypeIds.end(), imageTypeId) !=
atomicImageTypeIds.end();
if (sampled == 2 && hasFloatSampledType && !usedByAtomic) {
outputBinary[offset + 8] = static_cast<Uint32>(spv::ImageFormat::Unknown);
hasFloatStorageImage = true;
}
}
offset += wordCount;
}
if (!hasFloatStorageImage) {
return true;
}
Vector<Uint32> addedCapabilities;
const Uint32 capabilityInstruction =
(2u << 16u) | static_cast<Uint32>(spv::Op::OpCapability);
if (!hasReadWithoutFormatCapability) {
addedCapabilities.push_back(capabilityInstruction);
addedCapabilities.push_back(
static_cast<Uint32>(spv::Capability::StorageImageReadWithoutFormat));
}
if (!hasWriteWithoutFormatCapability) {
addedCapabilities.push_back(capabilityInstruction);
addedCapabilities.push_back(
static_cast<Uint32>(spv::Capability::StorageImageWriteWithoutFormat));
}
outputBinary.insert(outputBinary.begin() + static_cast<std::ptrdiff_t>(capabilityInsertOffset),
addedCapabilities.begin(), addedCapabilities.end());
return true;
}
Result<String> ShaderCompiler::DecompileShader(SpvcSession& session) {
spvc_compiler_options options;
session.CreateOptions(&options);
MobileGL/MG_Util/ShaderTranspiler/ShaderCompiler.h
+8
View File
@@ -46,6 +46,14 @@ namespace MobileGL {
// decoration. DirectVulkan only.
static bool DecoratePositionInvariantForVulkan(const Vector<Uint32>& inputBinary,
Vector<uint32_t>& outputBinary);
// Replaces the declared format of float storage images with Unknown and adds the
// matching SPIR-V capabilities. DirectVulkan uses this only when both Vulkan
// shaderStorageImage*WithoutFormat features are enabled, allowing the
// glBindImageTexture format to select the descriptor view at runtime. Integer
// storage images deliberately keep their declared format for GL-compatible bit
// reinterpretation paths (for example, R32F storage accessed as r32ui).
static bool UseUnformattedFloatStorageImagesForVulkan(
const Vector<Uint32>& inputBinary, Vector<uint32_t>& outputBinary);
static Result<String> DecompileShader(SpvcSession& session);
};
} // namespace ShaderTranspiler
MobileGL/MG_Util/ShaderTranspiler/glslang/TMglGlslIoResolver.cpp
+18 -1
View File
@@ -63,7 +63,17 @@ namespace MobileGL {
void TMglGlslIoResolver::reserverStorageSlot(glslang::TVarEntryInfo& ent, TInfoSink& infoSink) {
const glslang::TType& type = ent.symbol->getType();
const glslang::TString& name = ent.symbol->getAccessName();
if (currentStage == EShLangVertex && type.getQualifier().isPipeInput()) {
// OpenGL assigns generic vertex attribute locations only to active inputs. glslang gathers
// both live and dead declarations before mapping, so allowing the default collector to
// reserve a dead vertex input would make it consume a location that an active input should
// reuse. Other stage interfaces still need the default cross-stage matching behavior.
if (!ent.live && currentStage == EShLangVertex && type.getQualifier().isPipeInput()) {
return;
}
// glBindAttribLocation only affects active inputs in the linked program. Applying an API
// binding to an inactive declaration would reserve its slot in glslang's collector and
// incorrectly push an active, automatically mapped input to a different location.
if (ent.live && currentStage == EShLangVertex && type.getQualifier().isPipeInput()) {
auto it = m_explicitVertexIns.find(name.c_str());
if (it != m_explicitVertexIns.end()) {
auto& writableType = ent.symbol->getWritableType();
@@ -94,6 +104,13 @@ namespace MobileGL {
TDefaultGlslIoResolver::reserverStorageSlot(ent, infoSink);
}
int TMglGlslIoResolver::resolveInOutLocation(EShLanguage stage, glslang::TVarEntryInfo& ent) {
if (!ent.live && stage == EShLangVertex && ent.symbol->getType().getQualifier().isPipeInput()) {
return ent.newLocation = -1;
}
return TDefaultGlslIoResolver::resolveInOutLocation(stage, ent);
}
void TMglGlslIoResolver::reserverResourceSlot(glslang::TVarEntryInfo& ent, TInfoSink& infoSink) {
const glslang::TType& type = ent.symbol->getType();
if (m_explicitOpaqueUniformBindings != nullptr && type.getBasicType() == glslang::EbtSampler &&
MobileGL/MG_Util/ShaderTranspiler/glslang/TMglGlslIoResolver.h
+1
View File
@@ -37,6 +37,7 @@ namespace MobileGL {
opaqueUniformBindings) {}
void reserverStorageSlot(glslang::TVarEntryInfo& ent, TInfoSink& infoSink) override;
void reserverResourceSlot(glslang::TVarEntryInfo& ent, TInfoSink& infoSink) override;
int resolveInOutLocation(EShLanguage stage, glslang::TVarEntryInfo& ent) override;
int resolveUniformLocation(EShLanguage stage, glslang::TVarEntryInfo& ent) override;
protected: