Nixyri/godot
1
0
Fork 0
mirror of https://github.com/godotengine/godot.git synced 2024-09-20 14:32:42 +00:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity

Add Metal support for macOS (arm64) and iOS

Browse Source
This commit is contained in:
Stuart Carnie 2024-02-20 05:52:00 +11:00 committed by Rémi Verschelde
parent 826de7976a
commit 2d0165574d
No known key found for this signature in database
GPG Key ID: C3336907360768E1
72 changed files with 67923 additions and 28 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
.github/actions/godot-build/action.yml vendored
View File

@ -7,11 +7,14 @@ inputs:
tests:
description: Unit tests.
default: false
required: false
platform:
description: Target platform.
required: false
sconsflags:
description: Additional SCons flags.
default: ""
required: false
scons-cache:
description: The SCons cache path.
default: "${{ github.workspace }}/.scons-cache/"

5
COPYRIGHT.txt
View File

@ -475,6 +475,11 @@ Comment: RVO2
Copyright: 2016, University of North Carolina at Chapel Hill
License: Apache-2.0
Files: ./thirdparty/spirv-cross/
Comment: SPIRV-Cross
Copyright: 2015-2021, Arm Limited
License: Apache-2.0 or Expat
Files: ./thirdparty/spirv-reflect/
Comment: SPIRV-Reflect
Copyright: 2017-2022, Google Inc.

1
SConstruct
View File

@ -222,6 +222,7 @@ opts.Add(BoolVariable("xaudio2", "Enable the XAudio2 audio driver", False))
opts.Add(BoolVariable("vulkan", "Enable the vulkan rendering driver", True))
opts.Add(BoolVariable("opengl3", "Enable the OpenGL/GLES3 rendering driver", True))
opts.Add(BoolVariable("d3d12", "Enable the Direct3D 12 rendering driver", False))
opts.Add(BoolVariable("metal", "Enable the Metal rendering driver (Apple arm64 only)", False))
opts.Add(BoolVariable("openxr", "Enable the OpenXR driver", True))
opts.Add(BoolVariable("use_volk", "Use the volk library to load the Vulkan loader dynamically", True))
opts.Add(BoolVariable("disable_exceptions", "Force disabling exception handling code", True))

1
core/core_bind.cpp
View File

@ -692,6 +692,7 @@ void OS::_bind_methods() {
BIND_ENUM_CONSTANT(RENDERING_DRIVER_VULKAN);
BIND_ENUM_CONSTANT(RENDERING_DRIVER_OPENGL3);
BIND_ENUM_CONSTANT(RENDERING_DRIVER_D3D12);
BIND_ENUM_CONSTANT(RENDERING_DRIVER_METAL);
BIND_ENUM_CONSTANT(SYSTEM_DIR_DESKTOP);
BIND_ENUM_CONSTANT(SYSTEM_DIR_DCIM);

1
core/core_bind.h
View File

@ -132,6 +132,7 @@ public:
RENDERING_DRIVER_VULKAN,
RENDERING_DRIVER_OPENGL3,
RENDERING_DRIVER_D3D12,
RENDERING_DRIVER_METAL,
};
PackedByteArray get_entropy(int p_bytes);

3
doc/classes/OS.xml
View File

@ -802,6 +802,9 @@
<constant name="RENDERING_DRIVER_D3D12" value="2" enum="RenderingDriver">
The Direct3D 12 rendering driver.
</constant>
<constant name="RENDERING_DRIVER_METAL" value="3" enum="RenderingDriver">
The Metal rendering driver.
</constant>
<constant name="SYSTEM_DIR_DESKTOP" value="0" enum="SystemDir">
Refers to the Desktop directory path.
</constant>

2
drivers/SCsub
View File

@ -33,6 +33,8 @@ if env["opengl3"]:
SConscript("gl_context/SCsub")
SConscript("gles3/SCsub")
SConscript("egl/SCsub")
if env["metal"]:
SConscript("metal/SCsub")
# Core dependencies
SConscript("png/SCsub")

39
drivers/metal/README.md Normal file
View File

@ -0,0 +1,39 @@
# Metal Rendering Device
This document aims to describe the Metal rendering device implementation in Godot.
## Future work / ideas
* Use placement heaps
* Explicit hazard tracking
* [MetalFX] upscaling support?
## Acknowledgments
The Metal rendering owes a lot to the work of the [MoltenVK] project, which is a Vulkan implementation on top of Metal.
In accordance with the Apache 2.0 license, the following copyright notices have been included where applicable:
```
/**************************************************************************/
/* */
/* Portions of this code were derived from MoltenVK. */
/* */
/* Copyright (c) 2015-2023 The Brenwill Workshop Ltd. */
/* (http://www.brenwill.com) */
/* */
/* Licensed under the Apache License, Version 2.0 (the "License"); */
/* you may not use this file except in compliance with the License. */
/* You may obtain a copy of the License at */
/* */
/* http://www.apache.org/licenses/LICENSE-2.0 */
/* */
/* Unless required by applicable law or agreed to in writing, software */
/* distributed under the License is distributed on an "AS IS" BASIS, */
/* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or */
/* implied. See the License for the specific language governing */
/* permissions and limitations under the License. */
/**************************************************************************/
```
[MoltenVK]: https://github.com/KhronosGroup/MoltenVK
[MetalFX]: https://developer.apple.com/documentation/metalfx?language=objc

49
drivers/metal/SCsub Normal file
View File

@ -0,0 +1,49 @@
#!/usr/bin/env python
Import("env")
env_metal = env.Clone()
# Thirdparty source files
thirdparty_obj = []
thirdparty_dir = "#thirdparty/spirv-cross/"
thirdparty_sources = [
"spirv_cfg.cpp",
"spirv_cross_util.cpp",
"spirv_cross.cpp",
"spirv_parser.cpp",
"spirv_msl.cpp",
"spirv_reflect.cpp",
"spirv_glsl.cpp",
"spirv_cross_parsed_ir.cpp",
]
thirdparty_sources = [thirdparty_dir + file for file in thirdparty_sources]
env_metal.Prepend(CPPPATH=[thirdparty_dir, thirdparty_dir + "/include"])
# Must enable exceptions for SPIRV-Cross; otherwise, it will abort the process on errors.
if "-fno-exceptions" in env_metal["CXXFLAGS"]:
env_metal["CXXFLAGS"].remove("-fno-exceptions")
env_metal.Append(CXXFLAGS=["-fexceptions"])
env_thirdparty = env_metal.Clone()
env_thirdparty.disable_warnings()
env_thirdparty.add_source_files(thirdparty_obj, thirdparty_sources)
env_metal.drivers_sources += thirdparty_obj
# Enable C++20 for the Objective-C++ Metal code, which uses C++20 concepts.
if "-std=gnu++17" in env_metal["CXXFLAGS"]:
env_metal["CXXFLAGS"].remove("-std=gnu++17")
env_metal.Append(CXXFLAGS=["-std=c++20"])
# Driver source files
driver_obj = []
env_metal.add_source_files(driver_obj, "*.mm")
env.drivers_sources += driver_obj
# Needed to force rebuilding the driver files when the thirdparty library is updated.
env.Depends(driver_obj, thirdparty_obj)

141
drivers/metal/metal_device_properties.h Normal file
View File

@ -0,0 +1,141 @@
/**************************************************************************/
/* metal_device_properties.h */
/**************************************************************************/
/* This file is part of: */
/* GODOT ENGINE */
/* https://godotengine.org */
/**************************************************************************/
/* Copyright (c) 2014-present Godot Engine contributors (see AUTHORS.md). */
/* Copyright (c) 2007-2014 Juan Linietsky, Ariel Manzur. */
/* */
/* Permission is hereby granted, free of charge, to any person obtaining */
/* a copy of this software and associated documentation files (the */
/* "Software"), to deal in the Software without restriction, including */
/* without limitation the rights to use, copy, modify, merge, publish, */
/* distribute, sublicense, and/or sell copies of the Software, and to */
/* permit persons to whom the Software is furnished to do so, subject to */
/* the following conditions: */
/* */
/* The above copyright notice and this permission notice shall be */
/* included in all copies or substantial portions of the Software. */
/* */
/* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, */
/* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF */
/* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. */
/* IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY */
/* CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, */
/* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE */
/* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. */
/**************************************************************************/
/**************************************************************************/
/* */
/* Portions of this code were derived from MoltenVK. */
/* */
/* Copyright (c) 2015-2023 The Brenwill Workshop Ltd. */
/* (http://www.brenwill.com) */
/* */
/* Licensed under the Apache License, Version 2.0 (the "License"); */
/* you may not use this file except in compliance with the License. */
/* You may obtain a copy of the License at */
/* */
/* http://www.apache.org/licenses/LICENSE-2.0 */
/* */
/* Unless required by applicable law or agreed to in writing, software */
/* distributed under the License is distributed on an "AS IS" BASIS, */
/* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or */
/* implied. See the License for the specific language governing */
/* permissions and limitations under the License. */
/**************************************************************************/
#ifndef METAL_DEVICE_PROPERTIES_H
#define METAL_DEVICE_PROPERTIES_H
#import "servers/rendering/rendering_device.h"
#import <Foundation/Foundation.h>
#import <Metal/Metal.h>
/** The buffer index to use for vertex content. */
const static uint32_t VERT_CONTENT_BUFFER_INDEX = 0;
const static uint32_t MAX_COLOR_ATTACHMENT_COUNT = 8;
typedef NS_OPTIONS(NSUInteger, SampleCount) {
SampleCount1 = (1UL << 0),
SampleCount2 = (1UL << 1),
SampleCount4 = (1UL << 2),
SampleCount8 = (1UL << 3),
SampleCount16 = (1UL << 4),
SampleCount32 = (1UL << 5),
SampleCount64 = (1UL << 6),
};
struct API_AVAILABLE(macos(11.0), ios(14.0)) MetalFeatures {
uint32_t mslVersion;
MTLGPUFamily highestFamily;
MTLLanguageVersion mslVersionEnum;
SampleCount supportedSampleCounts;
long hostMemoryPageSize;
bool layeredRendering;
bool multisampleLayeredRendering;
bool quadPermute; /**< If true, quadgroup permutation functions (vote, ballot, shuffle) are supported in shaders. */
bool simdPermute; /**< If true, SIMD-group permutation functions (vote, ballot, shuffle) are supported in shaders. */
bool simdReduction; /**< If true, SIMD-group reduction functions (arithmetic) are supported in shaders. */
bool tessellationShader; /**< If true, tessellation shaders are supported. */
bool imageCubeArray; /**< If true, image cube arrays are supported. */
};
struct MetalLimits {
uint64_t maxImageArrayLayers;
uint64_t maxFramebufferHeight;
uint64_t maxFramebufferWidth;
uint64_t maxImageDimension1D;
uint64_t maxImageDimension2D;
uint64_t maxImageDimension3D;
uint64_t maxImageDimensionCube;
uint64_t maxViewportDimensionX;
uint64_t maxViewportDimensionY;
MTLSize maxThreadsPerThreadGroup;
MTLSize maxComputeWorkGroupCount;
uint64_t maxBoundDescriptorSets;
uint64_t maxColorAttachments;
uint64_t maxTexturesPerArgumentBuffer;
uint64_t maxSamplersPerArgumentBuffer;
uint64_t maxBuffersPerArgumentBuffer;
uint64_t maxBufferLength;
uint64_t minUniformBufferOffsetAlignment;
uint64_t maxVertexDescriptorLayoutStride;
uint16_t maxViewports;
uint32_t maxPerStageBufferCount; /**< The total number of per-stage Metal buffers available for shader uniform content and attributes. */
uint32_t maxPerStageTextureCount; /**< The total number of per-stage Metal textures available for shader uniform content. */
uint32_t maxPerStageSamplerCount; /**< The total number of per-stage Metal samplers available for shader uniform content. */
uint32_t maxVertexInputAttributes;
uint32_t maxVertexInputBindings;
uint32_t maxVertexInputBindingStride;
uint32_t maxDrawIndexedIndexValue;
uint32_t minSubgroupSize; /**< The minimum number of threads in a SIMD-group. */
uint32_t maxSubgroupSize; /**< The maximum number of threads in a SIMD-group. */
BitField<RDD::ShaderStage> subgroupSupportedShaderStages;
BitField<RD::SubgroupOperations> subgroupSupportedOperations; /**< The subgroup operations supported by the device. */
};
class API_AVAILABLE(macos(11.0), ios(14.0)) MetalDeviceProperties {
private:
void init_features(id<MTLDevice> p_device);
void init_limits(id<MTLDevice> p_device);
public:
MetalFeatures features;
MetalLimits limits;
SampleCount find_nearest_supported_sample_count(RenderingDevice::TextureSamples p_samples) const;
MetalDeviceProperties(id<MTLDevice> p_device);
~MetalDeviceProperties();
private:
static const SampleCount sample_count[RenderingDevice::TextureSamples::TEXTURE_SAMPLES_MAX];
};
#endif // METAL_DEVICE_PROPERTIES_H

327
drivers/metal/metal_device_properties.mm Normal file
View File

@ -0,0 +1,327 @@
/**************************************************************************/
/* metal_device_properties.mm */
/**************************************************************************/
/* This file is part of: */
/* GODOT ENGINE */
/* https://godotengine.org */
/**************************************************************************/
/* Copyright (c) 2014-present Godot Engine contributors (see AUTHORS.md). */
/* Copyright (c) 2007-2014 Juan Linietsky, Ariel Manzur. */
/* */
/* Permission is hereby granted, free of charge, to any person obtaining */
/* a copy of this software and associated documentation files (the */
/* "Software"), to deal in the Software without restriction, including */
/* without limitation the rights to use, copy, modify, merge, publish, */
/* distribute, sublicense, and/or sell copies of the Software, and to */
/* permit persons to whom the Software is furnished to do so, subject to */
/* the following conditions: */
/* */
/* The above copyright notice and this permission notice shall be */
/* included in all copies or substantial portions of the Software. */
/* */
/* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, */
/* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF */
/* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. */
/* IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY */
/* CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, */
/* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE */
/* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. */
/**************************************************************************/
/**************************************************************************/
/* */
/* Portions of this code were derived from MoltenVK. */
/* */
/* Copyright (c) 2015-2023 The Brenwill Workshop Ltd. */
/* (http://www.brenwill.com) */
/* */
/* Licensed under the Apache License, Version 2.0 (the "License"); */
/* you may not use this file except in compliance with the License. */
/* You may obtain a copy of the License at */
/* */
/* http://www.apache.org/licenses/LICENSE-2.0 */
/* */
/* Unless required by applicable law or agreed to in writing, software */
/* distributed under the License is distributed on an "AS IS" BASIS, */
/* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or */
/* implied. See the License for the specific language governing */
/* permissions and limitations under the License. */
/**************************************************************************/
#import "metal_device_properties.h"
#import <Metal/Metal.h>
#import <spirv_cross.hpp>
#import <spirv_msl.hpp>
// Common scaling multipliers.
#define KIBI (1024)
#define MEBI (KIBI * KIBI)
#if (TARGET_OS_OSX && __MAC_OS_X_VERSION_MAX_ALLOWED < 140000) || (TARGET_OS_IOS && __IPHONE_OS_VERSION_MAX_ALLOWED < 170000)
#define MTLGPUFamilyApple9 (MTLGPUFamily)1009
#endif
API_AVAILABLE(macos(11.0), ios(14.0))
MTLGPUFamily &operator--(MTLGPUFamily &p_family) {
p_family = static_cast<MTLGPUFamily>(static_cast<int>(p_family) - 1);
if (p_family < MTLGPUFamilyApple1) {
p_family = MTLGPUFamilyApple9;
}
return p_family;
}
void MetalDeviceProperties::init_features(id<MTLDevice> p_device) {
features = {};
features.highestFamily = MTLGPUFamilyApple1;
for (MTLGPUFamily family = MTLGPUFamilyApple9; family >= MTLGPUFamilyApple1; --family) {
if ([p_device supportsFamily:family]) {
features.highestFamily = family;
break;
}
}
features.hostMemoryPageSize = sysconf(_SC_PAGESIZE);
for (SampleCount sc = SampleCount1; sc <= SampleCount64; sc <<= 1) {
if ([p_device supportsTextureSampleCount:sc]) {
features.supportedSampleCounts |= sc;
}
}
features.layeredRendering = [p_device supportsFamily:MTLGPUFamilyApple5];
features.multisampleLayeredRendering = [p_device supportsFamily:MTLGPUFamilyApple7];
features.tessellationShader = [p_device supportsFamily:MTLGPUFamilyApple3];
features.imageCubeArray = [p_device supportsFamily:MTLGPUFamilyApple3];
features.quadPermute = [p_device supportsFamily:MTLGPUFamilyApple4];
features.simdPermute = [p_device supportsFamily:MTLGPUFamilyApple6];
features.simdReduction = [p_device supportsFamily:MTLGPUFamilyApple7];
MTLCompileOptions *opts = [MTLCompileOptions new];
features.mslVersionEnum = opts.languageVersion; // By default, Metal uses the most recent language version.
#define setMSLVersion(m_maj, m_min) \
features.mslVersion = SPIRV_CROSS_NAMESPACE::CompilerMSL::Options::make_msl_version(m_maj, m_min)
switch (features.mslVersionEnum) {
#if __MAC_OS_X_VERSION_MAX_ALLOWED >= 150000 || __IPHONE_OS_VERSION_MAX_ALLOWED >= 180000
case MTLLanguageVersion3_2:
setMSLVersion(3, 2);
break;
#endif
#if __MAC_OS_X_VERSION_MAX_ALLOWED >= 140000 || __IPHONE_OS_VERSION_MAX_ALLOWED >= 170000
case MTLLanguageVersion3_1:
setMSLVersion(3, 1);
break;
#endif
case MTLLanguageVersion3_0:
setMSLVersion(3, 0);
break;
case MTLLanguageVersion2_4:
setMSLVersion(2, 4);
break;
case MTLLanguageVersion2_3:
setMSLVersion(2, 3);
break;
case MTLLanguageVersion2_2:
setMSLVersion(2, 2);
break;
case MTLLanguageVersion2_1:
setMSLVersion(2, 1);
break;
case MTLLanguageVersion2_0:
setMSLVersion(2, 0);
break;
case MTLLanguageVersion1_2:
setMSLVersion(1, 2);
break;
case MTLLanguageVersion1_1:
setMSLVersion(1, 1);
break;
#if TARGET_OS_IPHONE && !TARGET_OS_MACCATALYST
case MTLLanguageVersion1_0:
setMSLVersion(1, 0);
break;
#endif
}
}
void MetalDeviceProperties::init_limits(id<MTLDevice> p_device) {
using std::max;
using std::min;
// FST: https://developer.apple.com/metal/Metal-Feature-Set-Tables.pdf
// FST: Maximum number of layers per 1D texture array, 2D texture array, or 3D texture.
limits.maxImageArrayLayers = 2048;
if ([p_device supportsFamily:MTLGPUFamilyApple3]) {
// FST: Maximum 2D texture width and height.
limits.maxFramebufferWidth = 16384;
limits.maxFramebufferHeight = 16384;
limits.maxViewportDimensionX = 16384;
limits.maxViewportDimensionY = 16384;
// FST: Maximum 1D texture width.
limits.maxImageDimension1D = 16384;
// FST: Maximum 2D texture width and height.
limits.maxImageDimension2D = 16384;
// FST: Maximum cube map texture width and height.
limits.maxImageDimensionCube = 16384;
} else {
// FST: Maximum 2D texture width and height.
limits.maxFramebufferWidth = 8192;
limits.maxFramebufferHeight = 8192;
limits.maxViewportDimensionX = 8192;
limits.maxViewportDimensionY = 8192;
// FST: Maximum 1D texture width.
limits.maxImageDimension1D = 8192;
// FST: Maximum 2D texture width and height.
limits.maxImageDimension2D = 8192;
// FST: Maximum cube map texture width and height.
limits.maxImageDimensionCube = 8192;
}
// FST: Maximum 3D texture width, height, and depth.
limits.maxImageDimension3D = 2048;
limits.maxThreadsPerThreadGroup = p_device.maxThreadsPerThreadgroup;
// No effective limits.
limits.maxComputeWorkGroupCount = { std::numeric_limits<uint32_t>::max(), std::numeric_limits<uint32_t>::max(), std::numeric_limits<uint32_t>::max() };
// https://github.com/KhronosGroup/MoltenVK/blob/568cc3acc0e2299931fdaecaaa1fc3ec5b4af281/MoltenVK/MoltenVK/GPUObjects/MVKDevice.h#L85
limits.maxBoundDescriptorSets = SPIRV_CROSS_NAMESPACE::kMaxArgumentBuffers;
// FST: Maximum number of color render targets per render pass descriptor.
limits.maxColorAttachments = 8;
// Maximum number of textures the device can access, per stage, from an argument buffer.
if ([p_device supportsFamily:MTLGPUFamilyApple6]) {
limits.maxTexturesPerArgumentBuffer = 1'000'000;
} else if ([p_device supportsFamily:MTLGPUFamilyApple4]) {
limits.maxTexturesPerArgumentBuffer = 96;
} else {
limits.maxTexturesPerArgumentBuffer = 31;
}
// Maximum number of samplers the device can access, per stage, from an argument buffer.
if ([p_device supportsFamily:MTLGPUFamilyApple6]) {
limits.maxSamplersPerArgumentBuffer = 1024;
} else {
limits.maxSamplersPerArgumentBuffer = 16;
}
// Maximum number of buffers the device can access, per stage, from an argument buffer.
if ([p_device supportsFamily:MTLGPUFamilyApple6]) {
limits.maxBuffersPerArgumentBuffer = std::numeric_limits<uint64_t>::max();
} else if ([p_device supportsFamily:MTLGPUFamilyApple4]) {
limits.maxBuffersPerArgumentBuffer = 96;
} else {
limits.maxBuffersPerArgumentBuffer = 31;
}
limits.minSubgroupSize = limits.maxSubgroupSize = 1;
// These values were taken from MoltenVK.
if (features.simdPermute) {
limits.minSubgroupSize = 4;
limits.maxSubgroupSize = 32;
} else if (features.quadPermute) {
limits.minSubgroupSize = limits.maxSubgroupSize = 4;
}
limits.subgroupSupportedShaderStages.set_flag(RDD::ShaderStage::SHADER_STAGE_COMPUTE_BIT);
if (features.tessellationShader) {
limits.subgroupSupportedShaderStages.set_flag(RDD::ShaderStage::SHADER_STAGE_TESSELATION_CONTROL_BIT);
}
limits.subgroupSupportedShaderStages.set_flag(RDD::ShaderStage::SHADER_STAGE_FRAGMENT_BIT);
limits.subgroupSupportedOperations.set_flag(RD::SubgroupOperations::SUBGROUP_BASIC_BIT);
if (features.simdPermute || features.quadPermute) {
limits.subgroupSupportedOperations.set_flag(RD::SubgroupOperations::SUBGROUP_VOTE_BIT);
limits.subgroupSupportedOperations.set_flag(RD::SubgroupOperations::SUBGROUP_BALLOT_BIT);
limits.subgroupSupportedOperations.set_flag(RD::SubgroupOperations::SUBGROUP_SHUFFLE_BIT);
limits.subgroupSupportedOperations.set_flag(RD::SubgroupOperations::SUBGROUP_SHUFFLE_RELATIVE_BIT);
}
if (features.simdReduction) {
limits.subgroupSupportedOperations.set_flag(RD::SubgroupOperations::SUBGROUP_ARITHMETIC_BIT);
}
if (features.quadPermute) {
limits.subgroupSupportedOperations.set_flag(RD::SubgroupOperations::SUBGROUP_QUAD_BIT);
}
limits.maxBufferLength = p_device.maxBufferLength;
// FST: Maximum size of vertex descriptor layout stride.
limits.maxVertexDescriptorLayoutStride = std::numeric_limits<uint64_t>::max();
// Maximum number of viewports.
if ([p_device supportsFamily:MTLGPUFamilyApple5]) {
limits.maxViewports = 16;
} else {
limits.maxViewports = 1;
}
limits.maxPerStageBufferCount = 31;
limits.maxPerStageSamplerCount = 16;
if ([p_device supportsFamily:MTLGPUFamilyApple6]) {
limits.maxPerStageTextureCount = 128;
} else if ([p_device supportsFamily:MTLGPUFamilyApple4]) {
limits.maxPerStageTextureCount = 96;
} else {
limits.maxPerStageTextureCount = 31;
}
limits.maxVertexInputAttributes = 31;
limits.maxVertexInputBindings = 31;
limits.maxVertexInputBindingStride = (2 * KIBI);
#if TARGET_OS_IOS && !TARGET_OS_MACCATALYST
limits.minUniformBufferOffsetAlignment = 64;
#endif
#if TARGET_OS_OSX
// This is Apple Silicon specific.
limits.minUniformBufferOffsetAlignment = 16;
#endif
limits.maxDrawIndexedIndexValue = std::numeric_limits<uint32_t>::max() - 1;
}
MetalDeviceProperties::MetalDeviceProperties(id<MTLDevice> p_device) {
init_features(p_device);
init_limits(p_device);
}
MetalDeviceProperties::~MetalDeviceProperties() {
}
SampleCount MetalDeviceProperties::find_nearest_supported_sample_count(RenderingDevice::TextureSamples p_samples) const {
SampleCount supported = features.supportedSampleCounts;
if (supported & sample_count[p_samples]) {
return sample_count[p_samples];
}
SampleCount requested_sample_count = sample_count[p_samples];
// Find the nearest supported sample count.
while (requested_sample_count > SampleCount1) {
if (supported & requested_sample_count) {
return requested_sample_count;
}
requested_sample_count = (SampleCount)(requested_sample_count >> 1);
}
return SampleCount1;
}
// region static members
const SampleCount MetalDeviceProperties::sample_count[RenderingDevice::TextureSamples::TEXTURE_SAMPLES_MAX] = {
SampleCount1,
SampleCount2,
SampleCount4,
SampleCount8,
SampleCount16,
SampleCount32,
SampleCount64,
};
// endregion

838
drivers/metal/metal_objects.h Normal file
View File

@ -0,0 +1,838 @@
/**************************************************************************/
/* metal_objects.h */
/**************************************************************************/
/* This file is part of: */
/* GODOT ENGINE */
/* https://godotengine.org */
/**************************************************************************/
/* Copyright (c) 2014-present Godot Engine contributors (see AUTHORS.md). */
/* Copyright (c) 2007-2014 Juan Linietsky, Ariel Manzur. */
/* */
/* Permission is hereby granted, free of charge, to any person obtaining */
/* a copy of this software and associated documentation files (the */
/* "Software"), to deal in the Software without restriction, including */
/* without limitation the rights to use, copy, modify, merge, publish, */
/* distribute, sublicense, and/or sell copies of the Software, and to */
/* permit persons to whom the Software is furnished to do so, subject to */
/* the following conditions: */
/* */
/* The above copyright notice and this permission notice shall be */
/* included in all copies or substantial portions of the Software. */
/* */
/* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, */
/* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF */
/* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. */
/* IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY */
/* CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, */
/* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE */
/* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. */
/**************************************************************************/
/**************************************************************************/
/* */
/* Portions of this code were derived from MoltenVK. */
/* */
/* Copyright (c) 2015-2023 The Brenwill Workshop Ltd. */
/* (http://www.brenwill.com) */
/* */
/* Licensed under the Apache License, Version 2.0 (the "License"); */
/* you may not use this file except in compliance with the License. */
/* You may obtain a copy of the License at */
/* */
/* http://www.apache.org/licenses/LICENSE-2.0 */
/* */
/* Unless required by applicable law or agreed to in writing, software */
/* distributed under the License is distributed on an "AS IS" BASIS, */
/* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or */
/* implied. See the License for the specific language governing */
/* permissions and limitations under the License. */
/**************************************************************************/
#ifndef METAL_OBJECTS_H
#define METAL_OBJECTS_H
#import "metal_device_properties.h"
#import "metal_utils.h"
#import "pixel_formats.h"
#import "servers/rendering/rendering_device_driver.h"
#import <Foundation/Foundation.h>
#import <Metal/Metal.h>
#import <QuartzCore/CAMetalLayer.h>
#import <simd/simd.h>
#import <initializer_list>
#import <optional>
#import <spirv.hpp>
// These types can be used in Vector and other containers that use
// pointer operations not supported by ARC.
namespace MTL {
#define MTL_CLASS(name) \
class name { \
public: \
name(id<MTL##name> obj = nil) : m_obj(obj) {} \
operator id<MTL##name>() const { return m_obj; } \
id<MTL##name> m_obj; \
};
MTL_CLASS(Texture)
} //namespace MTL
enum ShaderStageUsage : uint32_t {
None = 0,
Vertex = RDD::SHADER_STAGE_VERTEX_BIT,
Fragment = RDD::SHADER_STAGE_FRAGMENT_BIT,
TesselationControl = RDD::SHADER_STAGE_TESSELATION_CONTROL_BIT,
TesselationEvaluation = RDD::SHADER_STAGE_TESSELATION_EVALUATION_BIT,
Compute = RDD::SHADER_STAGE_COMPUTE_BIT,
};
_FORCE_INLINE_ ShaderStageUsage &operator|=(ShaderStageUsage &p_a, int p_b) {
p_a = ShaderStageUsage(uint32_t(p_a) | uint32_t(p_b));
return p_a;
}
enum class MDCommandBufferStateType {
None,
Render,
Compute,
Blit,
};
enum class MDPipelineType {
None,
Render,
Compute,
};
class MDRenderPass;
class MDPipeline;
class MDRenderPipeline;
class MDComputePipeline;
class MDFrameBuffer;
class RenderingDeviceDriverMetal;
class MDUniformSet;
class MDShader;
#pragma mark - Resource Factory
struct ClearAttKey {
const static uint32_t COLOR_COUNT = MAX_COLOR_ATTACHMENT_COUNT;
const static uint32_t DEPTH_INDEX = COLOR_COUNT;
const static uint32_t STENCIL_INDEX = DEPTH_INDEX + 1;
const static uint32_t ATTACHMENT_COUNT = STENCIL_INDEX + 1;
uint16_t sample_count = 0;
uint16_t pixel_formats[ATTACHMENT_COUNT] = { 0 };
_FORCE_INLINE_ void set_color_format(uint32_t p_idx, MTLPixelFormat p_fmt) { pixel_formats[p_idx] = p_fmt; }
_FORCE_INLINE_ void set_depth_format(MTLPixelFormat p_fmt) { pixel_formats[DEPTH_INDEX] = p_fmt; }
_FORCE_INLINE_ void set_stencil_format(MTLPixelFormat p_fmt) { pixel_formats[STENCIL_INDEX] = p_fmt; }
_FORCE_INLINE_ MTLPixelFormat depth_format() const { return (MTLPixelFormat)pixel_formats[DEPTH_INDEX]; }
_FORCE_INLINE_ MTLPixelFormat stencil_format() const { return (MTLPixelFormat)pixel_formats[STENCIL_INDEX]; }
_FORCE_INLINE_ bool is_enabled(uint32_t p_idx) const { return pixel_formats[p_idx] != 0; }
_FORCE_INLINE_ bool is_depth_enabled() const { return pixel_formats[DEPTH_INDEX] != 0; }
_FORCE_INLINE_ bool is_stencil_enabled() const { return pixel_formats[STENCIL_INDEX] != 0; }
_FORCE_INLINE_ bool operator==(const ClearAttKey &p_rhs) const {
return memcmp(this, &p_rhs, sizeof(ClearAttKey)) == 0;
}
uint32_t hash() const {
uint32_t h = hash_murmur3_one_32(sample_count);
h = hash_murmur3_buffer(pixel_formats, ATTACHMENT_COUNT * sizeof(pixel_formats[0]), h);
return h;
}
};
class API_AVAILABLE(macos(11.0), ios(14.0)) MDResourceFactory {
private:
RenderingDeviceDriverMetal *device_driver;
id<MTLFunction> new_func(NSString *p_source, NSString *p_name, NSError **p_error);
id<MTLFunction> new_clear_vert_func(ClearAttKey &p_key);
id<MTLFunction> new_clear_frag_func(ClearAttKey &p_key);
NSString *get_format_type_string(MTLPixelFormat p_fmt);
public:
id<MTLRenderPipelineState> new_clear_pipeline_state(ClearAttKey &p_key, NSError **p_error);
id<MTLDepthStencilState> new_depth_stencil_state(bool p_use_depth, bool p_use_stencil);
MDResourceFactory(RenderingDeviceDriverMetal *p_device_driver) :
device_driver(p_device_driver) {}
~MDResourceFactory() = default;
};
class API_AVAILABLE(macos(11.0), ios(14.0)) MDResourceCache {
private:
typedef HashMap<ClearAttKey, id<MTLRenderPipelineState>, HashableHasher<ClearAttKey>> HashMap;
std::unique_ptr<MDResourceFactory> resource_factory;
HashMap clear_states;
struct {
id<MTLDepthStencilState> all;
id<MTLDepthStencilState> depth_only;
id<MTLDepthStencilState> stencil_only;
id<MTLDepthStencilState> none;
} clear_depth_stencil_state;
public:
id<MTLRenderPipelineState> get_clear_render_pipeline_state(ClearAttKey &p_key, NSError **p_error);
id<MTLDepthStencilState> get_depth_stencil_state(bool p_use_depth, bool p_use_stencil);
explicit MDResourceCache(RenderingDeviceDriverMetal *p_device_driver) :
resource_factory(new MDResourceFactory(p_device_driver)) {}
~MDResourceCache() = default;
};
class API_AVAILABLE(macos(11.0), ios(14.0)) MDCommandBuffer {
private:
RenderingDeviceDriverMetal *device_driver = nullptr;
id<MTLCommandQueue> queue = nil;
id<MTLCommandBuffer> commandBuffer = nil;
void _end_compute_dispatch();
void _end_blit();
#pragma mark - Render
void _render_set_dirty_state();
void _render_bind_uniform_sets();
static void _populate_vertices(simd::float4 *p_vertices, Size2i p_fb_size, VectorView<Rect2i> p_rects);
static uint32_t _populate_vertices(simd::float4 *p_vertices, uint32_t p_index, Rect2i const &p_rect, Size2i p_fb_size);
void _end_render_pass();
void _render_clear_render_area();
public:
MDCommandBufferStateType type = MDCommandBufferStateType::None;
struct RenderState {
MDRenderPass *pass = nullptr;
MDFrameBuffer *frameBuffer = nullptr;
MDRenderPipeline *pipeline = nullptr;
LocalVector<RDD::RenderPassClearValue> clear_values;
LocalVector<MTLViewport> viewports;
LocalVector<MTLScissorRect> scissors;
std::optional<Color> blend_constants;
uint32_t current_subpass = UINT32_MAX;
Rect2i render_area = {};
bool is_rendering_entire_area = false;
MTLRenderPassDescriptor *desc = nil;
id<MTLRenderCommandEncoder> encoder = nil;
id<MTLBuffer> __unsafe_unretained index_buffer = nil; // Buffer is owned by RDD.
MTLIndexType index_type = MTLIndexTypeUInt16;
LocalVector<id<MTLBuffer> __unsafe_unretained> vertex_buffers;
LocalVector<NSUInteger> vertex_offsets;
// clang-format off
enum DirtyFlag: uint8_t {
DIRTY_NONE = 0b0000'0000,
DIRTY_PIPELINE = 0b0000'0001, //! pipeline state
DIRTY_UNIFORMS = 0b0000'0010, //! uniform sets
DIRTY_DEPTH = 0b0000'0100, //! depth / stenci state
DIRTY_VERTEX = 0b0000'1000, //! vertex buffers
DIRTY_VIEWPORT = 0b0001'0000, //! viewport rectangles
DIRTY_SCISSOR = 0b0010'0000, //! scissor rectangles
DIRTY_BLEND = 0b0100'0000, //! blend state
DIRTY_RASTER = 0b1000'0000, //! encoder state like cull mode
DIRTY_ALL = 0xff,
};
// clang-format on
BitField<DirtyFlag> dirty = DIRTY_NONE;
LocalVector<MDUniformSet *> uniform_sets;
// Bit mask of the uniform sets that are dirty, to prevent redundant binding.
uint64_t uniform_set_mask = 0;
_FORCE_INLINE_ void reset() {
pass = nil;
frameBuffer = nil;
pipeline = nil;
current_subpass = UINT32_MAX;
render_area = {};
is_rendering_entire_area = false;
desc = nil;
encoder = nil;
index_buffer = nil;
index_type = MTLIndexTypeUInt16;
dirty = DIRTY_NONE;
uniform_sets.clear();
uniform_set_mask = 0;
clear_values.clear();
viewports.clear();
scissors.clear();
blend_constants.reset();
vertex_buffers.clear();
vertex_offsets.clear();
}
_FORCE_INLINE_ void mark_viewport_dirty() {
if (viewports.is_empty()) {
return;
}
dirty.set_flag(DirtyFlag::DIRTY_VIEWPORT);
}
_FORCE_INLINE_ void mark_scissors_dirty() {
if (scissors.is_empty()) {
return;
}
dirty.set_flag(DirtyFlag::DIRTY_SCISSOR);
}
_FORCE_INLINE_ void mark_vertex_dirty() {
if (vertex_buffers.is_empty()) {
return;
}
dirty.set_flag(DirtyFlag::DIRTY_VERTEX);
}
_FORCE_INLINE_ void mark_uniforms_dirty(std::initializer_list<uint32_t> l) {
if (uniform_sets.is_empty()) {
return;
}
for (uint32_t i : l) {
if (i < uniform_sets.size() && uniform_sets[i] != nullptr) {
uniform_set_mask |= 1 << i;
}
}
dirty.set_flag(DirtyFlag::DIRTY_UNIFORMS);
}
_FORCE_INLINE_ void mark_uniforms_dirty(void) {
if (uniform_sets.is_empty()) {
return;
}
for (uint32_t i = 0; i < uniform_sets.size(); i++) {
if (uniform_sets[i] != nullptr) {
uniform_set_mask |= 1 << i;
}
}
dirty.set_flag(DirtyFlag::DIRTY_UNIFORMS);
}
MTLScissorRect clip_to_render_area(MTLScissorRect p_rect) const {
uint32_t raLeft = render_area.position.x;
uint32_t raRight = raLeft + render_area.size.width;
uint32_t raBottom = render_area.position.y;
uint32_t raTop = raBottom + render_area.size.height;
p_rect.x = CLAMP(p_rect.x, raLeft, MAX(raRight - 1, raLeft));
p_rect.y = CLAMP(p_rect.y, raBottom, MAX(raTop - 1, raBottom));
p_rect.width = MIN(p_rect.width, raRight - p_rect.x);
p_rect.height = MIN(p_rect.height, raTop - p_rect.y);
return p_rect;
}
Rect2i clip_to_render_area(Rect2i p_rect) const {
int32_t raLeft = render_area.position.x;
int32_t raRight = raLeft + render_area.size.width;
int32_t raBottom = render_area.position.y;
int32_t raTop = raBottom + render_area.size.height;
p_rect.position.x = CLAMP(p_rect.position.x, raLeft, MAX(raRight - 1, raLeft));
p_rect.position.y = CLAMP(p_rect.position.y, raBottom, MAX(raTop - 1, raBottom));
p_rect.size.width = MIN(p_rect.size.width, raRight - p_rect.position.x);
p_rect.size.height = MIN(p_rect.size.height, raTop - p_rect.position.y);
return p_rect;
}
} render;
// State specific for a compute pass.
struct {
MDComputePipeline *pipeline = nullptr;
id<MTLComputeCommandEncoder> encoder = nil;
_FORCE_INLINE_ void reset() {
pipeline = nil;
encoder = nil;
}
} compute;
// State specific to a blit pass.
struct {
id<MTLBlitCommandEncoder> encoder = nil;
_FORCE_INLINE_ void reset() {
encoder = nil;
}
} blit;
_FORCE_INLINE_ id<MTLCommandBuffer> get_command_buffer() const {
return commandBuffer;
}
void begin();
void commit();
void end();
id<MTLBlitCommandEncoder> blit_command_encoder();
void encodeRenderCommandEncoderWithDescriptor(MTLRenderPassDescriptor *p_desc, NSString *p_label);
void bind_pipeline(RDD::PipelineID p_pipeline);
#pragma mark - Render Commands
void render_bind_uniform_set(RDD::UniformSetID p_uniform_set, RDD::ShaderID p_shader, uint32_t p_set_index);
void render_clear_attachments(VectorView<RDD::AttachmentClear> p_attachment_clears, VectorView<Rect2i> p_rects);
void render_set_viewport(VectorView<Rect2i> p_viewports);
void render_set_scissor(VectorView<Rect2i> p_scissors);
void render_set_blend_constants(const Color &p_constants);
void render_begin_pass(RDD::RenderPassID p_render_pass,
RDD::FramebufferID p_frameBuffer,
RDD::CommandBufferType p_cmd_buffer_type,
const Rect2i &p_rect,
VectorView<RDD::RenderPassClearValue> p_clear_values);
void render_next_subpass();
void render_draw(uint32_t p_vertex_count,
uint32_t p_instance_count,
uint32_t p_base_vertex,
uint32_t p_first_instance);
void render_bind_vertex_buffers(uint32_t p_binding_count, const RDD::BufferID *p_buffers, const uint64_t *p_offsets);
void render_bind_index_buffer(RDD::BufferID p_buffer, RDD::IndexBufferFormat p_format, uint64_t p_offset);
void render_draw_indexed(uint32_t p_index_count,
uint32_t p_instance_count,
uint32_t p_first_index,
int32_t p_vertex_offset,
uint32_t p_first_instance);
void render_draw_indexed_indirect(RDD::BufferID p_indirect_buffer, uint64_t p_offset, uint32_t p_draw_count, uint32_t p_stride);
void render_draw_indexed_indirect_count(RDD::BufferID p_indirect_buffer, uint64_t p_offset, RDD::BufferID p_count_buffer, uint64_t p_count_buffer_offset, uint32_t p_max_draw_count, uint32_t p_stride);
void render_draw_indirect(RDD::BufferID p_indirect_buffer, uint64_t p_offset, uint32_t p_draw_count, uint32_t p_stride);
void render_draw_indirect_count(RDD::BufferID p_indirect_buffer, uint64_t p_offset, RDD::BufferID p_count_buffer, uint64_t p_count_buffer_offset, uint32_t p_max_draw_count, uint32_t p_stride);
void render_end_pass();
#pragma mark - Compute Commands
void compute_bind_uniform_set(RDD::UniformSetID p_uniform_set, RDD::ShaderID p_shader, uint32_t p_set_index);
void compute_dispatch(uint32_t p_x_groups, uint32_t p_y_groups, uint32_t p_z_groups);
void compute_dispatch_indirect(RDD::BufferID p_indirect_buffer, uint64_t p_offset);
MDCommandBuffer(id<MTLCommandQueue> p_queue, RenderingDeviceDriverMetal *p_device_driver) :
device_driver(p_device_driver), queue(p_queue) {
type = MDCommandBufferStateType::None;
}
MDCommandBuffer() = default;
};
#if (TARGET_OS_OSX && __MAC_OS_X_VERSION_MAX_ALLOWED < 140000) || (TARGET_OS_IOS && __IPHONE_OS_VERSION_MAX_ALLOWED < 170000)
#define MTLBindingAccess MTLArgumentAccess
#define MTLBindingAccessReadOnly MTLArgumentAccessReadOnly
#define MTLBindingAccessReadWrite MTLArgumentAccessReadWrite
#define MTLBindingAccessWriteOnly MTLArgumentAccessWriteOnly
#endif
struct API_AVAILABLE(macos(11.0), ios(14.0)) BindingInfo {
MTLDataType dataType = MTLDataTypeNone;
uint32_t index = 0;
MTLBindingAccess access = MTLBindingAccessReadOnly;
MTLResourceUsage usage = 0;
MTLTextureType textureType = MTLTextureType2D;
spv::ImageFormat imageFormat = spv::ImageFormatUnknown;
uint32_t arrayLength = 0;
bool isMultisampled = false;
inline MTLArgumentDescriptor *new_argument_descriptor() const {
MTLArgumentDescriptor *desc = MTLArgumentDescriptor.argumentDescriptor;
desc.dataType = dataType;
desc.index = index;
desc.access = access;
desc.textureType = textureType;
desc.arrayLength = arrayLength;
return desc;
}
size_t serialize_size() const {
return sizeof(uint32_t) * 8 /* 8 uint32_t fields */;
}
template <typename W>
void serialize(W &p_writer) const {
p_writer.write((uint32_t)dataType);
p_writer.write(index);
p_writer.write((uint32_t)access);
p_writer.write((uint32_t)usage);
p_writer.write((uint32_t)textureType);
p_writer.write(imageFormat);
p_writer.write(arrayLength);
p_writer.write(isMultisampled);
}
template <typename R>
void deserialize(R &p_reader) {
p_reader.read((uint32_t &)dataType);
p_reader.read(index);
p_reader.read((uint32_t &)access);
p_reader.read((uint32_t &)usage);
p_reader.read((uint32_t &)textureType);
p_reader.read((uint32_t &)imageFormat);
p_reader.read(arrayLength);
p_reader.read(isMultisampled);
}
};
using RDC = RenderingDeviceCommons;
typedef API_AVAILABLE(macos(11.0), ios(14.0)) HashMap<RDC::ShaderStage, BindingInfo> BindingInfoMap;
struct API_AVAILABLE(macos(11.0), ios(14.0)) UniformInfo {
uint32_t binding;
ShaderStageUsage active_stages = None;
BindingInfoMap bindings;
BindingInfoMap bindings_secondary;
};
struct API_AVAILABLE(macos(11.0), ios(14.0)) UniformSet {
LocalVector<UniformInfo> uniforms;
uint32_t buffer_size = 0;
HashMap<RDC::ShaderStage, uint32_t> offsets;
HashMap<RDC::ShaderStage, id<MTLArgumentEncoder>> encoders;
};
class API_AVAILABLE(macos(11.0), ios(14.0)) MDShader {
public:
CharString name;
Vector<UniformSet> sets;
virtual void encode_push_constant_data(VectorView<uint32_t> p_data, MDCommandBuffer *p_cb) = 0;
MDShader(CharString p_name, Vector<UniformSet> p_sets) :
name(p_name), sets(p_sets) {}
virtual ~MDShader() = default;
};
class API_AVAILABLE(macos(11.0), ios(14.0)) MDComputeShader final : public MDShader {
public:
struct {
uint32_t binding = -1;
uint32_t size = 0;
} push_constants;
MTLSize local = {};
id<MTLLibrary> kernel;
#if DEV_ENABLED
CharString kernel_source;
#endif
void encode_push_constant_data(VectorView<uint32_t> p_data, MDCommandBuffer *p_cb) final;
MDComputeShader(CharString p_name, Vector<UniformSet> p_sets, id<MTLLibrary> p_kernel);
~MDComputeShader() override = default;
};
class API_AVAILABLE(macos(11.0), ios(14.0)) MDRenderShader final : public MDShader {
public:
struct {
struct {
int32_t binding = -1;
uint32_t size = 0;
} vert;
struct {
int32_t binding = -1;
uint32_t size = 0;
} frag;
} push_constants;
id<MTLLibrary> vert;
id<MTLLibrary> frag;
#if DEV_ENABLED
CharString vert_source;
CharString frag_source;
#endif
void encode_push_constant_data(VectorView<uint32_t> p_data, MDCommandBuffer *p_cb) final;
MDRenderShader(CharString p_name, Vector<UniformSet> p_sets, id<MTLLibrary> p_vert, id<MTLLibrary> p_frag);
~MDRenderShader() override = default;
};
enum StageResourceUsage : uint32_t {
VertexRead = (MTLResourceUsageRead << RDD::SHADER_STAGE_VERTEX * 2),
VertexWrite = (MTLResourceUsageWrite << RDD::SHADER_STAGE_VERTEX * 2),
FragmentRead = (MTLResourceUsageRead << RDD::SHADER_STAGE_FRAGMENT * 2),
FragmentWrite = (MTLResourceUsageWrite << RDD::SHADER_STAGE_FRAGMENT * 2),
TesselationControlRead = (MTLResourceUsageRead << RDD::SHADER_STAGE_TESSELATION_CONTROL * 2),
TesselationControlWrite = (MTLResourceUsageWrite << RDD::SHADER_STAGE_TESSELATION_CONTROL * 2),
TesselationEvaluationRead = (MTLResourceUsageRead << RDD::SHADER_STAGE_TESSELATION_EVALUATION * 2),
TesselationEvaluationWrite = (MTLResourceUsageWrite << RDD::SHADER_STAGE_TESSELATION_EVALUATION * 2),
ComputeRead = (MTLResourceUsageRead << RDD::SHADER_STAGE_COMPUTE * 2),
ComputeWrite = (MTLResourceUsageWrite << RDD::SHADER_STAGE_COMPUTE * 2),
};
_FORCE_INLINE_ StageResourceUsage &operator|=(StageResourceUsage &p_a, uint32_t p_b) {
p_a = StageResourceUsage(uint32_t(p_a) | p_b);
return p_a;
}
_FORCE_INLINE_ StageResourceUsage stage_resource_usage(RDC::ShaderStage p_stage, MTLResourceUsage p_usage) {
return StageResourceUsage(p_usage << (p_stage * 2));
}
_FORCE_INLINE_ MTLResourceUsage resource_usage_for_stage(StageResourceUsage p_usage, RDC::ShaderStage p_stage) {
return MTLResourceUsage((p_usage >> (p_stage * 2)) & 0b11);
}
template <>
struct HashMapComparatorDefault<RDD::ShaderID> {
static bool compare(const RDD::ShaderID &p_lhs, const RDD::ShaderID &p_rhs) {
return p_lhs.id == p_rhs.id;
}
};
struct BoundUniformSet {
id<MTLBuffer> buffer;
HashMap<id<MTLResource>, StageResourceUsage> bound_resources;
};
class API_AVAILABLE(macos(11.0), ios(14.0)) MDUniformSet {
public:
uint32_t index;
LocalVector<RDD::BoundUniform> uniforms;
HashMap<MDShader *, BoundUniformSet> bound_uniforms;
BoundUniformSet &boundUniformSetForShader(MDShader *p_shader, id<MTLDevice> p_device);
};
enum class MDAttachmentType : uint8_t {
None = 0,
Color = 1 << 0,
Depth = 1 << 1,
Stencil = 1 << 2,
};
_FORCE_INLINE_ MDAttachmentType &operator|=(MDAttachmentType &p_a, MDAttachmentType p_b) {
flags::set(p_a, p_b);
return p_a;
}
_FORCE_INLINE_ bool operator&(MDAttachmentType p_a, MDAttachmentType p_b) {
return uint8_t(p_a) & uint8_t(p_b);
}
struct MDSubpass {
uint32_t subpass_index = 0;
LocalVector<RDD::AttachmentReference> input_references;
LocalVector<RDD::AttachmentReference> color_references;
RDD::AttachmentReference depth_stencil_reference;
LocalVector<RDD::AttachmentReference> resolve_references;
MTLFmtCaps getRequiredFmtCapsForAttachmentAt(uint32_t p_index) const;
};
struct API_AVAILABLE(macos(11.0), ios(14.0)) MDAttachment {
private:
uint32_t index = 0;
uint32_t firstUseSubpassIndex = 0;
uint32_t lastUseSubpassIndex = 0;
public:
MTLPixelFormat format = MTLPixelFormatInvalid;
MDAttachmentType type = MDAttachmentType::None;
MTLLoadAction loadAction = MTLLoadActionDontCare;
MTLStoreAction storeAction = MTLStoreActionDontCare;
MTLLoadAction stencilLoadAction = MTLLoadActionDontCare;
MTLStoreAction stencilStoreAction = MTLStoreActionDontCare;
uint32_t samples = 1;
/*!
* @brief Returns true if this attachment is first used in the given subpass.
* @param p_subpass
* @return
*/
_FORCE_INLINE_ bool isFirstUseOf(MDSubpass const &p_subpass) const {
return p_subpass.subpass_index == firstUseSubpassIndex;
}
/*!
* @brief Returns true if this attachment is last used in the given subpass.
* @param p_subpass
* @return
*/
_FORCE_INLINE_ bool isLastUseOf(MDSubpass const &p_subpass) const {
return p_subpass.subpass_index == lastUseSubpassIndex;
}
void linkToSubpass(MDRenderPass const &p_pass);
MTLStoreAction getMTLStoreAction(MDSubpass const &p_subpass,
bool p_is_rendering_entire_area,
bool p_has_resolve,
bool p_can_resolve,
bool p_is_stencil) const;
bool configureDescriptor(MTLRenderPassAttachmentDescriptor *p_desc,
PixelFormats &p_pf,
MDSubpass const &p_subpass,
id<MTLTexture> p_attachment,
bool p_is_rendering_entire_area,
bool p_has_resolve,
bool p_can_resolve,
bool p_is_stencil) const;
/** Returns whether this attachment should be cleared in the subpass. */
bool shouldClear(MDSubpass const &p_subpass, bool p_is_stencil) const;
};
class API_AVAILABLE(macos(11.0), ios(14.0)) MDRenderPass {
public:
Vector<MDAttachment> attachments;
Vector<MDSubpass> subpasses;
uint32_t get_sample_count() const {
return attachments.is_empty() ? 1 : attachments[0].samples;
}
MDRenderPass(Vector<MDAttachment> &p_attachments, Vector<MDSubpass> &p_subpasses);
};
class API_AVAILABLE(macos(11.0), ios(14.0)) MDPipeline {
public:
MDPipelineType type;
explicit MDPipeline(MDPipelineType p_type) :
type(p_type) {}
virtual ~MDPipeline() = default;
};
class API_AVAILABLE(macos(11.0), ios(14.0)) MDRenderPipeline final : public MDPipeline {
public:
id<MTLRenderPipelineState> state = nil;
id<MTLDepthStencilState> depth_stencil = nil;
uint32_t push_constant_size = 0;
uint32_t push_constant_stages_mask = 0;
SampleCount sample_count = SampleCount1;
struct {
MTLCullMode cull_mode = MTLCullModeNone;
MTLTriangleFillMode fill_mode = MTLTriangleFillModeFill;
MTLDepthClipMode clip_mode = MTLDepthClipModeClip;
MTLWinding winding = MTLWindingClockwise;
MTLPrimitiveType render_primitive = MTLPrimitiveTypePoint;
struct {
bool enabled = false;
} depth_test;
struct {
bool enabled = false;
float depth_bias = 0.0;
float slope_scale = 0.0;
float clamp = 0.0;
_FORCE_INLINE_ void apply(id<MTLRenderCommandEncoder> __unsafe_unretained p_enc) const {
if (!enabled) {
return;
}
[p_enc setDepthBias:depth_bias slopeScale:slope_scale clamp:clamp];
}
} depth_bias;
struct {
bool enabled = false;
uint32_t front_reference = 0;
uint32_t back_reference = 0;
_FORCE_INLINE_ void apply(id<MTLRenderCommandEncoder> __unsafe_unretained p_enc) const {
if (!enabled)
return;
[p_enc setStencilFrontReferenceValue:front_reference backReferenceValue:back_reference];
};
} stencil;
struct {
bool enabled = false;
float r = 0.0;
float g = 0.0;
float b = 0.0;
float a = 0.0;
_FORCE_INLINE_ void apply(id<MTLRenderCommandEncoder> __unsafe_unretained p_enc) const {
//if (!enabled)
// return;
[p_enc setBlendColorRed:r green:g blue:b alpha:a];
};
} blend;
_FORCE_INLINE_ void apply(id<MTLRenderCommandEncoder> __unsafe_unretained p_enc) const {
[p_enc setCullMode:cull_mode];
[p_enc setTriangleFillMode:fill_mode];
[p_enc setDepthClipMode:clip_mode];
[p_enc setFrontFacingWinding:winding];
depth_bias.apply(p_enc);
stencil.apply(p_enc);
blend.apply(p_enc);
}
} raster_state;
MDRenderShader *shader = nil;
MDRenderPipeline() :
MDPipeline(MDPipelineType::Render) {}
~MDRenderPipeline() final = default;
};
class API_AVAILABLE(macos(11.0), ios(14.0)) MDComputePipeline final : public MDPipeline {
public:
id<MTLComputePipelineState> state = nil;
struct {
MTLSize local = {};
} compute_state;
MDComputeShader *shader = nil;
explicit MDComputePipeline(id<MTLComputePipelineState> p_state) :
MDPipeline(MDPipelineType::Compute), state(p_state) {}
~MDComputePipeline() final = default;
};
class API_AVAILABLE(macos(11.0), ios(14.0)) MDFrameBuffer {
public:
Vector<MTL::Texture> textures;
Size2i size;
MDFrameBuffer(Vector<MTL::Texture> p_textures, Size2i p_size) :
textures(p_textures), size(p_size) {}
MDFrameBuffer() {}
virtual ~MDFrameBuffer() = default;
};
// These functions are used to convert between Objective-C objects and
// the RIDs used by Godot, respecting automatic reference counting.
namespace rid {
// Converts an Objective-C object to a pointer, and incrementing the
// reference count.
_FORCE_INLINE_
void *owned(id p_id) {
return (__bridge_retained void *)p_id;
}
#define MAKE_ID(FROM, TO) \
_FORCE_INLINE_ TO make(FROM p_obj) { return TO(owned(p_obj)); }
MAKE_ID(id<MTLTexture>, RDD::TextureID)
MAKE_ID(id<MTLBuffer>, RDD::BufferID)
MAKE_ID(id<MTLSamplerState>, RDD::SamplerID)
MAKE_ID(MTLVertexDescriptor *, RDD::VertexFormatID)
MAKE_ID(id<MTLCommandQueue>, RDD::CommandPoolID)
// Converts a pointer to an Objective-C object without changing the reference count.
_FORCE_INLINE_
auto get(RDD::ID p_id) {
return (p_id.id) ? (__bridge ::id)(void *)p_id.id : nil;
}
// Converts a pointer to an Objective-C object, and decrements the reference count.
_FORCE_INLINE_
auto release(RDD::ID p_id) {
return (__bridge_transfer ::id)(void *)p_id.id;
}
} // namespace rid
#endif // METAL_OBJECTS_H

1380
drivers/metal/metal_objects.mm Normal file
View File

File diff suppressed because it is too large Load Diff

81
drivers/metal/metal_utils.h Normal file
View File

@ -0,0 +1,81 @@
/**************************************************************************/
/* metal_utils.h */
/**************************************************************************/
/* This file is part of: */
/* GODOT ENGINE */
/* https://godotengine.org */
/**************************************************************************/
/* Copyright (c) 2014-present Godot Engine contributors (see AUTHORS.md). */
/* Copyright (c) 2007-2014 Juan Linietsky, Ariel Manzur. */
/* */
/* Permission is hereby granted, free of charge, to any person obtaining */
/* a copy of this software and associated documentation files (the */
/* "Software"), to deal in the Software without restriction, including */
/* without limitation the rights to use, copy, modify, merge, publish, */
/* distribute, sublicense, and/or sell copies of the Software, and to */
/* permit persons to whom the Software is furnished to do so, subject to */
/* the following conditions: */
/* */
/* The above copyright notice and this permission notice shall be */
/* included in all copies or substantial portions of the Software. */
/* */
/* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, */
/* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF */
/* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. */
/* IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY */
/* CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, */
/* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE */
/* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. */
/**************************************************************************/
#ifndef METAL_UTILS_H
#define METAL_UTILS_H
#pragma mark - Boolean flags
namespace flags {
/*! Sets the flags within the value parameter specified by the mask parameter. */
template <typename Tv, typename Tm>
void set(Tv &p_value, Tm p_mask) {
using T = std::underlying_type_t<Tv>;
p_value = static_cast<Tv>(static_cast<T>(p_value) | static_cast<T>(p_mask));
}
/*! Clears the flags within the value parameter specified by the mask parameter. */
template <typename Tv, typename Tm>
void clear(Tv &p_value, Tm p_mask) {
using T = std::underlying_type_t<Tv>;
p_value = static_cast<Tv>(static_cast<T>(p_value) & ~static_cast<T>(p_mask));
}
/*! Returns whether the specified value has any of the bits specified in mask set to 1. */
template <typename Tv, typename Tm>
static constexpr bool any(Tv p_value, const Tm p_mask) { return ((p_value & p_mask) != 0); }
/*! Returns whether the specified value has all of the bits specified in mask set to 1. */
template <typename Tv, typename Tm>
static constexpr bool all(Tv p_value, const Tm p_mask) { return ((p_value & p_mask) == p_mask); }
} //namespace flags
#pragma mark - Alignment and Offsets
static constexpr bool is_power_of_two(uint64_t p_value) {
return p_value && ((p_value & (p_value - 1)) == 0);
}
static constexpr uint64_t round_up_to_alignment(uint64_t p_value, uint64_t p_alignment) {
DEV_ASSERT(is_power_of_two(p_alignment));
if (p_alignment == 0) {
return p_value;
}
uint64_t mask = p_alignment - 1;
uint64_t aligned_value = (p_value + mask) & ~mask;
return aligned_value;
}
#endif // METAL_UTILS_H

416
drivers/metal/pixel_formats.h Normal file
View File

@ -0,0 +1,416 @@
/**************************************************************************/
/* pixel_formats.h */
/**************************************************************************/
/* This file is part of: */
/* GODOT ENGINE */
/* https://godotengine.org */
/**************************************************************************/
/* Copyright (c) 2014-present Godot Engine contributors (see AUTHORS.md). */
/* Copyright (c) 2007-2014 Juan Linietsky, Ariel Manzur. */
/* */
/* Permission is hereby granted, free of charge, to any person obtaining */
/* a copy of this software and associated documentation files (the */
/* "Software"), to deal in the Software without restriction, including */
/* without limitation the rights to use, copy, modify, merge, publish, */
/* distribute, sublicense, and/or sell copies of the Software, and to */
/* permit persons to whom the Software is furnished to do so, subject to */
/* the following conditions: */
/* */
/* The above copyright notice and this permission notice shall be */
/* included in all copies or substantial portions of the Software. */
/* */
/* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, */
/* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF */
/* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. */
/* IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY */
/* CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, */
/* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE */
/* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. */
/**************************************************************************/
/**************************************************************************/
/* */
/* Portions of this code were derived from MoltenVK. */
/* */
/* Copyright (c) 2015-2023 The Brenwill Workshop Ltd. */
/* (http://www.brenwill.com) */
/* */
/* Licensed under the Apache License, Version 2.0 (the "License"); */
/* you may not use this file except in compliance with the License. */
/* You may obtain a copy of the License at */
/* */
/* http://www.apache.org/licenses/LICENSE-2.0 */
/* */
/* Unless required by applicable law or agreed to in writing, software */
/* distributed under the License is distributed on an "AS IS" BASIS, */
/* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or */
/* implied. See the License for the specific language governing */
/* permissions and limitations under the License. */
/**************************************************************************/
#ifndef PIXEL_FORMATS_H
#define PIXEL_FORMATS_H
#pragma clang diagnostic push
#pragma clang diagnostic ignored "-Wdeprecated-declarations"
#import "servers/rendering/rendering_device.h"
#import <Metal/Metal.h>
static const uint32_t _mtlPixelFormatCount = 256;
static const uint32_t _mtlPixelFormatCoreCount = MTLPixelFormatX32_Stencil8 + 2; // The actual last enum value is not available on iOS.
static const uint32_t _mtlVertexFormatCount = MTLVertexFormatHalf + 1;
#pragma mark -
#pragma mark Metal format capabilities
typedef enum : uint16_t {
kMTLFmtCapsNone = 0,
/*! The format can be used in a shader read operation. */
kMTLFmtCapsRead = (1 << 0),
/*! The format can be used in a shader filter operation during sampling. */
kMTLFmtCapsFilter = (1 << 1),
/*! The format can be used in a shader write operation. */
kMTLFmtCapsWrite = (1 << 2),
/*! The format can be used with atomic operations. */
kMTLFmtCapsAtomic = (1 << 3),
/*! The format can be used as a color attachment. */
kMTLFmtCapsColorAtt = (1 << 4),
/*! The format can be used as a depth-stencil attachment. */
kMTLFmtCapsDSAtt = (1 << 5),
/*! The format can be used with blend operations. */
kMTLFmtCapsBlend = (1 << 6),
/*! The format can be used as a destination for multisample antialias (MSAA) data. */
kMTLFmtCapsMSAA = (1 << 7),
/*! The format can be used as a resolve attachment. */
kMTLFmtCapsResolve = (1 << 8),
kMTLFmtCapsVertex = (1 << 9),
kMTLFmtCapsRF = (kMTLFmtCapsRead | kMTLFmtCapsFilter),
kMTLFmtCapsRC = (kMTLFmtCapsRead | kMTLFmtCapsColorAtt),
kMTLFmtCapsRCB = (kMTLFmtCapsRC | kMTLFmtCapsBlend),
kMTLFmtCapsRCM = (kMTLFmtCapsRC | kMTLFmtCapsMSAA),
kMTLFmtCapsRCMB = (kMTLFmtCapsRCM | kMTLFmtCapsBlend),
kMTLFmtCapsRWC = (kMTLFmtCapsRC | kMTLFmtCapsWrite),
kMTLFmtCapsRWCB = (kMTLFmtCapsRWC | kMTLFmtCapsBlend),
kMTLFmtCapsRWCM = (kMTLFmtCapsRWC | kMTLFmtCapsMSAA),
kMTLFmtCapsRWCMB = (kMTLFmtCapsRWCM | kMTLFmtCapsBlend),
kMTLFmtCapsRFCMRB = (kMTLFmtCapsRCMB | kMTLFmtCapsFilter | kMTLFmtCapsResolve),
kMTLFmtCapsRFWCMB = (kMTLFmtCapsRWCMB | kMTLFmtCapsFilter),
kMTLFmtCapsAll = (kMTLFmtCapsRFWCMB | kMTLFmtCapsResolve),
kMTLFmtCapsDRM = (kMTLFmtCapsDSAtt | kMTLFmtCapsRead | kMTLFmtCapsMSAA),
kMTLFmtCapsDRFM = (kMTLFmtCapsDRM | kMTLFmtCapsFilter),
kMTLFmtCapsDRMR = (kMTLFmtCapsDRM | kMTLFmtCapsResolve),
kMTLFmtCapsDRFMR = (kMTLFmtCapsDRFM | kMTLFmtCapsResolve),
kMTLFmtCapsChromaSubsampling = kMTLFmtCapsRF,
kMTLFmtCapsMultiPlanar = kMTLFmtCapsChromaSubsampling,
} MTLFmtCaps;
inline MTLFmtCaps operator|(MTLFmtCaps p_left, MTLFmtCaps p_right) {
return static_cast<MTLFmtCaps>(static_cast<uint32_t>(p_left) | p_right);
}
inline MTLFmtCaps &operator|=(MTLFmtCaps &p_left, MTLFmtCaps p_right) {
return (p_left = p_left | p_right);
}
#pragma mark -
#pragma mark Metal view classes
enum class MTLViewClass : uint8_t {
None,
Color8,
Color16,
Color32,
Color64,
Color128,
PVRTC_RGB_2BPP,
PVRTC_RGB_4BPP,
PVRTC_RGBA_2BPP,
PVRTC_RGBA_4BPP,
EAC_R11,
EAC_RG11,
EAC_RGBA8,
ETC2_RGB8,
ETC2_RGB8A1,
ASTC_4x4,
ASTC_5x4,
ASTC_5x5,
ASTC_6x5,
ASTC_6x6,
ASTC_8x5,
ASTC_8x6,
ASTC_8x8,
ASTC_10x5,
ASTC_10x6,
ASTC_10x8,
ASTC_10x10,
ASTC_12x10,
ASTC_12x12,
BC1_RGBA,
BC2_RGBA,
BC3_RGBA,
BC4_R,
BC5_RG,
BC6H_RGB,
BC7_RGBA,
Depth24_Stencil8,
Depth32_Stencil8,
BGRA10_XR,
BGR10_XR
};
#pragma mark -
#pragma mark Format descriptors
/** Enumerates the data type of a format. */
enum class MTLFormatType {
None, /**< Format type is unknown. */
ColorHalf, /**< A 16-bit floating point color. */
ColorFloat, /**< A 32-bit floating point color. */
ColorInt8, /**< A signed 8-bit integer color. */
ColorUInt8, /**< An unsigned 8-bit integer color. */
ColorInt16, /**< A signed 16-bit integer color. */
ColorUInt16, /**< An unsigned 16-bit integer color. */
ColorInt32, /**< A signed 32-bit integer color. */
ColorUInt32, /**< An unsigned 32-bit integer color. */
DepthStencil, /**< A depth and stencil value. */
Compressed, /**< A block-compressed color. */
};
typedef struct Extent2D {
uint32_t width;
uint32_t height;
} Extent2D;
/** Describes the properties of a DataFormat, including the corresponding Metal pixel and vertex format. */
typedef struct DataFormatDesc {
RD::DataFormat dataFormat;
MTLPixelFormat mtlPixelFormat;
MTLPixelFormat mtlPixelFormatSubstitute;
MTLVertexFormat mtlVertexFormat;
MTLVertexFormat mtlVertexFormatSubstitute;
uint8_t chromaSubsamplingPlaneCount;
uint8_t chromaSubsamplingComponentBits;
Extent2D blockTexelSize;
uint32_t bytesPerBlock;
MTLFormatType formatType;
const char *name;
bool hasReportedSubstitution;
inline double bytesPerTexel() const { return (double)bytesPerBlock / (double)(blockTexelSize.width * blockTexelSize.height); }
inline bool isSupported() const { return (mtlPixelFormat != MTLPixelFormatInvalid || chromaSubsamplingPlaneCount > 1); }
inline bool isSupportedOrSubstitutable() const { return isSupported() || (mtlPixelFormatSubstitute != MTLPixelFormatInvalid); }
inline bool vertexIsSupported() const { return (mtlVertexFormat != MTLVertexFormatInvalid); }
inline bool vertexIsSupportedOrSubstitutable() const { return vertexIsSupported() || (mtlVertexFormatSubstitute != MTLVertexFormatInvalid); }
} DataFormatDesc;
/** Describes the properties of a MTLPixelFormat or MTLVertexFormat. */
typedef struct MTLFormatDesc {
union {
MTLPixelFormat mtlPixelFormat;
MTLVertexFormat mtlVertexFormat;
};
RD::DataFormat dataFormat;
MTLFmtCaps mtlFmtCaps;
MTLViewClass mtlViewClass;
MTLPixelFormat mtlPixelFormatLinear;
const char *name = nullptr;
inline bool isSupported() const { return (mtlPixelFormat != MTLPixelFormatInvalid) && (mtlFmtCaps != kMTLFmtCapsNone); }
} MTLFormatDesc;
class API_AVAILABLE(macos(11.0), ios(14.0)) PixelFormats {
using DataFormat = RD::DataFormat;
public:
/** Returns whether the DataFormat is supported by the GPU bound to this instance. */
bool isSupported(DataFormat p_format);
/** Returns whether the DataFormat is supported by this implementation, or can be substituted by one that is. */
bool isSupportedOrSubstitutable(DataFormat p_format);
/** Returns whether the specified Metal MTLPixelFormat can be used as a depth format. */
_FORCE_INLINE_ bool isDepthFormat(MTLPixelFormat p_format) {
switch (p_format) {
case MTLPixelFormatDepth32Float:
case MTLPixelFormatDepth16Unorm:
case MTLPixelFormatDepth32Float_Stencil8:
#if TARGET_OS_OSX
case MTLPixelFormatDepth24Unorm_Stencil8:
#endif
return true;
default:
return false;
}
}
/** Returns whether the specified Metal MTLPixelFormat can be used as a stencil format. */
_FORCE_INLINE_ bool isStencilFormat(MTLPixelFormat p_format) {
switch (p_format) {
case MTLPixelFormatStencil8:
#if TARGET_OS_OSX
case MTLPixelFormatDepth24Unorm_Stencil8:
case MTLPixelFormatX24_Stencil8:
#endif
case MTLPixelFormatDepth32Float_Stencil8:
case MTLPixelFormatX32_Stencil8:
return true;
default:
return false;
}
}
/** Returns whether the specified Metal MTLPixelFormat is a PVRTC format. */
bool isPVRTCFormat(MTLPixelFormat p_format);
/** Returns the format type corresponding to the specified Godot pixel format, */
MTLFormatType getFormatType(DataFormat p_format);
/** Returns the format type corresponding to the specified Metal MTLPixelFormat, */
MTLFormatType getFormatType(MTLPixelFormat p_formt);
/**
* Returns the Metal MTLPixelFormat corresponding to the specified Godot pixel
* or returns MTLPixelFormatInvalid if no corresponding MTLPixelFormat exists.
*/
MTLPixelFormat getMTLPixelFormat(DataFormat p_format);
/**
* Returns the DataFormat corresponding to the specified Metal MTLPixelFormat,
* or returns DATA_FORMAT_MAX if no corresponding DataFormat exists.
*/
DataFormat getDataFormat(MTLPixelFormat p_format);
/**
* Returns the size, in bytes, of a texel block of the specified Godot pixel.
* For uncompressed formats, the returned value corresponds to the size in bytes of a single texel.
*/
uint32_t getBytesPerBlock(DataFormat p_format);
/**
* Returns the size, in bytes, of a texel block of the specified Metal format.
* For uncompressed formats, the returned value corresponds to the size in bytes of a single texel.
*/
uint32_t getBytesPerBlock(MTLPixelFormat p_format);
/** Returns the number of planes of the specified chroma-subsampling (YCbCr) DataFormat */
uint8_t getChromaSubsamplingPlaneCount(DataFormat p_format);
/** Returns the number of bits per channel of the specified chroma-subsampling (YCbCr) DataFormat */
uint8_t getChromaSubsamplingComponentBits(DataFormat p_format);
/**
* Returns the size, in bytes, of a texel of the specified Godot format.
* The returned value may be fractional for certain compressed formats.
*/
float getBytesPerTexel(DataFormat p_format);
/**
* Returns the size, in bytes, of a texel of the specified Metal format.
* The returned value may be fractional for certain compressed formats.
*/
float getBytesPerTexel(MTLPixelFormat p_format);
/**
* Returns the size, in bytes, of a row of texels of the specified Godot pixel format.
*
* For compressed formats, this takes into consideration the compression block size,
* and p_texels_per_row should specify the width in texels, not blocks. The result is rounded
* up if p_texels_per_row is not an integer multiple of the compression block width.
*/
size_t getBytesPerRow(DataFormat p_format, uint32_t p_texels_per_row);
/**
* Returns the size, in bytes, of a row of texels of the specified Metal format.
*
* For compressed formats, this takes into consideration the compression block size,
* and texelsPerRow should specify the width in texels, not blocks. The result is rounded
* up if texelsPerRow is not an integer multiple of the compression block width.
*/
size_t getBytesPerRow(MTLPixelFormat p_format, uint32_t p_texels_per_row);
/**
* Returns the size, in bytes, of a texture layer of the specified Godot pixel format.
*
* For compressed formats, this takes into consideration the compression block size,
* and p_texel_rows_per_layer should specify the height in texels, not blocks. The result is
* rounded up if p_texel_rows_per_layer is not an integer multiple of the compression block height.
*/
size_t getBytesPerLayer(DataFormat p_format, size_t p_bytes_per_row, uint32_t p_texel_rows_per_layer);
/**
* Returns the size, in bytes, of a texture layer of the specified Metal format.
* For compressed formats, this takes into consideration the compression block size,
* and p_texel_rows_per_layer should specify the height in texels, not blocks. The result is
* rounded up if p_texel_rows_per_layer is not an integer multiple of the compression block height.
*/
size_t getBytesPerLayer(MTLPixelFormat p_format, size_t p_bytes_per_row, uint32_t p_texel_rows_per_layer);
/** Returns the Metal format capabilities supported by the specified Godot format, without substitution. */
MTLFmtCaps getCapabilities(DataFormat p_format, bool p_extended = false);
/** Returns the Metal format capabilities supported by the specified Metal format. */
MTLFmtCaps getCapabilities(MTLPixelFormat p_format, bool p_extended = false);
/**
* Returns the Metal MTLVertexFormat corresponding to the specified
* DataFormat as used as a vertex attribute format.
*/
MTLVertexFormat getMTLVertexFormat(DataFormat p_format);
#pragma mark Construction
explicit PixelFormats(id<MTLDevice> p_device);
protected:
id<MTLDevice> device;
DataFormatDesc &getDataFormatDesc(DataFormat p_format);
DataFormatDesc &getDataFormatDesc(MTLPixelFormat p_format);
MTLFormatDesc &getMTLPixelFormatDesc(MTLPixelFormat p_format);
MTLFormatDesc &getMTLVertexFormatDesc(MTLVertexFormat p_format);
void initDataFormatCapabilities();
void initMTLPixelFormatCapabilities();
void initMTLVertexFormatCapabilities();
void buildMTLFormatMaps();
void buildDFFormatMaps();
void modifyMTLFormatCapabilities();
void modifyMTLFormatCapabilities(id<MTLDevice> p_device);
void addMTLPixelFormatCapabilities(id<MTLDevice> p_device,
MTLFeatureSet p_feature_set,
MTLPixelFormat p_format,
MTLFmtCaps p_caps);
void addMTLPixelFormatCapabilities(id<MTLDevice> p_device,
MTLGPUFamily p_family,
MTLPixelFormat p_format,
MTLFmtCaps p_caps);
void disableMTLPixelFormatCapabilities(MTLPixelFormat p_format,
MTLFmtCaps p_caps);
void disableAllMTLPixelFormatCapabilities(MTLPixelFormat p_format);
void addMTLVertexFormatCapabilities(id<MTLDevice> p_device,
MTLFeatureSet p_feature_set,
MTLVertexFormat p_format,
MTLFmtCaps p_caps);
DataFormatDesc _dataFormatDescriptions[RD::DATA_FORMAT_MAX];
MTLFormatDesc _mtlPixelFormatDescriptions[_mtlPixelFormatCount];
MTLFormatDesc _mtlVertexFormatDescriptions[_mtlVertexFormatCount];
// Most Metal formats have small values and are mapped by simple lookup array.
// Outliers are mapped by a map.
uint16_t _mtlFormatDescIndicesByMTLPixelFormatsCore[_mtlPixelFormatCoreCount];
HashMap<uint32_t, uint32_t> _mtlFormatDescIndicesByMTLPixelFormatsExt;
uint16_t _mtlFormatDescIndicesByMTLVertexFormats[_mtlVertexFormatCount];
};
#pragma clang diagnostic pop
#endif // PIXEL_FORMATS_H

1298
drivers/metal/pixel_formats.mm Normal file
View File

File diff suppressed because it is too large Load Diff

206
drivers/metal/rendering_context_driver_metal.h Normal file
View File

@ -0,0 +1,206 @@
/**************************************************************************/
/* rendering_context_driver_metal.h */
/**************************************************************************/
/* This file is part of: */
/* GODOT ENGINE */
/* https://godotengine.org */
/**************************************************************************/
/* Copyright (c) 2014-present Godot Engine contributors (see AUTHORS.md). */
/* Copyright (c) 2007-2014 Juan Linietsky, Ariel Manzur. */
/* */
/* Permission is hereby granted, free of charge, to any person obtaining */
/* a copy of this software and associated documentation files (the */
/* "Software"), to deal in the Software without restriction, including */
/* without limitation the rights to use, copy, modify, merge, publish, */
/* distribute, sublicense, and/or sell copies of the Software, and to */
/* permit persons to whom the Software is furnished to do so, subject to */
/* the following conditions: */
/* */
/* The above copyright notice and this permission notice shall be */
/* included in all copies or substantial portions of the Software. */
/* */
/* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, */
/* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF */
/* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. */
/* IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY */
/* CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, */
/* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE */
/* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. */
/**************************************************************************/
#ifndef RENDERING_CONTEXT_DRIVER_METAL_H
#define RENDERING_CONTEXT_DRIVER_METAL_H
#ifdef METAL_ENABLED
#import "rendering_device_driver_metal.h"
#import "servers/rendering/rendering_context_driver.h"
#import <CoreGraphics/CGGeometry.h>
#import <Metal/Metal.h>
#import <QuartzCore/CALayer.h>
@class CAMetalLayer;
@protocol CAMetalDrawable;
class PixelFormats;
class MDResourceCache;
class API_AVAILABLE(macos(11.0), ios(14.0)) RenderingContextDriverMetal : public RenderingContextDriver {
protected:
id<MTLDevice> metal_device = nil;
Device device; // There is only one device on Apple Silicon.
public:
Error initialize() final override;
const Device &device_get(uint32_t p_device_index) const final override;
uint32_t device_get_count() const final override;
bool device_supports_present(uint32_t p_device_index, SurfaceID p_surface) const final override { return true; }
RenderingDeviceDriver *driver_create() final override;
void driver_free(RenderingDeviceDriver *p_driver) final override;
SurfaceID surface_create(const void *p_platform_data) final override;
void surface_set_size(SurfaceID p_surface, uint32_t p_width, uint32_t p_height) final override;
void surface_set_vsync_mode(SurfaceID p_surface, DisplayServer::VSyncMode p_vsync_mode) final override;
DisplayServer::VSyncMode surface_get_vsync_mode(SurfaceID p_surface) const final override;
uint32_t surface_get_width(SurfaceID p_surface) const final override;
uint32_t surface_get_height(SurfaceID p_surface) const final override;
void surface_set_needs_resize(SurfaceID p_surface, bool p_needs_resize) final override;
bool surface_get_needs_resize(SurfaceID p_surface) const final override;
void surface_destroy(SurfaceID p_surface) final override;
bool is_debug_utils_enabled() const final override { return true; }
#pragma mark - Metal-specific methods
// Platform-specific data for the Windows embedded in this driver.
struct WindowPlatformData {
CAMetalLayer *__unsafe_unretained layer;
};
class Surface {
protected:
id<MTLDevice> device;
public:
uint32_t width = 0;
uint32_t height = 0;
DisplayServer::VSyncMode vsync_mode = DisplayServer::VSYNC_ENABLED;
bool needs_resize = false;
Surface(id<MTLDevice> p_device) :
device(p_device) {}
virtual ~Surface() = default;
MTLPixelFormat get_pixel_format() const { return MTLPixelFormatBGRA8Unorm; }
virtual Error resize(uint32_t p_desired_framebuffer_count) = 0;
virtual RDD::FramebufferID acquire_next_frame_buffer() = 0;
virtual void present(MDCommandBuffer *p_cmd_buffer) = 0;
};
class SurfaceLayer : public Surface {
CAMetalLayer *__unsafe_unretained layer = nil;
LocalVector<MDFrameBuffer> frame_buffers;
LocalVector<id<MTLDrawable>> drawables;
uint32_t rear = -1;
uint32_t front = 0;
uint32_t count = 0;
public:
SurfaceLayer(CAMetalLayer *p_layer, id<MTLDevice> p_device) :
Surface(p_device), layer(p_layer) {
layer.allowsNextDrawableTimeout = YES;
layer.framebufferOnly = YES;
layer.opaque = OS::get_singleton()->is_layered_allowed() ? NO : YES;
layer.pixelFormat = get_pixel_format();
layer.device = p_device;
}
~SurfaceLayer() override {
layer = nil;
}
Error resize(uint32_t p_desired_framebuffer_count) override final {
if (width == 0 || height == 0) {
// Very likely the window is minimized, don't create a swap chain.
return ERR_SKIP;
}
CGSize drawableSize = CGSizeMake(width, height);
CGSize current = layer.drawableSize;
if (!CGSizeEqualToSize(current, drawableSize)) {
layer.drawableSize = drawableSize;
}
// Metal supports a maximum of 3 drawables.
p_desired_framebuffer_count = MIN(3U, p_desired_framebuffer_count);
layer.maximumDrawableCount = p_desired_framebuffer_count;
#if TARGET_OS_OSX
// Display sync is only supported on macOS.
switch (vsync_mode) {
case DisplayServer::VSYNC_MAILBOX:
case DisplayServer::VSYNC_ADAPTIVE:
case DisplayServer::VSYNC_ENABLED:
layer.displaySyncEnabled = YES;
break;
case DisplayServer::VSYNC_DISABLED:
layer.displaySyncEnabled = NO;
break;
}
#endif
drawables.resize(p_desired_framebuffer_count);
frame_buffers.resize(p_desired_framebuffer_count);
for (uint32_t i = 0; i < p_desired_framebuffer_count; i++) {
// Reserve space for the drawable texture.
frame_buffers[i].textures.resize(1);
}
return OK;
}
RDD::FramebufferID acquire_next_frame_buffer() override final {
if (count == frame_buffers.size()) {
return RDD::FramebufferID();
}
rear = (rear + 1) % frame_buffers.size();
count++;
MDFrameBuffer &frame_buffer = frame_buffers[rear];
frame_buffer.size = Size2i(width, height);
id<CAMetalDrawable> drawable = layer.nextDrawable;
ERR_FAIL_NULL_V_MSG(drawable, RDD::FramebufferID(), "no drawable available");
drawables[rear] = drawable;
frame_buffer.textures.write[0] = drawable.texture;
return RDD::FramebufferID(&frame_buffer);
}
void present(MDCommandBuffer *p_cmd_buffer) override final {
if (count == 0) {
return;
}
// Release texture and drawable.
frame_buffers[front].textures.write[0] = nil;
id<MTLDrawable> drawable = drawables[front];
drawables[front] = nil;
count--;
front = (front + 1) % frame_buffers.size();
[p_cmd_buffer->get_command_buffer() presentDrawable:drawable];
}
};
id<MTLDevice> get_metal_device() const { return metal_device; }
#pragma mark - Initialization
RenderingContextDriverMetal();
~RenderingContextDriverMetal() override;
};
#endif // METAL_ENABLED
#endif // RENDERING_CONTEXT_DRIVER_METAL_H

134
drivers/metal/rendering_context_driver_metal.mm Normal file
View File

@ -0,0 +1,134 @@
/**************************************************************************/
/* rendering_context_driver_metal.mm */
/**************************************************************************/
/* This file is part of: */
/* GODOT ENGINE */
/* https://godotengine.org */
/**************************************************************************/
/* Copyright (c) 2014-present Godot Engine contributors (see AUTHORS.md). */
/* Copyright (c) 2007-2014 Juan Linietsky, Ariel Manzur. */
/* */
/* Permission is hereby granted, free of charge, to any person obtaining */
/* a copy of this software and associated documentation files (the */
/* "Software"), to deal in the Software without restriction, including */
/* without limitation the rights to use, copy, modify, merge, publish, */
/* distribute, sublicense, and/or sell copies of the Software, and to */
/* permit persons to whom the Software is furnished to do so, subject to */
/* the following conditions: */
/* */
/* The above copyright notice and this permission notice shall be */
/* included in all copies or substantial portions of the Software. */
/* */
/* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, */
/* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF */
/* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. */
/* IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY */
/* CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, */
/* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE */
/* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. */
/**************************************************************************/
#import "rendering_context_driver_metal.h"
@protocol MTLDeviceEx <MTLDevice>
#if TARGET_OS_OSX && __MAC_OS_X_VERSION_MAX_ALLOWED < 130300
- (void)setShouldMaximizeConcurrentCompilation:(BOOL)v;
#endif
@end
RenderingContextDriverMetal::RenderingContextDriverMetal() {
}
RenderingContextDriverMetal::~RenderingContextDriverMetal() {
}
Error RenderingContextDriverMetal::initialize() {
metal_device = MTLCreateSystemDefaultDevice();
#if TARGET_OS_OSX
if (@available(macOS 13.3, *)) {
[id<MTLDeviceEx>(metal_device) setShouldMaximizeConcurrentCompilation:YES];
}
#endif
device.type = DEVICE_TYPE_INTEGRATED_GPU;
device.vendor = VENDOR_APPLE;
device.workarounds = Workarounds();
MetalDeviceProperties props(metal_device);
int version = (int)props.features.highestFamily - (int)MTLGPUFamilyApple1 + 1;
device.name = vformat("%s (Apple%d)", metal_device.name.UTF8String, version);
return OK;
}
const RenderingContextDriver::Device &RenderingContextDriverMetal::device_get(uint32_t p_device_index) const {
DEV_ASSERT(p_device_index < 1);
return device;
}
uint32_t RenderingContextDriverMetal::device_get_count() const {
return 1;
}
RenderingDeviceDriver *RenderingContextDriverMetal::driver_create() {
return memnew(RenderingDeviceDriverMetal(this));
}
void RenderingContextDriverMetal::driver_free(RenderingDeviceDriver *p_driver) {
memdelete(p_driver);
}
RenderingContextDriver::SurfaceID RenderingContextDriverMetal::surface_create(const void *p_platform_data) {
const WindowPlatformData *wpd = (const WindowPlatformData *)(p_platform_data);
Surface *surface = memnew(SurfaceLayer(wpd->layer, metal_device));
return SurfaceID(surface);
}
void RenderingContextDriverMetal::surface_set_size(SurfaceID p_surface, uint32_t p_width, uint32_t p_height) {
Surface *surface = (Surface *)(p_surface);
if (surface->width == p_width && surface->height == p_height) {
return;
}
surface->width = p_width;
surface->height = p_height;
surface->needs_resize = true;
}
void RenderingContextDriverMetal::surface_set_vsync_mode(SurfaceID p_surface, DisplayServer::VSyncMode p_vsync_mode) {
Surface *surface = (Surface *)(p_surface);
if (surface->vsync_mode == p_vsync_mode) {
return;
}
surface->vsync_mode = p_vsync_mode;
surface->needs_resize = true;
}
DisplayServer::VSyncMode RenderingContextDriverMetal::surface_get_vsync_mode(SurfaceID p_surface) const {
Surface *surface = (Surface *)(p_surface);
return surface->vsync_mode;
}
uint32_t RenderingContextDriverMetal::surface_get_width(SurfaceID p_surface) const {
Surface *surface = (Surface *)(p_surface);
return surface->width;
}
uint32_t RenderingContextDriverMetal::surface_get_height(SurfaceID p_surface) const {
Surface *surface = (Surface *)(p_surface);
return surface->height;
}
void RenderingContextDriverMetal::surface_set_needs_resize(SurfaceID p_surface, bool p_needs_resize) {
Surface *surface = (Surface *)(p_surface);
surface->needs_resize = p_needs_resize;
}
bool RenderingContextDriverMetal::surface_get_needs_resize(SurfaceID p_surface) const {
Surface *surface = (Surface *)(p_surface);
return surface->needs_resize;
}
void RenderingContextDriverMetal::surface_destroy(SurfaceID p_surface) {
Surface *surface = (Surface *)(p_surface);
memdelete(surface);
}

417
drivers/metal/rendering_device_driver_metal.h Normal file
View File

@ -0,0 +1,417 @@
/**************************************************************************/
/* rendering_device_driver_metal.h */
/**************************************************************************/
/* This file is part of: */
/* GODOT ENGINE */
/* https://godotengine.org */
/**************************************************************************/
/* Copyright (c) 2014-present Godot Engine contributors (see AUTHORS.md). */
/* Copyright (c) 2007-2014 Juan Linietsky, Ariel Manzur. */
/* */
/* Permission is hereby granted, free of charge, to any person obtaining */
/* a copy of this software and associated documentation files (the */
/* "Software"), to deal in the Software without restriction, including */
/* without limitation the rights to use, copy, modify, merge, publish, */
/* distribute, sublicense, and/or sell copies of the Software, and to */
/* permit persons to whom the Software is furnished to do so, subject to */
/* the following conditions: */
/* */
/* The above copyright notice and this permission notice shall be */
/* included in all copies or substantial portions of the Software. */
/* */
/* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, */
/* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF */
/* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. */
/* IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY */
/* CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, */
/* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE */
/* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. */
/**************************************************************************/
#ifndef RENDERING_DEVICE_DRIVER_METAL_H
#define RENDERING_DEVICE_DRIVER_METAL_H
#import "metal_objects.h"
#import "servers/rendering/rendering_device_driver.h"
#import <Metal/Metal.h>
#import <spirv.hpp>
#import <variant>
#ifdef DEBUG_ENABLED
#ifndef _DEBUG
#define _DEBUG
#endif
#endif
class RenderingContextDriverMetal;
class API_AVAILABLE(macos(11.0), ios(14.0)) RenderingDeviceDriverMetal : public RenderingDeviceDriver {
template <typename T>
using Result = std::variant<T, Error>;
#pragma mark - Generic
RenderingContextDriverMetal *context_driver = nullptr;
RenderingContextDriver::Device context_device;
id<MTLDevice> device = nil;
uint32_t version_major = 2;
uint32_t version_minor = 0;
MetalDeviceProperties *metal_device_properties = nullptr;
PixelFormats *pixel_formats = nullptr;
std::unique_ptr<MDResourceCache> resource_cache;
RDD::Capabilities capabilities;
RDD::MultiviewCapabilities multiview_capabilities;
id<MTLBinaryArchive> archive = nil;
uint32_t archive_count = 0;
id<MTLCommandQueue> device_queue = nil;
id<MTLCaptureScope> device_scope = nil;
String pipeline_cache_id;
Error _create_device();
Error _check_capabilities();
public:
Error initialize(uint32_t p_device_index, uint32_t p_frame_count) override final;
#pragma mark - Memory
#pragma mark - Buffers
public:
virtual BufferID buffer_create(uint64_t p_size, BitField<BufferUsageBits> p_usage, MemoryAllocationType p_allocation_type) override final;
virtual bool buffer_set_texel_format(BufferID p_buffer, DataFormat p_format) override final;
virtual void buffer_free(BufferID p_buffer) override final;
virtual uint64_t buffer_get_allocation_size(BufferID p_buffer) override final;
virtual uint8_t *buffer_map(BufferID p_buffer) override final;
virtual void buffer_unmap(BufferID p_buffer) override final;
#pragma mark - Texture
private:
// Returns true if the texture is a valid linear format.
Result<bool> is_valid_linear(TextureFormat const &p_format) const;
void _get_sub_resource(TextureID p_texture, const TextureSubresource &p_subresource, TextureCopyableLayout *r_layout) const;
public:
virtual TextureID texture_create(const TextureFormat &p_format, const TextureView &p_view) override final;
virtual TextureID texture_create_from_extension(uint64_t p_native_texture, TextureType p_type, DataFormat p_format, uint32_t p_array_layers, bool p_depth_stencil) override final;
virtual TextureID texture_create_shared(TextureID p_original_texture, const TextureView &p_view) override final;
virtual TextureID texture_create_shared_from_slice(TextureID p_original_texture, const TextureView &p_view, TextureSliceType p_slice_type, uint32_t p_layer, uint32_t p_layers, uint32_t p_mipmap, uint32_t p_mipmaps) override final;
virtual void texture_free(TextureID p_texture) override final;
virtual uint64_t texture_get_allocation_size(TextureID p_texture) override final;
virtual void texture_get_copyable_layout(TextureID p_texture, const TextureSubresource &p_subresource, TextureCopyableLayout *r_layout) override final;
virtual uint8_t *texture_map(TextureID p_texture, const TextureSubresource &p_subresource) override final;
virtual void texture_unmap(TextureID p_texture) override final;
virtual BitField<TextureUsageBits> texture_get_usages_supported_by_format(DataFormat p_format, bool p_cpu_readable) override final;
virtual bool texture_can_make_shared_with_format(TextureID p_texture, DataFormat p_format, bool &r_raw_reinterpretation) override final;
#pragma mark - Sampler
public:
virtual SamplerID sampler_create(const SamplerState &p_state) final override;
virtual void sampler_free(SamplerID p_sampler) final override;
virtual bool sampler_is_format_supported_for_filter(DataFormat p_format, SamplerFilter p_filter) override final;
#pragma mark - Vertex Array
private:
public:
virtual VertexFormatID vertex_format_create(VectorView<VertexAttribute> p_vertex_attribs) override final;
virtual void vertex_format_free(VertexFormatID p_vertex_format) override final;
#pragma mark - Barriers
virtual void command_pipeline_barrier(
CommandBufferID p_cmd_buffer,
BitField<PipelineStageBits> p_src_stages,
BitField<PipelineStageBits> p_dst_stages,
VectorView<MemoryBarrier> p_memory_barriers,
VectorView<BufferBarrier> p_buffer_barriers,
VectorView<TextureBarrier> p_texture_barriers) override final;
#pragma mark - Fences
private:
struct Fence {
dispatch_semaphore_t semaphore;
Fence() :
semaphore(dispatch_semaphore_create(0)) {}
};
public:
virtual FenceID fence_create() override final;
virtual Error fence_wait(FenceID p_fence) override final;
virtual void fence_free(FenceID p_fence) override final;
#pragma mark - Semaphores
public:
virtual SemaphoreID semaphore_create() override final;
virtual void semaphore_free(SemaphoreID p_semaphore) override final;
#pragma mark - Commands
// ----- QUEUE FAMILY -----
virtual CommandQueueFamilyID command_queue_family_get(BitField<CommandQueueFamilyBits> p_cmd_queue_family_bits, RenderingContextDriver::SurfaceID p_surface = 0) override final;
// ----- QUEUE -----
public:
virtual CommandQueueID command_queue_create(CommandQueueFamilyID p_cmd_queue_family, bool p_identify_as_main_queue = false) override final;
virtual Error command_queue_execute_and_present(CommandQueueID p_cmd_queue, VectorView<SemaphoreID> p_wait_semaphores, VectorView<CommandBufferID> p_cmd_buffers, VectorView<SemaphoreID> p_cmd_semaphores, FenceID p_cmd_fence, VectorView<SwapChainID> p_swap_chains) override final;
virtual void command_queue_free(CommandQueueID p_cmd_queue) override final;
// ----- POOL -----
virtual CommandPoolID command_pool_create(CommandQueueFamilyID p_cmd_queue_family, CommandBufferType p_cmd_buffer_type) override final;
virtual void command_pool_free(CommandPoolID p_cmd_pool) override final;
// ----- BUFFER -----
private:
// Used to maintain references.
Vector<MDCommandBuffer *> command_buffers;
public:
virtual CommandBufferID command_buffer_create(CommandPoolID p_cmd_pool) override final;
virtual bool command_buffer_begin(CommandBufferID p_cmd_buffer) override final;
virtual bool command_buffer_begin_secondary(CommandBufferID p_cmd_buffer, RenderPassID p_render_pass, uint32_t p_subpass, FramebufferID p_framebuffer) override final;
virtual void command_buffer_end(CommandBufferID p_cmd_buffer) override final;
virtual void command_buffer_execute_secondary(CommandBufferID p_cmd_buffer, VectorView<CommandBufferID> p_secondary_cmd_buffers) override final;
#pragma mark - Swapchain
private:
struct SwapChain {
RenderingContextDriver::SurfaceID surface = RenderingContextDriver::SurfaceID();
RenderPassID render_pass;
RDD::DataFormat data_format = DATA_FORMAT_MAX;
SwapChain() :
render_pass(nullptr) {}
};
void _swap_chain_release(SwapChain *p_swap_chain);
void _swap_chain_release_buffers(SwapChain *p_swap_chain);
public:
virtual SwapChainID swap_chain_create(RenderingContextDriver::SurfaceID p_surface) override final;
virtual Error swap_chain_resize(CommandQueueID p_cmd_queue, SwapChainID p_swap_chain, uint32_t p_desired_framebuffer_count) override final;
virtual FramebufferID swap_chain_acquire_framebuffer(CommandQueueID p_cmd_queue, SwapChainID p_swap_chain, bool &r_resize_required) override final;
virtual RenderPassID swap_chain_get_render_pass(SwapChainID p_swap_chain) override final;
virtual DataFormat swap_chain_get_format(SwapChainID p_swap_chain) override final;
virtual void swap_chain_free(SwapChainID p_swap_chain) override final;
#pragma mark - Frame Buffer
virtual FramebufferID framebuffer_create(RenderPassID p_render_pass, VectorView<TextureID> p_attachments, uint32_t p_width, uint32_t p_height) override final;
virtual void framebuffer_free(FramebufferID p_framebuffer) override final;
#pragma mark - Shader
private:
// Serialization types need access to private state.
friend struct ShaderStageData;
friend struct SpecializationConstantData;
friend struct UniformData;
friend struct ShaderBinaryData;
friend struct PushConstantData;
private:
Error _reflect_spirv16(VectorView<ShaderStageSPIRVData> p_spirv, ShaderReflection &r_reflection);
public:
virtual String shader_get_binary_cache_key() override final;
virtual Vector<uint8_t> shader_compile_binary_from_spirv(VectorView<ShaderStageSPIRVData> p_spirv, const String &p_shader_name) override final;
virtual ShaderID shader_create_from_bytecode(const Vector<uint8_t> &p_shader_binary, ShaderDescription &r_shader_desc, String &r_name) override final;
virtual void shader_free(ShaderID p_shader) override final;
#pragma mark - Uniform Set
public:
virtual UniformSetID uniform_set_create(VectorView<BoundUniform> p_uniforms, ShaderID p_shader, uint32_t p_set_index) override final;
virtual void uniform_set_free(UniformSetID p_uniform_set) override final;
#pragma mark - Commands
virtual void command_uniform_set_prepare_for_use(CommandBufferID p_cmd_buffer, UniformSetID p_uniform_set, ShaderID p_shader, uint32_t p_set_index) override final;
#pragma mark Transfer
private:
enum class CopySource {
Buffer,
Texture,
};
void _copy_texture_buffer(CommandBufferID p_cmd_buffer,
CopySource p_source,
TextureID p_texture,
BufferID p_buffer,
VectorView<BufferTextureCopyRegion> p_regions);
public:
virtual void command_clear_buffer(CommandBufferID p_cmd_buffer, BufferID p_buffer, uint64_t p_offset, uint64_t p_size) override final;
virtual void command_copy_buffer(CommandBufferID p_cmd_buffer, BufferID p_src_buffer, BufferID p_dst_buffer, VectorView<BufferCopyRegion> p_regions) override final;
virtual void command_copy_texture(CommandBufferID p_cmd_buffer, TextureID p_src_texture, TextureLayout p_src_texture_layout, TextureID p_dst_texture, TextureLayout p_dst_texture_layout, VectorView<TextureCopyRegion> p_regions) override final;
virtual void command_resolve_texture(CommandBufferID p_cmd_buffer, TextureID p_src_texture, TextureLayout p_src_texture_layout, uint32_t p_src_layer, uint32_t p_src_mipmap, TextureID p_dst_texture, TextureLayout p_dst_texture_layout, uint32_t p_dst_layer, uint32_t p_dst_mipmap) override final;
virtual void command_clear_color_texture(CommandBufferID p_cmd_buffer, TextureID p_texture, TextureLayout p_texture_layout, const Color &p_color, const TextureSubresourceRange &p_subresources) override final;
virtual void command_copy_buffer_to_texture(CommandBufferID p_cmd_buffer, BufferID p_src_buffer, TextureID p_dst_texture, TextureLayout p_dst_texture_layout, VectorView<BufferTextureCopyRegion> p_regions) override final;
virtual void command_copy_texture_to_buffer(CommandBufferID p_cmd_buffer, TextureID p_src_texture, TextureLayout p_src_texture_layout, BufferID p_dst_buffer, VectorView<BufferTextureCopyRegion> p_regions) override final;
#pragma mark Pipeline
private:
Result<id<MTLFunction>> _create_function(id<MTLLibrary> p_library, NSString *p_name, VectorView<PipelineSpecializationConstant> &p_specialization_constants);
public:
virtual void pipeline_free(PipelineID p_pipeline_id) override final;
// ----- BINDING -----
virtual void command_bind_push_constants(CommandBufferID p_cmd_buffer, ShaderID p_shader, uint32_t p_first_index, VectorView<uint32_t> p_data) override final;
// ----- CACHE -----
private:
String _pipeline_get_cache_path() const;
public:
virtual bool pipeline_cache_create(const Vector<uint8_t> &p_data) override final;
virtual void pipeline_cache_free() override final;
virtual size_t pipeline_cache_query_size() override final;
virtual Vector<uint8_t> pipeline_cache_serialize() override final;
#pragma mark Rendering
// ----- SUBPASS -----
virtual RenderPassID render_pass_create(VectorView<Attachment> p_attachments, VectorView<Subpass> p_subpasses, VectorView<SubpassDependency> p_subpass_dependencies, uint32_t p_view_count) override final;
virtual void render_pass_free(RenderPassID p_render_pass) override final;
// ----- COMMANDS -----
public:
virtual void command_begin_render_pass(CommandBufferID p_cmd_buffer, RenderPassID p_render_pass, FramebufferID p_framebuffer, CommandBufferType p_cmd_buffer_type, const Rect2i &p_rect, VectorView<RenderPassClearValue> p_clear_values) override final;
virtual void command_end_render_pass(CommandBufferID p_cmd_buffer) override final;
virtual void command_next_render_subpass(CommandBufferID p_cmd_buffer, CommandBufferType p_cmd_buffer_type) override final;
virtual void command_render_set_viewport(CommandBufferID p_cmd_buffer, VectorView<Rect2i> p_viewports) override final;
virtual void command_render_set_scissor(CommandBufferID p_cmd_buffer, VectorView<Rect2i> p_scissors) override final;
virtual void command_render_clear_attachments(CommandBufferID p_cmd_buffer, VectorView<AttachmentClear> p_attachment_clears, VectorView<Rect2i> p_rects) override final;
// Binding.
virtual void command_bind_render_pipeline(CommandBufferID p_cmd_buffer, PipelineID p_pipeline) override final;
virtual void command_bind_render_uniform_set(CommandBufferID p_cmd_buffer, UniformSetID p_uniform_set, ShaderID p_shader, uint32_t p_set_index) override final;
// Drawing.
virtual void command_render_draw(CommandBufferID p_cmd_buffer, uint32_t p_vertex_count, uint32_t p_instance_count, uint32_t p_base_vertex, uint32_t p_first_instance) override final;
virtual void command_render_draw_indexed(CommandBufferID p_cmd_buffer, uint32_t p_index_count, uint32_t p_instance_count, uint32_t p_first_index, int32_t p_vertex_offset, uint32_t p_first_instance) override final;
virtual void command_render_draw_indexed_indirect(CommandBufferID p_cmd_buffer, BufferID p_indirect_buffer, uint64_t p_offset, uint32_t p_draw_count, uint32_t p_stride) override final;
virtual void command_render_draw_indexed_indirect_count(CommandBufferID p_cmd_buffer, BufferID p_indirect_buffer, uint64_t p_offset, BufferID p_count_buffer, uint64_t p_count_buffer_offset, uint32_t p_max_draw_count, uint32_t p_stride) override final;
virtual void command_render_draw_indirect(CommandBufferID p_cmd_buffer, BufferID p_indirect_buffer, uint64_t p_offset, uint32_t p_draw_count, uint32_t p_stride) override final;
virtual void command_render_draw_indirect_count(CommandBufferID p_cmd_buffer, BufferID p_indirect_buffer, uint64_t p_offset, BufferID p_count_buffer, uint64_t p_count_buffer_offset, uint32_t p_max_draw_count, uint32_t p_stride) override final;
// Buffer binding.
virtual void command_render_bind_vertex_buffers(CommandBufferID p_cmd_buffer, uint32_t p_binding_count, const BufferID *p_buffers, const uint64_t *p_offsets) override final;
virtual void command_render_bind_index_buffer(CommandBufferID p_cmd_buffer, BufferID p_buffer, IndexBufferFormat p_format, uint64_t p_offset) override final;
// Dynamic state.
virtual void command_render_set_blend_constants(CommandBufferID p_cmd_buffer, const Color &p_constants) override final;
virtual void command_render_set_line_width(CommandBufferID p_cmd_buffer, float p_width) override final;
// ----- PIPELINE -----
virtual PipelineID render_pipeline_create(
ShaderID p_shader,
VertexFormatID p_vertex_format,
RenderPrimitive p_render_primitive,
PipelineRasterizationState p_rasterization_state,
PipelineMultisampleState p_multisample_state,
PipelineDepthStencilState p_depth_stencil_state,
PipelineColorBlendState p_blend_state,
VectorView<int32_t> p_color_attachments,
BitField<PipelineDynamicStateFlags> p_dynamic_state,
RenderPassID p_render_pass,
uint32_t p_render_subpass,
VectorView<PipelineSpecializationConstant> p_specialization_constants) override final;
#pragma mark - Compute
// ----- COMMANDS -----
// Binding.
virtual void command_bind_compute_pipeline(CommandBufferID p_cmd_buffer, PipelineID p_pipeline) override final;
virtual void command_bind_compute_uniform_set(CommandBufferID p_cmd_buffer, UniformSetID p_uniform_set, ShaderID p_shader, uint32_t p_set_index) override final;
// Dispatching.
virtual void command_compute_dispatch(CommandBufferID p_cmd_buffer, uint32_t p_x_groups, uint32_t p_y_groups, uint32_t p_z_groups) override final;
virtual void command_compute_dispatch_indirect(CommandBufferID p_cmd_buffer, BufferID p_indirect_buffer, uint64_t p_offset) override final;
// ----- PIPELINE -----
virtual PipelineID compute_pipeline_create(ShaderID p_shader, VectorView<PipelineSpecializationConstant> p_specialization_constants) override final;
#pragma mark - Queries
// ----- TIMESTAMP -----
// Basic.
virtual QueryPoolID timestamp_query_pool_create(uint32_t p_query_count) override final;
virtual void timestamp_query_pool_free(QueryPoolID p_pool_id) override final;
virtual void timestamp_query_pool_get_results(QueryPoolID p_pool_id, uint32_t p_query_count, uint64_t *r_results) override final;
virtual uint64_t timestamp_query_result_to_time(uint64_t p_result) override final;
// Commands.
virtual void command_timestamp_query_pool_reset(CommandBufferID p_cmd_buffer, QueryPoolID p_pool_id, uint32_t p_query_count) override final;
virtual void command_timestamp_write(CommandBufferID p_cmd_buffer, QueryPoolID p_pool_id, uint32_t p_index) override final;
#pragma mark - Labels
virtual void command_begin_label(CommandBufferID p_cmd_buffer, const char *p_label_name, const Color &p_color) override final;
virtual void command_end_label(CommandBufferID p_cmd_buffer) override final;
#pragma mark - Submission
virtual void begin_segment(uint32_t p_frame_index, uint32_t p_frames_drawn) override final;
virtual void end_segment() override final;
#pragma mark - Miscellaneous
virtual void set_object_name(ObjectType p_type, ID p_driver_id, const String &p_name) override final;
virtual uint64_t get_resource_native_handle(DriverResource p_type, ID p_driver_id) override final;
virtual uint64_t get_total_memory_used() override final;
virtual uint64_t limit_get(Limit p_limit) override final;
virtual uint64_t api_trait_get(ApiTrait p_trait) override final;
virtual bool has_feature(Features p_feature) override final;
virtual const MultiviewCapabilities &get_multiview_capabilities() override final;
virtual String get_api_name() const override final { return "Metal"; };
virtual String get_api_version() const override final;
virtual String get_pipeline_cache_uuid() const override final;
virtual const Capabilities &get_capabilities() const override final;
virtual bool is_composite_alpha_supported(CommandQueueID p_queue) const override final;
// Metal-specific.
id<MTLDevice> get_device() const { return device; }
PixelFormats &get_pixel_formats() const { return *pixel_formats; }
MDResourceCache &get_resource_cache() const { return *resource_cache; }
MetalDeviceProperties const &get_device_properties() const { return *metal_device_properties; }
_FORCE_INLINE_ uint32_t get_metal_buffer_index_for_vertex_attribute_binding(uint32_t p_binding) {
return (metal_device_properties->limits.maxPerStageBufferCount - 1) - p_binding;
}
size_t get_texel_buffer_alignment_for_format(RDD::DataFormat p_format) const;
size_t get_texel_buffer_alignment_for_format(MTLPixelFormat p_format) const;
/******************/
RenderingDeviceDriverMetal(RenderingContextDriverMetal *p_context_driver);
~RenderingDeviceDriverMetal();
};
#endif // RENDERING_DEVICE_DRIVER_METAL_H

3883
drivers/metal/rendering_device_driver_metal.mm Normal file
View File

File diff suppressed because it is too large Load Diff

2
editor/editor_node.cpp
View File

@ -5032,6 +5032,8 @@ String EditorNode::_get_system_info() const {
driver_name = "Vulkan";
} else if (driver_name.begins_with("opengl3")) {
driver_name = "GLES3";
} else if (driver_name == "metal") {
driver_name = "Metal";
}
// Join info.

15
main/main.cpp
View File

@ -1935,6 +1935,7 @@ Error Main::setup(const char *execpath, int argc, char *argv[], bool p_second_ph
{
String driver_hints = "";
String driver_hints_with_d3d12 = "";
String driver_hints_with_metal = "";
{
Vector<String> driver_hints_arr;
@ -1947,18 +1948,25 @@ Error Main::setup(const char *execpath, int argc, char *argv[], bool p_second_ph
driver_hints_arr.push_back("d3d12");
#endif
driver_hints_with_d3d12 = String(",").join(driver_hints_arr);
#ifdef METAL_ENABLED
// Make metal the preferred and default driver.
driver_hints_arr.insert(0, "metal");
#endif
driver_hints_with_metal = String(",").join(driver_hints_arr);
}
String default_driver = driver_hints.get_slice(",", 0);
String default_driver_with_d3d12 = driver_hints_with_d3d12.get_slice(",", 0);
String default_driver_with_metal = driver_hints_with_metal.get_slice(",", 0);
// For now everything defaults to vulkan when available. This can change in future updates.
GLOBAL_DEF_RST_NOVAL("rendering/rendering_device/driver", default_driver);
GLOBAL_DEF_RST_NOVAL(PropertyInfo(Variant::STRING, "rendering/rendering_device/driver.windows", PROPERTY_HINT_ENUM, driver_hints_with_d3d12), default_driver_with_d3d12);
GLOBAL_DEF_RST_NOVAL(PropertyInfo(Variant::STRING, "rendering/rendering_device/driver.linuxbsd", PROPERTY_HINT_ENUM, driver_hints), default_driver);
GLOBAL_DEF_RST_NOVAL(PropertyInfo(Variant::STRING, "rendering/rendering_device/driver.android", PROPERTY_HINT_ENUM, driver_hints), default_driver);
GLOBAL_DEF_RST_NOVAL(PropertyInfo(Variant::STRING, "rendering/rendering_device/driver.ios", PROPERTY_HINT_ENUM, driver_hints), default_driver);
GLOBAL_DEF_RST_NOVAL(PropertyInfo(Variant::STRING, "rendering/rendering_device/driver.macos", PROPERTY_HINT_ENUM, driver_hints), default_driver);
GLOBAL_DEF_RST_NOVAL(PropertyInfo(Variant::STRING, "rendering/rendering_device/driver.ios", PROPERTY_HINT_ENUM, driver_hints_with_metal), default_driver_with_metal);
GLOBAL_DEF_RST_NOVAL(PropertyInfo(Variant::STRING, "rendering/rendering_device/driver.macos", PROPERTY_HINT_ENUM, driver_hints_with_metal), default_driver_with_metal);
GLOBAL_DEF_RST("rendering/rendering_device/fallback_to_vulkan", true);
GLOBAL_DEF_RST("rendering/rendering_device/fallback_to_d3d12", true);
@ -2232,6 +2240,9 @@ Error Main::setup(const char *execpath, int argc, char *argv[], bool p_second_ph
#endif
#ifdef D3D12_ENABLED
available_drivers.push_back("d3d12");
#endif
#ifdef METAL_ENABLED
available_drivers.push_back("metal");
#endif
}
#ifdef GLES3_ENABLED

4
modules/glslang/config.py
View File

@ -1,7 +1,7 @@
def can_build(env, platform):
# glslang is only needed when Vulkan or Direct3D 12-based renderers are available,
# glslang is only needed when Vulkan, Direct3D 12 or Metal-based renderers are available,
# as OpenGL doesn't use glslang.
return env["vulkan"] or env["d3d12"]
return env["vulkan"] or env["d3d12"] or env["metal"]
def configure(env):

3
modules/glslang/register_types.cpp
View File

@ -73,6 +73,9 @@ static Vector<uint8_t> _compile_shader_glsl(RenderingDevice::ShaderStage p_stage
// - SPIRV-Reflect won't be able to parse the compute workgroup size.
// - We want to play it safe with NIR-DXIL.
TargetVersion = glslang::EShTargetSpv_1_3;
} else if (capabilities.device_family == RDD::DEVICE_METAL) {
ClientVersion = glslang::EShTargetVulkan_1_1;
TargetVersion = glslang::EShTargetSpv_1_6;
} else {
// once we support other backends we'll need to do something here
if (r_error) {

17
platform/ios/detect.py
View File

@ -51,6 +51,7 @@ def get_flags():
"arch": "arm64",
"target": "template_debug",
"use_volk": False,
"metal": True,
"supported": ["mono"],
"builtin_pcre2_with_jit": False,
}
@ -154,8 +155,22 @@ def configure(env: "SConsEnvironment"):
env.Prepend(CPPPATH=["#platform/ios"])
env.Append(CPPDEFINES=["IOS_ENABLED", "UNIX_ENABLED", "COREAUDIO_ENABLED"])
if env["metal"] and env["arch"] != "arm64":
# Only supported on arm64, so skip it for x86_64 builds.
env["metal"] = False
if env["metal"]:
env.AppendUnique(CPPDEFINES=["METAL_ENABLED", "RD_ENABLED"])
env.Prepend(
CPPPATH=[
"$IOS_SDK_PATH/System/Library/Frameworks/Metal.framework/Headers",
"$IOS_SDK_PATH/System/Library/Frameworks/QuartzCore.framework/Headers",
]
)
env.Prepend(CPPPATH=["#thirdparty/spirv-cross"])
if env["vulkan"]:
env.Append(CPPDEFINES=["VULKAN_ENABLED", "RD_ENABLED"])
env.AppendUnique(CPPDEFINES=["VULKAN_ENABLED", "RD_ENABLED"])
if env["opengl3"]:
env.Append(CPPDEFINES=["GLES3_ENABLED", "GLES_SILENCE_DEPRECATION"])

4
platform/ios/display_server_ios.h
View File

@ -47,6 +47,10 @@
#include <vulkan/vulkan.h>
#endif
#endif // VULKAN_ENABLED
#if defined(METAL_ENABLED)
#include "drivers/metal/rendering_context_driver_metal.h"
#endif // METAL_ENABLED
#endif // RD_ENABLED
#if defined(GLES3_ENABLED)

26
platform/ios/display_server_ios.mm
View File

@ -72,6 +72,13 @@ DisplayServerIOS::DisplayServerIOS(const String &p_rendering_driver, WindowMode
union {
#ifdef VULKAN_ENABLED
RenderingContextDriverVulkanIOS::WindowPlatformData vulkan;
#endif
#ifdef METAL_ENABLED
#pragma clang diagnostic push
#pragma clang diagnostic ignored "-Wunguarded-availability"
// Eliminate "RenderingContextDriverMetal is only available on iOS 14.0 or newer".
RenderingContextDriverMetal::WindowPlatformData metal;
#pragma clang diagnostic pop
#endif
} wpd;
@ -85,7 +92,19 @@ DisplayServerIOS::DisplayServerIOS(const String &p_rendering_driver, WindowMode
rendering_context = memnew(RenderingContextDriverVulkanIOS);
}
#endif
#ifdef METAL_ENABLED
if (rendering_driver == "metal") {
if (@available(iOS 14.0, *)) {
layer = [AppDelegate.viewController.godotView initializeRenderingForDriver:@"metal"];
wpd.metal.layer = (CAMetalLayer *)layer;
rendering_context = memnew(RenderingContextDriverMetal);
} else {
OS::get_singleton()->alert("Metal is only supported on iOS 14.0 and later.");
r_error = ERR_UNAVAILABLE;
return;
}
}
#endif
if (rendering_context) {
if (rendering_context->initialize() != OK) {
ERR_PRINT(vformat("Failed to initialize %s context", rendering_driver));
@ -172,6 +191,11 @@ Vector<String> DisplayServerIOS::get_rendering_drivers_func() {
#if defined(VULKAN_ENABLED)
drivers.push_back("vulkan");
#endif
#if defined(METAL_ENABLED)
if (@available(ios 14.0, *)) {
drivers.push_back("metal");
}
#endif
#if defined(GLES3_ENABLED)
drivers.push_back("opengl3");
#endif

8
platform/ios/export/export_plugin.cpp
View File

@ -282,6 +282,7 @@ void EditorExportPlatformIOS::get_export_options(List<ExportOption> *r_options)
r_options->push_back(ExportOption(PropertyInfo(Variant::STRING, "application/short_version", PROPERTY_HINT_PLACEHOLDER_TEXT, "Leave empty to use project version"), ""));
r_options->push_back(ExportOption(PropertyInfo(Variant::STRING, "application/version", PROPERTY_HINT_PLACEHOLDER_TEXT, "Leave empty to use project version"), ""));
// TODO(sgc): set to iOS 14.0 for Metal
r_options->push_back(ExportOption(PropertyInfo(Variant::STRING, "application/min_ios_version"), "12.0"));
r_options->push_back(ExportOption(PropertyInfo(Variant::STRING, "application/additional_plist_content", PROPERTY_HINT_MULTILINE_TEXT), ""));
@ -2656,6 +2657,13 @@ bool EditorExportPlatformIOS::has_valid_export_configuration(const Ref<EditorExp
}
}
if (GLOBAL_GET("rendering/rendering_device/driver.ios") == "metal") {
float version = p_preset->get("application/min_ios_version").operator String().to_float();
if (version < 14.0) {
err += TTR("Metal renderer require iOS 14+.") + "\n";
}
}
if (!err.is_empty()) {
r_error = err;
}

2
platform/ios/godot_view.mm
View File

@ -71,7 +71,7 @@ static const float earth_gravity = 9.80665;
CALayer<DisplayLayer> *layer;
if ([driverName isEqualToString:@"vulkan"]) {
if ([driverName isEqualToString:@"vulkan"] || [driverName isEqualToString:@"metal"]) {
#if defined(TARGET_OS_SIMULATOR) && TARGET_OS_SIMULATOR
if (@available(iOS 13, *)) {
layer = [GodotMetalLayer layer];

22
platform/macos/detect.py
View File

@ -56,6 +56,7 @@ def get_flags():
return {
"arch": detect_arch(),
"use_volk": False,
"metal": True,
"supported": ["mono"],
}
@ -239,9 +240,22 @@ def configure(env: "SConsEnvironment"):
env.Append(LINKFLAGS=["-rpath", "@executable_path/../Frameworks", "-rpath", "@executable_path"])
if env["metal"] and env["arch"] != "arm64":
# Only supported on arm64, so skip it for x86_64 builds.
env["metal"] = False
extra_frameworks = set()
if env["metal"]:
env.AppendUnique(CPPDEFINES=["METAL_ENABLED", "RD_ENABLED"])
extra_frameworks.add("Metal")
extra_frameworks.add("MetalKit")
env.Prepend(CPPPATH=["#thirdparty/spirv-cross"])
if env["vulkan"]:
env.Append(CPPDEFINES=["VULKAN_ENABLED", "RD_ENABLED"])
env.Append(LINKFLAGS=["-framework", "Metal", "-framework", "IOSurface"])
env.AppendUnique(CPPDEFINES=["VULKAN_ENABLED", "RD_ENABLED"])
extra_frameworks.add("Metal")
extra_frameworks.add("IOSurface")
if not env["use_volk"]:
env.Append(LINKFLAGS=["-lMoltenVK"])
@ -260,3 +274,7 @@ def configure(env: "SConsEnvironment"):
"MoltenVK SDK installation directory not found, use 'vulkan_sdk_path' SCons parameter to specify SDK path."
)
sys.exit(255)
if len(extra_frameworks) > 0:
frameworks = [item for key in extra_frameworks for item in ["-framework", key]]
env.Append(LINKFLAGS=frameworks)

3
platform/macos/display_server_macos.h
View File

@ -47,6 +47,9 @@
#if defined(VULKAN_ENABLED)
#include "rendering_context_driver_vulkan_macos.h"
#endif // VULKAN_ENABLED
#if defined(METAL_ENABLED)
#include "drivers/metal/rendering_context_driver_metal.h"
#endif
#endif // RD_ENABLED
#define BitMap _QDBitMap // Suppress deprecated QuickDraw definition.

20
platform/macos/display_server_macos.mm
View File

@ -138,12 +138,20 @@ DisplayServerMacOS::WindowID DisplayServerMacOS::_create_window(WindowMode p_mod
union {
#ifdef VULKAN_ENABLED
RenderingContextDriverVulkanMacOS::WindowPlatformData vulkan;
#endif
#ifdef METAL_ENABLED
RenderingContextDriverMetal::WindowPlatformData metal;
#endif
} wpd;
#ifdef VULKAN_ENABLED
if (rendering_driver == "vulkan") {
wpd.vulkan.layer_ptr = (CAMetalLayer *const *)&layer;
}
#endif
#ifdef METAL_ENABLED
if (rendering_driver == "metal") {
wpd.metal.layer = (CAMetalLayer *)layer;
}
#endif
Error err = rendering_context->window_create(window_id_counter, &wpd);
ERR_FAIL_COND_V_MSG(err != OK, INVALID_WINDOW_ID, vformat("Can't create a %s context", rendering_driver));
@ -2700,7 +2708,7 @@ void DisplayServerMacOS::window_set_vsync_mode(DisplayServer::VSyncMode p_vsync_
gl_manager_legacy->set_use_vsync(p_vsync_mode != DisplayServer::VSYNC_DISABLED);
}
#endif
#if defined(VULKAN_ENABLED)
#if defined(RD_ENABLED)
if (rendering_context) {
rendering_context->window_set_vsync_mode(p_window, p_vsync_mode);
}
@ -2717,7 +2725,7 @@ DisplayServer::VSyncMode DisplayServerMacOS::window_get_vsync_mode(WindowID p_wi
return (gl_manager_legacy->is_using_vsync() ? DisplayServer::VSyncMode::VSYNC_ENABLED : DisplayServer::VSyncMode::VSYNC_DISABLED);
}
#endif
#if defined(VULKAN_ENABLED)
#if defined(RD_ENABLED)
if (rendering_context) {
return rendering_context->window_get_vsync_mode(p_window);
}
@ -3301,6 +3309,9 @@ Vector<String> DisplayServerMacOS::get_rendering_drivers_func() {
#if defined(VULKAN_ENABLED)
drivers.push_back("vulkan");
#endif
#if defined(METAL_ENABLED)
drivers.push_back("metal");
#endif
#if defined(GLES3_ENABLED)
drivers.push_back("opengl3");
drivers.push_back("opengl3_angle");
@ -3623,6 +3634,11 @@ DisplayServerMacOS::DisplayServerMacOS(const String &p_rendering_driver, WindowM
rendering_context = memnew(RenderingContextDriverVulkanMacOS);
}
#endif
#if defined(METAL_ENABLED)
if (rendering_driver == "metal") {
rendering_context = memnew(RenderingContextDriverMetal);
}
#endif
if (rendering_context) {
if (rendering_context->initialize() != OK) {

1
platform/macos/export/export_plugin.cpp
View File

@ -458,6 +458,7 @@ void EditorExportPlatformMacOS::get_export_options(List<ExportOption> *r_options
r_options->push_back(ExportOption(PropertyInfo(Variant::STRING, "application/additional_plist_content", PROPERTY_HINT_MULTILINE_TEXT), ""));
r_options->push_back(ExportOption(PropertyInfo(Variant::STRING, "xcode/platform_build"), "14C18"));
// TODO(sgc): Need to set appropriate version when using Metal
r_options->push_back(ExportOption(PropertyInfo(Variant::STRING, "xcode/sdk_version"), "13.1"));
r_options->push_back(ExportOption(PropertyInfo(Variant::STRING, "xcode/sdk_build"), "22C55"));
r_options->push_back(ExportOption(PropertyInfo(Variant::STRING, "xcode/sdk_name"), "macosx13.1"));

7
servers/rendering/renderer_rd/cluster_builder_rd.h
View File

@ -185,7 +185,14 @@ private:
};
uint32_t cluster_size = 32;
#if defined(MACOS_ENABLED) || defined(IOS_ENABLED)
// Results in visual artifacts on macOS and iOS when using MSAA and subgroups.
// Using subgroups and disabling MSAA is the optimal solution for now and also works
// with MoltenVK.
bool use_msaa = false;
#else
bool use_msaa = true;
#endif
Divisor divisor = DIVISOR_4;
Size2i screen_size;

9
servers/rendering/renderer_rd/shader_rd.cpp
View File

@ -191,9 +191,14 @@ void ShaderRD::_build_variant_code(StringBuilder &builder, uint32_t p_variant, c
for (const KeyValue<StringName, CharString> &E : p_version->code_sections) {
builder.append(String("#define ") + String(E.key) + "_CODE_USED\n");
}
#if defined(MACOS_ENABLED) || defined(IOS_ENABLED)
builder.append("#define MOLTENVK_USED\n");
#if (defined(MACOS_ENABLED) || defined(IOS_ENABLED))
if (RD::get_singleton()->get_device_capabilities().device_family == RDD::DEVICE_VULKAN) {
builder.append("#define MOLTENVK_USED\n");
}
// Image atomics are supported on Metal 3.1 but no support in MoltenVK or SPIRV-Cross yet.
builder.append("#define NO_IMAGE_ATOMICS\n");
#endif
builder.append(String("#define RENDER_DRIVER_") + OS::get_singleton()->get_current_rendering_driver_name().to_upper() + "\n");
} break;
case StageTemplate::Chunk::TYPE_MATERIAL_UNIFORMS: {

16
servers/rendering/renderer_rd/shaders/environment/volumetric_fog.glsl
View File

@ -34,7 +34,7 @@ layout(push_constant, std430) uniform Params {
}
params;
#ifdef MOLTENVK_USED
#ifdef NO_IMAGE_ATOMICS
layout(set = 1, binding = 1) volatile buffer emissive_only_map_buffer {
uint emissive_only_map[];
};
@ -64,7 +64,7 @@ layout(set = 1, binding = 2, std140) uniform SceneParams {
}
scene_params;
#ifdef MOLTENVK_USED
#ifdef NO_IMAGE_ATOMICS
layout(set = 1, binding = 3) volatile buffer density_only_map_buffer {
uint density_only_map[];
};
@ -117,7 +117,7 @@ void main() {
if (any(greaterThanEqual(pos, scene_params.fog_volume_size))) {
return; //do not compute
}
#ifdef MOLTENVK_USED
#ifdef NO_IMAGE_ATOMICS
uint lpos = pos.z * scene_params.fog_volume_size.x * scene_params.fog_volume_size.y + pos.y * scene_params.fog_volume_size.x + pos.x;
#endif
@ -222,7 +222,7 @@ void main() {
density *= cull_mask;
if (abs(density) > 0.001) {
int final_density = int(density * DENSITY_SCALE);
#ifdef MOLTENVK_USED
#ifdef NO_IMAGE_ATOMICS
atomicAdd(density_only_map[lpos], uint(final_density));
#else
imageAtomicAdd(density_only_map, pos, uint(final_density));
@ -236,7 +236,7 @@ void main() {
uvec3 emission_u = uvec3(emission.r * 511.0, emission.g * 511.0, emission.b * 255.0);
// R and G have 11 bits each and B has 10. Then pack them into a 32 bit uint
uint final_emission = emission_u.r << 21 | emission_u.g << 10 | emission_u.b;
#ifdef MOLTENVK_USED
#ifdef NO_IMAGE_ATOMICS
uint prev_emission = atomicAdd(emissive_only_map[lpos], final_emission);
#else
uint prev_emission = imageAtomicAdd(emissive_only_map, pos, final_emission);
@ -252,7 +252,7 @@ void main() {
if (any(overflowing)) {
uvec3 overflow_factor = mix(uvec3(0), uvec3(2047 << 21, 2047 << 10, 1023), overflowing);
uint force_max = overflow_factor.r | overflow_factor.g | overflow_factor.b;
#ifdef MOLTENVK_USED
#ifdef NO_IMAGE_ATOMICS
atomicOr(emissive_only_map[lpos], force_max);
#else
imageAtomicOr(emissive_only_map, pos, force_max);
@ -267,7 +267,7 @@ void main() {
uvec3 scattering_u = uvec3(scattering.r * 2047.0, scattering.g * 2047.0, scattering.b * 1023.0);
// R and G have 11 bits each and B has 10. Then pack them into a 32 bit uint
uint final_scattering = scattering_u.r << 21 | scattering_u.g << 10 | scattering_u.b;
#ifdef MOLTENVK_USED
#ifdef NO_IMAGE_ATOMICS
uint prev_scattering = atomicAdd(light_only_map[lpos], final_scattering);
#else
uint prev_scattering = imageAtomicAdd(light_only_map, pos, final_scattering);
@ -283,7 +283,7 @@ void main() {
if (any(overflowing)) {
uvec3 overflow_factor = mix(uvec3(0), uvec3(2047 << 21, 2047 << 10, 1023), overflowing);
uint force_max = overflow_factor.r | overflow_factor.g | overflow_factor.b;
#ifdef MOLTENVK_USED
#ifdef NO_IMAGE_ATOMICS
atomicOr(light_only_map[lpos], force_max);
#else
imageAtomicOr(light_only_map, pos, force_max);

12
servers/rendering/renderer_rd/shaders/environment/volumetric_fog_process.glsl
View File

@ -190,7 +190,7 @@ params;
#ifndef MODE_COPY
layout(set = 0, binding = 15) uniform texture3D prev_density_texture;
#ifdef MOLTENVK_USED
#ifdef NO_IMAGE_ATOMICS
layout(set = 0, binding = 16) buffer density_only_map_buffer {
uint density_only_map[];
};
@ -287,7 +287,7 @@ void main() {
if (any(greaterThanEqual(pos, params.fog_volume_size))) {
return; //do not compute
}
#ifdef MOLTENVK_USED
#ifdef NO_IMAGE_ATOMICS
uint lpos = pos.z * params.fog_volume_size.x * params.fog_volume_size.y + pos.y * params.fog_volume_size.x + pos.x;
#endif
@ -353,7 +353,7 @@ void main() {
vec3 total_light = vec3(0.0);
float total_density = params.base_density;
#ifdef MOLTENVK_USED
#ifdef NO_IMAGE_ATOMICS
uint local_density = density_only_map[lpos];
#else
uint local_density = imageLoad(density_only_map, pos).x;
@ -362,7 +362,7 @@ void main() {
total_density += float(int(local_density)) / DENSITY_SCALE;
total_density = max(0.0, total_density);
#ifdef MOLTENVK_USED
#ifdef NO_IMAGE_ATOMICS
uint scattering_u = light_only_map[lpos];
#else
uint scattering_u = imageLoad(light_only_map, pos).x;
@ -370,7 +370,7 @@ void main() {
vec3 scattering = vec3(scattering_u >> 21, (scattering_u << 11) >> 21, scattering_u % 1024) / vec3(2047.0, 2047.0, 1023.0);
scattering += params.base_scattering * params.base_density;
#ifdef MOLTENVK_USED
#ifdef NO_IMAGE_ATOMICS
uint emission_u = emissive_only_map[lpos];
#else
uint emission_u = imageLoad(emissive_only_map, pos).x;
@ -710,7 +710,7 @@ void main() {
final_density = mix(final_density, reprojected_density, reproject_amount);
imageStore(density_map, pos, final_density);
#ifdef MOLTENVK_USED
#ifdef NO_IMAGE_ATOMICS
density_only_map[lpos] = 0;
light_only_map[lpos] = 0;
emissive_only_map[lpos] = 0;

2
servers/rendering/renderer_rd/shaders/forward_clustered/scene_forward_clustered.glsl
View File

@ -2374,7 +2374,7 @@ void fragment_shader(in SceneData scene_data) {
}
}
#ifdef MOLTENVK_USED
#ifdef NO_IMAGE_ATOMICS
imageStore(geom_facing_grid, grid_pos, uvec4(imageLoad(geom_facing_grid, grid_pos).r | facing_bits)); //store facing bits
#else
imageAtomicOr(geom_facing_grid, grid_pos, facing_bits); //store facing bits

2
servers/rendering/rendering_device.cpp
View File

@ -2826,6 +2826,7 @@ RID RenderingDevice::uniform_set_create(const Vector<Uniform> &p_uniforms, RID p
for (int j = 0; j < (int)uniform_count; j++) {
if (uniforms[j].binding == set_uniform.binding) {
uniform_idx = j;
break;
}
}
ERR_FAIL_COND_V_MSG(uniform_idx == -1, RID(),
@ -3240,6 +3241,7 @@ RID RenderingDevice::render_pipeline_create(RID p_shader, FramebufferFormatID p_
for (int j = 0; j < vd.vertex_formats.size(); j++) {
if (vd.vertex_formats[j].location == i) {
found = true;
break;
}
}

1
servers/rendering/rendering_device_driver.h
View File

@ -759,6 +759,7 @@ public:
DEVICE_OPENGL,
DEVICE_VULKAN,
DEVICE_DIRECTX,
DEVICE_METAL,
};
struct Capabilities {

16
thirdparty/README.md vendored
View File

@ -827,6 +827,22 @@ and solve conflicts and also enrich the feature set originally
proposed by these libraries and better integrate them with Godot.
## spirv-cross
- Upstream: https://github.com/KhronosGroup/SPIRV-Cross
- Version: vulkan-sdk-1.3.290.0 (5d127b917f080c6f052553c47170ec0ba702e54f, 2024)
- License: Apache 2.0
Files extracted from upstream source:
- All `.cpp`, `.hpp` and `.h` files, minus `main.cpp`, `spirv_cross_c.*`, `spirv_hlsl.*`, `spirv_cpp.*`
- `include/` folder
- `LICENSE` and `LICENSES/` folder, minus `CC-BY-4.0.txt`
Versions of this SDK do not have to match the `vulkan` section, as this SDK is required
to generate Metal source from Vulkan SPIR-V.
## spirv-reflect
- Upstream: https://github.com/KhronosGroup/SPIRV-Reflect

114
thirdparty/spirv-cross/GLSL.std.450.h vendored Normal file
View File

@ -0,0 +1,114 @@
/*
* Copyright 2014-2016,2021 The Khronos Group, Inc.
* SPDX-License-Identifier: MIT
*
* MODIFICATIONS TO THIS FILE MAY MEAN IT NO LONGER ACCURATELY REFLECTS KHRONOS
* STANDARDS. THE UNMODIFIED, NORMATIVE VERSIONS OF KHRONOS SPECIFICATIONS AND
* HEADER INFORMATION ARE LOCATED AT https://www.khronos.org/registry/
*/
#ifndef GLSLstd450_H
#define GLSLstd450_H
static const int GLSLstd450Version = 100;
static const int GLSLstd450Revision = 3;
enum GLSLstd450 {
GLSLstd450Bad = 0, // Don't use
GLSLstd450Round = 1,
GLSLstd450RoundEven = 2,
GLSLstd450Trunc = 3,
GLSLstd450FAbs = 4,
GLSLstd450SAbs = 5,
GLSLstd450FSign = 6,
GLSLstd450SSign = 7,
GLSLstd450Floor = 8,
GLSLstd450Ceil = 9,
GLSLstd450Fract = 10,
GLSLstd450Radians = 11,
GLSLstd450Degrees = 12,
GLSLstd450Sin = 13,
GLSLstd450Cos = 14,
GLSLstd450Tan = 15,
GLSLstd450Asin = 16,
GLSLstd450Acos = 17,
GLSLstd450Atan = 18,
GLSLstd450Sinh = 19,
GLSLstd450Cosh = 20,
GLSLstd450Tanh = 21,
GLSLstd450Asinh = 22,
GLSLstd450Acosh = 23,
GLSLstd450Atanh = 24,
GLSLstd450Atan2 = 25,
GLSLstd450Pow = 26,
GLSLstd450Exp = 27,
GLSLstd450Log = 28,
GLSLstd450Exp2 = 29,
GLSLstd450Log2 = 30,
GLSLstd450Sqrt = 31,
GLSLstd450InverseSqrt = 32,
GLSLstd450Determinant = 33,
GLSLstd450MatrixInverse = 34,
GLSLstd450Modf = 35, // second operand needs an OpVariable to write to
GLSLstd450ModfStruct = 36, // no OpVariable operand
GLSLstd450FMin = 37,
GLSLstd450UMin = 38,
GLSLstd450SMin = 39,
GLSLstd450FMax = 40,
GLSLstd450UMax = 41,
GLSLstd450SMax = 42,
GLSLstd450FClamp = 43,
GLSLstd450UClamp = 44,
GLSLstd450SClamp = 45,
GLSLstd450FMix = 46,
GLSLstd450IMix = 47, // Reserved
GLSLstd450Step = 48,
GLSLstd450SmoothStep = 49,
GLSLstd450Fma = 50,
GLSLstd450Frexp = 51, // second operand needs an OpVariable to write to
GLSLstd450FrexpStruct = 52, // no OpVariable operand
GLSLstd450Ldexp = 53,
GLSLstd450PackSnorm4x8 = 54,
GLSLstd450PackUnorm4x8 = 55,
GLSLstd450PackSnorm2x16 = 56,
GLSLstd450PackUnorm2x16 = 57,
GLSLstd450PackHalf2x16 = 58,
GLSLstd450PackDouble2x32 = 59,
GLSLstd450UnpackSnorm2x16 = 60,
GLSLstd450UnpackUnorm2x16 = 61,
GLSLstd450UnpackHalf2x16 = 62,
GLSLstd450UnpackSnorm4x8 = 63,
GLSLstd450UnpackUnorm4x8 = 64,
GLSLstd450UnpackDouble2x32 = 65,
GLSLstd450Length = 66,
GLSLstd450Distance = 67,
GLSLstd450Cross = 68,
GLSLstd450Normalize = 69,
GLSLstd450FaceForward = 70,
GLSLstd450Reflect = 71,
GLSLstd450Refract = 72,
GLSLstd450FindILsb = 73,
GLSLstd450FindSMsb = 74,
GLSLstd450FindUMsb = 75,
GLSLstd450InterpolateAtCentroid = 76,
GLSLstd450InterpolateAtSample = 77,
GLSLstd450InterpolateAtOffset = 78,
GLSLstd450NMin = 79,
GLSLstd450NMax = 80,
GLSLstd450NClamp = 81,
GLSLstd450Count
};
#endif // #ifndef GLSLstd450_H

202
thirdparty/spirv-cross/LICENSE vendored Normal file
View File

@ -0,0 +1,202 @@
Apache License
Version 2.0, January 2004
http://www.apache.org/licenses/
TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
1. Definitions.
"License" shall mean the terms and conditions for use, reproduction,
and distribution as defined by Sections 1 through 9 of this document.
"Licensor" shall mean the copyright owner or entity authorized by
the copyright owner that is granting the License.
"Legal Entity" shall mean the union of the acting entity and all
other entities that control, are controlled by, or are under common
control with that entity. For the purposes of this definition,
"control" means (i) the power, direct or indirect, to cause the
direction or management of such entity, whether by contract or
otherwise, or (ii) ownership of fifty percent (50%) or more of the
outstanding shares, or (iii) beneficial ownership of such entity.
"You" (or "Your") shall mean an individual or Legal Entity
exercising permissions granted by this License.
"Source" form shall mean the preferred form for making modifications,
including but not limited to software source code, documentation
source, and configuration files.
"Object" form shall mean any form resulting from mechanical
transformation or translation of a Source form, including but
not limited to compiled object code, generated documentation,
and conversions to other media types.
"Work" shall mean the work of authorship, whether in Source or
Object form, made available under the License, as indicated by a
copyright notice that is included in or attached to the work
(an example is provided in the Appendix below).
"Derivative Works" shall mean any work, whether in Source or Object
form, that is based on (or derived from) the Work and for which the
editorial revisions, annotations, elaborations, or other modifications
represent, as a whole, an original work of authorship. For the purposes
of this License, Derivative Works shall not include works that remain
separable from, or merely link (or bind by name) to the interfaces of,
the Work and Derivative Works thereof.
"Contribution" shall mean any work of authorship, including
the original version of the Work and any modifications or additions
to that Work or Derivative Works thereof, that is intentionally
submitted to Licensor for inclusion in the Work by the copyright owner
or by an individual or Legal Entity authorized to submit on behalf of
the copyright owner. For the purposes of this definition, "submitted"
means any form of electronic, verbal, or written communication sent
to the Licensor or its representatives, including but not limited to
communication on electronic mailing lists, source code control systems,
and issue tracking systems that are managed by, or on behalf of, the
Licensor for the purpose of discussing and improving the Work, but
excluding communication that is conspicuously marked or otherwise
designated in writing by the copyright owner as "Not a Contribution."
"Contributor" shall mean Licensor and any individual or Legal Entity
on behalf of whom a Contribution has been received by Licensor and
subsequently incorporated within the Work.
2. Grant of Copyright License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
copyright license to reproduce, prepare Derivative Works of,
publicly display, publicly perform, sublicense, and distribute the
Work and such Derivative Works in Source or Object form.
3. Grant of Patent License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
(except as stated in this section) patent license to make, have made,
use, offer to sell, sell, import, and otherwise transfer the Work,
where such license applies only to those patent claims licensable
by such Contributor that are necessarily infringed by their
Contribution(s) alone or by combination of their Contribution(s)
with the Work to which such Contribution(s) was submitted. If You
institute patent litigation against any entity (including a
cross-claim or counterclaim in a lawsuit) alleging that the Work
or a Contribution incorporated within the Work constitutes direct
or contributory patent infringement, then any patent licenses
granted to You under this License for that Work shall terminate
as of the date such litigation is filed.
4. Redistribution. You may reproduce and distribute copies of the
Work or Derivative Works thereof in any medium, with or without
modifications, and in Source or Object form, provided that You
meet the following conditions:
(a) You must give any other recipients of the Work or
Derivative Works a copy of this License; and
(b) You must cause any modified files to carry prominent notices
stating that You changed the files; and
(c) You must retain, in the Source form of any Derivative Works
that You distribute, all copyright, patent, trademark, and
attribution notices from the Source form of the Work,
excluding those notices that do not pertain to any part of
the Derivative Works; and
(d) If the Work includes a "NOTICE" text file as part of its
distribution, then any Derivative Works that You distribute must
include a readable copy of the attribution notices contained
within such NOTICE file, excluding those notices that do not
pertain to any part of the Derivative Works, in at least one
of the following places: within a NOTICE text file distributed
as part of the Derivative Works; within the Source form or
documentation, if provided along with the Derivative Works; or,
within a display generated by the Derivative Works, if and
wherever such third-party notices normally appear. The contents
of the NOTICE file are for informational purposes only and
do not modify the License. You may add Your own attribution
notices within Derivative Works that You distribute, alongside
or as an addendum to the NOTICE text from the Work, provided
that such additional attribution notices cannot be construed
as modifying the License.
You may add Your own copyright statement to Your modifications and
may provide additional or different license terms and conditions
for use, reproduction, or distribution of Your modifications, or
for any such Derivative Works as a whole, provided Your use,
reproduction, and distribution of the Work otherwise complies with
the conditions stated in this License.
5. Submission of Contributions. Unless You explicitly state otherwise,
any Contribution intentionally submitted for inclusion in the Work
by You to the Licensor shall be under the terms and conditions of
this License, without any additional terms or conditions.
Notwithstanding the above, nothing herein shall supersede or modify
the terms of any separate license agreement you may have executed
with Licensor regarding such Contributions.
6. Trademarks. This License does not grant permission to use the trade
names, trademarks, service marks, or product names of the Licensor,
except as required for reasonable and customary use in describing the
origin of the Work and reproducing the content of the NOTICE file.
7. Disclaimer of Warranty. Unless required by applicable law or
agreed to in writing, Licensor provides the Work (and each
Contributor provides its Contributions) on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
implied, including, without limitation, any warranties or conditions
of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
PARTICULAR PURPOSE. You are solely responsible for determining the
appropriateness of using or redistributing the Work and assume any
risks associated with Your exercise of permissions under this License.
8. Limitation of Liability. In no event and under no legal theory,
whether in tort (including negligence), contract, or otherwise,
unless required by applicable law (such as deliberate and grossly
negligent acts) or agreed to in writing, shall any Contributor be
liable to You for damages, including any direct, indirect, special,
incidental, or consequential damages of any character arising as a
result of this License or out of the use or inability to use the
Work (including but not limited to damages for loss of goodwill,
work stoppage, computer failure or malfunction, or any and all
other commercial damages or losses), even if such Contributor
has been advised of the possibility of such damages.
9. Accepting Warranty or Additional Liability. While redistributing
the Work or Derivative Works thereof, You may choose to offer,
and charge a fee for, acceptance of support, warranty, indemnity,
or other liability obligations and/or rights consistent with this
License. However, in accepting such obligations, You may act only
on Your own behalf and on Your sole responsibility, not on behalf
of any other Contributor, and only if You agree to indemnify,
defend, and hold each Contributor harmless for any liability
incurred by, or claims asserted against, such Contributor by reason
of your accepting any such warranty or additional liability.
END OF TERMS AND CONDITIONS
APPENDIX: How to apply the Apache License to your work.
To apply the Apache License to your work, attach the following
boilerplate notice, with the fields enclosed by brackets "[]"
replaced with your own identifying information. (Don't include
the brackets!) The text should be enclosed in the appropriate
comment syntax for the file format. We also recommend that a
file or class name and description of purpose be included on the
same "printed page" as the copyright notice for easier
identification within third-party archives.
Copyright [yyyy] [name of copyright owner]
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.

208
thirdparty/spirv-cross/LICENSES/Apache-2.0.txt vendored Normal file
View File

@ -0,0 +1,208 @@
Apache License
Version 2.0, January 2004
http://www.apache.org/licenses/ TERMS AND CONDITIONS FOR USE, REPRODUCTION,
AND DISTRIBUTION
1. Definitions.
"License" shall mean the terms and conditions for use, reproduction, and distribution
as defined by Sections 1 through 9 of this document.
"Licensor" shall mean the copyright owner or entity authorized by the copyright
owner that is granting the License.
"Legal Entity" shall mean the union of the acting entity and all other entities
that control, are controlled by, or are under common control with that entity.
For the purposes of this definition, "control" means (i) the power, direct
or indirect, to cause the direction or management of such entity, whether
by contract or otherwise, or (ii) ownership of fifty percent (50%) or more
of the outstanding shares, or (iii) beneficial ownership of such entity.
"You" (or "Your") shall mean an individual or Legal Entity exercising permissions
granted by this License.
"Source" form shall mean the preferred form for making modifications, including
but not limited to software source code, documentation source, and configuration
files.
"Object" form shall mean any form resulting from mechanical transformation
or translation of a Source form, including but not limited to compiled object
code, generated documentation, and conversions to other media types.
"Work" shall mean the work of authorship, whether in Source or Object form,
made available under the License, as indicated by a copyright notice that
is included in or attached to the work (an example is provided in the Appendix
below).
"Derivative Works" shall mean any work, whether in Source or Object form,
that is based on (or derived from) the Work and for which the editorial revisions,
annotations, elaborations, or other modifications represent, as a whole, an
original work of authorship. For the purposes of this License, Derivative
Works shall not include works that remain separable from, or merely link (or
bind by name) to the interfaces of, the Work and Derivative Works thereof.
"Contribution" shall mean any work of authorship, including the original version
of the Work and any modifications or additions to that Work or Derivative
Works thereof, that is intentionally submitted to Licensor for inclusion in
the Work by the copyright owner or by an individual or Legal Entity authorized
to submit on behalf of the copyright owner. For the purposes of this definition,
"submitted" means any form of electronic, verbal, or written communication
sent to the Licensor or its representatives, including but not limited to
communication on electronic mailing lists, source code control systems, and
issue tracking systems that are managed by, or on behalf of, the Licensor
for the purpose of discussing and improving the Work, but excluding communication
that is conspicuously marked or otherwise designated in writing by the copyright
owner as "Not a Contribution."
"Contributor" shall mean Licensor and any individual or Legal Entity on behalf
of whom a Contribution has been received by Licensor and subsequently incorporated
within the Work.
2. Grant of Copyright License. Subject to the terms and conditions of this
License, each Contributor hereby grants to You a perpetual, worldwide, non-exclusive,
no-charge, royalty-free, irrevocable copyright license to reproduce, prepare
Derivative Works of, publicly display, publicly perform, sublicense, and distribute
the Work and such Derivative Works in Source or Object form.
3. Grant of Patent License. Subject to the terms and conditions of this License,
each Contributor hereby grants to You a perpetual, worldwide, non-exclusive,
no-charge, royalty-free, irrevocable (except as stated in this section) patent
license to make, have made, use, offer to sell, sell, import, and otherwise
transfer the Work, where such license applies only to those patent claims
licensable by such Contributor that are necessarily infringed by their Contribution(s)
alone or by combination of their Contribution(s) with the Work to which such
Contribution(s) was submitted. If You institute patent litigation against
any entity (including a cross-claim or counterclaim in a lawsuit) alleging
that the Work or a Contribution incorporated within the Work constitutes direct
or contributory patent infringement, then any patent licenses granted to You
under this License for that Work shall terminate as of the date such litigation
is filed.
4. Redistribution. You may reproduce and distribute copies of the Work or
Derivative Works thereof in any medium, with or without modifications, and
in Source or Object form, provided that You meet the following conditions:
(a) You must give any other recipients of the Work or Derivative Works a copy
of this License; and
(b) You must cause any modified files to carry prominent notices stating that
You changed the files; and
(c) You must retain, in the Source form of any Derivative Works that You distribute,
all copyright, patent, trademark, and attribution notices from the Source
form of the Work, excluding those notices that do not pertain to any part
of the Derivative Works; and
(d) If the Work includes a "NOTICE" text file as part of its distribution,
then any Derivative Works that You distribute must include a readable copy
of the attribution notices contained within such NOTICE file, excluding those
notices that do not pertain to any part of the Derivative Works, in at least
one of the following places: within a NOTICE text file distributed as part
of the Derivative Works; within the Source form or documentation, if provided
along with the Derivative Works; or, within a display generated by the Derivative
Works, if and wherever such third-party notices normally appear. The contents
of the NOTICE file are for informational purposes only and do not modify the
License. You may add Your own attribution notices within Derivative Works
that You distribute, alongside or as an addendum to the NOTICE text from the
Work, provided that such additional attribution notices cannot be construed
as modifying the License.
You may add Your own copyright statement to Your modifications and may provide
additional or different license terms and conditions for use, reproduction,
or distribution of Your modifications, or for any such Derivative Works as
a whole, provided Your use, reproduction, and distribution of the Work otherwise
complies with the conditions stated in this License.
5. Submission of Contributions. Unless You explicitly state otherwise, any
Contribution intentionally submitted for inclusion in the Work by You to the
Licensor shall be under the terms and conditions of this License, without
any additional terms or conditions. Notwithstanding the above, nothing herein
shall supersede or modify the terms of any separate license agreement you
may have executed with Licensor regarding such Contributions.
6. Trademarks. This License does not grant permission to use the trade names,
trademarks, service marks, or product names of the Licensor, except as required
for reasonable and customary use in describing the origin of the Work and
reproducing the content of the NOTICE file.
7. Disclaimer of Warranty. Unless required by applicable law or agreed to
in writing, Licensor provides the Work (and each Contributor provides its
Contributions) on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
KIND, either express or implied, including, without limitation, any warranties
or conditions of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR
A PARTICULAR PURPOSE. You are solely responsible for determining the appropriateness
of using or redistributing the Work and assume any risks associated with Your
exercise of permissions under this License.
8. Limitation of Liability. In no event and under no legal theory, whether
in tort (including negligence), contract, or otherwise, unless required by
applicable law (such as deliberate and grossly negligent acts) or agreed to
in writing, shall any Contributor be liable to You for damages, including
any direct, indirect, special, incidental, or consequential damages of any
character arising as a result of this License or out of the use or inability
to use the Work (including but not limited to damages for loss of goodwill,
work stoppage, computer failure or malfunction, or any and all other commercial
damages or losses), even if such Contributor has been advised of the possibility
of such damages.
9. Accepting Warranty or Additional Liability. While redistributing the Work
or Derivative Works thereof, You may choose to offer, and charge a fee for,
acceptance of support, warranty, indemnity, or other liability obligations
and/or rights consistent with this License. However, in accepting such obligations,
You may act only on Your own behalf and on Your sole responsibility, not on
behalf of any other Contributor, and only if You agree to indemnify, defend,
and hold each Contributor harmless for any liability incurred by, or claims
asserted against, such Contributor by reason of your accepting any such warranty
or additional liability. END OF TERMS AND CONDITIONS
APPENDIX: How to apply the Apache License to your work.
To apply the Apache License to your work, attach the following boilerplate
notice, with the fields enclosed by brackets "[]" replaced with your own identifying
information. (Don't include the brackets!) The text should be enclosed in
the appropriate comment syntax for the file format. We also recommend that
a file or class name and description of purpose be included on the same "printed
page" as the copyright notice for easier identification within third-party
archives.
Copyright [yyyy] [name of copyright owner]
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.

23
thirdparty/spirv-cross/LICENSES/LicenseRef-KhronosFreeUse.txt vendored Normal file
View File

@ -0,0 +1,23 @@
Copyright (c) 2014-2020 The Khronos Group Inc.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and/or associated documentation files (the "Materials"),
to deal in the Materials without restriction, including without limitation
the rights to use, copy, modify, merge, publish, distribute, sublicense,
and/or sell copies of the Materials, and to permit persons to whom the
Materials are furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Materials.
MODIFICATIONS TO THIS FILE MAY MEAN IT NO LONGER ACCURATELY REFLECTS KHRONOS
STANDARDS. THE UNMODIFIED, NORMATIVE VERSIONS OF KHRONOS SPECIFICATIONS AND
HEADER INFORMATION ARE LOCATED AT https://www.khronos.org/registry/
THE MATERIALS ARE PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
FROM,OUT OF OR IN CONNECTION WITH THE MATERIALS OR THE USE OR OTHER DEALINGS
IN THE MATERIALS.

19
thirdparty/spirv-cross/LICENSES/MIT.txt vendored Normal file
View File

@ -0,0 +1,19 @@
MIT License Copyright (c) <year> <copyright holders>
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is furnished
to do so, subject to the following conditions:
The above copyright notice and this permission notice (including the next
paragraph) shall be included in all copies or substantial portions of the
Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS
FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS
OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF
OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

80
thirdparty/spirv-cross/include/spirv_cross/barrier.hpp vendored Normal file
View File

@ -0,0 +1,80 @@
/*
* Copyright 2015-2017 ARM Limited
* SPDX-License-Identifier: Apache-2.0
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#ifndef SPIRV_CROSS_BARRIER_HPP
#define SPIRV_CROSS_BARRIER_HPP
#include <atomic>
#include <thread>
namespace spirv_cross
{
class Barrier
{
public:
Barrier()
{
count.store(0);
iteration.store(0);
}
void set_release_divisor(unsigned divisor)
{
this->divisor = divisor;
}
static inline void memoryBarrier()
{
std::atomic_thread_fence(std::memory_order_seq_cst);
}
void reset_counter()
{
count.store(0);
iteration.store(0);
}
void wait()
{
unsigned target_iteration = iteration.load(std::memory_order_relaxed) + 1;
// Overflows cleanly.
unsigned target_count = divisor * target_iteration;
// Barriers don't enforce memory ordering.
// Be as relaxed about the barrier as we possibly can!
unsigned c = count.fetch_add(1u, std::memory_order_relaxed);
if (c + 1 == target_count)
{
iteration.store(target_iteration, std::memory_order_relaxed);
}
else
{
// If we have more threads than the CPU, don't hog the CPU for very long periods of time.
while (iteration.load(std::memory_order_relaxed) != target_iteration)
std::this_thread::yield();
}
}
private:
unsigned divisor = 1;
std::atomic<unsigned> count;
std::atomic<unsigned> iteration;
};
}
#endif

127
thirdparty/spirv-cross/include/spirv_cross/external_interface.h vendored Normal file
View File

@ -0,0 +1,127 @@
/*
* Copyright 2015-2017 ARM Limited
* SPDX-License-Identifier: Apache-2.0
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#ifndef SPIRV_CROSS_EXTERNAL_INTERFACE_H
#define SPIRV_CROSS_EXTERNAL_INTERFACE_H
#ifdef __cplusplus
extern "C" {
#endif
#include <stddef.h>
typedef struct spirv_cross_shader spirv_cross_shader_t;
struct spirv_cross_interface
{
spirv_cross_shader_t *(*construct)(void);
void (*destruct)(spirv_cross_shader_t *thiz);
void (*invoke)(spirv_cross_shader_t *thiz);
};
void spirv_cross_set_stage_input(spirv_cross_shader_t *thiz, unsigned location, void *data, size_t size);
void spirv_cross_set_stage_output(spirv_cross_shader_t *thiz, unsigned location, void *data, size_t size);
void spirv_cross_set_push_constant(spirv_cross_shader_t *thiz, void *data, size_t size);
void spirv_cross_set_uniform_constant(spirv_cross_shader_t *thiz, unsigned location, void *data, size_t size);
void spirv_cross_set_resource(spirv_cross_shader_t *thiz, unsigned set, unsigned binding, void **data, size_t size);
const struct spirv_cross_interface *spirv_cross_get_interface(void);
typedef enum spirv_cross_builtin {
SPIRV_CROSS_BUILTIN_POSITION = 0,
SPIRV_CROSS_BUILTIN_FRAG_COORD = 1,
SPIRV_CROSS_BUILTIN_WORK_GROUP_ID = 2,
SPIRV_CROSS_BUILTIN_NUM_WORK_GROUPS = 3,
SPIRV_CROSS_NUM_BUILTINS
} spirv_cross_builtin;
void spirv_cross_set_builtin(spirv_cross_shader_t *thiz, spirv_cross_builtin builtin, void *data, size_t size);
#define SPIRV_CROSS_NUM_DESCRIPTOR_SETS 4
#define SPIRV_CROSS_NUM_DESCRIPTOR_BINDINGS 16
#define SPIRV_CROSS_NUM_STAGE_INPUTS 16
#define SPIRV_CROSS_NUM_STAGE_OUTPUTS 16
#define SPIRV_CROSS_NUM_UNIFORM_CONSTANTS 32
enum spirv_cross_format
{
SPIRV_CROSS_FORMAT_R8_UNORM = 0,
SPIRV_CROSS_FORMAT_R8G8_UNORM = 1,
SPIRV_CROSS_FORMAT_R8G8B8_UNORM = 2,
SPIRV_CROSS_FORMAT_R8G8B8A8_UNORM = 3,
SPIRV_CROSS_NUM_FORMATS
};
enum spirv_cross_wrap
{
SPIRV_CROSS_WRAP_CLAMP_TO_EDGE = 0,
SPIRV_CROSS_WRAP_REPEAT = 1,
SPIRV_CROSS_NUM_WRAP
};
enum spirv_cross_filter
{
SPIRV_CROSS_FILTER_NEAREST = 0,
SPIRV_CROSS_FILTER_LINEAR = 1,
SPIRV_CROSS_NUM_FILTER
};
enum spirv_cross_mipfilter
{
SPIRV_CROSS_MIPFILTER_BASE = 0,
SPIRV_CROSS_MIPFILTER_NEAREST = 1,
SPIRV_CROSS_MIPFILTER_LINEAR = 2,
SPIRV_CROSS_NUM_MIPFILTER
};
struct spirv_cross_miplevel
{
const void *data;
unsigned width, height;
size_t stride;
};
struct spirv_cross_sampler_info
{
const struct spirv_cross_miplevel *mipmaps;
unsigned num_mipmaps;
enum spirv_cross_format format;
enum spirv_cross_wrap wrap_s;
enum spirv_cross_wrap wrap_t;
enum spirv_cross_filter min_filter;
enum spirv_cross_filter mag_filter;
enum spirv_cross_mipfilter mip_filter;
};
typedef struct spirv_cross_sampler_2d spirv_cross_sampler_2d_t;
spirv_cross_sampler_2d_t *spirv_cross_create_sampler_2d(const struct spirv_cross_sampler_info *info);
void spirv_cross_destroy_sampler_2d(spirv_cross_sampler_2d_t *samp);
#ifdef __cplusplus
}
#endif
#endif

63
thirdparty/spirv-cross/include/spirv_cross/image.hpp vendored Normal file
View File

@ -0,0 +1,63 @@
/*
* Copyright 2015-2017 ARM Limited
* SPDX-License-Identifier: Apache-2.0
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#ifndef SPIRV_CROSS_IMAGE_HPP
#define SPIRV_CROSS_IMAGE_HPP
#ifndef GLM_SWIZZLE
#define GLM_SWIZZLE
#endif
#ifndef GLM_FORCE_RADIANS
#define GLM_FORCE_RADIANS
#endif
#include <glm/glm.hpp>
namespace spirv_cross
{
template <typename T>
struct image2DBase
{
virtual ~image2DBase() = default;
inline virtual T load(glm::ivec2 coord) const
{
return T(0, 0, 0, 1);
}
inline virtual void store(glm::ivec2 coord, const T &v)
{
}
};
typedef image2DBase<glm::vec4> image2D;
typedef image2DBase<glm::ivec4> iimage2D;
typedef image2DBase<glm::uvec4> uimage2D;
template <typename T>
inline T imageLoad(const image2DBase<T> &image, glm::ivec2 coord)
{
return image.load(coord);
}
template <typename T>
void imageStore(image2DBase<T> &image, glm::ivec2 coord, const T &value)
{
image.store(coord, value);
}
}
#endif

604
thirdparty/spirv-cross/include/spirv_cross/internal_interface.hpp vendored Normal file
View File

@ -0,0 +1,604 @@
/*
* Copyright 2015-2017 ARM Limited
* SPDX-License-Identifier: Apache-2.0
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#ifndef SPIRV_CROSS_INTERNAL_INTERFACE_HPP
#define SPIRV_CROSS_INTERNAL_INTERFACE_HPP
// This file must only be included by the shader generated by spirv-cross!
#ifndef GLM_FORCE_SWIZZLE
#define GLM_FORCE_SWIZZLE
#endif
#ifndef GLM_FORCE_RADIANS
#define GLM_FORCE_RADIANS
#endif
#include <glm/glm.hpp>
#include "barrier.hpp"
#include "external_interface.h"
#include "image.hpp"
#include "sampler.hpp"
#include "thread_group.hpp"
#include <assert.h>
#include <stdint.h>
namespace internal
{
// Adaptor helpers to adapt GLSL access chain syntax to C++.
// Don't bother with arrays of arrays on uniforms ...
// Would likely need horribly complex variadic template munging.
template <typename T>
struct Interface
{
enum
{
ArraySize = 1,
Size = sizeof(T)
};
Interface()
: ptr(0)
{
}
T &get()
{
assert(ptr);
return *ptr;
}
T *ptr;
};
// For array types, return a pointer instead.
template <typename T, unsigned U>
struct Interface<T[U]>
{
enum
{
ArraySize = U,
Size = U * sizeof(T)
};
Interface()
: ptr(0)
{
}
T *get()
{
assert(ptr);
return ptr;
}
T *ptr;
};
// For case when array size is 1, avoid double dereference.
template <typename T>
struct PointerInterface
{
enum
{
ArraySize = 1,
Size = sizeof(T *)
};
enum
{
PreDereference = true
};
PointerInterface()
: ptr(0)
{
}
T &get()
{
assert(ptr);
return *ptr;
}
T *ptr;
};
// Automatically converts a pointer down to reference to match GLSL syntax.
template <typename T>
struct DereferenceAdaptor
{
DereferenceAdaptor(T **ptr)
: ptr(ptr)
{
}
T &operator[](unsigned index) const
{
return *(ptr[index]);
}
T **ptr;
};
// We can't have a linear array of T* since T* can be an abstract type in case of samplers.
// We also need a list of pointers since we can have run-time length SSBOs.
template <typename T, unsigned U>
struct PointerInterface<T[U]>
{
enum
{
ArraySize = U,
Size = sizeof(T *) * U
};
enum
{
PreDereference = false
};
PointerInterface()
: ptr(0)
{
}
DereferenceAdaptor<T> get()
{
assert(ptr);
return DereferenceAdaptor<T>(ptr);
}
T **ptr;
};
// Resources can be more abstract and be unsized,
// so we need to have an array of pointers for those cases.
template <typename T>
struct Resource : PointerInterface<T>
{
};
// POD with no unknown sizes, so we can express these as flat arrays.
template <typename T>
struct UniformConstant : Interface<T>
{
};
template <typename T>
struct StageInput : Interface<T>
{
};
template <typename T>
struct StageOutput : Interface<T>
{
};
template <typename T>
struct PushConstant : Interface<T>
{
};
}
struct spirv_cross_shader
{
struct PPSize
{
PPSize()
: ptr(0)
, size(0)
{
}
void **ptr;
size_t size;
};
struct PPSizeResource
{
PPSizeResource()
: ptr(0)
, size(0)
, pre_dereference(false)
{
}
void **ptr;
size_t size;
bool pre_dereference;
};
PPSizeResource resources[SPIRV_CROSS_NUM_DESCRIPTOR_SETS][SPIRV_CROSS_NUM_DESCRIPTOR_BINDINGS];
PPSize stage_inputs[SPIRV_CROSS_NUM_STAGE_INPUTS];
PPSize stage_outputs[SPIRV_CROSS_NUM_STAGE_OUTPUTS];
PPSize uniform_constants[SPIRV_CROSS_NUM_UNIFORM_CONSTANTS];
PPSize push_constant;
PPSize builtins[SPIRV_CROSS_NUM_BUILTINS];
template <typename U>
void register_builtin(spirv_cross_builtin builtin, const U &value)
{
assert(!builtins[builtin].ptr);
builtins[builtin].ptr = (void **)&value.ptr;
builtins[builtin].size = sizeof(*value.ptr) * U::ArraySize;
}
void set_builtin(spirv_cross_builtin builtin, void *data, size_t size)
{
assert(builtins[builtin].ptr);
assert(size >= builtins[builtin].size);
*builtins[builtin].ptr = data;
}
template <typename U>
void register_resource(const internal::Resource<U> &value, unsigned set, unsigned binding)
{
assert(set < SPIRV_CROSS_NUM_DESCRIPTOR_SETS);
assert(binding < SPIRV_CROSS_NUM_DESCRIPTOR_BINDINGS);
assert(!resources[set][binding].ptr);
resources[set][binding].ptr = (void **)&value.ptr;
resources[set][binding].size = internal::Resource<U>::Size;
resources[set][binding].pre_dereference = internal::Resource<U>::PreDereference;
}
template <typename U>
void register_stage_input(const internal::StageInput<U> &value, unsigned location)
{
assert(location < SPIRV_CROSS_NUM_STAGE_INPUTS);
assert(!stage_inputs[location].ptr);
stage_inputs[location].ptr = (void **)&value.ptr;
stage_inputs[location].size = internal::StageInput<U>::Size;
}
template <typename U>
void register_stage_output(const internal::StageOutput<U> &value, unsigned location)
{
assert(location < SPIRV_CROSS_NUM_STAGE_OUTPUTS);
assert(!stage_outputs[location].ptr);
stage_outputs[location].ptr = (void **)&value.ptr;
stage_outputs[location].size = internal::StageOutput<U>::Size;
}
template <typename U>
void register_uniform_constant(const internal::UniformConstant<U> &value, unsigned location)
{
assert(location < SPIRV_CROSS_NUM_UNIFORM_CONSTANTS);
assert(!uniform_constants[location].ptr);
uniform_constants[location].ptr = (void **)&value.ptr;
uniform_constants[location].size = internal::UniformConstant<U>::Size;
}
template <typename U>
void register_push_constant(const internal::PushConstant<U> &value)
{
assert(!push_constant.ptr);
push_constant.ptr = (void **)&value.ptr;
push_constant.size = internal::PushConstant<U>::Size;
}
void set_stage_input(unsigned location, void *data, size_t size)
{
assert(location < SPIRV_CROSS_NUM_STAGE_INPUTS);
assert(stage_inputs[location].ptr);
assert(size >= stage_inputs[location].size);
*stage_inputs[location].ptr = data;
}
void set_stage_output(unsigned location, void *data, size_t size)
{
assert(location < SPIRV_CROSS_NUM_STAGE_OUTPUTS);
assert(stage_outputs[location].ptr);
assert(size >= stage_outputs[location].size);
*stage_outputs[location].ptr = data;
}
void set_uniform_constant(unsigned location, void *data, size_t size)
{
assert(location < SPIRV_CROSS_NUM_UNIFORM_CONSTANTS);
assert(uniform_constants[location].ptr);
assert(size >= uniform_constants[location].size);
*uniform_constants[location].ptr = data;
}
void set_push_constant(void *data, size_t size)
{
assert(push_constant.ptr);
assert(size >= push_constant.size);
*push_constant.ptr = data;
}
void set_resource(unsigned set, unsigned binding, void **data, size_t size)
{
assert(set < SPIRV_CROSS_NUM_DESCRIPTOR_SETS);
assert(binding < SPIRV_CROSS_NUM_DESCRIPTOR_BINDINGS);
assert(resources[set][binding].ptr);
assert(size >= resources[set][binding].size);
// We're using the regular PointerInterface, dereference ahead of time.
if (resources[set][binding].pre_dereference)
*resources[set][binding].ptr = *data;
else
*resources[set][binding].ptr = data;
}
};
namespace spirv_cross
{
template <typename T>
struct BaseShader : spirv_cross_shader
{
void invoke()
{
static_cast<T *>(this)->main();
}
};
struct FragmentResources
{
internal::StageOutput<glm::vec4> gl_FragCoord;
void init(spirv_cross_shader &s)
{
s.register_builtin(SPIRV_CROSS_BUILTIN_FRAG_COORD, gl_FragCoord);
}
#define gl_FragCoord __res->gl_FragCoord.get()
};
template <typename T, typename Res>
struct FragmentShader : BaseShader<FragmentShader<T, Res>>
{
inline void main()
{
impl.main();
}
FragmentShader()
{
resources.init(*this);
impl.__res = &resources;
}
T impl;
Res resources;
};
struct VertexResources
{
internal::StageOutput<glm::vec4> gl_Position;
void init(spirv_cross_shader &s)
{
s.register_builtin(SPIRV_CROSS_BUILTIN_POSITION, gl_Position);
}
#define gl_Position __res->gl_Position.get()
};
template <typename T, typename Res>
struct VertexShader : BaseShader<VertexShader<T, Res>>
{
inline void main()
{
impl.main();
}
VertexShader()
{
resources.init(*this);
impl.__res = &resources;
}
T impl;
Res resources;
};
struct TessEvaluationResources
{
inline void init(spirv_cross_shader &)
{
}
};
template <typename T, typename Res>
struct TessEvaluationShader : BaseShader<TessEvaluationShader<T, Res>>
{
inline void main()
{
impl.main();
}
TessEvaluationShader()
{
resources.init(*this);
impl.__res = &resources;
}
T impl;
Res resources;
};
struct TessControlResources
{
inline void init(spirv_cross_shader &)
{
}
};
template <typename T, typename Res>
struct TessControlShader : BaseShader<TessControlShader<T, Res>>
{
inline void main()
{
impl.main();
}
TessControlShader()
{
resources.init(*this);
impl.__res = &resources;
}
T impl;
Res resources;
};
struct GeometryResources
{
inline void init(spirv_cross_shader &)
{
}
};
template <typename T, typename Res>
struct GeometryShader : BaseShader<GeometryShader<T, Res>>
{
inline void main()
{
impl.main();
}
GeometryShader()
{
resources.init(*this);
impl.__res = &resources;
}
T impl;
Res resources;
};
struct ComputeResources
{
internal::StageInput<glm::uvec3> gl_WorkGroupID__;
internal::StageInput<glm::uvec3> gl_NumWorkGroups__;
void init(spirv_cross_shader &s)
{
s.register_builtin(SPIRV_CROSS_BUILTIN_WORK_GROUP_ID, gl_WorkGroupID__);
s.register_builtin(SPIRV_CROSS_BUILTIN_NUM_WORK_GROUPS, gl_NumWorkGroups__);
}
#define gl_WorkGroupID __res->gl_WorkGroupID__.get()
#define gl_NumWorkGroups __res->gl_NumWorkGroups__.get()
Barrier barrier__;
#define barrier() __res->barrier__.wait()
};
struct ComputePrivateResources
{
uint32_t gl_LocalInvocationIndex__;
#define gl_LocalInvocationIndex __priv_res.gl_LocalInvocationIndex__
glm::uvec3 gl_LocalInvocationID__;
#define gl_LocalInvocationID __priv_res.gl_LocalInvocationID__
glm::uvec3 gl_GlobalInvocationID__;
#define gl_GlobalInvocationID __priv_res.gl_GlobalInvocationID__
};
template <typename T, typename Res, unsigned WorkGroupX, unsigned WorkGroupY, unsigned WorkGroupZ>
struct ComputeShader : BaseShader<ComputeShader<T, Res, WorkGroupX, WorkGroupY, WorkGroupZ>>
{
inline void main()
{
resources.barrier__.reset_counter();
for (unsigned z = 0; z < WorkGroupZ; z++)
for (unsigned y = 0; y < WorkGroupY; y++)
for (unsigned x = 0; x < WorkGroupX; x++)
impl[z][y][x].__priv_res.gl_GlobalInvocationID__ =
glm::uvec3(WorkGroupX, WorkGroupY, WorkGroupZ) * resources.gl_WorkGroupID__.get() +
glm::uvec3(x, y, z);
group.run();
group.wait();
}
ComputeShader()
: group(&impl[0][0][0])
{
resources.init(*this);
resources.barrier__.set_release_divisor(WorkGroupX * WorkGroupY * WorkGroupZ);
unsigned i = 0;
for (unsigned z = 0; z < WorkGroupZ; z++)
{
for (unsigned y = 0; y < WorkGroupY; y++)
{
for (unsigned x = 0; x < WorkGroupX; x++)
{
impl[z][y][x].__priv_res.gl_LocalInvocationID__ = glm::uvec3(x, y, z);
impl[z][y][x].__priv_res.gl_LocalInvocationIndex__ = i++;
impl[z][y][x].__res = &resources;
}
}
}
}
T impl[WorkGroupZ][WorkGroupY][WorkGroupX];
ThreadGroup<T, WorkGroupX * WorkGroupY * WorkGroupZ> group;
Res resources;
};
inline void memoryBarrierShared()
{
Barrier::memoryBarrier();
}
inline void memoryBarrier()
{
Barrier::memoryBarrier();
}
// TODO: Rest of the barriers.
// Atomics
template <typename T>
inline T atomicAdd(T &v, T a)
{
static_assert(sizeof(std::atomic<T>) == sizeof(T), "Cannot cast properly to std::atomic<T>.");
// We need explicit memory barriers in GLSL to enfore any ordering.
// FIXME: Can we really cast this? There is no other way I think ...
return std::atomic_fetch_add_explicit(reinterpret_cast<std::atomic<T> *>(&v), a, std::memory_order_relaxed);
}
}
void spirv_cross_set_stage_input(spirv_cross_shader_t *shader, unsigned location, void *data, size_t size)
{
shader->set_stage_input(location, data, size);
}
void spirv_cross_set_stage_output(spirv_cross_shader_t *shader, unsigned location, void *data, size_t size)
{
shader->set_stage_output(location, data, size);
}
void spirv_cross_set_uniform_constant(spirv_cross_shader_t *shader, unsigned location, void *data, size_t size)
{
shader->set_uniform_constant(location, data, size);
}
void spirv_cross_set_resource(spirv_cross_shader_t *shader, unsigned set, unsigned binding, void **data, size_t size)
{
shader->set_resource(set, binding, data, size);
}
void spirv_cross_set_push_constant(spirv_cross_shader_t *shader, void *data, size_t size)
{
shader->set_push_constant(data, size);
}
void spirv_cross_set_builtin(spirv_cross_shader_t *shader, spirv_cross_builtin builtin, void *data, size_t size)
{
shader->set_builtin(builtin, data, size);
}
#endif

106
thirdparty/spirv-cross/include/spirv_cross/sampler.hpp vendored Normal file
View File

@ -0,0 +1,106 @@
/*
* Copyright 2015-2017 ARM Limited
* SPDX-License-Identifier: Apache-2.0
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#ifndef SPIRV_CROSS_SAMPLER_HPP
#define SPIRV_CROSS_SAMPLER_HPP
#include <vector>
namespace spirv_cross
{
struct spirv_cross_sampler_2d
{
inline virtual ~spirv_cross_sampler_2d()
{
}
};
template <typename T>
struct sampler2DBase : spirv_cross_sampler_2d
{
sampler2DBase(const spirv_cross_sampler_info *info)
{
mips.insert(mips.end(), info->mipmaps, info->mipmaps + info->num_mipmaps);
format = info->format;
wrap_s = info->wrap_s;
wrap_t = info->wrap_t;
min_filter = info->min_filter;
mag_filter = info->mag_filter;
mip_filter = info->mip_filter;
}
inline virtual T sample(glm::vec2 uv, float bias)
{
return sampleLod(uv, bias);
}
inline virtual T sampleLod(glm::vec2 uv, float lod)
{
if (mag_filter == SPIRV_CROSS_FILTER_NEAREST)
{
uv.x = wrap(uv.x, wrap_s, mips[0].width);
uv.y = wrap(uv.y, wrap_t, mips[0].height);
glm::vec2 uv_full = uv * glm::vec2(mips[0].width, mips[0].height);
int x = int(uv_full.x);
int y = int(uv_full.y);
return sample(x, y, 0);
}
else
{
return T(0, 0, 0, 1);
}
}
inline float wrap(float v, spirv_cross_wrap wrap, unsigned size)
{
switch (wrap)
{
case SPIRV_CROSS_WRAP_REPEAT:
return v - glm::floor(v);
case SPIRV_CROSS_WRAP_CLAMP_TO_EDGE:
{
float half = 0.5f / size;
return glm::clamp(v, half, 1.0f - half);
}
default:
return 0.0f;
}
}
std::vector<spirv_cross_miplevel> mips;
spirv_cross_format format;
spirv_cross_wrap wrap_s;
spirv_cross_wrap wrap_t;
spirv_cross_filter min_filter;
spirv_cross_filter mag_filter;
spirv_cross_mipfilter mip_filter;
};
typedef sampler2DBase<glm::vec4> sampler2D;
typedef sampler2DBase<glm::ivec4> isampler2D;
typedef sampler2DBase<glm::uvec4> usampler2D;
template <typename T>
inline T texture(const sampler2DBase<T> &samp, const glm::vec2 &uv, float bias = 0.0f)
{
return samp.sample(uv, bias);
}
}
#endif

114
thirdparty/spirv-cross/include/spirv_cross/thread_group.hpp vendored Normal file
View File

@ -0,0 +1,114 @@
/*
* Copyright 2015-2017 ARM Limited
* SPDX-License-Identifier: Apache-2.0
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#ifndef SPIRV_CROSS_THREAD_GROUP_HPP
#define SPIRV_CROSS_THREAD_GROUP_HPP
#include <condition_variable>
#include <mutex>
#include <thread>
namespace spirv_cross
{
template <typename T, unsigned Size>
class ThreadGroup
{
public:
ThreadGroup(T *impl)
{
for (unsigned i = 0; i < Size; i++)
workers[i].start(&impl[i]);
}
void run()
{
for (auto &worker : workers)
worker.run();
}
void wait()
{
for (auto &worker : workers)
worker.wait();
}
private:
struct Thread
{
enum State
{
Idle,
Running,
Dying
};
State state = Idle;
void start(T *impl)
{
worker = std::thread([impl, this] {
for (;;)
{
{
std::unique_lock<std::mutex> l{ lock };
cond.wait(l, [this] { return state != Idle; });
if (state == Dying)
break;
}
impl->main();
std::lock_guard<std::mutex> l{ lock };
state = Idle;
cond.notify_one();
}
});
}
void wait()
{
std::unique_lock<std::mutex> l{ lock };
cond.wait(l, [this] { return state == Idle; });
}
void run()
{
std::lock_guard<std::mutex> l{ lock };
state = Running;
cond.notify_one();
}
~Thread()
{
if (worker.joinable())
{
{
std::lock_guard<std::mutex> l{ lock };
state = Dying;
cond.notify_one();
}
worker.join();
}
}
std::thread worker;
std::condition_variable cond;
std::mutex lock;
};
Thread workers[Size];
};
}
#endif

2592
thirdparty/spirv-cross/spirv.hpp vendored Normal file
View File

File diff suppressed because it is too large Load Diff

430
thirdparty/spirv-cross/spirv_cfg.cpp vendored Normal file
View File

@ -0,0 +1,430 @@
/*
* Copyright 2016-2021 Arm Limited
* SPDX-License-Identifier: Apache-2.0 OR MIT
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
/*
* At your option, you may choose to accept this material under either:
* 1. The Apache License, Version 2.0, found at <http://www.apache.org/licenses/LICENSE-2.0>, or
* 2. The MIT License, found at <http://opensource.org/licenses/MIT>.
*/
#include "spirv_cfg.hpp"
#include "spirv_cross.hpp"
#include <algorithm>
#include <assert.h>
using namespace std;
namespace SPIRV_CROSS_NAMESPACE
{
CFG::CFG(Compiler &compiler_, const SPIRFunction &func_)
: compiler(compiler_)
, func(func_)
{
build_post_order_visit_order();
build_immediate_dominators();
}
uint32_t CFG::find_common_dominator(uint32_t a, uint32_t b) const
{
while (a != b)
{
if (get_visit_order(a) < get_visit_order(b))
a = get_immediate_dominator(a);
else
b = get_immediate_dominator(b);
}
return a;
}
void CFG::build_immediate_dominators()
{
// Traverse the post-order in reverse and build up the immediate dominator tree.
immediate_dominators.clear();
immediate_dominators[func.entry_block] = func.entry_block;
for (auto i = post_order.size(); i; i--)
{
uint32_t block = post_order[i - 1];
auto &pred = preceding_edges[block];
if (pred.empty()) // This is for the entry block, but we've already set up the dominators.
continue;
for (auto &edge : pred)
{
if (immediate_dominators[block])
{
assert(immediate_dominators[edge]);
immediate_dominators[block] = find_common_dominator(immediate_dominators[block], edge);
}
else
immediate_dominators[block] = edge;
}
}
}
bool CFG::is_back_edge(uint32_t to) const
{
// We have a back edge if the visit order is set with the temporary magic value 0.
// Crossing edges will have already been recorded with a visit order.
auto itr = visit_order.find(to);
return itr != end(visit_order) && itr->second.get() == 0;
}
bool CFG::has_visited_forward_edge(uint32_t to) const
{
// If > 0, we have visited the edge already, and this is not a back edge branch.
auto itr = visit_order.find(to);
return itr != end(visit_order) && itr->second.get() > 0;
}
bool CFG::post_order_visit(uint32_t block_id)
{
// If we have already branched to this block (back edge), stop recursion.
// If our branches are back-edges, we do not record them.
// We have to record crossing edges however.
if (has_visited_forward_edge(block_id))
return true;
else if (is_back_edge(block_id))
return false;
// Block back-edges from recursively revisiting ourselves.
visit_order[block_id].get() = 0;
auto &block = compiler.get<SPIRBlock>(block_id);
// If this is a loop header, add an implied branch to the merge target.
// This is needed to avoid annoying cases with do { ... } while(false) loops often generated by inliners.
// To the CFG, this is linear control flow, but we risk picking the do/while scope as our dominating block.
// This makes sure that if we are accessing a variable outside the do/while, we choose the loop header as dominator.
// We could use has_visited_forward_edge, but this break code-gen where the merge block is unreachable in the CFG.
// Make a point out of visiting merge target first. This is to make sure that post visit order outside the loop
// is lower than inside the loop, which is going to be key for some traversal algorithms like post-dominance analysis.
// For selection constructs true/false blocks will end up visiting the merge block directly and it works out fine,
// but for loops, only the header might end up actually branching to merge block.
if (block.merge == SPIRBlock::MergeLoop && post_order_visit(block.merge_block))
add_branch(block_id, block.merge_block);
// First visit our branch targets.
switch (block.terminator)
{
case SPIRBlock::Direct:
if (post_order_visit(block.next_block))
add_branch(block_id, block.next_block);
break;
case SPIRBlock::Select:
if (post_order_visit(block.true_block))
add_branch(block_id, block.true_block);
if (post_order_visit(block.false_block))
add_branch(block_id, block.false_block);
break;
case SPIRBlock::MultiSelect:
{
const auto &cases = compiler.get_case_list(block);
for (const auto &target : cases)
{
if (post_order_visit(target.block))
add_branch(block_id, target.block);
}
if (block.default_block && post_order_visit(block.default_block))
add_branch(block_id, block.default_block);
break;
}
default:
break;
}
// If this is a selection merge, add an implied branch to the merge target.
// This is needed to avoid cases where an inner branch dominates the outer branch.
// This can happen if one of the branches exit early, e.g.:
// if (cond) { ...; break; } else { var = 100 } use_var(var);
// We can use the variable without a Phi since there is only one possible parent here.
// However, in this case, we need to hoist out the inner variable to outside the branch.
// Use same strategy as loops.
if (block.merge == SPIRBlock::MergeSelection && post_order_visit(block.next_block))
{
// If there is only one preceding edge to the merge block and it's not ourselves, we need a fixup.
// Add a fake branch so any dominator in either the if (), or else () block, or a lone case statement
// will be hoisted out to outside the selection merge.
// If size > 1, the variable will be automatically hoisted, so we should not mess with it.
// The exception here is switch blocks, where we can have multiple edges to merge block,
// all coming from same scope, so be more conservative in this case.
// Adding fake branches unconditionally breaks parameter preservation analysis,
// which looks at how variables are accessed through the CFG.
auto pred_itr = preceding_edges.find(block.next_block);
if (pred_itr != end(preceding_edges))
{
auto &pred = pred_itr->second;
auto succ_itr = succeeding_edges.find(block_id);
size_t num_succeeding_edges = 0;
if (succ_itr != end(succeeding_edges))
num_succeeding_edges = succ_itr->second.size();
if (block.terminator == SPIRBlock::MultiSelect && num_succeeding_edges == 1)
{
// Multiple branches can come from the same scope due to "break;", so we need to assume that all branches
// come from same case scope in worst case, even if there are multiple preceding edges.
// If we have more than one succeeding edge from the block header, it should be impossible
// to have a dominator be inside the block.
// Only case this can go wrong is if we have 2 or more edges from block header and
// 2 or more edges to merge block, and still have dominator be inside a case label.
if (!pred.empty())
add_branch(block_id, block.next_block);
}
else
{
if (pred.size() == 1 && *pred.begin() != block_id)
add_branch(block_id, block.next_block);
}
}
else
{
// If the merge block does not have any preceding edges, i.e. unreachable, hallucinate it.
// We're going to do code-gen for it, and domination analysis requires that we have at least one preceding edge.
add_branch(block_id, block.next_block);
}
}
// Then visit ourselves. Start counting at one, to let 0 be a magic value for testing back vs. crossing edges.
visit_order[block_id].get() = ++visit_count;
post_order.push_back(block_id);
return true;
}
void CFG::build_post_order_visit_order()
{
uint32_t block = func.entry_block;
visit_count = 0;
visit_order.clear();
post_order.clear();
post_order_visit(block);
}
void CFG::add_branch(uint32_t from, uint32_t to)
{
const auto add_unique = [](SmallVector<uint32_t> &l, uint32_t value) {
auto itr = find(begin(l), end(l), value);
if (itr == end(l))
l.push_back(value);
};
add_unique(preceding_edges[to], from);
add_unique(succeeding_edges[from], to);
}
uint32_t CFG::find_loop_dominator(uint32_t block_id) const
{
while (block_id != SPIRBlock::NoDominator)
{
auto itr = preceding_edges.find(block_id);
if (itr == end(preceding_edges))
return SPIRBlock::NoDominator;
if (itr->second.empty())
return SPIRBlock::NoDominator;
uint32_t pred_block_id = SPIRBlock::NoDominator;
bool ignore_loop_header = false;
// If we are a merge block, go directly to the header block.
// Only consider a loop dominator if we are branching from inside a block to a loop header.
// NOTE: In the CFG we forced an edge from header to merge block always to support variable scopes properly.
for (auto &pred : itr->second)
{
auto &pred_block = compiler.get<SPIRBlock>(pred);
if (pred_block.merge == SPIRBlock::MergeLoop && pred_block.merge_block == ID(block_id))
{
pred_block_id = pred;
ignore_loop_header = true;
break;
}
else if (pred_block.merge == SPIRBlock::MergeSelection && pred_block.next_block == ID(block_id))
{
pred_block_id = pred;
break;
}
}
// No merge block means we can just pick any edge. Loop headers dominate the inner loop, so any path we
// take will lead there.
if (pred_block_id == SPIRBlock::NoDominator)
pred_block_id = itr->second.front();
block_id = pred_block_id;
if (!ignore_loop_header && block_id)
{
auto &block = compiler.get<SPIRBlock>(block_id);
if (block.merge == SPIRBlock::MergeLoop)
return block_id;
}
}
return block_id;
}
bool CFG::node_terminates_control_flow_in_sub_graph(BlockID from, BlockID to) const
{
// Walk backwards, starting from "to" block.
// Only follow pred edges if they have a 1:1 relationship, or a merge relationship.
// If we cannot find a path to "from", we must assume that to is inside control flow in some way.
auto &from_block = compiler.get<SPIRBlock>(from);
BlockID ignore_block_id = 0;
if (from_block.merge == SPIRBlock::MergeLoop)
ignore_block_id = from_block.merge_block;
while (to != from)
{
auto pred_itr = preceding_edges.find(to);
if (pred_itr == end(preceding_edges))
return false;
DominatorBuilder builder(*this);
for (auto &edge : pred_itr->second)
builder.add_block(edge);
uint32_t dominator = builder.get_dominator();
if (dominator == 0)
return false;
auto &dom = compiler.get<SPIRBlock>(dominator);
bool true_path_ignore = false;
bool false_path_ignore = false;
bool merges_to_nothing = dom.merge == SPIRBlock::MergeNone ||
(dom.merge == SPIRBlock::MergeSelection && dom.next_block &&
compiler.get<SPIRBlock>(dom.next_block).terminator == SPIRBlock::Unreachable) ||
(dom.merge == SPIRBlock::MergeLoop && dom.merge_block &&
compiler.get<SPIRBlock>(dom.merge_block).terminator == SPIRBlock::Unreachable);
if (dom.self == from || merges_to_nothing)
{
// We can only ignore inner branchy paths if there is no merge,
// i.e. no code is generated afterwards. E.g. this allows us to elide continue:
// for (;;) { if (cond) { continue; } else { break; } }.
// Codegen here in SPIR-V will be something like either no merge if one path directly breaks, or
// we merge to Unreachable.
if (ignore_block_id && dom.terminator == SPIRBlock::Select)
{
auto &true_block = compiler.get<SPIRBlock>(dom.true_block);
auto &false_block = compiler.get<SPIRBlock>(dom.false_block);
auto &ignore_block = compiler.get<SPIRBlock>(ignore_block_id);
true_path_ignore = compiler.execution_is_branchless(true_block, ignore_block);
false_path_ignore = compiler.execution_is_branchless(false_block, ignore_block);
}
}
// Cases where we allow traversal. This serves as a proxy for post-dominance in a loop body.
// TODO: Might want to do full post-dominance analysis, but it's a lot of churn for something like this ...
// - We're the merge block of a selection construct. Jump to header.
// - We're the merge block of a loop. Jump to header.
// - Direct branch. Trivial.
// - Allow cases inside a branch if the header cannot merge execution before loop exit.
if ((dom.merge == SPIRBlock::MergeSelection && dom.next_block == to) ||
(dom.merge == SPIRBlock::MergeLoop && dom.merge_block == to) ||
(dom.terminator == SPIRBlock::Direct && dom.next_block == to) ||
(dom.terminator == SPIRBlock::Select && dom.true_block == to && false_path_ignore) ||
(dom.terminator == SPIRBlock::Select && dom.false_block == to && true_path_ignore))
{
// Allow walking selection constructs if the other branch reaches out of a loop construct.
// It cannot be in-scope anymore.
to = dominator;
}
else
return false;
}
return true;
}
DominatorBuilder::DominatorBuilder(const CFG &cfg_)
: cfg(cfg_)
{
}
void DominatorBuilder::add_block(uint32_t block)
{
if (!cfg.get_immediate_dominator(block))
{
// Unreachable block via the CFG, we will never emit this code anyways.
return;
}
if (!dominator)
{
dominator = block;
return;
}
if (block != dominator)
dominator = cfg.find_common_dominator(block, dominator);
}
void DominatorBuilder::lift_continue_block_dominator()
{
// It is possible for a continue block to be the dominator of a variable is only accessed inside the while block of a do-while loop.
// We cannot safely declare variables inside a continue block, so move any variable declared
// in a continue block to the entry block to simplify.
// It makes very little sense for a continue block to ever be a dominator, so fall back to the simplest
// solution.
if (!dominator)
return;
auto &block = cfg.get_compiler().get<SPIRBlock>(dominator);
auto post_order = cfg.get_visit_order(dominator);
// If we are branching to a block with a higher post-order traversal index (continue blocks), we have a problem
// since we cannot create sensible GLSL code for this, fallback to entry block.
bool back_edge_dominator = false;
switch (block.terminator)
{
case SPIRBlock::Direct:
if (cfg.get_visit_order(block.next_block) > post_order)
back_edge_dominator = true;
break;
case SPIRBlock::Select:
if (cfg.get_visit_order(block.true_block) > post_order)
back_edge_dominator = true;
if (cfg.get_visit_order(block.false_block) > post_order)
back_edge_dominator = true;
break;
case SPIRBlock::MultiSelect:
{
auto &cases = cfg.get_compiler().get_case_list(block);
for (auto &target : cases)
{
if (cfg.get_visit_order(target.block) > post_order)
back_edge_dominator = true;
}
if (block.default_block && cfg.get_visit_order(block.default_block) > post_order)
back_edge_dominator = true;
break;
}
default:
break;
}
if (back_edge_dominator)
dominator = cfg.get_function().entry_block;
}
} // namespace SPIRV_CROSS_NAMESPACE

168
thirdparty/spirv-cross/spirv_cfg.hpp vendored Normal file
View File

@ -0,0 +1,168 @@
/*
* Copyright 2016-2021 Arm Limited
* SPDX-License-Identifier: Apache-2.0 OR MIT
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
/*
* At your option, you may choose to accept this material under either:
* 1. The Apache License, Version 2.0, found at <http://www.apache.org/licenses/LICENSE-2.0>, or
* 2. The MIT License, found at <http://opensource.org/licenses/MIT>.
*/
#ifndef SPIRV_CROSS_CFG_HPP
#define SPIRV_CROSS_CFG_HPP
#include "spirv_common.hpp"
#include <assert.h>
namespace SPIRV_CROSS_NAMESPACE
{
class Compiler;
class CFG
{
public:
CFG(Compiler &compiler, const SPIRFunction &function);
Compiler &get_compiler()
{
return compiler;
}
const Compiler &get_compiler() const
{
return compiler;
}
const SPIRFunction &get_function() const
{
return func;
}
uint32_t get_immediate_dominator(uint32_t block) const
{
auto itr = immediate_dominators.find(block);
if (itr != std::end(immediate_dominators))
return itr->second;
else
return 0;
}
bool is_reachable(uint32_t block) const
{
return visit_order.count(block) != 0;
}
uint32_t get_visit_order(uint32_t block) const
{
auto itr = visit_order.find(block);
assert(itr != std::end(visit_order));
int v = itr->second.get();
assert(v > 0);
return uint32_t(v);
}
uint32_t find_common_dominator(uint32_t a, uint32_t b) const;
const SmallVector<uint32_t> &get_preceding_edges(uint32_t block) const
{
auto itr = preceding_edges.find(block);
if (itr != std::end(preceding_edges))
return itr->second;
else
return empty_vector;
}
const SmallVector<uint32_t> &get_succeeding_edges(uint32_t block) const
{
auto itr = succeeding_edges.find(block);
if (itr != std::end(succeeding_edges))
return itr->second;
else
return empty_vector;
}
template <typename Op>
void walk_from(std::unordered_set<uint32_t> &seen_blocks, uint32_t block, const Op &op) const
{
if (seen_blocks.count(block))
return;
seen_blocks.insert(block);
if (op(block))
{
for (auto b : get_succeeding_edges(block))
walk_from(seen_blocks, b, op);
}
}
uint32_t find_loop_dominator(uint32_t block) const;
bool node_terminates_control_flow_in_sub_graph(BlockID from, BlockID to) const;
private:
struct VisitOrder
{
int &get()
{
return v;
}
const int &get() const
{
return v;
}
int v = -1;
};
Compiler &compiler;
const SPIRFunction &func;
std::unordered_map<uint32_t, SmallVector<uint32_t>> preceding_edges;
std::unordered_map<uint32_t, SmallVector<uint32_t>> succeeding_edges;
std::unordered_map<uint32_t, uint32_t> immediate_dominators;
std::unordered_map<uint32_t, VisitOrder> visit_order;
SmallVector<uint32_t> post_order;
SmallVector<uint32_t> empty_vector;
void add_branch(uint32_t from, uint32_t to);
void build_post_order_visit_order();
void build_immediate_dominators();
bool post_order_visit(uint32_t block);
uint32_t visit_count = 0;
bool is_back_edge(uint32_t to) const;
bool has_visited_forward_edge(uint32_t to) const;
};
class DominatorBuilder
{
public:
DominatorBuilder(const CFG &cfg);
void add_block(uint32_t block);
uint32_t get_dominator() const
{
return dominator;
}
void lift_continue_block_dominator();
private:
const CFG &cfg;
uint32_t dominator = 0;
};
} // namespace SPIRV_CROSS_NAMESPACE
#endif

1943
thirdparty/spirv-cross/spirv_common.hpp vendored Normal file
View File

File diff suppressed because it is too large Load Diff

5668
thirdparty/spirv-cross/spirv_cross.cpp vendored Normal file
View File

File diff suppressed because it is too large Load Diff

1182
thirdparty/spirv-cross/spirv_cross.hpp vendored Normal file
View File

File diff suppressed because it is too large Load Diff

756
thirdparty/spirv-cross/spirv_cross_containers.hpp vendored Normal file
View File

@ -0,0 +1,756 @@
/*
* Copyright 2019-2021 Hans-Kristian Arntzen
* SPDX-License-Identifier: Apache-2.0 OR MIT
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
/*
* At your option, you may choose to accept this material under either:
* 1. The Apache License, Version 2.0, found at <http://www.apache.org/licenses/LICENSE-2.0>, or
* 2. The MIT License, found at <http://opensource.org/licenses/MIT>.
*/
#ifndef SPIRV_CROSS_CONTAINERS_HPP
#define SPIRV_CROSS_CONTAINERS_HPP
#include "spirv_cross_error_handling.hpp"
#include <algorithm>
#include <exception>
#include <functional>
#include <iterator>
#include <limits>
#include <memory>
#include <stack>
#include <stddef.h>
#include <stdint.h>
#include <stdlib.h>
#include <string.h>
#include <type_traits>
#include <unordered_map>
#include <unordered_set>
#include <utility>
#include <vector>
#ifdef SPIRV_CROSS_NAMESPACE_OVERRIDE
#define SPIRV_CROSS_NAMESPACE SPIRV_CROSS_NAMESPACE_OVERRIDE
#else
#define SPIRV_CROSS_NAMESPACE spirv_cross
#endif
namespace SPIRV_CROSS_NAMESPACE
{
#ifndef SPIRV_CROSS_FORCE_STL_TYPES
// std::aligned_storage does not support size == 0, so roll our own.
template <typename T, size_t N>
class AlignedBuffer
{
public:
T *data()
{
#if defined(_MSC_VER) && _MSC_VER < 1900
// MSVC 2013 workarounds, sigh ...
// Only use this workaround on MSVC 2013 due to some confusion around default initialized unions.
// Spec seems to suggest the memory will be zero-initialized, which is *not* what we want.
return reinterpret_cast<T *>(u.aligned_char);
#else
return reinterpret_cast<T *>(aligned_char);
#endif
}
private:
#if defined(_MSC_VER) && _MSC_VER < 1900
// MSVC 2013 workarounds, sigh ...
union
{
char aligned_char[sizeof(T) * N];
double dummy_aligner;
} u;
#else
alignas(T) char aligned_char[sizeof(T) * N];
#endif
};
template <typename T>
class AlignedBuffer<T, 0>
{
public:
T *data()
{
return nullptr;
}
};
// An immutable version of SmallVector which erases type information about storage.
template <typename T>
class VectorView
{
public:
T &operator[](size_t i) SPIRV_CROSS_NOEXCEPT
{
return ptr[i];
}
const T &operator[](size_t i) const SPIRV_CROSS_NOEXCEPT
{
return ptr[i];
}
bool empty() const SPIRV_CROSS_NOEXCEPT
{
return buffer_size == 0;
}
size_t size() const SPIRV_CROSS_NOEXCEPT
{
return buffer_size;
}
T *data() SPIRV_CROSS_NOEXCEPT
{
return ptr;
}
const T *data() const SPIRV_CROSS_NOEXCEPT
{
return ptr;
}
T *begin() SPIRV_CROSS_NOEXCEPT
{
return ptr;
}
T *end() SPIRV_CROSS_NOEXCEPT
{
return ptr + buffer_size;
}
const T *begin() const SPIRV_CROSS_NOEXCEPT
{
return ptr;
}
const T *end() const SPIRV_CROSS_NOEXCEPT
{
return ptr + buffer_size;
}
T &front() SPIRV_CROSS_NOEXCEPT
{
return ptr[0];
}
const T &front() const SPIRV_CROSS_NOEXCEPT
{
return ptr[0];
}
T &back() SPIRV_CROSS_NOEXCEPT
{
return ptr[buffer_size - 1];
}
const T &back() const SPIRV_CROSS_NOEXCEPT
{
return ptr[buffer_size - 1];
}
// Makes it easier to consume SmallVector.
#if defined(_MSC_VER) && _MSC_VER < 1900
explicit operator std::vector<T>() const
{
// Another MSVC 2013 workaround. It does not understand lvalue/rvalue qualified operations.
return std::vector<T>(ptr, ptr + buffer_size);
}
#else
// Makes it easier to consume SmallVector.
explicit operator std::vector<T>() const &
{
return std::vector<T>(ptr, ptr + buffer_size);
}
// If we are converting as an r-value, we can pilfer our elements.
explicit operator std::vector<T>() &&
{
return std::vector<T>(std::make_move_iterator(ptr), std::make_move_iterator(ptr + buffer_size));
}
#endif
// Avoid sliced copies. Base class should only be read as a reference.
VectorView(const VectorView &) = delete;
void operator=(const VectorView &) = delete;
protected:
VectorView() = default;
T *ptr = nullptr;
size_t buffer_size = 0;
};
// Simple vector which supports up to N elements inline, without malloc/free.
// We use a lot of throwaway vectors all over the place which triggers allocations.
// This class only implements the subset of std::vector we need in SPIRV-Cross.
// It is *NOT* a drop-in replacement in general projects.
template <typename T, size_t N = 8>
class SmallVector : public VectorView<T>
{
public:
SmallVector() SPIRV_CROSS_NOEXCEPT
{
this->ptr = stack_storage.data();
buffer_capacity = N;
}
template <typename U>
SmallVector(const U *arg_list_begin, const U *arg_list_end) SPIRV_CROSS_NOEXCEPT : SmallVector()
{
auto count = size_t(arg_list_end - arg_list_begin);
reserve(count);
for (size_t i = 0; i < count; i++, arg_list_begin++)
new (&this->ptr[i]) T(*arg_list_begin);
this->buffer_size = count;
}
template <typename U>
SmallVector(std::initializer_list<U> init) SPIRV_CROSS_NOEXCEPT : SmallVector(init.begin(), init.end())
{
}
template <typename U, size_t M>
explicit SmallVector(const U (&init)[M]) SPIRV_CROSS_NOEXCEPT : SmallVector(init, init + M)
{
}
SmallVector(SmallVector &&other) SPIRV_CROSS_NOEXCEPT : SmallVector()
{
*this = std::move(other);
}
SmallVector &operator=(SmallVector &&other) SPIRV_CROSS_NOEXCEPT
{
clear();
if (other.ptr != other.stack_storage.data())
{
// Pilfer allocated pointer.
if (this->ptr != stack_storage.data())
free(this->ptr);
this->ptr = other.ptr;
this->buffer_size = other.buffer_size;
buffer_capacity = other.buffer_capacity;
other.ptr = nullptr;
other.buffer_size = 0;
other.buffer_capacity = 0;
}
else
{
// Need to move the stack contents individually.
reserve(other.buffer_size);
for (size_t i = 0; i < other.buffer_size; i++)
{
new (&this->ptr[i]) T(std::move(other.ptr[i]));
other.ptr[i].~T();
}
this->buffer_size = other.buffer_size;
other.buffer_size = 0;
}
return *this;
}
SmallVector(const SmallVector &other) SPIRV_CROSS_NOEXCEPT : SmallVector()
{
*this = other;
}
SmallVector &operator=(const SmallVector &other) SPIRV_CROSS_NOEXCEPT
{
if (this == &other)
return *this;
clear();
reserve(other.buffer_size);
for (size_t i = 0; i < other.buffer_size; i++)
new (&this->ptr[i]) T(other.ptr[i]);
this->buffer_size = other.buffer_size;
return *this;
}
explicit SmallVector(size_t count) SPIRV_CROSS_NOEXCEPT : SmallVector()
{
resize(count);
}
~SmallVector()
{
clear();
if (this->ptr != stack_storage.data())
free(this->ptr);
}
void clear() SPIRV_CROSS_NOEXCEPT
{
for (size_t i = 0; i < this->buffer_size; i++)
this->ptr[i].~T();
this->buffer_size = 0;
}
void push_back(const T &t) SPIRV_CROSS_NOEXCEPT
{
reserve(this->buffer_size + 1);
new (&this->ptr[this->buffer_size]) T(t);
this->buffer_size++;
}
void push_back(T &&t) SPIRV_CROSS_NOEXCEPT
{
reserve(this->buffer_size + 1);
new (&this->ptr[this->buffer_size]) T(std::move(t));
this->buffer_size++;
}
void pop_back() SPIRV_CROSS_NOEXCEPT
{
// Work around false positive warning on GCC 8.3.
// Calling pop_back on empty vector is undefined.
if (!this->empty())
resize(this->buffer_size - 1);
}
template <typename... Ts>
void emplace_back(Ts &&... ts) SPIRV_CROSS_NOEXCEPT
{
reserve(this->buffer_size + 1);
new (&this->ptr[this->buffer_size]) T(std::forward<Ts>(ts)...);
this->buffer_size++;
}
void reserve(size_t count) SPIRV_CROSS_NOEXCEPT
{
if ((count > (std::numeric_limits<size_t>::max)() / sizeof(T)) ||
(count > (std::numeric_limits<size_t>::max)() / 2))
{
// Only way this should ever happen is with garbage input, terminate.
std::terminate();
}
if (count > buffer_capacity)
{
size_t target_capacity = buffer_capacity;
if (target_capacity == 0)
target_capacity = 1;
// Weird parens works around macro issues on Windows if NOMINMAX is not used.
target_capacity = (std::max)(target_capacity, N);
// Need to ensure there is a POT value of target capacity which is larger than count,
// otherwise this will overflow.
while (target_capacity < count)
target_capacity <<= 1u;
T *new_buffer =
target_capacity > N ? static_cast<T *>(malloc(target_capacity * sizeof(T))) : stack_storage.data();
// If we actually fail this malloc, we are hosed anyways, there is no reason to attempt recovery.
if (!new_buffer)
std::terminate();
// In case for some reason two allocations both come from same stack.
if (new_buffer != this->ptr)
{
// We don't deal with types which can throw in move constructor.
for (size_t i = 0; i < this->buffer_size; i++)
{
new (&new_buffer[i]) T(std::move(this->ptr[i]));
this->ptr[i].~T();
}
}
if (this->ptr != stack_storage.data())
free(this->ptr);
this->ptr = new_buffer;
buffer_capacity = target_capacity;
}
}
void insert(T *itr, const T *insert_begin, const T *insert_end) SPIRV_CROSS_NOEXCEPT
{
auto count = size_t(insert_end - insert_begin);
if (itr == this->end())
{
reserve(this->buffer_size + count);
for (size_t i = 0; i < count; i++, insert_begin++)
new (&this->ptr[this->buffer_size + i]) T(*insert_begin);
this->buffer_size += count;
}
else
{
if (this->buffer_size + count > buffer_capacity)
{
auto target_capacity = this->buffer_size + count;
if (target_capacity == 0)
target_capacity = 1;
if (target_capacity < N)
target_capacity = N;
while (target_capacity < count)
target_capacity <<= 1u;
// Need to allocate new buffer. Move everything to a new buffer.
T *new_buffer =
target_capacity > N ? static_cast<T *>(malloc(target_capacity * sizeof(T))) : stack_storage.data();
// If we actually fail this malloc, we are hosed anyways, there is no reason to attempt recovery.
if (!new_buffer)
std::terminate();
// First, move elements from source buffer to new buffer.
// We don't deal with types which can throw in move constructor.
auto *target_itr = new_buffer;
auto *original_source_itr = this->begin();
if (new_buffer != this->ptr)
{
while (original_source_itr != itr)
{
new (target_itr) T(std::move(*original_source_itr));
original_source_itr->~T();
++original_source_itr;
++target_itr;
}
}
// Copy-construct new elements.
for (auto *source_itr = insert_begin; source_itr != insert_end; ++source_itr, ++target_itr)
new (target_itr) T(*source_itr);
// Move over the other half.
if (new_buffer != this->ptr || insert_begin != insert_end)
{
while (original_source_itr != this->end())
{
new (target_itr) T(std::move(*original_source_itr));
original_source_itr->~T();
++original_source_itr;
++target_itr;
}
}
if (this->ptr != stack_storage.data())
free(this->ptr);
this->ptr = new_buffer;
buffer_capacity = target_capacity;
}
else
{
// Move in place, need to be a bit careful about which elements are constructed and which are not.
// Move the end and construct the new elements.
auto *target_itr = this->end() + count;
auto *source_itr = this->end();
while (target_itr != this->end() && source_itr != itr)
{
--target_itr;
--source_itr;
new (target_itr) T(std::move(*source_itr));
}
// For already constructed elements we can move-assign.
std::move_backward(itr, source_itr, target_itr);
// For the inserts which go to already constructed elements, we can do a plain copy.
while (itr != this->end() && insert_begin != insert_end)
*itr++ = *insert_begin++;
// For inserts into newly allocated memory, we must copy-construct instead.
while (insert_begin != insert_end)
{
new (itr) T(*insert_begin);
++itr;
++insert_begin;
}
}
this->buffer_size += count;
}
}
void insert(T *itr, const T &value) SPIRV_CROSS_NOEXCEPT
{
insert(itr, &value, &value + 1);
}
T *erase(T *itr) SPIRV_CROSS_NOEXCEPT
{
std::move(itr + 1, this->end(), itr);
this->ptr[--this->buffer_size].~T();
return itr;
}
void erase(T *start_erase, T *end_erase) SPIRV_CROSS_NOEXCEPT
{
if (end_erase == this->end())
{
resize(size_t(start_erase - this->begin()));
}
else
{
auto new_size = this->buffer_size - (end_erase - start_erase);
std::move(end_erase, this->end(), start_erase);
resize(new_size);
}
}
void resize(size_t new_size) SPIRV_CROSS_NOEXCEPT
{
if (new_size < this->buffer_size)
{
for (size_t i = new_size; i < this->buffer_size; i++)
this->ptr[i].~T();
}
else if (new_size > this->buffer_size)
{
reserve(new_size);
for (size_t i = this->buffer_size; i < new_size; i++)
new (&this->ptr[i]) T();
}
this->buffer_size = new_size;
}
private:
size_t buffer_capacity = 0;
AlignedBuffer<T, N> stack_storage;
};
// A vector without stack storage.
// Could also be a typedef-ed to std::vector,
// but might as well use the one we have.
template <typename T>
using Vector = SmallVector<T, 0>;
#else // SPIRV_CROSS_FORCE_STL_TYPES
template <typename T, size_t N = 8>
using SmallVector = std::vector<T>;
template <typename T>
using Vector = std::vector<T>;
template <typename T>
using VectorView = std::vector<T>;
#endif // SPIRV_CROSS_FORCE_STL_TYPES
// An object pool which we use for allocating IVariant-derived objects.
// We know we are going to allocate a bunch of objects of each type,
// so amortize the mallocs.
class ObjectPoolBase
{
public:
virtual ~ObjectPoolBase() = default;
virtual void deallocate_opaque(void *ptr) = 0;
};
template <typename T>
class ObjectPool : public ObjectPoolBase
{
public:
explicit ObjectPool(unsigned start_object_count_ = 16)
: start_object_count(start_object_count_)
{
}
template <typename... P>
T *allocate(P &&... p)
{
if (vacants.empty())
{
unsigned num_objects = start_object_count << memory.size();
T *ptr = static_cast<T *>(malloc(num_objects * sizeof(T)));
if (!ptr)
return nullptr;
vacants.reserve(num_objects);
for (unsigned i = 0; i < num_objects; i++)
vacants.push_back(&ptr[i]);
memory.emplace_back(ptr);
}
T *ptr = vacants.back();
vacants.pop_back();
new (ptr) T(std::forward<P>(p)...);
return ptr;
}
void deallocate(T *ptr)
{
ptr->~T();
vacants.push_back(ptr);
}
void deallocate_opaque(void *ptr) override
{
deallocate(static_cast<T *>(ptr));
}
void clear()
{
vacants.clear();
memory.clear();
}
protected:
Vector<T *> vacants;
struct MallocDeleter
{
void operator()(T *ptr)
{
::free(ptr);
}
};
SmallVector<std::unique_ptr<T, MallocDeleter>> memory;
unsigned start_object_count;
};
template <size_t StackSize = 4096, size_t BlockSize = 4096>
class StringStream
{
public:
StringStream()
{
reset();
}
~StringStream()
{
reset();
}
// Disable copies and moves. Makes it easier to implement, and we don't need it.
StringStream(const StringStream &) = delete;
void operator=(const StringStream &) = delete;
template <typename T, typename std::enable_if<!std::is_floating_point<T>::value, int>::type = 0>
StringStream &operator<<(const T &t)
{
auto s = std::to_string(t);
append(s.data(), s.size());
return *this;
}
// Only overload this to make float/double conversions ambiguous.
StringStream &operator<<(uint32_t v)
{
auto s = std::to_string(v);
append(s.data(), s.size());
return *this;
}
StringStream &operator<<(char c)
{
append(&c, 1);
return *this;
}
StringStream &operator<<(const std::string &s)
{
append(s.data(), s.size());
return *this;
}
StringStream &operator<<(const char *s)
{
append(s, strlen(s));
return *this;
}
template <size_t N>
StringStream &operator<<(const char (&s)[N])
{
append(s, strlen(s));
return *this;
}
std::string str() const
{
std::string ret;
size_t target_size = 0;
for (auto &saved : saved_buffers)
target_size += saved.offset;
target_size += current_buffer.offset;
ret.reserve(target_size);
for (auto &saved : saved_buffers)
ret.insert(ret.end(), saved.buffer, saved.buffer + saved.offset);
ret.insert(ret.end(), current_buffer.buffer, current_buffer.buffer + current_buffer.offset);
return ret;
}
void reset()
{
for (auto &saved : saved_buffers)
if (saved.buffer != stack_buffer)
free(saved.buffer);
if (current_buffer.buffer != stack_buffer)
free(current_buffer.buffer);
saved_buffers.clear();
current_buffer.buffer = stack_buffer;
current_buffer.offset = 0;
current_buffer.size = sizeof(stack_buffer);
}
private:
struct Buffer
{
char *buffer = nullptr;
size_t offset = 0;
size_t size = 0;
};
Buffer current_buffer;
char stack_buffer[StackSize];
SmallVector<Buffer> saved_buffers;
void append(const char *s, size_t len)
{
size_t avail = current_buffer.size - current_buffer.offset;
if (avail < len)
{
if (avail > 0)
{
memcpy(current_buffer.buffer + current_buffer.offset, s, avail);
s += avail;
len -= avail;
current_buffer.offset += avail;
}
saved_buffers.push_back(current_buffer);
size_t target_size = len > BlockSize ? len : BlockSize;
current_buffer.buffer = static_cast<char *>(malloc(target_size));
if (!current_buffer.buffer)
SPIRV_CROSS_THROW("Out of memory.");
memcpy(current_buffer.buffer, s, len);
current_buffer.offset = len;
current_buffer.size = target_size;
}
else
{
memcpy(current_buffer.buffer + current_buffer.offset, s, len);
current_buffer.offset += len;
}
}
};
} // namespace SPIRV_CROSS_NAMESPACE
#endif

99
thirdparty/spirv-cross/spirv_cross_error_handling.hpp vendored Normal file
View File

@ -0,0 +1,99 @@
/*
* Copyright 2015-2021 Arm Limited
* SPDX-License-Identifier: Apache-2.0 OR MIT
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
/*
* At your option, you may choose to accept this material under either:
* 1. The Apache License, Version 2.0, found at <http://www.apache.org/licenses/LICENSE-2.0>, or
* 2. The MIT License, found at <http://opensource.org/licenses/MIT>.
*/
#ifndef SPIRV_CROSS_ERROR_HANDLING
#define SPIRV_CROSS_ERROR_HANDLING
#include <stdio.h>
#include <stdlib.h>
#include <string>
#ifndef SPIRV_CROSS_EXCEPTIONS_TO_ASSERTIONS
#include <stdexcept>
#endif
#ifdef SPIRV_CROSS_NAMESPACE_OVERRIDE
#define SPIRV_CROSS_NAMESPACE SPIRV_CROSS_NAMESPACE_OVERRIDE
#else
#define SPIRV_CROSS_NAMESPACE spirv_cross
#endif
namespace SPIRV_CROSS_NAMESPACE
{
#ifdef SPIRV_CROSS_EXCEPTIONS_TO_ASSERTIONS
#if !defined(_MSC_VER) || defined(__clang__)
[[noreturn]]
#elif defined(_MSC_VER)
__declspec(noreturn)
#endif
inline void
report_and_abort(const std::string &msg)
{
#ifdef NDEBUG
(void)msg;
#else
fprintf(stderr, "There was a compiler error: %s\n", msg.c_str());
#endif
fflush(stderr);
abort();
}
#define SPIRV_CROSS_THROW(x) report_and_abort(x)
#else
class CompilerError : public std::runtime_error
{
public:
explicit CompilerError(const std::string &str)
: std::runtime_error(str)
{
}
explicit CompilerError(const char *str)
: std::runtime_error(str)
{
}
};
#define SPIRV_CROSS_THROW(x) throw CompilerError(x)
#endif
// MSVC 2013 does not have noexcept. We need this for Variant to get move constructor to work correctly
// instead of copy constructor.
// MSVC 2013 ignores that move constructors cannot throw in std::vector, so just don't define it.
#if defined(_MSC_VER) && _MSC_VER < 1900
#define SPIRV_CROSS_NOEXCEPT
#else
#define SPIRV_CROSS_NOEXCEPT noexcept
#endif
#if __cplusplus >= 201402l
#define SPIRV_CROSS_DEPRECATED(reason) [[deprecated(reason)]]
#elif defined(__GNUC__)
#define SPIRV_CROSS_DEPRECATED(reason) __attribute__((deprecated))
#elif defined(_MSC_VER)
#define SPIRV_CROSS_DEPRECATED(reason) __declspec(deprecated(reason))
#else
#define SPIRV_CROSS_DEPRECATED(reason)
#endif
} // namespace SPIRV_CROSS_NAMESPACE
#endif

1083
thirdparty/spirv-cross/spirv_cross_parsed_ir.cpp vendored Normal file
View File

File diff suppressed because it is too large Load Diff

256
thirdparty/spirv-cross/spirv_cross_parsed_ir.hpp vendored Normal file
View File

@ -0,0 +1,256 @@
/*
* Copyright 2018-2021 Arm Limited
* SPDX-License-Identifier: Apache-2.0 OR MIT
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
/*
* At your option, you may choose to accept this material under either:
* 1. The Apache License, Version 2.0, found at <http://www.apache.org/licenses/LICENSE-2.0>, or
* 2. The MIT License, found at <http://opensource.org/licenses/MIT>.
*/
#ifndef SPIRV_CROSS_PARSED_IR_HPP
#define SPIRV_CROSS_PARSED_IR_HPP
#include "spirv_common.hpp"
#include <stdint.h>
#include <unordered_map>
namespace SPIRV_CROSS_NAMESPACE
{
// This data structure holds all information needed to perform cross-compilation and reflection.
// It is the output of the Parser, but any implementation could create this structure.
// It is intentionally very "open" and struct-like with some helper functions to deal with decorations.
// Parser is the reference implementation of how this data structure should be filled in.
class ParsedIR
{
private:
// This must be destroyed after the "ids" vector.
std::unique_ptr<ObjectPoolGroup> pool_group;
public:
ParsedIR();
// Due to custom allocations from object pools, we cannot use a default copy constructor.
ParsedIR(const ParsedIR &other);
ParsedIR &operator=(const ParsedIR &other);
// Moves are unproblematic, but we need to implement it anyways, since MSVC 2013 does not understand
// how to default-implement these.
ParsedIR(ParsedIR &&other) SPIRV_CROSS_NOEXCEPT;
ParsedIR &operator=(ParsedIR &&other) SPIRV_CROSS_NOEXCEPT;
// Resizes ids, meta and block_meta.
void set_id_bounds(uint32_t bounds);
// The raw SPIR-V, instructions and opcodes refer to this by offset + count.
std::vector<uint32_t> spirv;
// Holds various data structures which inherit from IVariant.
SmallVector<Variant> ids;
// Various meta data for IDs, decorations, names, etc.
std::unordered_map<ID, Meta> meta;
// Holds all IDs which have a certain type.
// This is needed so we can iterate through a specific kind of resource quickly,
// and in-order of module declaration.
SmallVector<ID> ids_for_type[TypeCount];
// Special purpose lists which contain a union of types.
// This is needed so we can declare specialization constants and structs in an interleaved fashion,
// among other things.
// Constants can be undef or of struct type, and struct array sizes can use specialization constants.
SmallVector<ID> ids_for_constant_undef_or_type;
SmallVector<ID> ids_for_constant_or_variable;
// We need to keep track of the width the Ops that contains a type for the
// OpSwitch instruction, since this one doesn't contains the type in the
// instruction itself. And in some case we need to cast the condition to
// wider types. We only need the width to do the branch fixup since the
// type check itself can be done at runtime
std::unordered_map<ID, uint32_t> load_type_width;
// Declared capabilities and extensions in the SPIR-V module.
// Not really used except for reflection at the moment.
SmallVector<spv::Capability> declared_capabilities;
SmallVector<std::string> declared_extensions;
// Meta data about blocks. The cross-compiler needs to query if a block is either of these types.
// It is a bitset as there can be more than one tag per block.
enum BlockMetaFlagBits
{
BLOCK_META_LOOP_HEADER_BIT = 1 << 0,
BLOCK_META_CONTINUE_BIT = 1 << 1,
BLOCK_META_LOOP_MERGE_BIT = 1 << 2,
BLOCK_META_SELECTION_MERGE_BIT = 1 << 3,
BLOCK_META_MULTISELECT_MERGE_BIT = 1 << 4
};
using BlockMetaFlags = uint8_t;
SmallVector<BlockMetaFlags> block_meta;
std::unordered_map<BlockID, BlockID> continue_block_to_loop_header;
// Normally, we'd stick SPIREntryPoint in ids array, but it conflicts with SPIRFunction.
// Entry points can therefore be seen as some sort of meta structure.
std::unordered_map<FunctionID, SPIREntryPoint> entry_points;
FunctionID default_entry_point = 0;
struct Source
{
uint32_t version = 0;
bool es = false;
bool known = false;
bool hlsl = false;
Source() = default;
};
Source source;
spv::AddressingModel addressing_model = spv::AddressingModelMax;
spv::MemoryModel memory_model = spv::MemoryModelMax;
// Decoration handling methods.
// Can be useful for simple "raw" reflection.
// However, most members are here because the Parser needs most of these,
// and might as well just have the whole suite of decoration/name handling in one place.
void set_name(ID id, const std::string &name);
const std::string &get_name(ID id) const;
void set_decoration(ID id, spv::Decoration decoration, uint32_t argument = 0);
void set_decoration_string(ID id, spv::Decoration decoration, const std::string &argument);
bool has_decoration(ID id, spv::Decoration decoration) const;
uint32_t get_decoration(ID id, spv::Decoration decoration) const;
const std::string &get_decoration_string(ID id, spv::Decoration decoration) const;
const Bitset &get_decoration_bitset(ID id) const;
void unset_decoration(ID id, spv::Decoration decoration);
// Decoration handling methods (for members of a struct).
void set_member_name(TypeID id, uint32_t index, const std::string &name);
const std::string &get_member_name(TypeID id, uint32_t index) const;
void set_member_decoration(TypeID id, uint32_t index, spv::Decoration decoration, uint32_t argument = 0);
void set_member_decoration_string(TypeID id, uint32_t index, spv::Decoration decoration,
const std::string &argument);
uint32_t get_member_decoration(TypeID id, uint32_t index, spv::Decoration decoration) const;
const std::string &get_member_decoration_string(TypeID id, uint32_t index, spv::Decoration decoration) const;
bool has_member_decoration(TypeID id, uint32_t index, spv::Decoration decoration) const;
const Bitset &get_member_decoration_bitset(TypeID id, uint32_t index) const;
void unset_member_decoration(TypeID id, uint32_t index, spv::Decoration decoration);
void mark_used_as_array_length(ID id);
uint32_t increase_bound_by(uint32_t count);
Bitset get_buffer_block_flags(const SPIRVariable &var) const;
Bitset get_buffer_block_type_flags(const SPIRType &type) const;
void add_typed_id(Types type, ID id);
void remove_typed_id(Types type, ID id);
class LoopLock
{
public:
explicit LoopLock(uint32_t *counter);
LoopLock(const LoopLock &) = delete;
void operator=(const LoopLock &) = delete;
LoopLock(LoopLock &&other) SPIRV_CROSS_NOEXCEPT;
LoopLock &operator=(LoopLock &&other) SPIRV_CROSS_NOEXCEPT;
~LoopLock();
private:
uint32_t *lock = nullptr;
};
// This must be held while iterating over a type ID array.
// It is undefined if someone calls set<>() while we're iterating over a data structure, so we must
// make sure that this case is avoided.
// If we have a hard lock, it is an error to call set<>(), and an exception is thrown.
// If we have a soft lock, we silently ignore any additions to the typed arrays.
// This should only be used for physical ID remapping where we need to create an ID, but we will never
// care about iterating over them.
LoopLock create_loop_hard_lock() const;
LoopLock create_loop_soft_lock() const;
template <typename T, typename Op>
void for_each_typed_id(const Op &op)
{
auto loop_lock = create_loop_hard_lock();
for (auto &id : ids_for_type[T::type])
{
if (ids[id].get_type() == static_cast<Types>(T::type))
op(id, get<T>(id));
}
}
template <typename T, typename Op>
void for_each_typed_id(const Op &op) const
{
auto loop_lock = create_loop_hard_lock();
for (auto &id : ids_for_type[T::type])
{
if (ids[id].get_type() == static_cast<Types>(T::type))
op(id, get<T>(id));
}
}
template <typename T>
void reset_all_of_type()
{
reset_all_of_type(static_cast<Types>(T::type));
}
void reset_all_of_type(Types type);
Meta *find_meta(ID id);
const Meta *find_meta(ID id) const;
const std::string &get_empty_string() const
{
return empty_string;
}
void make_constant_null(uint32_t id, uint32_t type, bool add_to_typed_id_set);
void fixup_reserved_names();
static void sanitize_underscores(std::string &str);
static void sanitize_identifier(std::string &str, bool member, bool allow_reserved_prefixes);
static bool is_globally_reserved_identifier(std::string &str, bool allow_reserved_prefixes);
uint32_t get_spirv_version() const;
private:
template <typename T>
T &get(uint32_t id)
{
return variant_get<T>(ids[id]);
}
template <typename T>
const T &get(uint32_t id) const
{
return variant_get<T>(ids[id]);
}
mutable uint32_t loop_iteration_depth_hard = 0;
mutable uint32_t loop_iteration_depth_soft = 0;
std::string empty_string;
Bitset cleared_bitset;
std::unordered_set<uint32_t> meta_needing_name_fixup;
};
} // namespace SPIRV_CROSS_NAMESPACE
#endif

77
thirdparty/spirv-cross/spirv_cross_util.cpp vendored Normal file
View File

@ -0,0 +1,77 @@
/*
* Copyright 2015-2021 Arm Limited
* SPDX-License-Identifier: Apache-2.0 OR MIT
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
/*
* At your option, you may choose to accept this material under either:
* 1. The Apache License, Version 2.0, found at <http://www.apache.org/licenses/LICENSE-2.0>, or
* 2. The MIT License, found at <http://opensource.org/licenses/MIT>.
*/
#include "spirv_cross_util.hpp"
#include "spirv_common.hpp"
using namespace spv;
using namespace SPIRV_CROSS_NAMESPACE;
namespace spirv_cross_util
{
void rename_interface_variable(Compiler &compiler, const SmallVector<Resource> &resources, uint32_t location,
const std::string &name)
{
for (auto &v : resources)
{
if (!compiler.has_decoration(v.id, spv::DecorationLocation))
continue;
auto loc = compiler.get_decoration(v.id, spv::DecorationLocation);
if (loc != location)
continue;
auto &type = compiler.get_type(v.base_type_id);
// This is more of a friendly variant. If we need to rename interface variables, we might have to rename
// structs as well and make sure all the names match up.
if (type.basetype == SPIRType::Struct)
{
compiler.set_name(v.base_type_id, join("SPIRV_Cross_Interface_Location", location));
for (uint32_t i = 0; i < uint32_t(type.member_types.size()); i++)
compiler.set_member_name(v.base_type_id, i, join("InterfaceMember", i));
}
compiler.set_name(v.id, name);
}
}
void inherit_combined_sampler_bindings(Compiler &compiler)
{
auto &samplers = compiler.get_combined_image_samplers();
for (auto &s : samplers)
{
if (compiler.has_decoration(s.image_id, spv::DecorationDescriptorSet))
{
uint32_t set = compiler.get_decoration(s.image_id, spv::DecorationDescriptorSet);
compiler.set_decoration(s.combined_id, spv::DecorationDescriptorSet, set);
}
if (compiler.has_decoration(s.image_id, spv::DecorationBinding))
{
uint32_t binding = compiler.get_decoration(s.image_id, spv::DecorationBinding);
compiler.set_decoration(s.combined_id, spv::DecorationBinding, binding);
}
}
}
} // namespace spirv_cross_util

37
thirdparty/spirv-cross/spirv_cross_util.hpp vendored Normal file
View File

@ -0,0 +1,37 @@
/*
* Copyright 2015-2021 Arm Limited
* SPDX-License-Identifier: Apache-2.0 OR MIT
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
/*
* At your option, you may choose to accept this material under either:
* 1. The Apache License, Version 2.0, found at <http://www.apache.org/licenses/LICENSE-2.0>, or
* 2. The MIT License, found at <http://opensource.org/licenses/MIT>.
*/
#ifndef SPIRV_CROSS_UTIL_HPP
#define SPIRV_CROSS_UTIL_HPP
#include "spirv_cross.hpp"
namespace spirv_cross_util
{
void rename_interface_variable(SPIRV_CROSS_NAMESPACE::Compiler &compiler,
const SPIRV_CROSS_NAMESPACE::SmallVector<SPIRV_CROSS_NAMESPACE::Resource> &resources,
uint32_t location, const std::string &name);
void inherit_combined_sampler_bindings(SPIRV_CROSS_NAMESPACE::Compiler &compiler);
} // namespace spirv_cross_util
#endif

19109
thirdparty/spirv-cross/spirv_glsl.cpp vendored Normal file
View File

File diff suppressed because it is too large Load Diff

1074
thirdparty/spirv-cross/spirv_glsl.hpp vendored Normal file
View File

File diff suppressed because it is too large Load Diff

18810
thirdparty/spirv-cross/spirv_msl.cpp vendored Normal file
View File

File diff suppressed because it is too large Load Diff

1349
thirdparty/spirv-cross/spirv_msl.hpp vendored Normal file
View File

File diff suppressed because it is too large Load Diff

1337
thirdparty/spirv-cross/spirv_parser.cpp vendored Normal file
View File

File diff suppressed because it is too large Load Diff

103
thirdparty/spirv-cross/spirv_parser.hpp vendored Normal file
View File

@ -0,0 +1,103 @@
/*
* Copyright 2018-2021 Arm Limited
* SPDX-License-Identifier: Apache-2.0 OR MIT
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
/*
* At your option, you may choose to accept this material under either:
* 1. The Apache License, Version 2.0, found at <http://www.apache.org/licenses/LICENSE-2.0>, or
* 2. The MIT License, found at <http://opensource.org/licenses/MIT>.
*/
#ifndef SPIRV_CROSS_PARSER_HPP
#define SPIRV_CROSS_PARSER_HPP
#include "spirv_cross_parsed_ir.hpp"
#include <stdint.h>
namespace SPIRV_CROSS_NAMESPACE
{
class Parser
{
public:
Parser(const uint32_t *spirv_data, size_t word_count);
Parser(std::vector<uint32_t> spirv);
void parse();
ParsedIR &get_parsed_ir()
{
return ir;
}
private:
ParsedIR ir;
SPIRFunction *current_function = nullptr;
SPIRBlock *current_block = nullptr;
// For workarounds.
bool ignore_trailing_block_opcodes = false;
void parse(const Instruction &instr);
const uint32_t *stream(const Instruction &instr) const;
template <typename T, typename... P>
T &set(uint32_t id, P &&... args)
{
ir.add_typed_id(static_cast<Types>(T::type), id);
auto &var = variant_set<T>(ir.ids[id], std::forward<P>(args)...);
var.self = id;
return var;
}
template <typename T>
T &get(uint32_t id)
{
return variant_get<T>(ir.ids[id]);
}
template <typename T>
T *maybe_get(uint32_t id)
{
if (ir.ids[id].get_type() == static_cast<Types>(T::type))
return &get<T>(id);
else
return nullptr;
}
template <typename T>
const T &get(uint32_t id) const
{
return variant_get<T>(ir.ids[id]);
}
template <typename T>
const T *maybe_get(uint32_t id) const
{
if (ir.ids[id].get_type() == T::type)
return &get<T>(id);
else
return nullptr;
}
// This must be an ordered data structure so we always pick the same type aliases.
SmallVector<uint32_t> global_struct_cache;
SmallVector<std::pair<uint32_t, uint32_t>> forward_pointer_fixups;
bool types_are_logically_equivalent(const SPIRType &a, const SPIRType &b) const;
bool variable_storage_is_aliased(const SPIRVariable &v) const;
};
} // namespace SPIRV_CROSS_NAMESPACE
#endif

710
thirdparty/spirv-cross/spirv_reflect.cpp vendored Normal file
View File

@ -0,0 +1,710 @@
/*
* Copyright 2018-2021 Bradley Austin Davis
* SPDX-License-Identifier: Apache-2.0 OR MIT
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
/*
* At your option, you may choose to accept this material under either:
* 1. The Apache License, Version 2.0, found at <http://www.apache.org/licenses/LICENSE-2.0>, or
* 2. The MIT License, found at <http://opensource.org/licenses/MIT>.
*/
#include "spirv_reflect.hpp"
#include "spirv_glsl.hpp"
#include <iomanip>
using namespace spv;
using namespace SPIRV_CROSS_NAMESPACE;
using namespace std;
namespace simple_json
{
enum class Type
{
Object,
Array,
};
using State = std::pair<Type, bool>;
using Stack = std::stack<State>;
class Stream
{
Stack stack;
StringStream<> buffer;
uint32_t indent{ 0 };
char current_locale_radix_character = '.';
public:
void set_current_locale_radix_character(char c)
{
current_locale_radix_character = c;
}
void begin_json_object();
void end_json_object();
void emit_json_key(const std::string &key);
void emit_json_key_value(const std::string &key, const std::string &value);
void emit_json_key_value(const std::string &key, bool value);
void emit_json_key_value(const std::string &key, uint32_t value);
void emit_json_key_value(const std::string &key, int32_t value);
void emit_json_key_value(const std::string &key, float value);
void emit_json_key_object(const std::string &key);
void emit_json_key_array(const std::string &key);
void begin_json_array();
void end_json_array();
void emit_json_array_value(const std::string &value);
void emit_json_array_value(uint32_t value);
void emit_json_array_value(bool value);
std::string str() const
{
return buffer.str();
}
private:
inline void statement_indent()
{
for (uint32_t i = 0; i < indent; i++)
buffer << " ";
}
template <typename T>
inline void statement_inner(T &&t)
{
buffer << std::forward<T>(t);
}
template <typename T, typename... Ts>
inline void statement_inner(T &&t, Ts &&... ts)
{
buffer << std::forward<T>(t);
statement_inner(std::forward<Ts>(ts)...);
}
template <typename... Ts>
inline void statement(Ts &&... ts)
{
statement_indent();
statement_inner(std::forward<Ts>(ts)...);
buffer << '\n';
}
template <typename... Ts>
void statement_no_return(Ts &&... ts)
{
statement_indent();
statement_inner(std::forward<Ts>(ts)...);
}
};
} // namespace simple_json
using namespace simple_json;
// Hackery to emit JSON without using nlohmann/json C++ library (which requires a
// higher level of compiler compliance than is required by SPIRV-Cross
void Stream::begin_json_array()
{
if (!stack.empty() && stack.top().second)
{
statement_inner(",\n");
}
statement("[");
++indent;
stack.emplace(Type::Array, false);
}
void Stream::end_json_array()
{
if (stack.empty() || stack.top().first != Type::Array)
SPIRV_CROSS_THROW("Invalid JSON state");
if (stack.top().second)
{
statement_inner("\n");
}
--indent;
statement_no_return("]");
stack.pop();
if (!stack.empty())
{
stack.top().second = true;
}
}
void Stream::emit_json_array_value(const std::string &value)
{
if (stack.empty() || stack.top().first != Type::Array)
SPIRV_CROSS_THROW("Invalid JSON state");
if (stack.top().second)
statement_inner(",\n");
statement_no_return("\"", value, "\"");
stack.top().second = true;
}
void Stream::emit_json_array_value(uint32_t value)
{
if (stack.empty() || stack.top().first != Type::Array)
SPIRV_CROSS_THROW("Invalid JSON state");
if (stack.top().second)
statement_inner(",\n");
statement_no_return(std::to_string(value));
stack.top().second = true;
}
void Stream::emit_json_array_value(bool value)
{
if (stack.empty() || stack.top().first != Type::Array)
SPIRV_CROSS_THROW("Invalid JSON state");
if (stack.top().second)
statement_inner(",\n");
statement_no_return(value ? "true" : "false");
stack.top().second = true;
}
void Stream::begin_json_object()
{
if (!stack.empty() && stack.top().second)
{
statement_inner(",\n");
}
statement("{");
++indent;
stack.emplace(Type::Object, false);
}
void Stream::end_json_object()
{
if (stack.empty() || stack.top().first != Type::Object)
SPIRV_CROSS_THROW("Invalid JSON state");
if (stack.top().second)
{
statement_inner("\n");
}
--indent;
statement_no_return("}");
stack.pop();
if (!stack.empty())
{
stack.top().second = true;
}
}
void Stream::emit_json_key(const std::string &key)
{
if (stack.empty() || stack.top().first != Type::Object)
SPIRV_CROSS_THROW("Invalid JSON state");
if (stack.top().second)
statement_inner(",\n");
statement_no_return("\"", key, "\" : ");
stack.top().second = true;
}
void Stream::emit_json_key_value(const std::string &key, const std::string &value)
{
emit_json_key(key);
statement_inner("\"", value, "\"");
}
void Stream::emit_json_key_value(const std::string &key, uint32_t value)
{
emit_json_key(key);
statement_inner(value);
}
void Stream::emit_json_key_value(const std::string &key, int32_t value)
{
emit_json_key(key);
statement_inner(value);
}
void Stream::emit_json_key_value(const std::string &key, float value)
{
emit_json_key(key);
statement_inner(convert_to_string(value, current_locale_radix_character));
}
void Stream::emit_json_key_value(const std::string &key, bool value)
{
emit_json_key(key);
statement_inner(value ? "true" : "false");
}
void Stream::emit_json_key_object(const std::string &key)
{
emit_json_key(key);
statement_inner("{\n");
++indent;
stack.emplace(Type::Object, false);
}
void Stream::emit_json_key_array(const std::string &key)
{
emit_json_key(key);
statement_inner("[\n");
++indent;
stack.emplace(Type::Array, false);
}
void CompilerReflection::set_format(const std::string &format)
{
if (format != "json")
{
SPIRV_CROSS_THROW("Unsupported format");
}
}
string CompilerReflection::compile()
{
json_stream = std::make_shared<simple_json::Stream>();
json_stream->set_current_locale_radix_character(current_locale_radix_character);
json_stream->begin_json_object();
reorder_type_alias();
emit_entry_points();
emit_types();
emit_resources();
emit_specialization_constants();
json_stream->end_json_object();
return json_stream->str();
}
static bool naturally_emit_type(const SPIRType &type)
{
return type.basetype == SPIRType::Struct && !type.pointer && type.array.empty();
}
bool CompilerReflection::type_is_reference(const SPIRType &type) const
{
// Physical pointers and arrays of physical pointers need to refer to the pointee's type.
return is_physical_pointer(type) ||
(type_is_array_of_pointers(type) && type.storage == StorageClassPhysicalStorageBuffer);
}
void CompilerReflection::emit_types()
{
bool emitted_open_tag = false;
SmallVector<uint32_t> physical_pointee_types;
// If we have physical pointers or arrays of physical pointers, it's also helpful to emit the pointee type
// and chain the type hierarchy. For POD, arrays can emit the entire type in-place.
ir.for_each_typed_id<SPIRType>([&](uint32_t self, SPIRType &type) {
if (naturally_emit_type(type))
{
emit_type(self, emitted_open_tag);
}
else if (type_is_reference(type))
{
if (!naturally_emit_type(this->get<SPIRType>(type.parent_type)) &&
find(physical_pointee_types.begin(), physical_pointee_types.end(), type.parent_type) ==
physical_pointee_types.end())
{
physical_pointee_types.push_back(type.parent_type);
}
}
});
for (uint32_t pointee_type : physical_pointee_types)
emit_type(pointee_type, emitted_open_tag);
if (emitted_open_tag)
{
json_stream->end_json_object();
}
}
void CompilerReflection::emit_type(uint32_t type_id, bool &emitted_open_tag)
{
auto &type = get<SPIRType>(type_id);
auto name = type_to_glsl(type);
if (!emitted_open_tag)
{
json_stream->emit_json_key_object("types");
emitted_open_tag = true;
}
json_stream->emit_json_key_object("_" + std::to_string(type_id));
json_stream->emit_json_key_value("name", name);
if (is_physical_pointer(type))
{
json_stream->emit_json_key_value("type", "_" + std::to_string(type.parent_type));
json_stream->emit_json_key_value("physical_pointer", true);
}
else if (!type.array.empty())
{
emit_type_array(type);
json_stream->emit_json_key_value("type", "_" + std::to_string(type.parent_type));
json_stream->emit_json_key_value("array_stride", get_decoration(type_id, DecorationArrayStride));
}
else
{
json_stream->emit_json_key_array("members");
// FIXME ideally we'd like to emit the size of a structure as a
// convenience to people parsing the reflected JSON. The problem
// is that there's no implicit size for a type. It's final size
// will be determined by the top level declaration in which it's
// included. So there might be one size for the struct if it's
// included in a std140 uniform block and another if it's included
// in a std430 uniform block.
// The solution is to include *all* potential sizes as a map of
// layout type name to integer, but that will probably require
// some additional logic being written in this class, or in the
// parent CompilerGLSL class.
auto size = type.member_types.size();
for (uint32_t i = 0; i < size; ++i)
{
emit_type_member(type, i);
}
json_stream->end_json_array();
}
json_stream->end_json_object();
}
void CompilerReflection::emit_type_member(const SPIRType &type, uint32_t index)
{
auto &membertype = get<SPIRType>(type.member_types[index]);
json_stream->begin_json_object();
auto name = to_member_name(type, index);
// FIXME we'd like to emit the offset of each member, but such offsets are
// context dependent. See the comment above regarding structure sizes
json_stream->emit_json_key_value("name", name);
if (type_is_reference(membertype))
{
json_stream->emit_json_key_value("type", "_" + std::to_string(membertype.parent_type));
}
else if (membertype.basetype == SPIRType::Struct)
{
json_stream->emit_json_key_value("type", "_" + std::to_string(membertype.self));
}
else
{
json_stream->emit_json_key_value("type", type_to_glsl(membertype));
}
emit_type_member_qualifiers(type, index);
json_stream->end_json_object();
}
void CompilerReflection::emit_type_array(const SPIRType &type)
{
if (!is_physical_pointer(type) && !type.array.empty())
{
json_stream->emit_json_key_array("array");
// Note that we emit the zeros here as a means of identifying
// unbounded arrays. This is necessary as otherwise there would
// be no way of differentiating between float[4] and float[4][]
for (const auto &value : type.array)
json_stream->emit_json_array_value(value);
json_stream->end_json_array();
json_stream->emit_json_key_array("array_size_is_literal");
for (const auto &value : type.array_size_literal)
json_stream->emit_json_array_value(value);
json_stream->end_json_array();
}
}
void CompilerReflection::emit_type_member_qualifiers(const SPIRType &type, uint32_t index)
{
auto &membertype = get<SPIRType>(type.member_types[index]);
emit_type_array(membertype);
auto &memb = ir.meta[type.self].members;
if (index < memb.size())
{
auto &dec = memb[index];
if (dec.decoration_flags.get(DecorationLocation))
json_stream->emit_json_key_value("location", dec.location);
if (dec.decoration_flags.get(DecorationOffset))
json_stream->emit_json_key_value("offset", dec.offset);
// Array stride is a property of the array type, not the struct.
if (has_decoration(type.member_types[index], DecorationArrayStride))
json_stream->emit_json_key_value("array_stride",
get_decoration(type.member_types[index], DecorationArrayStride));
if (dec.decoration_flags.get(DecorationMatrixStride))
json_stream->emit_json_key_value("matrix_stride", dec.matrix_stride);
if (dec.decoration_flags.get(DecorationRowMajor))
json_stream->emit_json_key_value("row_major", true);
if (is_physical_pointer(membertype))
json_stream->emit_json_key_value("physical_pointer", true);
}
}
string CompilerReflection::execution_model_to_str(spv::ExecutionModel model)
{
switch (model)
{
case ExecutionModelVertex:
return "vert";
case ExecutionModelTessellationControl:
return "tesc";
case ExecutionModelTessellationEvaluation:
return "tese";
case ExecutionModelGeometry:
return "geom";
case ExecutionModelFragment:
return "frag";
case ExecutionModelGLCompute:
return "comp";
case ExecutionModelRayGenerationNV:
return "rgen";
case ExecutionModelIntersectionNV:
return "rint";
case ExecutionModelAnyHitNV:
return "rahit";
case ExecutionModelClosestHitNV:
return "rchit";
case ExecutionModelMissNV:
return "rmiss";
case ExecutionModelCallableNV:
return "rcall";
default:
return "???";
}
}
// FIXME include things like the local_size dimensions, geometry output vertex count, etc
void CompilerReflection::emit_entry_points()
{
auto entries = get_entry_points_and_stages();
if (!entries.empty())
{
// Needed to make output deterministic.
sort(begin(entries), end(entries), [](const EntryPoint &a, const EntryPoint &b) -> bool {
if (a.execution_model < b.execution_model)
return true;
else if (a.execution_model > b.execution_model)
return false;
else
return a.name < b.name;
});
json_stream->emit_json_key_array("entryPoints");
for (auto &e : entries)
{
json_stream->begin_json_object();
json_stream->emit_json_key_value("name", e.name);
json_stream->emit_json_key_value("mode", execution_model_to_str(e.execution_model));
if (e.execution_model == ExecutionModelGLCompute)
{
const auto &spv_entry = get_entry_point(e.name, e.execution_model);
SpecializationConstant spec_x, spec_y, spec_z;
get_work_group_size_specialization_constants(spec_x, spec_y, spec_z);
json_stream->emit_json_key_array("workgroup_size");
json_stream->emit_json_array_value(spec_x.id != ID(0) ? spec_x.constant_id :
spv_entry.workgroup_size.x);
json_stream->emit_json_array_value(spec_y.id != ID(0) ? spec_y.constant_id :
spv_entry.workgroup_size.y);
json_stream->emit_json_array_value(spec_z.id != ID(0) ? spec_z.constant_id :
spv_entry.workgroup_size.z);
json_stream->end_json_array();
json_stream->emit_json_key_array("workgroup_size_is_spec_constant_id");
json_stream->emit_json_array_value(spec_x.id != ID(0));
json_stream->emit_json_array_value(spec_y.id != ID(0));
json_stream->emit_json_array_value(spec_z.id != ID(0));
json_stream->end_json_array();
}
json_stream->end_json_object();
}
json_stream->end_json_array();
}
}
void CompilerReflection::emit_resources()
{
auto res = get_shader_resources();
emit_resources("subpass_inputs", res.subpass_inputs);
emit_resources("inputs", res.stage_inputs);
emit_resources("outputs", res.stage_outputs);
emit_resources("textures", res.sampled_images);
emit_resources("separate_images", res.separate_images);
emit_resources("separate_samplers", res.separate_samplers);
emit_resources("images", res.storage_images);
emit_resources("ssbos", res.storage_buffers);
emit_resources("ubos", res.uniform_buffers);
emit_resources("push_constants", res.push_constant_buffers);
emit_resources("counters", res.atomic_counters);
emit_resources("acceleration_structures", res.acceleration_structures);
}
void CompilerReflection::emit_resources(const char *tag, const SmallVector<Resource> &resources)
{
if (resources.empty())
{
return;
}
json_stream->emit_json_key_array(tag);
for (auto &res : resources)
{
auto &type = get_type(res.type_id);
auto typeflags = ir.meta[type.self].decoration.decoration_flags;
auto &mask = get_decoration_bitset(res.id);
// If we don't have a name, use the fallback for the type instead of the variable
// for SSBOs and UBOs since those are the only meaningful names to use externally.
// Push constant blocks are still accessed by name and not block name, even though they are technically Blocks.
bool is_push_constant = get_storage_class(res.id) == StorageClassPushConstant;
bool is_block = get_decoration_bitset(type.self).get(DecorationBlock) ||
get_decoration_bitset(type.self).get(DecorationBufferBlock);
ID fallback_id = !is_push_constant && is_block ? ID(res.base_type_id) : ID(res.id);
json_stream->begin_json_object();
if (type.basetype == SPIRType::Struct)
{
json_stream->emit_json_key_value("type", "_" + std::to_string(res.base_type_id));
}
else
{
json_stream->emit_json_key_value("type", type_to_glsl(type));
}
json_stream->emit_json_key_value("name", !res.name.empty() ? res.name : get_fallback_name(fallback_id));
{
bool ssbo_block = type.storage == StorageClassStorageBuffer ||
(type.storage == StorageClassUniform && typeflags.get(DecorationBufferBlock));
Bitset qualifier_mask = ssbo_block ? get_buffer_block_flags(res.id) : mask;
if (qualifier_mask.get(DecorationNonReadable))
json_stream->emit_json_key_value("writeonly", true);
if (qualifier_mask.get(DecorationNonWritable))
json_stream->emit_json_key_value("readonly", true);
if (qualifier_mask.get(DecorationRestrict))
json_stream->emit_json_key_value("restrict", true);
if (qualifier_mask.get(DecorationCoherent))
json_stream->emit_json_key_value("coherent", true);
if (qualifier_mask.get(DecorationVolatile))
json_stream->emit_json_key_value("volatile", true);
}
emit_type_array(type);
{
bool is_sized_block = is_block && (get_storage_class(res.id) == StorageClassUniform ||
get_storage_class(res.id) == StorageClassUniformConstant ||
get_storage_class(res.id) == StorageClassStorageBuffer);
if (is_sized_block)
{
uint32_t block_size = uint32_t(get_declared_struct_size(get_type(res.base_type_id)));
json_stream->emit_json_key_value("block_size", block_size);
}
}
if (type.storage == StorageClassPushConstant)
json_stream->emit_json_key_value("push_constant", true);
if (mask.get(DecorationLocation))
json_stream->emit_json_key_value("location", get_decoration(res.id, DecorationLocation));
if (mask.get(DecorationRowMajor))
json_stream->emit_json_key_value("row_major", true);
if (mask.get(DecorationColMajor))
json_stream->emit_json_key_value("column_major", true);
if (mask.get(DecorationIndex))
json_stream->emit_json_key_value("index", get_decoration(res.id, DecorationIndex));
if (type.storage != StorageClassPushConstant && mask.get(DecorationDescriptorSet))
json_stream->emit_json_key_value("set", get_decoration(res.id, DecorationDescriptorSet));
if (mask.get(DecorationBinding))
json_stream->emit_json_key_value("binding", get_decoration(res.id, DecorationBinding));
if (mask.get(DecorationInputAttachmentIndex))
json_stream->emit_json_key_value("input_attachment_index",
get_decoration(res.id, DecorationInputAttachmentIndex));
if (mask.get(DecorationOffset))
json_stream->emit_json_key_value("offset", get_decoration(res.id, DecorationOffset));
if (mask.get(DecorationWeightTextureQCOM))
json_stream->emit_json_key_value("WeightTextureQCOM", get_decoration(res.id, DecorationWeightTextureQCOM));
if (mask.get(DecorationBlockMatchTextureQCOM))
json_stream->emit_json_key_value("BlockMatchTextureQCOM", get_decoration(res.id, DecorationBlockMatchTextureQCOM));
// For images, the type itself adds a layout qualifer.
// Only emit the format for storage images.
if (type.basetype == SPIRType::Image && type.image.sampled == 2)
{
const char *fmt = format_to_glsl(type.image.format);
if (fmt != nullptr)
json_stream->emit_json_key_value("format", std::string(fmt));
}
json_stream->end_json_object();
}
json_stream->end_json_array();
}
void CompilerReflection::emit_specialization_constants()
{
auto specialization_constants = get_specialization_constants();
if (specialization_constants.empty())
return;
json_stream->emit_json_key_array("specialization_constants");
for (const auto &spec_const : specialization_constants)
{
auto &c = get<SPIRConstant>(spec_const.id);
auto type = get<SPIRType>(c.constant_type);
json_stream->begin_json_object();
json_stream->emit_json_key_value("name", get_name(spec_const.id));
json_stream->emit_json_key_value("id", spec_const.constant_id);
json_stream->emit_json_key_value("type", type_to_glsl(type));
json_stream->emit_json_key_value("variable_id", spec_const.id);
switch (type.basetype)
{
case SPIRType::UInt:
json_stream->emit_json_key_value("default_value", c.scalar());
break;
case SPIRType::Int:
json_stream->emit_json_key_value("default_value", c.scalar_i32());
break;
case SPIRType::Float:
json_stream->emit_json_key_value("default_value", c.scalar_f32());
break;
case SPIRType::Boolean:
json_stream->emit_json_key_value("default_value", c.scalar() != 0);
break;
default:
break;
}
json_stream->end_json_object();
}
json_stream->end_json_array();
}
string CompilerReflection::to_member_name(const SPIRType &type, uint32_t index) const
{
auto *type_meta = ir.find_meta(type.self);
if (type_meta)
{
auto &memb = type_meta->members;
if (index < memb.size() && !memb[index].alias.empty())
return memb[index].alias;
else
return join("_m", index);
}
else
return join("_m", index);
}

91
thirdparty/spirv-cross/spirv_reflect.hpp vendored Normal file
View File

@ -0,0 +1,91 @@
/*
* Copyright 2018-2021 Bradley Austin Davis
* SPDX-License-Identifier: Apache-2.0 OR MIT
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
/*
* At your option, you may choose to accept this material under either:
* 1. The Apache License, Version 2.0, found at <http://www.apache.org/licenses/LICENSE-2.0>, or
* 2. The MIT License, found at <http://opensource.org/licenses/MIT>.
*/
#ifndef SPIRV_CROSS_REFLECT_HPP
#define SPIRV_CROSS_REFLECT_HPP
#include "spirv_glsl.hpp"
#include <utility>
namespace simple_json
{
class Stream;
}
namespace SPIRV_CROSS_NAMESPACE
{
class CompilerReflection : public CompilerGLSL
{
using Parent = CompilerGLSL;
public:
explicit CompilerReflection(std::vector<uint32_t> spirv_)
: Parent(std::move(spirv_))
{
options.vulkan_semantics = true;
}
CompilerReflection(const uint32_t *ir_, size_t word_count)
: Parent(ir_, word_count)
{
options.vulkan_semantics = true;
}
explicit CompilerReflection(const ParsedIR &ir_)
: CompilerGLSL(ir_)
{
options.vulkan_semantics = true;
}
explicit CompilerReflection(ParsedIR &&ir_)
: CompilerGLSL(std::move(ir_))
{
options.vulkan_semantics = true;
}
void set_format(const std::string &format);
std::string compile() override;
private:
static std::string execution_model_to_str(spv::ExecutionModel model);
void emit_entry_points();
void emit_types();
void emit_resources();
void emit_specialization_constants();
void emit_type(uint32_t type_id, bool &emitted_open_tag);
void emit_type_member(const SPIRType &type, uint32_t index);
void emit_type_member_qualifiers(const SPIRType &type, uint32_t index);
void emit_type_array(const SPIRType &type);
void emit_resources(const char *tag, const SmallVector<Resource> &resources);
bool type_is_reference(const SPIRType &type) const;
std::string to_member_name(const SPIRType &type, uint32_t index) const;
std::shared_ptr<simple_json::Stream> json_stream;
};
} // namespace SPIRV_CROSS_NAMESPACE
#endif