seedlingattempt/godot
1
0
Fork 0
Code Issues Pull Requests Actions 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
72 changed files with 67923 additions and 28 deletions
Download Patch File Download Diff File Expand all files Collapse all files

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