spb/forge Public MIT
Forge — LLM training from scratch in pure C++20 + Metal on Apple Silicon.
C++ 61.2%
C 23%
Python 7.6%
TeX 7.2%
CMake 1.1%
1//-------------------------------------------------------------------------------------------------------------------------------------------------------------2//3// Metal/MTL4ComputePipeline.hpp4//5// Copyright 2020-2025 Apple Inc.6//7// Licensed under the Apache License, Version 2.0 (the "License");8// you may not use this file except in compliance with the License.9// You may obtain a copy of the License at10//11// http://www.apache.org/licenses/LICENSE-2.012//13// Unless required by applicable law or agreed to in writing, software14// distributed under the License is distributed on an "AS IS" BASIS,15// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.16// See the License for the specific language governing permissions and17// limitations under the License.18//19//-------------------------------------------------------------------------------------------------------------------------------------------------------------2021#pragma once2223#include "../Foundation/Foundation.hpp"24#include "MTL4PipelineState.hpp"25#include "MTLDefines.hpp"26#include "MTLHeaderBridge.hpp"27#include "MTLPrivate.hpp"28#include "MTLTypes.hpp"2930namespace MTL431{32class ComputePipelineDescriptor;33class FunctionDescriptor;34class StaticLinkingDescriptor;3536class ComputePipelineDescriptor : public NS::Copying<ComputePipelineDescriptor, PipelineDescriptor>37{38public:39 static ComputePipelineDescriptor* alloc();4041 FunctionDescriptor* computeFunctionDescriptor() const;4243 ComputePipelineDescriptor* init();4445 NS::UInteger maxTotalThreadsPerThreadgroup() const;4647 MTL::Size requiredThreadsPerThreadgroup() const;4849 void reset();5051 void setComputeFunctionDescriptor(const MTL4::FunctionDescriptor* computeFunctionDescriptor);5253 void setMaxTotalThreadsPerThreadgroup(NS::UInteger maxTotalThreadsPerThreadgroup);5455 void setRequiredThreadsPerThreadgroup(MTL::Size requiredThreadsPerThreadgroup);5657 void setStaticLinkingDescriptor(const MTL4::StaticLinkingDescriptor* staticLinkingDescriptor);5859 void setSupportBinaryLinking(bool supportBinaryLinking);6061 void setSupportIndirectCommandBuffers(MTL4::IndirectCommandBufferSupportState supportIndirectCommandBuffers);6263 void setThreadGroupSizeIsMultipleOfThreadExecutionWidth(bool threadGroupSizeIsMultipleOfThreadExecutionWidth);6465 StaticLinkingDescriptor* staticLinkingDescriptor() const;6667 bool supportBinaryLinking() const;6869 IndirectCommandBufferSupportState supportIndirectCommandBuffers() const;7071 bool threadGroupSizeIsMultipleOfThreadExecutionWidth() const;72};7374}75_MTL_INLINE MTL4::ComputePipelineDescriptor* MTL4::ComputePipelineDescriptor::alloc()76{77 return NS::Object::alloc<MTL4::ComputePipelineDescriptor>(_MTL_PRIVATE_CLS(MTL4ComputePipelineDescriptor));78}7980_MTL_INLINE MTL4::FunctionDescriptor* MTL4::ComputePipelineDescriptor::computeFunctionDescriptor() const81{82 return Object::sendMessage<MTL4::FunctionDescriptor*>(this, _MTL_PRIVATE_SEL(computeFunctionDescriptor));83}8485_MTL_INLINE MTL4::ComputePipelineDescriptor* MTL4::ComputePipelineDescriptor::init()86{87 return NS::Object::init<MTL4::ComputePipelineDescriptor>();88}8990_MTL_INLINE NS::UInteger MTL4::ComputePipelineDescriptor::maxTotalThreadsPerThreadgroup() const91{92 return Object::sendMessage<NS::UInteger>(this, _MTL_PRIVATE_SEL(maxTotalThreadsPerThreadgroup));93}9495_MTL_INLINE MTL::Size MTL4::ComputePipelineDescriptor::requiredThreadsPerThreadgroup() const96{97 return Object::sendMessage<MTL::Size>(this, _MTL_PRIVATE_SEL(requiredThreadsPerThreadgroup));98}99100_MTL_INLINE void MTL4::ComputePipelineDescriptor::reset()101{102 Object::sendMessage<void>(this, _MTL_PRIVATE_SEL(reset));103}104105_MTL_INLINE void MTL4::ComputePipelineDescriptor::setComputeFunctionDescriptor(const MTL4::FunctionDescriptor* computeFunctionDescriptor)106{107 Object::sendMessage<void>(this, _MTL_PRIVATE_SEL(setComputeFunctionDescriptor_), computeFunctionDescriptor);108}109110_MTL_INLINE void MTL4::ComputePipelineDescriptor::setMaxTotalThreadsPerThreadgroup(NS::UInteger maxTotalThreadsPerThreadgroup)111{112 Object::sendMessage<void>(this, _MTL_PRIVATE_SEL(setMaxTotalThreadsPerThreadgroup_), maxTotalThreadsPerThreadgroup);113}114115_MTL_INLINE void MTL4::ComputePipelineDescriptor::setRequiredThreadsPerThreadgroup(MTL::Size requiredThreadsPerThreadgroup)116{117 Object::sendMessage<void>(this, _MTL_PRIVATE_SEL(setRequiredThreadsPerThreadgroup_), requiredThreadsPerThreadgroup);118}119120_MTL_INLINE void MTL4::ComputePipelineDescriptor::setStaticLinkingDescriptor(const MTL4::StaticLinkingDescriptor* staticLinkingDescriptor)121{122 Object::sendMessage<void>(this, _MTL_PRIVATE_SEL(setStaticLinkingDescriptor_), staticLinkingDescriptor);123}124125_MTL_INLINE void MTL4::ComputePipelineDescriptor::setSupportBinaryLinking(bool supportBinaryLinking)126{127 Object::sendMessage<void>(this, _MTL_PRIVATE_SEL(setSupportBinaryLinking_), supportBinaryLinking);128}129130_MTL_INLINE void MTL4::ComputePipelineDescriptor::setSupportIndirectCommandBuffers(MTL4::IndirectCommandBufferSupportState supportIndirectCommandBuffers)131{132 Object::sendMessage<void>(this, _MTL_PRIVATE_SEL(setSupportIndirectCommandBuffers_), supportIndirectCommandBuffers);133}134135_MTL_INLINE void MTL4::ComputePipelineDescriptor::setThreadGroupSizeIsMultipleOfThreadExecutionWidth(bool threadGroupSizeIsMultipleOfThreadExecutionWidth)136{137 Object::sendMessage<void>(this, _MTL_PRIVATE_SEL(setThreadGroupSizeIsMultipleOfThreadExecutionWidth_), threadGroupSizeIsMultipleOfThreadExecutionWidth);138}139140_MTL_INLINE MTL4::StaticLinkingDescriptor* MTL4::ComputePipelineDescriptor::staticLinkingDescriptor() const141{142 return Object::sendMessage<MTL4::StaticLinkingDescriptor*>(this, _MTL_PRIVATE_SEL(staticLinkingDescriptor));143}144145_MTL_INLINE bool MTL4::ComputePipelineDescriptor::supportBinaryLinking() const146{147 return Object::sendMessageSafe<bool>(this, _MTL_PRIVATE_SEL(supportBinaryLinking));148}149150_MTL_INLINE MTL4::IndirectCommandBufferSupportState MTL4::ComputePipelineDescriptor::supportIndirectCommandBuffers() const151{152 return Object::sendMessage<MTL4::IndirectCommandBufferSupportState>(this, _MTL_PRIVATE_SEL(supportIndirectCommandBuffers));153}154155_MTL_INLINE bool MTL4::ComputePipelineDescriptor::threadGroupSizeIsMultipleOfThreadExecutionWidth() const156{157 return Object::sendMessage<bool>(this, _MTL_PRIVATE_SEL(threadGroupSizeIsMultipleOfThreadExecutionWidth));158}159