10b57cec5SDimitry Andric //===-- AMDGPUMachineFunctionInfo.cpp ---------------------------------------=// 20b57cec5SDimitry Andric // 30b57cec5SDimitry Andric // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 40b57cec5SDimitry Andric // See https://llvm.org/LICENSE.txt for license information. 50b57cec5SDimitry Andric // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 60b57cec5SDimitry Andric // 70b57cec5SDimitry Andric //===----------------------------------------------------------------------===// 80b57cec5SDimitry Andric 90b57cec5SDimitry Andric #include "AMDGPUMachineFunction.h" 100b57cec5SDimitry Andric #include "AMDGPUPerfHintAnalysis.h" 11*e8d8bef9SDimitry Andric #include "AMDGPUSubtarget.h" 120b57cec5SDimitry Andric #include "llvm/CodeGen/MachineModuleInfo.h" 13*e8d8bef9SDimitry Andric #include "llvm/Target/TargetMachine.h" 140b57cec5SDimitry Andric 150b57cec5SDimitry Andric using namespace llvm; 160b57cec5SDimitry Andric 17*e8d8bef9SDimitry Andric AMDGPUMachineFunction::AMDGPUMachineFunction(const MachineFunction &MF) 18*e8d8bef9SDimitry Andric : MachineFunctionInfo(), Mode(MF.getFunction()), 19*e8d8bef9SDimitry Andric IsEntryFunction( 20*e8d8bef9SDimitry Andric AMDGPU::isEntryFunctionCC(MF.getFunction().getCallingConv())), 21*e8d8bef9SDimitry Andric IsModuleEntryFunction( 22*e8d8bef9SDimitry Andric AMDGPU::isModuleEntryFunctionCC(MF.getFunction().getCallingConv())), 235ffd83dbSDimitry Andric NoSignedZerosFPMath(MF.getTarget().Options.NoSignedZerosFPMath) { 240b57cec5SDimitry Andric const AMDGPUSubtarget &ST = AMDGPUSubtarget::get(MF); 250b57cec5SDimitry Andric 260b57cec5SDimitry Andric // FIXME: Should initialize KernArgSize based on ExplicitKernelArgOffset, 270b57cec5SDimitry Andric // except reserved size is not correctly aligned. 280b57cec5SDimitry Andric const Function &F = MF.getFunction(); 290b57cec5SDimitry Andric 300b57cec5SDimitry Andric Attribute MemBoundAttr = F.getFnAttribute("amdgpu-memory-bound"); 310b57cec5SDimitry Andric MemoryBound = MemBoundAttr.isStringAttribute() && 320b57cec5SDimitry Andric MemBoundAttr.getValueAsString() == "true"; 330b57cec5SDimitry Andric 340b57cec5SDimitry Andric Attribute WaveLimitAttr = F.getFnAttribute("amdgpu-wave-limiter"); 350b57cec5SDimitry Andric WaveLimiter = WaveLimitAttr.isStringAttribute() && 360b57cec5SDimitry Andric WaveLimitAttr.getValueAsString() == "true"; 370b57cec5SDimitry Andric 380b57cec5SDimitry Andric CallingConv::ID CC = F.getCallingConv(); 390b57cec5SDimitry Andric if (CC == CallingConv::AMDGPU_KERNEL || CC == CallingConv::SPIR_KERNEL) 400b57cec5SDimitry Andric ExplicitKernArgSize = ST.getExplicitKernArgSize(F, MaxKernArgAlign); 410b57cec5SDimitry Andric } 420b57cec5SDimitry Andric 430b57cec5SDimitry Andric unsigned AMDGPUMachineFunction::allocateLDSGlobal(const DataLayout &DL, 445ffd83dbSDimitry Andric const GlobalVariable &GV) { 450b57cec5SDimitry Andric auto Entry = LocalMemoryObjects.insert(std::make_pair(&GV, 0)); 460b57cec5SDimitry Andric if (!Entry.second) 470b57cec5SDimitry Andric return Entry.first->second; 480b57cec5SDimitry Andric 495ffd83dbSDimitry Andric Align Alignment = 505ffd83dbSDimitry Andric DL.getValueOrABITypeAlignment(GV.getAlign(), GV.getValueType()); 510b57cec5SDimitry Andric 520b57cec5SDimitry Andric /// TODO: We should sort these to minimize wasted space due to alignment 530b57cec5SDimitry Andric /// padding. Currently the padding is decided by the first encountered use 540b57cec5SDimitry Andric /// during lowering. 55*e8d8bef9SDimitry Andric unsigned Offset = StaticLDSSize = alignTo(StaticLDSSize, Alignment); 560b57cec5SDimitry Andric 570b57cec5SDimitry Andric Entry.first->second = Offset; 58*e8d8bef9SDimitry Andric StaticLDSSize += DL.getTypeAllocSize(GV.getValueType()); 59*e8d8bef9SDimitry Andric 60*e8d8bef9SDimitry Andric // Update the LDS size considering the padding to align the dynamic shared 61*e8d8bef9SDimitry Andric // memory. 62*e8d8bef9SDimitry Andric LDSSize = alignTo(StaticLDSSize, DynLDSAlign); 630b57cec5SDimitry Andric 640b57cec5SDimitry Andric return Offset; 650b57cec5SDimitry Andric } 66*e8d8bef9SDimitry Andric 67*e8d8bef9SDimitry Andric void AMDGPUMachineFunction::setDynLDSAlign(const DataLayout &DL, 68*e8d8bef9SDimitry Andric const GlobalVariable &GV) { 69*e8d8bef9SDimitry Andric assert(DL.getTypeAllocSize(GV.getValueType()).isZero()); 70*e8d8bef9SDimitry Andric 71*e8d8bef9SDimitry Andric Align Alignment = 72*e8d8bef9SDimitry Andric DL.getValueOrABITypeAlignment(GV.getAlign(), GV.getValueType()); 73*e8d8bef9SDimitry Andric if (Alignment <= DynLDSAlign) 74*e8d8bef9SDimitry Andric return; 75*e8d8bef9SDimitry Andric 76*e8d8bef9SDimitry Andric LDSSize = alignTo(StaticLDSSize, Alignment); 77*e8d8bef9SDimitry Andric DynLDSAlign = Alignment; 78*e8d8bef9SDimitry Andric } 79