1 //===-- AMDGPUMachineFunctionInfo.cpp ---------------------------------------=//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8
9 #include "AMDGPUMachineFunction.h"
10 #include "AMDGPUSubtarget.h"
11 #include "AMDGPUPerfHintAnalysis.h"
12 #include "llvm/CodeGen/MachineModuleInfo.h"
13
14 using namespace llvm;
15
AMDGPUMachineFunction(const MachineFunction & MF)16 AMDGPUMachineFunction::AMDGPUMachineFunction(const MachineFunction &MF) :
17 MachineFunctionInfo(),
18 Mode(MF.getFunction()),
19 IsEntryFunction(AMDGPU::isEntryFunctionCC(MF.getFunction().getCallingConv())),
20 NoSignedZerosFPMath(MF.getTarget().Options.NoSignedZerosFPMath) {
21 const AMDGPUSubtarget &ST = AMDGPUSubtarget::get(MF);
22
23 // FIXME: Should initialize KernArgSize based on ExplicitKernelArgOffset,
24 // except reserved size is not correctly aligned.
25 const Function &F = MF.getFunction();
26
27 Attribute MemBoundAttr = F.getFnAttribute("amdgpu-memory-bound");
28 MemoryBound = MemBoundAttr.isStringAttribute() &&
29 MemBoundAttr.getValueAsString() == "true";
30
31 Attribute WaveLimitAttr = F.getFnAttribute("amdgpu-wave-limiter");
32 WaveLimiter = WaveLimitAttr.isStringAttribute() &&
33 WaveLimitAttr.getValueAsString() == "true";
34
35 CallingConv::ID CC = F.getCallingConv();
36 if (CC == CallingConv::AMDGPU_KERNEL || CC == CallingConv::SPIR_KERNEL)
37 ExplicitKernArgSize = ST.getExplicitKernArgSize(F, MaxKernArgAlign);
38 }
39
allocateLDSGlobal(const DataLayout & DL,const GlobalVariable & GV)40 unsigned AMDGPUMachineFunction::allocateLDSGlobal(const DataLayout &DL,
41 const GlobalVariable &GV) {
42 auto Entry = LocalMemoryObjects.insert(std::make_pair(&GV, 0));
43 if (!Entry.second)
44 return Entry.first->second;
45
46 Align Alignment =
47 DL.getValueOrABITypeAlignment(GV.getAlign(), GV.getValueType());
48
49 /// TODO: We should sort these to minimize wasted space due to alignment
50 /// padding. Currently the padding is decided by the first encountered use
51 /// during lowering.
52 unsigned Offset = StaticLDSSize = alignTo(StaticLDSSize, Alignment);
53
54 Entry.first->second = Offset;
55 StaticLDSSize += DL.getTypeAllocSize(GV.getValueType());
56
57 // Update the LDS size considering the padding to align the dynamic shared
58 // memory.
59 LDSSize = alignTo(StaticLDSSize, DynLDSAlign);
60
61 return Offset;
62 }
63
setDynLDSAlign(const DataLayout & DL,const GlobalVariable & GV)64 void AMDGPUMachineFunction::setDynLDSAlign(const DataLayout &DL,
65 const GlobalVariable &GV) {
66 assert(DL.getTypeAllocSize(GV.getValueType()).isZero());
67
68 Align Alignment =
69 DL.getValueOrABITypeAlignment(GV.getAlign(), GV.getValueType());
70 if (Alignment <= DynLDSAlign)
71 return;
72
73 LDSSize = alignTo(StaticLDSSize, Alignment);
74 DynLDSAlign = Alignment;
75 }
76