AMDGPUMachineFunction.cpp revision 360660 
1263348Sjmmv//===-- AMDGPUMachineFunctionInfo.cpp ---------------------------------------=//
2263348Sjmmv//
3263348Sjmmv// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4263348Sjmmv// See https://llvm.org/LICENSE.txt for license information.
5263348Sjmmv// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6263348Sjmmv//
7263348Sjmmv//===----------------------------------------------------------------------===//
8263348Sjmmv
9263348Sjmmv#include "AMDGPUMachineFunction.h"
10263348Sjmmv#include "AMDGPUSubtarget.h"
11263348Sjmmv#include "AMDGPUPerfHintAnalysis.h"
12263348Sjmmv#include "llvm/CodeGen/MachineModuleInfo.h"
13263348Sjmmv
14263348Sjmmvusing namespace llvm;
15263348Sjmmv
16263348SjmmvAMDGPUMachineFunction::AMDGPUMachineFunction(const MachineFunction &MF) :
17263348Sjmmv  MachineFunctionInfo(),
18263348Sjmmv  LocalMemoryObjects(),
19263348Sjmmv  ExplicitKernArgSize(0),
20263348Sjmmv  MaxKernArgAlign(0),
21263348Sjmmv  LDSSize(0),
22263348Sjmmv  IsEntryFunction(AMDGPU::isEntryFunctionCC(MF.getFunction().getCallingConv())),
23263348Sjmmv  NoSignedZerosFPMath(MF.getTarget().Options.NoSignedZerosFPMath),
24263348Sjmmv  MemoryBound(false),
25263348Sjmmv  WaveLimiter(false) {
26263348Sjmmv  const AMDGPUSubtarget &ST = AMDGPUSubtarget::get(MF);
27263348Sjmmv
28263348Sjmmv  // FIXME: Should initialize KernArgSize based on ExplicitKernelArgOffset,
29263348Sjmmv  // except reserved size is not correctly aligned.
30263348Sjmmv  const Function &F = MF.getFunction();
31263348Sjmmv
32263348Sjmmv  Attribute MemBoundAttr = F.getFnAttribute("amdgpu-memory-bound");
33  MemoryBound = MemBoundAttr.isStringAttribute() &&
34                MemBoundAttr.getValueAsString() == "true";
35
36  Attribute WaveLimitAttr = F.getFnAttribute("amdgpu-wave-limiter");
37  WaveLimiter = WaveLimitAttr.isStringAttribute() &&
38                WaveLimitAttr.getValueAsString() == "true";
39
40  CallingConv::ID CC = F.getCallingConv();
41  if (CC == CallingConv::AMDGPU_KERNEL || CC == CallingConv::SPIR_KERNEL)
42    ExplicitKernArgSize = ST.getExplicitKernArgSize(F, MaxKernArgAlign);
43}
44
45unsigned AMDGPUMachineFunction::allocateLDSGlobal(const DataLayout &DL,
46                                                  const GlobalValue &GV) {
47  auto Entry = LocalMemoryObjects.insert(std::make_pair(&GV, 0));
48  if (!Entry.second)
49    return Entry.first->second;
50
51  unsigned Align = GV.getAlignment();
52  if (Align == 0)
53    Align = DL.getABITypeAlignment(GV.getValueType());
54
55  /// TODO: We should sort these to minimize wasted space due to alignment
56  /// padding. Currently the padding is decided by the first encountered use
57  /// during lowering.
58  unsigned Offset = LDSSize = alignTo(LDSSize, Align);
59
60  Entry.first->second = Offset;
61  LDSSize += DL.getTypeAllocSize(GV.getValueType());
62
63  return Offset;
64}
65