• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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   LocalMemoryObjects(),
19   ExplicitKernArgSize(0),
20   LDSSize(0),
21   Mode(MF.getFunction(), MF.getSubtarget<GCNSubtarget>()),
22   IsEntryFunction(AMDGPU::isEntryFunctionCC(MF.getFunction().getCallingConv())),
23   NoSignedZerosFPMath(MF.getTarget().Options.NoSignedZerosFPMath),
24   MemoryBound(false),
25   WaveLimiter(false) {
26   const AMDGPUSubtarget &ST = AMDGPUSubtarget::get(MF);
27 
28   // FIXME: Should initialize KernArgSize based on ExplicitKernelArgOffset,
29   // except reserved size is not correctly aligned.
30   const Function &F = MF.getFunction();
31 
32   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 
allocateLDSGlobal(const DataLayout & DL,const GlobalValue & GV)45 unsigned 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