LLVM  7.0.0svn
AMDGPUMachineFunction.h
Go to the documentation of this file.
1 //===-- AMDGPUMachineFunctionInfo.h -------------------------------*- C++ -*-=//
2 //
3 // The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 
10 #ifndef LLVM_LIB_TARGET_AMDGPU_AMDGPUMACHINEFUNCTION_H
11 #define LLVM_LIB_TARGET_AMDGPU_AMDGPUMACHINEFUNCTION_H
12 
13 #include "llvm/ADT/DenseMap.h"
15 
16 namespace llvm {
17 
19  /// A map to keep track of local memory objects and their offsets within the
20  /// local memory space.
22 
23 protected:
24  uint64_t KernArgSize;
25  unsigned MaxKernArgAlign;
26 
27  /// Number of bytes in the LDS that are being used.
28  unsigned LDSSize;
29 
30  // FIXME: This should probably be removed.
31  /// Start of implicit kernel args
32  unsigned ABIArgOffset;
33 
34  // Kernels + shaders. i.e. functions called by the driver and not called
35  // by other functions.
37 
39 
40  // Function may be memory bound.
42 
43  // Kernel may need limited waves per EU for better performance.
45 
46 public:
48 
49  uint64_t allocateKernArg(uint64_t Size, unsigned Align) {
50  assert(isPowerOf2_32(Align));
51  KernArgSize = alignTo(KernArgSize, Align);
52 
53  uint64_t Result = KernArgSize;
54  KernArgSize += Size;
55 
56  MaxKernArgAlign = std::max(Align, MaxKernArgAlign);
57  return Result;
58  }
59 
60  uint64_t getKernArgSize() const {
61  return KernArgSize;
62  }
63 
64  unsigned getMaxKernArgAlign() const {
65  return MaxKernArgAlign;
66  }
67 
68  void setABIArgOffset(unsigned NewOffset) {
69  ABIArgOffset = NewOffset;
70  }
71 
72  unsigned getABIArgOffset() const {
73  return ABIArgOffset;
74  }
75 
76  unsigned getLDSSize() const {
77  return LDSSize;
78  }
79 
80  bool isEntryFunction() const {
81  return IsEntryFunction;
82  }
83 
84  bool hasNoSignedZerosFPMath() const {
85  return NoSignedZerosFPMath;
86  }
87 
88  bool isMemoryBound() const {
89  return MemoryBound;
90  }
91 
92  bool needsWaveLimiter() const {
93  return WaveLimiter;
94  }
95 
96  unsigned allocateLDSGlobal(const DataLayout &DL, const GlobalValue &GV);
97 };
98 
99 }
100 #endif
A parsed version of the target data layout string in and methods for querying it. ...
Definition: DataLayout.h:111
constexpr char Align[]
Key for Kernel::Arg::Metadata::mAlign.
unsigned allocateLDSGlobal(const DataLayout &DL, const GlobalValue &GV)
GCNRegPressure max(const GCNRegPressure &P1, const GCNRegPressure &P2)
Compute iterated dominance frontiers using a linear time algorithm.
Definition: AllocatorList.h:24
AMDGPUMachineFunction(const MachineFunction &MF)
uint64_t alignTo(uint64_t Value, uint64_t Align, uint64_t Skew=0)
Returns the next integer (mod 2**64) that is greater than or equal to Value and is a multiple of Alig...
Definition: MathExtras.h:685
MachineFunctionInfo - This class can be derived from and used by targets to hold private target-speci...
uint64_t allocateKernArg(uint64_t Size, unsigned Align)
constexpr bool isPowerOf2_32(uint32_t Value)
Return true if the argument is a power of two > 0.
Definition: MathExtras.h:429
unsigned LDSSize
Number of bytes in the LDS that are being used.
unsigned ABIArgOffset
Start of implicit kernel args.
void setABIArgOffset(unsigned NewOffset)
assert(ImpDefSCC.getReg()==AMDGPU::SCC &&ImpDefSCC.isDef())
constexpr char Size[]
Key for Kernel::Arg::Metadata::mSize.