xref: /freebsd-src/contrib/llvm-project/llvm/lib/Target/AMDGPU/AMDGPUMachineFunction.h (revision 753f127f3ace09432b2baeffd71a308760641a62)
1 //===-- AMDGPUMachineFunctionInfo.h -------------------------------*- C++ -*-=//
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 #ifndef LLVM_LIB_TARGET_AMDGPU_AMDGPUMACHINEFUNCTION_H
10 #define LLVM_LIB_TARGET_AMDGPU_AMDGPUMACHINEFUNCTION_H
11 
12 #include "Utils/AMDGPUBaseInfo.h"
13 #include "llvm/ADT/DenseMap.h"
14 #include "llvm/CodeGen/MachineFunction.h"
15 #include "llvm/IR/DataLayout.h"
16 #include "llvm/IR/GlobalValue.h"
17 #include "llvm/IR/GlobalVariable.h"
18 #include "llvm/IR/Function.h"
19 
20 namespace llvm {
21 
22 class AMDGPUMachineFunction : public MachineFunctionInfo {
23   /// A map to keep track of local memory objects and their offsets within the
24   /// local memory space.
25   SmallDenseMap<const GlobalValue *, unsigned, 4> LocalMemoryObjects;
26 
27 protected:
28   uint64_t ExplicitKernArgSize = 0; // Cache for this.
29   Align MaxKernArgAlign;        // Cache for this.
30 
31   /// Number of bytes in the LDS that are being used.
32   uint32_t LDSSize = 0;
33   uint32_t GDSSize = 0;
34 
35   /// Number of bytes in the LDS allocated statically. This field is only used
36   /// in the instruction selector and not part of the machine function info.
37   uint32_t StaticLDSSize = 0;
38   uint32_t StaticGDSSize = 0;
39 
40   /// Align for dynamic shared memory if any. Dynamic shared memory is
41   /// allocated directly after the static one, i.e., LDSSize. Need to pad
42   /// LDSSize to ensure that dynamic one is aligned accordingly.
43   /// The maximal alignment is updated during IR translation or lowering
44   /// stages.
45   Align DynLDSAlign;
46 
47   // State of MODE register, assumed FP mode.
48   AMDGPU::SIModeRegisterDefaults Mode;
49 
50   // Kernels + shaders. i.e. functions called by the hardware and not called
51   // by other functions.
52   bool IsEntryFunction = false;
53 
54   // Entry points called by other functions instead of directly by the hardware.
55   bool IsModuleEntryFunction = false;
56 
57   bool NoSignedZerosFPMath = false;
58 
59   // Function may be memory bound.
60   bool MemoryBound = false;
61 
62   // Kernel may need limited waves per EU for better performance.
63   bool WaveLimiter = false;
64 
65 public:
66   AMDGPUMachineFunction(const MachineFunction &MF);
67 
68   uint64_t getExplicitKernArgSize() const {
69     return ExplicitKernArgSize;
70   }
71 
72   Align getMaxKernArgAlign() const { return MaxKernArgAlign; }
73 
74   uint32_t getLDSSize() const {
75     return LDSSize;
76   }
77 
78   uint32_t getGDSSize() const {
79     return GDSSize;
80   }
81 
82   AMDGPU::SIModeRegisterDefaults getMode() const {
83     return Mode;
84   }
85 
86   bool isEntryFunction() const {
87     return IsEntryFunction;
88   }
89 
90   bool isModuleEntryFunction() const { return IsModuleEntryFunction; }
91 
92   bool hasNoSignedZerosFPMath() const {
93     return NoSignedZerosFPMath;
94   }
95 
96   bool isMemoryBound() const {
97     return MemoryBound;
98   }
99 
100   bool needsWaveLimiter() const {
101     return WaveLimiter;
102   }
103 
104   unsigned allocateLDSGlobal(const DataLayout &DL, const GlobalVariable &GV);
105   void allocateModuleLDSGlobal(const Function &F);
106 
107   Align getDynLDSAlign() const { return DynLDSAlign; }
108 
109   void setDynLDSAlign(const DataLayout &DL, const GlobalVariable &GV);
110 };
111 
112 }
113 #endif
114