1 //===- NVVMIntrRange.cpp - Set range attributes for NVVM intrinsics -------===//
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 // This pass adds appropriate range attributes for calls to NVVM
10 // intrinsics that return a limited range of values.
11 //
12 //===----------------------------------------------------------------------===//
13
14 #include "NVPTX.h"
15 #include "NVPTXUtilities.h"
16 #include "llvm/IR/InstIterator.h"
17 #include "llvm/IR/Instructions.h"
18 #include "llvm/IR/IntrinsicInst.h"
19 #include "llvm/IR/Intrinsics.h"
20 #include "llvm/IR/IntrinsicsNVPTX.h"
21 #include "llvm/IR/PassManager.h"
22 #include "llvm/Support/CommandLine.h"
23 #include <cstdint>
24
25 using namespace llvm;
26
27 #define DEBUG_TYPE "nvvm-intr-range"
28
29 namespace llvm { void initializeNVVMIntrRangePass(PassRegistry &); }
30
31 namespace {
32 class NVVMIntrRange : public FunctionPass {
33 public:
34 static char ID;
NVVMIntrRange()35 NVVMIntrRange() : FunctionPass(ID) {
36
37 initializeNVVMIntrRangePass(*PassRegistry::getPassRegistry());
38 }
39
40 bool runOnFunction(Function &) override;
41 };
42 } // namespace
43
createNVVMIntrRangePass()44 FunctionPass *llvm::createNVVMIntrRangePass() { return new NVVMIntrRange(); }
45
46 char NVVMIntrRange::ID = 0;
47 INITIALIZE_PASS(NVVMIntrRange, "nvvm-intr-range",
48 "Add !range metadata to NVVM intrinsics.", false, false)
49
50 // Adds the passed-in [Low,High) range information as metadata to the
51 // passed-in call instruction.
addRangeAttr(uint64_t Low,uint64_t High,IntrinsicInst * II)52 static bool addRangeAttr(uint64_t Low, uint64_t High, IntrinsicInst *II) {
53 if (II->getMetadata(LLVMContext::MD_range))
54 return false;
55
56 const uint64_t BitWidth = II->getType()->getIntegerBitWidth();
57 ConstantRange Range(APInt(BitWidth, Low), APInt(BitWidth, High));
58
59 if (auto CurrentRange = II->getRange())
60 Range = Range.intersectWith(CurrentRange.value());
61
62 II->addRangeRetAttr(Range);
63 return true;
64 }
65
runNVVMIntrRange(Function & F)66 static bool runNVVMIntrRange(Function &F) {
67 struct {
68 unsigned x, y, z;
69 } MaxBlockSize, MaxGridSize;
70
71 const unsigned MetadataNTID = getReqNTID(F).value_or(
72 getMaxNTID(F).value_or(std::numeric_limits<unsigned>::max()));
73
74 MaxBlockSize.x = std::min(1024u, MetadataNTID);
75 MaxBlockSize.y = std::min(1024u, MetadataNTID);
76 MaxBlockSize.z = std::min(64u, MetadataNTID);
77
78 MaxGridSize.x = 0x7fffffff;
79 MaxGridSize.y = 0xffff;
80 MaxGridSize.z = 0xffff;
81
82 // Go through the calls in this function.
83 bool Changed = false;
84 for (Instruction &I : instructions(F)) {
85 IntrinsicInst *II = dyn_cast<IntrinsicInst>(&I);
86 if (!II)
87 continue;
88
89 switch (II->getIntrinsicID()) {
90 // Index within block
91 case Intrinsic::nvvm_read_ptx_sreg_tid_x:
92 Changed |= addRangeAttr(0, MaxBlockSize.x, II);
93 break;
94 case Intrinsic::nvvm_read_ptx_sreg_tid_y:
95 Changed |= addRangeAttr(0, MaxBlockSize.y, II);
96 break;
97 case Intrinsic::nvvm_read_ptx_sreg_tid_z:
98 Changed |= addRangeAttr(0, MaxBlockSize.z, II);
99 break;
100
101 // Block size
102 case Intrinsic::nvvm_read_ptx_sreg_ntid_x:
103 Changed |= addRangeAttr(1, MaxBlockSize.x + 1, II);
104 break;
105 case Intrinsic::nvvm_read_ptx_sreg_ntid_y:
106 Changed |= addRangeAttr(1, MaxBlockSize.y + 1, II);
107 break;
108 case Intrinsic::nvvm_read_ptx_sreg_ntid_z:
109 Changed |= addRangeAttr(1, MaxBlockSize.z + 1, II);
110 break;
111
112 // Index within grid
113 case Intrinsic::nvvm_read_ptx_sreg_ctaid_x:
114 Changed |= addRangeAttr(0, MaxGridSize.x, II);
115 break;
116 case Intrinsic::nvvm_read_ptx_sreg_ctaid_y:
117 Changed |= addRangeAttr(0, MaxGridSize.y, II);
118 break;
119 case Intrinsic::nvvm_read_ptx_sreg_ctaid_z:
120 Changed |= addRangeAttr(0, MaxGridSize.z, II);
121 break;
122
123 // Grid size
124 case Intrinsic::nvvm_read_ptx_sreg_nctaid_x:
125 Changed |= addRangeAttr(1, MaxGridSize.x + 1, II);
126 break;
127 case Intrinsic::nvvm_read_ptx_sreg_nctaid_y:
128 Changed |= addRangeAttr(1, MaxGridSize.y + 1, II);
129 break;
130 case Intrinsic::nvvm_read_ptx_sreg_nctaid_z:
131 Changed |= addRangeAttr(1, MaxGridSize.z + 1, II);
132 break;
133
134 // warp size is constant 32.
135 case Intrinsic::nvvm_read_ptx_sreg_warpsize:
136 Changed |= addRangeAttr(32, 32 + 1, II);
137 break;
138
139 // Lane ID is [0..warpsize)
140 case Intrinsic::nvvm_read_ptx_sreg_laneid:
141 Changed |= addRangeAttr(0, 32, II);
142 break;
143
144 default:
145 break;
146 }
147 }
148
149 return Changed;
150 }
151
runOnFunction(Function & F)152 bool NVVMIntrRange::runOnFunction(Function &F) { return runNVVMIntrRange(F); }
153
run(Function & F,FunctionAnalysisManager & AM)154 PreservedAnalyses NVVMIntrRangePass::run(Function &F,
155 FunctionAnalysisManager &AM) {
156 return runNVVMIntrRange(F) ? PreservedAnalyses::none()
157 : PreservedAnalyses::all();
158 }
159