|
120
|
1 //===-- NVPTXLowerArgs.cpp - Lower arguments ------------------------------===//
|
|
|
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 //
|
|
|
11 // Arguments to kernel and device functions are passed via param space,
|
|
|
12 // which imposes certain restrictions:
|
|
|
13 // http://docs.nvidia.com/cuda/parallel-thread-execution/#state-spaces
|
|
|
14 //
|
|
|
15 // Kernel parameters are read-only and accessible only via ld.param
|
|
|
16 // instruction, directly or via a pointer. Pointers to kernel
|
|
|
17 // arguments can't be converted to generic address space.
|
|
|
18 //
|
|
|
19 // Device function parameters are directly accessible via
|
|
|
20 // ld.param/st.param, but taking the address of one returns a pointer
|
|
|
21 // to a copy created in local space which *can't* be used with
|
|
|
22 // ld.param/st.param.
|
|
|
23 //
|
|
|
24 // Copying a byval struct into local memory in IR allows us to enforce
|
|
|
25 // the param space restrictions, gives the rest of IR a pointer w/o
|
|
|
26 // param space restrictions, and gives us an opportunity to eliminate
|
|
|
27 // the copy.
|
|
|
28 //
|
|
|
29 // Pointer arguments to kernel functions need more work to be lowered:
|
|
|
30 //
|
|
|
31 // 1. Convert non-byval pointer arguments of CUDA kernels to pointers in the
|
|
|
32 // global address space. This allows later optimizations to emit
|
|
|
33 // ld.global.*/st.global.* for accessing these pointer arguments. For
|
|
|
34 // example,
|
|
|
35 //
|
|
|
36 // define void @foo(float* %input) {
|
|
|
37 // %v = load float, float* %input, align 4
|
|
|
38 // ...
|
|
|
39 // }
|
|
|
40 //
|
|
|
41 // becomes
|
|
|
42 //
|
|
|
43 // define void @foo(float* %input) {
|
|
|
44 // %input2 = addrspacecast float* %input to float addrspace(1)*
|
|
|
45 // %input3 = addrspacecast float addrspace(1)* %input2 to float*
|
|
|
46 // %v = load float, float* %input3, align 4
|
|
|
47 // ...
|
|
|
48 // }
|
|
|
49 //
|
|
|
50 // Later, NVPTXInferAddressSpaces will optimize it to
|
|
|
51 //
|
|
|
52 // define void @foo(float* %input) {
|
|
|
53 // %input2 = addrspacecast float* %input to float addrspace(1)*
|
|
|
54 // %v = load float, float addrspace(1)* %input2, align 4
|
|
|
55 // ...
|
|
|
56 // }
|
|
|
57 //
|
|
|
58 // 2. Convert pointers in a byval kernel parameter to pointers in the global
|
|
|
59 // address space. As #2, it allows NVPTX to emit more ld/st.global. E.g.,
|
|
|
60 //
|
|
|
61 // struct S {
|
|
|
62 // int *x;
|
|
|
63 // int *y;
|
|
|
64 // };
|
|
|
65 // __global__ void foo(S s) {
|
|
|
66 // int *b = s.y;
|
|
|
67 // // use b
|
|
|
68 // }
|
|
|
69 //
|
|
|
70 // "b" points to the global address space. In the IR level,
|
|
|
71 //
|
|
|
72 // define void @foo({i32*, i32*}* byval %input) {
|
|
|
73 // %b_ptr = getelementptr {i32*, i32*}, {i32*, i32*}* %input, i64 0, i32 1
|
|
|
74 // %b = load i32*, i32** %b_ptr
|
|
|
75 // ; use %b
|
|
|
76 // }
|
|
|
77 //
|
|
|
78 // becomes
|
|
|
79 //
|
|
|
80 // define void @foo({i32*, i32*}* byval %input) {
|
|
|
81 // %b_ptr = getelementptr {i32*, i32*}, {i32*, i32*}* %input, i64 0, i32 1
|
|
|
82 // %b = load i32*, i32** %b_ptr
|
|
|
83 // %b_global = addrspacecast i32* %b to i32 addrspace(1)*
|
|
|
84 // %b_generic = addrspacecast i32 addrspace(1)* %b_global to i32*
|
|
|
85 // ; use %b_generic
|
|
|
86 // }
|
|
|
87 //
|
|
|
88 // TODO: merge this pass with NVPTXInferAddressSpaces so that other passes don't
|
|
|
89 // cancel the addrspacecast pair this pass emits.
|
|
|
90 //===----------------------------------------------------------------------===//
|
|
|
91
|
|
|
92 #include "NVPTX.h"
|
|
121
|
93 #include "NVPTXTargetMachine.h"
|
|
120
|
94 #include "NVPTXUtilities.h"
|
|
|
95 #include "llvm/Analysis/ValueTracking.h"
|
|
|
96 #include "llvm/IR/Function.h"
|
|
|
97 #include "llvm/IR/Instructions.h"
|
|
|
98 #include "llvm/IR/Module.h"
|
|
|
99 #include "llvm/IR/Type.h"
|
|
|
100 #include "llvm/Pass.h"
|
|
|
101
|
|
|
102 using namespace llvm;
|
|
|
103
|
|
|
104 namespace llvm {
|
|
|
105 void initializeNVPTXLowerArgsPass(PassRegistry &);
|
|
|
106 }
|
|
|
107
|
|
|
108 namespace {
|
|
|
109 class NVPTXLowerArgs : public FunctionPass {
|
|
|
110 bool runOnFunction(Function &F) override;
|
|
|
111
|
|
|
112 bool runOnKernelFunction(Function &F);
|
|
|
113 bool runOnDeviceFunction(Function &F);
|
|
|
114
|
|
|
115 // handle byval parameters
|
|
|
116 void handleByValParam(Argument *Arg);
|
|
|
117 // Knowing Ptr must point to the global address space, this function
|
|
|
118 // addrspacecasts Ptr to global and then back to generic. This allows
|
|
|
119 // NVPTXInferAddressSpaces to fold the global-to-generic cast into
|
|
|
120 // loads/stores that appear later.
|
|
|
121 void markPointerAsGlobal(Value *Ptr);
|
|
|
122
|
|
|
123 public:
|
|
|
124 static char ID; // Pass identification, replacement for typeid
|
|
|
125 NVPTXLowerArgs(const NVPTXTargetMachine *TM = nullptr)
|
|
|
126 : FunctionPass(ID), TM(TM) {}
|
|
|
127 StringRef getPassName() const override {
|
|
|
128 return "Lower pointer arguments of CUDA kernels";
|
|
|
129 }
|
|
|
130
|
|
|
131 private:
|
|
|
132 const NVPTXTargetMachine *TM;
|
|
|
133 };
|
|
|
134 } // namespace
|
|
|
135
|
|
|
136 char NVPTXLowerArgs::ID = 1;
|
|
|
137
|
|
|
138 INITIALIZE_PASS(NVPTXLowerArgs, "nvptx-lower-args",
|
|
|
139 "Lower arguments (NVPTX)", false, false)
|
|
|
140
|
|
|
141 // =============================================================================
|
|
|
142 // If the function had a byval struct ptr arg, say foo(%struct.x* byval %d),
|
|
|
143 // then add the following instructions to the first basic block:
|
|
|
144 //
|
|
|
145 // %temp = alloca %struct.x, align 8
|
|
|
146 // %tempd = addrspacecast %struct.x* %d to %struct.x addrspace(101)*
|
|
|
147 // %tv = load %struct.x addrspace(101)* %tempd
|
|
|
148 // store %struct.x %tv, %struct.x* %temp, align 8
|
|
|
149 //
|
|
|
150 // The above code allocates some space in the stack and copies the incoming
|
|
|
151 // struct from param space to local space.
|
|
|
152 // Then replace all occurrences of %d by %temp.
|
|
|
153 // =============================================================================
|
|
|
154 void NVPTXLowerArgs::handleByValParam(Argument *Arg) {
|
|
|
155 Function *Func = Arg->getParent();
|
|
|
156 Instruction *FirstInst = &(Func->getEntryBlock().front());
|
|
|
157 PointerType *PType = dyn_cast<PointerType>(Arg->getType());
|
|
|
158
|
|
|
159 assert(PType && "Expecting pointer type in handleByValParam");
|
|
|
160
|
|
|
161 Type *StructType = PType->getElementType();
|
|
121
|
162 unsigned AS = Func->getParent()->getDataLayout().getAllocaAddrSpace();
|
|
|
163 AllocaInst *AllocA = new AllocaInst(StructType, AS, Arg->getName(), FirstInst);
|
|
120
|
164 // Set the alignment to alignment of the byval parameter. This is because,
|
|
|
165 // later load/stores assume that alignment, and we are going to replace
|
|
|
166 // the use of the byval parameter with this alloca instruction.
|
|
121
|
167 AllocA->setAlignment(Func->getParamAlignment(Arg->getArgNo()));
|
|
120
|
168 Arg->replaceAllUsesWith(AllocA);
|
|
|
169
|
|
|
170 Value *ArgInParam = new AddrSpaceCastInst(
|
|
|
171 Arg, PointerType::get(StructType, ADDRESS_SPACE_PARAM), Arg->getName(),
|
|
|
172 FirstInst);
|
|
|
173 LoadInst *LI = new LoadInst(ArgInParam, Arg->getName(), FirstInst);
|
|
|
174 new StoreInst(LI, AllocA, FirstInst);
|
|
|
175 }
|
|
|
176
|
|
|
177 void NVPTXLowerArgs::markPointerAsGlobal(Value *Ptr) {
|
|
|
178 if (Ptr->getType()->getPointerAddressSpace() == ADDRESS_SPACE_GLOBAL)
|
|
|
179 return;
|
|
|
180
|
|
|
181 // Deciding where to emit the addrspacecast pair.
|
|
|
182 BasicBlock::iterator InsertPt;
|
|
|
183 if (Argument *Arg = dyn_cast<Argument>(Ptr)) {
|
|
|
184 // Insert at the functon entry if Ptr is an argument.
|
|
|
185 InsertPt = Arg->getParent()->getEntryBlock().begin();
|
|
|
186 } else {
|
|
|
187 // Insert right after Ptr if Ptr is an instruction.
|
|
|
188 InsertPt = ++cast<Instruction>(Ptr)->getIterator();
|
|
|
189 assert(InsertPt != InsertPt->getParent()->end() &&
|
|
|
190 "We don't call this function with Ptr being a terminator.");
|
|
|
191 }
|
|
|
192
|
|
|
193 Instruction *PtrInGlobal = new AddrSpaceCastInst(
|
|
|
194 Ptr, PointerType::get(Ptr->getType()->getPointerElementType(),
|
|
|
195 ADDRESS_SPACE_GLOBAL),
|
|
|
196 Ptr->getName(), &*InsertPt);
|
|
|
197 Value *PtrInGeneric = new AddrSpaceCastInst(PtrInGlobal, Ptr->getType(),
|
|
|
198 Ptr->getName(), &*InsertPt);
|
|
|
199 // Replace with PtrInGeneric all uses of Ptr except PtrInGlobal.
|
|
|
200 Ptr->replaceAllUsesWith(PtrInGeneric);
|
|
|
201 PtrInGlobal->setOperand(0, Ptr);
|
|
|
202 }
|
|
|
203
|
|
|
204 // =============================================================================
|
|
|
205 // Main function for this pass.
|
|
|
206 // =============================================================================
|
|
|
207 bool NVPTXLowerArgs::runOnKernelFunction(Function &F) {
|
|
|
208 if (TM && TM->getDrvInterface() == NVPTX::CUDA) {
|
|
|
209 // Mark pointers in byval structs as global.
|
|
|
210 for (auto &B : F) {
|
|
|
211 for (auto &I : B) {
|
|
|
212 if (LoadInst *LI = dyn_cast<LoadInst>(&I)) {
|
|
|
213 if (LI->getType()->isPointerTy()) {
|
|
|
214 Value *UO = GetUnderlyingObject(LI->getPointerOperand(),
|
|
|
215 F.getParent()->getDataLayout());
|
|
|
216 if (Argument *Arg = dyn_cast<Argument>(UO)) {
|
|
|
217 if (Arg->hasByValAttr()) {
|
|
|
218 // LI is a load from a pointer within a byval kernel parameter.
|
|
|
219 markPointerAsGlobal(LI);
|
|
|
220 }
|
|
|
221 }
|
|
|
222 }
|
|
|
223 }
|
|
|
224 }
|
|
|
225 }
|
|
|
226 }
|
|
|
227
|
|
|
228 for (Argument &Arg : F.args()) {
|
|
|
229 if (Arg.getType()->isPointerTy()) {
|
|
|
230 if (Arg.hasByValAttr())
|
|
|
231 handleByValParam(&Arg);
|
|
|
232 else if (TM && TM->getDrvInterface() == NVPTX::CUDA)
|
|
|
233 markPointerAsGlobal(&Arg);
|
|
|
234 }
|
|
|
235 }
|
|
|
236 return true;
|
|
|
237 }
|
|
|
238
|
|
|
239 // Device functions only need to copy byval args into local memory.
|
|
|
240 bool NVPTXLowerArgs::runOnDeviceFunction(Function &F) {
|
|
|
241 for (Argument &Arg : F.args())
|
|
|
242 if (Arg.getType()->isPointerTy() && Arg.hasByValAttr())
|
|
|
243 handleByValParam(&Arg);
|
|
|
244 return true;
|
|
|
245 }
|
|
|
246
|
|
|
247 bool NVPTXLowerArgs::runOnFunction(Function &F) {
|
|
|
248 return isKernelFunction(F) ? runOnKernelFunction(F) : runOnDeviceFunction(F);
|
|
|
249 }
|
|
|
250
|
|
|
251 FunctionPass *
|
|
|
252 llvm::createNVPTXLowerArgsPass(const NVPTXTargetMachine *TM) {
|
|
|
253 return new NVPTXLowerArgs(TM);
|
|
|
254 }
|
