1 //===-- NVPTXLowerArgs.cpp - Lower arguments ------------------------------===// 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 // 10 // Arguments to kernel and device functions are passed via param space, 11 // which imposes certain restrictions: 12 // http://docs.nvidia.com/cuda/parallel-thread-execution/#state-spaces 13 // 14 // Kernel parameters are read-only and accessible only via ld.param 15 // instruction, directly or via a pointer. Pointers to kernel 16 // arguments can't be converted to generic address space. 17 // 18 // Device function parameters are directly accessible via 19 // ld.param/st.param, but taking the address of one returns a pointer 20 // to a copy created in local space which *can't* be used with 21 // ld.param/st.param. 22 // 23 // Copying a byval struct into local memory in IR allows us to enforce 24 // the param space restrictions, gives the rest of IR a pointer w/o 25 // param space restrictions, and gives us an opportunity to eliminate 26 // the copy. 27 // 28 // Pointer arguments to kernel functions need more work to be lowered: 29 // 30 // 1. Convert non-byval pointer arguments of CUDA kernels to pointers in the 31 // global address space. This allows later optimizations to emit 32 // ld.global.*/st.global.* for accessing these pointer arguments. For 33 // example, 34 // 35 // define void @foo(float* %input) { 36 // %v = load float, float* %input, align 4 37 // ... 38 // } 39 // 40 // becomes 41 // 42 // define void @foo(float* %input) { 43 // %input2 = addrspacecast float* %input to float addrspace(1)* 44 // %input3 = addrspacecast float addrspace(1)* %input2 to float* 45 // %v = load float, float* %input3, align 4 46 // ... 47 // } 48 // 49 // Later, NVPTXInferAddressSpaces will optimize it to 50 // 51 // define void @foo(float* %input) { 52 // %input2 = addrspacecast float* %input to float addrspace(1)* 53 // %v = load float, float addrspace(1)* %input2, align 4 54 // ... 55 // } 56 // 57 // 2. Convert pointers in a byval kernel parameter to pointers in the global 58 // address space. As #2, it allows NVPTX to emit more ld/st.global. E.g., 59 // 60 // struct S { 61 // int *x; 62 // int *y; 63 // }; 64 // __global__ void foo(S s) { 65 // int *b = s.y; 66 // // use b 67 // } 68 // 69 // "b" points to the global address space. In the IR level, 70 // 71 // define void @foo({i32*, i32*}* byval %input) { 72 // %b_ptr = getelementptr {i32*, i32*}, {i32*, i32*}* %input, i64 0, i32 1 73 // %b = load i32*, i32** %b_ptr 74 // ; use %b 75 // } 76 // 77 // becomes 78 // 79 // define void @foo({i32*, i32*}* byval %input) { 80 // %b_ptr = getelementptr {i32*, i32*}, {i32*, i32*}* %input, i64 0, i32 1 81 // %b = load i32*, i32** %b_ptr 82 // %b_global = addrspacecast i32* %b to i32 addrspace(1)* 83 // %b_generic = addrspacecast i32 addrspace(1)* %b_global to i32* 84 // ; use %b_generic 85 // } 86 // 87 // TODO: merge this pass with NVPTXInferAddressSpaces so that other passes don't 88 // cancel the addrspacecast pair this pass emits. 89 //===----------------------------------------------------------------------===// 90 91 #include "NVPTX.h" 92 #include "NVPTXTargetMachine.h" 93 #include "NVPTXUtilities.h" 94 #include "MCTargetDesc/NVPTXBaseInfo.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(); 162 const DataLayout &DL = Func->getParent()->getDataLayout(); 163 unsigned AS = DL.getAllocaAddrSpace(); 164 AllocaInst *AllocA = new AllocaInst(StructType, AS, Arg->getName(), FirstInst); 165 // Set the alignment to alignment of the byval parameter. This is because, 166 // later load/stores assume that alignment, and we are going to replace 167 // the use of the byval parameter with this alloca instruction. 168 AllocA->setAlignment(Func->getParamAlign(Arg->getArgNo()) 169 .getValueOr(DL.getPrefTypeAlign(StructType))); 170 Arg->replaceAllUsesWith(AllocA); 171 172 Value *ArgInParam = new AddrSpaceCastInst( 173 Arg, PointerType::get(StructType, ADDRESS_SPACE_PARAM), Arg->getName(), 174 FirstInst); 175 LoadInst *LI = 176 new LoadInst(StructType, ArgInParam, Arg->getName(), FirstInst); 177 new StoreInst(LI, AllocA, FirstInst); 178 } 179 180 void NVPTXLowerArgs::markPointerAsGlobal(Value *Ptr) { 181 if (Ptr->getType()->getPointerAddressSpace() == ADDRESS_SPACE_GLOBAL) 182 return; 183 184 // Deciding where to emit the addrspacecast pair. 185 BasicBlock::iterator InsertPt; 186 if (Argument *Arg = dyn_cast<Argument>(Ptr)) { 187 // Insert at the functon entry if Ptr is an argument. 188 InsertPt = Arg->getParent()->getEntryBlock().begin(); 189 } else { 190 // Insert right after Ptr if Ptr is an instruction. 191 InsertPt = ++cast<Instruction>(Ptr)->getIterator(); 192 assert(InsertPt != InsertPt->getParent()->end() && 193 "We don't call this function with Ptr being a terminator."); 194 } 195 196 Instruction *PtrInGlobal = new AddrSpaceCastInst( 197 Ptr, PointerType::get(Ptr->getType()->getPointerElementType(), 198 ADDRESS_SPACE_GLOBAL), 199 Ptr->getName(), &*InsertPt); 200 Value *PtrInGeneric = new AddrSpaceCastInst(PtrInGlobal, Ptr->getType(), 201 Ptr->getName(), &*InsertPt); 202 // Replace with PtrInGeneric all uses of Ptr except PtrInGlobal. 203 Ptr->replaceAllUsesWith(PtrInGeneric); 204 PtrInGlobal->setOperand(0, Ptr); 205 } 206 207 // ============================================================================= 208 // Main function for this pass. 209 // ============================================================================= 210 bool NVPTXLowerArgs::runOnKernelFunction(Function &F) { 211 if (TM && TM->getDrvInterface() == NVPTX::CUDA) { 212 // Mark pointers in byval structs as global. 213 for (auto &B : F) { 214 for (auto &I : B) { 215 if (LoadInst *LI = dyn_cast<LoadInst>(&I)) { 216 if (LI->getType()->isPointerTy()) { 217 Value *UO = GetUnderlyingObject(LI->getPointerOperand(), 218 F.getParent()->getDataLayout()); 219 if (Argument *Arg = dyn_cast<Argument>(UO)) { 220 if (Arg->hasByValAttr()) { 221 // LI is a load from a pointer within a byval kernel parameter. 222 markPointerAsGlobal(LI); 223 } 224 } 225 } 226 } 227 } 228 } 229 } 230 231 for (Argument &Arg : F.args()) { 232 if (Arg.getType()->isPointerTy()) { 233 if (Arg.hasByValAttr()) 234 handleByValParam(&Arg); 235 else if (TM && TM->getDrvInterface() == NVPTX::CUDA) 236 markPointerAsGlobal(&Arg); 237 } 238 } 239 return true; 240 } 241 242 // Device functions only need to copy byval args into local memory. 243 bool NVPTXLowerArgs::runOnDeviceFunction(Function &F) { 244 for (Argument &Arg : F.args()) 245 if (Arg.getType()->isPointerTy() && Arg.hasByValAttr()) 246 handleByValParam(&Arg); 247 return true; 248 } 249 250 bool NVPTXLowerArgs::runOnFunction(Function &F) { 251 return isKernelFunction(F) ? runOnKernelFunction(F) : runOnDeviceFunction(F); 252 } 253 254 FunctionPass * 255 llvm::createNVPTXLowerArgsPass(const NVPTXTargetMachine *TM) { 256 return new NVPTXLowerArgs(TM); 257 } 258