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
NVPTXLowerArgs(const NVPTXTargetMachine * TM=nullptr)125   NVPTXLowerArgs(const NVPTXTargetMachine *TM = nullptr)
126       : FunctionPass(ID), TM(TM) {}
getPassName() const127   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 // =============================================================================
handleByValParam(Argument * Arg)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   // Be sure to propagate alignment to this load; LLVM doesn't know that NVPTX
176   // addrspacecast preserves alignment.  Since params are constant, this load is
177   // definitely not volatile.
178   LoadInst *LI =
179       new LoadInst(StructType, ArgInParam, Arg->getName(),
180                    /*isVolatile=*/false, AllocA->getAlign(), FirstInst);
181   new StoreInst(LI, AllocA, FirstInst);
182 }
183 
markPointerAsGlobal(Value * Ptr)184 void NVPTXLowerArgs::markPointerAsGlobal(Value *Ptr) {
185   if (Ptr->getType()->getPointerAddressSpace() == ADDRESS_SPACE_GLOBAL)
186     return;
187 
188   // Deciding where to emit the addrspacecast pair.
189   BasicBlock::iterator InsertPt;
190   if (Argument *Arg = dyn_cast<Argument>(Ptr)) {
191     // Insert at the functon entry if Ptr is an argument.
192     InsertPt = Arg->getParent()->getEntryBlock().begin();
193   } else {
194     // Insert right after Ptr if Ptr is an instruction.
195     InsertPt = ++cast<Instruction>(Ptr)->getIterator();
196     assert(InsertPt != InsertPt->getParent()->end() &&
197            "We don't call this function with Ptr being a terminator.");
198   }
199 
200   Instruction *PtrInGlobal = new AddrSpaceCastInst(
201       Ptr, PointerType::get(Ptr->getType()->getPointerElementType(),
202                             ADDRESS_SPACE_GLOBAL),
203       Ptr->getName(), &*InsertPt);
204   Value *PtrInGeneric = new AddrSpaceCastInst(PtrInGlobal, Ptr->getType(),
205                                               Ptr->getName(), &*InsertPt);
206   // Replace with PtrInGeneric all uses of Ptr except PtrInGlobal.
207   Ptr->replaceAllUsesWith(PtrInGeneric);
208   PtrInGlobal->setOperand(0, Ptr);
209 }
210 
211 // =============================================================================
212 // Main function for this pass.
213 // =============================================================================
runOnKernelFunction(Function & F)214 bool NVPTXLowerArgs::runOnKernelFunction(Function &F) {
215   if (TM && TM->getDrvInterface() == NVPTX::CUDA) {
216     // Mark pointers in byval structs as global.
217     for (auto &B : F) {
218       for (auto &I : B) {
219         if (LoadInst *LI = dyn_cast<LoadInst>(&I)) {
220           if (LI->getType()->isPointerTy()) {
221             Value *UO = getUnderlyingObject(LI->getPointerOperand());
222             if (Argument *Arg = dyn_cast<Argument>(UO)) {
223               if (Arg->hasByValAttr()) {
224                 // LI is a load from a pointer within a byval kernel parameter.
225                 markPointerAsGlobal(LI);
226               }
227             }
228           }
229         }
230       }
231     }
232   }
233 
234   for (Argument &Arg : F.args()) {
235     if (Arg.getType()->isPointerTy()) {
236       if (Arg.hasByValAttr())
237         handleByValParam(&Arg);
238       else if (TM && TM->getDrvInterface() == NVPTX::CUDA)
239         markPointerAsGlobal(&Arg);
240     }
241   }
242   return true;
243 }
244 
245 // Device functions only need to copy byval args into local memory.
runOnDeviceFunction(Function & F)246 bool NVPTXLowerArgs::runOnDeviceFunction(Function &F) {
247   for (Argument &Arg : F.args())
248     if (Arg.getType()->isPointerTy() && Arg.hasByValAttr())
249       handleByValParam(&Arg);
250   return true;
251 }
252 
runOnFunction(Function & F)253 bool NVPTXLowerArgs::runOnFunction(Function &F) {
254   return isKernelFunction(F) ? runOnKernelFunction(F) : runOnDeviceFunction(F);
255 }
256 
257 FunctionPass *
createNVPTXLowerArgsPass(const NVPTXTargetMachine * TM)258 llvm::createNVPTXLowerArgsPass(const NVPTXTargetMachine *TM) {
259   return new NVPTXLowerArgs(TM);
260 }
261