annotate lib/Target/NVPTX/NVPTXLowerKernelArgs.cpp @ 95:afa8332a0e37 LLVM3.8

LLVM 3.8
author Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
date Tue, 13 Oct 2015 17:48:58 +0900
parents
children 7d135dc70f03
Ignore whitespace changes - Everywhere: Within whitespace: At end of lines:
rev   line source
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
1 //===-- NVPTXLowerKernelArgs.cpp - Lower kernel arguments -----------------===//
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
2 //
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
3 // The LLVM Compiler Infrastructure
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
4 //
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
5 // This file is distributed under the University of Illinois Open Source
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
6 // License. See LICENSE.TXT for details.
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
7 //
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
8 //===----------------------------------------------------------------------===//
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
9 //
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
10 // Pointer arguments to kernel functions need to be lowered specially.
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
11 //
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
12 // 1. Copy byval struct args to local memory. This is a preparation for handling
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
13 // cases like
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
14 //
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
15 // kernel void foo(struct A arg, ...)
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
16 // {
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
17 // struct A *p = &arg;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
18 // ...
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
19 // ... = p->filed1 ... (this is no generic address for .param)
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
20 // p->filed2 = ... (this is no write access to .param)
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
21 // }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
22 //
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
23 // 2. Convert non-byval pointer arguments of CUDA kernels to pointers in the
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
24 // global address space. This allows later optimizations to emit
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
25 // ld.global.*/st.global.* for accessing these pointer arguments. For
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
26 // example,
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
27 //
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
28 // define void @foo(float* %input) {
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
29 // %v = load float, float* %input, align 4
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
30 // ...
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
31 // }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
32 //
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
33 // becomes
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
34 //
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
35 // define void @foo(float* %input) {
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
36 // %input2 = addrspacecast float* %input to float addrspace(1)*
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
37 // %input3 = addrspacecast float addrspace(1)* %input2 to float*
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
38 // %v = load float, float* %input3, align 4
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
39 // ...
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
40 // }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
41 //
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
42 // Later, NVPTXFavorNonGenericAddrSpaces will optimize it to
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
43 //
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
44 // define void @foo(float* %input) {
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
45 // %input2 = addrspacecast float* %input to float addrspace(1)*
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
46 // %v = load float, float addrspace(1)* %input2, align 4
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
47 // ...
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
48 // }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
49 //
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
50 // 3. Convert pointers in a byval kernel parameter to pointers in the global
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
51 // address space. As #2, it allows NVPTX to emit more ld/st.global. E.g.,
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
52 //
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
53 // struct S {
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
54 // int *x;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
55 // int *y;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
56 // };
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
57 // __global__ void foo(S s) {
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
58 // int *b = s.y;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
59 // // use b
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
60 // }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
61 //
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
62 // "b" points to the global address space. In the IR level,
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
63 //
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
64 // define void @foo({i32*, i32*}* byval %input) {
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
65 // %b_ptr = getelementptr {i32*, i32*}, {i32*, i32*}* %input, i64 0, i32 1
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
66 // %b = load i32*, i32** %b_ptr
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
67 // ; use %b
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
68 // }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
69 //
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
70 // becomes
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
71 //
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
72 // define void @foo({i32*, i32*}* byval %input) {
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
73 // %b_ptr = getelementptr {i32*, i32*}, {i32*, i32*}* %input, i64 0, i32 1
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
74 // %b = load i32*, i32** %b_ptr
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
75 // %b_global = addrspacecast i32* %b to i32 addrspace(1)*
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
76 // %b_generic = addrspacecast i32 addrspace(1)* %b_global to i32*
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
77 // ; use %b_generic
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
78 // }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
79 //
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
80 // TODO: merge this pass with NVPTXFavorNonGenericAddrSpace so that other passes
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
81 // don't cancel the addrspacecast pair this pass emits.
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
82 //===----------------------------------------------------------------------===//
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
83
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
84 #include "NVPTX.h"
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
85 #include "NVPTXUtilities.h"
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
86 #include "NVPTXTargetMachine.h"
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
87 #include "llvm/Analysis/ValueTracking.h"
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
88 #include "llvm/IR/Function.h"
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
89 #include "llvm/IR/Instructions.h"
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
90 #include "llvm/IR/Module.h"
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
91 #include "llvm/IR/Type.h"
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
92 #include "llvm/Pass.h"
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
93
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
94 using namespace llvm;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
96 namespace llvm {
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
97 void initializeNVPTXLowerKernelArgsPass(PassRegistry &);
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
98 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
99
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
100 namespace {
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
101 class NVPTXLowerKernelArgs : public FunctionPass {
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
102 bool runOnFunction(Function &F) override;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
103
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
104 // handle byval parameters
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
105 void handleByValParam(Argument *Arg);
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
106 // Knowing Ptr must point to the global address space, this function
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
107 // addrspacecasts Ptr to global and then back to generic. This allows
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
108 // NVPTXFavorNonGenericAddrSpace to fold the global-to-generic cast into
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
109 // loads/stores that appear later.
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
110 void markPointerAsGlobal(Value *Ptr);
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
111
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
112 public:
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
113 static char ID; // Pass identification, replacement for typeid
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
114 NVPTXLowerKernelArgs(const NVPTXTargetMachine *TM = nullptr)
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
115 : FunctionPass(ID), TM(TM) {}
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
116 const char *getPassName() const override {
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
117 return "Lower pointer arguments of CUDA kernels";
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
118 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
119
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
120 private:
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
121 const NVPTXTargetMachine *TM;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
122 };
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
123 } // namespace
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
124
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
125 char NVPTXLowerKernelArgs::ID = 1;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
126
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
127 INITIALIZE_PASS(NVPTXLowerKernelArgs, "nvptx-lower-kernel-args",
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
128 "Lower kernel arguments (NVPTX)", false, false)
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
129
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
130 // =============================================================================
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
131 // If the function had a byval struct ptr arg, say foo(%struct.x *byval %d),
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
132 // then add the following instructions to the first basic block:
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
133 //
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
134 // %temp = alloca %struct.x, align 8
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
135 // %tempd = addrspacecast %struct.x* %d to %struct.x addrspace(101)*
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
136 // %tv = load %struct.x addrspace(101)* %tempd
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
137 // store %struct.x %tv, %struct.x* %temp, align 8
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
138 //
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
139 // The above code allocates some space in the stack and copies the incoming
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
140 // struct from param space to local space.
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
141 // Then replace all occurrences of %d by %temp.
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
142 // =============================================================================
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
143 void NVPTXLowerKernelArgs::handleByValParam(Argument *Arg) {
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
144 Function *Func = Arg->getParent();
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
145 Instruction *FirstInst = &(Func->getEntryBlock().front());
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
146 PointerType *PType = dyn_cast<PointerType>(Arg->getType());
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
147
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
148 assert(PType && "Expecting pointer type in handleByValParam");
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
149
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
150 Type *StructType = PType->getElementType();
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
151 AllocaInst *AllocA = new AllocaInst(StructType, Arg->getName(), FirstInst);
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
152 // Set the alignment to alignment of the byval parameter. This is because,
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
153 // later load/stores assume that alignment, and we are going to replace
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
154 // the use of the byval parameter with this alloca instruction.
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
155 AllocA->setAlignment(Func->getParamAlignment(Arg->getArgNo() + 1));
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
156 Arg->replaceAllUsesWith(AllocA);
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
157
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
158 Value *ArgInParam = new AddrSpaceCastInst(
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
159 Arg, PointerType::get(StructType, ADDRESS_SPACE_PARAM), Arg->getName(),
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
160 FirstInst);
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
161 LoadInst *LI = new LoadInst(ArgInParam, Arg->getName(), FirstInst);
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
162 new StoreInst(LI, AllocA, FirstInst);
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
163 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
164
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
165 void NVPTXLowerKernelArgs::markPointerAsGlobal(Value *Ptr) {
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
166 if (Ptr->getType()->getPointerAddressSpace() == ADDRESS_SPACE_GLOBAL)
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
167 return;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
168
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
169 // Deciding where to emit the addrspacecast pair.
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
170 BasicBlock::iterator InsertPt;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
171 if (Argument *Arg = dyn_cast<Argument>(Ptr)) {
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
172 // Insert at the functon entry if Ptr is an argument.
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
173 InsertPt = Arg->getParent()->getEntryBlock().begin();
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
174 } else {
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
175 // Insert right after Ptr if Ptr is an instruction.
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
176 InsertPt = cast<Instruction>(Ptr);
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
177 ++InsertPt;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
178 assert(InsertPt != InsertPt->getParent()->end() &&
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
179 "We don't call this function with Ptr being a terminator.");
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
180 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
181
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
182 Instruction *PtrInGlobal = new AddrSpaceCastInst(
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
183 Ptr, PointerType::get(Ptr->getType()->getPointerElementType(),
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
184 ADDRESS_SPACE_GLOBAL),
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
185 Ptr->getName(), InsertPt);
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
186 Value *PtrInGeneric = new AddrSpaceCastInst(PtrInGlobal, Ptr->getType(),
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
187 Ptr->getName(), InsertPt);
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
188 // Replace with PtrInGeneric all uses of Ptr except PtrInGlobal.
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
189 Ptr->replaceAllUsesWith(PtrInGeneric);
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
190 PtrInGlobal->setOperand(0, Ptr);
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
191 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
192
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
193 // =============================================================================
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
194 // Main function for this pass.
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
195 // =============================================================================
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
196 bool NVPTXLowerKernelArgs::runOnFunction(Function &F) {
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
197 // Skip non-kernels. See the comments at the top of this file.
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
198 if (!isKernelFunction(F))
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
199 return false;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
200
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
201 if (TM && TM->getDrvInterface() == NVPTX::CUDA) {
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
202 // Mark pointers in byval structs as global.
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
203 for (auto &B : F) {
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
204 for (auto &I : B) {
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
205 if (LoadInst *LI = dyn_cast<LoadInst>(&I)) {
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
206 if (LI->getType()->isPointerTy()) {
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
207 Value *UO = GetUnderlyingObject(LI->getPointerOperand(),
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
208 F.getParent()->getDataLayout());
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
209 if (Argument *Arg = dyn_cast<Argument>(UO)) {
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
210 if (Arg->hasByValAttr()) {
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
211 // LI is a load from a pointer within a byval kernel parameter.
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
212 markPointerAsGlobal(LI);
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
213 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
214 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
215 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
216 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
217 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
218 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
219 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
220
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
221 for (Argument &Arg : F.args()) {
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
222 if (Arg.getType()->isPointerTy()) {
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
223 if (Arg.hasByValAttr())
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
224 handleByValParam(&Arg);
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
225 else if (TM && TM->getDrvInterface() == NVPTX::CUDA)
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
226 markPointerAsGlobal(&Arg);
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
227 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
228 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
229 return true;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
230 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
231
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
232 FunctionPass *
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
233 llvm::createNVPTXLowerKernelArgsPass(const NVPTXTargetMachine *TM) {
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
234 return new NVPTXLowerKernelArgs(TM);
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
235 }