Go to most recent revision | Details | Last modification | View Log | RSS feed
Rev | Author | Line No. | Line |
---|---|---|---|
5564 | serge | 1 | /* |
2 | * Copyright 2012, 2013 Advanced Micro Devices, Inc. |
||
3 | * |
||
4 | * Permission is hereby granted, free of charge, to any person obtaining a |
||
5 | * copy of this software and associated documentation files (the "Software"), |
||
6 | * to deal in the Software without restriction, including without limitation |
||
7 | * the rights to use, copy, modify, merge, publish, distribute, sublicense, |
||
8 | * and/or sell copies of the Software, and to permit persons to whom the |
||
9 | * Software is furnished to do so, subject to the following conditions: |
||
10 | * |
||
11 | * The above copyright notice and this permission notice (including the next |
||
12 | * paragraph) shall be included in all copies or substantial portions of the |
||
13 | * Software. |
||
14 | * |
||
15 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
||
16 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
||
17 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL |
||
18 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
||
19 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, |
||
20 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE |
||
21 | * SOFTWARE. |
||
22 | * |
||
23 | * Authors: Tom Stellard |
||
24 | * |
||
25 | */ |
||
26 | |||
27 | #include "radeon_llvm_util.h" |
||
28 | #include "util/u_memory.h" |
||
29 | |||
30 | #include |
||
31 | #include |
||
32 | #include |
||
33 | #include |
||
34 | #include |
||
35 | |||
36 | LLVMModuleRef radeon_llvm_parse_bitcode(LLVMContextRef ctx, |
||
37 | const char * bitcode, unsigned bitcode_len) |
||
38 | { |
||
39 | LLVMMemoryBufferRef buf; |
||
40 | LLVMModuleRef module; |
||
41 | |||
42 | buf = LLVMCreateMemoryBufferWithMemoryRangeCopy((const char*)bitcode, |
||
43 | bitcode_len, "radeon"); |
||
44 | LLVMParseBitcodeInContext(ctx, buf, &module, NULL); |
||
45 | LLVMDisposeMemoryBuffer(buf); |
||
46 | return module; |
||
47 | } |
||
48 | |||
49 | unsigned radeon_llvm_get_num_kernels(LLVMContextRef ctx, |
||
50 | const char *bitcode, unsigned bitcode_len) |
||
51 | { |
||
52 | LLVMModuleRef mod = radeon_llvm_parse_bitcode(ctx, bitcode, bitcode_len); |
||
53 | return LLVMGetNamedMetadataNumOperands(mod, "opencl.kernels"); |
||
54 | } |
||
55 | |||
56 | static void radeon_llvm_optimize(LLVMModuleRef mod) |
||
57 | { |
||
58 | const char *data_layout = LLVMGetDataLayout(mod); |
||
59 | LLVMTargetDataRef TD = LLVMCreateTargetData(data_layout); |
||
60 | LLVMPassManagerBuilderRef builder = LLVMPassManagerBuilderCreate(); |
||
61 | LLVMPassManagerRef pass_manager = LLVMCreatePassManager(); |
||
62 | |||
63 | /* Functions calls are not supported yet, so we need to inline |
||
64 | * everything. The most efficient way to do this is to add |
||
65 | * the always_inline attribute to all non-kernel functions |
||
66 | * and then run the Always Inline pass. The Always Inline |
||
67 | * pass will automaically inline functions with this attribute |
||
68 | * and does not perform the expensive cost analysis that the normal |
||
69 | * inliner does. |
||
70 | */ |
||
71 | |||
72 | LLVMValueRef fn; |
||
73 | for (fn = LLVMGetFirstFunction(mod); fn; fn = LLVMGetNextFunction(fn)) { |
||
74 | /* All the non-kernel functions have internal linkage */ |
||
75 | if (LLVMGetLinkage(fn) == LLVMInternalLinkage) { |
||
76 | LLVMAddFunctionAttr(fn, LLVMAlwaysInlineAttribute); |
||
77 | } |
||
78 | } |
||
79 | |||
80 | LLVMAddTargetData(TD, pass_manager); |
||
81 | LLVMAddAlwaysInlinerPass(pass_manager); |
||
82 | LLVMPassManagerBuilderPopulateModulePassManager(builder, pass_manager); |
||
83 | |||
84 | LLVMRunPassManager(pass_manager, mod); |
||
85 | LLVMPassManagerBuilderDispose(builder); |
||
86 | LLVMDisposePassManager(pass_manager); |
||
87 | LLVMDisposeTargetData(TD); |
||
88 | } |
||
89 | |||
90 | LLVMModuleRef radeon_llvm_get_kernel_module(LLVMContextRef ctx, unsigned index, |
||
91 | const char *bitcode, unsigned bitcode_len) |
||
92 | { |
||
93 | LLVMModuleRef mod; |
||
94 | unsigned num_kernels; |
||
95 | LLVMValueRef *kernel_metadata; |
||
96 | unsigned i; |
||
97 | |||
98 | mod = radeon_llvm_parse_bitcode(ctx, bitcode, bitcode_len); |
||
99 | num_kernels = LLVMGetNamedMetadataNumOperands(mod, "opencl.kernels"); |
||
100 | kernel_metadata = MALLOC(num_kernels * sizeof(LLVMValueRef)); |
||
101 | LLVMGetNamedMetadataOperands(mod, "opencl.kernels", kernel_metadata); |
||
102 | for (i = 0; i < num_kernels; i++) { |
||
103 | LLVMValueRef kernel_signature, *kernel_function; |
||
104 | unsigned num_kernel_md_operands; |
||
105 | if (i == index) { |
||
106 | continue; |
||
107 | } |
||
108 | kernel_signature = kernel_metadata[i]; |
||
109 | num_kernel_md_operands = LLVMGetMDNodeNumOperands(kernel_signature); |
||
110 | kernel_function = MALLOC(num_kernel_md_operands * sizeof (LLVMValueRef)); |
||
111 | LLVMGetMDNodeOperands(kernel_signature, kernel_function); |
||
112 | LLVMDeleteFunction(*kernel_function); |
||
113 | FREE(kernel_function); |
||
114 | } |
||
115 | FREE(kernel_metadata); |
||
116 | radeon_llvm_optimize(mod); |
||
117 | return mod; |
||
118 | }> |