1 // Tests that ptxas and fatbinary are invoked correctly during CUDA
4 // Regular compiles with -O{0,1,2,3,4,fast}. -O4 and -Ofast map to ptxas O3.
5 // RUN: %clang -### --target=x86_64-linux-gnu -O0 -c %s 2>&1 \
6 // RUN: --offload-arch=sm_35 --cuda-path=%S/Inputs/CUDA/usr/local/cuda \
7 // RUN: | FileCheck -check-prefixes=CHECK,ARCH64,SM35,OPT0 %s
8 // RUN: %clang -### --target=x86_64-linux-gnu -O1 -c %s 2>&1 \
9 // RUN: --offload-arch=sm_35 --cuda-path=%S/Inputs/CUDA/usr/local/cuda \
10 // RUN: | FileCheck -check-prefixes=CHECK,ARCH64,SM35,OPT1 %s
11 // RUN: %clang -### --target=x86_64-linux-gnu -O2 -c %s 2>&1 \
12 // RUN: --offload-arch=sm_35 --cuda-path=%S/Inputs/CUDA/usr/local/cuda \
13 // RUN: | FileCheck -check-prefixes=CHECK,ARCH64,SM35,OPT2 %s
14 // RUN: %clang -### --target=x86_64-linux-gnu -O3 -c %s 2>&1 \
15 // RUN: --offload-arch=sm_35 --cuda-path=%S/Inputs/CUDA/usr/local/cuda \
16 // RUN: | FileCheck -check-prefixes=CHECK,ARCH64,SM35,OPT3 %s
17 // RUN: %clang -### --target=x86_64-linux-gnu -O4 -c %s 2>&1 \
18 // RUN: --offload-arch=sm_35 --cuda-path=%S/Inputs/CUDA/usr/local/cuda \
19 // RUN: | FileCheck -check-prefixes=CHECK,ARCH64,SM35,OPT3 %s
20 // RUN: %clang -### --target=x86_64-linux-gnu -Ofast -c %s 2>&1 \
21 // RUN: --offload-arch=sm_35 --cuda-path=%S/Inputs/CUDA/usr/local/cuda \
22 // RUN: | FileCheck -check-prefixes=CHECK,ARCH64,SM35,OPT3 %s
23 // Generating relocatable device code
24 // RUN: %clang -### --target=x86_64-linux-gnu -fgpu-rdc -c %s 2>&1 \
25 // RUN: --no-offload-new-driver --offload-arch=sm_35 --cuda-path=%S/Inputs/CUDA/usr/local/cuda \
26 // RUN: | FileCheck -check-prefixes=CHECK,ARCH64,SM35,RDC %s
28 // With debugging enabled, ptxas should be run with with no ptxas optimizations.
29 // RUN: %clang -### --target=x86_64-linux-gnu --cuda-noopt-device-debug -O2 -g -c %s 2>&1 \
30 // RUN: --offload-arch=sm_35 --cuda-path=%S/Inputs/CUDA/usr/local/cuda \
31 // RUN: | FileCheck -check-prefixes=CHECK,ARCH64,SM35,DBG %s
33 // --no-cuda-noopt-device-debug overrides --cuda-noopt-device-debug.
34 // RUN: %clang -### --target=x86_64-linux-gnu --cuda-noopt-device-debug \
35 // RUN: --offload-arch=sm_35 --cuda-path=%S/Inputs/CUDA/usr/local/cuda \
36 // RUN: --no-cuda-noopt-device-debug -O2 -c %s 2>&1 \
37 // RUN: | FileCheck -check-prefixes=CHECK,ARCH64,SM35,OPT2 %s
39 // Regular compile without -O. This should result in us passing -O0 to ptxas.
40 // RUN: %clang -### --target=x86_64-linux-gnu -c %s 2>&1 \
41 // RUN: --offload-arch=sm_35 --cuda-path=%S/Inputs/CUDA/usr/local/cuda \
42 // RUN: | FileCheck -check-prefixes=CHECK,ARCH64,SM35,OPT0 %s
44 // Regular compiles with -Os and -Oz. For lack of a better option, we map
45 // these to ptxas -O3.
46 // RUN: %clang -### --target=x86_64-linux-gnu -Os -c %s 2>&1 \
47 // RUN: --offload-arch=sm_35 --cuda-path=%S/Inputs/CUDA/usr/local/cuda \
48 // RUN: | FileCheck -check-prefixes=CHECK,ARCH64,SM35,OPT2 %s
49 // RUN: %clang -### --target=x86_64-linux-gnu -Oz -c %s 2>&1 \
50 // RUN: --offload-arch=sm_35 --cuda-path=%S/Inputs/CUDA/usr/local/cuda \
51 // RUN: | FileCheck -check-prefixes=CHECK,ARCH64,SM35,OPT2 %s
53 // Regular compile targeting sm_35.
54 // RUN: %clang -### --target=x86_64-linux-gnu --cuda-gpu-arch=sm_35 -c %s 2>&1 \
55 // RUN: --offload-arch=sm_35 --cuda-path=%S/Inputs/CUDA/usr/local/cuda \
56 // RUN: | FileCheck -check-prefixes=CHECK,ARCH64,SM35 %s
57 // Separate compilation targeting sm_35.
58 // RUN: %clang -### --target=x86_64-linux-gnu --cuda-gpu-arch=sm_35 -fgpu-rdc -c %s 2>&1 \
59 // RUN: --no-offload-new-driver --offload-arch=sm_35 --cuda-path=%S/Inputs/CUDA/usr/local/cuda \
60 // RUN: | FileCheck -check-prefixes=CHECK,ARCH64,SM35,RDC %s
63 // RUN: %clang -### --target=i386-linux-gnu -c %s 2>&1 \
64 // RUN: --offload-arch=sm_35 --cuda-path=%S/Inputs/CUDA/usr/local/cuda \
65 // RUN: | FileCheck -check-prefixes=CHECK,ARCH32,SM35 %s
66 // 32-bit compile when generating relocatable device code.
67 // RUN: %clang -### --target=i386-linux-gnu -fgpu-rdc -c %s 2>&1 \
68 // RUN: --no-offload-new-driver --offload-arch=sm_35 --cuda-path=%S/Inputs/CUDA/usr/local/cuda \
69 // RUN: | FileCheck -check-prefixes=CHECK,ARCH32,SM35,RDC %s
71 // Compile with -fintegrated-as. This should still cause us to invoke ptxas.
72 // RUN: %clang -### --target=x86_64-linux-gnu -fintegrated-as -c %s 2>&1 \
73 // RUN: --offload-arch=sm_35 --cuda-path=%S/Inputs/CUDA/usr/local/cuda \
74 // RUN: | FileCheck -check-prefixes=CHECK,ARCH64,SM35,OPT0 %s
75 // Check that we still pass -c when generating relocatable device code.
76 // RUN: %clang -### --target=x86_64-linux-gnu -fintegrated-as -fgpu-rdc -c %s 2>&1 \
77 // RUN: --no-offload-new-driver --offload-arch=sm_35 --cuda-path=%S/Inputs/CUDA/usr/local/cuda \
78 // RUN: | FileCheck -check-prefixes=CHECK,ARCH64,SM35,RDC %s
80 // Check -Xcuda-ptxas and -Xcuda-fatbinary
81 // RUN: %clang -### --target=x86_64-linux-gnu -c -Xcuda-ptxas -foo1 \
82 // RUN: --offload-arch=sm_35 --cuda-path=%S/Inputs/CUDA/usr/local/cuda \
83 // RUN: -Xcuda-fatbinary -bar1 -Xcuda-ptxas -foo2 -Xcuda-fatbinary -bar2 %s 2>&1 \
84 // RUN: | FileCheck -check-prefixes=CHECK,SM35,PTXAS-EXTRA,FATBINARY-EXTRA %s
86 // Check -Xcuda-ptxas with clang-cl
87 // RUN: %clang_cl -### -c -Xcuda-ptxas -foo1 \
88 // RUN: --offload-arch=sm_35 --cuda-path=%S/Inputs/CUDA/usr/local/cuda \
89 // RUN: -Xcuda-ptxas -foo2 -- %s 2>&1 \
90 // RUN: | FileCheck -check-prefixes=CHECK,SM35,PTXAS-EXTRA %s
93 // RUN: %clang -### --target=x86_64-apple-macosx -O0 -c %s 2>&1 \
94 // RUN: --offload-arch=sm_35 --cuda-path=%S/Inputs/CUDA/usr/local/cuda \
95 // RUN: | FileCheck -check-prefixes=CHECK,ARCH64,SM35,OPT0 %s
96 // RUN: %clang -### --target=x86_64-apple-macosx --cuda-gpu-arch=sm_35 -c %s 2>&1 \
97 // RUN: --offload-arch=sm_35 --cuda-path=%S/Inputs/CUDA/usr/local/cuda \
98 // RUN: | FileCheck -check-prefixes=CHECK,ARCH64,SM35 %s
99 // RUN: %clang -### --target=i386-apple-macosx -c %s 2>&1 \
100 // RUN: --offload-arch=sm_35 --cuda-path=%S/Inputs/CUDA/usr/local/cuda \
101 // RUN: | FileCheck -check-prefixes=CHECK,ARCH32,SM35 %s
103 // Check relocatable device code generation on MacOS.
104 // RUN: %clang -### --target=x86_64-apple-macosx -O0 -fgpu-rdc -c %s 2>&1 \
105 // RUN: --no-offload-new-driver --offload-arch=sm_35 --cuda-path=%S/Inputs/CUDA/usr/local/cuda \
106 // RUN: | FileCheck -check-prefixes=CHECK,ARCH64,SM35,RDC %s
107 // RUN: %clang -### --target=x86_64-apple-macosx --cuda-gpu-arch=sm_35 -fgpu-rdc -c %s 2>&1 \
108 // RUN: --no-offload-new-driver --offload-arch=sm_35 --cuda-path=%S/Inputs/CUDA/usr/local/cuda \
109 // RUN: | FileCheck -check-prefixes=CHECK,ARCH64,SM35,RDC %s
110 // RUN: %clang -### --target=i386-apple-macosx -fgpu-rdc -c %s 2>&1 \
111 // RUN: --no-offload-new-driver --offload-arch=sm_35 --cuda-path=%S/Inputs/CUDA/usr/local/cuda \
112 // RUN: | FileCheck -check-prefixes=CHECK,ARCH32,SM35,RDC %s
114 // Check that CLANG forwards the -v flag to PTXAS.
115 // RUN: %clang -### -save-temps -v %s 2>&1 \
116 // RUN: --offload-arch=sm_35 --cuda-path=%S/Inputs/CUDA/usr/local/cuda \
117 // RUN: | FileCheck -check-prefix=CHK-PTXAS-VERBOSE %s
119 // Match clang job that produces PTX assembly.
121 // ARCH64-SAME: "-triple" "nvptx64-nvidia-cuda"
122 // ARCH32-SAME: "-triple" "nvptx-nvidia-cuda"
123 // SM35-SAME: "-target-cpu" "sm_35"
124 // RDC-SAME: "-fgpu-rdc"
125 // CHECK-NOT: "-fgpu-rdc"
126 // SM35-SAME: "-o" "[[PTXFILE:[^"]*]]"
128 // Match the call to ptxas (which assembles PTX to SASS).
130 // ARCH64-SAME: "-m64"
131 // ARCH32-SAME: "-m32"
140 // DBG-SAME: "-g" "--dont-merge-basicblocks" "--return-at-end"
141 // SM35-SAME: "--gpu-name" "sm_35"
142 // SM35-SAME: "--output-file" "[[CUBINFILE:[^"]*]]"
143 // CHECK-SAME: "[[PTXFILE]]"
144 // PTXAS-EXTRA-SAME: "-foo1"
145 // PTXAS-EXTRA-SAME: "-foo2"
146 // CHECK-NOT: "-foo1"
147 // CHECK-NOT: "-foo2"
151 // Match the call to fatbinary (which combines all our PTX and SASS into one
154 // CHECK-SAME-DAG: "--cuda"
155 // ARCH64-SAME-DAG: "-64"
156 // ARCH32-SAME-DAG: "-32"
157 // CHECK-DAG: "--create" "[[FATBINARY:[^"]*]]"
158 // SM35-SAME-DAG: "--image=profile=compute_35,file=[[PTXFILE]]"
159 // SM35-SAME-DAG: "--image=profile=sm_35,file=[[CUBINFILE]]"
160 // FATBINARY-EXTRA-SAME: "-bar1"
161 // FATBINARY-EXTRA-SAME: "-bar2"
163 // Match the clang job for host compilation.
165 // ARCH64-SAME: "-triple" "x86_64-
166 // ARCH32-SAME: "-triple" "i386-
167 // CHECK-SAME: "-fcuda-include-gpubinary" "[[FATBINARY]]"
169 // CHK-PTXAS-VERBOSE: ptxas{{.*}}" "-v"