1 // Tests that ptxas and fatbinary are correctly during CUDA compilation. 2 // 3 // REQUIRES: clang-driver 4 // REQUIRES: x86-registered-target 5 // REQUIRES: nvptx-registered-target 6 7 // Regular compiles with -O{0,1,2,3,4,fast}. -O4 and -Ofast map to ptxas O3. 8 // RUN: %clang -### -target x86_64-linux-gnu -O0 -c %s 2>&1 \ 9 // RUN: | FileCheck -check-prefix ARCH64 -check-prefix SM20 -check-prefix OPT0 %s 10 // RUN: %clang -### -target x86_64-linux-gnu -O1 -c %s 2>&1 \ 11 // RUN: | FileCheck -check-prefix ARCH64 -check-prefix SM20 -check-prefix OPT1 %s 12 // RUN: %clang -### -target x86_64-linux-gnu -O2 -c %s 2>&1 \ 13 // RUN: | FileCheck -check-prefix ARCH64 -check-prefix SM20 -check-prefix OPT2 %s 14 // RUN: %clang -### -target x86_64-linux-gnu -O3 -c %s 2>&1 \ 15 // RUN: | FileCheck -check-prefix ARCH64 -check-prefix SM20 -check-prefix OPT3 %s 16 // RUN: %clang -### -target x86_64-linux-gnu -O4 -c %s 2>&1 \ 17 // RUN: | FileCheck -check-prefix ARCH64 -check-prefix SM20 -check-prefix OPT3 %s 18 // RUN: %clang -### -target x86_64-linux-gnu -Ofast -c %s 2>&1 \ 19 // RUN: | FileCheck -check-prefix ARCH64 -check-prefix SM20 -check-prefix OPT3 %s 20 21 // With debugging enabled, ptxas should be run with with no ptxas optimizations. 22 // RUN: %clang -### -target x86_64-linux-gnu --cuda-noopt-device-debug -O2 -c %s 2>&1 \ 23 // RUN: | FileCheck -check-prefix ARCH64 -check-prefix SM20 -check-prefix DBG %s 24 25 // --no-cuda-noopt-device-debug overrides --cuda-noopt-device-debug. 26 // RUN: %clang -### -target x86_64-linux-gnu --cuda-noopt-debug \ 27 // RUN: --no-cuda-noopt-debug -O2 -c %s 2>&1 \ 28 // RUN: | FileCheck -check-prefix ARCH64 -check-prefix SM20 -check-prefix OPT2 %s 29 30 // Regular compile without -O. This should result in us passing -O0 to ptxas. 31 // RUN: %clang -### -target x86_64-linux-gnu -c %s 2>&1 \ 32 // RUN: | FileCheck -check-prefix ARCH64 -check-prefix SM20 -check-prefix OPT0 %s 33 34 // Regular compiles with -Os and -Oz. For lack of a better option, we map 35 // these to ptxas -O3. 36 // RUN: %clang -### -target x86_64-linux-gnu -Os -c %s 2>&1 \ 37 // RUN: | FileCheck -check-prefix ARCH64 -check-prefix SM20 -check-prefix OPT2 %s 38 // RUN: %clang -### -target x86_64-linux-gnu -Oz -c %s 2>&1 \ 39 // RUN: | FileCheck -check-prefix ARCH64 -check-prefix SM20 -check-prefix OPT2 %s 40 41 // Regular compile targeting sm_35. 42 // RUN: %clang -### -target x86_64-linux-gnu --cuda-gpu-arch=sm_35 -c %s 2>&1 \ 43 // RUN: | FileCheck -check-prefix ARCH64 -check-prefix SM35 %s 44 45 // 32-bit compile. 46 // RUN: %clang -### -target x86_32-linux-gnu -c %s 2>&1 \ 47 // RUN: | FileCheck -check-prefix ARCH32 -check-prefix SM20 %s 48 49 // Compile with -fintegrated-as. This should still cause us to invoke ptxas. 50 // RUN: %clang -### -target x86_64-linux-gnu -fintegrated-as -c %s 2>&1 \ 51 // RUN: | FileCheck -check-prefix ARCH64 -check-prefix SM20 -check-prefix OPT0 %s 52 53 // Check -Xcuda-ptxas and -Xcuda-fatbinary 54 // RUN: %clang -### -target x86_64-linux-gnu -c -Xcuda-ptxas -foo1 \ 55 // RUN: -Xcuda-fatbinary -bar1 -Xcuda-ptxas -foo2 -Xcuda-fatbinary -bar2 %s 2>&1 \ 56 // RUN: | FileCheck -check-prefix SM20 -check-prefix PTXAS-EXTRA \ 57 // RUN: -check-prefix FATBINARY-EXTRA %s 58 59 // Match clang job that produces PTX assembly. 60 // CHECK: "-cc1" "-triple" "nvptx64-nvidia-cuda" 61 // SM20: "-target-cpu" "sm_20" 62 // SM35: "-target-cpu" "sm_35" 63 // SM20: "-o" "[[PTXFILE:[^"]*]]" 64 // SM35: "-o" "[[PTXFILE:[^"]*]]" 65 66 // Match the call to ptxas (which assembles PTX to SASS). 67 // CHECK:ptxas 68 // ARCH64: "-m64" 69 // ARCH32: "-m32" 70 // OPT0: "-O0" 71 // OPT0-NOT: "-g" 72 // OPT1: "-O1" 73 // OPT1-NOT: "-g" 74 // OPT2: "-O2" 75 // OPT2-NOT: "-g" 76 // OPT3: "-O3" 77 // OPT3-NOT: "-g" 78 // DBG: "-g" "--dont-merge-basicblocks" "--return-at-end" 79 // SM20: "--gpu-name" "sm_20" 80 // SM35: "--gpu-name" "sm_35" 81 // SM20: "--output-file" "[[CUBINFILE:[^"]*]]" 82 // SM35: "--output-file" "[[CUBINFILE:[^"]*]]" 83 // PTXAS-EXTRA: "-foo1" 84 // PTXAS-EXTRA-SAME: "-foo2" 85 // CHECK-SAME: "[[PTXFILE]]" 86 87 // Match the call to fatbinary (which combines all our PTX and SASS into one 88 // blob). 89 // CHECK:fatbinary 90 // CHECK-DAG: "--cuda" 91 // ARCH64-DAG: "-64" 92 // ARCH32-DAG: "-32" 93 // CHECK-DAG: "--create" "[[FATBINARY:[^"]*]]" 94 // SM20-DAG: "--image=profile=compute_20,file=[[PTXFILE]]" 95 // SM35-DAG: "--image=profile=compute_35,file=[[PTXFILE]]" 96 // SM20-DAG: "--image=profile=sm_20,file=[[CUBINFILE]]" 97 // SM35-DAG: "--image=profile=sm_35,file=[[CUBINFILE]]" 98 // FATBINARY-EXTRA: "-bar1" 99 // FATBINARY-EXTRA-SAME: "-bar2" 100 101 // Match the clang job for host compilation. 102 // CHECK: "-cc1" "-triple" "x86_64--linux-gnu" 103 // CHECK-SAME: "-fcuda-include-gpubinary" "[[FATBINARY]]" 104