1*67e74705SXin Li // Tests that ptxas and fatbinary are correctly during CUDA compilation. 2*67e74705SXin Li // 3*67e74705SXin Li // REQUIRES: clang-driver 4*67e74705SXin Li // REQUIRES: x86-registered-target 5*67e74705SXin Li // REQUIRES: nvptx-registered-target 6*67e74705SXin Li 7*67e74705SXin Li // Regular compiles with -O{0,1,2,3,4,fast}. -O4 and -Ofast map to ptxas O3. 8*67e74705SXin Li // RUN: %clang -### -target x86_64-linux-gnu -O0 -c %s 2>&1 \ 9*67e74705SXin Li // RUN: | FileCheck -check-prefix ARCH64 -check-prefix SM20 -check-prefix OPT0 %s 10*67e74705SXin Li // RUN: %clang -### -target x86_64-linux-gnu -O1 -c %s 2>&1 \ 11*67e74705SXin Li // RUN: | FileCheck -check-prefix ARCH64 -check-prefix SM20 -check-prefix OPT1 %s 12*67e74705SXin Li // RUN: %clang -### -target x86_64-linux-gnu -O2 -c %s 2>&1 \ 13*67e74705SXin Li // RUN: | FileCheck -check-prefix ARCH64 -check-prefix SM20 -check-prefix OPT2 %s 14*67e74705SXin Li // RUN: %clang -### -target x86_64-linux-gnu -O3 -c %s 2>&1 \ 15*67e74705SXin Li // RUN: | FileCheck -check-prefix ARCH64 -check-prefix SM20 -check-prefix OPT3 %s 16*67e74705SXin Li // RUN: %clang -### -target x86_64-linux-gnu -O4 -c %s 2>&1 \ 17*67e74705SXin Li // RUN: | FileCheck -check-prefix ARCH64 -check-prefix SM20 -check-prefix OPT3 %s 18*67e74705SXin Li // RUN: %clang -### -target x86_64-linux-gnu -Ofast -c %s 2>&1 \ 19*67e74705SXin Li // RUN: | FileCheck -check-prefix ARCH64 -check-prefix SM20 -check-prefix OPT3 %s 20*67e74705SXin Li 21*67e74705SXin Li // With debugging enabled, ptxas should be run with with no ptxas optimizations. 22*67e74705SXin Li // RUN: %clang -### -target x86_64-linux-gnu --cuda-noopt-device-debug -O2 -c %s 2>&1 \ 23*67e74705SXin Li // RUN: | FileCheck -check-prefix ARCH64 -check-prefix SM20 -check-prefix DBG %s 24*67e74705SXin Li 25*67e74705SXin Li // --no-cuda-noopt-device-debug overrides --cuda-noopt-device-debug. 26*67e74705SXin Li // RUN: %clang -### -target x86_64-linux-gnu --cuda-noopt-debug \ 27*67e74705SXin Li // RUN: --no-cuda-noopt-debug -O2 -c %s 2>&1 \ 28*67e74705SXin Li // RUN: | FileCheck -check-prefix ARCH64 -check-prefix SM20 -check-prefix OPT2 %s 29*67e74705SXin Li 30*67e74705SXin Li // Regular compile without -O. This should result in us passing -O0 to ptxas. 31*67e74705SXin Li // RUN: %clang -### -target x86_64-linux-gnu -c %s 2>&1 \ 32*67e74705SXin Li // RUN: | FileCheck -check-prefix ARCH64 -check-prefix SM20 -check-prefix OPT0 %s 33*67e74705SXin Li 34*67e74705SXin Li // Regular compiles with -Os and -Oz. For lack of a better option, we map 35*67e74705SXin Li // these to ptxas -O3. 36*67e74705SXin Li // RUN: %clang -### -target x86_64-linux-gnu -Os -c %s 2>&1 \ 37*67e74705SXin Li // RUN: | FileCheck -check-prefix ARCH64 -check-prefix SM20 -check-prefix OPT2 %s 38*67e74705SXin Li // RUN: %clang -### -target x86_64-linux-gnu -Oz -c %s 2>&1 \ 39*67e74705SXin Li // RUN: | FileCheck -check-prefix ARCH64 -check-prefix SM20 -check-prefix OPT2 %s 40*67e74705SXin Li 41*67e74705SXin Li // Regular compile targeting sm_35. 42*67e74705SXin Li // RUN: %clang -### -target x86_64-linux-gnu --cuda-gpu-arch=sm_35 -c %s 2>&1 \ 43*67e74705SXin Li // RUN: | FileCheck -check-prefix ARCH64 -check-prefix SM35 %s 44*67e74705SXin Li 45*67e74705SXin Li // 32-bit compile. 46*67e74705SXin Li // RUN: %clang -### -target x86_32-linux-gnu -c %s 2>&1 \ 47*67e74705SXin Li // RUN: | FileCheck -check-prefix ARCH32 -check-prefix SM20 %s 48*67e74705SXin Li 49*67e74705SXin Li // Compile with -fintegrated-as. This should still cause us to invoke ptxas. 50*67e74705SXin Li // RUN: %clang -### -target x86_64-linux-gnu -fintegrated-as -c %s 2>&1 \ 51*67e74705SXin Li // RUN: | FileCheck -check-prefix ARCH64 -check-prefix SM20 -check-prefix OPT0 %s 52*67e74705SXin Li 53*67e74705SXin Li // Check -Xcuda-ptxas and -Xcuda-fatbinary 54*67e74705SXin Li // RUN: %clang -### -target x86_64-linux-gnu -c -Xcuda-ptxas -foo1 \ 55*67e74705SXin Li // RUN: -Xcuda-fatbinary -bar1 -Xcuda-ptxas -foo2 -Xcuda-fatbinary -bar2 %s 2>&1 \ 56*67e74705SXin Li // RUN: | FileCheck -check-prefix SM20 -check-prefix PTXAS-EXTRA \ 57*67e74705SXin Li // RUN: -check-prefix FATBINARY-EXTRA %s 58*67e74705SXin Li 59*67e74705SXin Li // Match clang job that produces PTX assembly. 60*67e74705SXin Li // CHECK: "-cc1" "-triple" "nvptx64-nvidia-cuda" 61*67e74705SXin Li // SM20: "-target-cpu" "sm_20" 62*67e74705SXin Li // SM35: "-target-cpu" "sm_35" 63*67e74705SXin Li // SM20: "-o" "[[PTXFILE:[^"]*]]" 64*67e74705SXin Li // SM35: "-o" "[[PTXFILE:[^"]*]]" 65*67e74705SXin Li 66*67e74705SXin Li // Match the call to ptxas (which assembles PTX to SASS). 67*67e74705SXin Li // CHECK:ptxas 68*67e74705SXin Li // ARCH64: "-m64" 69*67e74705SXin Li // ARCH32: "-m32" 70*67e74705SXin Li // OPT0: "-O0" 71*67e74705SXin Li // OPT0-NOT: "-g" 72*67e74705SXin Li // OPT1: "-O1" 73*67e74705SXin Li // OPT1-NOT: "-g" 74*67e74705SXin Li // OPT2: "-O2" 75*67e74705SXin Li // OPT2-NOT: "-g" 76*67e74705SXin Li // OPT3: "-O3" 77*67e74705SXin Li // OPT3-NOT: "-g" 78*67e74705SXin Li // DBG: "-g" "--dont-merge-basicblocks" "--return-at-end" 79*67e74705SXin Li // SM20: "--gpu-name" "sm_20" 80*67e74705SXin Li // SM35: "--gpu-name" "sm_35" 81*67e74705SXin Li // SM20: "--output-file" "[[CUBINFILE:[^"]*]]" 82*67e74705SXin Li // SM35: "--output-file" "[[CUBINFILE:[^"]*]]" 83*67e74705SXin Li // PTXAS-EXTRA: "-foo1" 84*67e74705SXin Li // PTXAS-EXTRA-SAME: "-foo2" 85*67e74705SXin Li // CHECK-SAME: "[[PTXFILE]]" 86*67e74705SXin Li 87*67e74705SXin Li // Match the call to fatbinary (which combines all our PTX and SASS into one 88*67e74705SXin Li // blob). 89*67e74705SXin Li // CHECK:fatbinary 90*67e74705SXin Li // CHECK-DAG: "--cuda" 91*67e74705SXin Li // ARCH64-DAG: "-64" 92*67e74705SXin Li // ARCH32-DAG: "-32" 93*67e74705SXin Li // CHECK-DAG: "--create" "[[FATBINARY:[^"]*]]" 94*67e74705SXin Li // SM20-DAG: "--image=profile=compute_20,file=[[PTXFILE]]" 95*67e74705SXin Li // SM35-DAG: "--image=profile=compute_35,file=[[PTXFILE]]" 96*67e74705SXin Li // SM20-DAG: "--image=profile=sm_20,file=[[CUBINFILE]]" 97*67e74705SXin Li // SM35-DAG: "--image=profile=sm_35,file=[[CUBINFILE]]" 98*67e74705SXin Li // FATBINARY-EXTRA: "-bar1" 99*67e74705SXin Li // FATBINARY-EXTRA-SAME: "-bar2" 100*67e74705SXin Li 101*67e74705SXin Li // Match the clang job for host compilation. 102*67e74705SXin Li // CHECK: "-cc1" "-triple" "x86_64--linux-gnu" 103*67e74705SXin Li // CHECK-SAME: "-fcuda-include-gpubinary" "[[FATBINARY]]" 104