clang/test/Driver/hip-options.hip

   1 // REQUIRES: x86-registered-target
   2 // REQUIRES: amdgpu-registered-target
   3
   4 // RUN: %clang -### -x hip --gpu-max-threads-per-block=1024 %s 2>&1 | FileCheck %s
   5
   6 // Check that there are commands for both host- and device-side compilations.
   7 //
   8 // CHECK: "-cc1" {{.*}} "-fcuda-is-device"
   9 // CHECK-SAME: "--gpu-max-threads-per-block=1024"
  10
  11 // RUN: %clang -### -nogpuinc -nogpulib -fgpu-allow-device-init \
  12 // RUN:   %s 2>&1 | FileCheck -check-prefix=DEVINIT %s
  13 // DEVINIT: "-cc1" {{.*}}"-fgpu-allow-device-init"
  14 // DEVINIT: "-cc1" {{.*}}"-fgpu-allow-device-init"
  15
  16 // Check -fgpu-default-stream=per-thread.
  17 // RUN: %clang -### -nogpuinc -nogpulib -fgpu-default-stream=per-thread \
  18 // RUN:   %s -save-temps 2>&1 | FileCheck -check-prefix=PTH %s
  19 // PTH: "-cc1"{{.*}} "-E" {{.*}}"-fgpu-default-stream=per-thread"
  20 // PTH: "-cc1"{{.*}} "-fgpu-default-stream=per-thread" {{.*}}"-x" "hip-cpp-output"
  21 // PTH: "-cc1"{{.*}} "-E" {{.*}}"-fgpu-default-stream=per-thread"
  22 // PTH: "-cc1"{{.*}} "-fgpu-default-stream=per-thread" {{.*}}"-x" "hip-cpp-output"
  23
  24 // RUN: %clang -### -x hip --target=x86_64-pc-windows-msvc -fms-extensions \
  25 // RUN:   -mllvm -amdgpu-early-inline-all=true  %s 2>&1 | \
  26 // RUN:   FileCheck -check-prefix=MLLVM %s
  27 // MLLVM-NOT: "-mllvm"{{.*}}"-amdgpu-early-inline-all=true"{{.*}}"-mllvm"{{.*}}"-amdgpu-early-inline-all=true"
  28
  29 // RUN: %clang -### -Xarch_device -g -nogpulib --cuda-gpu-arch=gfx900 \
  30 // RUN:   -Xarch_device -fcf-protection=branch \
  31 // RUN:   --cuda-gpu-arch=gfx906  %s 2>&1 | FileCheck -check-prefix=DEV %s
  32 // DEV: "-cc1"{{.*}} "-fcuda-is-device" {{.*}} "-debug-info-kind={{.*}}" {{.*}} "-fcf-protection=branch"
  33 // DEV: "-cc1"{{.*}} "-fcuda-is-device" {{.*}} "-debug-info-kind={{.*}}" {{.*}} "-fcf-protection=branch"
  34 // DEV-NOT: clang{{.*}} {{.*}} "-debug-info-kind={{.*}}"
  35
  36 // RUN: %clang -### -Xarch_host -g -nogpulib --cuda-gpu-arch=gfx900 \
  37 // RUN:   --cuda-gpu-arch=gfx906  %s 2>&1 | FileCheck -check-prefix=HOST %s
  38 // HOST-NOT: "-fcuda-is-device" {{.*}} "-debug-info-kind={{.*}}"
  39 // HOST-NOT: "-fcuda-is-device" {{.*}} "-debug-info-kind={{.*}}"
  40 // HOST: clang{{.*}} "-debug-info-kind={{.*}}"
  41
  42 // RUN: %clang -### -nogpuinc -nogpulib -munsafe-fp-atomics \
  43 // RUN:   --cuda-gpu-arch=gfx906  %s 2>&1 | FileCheck -check-prefix=UNSAFE-FP-ATOMICS %s
  44 // UNSAFE-FP-ATOMICS: clang{{.*}} "-triple" "amdgcn-amd-amdhsa" {{.*}} "-munsafe-fp-atomics"
  45
  46 // RUN: %clang -### -nogpuinc -nogpulib \
  47 // RUN:   --cuda-gpu-arch=gfx906  %s 2>&1 | FileCheck -check-prefix=DEFAULT-UNSAFE-FP-ATOMICS %s
  48 // DEFAULT-UNSAFE-FP-ATOMICS-NOT: clang{{.*}} "-triple" "amdgcn-amd-amdhsa" {{.*}} "-munsafe-fp-atomics"
  49
  50 // RUN: %clang -### --target=x86_64-unknown-linux-gnu -nogpuinc -nogpulib -fgpu-exclude-wrong-side-overloads \
  51 // RUN:   --cuda-gpu-arch=gfx906  %s 2>&1 | FileCheck -check-prefix=FIX-OVERLOAD %s
  52 // FIX-OVERLOAD: "-cc1"{{.*}} "-triple" "amdgcn-amd-amdhsa" {{.*}} "-fgpu-exclude-wrong-side-overloads" "-fgpu-defer-diag"
  53 // FIX-OVERLOAD: "-cc1"{{.*}} "-triple" "x86_64-unknown-linux-gnu" {{.*}} "-fgpu-exclude-wrong-side-overloads" "-fgpu-defer-diag"
  54
  55 // Check -mconstructor-aliases is not passed to device compilation.
  56
  57 // RUN: %clang -### --target=x86_64-unknown-linux-gnu -nogpuinc -nogpulib \
  58 // RUN:   --cuda-gpu-arch=gfx906  %s 2>&1 | FileCheck -check-prefix=CTA %s
  59 // CTA: "-cc1"{{.*}} "-triple" "x86_64-unknown-linux-gnu" {{.*}} "-mconstructor-aliases"
  60 // CTA-NOT: "-cc1"{{.*}} "-triple" "amdgcn-amd-amdhsa" {{.*}} "-mconstructor-aliases"
  61
  62 // RUN: %clang -### --target=x86_64-unknown-linux-gnu -nogpuinc -nogpulib \
  63 // RUN:   --offload-arch=gfx906 -fgpu-inline-threshold=1000 %s 2>&1 | FileCheck -check-prefix=THRESH %s
  64 // THRESH: "-cc1"{{.*}} "-triple" "amdgcn-amd-amdhsa" {{.*}} "-mllvm" "-inline-threshold=1000"
  65 // THRESH-NOT: "-cc1"{{.*}} "-triple" "x86_64-unknown-linux-gnu" {{.*}} "-inline-threshold=1000"
  66
  67 // Check -foffload-lto=thin translated correctly.
  68
  69 // RUN: %clang -### --target=x86_64-unknown-linux-gnu -nogpuinc -nogpulib \
  70 // RUN:   --cuda-gpu-arch=gfx906 -foffload-lto=thin -fwhole-program-vtables %s 2>&1 \
  71 // RUN:   | FileCheck -check-prefix=HIPTHINLTO %s
  72
  73 // RUN: %clang -### --target=x86_64-unknown-linux-gnu -nogpuinc -nogpulib \
  74 // RUN:   --cuda-gpu-arch=gfx906 -fgpu-rdc -foffload-lto=thin -fwhole-program-vtables %s 2>&1 \
  75 // RUN:   | FileCheck -check-prefix=HIPTHINLTO %s
  76
  77 // Ensure we don't error about -fwhole-program-vtables for the non-device offload compile.
  78 // HIPTHINLTO-NOT: error: invalid argument '-fwhole-program-vtables' only allowed with '-flto'
  79 // HIPTHINLTO-NOT: "-cc1"{{.*}} "-triple" "x86_64-unknown-linux-gnu" {{.*}} "-flto-unit"
  80 // HIPTHINLTO: "-cc1"{{.*}} "-triple" "amdgcn-amd-amdhsa" {{.*}} "-flto=thin" "-flto-unit" {{.*}} "-fwhole-program-vtables"
  81 // HIPTHINLTO-NOT: "-cc1"{{.*}} "-triple" "x86_64-unknown-linux-gnu" {{.*}} "-flto-unit"
  82 // HIPTHINLTO: lld{{.*}}"-plugin-opt=mcpu=gfx906" "-plugin-opt=thinlto" "-plugin-opt=-force-import-all"
  83
  84 // Check that -flto=thin is handled correctly, particularly with -fwhole-program-vtables.
  85 //
  86 // RUN: %clang -### --target=x86_64-unknown-linux-gnu -nogpuinc -nogpulib \
  87 // RUN:   --cuda-gpu-arch=gfx906 -flto=thin -fwhole-program-vtables %s 2>&1 \
  88 // RUN:   | FileCheck -check-prefix=THINLTO %s
  89
  90 // Ensure we don't error about -fwhole-program-vtables for the device offload compile. We should
  91 // drop -fwhole-program-vtables for the device offload compile and pass it through for the
  92 // non-device offload compile along with -flto=thin.
  93 // THINLTO-NOT: error: invalid argument '-fwhole-program-vtables' only allowed with '-flto'
  94 // THINLTO-NOT: "-cc1"{{.*}}" "-triple" "amdgcn-amd-amdhsa" {{.*}} "-fwhole-program-vtables"
  95 // THINLTO: "-cc1"{{.*}} "-triple" "x86_64-unknown-linux-gnu" {{.*}} "-flto=thin" {{.*}} "-fwhole-program-vtables"
  96 // THINLTO-NOT: "-cc1"{{.*}}" "-triple" "amdgcn-amd-amdhsa" {{.*}} "-fwhole-program-vtables"
  97
  98 // Check -fopenmp is allowed with HIP but -fopenmp-targets= is not allowed.
  99
 100 // RUN: %clang -### --target=x86_64-unknown-linux-gnu -nogpuinc -nogpulib \
 101 // RUN:   --offload-arch=gfx906 -fopenmp %s 2>&1 | FileCheck -check-prefix=OMP %s
 102 // OMP-NOT: "-cc1"{{.*}} "-triple" "amdgcn-amd-amdhsa" {{.*}} "-fopenmp"
 103 // OMP: "-cc1"{{.*}} "-triple" "x86_64-unknown-linux-gnu" {{.*}} "-fopenmp"
 104
 105 // RUN: not %clang --target=x86_64-unknown-linux-gnu -nogpuinc -nogpulib \
 106 // RUN:   --offload-arch=gfx906 -fopenmp -fopenmp-targets=amdgcn %s 2>&1 \
 107 // RUN:   | FileCheck -check-prefix=OMPTGT %s
 108 // OMPTGT: unsupported option '-fopenmp-targets=' for language mode 'HIP'
 109
 110 // Check -Xoffload-linker option is passed to lld.
 111
 112 // RUN: %clang -### --target=x86_64-unknown-linux-gnu -nogpuinc -nogpulib \
 113 // RUN:   --cuda-gpu-arch=gfx906 -fgpu-rdc -Xoffload-linker --build-id=md5 %s 2>&1 \
 114 // RUN:   | FileCheck -check-prefix=OFL-LINK %s
 115 // RUN: %clang -### --target=x86_64-unknown-linux-gnu -nogpuinc -nogpulib \
 116 // RUN:   --cuda-gpu-arch=gfx906 -Xoffload-linker --build-id=md5 %s 2>&1 \
 117 // RUN:   | FileCheck -check-prefix=OFL-LINK %s
 118 // OFL-LINK: lld{{.*}}"--build-id=md5"
 119
 120 // RUN: %clang -### --target=x86_64-unknown-linux-gnu -nogpuinc -nogpulib \
 121 // RUN:   --offload-arch=gfx906 -fhip-kernel-arg-name %s 2>&1 \
 122 // RUN:   | FileCheck -check-prefix=KAN %s
 123 // KAN: "-cc1"{{.*}} "-triple" "amdgcn-amd-amdhsa" {{.*}} "-fhip-kernel-arg-name"
 124
 125 // RUN: %clang -### --target=x86_64-unknown-linux-gnu -nogpuinc -nogpulib \
 126 // RUN:   --offload-arch=gfx906 %s 2>&1 \
 127 // RUN:   | FileCheck -check-prefix=KANNEG %s
 128 // KANNEG-NOT: "-fhip-kernel-arg-name"