223 lines · plain
1// Tests the bindings generated for a CUDA offloading target for different2// combinations of:3// - Number of gpu architectures;4// - Host/device-only compilation;5// - User-requested final phase - binary or assembly.6// It parallels cuda-phases.cu test, but verifies whether output file is temporary or not.7 8// It's hard to check whether file name is temporary in a portable9// way. Instead we check whether we've generated a permanent name on10// device side, which appends '-device-cuda-<triple>' suffix.11 12//13// Test single gpu architecture with complete compilation.14// No intermediary device files should have "-device-cuda..." in the name.15//16// RUN: %clang -target powerpc64le-ibm-linux-gnu -ccc-print-bindings --cuda-gpu-arch=sm_30 %s 2>&1 \17// RUN: | FileCheck -check-prefix=BIN %s18// BIN: # "nvptx64-nvidia-cuda" - "clang",{{.*}} output:19// BIN-NOT: cuda-bindings-device-cuda-nvptx6420// BIN: # "nvptx64-nvidia-cuda" - "NVPTX::Assembler",{{.*}} output:21// BIN-NOT: cuda-bindings-device-cuda-nvptx6422// BIN: # "nvptx64-nvidia-cuda" - "NVPTX::Linker",{{.*}} output:23// BIN-NOT: cuda-bindings-device-cuda-nvptx6424// BIN: # "powerpc64le-ibm-linux-gnu" - "clang",{{.*}} output:25// BIN-NOT: cuda-bindings-device-cuda-nvptx6426// BIN: # "powerpc64le-ibm-linux-gnu" - "Offload::Linker", inputs:{{.*}}, output: "a.out"27 28//29// Test single gpu architecture up to the assemble phase.30//31// RUN: %clang -target powerpc64le-ibm-linux-gnu -ccc-print-bindings --cuda-gpu-arch=sm_30 %s -S 2>&1 \32// RUN: | FileCheck -check-prefix=ASM %s33// ASM-DAG: # "nvptx64-nvidia-cuda" - "clang",{{.*}} output: "[[PTX:.+]].s"34// ASM-DAG: # "powerpc64le-ibm-linux-gnu" - "clang",{{.*}} output: "cuda-bindings.s"35 36//37// Test two gpu architectures with complete compilation.38//39// RUN: %clang -target powerpc64le-ibm-linux-gnu -ccc-print-bindings \40// RUN: --cuda-gpu-arch=sm_30 --cuda-gpu-arch=sm_35 %s 2>&1 \41// RUN: | FileCheck -check-prefixes=BIN2,AOUT %s42// RUN: %clang -target powerpc64le-ibm-linux-gnu -ccc-print-bindings \43// RUN: --offload-arch=sm_30,sm_35 %s 2>&1 \44// RUN: | FileCheck -check-prefixes=BIN2,AOUT %s45// .. same, but with explicitly specified output.46// RUN: %clang -target powerpc64le-ibm-linux-gnu -ccc-print-bindings \47// RUN: --cuda-gpu-arch=sm_30 --cuda-gpu-arch=sm_35 %s -o %t/out 2>&1 \48// RUN: | FileCheck -check-prefixes=BIN2,TOUT %s49// RUN: %clang -target powerpc64le-ibm-linux-gnu -ccc-print-bindings \50// RUN: --offload-arch=sm_30,sm_35 %s -o %t/out 2>&1 \51// RUN: | FileCheck -check-prefixes=BIN2,TOUT %s52// BIN2: # "nvptx64-nvidia-cuda" - "clang",{{.*}} output:53// BIN2-NOT: cuda-bindings-device-cuda-nvptx6454// BIN2: # "nvptx64-nvidia-cuda" - "NVPTX::Assembler",{{.*}} output:55// BIN2-NOT: cuda-bindings-device-cuda-nvptx6456// BIN2: # "nvptx64-nvidia-cuda" - "clang",{{.*}} output:57// BIN2-NOT: cuda-bindings-device-cuda-nvptx6458// BIN2: # "nvptx64-nvidia-cuda" - "NVPTX::Assembler",{{.*}} output:59// BIN2-NOT: cuda-bindings-device-cuda-nvptx6460// BIN2: # "nvptx64-nvidia-cuda" - "NVPTX::Linker",{{.*}} output:61// BIN2-NOT: cuda-bindings-device-cuda-nvptx6462// BIN2: # "powerpc64le-ibm-linux-gnu" - "clang",{{.*}} output:63// BIN2-NOT: cuda-bindings-device-cuda-nvptx6464// AOUT: # "powerpc64le-ibm-linux-gnu" - "Offload::Linker", inputs:{{.*}}, output: "a.out"65// TOUT: # "powerpc64le-ibm-linux-gnu" - "Offload::Linker", inputs:{{.*}}, output: "{{.*}}/out"66 67// .. same, but with -fsyntax-only68// RUN: %clang -target powerpc64le-ibm-linux-gnu -ccc-print-bindings -fsyntax-only \69// RUN: --cuda-gpu-arch=sm_30 --cuda-gpu-arch=sm_35 %s 2>&1 \70// RUN: | FileCheck -check-prefix=NDSYN %s71// RUN: %clang -target powerpc64le-ibm-linux-gnu -ccc-print-bindings -fsyntax-only \72// RUN: --offload-arch=sm_30,sm_35 %s -o %t/out 2>&1 \73// RUN: | FileCheck -check-prefix=NDSYN %s74// RUN: %clang -target powerpc64le-ibm-linux-gnu -ccc-print-bindings -fsyntax-only \75// RUN: --cuda-gpu-arch=sm_30 --cuda-gpu-arch=sm_35 %s 2>&1 \76// RUN: | FileCheck -check-prefix=NDSYN %s77// RUN: %clang -target powerpc64le-ibm-linux-gnu -ccc-print-bindings -fsyntax-only \78// RUN: --offload-arch=sm_30,sm_35 %s -o %t/out 2>&1 \79// RUN: | FileCheck -check-prefix=NDSYN %s80// NDSYN-NOT: inputs:81// NDSYN: # "nvptx64-nvidia-cuda" - "clang", inputs: [{{.*}}], output: (nothing)82// NDSYN-NEXT: # "nvptx64-nvidia-cuda" - "clang", inputs: [{{.*}}], output: (nothing)83// NDSYN-NEXT: # "powerpc64le-ibm-linux-gnu" - "clang", inputs: [{{.*}}], output: (nothing)84// NDSYN-NOT: inputs:85 86 87//88// Test two gpu architectures up to the assemble phase.89//90// RUN: %clang -target powerpc64le-ibm-linux-gnu -ccc-print-bindings \91// RUN: --cuda-gpu-arch=sm_30 --cuda-gpu-arch=sm_35 %s -S 2>&1 \92// RUN: | FileCheck -check-prefix=ASM2 %s93// ASM2-DAG: # "nvptx64-nvidia-cuda" - "clang",{{.*}} output: "[[SM30:.+]].s"94// ASM2-DAG: # "nvptx64-nvidia-cuda" - "clang",{{.*}} output: "[[SM35:.+]].s"95// ASM2-DAG: # "powerpc64le-ibm-linux-gnu" - "clang",{{.*}} output: "cuda-bindings.s"96 97//98// Test one or more gpu architecture with complete compilation in host-only99// compilation mode.100//101// RUN: %clang -target powerpc64le-ibm-linux-gnu -ccc-print-bindings \102// RUN: --cuda-gpu-arch=sm_30 %s --cuda-host-only 2>&1 \103// RUN: | FileCheck -check-prefix=HBIN %s104// RUN: %clang -target powerpc64le-ibm-linux-gnu -ccc-print-bindings \105// RUN: --cuda-gpu-arch=sm_30 --cuda-gpu-arch=sm_35 %s --cuda-host-only 2>&1 \106// RUN: | FileCheck -check-prefix=HBIN %s107// HBIN: # "powerpc64le-ibm-linux-gnu" - "clang",{{.*}} output:108// HBIN-NOT: cuda-bindings-device-cuda-nvptx64109// HBIN: # "powerpc64le-ibm-linux-gnu" - "Offload::Linker", inputs:{{.*}}, output: "a.out"110 111//112// Test one or more gpu architecture up to the assemble phase in host-only113// compilation mode.114//115// RUN: %clang -target powerpc64le-ibm-linux-gnu -ccc-print-bindings \116// RUN: --cuda-gpu-arch=sm_30 %s --cuda-host-only -S 2>&1 \117// RUN: | FileCheck -check-prefix=HASM %s118// RUN: %clang -target powerpc64le-ibm-linux-gnu -ccc-print-bindings \119// RUN: --cuda-gpu-arch=sm_30 --cuda-gpu-arch=sm_35 %s --cuda-host-only -S 2>&1 \120// RUN: | FileCheck -check-prefix=HASM %s121// HASM: # "powerpc64le-ibm-linux-gnu" - "clang",{{.*}} output: "cuda-bindings.s"122 123//124// Test single gpu architecture with complete compilation in device-only125// compilation mode.126//127// RUN: %clang -target powerpc64le-ibm-linux-gnu -ccc-print-bindings \128// RUN: --cuda-gpu-arch=sm_30 %s --cuda-device-only 2>&1 \129// RUN: | FileCheck -check-prefix=DBIN %s130// DBIN: # "nvptx64-nvidia-cuda" - "clang",{{.*}} output:131// DBIN-NOT: cuda-bindings-device-cuda-nvptx64132// DBIN: # "nvptx64-nvidia-cuda" - "NVPTX::Assembler",{{.*}} output: "cuda-bindings-cuda-nvptx64-nvidia-cuda-sm_30.o"133 134//135// Test single gpu architecture up to the assemble phase in device-only136// compilation mode.137//138// RUN: %clang -target powerpc64le-ibm-linux-gnu -ccc-print-bindings \139// RUN: --cuda-gpu-arch=sm_30 %s --cuda-device-only -S 2>&1 \140// RUN: | FileCheck -check-prefix=DASM %s141// DASM: # "nvptx64-nvidia-cuda" - "clang",{{.*}} output: "cuda-bindings-cuda-nvptx64-nvidia-cuda-sm_30.s"142 143//144// Test two gpu architectures with complete compilation in device-only145// compilation mode.146//147// RUN: %clang -target powerpc64le-ibm-linux-gnu -ccc-print-bindings --no-offload-new-driver \148// RUN: --cuda-gpu-arch=sm_30 --cuda-gpu-arch=sm_35 %s --cuda-device-only 2>&1 \149// RUN: | FileCheck -check-prefix=DBIN2 %s150// DBIN2: # "nvptx64-nvidia-cuda" - "clang",{{.*}} output:151// DBIN2-NOT: cuda-bindings-device-cuda-nvptx64152// DBIN2: # "nvptx64-nvidia-cuda" - "NVPTX::Assembler",{{.*}} output: "cuda-bindings-cuda-nvptx64-nvidia-cuda-sm_30.o"153// DBIN2: # "nvptx64-nvidia-cuda" - "clang",{{.*}} output:154// DBIN2-NOT: cuda-bindings-device-cuda-nvptx64155// DBIN2: # "nvptx64-nvidia-cuda" - "NVPTX::Assembler",{{.*}} output: "cuda-bindings-cuda-nvptx64-nvidia-cuda-sm_35.o"156 157//158// Test two gpu architectures up to the assemble phase in device-only159// compilation mode.160//161// RUN: %clang -target powerpc64le-ibm-linux-gnu -ccc-print-bindings --no-offload-new-driver \162// RUN: --cuda-gpu-arch=sm_30 --cuda-gpu-arch=sm_35 %s --cuda-device-only -S 2>&1 \163// RUN: | FileCheck -check-prefix=DASM2 %s164// DASM2: # "nvptx64-nvidia-cuda" - "clang",{{.*}} output: "cuda-bindings-cuda-nvptx64-nvidia-cuda-sm_30.s"165// DASM2: # "nvptx64-nvidia-cuda" - "clang",{{.*}} output: "cuda-bindings-cuda-nvptx64-nvidia-cuda-sm_35.s"166 167//168// Ensure we output the user's specified name in device-only mode.169//170// RUN: %clang -target powerpc64le-ibm-linux-gnu -### \171// RUN: --cuda-gpu-arch=sm_52 --cuda-device-only -c -o foo.o --cuda-path=%S/Inputs/CUDA_80/usr/local/cuda %s 2>&1 \172// RUN: | FileCheck -check-prefix=D_ONLY %s173// RUN: %clang -target powerpc64le-ibm-linux-gnu -### --offload-new-driver \174// RUN: --cuda-gpu-arch=sm_52 --cuda-device-only -c -o foo.o --cuda-path=%S/Inputs/CUDA_80/usr/local/cuda %s 2>&1 \175// RUN: | FileCheck -check-prefix=D_ONLY %s176// D_ONLY: "foo.o"177 178//179// Check to make sure we can generate multiple outputs for device-only180// compilation and fail with '-o'.181//182// RUN: %clang -### -target powerpc64le-ibm-linux-gnu --offload-new-driver -ccc-print-bindings \183// RUN: --offload-arch=sm_70 --offload-arch=sm_52 --offload-device-only -c %s 2>&1 \184// RUN: | FileCheck -check-prefix=MULTI-D-ONLY %s185// MULTI-D-ONLY: # "nvptx64-nvidia-cuda" - "clang", inputs: ["[[INPUT:.+]]"], output: "[[PTX_70:.+]]"186// MULTI-D-ONLY-NEXT: # "nvptx64-nvidia-cuda" - "NVPTX::Assembler", inputs: ["[[PTX_70]]"], output: "[[CUBIN_70:.+]]"187// MULTI-D-ONLY-NEXT: # "nvptx64-nvidia-cuda" - "clang", inputs: ["[[INPUT]]"], output: "[[PTX_52:.+]]"188// MULTI-D-ONLY-NEXT: # "nvptx64-nvidia-cuda" - "NVPTX::Assembler", inputs: ["[[PTX_52]]"], output: "[[CUBIN_52:.+]]"189//190// RUN: not %clang -### --target=powerpc64le-ibm-linux-gnu --offload-new-driver -ccc-print-bindings \191// RUN: --offload-arch=sm_70 --offload-arch=sm_52 --offload-device-only -c -o %t %s 2>&1 \192// RUN: | FileCheck -check-prefix=MULTI-D-ONLY-O %s193// MULTI-D-ONLY-O: error: cannot specify -o when generating multiple output files194 195//196// Check to ensure that we can use '-fsyntax-only' for CUDA output with the new197// driver.198// 199// RUN: %clang -### -target powerpc64le-ibm-linux-gnu --offload-new-driver \200// RUN: -fsyntax-only --offload-arch=sm_70 --offload-arch=sm_52 -c --cuda-path=%S/Inputs/CUDA_111/usr/local/cuda %s 2>&1 \201// RUN: | FileCheck -check-prefix=SYNTAX-ONLY %s202// SYNTAX-ONLY: "-cc1" "-triple" "nvptx64-nvidia-cuda"{{.*}}"-fsyntax-only"203// SYNTAX-ONLY: "-cc1" "-triple" "nvptx64-nvidia-cuda"{{.*}}"-fsyntax-only"204// SYNTAX-ONLY: "-cc1" "-triple" "powerpc64le-ibm-linux-gnu"{{.*}}"-fsyntax-only"205 206//207// Check to ensure that we can use '-save-temps' when operating in RDC-mode.208//209// RUN: %clang -### -target powerpc64le-ibm-linux-gnu -save-temps --offload-new-driver \210// RUN: -fgpu-rdc --offload-arch=sm_70 --offload-arch=sm_52 -c --cuda-path=%S/Inputs/CUDA_111/usr/local/cuda %s 2>&1 \211// RUN: | FileCheck -check-prefix=SAVE-TEMPS %s212// SAVE-TEMPS: "-cc1" "-triple" "nvptx64-nvidia-cuda"{{.*}}"-target-cpu" "sm_52"213// SAVE-TEMPS: "-cc1" "-triple" "nvptx64-nvidia-cuda"{{.*}}"-target-cpu" "sm_70"214// SAVE-TEMPS: "-cc1" "-triple" "powerpc64le-ibm-linux-gnu"215 216//217// Check to ensure that we cannot use '-foffload' when not operating in RDC-mode.218//219// RUN: not %clang -### --target=powerpc64le-ibm-linux-gnu -fno-gpu-rdc --offload-new-driver \220// RUN: -foffload-lto --offload-arch=sm_70 --offload-arch=sm_52 -c %s 2>&1 \221// RUN: | FileCheck -check-prefix=LTO-NO-RDC %s222// LTO-NO-RDC: error: unsupported option '-foffload-lto' for language mode '-fno-gpu-rdc'223