clang/test/Driver/rocm-device-libs.cl

   1 // REQUIRES: amdgpu-registered-target
   2 // REQUIRES: !system-windows
   3
   4 // Test flush-denormals-to-zero enabled uses oclc_daz_opt_on
   5
   6 // RUN: %clang -### -target amdgcn-amd-amdhsa \
   7 // RUN:   -x cl -mcpu=gfx900 \
   8 // RUN:   --rocm-path=%S/Inputs/rocm \
   9 // RUN:   %s \
  10 // RUN: 2>&1 | FileCheck  --check-prefixes=COMMON,COMMON-DEFAULT,GFX900-DEFAULT,GFX900,WAVE64 %s
  11
  12
  13
  14 // Make sure the different denormal default is respected for gfx8
  15 // RUN: %clang -### -target amdgcn-amd-amdhsa \
  16 // RUN:   -x cl -mcpu=gfx803 \
  17 // RUN:   --rocm-path=%S/Inputs/rocm \
  18 // RUN:   %s \
  19 // RUN: 2>&1 | FileCheck  --check-prefixes=COMMON,COMMON-DEFAULT,GFX803-DEFAULT,GFX803,WAVE64 %s
  20
  21
  22
  23 // Make sure the non-canonical name works
  24 // RUN: %clang -### -target amdgcn-amd-amdhsa \
  25 // RUN:   -x cl -mcpu=fiji \
  26 // RUN:   --rocm-path=%S/Inputs/rocm \
  27 // RUN:   %s \
  28 // RUN: 2>&1 | FileCheck  --check-prefixes=COMMON,COMMON-DEFAULT,GFX803-DEFAULT,GFX803,WAVE64 %s
  29
  30
  31
  32 // RUN: %clang -### -target amdgcn-amd-amdhsa \
  33 // RUN:   -x cl -mcpu=gfx900 \
  34 // RUN:   -cl-denorms-are-zero \
  35 // RUN:   --rocm-path=%S/Inputs/rocm \
  36 // RUN:   %s \
  37 // RUN: 2>&1 | FileCheck  --check-prefixes=COMMON,COMMON-DAZ,GFX900,WAVE64 %s
  38
  39
  40 // RUN: %clang -### -target amdgcn-amd-amdhsa \
  41 // RUN:   -x cl -mcpu=gfx803 \
  42 // RUN:   -cl-denorms-are-zero \
  43 // RUN:   --rocm-path=%S/Inputs/rocm \
  44 // RUN:   %s \
  45 // RUN: 2>&1 | FileCheck  --check-prefixes=COMMON,COMMON-DAZ,GFX803,WAVE64 %s
  46
  47
  48
  49 // RUN: %clang -### -target amdgcn-amd-amdhsa \
  50 // RUN:   -x cl -mcpu=gfx803 \
  51 // RUN:   -cl-finite-math-only \
  52 // RUN:   --rocm-path=%S/Inputs/rocm \
  53 // RUN:   %s \
  54 // RUN: 2>&1 | FileCheck  --check-prefixes=COMMON,COMMON-FINITE-ONLY,GFX803,WAVE64 %s
  55
  56
  57
  58 // RUN: %clang -### -target amdgcn-amd-amdhsa    \
  59 // RUN:   -x cl -mcpu=gfx803                     \
  60 // RUN:   -cl-fp32-correctly-rounded-divide-sqrt \
  61 // RUN:   --rocm-path=%S/Inputs/rocm \
  62 // RUN:   %s \
  63 // RUN: 2>&1 | FileCheck  --check-prefixes=COMMON,COMMON-CORRECT-SQRT,GFX803,WAVE64 %s
  64
  65
  66
  67 // RUN: %clang -### -target amdgcn-amd-amdhsa    \
  68 // RUN:   -x cl -mcpu=gfx803                     \
  69 // RUN:   -cl-fast-relaxed-math \
  70 // RUN:   --rocm-path=%S/Inputs/rocm \
  71 // RUN:   %s \
  72 // RUN: 2>&1 | FileCheck  --check-prefixes=COMMON,COMMON-FAST-RELAXED,GFX803,WAVE64 %s
  73
  74
  75
  76 // RUN: %clang -### -target amdgcn-amd-amdhsa    \
  77 // RUN:   -x cl -mcpu=gfx803                     \
  78 // RUN:   -cl-unsafe-math-optimizations \
  79 // RUN:   --rocm-path=%S/Inputs/rocm \
  80 // RUN:   %s 2>&1 | \
  81 // RUN:   FileCheck  --check-prefixes=COMMON,GFX803,WAVE64 %s
  82
  83 // RUN: %clang -### -target amdgcn-amd-amdhsa    \
  84 // RUN:   -x cl -mcpu=gfx1010                    \
  85 // RUN:   --rocm-path=%S/Inputs/rocm \
  86 // RUN:   %s 2>&1 | \
  87 // RUN:   FileCheck  --check-prefix=WAVE32 %s
  88
  89 // RUN: %clang -### -target amdgcn-amd-amdhsa    \
  90 // RUN:   -x cl -mcpu=gfx1011                    \
  91 // RUN:   --rocm-path=%S/Inputs/rocm \
  92 // RUN:   %s 2>&1 | \
  93 // RUN:   FileCheck  --check-prefix=WAVE32 %s
  94
  95 // RUN: %clang -### -target amdgcn-amd-amdhsa    \
  96 // RUN:   -x cl -mcpu=gfx1012                    \
  97 // RUN:   --rocm-path=%S/Inputs/rocm \
  98 // RUN:   %s 2>&1 | \
  99 // RUN:   FileCheck  --check-prefix=WAVE32 %s
 100
 101
 102 // RUN: %clang -### -target amdgcn-amd-amdhsa    \
 103 // RUN:   -x cl -mcpu=gfx1010 -mwavefrontsize64  \
 104 // RUN:   --rocm-path=%S/Inputs/rocm \
 105 // RUN:   %s 2>&1 | \
 106 // RUN:   FileCheck  --check-prefix=WAVE64 %s
 107
 108 // RUN: %clang -### -target amdgcn-amd-amdhsa    \
 109 // RUN:   -x cl -mcpu=gfx1010 -mwavefrontsize64 -mno-wavefrontsize64  \
 110 // RUN:   --rocm-path=%S/Inputs/rocm \
 111 // RUN:   %s 2>&1 | \
 112 // RUN:   FileCheck  --check-prefix=WAVE32 %s
 113
 114 // Ignore -mno-wavefrontsize64 without wave32 support
 115 // RUN: %clang -### -target amdgcn-amd-amdhsa       \
 116 // RUN:   -x cl -mcpu=gfx803  -mno-wavefrontsize64  \
 117 // RUN:   --rocm-path=%S/Inputs/rocm    \
 118 // RUN:   %s 2>&1 | \
 119 // RUN:   FileCheck  --check-prefixes=GFX803,WAVE64 %s
 120
 121
 122
 123 // Test --hip-device-lib-path format
 124 // RUN: %clang -### -target amdgcn-amd-amdhsa \
 125 // RUN:   -x cl -mcpu=gfx900 \
 126 // RUN:   --hip-device-lib-path=%S/Inputs/rocm/amdgcn/bitcode \
 127 // RUN:   %S/opencl.cl \
 128 // RUN: 2>&1 | FileCheck  --check-prefixes=COMMON,COMMON-DEFAULT,GFX900-DEFAULT,GFX900,WAVE64 %s
 129
 130 // Test environment variable HIP_DEVICE_LIB_PATH
 131 // RUN: env HIP_DEVICE_LIB_PATH=%S/Inputs/rocm/amdgcn/bitcode %clang -### -target amdgcn-amd-amdhsa \
 132 // RUN:   -x cl -mcpu=gfx900 \
 133 // RUN:   %S/opencl.cl \
 134 // RUN: 2>&1 | FileCheck  --check-prefixes=COMMON,COMMON-DEFAULT,GFX900-DEFAULT,GFX900,WAVE64 %s
 135
 136
 137
 138 // COMMON: "-triple" "amdgcn-amd-amdhsa"
 139 // COMMON-SAME: "-mlink-builtin-bitcode" "{{.*}}/amdgcn/bitcode/opencl.bc"
 140 // COMMON-SAME: "-mlink-builtin-bitcode" "{{.*}}/amdgcn/bitcode/ocml.bc"
 141 // COMMON-SAME: "-mlink-builtin-bitcode" "{{.*}}/amdgcn/bitcode/ockl.bc"
 142
 143 // GFX900-DEFAULT-SAME: "-mlink-builtin-bitcode" "{{.*}}/amdgcn/bitcode/oclc_daz_opt_off.bc"
 144 // GFX803-DEFAULT-SAME: "-mlink-builtin-bitcode" "{{.*}}/amdgcn/bitcode/oclc_daz_opt_on.bc"
 145 // GFX700-DEFAULT-SAME: "-mlink-builtin-bitcode" "{{.*}}/amdgcn/bitcode/oclc_daz_opt_on.bc"
 146 // COMMON-DAZ-SAME: "-mlink-builtin-bitcode" "{{.*}}/amdgcn/bitcode/oclc_daz_opt_on.bc"
 147
 148
 149 // COMMON-DEFAULT-SAME: "-mlink-builtin-bitcode" "{{.*}}/amdgcn/bitcode/oclc_unsafe_math_off.bc"
 150 // COMMON-DEFAULT-SAME: "-mlink-builtin-bitcode" "{{.*}}/amdgcn/bitcode/oclc_finite_only_off.bc"
 151 // COMMON-DEFAULT-SAME: "-mlink-builtin-bitcode" "{{.*}}/amdgcn/bitcode/oclc_correctly_rounded_sqrt_off.bc"
 152
 153
 154 // COMMON-FINITE-ONLY-SAME: "-mlink-builtin-bitcode" "{{.*}}/amdgcn/bitcode/oclc_unsafe_math_off.bc"
 155 // COMMON-FINITE-ONLY-SAME: "-mlink-builtin-bitcode" "{{.*}}/amdgcn/bitcode/oclc_finite_only_on.bc"
 156 // COMMON-FINITE-ONLY-SAME: "-mlink-builtin-bitcode" "{{.*}}/amdgcn/bitcode/oclc_correctly_rounded_sqrt_off.bc"
 157
 158
 159 // COMMON-CORRECT-SQRT-SAME: "-mlink-builtin-bitcode" "{{.*}}/amdgcn/bitcode/oclc_unsafe_math_off.bc"
 160 // COMMON-CORRECT-SQRT-SAME: "-mlink-builtin-bitcode" "{{.*}}/amdgcn/bitcode/oclc_finite_only_off.bc"
 161 // COMMON-CORRECT-SQRT-SAME: "-mlink-builtin-bitcode" "{{.*}}/amdgcn/bitcode/oclc_correctly_rounded_sqrt_on.bc"
 162
 163
 164 // COMMON-FAST-RELAXED-SAME: "-mlink-builtin-bitcode" "{{.*}}/amdgcn/bitcode/oclc_unsafe_math_on.bc"
 165 // COMMON-FAST-RELAXED-SAME: "-mlink-builtin-bitcode" "{{.*}}/amdgcn/bitcode/oclc_finite_only_on.bc"
 166 // COMMON-FAST-RELAXED-SAME: "-mlink-builtin-bitcode" "{{.*}}/amdgcn/bitcode/oclc_correctly_rounded_sqrt_off.bc"
 167
 168
 169 // COMMON-UNSAFE-MATH-SAME: "-mlink-builtin-bitcode" "{{.*}}/amdgcn/bitcode/oclc_unsafe_math_on.bc"
 170 // COMMON-UNSAFE-MATH-SAME: "-mlink-builtin-bitcode" "{{.*}}/amdgcn/bitcode/oclc_finite_only_off.bc"
 171 // COMMON-UNSAFE-MATH-SAME: "-mlink-builtin-bitcode" "{{.*}}/amdgcn/bitcode/oclc_correctly_rounded_sqrt_off.bc"
 172
 173 // WAVE64: "-mlink-builtin-bitcode" "{{.*}}/amdgcn/bitcode/oclc_wavefrontsize64_on.bc"
 174 // WAVE32: "-mlink-builtin-bitcode" "{{.*}}/amdgcn/bitcode/oclc_wavefrontsize64_off.bc"
 175
 176 // GFX900: "-mlink-builtin-bitcode" "{{.*}}/amdgcn/bitcode/oclc_isa_version_900.bc"
 177 // GFX803: "-mlink-builtin-bitcode" "{{.*}}/amdgcn/bitcode/oclc_isa_version_803.bc"
 178
 179 kernel void func(void);