| 1 |
project('cudamodule', 'cuda', version : '1.0.0') |
| 2 |
|
| 3 |
nvcc = meson.get_compiler('cuda') |
| 4 |
cuda = import('unstable-cuda') |
| 5 |
|
| 6 |
arch_flags = cuda.nvcc_arch_flags(nvcc.version(), 'Auto', detected: ['6.0']) |
| 7 |
arch_readable = cuda.nvcc_arch_readable(nvcc.version(), 'Auto', detected: ['6.0']) |
| 8 |
driver_version = cuda.min_driver_version(nvcc.version()) |
| 9 |
|
| 10 |
message('NVCC version: ' + nvcc.version()) |
| 11 |
message('NVCC flags: ' + ' '.join(arch_flags)) |
| 12 |
message('NVCC readable: ' + ' '.join(arch_readable)) |
| 13 |
message('Driver version: >=' + driver_version) |
| 14 |
|
| 15 |
exe = executable('prog', 'prog.cu', cuda_args: arch_flags) |
| 16 |
test('cudatest', exe) |
| 17 |
|
| 18 |
|
| 19 |
# |
| 20 |
# Assert Series |
| 21 |
# |
| 22 |
|
| 23 |
# Sanity test. |
| 24 |
assert(' '.join(cuda.nvcc_arch_flags('11.1', '8.6')) == |
| 25 |
'-gencode arch=compute_86,code=sm_86') |
| 26 |
|
| 27 |
# CUDA Toolkit too old, flag filtered out. |
| 28 |
assert(' '.join(cuda.nvcc_arch_flags('11.0', '8.6')) == |
| 29 |
'') |
| 30 |
|
| 31 |
# Named architectures. |
| 32 |
assert(' '.join(cuda.nvcc_arch_flags('11.0', 'Ampere')) == |
| 33 |
'-gencode arch=compute_80,code=sm_80') |
| 34 |
|
| 35 |
# Splitting & deduplication. |
| 36 |
assert(' '.join(cuda.nvcc_arch_flags('11.0', 'Ampere;8.0,8.0')) == |
| 37 |
'-gencode arch=compute_80,code=sm_80') |
| 38 |
|
| 39 |
# Same, but list supplied as list. |
| 40 |
assert(' '.join(cuda.nvcc_arch_flags('11.0', ['Ampere', '8.0', '8.0'])) == |
| 41 |
'-gencode arch=compute_80,code=sm_80') |
| 42 |
|
| 43 |
# Same, but mode set to Auto with detected set to a string with a variety of separators. |
| 44 |
assert(' '.join(cuda.nvcc_arch_flags('11.0', 'Auto', detected: 'Ampere;8.0,8.0')) == |
| 45 |
'-gencode arch=compute_80,code=sm_80') |
| 46 |
|
| 47 |
# Same, but detected set to a list. |
| 48 |
assert(' '.join(cuda.nvcc_arch_flags('11.0', 'Auto', detected: ['Ampere', '8.0', '8.0'])) == |
| 49 |
'-gencode arch=compute_80,code=sm_80') |
| 50 |
|
| 51 |
# Ask for 8.6 binary with 8.0-level PTX. |
| 52 |
assert(' '.join(cuda.nvcc_arch_flags('11.1', '8.6(8.0)')) == |
| 53 |
'-gencode arch=compute_80,code=sm_86') |
| 54 |
|
| 55 |
# Same, but keep the 8.0 PTX. |
| 56 |
assert(' '.join(cuda.nvcc_arch_flags('11.1', '8.6(8.0)+PTX')) == |
| 57 |
'-gencode arch=compute_80,code=sm_86 -gencode arch=compute_80,code=compute_80') |
| 58 |
|
| 59 |
# Detected Ampere RTX 3090 on CUDA 10.2, saturate to 7.5+PTX |
| 60 |
assert(' '.join(cuda.nvcc_arch_flags('10.2', 'Auto', detected: ['8.0'])) == |
| 61 |
'-gencode arch=compute_75,code=sm_75 -gencode arch=compute_75,code=compute_75') |
| 62 |
|
| 63 |
# Failed to auto-detect with CUDA 10.2, default to common GPUs (3.0;3.5;5.0;5.2;6.0;6.1;7.0;7.5+PTX) |
| 64 |
assert(' '.join(cuda.nvcc_arch_flags('10.2', 'Auto', detected: [])) == |
| 65 |
'-gencode arch=compute_30,code=sm_30 -gencode arch=compute_35,code=sm_35 '+ |
| 66 |
'-gencode arch=compute_50,code=sm_50 -gencode arch=compute_52,code=sm_52 '+ |
| 67 |
'-gencode arch=compute_60,code=sm_60 -gencode arch=compute_61,code=sm_61 '+ |
| 68 |
'-gencode arch=compute_70,code=sm_70 -gencode arch=compute_75,code=sm_75 '+ |
| 69 |
'-gencode arch=compute_75,code=compute_75') |