mirrors/tinygrad
2
0
Fork 0
mirror of https://github.com/tinygrad/tinygrad.git synced 2026-06-24 02:14:17 +00:00
Code Issues Projects Releases Packages Wiki Activity

ci: try parallelizing tc tests (#16710)

Browse source
This commit is contained in:
Christopher Milan 2026-06-22 17:43:32 -07:00 • committed by GitHub
commit da98fae9e1
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 16 additions and 36 deletions
Download patch file Download diff file Expand all files Collapse all files

52
.github/workflows/test.yml vendored
View file

@ -133,46 +133,26 @@ jobs:
run: SKIP_SLOW_TEST=1 DEV=PYTHON python3 -m pytest -n=auto test/backend/test_dtype.py test/backend/test_dtype_alu.py test/backend/test_ops.py test/backend/test_uops.py test/backend/test_symbolic_ops.py test/backend/test_renderer_failures.py::TestRendererFailures --durations=20
- name: Test IMAGE support
run: IMAGE=1 DEV=PYTHON python3 test/backend/test_ops.py TestOps.test_gemm TestOps.test_simple_conv2d
- name: Test emulated METAL tensor cores
- name: Test emulated tensor cores
env:
DEV: 'PYTHON::METAL'
DEBUG: 2
N: 64
CNT: 1
SHOULD_USE_TC: 1
run: |
DEBUG=2 python3 test/backend/test_ops.py TestOps.test_big_gemm
python3 -m pytest -nauto test/opt/test_tensor_cores.py
- name: Test emulated AMD tensor cores
env:
DEV: 'PYTHON::gfx1100'
parallel -k --link --tagstring '[{1}]' '{2} python3 ./extra/gemm/simple_matmul.py' \
::: metal gfx950 gfx1100 gfx1100_acchalf gfx1201 gfx1201_acchalf sm_75 sm_80_half sm_80_tf32 \
::: 'DEV=PYTHON::METAL' 'DEV=PYTHON::gfx950 HALF=1 ACC_HALF=0' \
'DEV=PYTHON::gfx1100 HALF=1 ACC_HALF=0' 'DEV=PYTHON::gfx1100 HALF=1 ACC_HALF=1 ATOL=1e-3' \
'DEV=PYTHON::gfx1201 HALF=1 ACC_HALF=0' 'DEV=PYTHON::gfx1201 HALF=1 ACC_HALF=1 ATOL=1e-3' \
'DEV=PYTHON::sm_75 HALF=1' 'DEV=PYTHON::sm_80 HALF=1' 'DEV=PYTHON::sm_80 ALLOW_TF32=1'
- name: Run additional tensor core tests
run: |
DEBUG=2 N=16 HALF=1 ACC_HALF=0 python3 ./extra/gemm/simple_matmul.py
DEBUG=2 N=64 HALF=1 ACC_HALF=0 python3 ./extra/gemm/simple_matmul.py
DEBUG=2 N=16 HALF=1 ACC_HALF=1 ATOL=1e-3 python3 ./extra/gemm/simple_matmul.py
DEBUG=2 N=64 HALF=1 ACC_HALF=1 ATOL=1e-3 python3 ./extra/gemm/simple_matmul.py
python3 -m pytest -nauto test/opt/test_tensor_cores.py
- name: Test emulated AMD MFMA tensor cores
env:
DEV: 'PYTHON::gfx950'
run: |
DEBUG=2 N=64 HALF=1 ACC_HALF=0 python3 ./extra/gemm/simple_matmul.py
python3 -m pytest -nauto test/opt/test_tensor_cores.py
- name: Test emulated AMD RDNA4 tensor cores
env:
DEV: 'PYTHON::gfx1201'
run: |
DEBUG=2 N=16 HALF=1 ACC_HALF=0 python3 ./extra/gemm/simple_matmul.py
DEBUG=2 N=64 HALF=1 ACC_HALF=0 python3 ./extra/gemm/simple_matmul.py
DEBUG=2 N=16 HALF=1 ACC_HALF=1 ATOL=1e-3 python3 ./extra/gemm/simple_matmul.py
DEBUG=2 N=64 HALF=1 ACC_HALF=1 ATOL=1e-3 python3 ./extra/gemm/simple_matmul.py
python3 -m pytest -nauto test/opt/test_tensor_cores.py
- name: Test emulated CUDA tensor cores
run: |
DEBUG=2 DEV=PYTHON::sm_80 python3 test/backend/test_ops.py TestOps.test_gemm_fp16
DEBUG=2 ALLOW_TF32=1 DEV=PYTHON::sm_80 python3 test/backend/test_ops.py TestOps.test_gemm
DEBUG=2 DEV=PYTHON::sm_75 python3 test/backend/test_ops.py TestOps.test_gemm_fp16
DEV=PYTHON::METAL python3 -m pytest -nauto test/opt/test_tensor_cores.py test/null/test_uops_stats.py::TestUOpsStatsMatmulHalf
DEV=PYTHON::gfx1100 python3 -m pytest -nauto test/opt/test_tensor_cores.py test/null/test_uops_stats.py::TestUOpsStatsMatmulHalf
DEV=PYTHON::gfx950 python3 -m pytest -nauto test/opt/test_tensor_cores.py
DEV=PYTHON::gfx1201 python3 -m pytest -nauto test/opt/test_tensor_cores.py
ALLOW_TF32=1 DEV=PYTHON::sm_89 python3 -m pytest -nauto test/opt/test_tensor_cores.py
- name: Test device flop counts
run: |
DEBUG=2 DEV=PYTHON::METAL python3 ./test/null/test_uops_stats.py TestUOpsStatsMatmulHalf
DEBUG=2 DEV=PYTHON::gfx1100 python3 ./test/null/test_uops_stats.py TestUOpsStatsMatmulHalf
DEBUG=2 DEV=PYTHON::sm_80 python3 ./test/null/test_uops_stats.py TestUOpsStatsMatmulHalf
linter: