gaspersic/triton
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[BACKEND] Support optional mask in TritonGPUToLLVM (#80)

Browse Source 
Co-authored-by: gzhu <gzhu@nvidia.com>
This commit is contained in:
goostavz
2022-08-25 08:51:37 +08:00
committed by GitHub
parent 84aa7d025a
commit bedbf221c0
5 changed files with 69 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
python/tests/test_vecadd_no_scf.py
View File

@@ -27,7 +27,13 @@ def test_vecadd_no_scf():
z_ptrs = z_ptr + offset
tl.store(z_ptrs, z)
ptx, shem_size, kernel_name = triton.compile(kernel, "*fp32,i32,*fp32,i32,*fp32,i32", constants={"BLOCK_SIZE_N": 256}, num_warps=NUM_WARPS, device=0, output="ptx")
# TODO: add this to CI, to make sure the the compilation flow is at lease OK
# before we have GPU machines for CI.
# ptx, shem_size, kernel_name = triton.compile(kernel,
# "*fp32,i32,*fp32,i32,*fp32,i32",
# constants={"BLOCK_SIZE_N": 256},
# num_warps=NUM_WARPS,
# device=0, output="ptx")
torch.zeros([10], device=torch.device('cuda'))
device = torch.cuda.current_device()