Skip to content

Remove int_scaled_mm's dependency on triton for cpu #128

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 10 commits into from
Oct 29, 2024
Merged

Remove int_scaled_mm's dependency on triton for cpu #128

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
d7f77c5
Avoid int_scaled_mm's dependency on triton for cpu
Xia-Weiwen Apr 8, 2024
ded45c1
Merge branch 'main' into cpu_int_scaled_mm_2
cpuhrsch Apr 15, 2024
e5758e1
Merge branch 'main' into cpu_int_scaled_mm_2
Xia-Weiwen Apr 18, 2024
946374c
refine implementation
Xia-Weiwen Apr 18, 2024
43c25b6
Fall back to ref if torch._int_mm not available on CPU
Xia-Weiwen Apr 18, 2024
f3b979a
Merge branch 'main' into cpu_int_scaled_mm_2
msaroufim Apr 25, 2024
b5b9525
Merge branch 'main' into cpu_int_scaled_mm_2
Xia-Weiwen Oct 28, 2024
0930f71
Refine code; _int_mm on CPU requires torch 2.6
Xia-Weiwen Oct 28, 2024
9f366a9
Merge branch 'main' into cpu_int_scaled_mm_2
Xia-Weiwen Oct 29, 2024
97dfea8
Decompose scaled_int_mm for CPU
Xia-Weiwen Oct 29, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 8 additions & 1 deletion torchao/kernel/intmm.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,7 +2,7 @@
import os
import torch

from torchao.utils import TORCH_VERSION_AT_LEAST_2_2
from torchao.utils import TORCH_VERSION_AT_LEAST_2_2, TORCH_VERSION_AT_LEAST_2_6

try:
# Only works for torch2.2 or newer.
Expand Down Expand Up @@ -134,6 +134,13 @@ def int_scaled_matmul(a: torch.Tensor, b: torch.Tensor, scales1: torch.Tensor) -
assert scales1.is_contiguous()
scales1 = scales1.expand((M, N))
assert scales1.dim() == 2

if scales1.device.type == "cpu" and TORCH_VERSION_AT_LEAST_2_6:
# CPU prefers decomposed version of int_scaled_matmul
# to leverage the fusion capability of Inductor
c = torch._int_mm(a, b)
return c.to(scales1.dtype) * scales1

if intmm_triton is not None and AUTOTUNER_ENABLE:
return torch.ops.torchao.int_scaled_matmul(a, b, scales1)

Expand Down
6 changes: 0 additions & 6 deletions torchao/kernel/intmm_triton.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -356,9 +356,3 @@ def int_scaled_matmul_cuda(a, b, scales1):
int_scaled_matmul_kernel, [a, b, scales1, c], int8_mm_kernel_configs
)
return int_scaled_matmul_kernel(a, b, scales1, c, best_config)


@torch.library.impl(lib, "int_scaled_matmul", "CPU")
def int_scaled_matmul_cpu(a, b, scales1):
c = torch._int_mm(a, b)
return c.to(scales1.dtype) * scales1
Loading