[MLAS] Add 8-bit weights ARM64 Gemm implementation #25110
hariharans29
merged
75 commits into
main
from
hari/matmul8bits_arm
finished prepack
b52a1ce1
changed interface to support blocksum2
05231068
finished quantb for quant a unsigned
fd92ab89
finished quantize a
ed5cf8d2
finished Q8Int8GemmR2xC8Neon
b9b9691e
finished kernels
685baffa
fixed build
67473307
passed prepack
b0873174
finished ut for quant a
196c04c2
fixed build
353d460b
Merge remote-tracking branch 'origin/main' into hari/matmul8bits_arm
4d62e32c
github-actions
commented
on 2025-06-18
Comment out some 4 bit tests
e88e32d7
Apple I8MM check
58011b04
Tests
acc4b812
Tests 2
2700493c
Update onnxruntime/test/mlas/unittest/test_sq8bitgemm.cpp
76de326a
Changes
159d4d32
Fixes
e4bc74ea
Re-enable 4 bit tests
e92055bd
Stage
94f30224
Some tests work
61c18728
Git attempt
16da92b0
Lint attempt
3ce481d9
Update onnxruntime/test/mlas/unittest/test_sq8bitgemm.cpp
29f66bd4
More changesc
987574bc
Merge branch 'hari/matmul8bits_arm' of https://github.com/microsoft/o…
d921b06a
Fix tests
cf92e6f1
github-actions
commented
on 2025-06-25
Stage
8156fc79
Stage
9a1fe225
Update onnxruntime/test/mlas/unittest/test_sq8bitgemm.cpp
31c8f931
Update onnxruntime/test/mlas/unittest/test_sq8bitgemm.cpp
92ec5ff0
Try fix x86 builds
7159d5e7
Merge branch 'hari/matmul8bits_arm' of https://github.com/microsoft/o…
7ad1d36c
Try fix lint errors
03f29166
Yipee zero point tests are all passing
47420b51
Comments and Nits
2a5100de
hariharans29
changed the title
Enable MatmulNBits test
d64568b2
Fixes
0c557559
Merge remote-tracking branch 'origin/main' into hari/matmul8bits_arm
01d4a98a
a
c8188d44
I8MM support re-enable
635eec9b
Fix warning
f736faea
Enable tests with ZP = false
aa794671
hariharans29
changed the title
github-actions
commented
on 2025-06-28
hariharans29
changed the title
Update onnxruntime/contrib_ops/cpu/quantization/matmul_nbits.cc
10e3afa2
I8MM fixes
c4331e01
hariharans29
commented
on 2025-06-28
Remove unnecessary template
5b7c3af7
copilot-pull-request-reviewer
commented
on 2025-06-28
vraspar
commented
on 2025-07-07
Resolve conflicts and update PR with more fixes
9ae58eec
Fix warning
b6cd309d
Properly remove warning
98f5fe0f
Merge remote-tracking branch 'origin' into hari/matmul8bits_arm
0d9442b4
PR feedback
9c2faa66
Refine
47e2420e
Update onnxruntime/contrib_ops/cpu/quantization/matmul_nbits.cc
5eb9ed9c
Update onnxruntime/contrib_ops/cpu/quantization/matmul_nbits.cc
bb978f72
Update onnxruntime/contrib_ops/cpu/quantization/matmul_nbits.cc
9b5c3891
Update onnxruntime/contrib_ops/cpu/quantization/matmul_nbits.cc
76d085b1
Update onnxruntime/test/contrib_ops/matmul_8bits_test.cc
12e3a1d1
Update onnxruntime/contrib_ops/cpu/quantization/matmul_nbits.cc
3827317a
Update onnxruntime/contrib_ops/cpu/quantization/matmul_nbits.cc
d8f4235e
Ignore sending scales while pre-packing weights on ARM64
46aa3629
Fix warning
2c956ae7
4 bit fix
83296bba
Merge branch 'hari/matmul8bits_arm' of https://github.com/microsoft/o…
8f145007
Update onnxruntime/test/contrib_ops/matmul_8bits_test.cc
405105b2
Lint
303e8677
Merge branch 'hari/matmul8bits_arm' of https://github.com/microsoft/o…
91de908d
Fix lintrunner mess-up once and for all
890a046a
Update onnxruntime/contrib_ops/cpu/quantization/matmul_nbits.cc
ec0c8abe
Update onnxruntime/contrib_ops/cpu/quantization/matmul_nbits.cc
1f71f6cd
Lint
376fc1b5
jywu-msft
dismissed these changes
on 2025-09-03
hariharans29
changed the title
More fixes
eefa72c0
hariharans29
changed the title
edgchen1
dismissed these changes
on 2025-09-04
PR comments
e1da3d5c
Missed out on one
77dff226
edgchen1
dismissed these changes
on 2025-09-04
Remove guards
7404cb37
Merge remote-tracking branch 'origin/main' into hari/matmul8bits_arm
edb3d728
edgchen1
approved these changes
on 2025-09-04
Labels
cherry-picked
Milestone
No milestone
Login to write a write a comment.
Login via GitHub