Added SME1 support for int8 GEMM and IGEMM operations #8880
Conversation
qti-vaiskv
force-pushed
the
sme1/pqs8-qc8w-gemm-igemm
branch
2 times, most recently
from
ae4bf78 to
6d89c3a
Compare
qti-vaiskv
force-pushed
the
sme1/pqs8-qc8w-gemm-igemm
branch
from
6d89c3a to
ab26084
Compare
|
Hi @dsharlet - can you please help review this PR |
qti-vaiskv
force-pushed
the
sme1/pqs8-qc8w-gemm-igemm
branch
from
ab26084 to
e8b092c
Compare
test/qs8-qc8w-igemm-minmax-fp32.cc
Outdated
| @@ -741,6 +741,46 @@ std::vector<GemmTestParams> CreateTests2( | |||
|
|
|||
|
|
|||
| #if XNN_ENABLE_AVX512AMX && (XNN_ARCH_X86 || XNN_ARCH_X86_64) | |||
| INSTANTIATE_TEST_SUITE_P( | |||
There was a problem hiding this comment.
Why are there so many unrelated changes in this file?
I checked at HEAD, it doesn't appear that this file is stale.
cc @fbarchard
There was a problem hiding this comment.
Sorry, I used the generate-gemm-test.py to generate the test to wrong output file test/qs8-qc8w-igemm-minmax-fp32.cc.
For the SME2 the output file used was test/qs8-qc8w-igemm-minmax-fp32-3.cc. I have regenerated this file with SME1 test case. But still the generated file has unrelated changes. Not sure why this is happening.
To generate the file, I used the below command:
python tools/generate-gemm-test.py -s test/qs8-qc8w-igemm-minmax-fp32.yaml -o test/qs8-qc8w-igemm-minmax-fp32-3.cc
There was a problem hiding this comment.
The correct command can be found in this script (or just run the script): https://github.com/google/XNNPACK/blob/master/scripts/generate-tests.sh
qti-vaiskv
force-pushed
the
sme1/pqs8-qc8w-gemm-igemm
branch
from
e8b092c to
05f7168
Compare
2 participants
No description provided.