Skip to content

[blas][rocblas] Support int8 inputs with float output in gemm_batch #763

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

Open
zjin-lcf wants to merge 3 commits into
base: develop
Choose a base branch
from
Open

[blas][rocblas] Support int8 inputs with float output in gemm_batch #763

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
b7e5435
[blas][cublas] Support int8 inputs with float output in gemm_batch
zjin-lcf Aug 14, 2026
f95fdbb
[blas][tests] Guard the int8 error-model diagnostic against a zero mo…
zjin-lcf Aug 15, 2026
a081fe0
[blas][rocblas] Support int8 inputs with float output in gemm_batch
zjin-lcf Aug 15, 2026
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
8 changes: 5 additions & 3 deletions src/blas/backends/cublas/cublas_batch.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -199,6 +199,7 @@ inline void gemm_batch_impl(sycl::queue& queue, transpose transa, transpose tran

GEMM_STRIDED_BATCH_LAUNCHER(sycl::half, sycl::half, sycl::half, sycl::half)
GEMM_STRIDED_BATCH_LAUNCHER(sycl::half, sycl::half, float, float)
GEMM_STRIDED_BATCH_LAUNCHER(std::int8_t, std::int8_t, float, float)
GEMM_STRIDED_BATCH_LAUNCHER(float, float, float, float)
GEMM_STRIDED_BATCH_LAUNCHER(double, double, double, double)
GEMM_STRIDED_BATCH_LAUNCHER(std::complex<float>, std::complex<float>, std::complex<float>,
Expand All @@ -208,6 +209,8 @@ GEMM_STRIDED_BATCH_LAUNCHER(std::complex<double>, std::complex<double>, std::com

#undef GEMM_STRIDED_BATCH_LAUNCHER

// cuBLAS computes an int32 output only with CUBLAS_COMPUTE_32I, which requires int32 alpha and
// beta, whereas oneMath specifies float scalars for this combination.
#define GEMM_STRIDED_BATCH_LAUNCHER(TYPE_A, TYPE_B, TYPE_C, TYPE_S) \
void gemm_batch(sycl::queue& queue, transpose transa, transpose transb, int64_t m, int64_t n, \
int64_t k, TYPE_S alpha, sycl::buffer<TYPE_A, 1>& a, int64_t lda, \
Expand All @@ -220,7 +223,6 @@ GEMM_STRIDED_BATCH_LAUNCHER(std::complex<double>, std::complex<double>, std::com
dtype_string<TYPE_C>() + "," + dtype_string<TYPE_S>() + ">"); \
}

GEMM_STRIDED_BATCH_LAUNCHER(std::int8_t, std::int8_t, float, float)
GEMM_STRIDED_BATCH_LAUNCHER(std::int8_t, std::int8_t, std::int32_t, float)

#undef GEMM_STRIDED_BATCH_LAUNCHER
Expand Down Expand Up @@ -668,6 +670,7 @@ inline sycl::event gemm_batch_strided_usm_impl(sycl::queue& queue, transpose tra

GEMM_STRIDED_BATCH_LAUNCHER_USM(sycl::half, sycl::half, sycl::half, sycl::half)
GEMM_STRIDED_BATCH_LAUNCHER_USM(sycl::half, sycl::half, float, float)
GEMM_STRIDED_BATCH_LAUNCHER_USM(std::int8_t, std::int8_t, float, float)
GEMM_STRIDED_BATCH_LAUNCHER_USM(float, float, float, float)
GEMM_STRIDED_BATCH_LAUNCHER_USM(double, double, double, double)
GEMM_STRIDED_BATCH_LAUNCHER_USM(std::complex<float>, std::complex<float>, std::complex<float>,
Expand All @@ -689,7 +692,6 @@ GEMM_STRIDED_BATCH_LAUNCHER_USM(std::complex<double>, std::complex<double>, std:
dtype_string<TYPE_C>() + "," + dtype_string<TYPE_S>() + ">"); \
}

GEMM_STRIDED_BATCH_LAUNCHER_USM(std::int8_t, std::int8_t, float, float)
GEMM_STRIDED_BATCH_LAUNCHER_USM(std::int8_t, std::int8_t, std::int32_t, float)

#undef GEMM_STRIDED_BATCH_LAUNCHER_USM
Expand Down Expand Up @@ -761,6 +763,7 @@ inline sycl::event gemm_batch_usm_impl(sycl::queue& queue, transpose* transa, tr

GEMM_BATCH_LAUNCHER_USM(sycl::half, sycl::half, sycl::half, sycl::half)
GEMM_BATCH_LAUNCHER_USM(sycl::half, sycl::half, float, float)
GEMM_BATCH_LAUNCHER_USM(std::int8_t, std::int8_t, float, float)
GEMM_BATCH_LAUNCHER_USM(float, float, float, float)
GEMM_BATCH_LAUNCHER_USM(double, double, double, double)
GEMM_BATCH_LAUNCHER_USM(std::complex<float>, std::complex<float>, std::complex<float>,
Expand All @@ -782,7 +785,6 @@ GEMM_BATCH_LAUNCHER_USM(std::complex<double>, std::complex<double>, std::complex
dtype_string<TYPE_C>() + "," + dtype_string<TYPE_S>() + ">"); \
}

GEMM_BATCH_LAUNCHER_USM(std::int8_t, std::int8_t, float, float)
GEMM_BATCH_LAUNCHER_USM(std::int8_t, std::int8_t, std::int32_t, float)

#undef GEMM_BATCH_LAUNCHER_USM
Expand Down
Loading
Loading