vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

[SYCL] Add oneDNN primitive support (#9091)

Browse source 
* add onednn

* add sycl_f16

* add dnnl stream

* add engine map

* use dnnl for intel only

* use fp16fp16fp16

* update doc
This commit is contained in:
luoyu-intel 2024-08-22 12:50:10 +08:00 • committed by GitHub
parent a1631e53f6
commit 1731d4238f
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
6 changed files with 186 additions and 10 deletions
Download patch file Download diff file Expand all files Collapse all files

16
ggml/src/ggml-sycl.cpp
View file

@ -38,6 +38,7 @@
#include "ggml-sycl/backend.hpp"
#include "ggml-sycl/presets.hpp"
#include "ggml-sycl/gemm.hpp"
bool ggml_sycl_loaded(void);
void ggml_sycl_free_data(struct ggml_tensor * tensor);
@ -2482,6 +2483,7 @@ inline void ggml_sycl_op_mul_mat_sycl(
const sycl::half alpha_f16 = 1.0f;
const sycl::half beta_f16 = 0.0f;
#if !GGML_SYCL_DNNL
SYCL_CHECK(CHECK_TRY_ERROR(dpct::gemm(
*stream, oneapi::mkl::transpose::trans,
oneapi::mkl::transpose::nontrans, row_diff, src1_ncols, ne10,
@ -2491,6 +2493,13 @@ inline void ggml_sycl_op_mul_mat_sycl(
dpct::library_data_t::real_half)));
const to_fp32_sycl_t to_fp32_sycl = ggml_get_to_fp32_sycl(GGML_TYPE_F16);
to_fp32_sycl(dst_f16.get(), dst_dd_i, row_diff*src1_ncols, stream);
#else
auto dnnl_stream = ctx.stream_dnnl(stream);
DnnlGemmWrapper::row_gemm(dnnl_stream, false, true, src1_ncols, row_diff, ne10, src1_ptr, DnnlGemmWrapper::to_dt<sycl::half>(),
src0_ptr, DnnlGemmWrapper::to_dt<sycl::half>(), dst_f16.get(), DnnlGemmWrapper::to_dt<sycl::half>());
const to_fp32_sycl_t to_fp32_sycl = ggml_get_to_fp32_sycl(GGML_TYPE_F16);
to_fp32_sycl(dst_f16.get(), dst_dd_i, row_diff* src1_ncols, stream);
#endif
}
else {
// GGML_SYCL_DEBUG("ggml_sycl_op_mul_mat_sycl - fp32 path\n");
@ -2513,13 +2522,18 @@ inline void ggml_sycl_op_mul_mat_sycl(
const float alpha = 1.0f;
const float beta = 0.0f;
#if !GGML_SYCL_DNNL
SYCL_CHECK(CHECK_TRY_ERROR(oneapi::mkl::blas::column_major::gemm(
*stream, oneapi::mkl::transpose::trans,
oneapi::mkl::transpose::nontrans, row_diff, src1_ncols, ne10,
dpct::get_value(&alpha, *stream), src0_ddf_i, ne00,
src1_ddf1_i, ne10, dpct::get_value(&beta, *stream),
dst_dd_i, ldc)));
#else
auto dnnl_stream = ctx.stream_dnnl(stream);
DnnlGemmWrapper::row_gemm(dnnl_stream, false, true, src1_ncols, row_diff, ne10, src1_ddf1_i, DnnlGemmWrapper::to_dt<float>(),
src0_ddf_i, DnnlGemmWrapper::to_dt<float>(), dst_dd_i, DnnlGemmWrapper::to_dt<float>());
#endif
}
(void) dst;
(void) src1_ddq_i;