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

Maintain previous behaviour for igpu

Browse source
This commit is contained in:
Aidan 2024-03-21 11:56:03 +00:00
parent 10aef3680f
commit d1cb8fedbd
1 changed files with 28 additions and 9 deletions
Download patch file Download diff file Expand all files Collapse all files

27
ggml-sycl.cpp
View file

@ -15246,6 +15246,9 @@ static void ggml_sycl_mul_mat_batched_sycl(const ggml_tensor *src0,
SYCL_CHECK(ggml_sycl_set_device(g_main_device)); SYCL_CHECK(ggml_sycl_set_device(g_main_device));
dpct::queue_ptr main_stream = g_syclStreams[g_main_device][0]; dpct::queue_ptr main_stream = g_syclStreams[g_main_device][0];
bool no_mixed_dtypes = main_stream->get_backend() == sycl::backend::ext_oneapi_cuda ||
main_stream->get_backend() == sycl::backend::ext_oneapi_hip;
SYCL_CHECK( SYCL_CHECK(
CHECK_TRY_ERROR(g_sycl_handles[g_main_device] = main_stream)); CHECK_TRY_ERROR(g_sycl_handles[g_main_device] = main_stream));
@ -15274,26 +15277,40 @@ static void ggml_sycl_mul_mat_batched_sycl(const ggml_tensor *src0,
sycl_pool_alloc<sycl::half> dst_f16; sycl_pool_alloc<sycl::half> dst_f16;
char * dst_t; char * dst_t;
dpct::library_data_t cu_compute_type = dpct::library_data_t::real_half; dpct::library_data_t cu_compute_type = dpct::library_data_t::real_float;
dpct::library_data_t cu_data_type = dpct::library_data_t::real_half; dpct::library_data_t cu_data_type = dpct::library_data_t::real_float;
if (no_mixed_dtypes) {
cu_compute_type = dpct::library_data_t::real_half;
cu_data_type = dpct::library_data_t::real_half;
}
// dst strides // dst strides
size_t nbd2 = dst->nb[2]; size_t nbd2 = dst->nb[2];
size_t nbd3 = dst->nb[3]; size_t nbd3 = dst->nb[3];
const float alpha_f32 = 1.0f;
const float beta_f32 = 0.0f;
const sycl::half alpha_f16 = 1.0f; const sycl::half alpha_f16 = 1.0f;
const sycl::half beta_f16 = 0.0f; const sycl::half beta_f16 = 0.0f;
const void * alpha = &alpha_f16; const void * alpha = &alpha_f32;
const void * beta = &beta_f16; const void * beta = &beta_f32;
if (no_mixed_dtypes) {
alpha = &alpha_f16;
beta = &beta_f16;
}
// TODO: Renable (dst->op_params[0] =! GGML_PREC_DEFAULT) pathway // TODO: Renable (dst->op_params[0] =! GGML_PREC_DEFAULT) pathway
// when oneMKL open source supports half, half, float, float: datatypes // when oneMKL open source supports half, half, float, float: datatypes
dst_t = (char *) dst_ddf;
if (no_mixed_dtypes) {
dst_t = (char *) dst_f16.alloc(ne_dst); dst_t = (char *) dst_f16.alloc(ne_dst);
nbd2 /= sizeof(float) / sizeof(sycl::half); nbd2 /= sizeof(float) / sizeof(sycl::half);
nbd3 /= sizeof(float) / sizeof(sycl::half); nbd3 /= sizeof(float) / sizeof(sycl::half);
}
GGML_ASSERT(ne12 % ne02 == 0); GGML_ASSERT(ne12 % ne02 == 0);
GGML_ASSERT(ne13 % ne03 == 0); GGML_ASSERT(ne13 % ne03 == 0);
@ -15379,9 +15396,11 @@ static void ggml_sycl_mul_mat_batched_sycl(const ggml_tensor *src0,
} }
#endif #endif
if (no_mixed_dtypes) {
const to_fp32_sycl_t to_fp32_sycl = ggml_get_to_fp32_sycl(GGML_TYPE_F16); const to_fp32_sycl_t to_fp32_sycl = ggml_get_to_fp32_sycl(GGML_TYPE_F16);
to_fp32_sycl(dst_f16.get(), dst_ddf, ne_dst, main_stream); to_fp32_sycl(dst_f16.get(), dst_ddf, ne_dst, main_stream);
} }
}
catch (sycl::exception const &exc) { catch (sycl::exception const &exc) {
std::cerr << exc.what() << "Exception caught at file:" << __FILE__ std::cerr << exc.what() << "Exception caught at file:" << __FILE__
<< ", line:" << __LINE__ << std::endl; << ", line:" << __LINE__ << std::endl;