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

norm: add try catch sycl exception

Browse source
This commit is contained in:
Akarshan Biswas 2025-02-02 18:20:44 +05:30
parent 5c05a3eedc
commit d31c62d758
No known key found for this signature in database
GPG key ID: 52A578A14B32134D
3 changed files with 33 additions and 27 deletions
Download patch file Download diff file Expand all files Collapse all files

19
ggml/src/ggml-sycl/ggml-sycl.cpp
View file

@ -2106,25 +2106,6 @@ catch (sycl::exception const &exc) {
std::exit(1); std::exit(1);
} }
static void ggml_sycl_norm(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
GGML_SYCL_DEBUG("call %s\n", __func__);
ggml_sycl_op_norm(ctx, dst);
GGML_SYCL_DEBUG("call %s done\n", __func__);
}
static void ggml_sycl_rms_norm(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
GGML_SYCL_DEBUG("call %s\n", __func__);
ggml_sycl_op_rms_norm(ctx, dst);
GGML_SYCL_DEBUG("call %s done\n", __func__);
}
static void ggml_sycl_group_norm(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
GGML_SYCL_DEBUG("call %s\n", __func__);
ggml_sycl_op_group_norm(ctx, dst);
GGML_SYCL_DEBUG("call %s done\n", __func__);
}
static void ggml_sycl_mul_mat_vec_p021(ggml_backend_sycl_context & ctx, const ggml_tensor *src0, static void ggml_sycl_mul_mat_vec_p021(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
const ggml_tensor *src1, const ggml_tensor *src1,
ggml_tensor *dst) try { ggml_tensor *dst) try {

33
ggml/src/ggml-sycl/norm.cpp
View file

@ -311,7 +311,7 @@ static void rms_norm_f32_sycl(const float* x, float* dst, const int ncols,
} }
} }
void ggml_sycl_op_norm(ggml_backend_sycl_context & ctx, ggml_tensor * dst) { static void ggml_sycl_op_norm(ggml_backend_sycl_context & ctx, ggml_tensor * dst) try {
GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32); GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32);
GGML_ASSERT(dst->type == GGML_TYPE_F32); GGML_ASSERT(dst->type == GGML_TYPE_F32);
@ -328,9 +328,12 @@ void ggml_sycl_op_norm(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
dpct::queue_ptr main_stream = ctx.stream(); dpct::queue_ptr main_stream = ctx.stream();
norm_f32_sycl(src0_dd, dst_dd, ne00, nrows, eps, main_stream, ctx.device); norm_f32_sycl(src0_dd, dst_dd, ne00, nrows, eps, main_stream, ctx.device);
} catch (const sycl::exception & exc) {
std::cerr << exc.what() << "Exception caught at file:" << __FILE__ << ", line:" << __LINE__ << std::endl;
std::exit(1);
} }
void ggml_sycl_op_group_norm(ggml_backend_sycl_context& ctx, ggml_tensor* dst) { static void ggml_sycl_op_group_norm(ggml_backend_sycl_context& ctx, ggml_tensor* dst) try {
GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32); GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32);
GGML_ASSERT(dst->type == GGML_TYPE_F32); GGML_ASSERT(dst->type == GGML_TYPE_F32);
@ -346,9 +349,12 @@ void ggml_sycl_op_group_norm(ggml_backend_sycl_context& ctx, ggml_tensor* dst) {
float * dst_dd = static_cast<float *>(dst->data); float * dst_dd = static_cast<float *>(dst->data);
dpct::queue_ptr main_stream = ctx.stream(); dpct::queue_ptr main_stream = ctx.stream();
group_norm_f32_sycl(src0_dd, dst_dd, num_groups, eps, group_size, dst->src[0]->ne[0] * dst->src[0]->ne[1] * dst->src[0]->ne[2], main_stream, ctx.device); group_norm_f32_sycl(src0_dd, dst_dd, num_groups, eps, group_size, dst->src[0]->ne[0] * dst->src[0]->ne[1] * dst->src[0]->ne[2], main_stream, ctx.device);
} catch (const sycl::exception & exc) {
std::cerr << exc.what() << "Exception caught at file:" << __FILE__ << ", line:" << __LINE__ << std::endl;
std::exit(1);
} }
void ggml_sycl_op_rms_norm(ggml_backend_sycl_context & ctx, ggml_tensor * dst) { static void ggml_sycl_op_rms_norm(ggml_backend_sycl_context & ctx, ggml_tensor * dst) try {
GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32); GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32);
GGML_ASSERT(dst->type == GGML_TYPE_F32); GGML_ASSERT(dst->type == GGML_TYPE_F32);
@ -364,4 +370,25 @@ void ggml_sycl_op_rms_norm(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
dpct::queue_ptr main_stream = ctx.stream(); dpct::queue_ptr main_stream = ctx.stream();
rms_norm_f32_sycl(src0_dd, dst_dd, ne00, nrows, eps, main_stream, ctx.device); rms_norm_f32_sycl(src0_dd, dst_dd, ne00, nrows, eps, main_stream, ctx.device);
} catch (const sycl::exception & exc) {
std::cerr << exc.what() << "Exception caught at file:" << __FILE__ << ", line:" << __LINE__ << std::endl;
std::exit(1);
} }
void ggml_sycl_norm(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
GGML_SYCL_DEBUG("call %s\n", __func__);
ggml_sycl_op_norm(ctx, dst);
GGML_SYCL_DEBUG("call %s done\n", __func__);
}
void ggml_sycl_rms_norm(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
GGML_SYCL_DEBUG("call %s\n", __func__);
ggml_sycl_op_rms_norm(ctx, dst);
GGML_SYCL_DEBUG("call %s done\n", __func__);
}
void ggml_sycl_group_norm(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
GGML_SYCL_DEBUG("call %s\n", __func__);
ggml_sycl_op_group_norm(ctx, dst);
GGML_SYCL_DEBUG("call %s done\n", __func__);
}

8
ggml/src/ggml-sycl/norm.hpp
View file

@ -15,10 +15,8 @@
#include "common.hpp" #include "common.hpp"
void ggml_sycl_op_norm(ggml_backend_sycl_context& ctx, ggml_tensor * dst); void ggml_sycl_norm(ggml_backend_sycl_context & ctx, ggml_tensor * dst);
void ggml_sycl_rms_norm(ggml_backend_sycl_context & ctx, ggml_tensor * dst);
void ggml_sycl_op_rms_norm(ggml_backend_sycl_context& ctx, ggml_tensor * dst); void ggml_sycl_group_norm(ggml_backend_sycl_context & ctx, ggml_tensor * dst);
void ggml_sycl_op_group_norm(ggml_backend_sycl_context& ctx, ggml_tensor * dst);
#endif // GGML_SYCL_NORM_HPP #endif // GGML_SYCL_NORM_HPP