CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-09 10:17:06 +00:00
Code Issues Packages Projects Releases Wiki Activity

SYCL: Remove misleading ggml_sycl_op_flatten function (#12387)

Browse Source 
* SYCL: Remove misleading ggml_sycl_op_flatten function

* remove trailing whitespace

* Fix L2 norm from rebase

* remove try catch block from element_wise.cpp

* remove comment from common.hp

* ggml-sycl.cpp: Add try catch sycl::exception block in compute_forward

* norm.cpp: remove try catch exception block
This commit is contained in:
Akarshan Biswas
2025-03-31 14:55:24 +05:30
committed by GitHub
parent f52d59d771
commit 6c02a032fa
12 changed files with 369 additions and 586 deletions
Download Patch File Download Diff File Expand all files Collapse all files

44
ggml/src/ggml-sycl/getrows.cpp
View File

@@ -257,50 +257,54 @@ static void get_rows_sycl_float(ggml_backend_sycl_context & ctx, const ggml_tens
GGML_UNUSED(ctx);
}
void ggml_sycl_op_get_rows(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
const ggml_tensor *src1, ggml_tensor *dst,
const float *src0_d, const float *src1_d,
float *dst_d, const queue_ptr &stream) {
void ggml_sycl_op_get_rows(ggml_backend_sycl_context & ctx, ggml_tensor *dst) {
GGML_ASSERT(src1->type == GGML_TYPE_I32);
GGML_ASSERT(dst->src[1]->type == GGML_TYPE_I32);
GGML_ASSERT(dst->type == GGML_TYPE_F32);
GGML_ASSERT(src0->nb[0] == ggml_type_size(src0->type));
GGML_ASSERT(src1->nb[0] == ggml_type_size(src1->type));
GGML_ASSERT(dst->src[0]->nb[0] == ggml_type_size(dst->src[0]->type));
GGML_ASSERT(dst->src[1]->nb[0] == ggml_type_size(dst->src[1]->type));
GGML_ASSERT(dst->nb[0] == ggml_type_size(dst->type));
const int32_t * src1_i32 = (const int32_t *) src1_d;
switch (src0->type) {
const int32_t * src1_i32 = (const int32_t *) dst->src[1]->data;
/* TODO: Refactor and remove duplicates */
switch (dst->src[0]->type) {
case GGML_TYPE_F16:
get_rows_sycl_float(ctx, src0, src1, dst, (const sycl::half *)src0_d,
src1_i32, dst_d, stream);
get_rows_sycl_float(ctx, dst->src[0], dst->src[1], dst, (const sycl::half *)dst->src[0]->data,
src1_i32, (float *)dst->data, ctx.stream());
break;
case GGML_TYPE_F32:
get_rows_sycl_float(ctx, src0, src1, dst, src0_d, src1_i32, dst_d, stream);
get_rows_sycl_float(ctx, dst->src[0], dst->src[1], dst, (const float *)dst->src[0]->data,
src1_i32, (float *)dst->data, ctx.stream());
break;
case GGML_TYPE_Q4_0:
if (ctx.opt_feature.reorder && dst->op == GGML_OP_MUL_MAT) {
get_rows_sycl_reorder<QK4_0, QR4_0, dequantize_q4_0_reorder>(ctx, src0, src1, dst, src0_d, src1_i32, dst_d, stream);
get_rows_sycl_reorder<QK4_0, QR4_0, dequantize_q4_0_reorder>(ctx, dst->src[0], dst->src[1], dst, (const float *)dst->src[0]->data,
src1_i32, (float *)dst->data, ctx.stream());
} else {
get_rows_sycl<QK4_0, QR4_0, dequantize_q4_0>(ctx, src0, src1, dst, src0_d, src1_i32, dst_d, stream);
get_rows_sycl<QK4_0, QR4_0, dequantize_q4_0>(ctx, dst->src[0], dst->src[1], dst, (const float *)dst->src[0]->data,
src1_i32, (float *)dst->data, ctx.stream());
}
break;
case GGML_TYPE_Q4_1:
get_rows_sycl<QK4_1, QR4_1, dequantize_q4_1>(ctx, src0, src1, dst, src0_d, src1_i32, dst_d, stream);
get_rows_sycl<QK4_1, QR4_1, dequantize_q4_1>(ctx, dst->src[0], dst->src[1], dst, (const float *)dst->src[0]->data,
src1_i32, (float *)dst->data, ctx.stream());
break;
case GGML_TYPE_Q5_0:
get_rows_sycl<QK5_0, QR5_0, dequantize_q5_0>(ctx, src0, src1, dst, src0_d, src1_i32, dst_d, stream);
get_rows_sycl<QK5_0, QR5_0, dequantize_q5_0>(ctx, dst->src[0], dst->src[1], dst, (const float *)dst->src[0]->data,
src1_i32, (float *)dst->data, ctx.stream());
break;
case GGML_TYPE_Q5_1:
get_rows_sycl<QK5_1, QR5_1, dequantize_q5_1>(ctx, src0, src1, dst, src0_d, src1_i32, dst_d, stream);
get_rows_sycl<QK5_1, QR5_1, dequantize_q5_1>(ctx, dst->src[0], dst->src[1], dst, (const float *)dst->src[0]->data,
src1_i32, (float *)dst->data, ctx.stream());
break;
case GGML_TYPE_Q8_0:
get_rows_sycl<QK8_0, QR8_0, dequantize_q8_0>(ctx, src0, src1, dst, src0_d, src1_i32, dst_d, stream);
get_rows_sycl<QK8_0, QR8_0, dequantize_q8_0>(ctx, dst->src[0], dst->src[1], dst, (const float *)dst->src[0]->data,
src1_i32, (float *)dst->data, ctx.stream());
break;
default:
// TODO: k-quants
GGML_LOG_ERROR("%s: unsupported type: %s\n", __func__, ggml_type_name(src0->type));
GGML_LOG_ERROR("%s: unsupported type: %s\n", __func__, ggml_type_name(dst->src[0]->type));
GGML_ABORT("fatal error");
}
}