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

fix format

Browse source
This commit is contained in:
abhilash1910 2024-03-25 04:48:28 -07:00
parent 36c7f02abc
commit 7ea2e1574f
1 changed files with 8 additions and 11 deletions
Download patch file Download diff file Expand all files Collapse all files

19
ggml-sycl.cpp
View file

@ -15355,18 +15355,16 @@ static void ggml_sycl_mul_mat(const ggml_tensor * src0, const ggml_tensor * src1
min_compute_capability = g_device_caps[i].cc; min_compute_capability = g_device_caps[i].cc;
} }
} }
bool use_dequantize_mul_mat_vec = (ggml_is_quantized(src0->type) || src0->type == GGML_TYPE_F16) bool use_dequantize_mul_mat_vec = (ggml_is_quantized(src0->type) || src0->type == GGML_TYPE_F16)
&& src1->type == GGML_TYPE_F32 && dst->type == GGML_TYPE_F32 && src1->type == GGML_TYPE_F32 && dst->type == GGML_TYPE_F32
&& src0->ne[0] % GGML_SYCL_DMMV_X == 0 && src1->ne[1] == 1; && src0->ne[0] % GGML_SYCL_DMMV_X == 0 && src1->ne[1] == 1;
bool use_mul_mat_vec_q = min_compute_capability >= VER_4VEC && ggml_is_quantized(src0->type) bool use_mul_mat_vec_q = min_compute_capability >= VER_4VEC && ggml_is_quantized(src0->type)
&& src1->type == GGML_TYPE_F32 && dst->type == GGML_TYPE_F32 && src1->type == GGML_TYPE_F32 && dst->type == GGML_TYPE_F32
&& src1->ne[1] <= XMX_MAX_BATCH_SIZE; && src1->ne[1] <= XMX_MAX_BATCH_SIZE;
bool use_mul_mat_q = ggml_sycl_supports_mmq(src0->type) && src1->type == GGML_TYPE_F32 bool use_mul_mat_q = ggml_sycl_supports_mmq(src0->type) && src1->type == GGML_TYPE_F32
&& dst->type == GGML_TYPE_F32; && dst->type == GGML_TYPE_F32;
#ifdef SYCL_USE_XMX #ifdef SYCL_USE_XMX
const bool use_xmx = true; const bool use_xmx = true;
#else #else
@ -15405,13 +15403,12 @@ static void ggml_sycl_mul_mat(const ggml_tensor * src0, const ggml_tensor * src1
} else if (use_mul_mat_vec_q){ } else if (use_mul_mat_vec_q){
// use ggml_sycl_op_mul_mat_vec_q // use ggml_sycl_op_mul_mat_vec_q
// GGML_SYCL_DEBUG("ggml_sycl_mul_mat ggml_sycl_op_mul_mat_vec_q path\n"); // GGML_SYCL_DEBUG("ggml_sycl_mul_mat ggml_sycl_op_mul_mat_vec_q path\n");
ggml_sycl_op_mul_mat(src0, src1, dst, ggml_sycl_op_mul_mat_vec_q, true); ggml_sycl_op_mul_mat(src0, src1, dst, ggml_sycl_op_mul_mat_vec_q, true);
} else if (use_mul_mat_q){ } else if (use_mul_mat_q){
if (use_xmx && min_compute_capability >= VER_GEN9 && src1->ne[1] > XMX_MAX_BATCH_SIZE) { if (use_xmx && min_compute_capability >= VER_GEN9 && src1->ne[1] > XMX_MAX_BATCH_SIZE) {
use_mul_mat_q = false; use_mul_mat_q = false;
} }
if (use_mul_mat_q) { if (use_mul_mat_q) {
// GGML_SYCL_DEBUG("ggml_sycl_mul_mat ggml_sycl_op_mul_mat_q path\n"); // GGML_SYCL_DEBUG("ggml_sycl_mul_mat ggml_sycl_op_mul_mat_q path\n");