Skip to content

sycl: add usage of enqueue_functions extension #14244

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 12 commits into from
Jun 20, 2025
Merged

sycl: add usage of enqueue_functions extension #14244

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
12 commits
Select commit Hold shift + click to select a range
6ddef2c
Add header and namespace to use enqueue_functions extension
s-Nick Jun 12, 2025
113e3d4
Convert submit and parallel_for to use new extension in convert.cpp
s-Nick Jun 12, 2025
b4c1f85
Convert submit and parallel_for to use extension in ggml-sycl.cpp
s-Nick Jun 12, 2025
02b13eb
Convert submit and parallel_for to use extension in gla.cpp
s-Nick Jun 12, 2025
7f45b08
Convert submit and parallel_for in mmq.cpp
s-Nick Jun 12, 2025
da7059f
Convert submit and parallel_for in mmvq.cpp
s-Nick Jun 12, 2025
fd682bb
Convert submit and parallel_for in remaining files
s-Nick Jun 12, 2025
b483442
Convert all simple parallel_for to nd_launch from enqueue_functions
s-Nick Jun 12, 2025
740cf89
Wrapping extension in general function
s-Nick Jun 16, 2025
7cd58a6
Formatting
s-Nick Jun 17, 2025
49bda6d
Address PR comments
s-Nick Jun 18, 2025
dc74c5b
Move namespace shortening outside function body.
s-Nick Jun 18, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
11 changes: 5 additions & 6 deletions ggml/src/ggml-sycl/binbcast.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -225,9 +225,9 @@ struct bin_bcast_sycl {
dpct::has_capability_or_fail(stream->get_device(),
{sycl::aspect::fp16});

stream->parallel_for(
sycl::nd_range<3>(sycl::range<3>(1, 1, block_num) *
sycl::range<3>(1, 1, block_size),
sycl_parallel_for(
stream,
sycl::nd_range<3>(sycl::range<3>(1, 1, block_num) * sycl::range<3>(1, 1, block_size),
sycl::range<3>(1, 1, block_size)),
[=](sycl::nd_item<3> item_ct1) {
k_bin_bcast_unravel<bin_op>(
Expand All @@ -246,9 +246,8 @@ struct bin_bcast_sycl {
dpct::has_capability_or_fail(stream->get_device(),
{sycl::aspect::fp16});

stream->parallel_for(
sycl::nd_range<3>(block_nums * block_dims, block_dims),
[=](sycl::nd_item<3> item_ct1) {
sycl_parallel_for(
stream, sycl::nd_range<3>(block_nums * block_dims, block_dims), [=](sycl::nd_item<3> item_ct1) {
k_bin_bcast<bin_op>(src0_dd, src1_dd, dst_dd, ne0, ne1,
ne2, ne3, ne10, ne11, ne12, ne13,
s1, s2, s3, s01, s02, s03, s11, s12, s13,
Expand Down
69 changes: 28 additions & 41 deletions ggml/src/ggml-sycl/concat.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -89,33 +89,24 @@ static void concat_f32_sycl(const float *x, const float *y, float *dst,
sycl::range<3> gridDim(ne2, ne1, num_blocks);
switch (dim) {
case 0:
stream->parallel_for(
sycl::nd_range<3>(gridDim *
sycl::range<3>(1, 1, SYCL_CONCAT_BLOCK_SIZE),
sycl::range<3>(1, 1, SYCL_CONCAT_BLOCK_SIZE)),
[=](sycl::nd_item<3> item_ct1) {
concat_f32_dim0(x, y, dst, ne0, ne00, item_ct1);
});
break;
sycl_parallel_for(stream,
sycl::nd_range<3>(gridDim * sycl::range<3>(1, 1, SYCL_CONCAT_BLOCK_SIZE),
sycl::range<3>(1, 1, SYCL_CONCAT_BLOCK_SIZE)),
[=](sycl::nd_item<3> item_ct1) { concat_f32_dim0(x, y, dst, ne0, ne00, item_ct1); });
break;
case 1:
stream->parallel_for(
sycl::nd_range<3>(gridDim *
sycl::range<3>(1, 1, SYCL_CONCAT_BLOCK_SIZE),
sycl::range<3>(1, 1, SYCL_CONCAT_BLOCK_SIZE)),
[=](sycl::nd_item<3> item_ct1) {
concat_f32_dim1(x, y, dst, ne0, ne01, item_ct1);
});
break;
sycl_parallel_for(stream,
sycl::nd_range<3>(gridDim * sycl::range<3>(1, 1, SYCL_CONCAT_BLOCK_SIZE),
sycl::range<3>(1, 1, SYCL_CONCAT_BLOCK_SIZE)),
[=](sycl::nd_item<3> item_ct1) { concat_f32_dim1(x, y, dst, ne0, ne01, item_ct1); });
break;
// dim >=2 will be dispatched to the default path
default:
stream->parallel_for(
sycl::nd_range<3>(gridDim *
sycl::range<3>(1, 1, SYCL_CONCAT_BLOCK_SIZE),
sycl::range<3>(1, 1, SYCL_CONCAT_BLOCK_SIZE)),
[=](sycl::nd_item<3> item_ct1) {
concat_f32_dim2(x, y, dst, ne0, ne02, item_ct1);
});
break;
sycl_parallel_for(stream,
sycl::nd_range<3>(gridDim * sycl::range<3>(1, 1, SYCL_CONCAT_BLOCK_SIZE),
sycl::range<3>(1, 1, SYCL_CONCAT_BLOCK_SIZE)),
[=](sycl::nd_item<3> item_ct1) { concat_f32_dim2(x, y, dst, ne0, ne02, item_ct1); });
break;
}
}

Expand All @@ -129,33 +120,29 @@ static void concat_f32_sycl_non_cont(
int64_t ne2, int64_t ne3, uint64_t nb0, uint64_t nb1, uint64_t nb2,
uint64_t nb3, int32_t dim) {
sycl::range<3> gridDim(ne3, ne2, ne1);
stream->parallel_for(
sycl::nd_range<3>(gridDim, sycl::range<3>(1, 1, 1)),
[=](sycl::nd_item<3> item_ct1) {
int64_t i3 = item_ct1.get_group(0);
int64_t i2 = item_ct1.get_group(1);
int64_t i1 = item_ct1.get_group(2);
sycl_parallel_for(stream, sycl::nd_range<3>(gridDim, sycl::range<3>(1, 1, 1)), [=](sycl::nd_item<3> item_ct1) {
int64_t i3 = item_ct1.get_group(0);
int64_t i2 = item_ct1.get_group(1);
int64_t i1 = item_ct1.get_group(2);

int64_t o[4] = {0, 0, 0, 0};
o[dim] = dim == 0 ? ne00 : (dim == 1 ? ne01 : (dim == 2 ? ne02 : ne03));
int64_t o[4] = { 0, 0, 0, 0 };
o[dim] = dim == 0 ? ne00 : (dim == 1 ? ne01 : (dim == 2 ? ne02 : ne03));

const float *x;
const float * x;

for (int i0 = item_ct1.get_local_id(2); i0 < ne0;
i0 += item_ct1.get_local_range(2)) {
for (int i0 = item_ct1.get_local_id(2); i0 < ne0; i0 += item_ct1.get_local_range(2)) {
if (i0 < ne00 && i1 < ne01 && i2 < ne02 && i3 < ne03) {
x = (const float *)(src0 + (i3)*nb03 + (i2)*nb02 + (i1)*nb01 +
(i0)*nb00);
x = (const float *) (src0 + (i3) *nb03 + (i2) *nb02 + (i1) *nb01 + (i0) *nb00);
} else {
x = (const float *)(src1 + (i3 - o[3]) * nb13 + (i2 - o[2]) * nb12 +
(i1 - o[1]) * nb11 + (i0 - o[0]) * nb10);
x = (const float *) (src1 + (i3 - o[3]) * nb13 + (i2 - o[2]) * nb12 + (i1 - o[1]) * nb11 +
(i0 - o[0]) * nb10);
}

float *y = (float *)(dst + i3 * nb3 + i2 * nb2 + i1 * nb1 + i0 * nb0);

*y = *x;
}
});
}
});
}

void ggml_sycl_op_concat(ggml_backend_sycl_context & ctx, ggml_tensor *dst) {
Expand Down
14 changes: 4 additions & 10 deletions ggml/src/ggml-sycl/conv.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -59,16 +59,10 @@ static void conv_transpose_1d_f32_f32_sycl(
const int num_blocks = (output_size + SYCL_CONV_TRANPOSE_1D_BLOCK_SIZE - 1) / SYCL_CONV_TRANPOSE_1D_BLOCK_SIZE;
const sycl::range<3> block_dims(1, 1, SYCL_CONV_TRANPOSE_1D_BLOCK_SIZE);
const sycl::range<3> block_nums(1, 1, num_blocks);
stream->parallel_for(
sycl::nd_range<3>(
block_nums * block_dims, block_dims),
[=](sycl::nd_item<3> item_ct1) {
conv_transpose_1d_kernel(
s0, output_size,
src0_ne0, src0_ne1, src0_ne2,
src1_ne0, dst_ne0,
src0, src1, dst, item_ct1);
});
sycl_parallel_for(stream, sycl::nd_range<3>(block_nums * block_dims, block_dims), [=](sycl::nd_item<3> item_ct1) {
conv_transpose_1d_kernel(s0, output_size, src0_ne0, src0_ne1, src0_ne2, src1_ne0, dst_ne0, src0, src1, dst,
item_ct1);
});
}

void ggml_sycl_op_conv_transpose_1d(ggml_backend_sycl_context & ctx, ggml_tensor *dst) {
Expand Down
Loading
Loading