Skip to content

vulkan: Skip syncing for prealloc_y when it is reused #15544

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 1 commit into from
Aug 30, 2025
Merged

vulkan: Skip syncing for prealloc_y when it is reused #15544

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
35 changes: 15 additions & 20 deletions ggml/src/ggml-vulkan/ggml-vulkan.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5724,11 +5724,6 @@ static void ggml_vk_mul_mat_q_f16(ggml_backend_vk_context * ctx, vk_context& sub
ggml_vk_sync_buffers(ctx, subctx);
}
}
if (y_non_contig || quantize_y) {
if (ctx->prealloc_y_need_sync) {
ggml_vk_sync_buffers(ctx, subctx);
}
}

if (x_non_contig) {
ggml_vk_cpy_to_contiguous(ctx, subctx, to_fp16_vk_0, src0, { d_Qx, qx_buf_offset, VK_WHOLE_SIZE }, { d_X, 0, VK_WHOLE_SIZE });
Expand All @@ -5740,6 +5735,9 @@ static void ggml_vk_mul_mat_q_f16(ggml_backend_vk_context * ctx, vk_context& sub
if (y_non_contig) {
if (ctx->prealloc_y_last_pipeline_used != to_fp16_vk_1.get() ||
ctx->prealloc_y_last_tensor_used != src1) {
if (ctx->prealloc_y_need_sync) {
ggml_vk_sync_buffers(ctx, subctx);
}
ggml_vk_cpy_to_contiguous(ctx, subctx, to_fp16_vk_1, src1, { d_Qy, qy_buf_offset, VK_WHOLE_SIZE }, { d_Y, 0, VK_WHOLE_SIZE });
ctx->prealloc_y_last_pipeline_used = to_fp16_vk_1.get();
ctx->prealloc_y_last_tensor_used = src1;
Expand All @@ -5748,6 +5746,9 @@ static void ggml_vk_mul_mat_q_f16(ggml_backend_vk_context * ctx, vk_context& sub
if (quantize_y) {
if (ctx->prealloc_y_last_pipeline_used != to_q8_1.get() ||
ctx->prealloc_y_last_tensor_used != src1) {
if (ctx->prealloc_y_need_sync) {
ggml_vk_sync_buffers(ctx, subctx);
}
ggml_vk_quantize_q8_1(ctx, subctx, { d_Qy, qy_buf_offset, VK_WHOLE_SIZE }, { d_Y, 0, VK_WHOLE_SIZE }, y_ne * ne12 * ne13);
ctx->prealloc_y_last_pipeline_used = to_q8_1.get();
ctx->prealloc_y_last_tensor_used = src1;
Expand Down Expand Up @@ -5932,11 +5933,6 @@ static void ggml_vk_mul_mat_vec_q_f16(ggml_backend_vk_context * ctx, vk_context&
ggml_vk_sync_buffers(ctx, subctx);
}
}
if (y_non_contig) {
if (ctx->prealloc_y_need_sync) {
ggml_vk_sync_buffers(ctx, subctx);
}
}

if (x_non_contig) {
GGML_ASSERT(x_sz == ggml_vk_align_size(ggml_type_size(src0->type) * x_ne, ctx->device->properties.limits.minStorageBufferOffsetAlignment));
Expand All @@ -5946,6 +5942,9 @@ static void ggml_vk_mul_mat_vec_q_f16(ggml_backend_vk_context * ctx, vk_context&
GGML_ASSERT(y_sz == ggml_type_size(src1->type) * y_ne);
if (ctx->prealloc_y_last_pipeline_used != to_fp16_vk_1.get() ||
ctx->prealloc_y_last_tensor_used != src1) {
if (ctx->prealloc_y_need_sync) {
ggml_vk_sync_buffers(ctx, subctx);
}
ggml_vk_cpy_to_contiguous(ctx, subctx, to_fp16_vk_1, src1, { d_Qy, qy_buf_offset, VK_WHOLE_SIZE }, { d_Y, 0, VK_WHOLE_SIZE });
ctx->prealloc_y_last_pipeline_used = to_fp16_vk_1.get();
ctx->prealloc_y_last_tensor_used = src1;
Expand Down Expand Up @@ -6379,11 +6378,6 @@ static void ggml_vk_mul_mat_id_q_f16(ggml_backend_vk_context * ctx, vk_context&
ggml_vk_sync_buffers(ctx, subctx);
}
}
if (y_non_contig) {
if (ctx->prealloc_y_need_sync) {
ggml_vk_sync_buffers(ctx, subctx);
}
}

if (x_non_contig) {
ggml_vk_cpy_to_contiguous(ctx, subctx, to_fp16_vk_0, src0, { d_Qx, qx_buf_offset, VK_WHOLE_SIZE }, { d_X, 0, VK_WHOLE_SIZE });
Expand All @@ -6396,6 +6390,9 @@ static void ggml_vk_mul_mat_id_q_f16(ggml_backend_vk_context * ctx, vk_context&
if (y_non_contig) {
if (ctx->prealloc_y_last_pipeline_used != to_fp16_vk_1.get() ||
ctx->prealloc_y_last_tensor_used != src1) {
if (ctx->prealloc_y_need_sync) {
ggml_vk_sync_buffers(ctx, subctx);
}
ggml_vk_cpy_to_contiguous(ctx, subctx, to_fp16_vk_1, src1, { d_Qy, qy_buf_offset, VK_WHOLE_SIZE }, { d_Y, 0, VK_WHOLE_SIZE });
ctx->prealloc_y_last_pipeline_used = to_fp16_vk_1.get();
ctx->prealloc_y_last_tensor_used = src1;
Expand Down Expand Up @@ -6593,11 +6590,6 @@ static void ggml_vk_mul_mat_vec_id_q_f16(ggml_backend_vk_context * ctx, vk_conte
ggml_vk_sync_buffers(ctx, subctx);
}
}
if (y_non_contig) {
if (ctx->prealloc_y_need_sync) {
ggml_vk_sync_buffers(ctx, subctx);
}
}

if (x_non_contig) {
GGML_ASSERT(x_sz == ggml_vk_align_size(ggml_type_size(src0->type) * x_ne, ctx->device->properties.limits.minStorageBufferOffsetAlignment));
Expand All @@ -6607,6 +6599,9 @@ static void ggml_vk_mul_mat_vec_id_q_f16(ggml_backend_vk_context * ctx, vk_conte
GGML_ASSERT(y_sz == ggml_type_size(src1->type) * y_ne);
if (ctx->prealloc_y_last_pipeline_used != to_fp16_vk_1.get() ||
ctx->prealloc_y_last_tensor_used != src1) {
if (ctx->prealloc_y_need_sync) {
ggml_vk_sync_buffers(ctx, subctx);
}
ggml_vk_cpy_to_contiguous(ctx, subctx, to_fp16_vk_1, src1, { d_Qy, qy_buf_offset, VK_WHOLE_SIZE }, { d_Y, 0, VK_WHOLE_SIZE });
ctx->prealloc_y_last_pipeline_used = to_fp16_vk_1.get();
ctx->prealloc_y_last_tensor_used = src1;
Expand Down
Loading