Skip to content

vulkan : add fp16 support for the conv_2d kernel #14872

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 4 commits into from
Jul 27, 2025
Merged

vulkan : add fp16 support for the conv_2d kernel #14872

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
f0f7b73
vulkan : add fp16 support for the conv_2d kernel
Green-Sky Jul 25, 2025
74fc81f
fix reported type
Green-Sky Jul 25, 2025
4fa0331
add f16 to conv_2d testing
Green-Sky Jul 25, 2025
d6c0382
weaken conv2d test error threshold
Green-Sky Jul 27, 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
22 changes: 18 additions & 4 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 @@ -484,6 +484,7 @@ struct vk_device_struct {
vk_pipeline pipeline_rwkv_wkv7_f32;
vk_pipeline pipeline_opt_step_adamw_f32;
vk_pipeline pipeline_conv2d_f32;
vk_pipeline pipeline_conv2d_f16_f32;
vk_pipeline pipeline_conv2d_dw_whcn_f32;
vk_pipeline pipeline_conv2d_dw_cwhn_f32;

Expand Down Expand Up @@ -3074,12 +3075,21 @@ static void ggml_vk_load_shaders(vk_device& device) {
device, device->pipeline_conv2d_f32, "conv2d_f32", conv2d_f32_len, conv2d_f32_data, "main", 3,
sizeof(vk_op_conv2d_push_constants), { conv2d_BS_K, conv2d_BS_NPQ, 1 },
{ conv2d_WG_SIZE, conv2d_BS_K, conv2d_BS_CRS, conv2d_BS_NPQ, conv2d_TS_K, use_collectives }, 1, true, true);
ggml_vk_create_pipeline(
device, device->pipeline_conv2d_f16_f32, "conv2d_f16_f32", conv2d_f16_f32_len, conv2d_f16_f32_data, "main", 3,
sizeof(vk_op_conv2d_push_constants), { conv2d_BS_K, conv2d_BS_NPQ, 1 },
{ conv2d_WG_SIZE, conv2d_BS_K, conv2d_BS_CRS, conv2d_BS_NPQ, conv2d_TS_K, use_collectives }, 1, true, true);
} else {
ggml_vk_create_pipeline(
device, device->pipeline_conv2d_f32, "conv2d_f32", conv2d_f32_len, conv2d_f32_data, "main", 3,
sizeof(vk_op_conv2d_push_constants), { conv2d_BS_K, conv2d_BS_NPQ, 1 },
{ conv2d_WG_SIZE, conv2d_BS_K, conv2d_BS_CRS, conv2d_BS_NPQ, conv2d_TS_K, use_collectives }, 1, true,
false);
ggml_vk_create_pipeline(
device, device->pipeline_conv2d_f16_f32, "conv2d_f16_f32", conv2d_f16_f32_len, conv2d_f16_f32_data, "main", 3,
sizeof(vk_op_conv2d_push_constants), { conv2d_BS_K, conv2d_BS_NPQ, 1 },
{ conv2d_WG_SIZE, conv2d_BS_K, conv2d_BS_CRS, conv2d_BS_NPQ, conv2d_TS_K, use_collectives }, 1, true,
false);
}

ggml_vk_create_pipeline(device, device->pipeline_conv2d_dw_whcn_f32, "conv2d_dw_whcn_f32", conv2d_dw_whcn_f32_len, conv2d_dw_whcn_f32_data, "main", 3, sizeof(vk_op_conv2d_dw_push_constants), {512, 1, 1}, {}, 1);
Expand Down Expand Up @@ -6958,9 +6968,13 @@ static vk_pipeline ggml_vk_op_get_pipeline(ggml_backend_vk_context * ctx, const
}
return nullptr;
case GGML_OP_CONV_2D:
if (src0->type == GGML_TYPE_F32 && src1->type == GGML_TYPE_F32 && dst->type == GGML_TYPE_F32 &&
if (src1->type == GGML_TYPE_F32 && dst->type == GGML_TYPE_F32 &&
ggml_is_contiguous(src0) && ggml_is_contiguous(src1) && ggml_is_contiguous(dst)) {
return ctx->device->pipeline_conv2d_f32;
if (src0->type == GGML_TYPE_F32) {
return ctx->device->pipeline_conv2d_f32;
} else if (src0->type == GGML_TYPE_F16) {
return ctx->device->pipeline_conv2d_f16_f32;
}
}
return nullptr;
case GGML_OP_CONV_2D_DW:
Expand Down Expand Up @@ -8178,13 +8192,13 @@ static void ggml_vk_pool_2d(ggml_backend_vk_context * ctx, vk_context& subctx, c

static void ggml_vk_conv_2d(ggml_backend_vk_context * ctx, vk_context & subctx, const ggml_tensor * src0,
const ggml_tensor * src1, ggml_tensor * dst, bool dryrun = false) {
GGML_ASSERT(src0->type == GGML_TYPE_F32);
GGML_ASSERT(src0->type == GGML_TYPE_F32 || src0->type == GGML_TYPE_F16);
GGML_ASSERT(src1->type == GGML_TYPE_F32);
GGML_ASSERT(dst->type == GGML_TYPE_F32);

GGML_TENSOR_BINARY_OP_LOCALS

GGML_ASSERT(nb00 == sizeof(float));
GGML_ASSERT(nb00 == sizeof(float) || nb00 == sizeof(ggml_fp16_t));
GGML_ASSERT(nb10 == sizeof(float));
GGML_ASSERT(nb0 == sizeof(float));

Expand Down
1 change: 1 addition & 0 deletions ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -656,6 +656,7 @@ void process_shaders() {
string_to_spv("opt_step_adamw_f32", "opt_step_adamw.comp", merge_maps(base_dict, {{"A_TYPE", "float"}}));

string_to_spv("conv2d_f32", "conv2d_mm.comp", {{"A_TYPE", "float"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}, {"USE_COLLECTIVES", "1"}});
string_to_spv("conv2d_f16_f32", "conv2d_mm.comp", {{"A_TYPE", "float16_t"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}, {"USE_COLLECTIVES", "1"}});

string_to_spv("conv2d_dw_whcn_f32", "conv2d_dw.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}, {"WHCN", "1"}}));
string_to_spv("conv2d_dw_cwhn_f32", "conv2d_dw.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}, {"CWHN", "1"}}));
Expand Down
Loading