vulkan: add col2im_1d op (#24425)

* vulkan: add GGML_OP_COL2IM_1D, follow-up to the CPU op

* vulkan: col2im_1d bounded gather loop instead of full-K scan with modulo

* vulkan: col2im_1d address review from @jeffbolznv

* vulkan: col2im_1d return nullptr for unsupported types, address review from @0cc4m
This commit is contained in:
Pascal
2026-06-16 06:34:43 +02:00
committed by GitHub
parent 7dad2f1a17
commit ad39ccaa19
3 changed files with 133 additions and 0 deletions
@@ -1003,6 +1003,9 @@ void process_shaders() {
string_to_spv("timestep_embedding_f32", "timestep_embedding.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "float"}}));
string_to_spv("conv_transpose_1d_f32", "conv_transpose_1d.comp", {{"A_TYPE", "float"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}});
string_to_spv("col2im_1d_f32", "col2im_1d.comp", {{"DATA_A_F32", "1"}, {"A_TYPE", "float"}, {"D_TYPE", "float"}});
string_to_spv("col2im_1d_f16", "col2im_1d.comp", {{"DATA_A_F16", "1"}, {"A_TYPE", "float16_t"}, {"D_TYPE", "float16_t"}});
string_to_spv("col2im_1d_bf16", "col2im_1d.comp", {{"DATA_A_BF16", "1"}, {"A_TYPE", "uint16_t"}, {"D_TYPE", "uint16_t"}});
string_to_spv("snake_f32", "snake.comp", {{"DATA_A_F32", "1"}, {"A_TYPE", "float"}, {"D_TYPE", "float"}});
string_to_spv("snake_f16", "snake.comp", {{"DATA_A_F16", "1"}, {"A_TYPE", "float16_t"}, {"D_TYPE", "float16_t"}});