vulkan: add col2im_1d op#24425
Merged
ServeurpersoCom merged 4 commits intoJun 16, 2026
Merged
Conversation
github-actions
Bot
added
Vulkan
jeffbolznv
reviewed
Jun 10, 2026
0cc4m
approved these changes
Jun 15, 2026
jeffbolznv
approved these changes
Jun 15, 2026
papamoose
pushed a commit
to papamoose/llama.cpp
that referenced
this pull request
Jun 27, 2026
* 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
adrianhoehne
pushed a commit
to adrianhoehne/llama.cpp
that referenced
this pull request
Jul 5, 2026
* 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 file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Overview
vulkan: add GGML_OP_COL2IM_1D
Vulkan backend follow-up to the CPU op ( #24206 ), same formulation: a gather shader, one invocation per output, each reading only the ceil(K/stride) columns that scatter into it. F32 / F16 / BF16, the BF16 path stores as uint16_t and converts through bf16_to_fp32 so it runs even on devices without native bf16.
A 2D dispatch maps invocations directly to (t_out, oc), so there is no flat index decomposition and no div/mod to begin with, unlike the CUDA side.
Additional information
Validated against the test-backend-ops grid merged with the CPU op, zero additional test code: 33/33 on Vulkan0 across the eight geometries and three types, plus the three perf entries. Wiring sits next to conv_transpose_1d: shader registration in vulkan-shaders-gen, pipelines, push constants, dispatch and the supports_op entry.
Optimization (2nd commit):
The shader now walks the bounded t_in range instead of scanning all K with a per step modulo, so it touches only the ceil(K/stride) useful columns. Measured around 7x on the cache resident K heavy vocoder stage shapes (F32 and F16), bringing Vulkan to parity with CUDA there; on the DRAM bound long shape it is a no op, as expected.
Requirements