]> git.djapps.eu Git - pkg/ggml/sources/ggml/commit
ggml webgpu: faster matrix multiplication/matrix-vector multiplication (llama/17031)
authorReese Levine <redacted>
Sat, 8 Nov 2025 03:27:20 +0000 (19:27 -0800)
committerGeorgi Gerganov <redacted>
Sun, 9 Nov 2025 16:30:22 +0000 (18:30 +0200)
commite60fe4e3b0817f57b8bff159b4182bc1712b50c5
tree0f61048f31d4d2d83d04316f37f728593177b815
parented31fa7431a28ae83f4963aa3a09b747a9f84556
ggml webgpu: faster matrix multiplication/matrix-vector multiplication (llama/17031)

* Faster tensors (llama/8)

Add fast matrix and matrix/vector multiplication.

* Use map for shader replacements instead of pair of strings
src/ggml-webgpu/ggml-webgpu.cpp
src/ggml-webgpu/wgsl-shaders/embed_wgsl.py
src/ggml-webgpu/wgsl-shaders/mul_mat.tmpl.wgsl
src/ggml-webgpu/wgsl-shaders/mul_mat_decls.tmpl [new file with mode: 0644]
src/ggml-webgpu/wgsl-shaders/mul_mat_reg_tile.tmpl.wgsl [new file with mode: 0644]
src/ggml-webgpu/wgsl-shaders/mul_mat_subgroup_matrix.tmpl.wgsl [new file with mode: 0644]
src/ggml-webgpu/wgsl-shaders/mul_mat_vec.tmpl.wgsl [new file with mode: 0644]