onnxruntime
[js/webgpu] Provide a naive vectorized matmul algorithm
#18758
Merged

[js/webgpu] Provide a naive vectorized matmul algorithm #18758

guschmue merged 13 commits into microsoft:main from qjia7:opt_matmul_vec
qjia7
qjia7 [js/webgpu] Optimize the naive nhwc conv
5245593d
qjia7 output multiple col data
772321e2
qjia7 support maxInComponent
9676e23c
qjia7 [js/webgpu] Optimize matmul with large batch size
7c4a148b
qjia7 add components support for output
84e0b54d
qjia7 add A components
4f8c09d7
qjia7 output multiple results per thread
2bb39744
qjia7 fix rebase errors
26376073
qjia7 go to naive matmul path for special shape
88078ecd
qjia7 nits
ea9144f1
qjia7 remove the unused file
82b16b3e
qjia7 Use uniforms
dff4153f
qjia7
qjia7 use snake case instead of camel case
f8161ab4
guschmue guschmue added ep:WebGPU
guschmue
guschmue approved these changes on 2023-12-12
fs-eire
fs-eire approved these changes on 2023-12-13
fs-eire
fs-eire
fs-eire
azure-pipelines
azure-pipelines
azure-pipelines
fs-eire
fs-eire commented on 2023-12-13
guschmue guschmue merged b30e721d into main 2 years ago
qjia7 qjia7 deleted the opt_matmul_vec branch 2 years ago

Login to write a write a comment.

Login via GitHub

Reviewers
Assignees
No one assigned
Labels
Milestone