Skip to content

Pull requests: google/XNNPACK

New pull request New
116 Open 6,958 Closed
116 Open 6,958 Closed
Author
Filter by author
Loading
Label
Filter by label
Loading
Use alt + click/return to exclude labels
or ⇧ + click/return for logical OR
Projects
Filter by project
Loading
Milestones
Filter by milestone
Loading
Reviews
Filter by reviews
No reviews Review required Approved review Changes requested
Assignee
Filter by who’s assigned
Sort
Sort by
Newest Oldest Most commented Least commented Recently updated Least recently updated Best match
Most reactions
👍 👎 😄 🎉 😕 ❤️ 🚀 👀

Pull requests list

QS8 PACKW do 2 columns at a time to increase read size to 128 bits
#7320 opened Oct 22, 2024 by copybara-service bot Loading…
Update versions for cpuinfo and com_google_googletest.
#7319 opened Oct 22, 2024 by copybara-service bot Loading…
Refactor binary-elementwise-nd.cc and binary.cc to reduce the number of test suites via type erasure. This should make the test easier to read and maintain.
#7315 opened Oct 22, 2024 by copybara-service bot Loading…
std::numeric_limits<> always produces zero for the xnn_float16 type, meaning that we always took the "expected to overflow" path for f16 tests in binary-elementwise-nd.cc. This fixes that incorrect usage, and loosens the tolerance for SquaredDifference for f16, which (based on testing) requires a much looser tolerance than other tests.
#7314 opened Oct 22, 2024 by copybara-service bot Loading…
Fix BMM benchmark and add TRANSPOSE_B and threads
#7313 opened Oct 22, 2024 by copybara-service bot Loading…
Enable AVXVNNI microkernels for mobile platforms
#7311 opened Oct 22, 2024 by soniravi31 Loading…
Removed global average pooling from average-pooling-nhwc.
#7304 opened Oct 21, 2024 by copybara-service bot Loading…
1
Simplify kc remainder processing for qs8-x8c8 avxvnni
#7297 opened Oct 18, 2024 by xujuntwt95329 Loading…
Set requantization scale upper bound to 1.0
#7296 opened Oct 18, 2024 by copybara-service bot Loading…
Add qs8 c4 microkernels for wasmsdot
#7290 opened Oct 17, 2024 by yolanda15 Loading…
2
Add initial bits for RNDNU16 requantization.
#7286 opened Oct 17, 2024 by copybara-service bot Loading…
Put back missing packing optimization.
#7284 opened Oct 16, 2024 by copybara-service bot Loading…
Use the XNNPack static_reduce operation to delegate SUM and MEAN.
#7282 opened Oct 16, 2024 by copybara-service bot Loading…
Add Quantized version of the Attention benchmark to bench/models. (Note that the BMM and Softmax operations are still done in fp32, necessitating conversion to/from qs8; this appears to be the precedent in existing tests.)
#7273 opened Oct 15, 2024 by copybara-service bot Loading…
Use qs8-packw microkernels for tests
#7269 opened Oct 15, 2024 by copybara-service bot Loading…
Integration of Kleidi F32 SME kernels
#7260 opened Oct 14, 2024 by copybara-service bot Loading…
8
WIP - qc4w packw for avxvnni
#7259 opened Oct 13, 2024 by copybara-service bot Loading…
Replaced GlobalAveragePooling with static_mean.
#7253 opened Oct 10, 2024 by copybara-service bot Loading…
1
Link to all KleidiAI microkernels to prevent annoying build failures
#7249 opened Oct 10, 2024 by copybara-service bot Loading…
Add Emscripten support for Bazel benchmark builds
#7242 opened Oct 9, 2024 by SPY Loading…
Support qs8 5x8c8 wasmdot microkernels
#7240 opened Oct 9, 2024 by yolanda15 Loading…
4
QS8 / QU8 PReLU microkernels
#7235 opened Oct 8, 2024 by soniravi31 Loading…
2
Kleidi introduce new 16x4 kernels
#7211 opened Oct 1, 2024 by mcr229 Loading…
11
Replaced gavgpool and gsumpool with static_reduce.
#7202 opened Sep 30, 2024 by copybara-service bot Loading…
1
Xnn f32 reduce window
#7192 opened Sep 26, 2024 by LokeshReddyOVS-MCW Loading…
15
ProTip! Mix and match filters to narrow down what you’re looking for.