Skip to content

Pull requests: google/XNNPACK

New pull request New
84 Open 6,243 Closed
84 Open 6,243 Closed
Author
Filter by author
Label
Filter by label
Use alt + click/return to exclude labels
or ⇧ + click/return for logical OR
Projects
Filter by project
Milestones
Filter by milestone
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

per channel quantized transpose conv
#6561 opened Jun 12, 2024 by copybara-service bot Loading…
Add weights packing which handles scales and biases also.
#6560 opened Jun 12, 2024 by copybara-service bot Loading…
Update test and benchmark generation for blockwise kr>2
#6557 opened Jun 12, 2024 by GregoryComer Loading…
4
AVX512 VNNI microkernels 14x16c8, 14x8c8, 28x16c4
#6554 opened Jun 12, 2024 by copybara-service bot Loading…
gemm-benchmark remove unused variables
#6553 opened Jun 12, 2024 by copybara-service bot Loading…
Add wrappers for the KleidiAI qp8-qc4w GEMM microkernels.
#6551 opened Jun 11, 2024 by copybara-service bot Loading…
Add copy_bias helper functions and use it to refactor packing.c.
#6550 opened Jun 11, 2024 by copybara-service bot Loading…
Hexagon HVX qs8-vadd-minmax
#6544 opened Jun 10, 2024 by ejparkqc Loading…
21
Enable AMX using pytorch/cpuinfo cpuinfo_has_x86_amx_int8()
#6543 opened Jun 10, 2024 by copybara-service bot Loading…
Add f32 rsum RVV implementation microkernels, tests and config changes for LMUL 2, 4 and 8
#6533 opened Jun 6, 2024 by KaustubhIMG Loading…
4
LLM decode benchmarks fill the cache with a predefined number of tokens before starting decoding.
#6531 opened Jun 6, 2024 by copybara-service bot Loading…
Expose MR for qd8-f32-qc4w.
#6530 opened Jun 6, 2024 by copybara-service bot Loading…
Hexagon HVX implementation of F32 gemm
#6507 opened May 30, 2024 by mallick-qc Loading…
1
Add QS8_QC8W GEMM/IGEMM microkernels for Wasm Relaxed Unsigned and Signed …
#6505 opened May 30, 2024 by fanchenkong1 Loading…
5
Fixes XNNPACK profiling information logged from the xnnpack_delegate. Currently the xnnpack node_index is passed to the profiler, this change corrects that to the TFLite node id.
#6499 opened May 29, 2024 by copybara-service bot Loading…
Avoid benchmark link errors in Bazel for platforms that don't have specializations for ~all kernels. Benchmarks should prefer to use the test/bench microkernels -- *not* the prod microkernels -- therefore they should only depend on test_mode dependencies for those that exist. Unfortunately, the Bazel build rules for benchmarks didn't really follow these guidelines, so new appropriate targets were added and depended on as appropriate.
#6490 opened May 28, 2024 by copybara-service bot Loading…
use 4x16s4 gemm kernel for meteor lake
#6485 opened May 28, 2024 by xujuntwt95329 • Draft
7
QS8-RSUM AVX2 use vpmaddubsw
#6482 opened May 27, 2024 by copybara-service bot Loading…
Extend the convert operator for the new qp8 packed per-row dynamic quantization.
#6479 opened May 27, 2024 by copybara-service bot Loading…
AVX2 qs8 rsum use vpmovsxbd to read bytes as ints
#6444 opened May 21, 2024 by copybara-service bot Loading…
Prototype to integrate SW optimizations for Arm® CPUs
#6436 opened May 17, 2024 by gmiodice Loading…
20
Use a better error bound for fp16 tests of the rsum microkernel.
#6431 opened May 16, 2024 by copybara-service bot Loading…
2
F32-RMINMAXSUM - add reduction sum to f32-rminmax
#6427 opened May 16, 2024 by copybara-service bot Loading…
1
Enable fp16 for risc-v
#6426 opened May 16, 2024 by copybara-service bot Loading…
Add a new x8-packq microkernel that packs and per-row dynamically quantizes fp32 to qp8.
#6424 opened May 15, 2024 by copybara-service bot Loading…
5
ProTip! Add no:assignee to see everything that’s not assigned.