mirror_xnnpack

Зеркало из https://github.com/google/xnnpack.git

Форк

Quentin Khan
Pass down the original kernel and bias buffers when creating a fully connected operation.
2 часа назад
9e77b69

Описание

High-efficiency floating-point neural network inference operators for mobile, server, and Web

convolutional-neural-network

convolutional-neural-networks

cpu

inference

inference-optimization

matrix-multiplication

Языки

C78,6%
C++13,3%
Assembly5%
Shell1,2%
Starlark0,7%
CMake0,5%
Остальные0,7%

Added sum_squared reductions to arm.

11 дней назад

bench

Templated neondot int2 GEMMs.

6 дней назад

build_config

Make XNNPACK not use KleidiAI on Arm64 Windows

4 месяца назад

cmake

Added scalar kernels to qc2w gemm-config. Fixed the problem with missing row_sum variant in operator-run - this was only discovered when running subgraph and operator tests with 1x2 scalar kernel in the config.

16 часов назад

doc

Rename all non self-contained header files from `.h` to `.inc` to avoid confusing our tooling into trying to parse them in isolation.

7 месяцев назад

gemm_compiler

Add taller fma microkernels

4 месяца назад

gen

16 часов назад

include

Add a function to get fingerprints by index for testing.

день назад

scripts

Templated neondot int2 GEMMs.

6 дней назад

src

Pass down the original kernel and bias buffers when creating a fully connected operation.

2 часа назад

test

Attepmted to fix 2-bit fully-connected operator flakyness.

3 часа назад

third_party

tools

build: allow generated files to be read by GN and Bazel

15 дней назад

ynnpack

Add stencil_copy helper function

3 часа назад

.bazelrc

Set -ffp-contract=off

14 дней назад

.clang-format

.gitignore

[KleidiAI] new SDOT kernels for qb4

6 месяцев назад

BUILD.bazel

[roll-forward] Fingerprint gemm kernel packing operations.

22 дня назад

BUILD.md

build: add a basic DEPS file for GN

19 дней назад

CMakeLists.txt

Set -ffp-contract=off

14 дней назад

CONTRIBUTING.md

DEPS

build: add a basic DEPS file for GN

19 дней назад

LICENSE

MODULE.bazel

Merge pull request #9178 from qualcomm:sme1/f32_int8_int8

19 дней назад

README.md

Fix typo in README.md

6 месяцев назад

WORKSPACE

Update Slinky version.

19 дней назад

WORKSPACE.bzlmod

build_defs.bzl

Remove -Wimplicit-fallthrough from non-GCC compilers

месяц назад

build_params.bzl

Add YNNPACK to XNNPACK

2 месяца назад

build_srcs.bzl

Merge remote-tracking branch 'google/master' into pf32-igemm-sme1

14 дней назад

emscripten.bzl

generated_file.bzl

Add YNNPACK to XNNPACK

2 месяца назад

preamble.js.lds

register_extension_info.bzl

Add YNNPACK to XNNPACK

2 месяца назад

XNNPACK Supported Architectures Operator Coverage Performance Minimum build requirements Publications Ecosystem Acknowledgements

README.md

Описание

High-efficiency floating-point neural network inference operators for mobile, server, and Web

convolutional-neural-network

convolutional-neural-networks

cpu

inference

inference-optimization

matrix-multiplication

Языки

C78,6%
C++13,3%
Assembly5%
Shell1,2%
Starlark0,7%
CMake0,5%
Остальные0,7%

Сообщить о нарушении

mirror_xnnpack

Quentin KhanPass down the original kernel and bias buffers when creating a fully connected operation.2 часа назад9e77b69

Описание

Языки

Описание

Языки

Quentin Khan
Pass down the original kernel and bias buffers when creating a fully connected operation.
2 часа назад
9e77b69