| commit | bec99d61aa9d1e4d07a85f2be326821a1600593c | [log] [tgz] |
|---|---|---|
| author | Benoit Jacob <benoitjacob@google.com> | Wed May 13 13:38:14 2020 -0700 |
| committer | Copybara-Service <copybara-worker@google.com> | Wed May 13 13:38:33 2020 -0700 |
| tree | 53e7e5943ec777d42bc9c96a47dae8e8ebb7f1e0 | |
| parent | b991b7a2b8d17475231904290c6ef06e1a4089b7 [diff] |
Enable AVX512 in the open-source build, on the same compilers as other x86 SIMD paths are enabled (Clang>=8, GCC>=9). Mostly this involved changing some code to use equivalent (redundant) intrinsics that happen to be supported on GCC. GCC also found a `dst_stride` variable set but not used, so that's removed. And another `dst_stride` potentially used uninitialized. In principle, -march=skylake-avx512 should only require GCC 5.3. According to https://www.tensorflow.org/install/source#cpu, since TensorFlow 2.0, we should be able to assume GCC >= 7. PiperOrigin-RevId: 311395075
This is not an officially supported Google product.
ruy is a matrix multiplication library. Its focus is to cover the matrix multiplication needs of neural network inference engines. Its initial user has been TensorFlow Lite, where it is used by default on the ARM CPU architecture.
ruy supports both floating-point and 8bit-integer-quantized matrices.
ruy is designed to achieve maximal performance not just on very large sizes, as is the focus of many established libraries, but on whatever are the actual sizes and shapes of matrices most critical in current TensorFlow Lite applications. This often means quite small sizes, e.g. 100x100 or even 50x50, and all sorts of rectangular shapes.
ruy is currently only optimized for the ARM architectures (both 64-bit and 32-bit code). Optimization for the Intel x86 architecture is in progress.
ruy is currently optimized only for the following combination of storage orders: LHS = row-major, RHS = column-major, destination = column-major. All other combinations of storage orders fall back to slow reference code at the moment.