XNNPACK is a highly optimized library of floating-point neural network inference operators for ARM, WebAssembly, and x86 platforms. XNNPACK is not intended for direct use by deep learning practitioners and researchers; instead it provides low-level performance primitives for accelerating high-level machine learning frameworks, such as TensorFlow Lite, TensorFlow.js, PyTorch, and MediaPipe.
- ARM64 on Android, Linux, macOS, and iOS (including WatchOS and tvOS)
- ARMv7 (with NEON) on Android, Linux, and iOS (including WatchOS)
- x86 and x86-64 (up to AVX512) on Windows, Linux, macOS, Android, and iOS simulator
- WebAssembly MVP
- WebAssembly SIMD (experimental)
XNNPACK implements the following neural network operators:
- 2D Convolution (including grouped and depthwise)
- 2D Deconvolution (AKA Transposed Convolution)
- 2D Average Pooling
- 2D Max Pooling
- 2D ArgMax Pooling (Max Pooling + indices)
- 2D Unpooling
- 2D Bilinear Resize
- 2D Depth-to-Space (AKA Pixel Shuffle)
- Add (including broadcasting, two inputs only)
- Subtract (including broadcasting)
- Divide (including broadcasting)
- Maximum (including broadcasting)
- Minimum (including broadcasting)
- Multiply (including broadcasting)
- Squared Difference (including broadcasting)
- Global Average Pooling
- Channel Shuffle
- Fully Connected
- Abs (absolute value)
- Bankers' Rounding (rounding to nearest, ties to even)
- Ceiling (rounding to integer above)
- Clamp (includes ReLU and ReLU6)
- Copy
- ELU
- Floor (rounding to integer below)
- HardSwish
- Leaky ReLU
- Negate
- Sigmoid
- Softmax
- Square
- Truncation (rounding to integer towards zero)
- PReLU
All operators in XNNPACK support NHWC layout, but additionally allow custom stride along the Channel dimension. Thus, operators can consume a subset of channels in the input tensor, and produce a subset of channels in the output tensor, providing a zero-cost Channel Split and Channel Concatenation operations.
The table below presents single-threaded performance of XNNPACK library on three generations of MobileNet models and three generations of Pixel phones.
| Model | Pixel, ms | Pixel 2, ms | Pixel 3a, ms |
|---|---|---|---|
| MobileNet v1 1.0X | 82 | 86 | 88 |
| MobileNet v2 1.0X | 49 | 53 | 55 |
| MobileNet v3 Large | 39 | 42 | 44 |
| MobileNet v3 Small | 12 | 14 | 14 |
The following table presents multi-threaded (using as many threads as there are big cores) performance of XNNPACK library on three generations of MobileNet models and three generations of Pixel phones.
| Model | Pixel, ms | Pixel 2, ms | Pixel 3a, ms |
|---|---|---|---|
| MobileNet v1 1.0X | 43 | 27 | 46 |
| MobileNet v2 1.0X | 26 | 18 | 28 |
| MobileNet v3 Large | 22 | 16 | 24 |
| MobileNet v3 Small | 7 | 6 | 8 |
Benchmarked on March 27, 2020 with end2end_bench --benchmark_min_time=5 on an Android/ARM64 build with Android NDK r21 (bazel build -c opt --config android_arm64 :end2end_bench) and neural network models with randomized weights and inputs.
The table below presents multi-threaded performance of XNNPACK library on three generations of MobileNet models and three generations of Raspberry Pi boards.
| Model | RPi Zero W (BCM2835), ms | RPi 2 (BCM2836), ms | RPi 3+ (BCM2837B0), ms | RPi 4 (BCM2711), ms |
|---|---|---|---|---|
| MobileNet v1 1.0X | 4004 | 337 | 116 | 72 |
| MobileNet v2 1.0X | 2011 | 195 | 83 | 41 |
| MobileNet v3 Large | 1694 | 163 | 70 | 38 |
| MobileNet v3 Small | 482 | 52 | 23 | 13 |
Benchmarked on May 22, 2020 with end2end-bench --benchmark_min_time=5 on a Raspbian Buster build with CMake (./scripts/build-local.sh) and neural network models with randomized weights and inputs.
- Marat Dukhan "The Indirect Convolution Algorithm". Presented on Efficient Deep Learning for Compute Vision (ECV) 2019 workshop (slides, paper on ArXiv).
- Erich Elsen, Marat Dukhan, Trevor Gale, Karen Simonyan "Fast Sparse ConvNets". Paper on ArXiv, pre-trained sparse models.
- Marat Dukhan, Artsiom Ablavatski "The Two-Pass Softmax Algorithm". Paper on ArXiv.
- Yury Pisarchyk, Juhyun Lee "Efficient Memory Management for Deep Neural Net Inference". Paper on ArXiv.
- TensorFlow Lite.
- TensorFlow.js WebAssembly backend.
- PyTorch Mobile.
- MediaPipe for the Web.
- Alibaba HALO (Heterogeneity-Aware Lowering and Optimization)
- Samsung ONE (On-device Neural Engine)
XNNPACK is a based on QNNPACK library. Over time its codebase diverged a lot, and XNNPACK API is no longer compatible with QNNPACK.