Tensor Comprehensions (TC) is a fully-functional C++ library to automatically synthesize high-performance machine learning kernels using Halide, ISL and NVRTC or LLVM. TC additionally provides basic integration with Caffe2 and PyTorch. We provide more details in our paper on arXiv.

This library is designed to be highly portable, machine-learning-framework agnostic and only requires a simple tensor library with memory allocation, offloading and synchronization capabilities.

For now, we have integrated TC with Caffe2 and PyTorch.

A simple example

The following illustrates a short but powerful feature of the library: the capacity to JIT-compile high-performance machine learning kernels on demand, for specific sizes.

import tensor_comprehensions as tc
import torch
lang = """
def tensordot(float(N, C1, C2, H, W) I0, float(N, C2, C3, H, W) I1) -> (O) {
    O(n, c1, c3, h, w) +=! I0(n, c1, c2, h, w) * I1(n, c2, c3, h, w)
}
"""
N, C1, C2, C3, H, W = 32, 512, 8, 2, 28, 28
tensordot = tc.define(lang, name="tensordot")
I0, I1 = torch.randn(N, C1, C2, H, W).cuda(), torch.randn(N, C2, C3, H, W).cuda()
best_options = tensordot.autotune(I0, I1, cache=True)
out = tensordot(I0, I1, options=best_options)

After a few generations of autotuning on a 2-GPU P100 system, we see results resembling:

In C++ a minimal autotuning example resembles the following:

TEST(TensorDot, SimpleAutotune) {
  // 1. Define and setup the TC compilation unit with CUDA memory
  // management backed by ATen tensors.
  std::string tc = R"TC(
def tensordot(float(N, C1, C2, H, W) I0,
              float(N, C2, C3, H, W) I1)  -> (O)
{
    O(n, c1, c3, h, w) +=! I0(n, c1, r_c2, h, w) * I1(n, r_c2, c3, h, w)
}
  )TC";

  // 2. Allocate tensors with random data.
  at::Tensor I0 = at::CUDA(at::kFloat).rand({32,  8, 16, 17, 25});
  at::Tensor I1 = at::CUDA(at::kFloat).rand({32, 16, 2, 17, 25});

  // 3. Run autotuning with evolutionary search starting from a naive option.
  auto naiveOptions = Backend::MappingOptionsType::makeNaiveMappingOptions();
  tc::aten::ATenAutotuner<tc::CudaBackend, tc::autotune::GeneticSearch>
      geneticAutotuneATen(tc);
  auto bestOption =
      geneticAutotuneATen.tune("tensordot", {I0, I1}, {naiveOptions});

  // 4. Compile and run the TC with the best option after allocating output
  //    tensors.
  auto pExecutor =
      tc::aten::compile<Backend>(tc, "tensordot", {I0, I1}, bestOption[0]);
  auto outputs = tc::aten::prepareOutputs(tc, "tensordot", {I0, I1});
  auto timings = tc::aten::profile(*pExecutor, {I0, I1}, outputs);
  std::cout << "tensordot size I0: " << I0.sizes() << ", "
            << "size I1: " << I1.sizes()
            << " ran in: " << timings.kernelRuntime.toMicroSeconds() << "us\n";
}

Note that we only need to autotune a TC once to obtain reasonable mapping options that can translate to other problem sizes for a given TC as the following snippet illustrates:

// 5. Reuse bestOptions from autotuning on another kernel
for (auto sizes : std::vector<std::pair<at::IntList, at::IntList>>{
         {{4, 9, 7, 16, 14}, {4, 7, 3, 16, 14}},
         {{8, 5, 11, 10, 10}, {8, 11, 16, 10, 10}},
     }) {
  at::Tensor I0 = makeATenTensor<Backend>(sizes.first);
  at::Tensor I1 = makeATenTensor<Backend>(sizes.second);
  auto pExecutor =
      tc::aten::compile<Backend>(tc, "tensordot", {I0, I1}, bestOption[0]);
  auto outputs = tc::aten::prepareOutputs(tc, "tensordot", {I0, I1});
  auto timings = tc::aten::profile(*pExecutor, {I0, I1}, outputs);
  std::cout << "tensordot size I0: " << I0.sizes() << ", "
            << "size I1: " << I1.sizes()
            << " ran in: " << timings.kernelRuntime.toMicroSeconds()
            << "us\n";
}

Putting it all together, one may see:

> build$ ./examples/example_simple
[==========] Running 1 test from 1 test case.
[----------] Global test environment set-up.
[----------] 1 test from TensorDot
[ RUN      ] TensorDot.SimpleAutotune
Generation 0    Jobs(Compiled, GPU)/total  (10, 10)/10   (best/median/worst)us: 226/4238/7345
Generation 1    Jobs(Compiled, GPU)/total  (10, 10)/10   (best/median/worst)us: 220/221/233
Generation 2    Jobs(Compiled, GPU)/total  (10, 10)/10   (best/median/worst)us: 220/221/234
tensordot size I0: [16, 8, 16, 17, 25], size I1: [16, 16, 2, 17, 25] ran in: 239us
tensordot size I0: [4, 9, 7, 16, 14], size I1: [4, 7, 3, 16, 14] ran in: 56us
tensordot size I0: [8, 5, 11, 10, 10], size I1: [8, 11, 16, 10, 10] ran in: 210us
[       OK ] TensorDot.SimpleAutotune (27812 ms)
[----------] 1 test from TensorDot (27812 ms total)

[----------] Global test environment tear-down
[==========] 1 test from 1 test case ran. (27812 ms total)
[  PASSED  ] 1 test.

We have not yet characterized the precise fraction of peak performance we obtain but it is not uncommon to obtain 80%+ of peak shared memory bandwidth after autotuning. Solid register-level optimizations are still in the work but TC in its current form already addresses the productivity gap between the needs of research and the needs of production. Which is why we are excited to share it with the entire community and bring this collaborative effort in the open.

Documentation

General: You can find detailed information about Tensor Comprehensions here.

C++ API: We also provide documentation for our C++ API which can can be found here

Installation

Binaries

We provide conda package for making it easy to install and use TC binary. Please refer to our documentation here for instructions.

From Source

You can find documentation here which contains instructions for building TC via docker, conda packages or in non-conda environment.

Communication

• Email: tensorcomp@fb.com
• GitHub issues: bug reports, feature requests, install issues, RFCs, thoughts, etc.

Code of Conduct

See the CODE_OF_CONDUCT.md file for more details.

License

Tensor Comprehensions is distributed under a permissive Apache v2.0 license, see the LICENSE file for more details.

Contributing

See the CONTRIBUTING.md file for more details.