Code

Fast atan and atan2 functions.#8388

Code

Closed

mcourteaux wants to merge 28 commits intohalide:mainhalide/Halide:mainfrom mcourteaux:fast-atanmcourteaux/Halide:fast-atanCopy head branch name to clipboard

Commits

Commits on Feb 2, 2025

• Fast vectorizable atan and atan2 functions.

  mcourteaux

  committed
  4e99a3e

  Copy full SHA for 4e99a3e

• Default to not using fast atan versions if on CUDA.

  mcourteaux

  committed
  0f00cb1

  Copy full SHA for 0f00cb1

• Finished fast atan/atan2 functions and tests.

  mcourteaux

  committed
  6af7972

  Copy full SHA for 6af7972

• Correct attribution.

  mcourteaux

  committed
  7cabaa3

  Copy full SHA for 7cabaa3

• Clang-format

  mcourteaux

  committed
  2850da3

  Copy full SHA for 2850da3

• Weird WebAssembly limits...

  mcourteaux

  committed
  4297809

  Copy full SHA for 4297809

• Small improvements to the optimization script.

  mcourteaux

  committed
  18bc63c

  Copy full SHA for 18bc63c

• Polynomial optimization for log, exp, sin, cos with correct ranges.

  mcourteaux

  committed
  d73177c

  Copy full SHA for d73177c

• Improve fast atan performance tests for GPU.

  mcourteaux

  committed
  a146f57

  Copy full SHA for a146f57

• Bugfix fast_atan approximation. Fix correctness test to exceed the range (-1, 1) to test (-4, 4). Cleanup code/comments. Test performance for all approximations.

  mcourteaux

  committed
  4814c37

  Copy full SHA for 4814c37

• Cleanup

  mcourteaux

  committed
  803a762

  Copy full SHA for 803a762

• Enum class instead of enum for ApproximationPrecision.

  mcourteaux

  committed
  2a9b506

  Copy full SHA for 2a9b506

• Weird Metal limits. There should be a better way...

  mcourteaux

  committed
  dfb2264

  Copy full SHA for dfb2264

• Skip test for WebGPU.

  mcourteaux

  committed
  746337b

  Copy full SHA for 746337b

• Fast atan/atan2 polynomials reoptimized. New optimization strategy: ULP.

  mcourteaux

  committed
  614abc2

  Copy full SHA for 614abc2

• Feedback Steven.

  mcourteaux

  committed
  5163d78

  Copy full SHA for 5163d78

• More comments and test mantissa error.

  mcourteaux

  committed
  091ab5c

  Copy full SHA for 091ab5c

• Do not error when testing arctan performance on Metal / WebGPU.

  mcourteaux

  committed
  05ecbb6

  Copy full SHA for 05ecbb6

• Rework precision specification. Generalize towards using this for other functions.

  mcourteaux

  committed
  bc2ed82

  Copy full SHA for bc2ed82

• Clang-format.

  mcourteaux

  committed
  5a5cb27

  Copy full SHA for 5a5cb27

• Fix makefile and clang-tidy.

  mcourteaux

  committed
  58cb828

  Copy full SHA for 58cb828

• Fix incorrect approximation selection when required precision is not available.

  mcourteaux

  committed
  735c431

  Copy full SHA for 735c431

• Feedback from Steven.

  mcourteaux

  committed
  f598106

  Copy full SHA for f598106

Commits on Feb 4, 2025

• Implemented approximation tables for sin, cos, exp, log fast variants. Still needs cleanup.

  mcourteaux

  committed
  20f9187

  Copy full SHA for 20f9187

• Clang-format.

  mcourteaux

  committed
  05ebf8d

  Copy full SHA for 05ebf8d

• Move Polynomial Optimizer Python script to tools/ directory.

  mcourteaux

  committed
  cf4ef47

  Copy full SHA for cf4ef47

• Enable performance test for fast_atan and fast_atan2.

  mcourteaux

  committed
  7156130

  Copy full SHA for 7156130

• LLVM upper-limit 99 (CMake needs an upper limit).

  mcourteaux

  committed
  302534b

  Copy full SHA for 302534b