Skip to content

Navigation Menu

Appearance settings

View all features
- BY COMPANY SIZE
  Enterprises
  Small and medium teams
  Startups
  Nonprofits
- BY USE CASE
  App Modernization
  DevSecOps
  DevOps
  CI/CD
  View all use cases
- BY INDUSTRY
  Healthcare
  Financial services
  Manufacturing
  Government
  View all industries
View all solutions
- EXPLORE BY TOPIC
  AI
  Software Development
  DevOps
  Security
  View all topics
- EXPLORE BY TYPE
  Customer stories
  Events & webinars
  Ebooks & reports
  Business insights
  GitHub Skills
- SUPPORT & SERVICES
  Documentation
  Customer support
  Community forum
  Trust center
  Partners
View all resources
- COMMUNITY
  GitHub SponsorsFund open source developers
- PROGRAMS
  Security Lab
  Maintainer Community
  Accelerator
  Archive Program
- REPOSITORIES
  Topics
  Trending
  Collections
- ENTERPRISE SOLUTIONS
  Enterprise platformAI-powered developer platform
- AVAILABLE ADD-ONS
  GitHub Advanced SecurityEnterprise-grade security features
  Copilot for BusinessEnterprise-grade AI features
  Premium SupportEnterprise-grade 24/7 support
Pricing

Search code, repositories, users, issues, pull requests...

Search

Clear

Search syntax tips

Provide feedback

We read every piece of feedback, and take your input very seriously.

Include my email address so I can be contacted

Saved searches

Use saved searches to filter your results more quickly

Name

Query

To see all available qualifiers, see our documentation.

Appearance settings

You signed in with another tab or window. Reload to refresh your session. You signed out in another tab or window. Reload to refresh your session. You switched accounts on another tab or window. Reload to refresh your session.

Dismiss alert

halide / Halide Public

Notifications You must be signed in to change notification settings
Fork 1.1k
Star 6.6k

Code
Issues 1k
Pull requests 84
Discussions
Actions
Projects
Wiki
Security
Insights

Additional navigation options

Code
Issues
Pull requests
Discussions
Actions
Projects
Wiki
Security
Insights

Fast Function Approximations lowering.#8566

Open

mcourteaux wants to merge 86 commits intohalide:mainhalide/Halide:mainfrom

mcourteaux:fast-math-loweringmcourteaux/Halide:fast-math-loweringCopy head branch name to clipboard

Conversation Commits86 (86)Checks Files changed

Open

Fast Function Approximations lowering.#8566
mcourteaux wants to merge 86 commits intohalide:mainfrom
mcourteaux:fast-math-lowering

Commits

Commits on Jul 22, 2025

Fast vectorizable atan and atan2 functions.
mcourteaux
committed
Default to not using fast atan versions if on CUDA.
mcourteaux
committed
Finished fast atan/atan2 functions and tests.
mcourteaux
committed
Correct attribution.
mcourteaux
committed
Clang-format
mcourteaux
committed
Weird WebAssembly limits...
mcourteaux
committed
Small improvements to the optimization script.
mcourteaux
committed
Polynomial optimization for log, exp, sin, cos with correct ranges.
mcourteaux
committed
Improve fast atan performance tests for GPU.
mcourteaux
committed
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
Cleanup
mcourteaux
committed
Enum class instead of enum for ApproximationPrecision.
mcourteaux
committed
Weird Metal limits. There should be a better way...
mcourteaux
committed
Skip test for WebGPU.
mcourteaux
committed
Fast atan/atan2 polynomials reoptimized. New optimization strategy: ULP.
mcourteaux
committed
Feedback Steven.
mcourteaux
committed
More comments and test mantissa error.
mcourteaux
committed
Do not error when testing arctan performance on Metal / WebGPU.
mcourteaux
committed
Rework precision specification. Generalize towards using this for other functions.
mcourteaux
committed
Clang-format.
mcourteaux
committed
Fix makefile and clang-tidy.
mcourteaux
committed
Fix incorrect approximation selection when required precision is not available.
mcourteaux
committed
Feedback from Steven.
mcourteaux
committed
Implemented approximation tables for sin, cos, exp, log fast variants. Still needs cleanup.
mcourteaux
committed
Clang-format.
mcourteaux
committed
Move Polynomial Optimizer Python script to tools/ directory.
mcourteaux
committed
Enable performance test for fast_atan and fast_atan2.
mcourteaux
committed
LLVM upper-limit 99 (CMake needs an upper limit).
mcourteaux
committed
Add LLVM IR for PTX sin.approx, cos.approx, tanh.approx
mcourteaux
committed
Implemented tan. Improved polynomial optimizer performance for MULPE optimization. Greatly improved accuracy testing framework.
mcourteaux
committed
Implemented tanh, tan. Many improvements to accuracy test and performance test.
mcourteaux
committed
Clang-format.
mcourteaux
committed
WIP: Fiddle with strict_float behavior in CSE. Fix fast math precision test by precomputing arguments buffer.
mcourteaux
committed
Nuke MAE_MULPE. Separate optimized MULPE-corrected sin and cos.
mcourteaux
committed
Clang-format
mcourteaux
committed
Some cleanup.
mcourteaux
committed
Fix sine.
mcourteaux
committed
Fix clang-tidy. Mark OpenCL exp() as fast.
mcourteaux
committed
Clang format is annoying me.
mcourteaux
committed
Remove my experimental CSE step.
mcourteaux
committed
OpenCL performance of fast_exp forced poly is expected to be worse.
mcourteaux
committed
OpenCL fast functions selected for fast transcendentals.
mcourteaux
committed
Lower fast intrinsics on metal to the fast:: namespace versions.
mcourteaux
committed
Split tables for sin and cos, as metal has odd precision for sin. Add support for fast_tanh on all backends.
mcourteaux
committed
Move range_reduce_log to a header. Drive-by fix listing libOpenCL.so.1 as well.
mcourteaux
committed
Fix API documentation. Improve measuring accuracy. Fix vector_math test not touching input: prevents constant folding.
mcourteaux
committed
Also vectorize on GPU to make sure we test that.
mcourteaux
committed
Add FastMathFunctions.cpp to Makefile
mcourteaux
committed
Add support for derivatives for the fast_ intrinsics.
mcourteaux
committed
Remove unused helper function.
mcourteaux
committed
Add in a gracefactor for precision when the system does not support FMA.
mcourteaux
committed
Clang Format.
mcourteaux
committed
Windows doesn't print thousand separaters with printf. :(
mcourteaux
committed
Remove grace factor, and use safety factor of 5% when selecting a polynomial by default instead.
mcourteaux
committed
Use 50% tighter constraints when no FMA is available to compensate for lost precision. Also test accuracy of non-forced polynomials, i.e., potentially intrinsics.
mcourteaux
committed
Clang-format.
mcourteaux
committed
Working on better optimizations. Improving PR and code.
mcourteaux
committed
Implemented fast_asin() fast_acos(). Slowly redoing coefficients.
mcourteaux
committed
WIP: determine precision of the polynomials.
mcourteaux
committed
Revived all tests.
mcourteaux
committed
Clang format
mcourteaux
committed
Implement expm1. Fix accuracy of tanh. Fix lowering of tanh on CUDA. Selectively disable some tests that require strict_float on GPU backends.
mcourteaux
committed
Clang-format
mcourteaux
committed
Feedback, and remove expm1 test.
mcourteaux
committed
Fix compilation issues.
mcourteaux
committed
One more compilation issue.
mcourteaux
committed
Fixed a bracket.
mcourteaux
committed
Update some precision info on math intrinsics for Vulkan and Metal.
mcourteaux
committed
Fix makefile after I accidentally broke it by sorting files alphabetically.
mcourteaux
committed
Add fast math calls to new extern_function_name_map for OpenCL.
mcourteaux
committed
Move fast function calls to extern table for Metal.
mcourteaux
committed
Try to fix compile/test issues.
mcourteaux
committed
Fix Makefile and symbol visibility issue.
mcourteaux
committed
Clang-format
mcourteaux
committed
Make use of the new strict_float intrinsics for the fast math functions.
mcourteaux
committed
Relax performance tests for GPUs.
mcourteaux
committed
Clang-format
mcourteaux
committed
Fix incorrect forward declaration.
mcourteaux
committed
Fix acos on Metal. Relax perf-test for tanh on OpenCL.
mcourteaux
committed
Fix strict float behavior for the fast_tan function. Implemented split float calculations for f64 and f16.
mcourteaux
committed
Enable fp16 fast_math functions without promises.
mcourteaux
committed
Clear internal assert, as it assumed SSE floating point behavior, which failed on x87.
mcourteaux
committed
Let CodeGen_C handle all float-literal printing (also for Float(16) in case that's marked as supported by the GPU backend.)
mcourteaux
committed
Fix internal test for CodeGen_C given the scientific way of printing literals.
mcourteaux
committed

Commits on Feb 1, 2026

Merge branch 'main' into fast-math-lowering
mcourteaux
committed
Update.
mcourteaux
committed

Footer

© 2026 GitHub, Inc.

Footer navigation

Terms
Privacy
Security
Status
Community
Docs
Contact

You can’t perform that action at this time.