[all-commits] [llvm/llvm-project] cdf6a5: [libc] Use intrinsics for x86-64 fma and optimize ...

[lntue via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: cdf6a581b9277d1cf44e7a3f5979dcba3dbe1631
      https://github.com/llvm/llvm-project/commit/cdf6a581b9277d1cf44e7a3f5979dcba3dbe1631
  Author: Tue Ly <lntue at google.com>
  Date:   2021-12-09 (Thu, 09 Dec 2021)

  Changed paths:
    M libc/src/__support/FPUtil/PolyEval.h
    M libc/src/__support/FPUtil/x86_64/FMA.h
    A libc/src/__support/FPUtil/x86_64/PolyEval.h
    M libc/test/src/math/expm1f_test.cpp

  Log Message:
  -----------
  [libc] Use intrinsics for x86-64 fma and optimize PolyEval for x86-64 with degree 3 & 5 polynomials.

- Use intrinsics for x86-64 fma
- Optimize PolyEval for x86-64 with degree 3 & 5 polynomials.
- There might be a slight loss of accuracy compared to Horner's scheme due to usages of higher powers x^2 and x^3 in the computations.

Reviewed By: sivachandra

Differential Revision: https://reviews.llvm.org/D115347