[all-commits] [llvm/llvm-project] b73942: AMDGPU/EG, CM: Implement fsqrt using recip(rsqrt(x)...

[Jan Vesely via All-commits]

  Branch: refs/heads/release/10.x
  Home:   https://github.com/llvm/llvm-project
  Commit: b73942dbc144c11dc94fd32a7d8025a22e7e1d6b
      https://github.com/llvm/llvm-project/commit/b73942dbc144c11dc94fd32a7d8025a22e7e1d6b
  Author: Jan Vesely <jan.vesely at rutgers.edu>
  Date:   2020-02-10 (Mon, 10 Feb 2020)

  Changed paths:
    M llvm/lib/Target/AMDGPU/CaymanInstructions.td
    M llvm/lib/Target/AMDGPU/EvergreenInstructions.td
    M llvm/lib/Target/AMDGPU/R600Instructions.td
    M llvm/test/CodeGen/AMDGPU/fsqrt.ll

  Log Message:
  -----------
  AMDGPU/EG,CM: Implement fsqrt using recip(rsqrt(x)) instead of x * rsqrt(x)

The old version might be faster on EG (RECIP_IEEE is Trans only),
but it'd need extra corner case checks.
This gives correct corner case behaviour and saves a register.
Fixes OCL CTS sqrt test (1-thread, scalar) on Turks.

Reviewer: arsenm
Differential Revision: https://reviews.llvm.org/D74017

(cherry picked from commit e6686adf8a743564f0c455c34f04752ab08cf642)