tree 50c010db4850083111f0aa3fbd165a5321c5c942
parent 0ed3fa6a7f4dd6bf783dd88528489d4be20f7c9c
author Nicolas Capens <capn@google.com> 1656612120 -0400
committer Nicolas Capens <nicolascapens@google.com> 1656637370 +0000

Implement SIMD helper functions using 128-bit operations

Functions such as SignMask() are not straightforward to implement for
SIMD vectors larger than 128-bit (i.e. more than 4 lanes wide), but to
facilitate the transition to using rr::SIMD types we can simply
implement them using the legacy 128-bit functions operating only on the
lower 128-bit half of the SIMD vector.

This approach ensures we don't lose performance when replacing types
such as rr::Float4 with rr::SIMD::Float while keeping the width at 4.
ASSERT() expressions have been added to pinpoint which functions need
work when locally testing with wider SIMD width.

Bug: b/214583550
Change-Id: Ia644e38c9bb2a0c8175da0ddc5a861d5e62e7bd4
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/66811
Kokoro-Result: kokoro <noreply+kokoro@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>