tree 94b3f0675b3355ae6197f206c57b90186e68aaaa
parent 75665e588cbe7be2bd5acb31a06c0e7d23f28c38
author Nicolas Capens <capn@google.com> 1644039109 -0500
committer Nicolas Capens <nicolascapens@google.com> 1644262216 +0000

Use full-precision transcendental function implementations in Reactor

Lower-precision implementations of transcendental functions for SPIR-V
shaders have previously been moved out of Reactor. The ones in Reactor
should now be reference implementations.

The LLVM backend already used IR intrinsics which end up emitting calls
to the standard C/C++ math functions, or we did so ourselves. Only
Subzero is affected by this change (but not the Vulkan implementation,
since it now uses transcendental functions from ShaderCore).

Bug: b/169755552
Change-Id: I42aeb5fae6f42f2980401b3318084f9e8f328432
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/62288
Kokoro-Result: kokoro <noreply+kokoro@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>