tree b04528b17cec2a4731fd7dbfd72f6b81051641d7
parent b16655545f81ac9701015e1bba8f6c967927ff66
author Chris Forbes <chrisforbes@google.com> 1557520800 -0700
committer Chris Forbes <chrisforbes@google.com> 1557596959 +0000

Fix FP blending path to not make clamping assumptions

We had assumed in Context in various places that negative pixel values
would be clamped to zero, which is incorrect for floating-point color
attachments.

TODO: on the assumption that there is value in these blend optimizations
themselves, we should split the context blend state to be
per-attachment. This would give us the `independentBlend` optional
feature, and also allow us to keep the optimization for those
attachments which do have clamping in a mixed MRT scenario.

Bug: b/132433217
Test: dEQP-VK.pipeline.blend.format.r16*
Change-Id: I5434475aba5d21a8c8342b42cccc575b517f0900
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31050
Tested-by: Chris Forbes <chrisforbes@google.com>
Presubmit-Ready: Chris Forbes <chrisforbes@google.com>
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>