I had made a bad assumption earlier that these would choose the correct operation based on signedness. That's not the case -- LLVM has separate Max/UMax and Min/UMin RMW operations.

Test: dEQP-VK.glsl.atomic_operations.*
Change-Id: I92a9e56011c4ef8ac658a826c480b16f65d0704d
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31909
Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>
Tested-by: Chris Forbes <chrisforbes@google.com>
Presubmit-Ready: Chris Forbes <chrisforbes@google.com>