Fix denorm tests in float_controls
When binary operations on denorms are executed against Inf or Nan
SignedZeroInfNanPreserve mode should always be enabled as some
implementations may perform optimizations that assume that
arguments are not Nan or Inf. Before this change DenormPreserve
and DenormFlushToZero tests had only those modes enabled. Now
SignedZeroInfNanPreserve is added when second operand is Inf or Nan.
Components: Vulkan
VK-GL-CTS issue: 1517
Affects:
dEQP-VK.spirv_assembly.instruction.compute.float_controls.*
dEQP-VK.spirv_assembly.instruction.graphics.float_controls.*
Change-Id: I0b5e0ed010bde200154f58f9fe194fdf7c565803