Allow not flusing denorms in operands
authorSlawomir Cygan <slawomir.cygan@intel.com>
Fri, 28 Feb 2020 14:36:58 +0000 (15:36 +0100)
committerAlexander Galazin <Alexander.Galazin@arm.com>
Tue, 17 Mar 2020 09:30:01 +0000 (05:30 -0400)
commit0f323f593e9989190f0b8a5768aa25b594d7c1be
tree78c200524fee21f9bd7f97a906dfb7f93cf63941
parentf45f44b3be8d2d3fa4aedb8bccefb758ec2e268d
Allow not flusing denorms in operands

This fixes tests for denorm value flush to zero in Ceil(), Log(),
 Log2(), Sqrt(), Sqrt2() operations. The test was assuming, that
operands of denormal value are flushed before the operation, however
the specification does not require that:

"If the entry point is declared with the DenormFlushToZero execution mode then for the
affected instuctions the denormalized result must be flushed to zero and the denormalized
operands **may** be flushed to zero."

VK-GL-CTS Issue: 2242

Affects: dEQP-VK.spirv_assembly.instruction.compute.float_controls.*.input_args.*_denorm_flush_to_zero

Components: Vulkan

Change-Id: Iedeb0d18fde3749a83be3262dba5efabf365aea2
external/vulkancts/modules/vulkan/spirv_assembly/vktSpvAsmFloatControlsTests.cpp