Fix results-check for floating-point intermediate atomic operations
This commit fixes the results checking function for a set of tests that
verify atomic operations on floating-point formats. Previously when
compiled with GCC 4.8.4 the generated code would store incorrect values
into `resultValues[i]` rather than correctly interpreting the uint32_t
input data as a float and converting to a uint32_t before storing in
`resultValues[i]`.
Affects:
dEQP-VK.image.atomic_operations.add*r32f*
dEQP-VK.image.atomic_operations.exchange*r32f*
Components: Vulkan
VK-GL-CTS issue: 2632
Change-Id: I34e55d8ef109b5a14bda86d28821a0053c874546