add support for the 'fcmp false' instruction

[VyacheslavLevytskyy]

There is no direct equivalent for "fcmp false" in the specification [https://registry.khronos.org/SPIR-V/specs/unified1/SPIRV.html#_relational_and_logical_instructions], but we may return ConstantFalse in this case instead of reporting error.

This patch depends on #2216

[VyacheslavLevytskyy]

@MrSidims @bwlodarcz Can you please have a look

[bwlodarcz]

Unfortunately the 'fcmp' instruction can go through vector also. I prepared test:

define spir_func <4 x i1> @g(<4 x float> %0) {
entry:
 %1 = fcmp false <4 x float> %0, %0
 ret <4 x i1> %1
}

llvm-as and translator passes but spirv-val

error: line 28: OpReturnValue Value <id> '15[%false_0]'s type does not match OpFunction's return type.
  OpReturnValue %false_0

Without further investigation I interpret this error that ConstantFalse is constant type declaration. In SPIRV BoolType != Constant BoolType. So the return type of the function is not the same as the one returned.

[VyacheslavLevytskyy]

@bwlodarcz Thank you. Let's represent a vector of false values as a OpConstantNull ("The null value is type dependent, defined as follows: - Scalar Boolean: false ... Composites: Members are set recursively to the null constant according to the null value of their constituent types.")

[VyacheslavLevytskyy]

Please take a look @asudarsa @LU-JOHN

[bwlodarcz]

LGTM.

[asudarsa]

LGTM. Thanks