File Transforms/InstSimplify/canonicalize.ll has the following test:
define float @canonicalize_neg_denorm_positive_zero_output_preserve_sign_input() "denormal-fp-math"="positive-zero,preserve-sign" {
; CHECK-LABEL: @canonicalize_neg_denorm_positive_zero_output_preserve_sign_input(
; CHECK-NEXT: ret float 0.000000e+00
;
%ret = call float @llvm.canonicalize.f32(float bitcast (i32 -2139095041 to float))
ret float %ret
}The expected result is wrong. This function should return -0.0.
If you run the equivalent multiplication by 1.0, you get -0.0:
define float @src() "denormal-fp-math"="positive-zero,preserve-sign" {
%ret = fmul float bitcast (i32 -2139095041 to float), 1.0
ret float %ret
}
define float @tgt() "denormal-fp-math"="positive-zero,preserve-sign" {
ret float 0.0
}
Transformation doesn't verify!
ERROR: Value mismatch
Example:
Source:
float %ret = #x80000000 (-0.0)
Target:
Source value: #x80000000 (-0.0)
Target value: #x00000000 (+0.0)In summary, canonicalize should use just the input denormal flag and ignore the output one.
cc @arsenm
File Transforms/InstSimplify/canonicalize.ll has the following test:
The expected result is wrong. This function should return -0.0.
If you run the equivalent multiplication by 1.0, you get -0.0:
In summary, canonicalize should use just the input denormal flag and ignore the output one.
cc @arsenm