[MIPS] match llvm.{min,max}num with {min,max}.fmt for R6

[Cyanoxygen]

• The behavior is similar to UCOMISD on x86, which is also used to compare two fp values, specifically on handling of NaNs.
• Update related tests regarding this change.
• The further goal is to implement llvm.minimum and llvm.maximum intrinsics for MIPS R6 and Pre-R6.

Part of #64207

[wzssyqa]

We need a real regression test for fmax/fmin. Please add a new one.
Maybe RISCV/float-maximum-minimum.ll is a good example.

[Cyanoxygen]

Maybe RISCV/float-maximum-minimum.ll is a good example.

That test's for llvm.fminimum and llvm.fmaximum, which is different than fminnum and fmaxnum. I can extend the test in #89907.

For the test in this PR, we definitely need a new one.

[wzssyqa]

@Cyanoxygen MSA has fmax.fmt/fmin.fmt, too.
Consider to support them?

[Cyanoxygen]

@Cyanoxygen MSA has fmax.fmt/fmin.fmt, too. Consider to support them?

Yes, I would like to try to look into that.

[wzssyqa]

#90321 @Cyanoxygen

[wzssyqa]

Ohh, this patch is not correct.

In https://llvm.org/docs/LangRef.html#llvm-minnum-intrinsic :

Unlike the IEEE-754 2008 behavior, this does not distinguish between signaling and quiet NaN inputs. If a target’s implementation follows the standard and returns a quiet NaN if either input is a signaling NaN, the intrinsic lowering is responsible for quieting the inputs to correctly return the non-NaN input (e.g. by using the equivalent of llvm.canonicalize).