Fix @fastmath x^2 inlining regression for Float32 and Float16

[Yashagarwal9798]

Summary

This PR fixes the performance regression where @fastmath x^2 for Float32 was not being inlined to efficient LLVM code, unlike Float64.

Problem

As reported in #60639, @fastmath x^2 for Float32 was falling back to power_by_squaring instead of using the LLVM powi intrinsic. This resulted in:

• Unnecessary function calls instead of inline multiplication
• Potential type promotion to Float64
• Suboptimal generated code compared to Float64

Before this fix, @code_llvm @fastmath Float32(1.5)^2 would show calls to power_by_squaring, while Float64 correctly used the llvm.powi intrinsic.

Solution

Added the missing pow_fast methods for Float32 and Float16:

• pow_fast(::Float32, ::Int32) - uses llvm.powi.f32.i32 intrinsic directly
• pow_fast(::Float32, ::Integer) - wrapper that converts to Int32 when safe, matching the Float64 pattern
• pow_fast(::Float16, ::Integer) - converts to Float32, computes, and converts back

This mirrors the existing implementation for Float64 which already used llvm.powi.f64.i32.

Testing

Added a regression test that verifies @fastmath x^2 generates inline fmul instructions (not power_by_squaring calls) for Float16, Float32, and Float64.

Fixes #60639

[adienes]

@Yashagarwal9798 please be aware that all uses of AI must be disclosed

[DilumAluthge]

Build failure looks real?

error during bootstrap:
LoadError("sysimg.jl", 3, LoadError("Base.jl", 222, LoadError("fastmath.jl", 300, UndefVarError(:IEEEFloat, 0x0000000000005e69, Base.FastMath))))

[DilumAluthge]

@oscardssmith CI is all green now.

[oscardssmith]

@Yashagarwal9798 thanks for the PR!

[Yashagarwal9798]

Thanks! really glad I could help improve the project.

[eschnett]

This is a candidate for backporting. Although this concerns only an optimization, generating inefficient code for x^2 is a disaster performance-wise.

[oscardssmith]

Agreed. If we backport, we need to be careful not to back-port the Float16 version of this to 1.12 (LLVM only recently doesn't produce garbage on x86 with this intrinsic). 1.13 should receive the back-port unmodified though. I'll put up the 1.12 PR.

[vchuravy]

@oscardssmith did you open a PR to backport this to 1.12?

[oscardssmith]

I did, but I linked to https://github.com/JuliaLang/julia/pull/60640/changes so it didn't tag.