[InstCombine] fshl(x, 0, y) with in-range y not converted to shl x, y

[nikic]

https://alive2.llvm.org/ce/z/k4SnKK

define i32 @src(i32 %x, i32 range(i32 0, 32) %y) {
    %res = call i32 @llvm.fshl(i32 %x, i32 0, i32 %y)
    ret i32 %res
}

define i32 @tgt(i32 %x, i32 range(i32 0, 32) %y) {
    %res = shl i32 %x, %y
    ret i32 %res
}

We currently only do this if the shift amount is constant, but not for a dynamic shift amount that is < bitwidth based on known bits.

[nikic]

@dtcxzyw suggested masking the shift amount as an alternative:

https://alive2.llvm.org/ce/z/3oTEop

define i32 @src(i32 %x, i32 %y) {
  %res = call i32 @llvm.fshl.i32(i32 %x, i32 0, i32 %y)
  ret i32 %res
}

define i32 @tgt(i32 %x, i32 %y) {
  %rem = and i32 %y, 31
  %res = shl i32 %x, %rem
  ret i32 %res
}

[AmrDeveloper]

Interested, I can work on this