[X86][GlobalIsel] Missing Bitwise Combine Optimization

[mahesh-attarde]

I was working on one of G_IS_FPCLASS expansion and noticed we are not combining simple cases as compare to SDAG.

; RUN: llc < %s -mtriple=x86_64-linux -global-isel -global-isel-abort=1 | FileCheck %s -check-prefixes=X64,X64-GISEL
; RUN: llc < %s -mtriple=x86_64-linux | FileCheck %s -check-prefixes=X64,X64-SDAGISEL

; X64-SDAGISEL-LABEL: not_isinf_f:
; X64-SDAGISEL:       # %bb.0: # %entry
; X64-SDAGISEL-NEXT:    movd %xmm0, %eax
; X64-SDAGISEL-NEXT:    andl $2147483647, %eax # imm = 0x7FFFFFFF
; X64-SDAGISEL-NEXT:    cmpl $2139095040, %eax # imm = 0x7F800000
; X64-SDAGISEL-NEXT:    setne %al
; X64-SDAGISEL-NEXT:    retq

; X64-GISEL-LABEL: not_isinf_f:
; X64-GISEL:       # %bb.0: # %entry
; X64-GISEL-NEXT:    movd %xmm0, %eax
; X64-GISEL-NEXT:    andl $2147483647, %eax # imm = 0x7FFFFFFF
; X64-GISEL-NEXT:    xorl %ecx, %ecx
; X64-GISEL-NEXT:    cmpl $2139095040, %eax # imm = 0x7F800000
; X64-GISEL-NEXT:    setb %dl
; X64-GISEL-NEXT:    orb %cl, %dl
; X64-GISEL-NEXT:    cmpl $2139095040, %eax # imm = 0x7F800000
; X64-GISEL-NEXT:    seta %al
; X64-GISEL-NEXT:    orb %dl, %al
; X64-GISEL-NEXT:    retq

define i1 @not_isinf_f(float %x) nounwind {
entry:
  %0 = tail call i1 @llvm.is.fpclass.f32(float %x, i32 507)  ; ~0x204 = "~inf"
  ret i1 %0
}

In above example sequence "xorl %ecx, %ecx" and " orb %cl, %dl" is zero with OR. It is redundant.