[PATCH] D30651: [InlineCost, -Oz] Don't take into account the penalty of a fast call of frequently used functions

[Evgeny Astigeevich via Phabricator via llvm-commits]

eastig added a comment.

I checked if the GEPs from the provided IR are free on x86.  No, they are not:

          movq    %rdi, -8(%rbp)
          movq    %rsi, -16(%rbp)
          movl    %edx, -20(%rbp)
  .LBB0_1:                                # =>This Inner Loop Header: Depth=1
          movl    -20(%rbp), %eax
          movl    %eax, %ecx
          addl    $-1, %ecx
          movl    %ecx, -20(%rbp)
          cmpl    $0, %eax
          je      .LBB0_3
  # BB#2:                                 #   in Loop: Header=BB0_1 Depth=1
          movq    -16(%rbp), %rax
          movq    %rax, %rcx
          addq    $1, %rcx
          movq    %rcx, -16(%rbp)
          movb    (%rax), %dl
          movq    -8(%rbp), %rax
          movq    %rax, %rcx
          addq    $1, %rcx
          movq    %rcx, -8(%rbp)
          movb    %dl, (%rax)
          jmp     .LBB0_1
  .LBB0_3:

MOVQ + ADDQ are generated for them. I am writing a patch.


https://reviews.llvm.org/D30651