[llvm-bugs] [Bug 31754] New: replacing variable by a constant breaks subtract-with-borrow (sbb) optimization

[via llvm-bugs]

https://llvm.org/bugs/show_bug.cgi?id=31754

            Bug ID: 31754
           Summary: replacing variable by a constant breaks
                    subtract-with-borrow (sbb) optimization
           Product: new-bugs
           Version: 3.9
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: normal
          Priority: P
         Component: new bugs
          Assignee: unassignedbugs at nondot.org
          Reporter: vincent-llvm at vinc17.net
                CC: llvm-bugs at lists.llvm.org
    Classification: Unclassified

In some C code, the generated x86_64 code gets worse when I replace a variable
by a constant. Example:

typedef unsigned long T;

void sub (T *p, T u, T v, T w)
{
  p[0] = w - v;
  p[1] = u - (w < v);
}

void sub0 (T *p, T u, T v)
{
  T w = 0;
  p[0] = w - v;
  p[1] = u - (w < v);
}

void sub1 (T *p, T u, T v)
{
  T w = 1;
  p[0] = w - v;
  p[1] = u - (w < v);
}

sub0 and sub1 are like sub, except that w is 0 and 1 respectively.

With Clang 3.9.1 under Debian/unstable, I get from "clang-3.9 -O3 -S":

sub:                                    # @sub
        .cfi_startproc
# BB#0:
        subq    %rdx, %rcx
        movq    %rcx, (%rdi)
        sbbq    $0, %rsi
        movq    %rsi, 8(%rdi)
        retq

sub0:                                   # @sub0
        .cfi_startproc
# BB#0:
        movq    %rdx, %rax
        negq    %rax
        movq    %rax, (%rdi)
        cmpq    $1, %rdx
        adcq    $-1, %rsi
        movq    %rsi, 8(%rdi)
        retq

sub1:                                   # @sub1
        .cfi_startproc
# BB#0:
        movl    $1, %eax
        subq    %rdx, %rax
        movq    %rax, (%rdi)
        xorl    %eax, %eax
        cmpq    $1, %rdx
        seta    %al
        subq    %rax, %rsi
        movq    %rsi, 8(%rdi)
        retq

For sub0 and sub1, I would expect

        sbbq    $0, %rsi

like for sub.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20170125/7f8c9e14/attachment.html>