[LLVMbugs] [Bug 12567] New: Alias analysis confused by loop

bugzilla-daemon at llvm.org bugzilla-daemon at llvm.org
Mon Apr 16 14:49:36 PDT 2012


             Bug #: 12567
           Summary: Alias analysis confused by loop
           Product: libraries
           Version: trunk
          Platform: PC
        OS/Version: Linux
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: Global Analyses
        AssignedTo: unassignedbugs at nondot.org
        ReportedBy: jeroen.dobbelaere at synopsys.com
                CC: llvmbugs at cs.uiuc.edu
    Classification: Unclassified

Created attachment 8394
  --> http://llvm.org/bugs/attachment.cgi?id=8394
test file that shows the problem

Compile the attached file with:
  clang -O3 -S test_alias.c 

When p_a, p_b and p_c are initialized outside the loop, llvm is not able to
deduce that p_a, p_b and p_c will not alias. Because of that it will keep the
writes to '*p_c', resulting in slower code.

Note: If the initialization is done inside the loop, everything is fine.
Note: gcc is able to deduce that these will not alias in both cases.

Code that clang/llvm produces:
    .file    "test_alias_002.c"
    .globl    foo_002
    .align    16, 0x90
    .type    foo_002, at function
foo_002:                                # @foo_002
# BB#0:                                 # %entry
    xorl    %eax, %eax
    .align    16, 0x90
.LBB0_1:                                # %for.body
                                        # =>This Inner Loop Header: Depth=1
    movl    $0, c(%rax)
    movl    b(,%rax,4), %ecx
    imull    a(,%rax,4), %ecx
    movl    %ecx, c(%rax)          #### Unnecessary store
    movl    b+4(,%rax,4), %edx
    imull    a+4(,%rax,4), %edx
    addl    %ecx, %edx
    leaq    4(%rax), %rcx
    movl    %edx, c(%rax)          #### Unnecessary store
    movl    b+8(,%rax,4), %esi
    imull    a+8(,%rax,4), %esi
    addl    %edx, %esi
    movl    %esi, c(%rax)          #### Unnecessary store
    movl    b+12(,%rax,4), %edx
    imull    a+12(,%rax,4), %edx
    addl    %esi, %edx
    cmpl    $128, %ecx
    movl    %edx, c(%rax)
    movq    %rcx, %rax
    jne    .LBB0_1
# BB#2:                                 # %for.end



Configure bugmail: http://llvm.org/bugs/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.

More information about the llvm-bugs mailing list