[LLVMbugs] [Bug 20311] New: Header cpuid.h assembly breaks

bugzilla-daemon at llvm.org bugzilla-daemon at llvm.org
Tue Jul 15 15:43:11 PDT 2014


http://llvm.org/bugs/show_bug.cgi?id=20311

            Bug ID: 20311
           Summary: Header cpuid.h assembly breaks
           Product: clang
           Version: unspecified
          Hardware: PC
                OS: All
            Status: NEW
          Severity: normal
          Priority: P
         Component: -New Bugs
          Assignee: unassignedclangbugs at nondot.org
          Reporter: gjasny at googlemail.com
                CC: llvmbugs at cs.uiuc.edu
    Classification: Unclassified

Hello,

the following occured with Xcode 5.1.1 but I don't see why it cannot occur also
with the header file shipped with clang.

I put the following into radar bug #17686779

Summary:
Hello,

I'm using Xcode 5.1.1(clang-503.0.40) on 10.10 DP3 and build 32bit i386
binaries.

The __cpuid macro from cupid.h:
    __asm("  pushl  %%ebx\n" \
          "  cpuid\n" \
          "  mov    %%ebx,%1\n" \
          "  popl   %%ebx" \
        : "=a"(__eax), "=r" (__ebx), "=c"(__ecx), "=d"(__edx) \
        : "0"(__level))

gets compiled into:
0x3921f5:  cpuid  
0x3921f7:  movl   %ebx, %ebx
0x3921f9:  popl   %ebx

The push/pop is used to preserve the pic register ebx across calls, but this
fails if ebx itself is picked as alternative register (=r) because the cpuid
output in ebx never reaches the target.

Steps to Reproduce:
I was not able to produce a minimal testcase because there esi was used instead
of ebx and the precondition for this bug was not met.

Expected Results:
The compiler must be told to not use ebx for the =r constraint.
Replacing =r in this macro with =S fixes the bug for me.

Actual Results:
Broken code

Version:
Xcode Version 5.1.1 (5B1008) (clang-503.0.40) on 10.10 DP3

-- 
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/20140715/c853e6d5/attachment.html>


More information about the llvm-bugs mailing list