[LLVMbugs] [Bug 19412] New: Errorneous assumption of 8 bit argument widened to 32 bit

[bugzilla-daemon at llvm.org]

http://llvm.org/bugs/show_bug.cgi?id=19412

            Bug ID: 19412
           Summary: Errorneous assumption of 8 bit argument widened to 32
                    bit
           Product: clang
           Version: 3.4
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: normal
          Priority: P
         Component: LLVM Codegen
          Assignee: unassignedclangbugs at nondot.org
          Reporter: sebastiangraf at t-online.de
                CC: llvmbugs at cs.uiuc.edu
    Classification: Unclassified

When using https://github.com/vmt/udis86, I hit this both on my Mac and my
Linux PC using clang as compiler targeting x64.

To reproduce (python 2.7 is needed as default python, I submitted a pull
request to fix that but it wasn't accepted yet):
1. Download the repository
2. $ ./configure --prefix=`pwd`/bin CC="clang" from the root of the repository
3. $ make && make install   (will not make if it couldn't use python 2.7 for
codegen)
4. Examine ud_set_mode in ./bin/lib/libudis.so with gdb
5. $ gdb ./bin/lib/libudis86.so
6. (gdb) info line ud_set_mode
7. (gdb) x/13i 0x11100    (start of ud_set_mode was 0x11100 for me)

This results in the assembly listing for ud_set_mode in ./libudis/udis86.c:86:
(gdb) x/13i 0x11100
   0x11100 <ud_set_mode>:    lea    -0x10(%rsi),%eax
   0x11103 <ud_set_mode+3>:    cmp    $0x30,%eax
   0x11106 <ud_set_mode+6>:    ja     0x11120 <ud_set_mode+32>
   0x11108 <ud_set_mode+8>:    movabs $0x1000000010001,%rcx
   0x11112 <ud_set_mode+18>:    bt     %rax,%rcx
   0x11116 <ud_set_mode+22>:    jae    0x11120 <ud_set_mode+32>
   0x11118 <ud_set_mode+24>:    mov    %sil,0x170(%rdi)
   0x1111f <ud_set_mode+31>:    retq   
   0x11120 <ud_set_mode+32>:    movb   $0x10,0x170(%rdi)
   0x11127 <ud_set_mode+39>:    retq   
   0x11128:    nopl   0x0(%rax,%rax,1)
   0x11130 <ud_set_pc>:    mov    %rsi,0x178(%rdi)
   0x11137 <ud_set_pc+7>:    retq


Note that the 8 bit argument m is assumed to be 32 bits wide in 0x11103. For
completeness sake, the assembly output from udis86.s generated by using clang
-S ...:
ud_set_mode:                            # @ud_set_mode
    .cfi_startproc
# BB#0:
                                        # kill: ESI<def> ESI<kill> RSI<def>
    leal    -16(%rsi), %eax
    cmpl    $48, %eax
    ja    .LBB1_3
# BB#1:
    movabsq    $281474976776193, %rcx  # imm = 0x1000000010001
    btq    %rax, %rcx
    jae    .LBB1_3
# BB#2:
    movb    %sil, 368(%rdi)
    ret


The respective IR:
 Function Attrs: nounwind uwtable
define void @ud_set_mode(%struct.ud* nocapture %u, i8 zeroext %m) #0 {
  %1 = zext i8 %m to i32
  switch i32 %1, label %4 [
    i32 16, label %2
    i32 32, label %2
    i32 64, label %2
  ]

; <label>:2                                       ; preds = %0, %0, %0
  %3 = getelementptr inbounds %struct.ud* %u, i64 0, i32 18
  store i8 %m, i8* %3, align 1, !tbaa !1
  br label %6

; <label>:4                                       ; preds = %0
  %5 = getelementptr inbounds %struct.ud* %u, i64 0, i32 18
  store i8 16, i8* %5, align 1, !tbaa !1
  br label %6

; <label>:6                                       ; preds = %4, %2
  ret void
}


And the C function:
extern void 
ud_set_mode(struct ud* u, uint8_t m)
{
  switch(m) {
  case 16:
  case 32:
  case 64: u->dis_mode = m ; return;
  default: u->dis_mode = 16; return;
  }
}


The IR actually seems to be correct, so maybe this is more of a backend issue.
The zero extension obviously doesn't happen. I wasn't able to reproduce it with
a more simple case, so I'm sorry for that rather long reproduction recipe.

-- 
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/20140412/38bdde5b/attachment.html>