[Qemu-devel] Performance bug in tcg/i386 output:

[Qemu-devel] Performance bug in tcg/i386 output:

[Chad]

[-- Attachment #1: Type: text/plain, Size: 417 bytes --]

Running qemu[-i386] with -d out_asm and looking at qemu.out shows this
test/j**/jmp pattern repeatedly:

0xb1aa40f6:  test   %ecx,%ecx
0xb1aa40f8:  jne    0xb1aa4115
0xb1aa40fe:  jmp    0xb1aa4103
0xb1aa4103:  mov    $0xc0bf,%eax

Therefore there's a bug in the tcg generator creating many unnecessary
jmps.  It might get optimized out by the CPU, but it might not and it bloats
the asm output in any case...

- Chad

[-- Attachment #2: Type: text/html, Size: 455 bytes --]

Re: [Qemu-devel] Performance bug in tcg/i386 output:

[Laurent Desnogues]

On Sun, Nov 8, 2009 at 6:47 PM, Chad <chad.page@gmail.com> wrote:
> Running qemu[-i386] with -d out_asm and looking at qemu.out shows this
> test/j**/jmp pattern repeatedly:
>
> 0xb1aa40f6:  test   %ecx,%ecx
> 0xb1aa40f8:  jne    0xb1aa4115
> 0xb1aa40fe:  jmp    0xb1aa4103
> 0xb1aa4103:  mov    $0xc0bf,%eax
>
> Therefore there's a bug in the tcg generator creating many unnecessary
> jmps.  It might get optimized out by the CPU, but it might not and it bloats
> the asm output in any case...

Some of these jumps are patched later when the target
block is translated (block chaining).

If you want to see the real output, you'll have to disassemble
after execution.


Laurent