[FFmpeg-devel] [PATCH] Fix apply_welch_window_sse2 compilation on Mac OS X/x86

[Pierre d'Herbemont]

On Oct 17, 2007, at 10:05 PM, Loren Merritt wrote:

> But I don't think the fix is correct: It's not valid to jmp from  
> one asm block to another. If you split the asm block, you also have  
> to write the loop part in C. And the reason I put it in asm is that  
> gcc doesn't generate as efficient a loop.

There is no problem in jumping from one asm block to an other.  
However there can be problem with labels.

Here, the only reason of jmp failing would be the definition of the  
1: label by gcc between the two blocks. If I recall well gcc does  
name it's label Lxxx, so no problem here. Moreover there shouldn't be  
any jmp in the in-between code.

Here is the assembly output I get:

   22531     1:
   22532     movapd   %xmm7,  %xmm1
   22533     mulpd    %xmm1,  %xmm1
   22534     movapd   %xmm6,  %xmm0
   22535     subpd    %xmm1,  %xmm0
   22536     pshufd   $0x4e,   %xmm0, %xmm1
   22537     cvtpi2pd (%edx,%esi), %xmm2
   22538     cvtpi2pd (%eax,%ecx), %xmm3
   22539
   22540     sall    $3, %edi
   22541     movl    %edi, -64(%ebp)
   22542     movl    -48(%ebp), %edx
   22543     addl    %edi, %edx
   22544     movl    12(%ebp), %edi
   22545     movl    -48(%ebp), %eax
   22546     leal    (%eax,%edi,8), %edi
   22547     movl    %edi, -60(%ebp)
   22548     movl    -64(%ebp), %eax
   22549     subl    %eax, %edi
   22550     movl    %edi, %eax
   22551     subl    $16, %eax
   22552     mulpd    %xmm0,  %xmm2
   22553     mulpd    %xmm1,  %xmm3
   22554     movapd   %xmm2, (%edx,%esi,2)
   22555     movupd    %xmm3, (%eax,%ecx,2)
   22556     subpd    %xmm5,  %xmm7
   22557     sub      $8,      %ecx
   22558     add      $8,      %esi
   22559     jl 1b

Thanks for the review,

Pierre.