[FFmpeg-devel] [PATCH] Make swscale x86 ASM depend less on registers been preserved across blocks

[Vitor Sessak]

On Wed, Sep 17, 2014 at 10:31 PM, Michael Niedermayer <michaelni at gmx.at> wrote:
> On Wed, Sep 17, 2014 at 09:21:08PM +0200, Vitor Sessak wrote:
>> Hi!
>>
>> I'm not sure we can count on registers been preserved across ASM
>> blocks. Moreover, this causes problems with utils that instrument the
>> code by inserting function calls between lines (like ThreadSanitizer).
>>
>> The attached patch fix one instance of this problem in swscale.
>>
>> -Vitor
>
>>  swscale.c |   82 +++++++++++++++++++++++++++++++-------------------------------
>>  1 file changed, 42 insertions(+), 40 deletions(-)
>> 7fe9f7a0c1a7c1ed843b891f5f810fbd94a2e03f  0001-swscale-x86-do-not-expect-registers-to-be-preserved-.patch
>> From f1c59628b2baeb9994bed8947c0a2c228610bf4f Mon Sep 17 00:00:00 2001
>> From: Vitor Sessak <vsessak at google.com>
>> Date: Wed, 17 Sep 2014 21:10:16 +0200
>> Subject: [PATCH] swscale/x86: do not expect registers to be preserved across
>>  inline ASM blocks
>>
>> ---
>>  libswscale/x86/swscale.c | 82 +++++++++++++++++++++++++-----------------------
>>  1 file changed, 42 insertions(+), 40 deletions(-)
>>
>> diff --git a/libswscale/x86/swscale.c b/libswscale/x86/swscale.c
>> index c4c0e28..38157c8 100644
>> --- a/libswscale/x86/swscale.c
>> +++ b/libswscale/x86/swscale.c
>> @@ -205,36 +205,19 @@ static void yuv2yuvX_sse3(const int16_t *filter, int filterSize,
>>          yuv2yuvX_mmxext(filter, filterSize, src, dest, dstW, dither, offset);
>>          return;
>>      }
>> -    if (offset) {
>> -        __asm__ volatile("movq       (%0), %%xmm3\n\t"
>> -                         "movdqa    %%xmm3, %%xmm4\n\t"
>> -                         "psrlq       $24, %%xmm3\n\t"
>> -                         "psllq       $40, %%xmm4\n\t"
>> -                         "por       %%xmm4, %%xmm3\n\t"
>> -                         :: "r"(dither)
>> -                         );
>> -    } else {
>> -        __asm__ volatile("movq       (%0), %%xmm3\n\t"
>> -                         :: "r"(dither)
>> -                         );
>> -    }
>
>> -    filterSize--;
>
> not sure i missed it but it seems this is lost in teh new code

Good catch. Fixed!

>> -    __asm__ volatile(
>> -        "pxor      %%xmm0, %%xmm0\n\t"
>> -        "punpcklbw %%xmm0, %%xmm3\n\t"
>> -        "movd          %0, %%xmm1\n\t"
>> -        "punpcklwd %%xmm1, %%xmm1\n\t"
>> -        "punpckldq %%xmm1, %%xmm1\n\t"
>> -        "punpcklqdq %%xmm1, %%xmm1\n\t"
>> -        "psllw         $3, %%xmm1\n\t"
>> -        "paddw     %%xmm1, %%xmm3\n\t"
>> -        "psraw         $4, %%xmm3\n\t"
>> -        ::"m"(filterSize)
>> -     );
>> -    __asm__ volatile(
>> -        "movdqa    %%xmm3, %%xmm4\n\t"
>> -        "movdqa    %%xmm3, %%xmm7\n\t"
>> -        "movl %3, %%ecx\n\t"
>> +#define MAIN_FUNCTION \
>> +        "pxor       %%xmm0, %%xmm0 \n\t" \
>> +        "punpcklbw  %%xmm0, %%xmm3 \n\t" \
>> +        "movd           %4, %%xmm1 \n\t" \
>> +        "punpcklwd  %%xmm1, %%xmm1 \n\t" \
>> +        "punpckldq  %%xmm1, %%xmm1 \n\t" \
>> +        "punpcklqdq %%xmm1, %%xmm1 \n\t" \
>> +        "psllw          $3, %%xmm1 \n\t" \
>> +        "paddw      %%xmm1, %%xmm3 \n\t" \
>> +        "psraw          $4, %%xmm3 \n\t" \
>> +        "movdqa     %%xmm3, %%xmm4 \n\t" \
>> +        "movdqa     %%xmm3, %%xmm7 \n\t" \
>> +        "movl           %3, %%ecx  \n\t" \
>>          "mov                                 %0, %%"REG_d"  \n\t"\
>>          "mov                        (%%"REG_d"), %%"REG_S"  \n\t"\
>>          ".p2align                             4             \n\t" /* FIXME Unroll? */\
>> @@ -252,20 +235,39 @@ static void yuv2yuvX_sse3(const int16_t *filter, int filterSize,
>>          " jnz                                1b             \n\t"\
>>          "psraw                               $3, %%xmm3      \n\t"\
>>          "psraw                               $3, %%xmm4      \n\t"\
>> -        "packuswb                         %%xmm4, %%xmm3      \n\t"
>> -        "movntdq                          %%xmm3, (%1, %%"REG_c")\n\t"
>> +        "packuswb                         %%xmm4, %%xmm3      \n\t"\
>> +        "movntdq                          %%xmm3, (%1, %%"REG_c")\n\t"\
>>          "add                         $16, %%"REG_c"         \n\t"\
>>          "cmp                          %2, %%"REG_c"         \n\t"\
>> -        "movdqa    %%xmm7, %%xmm3\n\t"
>> -        "movdqa    %%xmm7, %%xmm4\n\t"
>> +        "movdqa                   %%xmm7, %%xmm3            \n\t" \
>> +        "movdqa                   %%xmm7, %%xmm4            \n\t" \
>>          "mov                                 %0, %%"REG_d"  \n\t"\
>>          "mov                        (%%"REG_d"), %%"REG_S"  \n\t"\
>> -        "jb                                  1b             \n\t"\
>> -        :: "g" (filter),
>> -           "r" (dest-offset), "g" ((x86_reg)(dstW+offset)), "m" (offset)
>> -        : XMM_CLOBBERS("%xmm0" , "%xmm1" , "%xmm2" , "%xmm3" , "%xmm4" , "%xmm5" , "%xmm7" ,)
>> -         "%"REG_d, "%"REG_S, "%"REG_c
>> -    );
>> +        "jb                                  1b             \n\t"
>> +
>> +    if (offset) {
>> +        __asm__ volatile(
>> +            "movq          %5, %%xmm3  \n\t"
>> +            "movdqa    %%xmm3, %%xmm4  \n\t"
>> +            "psrlq        $24, %%xmm3  \n\t"
>> +            "psllq        $40, %%xmm4  \n\t"
>> +            "por       %%xmm4, %%xmm3  \n\t"
>> +            MAIN_FUNCTION
>> +           :: "g" (filter),
>> +           "r" (dest-offset), "g" ((x86_reg)(dstW+offset)), "m" (offset),
>> +           "m"(filterSize), "m"(((uint64_t *) dither)[0])
>> +           : "%"REG_d, "%"REG_S, "%"REG_c
>> +           );
>
> you lost the XMM_CLOBBERS()

True. Restored it.

>> +    } else {
>> +        __asm__ volatile(
>> +            "movq          %5, %%xmm3   \n\t"
>> +            MAIN_FUNCTION
>> +           :: "g" (filter),
>> +           "r" (dest-offset), "g" ((x86_reg)(dstW+offset)), "m" (offset),
>> +           "m"(filterSize), "m"(((uint64_t *) dither)[0])
>> +           : "%"REG_d, "%"REG_S, "%"REG_c
>> +           );
>
> tabs

Why the hell the default emacs profile is so bad :-(

New patch attached.

-Vitor
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0001-swscale-x86-do-not-expect-registers-to-be-preserved-.patch
Type: application/octet-stream
Size: 5181 bytes
Desc: not available
URL: <https://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20140917/89130fde/attachment.obj>