Re: [FFmpeg-devel] [PATCH] apv_decode: Replace division with shift

[Mark Thompson <sw@jkqxz.net>]

On 27/04/2025 23:21, James Almer wrote:
> On 4/27/2025 3:49 PM, Mark Thompson wrote:
>> The compiler can't see that this should be a shift and generates a real
>> division which is slow enough to appear in profiles on its own.
>> ---
>>   libavcodec/apv_decode.c | 12 ++++++------
>>   1 file changed, 6 insertions(+), 6 deletions(-)
>>
>> diff --git a/libavcodec/apv_decode.c b/libavcodec/apv_decode.c
>> index e28bc29c8f..e2c3161978 100644
>> --- a/libavcodec/apv_decode.c
>> +++ b/libavcodec/apv_decode.c
>> @@ -173,8 +173,8 @@ static int apv_decode_tile_component(AVCodecContext *avctx, void *data,
>>       const AVPixFmtDescriptor *pix_fmt_desc =
>>           av_pix_fmt_desc_get(avctx->pix_fmt);
>>   -    int sub_w = comp_index == 0 ? 1 : pix_fmt_desc->log2_chroma_w + 1;
>> -    int sub_h = comp_index == 0 ? 1 : pix_fmt_desc->log2_chroma_h + 1;
>> +    int sub_w_shift = comp_index == 0 ? 0 : pix_fmt_desc->log2_chroma_w;
>> +    int sub_h_shift = comp_index == 0 ? 0 : pix_fmt_desc->log2_chroma_h;
>>         APVRawTile *tile = &input->tile[tile_index];
>>   @@ -190,8 +190,8 @@ static int apv_decode_tile_component(AVCodecContext *avctx, void *data,
>>       int tile_mb_width  = tile_width  / APV_MB_WIDTH;
>>       int tile_mb_height = tile_height / APV_MB_HEIGHT;
>>   -    int blk_mb_width  = 2 / sub_w;
>> -    int blk_mb_height = 2 / sub_h;
>> +    int blk_mb_width  = 2 >> sub_w_shift;
>> +    int blk_mb_height = 2 >> sub_h_shift;
>>         int bit_depth;
>>       int qp_shift;
>> @@ -234,10 +234,10 @@ static int apv_decode_tile_component(AVCodecContext *avctx, void *data,
>>                   for (int blk_x = 0; blk_x < blk_mb_width; blk_x++) {
>>                       int frame_y = (tile_start_y +
>>                                      APV_MB_HEIGHT * mb_y +
>> -                                   APV_TR_SIZE * blk_y) / sub_h;
>> +                                   APV_TR_SIZE * blk_y) >> sub_h_shift;
>>                       int frame_x = (tile_start_x +
>>                                      APV_MB_WIDTH * mb_x +
>> -                                   APV_TR_SIZE * blk_x) / sub_w;
>> +                                   APV_TR_SIZE * blk_x) >> sub_w_shift;
>>                         ptrdiff_t frame_pitch = apv->output_frame->linesize[comp_index];
>>                       uint8_t  *block_start = apv->output_frame->data[comp_index] +
> 
> LGTM.

Applied.

Thanks,

- Mark

_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".