[FFmpeg-devel] [PATCH] [PATCH] POWER8 VSX vectorization libswscale/input.c Track ticket 5570

[Pestov Vyacheslav]

On 25.03.2020 1:12, Carl Eugen Hoyos wrote:

> Am Di., 24. März 2020 um 14:28 Uhr schrieb Pestov Vyacheslav
> <pestov.vyach at yandex.ru>:
>> yuy2ToY_c: 10157
>> yuy2ToY_c_vsx: 2353
>>
>> yuy2ToUV_c: 4907
>> yuy2ToUV_c_vsx: 1357
>>
>> rgb24ToY_c: 21172
>> rgb24ToY_c_vsx: 9191
>>
>> rgb24ToUV_c: 33568
>> rgb24ToUV_c_vsx: 12746
>>
>> bgr24ToY_c: 20983
>> bgr24ToY_c_vsx: 9381
>>
>> bgr24ToUV_c: 34513
>> bgr24ToUV_c_vsx: 12708
>>
>> monowhite2Y_c: 5247
>> monowhite2Y_c_vsx: 2099
>>
>> monoblack2Y_c: 5584
>> monoblack2Y_c_vsx: 1993
>>
>> uyvyToY_c: 10111
>> uyvyToY_c_vsx: 1780
>>
>> uyvyToUV_c: 4872
>> uyvyToUV_c_vsx: 1284
>>
>> nvXXtoUV_c: 5128
>> nvXXtoUV_c_vsx: 1456
>>
>> rgbaToA_c: 9930
>> rgbaToA_c_vsx: 2599
>>
>> bswap16Y_c: 10399
>> bswap16Y_c_vsx: 2451
>>
>> rgb16_32ToUV_half_c_template: 42350
>> rgb16_32ToUV_half_c_template_vsx: 18583
>>
>> bswap16UV_c: 11784
>> bswap16UV_c_vsx: 2873
>>
>> planar_rgb_to_y: 24602
>> planar_rgb_to_y_vsx: 10792
>>
>> planar_rgb_to_uv: 35601
>> planar_rgb_to_uv_vsx: 14112
>>
>> planar_rgb16_to_y: 25686
>> planar_rgb16_to_y_vsx: 10293
>>
>> planar_rgb16_to_uv: 36367
>> planar_rgb16_to_uv_vsx: 13575
>>
>> yvy2ToUV_c: 4879
>> yvy2ToUV_c_vsx: 1239
>>
>> read_ya16be_gray_c: 9591
>> read_ya16be_gray_c_vsx: 4164
>>
>> read_ya16be_alpha_c: 9390
>> read_ya16be_alpha_c_vsx: 1874
>>
>> read_ya16le_gray_c: 9884
>> read_ya16le_gray_c_vsx: 4224
>>
>> read_ya16le_alpha_c: 9403
>> read_ya16le_alpha_c_vsx: 2026
>>
>> planar_rgb_to_a: 10262
>> planar_rgb_to_a_vsx: 9361
>>
>> planar_rgb16_to_a: 9554
>> planar_rgb16_to_a_vsx: 9393
>>
>> read_ayuv64le_Y_c: 10457
>> read_ayuv64le_Y_c_vsx: 7703
>>
>> read_ayuv64le_A_c: 9404
>> read_ayuv64le_A_c_vsx: 2797
>>
>> read_ayuv64le_UV_c: 9464
>> read_ayuv64le_UV_c_vsx: 3781
>>
>> p010LEToY_c: 9546
>> p010LEToY_c_vsx: 2422
>>
>> p010LEToUV_c: 6390
>> p010LEToUV_c_vsx: 2681
>>
>> p010BEToY_c: 9836
>> p010BEToY_c_vsx: 2572
>>
>> p010BEToUV_c: 7022
>> p010BEToUV_c_vsx: 2660
>>
>> p016LEToUV_c: 5022
>> p016LEToUV_c_vsx: 2447
>>
>> p016BEToUV_c: 5293
>> p016BEToUV_c_vsx: 2307
> To make our lives a little easier, could you tell us what you tested
> and how we can reproduce your results?
>
> Also: Is your patch expected to be bit-exact? If yes, do you
> have a script that allows to compare C and vsx code?
> If not, how did you test your code?
> (Or does fate cover these conversions? I wouldn't expect so.)
>
> Thank you, Carl Eugen
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".

Hi, yes I am using some scripts(see attached archive):

1) Put macros into tested functions in input_vsx.c and input.c   (START_TIMER and STOP_TIMET). 
You can take files with macros from my dropbox 
https://www.dropbox.com/sh/eoed5pjp9a9psx0/AACpsa6PKGAIl5pYF58sHeRda?dl=0

2) I am generating some random input file with dd utility. For example:

dd if=/dev/urandom of=/tmp/test.raw bs=1024 count=40900

3) Run script script ffmpeg_bench.sh from directory where created test.raw

cd /tmp

bash ./ffmpeg_bench.sh

4) Run python3 script print_result.py.  This script takes the ffmpeg_bench results from files bench_cpu.txt and bench_vsx.txt, calculates the average values and displays in a convenient form 

With best regards, Pestov Vyacheslav

-------------- next part --------------
A non-text attachment was scrubbed...
Name: attach.zip
Type: application/zip
Size: 1866 bytes
Desc: not available
URL: <https://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20200325/e5b4468b/attachment.zip>