G
Guest
Guest
Despite the heat of getting a more optimized piece of FlaskMPEG code, I have ran a few benchmark test based on some alternatives program besides FlaskMPEG. Belows are my conclusion to my test.
I think FlaskMPEG is based on a rather unoptimized piece of code. Believe me, I tried on my dual Celeron 300A@450Mhz configuration, MPEG2AVI gave me 14.0 fps, already smoke out the fastest Thunderbird 1.2 Ghz SDR configuration. (I have tried on 32 bit MMX iDCT, 16 bit Chen MMX idCT)
And please note, MPEG2AVI is not written to fully benefit from dual processor combo, I only get 60% of my CPU time occupied on average. Which means if P4 1.5 Ghz is only getting 19 fps or Thunderbird 1.2 Ghz only getting 13.6 fps, there must be something SERIOUSLY wrong with the FlaskMPEG code. Extremely unoptimized piece of code I should say.
To those who try to draw conclusion on the current FlaskMPEG benchmark, I should warn you FlaskMPEG is not fully optimized, not even the MMX instruction, needless to say any of the 3DNow! or SSE2 instructions set. Drawing conclusion as which is faster is certainly too early.
I am still studying FlaskMPEG source code. If time allows, I will try to optimize the MMX iDCT code further to see what P4 and Athlon is capable at.
I think FlaskMPEG is based on a rather unoptimized piece of code. Believe me, I tried on my dual Celeron 300A@450Mhz configuration, MPEG2AVI gave me 14.0 fps, already smoke out the fastest Thunderbird 1.2 Ghz SDR configuration. (I have tried on 32 bit MMX iDCT, 16 bit Chen MMX idCT)
And please note, MPEG2AVI is not written to fully benefit from dual processor combo, I only get 60% of my CPU time occupied on average. Which means if P4 1.5 Ghz is only getting 19 fps or Thunderbird 1.2 Ghz only getting 13.6 fps, there must be something SERIOUSLY wrong with the FlaskMPEG code. Extremely unoptimized piece of code I should say.
To those who try to draw conclusion on the current FlaskMPEG benchmark, I should warn you FlaskMPEG is not fully optimized, not even the MMX instruction, needless to say any of the 3DNow! or SSE2 instructions set. Drawing conclusion as which is faster is certainly too early.
I am still studying FlaskMPEG source code. If time allows, I will try to optimize the MMX iDCT code further to see what P4 and Athlon is capable at.