Brotli Compression Algorithm Combines High Compression Ratio, and Fast Decompression

It’s worth noting that, unlike other compression algorithms, brotli uses a static dictionary based on common English, Spanish, HTML etc. That’s not a bad idea but it is sort-of cheating when you’re doing a comparison against other algorithms.

Using the term quality to compress a video file with brotli, kind of thru me off… should be called compression strength or something like that, no?

Hi, a good review and good for us all if Brotli manages to boost the browsing. @Jean-Luc Aufranc (CNXSoft) Please run your test with ‘dickens’ (from Silesia Corpus) too, it is one of the most used testdatafiles in compression community for a reason, it represents English texts in its fat book form (2 Bibles). Just use ‘-v’ option in order we all to see what is the Brotli’s decompression speed on your Vishera, on my old Core 2 laptop I reached 145 MB/s, for more info: https://github.com/google/brotli/issues/174 Also, I shared my opinion with Brotli’s team about their current defaults and… Read more »

@Jean-Luc Aufranc (CNXSoft) >I can test it, but could you provide the download link for dickens? Bravo, I very much want to see how “low caliber” CPUs execute them, ‘dickens’ is inhere: http://sun.aei.polsl.pl/~sdeor/index.php?page=silesia http://sun.aei.polsl.pl/~sdeor/corpus/dickens.bz2 >A single script that I can run to provide the results you want would be even nicer. Sure, but I am lame in *nix prompt, so an “estimate” for *nix: Step #1: Download the whole package all-in-one: https://mega.nz/#!c0hBkCYQ!U2irEQPbiig6xcUD0mXIVn7P-suzTcku6COKtkvKHmo Step #2: Download the C source: https://software.intel.com/sites/default/files/managed/6d/27/Nakamichi_Shifune.zip Step #3: Compile the C source: ./gcc -O3 -fomit-frame-pointer Nakamichi_Shifune.c -o Nakamichi_Shifune.elf -D_N_XMM -D_N_prefetch_4096 -D_N_Branchfull Step #4: ./Nakamichi_Shifune.elf dickens.Nakamichi /report Sorry… Read more »

@Jean-Luc Aufranc (CNXSoft) Oh, there is difference between GCC Windows and GCC Linux, maybe GCC is not fully installed. Just comment the line #940, SSE4.1 is actually not used just SSE2: 938 #ifdef _N_XMM 939 #include // SSE2 intrinsics //940 #include // SSE4.1 intrinsics 941 #endif It should work this way. I was lazy to make time reporting portable enough: k = (((float)1000*TargetSize/(clocks2 – clocks1 + 1))); k=k>>20; k=k*Trials; printf(“RAM-to-RAM performance: %d MB/s.\n”, k); If clocks_per_sec is not 1000 as in Windows, the MB/s would be unusable. Anyway, thanks for your FX8350 result, 195MB/s is still far from my ~512MB/s… Read more »

@Jean-Luc Aufranc (CNXSoft) Thank you for wrestling with my code 😉 If you still care, just replace the ‘1000’ with ‘CLOCKS_PER_SEC’ in next three lines and voila: //k = (((float)1000*TargetSize/(clocks2 - clocks1 + 1))); k=k>>20; k=k*Trials; k = (((float)CLOCKS_PER_SEC*TargetSize/(clocks2 - clocks1 + 1))); k=k>>20; k=k*Trials; ... //k = (((float)1000*SourceSize/(clocks2 - clocks1 + 1))); //k=k>>10; k = (((float)CLOCKS_PER_SEC*SourceSize/(clocks2 - clocks1 + 1))); //k=k>>10; ... //j = (float)1000*1024*( 512 )/ ((int) durationGENERIC); j = (float)CLOCKS_PER_SEC*1024*( 512 )/ ((int) durationGENERIC); This is the dump from my MinGW console: D:\Internetdownloads\Nakamichi_Shifune>gcc --version gcc (x86_64-posix-seh-rev0, Built by MinGW-W64 project) 5.1.0 Copyright (C) 2015 Free Software Foundation,… Read more »

@Jean-Luc Aufranc (CNXSoft)
Ugh, stupid of me, you said that it stuck still decompressing, don’t know what is the cause, maybe the XMM macro is the culprit, you could comment it and put below it standard memcpy() as below:


#ifdef _N_XMM
//SlowCopy128bit( (const char *)( (uint64_t)(srcLOCAL+1) ), retLOCAL );
memcpy(retLOCAL, (const char *)( (uint64_t)(srcLOCAL+1) ), 16);


Being little endian (as x86) it should work.

@Jean-Luc Aufranc (CNXSoft) >I had only compiled the code with: >gcc -O3 -fomit-frame-pointer Nakamichi_Shifune.c -o Nakamichi_Shifune.elf >maybe that’s why it looped. If this was the full compile line no wonder, you cannot omit ‘-D_N_XMM’, it should be like this: gcc -O3 -fomit-frame-pointer Nakamichi_Shifune.c -o Nakamichi_Shifune.elf -D_N_XMM -march=armv7v >Anyway, I’ve done the modifications with memcpy, and built it as on x86, but it will just segfault as I run the program. Never compiled for ARM, but shouldn’t you specify ARM target in GCC command options?! https://gcc.gnu.org/onlinedocs/gcc/ARM-Options.html It says that “not all architectures are recognized” automatically: -march=name This specifies the name of… Read more »