4 ms·

SSE SIMD instructions should be unencumbered by patents at this point. So it's worthwhile to "bite the bullet" rather than do something like -mgeneral-regs-only

by jart 5y ago

SSE SIMD instructions should be unencumbered by patents at this point. So it's worthwhile to "bite the bullet" rather than do something like -mgeneral-regs-only. SIMD opcodes frequently give me 10x or even 54x speedups for the functions I write. I know this isn't Rust, but if anyone needs a more readable explanation of what the SIMD opcodes actually mean (i.e. more readable than intel manual) then check out the cosmopolitan codebase. I implemented all of them in ANSI C and called it ANSITRINSICS just so I didn't have to use Intel's ugly intrinsics API and wouldn't have to feel guilty about muh portability. It's well-tested and fuzzed against the real thing, so I hope it can serve as excellent documentation to anyone who wants to understand the great mystery of vector instructions: https://github.com/jart/cosmopolitan/tree/master/libc/intrin https://github.com/jart/cosmopolitan/tree/master/libc/intrin A good example is PUNPCKLBW which is such an important instruction but a name like that is very intimidating to someone who hasn't rote memorized its meaning. I think ANSI C does a much better job explaining what these instructions do than the pictures in the Intel manual do. https://github.com/jart/cosmopolitan/blob/48645651983ca46a31b3e85fb7b7e61500759f52/libc/intrin/punpcklbw.c#L21-L56 https://github.com/jart/cosmopolitan/blob/48645651983ca46a31... Particularly when it comes to ones like PMADDUBSW which is without a doubt the greatest of the SSE instructions. https://github.com/jart/cosmopolitan/blob/48645651983ca46a31b3e85fb7b7e61500759f52/libc/intrin/pmaddubsw.c#L24-L42 https://github.com/jart/cosmopolitan/blob/48645651983ca46a31... In many cases if you compile these ansitrinsics with -ftree-vectorize you'll notice that the compiler recognizes the ANSI C code boilerplate and compiles it down into the underlying simd instruction! I think GCC and Clang should be hard coded to recognize all of them. That way we can write truly portable code that's fast too.

jdonaldson 5y ago
Thank you for this, you're doing the Lord's work.
nwmcsween 5y ago
Trying to understand what does SIMD intrinsics offer in a 'libc' that SWAR couldn't do? For absoultely massive values to strlen, etc sure but IIRC the average strlen was something like 10 or less.
- jart 5y ago
  In the case of strlen, SSE goes a few cycles slower for n<8, gains a slight advantage over SWAR for n>8, and starts going ~twice as fast for n>255. Benchmarks: https://justine.lol/swarsimd.txt https://justine.lol/swarsimd.txt The difference is probably more pronounced in functions that do block copying like https://github.com/jart/cosmopolitan/blob/master/libc/str/stpcpy.c https://github.com/jart/cosmopolitan/blob/master/libc/str/st... (which actually uses the ANSITRINSICS library). For more complicated functions like this: https://github.com/jart/cosmopolitan/blob/b3838173ecbb67fae548008ae9a24ce7289fed1c/libc/str/tprecode8to16.c#L27-L44 https://github.com/jart/cosmopolitan/blob/b3838173ecbb67fae5... SIMD has the advantage that figuring out how to do the SWAR arithmetic correctly might be more difficult than figuring out how to use the hardware API. It goes without saying that the most compelling benefits of SSE are the instructions like PMADDUBSW which can be used to great effect for signal decimation https://github.com/jart/cosmopolitan/blob/b3838173ecbb67fae548008ae9a24ce7289fed1c/dsp/scale/cdecimate2xuint8x8.c#L41-L98 https://github.com/jart/cosmopolitan/blob/b3838173ecbb67fae5... assuming you use The Perfect Kernel {-1, -3, 3, 17, 17, 3, -3, -1} http://www.johncostella.com/magic/ http://www.johncostella.com/magic/ or PADDSW which can mix audio samples real fast doing saturating arithmetic. You can't SWAR that. Another great example is carryless multiplication instructions which literally make CRC32 computation go 54x faster.

by jart 5y ago

jdonaldson 5y ago

Thank you for this, you're doing the Lord's work.

nwmcsween 5y ago

Trying to understand what does SIMD intrinsics offer in a 'libc' that SWAR couldn't do? For absoultely massive values to strlen, etc sure but IIRC the average strlen was something like 10 or less.

jart 5y ago
In the case of strlen, SSE goes a few cycles slower for n<8, gains a slight advantage over SWAR for n>8, and starts going ~twice as fast for n>255. Benchmarks: https://justine.lol/swarsimd.txt https://justine.lol/swarsimd.txt The difference is probably more pronounced in functions that do block copying like https://github.com/jart/cosmopolitan/blob/master/libc/str/stpcpy.c https://github.com/jart/cosmopolitan/blob/master/libc/str/st... (which actually uses the ANSITRINSICS library). For more complicated functions like this: https://github.com/jart/cosmopolitan/blob/b3838173ecbb67fae548008ae9a24ce7289fed1c/libc/str/tprecode8to16.c#L27-L44 https://github.com/jart/cosmopolitan/blob/b3838173ecbb67fae5... SIMD has the advantage that figuring out how to do the SWAR arithmetic correctly might be more difficult than figuring out how to use the hardware API. It goes without saying that the most compelling benefits of SSE are the instructions like PMADDUBSW which can be used to great effect for signal decimation https://github.com/jart/cosmopolitan/blob/b3838173ecbb67fae548008ae9a24ce7289fed1c/dsp/scale/cdecimate2xuint8x8.c#L41-L98 https://github.com/jart/cosmopolitan/blob/b3838173ecbb67fae5... assuming you use The Perfect Kernel {-1, -3, 3, 17, 17, 3, -3, -1} http://www.johncostella.com/magic/ http://www.johncostella.com/magic/ or PADDSW which can mix audio samples real fast doing saturating arithmetic. You can't SWAR that. Another great example is carryless multiplication instructions which literally make CRC32 computation go 54x faster.