I knew a guy who worked at MS when they were developing the Barney doll. He signed up to beta/play-test the doll since he had a son in the target age range.
He left work on Friday with the new Barney doll.
When he came into work on the following Monday, he told his co-workers, "Looks like I'm going to HAVE to get a Barney doll for my son when they're released."
The power of Barney...
He also mentioned that when all the Actimates dolls and other consumer-related products were released, the internal-only Microsoft store looked like a techie-version of FAO Schwarz instead of a Microsoft-leaning Egghead software store.
The bean counters at MS killed a lot of product ideas when they came up with the high revenue bar for any possible new products - as if anyone could predict that stuff accurately.
The code in question has to process a string of variable length.
But the compiler/CPU can process bytes one at a time or much faster in groups. The code is trying to process as much as possible in groups of 128.
But since the caller can pass in a string which is not a mulitple of 128 chars, the first for-loop (& 127) will figure out how much of the string to process such that the remaining string length is a multiple of 128.
The second for-loop (>> 7) calculates divides by 128 (>> 7) to find out how many multiples of 128 there are to process. The inner for-loop processes 128 chars looking for 's' chars.
Now the for-loop within a for-loop doesn't look any faster than the plain single for-loop, but I'd assume that the heuristics of certain compilers can intuit that it can generate code to operate on multiple chars at the same time (SIMD instructions), since the result of one operation are independent of others.
On a compiler that cannot generate SIMD code, the code won't be much faster, if at all, than the naive straightforward manner.
If this code only runs on one compiler version/CPU arch, then ASSUMING the compiler will do the RIGHT THING and auto-vectorize the code is okay.
But if your code will be cross-platform/run on different OSes/CPU arch's, then a SWAR version may be more consistently performant - no need to guess if the compiler's optimization heuristics decided to go with the general purpose CPU registers or faster SIMD registers.
Downside is that the devs are exposed to the gnarly optimized code.
I assume the M1's SIMD registers are wider/more numerous than just the couple of size_t registers used for the loading/masking/accumulating inner loop in your run_swtches().
You can speedup the code by unrolling your inner loop a few times (try 4x or 8x) - it does mean that your overflow prevention limit is lowered (to a multiple of the unrolled grouping number) and run a few more times. But the speedup offsets the increased bookkeeping.
A version I played with showed increased speed by saving the in-progress accumulation in an array and then doing the final accumulation after the main loop is done. But that may be due to the CPU arch/compiler I'm using.
If I unroll the 64-bit SWAR version by 8x instead of 4x, the runtime is reduced by another 10% over the 4x-unrolled SWAR version. Diminishing returns...
Almost the same as my SWAR version - which is what you're doing.
But aren't you reading off the end of the buffer in your memcpy(&w...)? Say with an empty input string whose start address is aligned to sizeof(size_t) bytes?
I just passed in the string length since the caller had that info, otherwise you'd scan the whole string again looking for the zero terminator.
My SWAR version almost does what your vectorization algorithm description does - just that the SWAR-code looks rather gnarly because the compiler isn't auto-generating the vector code for you, it's hand-coded in C by me and I'm limited to 64 bits at a time.
He left work on Friday with the new Barney doll.
When he came into work on the following Monday, he told his co-workers, "Looks like I'm going to HAVE to get a Barney doll for my son when they're released."
The power of Barney...
He also mentioned that when all the Actimates dolls and other consumer-related products were released, the internal-only Microsoft store looked like a techie-version of FAO Schwarz instead of a Microsoft-leaning Egghead software store.
The bean counters at MS killed a lot of product ideas when they came up with the high revenue bar for any possible new products - as if anyone could predict that stuff accurately.