Help the compiler vectorize `adjacent_difference`

[AlexGuteniev]

📜 The approach

The following things prevented the original algorithm from vectorization:

• Loop-carried dependency, the previous input is used as one of operands.
  • This seems expected that the compiler doesn't transform such code to eliminate this automagically, too much of a transformation.
  • This was addressed by transforming the code to read the input array twice per iteration instead of carrying the values through the loop.
• Odd iterator pattern where the compiler cannot understand the iteration.
  • This seemed to me a strange limitation, so it was reported as DevCom-10742868.
  • This was addressed by using integer index.

🛑 Correctness concern

The standard defines exact steps for this algorithm. The optimization alters the steps. In particular the standard wants the subtracted value to be saved from the previous iteration, rather than being read again. The two below sections explain what precautions are made to make the change unobservable, so I hope the change is correct.

✅ Checks for eligibility

The following checks were added:

• No Aliasing (see below)
• Iterators can be pointers
• Source iterator is not volatile (read order is altered)
• Trivially copyable (we skip copying where the standard asks for it)

There's no need in check for integral types or so, since the compiler makes the final decision anyway, and it may be able optimize even something that wouldn't pass a strict check.

⚠️ No Aliasing

Apparently there's no rule that the source and the destination ranges may not overlap. We should handle aliasing.

Unlike the #4431 precedent, we can't yield to the compiler here. The compiler is able to insert overlaps check that prevents vectorization and go to the scalar fallback in case of checks failure, but:

• We apply transformation that would change the meaning of the program in case of overlapping range, and the meaning would be changed no matter if vectorization happens
• The checks that compiler inserts may be too loose, it may allow like equal source and destination pointer, as these are thc checks if the transformed algorithm would not change the meaning

So we do our own checks.

Then we tell the compiler with __restrict that we already checked, and it should not bother. This is done in a separate function, because the __restrict is not aliased within scope, so saying __restrict within the original algorithm would apparently be a lie.

The extra check by the compiler, if not prevented would slightly add run time and dead code size.

😾 Compiler warnings

We have a great feature called integral promotion, or whatever it is called. Smaller types are converted to integers, and there is a warning about converting them back. Local pragma does suppresses them in benchmark, but not in the test.

I don't know what to do with this.

⏱️ Benchmark results

Benchmark	main	this	this + AVX2
bm<uint8_t>/2255	745 ns	563 ns	562 ns
bm<uint16_t>/2255	799 ns	83.3 ns	75.1 ns
bm<uint32_t>/2255	731 ns	154 ns	141 ns
bm<uint64_t>/2255	805 ns	293 ns	272 ns
bm/2255	751 ns	154 ns	123 ns
bm/2255	753 ns	304 ns	233 ns

🥇 Results interpretation

• Overall, we're good 😸
• 8-bit case failed to vectorize for no reason (didn't look up if it is known compiler issue, or to be reported)
• Still 8-bit case is noticeably better. I didn't analyze that, but looks like this a consistent thing, not codegen gremlins. I think it is a side effect of eliminating loop-carried dependency, so the processor can parallelize and overlap iterations
• AVX2 is only slightly faster. I did not analyze, but think that memory wall is being hit here 🧱

[AlexGuteniev]

• 8-bit case failed to vectorize for no reason (didn't look up if it is known compiler issue, or to be reported)

Reported DevCom-10745948

[CaseyCarter]

• 8-bit case failed to vectorize for no reason (didn't look up if it is known compiler issue, or to be reported)

Interestingly it vectorizes if we use - directly instead of indirecting through std::minus, or if the output is a pointer to int. Something to do with narrowing the result of the promoted operation, maybe?

[StephanTLavavej]

Thanks! :heart_eyes_cat: I pushed minor nitpicks and a significant fix for C++14/17. Speedups look good on my 5950X:

Benchmark	Before	After	Speedup
bm<uint8_t>/2255	968 ns	967 ns	1.00
bm<uint16_t>/2255	917 ns	97.2 ns	9.43
bm<uint32_t>/2255	648 ns	158 ns	4.10
bm<uint64_t>/2255	689 ns	331 ns	2.08
bm<float>/2255	646 ns	158 ns	4.09
bm<double>/2255	652 ns	332 ns	1.96

[StephanTLavavej]

I'm mirroring this to the MSVC-internal repo - please notify me if any further changes are pushed.

[StephanTLavavej]

I had to push an additional commit to fix the overlap check for heterogeneous types.

[StephanTLavavej]

Thanks for helping the compiler, said the author of the presentation, Don't Help The Compiler :joy_cat: :heart_eyes_cat: :tada:

[AlexGuteniev]

Good news -- the 8 bit case is auto-vectorized now, despite DevCom-10745948 is not marked as resolved.