Vectorize basic_string::rfind (the string needle overload)

[AlexGuteniev]

Towards #5036.

Similar to #5048, the differences:

• The parameters calculations that use _Start_at are trickier
• The input_needle.empty() special case in the tests is really special case this time
• Used intentionally not-very-aligned strings, towards Provide an intentional (mis)alignment that corresponds to typical usage in benchmarks for plain arrays #5035

Benchmark	main	this
member_rfind<not_highly_aligned_string>/0	108 ns	23.2 ns
member_rfind<not_highly_aligned_string>/1	101 ns	22.6 ns
member_rfind<not_highly_aligned_string>/2	1298 ns	305 ns
member_rfind<not_highly_aligned_string>/3	1554 ns	347 ns
member_rfind<not_highly_aligned_string>/4	7914 ns	4870 ns
member_rfind<not_highly_aligned_string>/5	15104 ns	15482 ns
member_rfind<not_highly_aligned_wstring>/0	98.1 ns	39.7 ns
member_rfind<not_highly_aligned_wstring>/1	101 ns	38.9 ns
member_rfind<not_highly_aligned_wstring>/2	1288 ns	624 ns
member_rfind<not_highly_aligned_wstring>/3	1559 ns	722 ns
member_rfind<not_highly_aligned_wstring>/4	5645 ns	9777 ns
member_rfind<not_highly_aligned_wstring>/5	15808 ns	20634 ns

STL/benchmarks/src/search.cpp

Lines 41 to 49 in a1bc126

	constexpr data_and_pattern patterns[] = {
	/* 0. Small, closer to end */ {lorem_ipsum, "aliquet"sv},
	/* 1. Large, closer to end */ {lorem_ipsum, "aliquet malesuada"sv},
	/* 2. Small, closer to begin */ {lorem_ipsum, "pulvinar"sv},
	/* 3. Large, closer to begin */ {lorem_ipsum, "dapibus elit interdum"sv},
	/* 4. Small, evil */ {fill_pattern_view<3000, false>, fill_pattern_view<7, true>},
	/* 5. Large, evil */ {fill_pattern_view<3000, false>, fill_pattern_view<20, true>},
	};

[StephanTLavavej]

Results on my 5950X:

Benchmark	Before	After	Speedup	Notes
member_rfind<not_highly_aligned_string>/0	84.4 ns	21.5 ns	3.93	0: Small, closer to end
member_rfind<not_highly_aligned_string>/1	78.9 ns	21.1 ns	3.74	1: Large, closer to end
member_rfind<not_highly_aligned_string>/2	1336 ns	196 ns	6.82	2: Small, closer to begin
member_rfind<not_highly_aligned_string>/3	1590 ns	253 ns	6.28	3: Large, closer to begin
member_rfind<not_highly_aligned_string>/4	10408 ns	6720 ns	1.55	4: Small, evil
member_rfind<not_highly_aligned_string>/5	12419 ns	20189 ns	0.62	5: Large, evil
member_rfind<not_highly_aligned_wstring>/0	108 ns	32.3 ns	3.34	0: Small, closer to end
member_rfind<not_highly_aligned_wstring>/1	104 ns	31.5 ns	3.30	1: Large, closer to end
member_rfind<not_highly_aligned_wstring>/2	1845 ns	441 ns	4.18	2: Small, closer to begin
member_rfind<not_highly_aligned_wstring>/3	2148 ns	506 ns	4.25	3: Large, closer to begin
member_rfind<not_highly_aligned_wstring>/4	5756 ns	13644 ns	0.42	4: Small, evil
member_rfind<not_highly_aligned_wstring>/5	15833 ns	24849 ns	0.64	5: Large, evil

These are great speedups with only the evil cases being pessimized, and they aren't absolutely horrible.

[StephanTLavavej]

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

[StephanTLavavej]

😹 🚀 😻 function! member basic_string important another vectorizing for Thanks