[ptweezy]

This PR makes the AVX2 inflate decoder (decode_huffman_code_block_stateless_04) copy matches 32 bytes at a time with ymm registers instead of 16 with xmm. A match of up to 32 bytes whose distance is at least its length takes one store; a match whose distance is less than both 32 and its length still goes through small_byte_copy. The main-loop output slop grows from 274 to 290 bytes, exactly enough for the longest stores.

This applies to every x86-64 CPU with AVX2; _01 assembles to the same code as on master.

Output

Decoded output does not change. On main-loop ISAL_INVALID_LOOKBACK and ISAL_INVALID_SYMBOL failures, where master already miscounts (#446), stateless avail_out comes out 290 bytes short instead of 274, and the leftover bytes in the overcounted tail of the output differ.

Conflicts

This PR conflicts with #446 in igzip/igzip_decode_block_stateless.asm.

Results

Inflate on GitHub runners, in-process A/B, two runs per runner. Each cell is master's best-of-11 MB/s and the median over runs of this PR's change in best-of-11 throughput (harness, runs 1, 2, 3). 4 KB and 1 KB isa-l L1 slices change by 1.2% or less.

CPU (runners)	Silesia, zlib -6	64 KB slices, isa-l L1
Zen 3 EPYC 7763 (4)	850, +1.0%	686, +0.7%
Zen 4 EPYC 9V74 (6)	986, +1.6%	806, +1.3%
Zen 5 EPYC 9V45 (6)	1159, +2.7%	936, +1.9%
Ice Lake Xeon 8370C (1)	818, +1.5%	666, +1.2%
Emerald Rapids Xeon 8573C (5)	852, +2.0%	709, +1.8%
Granite Rapids Xeon 6973P-C (2)	958, +1.6%	792, +1.5%

Testing

• make check passes with _04, with _01 forced, and natively on Windows (mingw-w64), where 2.2 million win64 calls per decoder leave callee-saved registers and the upper ymm state clean.
• Worst-case streams inflated into every output size, ending at a guard page, cause no faults; with a slop one byte smaller, _04 faults.
• 8,000 valid and corrupted streams give master's results except in the failures above; with igzip: fix x86 inflate counts when the main loop finds an error #446 applied to both, all match.