Looks good, simple and nice. :-)

Looks good

Included one commit too many last time :)

bcopy and bzero are now also SIMDified

• Revert to previous strlen as the other version failed a test when running entire test suite

I dont think this one is gonna be worth including as I'm unable to comfortably beat the perf of the existing implementation.

remove manpage update, lets handle that in a separate commit

• fix Makefile

• Update based on review

• Use unsigned comparisons for main loop

strcat in C instead of assembly

alright, that makes sense. I'll replace it.

• revert accidental edit of the makefile

• Update based on review

• Update based on review

• amd64 -> aarch64 :)

Follow __$FUNC convention
ifdef's to simulate bcmp

follow __$FUNC convention

follow __$FUNC convention

Use unsigned comparisons for limit
follow __$FUNC convention

Update based on review.

• unsigned comparison for limit (b.mi -> b.lo)
• label function using __$FUNC convention

Use correct style for SPDX identifier

Pad .rodata with -1 and use unsigned comparisons

Use unsigned comparisons everywhere

Use unsigned comparisons everywhere

New method for handling short strings based on D46052

will rebase on D46052 to improve performance for the short case

Looks good, gonna revise the aarch64 port of this to handle it the same way.

Introduces a null "match" when limit is reached for short strings near a page boundary

• Revert to complicated handling of short string near a page boundary

I extended the unit tests and it was failing for short unaligned strings near a page boundary.

• Revert last commit, it wasn't unnecessary :)

In D45839#1047448, @fuz wrote:

Assuming this passes the test suite now, I'd say LGTM.

• Remove unnecessary mov, from strncmp review

Updated based on review

• Use correct offset for the loop in the normal case

• Updated based on review

Thanks for review! Changes appear to increase performance by a percent or two.

Fixes the page crossing for tails.

I'll write a proper new test instead testing more cases for the bounds

• Revert to branched handling of strings near the end of a page.

Hit the submit button too early, removing the branches for the tbl maneuver didn't work as I had hoped.

Remove branching for tbl maneuver.
Applies suggested microoptimizations for arithmetic.

• Remove unnecessary mov

• Avoid page crossing into unmapped pages

Passes all tests, need to extend memcmp tests to verify correct
behaviour for small buffers at page boundaries.

• Address comments

interleaves independent instructions for paralellism.
finds the offset in a branch free manner after a match.
simplifies the code for avoiding overreads in the main loop

• Avoid overreads in main loop

We re-compare some of the bytes we already compared to avoid crossing
over into an unmapped page.

I'm working on solving that right now, I mentioned it in the description.
The problem is that a buffer less than 32 bytes can be placed at the end of a page causing an overread.
I'll update it later today. I was mostly looking for some feedback at first.

• simd(7): also mention (NEON)

• simd(7): ASIMD instead of Neon