Re: [V3] powerpc: Optimized strncmp for power10

[Peter Bergner <bergner@linux.ibm.com>]

On 4/29/24 4:58 AM, Amrita H S wrote:
> +++ b/sysdeps/powerpc/powerpc64/le/power10/strncmp.S
> @@ -0,0 +1,271 @@
> +/* Optimized strncmp implementation for PowerPC64/POWER10.
> +   Copyright (C) 2021-2023 Free Software Foundation, Inc.

This is a new file, so I believe the Copyright date should just be "2024".


> +++ b/sysdeps/powerpc/powerpc64/multiarch/strncmp-power10.S
> @@ -0,0 +1,25 @@
> +/* Copyright (C) 2016-2023 Free Software Foundation, Inc.

Likewise.



> +ENTRY_TOCLESS (STRNCMP, 4)
> +	/* Check if size is 0.  */
> +	cmpdi	 cr0,r5,0
> +	beq	 cr0,L(ret0)
> +	andi.   r7,r3,4095
> +	andi.   r8,r4,4095
> +	cmpldi  cr0,r7,4096-16
> +	cmpldi  cr1,r8,4096-16
> +	bgt     cr0,L(crosses)
> +	bgt     cr1,L(crosses)
> +	COMPARE_16(v4,v5,0)
> +	addi	r3,r3,16
> +	addi	r4,r4,16

This code looks like it assumes the kernel is using a 4k page size.
All distros that I know of and the default kernel config for ppc64
and ppc64le kernels is to use a 64K HW page size.  Is there a reason
we're not checking for a 64k cache boundary here?

Adhemerval, you seem to have added the first power8 strncmp.S optimized
routine (sysdeps/powerpc/powerpc64/power8/strncmp.S) and that also uses
a 4k page boundary.  Do you remember the history of why we checked for
a 4k page boundary rather than 64k?  Was is a matter of using 64k showed
no improvement over 4k and using 4k meant we didn't have to worry about
some system maybe running in 4k page size kernels?

Peter