powerpc: optimize strcpy/stpcpy for POWER9/10

  This patch modifies the current POWER9 implementation of strcpy and
stpcpy to optimize it for POWER9/10.

Since no new POWER10 instructions are used, the original POWER9 strcpy is
modified instead of creating a new implementation for POWER10.  This
implementation is based on both the original POWER9 implementation of
strcpy and the preamble of the new POWER10 implementation of strlen.

The changes also affect stpcpy, which uses the same implementation with
some additional code before returning.

On POWER9, averaging improvements across the benchmark
inputs (length/source alignement/destination alignment), for an
expermient that ran the benchmark five times, bench-strcpy showed an
improvement of 5.23%, and bench-stpcpy showed an improvement of 6.59%.

On POWER10, bench-strcpy showed 13.16%, and bench-stpcpy showed 13.59%.

The changes are:

1. Removed the null string optimization.

   Although this results in a few extra cycles for the null string, in
   combination with the second change, this resulted in improvements for
   for other cases.

2. Adapted the preamble from strlen for POWER10.

   This is the part of the function that handles up to the first 16 bytes
   of the string.

3. Increased number of unrolled iterations in the main loop to 6.
---
 sysdeps/powerpc/powerpc64/le/power9/strcpy.S | 160 +++++++++++--------
 1 file changed, 89 insertions(+), 71 deletions(-)
  

LGTM. There are just two minor typos in the commit message.

I ran the testsuite on POWER8, POWER9, and POWER10, each with and
without --disable-multiarch. No test failures caused by this change.

Reviewed-by: Matheus Castanho <msc@linux.ibm.com>

I can commit it for you with the typo fixes later today if there are no
objections.

Pedro Franco de Carvalho via Libc-alpha <libc-alpha@sourceware.org> writes:

> This patch modifies the current POWER9 implementation of strcpy and
> stpcpy to optimize it for POWER9/10.
>
> Since no new POWER10 instructions are used, the original POWER9 strcpy is
> modified instead of creating a new implementation for POWER10.  This
> implementation is based on both the original POWER9 implementation of
> strcpy and the preamble of the new POWER10 implementation of strlen.
>
> The changes also affect stpcpy, which uses the same implementation with
> some additional code before returning.
>
> On POWER9, averaging improvements across the benchmark
> inputs (length/source alignement/destination alignment), for an

alignement -> alignment

> expermient that ran the benchmark five times, bench-strcpy showed an

expermient -> experiment

> improvement of 5.23%, and bench-stpcpy showed an improvement of 6.59%.
>
> On POWER10, bench-strcpy showed 13.16%, and bench-stpcpy showed 13.59%.
>
> The changes are:
>
> 1. Removed the null string optimization.
>
>    Although this results in a few extra cycles for the null string, in
>    combination with the second change, this resulted in improvements for
>    for other cases.
>

Ok.

> 2. Adapted the preamble from strlen for POWER10.
>
>    This is the part of the function that handles up to the first 16 bytes
>    of the string.
>
> 3. Increased number of unrolled iterations in the main loop to 6.
> ---
>  sysdeps/powerpc/powerpc64/le/power9/strcpy.S | 160 +++++++++++--------
>  1 file changed, 89 insertions(+), 71 deletions(-)
>
> diff --git a/sysdeps/powerpc/powerpc64/le/power9/strcpy.S b/sysdeps/powerpc/powerpc64/le/power9/strcpy.S
> index 76cfcae200..b9e5afdcbe 100644
> --- a/sysdeps/powerpc/powerpc64/le/power9/strcpy.S
> +++ b/sysdeps/powerpc/powerpc64/le/power9/strcpy.S
> @@ -45,91 +45,78 @@
>     The implementation can load bytes past a null terminator, but only
>     up to the next 16B boundary, so it never crosses a page.  */
>
> +/* Load quadword at addr+offset to vreg, check for null bytes,
> +   and branch to label if any are found.  */
> +#define CHECK16(vreg,offset,addr,label) \
> +	lxv	vreg+32,offset(addr);	\
> +	vcmpequb. v6,vreg,v18;	\
> +	bne	cr6,L(label);
> +

Ok, result of comparison will always be saved to v6.

>  .machine power9
>  ENTRY_TOCLESS (FUNC_NAME, 4)
>  	CALL_MCOUNT 2
>
> -	/* NULL string optimisation  */
> -	lbz	r0,0(r4)
> -	stb	r0,0(r3)
> -	cmpwi	r0,0
> -	beqlr
> -
> -	addi	r4,r4,1
> -	addi	r11,r3,1
> -

Ok. Removed null string optimization.

>  	vspltisb v18,0		/* Zeroes in v18  */
> +	vspltisb v19,-1 	/* 0xFF bytes in v19  */
>
> -	neg	r5,r4
> -	rldicl	r9,r5,0,60	/* How many bytes to get source 16B aligned?  */
> +	/* Next 16B-aligned address. Prepare address for L(loop).  */
> +	addi	r5,r4,16
> +	clrrdi	r5,r5,4
> +	subf	r8,r4,r5
> +	add	r11,r3,r8
>

Ok. Align source to next 16 B boundary and move dest the same amount.

> -	/* Get source 16B aligned  */
> +	/* Align data and fill bytes not loaded with non matching char.  */
>  	lvx	v0,0,r4
>  	lvsr	v1,0,r4
> -	vperm	v0,v18,v0,v1
> -
> -	vcmpequb v6,v0,v18	/* 0xff if byte is NULL, 0x00 otherwise  */
> -	vctzlsbb r7,v6		/* Number of trailing zeroes  */
> -	addi	r8,r7,1		/* Add null terminator  */
> +	vperm	v0,v19,v0,v1
>
> -	/* r8 = bytes including null
> -	   r9 = bytes to get source 16B aligned
> -	   if r8 > r9
> -	      no null, copy r9 bytes
> -	   else
> -	      there is a null, copy r8 bytes and return.  */
> -	cmpd	r8,r9
> -	bgt	L(no_null)
> +	vcmpequb. v6,v0,v18	/* 0xff if byte is NULL, 0x00 otherwise  */
> +	beq	cr6,L(no_null)

Ok. Will branch out if all elements are != 0.
>
> -	sldi	r10,r8,56	/* stxvl wants size in top 8 bits  */
> -	stxvl	32+v0,r11,r10	/* Partial store  */
> +	/* There's a null byte.  */
> +	vctzlsbb r8,v6		/* Number of trailing zeroes  */
> +	addi	r9,r8,1 	/* Add null byte.  */
> +	sldi	r10,r9,56	/* stxvl wants size in top 8 bits.  */
> +	stxvl	32+v0,r3,r10	/* Partial store  */
>
>  #ifdef USE_AS_STPCPY
>  	/* stpcpy returns the dest address plus the size not counting the
>  	   final '\0'.  */
> -	add	r3,r11,r7
> +	add	r3,r3,r8
>  #endif
>  	blr

Ok.

>
>  L(no_null):
> -	sldi	r10,r9,56	/* stxvl wants size in top 8 bits  */
> -	stxvl	32+v0,r11,r10	/* Partial store  */
> -
> -	add	r4,r4,r9
> -	add	r11,r11,r9
> +	sldi	r10,r8,56	/* stxvl wants size in top 8 bits  */
> +	stxvl	32+v0,r3,r10	/* Partial store  */
>

Ok.

> +	.p2align 4
>  L(loop):
> -	lxv	32+v0,0(r4)
> -	vcmpequb. v6,v0,v18	/* Any zero bytes?  */
> -	bne	cr6,L(tail1)
> -
> -	lxv	32+v1,16(r4)
> -	vcmpequb. v6,v1,v18	/* Any zero bytes?  */
> -	bne	cr6,L(tail2)
> -
> -	lxv	32+v2,32(r4)
> -	vcmpequb. v6,v2,v18	/* Any zero bytes?  */
> -	bne	cr6,L(tail3)
> -
> -	lxv	32+v3,48(r4)
> -	vcmpequb. v6,v3,v18	/* Any zero bytes?  */
> -	bne	cr6,L(tail4)
> +	CHECK16(v0,0,r5,tail1)
> +	CHECK16(v1,16,r5,tail2)
> +	CHECK16(v2,32,r5,tail3)
> +	CHECK16(v3,48,r5,tail4)
> +	CHECK16(v4,64,r5,tail5)
> +	CHECK16(v5,80,r5,tail6)
>

Ok. Unroll 2 more iterations.

>  	stxv	32+v0,0(r11)
>  	stxv	32+v1,16(r11)
>  	stxv	32+v2,32(r11)
>  	stxv	32+v3,48(r11)
> +	stxv	32+v4,64(r11)
> +	stxv	32+v5,80(r11)
>
> -	addi	r4,r4,64
> -	addi	r11,r11,64
> +	addi	r5,r5,96
> +	addi	r11,r11,96

Ok. Addresses adjusted accordingly.

>
>  	b	L(loop)
>
> +	.p2align 4
>  L(tail1):
> -	vctzlsbb r8,v6
> -	addi	r9,r8,1
> +	vctzlsbb r8,v6		/* Number of trailing zeroes  */
> +	addi	r9,r8,1		/* Add null terminator  */
>  	sldi	r9,r9,56	/* stxvl wants size in top 8 bits  */
> -	stxvl	32+v0,r11,r9
> +	stxvl	32+v0,r11,r9	/* Partial store  */
>  #ifdef USE_AS_STPCPY
>  	/* stpcpy returns the dest address plus the size not counting the
>  	   final '\0'.  */
> @@ -137,50 +124,81 @@ L(tail1):
>  #endif
>  	blr
>

Ok. Besides new instruction alignment, nothing really changed, just
adding comments.

> +	.p2align 4
>  L(tail2):
>  	stxv	32+v0,0(r11)
> -	vctzlsbb r8,v6		/* Number of trailing zeroes  */
> -	addi	r9,r8,1		/* Add null terminator  */
> -	sldi	r10,r9,56	/* stxvl wants size in top 8 bits  */
> +	vctzlsbb r8,v6
> +	addi	r9,r8,1
> +	sldi	r9,r9,56
>  	addi	r11,r11,16
> -	stxvl	32+v1,r11,r10	/* Partial store  */
> +	stxvl	32+v1,r11,r9
>  #ifdef USE_AS_STPCPY
> -	/* stpcpy returns the dest address plus the size not counting the
> -	   final '\0'.  */
>  	add	r3,r11,r8
>  #endif
>  	blr
>
> +	.p2align 4
>  L(tail3):
>  	stxv	32+v0,0(r11)
>  	stxv	32+v1,16(r11)
> -	vctzlsbb r8,v6		/* Number of trailing zeroes  */
> -	addi	r9,r8,1		/* Add null terminator  */
> -	sldi	r10,r9,56	/* stxvl wants size in top 8 bits  */
> +	vctzlsbb r8,v6
> +	addi	r9,r8,1
> +	sldi	r9,r9,56
>  	addi	r11,r11,32
> -	stxvl	32+v2,r11,r10	/* Partial store  */
> +	stxvl	32+v2,r11,r9
>  #ifdef USE_AS_STPCPY
> -	/* stpcpy returns the dest address plus the size not counting the
> -	   final '\0'.  */
>  	add	r3,r11,r8
>  #endif
>  	blr
>
> +	.p2align 4
>  L(tail4):
>  	stxv	32+v0,0(r11)
>  	stxv	32+v1,16(r11)
>  	stxv	32+v2,32(r11)
> -	vctzlsbb r8,v6		/* Number of trailing zeroes  */
> -	addi	r9,r8,1		/* Add null terminator  */
> -	sldi	r10,r9,56	/* stxvl wants size in top 8 bits  */
> +	vctzlsbb r8,v6
> +	addi	r9,r8,1
> +	sldi	r9,r9,56
>  	addi	r11,r11,48
> -	stxvl	32+v3,r11,r10	/* Partial store  */
> +	stxvl	32+v3,r11,r9
>  #ifdef USE_AS_STPCPY
> -	/* stpcpy returns the dest address plus the size not counting the
> -	   final '\0'.  */
>  	add	r3,r11,r8
>  #endif
>  	blr

Ok. Changes to existing tails consist of adding instruction alignment
and using r9 instead of r10 to hold the length for stxvl. Code style is
more consistent between the tails now.

> +
> +	.p2align 4
> +L(tail5):
> +	stxv	32+v0,0(r11)
> +	stxv	32+v1,16(r11)
> +	stxv	32+v2,32(r11)
> +	stxv	32+v3,48(r11)
> +	vctzlsbb r8,v6
> +	addi	r9,r8,1
> +	sldi	r9,r9,56
> +	addi	r11,r11,64
> +	stxvl	32+v4,r11,r9
> +#ifdef USE_AS_STPCPY
> +	add	r3,r11,r8
> +#endif
> +	blr
> +
> +	.p2align 4
> +L(tail6):
> +	stxv	32+v0,0(r11)
> +	stxv	32+v1,16(r11)
> +	stxv	32+v2,32(r11)
> +	stxv	32+v3,48(r11)
> +	stxv	32+v4,64(r11)
> +	vctzlsbb r8,v6
> +	addi	r9,r8,1
> +	sldi	r9,r9,56
> +	addi	r11,r11,80
> +	stxvl	32+v5,r11,r9
> +#ifdef USE_AS_STPCPY
> +	add	r3,r11,r8
> +#endif
> +	blr
> +

Ok. Add code for 2 more tails.

>  END (FUNC_NAME)
>  #ifndef USE_AS_STPCPY
>  libc_hidden_builtin_def (strcpy)


--
Matheus Castanho
  

Matheus Castanho <msc@linux.ibm.com> writes:

> I can commit it for you with the typo fixes later today if there are no
> objections.

Yes, please, feel free to correct my typos.

Thanks a lot for the review.

--
Pedro Franco de Carvalho
  

Pedro Franco de Carvalho <pedromfc@linux.ibm.com> writes:

> Matheus Castanho <msc@linux.ibm.com> writes:
>
>> I can commit it for you with the typo fixes later today if there are no
>> objections.
>
> Yes, please, feel free to correct my typos.
>
> Thanks a lot for the review.

Pushed as 813c6ec808556553be9d39e900a3fc97ceb32330

Thanks,
Matheus Castanho