[PATCH] S390: Optimize __memset_z196.

[Stefan Liebler]

Just as information, if nobody opposes, I'll commit this patch tomorrow.

On 6/19/20 3:51 PM, Stefan Liebler wrote:
> It turned out that an 256b-mvc instruction which depends on the
> result of a previous 256b-mvc instruction is counterproductive.
> Therefore this patch adjusts the 256b-loop by storing the
> first byte with stc and setting the remaining 255b with mvc.
> Now the 255b-mvc instruction depends on the stc instruction.
> ---
>  sysdeps/s390/memset-z900.S | 19 +++++++++----------
>  1 file changed, 9 insertions(+), 10 deletions(-)
> 
> diff --git a/sysdeps/s390/memset-z900.S b/sysdeps/s390/memset-z900.S
> index ca3eac0522..1e0c334156 100644
> --- a/sysdeps/s390/memset-z900.S
> +++ b/sysdeps/s390/memset-z900.S
> @@ -157,28 +157,27 @@ ENTRY(MEMSET_Z196)
>  # if !defined __s390x__
>  	llgfr	%r4,%r4
>  # endif /* !defined __s390x__  */
> -	ltgr    %r4,%r4
> -	je      .L_Z196_4
> +	clgfi	%r4,1
> +	jl	.L_Z196_4	    # n == 0
>  	stc     %r3,0(%r2)
> +	je      .L_Z196_4	    # n == 1
> +	aghi	%r4,-2
>  	lgr     %r1,%r2
> -	cghi    %r4,1
> -	je      .L_Z196_4
> -	aghi    %r4,-2
> -	srlg    %r5,%r4,8
> -	ltgr    %r5,%r5
> -	jne     .L_Z196_1
> +	risbg	%r5,%r4,8,128+63,56 # r5 = n / 256
> +	jne     .L_Z196_1	    # Jump away if r5 != 0
>  .L_Z196_3:
>  	exrl    %r4,.L_Z196_17
>  .L_Z196_4:
>  	br      %r14
>  .L_Z196_1:
>  	cgfi	%r5,1048576
> -	jh	__memset_mvcle	   # Switch to mvcle for >256MB
> +	jh	__memset_mvcle	    # Switch to mvcle for >256MB
>  .L_Z196_2:
>  	pfd     2,1024(%r1)
> -	mvc     1(256,%r1),0(%r1)
> +	mvc     1(255,%r1),0(%r1)
>  	aghi    %r5,-1
>  	la      %r1,256(%r1)
> +	stc     %r3,0(%r1)
>  	jne     .L_Z196_2
>  	j       .L_Z196_3
>  .L_Z196_17:
>