On Fri, Jun 08, 2018 at 10:20:41AM +0000, Christophe Leroy wrote:
> The generic implementation of strlen() reads strings byte per byte.
>
> This patch implements strlen() in assembly based on a read of entire
> words, in the same spirit as what some other arches and glibc do.
>
> On a 8xx the time spent in strlen is reduced by 2/3 for long strings.
>
> strlen() selftest on an 8xx provides the following values:
>
> Before the patch (ie with the generic strlen() in lib/string.c):
>
> len 256 : time = 0.803648
> len 16 : time = 0.062989
> len 4 : time = 0.026269
>
> After the patch:
>
> len 256 : time = 0.267791 ==> 66% improvment
> len 16 : time = 0.037902 ==> 41% improvment
> len 4 : time = 0.026124 ==> no degradation
>
> Signed-off-by: Christophe Leroy <christophe.le...@c-s.fr>
> ---
> Not tested on PPC64.
>
> Changes in v4:
> - Added alignment of the loop
> - doing the andc only if still not 0 as it happends only for bytes above
> 0x7f which is pretty rare in a string
>
> Changes in v3:
> - Made it common to PPC32 and PPC64
>
> Changes in v2:
> - Moved handling of unaligned strings outside of the main path as it is very
> unlikely.
> - Removed the verification of the fourth byte in case none of the three
> first ones are NUL.
>
>
> arch/powerpc/include/asm/asm-compat.h | 4 +++
> arch/powerpc/include/asm/string.h | 1 +
> arch/powerpc/lib/string.S | 57
> +++++++++++++++++++++++++++++++++++
> 3 files changed, 62 insertions(+)
>
> diff --git a/arch/powerpc/include/asm/asm-compat.h
> b/arch/powerpc/include/asm/asm-compat.h
> index 7f2a7702596c..0e99fe7570c0 100644
> --- a/arch/powerpc/include/asm/asm-compat.h
> +++ b/arch/powerpc/include/asm/asm-compat.h
> @@ -20,8 +20,10 @@
>
> /* operations for longs and pointers */
> #define PPC_LL stringify_in_c(ld)
> +#define PPC_LLU stringify_in_c(ldu)
> #define PPC_STL stringify_in_c(std)
> #define PPC_STLU stringify_in_c(stdu)
> +#define PPC_ROTLI stringify_in_c(rotldi)
> #define PPC_LCMPI stringify_in_c(cmpdi)
> #define PPC_LCMPLI stringify_in_c(cmpldi)
> #define PPC_LCMP stringify_in_c(cmpd)
> @@ -53,8 +55,10 @@
>
> /* operations for longs and pointers */
> #define PPC_LL stringify_in_c(lwz)
> +#define PPC_LLU stringify_in_c(lwzu)
> #define PPC_STL stringify_in_c(stw)
> #define PPC_STLU stringify_in_c(stwu)
> +#define PPC_ROTLI stringify_in_c(rotlwi)
> #define PPC_LCMPI stringify_in_c(cmpwi)
> #define PPC_LCMPLI stringify_in_c(cmplwi)
> #define PPC_LCMP stringify_in_c(cmpw)
> diff --git a/arch/powerpc/include/asm/string.h
> b/arch/powerpc/include/asm/string.h
> index 9b8cedf618f4..8fdcb532de72 100644
> --- a/arch/powerpc/include/asm/string.h
> +++ b/arch/powerpc/include/asm/string.h
> @@ -13,6 +13,7 @@
> #define __HAVE_ARCH_MEMCHR
> #define __HAVE_ARCH_MEMSET16
> #define __HAVE_ARCH_MEMCPY_FLUSHCACHE
> +#define __HAVE_ARCH_STRLEN
>
> extern char * strcpy(char *,const char *);
> extern char * strncpy(char *,const char *, __kernel_size_t);
> diff --git a/arch/powerpc/lib/string.S b/arch/powerpc/lib/string.S
> index 4b41970e9ed8..238f61e2024f 100644
> --- a/arch/powerpc/lib/string.S
> +++ b/arch/powerpc/lib/string.S
> @@ -67,3 +67,60 @@ _GLOBAL(memchr)
> 2: li r3,0
> blr
> EXPORT_SYMBOL(memchr)
> +
> +_GLOBAL(strlen)
> + andi. r9, r3, (SZL - 1)
> + addi r10, r3, -SZL
> + bne- 1f
> +2: lis r6, 0x8080
> + ori r6, r6, 0x8080 /* r6 = 0x80808080 (himagic) */
> +#ifdef CONFIG_PPC64
> + rldimi r6, r6, 32, 0 /* r6 = 0x8080808080808080 (himagic) */
> +#endif
> + PPC_ROTLI r7, r6, 1 /* r7 = 0x01010101(01010101) (lomagic)*/
> + .balign IFETCH_ALIGN_BYTES
> +3: PPC_LLU r9, SZL(r10)
> + /* ((x - lomagic) & ~x & himagic) == 0 means no byte in x is NUL */
> + subf r8, r7, r9
> + and. r8, r8, r6
> + beq+ 3b
> + andc. r8, r8, r9
> + beq+ 3b
> +#ifdef CONFIG_PPC64
> + rldicl. r8, r9, 8, 56
> + beq 20f
> + rldicl. r8, r9, 16, 56
> + beq 21f
> + rldicl. r8, r9, 24, 56
> + beq 22f
> + rldicl. r8, r9, 32, 56
> + beq 23f
> + addi r10, r10, 4
> +#endif
> + rlwinm. r8, r9, 0, 0xff000000
> + beq 20f
> + rlwinm. r8, r9, 0, 0x00ff0000
> + beq 21f
> + rlwinm. r8, r9, 0, 0x0000ff00
> + beq 22f
> +23: subf r3, r3, r10
Actually these rlwinm. can likely be replaced by a single
cntlzw /cntlzd; for 32 bit something like:
cntlzw r8,r9
subf r3,r3,r10
srwi r8,r8,3
add r3,r3,r8
blr
and similar for 64 bit but with cntlzd.
Gabriel
> + addi r3, r3, 3
> + blr
> +22: subf r3, r3, r10
> + addi r3, r3, 2
> + blr
> +21: subf r3, r3, r10
> + addi r3, r3, 1
> + blr
> +19: addi r10, r10, (SZL - 1)
> +20: subf r3, r3, r10
> + blr
> +
> +1: lbz r9, SZL(r10)
> + addi r10, r10, 1
> + cmpwi cr1, r9, 0
> + andi. r9, r10, (SZL - 1)
> + beq cr1, 19b
> + bne 1b
> + b 2b
> +EXPORT_SYMBOL(strlen)
> --
> 2.13.3
>
