From 0f67de14e605f519d74ed6fff67c6712158459d3 Mon Sep 17 00:00:00 2001 From: Johannes Carlsson Date: Thu, 3 Feb 2011 15:16:15 +0100 Subject: [PATCH] Use more optimized version of memmove On ARM there is currently no assembler optimized memmove in libc. There is however a more optimized bcopy which copies long instead of bytes where possible. This almost doubles the performance in best case. Change-Id: I1f1cd27529443358047c385730deaf938ce4e642 --- libc/string/memmove.c | 37 +++---------------------------------- 1 file changed, 3 insertions(+), 34 deletions(-) diff --git a/libc/string/memmove.c b/libc/string/memmove.c index 98ecfc90b..072104b6c 100644 --- a/libc/string/memmove.c +++ b/libc/string/memmove.c @@ -26,6 +26,7 @@ * SUCH DAMAGE. */ #include +#include void *memmove(void *dst, const void *src, size_t n) { @@ -37,39 +38,7 @@ void *memmove(void *dst, const void *src, size_t n) if (__builtin_expect((q < p) || ((size_t)(q - p) >= n), 1)) { return memcpy(dst, src, n); } else { -#define PRELOAD_DISTANCE 64 - /* a semi-optimized memmove(). we're preloading the src and dst buffers - * as we go */ - size_t c0, c1, i; - p += n; - q += n; - /* note: we preload the destination as well, because the 1-byte at a time - * copy below doesn't take advantage of the write-buffer, we need - * to use the cache instead as a poor man's write-combiner */ - __builtin_prefetch(p-1); - __builtin_prefetch(q-1); - if (PRELOAD_DISTANCE > 32) { - __builtin_prefetch(p-(32+1)); - __builtin_prefetch(q-(32+1)); - } - /* do the prefetech as soon as possible, prevent the compiler to - * reorder the instructions above the prefetch */ - asm volatile("":::"memory"); - c0 = n & 0x1F; /* cache-line is 32 bytes */ - c1 = n >> 5; - while ( c1-- ) { - /* ARMv6 can have up to 3 memory access outstanding */ - __builtin_prefetch(p - (PRELOAD_DISTANCE+1)); - __builtin_prefetch(q - (PRELOAD_DISTANCE+1)); - asm volatile("":::"memory"); - for (i=0 ; i<32 ; i++) { - *--q = *--p; - } - } - while ( c0-- ) { - *--q = *--p; - } + bcopy(src, dst, n); + return dst; } - - return dst; }