memcpy.c

00001 /* Copyright (C) 2013 - Adam Green (https://github.com/adamgreen)
00002 
00003    Licensed under the Apache License, Version 2.0 (the "License");
00004    you may not use this file except in compliance with the License.
00005    You may obtain a copy of the License at
00006 
00007        http://www.apache.org/licenses/LICENSE-2.0
00008 
00009    Unless required by applicable law or agreed to in writing, software
00010    distributed under the License is distributed on an "AS IS" BASIS,
00011    WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
00012    See the License for the specific language governing permissions and
00013    limitations under the License.
00014 */
00015 #if defined(TOOLCHAIN_GCC) && defined(__thumb2__)
00016 
00017 #include <stdio.h>
00018 
00019 
00020 /* This is a hand written Thumb-2 assembly language version of the
00021    standard C memcpy() function that can be used by the lwIP networking
00022    stack to improve its performance.  It copies 4 bytes at a time and
00023    unrolls the loop to perform 4 of these copies per loop iteration.
00024 */
00025 __attribute__((naked)) void thumb2_memcpy(void* pDest, const void* pSource, size_t length)
00026 {
00027     __asm (
00028         ".syntax unified\n"
00029         ".thumb\n"
00030 
00031         // Copy 16 bytes at a time first.
00032         "    lsrs    r3, r2, #4\n"
00033         "    beq.n   2$\n"
00034         "1$: ldr     r12, [r1], #4\n"
00035         "    str     r12, [r0], #4\n"
00036         "    ldr     r12, [r1], #4\n"
00037         "    str     r12, [r0], #4\n"
00038         "    ldr     r12, [r1], #4\n"
00039         "    str     r12, [r0], #4\n"
00040         "    ldr     r12, [r1], #4\n"
00041         "    str     r12, [r0], #4\n"
00042         "    subs    r3, #1\n"
00043         "    bne     1$\n"
00044 
00045         // Copy byte by byte for what is left.
00046         "2$:\n"
00047         "    ands    r3, r2, #0xf\n"
00048         "    beq.n   4$\n"
00049         "3$: ldrb    r12, [r1], #1\n"
00050         "    strb    r12, [r0], #1\n"
00051         "    subs    r3, #1\n"
00052         "    bne     3$\n"
00053 
00054         // Return to caller.
00055         "4$: bx      lr\n"
00056     );
00057 }
00058 
00059 #endif