diff options
Diffstat (limited to 'newlib/libc/machine/xscale/memset.c')
-rw-r--r-- | newlib/libc/machine/xscale/memset.c | 119 |
1 files changed, 55 insertions, 64 deletions
diff --git a/newlib/libc/machine/xscale/memset.c b/newlib/libc/machine/xscale/memset.c index 4ff8d01e0..bfd0f0d47 100644 --- a/newlib/libc/machine/xscale/memset.c +++ b/newlib/libc/machine/xscale/memset.c @@ -11,78 +11,69 @@ void * memset (void *dst, int c, size_t len) { int dummy; - asm volatile ("tst %0, #0x3" #ifndef __OPTIMIZE_SIZE__ -"\n\ - beq 1f\n\ - b 2f\n\ -0:\n\ - strb %1, [%0], #1\n\ - tst %0, #0x3\n\ - beq 1f\n\ -2:\n\ - movs r3, %2\n\ - sub %2, %2, #1\n\ - bne 0b\n\ -# At this point we know that %2 == len == -1 (since the SUB has already taken\n\ -# place). If we fall through to the 1: label (as the code used to do), the\n\ -# CMP will detect this negative value and branch to the 2: label. This will\n\ -# test %2 again, but this time against 0. The test will fail and the loop\n\ -# at 2: will go on for (almost) ever. Hence the explicit branch to the end\n\ -# of the hand written assembly code.\n\ - b 4f\n\ -1:\n\ - cmp %2, #0x3\n\ - bls 2f\n\ - and %1, %1, #0xff\n\ - orr lr, %1, %1, asl #8\n\ - cmp %2, #0xf\n\ - orr lr, lr, lr, asl #16\n\ - bls 1f\n\ - mov r3, lr\n\ - mov r4, lr\n\ - mov r5, lr\n\ -0:\n\ - sub %2, %2, #16\n\ - stmia %0!, { r3, r4, r5, lr }\n\ - cmp %2, #0xf\n\ - bhi 0b\n\ -1:\n\ - cmp %2, #0x7\n\ - bls 1f\n\ - mov r3, lr\n\ -0:\n\ - sub %2, %2, #8\n\ - stmia %0!, { r3, lr }\n\ - cmp %2, #0x7\n\ - bhi 0b\n\ -1:\n\ - cmp %2, #0x3\n\ - bls 2f\n\ -0:\n\ - sub %2, %2, #4\n\ - str lr, [%0], #4\n\ - cmp %2, #0x3\n\ - bhi 0b\n\ +" + beq 1f + b 2f +0: + strb %1, [%0], #1 + tst %0, #0x3 + beq 1f +2: + movs r3, %2 + sub %2, %2, #1 + bne 0b +1: + cmp %2, #0x3 + bls 2f + and %1, %1, #0xff + orr lr, %1, %1, asl #8 + cmp %2, #0xf + orr lr, lr, lr, asl #16 + bls 1f + mov r3, lr + mov r4, lr + mov r5, lr +0: + sub %2, %2, #16 + stmia %0!, { r3, r4, r5, lr } + cmp %2, #0xf + bhi 0b +1: + cmp %2, #0x7 + bls 1f + mov r3, lr +0: + sub %2, %2, #8 + stmia %0!, { r3, lr } + cmp %2, #0x7 + bhi 0b +1: + cmp %2, #0x3 + bls 2f +0: + sub %2, %2, #4 + str lr, [%0], #4 + cmp %2, #0x3 + bhi 0b " #endif /* !__OPTIMIZE_SIZE__ */ -"\n\ -2:\n\ - movs r3, %2\n\ - sub %2, %2, #1\n\ - beq 4f\n\ -0:\n\ - movs r3, %2\n\ - sub %2, %2, #1\n\ - strb %1, [%0], #1\n\ - bne 0b\n\ -4:" +" +2: + movs r3, %2 + sub %2, %2, #1 + beq 1f +0: + movs r3, %2 + sub %2, %2, #1 + strb %1, [%0], #1 + bne 0b +1:" : "=&r" (dummy), "=&r" (c), "=&r" (len) : "0" (dst), "1" (c), "2" (len) : "memory", "r3", "r4", "r5", "lr"); - return dst; } |