From b628072a9d9d88e7adc35ec20fe87552f0ca4a7c Mon Sep 17 00:00:00 2001 From: matt Date: Tue, 20 Aug 2013 08:05:49 +0000 Subject: [PATCH] swap r1 & ip use adds, eors, etc. teq -> cmp --- common/lib/libc/arch/arm/string/strrchr_arm.S | 72 ++++++++++--------- 1 file changed, 39 insertions(+), 33 deletions(-) diff --git a/common/lib/libc/arch/arm/string/strrchr_arm.S b/common/lib/libc/arch/arm/string/strrchr_arm.S index ac2fb769ef67..c3e3210d38a4 100644 --- a/common/lib/libc/arch/arm/string/strrchr_arm.S +++ b/common/lib/libc/arch/arm/string/strrchr_arm.S @@ -29,7 +29,7 @@ #include -RCSID("$NetBSD: strrchr_arm.S,v 1.4 2013/08/11 04:56:32 matt Exp $") +RCSID("$NetBSD: strrchr_arm.S,v 1.5 2013/08/20 08:05:49 matt Exp $") #ifdef __ARMEL__ #define BYTE0 0x000000ff @@ -37,49 +37,52 @@ RCSID("$NetBSD: strrchr_arm.S,v 1.4 2013/08/11 04:56:32 matt Exp $") #define BYTE2 0x00ff0000 #define BYTE3 0xff000000 #define lshi lsl +#define lshis lsls #else #define BYTE0 0xff000000 #define BYTE1 0x00ff0000 #define BYTE2 0x0000ff00 #define BYTE3 0x000000ff #define lshi lsr +#define lshis lsrs #endif - .text ENTRY(strrchr) - teq r1, #0 /* searching for NUL? */ + ands r2, r1, #0xff /* is the byte value NUL? */ bne 1f /* no, do it the hard way */ push {r0, lr} /* save pointer and return addr */ bl PLT_SYM(strlen) /* get length */ - pop {r1, lr} /* restore pointer and returna addr */ - add r0, r0, r1 /* add pointer to length */ - RET /* return */ + pop {r1, r2} /* restore pointer / return addr */ + adds r0, r0, r1 /* add pointer to length */ + RETr(r2) /* return */ -1: mov ip, r0 /* we use r0 at the return value */ - mov r0, #0 /* return NULL by default */ - and r2, r1, #0xff /* restrict to byte value */ -2: tst ip, #3 /* test for word alignment */ +1: mov r1, r0 /* we use r0 at the return value */ + movs r0, #0 /* return NULL by default */ +2: tst r1, #3 /* test for word alignment */ beq .Lpre_main_loop /* finally word aligned */ - ldrb r3, [ip], #1 /* load a byte */ + ldrb r3, [r1], #1 /* load a byte */ cmp r3, r2 /* did it match? */ - subeq r0, ip, #1 /* yes, remember that it did */ - teq r3, #0 /* was it NUL? */ +#ifdef __thumb__ + it eq + subeq r0, r1, #1 /* yes, remember that it did */ +#endif + cmp r3, #0 /* was it NUL? */ bne 2b /* no, try next byte */ RET /* return */ .Lpre_main_loop: push {r4, r5} /* save some registers */ #if defined(_ARM_ARCH_7) - movw r1, #0xfefe /* magic constant; 254 in each byte */ - movt r1, #0xfefe /* magic constant; 254 in each byte */ + movw ip, #0xfefe /* magic constant; 254 in each byte */ + movt ip, #0xfefe /* magic constant; 254 in each byte */ #elif defined(_ARM_ARCH_6) - mov r1, #0xfe /* put 254 in low byte */ - orr r1, r1, r1, lsl #8 /* move to next byte */ - orr r1, r1, r1, lsl #16 /* move to next halfword */ + mov ip, #0xfe /* put 254 in low byte */ + orr ip, ip, ip, lsl #8 /* move to next byte */ + orr ip, ip, ip, lsl #16 /* move to next halfword */ #endif /* _ARM_ARCH_6 */ orr r2, r2, r2, lsl #8 /* move to next byte */ orr r2, r2, r2, lsl #16 /* move to next halfword */ .Lmain_loop: - ldr r3, [ip], #4 /* load next word */ + ldr r3, [r1], #4 /* load next word */ #if defined(_ARM_ARCH_6) /* * Add 254 to each byte using the UQADD8 (unsigned saturating add 8) @@ -87,10 +90,10 @@ ENTRY(strrchr) * become 255. For NUL, it will be 254. When we complement the * result, if the result is non-0 then we must have encountered a NUL. */ - uqadd8 r4, r3, r1 /* NUL detection happens here */ + uqadd8 r4, r3, ip /* NUL detection happens here */ usub8 r3, r3, r2 /* bias for char looked for? */ - uqadd8 r5, r3, r1 /* char detection happens here */ - and r3, r4, r5 /* merge results */ + uqadd8 r5, r3, ip /* char detection happens here */ + ands r3, r4, r5 /* merge results */ mvns r3, r3 /* is the complement non-0? */ beq .Lmain_loop /* no, then keep going */ @@ -104,7 +107,10 @@ ENTRY(strrchr) */ bics r5, r5, r4 /* clear any NUL match bits */ beq .Ldone /* no remaining matches, we're done */ - movs r3, r4, lshi #8 /* shift up a byte */ + lshis r3, r4, #8 /* shift up a byte */ +#ifdef __thumb__ + itt ne +#endif orrsne r3, r3, r3, lshi #8 /* if non 0, copy up to next byte */ orrsne r3, r3, r3, lshi #8 /* if non 0, copy up to last byte */ bics r5, r5, r3 /* clear match bits */ @@ -117,13 +123,13 @@ ENTRY(strrchr) * If we have multiple matches, we want to the select the "last" match * in the word which will be the lowest bit set. */ - sub r3, r5, #1 /* subtract 1 */ - and r3, r3, r5 /* and with mask */ - eor r5, r5, r3 /* only have the lowest bit set left */ + subs r3, r5, #1 /* subtract 1 */ + ands r3, r3, r5 /* and with mask */ + eors r5, r5, r3 /* only have the lowest bit set left */ clz r5, r5 /* count how many leading zeros */ - add r0, ip, r5, lsr #3 /* divide that by 8 and add to count */ - sub r0, r0, #4 /* compensate for the post-inc */ - teq r4, #0 /* did we read any NULs? */ + add r0, r1, r5, lsr #3 /* divide that by 8 and add to count */ + subs r0, r0, #4 /* compensate for the post-inc */ + cmp r4, #0 /* did we read any NULs? */ beq .Lmain_loop /* no, get next word */ #else /* @@ -134,19 +140,19 @@ ENTRY(strrchr) tst r3, #BYTE0 /* is byte 0 a NUL? */ beq .Ldone /* yes, then we're done */ tst r4, #BYTE0 /* is byte 0 a match? */ - subeq r0, ip, #4 /* yes, remember its location */ + subeq r0, r1, #4 /* yes, remember its location */ tst r3, #BYTE1 /* is byte 1 a NUL? */ beq .Ldone /* yes, then we're done */ tst r4, #BYTE1 /* is byte 1 a match? */ - subeq r0, ip, #3 /* yes, remember its location */ + subeq r0, r1, #3 /* yes, remember its location */ tst r3, #BYTE2 /* is byte 2 a NUL? */ beq .Ldone /* yes, then we're done */ tst r4, #BYTE2 /* is byte 2 a match? */ - subeq r0, ip, #2 /* yes, remember its location */ + subeq r0, r1, #2 /* yes, remember its location */ tst r3, #BYTE3 /* is byte 3 a NUL? */ beq .Ldone /* yes, then we're done */ tst r4, #BYTE3 /* is byte 3 a match? */ - subeq r0, ip, #1 /* yes, remember its location */ + subeq r0, r1, #1 /* yes, remember its location */ b .Lmain_loop #endif /* _ARM_ARCH_6 */ .Ldone: