# commit 8a7413f9b036da83ffde491a37d9d2340bc321a7 # Author: Alan Modra # Date: Sat Aug 17 18:41:17 2013 +0930 # # PowerPC LE strcmp and strncmp # http://sourceware.org/ml/libc-alpha/2013-08/msg00099.html # # More little-endian support. I leave the main strcmp loops unchanged, # (well, except for renumbering rTMP to something other than r0 since # it's needed in an addi insn) and modify the tail for little-endian. # # I noticed some of the big-endian tail code was a little untidy so have # cleaned that up too. # # * sysdeps/powerpc/powerpc64/strcmp.S (rTMP2): Define as r0. # (rTMP): Define as r11. # (strcmp): Add little-endian support. Optimise tail. # * sysdeps/powerpc/powerpc32/strcmp.S: Similarly. # * sysdeps/powerpc/powerpc64/strncmp.S: Likewise. # * sysdeps/powerpc/powerpc32/strncmp.S: Likewise. # * sysdeps/powerpc/powerpc64/power4/strncmp.S: Likewise. # * sysdeps/powerpc/powerpc32/power4/strncmp.S: Likewise. # * sysdeps/powerpc/powerpc64/power7/strncmp.S: Likewise. # * sysdeps/powerpc/powerpc32/power7/strncmp.S: Likewise. # --- # sysdeps/powerpc/powerpc32/power4/strncmp.S | 56 ++++++++++++++++++------ # sysdeps/powerpc/powerpc32/power7/strncmp.S | 55 +++++++++++++++++------- # sysdeps/powerpc/powerpc32/strcmp.S | 42 +++++++++++++++++- # sysdeps/powerpc/powerpc32/strncmp.S | 56 ++++++++++++++++++------ # sysdeps/powerpc/powerpc64/power4/strncmp.S | 63 ++++++++++++++++++++++++---- # sysdeps/powerpc/powerpc64/power7/strncmp.S | 61 +++++++++++++++++++++++---- # sysdeps/powerpc/powerpc64/strcmp.S | 65 ++++++++++++++++++++++++----- # sysdeps/powerpc/powerpc64/strncmp.S | 65 ++++++++++++++++++++++++----- # 8 files changed, 381 insertions(+), 82 deletions(-) # --- a/sysdeps/powerpc/powerpc32/power4/strncmp.S +++ b/sysdeps/powerpc/powerpc32/power4/strncmp.S @@ -26,7 +26,7 @@ EALIGN (BP_SYM(strncmp), 4, 0) -#define rTMP r0 +#define rTMP2 r0 #define rRTN r3 #define rSTR1 r3 /* first string arg */ #define rSTR2 r4 /* second string arg */ @@ -42,6 +42,7 @@ #define r7F7F r9 /* constant 0x7f7f7f7f */ #define rNEG r10 /* ~(word in s1 | 0x7f7f7f7f) */ #define rBITDIF r11 /* bits that differ in s1 & s2 words */ +#define rTMP r12 dcbt 0,rSTR1 or rTMP, rSTR2, rSTR1 @@ -80,12 +81,45 @@ we don't compare two strings as different because of gunk beyond the end of the strings... */ +#ifdef __LITTLE_ENDIAN__ +L(endstring): + slwi rTMP, rTMP, 1 + addi rTMP2, rTMP, -1 + andc rTMP2, rTMP2, rTMP + and rWORD2, rWORD2, rTMP2 /* Mask off gunk. */ + and rWORD1, rWORD1, rTMP2 + rlwinm rTMP2, rWORD2, 8, 0xffffffff /* Byte reverse word. */ + rlwinm rTMP, rWORD1, 8, 0xffffffff + rldimi rTMP2, rWORD2, 24, 32 + rldimi rTMP, rWORD1, 24, 32 + rlwimi rTMP2, rWORD2, 24, 16, 23 + rlwimi rTMP, rWORD1, 24, 16, 23 + xor. rBITDIF, rTMP, rTMP2 + sub rRTN, rTMP, rTMP2 + bgelr+ + ori rRTN, rTMP2, 1 + blr + +L(different): + lwz rWORD1, -4(rSTR1) + rlwinm rTMP2, rWORD2, 8, 0xffffffff /* Byte reverse word. */ + rlwinm rTMP, rWORD1, 8, 0xffffffff + rldimi rTMP2, rWORD2, 24, 32 + rldimi rTMP, rWORD1, 24, 32 + rlwimi rTMP2, rWORD2, 24, 16, 23 + rlwimi rTMP, rWORD1, 24, 16, 23 + xor. rBITDIF, rTMP, rTMP2 + sub rRTN, rTMP, rTMP2 + bgelr+ + ori rRTN, rTMP2, 1 + blr + +#else L(endstring): and rTMP, r7F7F, rWORD1 beq cr1, L(equal) add rTMP, rTMP, r7F7F xor. rBITDIF, rWORD1, rWORD2 - andc rNEG, rNEG, rTMP blt- L(highbit) cntlzw rBITDIF, rBITDIF @@ -93,28 +127,20 @@ addi rNEG, rNEG, 7 cmpw cr1, rNEG, rBITDIF sub rRTN, rWORD1, rWORD2 - blt- cr1, L(equal) - srawi rRTN, rRTN, 31 - ori rRTN, rRTN, 1 - blr + bgelr+ cr1 L(equal): li rRTN, 0 blr L(different): - lwzu rWORD1, -4(rSTR1) + lwz rWORD1, -4(rSTR1) xor. rBITDIF, rWORD1, rWORD2 sub rRTN, rWORD1, rWORD2 - blt- L(highbit) - srawi rRTN, rRTN, 31 - ori rRTN, rRTN, 1 - blr + bgelr+ L(highbit): - srwi rWORD2, rWORD2, 24 - srwi rWORD1, rWORD1, 24 - sub rRTN, rWORD1, rWORD2 + ori rRTN, rWORD2, 1 blr - +#endif /* Oh well. In this case, we just do a byte-by-byte comparison. */ .align 4 --- a/sysdeps/powerpc/powerpc32/power7/strncmp.S +++ b/sysdeps/powerpc/powerpc32/power7/strncmp.S @@ -28,7 +28,7 @@ EALIGN (BP_SYM(strncmp),5,0) -#define rTMP r0 +#define rTMP2 r0 #define rRTN r3 #define rSTR1 r3 /* first string arg */ #define rSTR2 r4 /* second string arg */ @@ -44,6 +44,7 @@ #define r7F7F r9 /* constant 0x7f7f7f7f */ #define rNEG r10 /* ~(word in s1 | 0x7f7f7f7f) */ #define rBITDIF r11 /* bits that differ in s1 & s2 words */ +#define rTMP r12 dcbt 0,rSTR1 nop @@ -83,13 +84,45 @@ /* OK. We've hit the end of the string. We need to be careful that we don't compare two strings as different because of gunk beyond the end of the strings... */ +#ifdef __LITTLE_ENDIAN__ +L(endstring): + slwi rTMP, rTMP, 1 + addi rTMP2, rTMP, -1 + andc rTMP2, rTMP2, rTMP + and rWORD2, rWORD2, rTMP2 /* Mask off gunk. */ + and rWORD1, rWORD1, rTMP2 + rlwinm rTMP2, rWORD2, 8, 0xffffffff /* Byte reverse word. */ + rlwinm rTMP, rWORD1, 8, 0xffffffff + rldimi rTMP2, rWORD2, 24, 32 + rldimi rTMP, rWORD1, 24, 32 + rlwimi rTMP2, rWORD2, 24, 16, 23 + rlwimi rTMP, rWORD1, 24, 16, 23 + xor. rBITDIF, rTMP, rTMP2 + sub rRTN, rTMP, rTMP2 + bgelr + ori rRTN, rTMP2, 1 + blr + +L(different): + lwz rWORD1, -4(rSTR1) + rlwinm rTMP2, rWORD2, 8, 0xffffffff /* Byte reverse word. */ + rlwinm rTMP, rWORD1, 8, 0xffffffff + rldimi rTMP2, rWORD2, 24, 32 + rldimi rTMP, rWORD1, 24, 32 + rlwimi rTMP2, rWORD2, 24, 16, 23 + rlwimi rTMP, rWORD1, 24, 16, 23 + xor. rBITDIF, rTMP, rTMP2 + sub rRTN, rTMP, rTMP2 + bgelr + ori rRTN, rTMP2, 1 + blr +#else L(endstring): and rTMP,r7F7F,rWORD1 beq cr1,L(equal) add rTMP,rTMP,r7F7F xor. rBITDIF,rWORD1,rWORD2 - andc rNEG,rNEG,rTMP blt L(highbit) cntlzw rBITDIF,rBITDIF @@ -97,28 +130,20 @@ addi rNEG,rNEG,7 cmpw cr1,rNEG,rBITDIF sub rRTN,rWORD1,rWORD2 - blt cr1,L(equal) - srawi rRTN,rRTN,31 - ori rRTN,rRTN,1 - blr + bgelr cr1 L(equal): li rRTN,0 blr L(different): - lwzu rWORD1,-4(rSTR1) + lwz rWORD1,-4(rSTR1) xor. rBITDIF,rWORD1,rWORD2 sub rRTN,rWORD1,rWORD2 - blt L(highbit) - srawi rRTN,rRTN,31 - ori rRTN,rRTN,1 - blr + bgelr L(highbit): - srwi rWORD2,rWORD2,24 - srwi rWORD1,rWORD1,24 - sub rRTN,rWORD1,rWORD2 + ori rRTN, rWORD2, 1 blr - +#endif /* Oh well. In this case, we just do a byte-by-byte comparison. */ .align 4 --- a/sysdeps/powerpc/powerpc32/strcmp.S +++ b/sysdeps/powerpc/powerpc32/strcmp.S @@ -26,7 +26,7 @@ EALIGN (BP_SYM (strcmp), 4, 0) -#define rTMP r0 +#define rTMP2 r0 #define rRTN r3 #define rSTR1 r3 /* first string arg */ #define rSTR2 r4 /* second string arg */ @@ -40,6 +40,7 @@ #define r7F7F r8 /* constant 0x7f7f7f7f */ #define rNEG r9 /* ~(word in s1 | 0x7f7f7f7f) */ #define rBITDIF r10 /* bits that differ in s1 & s2 words */ +#define rTMP r11 CHECK_BOUNDS_LOW (rSTR1, rTMP, rHIGH1) CHECK_BOUNDS_LOW (rSTR2, rTMP, rHIGH2) @@ -64,10 +65,45 @@ and. rTMP, rTMP, rNEG cmpw cr1, rWORD1, rWORD2 beq+ L(g0) -L(endstring): + /* OK. We've hit the end of the string. We need to be careful that we don't compare two strings as different because of gunk beyond the end of the strings... */ +#ifdef __LITTLE_ENDIAN__ +L(endstring): + addi rTMP2, rTMP, -1 + andc rTMP2, rTMP2, rTMP + rlwimi rTMP2, rTMP2, 1, 0, 30 + and rWORD2, rWORD2, rTMP2 /* Mask off gunk. */ + and rWORD1, rWORD1, rTMP2 + rlwinm rTMP2, rWORD2, 8, 0xffffffff /* Byte reverse word. */ + rlwinm rTMP, rWORD1, 8, 0xffffffff + rlwimi rTMP2, rWORD2, 24, 0, 7 + rlwimi rTMP, rWORD1, 24, 0, 7 + rlwimi rTMP2, rWORD2, 24, 16, 23 + rlwimi rTMP, rWORD1, 24, 16, 23 + xor. rBITDIF, rTMP, rTMP2 + sub rRTN, rTMP, rTMP2 + bgelr+ + ori rRTN, rTMP2, 1 + blr + +L(different): + lwz rWORD1, -4(rSTR1) + rlwinm rTMP2, rWORD2, 8, 0xffffffff /* Byte reverse word. */ + rlwinm rTMP, rWORD1, 8, 0xffffffff + rlwimi rTMP2, rWORD2, 24, 0, 7 + rlwimi rTMP, rWORD1, 24, 0, 7 + rlwimi rTMP2, rWORD2, 24, 16, 23 + rlwimi rTMP, rWORD1, 24, 16, 23 + xor. rBITDIF, rTMP, rTMP2 + sub rRTN, rTMP, rTMP2 + bgelr+ + ori rRTN, rTMP2, 1 + blr + +#else +L(endstring): and rTMP, r7F7F, rWORD1 beq cr1, L(equal) add rTMP, rTMP, r7F7F @@ -94,7 +130,7 @@ ori rRTN, rWORD2, 1 /* GKM FIXME: check high bounds. */ blr - +#endif /* Oh well. In this case, we just do a byte-by-byte comparison. */ .align 4 --- a/sysdeps/powerpc/powerpc32/strncmp.S +++ b/sysdeps/powerpc/powerpc32/strncmp.S @@ -26,7 +26,7 @@ EALIGN (BP_SYM(strncmp), 4, 0) -#define rTMP r0 +#define rTMP2 r0 #define rRTN r3 #define rSTR1 r3 /* first string arg */ #define rSTR2 r4 /* second string arg */ @@ -40,6 +40,7 @@ #define r7F7F r9 /* constant 0x7f7f7f7f */ #define rNEG r10 /* ~(word in s1 | 0x7f7f7f7f) */ #define rBITDIF r11 /* bits that differ in s1 & s2 words */ +#define rTMP r12 dcbt 0,rSTR1 or rTMP, rSTR2, rSTR1 @@ -78,12 +79,45 @@ we don't compare two strings as different because of gunk beyond the end of the strings... */ +#ifdef __LITTLE_ENDIAN__ +L(endstring): + slwi rTMP, rTMP, 1 + addi rTMP2, rTMP, -1 + andc rTMP2, rTMP2, rTMP + and rWORD2, rWORD2, rTMP2 /* Mask off gunk. */ + and rWORD1, rWORD1, rTMP2 + rlwinm rTMP2, rWORD2, 8, 0xffffffff /* Byte reverse word. */ + rlwinm rTMP, rWORD1, 8, 0xffffffff + rlwimi rTMP2, rWORD2, 24, 0, 7 + rlwimi rTMP, rWORD1, 24, 0, 7 + rlwimi rTMP2, rWORD2, 24, 16, 23 + rlwimi rTMP, rWORD1, 24, 16, 23 + xor. rBITDIF, rTMP, rTMP2 + sub rRTN, rTMP, rTMP2 + bgelr+ + ori rRTN, rTMP2, 1 + blr + +L(different): + lwz rWORD1, -4(rSTR1) + rlwinm rTMP2, rWORD2, 8, 0xffffffff /* Byte reverse word. */ + rlwinm rTMP, rWORD1, 8, 0xffffffff + rlwimi rTMP2, rWORD2, 24, 0, 7 + rlwimi rTMP, rWORD1, 24, 0, 7 + rlwimi rTMP2, rWORD2, 24, 16, 23 + rlwimi rTMP, rWORD1, 24, 16, 23 + xor. rBITDIF, rTMP, rTMP2 + sub rRTN, rTMP, rTMP2 + bgelr+ + ori rRTN, rTMP2, 1 + blr + +#else L(endstring): and rTMP, r7F7F, rWORD1 beq cr1, L(equal) add rTMP, rTMP, r7F7F xor. rBITDIF, rWORD1, rWORD2 - andc rNEG, rNEG, rTMP blt- L(highbit) cntlzw rBITDIF, rBITDIF @@ -91,28 +125,20 @@ addi rNEG, rNEG, 7 cmpw cr1, rNEG, rBITDIF sub rRTN, rWORD1, rWORD2 - blt- cr1, L(equal) - srawi rRTN, rRTN, 31 - ori rRTN, rRTN, 1 - blr + bgelr+ cr1 L(equal): li rRTN, 0 blr L(different): - lwzu rWORD1, -4(rSTR1) + lwz rWORD1, -4(rSTR1) xor. rBITDIF, rWORD1, rWORD2 sub rRTN, rWORD1, rWORD2 - blt- L(highbit) - srawi rRTN, rRTN, 31 - ori rRTN, rRTN, 1 - blr + bgelr+ L(highbit): - srwi rWORD2, rWORD2, 24 - srwi rWORD1, rWORD1, 24 - sub rRTN, rWORD1, rWORD2 + ori rRTN, rWORD2, 1 blr - +#endif /* Oh well. In this case, we just do a byte-by-byte comparison. */ .align 4 --- a/sysdeps/powerpc/powerpc64/power4/strncmp.S +++ b/sysdeps/powerpc/powerpc64/power4/strncmp.S @@ -27,7 +27,7 @@ EALIGN (BP_SYM(strncmp), 4, 0) CALL_MCOUNT 3 -#define rTMP r0 +#define rTMP2 r0 #define rRTN r3 #define rSTR1 r3 /* first string arg */ #define rSTR2 r4 /* second string arg */ @@ -43,6 +43,7 @@ #define r7F7F r9 /* constant 0x7f7f7f7f7f7f7f7f */ #define rNEG r10 /* ~(word in s1 | 0x7f7f7f7f7f7f7f7f) */ #define rBITDIF r11 /* bits that differ in s1 & s2 words */ +#define rTMP r12 dcbt 0,rSTR1 or rTMP, rSTR2, rSTR1 @@ -84,12 +85,59 @@ we don't compare two strings as different because of gunk beyond the end of the strings... */ +#ifdef __LITTLE_ENDIAN__ +L(endstring): + addi rTMP2, rTMP, -1 + beq cr1, L(equal) + andc rTMP2, rTMP2, rTMP + rldimi rTMP2, rTMP2, 1, 0 + and rWORD2, rWORD2, rTMP2 /* Mask off gunk. */ + and rWORD1, rWORD1, rTMP2 + cmpd cr1, rWORD1, rWORD2 + beq cr1, L(equal) + xor rBITDIF, rWORD1, rWORD2 /* rBITDIF has bits that differ. */ + neg rNEG, rBITDIF + and rNEG, rNEG, rBITDIF /* rNEG has LS bit that differs. */ + cntlzd rNEG, rNEG /* bitcount of the bit. */ + andi. rNEG, rNEG, 56 /* bitcount to LS byte that differs. */ + sld rWORD1, rWORD1, rNEG /* shift left to clear MS bytes. */ + sld rWORD2, rWORD2, rNEG + xor. rBITDIF, rWORD1, rWORD2 + sub rRTN, rWORD1, rWORD2 + blt- L(highbit) + sradi rRTN, rRTN, 63 /* must return an int. */ + ori rRTN, rRTN, 1 + blr +L(equal): + li rRTN, 0 + blr + +L(different): + ld rWORD1, -8(rSTR1) + xor rBITDIF, rWORD1, rWORD2 /* rBITDIF has bits that differ. */ + neg rNEG, rBITDIF + and rNEG, rNEG, rBITDIF /* rNEG has LS bit that differs. */ + cntlzd rNEG, rNEG /* bitcount of the bit. */ + andi. rNEG, rNEG, 56 /* bitcount to LS byte that differs. */ + sld rWORD1, rWORD1, rNEG /* shift left to clear MS bytes. */ + sld rWORD2, rWORD2, rNEG + xor. rBITDIF, rWORD1, rWORD2 + sub rRTN, rWORD1, rWORD2 + blt- L(highbit) + sradi rRTN, rRTN, 63 + ori rRTN, rRTN, 1 + blr +L(highbit): + sradi rRTN, rWORD2, 63 + ori rRTN, rRTN, 1 + blr + +#else L(endstring): and rTMP, r7F7F, rWORD1 beq cr1, L(equal) add rTMP, rTMP, r7F7F xor. rBITDIF, rWORD1, rWORD2 - andc rNEG, rNEG, rTMP blt- L(highbit) cntlzd rBITDIF, rBITDIF @@ -98,7 +146,7 @@ cmpd cr1, rNEG, rBITDIF sub rRTN, rWORD1, rWORD2 blt- cr1, L(equal) - sradi rRTN, rRTN, 63 + sradi rRTN, rRTN, 63 /* must return an int. */ ori rRTN, rRTN, 1 blr L(equal): @@ -106,7 +154,7 @@ blr L(different): - ldu rWORD1, -8(rSTR1) + ld rWORD1, -8(rSTR1) xor. rBITDIF, rWORD1, rWORD2 sub rRTN, rWORD1, rWORD2 blt- L(highbit) @@ -114,11 +162,10 @@ ori rRTN, rRTN, 1 blr L(highbit): - srdi rWORD2, rWORD2, 56 - srdi rWORD1, rWORD1, 56 - sub rRTN, rWORD1, rWORD2 + sradi rRTN, rWORD2, 63 + ori rRTN, rRTN, 1 blr - +#endif /* Oh well. In this case, we just do a byte-by-byte comparison. */ .align 4 --- a/sysdeps/powerpc/powerpc64/power7/strncmp.S +++ b/sysdeps/powerpc/powerpc64/power7/strncmp.S @@ -29,7 +29,7 @@ EALIGN (BP_SYM(strncmp),5,0) CALL_MCOUNT 3 -#define rTMP r0 +#define rTMP2 r0 #define rRTN r3 #define rSTR1 r3 /* first string arg */ #define rSTR2 r4 /* second string arg */ @@ -45,6 +45,7 @@ #define r7F7F r9 /* constant 0x7f7f7f7f7f7f7f7f */ #define rNEG r10 /* ~(word in s1 | 0x7f7f7f7f7f7f7f7f) */ #define rBITDIF r11 /* bits that differ in s1 & s2 words */ +#define rTMP r12 dcbt 0,rSTR1 nop @@ -88,12 +89,57 @@ we don't compare two strings as different because of gunk beyond the end of the strings... */ +#ifdef __LITTLE_ENDIAN__ +L(endstring): + addi rTMP2, rTMP, -1 + beq cr1, L(equal) + andc rTMP2, rTMP2, rTMP + rldimi rTMP2, rTMP2, 1, 0 + and rWORD2, rWORD2, rTMP2 /* Mask off gunk. */ + and rWORD1, rWORD1, rTMP2 + cmpd cr1, rWORD1, rWORD2 + beq cr1, L(equal) + cmpb rBITDIF, rWORD1, rWORD2 /* 0xff on equal bytes. */ + addi rNEG, rBITDIF, 1 + orc rNEG, rNEG, rBITDIF /* 0's below LS differing byte. */ + sldi rNEG, rNEG, 8 /* 1's above LS differing byte. */ + andc rWORD1, rWORD1, rNEG /* mask off MS bytes. */ + andc rWORD2, rWORD2, rNEG + xor. rBITDIF, rWORD1, rWORD2 + sub rRTN, rWORD1, rWORD2 + blt L(highbit) + sradi rRTN, rRTN, 63 /* must return an int. */ + ori rRTN, rRTN, 1 + blr +L(equal): + li rRTN, 0 + blr + +L(different): + ld rWORD1, -8(rSTR1) + cmpb rBITDIF, rWORD1, rWORD2 /* 0xff on equal bytes. */ + addi rNEG, rBITDIF, 1 + orc rNEG, rNEG, rBITDIF /* 0's below LS differing byte. */ + sldi rNEG, rNEG, 8 /* 1's above LS differing byte. */ + andc rWORD1, rWORD1, rNEG /* mask off MS bytes. */ + andc rWORD2, rWORD2, rNEG + xor. rBITDIF, rWORD1, rWORD2 + sub rRTN, rWORD1, rWORD2 + blt L(highbit) + sradi rRTN, rRTN, 63 + ori rRTN, rRTN, 1 + blr +L(highbit): + sradi rRTN, rWORD2, 63 + ori rRTN, rRTN, 1 + blr + +#else L(endstring): and rTMP,r7F7F,rWORD1 beq cr1,L(equal) add rTMP,rTMP,r7F7F xor. rBITDIF,rWORD1,rWORD2 - andc rNEG,rNEG,rTMP blt L(highbit) cntlzd rBITDIF,rBITDIF @@ -102,7 +148,7 @@ cmpd cr1,rNEG,rBITDIF sub rRTN,rWORD1,rWORD2 blt cr1,L(equal) - sradi rRTN,rRTN,63 + sradi rRTN,rRTN,63 /* must return an int. */ ori rRTN,rRTN,1 blr L(equal): @@ -110,7 +156,7 @@ blr L(different): - ldu rWORD1,-8(rSTR1) + ld rWORD1,-8(rSTR1) xor. rBITDIF,rWORD1,rWORD2 sub rRTN,rWORD1,rWORD2 blt L(highbit) @@ -118,11 +164,10 @@ ori rRTN,rRTN,1 blr L(highbit): - srdi rWORD2,rWORD2,56 - srdi rWORD1,rWORD1,56 - sub rRTN,rWORD1,rWORD2 + sradi rRTN,rWORD2,63 + ori rRTN,rRTN,1 blr - +#endif /* Oh well. In this case, we just do a byte-by-byte comparison. */ .align 4 --- a/sysdeps/powerpc/powerpc64/strcmp.S +++ b/sysdeps/powerpc/powerpc64/strcmp.S @@ -27,7 +27,7 @@ EALIGN (BP_SYM(strcmp), 4, 0) CALL_MCOUNT 2 -#define rTMP r0 +#define rTMP2 r0 #define rRTN r3 #define rSTR1 r3 /* first string arg */ #define rSTR2 r4 /* second string arg */ @@ -46,6 +46,7 @@ #define r7F7F r8 /* constant 0x7f7f7f7f7f7f7f7f */ #define rNEG r9 /* ~(word in s1 | 0x7f7f7f7f7f7f7f7f) */ #define rBITDIF r10 /* bits that differ in s1 & s2 words */ +#define rTMP r11 CHECK_BOUNDS_LOW (rSTR1, rTMP, rHIGH1) CHECK_BOUNDS_LOW (rSTR2, rTMP, rHIGH2) @@ -72,19 +73,66 @@ ldu rWORD2, 8(rSTR2) L(g1): add rTMP, rFEFE, rWORD1 nor rNEG, r7F7F, rWORD1 - and. rTMP, rTMP, rNEG cmpd cr1, rWORD1, rWORD2 beq+ L(g0) -L(endstring): + /* OK. We've hit the end of the string. We need to be careful that we don't compare two strings as different because of gunk beyond the end of the strings... */ +#ifdef __LITTLE_ENDIAN__ +L(endstring): + addi rTMP2, rTMP, -1 + beq cr1, L(equal) + andc rTMP2, rTMP2, rTMP + rldimi rTMP2, rTMP2, 1, 0 + and rWORD2, rWORD2, rTMP2 /* Mask off gunk. */ + and rWORD1, rWORD1, rTMP2 + cmpd cr1, rWORD1, rWORD2 + beq cr1, L(equal) + xor rBITDIF, rWORD1, rWORD2 /* rBITDIF has bits that differ. */ + neg rNEG, rBITDIF + and rNEG, rNEG, rBITDIF /* rNEG has LS bit that differs. */ + cntlzd rNEG, rNEG /* bitcount of the bit. */ + andi. rNEG, rNEG, 56 /* bitcount to LS byte that differs. */ + sld rWORD1, rWORD1, rNEG /* shift left to clear MS bytes. */ + sld rWORD2, rWORD2, rNEG + xor. rBITDIF, rWORD1, rWORD2 + sub rRTN, rWORD1, rWORD2 + blt- L(highbit) + sradi rRTN, rRTN, 63 /* must return an int. */ + ori rRTN, rRTN, 1 + blr +L(equal): + li rRTN, 0 + blr + +L(different): + ld rWORD1, -8(rSTR1) + xor rBITDIF, rWORD1, rWORD2 /* rBITDIF has bits that differ. */ + neg rNEG, rBITDIF + and rNEG, rNEG, rBITDIF /* rNEG has LS bit that differs. */ + cntlzd rNEG, rNEG /* bitcount of the bit. */ + andi. rNEG, rNEG, 56 /* bitcount to LS byte that differs. */ + sld rWORD1, rWORD1, rNEG /* shift left to clear MS bytes. */ + sld rWORD2, rWORD2, rNEG + xor. rBITDIF, rWORD1, rWORD2 + sub rRTN, rWORD1, rWORD2 + blt- L(highbit) + sradi rRTN, rRTN, 63 + ori rRTN, rRTN, 1 + blr +L(highbit): + sradi rRTN, rWORD2, 63 + ori rRTN, rRTN, 1 + blr + +#else +L(endstring): and rTMP, r7F7F, rWORD1 beq cr1, L(equal) add rTMP, rTMP, r7F7F xor. rBITDIF, rWORD1, rWORD2 - andc rNEG, rNEG, rTMP blt- L(highbit) cntlzd rBITDIF, rBITDIF @@ -93,7 +141,7 @@ cmpd cr1, rNEG, rBITDIF sub rRTN, rWORD1, rWORD2 blt- cr1, L(equal) - sradi rRTN, rRTN, 63 + sradi rRTN, rRTN, 63 /* must return an int. */ ori rRTN, rRTN, 1 blr L(equal): @@ -110,12 +158,11 @@ ori rRTN, rRTN, 1 blr L(highbit): - srdi rWORD2, rWORD2, 56 - srdi rWORD1, rWORD1, 56 - sub rRTN, rWORD1, rWORD2 + sradi rRTN, rWORD2, 63 + ori rRTN, rRTN, 1 /* GKM FIXME: check high bounds. */ blr - +#endif /* Oh well. In this case, we just do a byte-by-byte comparison. */ .align 4 --- a/sysdeps/powerpc/powerpc64/strncmp.S +++ b/sysdeps/powerpc/powerpc64/strncmp.S @@ -27,7 +27,7 @@ EALIGN (BP_SYM(strncmp), 4, 0) CALL_MCOUNT 3 -#define rTMP r0 +#define rTMP2 r0 #define rRTN r3 #define rSTR1 r3 /* first string arg */ #define rSTR2 r4 /* second string arg */ @@ -41,6 +41,7 @@ #define r7F7F r9 /* constant 0x7f7f7f7f7f7f7f7f */ #define rNEG r10 /* ~(word in s1 | 0x7f7f7f7f7f7f7f7f) */ #define rBITDIF r11 /* bits that differ in s1 & s2 words */ +#define rTMP r12 dcbt 0,rSTR1 or rTMP, rSTR2, rSTR1 @@ -81,13 +82,60 @@ /* OK. We've hit the end of the string. We need to be careful that we don't compare two strings as different because of gunk beyond the end of the strings... */ - + +#ifdef __LITTLE_ENDIAN__ +L(endstring): + addi rTMP2, rTMP, -1 + beq cr1, L(equal) + andc rTMP2, rTMP2, rTMP + rldimi rTMP2, rTMP2, 1, 0 + and rWORD2, rWORD2, rTMP2 /* Mask off gunk. */ + and rWORD1, rWORD1, rTMP2 + cmpd cr1, rWORD1, rWORD2 + beq cr1, L(equal) + xor rBITDIF, rWORD1, rWORD2 /* rBITDIF has bits that differ. */ + neg rNEG, rBITDIF + and rNEG, rNEG, rBITDIF /* rNEG has LS bit that differs. */ + cntlzd rNEG, rNEG /* bitcount of the bit. */ + andi. rNEG, rNEG, 56 /* bitcount to LS byte that differs. */ + sld rWORD1, rWORD1, rNEG /* shift left to clear MS bytes. */ + sld rWORD2, rWORD2, rNEG + xor. rBITDIF, rWORD1, rWORD2 + sub rRTN, rWORD1, rWORD2 + blt- L(highbit) + sradi rRTN, rRTN, 63 /* must return an int. */ + ori rRTN, rRTN, 1 + blr +L(equal): + li rRTN, 0 + blr + +L(different): + ld rWORD1, -8(rSTR1) + xor rBITDIF, rWORD1, rWORD2 /* rBITDIF has bits that differ. */ + neg rNEG, rBITDIF + and rNEG, rNEG, rBITDIF /* rNEG has LS bit that differs. */ + cntlzd rNEG, rNEG /* bitcount of the bit. */ + andi. rNEG, rNEG, 56 /* bitcount to LS byte that differs. */ + sld rWORD1, rWORD1, rNEG /* shift left to clear MS bytes. */ + sld rWORD2, rWORD2, rNEG + xor. rBITDIF, rWORD1, rWORD2 + sub rRTN, rWORD1, rWORD2 + blt- L(highbit) + sradi rRTN, rRTN, 63 + ori rRTN, rRTN, 1 + blr +L(highbit): + sradi rRTN, rWORD2, 63 + ori rRTN, rRTN, 1 + blr + +#else L(endstring): and rTMP, r7F7F, rWORD1 beq cr1, L(equal) add rTMP, rTMP, r7F7F xor. rBITDIF, rWORD1, rWORD2 - andc rNEG, rNEG, rTMP blt- L(highbit) cntlzd rBITDIF, rBITDIF @@ -96,7 +144,7 @@ cmpd cr1, rNEG, rBITDIF sub rRTN, rWORD1, rWORD2 blt- cr1, L(equal) - sradi rRTN, rRTN, 63 + sradi rRTN, rRTN, 63 /* must return an int. */ ori rRTN, rRTN, 1 blr L(equal): @@ -104,7 +152,7 @@ blr L(different): - ldu rWORD1, -8(rSTR1) + ld rWORD1, -8(rSTR1) xor. rBITDIF, rWORD1, rWORD2 sub rRTN, rWORD1, rWORD2 blt- L(highbit) @@ -112,11 +160,10 @@ ori rRTN, rRTN, 1 blr L(highbit): - srdi rWORD2, rWORD2, 56 - srdi rWORD1, rWORD1, 56 - sub rRTN, rWORD1, rWORD2 + sradi rRTN, rWORD2, 63 + ori rRTN, rRTN, 1 blr - +#endif /* Oh well. In this case, we just do a byte-by-byte comparison. */ .align 4