diff options
Diffstat (limited to 'sysdeps/powerpc/powerpc32/power7')
| -rw-r--r-- | sysdeps/powerpc/powerpc32/power7/strchr.S | 51 | ||||
| -rw-r--r-- | sysdeps/powerpc/powerpc32/power7/strchrnul.S | 27 |
2 files changed, 58 insertions, 20 deletions
diff --git a/sysdeps/powerpc/powerpc32/power7/strchr.S b/sysdeps/powerpc/powerpc32/power7/strchr.S index 0ecadb271a..b662659671 100644 --- a/sysdeps/powerpc/powerpc32/power7/strchr.S +++ b/sysdeps/powerpc/powerpc32/power7/strchr.S @@ -35,8 +35,8 @@ ENTRY (strchr) beq cr7,L(null_match) /* Replicate byte to word. */ - rlwimi r4,r4,8,16,23 - rlwimi r4,r4,16,0,15 + insrdi r4,r4,8,48 + insrdi r4,r4,16,32 /* Now r4 has a word of c bytes and r0 has a word of null bytes. */ @@ -46,11 +46,17 @@ ENTRY (strchr) /* Move the words left and right to discard the bits that are not part of the string and to bring them back as zeros. */ - +#ifdef __LITTLE_ENDIAN__ + srw r10,r10,r6 + srw r11,r11,r6 + slw r10,r10,r6 + slw r11,r11,r6 +#else slw r10,r10,r6 slw r11,r11,r6 srw r10,r10,r6 srw r11,r11,r6 +#endif or r5,r10,r11 /* OR the results to speed things up. */ cmpwi cr7,r5,0 /* If r5 == 0, no c or null bytes have been found. */ @@ -65,7 +71,7 @@ ENTRY (strchr) /* Handle WORD2 of pair. */ lwzu r12,4(r8) - cmpb r10,r12,r4 + cmpb r10,r12,r4 cmpb r11,r12,r0 or r5,r10,r11 cmpwi cr7,r5,0 @@ -100,22 +106,31 @@ L(loop): bne cr6,L(done) /* The c/null byte must be in the second word. Adjust the address - again and move the result of cmpb to r10 so we can calculate the - pointer. */ + again and move the result of cmpb to r10/r11 so we can calculate + the pointer. */ mr r10,r6 mr r11,r7 addi r8,r8,4 - /* r5 has the output of the cmpb instruction, that is, it contains + /* r10/r11 have the output of the cmpb instructions, that is, 0xff in the same position as the c/null byte in the original word from the string. Use that to calculate the pointer. */ L(done): - cntlzw r4,r10 /* Count leading zeroes before c matches. */ - cntlzw r0,r11 /* Count leading zeroes before null matches. */ - cmplw cr7,r4,r0 +#ifdef __LITTLE_ENDIAN__ + addi r3,r10,-1 + andc r3,r3,r10 + popcntw r0,r3 + addi r4,r11,-1 + andc r4,r4,r11 + cmplw cr7,r3,r4 + bgt cr7,L(no_match) +#else + cntlzw r0,r10 /* Count leading zeros before c matches. */ + cmplw cr7,r11,r10 bgt cr7,L(no_match) - srwi r0,r4,3 /* Convert leading zeroes to bytes. */ +#endif + srwi r0,r0,3 /* Convert leading zeros to bytes. */ add r3,r8,r0 /* Return address of the matching c byte or null in case c was not found. */ blr @@ -133,10 +148,14 @@ L(null_match): cmpb r5,r12,r0 /* Compare each byte against null bytes. */ /* Move the words left and right to discard the bits that are - not part of the string and to bring them back as zeros. */ - + not part of the string and bring them back as zeros. */ +#ifdef __LITTLE_ENDIAN__ + srw r5,r5,r6 + slw r5,r5,r6 +#else slw r5,r5,r6 srw r5,r5,r6 +#endif cmpwi cr7,r5,0 /* If r10 == 0, no c or null bytes have been found. */ bne cr7,L(done_null) @@ -191,7 +210,13 @@ L(loop_null): 0xff in the same position as the null byte in the original word from the string. Use that to calculate the pointer. */ L(done_null): +#ifdef __LITTLE_ENDIAN__ + addi r0,r5,-1 + andc r0,r0,r5 + popcntw r0,r0 +#else cntlzw r0,r5 /* Count leading zeros before the match. */ +#endif srwi r0,r0,3 /* Convert leading zeros to bytes. */ add r3,r8,r0 /* Return address of the matching null byte. */ blr diff --git a/sysdeps/powerpc/powerpc32/power7/strchrnul.S b/sysdeps/powerpc/powerpc32/power7/strchrnul.S index d4cacab60b..f5d24d4340 100644 --- a/sysdeps/powerpc/powerpc32/power7/strchrnul.S +++ b/sysdeps/powerpc/powerpc32/power7/strchrnul.S @@ -27,8 +27,8 @@ ENTRY (__strchrnul) clrrwi r8,r3,2 /* Align the address to word boundary. */ /* Replicate byte to word. */ - rlwimi r4,r4,8,16,23 - rlwimi r4,r4,16,0,15 + insrdi r4,r4,8,48 + insrdi r4,r4,16,32 rlwinm r6,r3,3,27,28 /* Calculate padding. */ lwz r12,0(r8) /* Load word from memory. */ @@ -43,10 +43,17 @@ ENTRY (__strchrnul) /* Move the words left and right to discard the bits that are not part of the string and bring them back as zeros. */ +#ifdef __LITTLE_ENDIAN__ + srw r10,r10,r6 + srw r9,r9,r6 + slw r10,r10,r6 + slw r9,r9,r6 +#else slw r10,r10,r6 slw r9,r9,r6 srw r10,r10,r6 srw r9,r9,r6 +#endif or r5,r9,r10 /* OR the results to speed things up. */ cmpwi cr7,r5,0 /* If r5 == 0, no c or null bytes have been found. */ @@ -54,7 +61,7 @@ ENTRY (__strchrnul) mtcrf 0x01,r8 - /* Are we now aligned to a quadword boundary? If so, skip to + /* Are we now aligned to a doubleword boundary? If so, skip to the main loop. Otherwise, go through the alignment code. */ bt 29,L(loop) @@ -76,7 +83,7 @@ L(loop): single register for speed. This is an attempt to speed up the null-checking process for bigger strings. */ lwz r12,4(r8) - lwzu r11,8(r8) + lwzu r11,8(r8) cmpb r10,r12,r0 cmpb r9,r12,r4 cmpb r6,r11,r0 @@ -95,9 +102,9 @@ L(loop): addi r8,r8,-4 bne cr6,L(done) - /* The c/null byte must be in the second word. Adjust the - address again and move the result of cmpb to r10 so we can calculate - the pointer. */ + /* The c/null byte must be in the second word. Adjust the address + again and move the result of cmpb to r5 so we can calculate the + pointer. */ mr r5,r10 addi r8,r8,4 @@ -105,7 +112,13 @@ L(loop): 0xff in the same position as the c/null byte in the original word from the string. Use that to calculate the pointer. */ L(done): +#ifdef __LITTLE_ENDIAN__ + addi r0,r5,-1 + andc r0,r0,r5 + popcntw r0,r0 +#else cntlzw r0,r5 /* Count leading zeros before the match. */ +#endif srwi r0,r0,3 /* Convert leading zeros to bytes. */ add r3,r8,r0 /* Return address of matching c/null byte. */ blr |