The alpha assembly version of various string functions do not work correctly on some corner cases, and thus doesn't pass the testsuite. This patch removes them. --- ports/sysdeps/alpha/alphaev67/stpncpy.S | 115 ------------- ports/sysdeps/alpha/alphaev67/strncat.S | 87 ---------- ports/sysdeps/alpha/stpncpy.S | 106 ------------ ports/sysdeps/alpha/strcmp.S | 194 ---------------------- ports/sysdeps/alpha/strncat.S | 94 ---------- ports/sysdeps/alpha/strncmp.S | 277 -------------------------------- ports/sysdeps/alpha/strncpy.S | 87 ---------- 7 files changed, 960 deletions(-) --- a/ports/sysdeps/alpha/strncat.S +++ /dev/null @@ -1,94 +0,0 @@ -/* Copyright (C) 1996-2014 Free Software Foundation, Inc. - This file is part of the GNU C Library. - Contributed by Richard Henderson , 1996. - - The GNU C Library is free software; you can redistribute it and/or - modify it under the terms of the GNU Lesser General Public - License as published by the Free Software Foundation; either - version 2.1 of the License, or (at your option) any later version. - - The GNU C Library is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - Lesser General Public License for more details. - - You should have received a copy of the GNU Lesser General Public - License along with the GNU C Library. If not, see - . */ - -/* Append no more than COUNT characters from the null-terminated string SRC - to the null-terminated string DST. Always null-terminate the new DST. */ - -#include - - .text - -ENTRY(strncat) - ldgp gp, 0(pv) -#ifdef PROF - .set noat - lda AT, _mcount - jsr AT, (AT), _mcount - .set at -#endif - .prologue 1 - - mov a0, v0 # set up return value - beq a2, $zerocount - - /* Find the end of the string. */ - - ldq_u t0, 0(a0) # load first quadword (a0 may be misaligned) - lda t1, -1(zero) - insqh t1, a0, t1 - andnot a0, 7, a0 - or t1, t0, t0 - cmpbge zero, t0, t1 # t1 <- bitmask: bit i == 1 <==> i-th byte == 0 - bne t1, $found - -$loop: ldq t0, 8(a0) - addq a0, 8, a0 # addr += 8 - cmpbge zero, t0, t1 - beq t1, $loop - -$found: negq t1, t2 # clear all but least set bit - and t1, t2, t1 - - and t1, 0xf0, t2 # binary search for that set bit - and t1, 0xcc, t3 - and t1, 0xaa, t4 - cmovne t2, 4, t2 - cmovne t3, 2, t3 - cmovne t4, 1, t4 - addq t2, t3, t2 - addq a0, t4, a0 - addq a0, t2, a0 - - /* Now do the append. */ - - jsr t9, __stxncpy - - /* Worry about the null termination. */ - - zapnot t0, t8, t1 # was last byte a null? - bne t1, 0f - ret - -0: and t10, 0x80, t1 - bne t1, 1f - - /* Here there are bytes left in the current word. Clear one. */ - addq t10, t10, t10 # end-of-count bit <<= 1 - zap t0, t10, t0 - stq_u t0, 0(a0) - ret - -1: /* Here we must read the next DST word and clear the first byte. */ - ldq_u t0, 8(a0) - zap t0, 1, t0 - stq_u t0, 8(a0) - -$zerocount: - ret - - END(strncat) --- a/ports/sysdeps/alpha/strncpy.S +++ /dev/null @@ -1,87 +0,0 @@ -/* Copyright (C) 1996-2014 Free Software Foundation, Inc. - Contributed by Richard Henderson (rth@tamu.edu) - This file is part of the GNU C Library. - - The GNU C Library is free software; you can redistribute it and/or - modify it under the terms of the GNU Lesser General Public - License as published by the Free Software Foundation; either - version 2.1 of the License, or (at your option) any later version. - - The GNU C Library is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - Lesser General Public License for more details. - - You should have received a copy of the GNU Lesser General Public - License along with the GNU C Library. If not, see - . */ - -/* Copy no more than COUNT bytes of the null-terminated string from - SRC to DST. If SRC does not cover all of COUNT, the balance is - zeroed. */ - -#include - - .set noat - .set noreorder - - .text - -ENTRY(strncpy) - ldgp gp, 0(pv) -#ifdef PROF - lda AT, _mcount - jsr AT, (AT), _mcount -#endif - .prologue 1 - - mov a0, v0 # set return value now - beq a2, $zerocount - jsr t9, __stxncpy # do the work of the copy - - bne a2, $multiword # do we have full words left? - - .align 3 - subq t8, 1, t2 # e0 : guess not - subq t10, 1, t3 # .. e1 : - or t2, t8, t2 # e0 : clear the bits between the last - or t3, t10, t3 # .. e1 : written byte and the last byte in - andnot t3, t2, t3 # e0 : COUNT - zap t0, t3, t0 # e1 : - stq_u t0, 0(a0) # e0 : - ret # .. e1 : - -$multiword: - subq t8, 1, t7 # e0 : clear the final bits in the prev - or t7, t8, t7 # e1 : word - zapnot t0, t7, t0 # e0 : - subq a2, 1, a2 # .. e1 : - stq_u t0, 0(a0) # e0 : - addq a0, 8, a0 # .. e1 : - - beq a2, 1f # e1 : - blbc a2, 0f # e1 : - - stq_u zero, 0(a0) # e0 : zero one word - subq a2, 1, a2 # .. e1 : - addq a0, 8, a0 # e0 : - beq a2, 1f # .. e1 : - -0: stq_u zero, 0(a0) # e0 : zero two words - subq a2, 2, a2 # .. e1 : - stq_u zero, 8(a0) # e0 : - addq a0, 16, a0 # .. e1 : - bne a2, 0b # e1 : - unop - -1: ldq_u t0, 0(a0) # e0 : clear the leading bits in the final - subq t10, 1, t7 # .. e1 : word - or t7, t10, t7 # e0 : - zap t0, t7, t0 # e1 (stall) - stq_u t0, 0(a0) # e0 : - -$zerocount: - ret # .. e1 : - - END(strncpy) -libc_hidden_builtin_def (strncpy) --- a/ports/sysdeps/alpha/alphaev67/strncat.S +++ /dev/null @@ -1,87 +0,0 @@ -/* Copyright (C) 2000-2014 Free Software Foundation, Inc. - Contributed by Richard Henderson , 1996. - EV67 optimized by Rick Gorton . - This file is part of the GNU C Library. - - The GNU C Library is free software; you can redistribute it and/or - modify it under the terms of the GNU Lesser General Public - License as published by the Free Software Foundation; either - version 2.1 of the License, or (at your option) any later version. - - The GNU C Library is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - Lesser General Public License for more details. - - You should have received a copy of the GNU Lesser General Public - License along with the GNU C Library. If not, see - . */ - -/* Append no more than COUNT characters from the null-terminated string SRC - to the null-terminated string DST. Always null-terminate the new DST. */ - -#include - - .arch ev6 - .set noreorder - .text - -ENTRY(strncat) - ldgp gp, 0(pv) -#ifdef PROF - .set noat - lda AT, _mcount - jsr AT, (AT), _mcount - .set at -#endif - .prologue 1 - - mov a0, v0 # set up return value - beq a2, $zerocount # U : - /* Find the end of the string. */ - ldq_u t0, 0(a0) # L : load first quadword (a0 may be misaligned) - lda t1, -1 # E : - - insqh t1, v0, t1 # U : - andnot a0, 7, a0 # E : - nop # E : - or t1, t0, t0 # E : - - nop # E : - nop # E : - cmpbge zero, t0, t1 # E : bits set iff byte == 0 - bne t1, $found # U : - -$loop: ldq t0, 8(a0) # L : - addq a0, 8, a0 # E : - cmpbge zero, t0, t1 # E : - beq t1, $loop # U : - -$found: cttz t1, t2 # U0 : - addq a0, t2, a0 # E : - jsr t9, __stxncpy # L0 : Now do the append. - - /* Worry about the null termination. */ - - cttz t10, t2 # U0: byte offset of end-of-count. - bic a0, 7, a0 # E : word align the last write address. - zapnot t0, t8, t1 # U : was last byte a null? - nop # E : - - bne t1, 0f # U : - nop # E : - nop # E : - ret # L0 : - -0: addq t2, a0, a0 # E : address of end-of-count - stb zero, 1(a0) # L : - nop # E : - ret # L0 : - -$zerocount: - nop # E : - nop # E : - nop # E : - ret # L0 : - - END(strncat) --- a/ports/sysdeps/alpha/stpncpy.S +++ /dev/null @@ -1,106 +0,0 @@ -/* Copyright (C) 1996-2014 Free Software Foundation, Inc. - This file is part of the GNU C Library. - Contributed by Richard Henderson (rth@tamu.edu) - - The GNU C Library is free software; you can redistribute it and/or - modify it under the terms of the GNU Lesser General Public - License as published by the Free Software Foundation; either - version 2.1 of the License, or (at your option) any later version. - - The GNU C Library is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - Lesser General Public License for more details. - - You should have received a copy of the GNU Lesser General Public - License along with the GNU C Library. If not, see - . */ - -/* Copy no more than COUNT bytes of the null-terminated string from - SRC to DST. If SRC does not cover all of COUNT, the balance is - zeroed. Return the address of the terminating null in DEST, if - any, else DEST + COUNT. */ - -#include - - .set noat - .set noreorder - - .text - -ENTRY(__stpncpy) - ldgp gp, 0(pv) -#ifdef PROF - lda AT, _mcount - jsr AT, (AT), _mcount -#endif - .prologue 1 - - beq a2, $zerocount - jsr t9, __stxncpy # do the work of the copy - - and t8, 0xf0, t3 # binary search for byte offset of the - and t8, 0xcc, t2 # last byte written. - and t8, 0xaa, t1 - andnot a0, 7, v0 - cmovne t3, 4, t3 - cmovne t2, 2, t2 - cmovne t1, 1, t1 - addq v0, t3, v0 - addq t1, t2, t1 - addq v0, t1, v0 - - bne a2, $multiword # do we have full words left? - - .align 3 - zapnot t0, t8, t4 # e0 : was last byte a null? - subq t8, 1, t2 # .. e1 : - addq v0, 1, t5 # e0 : - subq t10, 1, t3 # .. e1 : - or t2, t8, t2 # e0 : clear the bits between the last - or t3, t10, t3 # .. e1 : written byte and the last byte in - andnot t3, t2, t3 # e0 : COUNT - cmovne t4, t5, v0 # .. e1 : if last written wasnt null, inc v0 - zap t0, t3, t0 # e0 : - stq_u t0, 0(a0) # e1 : - ret # .. e1 : - - .align 3 -$multiword: - subq t8, 1, t7 # e0 : clear the final bits in the prev - or t7, t8, t7 # e1 : word - zapnot t0, t7, t0 # e0 : - subq a2, 1, a2 # .. e1 : - stq_u t0, 0(a0) # e0 : - addq a0, 8, a0 # .. e1 : - - beq a2, 1f # e1 : - blbc a2, 0f # e1 : - - stq_u zero, 0(a0) # e0 : zero one word - subq a2, 1, a2 # .. e1 : - addq a0, 8, a0 # e0 : - beq a2, 1f # .. e1 : - -0: stq_u zero, 0(a0) # e0 : zero two words - subq a2, 2, a2 # .. e1 : - stq_u zero, 8(a0) # e0 : - addq a0, 16, a0 # .. e1 : - bne a2, 0b # e1 : - unop - -1: ldq_u t0, 0(a0) # e0 : clear the leading bits in the final - subq t10, 1, t7 # .. e1 : word - or t7, t10, t7 # e0 : - zap t0, t7, t0 # e1 (stall) - stq_u t0, 0(a0) # e0 : - ret # .. e1 : - -$zerocount: - mov a0, v0 - ret - - END(__stpncpy) - -libc_hidden_def (__stpncpy) -weak_alias (__stpncpy, stpncpy) --- a/ports/sysdeps/alpha/alphaev67/stpncpy.S +++ /dev/null @@ -1,115 +0,0 @@ -/* Copyright (C) 2000-2014 Free Software Foundation, Inc. - Contributed by Richard Henderson (rth@redhat.com) - This file is part of the GNU C Library. - - The GNU C Library is free software; you can redistribute it and/or - modify it under the terms of the GNU Lesser General Public - License as published by the Free Software Foundation; either - version 2.1 of the License, or (at your option) any later version. - - The GNU C Library is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - Lesser General Public License for more details. - - You should have received a copy of the GNU Lesser General Public - License along with the GNU C Library. If not, see - . */ - -/* Copy no more than N bytes from SRC to DEST, returning the address of - the terminating '\0' in DEST. */ - -#include - - .arch ev6 - .set noat - .set noreorder - .text - -ENTRY(__stpncpy) - ldgp gp, 0(pv) -#ifdef PROF - lda AT, _mcount - jsr AT, (AT), _mcount -#endif - .prologue 1 - - mov a0, v0 - beq a2, $zerocount - - .align 4 - nop - nop - jsr t9, __stxncpy # do the work of the copy - - cttz t8, t4 - zapnot t0, t8, t5 - andnot a0, 7, a0 - bne a2, $multiword # do we have full words left? - - subq t8, 1, t2 - subq t10, 1, t3 - cmpult zero, t5, t5 - addq a0, t4, v0 - - or t2, t8, t2 - or t3, t10, t3 - addq v0, t5, v0 - andnot t3, t2, t3 - - zap t0, t3, t0 - nop - stq t0, 0(a0) - ret - -$multiword: - subq t8, 1, t7 # clear the final bits in the prev word - cmpult zero, t5, t5 - or t7, t8, t7 - zapnot t0, t7, t0 - - subq a2, 1, a2 - stq t0, 0(a0) - addq a0, 8, a1 - beq a2, 1f # loop over full words remaining - - nop - nop - nop - blbc a2, 0f - - stq zero, 0(a1) - subq a2, 1, a2 - addq a1, 8, a1 - beq a2, 1f - -0: stq zero, 0(a1) - subq a2, 2, a2 - nop - nop - - stq zero, 8(a1) - addq a1, 16, a1 - nop - bne a2, 0b - -1: ldq t0, 0(a1) # clear the leading bits in the final word - subq t10, 1, t7 - addq a0, t4, v0 - nop - - or t7, t10, t7 - addq v0, t5, v0 - zap t0, t7, t0 - stq t0, 0(a1) - -$zerocount: - nop - nop - nop - ret - - END(__stpncpy) - -libc_hidden_def (__stpncpy) -weak_alias (__stpncpy, stpncpy) --- a/ports/sysdeps/alpha/strcmp.S +++ /dev/null @@ -1,194 +0,0 @@ -/* Copyright (C) 1996-2014 Free Software Foundation, Inc. - Contributed by Richard Henderson (rth@tamu.edu) - This file is part of the GNU C Library. - - The GNU C Library is free software; you can redistribute it and/or - modify it under the terms of the GNU Lesser General Public - License as published by the Free Software Foundation; either - version 2.1 of the License, or (at your option) any later version. - - The GNU C Library is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - Lesser General Public License for more details. - - You should have received a copy of the GNU Lesser General Public - License along with the GNU C Library. If not, see - . */ - -/* Bytewise compare two null-terminated strings. */ - -#include - - .set noat - .set noreorder - - .text - -ENTRY(strcmp) -#ifdef PROF - ldgp gp, 0(pv) - lda AT, _mcount - jmp AT, (AT), _mcount - .prologue 1 -#else - .prologue 0 -#endif - - ldq_u t0, 0(a0) # e0 : give cache time to catch up - xor a0, a1, t2 # .. e1 : are s1 and s2 co-aligned? - ldq_u t1, 0(a1) # e0 : - and t2, 7, t2 # .. e1 : - lda t3, -1 # e0 : - bne t2, $unaligned # .. e1 : - - /* On entry to this basic block: - t0 == the first destination word for masking back in - t1 == the first source word. - t3 == -1. */ - -$aligned: - mskqh t3, a0, t3 # e0 : - nop # .. e1 : - ornot t1, t3, t1 # e0 : - ornot t0, t3, t0 # .. e1 : - cmpbge zero, t1, t7 # e0 : bits set iff null found - bne t7, $eos # e1 (zdb) - - /* Aligned compare main loop. - On entry to this basic block: - t0 == an s1 word. - t1 == an s2 word not containing a null. */ - -$a_loop: - xor t0, t1, t2 # e0 : - bne t2, $wordcmp # .. e1 (zdb) - ldq_u t1, 8(a1) # e0 : - ldq_u t0, 8(a0) # .. e1 : - addq a1, 8, a1 # e0 : - addq a0, 8, a0 # .. e1 : - cmpbge zero, t1, t7 # e0 : - beq t7, $a_loop # .. e1 (zdb) - br $eos # e1 : - - /* The two strings are not co-aligned. Align s1 and cope. */ - -$unaligned: - and a0, 7, t4 # e0 : find s1 misalignment - and a1, 7, t5 # .. e1 : find s2 misalignment - subq a1, t4, a1 # e0 : - - /* If s2 misalignment is larger than s2 misalignment, we need - extra startup checks to avoid SEGV. */ - - cmplt t4, t5, t8 # .. e1 : - beq t8, $u_head # e1 : - - mskqh t3, t5, t3 # e0 : - ornot t1, t3, t3 # e0 : - cmpbge zero, t3, t7 # e1 : is there a zero? - beq t7, $u_head # e1 : - - /* We've found a zero in the first partial word of s2. Align - our current s1 and s2 words and compare what we've got. */ - - extql t1, t5, t1 # e0 : - extql t0, a0, t0 # e0 : - cmpbge zero, t1, t7 # .. e1 : find that zero again - br $eos # e1 : and finish up - - .align 3 -$u_head: - /* We know just enough now to be able to assemble the first - full word of s2. We can still find a zero at the end of it. - - On entry to this basic block: - t0 == first word of s1 - t1 == first partial word of s2. */ - - ldq_u t2, 8(a1) # e0 : load second partial s2 word - lda t3, -1 # .. e1 : create leading garbage mask - extql t1, a1, t1 # e0 : create first s2 word - mskqh t3, a0, t3 # e0 : - extqh t2, a1, t4 # e0 : - ornot t0, t3, t0 # .. e1 : kill s1 garbage - or t1, t4, t1 # e0 : s2 word now complete - cmpbge zero, t0, t7 # .. e1 : find zero in first s1 word - ornot t1, t3, t1 # e0 : kill s2 garbage - lda t3, -1 # .. e1 : - mskql t3, a1, t3 # e0 : mask for s2[1] bits we have seen - bne t7, $eos # .. e1 : - xor t0, t1, t4 # e0 : compare aligned words - bne t4, $wordcmp # .. e1 (zdb) - or t2, t3, t3 # e0 : - cmpbge zero, t3, t7 # e1 : - bne t7, $u_final # e1 : - - /* Unaligned copy main loop. In order to avoid reading too much, - the loop is structured to detect zeros in aligned words from s2. - This has, unfortunately, effectively pulled half of a loop - iteration out into the head and half into the tail, but it does - prevent nastiness from accumulating in the very thing we want - to run as fast as possible. - - On entry to this basic block: - t2 == the unshifted low-bits from the next s2 word. */ - - .align 3 -$u_loop: - extql t2, a1, t3 # e0 : - ldq_u t2, 16(a1) # .. e1 : load next s2 high bits - ldq_u t0, 8(a0) # e0 : load next s1 word - addq a1, 8, a1 # .. e1 : - addq a0, 8, a0 # e0 : - nop # .. e1 : - extqh t2, a1, t1 # e0 : - cmpbge zero, t0, t7 # .. e1 : find zero in current s1 word - or t1, t3, t1 # e0 : - bne t7, $eos # .. e1 : - xor t0, t1, t4 # e0 : compare the words - bne t4, $wordcmp # .. e1 (zdb) - cmpbge zero, t2, t4 # e0 : find zero in next low bits - beq t4, $u_loop # .. e1 (zdb) - - /* We've found a zero in the low bits of the last s2 word. Get - the next s1 word and align them. */ -$u_final: - ldq_u t0, 8(a0) # e1 : - extql t2, a1, t1 # .. e0 : - cmpbge zero, t1, t7 # e0 : - - /* We've found a zero somewhere in a word we just read. - On entry to this basic block: - t0 == s1 word - t1 == s2 word - t7 == cmpbge mask containing the zero. */ - - .align 3 -$eos: - negq t7, t6 # e0 : create bytemask of valid data - and t6, t7, t8 # e1 : - subq t8, 1, t6 # e0 : - or t6, t8, t7 # e1 : - zapnot t0, t7, t0 # e0 : kill the garbage - zapnot t1, t7, t1 # .. e1 : - xor t0, t1, v0 # e0 : and compare - beq v0, $done # .. e1 : - - /* Here we have two differing co-aligned words in t0 & t1. - Bytewise compare them and return (t0 > t1 ? 1 : -1). */ -$wordcmp: - cmpbge t0, t1, t2 # e0 : comparison yields bit mask of ge - cmpbge t1, t0, t3 # .. e1 : - xor t2, t3, t0 # e0 : bits set iff t0/t1 bytes differ - negq t0, t1 # e1 : clear all but least bit - and t0, t1, t0 # e0 : - lda v0, -1 # .. e1 : - and t0, t2, t1 # e0 : was bit set in t0 > t1? - cmovne t1, 1, v0 # .. e1 (zdb) - -$done: - ret # e1 : - - END(strcmp) -libc_hidden_builtin_def (strcmp) --- a/ports/sysdeps/alpha/strncmp.S +++ /dev/null @@ -1,277 +0,0 @@ -/* Copyright (C) 1996-2014 Free Software Foundation, Inc. - Contributed by Richard Henderson (rth@tamu.edu) - This file is part of the GNU C Library. - - The GNU C Library is free software; you can redistribute it and/or - modify it under the terms of the GNU Lesser General Public - License as published by the Free Software Foundation; either - version 2.1 of the License, or (at your option) any later version. - - The GNU C Library is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - Lesser General Public License for more details. - - You should have received a copy of the GNU Lesser General Public - License along with the GNU C Library. If not, see - . */ - -/* Bytewise compare two null-terminated strings of length no longer than N. */ - -#include - - .set noat - .set noreorder - -/* EV6 only predicts one branch per octaword. We'll use these to push - subsequent branches back to the next bundle. This will generally add - a fetch+decode cycle to older machines, so skip in that case. */ -#ifdef __alpha_fix__ -# define ev6_unop unop -#else -# define ev6_unop -#endif - - .text - -ENTRY(strncmp) -#ifdef PROF - ldgp gp, 0(pv) - lda AT, _mcount - jsr AT, (AT), _mcount - .prologue 1 -#else - .prologue 0 -#endif - - xor a0, a1, t2 # are s1 and s2 co-aligned? - beq a2, $zerolength - ldq_u t0, 0(a0) # load asap to give cache time to catch up - ldq_u t1, 0(a1) - lda t3, -1 - and t2, 7, t2 - srl t3, 1, t6 - and a0, 7, t4 # find s1 misalignment - and a1, 7, t5 # find s2 misalignment - cmovlt a2, t6, a2 # bound neg count to LONG_MAX - addq a1, a2, a3 # s2+count - addq a2, t4, a2 # bias count by s1 misalignment - and a2, 7, t10 # ofs of last byte in s1 last word - srl a2, 3, a2 # remaining full words in s1 count - bne t2, $unaligned - - /* On entry to this basic block: - t0 == the first word of s1. - t1 == the first word of s2. - t3 == -1. */ -$aligned: - mskqh t3, a1, t8 # mask off leading garbage - ornot t1, t8, t1 - ornot t0, t8, t0 - cmpbge zero, t1, t7 # bits set iff null found - beq a2, $eoc # check end of count - bne t7, $eos - beq t10, $ant_loop - - /* Aligned compare main loop. - On entry to this basic block: - t0 == an s1 word. - t1 == an s2 word not containing a null. */ - - .align 4 -$a_loop: - xor t0, t1, t2 # e0 : - bne t2, $wordcmp # .. e1 (zdb) - ldq_u t1, 8(a1) # e0 : - ldq_u t0, 8(a0) # .. e1 : - - subq a2, 1, a2 # e0 : - addq a1, 8, a1 # .. e1 : - addq a0, 8, a0 # e0 : - beq a2, $eoc # .. e1 : - - cmpbge zero, t1, t7 # e0 : - beq t7, $a_loop # .. e1 : - - br $eos - - /* Alternate aligned compare loop, for when there's no trailing - bytes on the count. We have to avoid reading too much data. */ - .align 4 -$ant_loop: - xor t0, t1, t2 # e0 : - ev6_unop - ev6_unop - bne t2, $wordcmp # .. e1 (zdb) - - subq a2, 1, a2 # e0 : - beq a2, $zerolength # .. e1 : - ldq_u t1, 8(a1) # e0 : - ldq_u t0, 8(a0) # .. e1 : - - addq a1, 8, a1 # e0 : - addq a0, 8, a0 # .. e1 : - cmpbge zero, t1, t7 # e0 : - beq t7, $ant_loop # .. e1 : - - br $eos - - /* The two strings are not co-aligned. Align s1 and cope. */ - /* On entry to this basic block: - t0 == the first word of s1. - t1 == the first word of s2. - t3 == -1. - t4 == misalignment of s1. - t5 == misalignment of s2. - t10 == misalignment of s1 end. */ - .align 4 -$unaligned: - /* If s1 misalignment is larger than s2 misalignment, we need - extra startup checks to avoid SEGV. */ - subq a1, t4, a1 # adjust s2 for s1 misalignment - cmpult t4, t5, t9 - subq a3, 1, a3 # last byte of s2 - bic a1, 7, t8 - mskqh t3, t5, t7 # mask garbage in s2 - subq a3, t8, a3 - ornot t1, t7, t7 - srl a3, 3, a3 # remaining full words in s2 count - beq t9, $u_head - - /* Failing that, we need to look for both eos and eoc within the - first word of s2. If we find either, we can continue by - pretending that the next word of s2 is all zeros. */ - lda t2, 0 # next = zero - cmpeq a3, 0, t8 # eoc in the first word of s2? - cmpbge zero, t7, t7 # eos in the first word of s2? - or t7, t8, t8 - bne t8, $u_head_nl - - /* We know just enough now to be able to assemble the first - full word of s2. We can still find a zero at the end of it. - - On entry to this basic block: - t0 == first word of s1 - t1 == first partial word of s2. - t3 == -1. - t10 == ofs of last byte in s1 last word. - t11 == ofs of last byte in s2 last word. */ -$u_head: - ldq_u t2, 8(a1) # load second partial s2 word - subq a3, 1, a3 -$u_head_nl: - extql t1, a1, t1 # create first s2 word - mskqh t3, a0, t8 - extqh t2, a1, t4 - ornot t0, t8, t0 # kill s1 garbage - or t1, t4, t1 # s2 word now complete - cmpbge zero, t0, t7 # find eos in first s1 word - ornot t1, t8, t1 # kill s2 garbage - beq a2, $eoc - subq a2, 1, a2 - bne t7, $eos - mskql t3, a1, t8 # mask out s2[1] bits we have seen - xor t0, t1, t4 # compare aligned words - or t2, t8, t8 - bne t4, $wordcmp - cmpbge zero, t8, t7 # eos in high bits of s2[1]? - cmpeq a3, 0, t8 # eoc in s2[1]? - or t7, t8, t7 - bne t7, $u_final - - /* Unaligned copy main loop. In order to avoid reading too much, - the loop is structured to detect zeros in aligned words from s2. - This has, unfortunately, effectively pulled half of a loop - iteration out into the head and half into the tail, but it does - prevent nastiness from accumulating in the very thing we want - to run as fast as possible. - - On entry to this basic block: - t2 == the unshifted low-bits from the next s2 word. - t10 == ofs of last byte in s1 last word. - t11 == ofs of last byte in s2 last word. */ - .align 4 -$u_loop: - extql t2, a1, t3 # e0 : - ldq_u t2, 16(a1) # .. e1 : load next s2 high bits - ldq_u t0, 8(a0) # e0 : load next s1 word - addq a1, 8, a1 # .. e1 : - - addq a0, 8, a0 # e0 : - subq a3, 1, a3 # .. e1 : - extqh t2, a1, t1 # e0 : - cmpbge zero, t0, t7 # .. e1 : eos in current s1 word - - or t1, t3, t1 # e0 : - beq a2, $eoc # .. e1 : eoc in current s1 word - subq a2, 1, a2 # e0 : - cmpbge zero, t2, t4 # .. e1 : eos in s2[1] - - xor t0, t1, t3 # e0 : compare the words - ev6_unop - ev6_unop - bne t7, $eos # .. e1 : - - cmpeq a3, 0, t5 # e0 : eoc in s2[1] - ev6_unop - ev6_unop - bne t3, $wordcmp # .. e1 : - - or t4, t5, t4 # e0 : eos or eoc in s2[1]. - beq t4, $u_loop # .. e1 (zdb) - - /* We've found a zero in the low bits of the last s2 word. Get - the next s1 word and align them. */ - .align 3 -$u_final: - ldq_u t0, 8(a0) - extql t2, a1, t1 - cmpbge zero, t1, t7 - bne a2, $eos - - /* We've hit end of count. Zero everything after the count - and compare whats left. */ - .align 3 -$eoc: - mskql t0, t10, t0 - mskql t1, t10, t1 - cmpbge zero, t1, t7 - - /* We've found a zero somewhere in a word we just read. - On entry to this basic block: - t0 == s1 word - t1 == s2 word - t7 == cmpbge mask containing the zero. */ - .align 3 -$eos: - negq t7, t6 # create bytemask of valid data - and t6, t7, t8 - subq t8, 1, t6 - or t6, t8, t7 - zapnot t0, t7, t0 # kill the garbage - zapnot t1, t7, t1 - xor t0, t1, v0 # ... and compare - beq v0, $done - - /* Here we have two differing co-aligned words in t0 & t1. - Bytewise compare them and return (t0 > t1 ? 1 : -1). */ - .align 3 -$wordcmp: - cmpbge t0, t1, t2 # comparison yields bit mask of ge - cmpbge t1, t0, t3 - xor t2, t3, t0 # bits set iff t0/t1 bytes differ - negq t0, t1 # clear all but least bit - and t0, t1, t0 - lda v0, -1 - and t0, t2, t1 # was bit set in t0 > t1? - cmovne t1, 1, v0 -$done: - ret - - .align 3 -$zerolength: - clr v0 - ret - - END(strncmp) -libc_hidden_builtin_def (strncmp)