diff options
Diffstat (limited to 'arch/sparc/lib')
32 files changed, 7186 insertions, 0 deletions
diff --git a/arch/sparc/lib/COPYING.LIB b/arch/sparc/lib/COPYING.LIB new file mode 100644 index 00000000000..eb685a5ec98 --- /dev/null +++ b/arch/sparc/lib/COPYING.LIB @@ -0,0 +1,481 @@ + GNU LIBRARY GENERAL PUBLIC LICENSE + Version 2, June 1991 + + Copyright (C) 1991 Free Software Foundation, Inc. + 675 Mass Ave, Cambridge, MA 02139, USA + Everyone is permitted to copy and distribute verbatim copies + of this license document, but changing it is not allowed. + +[This is the first released version of the library GPL. It is + numbered 2 because it goes with version 2 of the ordinary GPL.] + + Preamble + + The licenses for most software are designed to take away your +freedom to share and change it. By contrast, the GNU General Public +Licenses are intended to guarantee your freedom to share and change +free software--to make sure the software is free for all its users. + + This license, the Library General Public License, applies to some +specially designated Free Software Foundation software, and to any +other libraries whose authors decide to use it. You can use it for +your libraries, too. + + When we speak of free software, we are referring to freedom, not +price. Our General Public Licenses are designed to make sure that you +have the freedom to distribute copies of free software (and charge for +this service if you wish), that you receive source code or can get it +if you want it, that you can change the software or use pieces of it +in new free programs; and that you know you can do these things. + + To protect your rights, we need to make restrictions that forbid +anyone to deny you these rights or to ask you to surrender the rights. +These restrictions translate to certain responsibilities for you if +you distribute copies of the library, or if you modify it. + + For example, if you distribute copies of the library, whether gratis +or for a fee, you must give the recipients all the rights that we gave +you. You must make sure that they, too, receive or can get the source +code. If you link a program with the library, you must provide +complete object files to the recipients so that they can relink them +with the library, after making changes to the library and recompiling +it. And you must show them these terms so they know their rights. + + Our method of protecting your rights has two steps: (1) copyright +the library, and (2) offer you this license which gives you legal +permission to copy, distribute and/or modify the library. + + Also, for each distributor's protection, we want to make certain +that everyone understands that there is no warranty for this free +library. If the library is modified by someone else and passed on, we +want its recipients to know that what they have is not the original +version, so that any problems introduced by others will not reflect on +the original authors' reputations. + + Finally, any free program is threatened constantly by software +patents. We wish to avoid the danger that companies distributing free +software will individually obtain patent licenses, thus in effect +transforming the program into proprietary software. To prevent this, +we have made it clear that any patent must be licensed for everyone's +free use or not licensed at all. + + Most GNU software, including some libraries, is covered by the ordinary +GNU General Public License, which was designed for utility programs. This +license, the GNU Library General Public License, applies to certain +designated libraries. This license is quite different from the ordinary +one; be sure to read it in full, and don't assume that anything in it is +the same as in the ordinary license. + + The reason we have a separate public license for some libraries is that +they blur the distinction we usually make between modifying or adding to a +program and simply using it. Linking a program with a library, without +changing the library, is in some sense simply using the library, and is +analogous to running a utility program or application program. However, in +a textual and legal sense, the linked executable is a combined work, a +derivative of the original library, and the ordinary General Public License +treats it as such. + + Because of this blurred distinction, using the ordinary General +Public License for libraries did not effectively promote software +sharing, because most developers did not use the libraries. We +concluded that weaker conditions might promote sharing better. + + However, unrestricted linking of non-free programs would deprive the +users of those programs of all benefit from the free status of the +libraries themselves. This Library General Public License is intended to +permit developers of non-free programs to use free libraries, while +preserving your freedom as a user of such programs to change the free +libraries that are incorporated in them. (We have not seen how to achieve +this as regards changes in header files, but we have achieved it as regards +changes in the actual functions of the Library.) The hope is that this +will lead to faster development of free libraries. + + The precise terms and conditions for copying, distribution and +modification follow. Pay close attention to the difference between a +"work based on the library" and a "work that uses the library". The +former contains code derived from the library, while the latter only +works together with the library. + + Note that it is possible for a library to be covered by the ordinary +General Public License rather than by this special one. + + GNU LIBRARY GENERAL PUBLIC LICENSE + TERMS AND CONDITIONS FOR COPYING, DISTRIBUTION AND MODIFICATION + + 0. This License Agreement applies to any software library which +contains a notice placed by the copyright holder or other authorized +party saying it may be distributed under the terms of this Library +General Public License (also called "this License"). Each licensee is +addressed as "you". + + A "library" means a collection of software functions and/or data +prepared so as to be conveniently linked with application programs +(which use some of those functions and data) to form executables. + + The "Library", below, refers to any such software library or work +which has been distributed under these terms. A "work based on the +Library" means either the Library or any derivative work under +copyright law: that is to say, a work containing the Library or a +portion of it, either verbatim or with modifications and/or translated +straightforwardly into another language. (Hereinafter, translation is +included without limitation in the term "modification".) + + "Source code" for a work means the preferred form of the work for +making modifications to it. For a library, complete source code means +all the source code for all modules it contains, plus any associated +interface definition files, plus the scripts used to control compilation +and installation of the library. + + Activities other than copying, distribution and modification are not +covered by this License; they are outside its scope. The act of +running a program using the Library is not restricted, and output from +such a program is covered only if its contents constitute a work based +on the Library (independent of the use of the Library in a tool for +writing it). Whether that is true depends on what the Library does +and what the program that uses the Library does. + + 1. You may copy and distribute verbatim copies of the Library's +complete source code as you receive it, in any medium, provided that +you conspicuously and appropriately publish on each copy an +appropriate copyright notice and disclaimer of warranty; keep intact +all the notices that refer to this License and to the absence of any +warranty; and distribute a copy of this License along with the +Library. + + You may charge a fee for the physical act of transferring a copy, +and you may at your option offer warranty protection in exchange for a +fee. + + 2. You may modify your copy or copies of the Library or any portion +of it, thus forming a work based on the Library, and copy and +distribute such modifications or work under the terms of Section 1 +above, provided that you also meet all of these conditions: + + a) The modified work must itself be a software library. + + b) You must cause the files modified to carry prominent notices + stating that you changed the files and the date of any change. + + c) You must cause the whole of the work to be licensed at no + charge to all third parties under the terms of this License. + + d) If a facility in the modified Library refers to a function or a + table of data to be supplied by an application program that uses + the facility, other than as an argument passed when the facility + is invoked, then you must make a good faith effort to ensure that, + in the event an application does not supply such function or + table, the facility still operates, and performs whatever part of + its purpose remains meaningful. + + (For example, a function in a library to compute square roots has + a purpose that is entirely well-defined independent of the + application. Therefore, Subsection 2d requires that any + application-supplied function or table used by this function must + be optional: if the application does not supply it, the square + root function must still compute square roots.) + +These requirements apply to the modified work as a whole. If +identifiable sections of that work are not derived from the Library, +and can be reasonably considered independent and separate works in +themselves, then this License, and its terms, do not apply to those +sections when you distribute them as separate works. But when you +distribute the same sections as part of a whole which is a work based +on the Library, the distribution of the whole must be on the terms of +this License, whose permissions for other licensees extend to the +entire whole, and thus to each and every part regardless of who wrote +it. + +Thus, it is not the intent of this section to claim rights or contest +your rights to work written entirely by you; rather, the intent is to +exercise the right to control the distribution of derivative or +collective works based on the Library. + +In addition, mere aggregation of another work not based on the Library +with the Library (or with a work based on the Library) on a volume of +a storage or distribution medium does not bring the other work under +the scope of this License. + + 3. You may opt to apply the terms of the ordinary GNU General Public +License instead of this License to a given copy of the Library. To do +this, you must alter all the notices that refer to this License, so +that they refer to the ordinary GNU General Public License, version 2, +instead of to this License. (If a newer version than version 2 of the +ordinary GNU General Public License has appeared, then you can specify +that version instead if you wish.) Do not make any other change in +these notices. + + Once this change is made in a given copy, it is irreversible for +that copy, so the ordinary GNU General Public License applies to all +subsequent copies and derivative works made from that copy. + + This option is useful when you wish to copy part of the code of +the Library into a program that is not a library. + + 4. You may copy and distribute the Library (or a portion or +derivative of it, under Section 2) in object code or executable form +under the terms of Sections 1 and 2 above provided that you accompany +it with the complete corresponding machine-readable source code, which +must be distributed under the terms of Sections 1 and 2 above on a +medium customarily used for software interchange. + + If distribution of object code is made by offering access to copy +from a designated place, then offering equivalent access to copy the +source code from the same place satisfies the requirement to +distribute the source code, even though third parties are not +compelled to copy the source along with the object code. + + 5. A program that contains no derivative of any portion of the +Library, but is designed to work with the Library by being compiled or +linked with it, is called a "work that uses the Library". Such a +work, in isolation, is not a derivative work of the Library, and +therefore falls outside the scope of this License. + + However, linking a "work that uses the Library" with the Library +creates an executable that is a derivative of the Library (because it +contains portions of the Library), rather than a "work that uses the +library". The executable is therefore covered by this License. +Section 6 states terms for distribution of such executables. + + When a "work that uses the Library" uses material from a header file +that is part of the Library, the object code for the work may be a +derivative work of the Library even though the source code is not. +Whether this is true is especially significant if the work can be +linked without the Library, or if the work is itself a library. The +threshold for this to be true is not precisely defined by law. + + If such an object file uses only numerical parameters, data +structure layouts and accessors, and small macros and small inline +functions (ten lines or less in length), then the use of the object +file is unrestricted, regardless of whether it is legally a derivative +work. (Executables containing this object code plus portions of the +Library will still fall under Section 6.) + + Otherwise, if the work is a derivative of the Library, you may +distribute the object code for the work under the terms of Section 6. +Any executables containing that work also fall under Section 6, +whether or not they are linked directly with the Library itself. + + 6. As an exception to the Sections above, you may also compile or +link a "work that uses the Library" with the Library to produce a +work containing portions of the Library, and distribute that work +under terms of your choice, provided that the terms permit +modification of the work for the customer's own use and reverse +engineering for debugging such modifications. + + You must give prominent notice with each copy of the work that the +Library is used in it and that the Library and its use are covered by +this License. You must supply a copy of this License. If the work +during execution displays copyright notices, you must include the +copyright notice for the Library among them, as well as a reference +directing the user to the copy of this License. Also, you must do one +of these things: + + a) Accompany the work with the complete corresponding + machine-readable source code for the Library including whatever + changes were used in the work (which must be distributed under + Sections 1 and 2 above); and, if the work is an executable linked + with the Library, with the complete machine-readable "work that + uses the Library", as object code and/or source code, so that the + user can modify the Library and then relink to produce a modified + executable containing the modified Library. (It is understood + that the user who changes the contents of definitions files in the + Library will not necessarily be able to recompile the application + to use the modified definitions.) + + b) Accompany the work with a written offer, valid for at + least three years, to give the same user the materials + specified in Subsection 6a, above, for a charge no more + than the cost of performing this distribution. + + c) If distribution of the work is made by offering access to copy + from a designated place, offer equivalent access to copy the above + specified materials from the same place. + + d) Verify that the user has already received a copy of these + materials or that you have already sent this user a copy. + + For an executable, the required form of the "work that uses the +Library" must include any data and utility programs needed for +reproducing the executable from it. However, as a special exception, +the source code distributed need not include anything that is normally +distributed (in either source or binary form) with the major +components (compiler, kernel, and so on) of the operating system on +which the executable runs, unless that component itself accompanies +the executable. + + It may happen that this requirement contradicts the license +restrictions of other proprietary libraries that do not normally +accompany the operating system. Such a contradiction means you cannot +use both them and the Library together in an executable that you +distribute. + + 7. You may place library facilities that are a work based on the +Library side-by-side in a single library together with other library +facilities not covered by this License, and distribute such a combined +library, provided that the separate distribution of the work based on +the Library and of the other library facilities is otherwise +permitted, and provided that you do these two things: + + a) Accompany the combined library with a copy of the same work + based on the Library, uncombined with any other library + facilities. This must be distributed under the terms of the + Sections above. + + b) Give prominent notice with the combined library of the fact + that part of it is a work based on the Library, and explaining + where to find the accompanying uncombined form of the same work. + + 8. You may not copy, modify, sublicense, link with, or distribute +the Library except as expressly provided under this License. Any +attempt otherwise to copy, modify, sublicense, link with, or +distribute the Library is void, and will automatically terminate your +rights under this License. However, parties who have received copies, +or rights, from you under this License will not have their licenses +terminated so long as such parties remain in full compliance. + + 9. You are not required to accept this License, since you have not +signed it. However, nothing else grants you permission to modify or +distribute the Library or its derivative works. These actions are +prohibited by law if you do not accept this License. Therefore, by +modifying or distributing the Library (or any work based on the +Library), you indicate your acceptance of this License to do so, and +all its terms and conditions for copying, distributing or modifying +the Library or works based on it. + + 10. Each time you redistribute the Library (or any work based on the +Library), the recipient automatically receives a license from the +original licensor to copy, distribute, link with or modify the Library +subject to these terms and conditions. You may not impose any further +restrictions on the recipients' exercise of the rights granted herein. +You are not responsible for enforcing compliance by third parties to +this License. + + 11. If, as a consequence of a court judgment or allegation of patent +infringement or for any other reason (not limited to patent issues), +conditions are imposed on you (whether by court order, agreement or +otherwise) that contradict the conditions of this License, they do not +excuse you from the conditions of this License. If you cannot +distribute so as to satisfy simultaneously your obligations under this +License and any other pertinent obligations, then as a consequence you +may not distribute the Library at all. For example, if a patent +license would not permit royalty-free redistribution of the Library by +all those who receive copies directly or indirectly through you, then +the only way you could satisfy both it and this License would be to +refrain entirely from distribution of the Library. + +If any portion of this section is held invalid or unenforceable under any +particular circumstance, the balance of the section is intended to apply, +and the section as a whole is intended to apply in other circumstances. + +It is not the purpose of this section to induce you to infringe any +patents or other property right claims or to contest validity of any +such claims; this section has the sole purpose of protecting the +integrity of the free software distribution system which is +implemented by public license practices. Many people have made +generous contributions to the wide range of software distributed +through that system in reliance on consistent application of that +system; it is up to the author/donor to decide if he or she is willing +to distribute software through any other system and a licensee cannot +impose that choice. + +This section is intended to make thoroughly clear what is believed to +be a consequence of the rest of this License. + + 12. If the distribution and/or use of the Library is restricted in +certain countries either by patents or by copyrighted interfaces, the +original copyright holder who places the Library under this License may add +an explicit geographical distribution limitation excluding those countries, +so that distribution is permitted only in or among countries not thus +excluded. In such case, this License incorporates the limitation as if +written in the body of this License. + + 13. The Free Software Foundation may publish revised and/or new +versions of the Library General Public License from time to time. +Such new versions will be similar in spirit to the present version, +but may differ in detail to address new problems or concerns. + +Each version is given a distinguishing version number. If the Library +specifies a version number of this License which applies to it and +"any later version", you have the option of following the terms and +conditions either of that version or of any later version published by +the Free Software Foundation. If the Library does not specify a +license version number, you may choose any version ever published by +the Free Software Foundation. + + 14. If you wish to incorporate parts of the Library into other free +programs whose distribution conditions are incompatible with these, +write to the author to ask for permission. For software which is +copyrighted by the Free Software Foundation, write to the Free +Software Foundation; we sometimes make exceptions for this. Our +decision will be guided by the two goals of preserving the free status +of all derivatives of our free software and of promoting the sharing +and reuse of software generally. + + NO WARRANTY + + 15. BECAUSE THE LIBRARY IS LICENSED FREE OF CHARGE, THERE IS NO +WARRANTY FOR THE LIBRARY, TO THE EXTENT PERMITTED BY APPLICABLE LAW. +EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT HOLDERS AND/OR +OTHER PARTIES PROVIDE THE LIBRARY "AS IS" WITHOUT WARRANTY OF ANY +KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR +PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE +LIBRARY IS WITH YOU. SHOULD THE LIBRARY PROVE DEFECTIVE, YOU ASSUME +THE COST OF ALL NECESSARY SERVICING, REPAIR OR CORRECTION. + + 16. IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN +WRITING WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MAY MODIFY +AND/OR REDISTRIBUTE THE LIBRARY AS PERMITTED ABOVE, BE LIABLE TO YOU +FOR DAMAGES, INCLUDING ANY GENERAL, SPECIAL, INCIDENTAL OR +CONSEQUENTIAL DAMAGES ARISING OUT OF THE USE OR INABILITY TO USE THE +LIBRARY (INCLUDING BUT NOT LIMITED TO LOSS OF DATA OR DATA BEING +RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD PARTIES OR A +FAILURE OF THE LIBRARY TO OPERATE WITH ANY OTHER SOFTWARE), EVEN IF +SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH +DAMAGES. + + END OF TERMS AND CONDITIONS + + Appendix: How to Apply These Terms to Your New Libraries + + If you develop a new library, and you want it to be of the greatest +possible use to the public, we recommend making it free software that +everyone can redistribute and change. You can do so by permitting +redistribution under these terms (or, alternatively, under the terms of the +ordinary General Public License). + + To apply these terms, attach the following notices to the library. It is +safest to attach them to the start of each source file to most effectively +convey the exclusion of warranty; and each file should have at least the +"copyright" line and a pointer to where the full notice is found. + + <one line to give the library's name and a brief idea of what it does.> + Copyright (C) <year> <name of author> + + This library is free software; you can redistribute it and/or + modify it under the terms of the GNU Library General Public + License as published by the Free Software Foundation; either + version 2 of the License, or (at your option) any later version. + + This 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 + Library General Public License for more details. + + You should have received a copy of the GNU Library General Public + License along with this library; if not, write to the Free + Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. + +Also add information on how to contact you by electronic and paper mail. + +You should also get your employer (if you work as a programmer) or your +school, if any, to sign a "copyright disclaimer" for the library, if +necessary. Here is a sample; alter the names: + + Yoyodyne, Inc., hereby disclaims all copyright interest in the + library `Frob' (a library for tweaking knobs) written by James Random Hacker. + + <signature of Ty Coon>, 1 April 1990 + Ty Coon, President of Vice + +That's all there is to it! diff --git a/arch/sparc/lib/Makefile b/arch/sparc/lib/Makefile new file mode 100644 index 00000000000..2296ff9dc47 --- /dev/null +++ b/arch/sparc/lib/Makefile @@ -0,0 +1,13 @@ +# $Id: Makefile,v 1.35 2000/12/15 00:41:18 davem Exp $ +# Makefile for Sparc library files.. +# + +EXTRA_AFLAGS := -ansi -DST_DIV0=0x02 + +lib-y := mul.o rem.o sdiv.o udiv.o umul.o urem.o ashrdi3.o memcpy.o memset.o \ + strlen.o checksum.o blockops.o memscan.o memcmp.o strncmp.o \ + strncpy_from_user.o divdi3.o udivdi3.o strlen_user.o \ + copy_user.o locks.o atomic.o atomic32.o bitops.o \ + lshrdi3.o ashldi3.o rwsem.o muldi3.o bitext.o + +lib-$(CONFIG_DEBUG_SPINLOCK) += debuglocks.o diff --git a/arch/sparc/lib/ashldi3.S b/arch/sparc/lib/ashldi3.S new file mode 100644 index 00000000000..52418a0cb3d --- /dev/null +++ b/arch/sparc/lib/ashldi3.S @@ -0,0 +1,34 @@ +/* $Id: ashldi3.S,v 1.2 1999/11/19 04:11:46 davem Exp $ + * ashldi3.S: GCC emits these for certain drivers playing + * with long longs. + * + * Copyright (C) 1999 David S. Miller (davem@redhat.com) + */ + + .text + .align 4 + .globl __ashldi3 +__ashldi3: + cmp %o2, 0 + be 9f + mov 0x20, %g2 + + sub %g2, %o2, %g2 + cmp %g2, 0 + bg 7f + sll %o0, %o2, %g3 + + neg %g2 + clr %o5 + b 8f + sll %o1, %g2, %o4 +7: + srl %o1, %g2, %g2 + sll %o1, %o2, %o5 + or %g3, %g2, %o4 +8: + mov %o4, %o0 + mov %o5, %o1 +9: + retl + nop diff --git a/arch/sparc/lib/ashrdi3.S b/arch/sparc/lib/ashrdi3.S new file mode 100644 index 00000000000..2848237598a --- /dev/null +++ b/arch/sparc/lib/ashrdi3.S @@ -0,0 +1,36 @@ +/* $Id: ashrdi3.S,v 1.4 1999/11/19 04:11:49 davem Exp $ + * ashrdi3.S: The filesystem code creates all kinds of references to + * this little routine on the sparc with gcc. + * + * Copyright (C) 1995 David S. Miller (davem@caip.rutgers.edu) + */ + + .text + .align 4 + .globl __ashrdi3 +__ashrdi3: + tst %o2 + be 3f + or %g0, 32, %g2 + + sub %g2, %o2, %g2 + + tst %g2 + bg 1f + sra %o0, %o2, %o4 + + sra %o0, 31, %o4 + sub %g0, %g2, %g2 + ba 2f + sra %o0, %g2, %o5 + +1: + sll %o0, %g2, %g3 + srl %o1, %o2, %g2 + or %g2, %g3, %o5 +2: + or %g0, %o4, %o0 + or %g0, %o5, %o1 +3: + jmpl %o7 + 8, %g0 + nop diff --git a/arch/sparc/lib/atomic.S b/arch/sparc/lib/atomic.S new file mode 100644 index 00000000000..f48ad0c4dad --- /dev/null +++ b/arch/sparc/lib/atomic.S @@ -0,0 +1,100 @@ +/* atomic.S: Move this stuff here for better ICACHE hit rates. + * + * Copyright (C) 1996 David S. Miller (davem@caipfs.rutgers.edu) + */ + +#include <linux/config.h> +#include <asm/ptrace.h> +#include <asm/psr.h> + + .text + .align 4 + + .globl __atomic_begin +__atomic_begin: + +#ifndef CONFIG_SMP + .globl ___xchg32_sun4c +___xchg32_sun4c: + rd %psr, %g3 + andcc %g3, PSR_PIL, %g0 + bne 1f + nop + wr %g3, PSR_PIL, %psr + nop; nop; nop +1: + andcc %g3, PSR_PIL, %g0 + ld [%g1], %g7 + bne 1f + st %g2, [%g1] + wr %g3, 0x0, %psr + nop; nop; nop +1: + mov %g7, %g2 + jmpl %o7 + 8, %g0 + mov %g4, %o7 + + .globl ___xchg32_sun4md +___xchg32_sun4md: + swap [%g1], %g2 + jmpl %o7 + 8, %g0 + mov %g4, %o7 +#endif + + /* Read asm-sparc/atomic.h carefully to understand how this works for SMP. + * Really, some things here for SMP are overly clever, go read the header. + */ + .globl ___atomic24_add +___atomic24_add: + rd %psr, %g3 ! Keep the code small, old way was stupid + nop; nop; nop; ! Let the bits set + or %g3, PSR_PIL, %g7 ! Disable interrupts + wr %g7, 0x0, %psr ! Set %psr + nop; nop; nop; ! Let the bits set +#ifdef CONFIG_SMP +1: ldstub [%g1 + 3], %g7 ! Spin on the byte lock for SMP. + orcc %g7, 0x0, %g0 ! Did we get it? + bne 1b ! Nope... + ld [%g1], %g7 ! Load locked atomic24_t + sra %g7, 8, %g7 ! Get signed 24-bit integer + add %g7, %g2, %g2 ! Add in argument + sll %g2, 8, %g7 ! Transpose back to atomic24_t + st %g7, [%g1] ! Clever: This releases the lock as well. +#else + ld [%g1], %g7 ! Load locked atomic24_t + add %g7, %g2, %g2 ! Add in argument + st %g2, [%g1] ! Store it back +#endif + wr %g3, 0x0, %psr ! Restore original PSR_PIL + nop; nop; nop; ! Let the bits set + jmpl %o7, %g0 ! NOTE: not + 8, see callers in atomic.h + mov %g4, %o7 ! Restore %o7 + + .globl ___atomic24_sub +___atomic24_sub: + rd %psr, %g3 ! Keep the code small, old way was stupid + nop; nop; nop; ! Let the bits set + or %g3, PSR_PIL, %g7 ! Disable interrupts + wr %g7, 0x0, %psr ! Set %psr + nop; nop; nop; ! Let the bits set +#ifdef CONFIG_SMP +1: ldstub [%g1 + 3], %g7 ! Spin on the byte lock for SMP. + orcc %g7, 0x0, %g0 ! Did we get it? + bne 1b ! Nope... + ld [%g1], %g7 ! Load locked atomic24_t + sra %g7, 8, %g7 ! Get signed 24-bit integer + sub %g7, %g2, %g2 ! Subtract argument + sll %g2, 8, %g7 ! Transpose back to atomic24_t + st %g7, [%g1] ! Clever: This releases the lock as well +#else + ld [%g1], %g7 ! Load locked atomic24_t + sub %g7, %g2, %g2 ! Subtract argument + st %g2, [%g1] ! Store it back +#endif + wr %g3, 0x0, %psr ! Restore original PSR_PIL + nop; nop; nop; ! Let the bits set + jmpl %o7, %g0 ! NOTE: not + 8, see callers in atomic.h + mov %g4, %o7 ! Restore %o7 + + .globl __atomic_end +__atomic_end: diff --git a/arch/sparc/lib/atomic32.c b/arch/sparc/lib/atomic32.c new file mode 100644 index 00000000000..19724c5800a --- /dev/null +++ b/arch/sparc/lib/atomic32.c @@ -0,0 +1,53 @@ +/* + * atomic32.c: 32-bit atomic_t implementation + * + * Copyright (C) 2004 Keith M Wesolowski + * + * Based on asm-parisc/atomic.h Copyright (C) 2000 Philipp Rumpf + */ + +#include <asm/atomic.h> +#include <linux/spinlock.h> +#include <linux/module.h> + +#ifdef CONFIG_SMP +#define ATOMIC_HASH_SIZE 4 +#define ATOMIC_HASH(a) (&__atomic_hash[(((unsigned long)a)>>8) & (ATOMIC_HASH_SIZE-1)]) + +spinlock_t __atomic_hash[ATOMIC_HASH_SIZE] = { + [0 ... (ATOMIC_HASH_SIZE-1)] = SPIN_LOCK_UNLOCKED +}; + +#else /* SMP */ + +static spinlock_t dummy = SPIN_LOCK_UNLOCKED; +#define ATOMIC_HASH_SIZE 1 +#define ATOMIC_HASH(a) (&dummy) + +#endif /* SMP */ + +int __atomic_add_return(int i, atomic_t *v) +{ + int ret; + unsigned long flags; + spin_lock_irqsave(ATOMIC_HASH(v), flags); + + ret = (v->counter += i); + + spin_unlock_irqrestore(ATOMIC_HASH(v), flags); + return ret; +} + +void atomic_set(atomic_t *v, int i) +{ + unsigned long flags; + spin_lock_irqsave(ATOMIC_HASH(v), flags); + + v->counter = i; + + spin_unlock_irqrestore(ATOMIC_HASH(v), flags); +} + +EXPORT_SYMBOL(__atomic_add_return); +EXPORT_SYMBOL(atomic_set); + diff --git a/arch/sparc/lib/bitext.c b/arch/sparc/lib/bitext.c new file mode 100644 index 00000000000..94b05e8c906 --- /dev/null +++ b/arch/sparc/lib/bitext.c @@ -0,0 +1,132 @@ +/* + * bitext.c: kernel little helper (of bit shuffling variety). + * + * Copyright (C) 2002 Pete Zaitcev <zaitcev@yahoo.com> + * + * The algorithm to search a zero bit string is geared towards its application. + * We expect a couple of fixed sizes of requests, so a rotating counter, reset + * by align size, should provide fast enough search while maintaining low + * fragmentation. + */ + +#include <linux/smp_lock.h> +#include <linux/bitops.h> + +#include <asm/bitext.h> + +/** + * bit_map_string_get - find and set a bit string in bit map. + * @t: the bit map. + * @len: requested string length + * @align: requested alignment + * + * Returns offset in the map or -1 if out of space. + * + * Not safe to call from an interrupt (uses spin_lock). + */ +int bit_map_string_get(struct bit_map *t, int len, int align) +{ + int offset, count; /* siamese twins */ + int off_new; + int align1; + int i, color; + + if (t->num_colors) { + /* align is overloaded to be the page color */ + color = align; + align = t->num_colors; + } else { + color = 0; + if (align == 0) + align = 1; + } + align1 = align - 1; + if ((align & align1) != 0) + BUG(); + if (align < 0 || align >= t->size) + BUG(); + if (len <= 0 || len > t->size) + BUG(); + color &= align1; + + spin_lock(&t->lock); + if (len < t->last_size) + offset = t->first_free; + else + offset = t->last_off & ~align1; + count = 0; + for (;;) { + off_new = find_next_zero_bit(t->map, t->size, offset); + off_new = ((off_new + align1) & ~align1) + color; + count += off_new - offset; + offset = off_new; + if (offset >= t->size) + offset = 0; + if (count + len > t->size) { + spin_unlock(&t->lock); +/* P3 */ printk(KERN_ERR + "bitmap out: size %d used %d off %d len %d align %d count %d\n", + t->size, t->used, offset, len, align, count); + return -1; + } + + if (offset + len > t->size) { + count += t->size - offset; + offset = 0; + continue; + } + + i = 0; + while (test_bit(offset + i, t->map) == 0) { + i++; + if (i == len) { + for (i = 0; i < len; i++) + __set_bit(offset + i, t->map); + if (offset == t->first_free) + t->first_free = find_next_zero_bit + (t->map, t->size, + t->first_free + len); + if ((t->last_off = offset + len) >= t->size) + t->last_off = 0; + t->used += len; + t->last_size = len; + spin_unlock(&t->lock); + return offset; + } + } + count += i + 1; + if ((offset += i + 1) >= t->size) + offset = 0; + } +} + +void bit_map_clear(struct bit_map *t, int offset, int len) +{ + int i; + + if (t->used < len) + BUG(); /* Much too late to do any good, but alas... */ + spin_lock(&t->lock); + for (i = 0; i < len; i++) { + if (test_bit(offset + i, t->map) == 0) + BUG(); + __clear_bit(offset + i, t->map); + } + if (offset < t->first_free) + t->first_free = offset; + t->used -= len; + spin_unlock(&t->lock); +} + +void bit_map_init(struct bit_map *t, unsigned long *map, int size) +{ + + if ((size & 07) != 0) + BUG(); + memset(map, 0, size>>3); + + memset(t, 0, sizeof *t); + spin_lock_init(&t->lock); + t->map = map; + t->size = size; +} diff --git a/arch/sparc/lib/bitops.S b/arch/sparc/lib/bitops.S new file mode 100644 index 00000000000..3e939976907 --- /dev/null +++ b/arch/sparc/lib/bitops.S @@ -0,0 +1,110 @@ +/* bitops.S: Low level assembler bit operations. + * + * Copyright (C) 1996 David S. Miller (davem@caip.rutgers.edu) + */ + +#include <linux/config.h> +#include <asm/ptrace.h> +#include <asm/psr.h> + + .text + .align 4 + + .globl __bitops_begin +__bitops_begin: + + /* Take bits in %g2 and set them in word at %g1, + * return whether bits were set in original value + * in %g2. %g4 holds value to restore into %o7 + * in delay slot of jmpl return, %g3 + %g5 + %g7 can be + * used as temporaries and thus is considered clobbered + * by all callers. + */ + .globl ___set_bit +___set_bit: + rd %psr, %g3 + nop; nop; nop; + or %g3, PSR_PIL, %g5 + wr %g5, 0x0, %psr + nop; nop; nop +#ifdef CONFIG_SMP + set bitops_spinlock, %g5 +2: ldstub [%g5], %g7 ! Spin on the byte lock for SMP. + orcc %g7, 0x0, %g0 ! Did we get it? + bne 2b ! Nope... +#endif + ld [%g1], %g7 + or %g7, %g2, %g5 + and %g7, %g2, %g2 +#ifdef CONFIG_SMP + st %g5, [%g1] + set bitops_spinlock, %g5 + stb %g0, [%g5] +#else + st %g5, [%g1] +#endif + wr %g3, 0x0, %psr + nop; nop; nop + jmpl %o7, %g0 + mov %g4, %o7 + + /* Same as above, but clears the bits from %g2 instead. */ + .globl ___clear_bit +___clear_bit: + rd %psr, %g3 + nop; nop; nop + or %g3, PSR_PIL, %g5 + wr %g5, 0x0, %psr + nop; nop; nop +#ifdef CONFIG_SMP + set bitops_spinlock, %g5 +2: ldstub [%g5], %g7 ! Spin on the byte lock for SMP. + orcc %g7, 0x0, %g0 ! Did we get it? + bne 2b ! Nope... +#endif + ld [%g1], %g7 + andn %g7, %g2, %g5 + and %g7, %g2, %g2 +#ifdef CONFIG_SMP + st %g5, [%g1] + set bitops_spinlock, %g5 + stb %g0, [%g5] +#else + st %g5, [%g1] +#endif + wr %g3, 0x0, %psr + nop; nop; nop + jmpl %o7, %g0 + mov %g4, %o7 + + /* Same thing again, but this time toggles the bits from %g2. */ + .globl ___change_bit +___change_bit: + rd %psr, %g3 + nop; nop; nop + or %g3, PSR_PIL, %g5 + wr %g5, 0x0, %psr + nop; nop; nop +#ifdef CONFIG_SMP + set bitops_spinlock, %g5 +2: ldstub [%g5], %g7 ! Spin on the byte lock for SMP. + orcc %g7, 0x0, %g0 ! Did we get it? + bne 2b ! Nope... +#endif + ld [%g1], %g7 + xor %g7, %g2, %g5 + and %g7, %g2, %g2 +#ifdef CONFIG_SMP + st %g5, [%g1] + set bitops_spinlock, %g5 + stb %g0, [%g5] +#else + st %g5, [%g1] +#endif + wr %g3, 0x0, %psr + nop; nop; nop + jmpl %o7, %g0 + mov %g4, %o7 + + .globl __bitops_end +__bitops_end: diff --git a/arch/sparc/lib/blockops.S b/arch/sparc/lib/blockops.S new file mode 100644 index 00000000000..a7c7ffaa4a9 --- /dev/null +++ b/arch/sparc/lib/blockops.S @@ -0,0 +1,89 @@ +/* $Id: blockops.S,v 1.8 1998/01/30 10:58:44 jj Exp $ + * blockops.S: Common block zero optimized routines. + * + * Copyright (C) 1996 David S. Miller (davem@caip.rutgers.edu) + */ + +#include <asm/page.h> + + /* Zero out 64 bytes of memory at (buf + offset). + * Assumes %g1 contains zero. + */ +#define BLAST_BLOCK(buf, offset) \ + std %g0, [buf + offset + 0x38]; \ + std %g0, [buf + offset + 0x30]; \ + std %g0, [buf + offset + 0x28]; \ + std %g0, [buf + offset + 0x20]; \ + std %g0, [buf + offset + 0x18]; \ + std %g0, [buf + offset + 0x10]; \ + std %g0, [buf + offset + 0x08]; \ + std %g0, [buf + offset + 0x00]; + + /* Copy 32 bytes of memory at (src + offset) to + * (dst + offset). + */ +#define MIRROR_BLOCK(dst, src, offset, t0, t1, t2, t3, t4, t5, t6, t7) \ + ldd [src + offset + 0x18], t0; \ + ldd [src + offset + 0x10], t2; \ + ldd [src + offset + 0x08], t4; \ + ldd [src + offset + 0x00], t6; \ + std t0, [dst + offset + 0x18]; \ + std t2, [dst + offset + 0x10]; \ + std t4, [dst + offset + 0x08]; \ + std t6, [dst + offset + 0x00]; + + /* Profiling evidence indicates that memset() is + * commonly called for blocks of size PAGE_SIZE, + * and (2 * PAGE_SIZE) (for kernel stacks) + * and with a second arg of zero. We assume in + * all of these cases that the buffer is aligned + * on at least an 8 byte boundary. + * + * Therefore we special case them to make them + * as fast as possible. + */ + + .text + .align 4 + .globl bzero_1page, __copy_1page + +bzero_1page: +/* NOTE: If you change the number of insns of this routine, please check + * arch/sparc/mm/hypersparc.S */ + /* %o0 = buf */ + or %g0, %g0, %g1 + or %o0, %g0, %o1 + or %g0, (PAGE_SIZE >> 8), %g2 +1: + BLAST_BLOCK(%o0, 0x00) + BLAST_BLOCK(%o0, 0x40) + BLAST_BLOCK(%o0, 0x80) + BLAST_BLOCK(%o0, 0xc0) + subcc %g2, 1, %g2 + bne 1b + add %o0, 0x100, %o0 + + retl + nop + +__copy_1page: +/* NOTE: If you change the number of insns of this routine, please check + * arch/sparc/mm/hypersparc.S */ + /* %o0 = dst, %o1 = src */ + or %g0, (PAGE_SIZE >> 8), %g1 +1: + MIRROR_BLOCK(%o0, %o1, 0x00, %o2, %o3, %o4, %o5, %g2, %g3, %g4, %g5) + MIRROR_BLOCK(%o0, %o1, 0x20, %o2, %o3, %o4, %o5, %g2, %g3, %g4, %g5) + MIRROR_BLOCK(%o0, %o1, 0x40, %o2, %o3, %o4, %o5, %g2, %g3, %g4, %g5) + MIRROR_BLOCK(%o0, %o1, 0x60, %o2, %o3, %o4, %o5, %g2, %g3, %g4, %g5) + MIRROR_BLOCK(%o0, %o1, 0x80, %o2, %o3, %o4, %o5, %g2, %g3, %g4, %g5) + MIRROR_BLOCK(%o0, %o1, 0xa0, %o2, %o3, %o4, %o5, %g2, %g3, %g4, %g5) + MIRROR_BLOCK(%o0, %o1, 0xc0, %o2, %o3, %o4, %o5, %g2, %g3, %g4, %g5) + MIRROR_BLOCK(%o0, %o1, 0xe0, %o2, %o3, %o4, %o5, %g2, %g3, %g4, %g5) + subcc %g1, 1, %g1 + add %o0, 0x100, %o0 + bne 1b + add %o1, 0x100, %o1 + + retl + nop diff --git a/arch/sparc/lib/checksum.S b/arch/sparc/lib/checksum.S new file mode 100644 index 00000000000..77f228533d4 --- /dev/null +++ b/arch/sparc/lib/checksum.S @@ -0,0 +1,583 @@ +/* checksum.S: Sparc optimized checksum code. + * + * Copyright(C) 1995 Linus Torvalds + * Copyright(C) 1995 Miguel de Icaza + * Copyright(C) 1996 David S. Miller + * Copyright(C) 1997 Jakub Jelinek + * + * derived from: + * Linux/Alpha checksum c-code + * Linux/ix86 inline checksum assembly + * RFC1071 Computing the Internet Checksum (esp. Jacobsons m68k code) + * David Mosberger-Tang for optimized reference c-code + * BSD4.4 portable checksum routine + */ + +#include <asm/errno.h> + +#define CSUM_BIGCHUNK(buf, offset, sum, t0, t1, t2, t3, t4, t5) \ + ldd [buf + offset + 0x00], t0; \ + ldd [buf + offset + 0x08], t2; \ + addxcc t0, sum, sum; \ + addxcc t1, sum, sum; \ + ldd [buf + offset + 0x10], t4; \ + addxcc t2, sum, sum; \ + addxcc t3, sum, sum; \ + ldd [buf + offset + 0x18], t0; \ + addxcc t4, sum, sum; \ + addxcc t5, sum, sum; \ + addxcc t0, sum, sum; \ + addxcc t1, sum, sum; + +#define CSUM_LASTCHUNK(buf, offset, sum, t0, t1, t2, t3) \ + ldd [buf - offset - 0x08], t0; \ + ldd [buf - offset - 0x00], t2; \ + addxcc t0, sum, sum; \ + addxcc t1, sum, sum; \ + addxcc t2, sum, sum; \ + addxcc t3, sum, sum; + + /* Do end cruft out of band to get better cache patterns. */ +csum_partial_end_cruft: + be 1f ! caller asks %o1 & 0x8 + andcc %o1, 4, %g0 ! nope, check for word remaining + ldd [%o0], %g2 ! load two + addcc %g2, %o2, %o2 ! add first word to sum + addxcc %g3, %o2, %o2 ! add second word as well + add %o0, 8, %o0 ! advance buf ptr + addx %g0, %o2, %o2 ! add in final carry + andcc %o1, 4, %g0 ! check again for word remaining +1: be 1f ! nope, skip this code + andcc %o1, 3, %o1 ! check for trailing bytes + ld [%o0], %g2 ! load it + addcc %g2, %o2, %o2 ! add to sum + add %o0, 4, %o0 ! advance buf ptr + addx %g0, %o2, %o2 ! add in final carry + andcc %o1, 3, %g0 ! check again for trailing bytes +1: be 1f ! no trailing bytes, return + addcc %o1, -1, %g0 ! only one byte remains? + bne 2f ! at least two bytes more + subcc %o1, 2, %o1 ! only two bytes more? + b 4f ! only one byte remains + or %g0, %g0, %o4 ! clear fake hword value +2: lduh [%o0], %o4 ! get hword + be 6f ! jmp if only hword remains + add %o0, 2, %o0 ! advance buf ptr either way + sll %o4, 16, %o4 ! create upper hword +4: ldub [%o0], %o5 ! get final byte + sll %o5, 8, %o5 ! put into place + or %o5, %o4, %o4 ! coalese with hword (if any) +6: addcc %o4, %o2, %o2 ! add to sum +1: retl ! get outta here + addx %g0, %o2, %o0 ! add final carry into retval + + /* Also do alignment out of band to get better cache patterns. */ +csum_partial_fix_alignment: + cmp %o1, 6 + bl cpte - 0x4 + andcc %o0, 0x2, %g0 + be 1f + andcc %o0, 0x4, %g0 + lduh [%o0 + 0x00], %g2 + sub %o1, 2, %o1 + add %o0, 2, %o0 + sll %g2, 16, %g2 + addcc %g2, %o2, %o2 + srl %o2, 16, %g3 + addx %g0, %g3, %g2 + sll %o2, 16, %o2 + sll %g2, 16, %g3 + srl %o2, 16, %o2 + andcc %o0, 0x4, %g0 + or %g3, %o2, %o2 +1: be cpa + andcc %o1, 0xffffff80, %o3 + ld [%o0 + 0x00], %g2 + sub %o1, 4, %o1 + addcc %g2, %o2, %o2 + add %o0, 4, %o0 + addx %g0, %o2, %o2 + b cpa + andcc %o1, 0xffffff80, %o3 + + /* The common case is to get called with a nicely aligned + * buffer of size 0x20. Follow the code path for that case. + */ + .globl csum_partial +csum_partial: /* %o0=buf, %o1=len, %o2=sum */ + andcc %o0, 0x7, %g0 ! alignment problems? + bne csum_partial_fix_alignment ! yep, handle it + sethi %hi(cpte - 8), %g7 ! prepare table jmp ptr + andcc %o1, 0xffffff80, %o3 ! num loop iterations +cpa: be 3f ! none to do + andcc %o1, 0x70, %g1 ! clears carry flag too +5: CSUM_BIGCHUNK(%o0, 0x00, %o2, %o4, %o5, %g2, %g3, %g4, %g5) + CSUM_BIGCHUNK(%o0, 0x20, %o2, %o4, %o5, %g2, %g3, %g4, %g5) + CSUM_BIGCHUNK(%o0, 0x40, %o2, %o4, %o5, %g2, %g3, %g4, %g5) + CSUM_BIGCHUNK(%o0, 0x60, %o2, %o4, %o5, %g2, %g3, %g4, %g5) + addx %g0, %o2, %o2 ! sink in final carry + subcc %o3, 128, %o3 ! detract from loop iters + bne 5b ! more to do + add %o0, 128, %o0 ! advance buf ptr + andcc %o1, 0x70, %g1 ! clears carry flag too +3: be cpte ! nope + andcc %o1, 0xf, %g0 ! anything left at all? + srl %g1, 1, %o4 ! compute offset + sub %g7, %g1, %g7 ! adjust jmp ptr + sub %g7, %o4, %g7 ! final jmp ptr adjust + jmp %g7 + %lo(cpte - 8) ! enter the table + add %o0, %g1, %o0 ! advance buf ptr +cptbl: CSUM_LASTCHUNK(%o0, 0x68, %o2, %g2, %g3, %g4, %g5) + CSUM_LASTCHUNK(%o0, 0x58, %o2, %g2, %g3, %g4, %g5) + CSUM_LASTCHUNK(%o0, 0x48, %o2, %g2, %g3, %g4, %g5) + CSUM_LASTCHUNK(%o0, 0x38, %o2, %g2, %g3, %g4, %g5) + CSUM_LASTCHUNK(%o0, 0x28, %o2, %g2, %g3, %g4, %g5) + CSUM_LASTCHUNK(%o0, 0x18, %o2, %g2, %g3, %g4, %g5) + CSUM_LASTCHUNK(%o0, 0x08, %o2, %g2, %g3, %g4, %g5) + addx %g0, %o2, %o2 ! fetch final carry + andcc %o1, 0xf, %g0 ! anything left at all? +cpte: bne csum_partial_end_cruft ! yep, handle it + andcc %o1, 8, %g0 ! check how much +cpout: retl ! get outta here + mov %o2, %o0 ! return computed csum + + .globl __csum_partial_copy_start, __csum_partial_copy_end +__csum_partial_copy_start: + +/* Work around cpp -rob */ +#define ALLOC #alloc +#define EXECINSTR #execinstr +#define EX(x,y,a,b) \ +98: x,y; \ + .section .fixup,ALLOC,EXECINSTR; \ + .align 4; \ +99: ba 30f; \ + a, b, %o3; \ + .section __ex_table,ALLOC; \ + .align 4; \ + .word 98b, 99b; \ + .text; \ + .align 4 + +#define EX2(x,y) \ +98: x,y; \ + .section __ex_table,ALLOC; \ + .align 4; \ + .word 98b, 30f; \ + .text; \ + .align 4 + +#define EX3(x,y) \ +98: x,y; \ + .section __ex_table,ALLOC; \ + .align 4; \ + .word 98b, 96f; \ + .text; \ + .align 4 + +#define EXT(start,end,handler) \ + .section __ex_table,ALLOC; \ + .align 4; \ + .word start, 0, end, handler; \ + .text; \ + .align 4 + + /* This aligned version executes typically in 8.5 superscalar cycles, this + * is the best I can do. I say 8.5 because the final add will pair with + * the next ldd in the main unrolled loop. Thus the pipe is always full. + * If you change these macros (including order of instructions), + * please check the fixup code below as well. + */ +#define CSUMCOPY_BIGCHUNK_ALIGNED(src, dst, sum, off, t0, t1, t2, t3, t4, t5, t6, t7) \ + ldd [src + off + 0x00], t0; \ + ldd [src + off + 0x08], t2; \ + addxcc t0, sum, sum; \ + ldd [src + off + 0x10], t4; \ + addxcc t1, sum, sum; \ + ldd [src + off + 0x18], t6; \ + addxcc t2, sum, sum; \ + std t0, [dst + off + 0x00]; \ + addxcc t3, sum, sum; \ + std t2, [dst + off + 0x08]; \ + addxcc t4, sum, sum; \ + std t4, [dst + off + 0x10]; \ + addxcc t5, sum, sum; \ + std t6, [dst + off + 0x18]; \ + addxcc t6, sum, sum; \ + addxcc t7, sum, sum; + + /* 12 superscalar cycles seems to be the limit for this case, + * because of this we thus do all the ldd's together to get + * Viking MXCC into streaming mode. Ho hum... + */ +#define CSUMCOPY_BIGCHUNK(src, dst, sum, off, t0, t1, t2, t3, t4, t5, t6, t7) \ + ldd [src + off + 0x00], t0; \ + ldd [src + off + 0x08], t2; \ + ldd [src + off + 0x10], t4; \ + ldd [src + off + 0x18], t6; \ + st t0, [dst + off + 0x00]; \ + addxcc t0, sum, sum; \ + st t1, [dst + off + 0x04]; \ + addxcc t1, sum, sum; \ + st t2, [dst + off + 0x08]; \ + addxcc t2, sum, sum; \ + st t3, [dst + off + 0x0c]; \ + addxcc t3, sum, sum; \ + st t4, [dst + off + 0x10]; \ + addxcc t4, sum, sum; \ + st t5, [dst + off + 0x14]; \ + addxcc t5, sum, sum; \ + st t6, [dst + off + 0x18]; \ + addxcc t6, sum, sum; \ + st t7, [dst + off + 0x1c]; \ + addxcc t7, sum, sum; + + /* Yuck, 6 superscalar cycles... */ +#define CSUMCOPY_LASTCHUNK(src, dst, sum, off, t0, t1, t2, t3) \ + ldd [src - off - 0x08], t0; \ + ldd [src - off - 0x00], t2; \ + addxcc t0, sum, sum; \ + st t0, [dst - off - 0x08]; \ + addxcc t1, sum, sum; \ + st t1, [dst - off - 0x04]; \ + addxcc t2, sum, sum; \ + st t2, [dst - off - 0x00]; \ + addxcc t3, sum, sum; \ + st t3, [dst - off + 0x04]; + + /* Handle the end cruft code out of band for better cache patterns. */ +cc_end_cruft: + be 1f + andcc %o3, 4, %g0 + EX(ldd [%o0 + 0x00], %g2, and %o3, 0xf) + add %o1, 8, %o1 + addcc %g2, %g7, %g7 + add %o0, 8, %o0 + addxcc %g3, %g7, %g7 + EX2(st %g2, [%o1 - 0x08]) + addx %g0, %g7, %g7 + andcc %o3, 4, %g0 + EX2(st %g3, [%o1 - 0x04]) +1: be 1f + andcc %o3, 3, %o3 + EX(ld [%o0 + 0x00], %g2, add %o3, 4) + add %o1, 4, %o1 + addcc %g2, %g7, %g7 + EX2(st %g2, [%o1 - 0x04]) + addx %g0, %g7, %g7 + andcc %o3, 3, %g0 + add %o0, 4, %o0 +1: be 1f + addcc %o3, -1, %g0 + bne 2f + subcc %o3, 2, %o3 + b 4f + or %g0, %g0, %o4 +2: EX(lduh [%o0 + 0x00], %o4, add %o3, 2) + add %o0, 2, %o0 + EX2(sth %o4, [%o1 + 0x00]) + be 6f + add %o1, 2, %o1 + sll %o4, 16, %o4 +4: EX(ldub [%o0 + 0x00], %o5, add %g0, 1) + EX2(stb %o5, [%o1 + 0x00]) + sll %o5, 8, %o5 + or %o5, %o4, %o4 +6: addcc %o4, %g7, %g7 +1: retl + addx %g0, %g7, %o0 + + /* Also, handle the alignment code out of band. */ +cc_dword_align: + cmp %g1, 6 + bl,a ccte + andcc %g1, 0xf, %o3 + andcc %o0, 0x1, %g0 + bne ccslow + andcc %o0, 0x2, %g0 + be 1f + andcc %o0, 0x4, %g0 + EX(lduh [%o0 + 0x00], %g4, add %g1, 0) + sub %g1, 2, %g1 + EX2(sth %g4, [%o1 + 0x00]) + add %o0, 2, %o0 + sll %g4, 16, %g4 + addcc %g4, %g7, %g7 + add %o1, 2, %o1 + srl %g7, 16, %g3 + addx %g0, %g3, %g4 + sll %g7, 16, %g7 + sll %g4, 16, %g3 + srl %g7, 16, %g7 + andcc %o0, 0x4, %g0 + or %g3, %g7, %g7 +1: be 3f + andcc %g1, 0xffffff80, %g0 + EX(ld [%o0 + 0x00], %g4, add %g1, 0) + sub %g1, 4, %g1 + EX2(st %g4, [%o1 + 0x00]) + add %o0, 4, %o0 + addcc %g4, %g7, %g7 + add %o1, 4, %o1 + addx %g0, %g7, %g7 + b 3f + andcc %g1, 0xffffff80, %g0 + + /* Sun, you just can't beat me, you just can't. Stop trying, + * give up. I'm serious, I am going to kick the living shit + * out of you, game over, lights out. + */ + .align 8 + .globl __csum_partial_copy_sparc_generic +__csum_partial_copy_sparc_generic: + /* %o0=src, %o1=dest, %g1=len, %g7=sum */ + xor %o0, %o1, %o4 ! get changing bits + andcc %o4, 3, %g0 ! check for mismatched alignment + bne ccslow ! better this than unaligned/fixups + andcc %o0, 7, %g0 ! need to align things? + bne cc_dword_align ! yes, we check for short lengths there + andcc %g1, 0xffffff80, %g0 ! can we use unrolled loop? +3: be 3f ! nope, less than one loop remains + andcc %o1, 4, %g0 ! dest aligned on 4 or 8 byte boundary? + be ccdbl + 4 ! 8 byte aligned, kick ass +5: CSUMCOPY_BIGCHUNK(%o0,%o1,%g7,0x00,%o4,%o5,%g2,%g3,%g4,%g5,%o2,%o3) + CSUMCOPY_BIGCHUNK(%o0,%o1,%g7,0x20,%o4,%o5,%g2,%g3,%g4,%g5,%o2,%o3) + CSUMCOPY_BIGCHUNK(%o0,%o1,%g7,0x40,%o4,%o5,%g2,%g3,%g4,%g5,%o2,%o3) + CSUMCOPY_BIGCHUNK(%o0,%o1,%g7,0x60,%o4,%o5,%g2,%g3,%g4,%g5,%o2,%o3) +10: EXT(5b, 10b, 20f) ! note for exception handling + sub %g1, 128, %g1 ! detract from length + addx %g0, %g7, %g7 ! add in last carry bit + andcc %g1, 0xffffff80, %g0 ! more to csum? + add %o0, 128, %o0 ! advance src ptr + bne 5b ! we did not go negative, continue looping + add %o1, 128, %o1 ! advance dest ptr +3: andcc %g1, 0x70, %o2 ! can use table? +ccmerge:be ccte ! nope, go and check for end cruft + andcc %g1, 0xf, %o3 ! get low bits of length (clears carry btw) + srl %o2, 1, %o4 ! begin negative offset computation + sethi %hi(12f), %o5 ! set up table ptr end + add %o0, %o2, %o0 ! advance src ptr + sub %o5, %o4, %o5 ! continue table calculation + sll %o2, 1, %g2 ! constant multiplies are fun... + sub %o5, %g2, %o5 ! some more adjustments + jmp %o5 + %lo(12f) ! jump into it, duff style, wheee... + add %o1, %o2, %o1 ! advance dest ptr (carry is clear btw) +cctbl: CSUMCOPY_LASTCHUNK(%o0,%o1,%g7,0x68,%g2,%g3,%g4,%g5) + CSUMCOPY_LASTCHUNK(%o0,%o1,%g7,0x58,%g2,%g3,%g4,%g5) + CSUMCOPY_LASTCHUNK(%o0,%o1,%g7,0x48,%g2,%g3,%g4,%g5) + CSUMCOPY_LASTCHUNK(%o0,%o1,%g7,0x38,%g2,%g3,%g4,%g5) + CSUMCOPY_LASTCHUNK(%o0,%o1,%g7,0x28,%g2,%g3,%g4,%g5) + CSUMCOPY_LASTCHUNK(%o0,%o1,%g7,0x18,%g2,%g3,%g4,%g5) + CSUMCOPY_LASTCHUNK(%o0,%o1,%g7,0x08,%g2,%g3,%g4,%g5) +12: EXT(cctbl, 12b, 22f) ! note for exception table handling + addx %g0, %g7, %g7 + andcc %o3, 0xf, %g0 ! check for low bits set +ccte: bne cc_end_cruft ! something left, handle it out of band + andcc %o3, 8, %g0 ! begin checks for that code + retl ! return + mov %g7, %o0 ! give em the computed checksum +ccdbl: CSUMCOPY_BIGCHUNK_ALIGNED(%o0,%o1,%g7,0x00,%o4,%o5,%g2,%g3,%g4,%g5,%o2,%o3) + CSUMCOPY_BIGCHUNK_ALIGNED(%o0,%o1,%g7,0x20,%o4,%o5,%g2,%g3,%g4,%g5,%o2,%o3) + CSUMCOPY_BIGCHUNK_ALIGNED(%o0,%o1,%g7,0x40,%o4,%o5,%g2,%g3,%g4,%g5,%o2,%o3) + CSUMCOPY_BIGCHUNK_ALIGNED(%o0,%o1,%g7,0x60,%o4,%o5,%g2,%g3,%g4,%g5,%o2,%o3) +11: EXT(ccdbl, 11b, 21f) ! note for exception table handling + sub %g1, 128, %g1 ! detract from length + addx %g0, %g7, %g7 ! add in last carry bit + andcc %g1, 0xffffff80, %g0 ! more to csum? + add %o0, 128, %o0 ! advance src ptr + bne ccdbl ! we did not go negative, continue looping + add %o1, 128, %o1 ! advance dest ptr + b ccmerge ! finish it off, above + andcc %g1, 0x70, %o2 ! can use table? (clears carry btw) + +ccslow: cmp %g1, 0 + mov 0, %g5 + bleu 4f + andcc %o0, 1, %o5 + be,a 1f + srl %g1, 1, %g4 + sub %g1, 1, %g1 + EX(ldub [%o0], %g5, add %g1, 1) + add %o0, 1, %o0 + EX2(stb %g5, [%o1]) + srl %g1, 1, %g4 + add %o1, 1, %o1 +1: cmp %g4, 0 + be,a 3f + andcc %g1, 1, %g0 + andcc %o0, 2, %g0 + be,a 1f + srl %g4, 1, %g4 + EX(lduh [%o0], %o4, add %g1, 0) + sub %g1, 2, %g1 + srl %o4, 8, %g2 + sub %g4, 1, %g4 + EX2(stb %g2, [%o1]) + add %o4, %g5, %g5 + EX2(stb %o4, [%o1 + 1]) + add %o0, 2, %o0 + srl %g4, 1, %g4 + add %o1, 2, %o1 +1: cmp %g4, 0 + be,a 2f + andcc %g1, 2, %g0 + EX3(ld [%o0], %o4) +5: srl %o4, 24, %g2 + srl %o4, 16, %g3 + EX2(stb %g2, [%o1]) + srl %o4, 8, %g2 + EX2(stb %g3, [%o1 + 1]) + add %o0, 4, %o0 + EX2(stb %g2, [%o1 + 2]) + addcc %o4, %g5, %g5 + EX2(stb %o4, [%o1 + 3]) + addx %g5, %g0, %g5 ! I am now to lazy to optimize this (question it + add %o1, 4, %o1 ! is worthy). Maybe some day - with the sll/srl + subcc %g4, 1, %g4 ! tricks + bne,a 5b + EX3(ld [%o0], %o4) + sll %g5, 16, %g2 + srl %g5, 16, %g5 + srl %g2, 16, %g2 + andcc %g1, 2, %g0 + add %g2, %g5, %g5 +2: be,a 3f + andcc %g1, 1, %g0 + EX(lduh [%o0], %o4, and %g1, 3) + andcc %g1, 1, %g0 + srl %o4, 8, %g2 + add %o0, 2, %o0 + EX2(stb %g2, [%o1]) + add %g5, %o4, %g5 + EX2(stb %o4, [%o1 + 1]) + add %o1, 2, %o1 +3: be,a 1f + sll %g5, 16, %o4 + EX(ldub [%o0], %g2, add %g0, 1) + sll %g2, 8, %o4 + EX2(stb %g2, [%o1]) + add %g5, %o4, %g5 + sll %g5, 16, %o4 +1: addcc %o4, %g5, %g5 + srl %g5, 16, %o4 + addx %g0, %o4, %g5 + orcc %o5, %g0, %g0 + be 4f + srl %g5, 8, %o4 + and %g5, 0xff, %g2 + and %o4, 0xff, %o4 + sll %g2, 8, %g2 + or %g2, %o4, %g5 +4: addcc %g7, %g5, %g7 + retl + addx %g0, %g7, %o0 +__csum_partial_copy_end: + +/* We do these strange calculations for the csum_*_from_user case only, ie. + * we only bother with faults on loads... */ + +/* o2 = ((g2%20)&3)*8 + * o3 = g1 - (g2/20)*32 - o2 */ +20: + cmp %g2, 20 + blu,a 1f + and %g2, 3, %o2 + sub %g1, 32, %g1 + b 20b + sub %g2, 20, %g2 +1: + sll %o2, 3, %o2 + b 31f + sub %g1, %o2, %o3 + +/* o2 = (!(g2 & 15) ? 0 : (((g2 & 15) + 1) & ~1)*8) + * o3 = g1 - (g2/16)*32 - o2 */ +21: + andcc %g2, 15, %o3 + srl %g2, 4, %g2 + be,a 1f + clr %o2 + add %o3, 1, %o3 + and %o3, 14, %o3 + sll %o3, 3, %o2 +1: + sll %g2, 5, %g2 + sub %g1, %g2, %o3 + b 31f + sub %o3, %o2, %o3 + +/* o0 += (g2/10)*16 - 0x70 + * 01 += (g2/10)*16 - 0x70 + * o2 = (g2 % 10) ? 8 : 0 + * o3 += 0x70 - (g2/10)*16 - o2 */ +22: + cmp %g2, 10 + blu,a 1f + sub %o0, 0x70, %o0 + add %o0, 16, %o0 + add %o1, 16, %o1 + sub %o3, 16, %o3 + b 22b + sub %g2, 10, %g2 +1: + sub %o1, 0x70, %o1 + add %o3, 0x70, %o3 + clr %o2 + tst %g2 + bne,a 1f + mov 8, %o2 +1: + b 31f + sub %o3, %o2, %o3 +96: + and %g1, 3, %g1 + sll %g4, 2, %g4 + add %g1, %g4, %o3 +30: +/* %o1 is dst + * %o3 is # bytes to zero out + * %o4 is faulting address + * %o5 is %pc where fault occurred */ + clr %o2 +31: +/* %o0 is src + * %o1 is dst + * %o2 is # of bytes to copy from src to dst + * %o3 is # bytes to zero out + * %o4 is faulting address + * %o5 is %pc where fault occurred */ + save %sp, -104, %sp + mov %i5, %o0 + mov %i7, %o1 + mov %i4, %o2 + call lookup_fault + mov %g7, %i4 + cmp %o0, 2 + bne 1f + add %g0, -EFAULT, %i5 + tst %i2 + be 2f + mov %i0, %o1 + mov %i1, %o0 +5: + call __memcpy + mov %i2, %o2 + tst %o0 + bne,a 2f + add %i3, %i2, %i3 + add %i1, %i2, %i1 +2: + mov %i1, %o0 +6: + call __bzero + mov %i3, %o1 +1: + ld [%sp + 168], %o2 ! struct_ptr of parent + st %i5, [%o2] + ret + restore + + .section __ex_table,#alloc + .align 4 + .word 5b,2 + .word 6b,2 diff --git a/arch/sparc/lib/copy_user.S b/arch/sparc/lib/copy_user.S new file mode 100644 index 00000000000..577505b692a --- /dev/null +++ b/arch/sparc/lib/copy_user.S @@ -0,0 +1,492 @@ +/* copy_user.S: Sparc optimized copy_from_user and copy_to_user code. + * + * Copyright(C) 1995 Linus Torvalds + * Copyright(C) 1996 David S. Miller + * Copyright(C) 1996 Eddie C. Dost + * Copyright(C) 1996,1998 Jakub Jelinek + * + * derived from: + * e-mail between David and Eddie. + * + * Returns 0 if successful, otherwise count of bytes not copied yet + */ + +#include <asm/ptrace.h> +#include <asm/asmmacro.h> +#include <asm/page.h> + +/* Work around cpp -rob */ +#define ALLOC #alloc +#define EXECINSTR #execinstr +#define EX(x,y,a,b) \ +98: x,y; \ + .section .fixup,ALLOC,EXECINSTR; \ + .align 4; \ +99: ba fixupretl; \ + a, b, %g3; \ + .section __ex_table,ALLOC; \ + .align 4; \ + .word 98b, 99b; \ + .text; \ + .align 4 + +#define EX2(x,y,c,d,e,a,b) \ +98: x,y; \ + .section .fixup,ALLOC,EXECINSTR; \ + .align 4; \ +99: c, d, e; \ + ba fixupretl; \ + a, b, %g3; \ + .section __ex_table,ALLOC; \ + .align 4; \ + .word 98b, 99b; \ + .text; \ + .align 4 + +#define EXO2(x,y) \ +98: x, y; \ + .section __ex_table,ALLOC; \ + .align 4; \ + .word 98b, 97f; \ + .text; \ + .align 4 + +#define EXT(start,end,handler) \ + .section __ex_table,ALLOC; \ + .align 4; \ + .word start, 0, end, handler; \ + .text; \ + .align 4 + +/* Please do not change following macros unless you change logic used + * in .fixup at the end of this file as well + */ + +/* Both these macros have to start with exactly the same insn */ +#define MOVE_BIGCHUNK(src, dst, offset, t0, t1, t2, t3, t4, t5, t6, t7) \ + ldd [%src + (offset) + 0x00], %t0; \ + ldd [%src + (offset) + 0x08], %t2; \ + ldd [%src + (offset) + 0x10], %t4; \ + ldd [%src + (offset) + 0x18], %t6; \ + st %t0, [%dst + (offset) + 0x00]; \ + st %t1, [%dst + (offset) + 0x04]; \ + st %t2, [%dst + (offset) + 0x08]; \ + st %t3, [%dst + (offset) + 0x0c]; \ + st %t4, [%dst + (offset) + 0x10]; \ + st %t5, [%dst + (offset) + 0x14]; \ + st %t6, [%dst + (offset) + 0x18]; \ + st %t7, [%dst + (offset) + 0x1c]; + +#define MOVE_BIGALIGNCHUNK(src, dst, offset, t0, t1, t2, t3, t4, t5, t6, t7) \ + ldd [%src + (offset) + 0x00], %t0; \ + ldd [%src + (offset) + 0x08], %t2; \ + ldd [%src + (offset) + 0x10], %t4; \ + ldd [%src + (offset) + 0x18], %t6; \ + std %t0, [%dst + (offset) + 0x00]; \ + std %t2, [%dst + (offset) + 0x08]; \ + std %t4, [%dst + (offset) + 0x10]; \ + std %t6, [%dst + (offset) + 0x18]; + +#define MOVE_LASTCHUNK(src, dst, offset, t0, t1, t2, t3) \ + ldd [%src - (offset) - 0x10], %t0; \ + ldd [%src - (offset) - 0x08], %t2; \ + st %t0, [%dst - (offset) - 0x10]; \ + st %t1, [%dst - (offset) - 0x0c]; \ + st %t2, [%dst - (offset) - 0x08]; \ + st %t3, [%dst - (offset) - 0x04]; + +#define MOVE_HALFCHUNK(src, dst, offset, t0, t1, t2, t3) \ + lduh [%src + (offset) + 0x00], %t0; \ + lduh [%src + (offset) + 0x02], %t1; \ + lduh [%src + (offset) + 0x04], %t2; \ + lduh [%src + (offset) + 0x06], %t3; \ + sth %t0, [%dst + (offset) + 0x00]; \ + sth %t1, [%dst + (offset) + 0x02]; \ + sth %t2, [%dst + (offset) + 0x04]; \ + sth %t3, [%dst + (offset) + 0x06]; + +#define MOVE_SHORTCHUNK(src, dst, offset, t0, t1) \ + ldub [%src - (offset) - 0x02], %t0; \ + ldub [%src - (offset) - 0x01], %t1; \ + stb %t0, [%dst - (offset) - 0x02]; \ + stb %t1, [%dst - (offset) - 0x01]; + + .text + .align 4 + + .globl __copy_user_begin +__copy_user_begin: + + .globl __copy_user +dword_align: + andcc %o1, 1, %g0 + be 4f + andcc %o1, 2, %g0 + + EXO2(ldub [%o1], %g2) + add %o1, 1, %o1 + EXO2(stb %g2, [%o0]) + sub %o2, 1, %o2 + bne 3f + add %o0, 1, %o0 + + EXO2(lduh [%o1], %g2) + add %o1, 2, %o1 + EXO2(sth %g2, [%o0]) + sub %o2, 2, %o2 + b 3f + add %o0, 2, %o0 +4: + EXO2(lduh [%o1], %g2) + add %o1, 2, %o1 + EXO2(sth %g2, [%o0]) + sub %o2, 2, %o2 + b 3f + add %o0, 2, %o0 + +__copy_user: /* %o0=dst %o1=src %o2=len */ + xor %o0, %o1, %o4 +1: + andcc %o4, 3, %o5 +2: + bne cannot_optimize + cmp %o2, 15 + + bleu short_aligned_end + andcc %o1, 3, %g0 + + bne dword_align +3: + andcc %o1, 4, %g0 + + be 2f + mov %o2, %g1 + + EXO2(ld [%o1], %o4) + sub %g1, 4, %g1 + EXO2(st %o4, [%o0]) + add %o1, 4, %o1 + add %o0, 4, %o0 +2: + andcc %g1, 0xffffff80, %g7 + be 3f + andcc %o0, 4, %g0 + + be ldd_std + 4 +5: + MOVE_BIGCHUNK(o1, o0, 0x00, o2, o3, o4, o5, g2, g3, g4, g5) + MOVE_BIGCHUNK(o1, o0, 0x20, o2, o3, o4, o5, g2, g3, g4, g5) + MOVE_BIGCHUNK(o1, o0, 0x40, o2, o3, o4, o5, g2, g3, g4, g5) + MOVE_BIGCHUNK(o1, o0, 0x60, o2, o3, o4, o5, g2, g3, g4, g5) +80: + EXT(5b, 80b, 50f) + subcc %g7, 128, %g7 + add %o1, 128, %o1 + bne 5b + add %o0, 128, %o0 +3: + andcc %g1, 0x70, %g7 + be copy_user_table_end + andcc %g1, 8, %g0 + + sethi %hi(copy_user_table_end), %o5 + srl %g7, 1, %o4 + add %g7, %o4, %o4 + add %o1, %g7, %o1 + sub %o5, %o4, %o5 + jmpl %o5 + %lo(copy_user_table_end), %g0 + add %o0, %g7, %o0 + +copy_user_table: + MOVE_LASTCHUNK(o1, o0, 0x60, g2, g3, g4, g5) + MOVE_LASTCHUNK(o1, o0, 0x50, g2, g3, g4, g5) + MOVE_LASTCHUNK(o1, o0, 0x40, g2, g3, g4, g5) + MOVE_LASTCHUNK(o1, o0, 0x30, g2, g3, g4, g5) + MOVE_LASTCHUNK(o1, o0, 0x20, g2, g3, g4, g5) + MOVE_LASTCHUNK(o1, o0, 0x10, g2, g3, g4, g5) + MOVE_LASTCHUNK(o1, o0, 0x00, g2, g3, g4, g5) +copy_user_table_end: + EXT(copy_user_table, copy_user_table_end, 51f) + be copy_user_last7 + andcc %g1, 4, %g0 + + EX(ldd [%o1], %g2, and %g1, 0xf) + add %o0, 8, %o0 + add %o1, 8, %o1 + EX(st %g2, [%o0 - 0x08], and %g1, 0xf) + EX2(st %g3, [%o0 - 0x04], and %g1, 0xf, %g1, sub %g1, 4) +copy_user_last7: + be 1f + andcc %g1, 2, %g0 + + EX(ld [%o1], %g2, and %g1, 7) + add %o1, 4, %o1 + EX(st %g2, [%o0], and %g1, 7) + add %o0, 4, %o0 +1: + be 1f + andcc %g1, 1, %g0 + + EX(lduh [%o1], %g2, and %g1, 3) + add %o1, 2, %o1 + EX(sth %g2, [%o0], and %g1, 3) + add %o0, 2, %o0 +1: + be 1f + nop + + EX(ldub [%o1], %g2, add %g0, 1) + EX(stb %g2, [%o0], add %g0, 1) +1: + retl + clr %o0 + +ldd_std: + MOVE_BIGALIGNCHUNK(o1, o0, 0x00, o2, o3, o4, o5, g2, g3, g4, g5) + MOVE_BIGALIGNCHUNK(o1, o0, 0x20, o2, o3, o4, o5, g2, g3, g4, g5) + MOVE_BIGALIGNCHUNK(o1, o0, 0x40, o2, o3, o4, o5, g2, g3, g4, g5) + MOVE_BIGALIGNCHUNK(o1, o0, 0x60, o2, o3, o4, o5, g2, g3, g4, g5) +81: + EXT(ldd_std, 81b, 52f) + subcc %g7, 128, %g7 + add %o1, 128, %o1 + bne ldd_std + add %o0, 128, %o0 + + andcc %g1, 0x70, %g7 + be copy_user_table_end + andcc %g1, 8, %g0 + + sethi %hi(copy_user_table_end), %o5 + srl %g7, 1, %o4 + add %g7, %o4, %o4 + add %o1, %g7, %o1 + sub %o5, %o4, %o5 + jmpl %o5 + %lo(copy_user_table_end), %g0 + add %o0, %g7, %o0 + +cannot_optimize: + bleu short_end + cmp %o5, 2 + + bne byte_chunk + and %o2, 0xfffffff0, %o3 + + andcc %o1, 1, %g0 + be 10f + nop + + EXO2(ldub [%o1], %g2) + add %o1, 1, %o1 + EXO2(stb %g2, [%o0]) + sub %o2, 1, %o2 + andcc %o2, 0xfffffff0, %o3 + be short_end + add %o0, 1, %o0 +10: + MOVE_HALFCHUNK(o1, o0, 0x00, g2, g3, g4, g5) + MOVE_HALFCHUNK(o1, o0, 0x08, g2, g3, g4, g5) +82: + EXT(10b, 82b, 53f) + subcc %o3, 0x10, %o3 + add %o1, 0x10, %o1 + bne 10b + add %o0, 0x10, %o0 + b 2f + and %o2, 0xe, %o3 + +byte_chunk: + MOVE_SHORTCHUNK(o1, o0, -0x02, g2, g3) + MOVE_SHORTCHUNK(o1, o0, -0x04, g2, g3) + MOVE_SHORTCHUNK(o1, o0, -0x06, g2, g3) + MOVE_SHORTCHUNK(o1, o0, -0x08, g2, g3) + MOVE_SHORTCHUNK(o1, o0, -0x0a, g2, g3) + MOVE_SHORTCHUNK(o1, o0, -0x0c, g2, g3) + MOVE_SHORTCHUNK(o1, o0, -0x0e, g2, g3) + MOVE_SHORTCHUNK(o1, o0, -0x10, g2, g3) +83: + EXT(byte_chunk, 83b, 54f) + subcc %o3, 0x10, %o3 + add %o1, 0x10, %o1 + bne byte_chunk + add %o0, 0x10, %o0 + +short_end: + and %o2, 0xe, %o3 +2: + sethi %hi(short_table_end), %o5 + sll %o3, 3, %o4 + add %o0, %o3, %o0 + sub %o5, %o4, %o5 + add %o1, %o3, %o1 + jmpl %o5 + %lo(short_table_end), %g0 + andcc %o2, 1, %g0 +84: + MOVE_SHORTCHUNK(o1, o0, 0x0c, g2, g3) + MOVE_SHORTCHUNK(o1, o0, 0x0a, g2, g3) + MOVE_SHORTCHUNK(o1, o0, 0x08, g2, g3) + MOVE_SHORTCHUNK(o1, o0, 0x06, g2, g3) + MOVE_SHORTCHUNK(o1, o0, 0x04, g2, g3) + MOVE_SHORTCHUNK(o1, o0, 0x02, g2, g3) + MOVE_SHORTCHUNK(o1, o0, 0x00, g2, g3) +short_table_end: + EXT(84b, short_table_end, 55f) + be 1f + nop + EX(ldub [%o1], %g2, add %g0, 1) + EX(stb %g2, [%o0], add %g0, 1) +1: + retl + clr %o0 + +short_aligned_end: + bne short_end + andcc %o2, 8, %g0 + + be 1f + andcc %o2, 4, %g0 + + EXO2(ld [%o1 + 0x00], %g2) + EXO2(ld [%o1 + 0x04], %g3) + add %o1, 8, %o1 + EXO2(st %g2, [%o0 + 0x00]) + EX(st %g3, [%o0 + 0x04], sub %o2, 4) + add %o0, 8, %o0 +1: + b copy_user_last7 + mov %o2, %g1 + + .section .fixup,#alloc,#execinstr + .align 4 +97: + mov %o2, %g3 +fixupretl: + sethi %hi(PAGE_OFFSET), %g1 + cmp %o0, %g1 + blu 1f + cmp %o1, %g1 + bgeu 1f + nop + save %sp, -64, %sp + mov %i0, %o0 + call __bzero + mov %g3, %o1 + restore +1: retl + mov %g3, %o0 + +/* exception routine sets %g2 to (broken_insn - first_insn)>>2 */ +50: +/* This magic counts how many bytes are left when crash in MOVE_BIGCHUNK + * happens. This is derived from the amount ldd reads, st stores, etc. + * x = g2 % 12; + * g3 = g1 + g7 - ((g2 / 12) * 32 + (x < 4) ? 0 : (x - 4) * 4); + * o0 += (g2 / 12) * 32; + */ + cmp %g2, 12 + add %o0, %g7, %o0 + bcs 1f + cmp %g2, 24 + bcs 2f + cmp %g2, 36 + bcs 3f + nop + sub %g2, 12, %g2 + sub %g7, 32, %g7 +3: sub %g2, 12, %g2 + sub %g7, 32, %g7 +2: sub %g2, 12, %g2 + sub %g7, 32, %g7 +1: cmp %g2, 4 + bcs,a 60f + clr %g2 + sub %g2, 4, %g2 + sll %g2, 2, %g2 +60: and %g1, 0x7f, %g3 + sub %o0, %g7, %o0 + add %g3, %g7, %g3 + ba fixupretl + sub %g3, %g2, %g3 +51: +/* i = 41 - g2; j = i % 6; + * g3 = (g1 & 15) + (i / 6) * 16 + (j < 4) ? (j + 1) * 4 : 16; + * o0 -= (i / 6) * 16 + 16; + */ + neg %g2 + and %g1, 0xf, %g1 + add %g2, 41, %g2 + add %o0, %g1, %o0 +1: cmp %g2, 6 + bcs,a 2f + cmp %g2, 4 + add %g1, 16, %g1 + b 1b + sub %g2, 6, %g2 +2: bcc,a 2f + mov 16, %g2 + inc %g2 + sll %g2, 2, %g2 +2: add %g1, %g2, %g3 + ba fixupretl + sub %o0, %g3, %o0 +52: +/* g3 = g1 + g7 - (g2 / 8) * 32 + (g2 & 4) ? (g2 & 3) * 8 : 0; + o0 += (g2 / 8) * 32 */ + andn %g2, 7, %g4 + add %o0, %g7, %o0 + andcc %g2, 4, %g0 + and %g2, 3, %g2 + sll %g4, 2, %g4 + sll %g2, 3, %g2 + bne 60b + sub %g7, %g4, %g7 + ba 60b + clr %g2 +53: +/* g3 = o3 + (o2 & 15) - (g2 & 8) - (g2 & 4) ? (g2 & 3) * 2 : 0; + o0 += (g2 & 8) */ + and %g2, 3, %g4 + andcc %g2, 4, %g0 + and %g2, 8, %g2 + sll %g4, 1, %g4 + be 1f + add %o0, %g2, %o0 + add %g2, %g4, %g2 +1: and %o2, 0xf, %g3 + add %g3, %o3, %g3 + ba fixupretl + sub %g3, %g2, %g3 +54: +/* g3 = o3 + (o2 & 15) - (g2 / 4) * 2 - (g2 & 2) ? (g2 & 1) : 0; + o0 += (g2 / 4) * 2 */ + srl %g2, 2, %o4 + and %g2, 1, %o5 + srl %g2, 1, %g2 + add %o4, %o4, %o4 + and %o5, %g2, %o5 + and %o2, 0xf, %o2 + add %o0, %o4, %o0 + sub %o3, %o5, %o3 + sub %o2, %o4, %o2 + ba fixupretl + add %o2, %o3, %g3 +55: +/* i = 27 - g2; + g3 = (o2 & 1) + i / 4 * 2 + !(i & 3); + o0 -= i / 4 * 2 + 1 */ + neg %g2 + and %o2, 1, %o2 + add %g2, 27, %g2 + srl %g2, 2, %o5 + andcc %g2, 3, %g0 + mov 1, %g2 + add %o5, %o5, %o5 + be,a 1f + clr %g2 +1: add %g2, %o5, %g3 + sub %o0, %g3, %o0 + ba fixupretl + add %g3, %o2, %g3 + + .globl __copy_user_end +__copy_user_end: diff --git a/arch/sparc/lib/debuglocks.c b/arch/sparc/lib/debuglocks.c new file mode 100644 index 00000000000..fb182352782 --- /dev/null +++ b/arch/sparc/lib/debuglocks.c @@ -0,0 +1,202 @@ +/* $Id: debuglocks.c,v 1.11 2001/09/20 00:35:31 davem Exp $ + * debuglocks.c: Debugging versions of SMP locking primitives. + * + * Copyright (C) 1997 David S. Miller (davem@caip.rutgers.edu) + * Copyright (C) 1998-99 Anton Blanchard (anton@progsoc.uts.edu.au) + */ + +#include <linux/kernel.h> +#include <linux/sched.h> +#include <linux/threads.h> /* For NR_CPUS */ +#include <linux/spinlock.h> +#include <asm/psr.h> +#include <asm/system.h> + +#ifdef CONFIG_SMP + +/* Some notes on how these debugging routines work. When a lock is acquired + * an extra debugging member lock->owner_pc is set to the caller of the lock + * acquisition routine. Right before releasing a lock, the debugging program + * counter is cleared to zero. + * + * Furthermore, since PC's are 4 byte aligned on Sparc, we stuff the CPU + * number of the owner in the lowest two bits. + */ + +#define STORE_CALLER(A) __asm__ __volatile__("mov %%i7, %0" : "=r" (A)); + +static inline void show(char *str, spinlock_t *lock, unsigned long caller) +{ + int cpu = smp_processor_id(); + + printk("%s(%p) CPU#%d stuck at %08lx, owner PC(%08lx):CPU(%lx)\n",str, + lock, cpu, caller, lock->owner_pc & ~3, lock->owner_pc & 3); +} + +static inline void show_read(char *str, rwlock_t *lock, unsigned long caller) +{ + int cpu = smp_processor_id(); + + printk("%s(%p) CPU#%d stuck at %08lx, owner PC(%08lx):CPU(%lx)\n", str, + lock, cpu, caller, lock->owner_pc & ~3, lock->owner_pc & 3); +} + +static inline void show_write(char *str, rwlock_t *lock, unsigned long caller) +{ + int cpu = smp_processor_id(); + int i; + + printk("%s(%p) CPU#%d stuck at %08lx, owner PC(%08lx):CPU(%lx)", str, + lock, cpu, caller, lock->owner_pc & ~3, lock->owner_pc & 3); + + for(i = 0; i < NR_CPUS; i++) + printk(" reader[%d]=%08lx", i, lock->reader_pc[i]); + + printk("\n"); +} + +#undef INIT_STUCK +#define INIT_STUCK 100000000 + +void _do_spin_lock(spinlock_t *lock, char *str) +{ + unsigned long caller; + unsigned long val; + int cpu = smp_processor_id(); + int stuck = INIT_STUCK; + + STORE_CALLER(caller); + +again: + __asm__ __volatile__("ldstub [%1], %0" : "=r" (val) : "r" (&(lock->lock))); + if(val) { + while(lock->lock) { + if (!--stuck) { + show(str, lock, caller); + stuck = INIT_STUCK; + } + barrier(); + } + goto again; + } + lock->owner_pc = (cpu & 3) | (caller & ~3); +} + +int _spin_trylock(spinlock_t *lock) +{ + unsigned long val; + unsigned long caller; + int cpu = smp_processor_id(); + + STORE_CALLER(caller); + + __asm__ __volatile__("ldstub [%1], %0" : "=r" (val) : "r" (&(lock->lock))); + if(!val) { + /* We got it, record our identity for debugging. */ + lock->owner_pc = (cpu & 3) | (caller & ~3); + } + return val == 0; +} + +void _do_spin_unlock(spinlock_t *lock) +{ + lock->owner_pc = 0; + barrier(); + lock->lock = 0; +} + +void _do_read_lock(rwlock_t *rw, char *str) +{ + unsigned long caller; + unsigned long val; + int cpu = smp_processor_id(); + int stuck = INIT_STUCK; + + STORE_CALLER(caller); + +wlock_again: + __asm__ __volatile__("ldstub [%1 + 3], %0" : "=r" (val) : "r" (&(rw->lock))); + if(val) { + while(rw->lock & 0xff) { + if (!--stuck) { + show_read(str, rw, caller); + stuck = INIT_STUCK; + } + barrier(); + } + goto wlock_again; + } + + rw->reader_pc[cpu] = caller; + barrier(); + rw->lock++; +} + +void _do_read_unlock(rwlock_t *rw, char *str) +{ + unsigned long caller; + unsigned long val; + int cpu = smp_processor_id(); + int stuck = INIT_STUCK; + + STORE_CALLER(caller); + +wlock_again: + __asm__ __volatile__("ldstub [%1 + 3], %0" : "=r" (val) : "r" (&(rw->lock))); + if(val) { + while(rw->lock & 0xff) { + if (!--stuck) { + show_read(str, rw, caller); + stuck = INIT_STUCK; + } + barrier(); + } + goto wlock_again; + } + + rw->reader_pc[cpu] = 0; + barrier(); + rw->lock -= 0x1ff; +} + +void _do_write_lock(rwlock_t *rw, char *str) +{ + unsigned long caller; + unsigned long val; + int cpu = smp_processor_id(); + int stuck = INIT_STUCK; + + STORE_CALLER(caller); + +wlock_again: + __asm__ __volatile__("ldstub [%1 + 3], %0" : "=r" (val) : "r" (&(rw->lock))); + if(val) { +wlock_wait: + while(rw->lock) { + if (!--stuck) { + show_write(str, rw, caller); + stuck = INIT_STUCK; + } + barrier(); + } + goto wlock_again; + } + + if (rw->lock & ~0xff) { + *(((unsigned char *)&rw->lock)+3) = 0; + barrier(); + goto wlock_wait; + } + + barrier(); + rw->owner_pc = (cpu & 3) | (caller & ~3); +} + +void _do_write_unlock(rwlock_t *rw) +{ + rw->owner_pc = 0; + barrier(); + rw->lock = 0; +} + +#endif /* SMP */ diff --git a/arch/sparc/lib/divdi3.S b/arch/sparc/lib/divdi3.S new file mode 100644 index 00000000000..681b3683da9 --- /dev/null +++ b/arch/sparc/lib/divdi3.S @@ -0,0 +1,295 @@ +/* Copyright (C) 1989, 1992, 1993, 1994, 1995 Free Software Foundation, Inc. + +This file is part of GNU CC. + +GNU CC is free software; you can redistribute it and/or modify +it under the terms of the GNU General Public License as published by +the Free Software Foundation; either version 2, or (at your option) +any later version. + +GNU CC 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 General Public License for more details. + +You should have received a copy of the GNU General Public License +along with GNU CC; see the file COPYING. If not, write to +the Free Software Foundation, 59 Temple Place - Suite 330, +Boston, MA 02111-1307, USA. */ + + .data + .align 8 + .globl __clz_tab +__clz_tab: + .byte 0,1,2,2,3,3,3,3,4,4,4,4,4,4,4,4,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5 + .byte 6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6 + .byte 7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7 + .byte 7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7 + .byte 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8 + .byte 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8 + .byte 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8 + .byte 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8 + .size __clz_tab,256 + .global .udiv + + .text + .align 4 + .globl __divdi3 +__divdi3: + save %sp,-104,%sp + cmp %i0,0 + bge .LL40 + mov 0,%l4 + mov -1,%l4 + sub %g0,%i1,%o0 + mov %o0,%o5 + subcc %g0,%o0,%g0 + sub %g0,%i0,%o0 + subx %o0,0,%o4 + mov %o4,%i0 + mov %o5,%i1 +.LL40: + cmp %i2,0 + bge .LL84 + mov %i3,%o4 + xnor %g0,%l4,%l4 + sub %g0,%i3,%o0 + mov %o0,%o3 + subcc %g0,%o0,%g0 + sub %g0,%i2,%o0 + subx %o0,0,%o2 + mov %o2,%i2 + mov %o3,%i3 + mov %i3,%o4 +.LL84: + cmp %i2,0 + bne .LL45 + mov %i1,%i3 + cmp %o4,%i0 + bleu .LL46 + mov %i3,%o1 + mov 32,%g1 + subcc %i0,%o4,%g0 +1: bcs 5f + addxcc %o1,%o1,%o1 ! shift n1n0 and a q-bit in lsb + sub %i0,%o4,%i0 ! this kills msb of n + addx %i0,%i0,%i0 ! so this cannot give carry + subcc %g1,1,%g1 +2: bne 1b + subcc %i0,%o4,%g0 + bcs 3f + addxcc %o1,%o1,%o1 ! shift n1n0 and a q-bit in lsb + b 3f + sub %i0,%o4,%i0 ! this kills msb of n +4: sub %i0,%o4,%i0 +5: addxcc %i0,%i0,%i0 + bcc 2b + subcc %g1,1,%g1 +! Got carry from n. Subtract next step to cancel this carry. + bne 4b + addcc %o1,%o1,%o1 ! shift n1n0 and a 0-bit in lsb + sub %i0,%o4,%i0 +3: xnor %o1,0,%o1 + b .LL50 + mov 0,%o2 +.LL46: + cmp %o4,0 + bne .LL85 + mov %i0,%o2 + mov 1,%o0 + call .udiv,0 + mov 0,%o1 + mov %o0,%o4 + mov %i0,%o2 +.LL85: + mov 0,%g3 + mov 32,%g1 + subcc %g3,%o4,%g0 +1: bcs 5f + addxcc %o2,%o2,%o2 ! shift n1n0 and a q-bit in lsb + sub %g3,%o4,%g3 ! this kills msb of n + addx %g3,%g3,%g3 ! so this cannot give carry + subcc %g1,1,%g1 +2: bne 1b + subcc %g3,%o4,%g0 + bcs 3f + addxcc %o2,%o2,%o2 ! shift n1n0 and a q-bit in lsb + b 3f + sub %g3,%o4,%g3 ! this kills msb of n +4: sub %g3,%o4,%g3 +5: addxcc %g3,%g3,%g3 + bcc 2b + subcc %g1,1,%g1 +! Got carry from n. Subtract next step to cancel this carry. + bne 4b + addcc %o2,%o2,%o2 ! shift n1n0 and a 0-bit in lsb + sub %g3,%o4,%g3 +3: xnor %o2,0,%o2 + mov %g3,%i0 + mov %i3,%o1 + mov 32,%g1 + subcc %i0,%o4,%g0 +1: bcs 5f + addxcc %o1,%o1,%o1 ! shift n1n0 and a q-bit in lsb + sub %i0,%o4,%i0 ! this kills msb of n + addx %i0,%i0,%i0 ! so this cannot give carry + subcc %g1,1,%g1 +2: bne 1b + subcc %i0,%o4,%g0 + bcs 3f + addxcc %o1,%o1,%o1 ! shift n1n0 and a q-bit in lsb + b 3f + sub %i0,%o4,%i0 ! this kills msb of n +4: sub %i0,%o4,%i0 +5: addxcc %i0,%i0,%i0 + bcc 2b + subcc %g1,1,%g1 +! Got carry from n. Subtract next step to cancel this carry. + bne 4b + addcc %o1,%o1,%o1 ! shift n1n0 and a 0-bit in lsb + sub %i0,%o4,%i0 +3: xnor %o1,0,%o1 + b .LL86 + mov %o1,%l1 +.LL45: + cmp %i2,%i0 + bleu .LL51 + sethi %hi(65535),%o0 + b .LL78 + mov 0,%o1 +.LL51: + or %o0,%lo(65535),%o0 + cmp %i2,%o0 + bgu .LL58 + mov %i2,%o1 + cmp %i2,256 + addx %g0,-1,%o0 + b .LL64 + and %o0,8,%o2 +.LL58: + sethi %hi(16777215),%o0 + or %o0,%lo(16777215),%o0 + cmp %i2,%o0 + bgu .LL64 + mov 24,%o2 + mov 16,%o2 +.LL64: + srl %o1,%o2,%o0 + sethi %hi(__clz_tab),%o1 + or %o1,%lo(__clz_tab),%o1 + ldub [%o0+%o1],%o0 + add %o0,%o2,%o0 + mov 32,%o1 + subcc %o1,%o0,%o3 + bne,a .LL72 + sub %o1,%o3,%o1 + cmp %i0,%i2 + bgu .LL74 + cmp %i3,%o4 + blu .LL78 + mov 0,%o1 +.LL74: + b .LL78 + mov 1,%o1 +.LL72: + sll %i2,%o3,%o2 + srl %o4,%o1,%o0 + or %o2,%o0,%i2 + sll %o4,%o3,%o4 + srl %i0,%o1,%o2 + sll %i0,%o3,%o0 + srl %i3,%o1,%o1 + or %o0,%o1,%i0 + sll %i3,%o3,%i3 + mov %i0,%o1 + mov 32,%g1 + subcc %o2,%i2,%g0 +1: bcs 5f + addxcc %o1,%o1,%o1 ! shift n1n0 and a q-bit in lsb + sub %o2,%i2,%o2 ! this kills msb of n + addx %o2,%o2,%o2 ! so this cannot give carry + subcc %g1,1,%g1 +2: bne 1b + subcc %o2,%i2,%g0 + bcs 3f + addxcc %o1,%o1,%o1 ! shift n1n0 and a q-bit in lsb + b 3f + sub %o2,%i2,%o2 ! this kills msb of n +4: sub %o2,%i2,%o2 +5: addxcc %o2,%o2,%o2 + bcc 2b + subcc %g1,1,%g1 +! Got carry from n. Subtract next step to cancel this carry. + bne 4b + addcc %o1,%o1,%o1 ! shift n1n0 and a 0-bit in lsb + sub %o2,%i2,%o2 +3: xnor %o1,0,%o1 + mov %o2,%i0 + wr %g0,%o1,%y ! SPARC has 0-3 delay insn after a wr + sra %o4,31,%g2 ! Do not move this insn + and %o1,%g2,%g2 ! Do not move this insn + andcc %g0,0,%g1 ! Do not move this insn + mulscc %g1,%o4,%g1 + mulscc %g1,%o4,%g1 + mulscc %g1,%o4,%g1 + mulscc %g1,%o4,%g1 + mulscc %g1,%o4,%g1 + mulscc %g1,%o4,%g1 + mulscc %g1,%o4,%g1 + mulscc %g1,%o4,%g1 + mulscc %g1,%o4,%g1 + mulscc %g1,%o4,%g1 + mulscc %g1,%o4,%g1 + mulscc %g1,%o4,%g1 + mulscc %g1,%o4,%g1 + mulscc %g1,%o4,%g1 + mulscc %g1,%o4,%g1 + mulscc %g1,%o4,%g1 + mulscc %g1,%o4,%g1 + mulscc %g1,%o4,%g1 + mulscc %g1,%o4,%g1 + mulscc %g1,%o4,%g1 + mulscc %g1,%o4,%g1 + mulscc %g1,%o4,%g1 + mulscc %g1,%o4,%g1 + mulscc %g1,%o4,%g1 + mulscc %g1,%o4,%g1 + mulscc %g1,%o4,%g1 + mulscc %g1,%o4,%g1 + mulscc %g1,%o4,%g1 + mulscc %g1,%o4,%g1 + mulscc %g1,%o4,%g1 + mulscc %g1,%o4,%g1 + mulscc %g1,%o4,%g1 + mulscc %g1,0,%g1 + add %g1,%g2,%o0 + rd %y,%o2 + cmp %o0,%i0 + bgu,a .LL78 + add %o1,-1,%o1 + bne,a .LL50 + mov 0,%o2 + cmp %o2,%i3 + bleu .LL50 + mov 0,%o2 + add %o1,-1,%o1 +.LL78: + mov 0,%o2 +.LL50: + mov %o1,%l1 +.LL86: + mov %o2,%l0 + mov %l0,%i0 + mov %l1,%i1 + cmp %l4,0 + be .LL81 + sub %g0,%i1,%o0 + mov %o0,%l3 + subcc %g0,%o0,%g0 + sub %g0,%i0,%o0 + subx %o0,0,%l2 + mov %l2,%i0 + mov %l3,%i1 +.LL81: + ret + restore diff --git a/arch/sparc/lib/locks.S b/arch/sparc/lib/locks.S new file mode 100644 index 00000000000..95fa4842496 --- /dev/null +++ b/arch/sparc/lib/locks.S @@ -0,0 +1,72 @@ +/* $Id: locks.S,v 1.16 2000/02/26 11:02:47 anton Exp $ + * locks.S: SMP low-level lock primitives on Sparc. + * + * Copyright (C) 1996 David S. Miller (davem@caip.rutgers.edu) + * Copyright (C) 1998 Anton Blanchard (anton@progsoc.uts.edu.au) + * Copyright (C) 1998 Jakub Jelinek (jj@ultra.linux.cz) + */ + +#include <asm/ptrace.h> +#include <asm/psr.h> +#include <asm/smp.h> +#include <asm/spinlock.h> + + .text + .align 4 + + /* Read/writer locks, as usual this is overly clever to make it + * as fast as possible. + */ + + /* caches... */ +___rw_read_enter_spin_on_wlock: + orcc %g2, 0x0, %g0 + be,a ___rw_read_enter + ldstub [%g1 + 3], %g2 + b ___rw_read_enter_spin_on_wlock + ldub [%g1 + 3], %g2 +___rw_read_exit_spin_on_wlock: + orcc %g2, 0x0, %g0 + be,a ___rw_read_exit + ldstub [%g1 + 3], %g2 + b ___rw_read_exit_spin_on_wlock + ldub [%g1 + 3], %g2 +___rw_write_enter_spin_on_wlock: + orcc %g2, 0x0, %g0 + be,a ___rw_write_enter + ldstub [%g1 + 3], %g2 + b ___rw_write_enter_spin_on_wlock + ld [%g1], %g2 + + .globl ___rw_read_enter +___rw_read_enter: + orcc %g2, 0x0, %g0 + bne,a ___rw_read_enter_spin_on_wlock + ldub [%g1 + 3], %g2 + ld [%g1], %g2 + add %g2, 1, %g2 + st %g2, [%g1] + retl + mov %g4, %o7 + + .globl ___rw_read_exit +___rw_read_exit: + orcc %g2, 0x0, %g0 + bne,a ___rw_read_exit_spin_on_wlock + ldub [%g1 + 3], %g2 + ld [%g1], %g2 + sub %g2, 0x1ff, %g2 + st %g2, [%g1] + retl + mov %g4, %o7 + + .globl ___rw_write_enter +___rw_write_enter: + orcc %g2, 0x0, %g0 + bne ___rw_write_enter_spin_on_wlock + ld [%g1], %g2 + andncc %g2, 0xff, %g0 + bne,a ___rw_write_enter_spin_on_wlock + stb %g0, [%g1 + 3] + retl + mov %g4, %o7 diff --git a/arch/sparc/lib/lshrdi3.S b/arch/sparc/lib/lshrdi3.S new file mode 100644 index 00000000000..35abf5b2bd1 --- /dev/null +++ b/arch/sparc/lib/lshrdi3.S @@ -0,0 +1,27 @@ +/* $Id: lshrdi3.S,v 1.1 1999/03/21 06:37:45 davem Exp $ */ + + .globl __lshrdi3 +__lshrdi3: + cmp %o2, 0 + be 3f + mov 0x20, %g2 + + sub %g2, %o2, %g2 + cmp %g2, 0 + bg 1f + srl %o0, %o2, %o4 + + clr %o4 + neg %g2 + b 2f + srl %o0, %g2, %o5 +1: + sll %o0, %g2, %g3 + srl %o1, %o2, %g2 + or %g2, %g3, %o5 +2: + mov %o4, %o0 + mov %o5, %o1 +3: + retl + nop diff --git a/arch/sparc/lib/memcmp.S b/arch/sparc/lib/memcmp.S new file mode 100644 index 00000000000..cb4bdb0cc2a --- /dev/null +++ b/arch/sparc/lib/memcmp.S @@ -0,0 +1,312 @@ + .text + .align 4 + .global __memcmp, memcmp +__memcmp: +memcmp: +#if 1 + cmp %o2, 0 + ble L3 + mov 0, %g3 +L5: + ldub [%o0], %g2 + ldub [%o1], %g3 + sub %g2, %g3, %g2 + mov %g2, %g3 + sll %g2, 24, %g2 + + cmp %g2, 0 + bne L3 + add %o0, 1, %o0 + + add %o2, -1, %o2 + + cmp %o2, 0 + bg L5 + add %o1, 1, %o1 +L3: + sll %g3, 24, %o0 + sra %o0, 24, %o0 + + retl + nop +#else + save %sp, -104, %sp + mov %i2, %o4 + mov %i0, %o0 + + cmp %o4, 15 + ble L72 + mov %i1, %i2 + + andcc %i2, 3, %g0 + be L161 + andcc %o0, 3, %g2 +L75: + ldub [%o0], %g3 + ldub [%i2], %g2 + add %o0,1, %o0 + + subcc %g3, %g2, %i0 + bne L156 + add %i2, 1, %i2 + + andcc %i2, 3, %g0 + bne L75 + add %o4, -1, %o4 + + andcc %o0, 3, %g2 +L161: + bne,a L78 + mov %i2, %i1 + + mov %o0, %i5 + mov %i2, %i3 + srl %o4, 2, %i4 + + cmp %i4, 0 + bge L93 + mov %i4, %g2 + + add %i4, 3, %g2 +L93: + sra %g2, 2, %g2 + sll %g2, 2, %g2 + sub %i4, %g2, %g2 + + cmp %g2, 1 + be,a L88 + add %o0, 4, %i5 + + bg L94 + cmp %g2, 2 + + cmp %g2, 0 + be,a L86 + ld [%o0], %g3 + + b L162 + ld [%i5], %g3 +L94: + be L81 + cmp %g2, 3 + + be,a L83 + add %o0, -4, %i5 + + b L162 + ld [%i5], %g3 +L81: + add %o0, -8, %i5 + ld [%o0], %g3 + add %i2, -8, %i3 + ld [%i2], %g2 + + b L82 + add %i4, 2, %i4 +L83: + ld [%o0], %g4 + add %i2, -4, %i3 + ld [%i2], %g1 + + b L84 + add %i4, 1, %i4 +L86: + b L87 + ld [%i2], %g2 +L88: + add %i2, 4, %i3 + ld [%o0], %g4 + add %i4, -1, %i4 + ld [%i2], %g1 +L95: + ld [%i5], %g3 +L162: + cmp %g4, %g1 + be L87 + ld [%i3], %g2 + + cmp %g4, %g1 +L163: + bleu L114 + mov -1, %i0 + + b L114 + mov 1, %i0 +L87: + ld [%i5 + 4], %g4 + cmp %g3, %g2 + bne L163 + ld [%i3 + 4], %g1 +L84: + ld [%i5 + 8], %g3 + + cmp %g4, %g1 + bne L163 + ld [%i3 + 8], %g2 +L82: + ld [%i5 + 12], %g4 + cmp %g3, %g2 + bne L163 + ld [%i3 + 12], %g1 + + add %i5, 16, %i5 + + addcc %i4, -4, %i4 + bne L95 + add %i3, 16, %i3 + + cmp %g4, %g1 + bne L163 + nop + + b L114 + mov 0, %i0 +L78: + srl %o4, 2, %i0 + and %o0, -4, %i3 + orcc %i0, %g0, %g3 + sll %g2, 3, %o7 + mov 32, %g2 + + bge L129 + sub %g2, %o7, %o1 + + add %i0, 3, %g3 +L129: + sra %g3, 2, %g2 + sll %g2, 2, %g2 + sub %i0, %g2, %g2 + + cmp %g2, 1 + be,a L124 + ld [%i3], %o3 + + bg L130 + cmp %g2, 2 + + cmp %g2, 0 + be,a L122 + ld [%i3], %o2 + + b L164 + sll %o3, %o7, %g3 +L130: + be L117 + cmp %g2, 3 + + be,a L119 + ld [%i3], %g1 + + b L164 + sll %o3, %o7, %g3 +L117: + ld [%i3], %g4 + add %i2, -8, %i1 + ld [%i3 + 4], %o3 + add %i0, 2, %i0 + ld [%i2], %i4 + + b L118 + add %i3, -4, %i3 +L119: + ld [%i3 + 4], %g4 + add %i2, -4, %i1 + ld [%i2], %i5 + + b L120 + add %i0, 1, %i0 +L122: + ld [%i3 + 4], %g1 + ld [%i2], %i4 + + b L123 + add %i3, 4, %i3 +L124: + add %i2, 4, %i1 + ld [%i3 + 4], %o2 + add %i0, -1, %i0 + ld [%i2], %i5 + add %i3, 8, %i3 +L131: + sll %o3, %o7, %g3 +L164: + srl %o2, %o1, %g2 + ld [%i3], %g1 + or %g3, %g2, %g3 + + cmp %g3, %i5 + bne L163 + ld [%i1], %i4 +L123: + sll %o2, %o7, %g3 + srl %g1, %o1, %g2 + ld [%i3 + 4], %g4 + or %g3, %g2, %g3 + + cmp %g3, %i4 + bne L163 + ld [%i1 + 4], %i5 +L120: + sll %g1, %o7, %g3 + srl %g4, %o1, %g2 + ld [%i3 + 8], %o3 + or %g3, %g2, %g3 + + cmp %g3, %i5 + bne L163 + ld [%i1 + 8], %i4 +L118: + sll %g4, %o7, %g3 + srl %o3, %o1, %g2 + ld [%i3 + 12], %o2 + or %g3, %g2, %g3 + + cmp %g3, %i4 + bne L163 + ld [%i1 + 12], %i5 + + add %i3, 16, %i3 + addcc %i0, -4, %i0 + bne L131 + add %i1, 16, %i1 + + sll %o3, %o7, %g3 + srl %o2, %o1, %g2 + or %g3, %g2, %g3 + + cmp %g3, %i5 + be,a L114 + mov 0, %i0 + + b,a L163 +L114: + cmp %i0, 0 + bne L156 + and %o4, -4, %g2 + + add %o0, %g2, %o0 + add %i2, %g2, %i2 + and %o4, 3, %o4 +L72: + cmp %o4, 0 + be L156 + mov 0, %i0 + + ldub [%o0], %g3 +L165: + ldub [%i2], %g2 + add %o0, 1, %o0 + + subcc %g3, %g2, %i0 + bne L156 + add %i2, 1, %i2 + + addcc %o4, -1, %o4 + bne,a L165 + ldub [%o0], %g3 + + mov 0, %i0 +L156: + ret + restore +#endif diff --git a/arch/sparc/lib/memcpy.S b/arch/sparc/lib/memcpy.S new file mode 100644 index 00000000000..ce10bc869af --- /dev/null +++ b/arch/sparc/lib/memcpy.S @@ -0,0 +1,1150 @@ +/* memcpy.S: Sparc optimized memcpy and memmove code + * Hand optimized from GNU libc's memcpy and memmove + * Copyright (C) 1991,1996 Free Software Foundation + * Copyright (C) 1995 Linus Torvalds (Linus.Torvalds@helsinki.fi) + * Copyright (C) 1996 David S. Miller (davem@caip.rutgers.edu) + * Copyright (C) 1996 Eddie C. Dost (ecd@skynet.be) + * Copyright (C) 1996 Jakub Jelinek (jj@sunsite.mff.cuni.cz) + */ + +#ifdef __KERNEL__ + +#define FUNC(x) \ + .globl x; \ + .type x,@function; \ + .align 4; \ +x: + +#undef FASTER_REVERSE +#undef FASTER_NONALIGNED +#define FASTER_ALIGNED + +/* In kernel these functions don't return a value. + * One should use macros in asm/string.h for that purpose. + * We return 0, so that bugs are more apparent. + */ +#define SETUP_RETL +#define RETL_INSN clr %o0 + +#else + +/* libc */ + +#include "DEFS.h" + +#define FASTER_REVERSE +#define FASTER_NONALIGNED +#define FASTER_ALIGNED + +#define SETUP_RETL mov %o0, %g6 +#define RETL_INSN mov %g6, %o0 + +#endif + +/* Both these macros have to start with exactly the same insn */ +#define MOVE_BIGCHUNK(src, dst, offset, t0, t1, t2, t3, t4, t5, t6, t7) \ + ldd [%src + (offset) + 0x00], %t0; \ + ldd [%src + (offset) + 0x08], %t2; \ + ldd [%src + (offset) + 0x10], %t4; \ + ldd [%src + (offset) + 0x18], %t6; \ + st %t0, [%dst + (offset) + 0x00]; \ + st %t1, [%dst + (offset) + 0x04]; \ + st %t2, [%dst + (offset) + 0x08]; \ + st %t3, [%dst + (offset) + 0x0c]; \ + st %t4, [%dst + (offset) + 0x10]; \ + st %t5, [%dst + (offset) + 0x14]; \ + st %t6, [%dst + (offset) + 0x18]; \ + st %t7, [%dst + (offset) + 0x1c]; + +#define MOVE_BIGALIGNCHUNK(src, dst, offset, t0, t1, t2, t3, t4, t5, t6, t7) \ + ldd [%src + (offset) + 0x00], %t0; \ + ldd [%src + (offset) + 0x08], %t2; \ + ldd [%src + (offset) + 0x10], %t4; \ + ldd [%src + (offset) + 0x18], %t6; \ + std %t0, [%dst + (offset) + 0x00]; \ + std %t2, [%dst + (offset) + 0x08]; \ + std %t4, [%dst + (offset) + 0x10]; \ + std %t6, [%dst + (offset) + 0x18]; + +#define MOVE_LASTCHUNK(src, dst, offset, t0, t1, t2, t3) \ + ldd [%src - (offset) - 0x10], %t0; \ + ldd [%src - (offset) - 0x08], %t2; \ + st %t0, [%dst - (offset) - 0x10]; \ + st %t1, [%dst - (offset) - 0x0c]; \ + st %t2, [%dst - (offset) - 0x08]; \ + st %t3, [%dst - (offset) - 0x04]; + +#define MOVE_LASTALIGNCHUNK(src, dst, offset, t0, t1, t2, t3) \ + ldd [%src - (offset) - 0x10], %t0; \ + ldd [%src - (offset) - 0x08], %t2; \ + std %t0, [%dst - (offset) - 0x10]; \ + std %t2, [%dst - (offset) - 0x08]; + +#define MOVE_SHORTCHUNK(src, dst, offset, t0, t1) \ + ldub [%src - (offset) - 0x02], %t0; \ + ldub [%src - (offset) - 0x01], %t1; \ + stb %t0, [%dst - (offset) - 0x02]; \ + stb %t1, [%dst - (offset) - 0x01]; + +/* Both these macros have to start with exactly the same insn */ +#define RMOVE_BIGCHUNK(src, dst, offset, t0, t1, t2, t3, t4, t5, t6, t7) \ + ldd [%src - (offset) - 0x20], %t0; \ + ldd [%src - (offset) - 0x18], %t2; \ + ldd [%src - (offset) - 0x10], %t4; \ + ldd [%src - (offset) - 0x08], %t6; \ + st %t0, [%dst - (offset) - 0x20]; \ + st %t1, [%dst - (offset) - 0x1c]; \ + st %t2, [%dst - (offset) - 0x18]; \ + st %t3, [%dst - (offset) - 0x14]; \ + st %t4, [%dst - (offset) - 0x10]; \ + st %t5, [%dst - (offset) - 0x0c]; \ + st %t6, [%dst - (offset) - 0x08]; \ + st %t7, [%dst - (offset) - 0x04]; + +#define RMOVE_BIGALIGNCHUNK(src, dst, offset, t0, t1, t2, t3, t4, t5, t6, t7) \ + ldd [%src - (offset) - 0x20], %t0; \ + ldd [%src - (offset) - 0x18], %t2; \ + ldd [%src - (offset) - 0x10], %t4; \ + ldd [%src - (offset) - 0x08], %t6; \ + std %t0, [%dst - (offset) - 0x20]; \ + std %t2, [%dst - (offset) - 0x18]; \ + std %t4, [%dst - (offset) - 0x10]; \ + std %t6, [%dst - (offset) - 0x08]; + +#define RMOVE_LASTCHUNK(src, dst, offset, t0, t1, t2, t3) \ + ldd [%src + (offset) + 0x00], %t0; \ + ldd [%src + (offset) + 0x08], %t2; \ + st %t0, [%dst + (offset) + 0x00]; \ + st %t1, [%dst + (offset) + 0x04]; \ + st %t2, [%dst + (offset) + 0x08]; \ + st %t3, [%dst + (offset) + 0x0c]; + +#define RMOVE_SHORTCHUNK(src, dst, offset, t0, t1) \ + ldub [%src + (offset) + 0x00], %t0; \ + ldub [%src + (offset) + 0x01], %t1; \ + stb %t0, [%dst + (offset) + 0x00]; \ + stb %t1, [%dst + (offset) + 0x01]; + +#define SMOVE_CHUNK(src, dst, offset, t0, t1, t2, t3, t4, t5, t6, prev, shil, shir, offset2) \ + ldd [%src + (offset) + 0x00], %t0; \ + ldd [%src + (offset) + 0x08], %t2; \ + srl %t0, shir, %t5; \ + srl %t1, shir, %t6; \ + sll %t0, shil, %t0; \ + or %t5, %prev, %t5; \ + sll %t1, shil, %prev; \ + or %t6, %t0, %t0; \ + srl %t2, shir, %t1; \ + srl %t3, shir, %t6; \ + sll %t2, shil, %t2; \ + or %t1, %prev, %t1; \ + std %t4, [%dst + (offset) + (offset2) - 0x04]; \ + std %t0, [%dst + (offset) + (offset2) + 0x04]; \ + sll %t3, shil, %prev; \ + or %t6, %t2, %t4; + +#define SMOVE_ALIGNCHUNK(src, dst, offset, t0, t1, t2, t3, t4, t5, t6, prev, shil, shir, offset2) \ + ldd [%src + (offset) + 0x00], %t0; \ + ldd [%src + (offset) + 0x08], %t2; \ + srl %t0, shir, %t4; \ + srl %t1, shir, %t5; \ + sll %t0, shil, %t6; \ + or %t4, %prev, %t0; \ + sll %t1, shil, %prev; \ + or %t5, %t6, %t1; \ + srl %t2, shir, %t4; \ + srl %t3, shir, %t5; \ + sll %t2, shil, %t6; \ + or %t4, %prev, %t2; \ + sll %t3, shil, %prev; \ + or %t5, %t6, %t3; \ + std %t0, [%dst + (offset) + (offset2) + 0x00]; \ + std %t2, [%dst + (offset) + (offset2) + 0x08]; + + .text + .align 4 + +#ifdef FASTER_REVERSE + +70: /* rdword_align */ + + andcc %o1, 1, %g0 + be 4f + andcc %o1, 2, %g0 + + ldub [%o1 - 1], %g2 + sub %o1, 1, %o1 + stb %g2, [%o0 - 1] + sub %o2, 1, %o2 + be 3f + sub %o0, 1, %o0 +4: + lduh [%o1 - 2], %g2 + sub %o1, 2, %o1 + sth %g2, [%o0 - 2] + sub %o2, 2, %o2 + b 3f + sub %o0, 2, %o0 + +#endif /* FASTER_REVERSE */ + +0: + retl + nop ! Only bcopy returns here and it retuns void... + +#ifdef __KERNEL__ +FUNC(amemmove) +FUNC(__memmove) +#endif +FUNC(memmove) + cmp %o0, %o1 + SETUP_RETL + bleu 9f + sub %o0, %o1, %o4 + + add %o1, %o2, %o3 + cmp %o3, %o0 + bleu 0f + andcc %o4, 3, %o5 + +#ifndef FASTER_REVERSE + + add %o1, %o2, %o1 + add %o0, %o2, %o0 + sub %o1, 1, %o1 + sub %o0, 1, %o0 + +1: /* reverse_bytes */ + + ldub [%o1], %o4 + subcc %o2, 1, %o2 + stb %o4, [%o0] + sub %o1, 1, %o1 + bne 1b + sub %o0, 1, %o0 + + retl + RETL_INSN + +#else /* FASTER_REVERSE */ + + add %o1, %o2, %o1 + add %o0, %o2, %o0 + bne 77f + cmp %o2, 15 + bleu 91f + andcc %o1, 3, %g0 + bne 70b +3: + andcc %o1, 4, %g0 + + be 2f + mov %o2, %g1 + + ld [%o1 - 4], %o4 + sub %g1, 4, %g1 + st %o4, [%o0 - 4] + sub %o1, 4, %o1 + sub %o0, 4, %o0 +2: + andcc %g1, 0xffffff80, %g7 + be 3f + andcc %o0, 4, %g0 + + be 74f + 4 +5: + RMOVE_BIGCHUNK(o1, o0, 0x00, o2, o3, o4, o5, g2, g3, g4, g5) + RMOVE_BIGCHUNK(o1, o0, 0x20, o2, o3, o4, o5, g2, g3, g4, g5) + RMOVE_BIGCHUNK(o1, o0, 0x40, o2, o3, o4, o5, g2, g3, g4, g5) + RMOVE_BIGCHUNK(o1, o0, 0x60, o2, o3, o4, o5, g2, g3, g4, g5) + subcc %g7, 128, %g7 + sub %o1, 128, %o1 + bne 5b + sub %o0, 128, %o0 +3: + andcc %g1, 0x70, %g7 + be 72f + andcc %g1, 8, %g0 + + sethi %hi(72f), %o5 + srl %g7, 1, %o4 + add %g7, %o4, %o4 + sub %o1, %g7, %o1 + sub %o5, %o4, %o5 + jmpl %o5 + %lo(72f), %g0 + sub %o0, %g7, %o0 + +71: /* rmemcpy_table */ + RMOVE_LASTCHUNK(o1, o0, 0x60, g2, g3, g4, g5) + RMOVE_LASTCHUNK(o1, o0, 0x50, g2, g3, g4, g5) + RMOVE_LASTCHUNK(o1, o0, 0x40, g2, g3, g4, g5) + RMOVE_LASTCHUNK(o1, o0, 0x30, g2, g3, g4, g5) + RMOVE_LASTCHUNK(o1, o0, 0x20, g2, g3, g4, g5) + RMOVE_LASTCHUNK(o1, o0, 0x10, g2, g3, g4, g5) + RMOVE_LASTCHUNK(o1, o0, 0x00, g2, g3, g4, g5) + +72: /* rmemcpy_table_end */ + + be 73f + andcc %g1, 4, %g0 + + ldd [%o1 - 0x08], %g2 + sub %o0, 8, %o0 + sub %o1, 8, %o1 + st %g2, [%o0] + st %g3, [%o0 + 0x04] + +73: /* rmemcpy_last7 */ + + be 1f + andcc %g1, 2, %g0 + + ld [%o1 - 4], %g2 + sub %o1, 4, %o1 + st %g2, [%o0 - 4] + sub %o0, 4, %o0 +1: + be 1f + andcc %g1, 1, %g0 + + lduh [%o1 - 2], %g2 + sub %o1, 2, %o1 + sth %g2, [%o0 - 2] + sub %o0, 2, %o0 +1: + be 1f + nop + + ldub [%o1 - 1], %g2 + stb %g2, [%o0 - 1] +1: + retl + RETL_INSN + +74: /* rldd_std */ + RMOVE_BIGALIGNCHUNK(o1, o0, 0x00, o2, o3, o4, o5, g2, g3, g4, g5) + RMOVE_BIGALIGNCHUNK(o1, o0, 0x20, o2, o3, o4, o5, g2, g3, g4, g5) + RMOVE_BIGALIGNCHUNK(o1, o0, 0x40, o2, o3, o4, o5, g2, g3, g4, g5) + RMOVE_BIGALIGNCHUNK(o1, o0, 0x60, o2, o3, o4, o5, g2, g3, g4, g5) + subcc %g7, 128, %g7 + sub %o1, 128, %o1 + bne 74b + sub %o0, 128, %o0 + + andcc %g1, 0x70, %g7 + be 72b + andcc %g1, 8, %g0 + + sethi %hi(72b), %o5 + srl %g7, 1, %o4 + add %g7, %o4, %o4 + sub %o1, %g7, %o1 + sub %o5, %o4, %o5 + jmpl %o5 + %lo(72b), %g0 + sub %o0, %g7, %o0 + +75: /* rshort_end */ + + and %o2, 0xe, %o3 +2: + sethi %hi(76f), %o5 + sll %o3, 3, %o4 + sub %o0, %o3, %o0 + sub %o5, %o4, %o5 + sub %o1, %o3, %o1 + jmpl %o5 + %lo(76f), %g0 + andcc %o2, 1, %g0 + + RMOVE_SHORTCHUNK(o1, o0, 0x0c, g2, g3) + RMOVE_SHORTCHUNK(o1, o0, 0x0a, g2, g3) + RMOVE_SHORTCHUNK(o1, o0, 0x08, g2, g3) + RMOVE_SHORTCHUNK(o1, o0, 0x06, g2, g3) + RMOVE_SHORTCHUNK(o1, o0, 0x04, g2, g3) + RMOVE_SHORTCHUNK(o1, o0, 0x02, g2, g3) + RMOVE_SHORTCHUNK(o1, o0, 0x00, g2, g3) + +76: /* rshort_table_end */ + + be 1f + nop + ldub [%o1 - 1], %g2 + stb %g2, [%o0 - 1] +1: + retl + RETL_INSN + +91: /* rshort_aligned_end */ + + bne 75b + andcc %o2, 8, %g0 + + be 1f + andcc %o2, 4, %g0 + + ld [%o1 - 0x08], %g2 + ld [%o1 - 0x04], %g3 + sub %o1, 8, %o1 + st %g2, [%o0 - 0x08] + st %g3, [%o0 - 0x04] + sub %o0, 8, %o0 +1: + b 73b + mov %o2, %g1 + +77: /* rnon_aligned */ + cmp %o2, 15 + bleu 75b + andcc %o0, 3, %g0 + be 64f + andcc %o0, 1, %g0 + be 63f + andcc %o0, 2, %g0 + ldub [%o1 - 1], %g5 + sub %o1, 1, %o1 + stb %g5, [%o0 - 1] + sub %o0, 1, %o0 + be 64f + sub %o2, 1, %o2 +63: + ldub [%o1 - 1], %g5 + sub %o1, 2, %o1 + stb %g5, [%o0 - 1] + sub %o0, 2, %o0 + ldub [%o1], %g5 + sub %o2, 2, %o2 + stb %g5, [%o0] +64: + and %o1, 3, %g2 + and %o1, -4, %o1 + and %o2, 0xc, %g3 + add %o1, 4, %o1 + cmp %g3, 4 + sll %g2, 3, %g4 + mov 32, %g2 + be 4f + sub %g2, %g4, %g7 + + blu 3f + cmp %g3, 8 + + be 2f + srl %o2, 2, %g3 + + ld [%o1 - 4], %o3 + add %o0, -8, %o0 + ld [%o1 - 8], %o4 + add %o1, -16, %o1 + b 7f + add %g3, 1, %g3 +2: + ld [%o1 - 4], %o4 + add %o0, -4, %o0 + ld [%o1 - 8], %g1 + add %o1, -12, %o1 + b 8f + add %g3, 2, %g3 +3: + ld [%o1 - 4], %o5 + add %o0, -12, %o0 + ld [%o1 - 8], %o3 + add %o1, -20, %o1 + b 6f + srl %o2, 2, %g3 +4: + ld [%o1 - 4], %g1 + srl %o2, 2, %g3 + ld [%o1 - 8], %o5 + add %o1, -24, %o1 + add %o0, -16, %o0 + add %g3, -1, %g3 + + ld [%o1 + 12], %o3 +5: + sll %o5, %g4, %g2 + srl %g1, %g7, %g5 + or %g2, %g5, %g2 + st %g2, [%o0 + 12] +6: + ld [%o1 + 8], %o4 + sll %o3, %g4, %g2 + srl %o5, %g7, %g5 + or %g2, %g5, %g2 + st %g2, [%o0 + 8] +7: + ld [%o1 + 4], %g1 + sll %o4, %g4, %g2 + srl %o3, %g7, %g5 + or %g2, %g5, %g2 + st %g2, [%o0 + 4] +8: + ld [%o1], %o5 + sll %g1, %g4, %g2 + srl %o4, %g7, %g5 + addcc %g3, -4, %g3 + or %g2, %g5, %g2 + add %o1, -16, %o1 + st %g2, [%o0] + add %o0, -16, %o0 + bne,a 5b + ld [%o1 + 12], %o3 + sll %o5, %g4, %g2 + srl %g1, %g7, %g5 + srl %g4, 3, %g3 + or %g2, %g5, %g2 + add %o1, %g3, %o1 + andcc %o2, 2, %g0 + st %g2, [%o0 + 12] + be 1f + andcc %o2, 1, %g0 + + ldub [%o1 + 15], %g5 + add %o1, -2, %o1 + stb %g5, [%o0 + 11] + add %o0, -2, %o0 + ldub [%o1 + 16], %g5 + stb %g5, [%o0 + 12] +1: + be 1f + nop + ldub [%o1 + 15], %g5 + stb %g5, [%o0 + 11] +1: + retl + RETL_INSN + +#endif /* FASTER_REVERSE */ + +/* NOTE: This code is executed just for the cases, + where %src (=%o1) & 3 is != 0. + We need to align it to 4. So, for (%src & 3) + 1 we need to do ldub,lduh + 2 lduh + 3 just ldub + so even if it looks weird, the branches + are correct here. -jj + */ +78: /* dword_align */ + + andcc %o1, 1, %g0 + be 4f + andcc %o1, 2, %g0 + + ldub [%o1], %g2 + add %o1, 1, %o1 + stb %g2, [%o0] + sub %o2, 1, %o2 + bne 3f + add %o0, 1, %o0 +4: + lduh [%o1], %g2 + add %o1, 2, %o1 + sth %g2, [%o0] + sub %o2, 2, %o2 + b 3f + add %o0, 2, %o0 + +#ifdef __KERNEL__ +FUNC(__memcpy) +#endif +FUNC(memcpy) /* %o0=dst %o1=src %o2=len */ + + sub %o0, %o1, %o4 + SETUP_RETL +9: + andcc %o4, 3, %o5 +0: + bne 86f + cmp %o2, 15 + + bleu 90f + andcc %o1, 3, %g0 + + bne 78b +3: + andcc %o1, 4, %g0 + + be 2f + mov %o2, %g1 + + ld [%o1], %o4 + sub %g1, 4, %g1 + st %o4, [%o0] + add %o1, 4, %o1 + add %o0, 4, %o0 +2: + andcc %g1, 0xffffff80, %g7 + be 3f + andcc %o0, 4, %g0 + + be 82f + 4 +5: + MOVE_BIGCHUNK(o1, o0, 0x00, o2, o3, o4, o5, g2, g3, g4, g5) + MOVE_BIGCHUNK(o1, o0, 0x20, o2, o3, o4, o5, g2, g3, g4, g5) + MOVE_BIGCHUNK(o1, o0, 0x40, o2, o3, o4, o5, g2, g3, g4, g5) + MOVE_BIGCHUNK(o1, o0, 0x60, o2, o3, o4, o5, g2, g3, g4, g5) + subcc %g7, 128, %g7 + add %o1, 128, %o1 + bne 5b + add %o0, 128, %o0 +3: + andcc %g1, 0x70, %g7 + be 80f + andcc %g1, 8, %g0 + + sethi %hi(80f), %o5 + srl %g7, 1, %o4 + add %g7, %o4, %o4 + add %o1, %g7, %o1 + sub %o5, %o4, %o5 + jmpl %o5 + %lo(80f), %g0 + add %o0, %g7, %o0 + +79: /* memcpy_table */ + + MOVE_LASTCHUNK(o1, o0, 0x60, g2, g3, g4, g5) + MOVE_LASTCHUNK(o1, o0, 0x50, g2, g3, g4, g5) + MOVE_LASTCHUNK(o1, o0, 0x40, g2, g3, g4, g5) + MOVE_LASTCHUNK(o1, o0, 0x30, g2, g3, g4, g5) + MOVE_LASTCHUNK(o1, o0, 0x20, g2, g3, g4, g5) + MOVE_LASTCHUNK(o1, o0, 0x10, g2, g3, g4, g5) + MOVE_LASTCHUNK(o1, o0, 0x00, g2, g3, g4, g5) + +80: /* memcpy_table_end */ + be 81f + andcc %g1, 4, %g0 + + ldd [%o1], %g2 + add %o0, 8, %o0 + st %g2, [%o0 - 0x08] + add %o1, 8, %o1 + st %g3, [%o0 - 0x04] + +81: /* memcpy_last7 */ + + be 1f + andcc %g1, 2, %g0 + + ld [%o1], %g2 + add %o1, 4, %o1 + st %g2, [%o0] + add %o0, 4, %o0 +1: + be 1f + andcc %g1, 1, %g0 + + lduh [%o1], %g2 + add %o1, 2, %o1 + sth %g2, [%o0] + add %o0, 2, %o0 +1: + be 1f + nop + + ldub [%o1], %g2 + stb %g2, [%o0] +1: + retl + RETL_INSN + +82: /* ldd_std */ + MOVE_BIGALIGNCHUNK(o1, o0, 0x00, o2, o3, o4, o5, g2, g3, g4, g5) + MOVE_BIGALIGNCHUNK(o1, o0, 0x20, o2, o3, o4, o5, g2, g3, g4, g5) + MOVE_BIGALIGNCHUNK(o1, o0, 0x40, o2, o3, o4, o5, g2, g3, g4, g5) + MOVE_BIGALIGNCHUNK(o1, o0, 0x60, o2, o3, o4, o5, g2, g3, g4, g5) + subcc %g7, 128, %g7 + add %o1, 128, %o1 + bne 82b + add %o0, 128, %o0 + +#ifndef FASTER_ALIGNED + + andcc %g1, 0x70, %g7 + be 80b + andcc %g1, 8, %g0 + + sethi %hi(80b), %o5 + srl %g7, 1, %o4 + add %g7, %o4, %o4 + add %o1, %g7, %o1 + sub %o5, %o4, %o5 + jmpl %o5 + %lo(80b), %g0 + add %o0, %g7, %o0 + +#else /* FASTER_ALIGNED */ + + andcc %g1, 0x70, %g7 + be 84f + andcc %g1, 8, %g0 + + sethi %hi(84f), %o5 + add %o1, %g7, %o1 + sub %o5, %g7, %o5 + jmpl %o5 + %lo(84f), %g0 + add %o0, %g7, %o0 + +83: /* amemcpy_table */ + + MOVE_LASTALIGNCHUNK(o1, o0, 0x60, g2, g3, g4, g5) + MOVE_LASTALIGNCHUNK(o1, o0, 0x50, g2, g3, g4, g5) + MOVE_LASTALIGNCHUNK(o1, o0, 0x40, g2, g3, g4, g5) + MOVE_LASTALIGNCHUNK(o1, o0, 0x30, g2, g3, g4, g5) + MOVE_LASTALIGNCHUNK(o1, o0, 0x20, g2, g3, g4, g5) + MOVE_LASTALIGNCHUNK(o1, o0, 0x10, g2, g3, g4, g5) + MOVE_LASTALIGNCHUNK(o1, o0, 0x00, g2, g3, g4, g5) + +84: /* amemcpy_table_end */ + be 85f + andcc %g1, 4, %g0 + + ldd [%o1], %g2 + add %o0, 8, %o0 + std %g2, [%o0 - 0x08] + add %o1, 8, %o1 +85: /* amemcpy_last7 */ + be 1f + andcc %g1, 2, %g0 + + ld [%o1], %g2 + add %o1, 4, %o1 + st %g2, [%o0] + add %o0, 4, %o0 +1: + be 1f + andcc %g1, 1, %g0 + + lduh [%o1], %g2 + add %o1, 2, %o1 + sth %g2, [%o0] + add %o0, 2, %o0 +1: + be 1f + nop + + ldub [%o1], %g2 + stb %g2, [%o0] +1: + retl + RETL_INSN + +#endif /* FASTER_ALIGNED */ + +86: /* non_aligned */ + cmp %o2, 6 + bleu 88f + +#ifdef FASTER_NONALIGNED + + cmp %o2, 256 + bcc 87f + +#endif /* FASTER_NONALIGNED */ + + andcc %o0, 3, %g0 + be 61f + andcc %o0, 1, %g0 + be 60f + andcc %o0, 2, %g0 + + ldub [%o1], %g5 + add %o1, 1, %o1 + stb %g5, [%o0] + sub %o2, 1, %o2 + bne 61f + add %o0, 1, %o0 +60: + ldub [%o1], %g3 + add %o1, 2, %o1 + stb %g3, [%o0] + sub %o2, 2, %o2 + ldub [%o1 - 1], %g3 + add %o0, 2, %o0 + stb %g3, [%o0 - 1] +61: + and %o1, 3, %g2 + and %o2, 0xc, %g3 + and %o1, -4, %o1 + cmp %g3, 4 + sll %g2, 3, %g4 + mov 32, %g2 + be 4f + sub %g2, %g4, %g7 + + blu 3f + cmp %g3, 0x8 + + be 2f + srl %o2, 2, %g3 + + ld [%o1], %o3 + add %o0, -8, %o0 + ld [%o1 + 4], %o4 + b 8f + add %g3, 1, %g3 +2: + ld [%o1], %o4 + add %o0, -12, %o0 + ld [%o1 + 4], %o5 + add %g3, 2, %g3 + b 9f + add %o1, -4, %o1 +3: + ld [%o1], %g1 + add %o0, -4, %o0 + ld [%o1 + 4], %o3 + srl %o2, 2, %g3 + b 7f + add %o1, 4, %o1 +4: + ld [%o1], %o5 + cmp %o2, 7 + ld [%o1 + 4], %g1 + srl %o2, 2, %g3 + bleu 10f + add %o1, 8, %o1 + + ld [%o1], %o3 + add %g3, -1, %g3 +5: + sll %o5, %g4, %g2 + srl %g1, %g7, %g5 + or %g2, %g5, %g2 + st %g2, [%o0] +7: + ld [%o1 + 4], %o4 + sll %g1, %g4, %g2 + srl %o3, %g7, %g5 + or %g2, %g5, %g2 + st %g2, [%o0 + 4] +8: + ld [%o1 + 8], %o5 + sll %o3, %g4, %g2 + srl %o4, %g7, %g5 + or %g2, %g5, %g2 + st %g2, [%o0 + 8] +9: + ld [%o1 + 12], %g1 + sll %o4, %g4, %g2 + srl %o5, %g7, %g5 + addcc %g3, -4, %g3 + or %g2, %g5, %g2 + add %o1, 16, %o1 + st %g2, [%o0 + 12] + add %o0, 16, %o0 + bne,a 5b + ld [%o1], %o3 +10: + sll %o5, %g4, %g2 + srl %g1, %g7, %g5 + srl %g7, 3, %g3 + or %g2, %g5, %g2 + sub %o1, %g3, %o1 + andcc %o2, 2, %g0 + st %g2, [%o0] + be 1f + andcc %o2, 1, %g0 + + ldub [%o1], %g2 + add %o1, 2, %o1 + stb %g2, [%o0 + 4] + add %o0, 2, %o0 + ldub [%o1 - 1], %g2 + stb %g2, [%o0 + 3] +1: + be 1f + nop + ldub [%o1], %g2 + stb %g2, [%o0 + 4] +1: + retl + RETL_INSN + +#ifdef FASTER_NONALIGNED + +87: /* faster_nonaligned */ + + andcc %o1, 3, %g0 + be 3f + andcc %o1, 1, %g0 + + be 4f + andcc %o1, 2, %g0 + + ldub [%o1], %g2 + add %o1, 1, %o1 + stb %g2, [%o0] + sub %o2, 1, %o2 + bne 3f + add %o0, 1, %o0 +4: + lduh [%o1], %g2 + add %o1, 2, %o1 + srl %g2, 8, %g3 + sub %o2, 2, %o2 + stb %g3, [%o0] + add %o0, 2, %o0 + stb %g2, [%o0 - 1] +3: + andcc %o1, 4, %g0 + + bne 2f + cmp %o5, 1 + + ld [%o1], %o4 + srl %o4, 24, %g2 + stb %g2, [%o0] + srl %o4, 16, %g3 + stb %g3, [%o0 + 1] + srl %o4, 8, %g2 + stb %g2, [%o0 + 2] + sub %o2, 4, %o2 + stb %o4, [%o0 + 3] + add %o1, 4, %o1 + add %o0, 4, %o0 +2: + be 33f + cmp %o5, 2 + be 32f + sub %o2, 4, %o2 +31: + ld [%o1], %g2 + add %o1, 4, %o1 + srl %g2, 24, %g3 + and %o0, 7, %g5 + stb %g3, [%o0] + cmp %g5, 7 + sll %g2, 8, %g1 + add %o0, 4, %o0 + be 41f + and %o2, 0xffffffc0, %o3 + ld [%o0 - 7], %o4 +4: + SMOVE_CHUNK(o1, o0, 0x00, g2, g3, g4, g5, o4, o5, g7, g1, 8, 24, -3) + SMOVE_CHUNK(o1, o0, 0x10, g2, g3, g4, g5, o4, o5, g7, g1, 8, 24, -3) + SMOVE_CHUNK(o1, o0, 0x20, g2, g3, g4, g5, o4, o5, g7, g1, 8, 24, -3) + SMOVE_CHUNK(o1, o0, 0x30, g2, g3, g4, g5, o4, o5, g7, g1, 8, 24, -3) + subcc %o3, 64, %o3 + add %o1, 64, %o1 + bne 4b + add %o0, 64, %o0 + + andcc %o2, 0x30, %o3 + be,a 1f + srl %g1, 16, %g2 +4: + SMOVE_CHUNK(o1, o0, 0x00, g2, g3, g4, g5, o4, o5, g7, g1, 8, 24, -3) + subcc %o3, 16, %o3 + add %o1, 16, %o1 + bne 4b + add %o0, 16, %o0 + + srl %g1, 16, %g2 +1: + st %o4, [%o0 - 7] + sth %g2, [%o0 - 3] + srl %g1, 8, %g4 + b 88f + stb %g4, [%o0 - 1] +32: + ld [%o1], %g2 + add %o1, 4, %o1 + srl %g2, 16, %g3 + and %o0, 7, %g5 + sth %g3, [%o0] + cmp %g5, 6 + sll %g2, 16, %g1 + add %o0, 4, %o0 + be 42f + and %o2, 0xffffffc0, %o3 + ld [%o0 - 6], %o4 +4: + SMOVE_CHUNK(o1, o0, 0x00, g2, g3, g4, g5, o4, o5, g7, g1, 16, 16, -2) + SMOVE_CHUNK(o1, o0, 0x10, g2, g3, g4, g5, o4, o5, g7, g1, 16, 16, -2) + SMOVE_CHUNK(o1, o0, 0x20, g2, g3, g4, g5, o4, o5, g7, g1, 16, 16, -2) + SMOVE_CHUNK(o1, o0, 0x30, g2, g3, g4, g5, o4, o5, g7, g1, 16, 16, -2) + subcc %o3, 64, %o3 + add %o1, 64, %o1 + bne 4b + add %o0, 64, %o0 + + andcc %o2, 0x30, %o3 + be,a 1f + srl %g1, 16, %g2 +4: + SMOVE_CHUNK(o1, o0, 0x00, g2, g3, g4, g5, o4, o5, g7, g1, 16, 16, -2) + subcc %o3, 16, %o3 + add %o1, 16, %o1 + bne 4b + add %o0, 16, %o0 + + srl %g1, 16, %g2 +1: + st %o4, [%o0 - 6] + b 88f + sth %g2, [%o0 - 2] +33: + ld [%o1], %g2 + sub %o2, 4, %o2 + srl %g2, 24, %g3 + and %o0, 7, %g5 + stb %g3, [%o0] + cmp %g5, 5 + srl %g2, 8, %g4 + sll %g2, 24, %g1 + sth %g4, [%o0 + 1] + add %o1, 4, %o1 + be 43f + and %o2, 0xffffffc0, %o3 + + ld [%o0 - 1], %o4 + add %o0, 4, %o0 +4: + SMOVE_CHUNK(o1, o0, 0x00, g2, g3, g4, g5, o4, o5, g7, g1, 24, 8, -1) + SMOVE_CHUNK(o1, o0, 0x10, g2, g3, g4, g5, o4, o5, g7, g1, 24, 8, -1) + SMOVE_CHUNK(o1, o0, 0x20, g2, g3, g4, g5, o4, o5, g7, g1, 24, 8, -1) + SMOVE_CHUNK(o1, o0, 0x30, g2, g3, g4, g5, o4, o5, g7, g1, 24, 8, -1) + subcc %o3, 64, %o3 + add %o1, 64, %o1 + bne 4b + add %o0, 64, %o0 + + andcc %o2, 0x30, %o3 + be,a 1f + srl %g1, 24, %g2 +4: + SMOVE_CHUNK(o1, o0, 0x00, g2, g3, g4, g5, o4, o5, g7, g1, 24, 8, -1) + subcc %o3, 16, %o3 + add %o1, 16, %o1 + bne 4b + add %o0, 16, %o0 + + srl %g1, 24, %g2 +1: + st %o4, [%o0 - 5] + b 88f + stb %g2, [%o0 - 1] +41: + SMOVE_ALIGNCHUNK(o1, o0, 0x00, g2, g3, g4, g5, o4, o5, g7, g1, 8, 24, -3) + SMOVE_ALIGNCHUNK(o1, o0, 0x10, g2, g3, g4, g5, o4, o5, g7, g1, 8, 24, -3) + SMOVE_ALIGNCHUNK(o1, o0, 0x20, g2, g3, g4, g5, o4, o5, g7, g1, 8, 24, -3) + SMOVE_ALIGNCHUNK(o1, o0, 0x30, g2, g3, g4, g5, o4, o5, g7, g1, 8, 24, -3) + subcc %o3, 64, %o3 + add %o1, 64, %o1 + bne 41b + add %o0, 64, %o0 + + andcc %o2, 0x30, %o3 + be,a 1f + srl %g1, 16, %g2 +4: + SMOVE_ALIGNCHUNK(o1, o0, 0x00, g2, g3, g4, g5, o4, o5, g7, g1, 8, 24, -3) + subcc %o3, 16, %o3 + add %o1, 16, %o1 + bne 4b + add %o0, 16, %o0 + + srl %g1, 16, %g2 +1: + sth %g2, [%o0 - 3] + srl %g1, 8, %g4 + b 88f + stb %g4, [%o0 - 1] +43: + SMOVE_ALIGNCHUNK(o1, o0, 0x00, g2, g3, g4, g5, o4, o5, g7, g1, 24, 8, 3) + SMOVE_ALIGNCHUNK(o1, o0, 0x10, g2, g3, g4, g5, o4, o5, g7, g1, 24, 8, 3) + SMOVE_ALIGNCHUNK(o1, o0, 0x20, g2, g3, g4, g5, o4, o5, g7, g1, 24, 8, 3) + SMOVE_ALIGNCHUNK(o1, o0, 0x30, g2, g3, g4, g5, o4, o5, g7, g1, 24, 8, 3) + subcc %o3, 64, %o3 + add %o1, 64, %o1 + bne 43b + add %o0, 64, %o0 + + andcc %o2, 0x30, %o3 + be,a 1f + srl %g1, 24, %g2 +4: + SMOVE_ALIGNCHUNK(o1, o0, 0x00, g2, g3, g4, g5, o4, o5, g7, g1, 24, 8, 3) + subcc %o3, 16, %o3 + add %o1, 16, %o1 + bne 4b + add %o0, 16, %o0 + + srl %g1, 24, %g2 +1: + stb %g2, [%o0 + 3] + b 88f + add %o0, 4, %o0 +42: + SMOVE_ALIGNCHUNK(o1, o0, 0x00, g2, g3, g4, g5, o4, o5, g7, g1, 16, 16, -2) + SMOVE_ALIGNCHUNK(o1, o0, 0x10, g2, g3, g4, g5, o4, o5, g7, g1, 16, 16, -2) + SMOVE_ALIGNCHUNK(o1, o0, 0x20, g2, g3, g4, g5, o4, o5, g7, g1, 16, 16, -2) + SMOVE_ALIGNCHUNK(o1, o0, 0x30, g2, g3, g4, g5, o4, o5, g7, g1, 16, 16, -2) + subcc %o3, 64, %o3 + add %o1, 64, %o1 + bne 42b + add %o0, 64, %o0 + + andcc %o2, 0x30, %o3 + be,a 1f + srl %g1, 16, %g2 +4: + SMOVE_ALIGNCHUNK(o1, o0, 0x00, g2, g3, g4, g5, o4, o5, g7, g1, 16, 16, -2) + subcc %o3, 16, %o3 + add %o1, 16, %o1 + bne 4b + add %o0, 16, %o0 + + srl %g1, 16, %g2 +1: + sth %g2, [%o0 - 2] + + /* Fall through */ + +#endif /* FASTER_NONALIGNED */ + +88: /* short_end */ + + and %o2, 0xe, %o3 +20: + sethi %hi(89f), %o5 + sll %o3, 3, %o4 + add %o0, %o3, %o0 + sub %o5, %o4, %o5 + add %o1, %o3, %o1 + jmpl %o5 + %lo(89f), %g0 + andcc %o2, 1, %g0 + + MOVE_SHORTCHUNK(o1, o0, 0x0c, g2, g3) + MOVE_SHORTCHUNK(o1, o0, 0x0a, g2, g3) + MOVE_SHORTCHUNK(o1, o0, 0x08, g2, g3) + MOVE_SHORTCHUNK(o1, o0, 0x06, g2, g3) + MOVE_SHORTCHUNK(o1, o0, 0x04, g2, g3) + MOVE_SHORTCHUNK(o1, o0, 0x02, g2, g3) + MOVE_SHORTCHUNK(o1, o0, 0x00, g2, g3) + +89: /* short_table_end */ + + be 1f + nop + + ldub [%o1], %g2 + stb %g2, [%o0] +1: + retl + RETL_INSN + +90: /* short_aligned_end */ + bne 88b + andcc %o2, 8, %g0 + + be 1f + andcc %o2, 4, %g0 + + ld [%o1 + 0x00], %g2 + ld [%o1 + 0x04], %g3 + add %o1, 8, %o1 + st %g2, [%o0 + 0x00] + st %g3, [%o0 + 0x04] + add %o0, 8, %o0 +1: + b 81b + mov %o2, %g1 diff --git a/arch/sparc/lib/memscan.S b/arch/sparc/lib/memscan.S new file mode 100644 index 00000000000..28e78ff090a --- /dev/null +++ b/arch/sparc/lib/memscan.S @@ -0,0 +1,133 @@ +/* $Id: memscan.S,v 1.4 1996/09/08 02:01:20 davem Exp $ + * memscan.S: Optimized memscan for the Sparc. + * + * Copyright (C) 1996 David S. Miller (davem@caip.rutgers.edu) + */ + +/* In essence, this is just a fancy strlen. */ + +#define LO_MAGIC 0x01010101 +#define HI_MAGIC 0x80808080 + + .text + .align 4 + .globl __memscan_zero, __memscan_generic + .globl memscan +__memscan_zero: + /* %o0 = addr, %o1 = size */ + cmp %o1, 0 + bne,a 1f + andcc %o0, 3, %g0 + + retl + nop + +1: + be mzero_scan_word + sethi %hi(HI_MAGIC), %g2 + + ldsb [%o0], %g3 +mzero_still_not_word_aligned: + cmp %g3, 0 + bne 1f + add %o0, 1, %o0 + + retl + sub %o0, 1, %o0 + +1: + subcc %o1, 1, %o1 + bne,a 1f + andcc %o0, 3, %g0 + + retl + nop + +1: + bne,a mzero_still_not_word_aligned + ldsb [%o0], %g3 + + sethi %hi(HI_MAGIC), %g2 +mzero_scan_word: + or %g2, %lo(HI_MAGIC), %o3 + sethi %hi(LO_MAGIC), %g3 + or %g3, %lo(LO_MAGIC), %o2 +mzero_next_word: + ld [%o0], %g2 +mzero_next_word_preloaded: + sub %g2, %o2, %g2 +mzero_next_word_preloaded_next: + andcc %g2, %o3, %g0 + bne mzero_byte_zero + add %o0, 4, %o0 + +mzero_check_out_of_fuel: + subcc %o1, 4, %o1 + bg,a 1f + ld [%o0], %g2 + + retl + nop + +1: + b mzero_next_word_preloaded_next + sub %g2, %o2, %g2 + + /* Check every byte. */ +mzero_byte_zero: + ldsb [%o0 - 4], %g2 + cmp %g2, 0 + bne mzero_byte_one + sub %o0, 4, %g3 + + retl + mov %g3, %o0 + +mzero_byte_one: + ldsb [%o0 - 3], %g2 + cmp %g2, 0 + bne,a mzero_byte_two_and_three + ldsb [%o0 - 2], %g2 + + retl + sub %o0, 3, %o0 + +mzero_byte_two_and_three: + cmp %g2, 0 + bne,a 1f + ldsb [%o0 - 1], %g2 + + retl + sub %o0, 2, %o0 + +1: + cmp %g2, 0 + bne,a mzero_next_word_preloaded + ld [%o0], %g2 + + retl + sub %o0, 1, %o0 + +mzero_found_it: + retl + sub %o0, 2, %o0 + +memscan: +__memscan_generic: + /* %o0 = addr, %o1 = c, %o2 = size */ + cmp %o2, 0 + bne,a 0f + ldub [%o0], %g2 + + b,a 2f +1: + ldub [%o0], %g2 +0: + cmp %g2, %o1 + be 2f + addcc %o2, -1, %o2 + bne 1b + add %o0, 1, %o0 +2: + retl + nop diff --git a/arch/sparc/lib/memset.S b/arch/sparc/lib/memset.S new file mode 100644 index 00000000000..a65eba41097 --- /dev/null +++ b/arch/sparc/lib/memset.S @@ -0,0 +1,203 @@ +/* linux/arch/sparc/lib/memset.S: Sparc optimized memset, bzero and clear_user code + * Copyright (C) 1991,1996 Free Software Foundation + * Copyright (C) 1996,1997 Jakub Jelinek (jj@sunsite.mff.cuni.cz) + * Copyright (C) 1996 David S. Miller (davem@caip.rutgers.edu) + * + * Returns 0, if ok, and number of bytes not yet set if exception + * occurs and we were called as clear_user. + */ + +#include <asm/ptrace.h> + +/* Work around cpp -rob */ +#define ALLOC #alloc +#define EXECINSTR #execinstr +#define EX(x,y,a,b) \ +98: x,y; \ + .section .fixup,ALLOC,EXECINSTR; \ + .align 4; \ +99: ba 30f; \ + a, b, %o0; \ + .section __ex_table,ALLOC; \ + .align 4; \ + .word 98b, 99b; \ + .text; \ + .align 4 + +#define EXT(start,end,handler) \ + .section __ex_table,ALLOC; \ + .align 4; \ + .word start, 0, end, handler; \ + .text; \ + .align 4 + +/* Please don't change these macros, unless you change the logic + * in the .fixup section below as well. + * Store 64 bytes at (BASE + OFFSET) using value SOURCE. */ +#define ZERO_BIG_BLOCK(base, offset, source) \ + std source, [base + offset + 0x00]; \ + std source, [base + offset + 0x08]; \ + std source, [base + offset + 0x10]; \ + std source, [base + offset + 0x18]; \ + std source, [base + offset + 0x20]; \ + std source, [base + offset + 0x28]; \ + std source, [base + offset + 0x30]; \ + std source, [base + offset + 0x38]; + +#define ZERO_LAST_BLOCKS(base, offset, source) \ + std source, [base - offset - 0x38]; \ + std source, [base - offset - 0x30]; \ + std source, [base - offset - 0x28]; \ + std source, [base - offset - 0x20]; \ + std source, [base - offset - 0x18]; \ + std source, [base - offset - 0x10]; \ + std source, [base - offset - 0x08]; \ + std source, [base - offset - 0x00]; + + .text + .align 4 + + .globl __bzero_begin +__bzero_begin: + + .globl __bzero, __memset, + .globl memset + .globl __memset_start, __memset_end +__memset_start: +__memset: +memset: + and %o1, 0xff, %g3 + sll %g3, 8, %g2 + or %g3, %g2, %g3 + sll %g3, 16, %g2 + or %g3, %g2, %g3 + b 1f + mov %o2, %o1 +3: + cmp %o2, 3 + be 2f + EX(stb %g3, [%o0], sub %o1, 0) + + cmp %o2, 2 + be 2f + EX(stb %g3, [%o0 + 0x01], sub %o1, 1) + + EX(stb %g3, [%o0 + 0x02], sub %o1, 2) +2: + sub %o2, 4, %o2 + add %o1, %o2, %o1 + b 4f + sub %o0, %o2, %o0 + +__bzero: + mov %g0, %g3 +1: + cmp %o1, 7 + bleu 7f + andcc %o0, 3, %o2 + + bne 3b +4: + andcc %o0, 4, %g0 + + be 2f + mov %g3, %g2 + + EX(st %g3, [%o0], sub %o1, 0) + sub %o1, 4, %o1 + add %o0, 4, %o0 +2: + andcc %o1, 0xffffff80, %o3 ! Now everything is 8 aligned and o1 is len to run + be 9f + andcc %o1, 0x78, %o2 +10: + ZERO_BIG_BLOCK(%o0, 0x00, %g2) + subcc %o3, 128, %o3 + ZERO_BIG_BLOCK(%o0, 0x40, %g2) +11: + EXT(10b, 11b, 20f) + bne 10b + add %o0, 128, %o0 + + orcc %o2, %g0, %g0 +9: + be 13f + andcc %o1, 7, %o1 + + srl %o2, 1, %o3 + set 13f, %o4 + sub %o4, %o3, %o4 + jmp %o4 + add %o0, %o2, %o0 + +12: + ZERO_LAST_BLOCKS(%o0, 0x48, %g2) + ZERO_LAST_BLOCKS(%o0, 0x08, %g2) +13: + be 8f + andcc %o1, 4, %g0 + + be 1f + andcc %o1, 2, %g0 + + EX(st %g3, [%o0], and %o1, 7) + add %o0, 4, %o0 +1: + be 1f + andcc %o1, 1, %g0 + + EX(sth %g3, [%o0], and %o1, 3) + add %o0, 2, %o0 +1: + bne,a 8f + EX(stb %g3, [%o0], and %o1, 1) +8: + retl + clr %o0 +7: + be 13b + orcc %o1, 0, %g0 + + be 0f +8: + add %o0, 1, %o0 + subcc %o1, 1, %o1 + bne,a 8b + EX(stb %g3, [%o0 - 1], add %o1, 1) +0: + retl + clr %o0 +__memset_end: + + .section .fixup,#alloc,#execinstr + .align 4 +20: + cmp %g2, 8 + bleu 1f + and %o1, 0x7f, %o1 + sub %g2, 9, %g2 + add %o3, 64, %o3 +1: + sll %g2, 3, %g2 + add %o3, %o1, %o0 + b 30f + sub %o0, %g2, %o0 +21: + mov 8, %o0 + and %o1, 7, %o1 + sub %o0, %g2, %o0 + sll %o0, 3, %o0 + b 30f + add %o0, %o1, %o0 +30: +/* %o4 is faulting address, %o5 is %pc where fault occurred */ + save %sp, -104, %sp + mov %i5, %o0 + mov %i7, %o1 + call lookup_fault + mov %i4, %o2 + ret + restore + + .globl __bzero_end +__bzero_end: diff --git a/arch/sparc/lib/mul.S b/arch/sparc/lib/mul.S new file mode 100644 index 00000000000..83dffbc2f62 --- /dev/null +++ b/arch/sparc/lib/mul.S @@ -0,0 +1,135 @@ +/* $Id: mul.S,v 1.4 1996/09/30 02:22:32 davem Exp $ + * mul.S: This routine was taken from glibc-1.09 and is covered + * by the GNU Library General Public License Version 2. + */ + +/* + * Signed multiply, from Appendix E of the Sparc Version 8 + * Architecture Manual. + */ + +/* + * Returns %o0 * %o1 in %o1%o0 (i.e., %o1 holds the upper 32 bits of + * the 64-bit product). + * + * This code optimizes short (less than 13-bit) multiplies. + */ + + .globl .mul +.mul: + mov %o0, %y ! multiplier -> Y + andncc %o0, 0xfff, %g0 ! test bits 12..31 + be Lmul_shortway ! if zero, can do it the short way + andcc %g0, %g0, %o4 ! zero the partial product and clear N and V + + /* + * Long multiply. 32 steps, followed by a final shift step. + */ + mulscc %o4, %o1, %o4 ! 1 + mulscc %o4, %o1, %o4 ! 2 + mulscc %o4, %o1, %o4 ! 3 + mulscc %o4, %o1, %o4 ! 4 + mulscc %o4, %o1, %o4 ! 5 + mulscc %o4, %o1, %o4 ! 6 + mulscc %o4, %o1, %o4 ! 7 + mulscc %o4, %o1, %o4 ! 8 + mulscc %o4, %o1, %o4 ! 9 + mulscc %o4, %o1, %o4 ! 10 + mulscc %o4, %o1, %o4 ! 11 + mulscc %o4, %o1, %o4 ! 12 + mulscc %o4, %o1, %o4 ! 13 + mulscc %o4, %o1, %o4 ! 14 + mulscc %o4, %o1, %o4 ! 15 + mulscc %o4, %o1, %o4 ! 16 + mulscc %o4, %o1, %o4 ! 17 + mulscc %o4, %o1, %o4 ! 18 + mulscc %o4, %o1, %o4 ! 19 + mulscc %o4, %o1, %o4 ! 20 + mulscc %o4, %o1, %o4 ! 21 + mulscc %o4, %o1, %o4 ! 22 + mulscc %o4, %o1, %o4 ! 23 + mulscc %o4, %o1, %o4 ! 24 + mulscc %o4, %o1, %o4 ! 25 + mulscc %o4, %o1, %o4 ! 26 + mulscc %o4, %o1, %o4 ! 27 + mulscc %o4, %o1, %o4 ! 28 + mulscc %o4, %o1, %o4 ! 29 + mulscc %o4, %o1, %o4 ! 30 + mulscc %o4, %o1, %o4 ! 31 + mulscc %o4, %o1, %o4 ! 32 + mulscc %o4, %g0, %o4 ! final shift + + ! If %o0 was negative, the result is + ! (%o0 * %o1) + (%o1 << 32)) + ! We fix that here. + +#if 0 + tst %o0 + bge 1f + rd %y, %o0 + + ! %o0 was indeed negative; fix upper 32 bits of result by subtracting + ! %o1 (i.e., return %o4 - %o1 in %o1). + retl + sub %o4, %o1, %o1 + +1: + retl + mov %o4, %o1 +#else + /* Faster code adapted from tege@sics.se's code for umul.S. */ + sra %o0, 31, %o2 ! make mask from sign bit + and %o1, %o2, %o2 ! %o2 = 0 or %o1, depending on sign of %o0 + rd %y, %o0 ! get lower half of product + retl + sub %o4, %o2, %o1 ! subtract compensation + ! and put upper half in place +#endif + +Lmul_shortway: + /* + * Short multiply. 12 steps, followed by a final shift step. + * The resulting bits are off by 12 and (32-12) = 20 bit positions, + * but there is no problem with %o0 being negative (unlike above). + */ + mulscc %o4, %o1, %o4 ! 1 + mulscc %o4, %o1, %o4 ! 2 + mulscc %o4, %o1, %o4 ! 3 + mulscc %o4, %o1, %o4 ! 4 + mulscc %o4, %o1, %o4 ! 5 + mulscc %o4, %o1, %o4 ! 6 + mulscc %o4, %o1, %o4 ! 7 + mulscc %o4, %o1, %o4 ! 8 + mulscc %o4, %o1, %o4 ! 9 + mulscc %o4, %o1, %o4 ! 10 + mulscc %o4, %o1, %o4 ! 11 + mulscc %o4, %o1, %o4 ! 12 + mulscc %o4, %g0, %o4 ! final shift + + /* + * %o4 has 20 of the bits that should be in the low part of the + * result; %y has the bottom 12 (as %y's top 12). That is: + * + * %o4 %y + * +----------------+----------------+ + * | -12- | -20- | -12- | -20- | + * +------(---------+------)---------+ + * --hi-- ----low-part---- + * + * The upper 12 bits of %o4 should be sign-extended to form the + * high part of the product (i.e., highpart = %o4 >> 20). + */ + + rd %y, %o5 + sll %o4, 12, %o0 ! shift middle bits left 12 + srl %o5, 20, %o5 ! shift low bits right 20, zero fill at left + or %o5, %o0, %o0 ! construct low part of result + retl + sra %o4, 20, %o1 ! ... and extract high part of result + + .globl .mul_patch +.mul_patch: + smul %o0, %o1, %o0 + retl + rd %y, %o1 + nop diff --git a/arch/sparc/lib/muldi3.S b/arch/sparc/lib/muldi3.S new file mode 100644 index 00000000000..7f17872d060 --- /dev/null +++ b/arch/sparc/lib/muldi3.S @@ -0,0 +1,76 @@ +/* Copyright (C) 1989, 1992, 1993, 1994, 1995 Free Software Foundation, Inc. + +This file is part of GNU CC. + +GNU CC is free software; you can redistribute it and/or modify +it under the terms of the GNU General Public License as published by +the Free Software Foundation; either version 2, or (at your option) +any later version. + +GNU CC 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 General Public License for more details. + +You should have received a copy of the GNU General Public License +along with GNU CC; see the file COPYING. If not, write to +the Free Software Foundation, 59 Temple Place - Suite 330, +Boston, MA 02111-1307, USA. */ + + .text + .align 4 + .globl __muldi3 +__muldi3: + save %sp, -104, %sp + wr %g0, %i1, %y + sra %i3, 0x1f, %g2 + and %i1, %g2, %g2 + andcc %g0, 0, %g1 + mulscc %g1, %i3, %g1 + mulscc %g1, %i3, %g1 + mulscc %g1, %i3, %g1 + mulscc %g1, %i3, %g1 + mulscc %g1, %i3, %g1 + mulscc %g1, %i3, %g1 + mulscc %g1, %i3, %g1 + mulscc %g1, %i3, %g1 + mulscc %g1, %i3, %g1 + mulscc %g1, %i3, %g1 + mulscc %g1, %i3, %g1 + mulscc %g1, %i3, %g1 + mulscc %g1, %i3, %g1 + mulscc %g1, %i3, %g1 + mulscc %g1, %i3, %g1 + mulscc %g1, %i3, %g1 + mulscc %g1, %i3, %g1 + mulscc %g1, %i3, %g1 + mulscc %g1, %i3, %g1 + mulscc %g1, %i3, %g1 + mulscc %g1, %i3, %g1 + mulscc %g1, %i3, %g1 + mulscc %g1, %i3, %g1 + mulscc %g1, %i3, %g1 + mulscc %g1, %i3, %g1 + mulscc %g1, %i3, %g1 + mulscc %g1, %i3, %g1 + mulscc %g1, %i3, %g1 + mulscc %g1, %i3, %g1 + mulscc %g1, %i3, %g1 + mulscc %g1, %i3, %g1 + mulscc %g1, %i3, %g1 + mulscc %g1, 0, %g1 + add %g1, %g2, %l2 + rd %y, %o1 + mov %o1, %l3 + mov %i1, %o0 + call .umul + mov %i2, %o1 + mov %o0, %l0 + mov %i0, %o0 + call .umul + mov %i3, %o1 + add %l0, %o0, %l0 + mov %l2, %i0 + add %l2, %l0, %i0 + ret + restore %g0, %l3, %o1 diff --git a/arch/sparc/lib/rem.S b/arch/sparc/lib/rem.S new file mode 100644 index 00000000000..44508148d05 --- /dev/null +++ b/arch/sparc/lib/rem.S @@ -0,0 +1,382 @@ +/* $Id: rem.S,v 1.7 1996/09/30 02:22:34 davem Exp $ + * rem.S: This routine was taken from glibc-1.09 and is covered + * by the GNU Library General Public License Version 2. + */ + + +/* This file is generated from divrem.m4; DO NOT EDIT! */ +/* + * Division and remainder, from Appendix E of the Sparc Version 8 + * Architecture Manual, with fixes from Gordon Irlam. + */ + +/* + * Input: dividend and divisor in %o0 and %o1 respectively. + * + * m4 parameters: + * .rem name of function to generate + * rem rem=div => %o0 / %o1; rem=rem => %o0 % %o1 + * true true=true => signed; true=false => unsigned + * + * Algorithm parameters: + * N how many bits per iteration we try to get (4) + * WORDSIZE total number of bits (32) + * + * Derived constants: + * TOPBITS number of bits in the top decade of a number + * + * Important variables: + * Q the partial quotient under development (initially 0) + * R the remainder so far, initially the dividend + * ITER number of main division loop iterations required; + * equal to ceil(log2(quotient) / N). Note that this + * is the log base (2^N) of the quotient. + * V the current comparand, initially divisor*2^(ITER*N-1) + * + * Cost: + * Current estimate for non-large dividend is + * ceil(log2(quotient) / N) * (10 + 7N/2) + C + * A large dividend is one greater than 2^(31-TOPBITS) and takes a + * different path, as the upper bits of the quotient must be developed + * one bit at a time. + */ + + + .globl .rem +.rem: + ! compute sign of result; if neither is negative, no problem + orcc %o1, %o0, %g0 ! either negative? + bge 2f ! no, go do the divide + mov %o0, %g2 ! compute sign in any case + + tst %o1 + bge 1f + tst %o0 + ! %o1 is definitely negative; %o0 might also be negative + bge 2f ! if %o0 not negative... + sub %g0, %o1, %o1 ! in any case, make %o1 nonneg +1: ! %o0 is negative, %o1 is nonnegative + sub %g0, %o0, %o0 ! make %o0 nonnegative +2: + + ! Ready to divide. Compute size of quotient; scale comparand. + orcc %o1, %g0, %o5 + bne 1f + mov %o0, %o3 + + ! Divide by zero trap. If it returns, return 0 (about as + ! wrong as possible, but that is what SunOS does...). + ta ST_DIV0 + retl + clr %o0 + +1: + cmp %o3, %o5 ! if %o1 exceeds %o0, done + blu Lgot_result ! (and algorithm fails otherwise) + clr %o2 + + sethi %hi(1 << (32 - 4 - 1)), %g1 + + cmp %o3, %g1 + blu Lnot_really_big + clr %o4 + + ! Here the dividend is >= 2**(31-N) or so. We must be careful here, + ! as our usual N-at-a-shot divide step will cause overflow and havoc. + ! The number of bits in the result here is N*ITER+SC, where SC <= N. + ! Compute ITER in an unorthodox manner: know we need to shift V into + ! the top decade: so do not even bother to compare to R. + 1: + cmp %o5, %g1 + bgeu 3f + mov 1, %g7 + + sll %o5, 4, %o5 + + b 1b + add %o4, 1, %o4 + + ! Now compute %g7. + 2: + addcc %o5, %o5, %o5 + + bcc Lnot_too_big + add %g7, 1, %g7 + + ! We get here if the %o1 overflowed while shifting. + ! This means that %o3 has the high-order bit set. + ! Restore %o5 and subtract from %o3. + sll %g1, 4, %g1 ! high order bit + srl %o5, 1, %o5 ! rest of %o5 + add %o5, %g1, %o5 + + b Ldo_single_div + sub %g7, 1, %g7 + + Lnot_too_big: + 3: + cmp %o5, %o3 + blu 2b + nop + + be Ldo_single_div + nop + /* NB: these are commented out in the V8-Sparc manual as well */ + /* (I do not understand this) */ + ! %o5 > %o3: went too far: back up 1 step + ! srl %o5, 1, %o5 + ! dec %g7 + ! do single-bit divide steps + ! + ! We have to be careful here. We know that %o3 >= %o5, so we can do the + ! first divide step without thinking. BUT, the others are conditional, + ! and are only done if %o3 >= 0. Because both %o3 and %o5 may have the high- + ! order bit set in the first step, just falling into the regular + ! division loop will mess up the first time around. + ! So we unroll slightly... + Ldo_single_div: + subcc %g7, 1, %g7 + bl Lend_regular_divide + nop + + sub %o3, %o5, %o3 + mov 1, %o2 + + b Lend_single_divloop + nop + Lsingle_divloop: + sll %o2, 1, %o2 + + bl 1f + srl %o5, 1, %o5 + ! %o3 >= 0 + sub %o3, %o5, %o3 + + b 2f + add %o2, 1, %o2 + 1: ! %o3 < 0 + add %o3, %o5, %o3 + sub %o2, 1, %o2 + 2: + Lend_single_divloop: + subcc %g7, 1, %g7 + bge Lsingle_divloop + tst %o3 + + b,a Lend_regular_divide + +Lnot_really_big: +1: + sll %o5, 4, %o5 + cmp %o5, %o3 + bleu 1b + addcc %o4, 1, %o4 + be Lgot_result + sub %o4, 1, %o4 + + tst %o3 ! set up for initial iteration +Ldivloop: + sll %o2, 4, %o2 + ! depth 1, accumulated bits 0 + bl L.1.16 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 2, accumulated bits 1 + bl L.2.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 3, accumulated bits 3 + bl L.3.19 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits 7 + bl L.4.23 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + + b 9f + add %o2, (7*2+1), %o2 + +L.4.23: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (7*2-1), %o2 + +L.3.19: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits 5 + bl L.4.21 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (5*2+1), %o2 + +L.4.21: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (5*2-1), %o2 + +L.2.17: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 3, accumulated bits 1 + bl L.3.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits 3 + bl L.4.19 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (3*2+1), %o2 + +L.4.19: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (3*2-1), %o2 + +L.3.17: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits 1 + bl L.4.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (1*2+1), %o2 + +L.4.17: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (1*2-1), %o2 + +L.1.16: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 2, accumulated bits -1 + bl L.2.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 3, accumulated bits -1 + bl L.3.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits -1 + bl L.4.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-1*2+1), %o2 + +L.4.15: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-1*2-1), %o2 + +L.3.15: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits -3 + bl L.4.13 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-3*2+1), %o2 + +L.4.13: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-3*2-1), %o2 + +L.2.15: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 3, accumulated bits -3 + bl L.3.13 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits -5 + bl L.4.11 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-5*2+1), %o2 + +L.4.11: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-5*2-1), %o2 + + +L.3.13: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits -7 + bl L.4.9 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-7*2+1), %o2 + +L.4.9: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-7*2-1), %o2 + + 9: +Lend_regular_divide: + subcc %o4, 1, %o4 + bge Ldivloop + tst %o3 + + bl,a Lgot_result + ! non-restoring fixup here (one instruction only!) + add %o3, %o1, %o3 + +Lgot_result: + ! check to see if answer should be < 0 + tst %g2 + bl,a 1f + sub %g0, %o3, %o3 +1: + retl + mov %o3, %o0 + + .globl .rem_patch +.rem_patch: + sra %o0, 0x1f, %o4 + wr %o4, 0x0, %y + nop + nop + nop + sdivcc %o0, %o1, %o2 + bvs,a 1f + xnor %o2, %g0, %o2 +1: smul %o2, %o1, %o2 + retl + sub %o0, %o2, %o0 + nop diff --git a/arch/sparc/lib/rwsem.S b/arch/sparc/lib/rwsem.S new file mode 100644 index 00000000000..e7578dc600b --- /dev/null +++ b/arch/sparc/lib/rwsem.S @@ -0,0 +1,205 @@ +/* $Id: rwsem.S,v 1.5 2000/05/09 17:40:13 davem Exp $ + * Assembly part of rw semaphores. + * + * Copyright (C) 1999 Jakub Jelinek (jakub@redhat.com) + */ + +#include <linux/config.h> +#include <asm/ptrace.h> +#include <asm/psr.h> + + .section .sched.text + .align 4 + + .globl ___down_read +___down_read: + rd %psr, %g3 + nop + nop + nop + or %g3, PSR_PIL, %g7 + wr %g7, 0, %psr + nop + nop + nop +#ifdef CONFIG_SMP +1: ldstub [%g1 + 4], %g7 + tst %g7 + bne 1b + ld [%g1], %g7 + sub %g7, 1, %g7 + st %g7, [%g1] + stb %g0, [%g1 + 4] +#else + ld [%g1], %g7 + sub %g7, 1, %g7 + st %g7, [%g1] +#endif + wr %g3, 0, %psr + add %g7, 1, %g7 + nop + nop + subcc %g7, 1, %g7 + bneg 3f + nop +2: jmpl %o7, %g0 + mov %g4, %o7 +3: save %sp, -64, %sp + mov %g1, %l1 + mov %g4, %l4 + bcs 4f + mov %g5, %l5 + call down_read_failed + mov %l1, %o0 + mov %l1, %g1 + mov %l4, %g4 + ba ___down_read + restore %l5, %g0, %g5 +4: call down_read_failed_biased + mov %l1, %o0 + mov %l1, %g1 + mov %l4, %g4 + ba 2b + restore %l5, %g0, %g5 + + .globl ___down_write +___down_write: + rd %psr, %g3 + nop + nop + nop + or %g3, PSR_PIL, %g7 + wr %g7, 0, %psr + sethi %hi(0x01000000), %g2 + nop + nop +#ifdef CONFIG_SMP +1: ldstub [%g1 + 4], %g7 + tst %g7 + bne 1b + ld [%g1], %g7 + sub %g7, %g2, %g7 + st %g7, [%g1] + stb %g0, [%g1 + 4] +#else + ld [%g1], %g7 + sub %g7, %g2, %g7 + st %g7, [%g1] +#endif + wr %g3, 0, %psr + add %g7, %g2, %g7 + nop + nop + subcc %g7, %g2, %g7 + bne 3f + nop +2: jmpl %o7, %g0 + mov %g4, %o7 +3: save %sp, -64, %sp + mov %g1, %l1 + mov %g4, %l4 + bcs 4f + mov %g5, %l5 + call down_write_failed + mov %l1, %o0 + mov %l1, %g1 + mov %l4, %g4 + ba ___down_write + restore %l5, %g0, %g5 +4: call down_write_failed_biased + mov %l1, %o0 + mov %l1, %g1 + mov %l4, %g4 + ba 2b + restore %l5, %g0, %g5 + + .text + .globl ___up_read +___up_read: + rd %psr, %g3 + nop + nop + nop + or %g3, PSR_PIL, %g7 + wr %g7, 0, %psr + nop + nop + nop +#ifdef CONFIG_SMP +1: ldstub [%g1 + 4], %g7 + tst %g7 + bne 1b + ld [%g1], %g7 + add %g7, 1, %g7 + st %g7, [%g1] + stb %g0, [%g1 + 4] +#else + ld [%g1], %g7 + add %g7, 1, %g7 + st %g7, [%g1] +#endif + wr %g3, 0, %psr + nop + nop + nop + cmp %g7, 0 + be 3f + nop +2: jmpl %o7, %g0 + mov %g4, %o7 +3: save %sp, -64, %sp + mov %g1, %l1 + mov %g4, %l4 + mov %g5, %l5 + clr %o1 + call __rwsem_wake + mov %l1, %o0 + mov %l1, %g1 + mov %l4, %g4 + ba 2b + restore %l5, %g0, %g5 + + .globl ___up_write +___up_write: + rd %psr, %g3 + nop + nop + nop + or %g3, PSR_PIL, %g7 + wr %g7, 0, %psr + sethi %hi(0x01000000), %g2 + nop + nop +#ifdef CONFIG_SMP +1: ldstub [%g1 + 4], %g7 + tst %g7 + bne 1b + ld [%g1], %g7 + add %g7, %g2, %g7 + st %g7, [%g1] + stb %g0, [%g1 + 4] +#else + ld [%g1], %g7 + add %g7, %g2, %g7 + st %g7, [%g1] +#endif + wr %g3, 0, %psr + sub %g7, %g2, %g7 + nop + nop + addcc %g7, %g2, %g7 + bcs 3f + nop +2: jmpl %o7, %g0 + mov %g4, %o7 +3: save %sp, -64, %sp + mov %g1, %l1 + mov %g4, %l4 + mov %g5, %l5 + mov %g7, %o1 + call __rwsem_wake + mov %l1, %o0 + mov %l1, %g1 + mov %l4, %g4 + ba 2b + restore %l5, %g0, %g5 diff --git a/arch/sparc/lib/sdiv.S b/arch/sparc/lib/sdiv.S new file mode 100644 index 00000000000..e0ad80b6f63 --- /dev/null +++ b/arch/sparc/lib/sdiv.S @@ -0,0 +1,379 @@ +/* $Id: sdiv.S,v 1.6 1996/10/02 17:37:00 davem Exp $ + * sdiv.S: This routine was taken from glibc-1.09 and is covered + * by the GNU Library General Public License Version 2. + */ + + +/* This file is generated from divrem.m4; DO NOT EDIT! */ +/* + * Division and remainder, from Appendix E of the Sparc Version 8 + * Architecture Manual, with fixes from Gordon Irlam. + */ + +/* + * Input: dividend and divisor in %o0 and %o1 respectively. + * + * m4 parameters: + * .div name of function to generate + * div div=div => %o0 / %o1; div=rem => %o0 % %o1 + * true true=true => signed; true=false => unsigned + * + * Algorithm parameters: + * N how many bits per iteration we try to get (4) + * WORDSIZE total number of bits (32) + * + * Derived constants: + * TOPBITS number of bits in the top decade of a number + * + * Important variables: + * Q the partial quotient under development (initially 0) + * R the remainder so far, initially the dividend + * ITER number of main division loop iterations required; + * equal to ceil(log2(quotient) / N). Note that this + * is the log base (2^N) of the quotient. + * V the current comparand, initially divisor*2^(ITER*N-1) + * + * Cost: + * Current estimate for non-large dividend is + * ceil(log2(quotient) / N) * (10 + 7N/2) + C + * A large dividend is one greater than 2^(31-TOPBITS) and takes a + * different path, as the upper bits of the quotient must be developed + * one bit at a time. + */ + + + .globl .div +.div: + ! compute sign of result; if neither is negative, no problem + orcc %o1, %o0, %g0 ! either negative? + bge 2f ! no, go do the divide + xor %o1, %o0, %g2 ! compute sign in any case + + tst %o1 + bge 1f + tst %o0 + ! %o1 is definitely negative; %o0 might also be negative + bge 2f ! if %o0 not negative... + sub %g0, %o1, %o1 ! in any case, make %o1 nonneg +1: ! %o0 is negative, %o1 is nonnegative + sub %g0, %o0, %o0 ! make %o0 nonnegative +2: + + ! Ready to divide. Compute size of quotient; scale comparand. + orcc %o1, %g0, %o5 + bne 1f + mov %o0, %o3 + + ! Divide by zero trap. If it returns, return 0 (about as + ! wrong as possible, but that is what SunOS does...). + ta ST_DIV0 + retl + clr %o0 + +1: + cmp %o3, %o5 ! if %o1 exceeds %o0, done + blu Lgot_result ! (and algorithm fails otherwise) + clr %o2 + + sethi %hi(1 << (32 - 4 - 1)), %g1 + + cmp %o3, %g1 + blu Lnot_really_big + clr %o4 + + ! Here the dividend is >= 2**(31-N) or so. We must be careful here, + ! as our usual N-at-a-shot divide step will cause overflow and havoc. + ! The number of bits in the result here is N*ITER+SC, where SC <= N. + ! Compute ITER in an unorthodox manner: know we need to shift V into + ! the top decade: so do not even bother to compare to R. + 1: + cmp %o5, %g1 + bgeu 3f + mov 1, %g7 + + sll %o5, 4, %o5 + + b 1b + add %o4, 1, %o4 + + ! Now compute %g7. + 2: + addcc %o5, %o5, %o5 + bcc Lnot_too_big + add %g7, 1, %g7 + + ! We get here if the %o1 overflowed while shifting. + ! This means that %o3 has the high-order bit set. + ! Restore %o5 and subtract from %o3. + sll %g1, 4, %g1 ! high order bit + srl %o5, 1, %o5 ! rest of %o5 + add %o5, %g1, %o5 + + b Ldo_single_div + sub %g7, 1, %g7 + + Lnot_too_big: + 3: + cmp %o5, %o3 + blu 2b + nop + + be Ldo_single_div + nop + /* NB: these are commented out in the V8-Sparc manual as well */ + /* (I do not understand this) */ + ! %o5 > %o3: went too far: back up 1 step + ! srl %o5, 1, %o5 + ! dec %g7 + ! do single-bit divide steps + ! + ! We have to be careful here. We know that %o3 >= %o5, so we can do the + ! first divide step without thinking. BUT, the others are conditional, + ! and are only done if %o3 >= 0. Because both %o3 and %o5 may have the high- + ! order bit set in the first step, just falling into the regular + ! division loop will mess up the first time around. + ! So we unroll slightly... + Ldo_single_div: + subcc %g7, 1, %g7 + bl Lend_regular_divide + nop + + sub %o3, %o5, %o3 + mov 1, %o2 + + b Lend_single_divloop + nop + Lsingle_divloop: + sll %o2, 1, %o2 + + bl 1f + srl %o5, 1, %o5 + ! %o3 >= 0 + sub %o3, %o5, %o3 + + b 2f + add %o2, 1, %o2 + 1: ! %o3 < 0 + add %o3, %o5, %o3 + sub %o2, 1, %o2 + 2: + Lend_single_divloop: + subcc %g7, 1, %g7 + bge Lsingle_divloop + tst %o3 + + b,a Lend_regular_divide + +Lnot_really_big: +1: + sll %o5, 4, %o5 + cmp %o5, %o3 + bleu 1b + addcc %o4, 1, %o4 + + be Lgot_result + sub %o4, 1, %o4 + + tst %o3 ! set up for initial iteration +Ldivloop: + sll %o2, 4, %o2 + ! depth 1, accumulated bits 0 + bl L.1.16 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 2, accumulated bits 1 + bl L.2.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 3, accumulated bits 3 + bl L.3.19 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits 7 + bl L.4.23 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (7*2+1), %o2 + +L.4.23: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (7*2-1), %o2 + +L.3.19: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits 5 + bl L.4.21 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (5*2+1), %o2 + +L.4.21: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (5*2-1), %o2 + +L.2.17: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 3, accumulated bits 1 + bl L.3.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits 3 + bl L.4.19 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (3*2+1), %o2 + +L.4.19: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (3*2-1), %o2 + + +L.3.17: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits 1 + bl L.4.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (1*2+1), %o2 + +L.4.17: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (1*2-1), %o2 + +L.1.16: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 2, accumulated bits -1 + bl L.2.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 3, accumulated bits -1 + bl L.3.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits -1 + bl L.4.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-1*2+1), %o2 + +L.4.15: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-1*2-1), %o2 + +L.3.15: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits -3 + bl L.4.13 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-3*2+1), %o2 + +L.4.13: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-3*2-1), %o2 + +L.2.15: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 3, accumulated bits -3 + bl L.3.13 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits -5 + bl L.4.11 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-5*2+1), %o2 + +L.4.11: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-5*2-1), %o2 + +L.3.13: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits -7 + bl L.4.9 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-7*2+1), %o2 + +L.4.9: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-7*2-1), %o2 + + 9: +Lend_regular_divide: + subcc %o4, 1, %o4 + bge Ldivloop + tst %o3 + + bl,a Lgot_result + ! non-restoring fixup here (one instruction only!) + sub %o2, 1, %o2 + +Lgot_result: + ! check to see if answer should be < 0 + tst %g2 + bl,a 1f + sub %g0, %o2, %o2 +1: + retl + mov %o2, %o0 + + .globl .div_patch +.div_patch: + sra %o0, 0x1f, %o2 + wr %o2, 0x0, %y + nop + nop + nop + sdivcc %o0, %o1, %o0 + bvs,a 1f + xnor %o0, %g0, %o0 +1: retl + nop diff --git a/arch/sparc/lib/strlen.S b/arch/sparc/lib/strlen.S new file mode 100644 index 00000000000..ed9a763368c --- /dev/null +++ b/arch/sparc/lib/strlen.S @@ -0,0 +1,81 @@ +/* strlen.S: Sparc optimized strlen code + * Hand optimized from GNU libc's strlen + * Copyright (C) 1991,1996 Free Software Foundation + * Copyright (C) 1996 David S. Miller (davem@caip.rutgers.edu) + * Copyright (C) 1996 Jakub Jelinek (jj@sunsite.mff.cuni.cz) + */ + +#define LO_MAGIC 0x01010101 +#define HI_MAGIC 0x80808080 + +0: + ldub [%o0], %o5 + cmp %o5, 0 + be 1f + add %o0, 1, %o0 + andcc %o0, 3, %g0 + be 4f + or %o4, %lo(HI_MAGIC), %o3 + ldub [%o0], %o5 + cmp %o5, 0 + be 2f + add %o0, 1, %o0 + andcc %o0, 3, %g0 + be 5f + sethi %hi(LO_MAGIC), %o4 + ldub [%o0], %o5 + cmp %o5, 0 + be 3f + add %o0, 1, %o0 + b 8f + or %o4, %lo(LO_MAGIC), %o2 +1: + retl + mov 0, %o0 +2: + retl + mov 1, %o0 +3: + retl + mov 2, %o0 + + .align 4 + .global strlen +strlen: + mov %o0, %o1 + andcc %o0, 3, %g0 + bne 0b + sethi %hi(HI_MAGIC), %o4 + or %o4, %lo(HI_MAGIC), %o3 +4: + sethi %hi(LO_MAGIC), %o4 +5: + or %o4, %lo(LO_MAGIC), %o2 +8: + ld [%o0], %o5 +2: + sub %o5, %o2, %o4 + andcc %o4, %o3, %g0 + be 8b + add %o0, 4, %o0 + + /* Check every byte. */ + srl %o5, 24, %g5 + andcc %g5, 0xff, %g0 + be 1f + add %o0, -4, %o4 + srl %o5, 16, %g5 + andcc %g5, 0xff, %g0 + be 1f + add %o4, 1, %o4 + srl %o5, 8, %g5 + andcc %g5, 0xff, %g0 + be 1f + add %o4, 1, %o4 + andcc %o5, 0xff, %g0 + bne,a 2b + ld [%o0], %o5 + add %o4, 1, %o4 +1: + retl + sub %o4, %o1, %o0 diff --git a/arch/sparc/lib/strlen_user.S b/arch/sparc/lib/strlen_user.S new file mode 100644 index 00000000000..8c8a371df3c --- /dev/null +++ b/arch/sparc/lib/strlen_user.S @@ -0,0 +1,109 @@ +/* strlen_user.S: Sparc optimized strlen_user code + * + * Return length of string in userspace including terminating 0 + * or 0 for error + * + * Copyright (C) 1991,1996 Free Software Foundation + * Copyright (C) 1996 David S. Miller (davem@caip.rutgers.edu) + * Copyright (C) 1996 Jakub Jelinek (jj@sunsite.mff.cuni.cz) + */ + +#define LO_MAGIC 0x01010101 +#define HI_MAGIC 0x80808080 + +10: + ldub [%o0], %o5 + cmp %o5, 0 + be 1f + add %o0, 1, %o0 + andcc %o0, 3, %g0 + be 4f + or %o4, %lo(HI_MAGIC), %o3 +11: + ldub [%o0], %o5 + cmp %o5, 0 + be 2f + add %o0, 1, %o0 + andcc %o0, 3, %g0 + be 5f + sethi %hi(LO_MAGIC), %o4 +12: + ldub [%o0], %o5 + cmp %o5, 0 + be 3f + add %o0, 1, %o0 + b 13f + or %o4, %lo(LO_MAGIC), %o2 +1: + retl + mov 1, %o0 +2: + retl + mov 2, %o0 +3: + retl + mov 3, %o0 + + .align 4 + .global __strlen_user, __strnlen_user +__strlen_user: + sethi %hi(32768), %o1 +__strnlen_user: + mov %o1, %g1 + mov %o0, %o1 + andcc %o0, 3, %g0 + bne 10b + sethi %hi(HI_MAGIC), %o4 + or %o4, %lo(HI_MAGIC), %o3 +4: + sethi %hi(LO_MAGIC), %o4 +5: + or %o4, %lo(LO_MAGIC), %o2 +13: + ld [%o0], %o5 +2: + sub %o5, %o2, %o4 + andcc %o4, %o3, %g0 + bne 82f + add %o0, 4, %o0 + sub %o0, %o1, %g2 +81: cmp %g2, %g1 + blu 13b + mov %o0, %o4 + ba,a 1f + + /* Check every byte. */ +82: srl %o5, 24, %g5 + andcc %g5, 0xff, %g0 + be 1f + add %o0, -3, %o4 + srl %o5, 16, %g5 + andcc %g5, 0xff, %g0 + be 1f + add %o4, 1, %o4 + srl %o5, 8, %g5 + andcc %g5, 0xff, %g0 + be 1f + add %o4, 1, %o4 + andcc %o5, 0xff, %g0 + bne 81b + sub %o0, %o1, %g2 + + add %o4, 1, %o4 +1: + retl + sub %o4, %o1, %o0 + + .section .fixup,#alloc,#execinstr + .align 4 +9: + retl + clr %o0 + + .section __ex_table,#alloc + .align 4 + + .word 10b, 9b + .word 11b, 9b + .word 12b, 9b + .word 13b, 9b diff --git a/arch/sparc/lib/strncmp.S b/arch/sparc/lib/strncmp.S new file mode 100644 index 00000000000..615626805d4 --- /dev/null +++ b/arch/sparc/lib/strncmp.S @@ -0,0 +1,118 @@ +/* $Id: strncmp.S,v 1.2 1996/09/09 02:47:20 davem Exp $ + * strncmp.S: Hand optimized Sparc assembly of GCC output from GNU libc + * generic strncmp routine. + */ + + .text + .align 4 + .global __strncmp, strncmp +__strncmp: +strncmp: + mov %o0, %g3 + mov 0, %o3 + + cmp %o2, 3 + ble 7f + mov 0, %g2 + + sra %o2, 2, %o4 + ldub [%g3], %o3 + +0: + ldub [%o1], %g2 + add %g3, 1, %g3 + and %o3, 0xff, %o0 + + cmp %o0, 0 + be 8f + add %o1, 1, %o1 + + cmp %o0, %g2 + be,a 1f + ldub [%g3], %o3 + + retl + sub %o0, %g2, %o0 + +1: + ldub [%o1], %g2 + add %g3,1, %g3 + and %o3, 0xff, %o0 + + cmp %o0, 0 + be 8f + add %o1, 1, %o1 + + cmp %o0, %g2 + be,a 1f + ldub [%g3], %o3 + + retl + sub %o0, %g2, %o0 + +1: + ldub [%o1], %g2 + add %g3, 1, %g3 + and %o3, 0xff, %o0 + + cmp %o0, 0 + be 8f + add %o1, 1, %o1 + + cmp %o0, %g2 + be,a 1f + ldub [%g3], %o3 + + retl + sub %o0, %g2, %o0 + +1: + ldub [%o1], %g2 + add %g3, 1, %g3 + and %o3, 0xff, %o0 + + cmp %o0, 0 + be 8f + add %o1, 1, %o1 + + cmp %o0, %g2 + be 1f + add %o4, -1, %o4 + + retl + sub %o0, %g2, %o0 + +1: + + cmp %o4, 0 + bg,a 0b + ldub [%g3], %o3 + + b 7f + and %o2, 3, %o2 + +9: + ldub [%o1], %g2 + add %g3, 1, %g3 + and %o3, 0xff, %o0 + + cmp %o0, 0 + be 8f + add %o1, 1, %o1 + + cmp %o0, %g2 + be 7f + add %o2, -1, %o2 + +8: + retl + sub %o0, %g2, %o0 + +7: + cmp %o2, 0 + bg,a 9b + ldub [%g3], %o3 + + and %g2, 0xff, %o0 + retl + sub %o3, %o0, %o0 diff --git a/arch/sparc/lib/strncpy_from_user.S b/arch/sparc/lib/strncpy_from_user.S new file mode 100644 index 00000000000..d77198976a6 --- /dev/null +++ b/arch/sparc/lib/strncpy_from_user.S @@ -0,0 +1,47 @@ +/* strncpy_from_user.S: Sparc strncpy from userspace. + * + * Copyright(C) 1996 David S. Miller + */ + +#include <asm/ptrace.h> +#include <asm/errno.h> + + .text + .align 4 + + /* Must return: + * + * -EFAULT for an exception + * count if we hit the buffer limit + * bytes copied if we hit a null byte + */ + + .globl __strncpy_from_user +__strncpy_from_user: + /* %o0=dest, %o1=src, %o2=count */ + mov %o2, %o3 +1: + subcc %o2, 1, %o2 + bneg 2f + nop +10: + ldub [%o1], %o4 + add %o0, 1, %o0 + cmp %o4, 0 + add %o1, 1, %o1 + bne 1b + stb %o4, [%o0 - 1] +2: + add %o2, 1, %o0 + retl + sub %o3, %o0, %o0 + + .section .fixup,#alloc,#execinstr + .align 4 +4: + retl + mov -EFAULT, %o0 + + .section __ex_table,#alloc + .align 4 + .word 10b, 4b diff --git a/arch/sparc/lib/udiv.S b/arch/sparc/lib/udiv.S new file mode 100644 index 00000000000..2abfc6b0f3e --- /dev/null +++ b/arch/sparc/lib/udiv.S @@ -0,0 +1,355 @@ +/* $Id: udiv.S,v 1.4 1996/09/30 02:22:38 davem Exp $ + * udiv.S: This routine was taken from glibc-1.09 and is covered + * by the GNU Library General Public License Version 2. + */ + + +/* This file is generated from divrem.m4; DO NOT EDIT! */ +/* + * Division and remainder, from Appendix E of the Sparc Version 8 + * Architecture Manual, with fixes from Gordon Irlam. + */ + +/* + * Input: dividend and divisor in %o0 and %o1 respectively. + * + * m4 parameters: + * .udiv name of function to generate + * div div=div => %o0 / %o1; div=rem => %o0 % %o1 + * false false=true => signed; false=false => unsigned + * + * Algorithm parameters: + * N how many bits per iteration we try to get (4) + * WORDSIZE total number of bits (32) + * + * Derived constants: + * TOPBITS number of bits in the top decade of a number + * + * Important variables: + * Q the partial quotient under development (initially 0) + * R the remainder so far, initially the dividend + * ITER number of main division loop iterations required; + * equal to ceil(log2(quotient) / N). Note that this + * is the log base (2^N) of the quotient. + * V the current comparand, initially divisor*2^(ITER*N-1) + * + * Cost: + * Current estimate for non-large dividend is + * ceil(log2(quotient) / N) * (10 + 7N/2) + C + * A large dividend is one greater than 2^(31-TOPBITS) and takes a + * different path, as the upper bits of the quotient must be developed + * one bit at a time. + */ + + + .globl .udiv +.udiv: + + ! Ready to divide. Compute size of quotient; scale comparand. + orcc %o1, %g0, %o5 + bne 1f + mov %o0, %o3 + + ! Divide by zero trap. If it returns, return 0 (about as + ! wrong as possible, but that is what SunOS does...). + ta ST_DIV0 + retl + clr %o0 + +1: + cmp %o3, %o5 ! if %o1 exceeds %o0, done + blu Lgot_result ! (and algorithm fails otherwise) + clr %o2 + + sethi %hi(1 << (32 - 4 - 1)), %g1 + + cmp %o3, %g1 + blu Lnot_really_big + clr %o4 + + ! Here the dividend is >= 2**(31-N) or so. We must be careful here, + ! as our usual N-at-a-shot divide step will cause overflow and havoc. + ! The number of bits in the result here is N*ITER+SC, where SC <= N. + ! Compute ITER in an unorthodox manner: know we need to shift V into + ! the top decade: so do not even bother to compare to R. + 1: + cmp %o5, %g1 + bgeu 3f + mov 1, %g7 + + sll %o5, 4, %o5 + + b 1b + add %o4, 1, %o4 + + ! Now compute %g7. + 2: + addcc %o5, %o5, %o5 + bcc Lnot_too_big + add %g7, 1, %g7 + + ! We get here if the %o1 overflowed while shifting. + ! This means that %o3 has the high-order bit set. + ! Restore %o5 and subtract from %o3. + sll %g1, 4, %g1 ! high order bit + srl %o5, 1, %o5 ! rest of %o5 + add %o5, %g1, %o5 + + b Ldo_single_div + sub %g7, 1, %g7 + + Lnot_too_big: + 3: + cmp %o5, %o3 + blu 2b + nop + + be Ldo_single_div + nop + /* NB: these are commented out in the V8-Sparc manual as well */ + /* (I do not understand this) */ + ! %o5 > %o3: went too far: back up 1 step + ! srl %o5, 1, %o5 + ! dec %g7 + ! do single-bit divide steps + ! + ! We have to be careful here. We know that %o3 >= %o5, so we can do the + ! first divide step without thinking. BUT, the others are conditional, + ! and are only done if %o3 >= 0. Because both %o3 and %o5 may have the high- + ! order bit set in the first step, just falling into the regular + ! division loop will mess up the first time around. + ! So we unroll slightly... + Ldo_single_div: + subcc %g7, 1, %g7 + bl Lend_regular_divide + nop + + sub %o3, %o5, %o3 + mov 1, %o2 + + b Lend_single_divloop + nop + Lsingle_divloop: + sll %o2, 1, %o2 + bl 1f + srl %o5, 1, %o5 + ! %o3 >= 0 + sub %o3, %o5, %o3 + b 2f + add %o2, 1, %o2 + 1: ! %o3 < 0 + add %o3, %o5, %o3 + sub %o2, 1, %o2 + 2: + Lend_single_divloop: + subcc %g7, 1, %g7 + bge Lsingle_divloop + tst %o3 + + b,a Lend_regular_divide + +Lnot_really_big: +1: + sll %o5, 4, %o5 + + cmp %o5, %o3 + bleu 1b + addcc %o4, 1, %o4 + + be Lgot_result + sub %o4, 1, %o4 + + tst %o3 ! set up for initial iteration +Ldivloop: + sll %o2, 4, %o2 + ! depth 1, accumulated bits 0 + bl L.1.16 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 2, accumulated bits 1 + bl L.2.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 3, accumulated bits 3 + bl L.3.19 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits 7 + bl L.4.23 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (7*2+1), %o2 + +L.4.23: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (7*2-1), %o2 + +L.3.19: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits 5 + bl L.4.21 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (5*2+1), %o2 + +L.4.21: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (5*2-1), %o2 + +L.2.17: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 3, accumulated bits 1 + bl L.3.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits 3 + bl L.4.19 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (3*2+1), %o2 + +L.4.19: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (3*2-1), %o2 + +L.3.17: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits 1 + bl L.4.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (1*2+1), %o2 + +L.4.17: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (1*2-1), %o2 + +L.1.16: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 2, accumulated bits -1 + bl L.2.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 3, accumulated bits -1 + bl L.3.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits -1 + bl L.4.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-1*2+1), %o2 + +L.4.15: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-1*2-1), %o2 + +L.3.15: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits -3 + bl L.4.13 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-3*2+1), %o2 + +L.4.13: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-3*2-1), %o2 + +L.2.15: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 3, accumulated bits -3 + bl L.3.13 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits -5 + bl L.4.11 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-5*2+1), %o2 + +L.4.11: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-5*2-1), %o2 + +L.3.13: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits -7 + bl L.4.9 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-7*2+1), %o2 + +L.4.9: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-7*2-1), %o2 + + 9: +Lend_regular_divide: + subcc %o4, 1, %o4 + bge Ldivloop + tst %o3 + + bl,a Lgot_result + ! non-restoring fixup here (one instruction only!) + sub %o2, 1, %o2 + +Lgot_result: + + retl + mov %o2, %o0 + + .globl .udiv_patch +.udiv_patch: + wr %g0, 0x0, %y + nop + nop + retl + udiv %o0, %o1, %o0 + nop diff --git a/arch/sparc/lib/udivdi3.S b/arch/sparc/lib/udivdi3.S new file mode 100644 index 00000000000..b430f1f0ef6 --- /dev/null +++ b/arch/sparc/lib/udivdi3.S @@ -0,0 +1,258 @@ +/* Copyright (C) 1989, 1992, 1993, 1994, 1995 Free Software Foundation, Inc. + +This file is part of GNU CC. + +GNU CC is free software; you can redistribute it and/or modify +it under the terms of the GNU General Public License as published by +the Free Software Foundation; either version 2, or (at your option) +any later version. + +GNU CC 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 General Public License for more details. + +You should have received a copy of the GNU General Public License +along with GNU CC; see the file COPYING. If not, write to +the Free Software Foundation, 59 Temple Place - Suite 330, +Boston, MA 02111-1307, USA. */ + + .text + .align 4 + .globl __udivdi3 +__udivdi3: + save %sp,-104,%sp + mov %i3,%o3 + cmp %i2,0 + bne .LL40 + mov %i1,%i3 + cmp %o3,%i0 + bleu .LL41 + mov %i3,%o1 + ! Inlined udiv_qrnnd + mov 32,%g1 + subcc %i0,%o3,%g0 +1: bcs 5f + addxcc %o1,%o1,%o1 ! shift n1n0 and a q-bit in lsb + sub %i0,%o3,%i0 ! this kills msb of n + addx %i0,%i0,%i0 ! so this cannot give carry + subcc %g1,1,%g1 +2: bne 1b + subcc %i0,%o3,%g0 + bcs 3f + addxcc %o1,%o1,%o1 ! shift n1n0 and a q-bit in lsb + b 3f + sub %i0,%o3,%i0 ! this kills msb of n +4: sub %i0,%o3,%i0 +5: addxcc %i0,%i0,%i0 + bcc 2b + subcc %g1,1,%g1 +! Got carry from n. Subtract next step to cancel this carry. + bne 4b + addcc %o1,%o1,%o1 ! shift n1n0 and a 0-bit in lsb + sub %i0,%o3,%i0 +3: xnor %o1,0,%o1 + ! End of inline udiv_qrnnd + b .LL45 + mov 0,%o2 +.LL41: + cmp %o3,0 + bne .LL77 + mov %i0,%o2 + mov 1,%o0 + call .udiv,0 + mov 0,%o1 + mov %o0,%o3 + mov %i0,%o2 +.LL77: + mov 0,%o4 + ! Inlined udiv_qrnnd + mov 32,%g1 + subcc %o4,%o3,%g0 +1: bcs 5f + addxcc %o2,%o2,%o2 ! shift n1n0 and a q-bit in lsb + sub %o4,%o3,%o4 ! this kills msb of n + addx %o4,%o4,%o4 ! so this cannot give carry + subcc %g1,1,%g1 +2: bne 1b + subcc %o4,%o3,%g0 + bcs 3f + addxcc %o2,%o2,%o2 ! shift n1n0 and a q-bit in lsb + b 3f + sub %o4,%o3,%o4 ! this kills msb of n +4: sub %o4,%o3,%o4 +5: addxcc %o4,%o4,%o4 + bcc 2b + subcc %g1,1,%g1 +! Got carry from n. Subtract next step to cancel this carry. + bne 4b + addcc %o2,%o2,%o2 ! shift n1n0 and a 0-bit in lsb + sub %o4,%o3,%o4 +3: xnor %o2,0,%o2 + ! End of inline udiv_qrnnd + mov %o4,%i0 + mov %i3,%o1 + ! Inlined udiv_qrnnd + mov 32,%g1 + subcc %i0,%o3,%g0 +1: bcs 5f + addxcc %o1,%o1,%o1 ! shift n1n0 and a q-bit in lsb + sub %i0,%o3,%i0 ! this kills msb of n + addx %i0,%i0,%i0 ! so this cannot give carry + subcc %g1,1,%g1 +2: bne 1b + subcc %i0,%o3,%g0 + bcs 3f + addxcc %o1,%o1,%o1 ! shift n1n0 and a q-bit in lsb + b 3f + sub %i0,%o3,%i0 ! this kills msb of n +4: sub %i0,%o3,%i0 +5: addxcc %i0,%i0,%i0 + bcc 2b + subcc %g1,1,%g1 +! Got carry from n. Subtract next step to cancel this carry. + bne 4b + addcc %o1,%o1,%o1 ! shift n1n0 and a 0-bit in lsb + sub %i0,%o3,%i0 +3: xnor %o1,0,%o1 + ! End of inline udiv_qrnnd + b .LL78 + mov %o1,%l1 +.LL40: + cmp %i2,%i0 + bleu .LL46 + sethi %hi(65535),%o0 + b .LL73 + mov 0,%o1 +.LL46: + or %o0,%lo(65535),%o0 + cmp %i2,%o0 + bgu .LL53 + mov %i2,%o1 + cmp %i2,256 + addx %g0,-1,%o0 + b .LL59 + and %o0,8,%o2 +.LL53: + sethi %hi(16777215),%o0 + or %o0,%lo(16777215),%o0 + cmp %o1,%o0 + bgu .LL59 + mov 24,%o2 + mov 16,%o2 +.LL59: + srl %o1,%o2,%o1 + sethi %hi(__clz_tab),%o0 + or %o0,%lo(__clz_tab),%o0 + ldub [%o1+%o0],%o0 + add %o0,%o2,%o0 + mov 32,%o1 + subcc %o1,%o0,%o2 + bne,a .LL67 + mov 32,%o0 + cmp %i0,%i2 + bgu .LL69 + cmp %i3,%o3 + blu .LL73 + mov 0,%o1 +.LL69: + b .LL73 + mov 1,%o1 +.LL67: + sub %o0,%o2,%o0 + sll %i2,%o2,%i2 + srl %o3,%o0,%o1 + or %i2,%o1,%i2 + sll %o3,%o2,%o3 + srl %i0,%o0,%o1 + sll %i0,%o2,%i0 + srl %i3,%o0,%o0 + or %i0,%o0,%i0 + sll %i3,%o2,%i3 + mov %i0,%o5 + mov %o1,%o4 + ! Inlined udiv_qrnnd + mov 32,%g1 + subcc %o4,%i2,%g0 +1: bcs 5f + addxcc %o5,%o5,%o5 ! shift n1n0 and a q-bit in lsb + sub %o4,%i2,%o4 ! this kills msb of n + addx %o4,%o4,%o4 ! so this cannot give carry + subcc %g1,1,%g1 +2: bne 1b + subcc %o4,%i2,%g0 + bcs 3f + addxcc %o5,%o5,%o5 ! shift n1n0 and a q-bit in lsb + b 3f + sub %o4,%i2,%o4 ! this kills msb of n +4: sub %o4,%i2,%o4 +5: addxcc %o4,%o4,%o4 + bcc 2b + subcc %g1,1,%g1 +! Got carry from n. Subtract next step to cancel this carry. + bne 4b + addcc %o5,%o5,%o5 ! shift n1n0 and a 0-bit in lsb + sub %o4,%i2,%o4 +3: xnor %o5,0,%o5 + ! End of inline udiv_qrnnd + mov %o4,%i0 + mov %o5,%o1 + ! Inlined umul_ppmm + wr %g0,%o1,%y ! SPARC has 0-3 delay insn after a wr + sra %o3,31,%g2 ! Do not move this insn + and %o1,%g2,%g2 ! Do not move this insn + andcc %g0,0,%g1 ! Do not move this insn + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,%o3,%g1 + mulscc %g1,0,%g1 + add %g1,%g2,%o0 + rd %y,%o2 + cmp %o0,%i0 + bgu,a .LL73 + add %o1,-1,%o1 + bne,a .LL45 + mov 0,%o2 + cmp %o2,%i3 + bleu .LL45 + mov 0,%o2 + add %o1,-1,%o1 +.LL73: + mov 0,%o2 +.LL45: + mov %o1,%l1 +.LL78: + mov %o2,%l0 + mov %l0,%i0 + mov %l1,%i1 + ret + restore diff --git a/arch/sparc/lib/umul.S b/arch/sparc/lib/umul.S new file mode 100644 index 00000000000..a784720a8a2 --- /dev/null +++ b/arch/sparc/lib/umul.S @@ -0,0 +1,169 @@ +/* $Id: umul.S,v 1.4 1996/09/30 02:22:39 davem Exp $ + * umul.S: This routine was taken from glibc-1.09 and is covered + * by the GNU Library General Public License Version 2. + */ + + +/* + * Unsigned multiply. Returns %o0 * %o1 in %o1%o0 (i.e., %o1 holds the + * upper 32 bits of the 64-bit product). + * + * This code optimizes short (less than 13-bit) multiplies. Short + * multiplies require 25 instruction cycles, and long ones require + * 45 instruction cycles. + * + * On return, overflow has occurred (%o1 is not zero) if and only if + * the Z condition code is clear, allowing, e.g., the following: + * + * call .umul + * nop + * bnz overflow (or tnz) + */ + + .globl .umul +.umul: + or %o0, %o1, %o4 + mov %o0, %y ! multiplier -> Y + + andncc %o4, 0xfff, %g0 ! test bits 12..31 of *both* args + be Lmul_shortway ! if zero, can do it the short way + andcc %g0, %g0, %o4 ! zero the partial product and clear N and V + + /* + * Long multiply. 32 steps, followed by a final shift step. + */ + mulscc %o4, %o1, %o4 ! 1 + mulscc %o4, %o1, %o4 ! 2 + mulscc %o4, %o1, %o4 ! 3 + mulscc %o4, %o1, %o4 ! 4 + mulscc %o4, %o1, %o4 ! 5 + mulscc %o4, %o1, %o4 ! 6 + mulscc %o4, %o1, %o4 ! 7 + mulscc %o4, %o1, %o4 ! 8 + mulscc %o4, %o1, %o4 ! 9 + mulscc %o4, %o1, %o4 ! 10 + mulscc %o4, %o1, %o4 ! 11 + mulscc %o4, %o1, %o4 ! 12 + mulscc %o4, %o1, %o4 ! 13 + mulscc %o4, %o1, %o4 ! 14 + mulscc %o4, %o1, %o4 ! 15 + mulscc %o4, %o1, %o4 ! 16 + mulscc %o4, %o1, %o4 ! 17 + mulscc %o4, %o1, %o4 ! 18 + mulscc %o4, %o1, %o4 ! 19 + mulscc %o4, %o1, %o4 ! 20 + mulscc %o4, %o1, %o4 ! 21 + mulscc %o4, %o1, %o4 ! 22 + mulscc %o4, %o1, %o4 ! 23 + mulscc %o4, %o1, %o4 ! 24 + mulscc %o4, %o1, %o4 ! 25 + mulscc %o4, %o1, %o4 ! 26 + mulscc %o4, %o1, %o4 ! 27 + mulscc %o4, %o1, %o4 ! 28 + mulscc %o4, %o1, %o4 ! 29 + mulscc %o4, %o1, %o4 ! 30 + mulscc %o4, %o1, %o4 ! 31 + mulscc %o4, %o1, %o4 ! 32 + mulscc %o4, %g0, %o4 ! final shift + + + /* + * Normally, with the shift-and-add approach, if both numbers are + * positive you get the correct result. With 32-bit two's-complement + * numbers, -x is represented as + * + * x 32 + * ( 2 - ------ ) mod 2 * 2 + * 32 + * 2 + * + * (the `mod 2' subtracts 1 from 1.bbbb). To avoid lots of 2^32s, + * we can treat this as if the radix point were just to the left + * of the sign bit (multiply by 2^32), and get + * + * -x = (2 - x) mod 2 + * + * Then, ignoring the `mod 2's for convenience: + * + * x * y = xy + * -x * y = 2y - xy + * x * -y = 2x - xy + * -x * -y = 4 - 2x - 2y + xy + * + * For signed multiplies, we subtract (x << 32) from the partial + * product to fix this problem for negative multipliers (see mul.s). + * Because of the way the shift into the partial product is calculated + * (N xor V), this term is automatically removed for the multiplicand, + * so we don't have to adjust. + * + * But for unsigned multiplies, the high order bit wasn't a sign bit, + * and the correction is wrong. So for unsigned multiplies where the + * high order bit is one, we end up with xy - (y << 32). To fix it + * we add y << 32. + */ +#if 0 + tst %o1 + bl,a 1f ! if %o1 < 0 (high order bit = 1), + add %o4, %o0, %o4 ! %o4 += %o0 (add y to upper half) + +1: + rd %y, %o0 ! get lower half of product + retl + addcc %o4, %g0, %o1 ! put upper half in place and set Z for %o1==0 +#else + /* Faster code from tege@sics.se. */ + sra %o1, 31, %o2 ! make mask from sign bit + and %o0, %o2, %o2 ! %o2 = 0 or %o0, depending on sign of %o1 + rd %y, %o0 ! get lower half of product + retl + addcc %o4, %o2, %o1 ! add compensation and put upper half in place +#endif + +Lmul_shortway: + /* + * Short multiply. 12 steps, followed by a final shift step. + * The resulting bits are off by 12 and (32-12) = 20 bit positions, + * but there is no problem with %o0 being negative (unlike above), + * and overflow is impossible (the answer is at most 24 bits long). + */ + mulscc %o4, %o1, %o4 ! 1 + mulscc %o4, %o1, %o4 ! 2 + mulscc %o4, %o1, %o4 ! 3 + mulscc %o4, %o1, %o4 ! 4 + mulscc %o4, %o1, %o4 ! 5 + mulscc %o4, %o1, %o4 ! 6 + mulscc %o4, %o1, %o4 ! 7 + mulscc %o4, %o1, %o4 ! 8 + mulscc %o4, %o1, %o4 ! 9 + mulscc %o4, %o1, %o4 ! 10 + mulscc %o4, %o1, %o4 ! 11 + mulscc %o4, %o1, %o4 ! 12 + mulscc %o4, %g0, %o4 ! final shift + + /* + * %o4 has 20 of the bits that should be in the result; %y has + * the bottom 12 (as %y's top 12). That is: + * + * %o4 %y + * +----------------+----------------+ + * | -12- | -20- | -12- | -20- | + * +------(---------+------)---------+ + * -----result----- + * + * The 12 bits of %o4 left of the `result' area are all zero; + * in fact, all top 20 bits of %o4 are zero. + */ + + rd %y, %o5 + sll %o4, 12, %o0 ! shift middle bits left 12 + srl %o5, 20, %o5 ! shift low bits right 20 + or %o5, %o0, %o0 + retl + addcc %g0, %g0, %o1 ! %o1 = zero, and set Z + + .globl .umul_patch +.umul_patch: + umul %o0, %o1, %o0 + retl + rd %y, %o1 + nop diff --git a/arch/sparc/lib/urem.S b/arch/sparc/lib/urem.S new file mode 100644 index 00000000000..ec7f0c502c5 --- /dev/null +++ b/arch/sparc/lib/urem.S @@ -0,0 +1,355 @@ +/* $Id: urem.S,v 1.4 1996/09/30 02:22:42 davem Exp $ + * urem.S: This routine was taken from glibc-1.09 and is covered + * by the GNU Library General Public License Version 2. + */ + +/* This file is generated from divrem.m4; DO NOT EDIT! */ +/* + * Division and remainder, from Appendix E of the Sparc Version 8 + * Architecture Manual, with fixes from Gordon Irlam. + */ + +/* + * Input: dividend and divisor in %o0 and %o1 respectively. + * + * m4 parameters: + * .urem name of function to generate + * rem rem=div => %o0 / %o1; rem=rem => %o0 % %o1 + * false false=true => signed; false=false => unsigned + * + * Algorithm parameters: + * N how many bits per iteration we try to get (4) + * WORDSIZE total number of bits (32) + * + * Derived constants: + * TOPBITS number of bits in the top decade of a number + * + * Important variables: + * Q the partial quotient under development (initially 0) + * R the remainder so far, initially the dividend + * ITER number of main division loop iterations required; + * equal to ceil(log2(quotient) / N). Note that this + * is the log base (2^N) of the quotient. + * V the current comparand, initially divisor*2^(ITER*N-1) + * + * Cost: + * Current estimate for non-large dividend is + * ceil(log2(quotient) / N) * (10 + 7N/2) + C + * A large dividend is one greater than 2^(31-TOPBITS) and takes a + * different path, as the upper bits of the quotient must be developed + * one bit at a time. + */ + + .globl .urem +.urem: + + ! Ready to divide. Compute size of quotient; scale comparand. + orcc %o1, %g0, %o5 + bne 1f + mov %o0, %o3 + + ! Divide by zero trap. If it returns, return 0 (about as + ! wrong as possible, but that is what SunOS does...). + ta ST_DIV0 + retl + clr %o0 + +1: + cmp %o3, %o5 ! if %o1 exceeds %o0, done + blu Lgot_result ! (and algorithm fails otherwise) + clr %o2 + + sethi %hi(1 << (32 - 4 - 1)), %g1 + + cmp %o3, %g1 + blu Lnot_really_big + clr %o4 + + ! Here the dividend is >= 2**(31-N) or so. We must be careful here, + ! as our usual N-at-a-shot divide step will cause overflow and havoc. + ! The number of bits in the result here is N*ITER+SC, where SC <= N. + ! Compute ITER in an unorthodox manner: know we need to shift V into + ! the top decade: so do not even bother to compare to R. + 1: + cmp %o5, %g1 + bgeu 3f + mov 1, %g7 + + sll %o5, 4, %o5 + + b 1b + add %o4, 1, %o4 + + ! Now compute %g7. + 2: + addcc %o5, %o5, %o5 + bcc Lnot_too_big + add %g7, 1, %g7 + + ! We get here if the %o1 overflowed while shifting. + ! This means that %o3 has the high-order bit set. + ! Restore %o5 and subtract from %o3. + sll %g1, 4, %g1 ! high order bit + srl %o5, 1, %o5 ! rest of %o5 + add %o5, %g1, %o5 + + b Ldo_single_div + sub %g7, 1, %g7 + + Lnot_too_big: + 3: + cmp %o5, %o3 + blu 2b + nop + + be Ldo_single_div + nop + /* NB: these are commented out in the V8-Sparc manual as well */ + /* (I do not understand this) */ + ! %o5 > %o3: went too far: back up 1 step + ! srl %o5, 1, %o5 + ! dec %g7 + ! do single-bit divide steps + ! + ! We have to be careful here. We know that %o3 >= %o5, so we can do the + ! first divide step without thinking. BUT, the others are conditional, + ! and are only done if %o3 >= 0. Because both %o3 and %o5 may have the high- + ! order bit set in the first step, just falling into the regular + ! division loop will mess up the first time around. + ! So we unroll slightly... + Ldo_single_div: + subcc %g7, 1, %g7 + bl Lend_regular_divide + nop + + sub %o3, %o5, %o3 + mov 1, %o2 + + b Lend_single_divloop + nop + Lsingle_divloop: + sll %o2, 1, %o2 + bl 1f + srl %o5, 1, %o5 + ! %o3 >= 0 + sub %o3, %o5, %o3 + b 2f + add %o2, 1, %o2 + 1: ! %o3 < 0 + add %o3, %o5, %o3 + sub %o2, 1, %o2 + 2: + Lend_single_divloop: + subcc %g7, 1, %g7 + bge Lsingle_divloop + tst %o3 + + b,a Lend_regular_divide + +Lnot_really_big: +1: + sll %o5, 4, %o5 + + cmp %o5, %o3 + bleu 1b + addcc %o4, 1, %o4 + + be Lgot_result + sub %o4, 1, %o4 + + tst %o3 ! set up for initial iteration +Ldivloop: + sll %o2, 4, %o2 + ! depth 1, accumulated bits 0 + bl L.1.16 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 2, accumulated bits 1 + bl L.2.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 3, accumulated bits 3 + bl L.3.19 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits 7 + bl L.4.23 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (7*2+1), %o2 + +L.4.23: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (7*2-1), %o2 + +L.3.19: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits 5 + bl L.4.21 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (5*2+1), %o2 + +L.4.21: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (5*2-1), %o2 + +L.2.17: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 3, accumulated bits 1 + bl L.3.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits 3 + bl L.4.19 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (3*2+1), %o2 + +L.4.19: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (3*2-1), %o2 + +L.3.17: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits 1 + bl L.4.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (1*2+1), %o2 + +L.4.17: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (1*2-1), %o2 + +L.1.16: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 2, accumulated bits -1 + bl L.2.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 3, accumulated bits -1 + bl L.3.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits -1 + bl L.4.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-1*2+1), %o2 + +L.4.15: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-1*2-1), %o2 + +L.3.15: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits -3 + bl L.4.13 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-3*2+1), %o2 + +L.4.13: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-3*2-1), %o2 + +L.2.15: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 3, accumulated bits -3 + bl L.3.13 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits -5 + bl L.4.11 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-5*2+1), %o2 + +L.4.11: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-5*2-1), %o2 + +L.3.13: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits -7 + bl L.4.9 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-7*2+1), %o2 + +L.4.9: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-7*2-1), %o2 + + 9: +Lend_regular_divide: + subcc %o4, 1, %o4 + bge Ldivloop + tst %o3 + + bl,a Lgot_result + ! non-restoring fixup here (one instruction only!) + add %o3, %o1, %o3 + +Lgot_result: + + retl + mov %o3, %o0 + + .globl .urem_patch +.urem_patch: + wr %g0, 0x0, %y + nop + nop + nop + udiv %o0, %o1, %o2 + umul %o2, %o1, %o2 + retl + sub %o0, %o2, %o0 |