From 0413b54c02d31cae62e8d13b3ee7ac109ceeff0e Mon Sep 17 00:00:00 2001 From: Ulrich Drepper Date: Sun, 10 Aug 1997 18:37:15 +0000 Subject: Update. 1997-08-10 18:48 Ulrich Drepper * Makeconfig: Define build-programs to no if not set and $(config-os) is none. * Makerules (+depfiles): Don't add $(others) if not build-programs. * Rules (others): Don't depend on $(install-bin), $(install-rootbin) and $(install-sbin) if not build-programs. * Makefile (subdirs): Remove login. * sysdeps/mach/hurd/Subdirs: Add login. * sysdeps/unix/Subdirs: Add login. * sysdeps/generic/sys/mman.h: Test feature macros before defining non-POSIX things. Add MAP_FAILED. * sysdeps/unix/bsd/osf/sys/mman.h: Likewise. * sysdeps/unix/bsd/sun/sunos4/sys/mman.h: Likewise. * sysdeps/unix/bsd/ultrix4/sys/mman.h: Likewise. * sysdeps/unix/sysv/irix4/sys/mman.h: Likewise. * sysdeps/unix/sysv/linux/sys/mman.h: Rewrite to not use kernel header but bits/mman.h. Test feature macros before definitions non-POSIX functions. * sysdeps/unix/sysv/linux/alpha/bits/mman.h: New file. * sysdeps/unix/sysv/linux/i386/bits/mman.h: New file. * sysdeps/unix/sysv/linux/m68k/bits/mman.h: New file. * sysdeps/unix/sysv/linux/mips/bits/mman.h: New file. * sysdeps/unix/sysv/linux/powerpc/bits/mman.h: New file. * sysdeps/unix/sysv/linux/sparc/bits/mman.h: New file. * sysdeps/unix/sysv/linux/sparc64/bits/mman.h: New file. * sysdeps/unix/sysv/linux/alpha/Dist: Add bits/mman.h. * sysdeps/unix/sysv/linux/i386/Dist: Likewise. * sysdeps/unix/sysv/linux/m68k/Dist: Likewise. * sysdeps/unix/sysv/linux/mips/Dist: Likewise. * sysdeps/unix/sysv/linux/powerpc/Dist: Likewise. * sysdeps/unix/sysv/linux/sparc/Dist: Likewise. * sysdeps/unix/sysv/linux/sparc64/Dist: New file. Likewise. * catgets/open_catalog.c (__open_catalog): Compare result of mmap with MAP_FAILED and not -1. * elf/dl-load.c (_dl_map_object_from_fd): Likewise. * elf/dl-minimal.c (malloc): Likewise. * elf/dl-misc.c (_dl_sysdep_read_whole_file): Likewise. * locale/loadlocale.c (_nl_load_locale): Likewise. * locale/programs/localedef.c: Likewise. * malloc/malloc.c (mmap_chunk): Likewise. (new_heap): Likewise. * string/stratcliff.c: Likewise. * string/tester.c: Add tests for stpcpy. * gmon/gmon.c (__monstartup): Use calloc instead of malloc+memset. * gmon/sys/gmon.h: Add prototype for write_profiling. * elf/dl-profile.c: Add decsriptions and implement reading of existing data in profiling file. * gmon/sys/gmon_out.h (GMON_SHOBJ_VERSION): New macro for shared lib profiling data. * sysdeps/stub/bits/libc-lock.h (__libc_once_define): Add new first parameter for class. * inet/inet_ntoa.c: Make once control variable static. Correctly handle case where key cannot be retrieved. * inet/Makefile (tests): Add tst-ntoa. * inet/tst-ntoa.c: New file. * manual/time.texi: Document localtime_r, gmtime_r, asctime_r, and ctime_r. Mention that tzname values must not be changed. * manual/users.texi: Document UTMP/WTMP handling functions. Mainly written by Mark Kettenis. * math/libm-test.c (cbrt_test): Add test for number with aboslute value < 1.0. * nss/nss_files/files-XXX.c (internal_setent): If opening of file failed with EGAIN return NSS_STATUS_TRYAGAIN. * nss/nss_files/files-alias.c (internal_setent): Likewise. * sysdeps/generic/bits/utmp.h: Pretty print. * sysdeps/i386/i586/strcpy.S: New file. Much faster implementation. * sysdeps/i386/i586/stpcpy.S: New file. * sysdeps/i386/i686/Implies: Don't fall back on i586 versions since these are less efficient than i486/i386 versions. * sysdeps/libm-i387/e_rem_pio2.c: Add empty file to prevent the version from libm-ieee754 be compiled since the later is not needed. * sysdeps/libm-i387/e_rem_pio2f.c: Likewise. * sysdeps/libm-i387/e_rem_pio2l.c: Likewise. * sysdeps/libm-i387/k_rem_pio2.c: Likewise. * sysdeps/libm-i387/k_rem_pio2f.c: Likewise. * sysdeps/libm-i387/k_rem_pio2l.c: Likewise. * sysdeps/m68k/fpu/e_rem_pio2.c: Likewise. * sysdeps/m68k/fpu/e_rem_pio2f.c: Likewise. * sysdeps/m68k/fpu/e_rem_pio2l.c: Likewise. * sysdeps/m68k/fpu/k_rem_pio2.c: Likewise. * sysdeps/m68k/fpu/k_rem_pio2f.c: Likewise. * sysdeps/m68k/fpu/k_rem_pio2l.c: Likewise. * sysdeps/libm-i387/s_cbrt.S: Fix several bugs. * sysdeps/libm-i387/s_cbrtf.S: Likewise. * sysdeps/libm-i387/s_cbrtl.S: Likewise. * sysdeps/unix/sysv/linux/if_index.c (if_indextoname): Correct little bug. * sysdeps/unix/sysv/linux/bits/socket.h (struct msghdr): Make field msg_iovlen of type int. * time/tzfile.c: Change return type of __tzstring to char *. * time/tzset.c: Likewise. Change definition of __tzstring. * time/tzset.c: Interpret no DST information in user provided time zone specification as it is meant: no DST. Patch by Paul Eggert . 1997-07-20 03:01 Geoff Keating * sysdeps/elf/dl-load.c (add_name_to_object): New procedure. (_dl_map_object_from_fd): Use add_name_to_object. (_dl_map_object): If an object's SONAME is used to resolve a dependency, add it to the list of the object's names. * inet/getnameinfo.c: Make `domain' non-const. * sysdeps/unix/sysv/linux/powerpc/kernel_termios.c: Clean up. * math/test-fenv.c (feenv_nomask_test): Don't execute if FE_NOMASK_ENV is not defined, or if fesetenv(FE_NOMASK_ENV) sets errno to ENOSYS. * sysdeps/powerpc/dl-machine.h: Print proper error message on unknown reloc type (rather than using assert). * sysdeps/unix/sysv/linux/powerpc/profil-counter.h: New file. * sysdeps/unix/sysv/linux/powerpc/profil.c: Deleted. 1997-07-16 12:47 Geoff Keating * sysdeps/powerpc/bits/fenv.h (feraiseexcept): New optimising macro. (feclearexcept): New optimising macro. (FE_NOMASK_ENV): Change to procedure so it can enable exceptions. (FE_ENABLED_ENV): Add. (FE_NONIEEE_ENV): Add. * sysdeps/powerpc/bits/fenv_const.c: Add __fe_enabled_env, __fe_nonieee_env; delete __fe_nomask_env. * sysdeps/powerpc/bits/fe_nomask.c: New file (stub warning until it gets implemented in the kernel). * sysdeps/powerpc/fraiseenv.c: Deal with chips that don't have FE_INVALID_SOFTWARE implemented. Use macros for bit names for clarity. * sysdeps/powerpc/fsetexcptflag.c: Likewise. * io/ftw.c: Don't compare holes in structures. * sysdeps/unix/sysv/linux/sys/sysmacros.h: Cast the result of the macros to `int', because otherwise it might be `long long' which the calling code is probably not expecting. * sysdeps/libm-ieee754/s_lround.c [NO_LONG_DOUBLE]: Fix a few bugs, document the existence of some more. * sysdeps/powerpc/s_llrint.c: New file. * sysdeps/powerpc/s_lrint.c: New file. * sysdeps/powerpc/s_llround.c: New file. * sysdeps/powerpc/s_lround.c: New file. * sysdeps/powerpc/s_sqrt.c: New file. * sysdeps/powerpc/s_sqrtf.c: New file. * sysdeps/powerpc/w_sqrt.s: New empty file. * sysdeps/powerpc/w_sqrtf.s: New empty file. * sysdeps/powerpc/t_sqrt.c: New file. * sysdeps/powerpc/test-arithf.c: New file. * sysdeps/powerpc/Makefile [subdir=math]: Add t_sqrt to support routines. Add test-arithf to test programs. * sysdeps/powerpc/bits/mathdef.h: Add FP_ILOGB0, FP_ILOGBNAN. * sysdeps/powerpc/strcmp.s: Simplify drastically. Now much neater, and possibly faster (or possibly slower, depending on input). 1997-06-08 22:55 Geoff Keating * sysdeps/powerpc/fenvbits.h: Correct FE_DFL_ENV and FE_NOMASK_ENV macros. * sysdeps/powerpc/s_rint.c: New file. 1997-05-22 08:47 Geoff Keating * sysdeps/powerpc/Makefile [subdir=math]: Add q_* routines. * sysdeps/powerpc/Dist: Add quad_float.h. * sysdeps/powerpc/q_dtoq.c: New file. * sysdeps/powerpc/q_itoq.c: New file. * sysdeps/powerpc/q_lltoq.c: New file. * sysdeps/powerpc/q_neg.c: New file. * sysdeps/powerpc/q_qtoi.c: New file. * sysdeps/powerpc/q_qtoll.c: New file. * sysdeps/powerpc/q_qtos.c: New file. * sysdeps/powerpc/q_qtou.c: New file. * sysdeps/powerpc/q_qtoull.c: New file. * sysdeps/powerpc/q_stoq.c: New file. * sysdeps/powerpc/q_ulltoq.c: New file. * sysdeps/powerpc/q_utoq.c: New file. * sysdeps/powerpc/quad_float.h: New file. * sysdeps/powerpc/test-arith.c: New file. * sysdeps/powerpc/fpu_control.h: Fix _FPU_GETCW. * sysdeps/powerpc/fegetround.c: Use mcrfs to be faster and not require a stack frame. * sysdeps/powerpc/bits/fenv.h: Include inline macro for fegetround. 1997-05-18 05:55 Geoff Keating * sysdeps/powerpc/fenv_libc.h (fegetenv_register, fesetenv_register): Add 'volatile'. (set_fpscr_bit, reset_fpscr_bit): New macros, FPSCR_* constants to use with them. * sysdeps/powerpc/s_copysign.S: New file. * sysdeps/powerpc/s_copysignf.s: New file. * sysdeps/powerpc/s_fabs.S: New file. * sysdeps/powerpc/s_fabsf.s: New file. * sysdeps/powerpc/s_isnan.c: New file. * sysdeps/powerpc/s_isnanf.s: New file. * sysdeps/powerpc/s_rintf.c: New file. * sysdeps/powerpc/fenvbits.h: Make FE_INVALID the summary bit in the FPSCR, not the enable bit. * sysdeps/powerpc/fraiseexcpt.c: Consequent change to the above. * sysdeps/powerpc/fclrexcpt.c: Correct. * sysdeps/powerpc/fsetexcptflag.c: Correct. * sysdeps/powerpc/ftestexcpt.c: Is now much simpler. * sysdeps/powerpc/fgetexcptflg.c: Simplify. * sysdeps/powerpc/strlen.s: Schedule better, save 3 clocks :-). * sysdeps/powerpc/dl-machine.h (elf_machine_rela): Add comment explaining some reentrancy issues with lazy PLT entries. 1997-08-09 13:04 Mark Kettenis * login/logout.c (logout): utmpname returns -1 on error. * login/libutil.map: Remove updwtmp. * login/getutline.c: Rename getutline to __getutline and make getutline a weak alias. Make getutxline a weak alias for __getutline. * login/getutid.c: Rename getutid to __getutid and make getutid a weak alias. Make getutxid a weak alias for __getutid. * libc.map: Add getutxid, getutxline. * login/utmpname.c (__utmpname): Reset backend right after backend endutent call. * login/utmp_file.c: Reordered functions. Remove unecessary header files. (getutent_r_file, pututline_file): Do not call setutent_file. At this point the file is guaranteed to be open (assert!). (getutid_r_file, getutline_r_file): Remove check for validity of file descriptor. At this point the file is guaranteed to be open. (getutent_r_file, internal_getut_r, getutline_r_file, pututline_file, updwtmp_file): Do not wait when unlocking file. * login/utmp_daemon.c: General cleanup and a bit of reordering. (getutent_r_daemon, pututline_daemon): Do not call setutent_daemon. At this point the socket is guaranteed to be open (assert!). (getutid_r_daemon, getutline_r_daemon): Do not check if daemon_sock is valid. At this point the socket is guaranteed to be open (assert!). * login/getutline_r.c: Remove unnecessary header files. (__getutline_r): Do not call backend setutent. * login/getutid_r.c: Remove unnecessary header files. (__getutid_r): Do not call backend setutent. * login/getutent_r.c: Remove unneccesary header files. (__libc_utmp_unknown_functions): Added getutid_r_unknown, getutline_r_unknown. (setutent_unknown): Only set file backend if setutent for the file backend was successful. (getutent_r_unknown, pututline_unknown): Call setutent_unknown instead of __setutent. Report failure if no backend was selected. (getutid_r_unknown): New function. (getutline_r_unknown): New function. (__endutent): Reset backend. This makes sure all backends are checked on the next setutent call. 1997-08-08 20:20 Thorsten Kukuk * nis_cache.c: Replace dummy functions. * libc.map: Add xdr_sizeof symbol. * sunrpc/Makefile: Add xdr_sizeof to routines. * sunrpc/rpc/xdr.h: Add xdr_sizeof prototype. * sunrpc/xdr_sizeof.c: New, from tirpc 2.3. 1997-08-08 Thomas Bushnell, n/BSG * sysdeps/mach/bits/libc-lock.h (__libc_once): Define correctly. 1997-08-07 Thomas Bushnell, n/BSG * sysdeps/mach/hurd/profil.c (fetch_samples): Put parens in the right place. 1997-08-06 Thomas Bushnell, n/BSG * sysdeps/mach/hurd/profil.c (fetch_samples): Do arithmetic on PC's in long long to avoid overflow. 1997-08-07 Thomas Bushnell, n/BSG * sysdeps/mach/bits/libc-lock.h (__libc_once, __libc_once_define): New macros. 1997-08-06 Andreas Jaeger * nis/Makefile (headers): Remove bits/nislib.h. 1997-08-06 14:54 Ulrich Drepper * sysdeps/mach/hurd/Subdirs: Add login. 1997-08-06 14:23 Klaus Espenlaub * db/hash/hash.c (init_hash): Don't use stat() if it doesn't provide the preferred block size. * login/programs/database.c (store_state_entry): Don't compile if there is no ut_type field. (store_state_entry, store_process_entry): Use the ut_tv field for timestamps if supported. * login/programs/utmpdump.c (print_entry): Always use ut_tv field. * login/programs/xtmp.c: Fix numerous xtmp/utmp typos. Use the ut_tv field for timestamps if supported. * login/programs/xtmp.h: Fix xtmp/utmp typo. * sysdeps/posix/defs.c (stdstream): Change (PTR) to (void *). * sysdeps/stub/connect.c (connect): Change to __connect, add alias. * sysdeps/stub/send.c (send): Likewise. * sysdeps/stub/s_exp2f.c: Emit correct stub_warning(). * sysdeps/stub/statfs.c: Move stub_warning() to the usual place. * sysdeps/stub/init-first.c: Add definition of __libc_pid. 1997-08-05 13:28 Philip Blundell * sysdeps/standalone/arm/bits/errno.h: Add EISDIR, EOPNOTSUPP; tidy up formatting. * Makefile (subdirs): Remove `login'. * sysdeps/unix/Subdirs: New file; build `login' subdirectory for Unix systems. 1997-08-05 Andreas Schwab * sysdeps/generic/bits/utmpx.h: New file. * sysdeps/mach/hurd/Dist: Add some files. * sysdeps/mips/Dist: Likewise. * sysdeps/mips/mips64/Dist: Likewise. * sysdeps/sparc/Dist: Likewise. * sysdeps/unix/sysv/linux/mips/Dist: Likewise. * sysdeps/unix/sysv/linux/sparc/Dist: Likewise. * sysdeps/mips/mipsel/Dist: New file. * sysdeps/sparc64/elf/Dist: New file. * sysdeps/unix/sysv/linux/sparc64/Dist: New file. 1997-08-05 Andreas Schwab * libc.map: Add missing symbols. 1997-08-05 Andreas Jaeger * manual/socket.texi: Correct typos. * manual/math.texi: Correct typos. * manual/time.texi (Formatting Date and Time): Likewise. 1997-08-04 13:06 Thomas Bushnell, n/BSG * gmon/gmon.c (write_gmon): New function; guts from _mcleanup. (_mcleanup): Use write_gmon. (write_profiling): This function can be safely called at any time to write a current histogram without interfering with ongoing profiling. * sysdeps/mach/hurd/profil.c (fetch_samples): Initialize NSAMPLES. 1997-08-01 17:53 Thomas Bushnell, n/BSG * sysdeps/mach/hurd/profil.c (fetch_samples): Sample buffer need not be vm_deallocated; it's a stack buffer. (profil_vm_deallocate): Delete prototype. (#include <../mach/RPC_vm_deallocate_rpc.c>): Drop this inclusion. * sysdeps/mach/hurd/Makefile ($(common-objpfx)hurd/../mach/RPC_vm_deallocate_rpc.c): Delete this rule. * sysdeps/mach/hurd/profil.c (fetch_samples): New function, guts from profile_waiter. (profile_waiter): Use fetch_samples. (profil): When turning off profiling, fetch the last bunch of samples before disabling PC sampling. (fetch_samples): Add prototype. 1997-07-30 12:53 Thomas Bushnell, n/BSG * sysdeps/mach/hurd/Makefile: Give normal rules for the mach RPC source included by profil.c instead of trying to use before-compile grot. 1997-07-23 15:04 Thomas Bushnell, n/BSG * sysdeps/mach/hurd/profil.c (profile_waiter): Do vm_deallocate after releasing lock. Use special RPC stubs and avoid assert, thus making this function entirely independent of the threadvar mechanism and of cthreads. (lock): Convert to be a spin lock instead of a mutex, so that the waiter thread need not use cthreads. Use a fork prepare hook instead of _hurd_fork_locks to make sure we are clean before forking. (fork_profil_child): Renamed from fork_profil. (profil_reply_port): New variable. (profil_vm_deallocate, profil_task_get_sampled_pcs): Special RPC stubs made by special hacks. * sysdeps/mach/hurd/Makefile (before-compile): Add the mach RPC source files that profil.c needs if we are in the gmon directory. * mach/setup-thread.c (__mach_setup_thread): Delete avoidance of a cthreads bug that hasn't existed for two years. * stdio-common/printf_fp.c (__printf_fp): Correct rounding of numbers * locale/programs/locale.c (print_escaped): New function. * time/australasia: Likewise. (open_database, synchronize_database, initialize_database): * nss/nss_files/files-netgrp.c (_nss_netgroup_parseline): Don't * Makerules (load-map-file): Currectly handle missing map file. cannot rely on M_SQRT2 being defined. (log1p): Use __M_SQRT2 not M_SQRT2. --- sysdeps/libm-i387/s_cbrtl.S | 143 ++++++++++++++++++++++++++++---------------- 1 file changed, 92 insertions(+), 51 deletions(-) (limited to 'sysdeps/libm-i387/s_cbrtl.S') diff --git a/sysdeps/libm-i387/s_cbrtl.S b/sysdeps/libm-i387/s_cbrtl.S index b2023d1991..6a3b9a8dc5 100644 --- a/sysdeps/libm-i387/s_cbrtl.S +++ b/sysdeps/libm-i387/s_cbrtl.S @@ -28,52 +28,69 @@ #endif .align ALIGNARG(4) - ASM_TYPE_DIRECTIVE(f1,@object) -f1: .double 0.338058687610520237 - ASM_SIZE_DIRECTIVE(f1) - ASM_TYPE_DIRECTIVE(f2,@object) -f2: .double 1.67595307700780102 - ASM_SIZE_DIRECTIVE(f2) - ASM_TYPE_DIRECTIVE(f3,@object) -f3: .double -2.82414939754975962 - ASM_SIZE_DIRECTIVE(f3) - ASM_TYPE_DIRECTIVE(f4,@object) -f4: .double 4.09559907378707839 - ASM_SIZE_DIRECTIVE(f4) - ASM_TYPE_DIRECTIVE(f5,@object) -f5: .double -4.11151425200350531 - ASM_SIZE_DIRECTIVE(f5) - ASM_TYPE_DIRECTIVE(f6,@object) -f6: .double 2.65298938441952296 - ASM_SIZE_DIRECTIVE(f6) - ASM_TYPE_DIRECTIVE(f7,@object) -f7: .double -0.988553671195413709 - ASM_SIZE_DIRECTIVE(f7) ASM_TYPE_DIRECTIVE(f8,@object) -f8: .double 0.161617097923756032 +f8: .tfloat 0.161617097923756032 ASM_SIZE_DIRECTIVE(f8) + .align ALIGNARG(4) + ASM_TYPE_DIRECTIVE(f7,@object) +f7: .tfloat -0.988553671195413709 + ASM_SIZE_DIRECTIVE(f7) + .align ALIGNARG(4) + ASM_TYPE_DIRECTIVE(f6,@object) +f6: .tfloat 2.65298938441952296 + ASM_SIZE_DIRECTIVE(f6) + .align ALIGNARG(4) + ASM_TYPE_DIRECTIVE(f5,@object) +f5: .tfloat -4.11151425200350531 + ASM_SIZE_DIRECTIVE(f5) + .align ALIGNARG(4) + ASM_TYPE_DIRECTIVE(f4,@object) +f4: .tfloat 4.09559907378707839 + ASM_SIZE_DIRECTIVE(f4) + .align ALIGNARG(4) + ASM_TYPE_DIRECTIVE(f3,@object) +f3: .tfloat -2.82414939754975962 + ASM_SIZE_DIRECTIVE(f3) + .align ALIGNARG(4) + ASM_TYPE_DIRECTIVE(f2,@object) +f2: .tfloat 1.67595307700780102 + ASM_SIZE_DIRECTIVE(f2) + .align ALIGNARG(4) + ASM_TYPE_DIRECTIVE(f1,@object) +f1: .tfloat 0.338058687610520237 + ASM_SIZE_DIRECTIVE(f1) -#define CBRT2 1.2599210498948731648 -#define SQR_CBRT2 1.5874010519681994748 +#define CBRT2 1.2599210498948731648 +#define ONE_CBRT2 0.793700525984099737355196796584 +#define SQR_CBRT2 1.5874010519681994748 +#define ONE_SQR_CBRT2 0.629960524947436582364439673883 + /* We make the entries in the following table all 16 bytes + wide to avoid having to implement a multiplication by 10. */ ASM_TYPE_DIRECTIVE(factor,@object) -factor: .double 1.0 / SQR_CBRT2 - .double 1.0 / CBRT2 - .double 1.0 - .double CBRT2 - .double SQR_CBRT2 + .align ALIGNARG(4) +factor: .tfloat ONE_SQR_CBRT2 + .byte 0, 0, 0, 0, 0, 0 + .tfloat ONE_CBRT2 + .byte 0, 0, 0, 0, 0, 0 + .tfloat 1.0 + .byte 0, 0, 0, 0, 0, 0 + .tfloat CBRT2 + .byte 0, 0, 0, 0, 0, 0 + .tfloat SQR_CBRT2 ASM_SIZE_DIRECTIVE(factor) ASM_TYPE_DIRECTIVE(two64,@object) + .align ALIGNARG(4) two64: .byte 0, 0, 0, 0, 0, 0, 0xf0, 0x43 ASM_SIZE_DIRECTIVE(two64) #ifdef PIC #define MO(op) op##@GOTOFF(%ebx) -#define MOX(op,x,f) op##@GOTOFF(%ebx,x,f) +#define MOX(op,x) op##@GOTOFF(%ebx,x,1) #else #define MO(op) op -#define MOX(op,x,f) op(,x,f) +#define MOX(op,x) op(x) #endif .text @@ -97,7 +114,7 @@ ENTRY(__cbrtl) #endif cmpl $0, %eax - je 2f + jne 2f #ifdef PIC fldt 8(%esp) @@ -106,8 +123,13 @@ ENTRY(__cbrtl) #endif fmull MO(two64) movl $-64, %ecx +#ifdef PIC + fstpt 8(%esp) + movl 16(%esp), %eax +#else fstpt 4(%esp) movl 12(%esp), %eax +#endif movl %eax, %edx andl $0x7fff, %eax @@ -126,31 +148,45 @@ ENTRY(__cbrtl) #endif fabs - /* The following code has two track: + /* The following code has two tracks: a) compute the normalized cbrt value b) compute xe/3 and xe%3 The right track computes the value for b) and this is done - in an optimized way by avoiding division. */ + in an optimized way by avoiding division. - fld %st(0) /* xm : xm */ + But why two tracks at all? Very easy: efficiency. Some FP + instruction can overlap with a certain amount of integer (and + FP) instructions. So we get (except for the imull) all + instructions for free. */ - fmull MO(f7) /* f7*xm : xm */ + fldt MO(f8) /* f8 : xm */ + fmul %st(1) /* f8*xm : xm */ + + fldt MO(f7) + faddp /* f7+f8*xm : xm */ + fmul %st(1) /* (f7+f8*xm)*xm : xm */ movl $1431655766, %eax - faddl MO(f6) /* f6+f7*xm : xm */ + fldt MO(f6) + faddp /* f6+(f7+f8*xm)*xm : xm */ imull %ecx - fmul %st(1) /* (f6+f7*xm)*xm : xm */ + fmul %st(1) /* (f6+(f7+f8*xm)*xm)*xm : xm */ movl %ecx, %eax - faddl MO(f5) /* f5+(f6+f7*xm)*xm : xm */ + fldt MO(f5) + faddp /* f5+(f6+(f7+f8*xm)*xm)*xm : xm */ sarl $31, %eax - fmul %st(1) /* (f5+(f6+f7*xm)*xm)*xm : xm */ + fmul %st(1) /* (f5+(f6+(f7+f8*xm)*xm)*xm)*xm : xm */ subl %eax, %edx - faddl MO(f4) /* f4+(f5+(f6+f7*xm)*xm)*xm : xm */ - fmul %st(1) /* (f4+(f5+(f6+f7*xm)*xm)*xm)*xm : xm */ - faddl MO(f3) /* f3+(f4+(f5+(f6+f7*xm)*xm)*xm)*xm : xm */ - fmul %st(1) /* (f3+(f4+(f5+(f6+f7*xm)*xm)*xm)*xm)*xm : xm */ - faddl MO(f2) /* f2+(f3+(f4+(f5+(f6+f7*xm)*xm)*xm)*xm)*xm : xm */ - fmul %st(1) /* (f2+(f3+(f4+(f5+(f6+f7*xm)*xm)*xm)*xm)*xm)*xm : xm */ - faddl MO(f1) /* u:=f1+(f2+(f3+(f4+(f5+(f6+f7*xm)*xm)*xm)*xm)*xm)*xm : xm */ + fldt MO(f4) + faddp /* f4+(f5+(f6+(f7+f8*xm)*xm)*xm)*xm : xm */ + fmul %st(1) /* (f4+(f5+(f6+(f7+f8*xm)*xm)*xm)*xm)*xm : xm */ + fldt MO(f3) + faddp /* f3+(f4+(f5+(f6+(f7+f8*xm)*xm)*xm)*xm)*xm : xm */ + fmul %st(1) /* (f3+(f4+(f5+(f6+(f7+f8*xm)*xm)*xm)*xm)*xm)*xm : xm */ + fldt MO(f2) + faddp /* f2+(f3+(f4+(f5+(f6+(f7+f8*xm)*xm)*xm)*xm)*xm)*xm : xm */ + fmul %st(1) /* (f2+(f3+(f4+(f5+(f6+(f7+f8*xm)*xm)*xm)*xm)*xm)*xm)*xm : xm */ + fldt MO(f1) + faddp /* u:=f1+(f2+(f3+(f4+(f5+(f6+(f7+f8*xm)*xm)*xm)*xm)*xm)*xm)*xm : xm */ fld %st /* u : u : xm */ fmul %st(1) /* u*u : u : xm */ @@ -164,19 +200,24 @@ ENTRY(__cbrtl) fadd %st(0) /* 2*t2 : t2+2*xm : u : xm */ subl %edx, %ecx faddp %st, %st(3) /* t2+2*xm : u : 2*t2+xm */ + shll $4, %ecx fmulp /* u*(t2+2*xm) : 2*t2+xm */ fdivp %st, %st(1) /* u*(t2+2*xm)/(2*t2+xm) */ - fmull MOX(16+factor,%ecx,8) /* u*(t2+2*xm)/(2*t2+xm)*FACT */ + fldt MOX(32+factor,%ecx) + fmulp /* u*(t2+2*xm)/(2*t2+xm)*FACT */ pushl %eax fildl (%esp) /* xe/3 : u*(t2+2*xm)/(2*t2+xm)*FACT */ fxch /* u*(t2+2*xm)/(2*t2+xm)*FACT : xe/3 */ - popl %eax fscale /* u*(t2+2*xm)/(2*t2+xm)*FACT*2^xe/3 */ - fstp %st(1) + popl %edx #ifdef PIC + movl 16(%esp), %eax popl %ebx +#else + movl 12(%esp), %eax #endif - testl $0x8000, 12(%esp) + testl $0x8000, %eax + fstp %st(1) jz 4f fchs 4: ret -- cgit v1.2.3