mirror of
				https://sourceware.org/git/glibc.git
				synced 2025-11-03 20:53:13 +03:00 
			
		
		
		
	1997-08-24 12:24 Ulrich Drepper <drepper@cygnus.com> * configure.in (INSTALL): Quote `$'. * libc.map: Add __xpg_basename. * csu/Makefile (initfini.s): Disable optimization. * elf/dl-deps.c: Implement handling of DL_FILTER. * elf/dl-load.c (_dl_init_paths): Add error check. * intl/finddomain.c (_nl_find_domain): Correct comment. * intl/localealias.c: Include <bits/libc-lock.h> not <libc-lock.h>. * libio/stdio.h: Make {,v}snprintf available if __USE_BSD. Change extern inline functions to work correctly in C++. * locale/iso-4217.def: Update for more recent ISO 4217 version. * locale/loadlocale.c (_nl_load_locale): Add cast. * manual/message.texi: Finish gettext section. * posix/getopt_init.c: Don't use relative #include path. (__getopt_clean_environment): Change function to take pointer to environment as argument. Optimize generation of test string a bit. * sysdeps/unix/sysv/linux/init-first.c: Call __getopt_clean_environment with additional argument. * poisx/glob.c: Add prototype for next_brace_sub. * sysdeps/generic/dl-sysdep.c: Recognize AT_BASE value on auxiliary vector. * sysdeps/i386/dl-machine.h (elf_machine_load_address): Rewrite to not generate relocation entry. Suggested by Richard Henderson. (ELF_MACHINE_BEFORE_RTLD_RELOC): Removed. (elf_machine_runtime_setup): Add .aligns. * sysdeps/i386/fpu/fraiseexcpt.c: Add volatile to asms. * sysdeps/i386/fpu/bits/mathinline.h: Partially undo change of 1997-08-14 03:14. gcc 2.7.2* is really broken in some aspects. * sysdeps/standalone/i386/i386.h: Clean up asm statements a bit. * sysdeps/standalone/i960/i960ca.h: Likewise. 1997-08-22 19:04 Richard Henderson <rth@cygnus.com> * elf/rtld.c (_dl_start): Init _dl_rtld_map.l_opencount due to undocumented test addition in _dl_map_object. Support ET_EXEC versions of ld.so, for debugging at least: * elf/dl-load.c (_dl_map_object): Add_name_to_object could get called despite the DT_SONAME != NULL test, segfaulting. Simplify the code here as well. * elf/dl-lookup.c (do_lookup): Skip objects with no symtab. (_dl_setup_hash): Likewise for hash tables. * elf/dl-version.c (_dl_check_map_versions): Likewise for strtabs. * elf/rtld.c (_dl_start): Likewise for rpath. (_dl_rtld_libname2): New variable. (dl_main): Use it to add an soname for ourselves when we don't have one of our own. Base it on the target's .interp. (dl_main): Again, skip printing of objects that don't have strtabs. Sparc 32 merge: * elf/dl-runtime.c (ELF_FIXUP_RETURN_VALUE): Provide default value. (fixup): Simplify code. Use ELF_FIXUP_RETURN_VALUE. (profile_fixup): Likewise, though this still needs fixing for Sparc32 and PPC. * sysdeps/powerpc/dl-machine.h: Transmute ELF_FIXUP_RETURNS_ADDRESS to ELF_FIXUP_RETURN_VALUE. * sysdeps/sparc/sparc32/dl-machine.h: Implement lazy relocation. Fix up _dl_start_user to handle _dl_skip_args properly. Use _dl_hwcap to determine if "flush" is available/needed. * sysdeps/sparc/configure.in: Remove. It doesn't actually do anything anymore, and what it did do is done somewhere else. * sysdeps/sparc/configure: Likewise. * sysdeps/sparc/fpu/bits/mathdef.h (FP_ILOGB0, FP_ILOGBNAN): New. * sysdeps/sparc/fpu/fraiseexcpt.c: Rearrange for smaller code. * sysdeps/sparc/sparc32/Makefile: Fix sparc->sparc/sparc32 bits in divrem expansions. * sysdeps/unix/sysv/linux/sparc/sparc32/sysdep.h (END, LOC): New definitions for assembly syntax differences. * sysdeps/sparc/sparc32/__longjmp.S: %g6,%g7 are reserved to the "system". Use %g2,%g3 instead. Use new local label macro. * sysdeps/sparc/sparc32/add_n.S: Use <sysdep.h> and ENTRY, END, and LOC for proper assembly headers/footers. * sysdeps/sparc/sparc32/addmul_1.S: Likewise. * sysdeps/sparc/sparc32/alloca.S: Likewise. * sysdeps/sparc/sparc32/dotmul.S: Likewise. * sysdeps/sparc/sparc32/lshift.S: Likewise. * sysdeps/sparc/sparc32/mul_1.S: Likewise. * sysdeps/sparc/sparc32/rshift.S: Likewise. * sysdeps/sparc/sparc32/sparcv8/addmul_1.S: Likewise. * sysdeps/sparc/sparc32/sparcv8/mul_1.S: Likewise. * sysdeps/sparc/sparc32/sparcv8/submul_1.S: Likewise. * sysdeps/sparc/sparc32/sparcv8/udiv_qrnnd.S: Likewise. * sysdeps/sparc/sparc32/sub_n.S: Likewise. * sysdeps/sparc/sparc32/submul_1.S: Likewise. * sysdeps/sparc/sparc32/udiv_qrnnd.S: Likewise. * sysdeps/sparc/sparc32/umul.S: Likewise. * sysdeps/sparc/sparc32/divrem.m4: Likewise. * sysdeps/sparc/sparc32/rem.S: Regenerate. * sysdeps/sparc/sparc32/sdiv.S: Regenerate. * sysdeps/sparc/sparc32/udiv.S: Regenerate. * sysdeps/sparc/sparc32/urem.S: Regenerate. * sysdeps/sparc/sparc32/sparcv8/dotmul.S: New file. * sysdeps/sparc/sparc32/sparcv8/rem.S: New file. * sysdeps/sparc/sparc32/sparcv8/sdiv.S: New file. * sysdeps/sparc/sparc32/sparcv8/udiv.S: New file. * sysdeps/sparc/sparc32/sparcv8/umul.S: New file. * sysdeps/sparc/sparc32/sparcv8/urem.S: New file. * sysdeps/sparc/sparc32/bsd-_setjmp.S: Dike out. * sysdeps/sparc/sparc32/bsd-setjmp.S: Likewise. * sysdeps/sparc/sparc32/setjmp.S: Add _setjmp and setjmp entry points. * sysdeps/unix/sysv/linux/sparc/sparc32/__sigtrampoline.S: Clean up PIC code. * sysdeps/sparc/sparc32/elf/start.S: New file, slightly modified from the sparc64 version. * sysdeps/sparc/sparc32/elf/start.c: Removed. * sysdeps/unix/sysv/linux/sparc/sparc32/init-first.h: Rewrite in assembly based on the sparc64 version. * sysdeps/sparc/sparc32/fpu/bits/fenv.h: Duh. Use proper syntax for manipulating %fsr. * sysdeps/sparc/sparc32/fpu/fpu_control.h: Make IEEE conformance be the default. * elf/elf.h (HWCAP_SPARC_*): New definitions. * elf/rtld.c (_dl_hwcap): New variable. * sysdeps/generic/dl-sysdep.c (_dl_sysdep_start): Record AT_HWCAP. * sysdeps/unix/sysv/linux/sparc/sparc32/getpagesize.c: New file. Attempt to get hold of the page size based on what we might have been told at startup time in _dl_pagesize. This will be obsolete when I finish the kernel hooks for a proper sysconf(), stay tuned. Sparc 64 merge: * sysdeps/sparc/sparc64/dl-machine.h (ELF_FIXUP_RETURN_VALUE): New. Figure out the right thing to return based on the .plt format. * sysdeps/sparc/sparc64/fpu/fpu_control.h: Update comment. * sysdeps/unix/sysv/linux/sparc/sparc64/bits/types.h (__dev_t): Should have been 64-bits wide. * sysdeps/unix/sysv/linux/sparc/sparc64/init-first.h: sll->sllx, optimize for branch delay slot usage. 1997-08-22 Andreas Schwab <schwab@issan.informatik.uni-dortmund.de> * csu/Makefile ($(objpfx)crt%.o): Fix a missing *.so -> *.os change. 1997-08-20 Andreas Jaeger <aj@arthur.rhein-neckar.de> * math/libm-test.c (identities): Change epsilon. * sysdeps/i386/fpu/bits/mathinline.h: Correct arguments to fabs, fabsf, fabsl, __fabsl. * sysdeps/libm-i387/e_remainderl.S: Pop extra value from FPU stack. * sysdeps/libm-ieee754/s_csinhl.c: Include <fenv.h>.
		
			
				
	
	
		
			233 lines
		
	
	
		
			5.9 KiB
		
	
	
	
		
			Plaintext
		
	
	
	
	
	
			
		
		
	
	
			233 lines
		
	
	
		
			5.9 KiB
		
	
	
	
		
			Plaintext
		
	
	
	
	
	
/*
 | 
						|
 * 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:
 | 
						|
 *  NAME	name of function to generate
 | 
						|
 *  OP		OP=div => %o0 / %o1; OP=rem => %o0 % %o1
 | 
						|
 *  S		S=true => signed; S=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.
 | 
						|
 */
 | 
						|
 | 
						|
define(N, `4')dnl
 | 
						|
define(WORDSIZE, `32')dnl
 | 
						|
define(TOPBITS, eval(WORDSIZE - N*((WORDSIZE-1)/N)))dnl
 | 
						|
dnl
 | 
						|
define(dividend, `%o0')dnl
 | 
						|
define(divisor, `%o1')dnl
 | 
						|
define(Q, `%o2')dnl
 | 
						|
define(R, `%o3')dnl
 | 
						|
define(ITER, `%o4')dnl
 | 
						|
define(V, `%o5')dnl
 | 
						|
dnl
 | 
						|
dnl m4 reminder: ifelse(a,b,c,d) => if a is b, then c, else d
 | 
						|
define(T, `%g1')dnl
 | 
						|
define(SC, `%g2')dnl
 | 
						|
ifelse(S, `true', `define(SIGN, `%g3')')dnl
 | 
						|
 | 
						|
dnl
 | 
						|
dnl This is the recursive definition for developing quotient digits.
 | 
						|
dnl
 | 
						|
dnl Parameters:
 | 
						|
dnl  $1	the current depth, 1 <= $1 <= N
 | 
						|
dnl  $2	the current accumulation of quotient bits
 | 
						|
dnl  N	max depth
 | 
						|
dnl
 | 
						|
dnl We add a new bit to $2 and either recurse or insert the bits in
 | 
						|
dnl the quotient.  R, Q, and V are inputs and outputs as defined above;
 | 
						|
dnl the condition codes are expected to reflect the input R, and are
 | 
						|
dnl modified to reflect the output R.
 | 
						|
dnl
 | 
						|
define(DEVELOP_QUOTIENT_BITS,
 | 
						|
`	! depth $1, accumulated bits $2
 | 
						|
	bl	LOC($1.eval(2**N+$2))
 | 
						|
	srl	V,1,V
 | 
						|
	! remainder is positive
 | 
						|
	subcc	R,V,R
 | 
						|
	ifelse($1, N,
 | 
						|
	`	b	9f
 | 
						|
		add	Q, ($2*2+1), Q
 | 
						|
	', `	DEVELOP_QUOTIENT_BITS(incr($1), `eval(2*$2+1)')')
 | 
						|
LOC($1.eval(2**N+$2)):
 | 
						|
	! remainder is negative
 | 
						|
	addcc	R,V,R
 | 
						|
	ifelse($1, N,
 | 
						|
	`	b	9f
 | 
						|
		add	Q, ($2*2-1), Q
 | 
						|
	', `	DEVELOP_QUOTIENT_BITS(incr($1), `eval(2*$2-1)')')
 | 
						|
	ifelse($1, 1, `9:')')dnl
 | 
						|
 | 
						|
#include <sysdep.h>
 | 
						|
#include <sys/trap.h>
 | 
						|
 | 
						|
ENTRY(NAME)
 | 
						|
ifelse(S, `true',
 | 
						|
`	! compute sign of result; if neither is negative, no problem
 | 
						|
	orcc	divisor, dividend, %g0	! either negative?
 | 
						|
	bge	2f			! no, go do the divide
 | 
						|
ifelse(OP, `div',
 | 
						|
`	xor	divisor, dividend, SIGN	! compute sign in any case',
 | 
						|
`	mov	dividend, SIGN		! sign of remainder matches dividend')
 | 
						|
	tst	divisor
 | 
						|
	bge	1f
 | 
						|
	tst	dividend
 | 
						|
	! divisor is definitely negative; dividend might also be negative
 | 
						|
	bge	2f			! if dividend not negative...
 | 
						|
	sub	%g0, divisor, divisor	! in any case, make divisor nonneg
 | 
						|
1:	! dividend is negative, divisor is nonnegative
 | 
						|
	sub	%g0, dividend, dividend	! make dividend nonnegative
 | 
						|
2:
 | 
						|
')
 | 
						|
	! Ready to divide.  Compute size of quotient; scale comparand.
 | 
						|
	orcc	divisor, %g0, V
 | 
						|
	bne	1f
 | 
						|
	mov	dividend, R
 | 
						|
 | 
						|
		! 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	R, V			! if divisor exceeds dividend, done
 | 
						|
	blu	LOC(got_result)		! (and algorithm fails otherwise)
 | 
						|
	clr	Q
 | 
						|
	sethi	%hi(1 << (WORDSIZE - TOPBITS - 1)), T
 | 
						|
	cmp	R, T
 | 
						|
	blu	LOC(not_really_big)
 | 
						|
	clr	ITER
 | 
						|
 | 
						|
	! `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	V, T
 | 
						|
		bgeu	3f
 | 
						|
		mov	1, SC
 | 
						|
		sll	V, N, V
 | 
						|
		b	1b
 | 
						|
		add	ITER, 1, ITER
 | 
						|
 | 
						|
	! Now compute SC.
 | 
						|
	2:	addcc	V, V, V
 | 
						|
		bcc	LOC(not_too_big)
 | 
						|
		add	SC, 1, SC
 | 
						|
 | 
						|
		! We get here if the divisor overflowed while shifting.
 | 
						|
		! This means that R has the high-order bit set.
 | 
						|
		! Restore V and subtract from R.
 | 
						|
		sll	T, TOPBITS, T	! high order bit
 | 
						|
		srl	V, 1, V		! rest of V
 | 
						|
		add	V, T, V
 | 
						|
		b	LOC(do_single_div)
 | 
						|
		sub	SC, 1, SC
 | 
						|
 | 
						|
	LOC(not_too_big):
 | 
						|
	3:	cmp	V, R
 | 
						|
		blu	2b
 | 
						|
		nop
 | 
						|
		be	LOC(do_single_div)
 | 
						|
		nop
 | 
						|
	/* NB: these are commented out in the V8-Sparc manual as well */
 | 
						|
	/* (I do not understand this) */
 | 
						|
	! V > R: went too far: back up 1 step
 | 
						|
	!	srl	V, 1, V
 | 
						|
	!	dec	SC
 | 
						|
	! do single-bit divide steps
 | 
						|
	!
 | 
						|
	! We have to be careful here.  We know that R >= V, so we can do the
 | 
						|
	! first divide step without thinking.  BUT, the others are conditional,
 | 
						|
	! and are only done if R >= 0.  Because both R and V 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...
 | 
						|
	LOC(do_single_div):
 | 
						|
		subcc	SC, 1, SC
 | 
						|
		bl	LOC(end_regular_divide)
 | 
						|
		nop
 | 
						|
		sub	R, V, R
 | 
						|
		mov	1, Q
 | 
						|
		b	LOC(end_single_divloop)
 | 
						|
		nop
 | 
						|
	LOC(single_divloop):
 | 
						|
		sll	Q, 1, Q
 | 
						|
		bl	1f
 | 
						|
		srl	V, 1, V
 | 
						|
		! R >= 0
 | 
						|
		sub	R, V, R
 | 
						|
		b	2f
 | 
						|
		add	Q, 1, Q
 | 
						|
	1:	! R < 0
 | 
						|
		add	R, V, R
 | 
						|
		sub	Q, 1, Q
 | 
						|
	2:
 | 
						|
	LOC(end_single_divloop):
 | 
						|
		subcc	SC, 1, SC
 | 
						|
		bge	LOC(single_divloop)
 | 
						|
		tst	R
 | 
						|
		b,a	LOC(end_regular_divide)
 | 
						|
 | 
						|
LOC(not_really_big):
 | 
						|
1:
 | 
						|
	sll	V, N, V
 | 
						|
	cmp	V, R
 | 
						|
	bleu	1b
 | 
						|
	addcc	ITER, 1, ITER
 | 
						|
	be	LOC(got_result)
 | 
						|
	sub	ITER, 1, ITER
 | 
						|
 | 
						|
	tst	R	! set up for initial iteration
 | 
						|
LOC(divloop):
 | 
						|
	sll	Q, N, Q
 | 
						|
	DEVELOP_QUOTIENT_BITS(1, 0)
 | 
						|
LOC(end_regular_divide):
 | 
						|
	subcc	ITER, 1, ITER
 | 
						|
	bge	LOC(divloop)
 | 
						|
	tst	R
 | 
						|
	bl,a	LOC(got_result)
 | 
						|
	! non-restoring fixup here (one instruction only!)
 | 
						|
ifelse(OP, `div',
 | 
						|
`	sub	Q, 1, Q
 | 
						|
', `	add	R, divisor, R
 | 
						|
')
 | 
						|
 | 
						|
LOC(got_result):
 | 
						|
ifelse(S, `true',
 | 
						|
`	! check to see if answer should be < 0
 | 
						|
	tst	SIGN
 | 
						|
	bl,a	1f
 | 
						|
	ifelse(OP, `div', `sub %g0, Q, Q', `sub %g0, R, R')
 | 
						|
1:')
 | 
						|
	retl
 | 
						|
	ifelse(OP, `div', `mov Q, %o0', `mov R, %o0')
 | 
						|
 | 
						|
END(NAME)
 |