From 1b35a57b1c1781f0fc8fc554f732b3a5408c5244 Mon Sep 17 00:00:00 2001 From: "David S. Miller" Date: Tue, 15 May 2012 11:23:01 -0700 Subject: sparc32: Kill off software 32-bit multiply/divide routines. For the explicit calls to .udiv/.umul in assembler, I made a mechanical (read as: safe) transformation. I didn't attempt to make any simplifications. In particular, __ndelay and __udelay can be simplified significantly. Some of the %y reads are unnecessary and these routines have no need any longer for allocating a register window, they can be leaf functions. Signed-off-by: David S. Miller --- arch/sparc/lib/Makefile | 2 +- arch/sparc/lib/divdi3.S | 4 +- arch/sparc/lib/ksyms.c | 17 --- arch/sparc/lib/mul.S | 137 ----------------- arch/sparc/lib/muldi3.S | 4 +- arch/sparc/lib/rem.S | 384 ----------------------------------------------- arch/sparc/lib/sdiv.S | 381 ---------------------------------------------- arch/sparc/lib/udiv.S | 357 ------------------------------------------- arch/sparc/lib/udivdi3.S | 3 +- arch/sparc/lib/umul.S | 171 --------------------- arch/sparc/lib/urem.S | 357 ------------------------------------------- 11 files changed, 7 insertions(+), 1810 deletions(-) delete mode 100644 arch/sparc/lib/mul.S delete mode 100644 arch/sparc/lib/rem.S delete mode 100644 arch/sparc/lib/sdiv.S delete mode 100644 arch/sparc/lib/udiv.S delete mode 100644 arch/sparc/lib/umul.S delete mode 100644 arch/sparc/lib/urem.S (limited to 'arch/sparc/lib') diff --git a/arch/sparc/lib/Makefile b/arch/sparc/lib/Makefile index 33d8d85..ead6df2 100644 --- a/arch/sparc/lib/Makefile +++ b/arch/sparc/lib/Makefile @@ -4,7 +4,7 @@ asflags-y := -ansi -DST_DIV0=0x02 ccflags-y := -Werror -lib-$(CONFIG_SPARC32) += mul.o rem.o sdiv.o udiv.o umul.o urem.o ashrdi3.o +lib-$(CONFIG_SPARC32) += ashrdi3.o lib-$(CONFIG_SPARC32) += memcpy.o memset.o lib-y += strlen.o lib-y += checksum_$(BITS).o diff --git a/arch/sparc/lib/divdi3.S b/arch/sparc/lib/divdi3.S index d74bc09..9614b48 100644 --- a/arch/sparc/lib/divdi3.S +++ b/arch/sparc/lib/divdi3.S @@ -19,7 +19,6 @@ Boston, MA 02111-1307, USA. */ .text .align 4 - .global .udiv .globl __divdi3 __divdi3: save %sp,-104,%sp @@ -83,8 +82,9 @@ __divdi3: bne .LL85 mov %i0,%o2 mov 1,%o0 - call .udiv,0 mov 0,%o1 + wr %g0, 0, %y + udiv %o0, %o1, %o0 mov %o0,%o4 mov %i0,%o2 .LL85: diff --git a/arch/sparc/lib/ksyms.c b/arch/sparc/lib/ksyms.c index 1bc8972..2dc3087 100644 --- a/arch/sparc/lib/ksyms.c +++ b/arch/sparc/lib/ksyms.c @@ -61,16 +61,6 @@ extern void ___rw_read_try(void); extern void ___rw_read_exit(void); extern void ___rw_write_enter(void); -/* Alias functions whose names begin with "." and export the aliases. - * The module references will be fixed up by module_frob_arch_sections. - */ -extern int _Div(int, int); -extern int _Mul(int, int); -extern int _Rem(int, int); -extern unsigned _Udiv(unsigned, unsigned); -extern unsigned _Umul(unsigned, unsigned); -extern unsigned _Urem(unsigned, unsigned); - /* Networking helper routines. */ EXPORT_SYMBOL(__csum_partial_copy_sparc_generic); @@ -95,13 +85,6 @@ EXPORT_SYMBOL(__ashldi3); EXPORT_SYMBOL(__lshrdi3); EXPORT_SYMBOL(__muldi3); EXPORT_SYMBOL(__divdi3); - -EXPORT_SYMBOL(_Rem); -EXPORT_SYMBOL(_Urem); -EXPORT_SYMBOL(_Mul); -EXPORT_SYMBOL(_Umul); -EXPORT_SYMBOL(_Div); -EXPORT_SYMBOL(_Udiv); #endif /* diff --git a/arch/sparc/lib/mul.S b/arch/sparc/lib/mul.S deleted file mode 100644 index c45470d..0000000 --- a/arch/sparc/lib/mul.S +++ /dev/null @@ -1,137 +0,0 @@ -/* - * 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 - .globl _Mul -.mul: -_Mul: /* needed for export */ - 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 index 7f17872..9794939 100644 --- a/arch/sparc/lib/muldi3.S +++ b/arch/sparc/lib/muldi3.S @@ -63,12 +63,12 @@ __muldi3: rd %y, %o1 mov %o1, %l3 mov %i1, %o0 - call .umul mov %i2, %o1 + umul %o0, %o1, %o0 mov %o0, %l0 mov %i0, %o0 - call .umul mov %i3, %o1 + umul %o0, %o1, %o0 add %l0, %o0, %l0 mov %l2, %i0 add %l2, %l0, %i0 diff --git a/arch/sparc/lib/rem.S b/arch/sparc/lib/rem.S deleted file mode 100644 index 42fb862..0000000 --- a/arch/sparc/lib/rem.S +++ /dev/null @@ -1,384 +0,0 @@ -/* - * 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 - .globl _Rem -.rem: -_Rem: /* needed for export */ - ! 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/sdiv.S b/arch/sparc/lib/sdiv.S deleted file mode 100644 index f0a0d4e..0000000 --- a/arch/sparc/lib/sdiv.S +++ /dev/null @@ -1,381 +0,0 @@ -/* - * 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 - .globl _Div -.div: -_Div: /* needed for export */ - ! 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/udiv.S b/arch/sparc/lib/udiv.S deleted file mode 100644 index 2101405..0000000 --- a/arch/sparc/lib/udiv.S +++ /dev/null @@ -1,357 +0,0 @@ -/* - * 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 - .globl _Udiv -.udiv: -_Udiv: /* needed for export */ - - ! 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 index b430f1f..24e0a35 100644 --- a/arch/sparc/lib/udivdi3.S +++ b/arch/sparc/lib/udivdi3.S @@ -60,8 +60,9 @@ __udivdi3: bne .LL77 mov %i0,%o2 mov 1,%o0 - call .udiv,0 mov 0,%o1 + wr %g0, 0, %y + udiv %o0, %o1, %o0 mov %o0,%o3 mov %i0,%o2 .LL77: diff --git a/arch/sparc/lib/umul.S b/arch/sparc/lib/umul.S deleted file mode 100644 index 1f36ae6..0000000 --- a/arch/sparc/lib/umul.S +++ /dev/null @@ -1,171 +0,0 @@ -/* - * 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 - .globl _Umul -.umul: -_Umul: /* needed for export */ - 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 deleted file mode 100644 index 77123eb..0000000 --- a/arch/sparc/lib/urem.S +++ /dev/null @@ -1,357 +0,0 @@ -/* - * 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 - .globl _Urem -.urem: -_Urem: /* needed for export */ - - ! 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 -- cgit v1.1