diff options
Diffstat (limited to 'sysdeps/alpha/ldiv.S')
| -rw-r--r-- | sysdeps/alpha/ldiv.S | 218 |
1 files changed, 218 insertions, 0 deletions
diff --git a/sysdeps/alpha/ldiv.S b/sysdeps/alpha/ldiv.S new file mode 100644 index 0000000000..63b0fd8d71 --- /dev/null +++ b/sysdeps/alpha/ldiv.S @@ -0,0 +1,218 @@ +/* Copyright (C) 1996-2014 Free Software Foundation, Inc. + This file is part of the GNU C Library. + Contributed by Richard Henderson <rth@tamu.edu>. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library. If not, see + <http://www.gnu.org/licenses/>. */ + +#include "div_libc.h" + +#undef FRAME +#ifdef __alpha_fix__ +#define FRAME 0 +#else +#define FRAME 16 +#endif + +#undef X +#undef Y +#define X $17 +#define Y $18 + + .set noat + + .align 4 + .globl ldiv + .ent ldiv +ldiv: + .frame sp, FRAME, ra +#if FRAME > 0 + lda sp, -FRAME(sp) +#endif +#ifdef PROF + .set macro + ldgp gp, 0(pv) + lda AT, _mcount + jsr AT, (AT), _mcount + .set nomacro + .prologue 1 +#else + .prologue 0 +#endif + + beq Y, $divbyzero + excb + mf_fpcr $f10 + + _ITOFT2 X, $f0, 0, Y, $f1, 8 + + .align 4 + cvtqt $f0, $f0 + cvtqt $f1, $f1 + divt/c $f0, $f1, $f0 + unop + + /* Check to see if X fit in the double as an exact value. */ + sll X, (64-53), AT + sra AT, (64-53), AT + cmpeq X, AT, AT + beq AT, $x_big + + /* If we get here, we're expecting exact results from the division. + Do nothing else besides convert and clean up. */ + cvttq/c $f0, $f0 + excb + mt_fpcr $f10 + _FTOIT $f0, $0, 0 + +$egress: + mulq $0, Y, $1 + subq X, $1, $1 + + stq $0, 0($16) + stq $1, 8($16) + mov $16, $0 + +#if FRAME > 0 + lda sp, FRAME(sp) +#endif + ret + + .align 4 +$x_big: + /* If we get here, X is large enough that we don't expect exact + results, and neither X nor Y got mis-translated for the fp + division. Our task is to take the fp result, figure out how + far it's off from the correct result and compute a fixup. */ + +#define Q v0 /* quotient */ +#define R t0 /* remainder */ +#define SY t1 /* scaled Y */ +#define S t2 /* scalar */ +#define QY t3 /* Q*Y */ + + /* The fixup code below can only handle unsigned values. */ + or X, Y, AT + mov $31, t5 + blt AT, $fix_sign_in +$fix_sign_in_ret1: + cvttq/c $f0, $f0 + + _FTOIT $f0, Q, 8 +$fix_sign_in_ret2: + mulq Q, Y, QY + excb + mt_fpcr $f10 + + .align 4 + subq QY, X, R + mov Y, SY + mov 1, S + bgt R, $q_high + +$q_high_ret: + subq X, QY, R + mov Y, SY + mov 1, S + bgt R, $q_low + +$q_low_ret: + negq Q, t4 + cmovlbs t5, t4, Q + br $egress + + .align 4 + /* The quotient that we computed was too large. We need to reduce + it by S such that Y*S >= R. Obviously the closer we get to the + correct value the better, but overshooting high is ok, as we'll + fix that up later. */ +0: + addq SY, SY, SY + addq S, S, S +$q_high: + cmpult SY, R, AT + bne AT, 0b + + subq Q, S, Q + unop + subq QY, SY, QY + br $q_high_ret + + .align 4 + /* The quotient that we computed was too small. Divide Y by the + current remainder (R) and add that to the existing quotient (Q). + The expectation, of course, is that R is much smaller than X. */ + /* Begin with a shift-up loop. Compute S such that Y*S >= R. We + already have a copy of Y in SY and the value 1 in S. */ +0: + addq SY, SY, SY + addq S, S, S +$q_low: + cmpult SY, R, AT + bne AT, 0b + + /* Shift-down and subtract loop. Each iteration compares our scaled + Y (SY) with the remainder (R); if SY <= R then X is divisible by + Y's scalar (S) so add it to the quotient (Q). */ +2: addq Q, S, t3 + srl S, 1, S + cmpule SY, R, AT + subq R, SY, t4 + + cmovne AT, t3, Q + cmovne AT, t4, R + srl SY, 1, SY + bne S, 2b + + br $q_low_ret + + .align 4 +$fix_sign_in: + /* If we got here, then X|Y is negative. Need to adjust everything + such that we're doing unsigned division in the fixup loop. */ + /* T5 is true if result should be negative. */ + xor X, Y, AT + cmplt AT, 0, t5 + cmplt X, 0, AT + negq X, t0 + + cmovne AT, t0, X + cmplt Y, 0, AT + negq Y, t0 + + cmovne AT, t0, Y + blbc t5, $fix_sign_in_ret1 + + cvttq/c $f0, $f0 + _FTOIT $f0, Q, 8 + .align 3 + negq Q, Q + br $fix_sign_in_ret2 + +$divbyzero: + mov a0, v0 + lda a0, GEN_INTDIV + call_pal PAL_gentrap + stq zero, 0(v0) + stq zero, 8(v0) + +#if FRAME > 0 + lda sp, FRAME(sp) +#endif + ret + + .end ldiv + +weak_alias (ldiv, lldiv) +weak_alias (ldiv, imaxdiv) |