diff options
| -rw-r--r-- | ChangeLog | 12 | ||||
| -rw-r--r-- | arith.c | 61 | ||||
| -rw-r--r-- | lib.c | 14 | ||||
| -rw-r--r-- | mpi-patches/fix-bad-shifts | 49 |
4 files changed, 122 insertions, 14 deletions
@@ -1,5 +1,17 @@ 2011-12-11 Kaz Kylheku <kaz@kylheku.com> + Bignum division implemented. More portability bugs found in MPI: + code like 1 << n, where n exceeds the width of the type int. + + * arith.c (trunc): New function, reimplementation of removed + trunc from lib.c. + + * lib.c (trunc): Removed. + + * mpi-patches/fix-bad-shifts: New file. + +2011-12-11 Kaz Kylheku <kaz@kylheku.com> + * arith.c (ABS): New macro. (plus, minus): Bugfix: must not pass signed values to mp_add_d and mp_sub_d functions. @@ -497,6 +497,67 @@ val mul(val anum, val bnum) abort(); } +val trunc(val anum, val bnum) +{ + int tag_a = tag(anum); + int tag_b = tag(bnum); + + switch (TAG_PAIR(tag_a, tag_b)) { + case TAG_PAIR(TAG_NUM, TAG_NUM): + { + cnum a = c_num(anum); + cnum b = c_num(bnum); + cnum ap = ABS(a); + cnum bp = ABS(b); + int neg = ((a < 0 && b > 0) || (a > 0 && b < 0)); + + if (b == 0) + uw_throw(numeric_error_s, lit("division by zero")); + + { + cnum quot = ap / bp; + return num(neg ? -quot : quot); + } + } + case TAG_PAIR(TAG_NUM, TAG_PTR): + type_check(bnum, BGNUM); + return zero; + case TAG_PAIR(TAG_PTR, TAG_NUM): + { + val n; + type_check(anum, BGNUM); + n = make_bignum(); + if (sizeof (int_ptr_t) <= sizeof (mp_digit)) { + cnum b = c_num(bnum); + cnum bp = ABS(b); + if (mp_div_d(mp(anum), bp, mp(n), 0) != MP_OKAY) + uw_throw(numeric_error_s, lit("division by zero")); + if (b < 0) + mp_neg(mp(n), mp(n)); + } else { + mp_int tmp; + mp_init(&tmp); + mp_set_intptr(&tmp, c_num(bnum)); + if (mp_div(mp(anum), &tmp, mp(n), 0) != MP_OKAY) + uw_throw(numeric_error_s, lit("division by zero")); + } + return normalize(n); + } + case TAG_PAIR(TAG_PTR, TAG_PTR): + { + val n; + type_check(anum, BGNUM); + type_check(bnum, BGNUM); + n = make_bignum(); + if (mp_div(mp(anum), mp(bnum), mp(n), 0) != MP_OKAY) + uw_throw(numeric_error_s, lit("division by zero")); + return normalize(n); + } + } + uw_throwf(error_s, lit("trunc: invalid operands ~s ~s"), anum, bnum, nao); + abort(); +} + void arith_init(void) { mp_init(&NUM_MAX_MP); @@ -857,20 +857,6 @@ val mulv(val nlist) return reduce_left(func_n2(mul), cdr(nlist), car(nlist), nil); } -val trunc(val anum, val bnum) -{ - cnum a = c_num(anum); - cnum b = c_num(bnum); - - numeric_assert (b != 0); - - { - cnum result = a / b; - numeric_assert (result <= NUM_MAX); - return num(result); - } -} - val mod(val anum, val bnum) { cnum a = c_num(anum); diff --git a/mpi-patches/fix-bad-shifts b/mpi-patches/fix-bad-shifts new file mode 100644 index 00000000..f6e7c979 --- /dev/null +++ b/mpi-patches/fix-bad-shifts @@ -0,0 +1,49 @@ +Index: mpi-1.8.6/mpi.c +=================================================================== +--- mpi-1.8.6.orig/mpi.c 2011-12-11 19:52:15.000000000 -0800 ++++ mpi-1.8.6/mpi.c 2011-12-11 19:53:09.000000000 -0800 +@@ -764,7 +764,7 @@ + if((pow = s_mp_ispow2d(d)) >= 0) { + mp_digit mask; + +- mask = (1 << pow) - 1; ++ mask = ((mp_digit) 1 << pow) - 1; + rem = DIGIT(a, 0) & mask; + + if(q) { +@@ -3068,7 +3068,7 @@ + return; + + /* Flush all the bits above 2^d in its digit */ +- dmask = (1 << nbit) - 1; ++ dmask = ((mp_digit) 1 << nbit) - 1; + dp[ndig] &= dmask; + + /* Flush all digits above the one with 2^d in it */ +@@ -3101,7 +3101,7 @@ + dp = DIGITS(mp); used = USED(mp); + d %= DIGIT_BIT; + +- mask = (1 << d) - 1; ++ mask = ((mp_digit) 1 << d) - 1; + + /* If the shift requires another digit, make sure we've got one to + work with */ +@@ -3149,7 +3149,7 @@ + s_mp_rshd(mp, d / DIGIT_BIT); + d %= DIGIT_BIT; + +- mask = (1 << d) - 1; ++ mask = ((mp_digit) 1 << d) - 1; + + save = 0; + for(ix = USED(mp) - 1; ix >= 0; ix--) { +@@ -3829,7 +3829,7 @@ + if((res = s_mp_pad(a, dig + 1)) != MP_OKAY) + return res; + +- DIGIT(a, dig) |= (1 << bit); ++ DIGIT(a, dig) |= ((mp_digit) 1 << bit); + + return MP_OKAY; + |