This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
[PATCH] SPU float format: Round towards zero
- From: "Ulrich Weigand" <uweigand at de dot ibm dot com>
- To: gcc-patches at gcc dot gnu dot org
- Cc: trevor_smigiel at playstation dot sony dot com, andrew_pinski at playstation dot sony dot com, rguenther at suse dot de, joseph at codesourcery dot com
- Date: Mon, 4 Aug 2008 21:29:01 +0200 (CEST)
- Subject: [PATCH] SPU float format: Round towards zero
Hello,
this patch addresses a second feature of SPU single-precision floating-point:
all operations round towards zero.
I've added a new element of the real_format structure to express this
property. This is taken into account in round_for_format for the standard
arithmetic operations, and in builtins.c when choosing the rounding mode
to use for the MPFR library for compile-time computations.
One problem with always round towards zero is generation of floating-point
constants like __FLT_MIN__. To ensure that when parsing the decimal
representation of such constants back in while using round-towards-zero,
the correct value in the target binary floating-point format results,
we have to round *away* from zero when generating the decimal value.
This patch fixes this using the method suggested by Joseph Myers here:
http://gcc.gnu.org/ml/gcc-patches/2007-11/msg00082.html
In addition, it turned out that the SPU back-end didn't always follow
round-towards-zero itself: for DImode to SFmode conversions, the default
libgcc routine was used (which implements round-to-nearest). The patch
adds inline implementations of those conversions that round towards zero,
as already specified in the SPU Language Extensions specification.
Finally, a number of test cases implicitly (or explicitly) tested
for the round-to-nearest behavious. This patch also fixes those
tests to check for round-towards-zero on the SPU.
Tested on spu-elf on mainline and 4.3, fixes the following FAILs:
FAIL: gcc.c-torture/execute/conversion.c execution, -O0
FAIL: gcc.c-torture/execute/conversion.c execution, -O1
FAIL: gcc.c-torture/execute/floatunsisf-1.c execution, -O0
FAIL: gcc.c-torture/execute/floatunsisf-1.c execution, -O1
FAIL: gcc.c-torture/execute/floatunsisf-1.c execution, -O2
FAIL: gcc.c-torture/execute/floatunsisf-1.c execution, -O3 -fomit-frame-pointer
FAIL: gcc.c-torture/execute/floatunsisf-1.c execution, -O3 -g
FAIL: gcc.c-torture/execute/floatunsisf-1.c execution, -Os
OK to commit?
Bye,
Ulrich
ChangeLog:
* real.h (struct real_format): New member round_towards_zero.
* real.c (round_for_format): Respect fmt->round_towards_zero.
(ieee_single_format, mips_single_format, motorola_single_format,
spu_single_format, ieee_double_format, mips_double_format,
motorola_double_format, ieee_extended_motorola_format,
ieee_extended_intel_96_format, ieee_extended_intel_128_format,
ieee_extended_intel_96_round_53_format, ibm_extended_format,
mips_extended_format, ieee_quad_format, mips_quad_format,
vax_f_format, vax_d_format, vax_g_format): Initialize it.
* config/pdp11/pdp11.c (pdp11_f_format, pdp11_d_format): Likewise.
* builtins.s (do_mpfr_arg1): Consider round_towards_zero member of
real_format to choose rounding mode when calling MPFR functions.
(do_mpfr_arg2, do_mpfr_arg3, do_mpfr_sincos): Likewise.
(do_mpfr_bessel_n, do_mpfr_remquo, do_mpfr_lgamma_r): Likewise.
* real.h (real_to_decimal_for_mode): Add prototype.
* real.c (real_to_decimal_for_mode): Renames old real_to_decimal.
Respect target rounding mode when generating decimal representation.
(real_to_decimal): New stub for backwards compatibility.
* c-cppbuiltin.c (builtin_define_with_hex_fp_value): Use
real_to_decimal_for_mode instead of real_to_decimal.
* config/spu/spu.md ("floatdisf2", "floatunsdisf2"): New.
testsuite/ChangeLog:
* gcc.c-torture/execute/20031003-1.c (main): Update test to
accommodate SPU single-precision rounding mode.
* gcc.c-torture/execute/conversion.c (test_float_to_integer,
test_float_to_longlong_integer): Likewise.
* gcc.c-torture/execute/ieee/rbug.c (main): Likewise.
* gcc.dg/hex-round-1.c: Skip test on SPU.
* gcc.dg/hex-round-2.c: Likewise.
diff -x .svn -urNp gcc-head-orig/gcc/builtins.c gcc-head/gcc/builtins.c
--- gcc-head-orig/gcc/builtins.c 2008-07-30 17:33:38.000000000 +0200
+++ gcc-head/gcc/builtins.c 2008-08-03 20:21:45.000000000 +0200
@@ -12825,14 +12825,16 @@ do_mpfr_arg1 (tree arg, tree type, int (
&& (!min || real_compare (inclusive ? GE_EXPR: GT_EXPR , ra, min))
&& (!max || real_compare (inclusive ? LE_EXPR: LT_EXPR , ra, max)))
{
- const int prec = REAL_MODE_FORMAT (TYPE_MODE (type))->p;
+ const struct real_format *fmt = REAL_MODE_FORMAT (TYPE_MODE (type));
+ const int prec = fmt->p;
+ const mp_rnd_t rnd = fmt->round_towards_zero? GMP_RNDZ : GMP_RNDN;
int inexact;
mpfr_t m;
mpfr_init2 (m, prec);
mpfr_from_real (m, ra, GMP_RNDN);
mpfr_clear_flags ();
- inexact = func (m, m, GMP_RNDN);
+ inexact = func (m, m, rnd);
result = do_mpfr_ckconv (m, type, inexact);
mpfr_clear (m);
}
@@ -12867,7 +12869,9 @@ do_mpfr_arg2 (tree arg1, tree arg2, tree
if (real_isfinite (ra1) && real_isfinite (ra2))
{
- const int prec = REAL_MODE_FORMAT (TYPE_MODE (type))->p;
+ const struct real_format *fmt = REAL_MODE_FORMAT (TYPE_MODE (type));
+ const int prec = fmt->p;
+ const mp_rnd_t rnd = fmt->round_towards_zero? GMP_RNDZ : GMP_RNDN;
int inexact;
mpfr_t m1, m2;
@@ -12875,7 +12879,7 @@ do_mpfr_arg2 (tree arg1, tree arg2, tree
mpfr_from_real (m1, ra1, GMP_RNDN);
mpfr_from_real (m2, ra2, GMP_RNDN);
mpfr_clear_flags ();
- inexact = func (m1, m1, m2, GMP_RNDN);
+ inexact = func (m1, m1, m2, rnd);
result = do_mpfr_ckconv (m1, type, inexact);
mpfr_clears (m1, m2, NULL);
}
@@ -12913,7 +12917,9 @@ do_mpfr_arg3 (tree arg1, tree arg2, tree
if (real_isfinite (ra1) && real_isfinite (ra2) && real_isfinite (ra3))
{
- const int prec = REAL_MODE_FORMAT (TYPE_MODE (type))->p;
+ const struct real_format *fmt = REAL_MODE_FORMAT (TYPE_MODE (type));
+ const int prec = fmt->p;
+ const mp_rnd_t rnd = fmt->round_towards_zero? GMP_RNDZ : GMP_RNDN;
int inexact;
mpfr_t m1, m2, m3;
@@ -12922,7 +12928,7 @@ do_mpfr_arg3 (tree arg1, tree arg2, tree
mpfr_from_real (m2, ra2, GMP_RNDN);
mpfr_from_real (m3, ra3, GMP_RNDN);
mpfr_clear_flags ();
- inexact = func (m1, m1, m2, m3, GMP_RNDN);
+ inexact = func (m1, m1, m2, m3, rnd);
result = do_mpfr_ckconv (m1, type, inexact);
mpfr_clears (m1, m2, m3, NULL);
}
@@ -12956,7 +12962,9 @@ do_mpfr_sincos (tree arg, tree arg_sinp,
if (real_isfinite (ra))
{
- const int prec = REAL_MODE_FORMAT (TYPE_MODE (type))->p;
+ const struct real_format *fmt = REAL_MODE_FORMAT (TYPE_MODE (type));
+ const int prec = fmt->p;
+ const mp_rnd_t rnd = fmt->round_towards_zero? GMP_RNDZ : GMP_RNDN;
tree result_s, result_c;
int inexact;
mpfr_t m, ms, mc;
@@ -12964,7 +12972,7 @@ do_mpfr_sincos (tree arg, tree arg_sinp,
mpfr_inits2 (prec, m, ms, mc, NULL);
mpfr_from_real (m, ra, GMP_RNDN);
mpfr_clear_flags ();
- inexact = mpfr_sin_cos (ms, mc, m, GMP_RNDN);
+ inexact = mpfr_sin_cos (ms, mc, m, rnd);
result_s = do_mpfr_ckconv (ms, type, inexact);
result_c = do_mpfr_ckconv (mc, type, inexact);
mpfr_clears (m, ms, mc, NULL);
@@ -13029,14 +13037,17 @@ do_mpfr_bessel_n (tree arg1, tree arg2,
&& real_isfinite (ra)
&& (!min || real_compare (inclusive ? GE_EXPR: GT_EXPR , ra, min)))
{
- const int prec = REAL_MODE_FORMAT (TYPE_MODE (type))->p;
+ const struct real_format *fmt = REAL_MODE_FORMAT (TYPE_MODE (type));
+ const int prec = fmt->p;
+ const mp_rnd_t rnd = fmt->round_towards_zero? GMP_RNDZ : GMP_RNDN;
+ tree result_s, result_c;
int inexact;
mpfr_t m;
mpfr_init2 (m, prec);
mpfr_from_real (m, ra, GMP_RNDN);
mpfr_clear_flags ();
- inexact = func (m, n, m, GMP_RNDN);
+ inexact = func (m, n, m, rnd);
result = do_mpfr_ckconv (m, type, inexact);
mpfr_clear (m);
}
@@ -13070,7 +13081,9 @@ do_mpfr_remquo (tree arg0, tree arg1, tr
if (real_isfinite (ra0) && real_isfinite (ra1))
{
- const int prec = REAL_MODE_FORMAT (TYPE_MODE (type))->p;
+ const struct real_format *fmt = REAL_MODE_FORMAT (TYPE_MODE (type));
+ const int prec = fmt->p;
+ const mp_rnd_t rnd = fmt->round_towards_zero? GMP_RNDZ : GMP_RNDN;
tree result_rem;
long integer_quo;
mpfr_t m0, m1;
@@ -13079,7 +13092,7 @@ do_mpfr_remquo (tree arg0, tree arg1, tr
mpfr_from_real (m0, ra0, GMP_RNDN);
mpfr_from_real (m1, ra1, GMP_RNDN);
mpfr_clear_flags ();
- mpfr_remquo (m0, &integer_quo, m0, m1, GMP_RNDN);
+ mpfr_remquo (m0, &integer_quo, m0, m1, rnd);
/* Remquo is independent of the rounding mode, so pass
inexact=0 to do_mpfr_ckconv(). */
result_rem = do_mpfr_ckconv (m0, type, /*inexact=*/ 0);
@@ -13147,7 +13160,9 @@ do_mpfr_lgamma_r (tree arg, tree arg_sg,
&& ra->cl != rvc_zero
&& !(real_isneg(ra) && real_isinteger(ra, TYPE_MODE (type))))
{
- const int prec = REAL_MODE_FORMAT (TYPE_MODE (type))->p;
+ const struct real_format *fmt = REAL_MODE_FORMAT (TYPE_MODE (type));
+ const int prec = fmt->p;
+ const mp_rnd_t rnd = fmt->round_towards_zero? GMP_RNDZ : GMP_RNDN;
int inexact, sg;
mpfr_t m;
tree result_lg;
@@ -13155,7 +13170,7 @@ do_mpfr_lgamma_r (tree arg, tree arg_sg,
mpfr_init2 (m, prec);
mpfr_from_real (m, ra, GMP_RNDN);
mpfr_clear_flags ();
- inexact = mpfr_lgamma (m, &sg, m, GMP_RNDN);
+ inexact = mpfr_lgamma (m, &sg, m, rnd);
result_lg = do_mpfr_ckconv (m, type, inexact);
mpfr_clear (m);
if (result_lg)
diff -x .svn -urNp gcc-head-orig/gcc/c-cppbuiltin.c gcc-head/gcc/c-cppbuiltin.c
--- gcc-head-orig/gcc/c-cppbuiltin.c 2008-07-30 17:33:40.000000000 +0200
+++ gcc-head/gcc/c-cppbuiltin.c 2008-08-03 20:21:46.000000000 +0200
@@ -846,7 +846,7 @@ builtin_define_with_int_value (const cha
/* Pass an object-like macro a hexadecimal floating-point value. */
static void
builtin_define_with_hex_fp_value (const char *macro,
- tree type ATTRIBUTE_UNUSED, int digits,
+ tree type, int digits,
const char *hex_str,
const char *fp_suffix,
const char *fp_cast)
@@ -865,7 +865,8 @@ builtin_define_with_hex_fp_value (const
then print it back out as decimal. */
real_from_string (&real, hex_str);
- real_to_decimal (dec_str, &real, sizeof (dec_str), digits, 0);
+ real_to_decimal_for_mode (dec_str, &real, sizeof (dec_str), digits, 0,
+ TYPE_MODE (type));
/* Assemble the macro in the following fashion
macro = fp_cast [dec_str fp_suffix] */
diff -x .svn -urNp gcc-head-orig/gcc/config/pdp11/pdp11.c gcc-head/gcc/config/pdp11/pdp11.c
--- gcc-head-orig/gcc/config/pdp11/pdp11.c 2008-06-10 23:22:37.000000000 +0200
+++ gcc-head/gcc/config/pdp11/pdp11.c 2008-08-03 20:21:46.000000000 +0200
@@ -78,6 +78,7 @@ const struct real_format pdp11_f_format
false,
false,
false,
+ false,
false
};
@@ -97,6 +98,7 @@ const struct real_format pdp11_d_format
false,
false,
false,
+ false,
false
};
diff -x .svn -urNp gcc-head-orig/gcc/config/spu/spu.md gcc-head/gcc/config/spu/spu.md
--- gcc-head-orig/gcc/config/spu/spu.md 2008-08-03 20:21:28.000000000 +0200
+++ gcc-head/gcc/config/spu/spu.md 2008-08-03 20:21:48.000000000 +0200
@@ -664,6 +664,65 @@
"frds\t%0,%1"
[(set_attr "type" "fpd")])
+(define_expand "floatdisf2"
+ [(set (match_operand:SF 0 "register_operand" "")
+ (float:SF (match_operand:DI 1 "register_operand" "")))]
+ ""
+ {
+ rtx c0 = gen_reg_rtx (SImode);
+ rtx r0 = gen_reg_rtx (DImode);
+ rtx r1 = gen_reg_rtx (SFmode);
+ rtx r2 = gen_reg_rtx (SImode);
+ rtx setneg = gen_reg_rtx (SImode);
+ rtx isneg = gen_reg_rtx (SImode);
+ rtx neg = gen_reg_rtx (DImode);
+ rtx mask = gen_reg_rtx (DImode);
+
+ emit_move_insn (c0, GEN_INT (-0x80000000ll));
+
+ emit_insn (gen_negdi2 (neg, operands[1]));
+ emit_insn (gen_cgt_di_m1 (isneg, operands[1]));
+ emit_insn (gen_extend_compare (mask, isneg));
+ emit_insn (gen_selb (r0, neg, operands[1], mask));
+ emit_insn (gen_andc_si (setneg, c0, isneg));
+
+ emit_insn (gen_floatunsdisf2 (r1, r0));
+
+ emit_insn (gen_iorsi3 (r2, gen_rtx_SUBREG (SImode, r1, 0), setneg));
+ emit_move_insn (operands[0], gen_rtx_SUBREG (SFmode, r2, 0));
+ DONE;
+ })
+
+(define_insn_and_split "floatunsdisf2"
+ [(set (match_operand:SF 0 "register_operand" "=r")
+ (unsigned_float:SF (match_operand:DI 1 "register_operand" "r")))
+ (clobber (match_scratch:SF 2 "=r"))
+ (clobber (match_scratch:SF 3 "=r"))
+ (clobber (match_scratch:SF 4 "=r"))]
+ ""
+ "#"
+ "reload_completed"
+ [(set (match_dup:SF 0)
+ (unsigned_float:SF (match_dup:DI 1)))]
+ {
+ rtx op1_v4si = gen_rtx_REG (V4SImode, REGNO (operands[1]));
+ rtx op2_v4sf = gen_rtx_REG (V4SFmode, REGNO (operands[2]));
+ rtx op2_ti = gen_rtx_REG (TImode, REGNO (operands[2]));
+ rtx op3_ti = gen_rtx_REG (TImode, REGNO (operands[3]));
+
+ REAL_VALUE_TYPE scale;
+ real_2expN (&scale, 32, SFmode);
+
+ emit_insn (gen_floatunsv4siv4sf2 (op2_v4sf, op1_v4si));
+ emit_insn (gen_shlqby_ti (op3_ti, op2_ti, GEN_INT (4)));
+
+ emit_move_insn (operands[4],
+ CONST_DOUBLE_FROM_REAL_VALUE (scale, SFmode));
+ emit_insn (gen_fma_sf (operands[0],
+ operands[2], operands[4], operands[3]));
+ DONE;
+ })
+
;; Do (double)(operands[1]+0x80000000u)-(double)0x80000000
(define_expand "floatsidf2"
[(set (match_operand:DF 0 "register_operand" "")
diff -x .svn -urNp gcc-head-orig/gcc/real.c gcc-head/gcc/real.c
--- gcc-head-orig/gcc/real.c 2008-08-03 20:21:29.000000000 +0200
+++ gcc-head/gcc/real.c 2008-08-03 20:21:48.000000000 +0200
@@ -1443,20 +1443,30 @@ rtd_divmod (REAL_VALUE_TYPE *num, REAL_V
/* Render R as a decimal floating point constant. Emit DIGITS significant
digits in the result, bounded by BUF_SIZE. If DIGITS is 0, choose the
maximum for the representation. If CROP_TRAILING_ZEROS, strip trailing
- zeros. */
+ zeros. If MODE is VOIDmode, round to nearest value. Otherwise, round
+ to a string that, when parsed back in mode MODE, yields the same value. */
#define M_LOG10_2 0.30102999566398119521
void
-real_to_decimal (char *str, const REAL_VALUE_TYPE *r_orig, size_t buf_size,
- size_t digits, int crop_trailing_zeros)
+real_to_decimal_for_mode (char *str, const REAL_VALUE_TYPE *r_orig,
+ size_t buf_size, size_t digits,
+ int crop_trailing_zeros, enum machine_mode mode)
{
+ const struct real_format *fmt = NULL;
const REAL_VALUE_TYPE *one, *ten;
REAL_VALUE_TYPE r, pten, u, v;
int dec_exp, cmp_one, digit;
size_t max_digits;
char *p, *first, *last;
bool sign;
+ bool round_up;
+
+ if (mode != VOIDmode)
+ {
+ fmt = REAL_MODE_FORMAT (mode);
+ gcc_assert (fmt);
+ }
r = *r_orig;
switch (r.cl)
@@ -1672,17 +1682,31 @@ real_to_decimal (char *str, const REAL_V
digit = rtd_divmod (&r, &pten);
/* Round the result. */
- if (digit == 5)
+ if (fmt && fmt->round_towards_zero)
{
- /* Round to nearest. If R is nonzero there are additional
- nonzero digits to be extracted. */
+ /* If the format uses round towards zero when parsing the string
+ back in, we need to always round away from zero here. */
if (cmp_significand_0 (&r))
digit++;
- /* Round to even. */
- else if ((p[-1] - '0') & 1)
- digit++;
+ round_up = digit > 0;
}
- if (digit > 5)
+ else
+ {
+ if (digit == 5)
+ {
+ /* Round to nearest. If R is nonzero there are additional
+ nonzero digits to be extracted. */
+ if (cmp_significand_0 (&r))
+ digit++;
+ /* Round to even. */
+ else if ((p[-1] - '0') & 1)
+ digit++;
+ }
+
+ round_up = digit > 5;
+ }
+
+ if (round_up)
{
while (p > first)
{
@@ -1716,6 +1740,24 @@ real_to_decimal (char *str, const REAL_V
/* Append the exponent. */
sprintf (last, "e%+d", dec_exp);
+
+ /* Verify that we can read the original value back in. */
+ if (mode != VOIDmode)
+ {
+ real_from_string (&r, str);
+ real_convert (&r, mode, &r);
+ gcc_assert (real_identical (&r, r_orig));
+ }
+}
+
+/* Likewise, except always uses round-to-nearest. */
+
+void
+real_to_decimal (char *str, const REAL_VALUE_TYPE *r_orig, size_t buf_size,
+ size_t digits, int crop_trailing_zeros)
+{
+ real_to_decimal_for_mode (str, r_orig, buf_size,
+ digits, crop_trailing_zeros, VOIDmode);
}
/* Render R as a hexadecimal floating point constant. Emit DIGITS
@@ -2389,9 +2431,8 @@ static void
round_for_format (const struct real_format *fmt, REAL_VALUE_TYPE *r)
{
int p2, np2, i, w;
- unsigned long sticky;
- bool guard, lsb;
int emin2m1, emax2;
+ bool round_up = false;
if (r->decimal)
{
@@ -2463,21 +2504,28 @@ round_for_format (const struct real_form
}
}
- /* There are P2 true significand bits, followed by one guard bit,
- followed by one sticky bit, followed by stuff. Fold nonzero
- stuff into the sticky bit. */
-
- sticky = 0;
- for (i = 0, w = (np2 - 1) / HOST_BITS_PER_LONG; i < w; ++i)
- sticky |= r->sig[i];
- sticky |=
- r->sig[w] & (((unsigned long)1 << ((np2 - 1) % HOST_BITS_PER_LONG)) - 1);
+ if (!fmt->round_towards_zero)
+ {
+ /* There are P2 true significand bits, followed by one guard bit,
+ followed by one sticky bit, followed by stuff. Fold nonzero
+ stuff into the sticky bit. */
+ unsigned long sticky;
+ bool guard, lsb;
+
+ sticky = 0;
+ for (i = 0, w = (np2 - 1) / HOST_BITS_PER_LONG; i < w; ++i)
+ sticky |= r->sig[i];
+ sticky |= r->sig[w]
+ & (((unsigned long)1 << ((np2 - 1) % HOST_BITS_PER_LONG)) - 1);
+
+ guard = test_significand_bit (r, np2 - 1);
+ lsb = test_significand_bit (r, np2);
- guard = test_significand_bit (r, np2 - 1);
- lsb = test_significand_bit (r, np2);
+ /* Round to even. */
+ round_up = guard && (sticky || lsb);
+ }
- /* Round to even. */
- if (guard && (sticky || lsb))
+ if (round_up)
{
REAL_VALUE_TYPE u;
get_zero (&u, 0);
@@ -2817,6 +2865,7 @@ const struct real_format ieee_single_for
128,
31,
31,
+ false,
true,
true,
true,
@@ -2836,6 +2885,7 @@ const struct real_format mips_single_for
128,
31,
31,
+ false,
true,
true,
true,
@@ -2855,6 +2905,7 @@ const struct real_format motorola_single
128,
31,
31,
+ false,
true,
true,
true,
@@ -2885,6 +2936,7 @@ const struct real_format spu_single_form
129,
31,
31,
+ true,
false,
false,
true,
@@ -3092,6 +3144,7 @@ const struct real_format ieee_double_for
1024,
63,
63,
+ false,
true,
true,
true,
@@ -3111,6 +3164,7 @@ const struct real_format mips_double_for
1024,
63,
63,
+ false,
true,
true,
true,
@@ -3130,6 +3184,7 @@ const struct real_format motorola_double
1024,
63,
63,
+ false,
true,
true,
true,
@@ -3467,6 +3522,7 @@ const struct real_format ieee_extended_m
16384,
95,
95,
+ false,
true,
true,
true,
@@ -3486,6 +3542,7 @@ const struct real_format ieee_extended_i
16384,
79,
79,
+ false,
true,
true,
true,
@@ -3505,6 +3562,7 @@ const struct real_format ieee_extended_i
16384,
79,
79,
+ false,
true,
true,
true,
@@ -3526,6 +3584,7 @@ const struct real_format ieee_extended_i
16384,
79,
79,
+ false,
true,
true,
true,
@@ -3612,6 +3671,7 @@ const struct real_format ibm_extended_fo
1024,
127,
-1,
+ false,
true,
true,
true,
@@ -3631,6 +3691,7 @@ const struct real_format mips_extended_f
1024,
127,
-1,
+ false,
true,
true,
true,
@@ -3892,6 +3953,7 @@ const struct real_format ieee_quad_forma
16384,
127,
127,
+ false,
true,
true,
true,
@@ -3911,6 +3973,7 @@ const struct real_format mips_quad_forma
16384,
127,
127,
+ false,
true,
true,
true,
@@ -4214,6 +4277,7 @@ const struct real_format vax_f_format =
false,
false,
false,
+ false,
false
};
@@ -4233,6 +4297,7 @@ const struct real_format vax_d_format =
false,
false,
false,
+ false,
false
};
@@ -4252,6 +4317,7 @@ const struct real_format vax_g_format =
false,
false,
false,
+ false,
false
};
@@ -4321,6 +4387,7 @@ const struct real_format decimal_single_
96,
31,
31,
+ false,
true,
true,
true,
@@ -4341,6 +4408,7 @@ const struct real_format decimal_double_
384,
63,
63,
+ false,
true,
true,
true,
@@ -4361,6 +4429,7 @@ const struct real_format decimal_quad_fo
6144,
127,
127,
+ false,
true,
true,
true,
@@ -4404,6 +4473,7 @@ const struct real_format real_internal_f
MAX_EXP,
-1,
-1,
+ false,
true,
true,
false,
diff -x .svn -urNp gcc-head-orig/gcc/real.h gcc-head/gcc/real.h
--- gcc-head-orig/gcc/real.h 2008-08-03 20:21:29.000000000 +0200
+++ gcc-head/gcc/real.h 2008-08-03 20:21:48.000000000 +0200
@@ -147,6 +147,9 @@ struct real_format
or -1 for a complex encoding. */
int signbit_rw;
+ /* Default rounding mode for operations on this format. */
+ bool round_towards_zero;
+
/* Properties of the format. */
bool has_nans;
bool has_inf;
@@ -216,6 +219,11 @@ extern bool exact_real_truncate (enum ma
extern void real_to_decimal (char *, const REAL_VALUE_TYPE *, size_t,
size_t, int);
+/* Render R as a decimal floating point constant, rounded so as to be
+ parsed back to the same value when interpreted in mode MODE. */
+extern void real_to_decimal_for_mode (char *, const REAL_VALUE_TYPE *, size_t,
+ size_t, int, enum machine_mode);
+
/* Render R as a hexadecimal floating point constant. */
extern void real_to_hexadecimal (char *, const REAL_VALUE_TYPE *,
size_t, size_t, int);
diff -x .svn -urNp gcc-head-orig/gcc/testsuite/gcc.c-torture/execute/20031003-1.c gcc-head/gcc/testsuite/gcc.c-torture/execute/20031003-1.c
--- gcc-head-orig/gcc/testsuite/gcc.c-torture/execute/20031003-1.c 2008-06-10 21:54:42.000000000 +0200
+++ gcc-head/gcc/testsuite/gcc.c-torture/execute/20031003-1.c 2008-08-03 20:21:48.000000000 +0200
@@ -19,9 +19,15 @@ int main()
#if INT_MAX == 2147483647
if (f1() != 2147483647)
abort ();
+#ifdef __SPU__
+ /* SPU float rounds towards zero. */
+ if (f2() != 0x7fffff80)
+ abort ();
+#else
if (f2() != 2147483647)
abort ();
#endif
+#endif
return 0;
}
diff -x .svn -urNp gcc-head-orig/gcc/testsuite/gcc.c-torture/execute/conversion.c gcc-head/gcc/testsuite/gcc.c-torture/execute/conversion.c
--- gcc-head-orig/gcc/testsuite/gcc.c-torture/execute/conversion.c 2008-06-10 21:57:42.000000000 +0200
+++ gcc-head/gcc/testsuite/gcc.c-torture/execute/conversion.c 2008-08-03 20:21:50.000000000 +0200
@@ -284,9 +284,15 @@ test_float_to_integer()
abort();
if (f2u(1.99) != 1)
abort();
+#ifdef __SPU__
+ /* SPU float rounds towards zero. */
+ if (f2u((float) ((~0U) >> 1)) != 0x7fffff80)
+ abort();
+#else
if (f2u((float) ((~0U) >> 1)) != (~0U) >> 1 && /* 0x7fffffff */
f2u((float) ((~0U) >> 1)) != ((~0U) >> 1) + 1)
abort();
+#endif
if (f2u((float) ~((~0U) >> 1)) != ~((~0U) >> 1)) /* 0x80000000 */
abort();
@@ -439,9 +445,15 @@ test_float_to_longlong_integer()
abort();
if (f2ull(1.99) != 1LL)
abort();
+#ifdef __SPU__
+ /* SPU float rounds towards zero. */
+ if (f2ull((float) ((~0ULL) >> 1)) != 0x7fffff8000000000ULL)
+ abort();
+#else
if (f2ull((float) ((~0ULL) >> 1)) != (~0ULL) >> 1 && /* 0x7fffffff */
f2ull((float) ((~0ULL) >> 1)) != ((~0ULL) >> 1) + 1)
abort();
+#endif
if (f2ull((float) ~((~0ULL) >> 1)) != ~((~0ULL) >> 1)) /* 0x80000000 */
abort();
diff -x .svn -urNp gcc-head-orig/gcc/testsuite/gcc.c-torture/execute/ieee/rbug.c gcc-head/gcc/testsuite/gcc.c-torture/execute/ieee/rbug.c
--- gcc-head-orig/gcc/testsuite/gcc.c-torture/execute/ieee/rbug.c 2008-06-10 21:53:18.000000000 +0200
+++ gcc-head/gcc/testsuite/gcc.c-torture/execute/ieee/rbug.c 2008-08-03 20:21:50.000000000 +0200
@@ -41,8 +41,14 @@ main ()
k = 0x8234508000000001ULL;
x = s (k);
k = (unsigned long long) x;
+#ifdef __SPU__
+ /* SPU float rounds towards zero. */
+ if (k != 0x8234500000000000ULL)
+ abort ();
+#else
if (k != 0x8234510000000000ULL)
abort ();
+#endif
exit (0);
}
diff -x .svn -urNp gcc-head-orig/gcc/testsuite/gcc.dg/hex-round-1.c gcc-head/gcc/testsuite/gcc.dg/hex-round-1.c
--- gcc-head-orig/gcc/testsuite/gcc.dg/hex-round-1.c 2008-06-10 22:34:44.000000000 +0200
+++ gcc-head/gcc/testsuite/gcc.dg/hex-round-1.c 2008-08-03 20:21:54.000000000 +0200
@@ -1,6 +1,7 @@
/* Test for hexadecimal float rounding: bug 21720. */
/* { dg-do link } */
/* { dg-options "-O -std=gnu99" } */
+/* { dg-skip-if "SPU float rounds towards zero" { spu-*-* } } */
#include <float.h>
diff -x .svn -urNp gcc-head-orig/gcc/testsuite/gcc.dg/hex-round-2.c gcc-head/gcc/testsuite/gcc.dg/hex-round-2.c
--- gcc-head-orig/gcc/testsuite/gcc.dg/hex-round-2.c 2008-06-10 22:31:26.000000000 +0200
+++ gcc-head/gcc/testsuite/gcc.dg/hex-round-2.c 2008-08-03 20:21:55.000000000 +0200
@@ -2,6 +2,7 @@
in number. */
/* { dg-do link } */
/* { dg-options "-O -std=gnu99" } */
+/* { dg-skip-if "SPU float rounds towards zero" { spu-*-* } } */
#include <float.h>
--
Dr. Ulrich Weigand
GNU Toolchain for Linux on System z and Cell BE
Ulrich.Weigand@de.ibm.com