Backport ia64 fix for PR target/42542 from mainline.
gcc/ 2010-01-17 H.J. Lu <hongjiu.lu@intel.com> Backport from mainline: 2010-01-13 Steve Ellcey <sje@cup.hp.com> PR target/42542 * config/ia64/ia64.c (ia64_expand_vecint_compare): Convert GTU to GT for V2SI by subtracting (-(INT MAX) - 1) from both operands to make them signed. gcc/testsuite/ 2010-01-17 H.J. Lu <hongjiu.lu@intel.com> Backport from mainline: 2010-01-13 Steve Ellcey <sje@cup.hp.com> PR target/42542 * gcc.target/ia64/pr42542-1.c: New. * gcc.target/ia64/pr42542-2.c: New. * gcc.target/ia64/pr42542-3.c: New. From-SVN: r155990
This commit is contained in:
H.J. Lu
H.J. Lu
parent
273281e07b
commit
ff1bcfda6b
|
|
@ -1,3 +1,13 @@
|
|||
2010-01-17 H.J. Lu <hongjiu.lu@intel.com>
|
||||
|
||||
Backport from mainline:
|
||||
2010-01-13 Steve Ellcey <sje@cup.hp.com>
|
||||
|
||||
PR target/42542
|
||||
* config/ia64/ia64.c (ia64_expand_vecint_compare): Convert GTU to GT
|
||||
for V2SI by subtracting (-(INT MAX) - 1) from both operands to make
|
||||
them signed.
|
||||
|
||||
2010-01-17 Richard Guenther <rguenther@suse.de>
|
||||
|
||||
PR tree-optimization/42773
|
||||
|
|
|
|||
|
|
@ -1628,25 +1628,18 @@ ia64_expand_vecint_compare (enum rtx_code code, enum machine_mode mode,
|
|||
{
|
||||
rtx t1, t2, mask;
|
||||
|
||||
/* Perform a parallel modulo subtraction. */
|
||||
t1 = gen_reg_rtx (V2SImode);
|
||||
emit_insn (gen_subv2si3 (t1, op0, op1));
|
||||
|
||||
/* Extract the original sign bit of op0. */
|
||||
mask = GEN_INT (-0x80000000);
|
||||
/* Subtract (-(INT MAX) - 1) from both operands to make
|
||||
them signed. */
|
||||
mask = GEN_INT (0x80000000);
|
||||
mask = gen_rtx_CONST_VECTOR (V2SImode, gen_rtvec (2, mask, mask));
|
||||
mask = force_reg (V2SImode, mask);
|
||||
t2 = gen_reg_rtx (V2SImode);
|
||||
emit_insn (gen_andv2si3 (t2, op0, mask));
|
||||
|
||||
/* XOR it back into the result of the subtraction. This results
|
||||
in the sign bit set iff we saw unsigned underflow. */
|
||||
x = gen_reg_rtx (V2SImode);
|
||||
emit_insn (gen_xorv2si3 (x, t1, t2));
|
||||
|
||||
mask = force_reg (mode, mask);
|
||||
t1 = gen_reg_rtx (mode);
|
||||
emit_insn (gen_subv2si3 (t1, op0, mask));
|
||||
t2 = gen_reg_rtx (mode);
|
||||
emit_insn (gen_subv2si3 (t2, op1, mask));
|
||||
op0 = t1;
|
||||
op1 = t2;
|
||||
code = GT;
|
||||
op0 = x;
|
||||
op1 = CONST0_RTX (mode);
|
||||
}
|
||||
break;
|
||||
|
||||
|
|
|
|||
|
|
@ -1,3 +1,13 @@
|
|||
2010-01-17 H.J. Lu <hongjiu.lu@intel.com>
|
||||
|
||||
Backport from mainline:
|
||||
2010-01-13 Steve Ellcey <sje@cup.hp.com>
|
||||
|
||||
PR target/42542
|
||||
* gcc.target/ia64/pr42542-1.c: New.
|
||||
* gcc.target/ia64/pr42542-2.c: New.
|
||||
* gcc.target/ia64/pr42542-3.c: New.
|
||||
|
||||
2010-01-17 Richard Guenther <rguenther@suse.de>
|
||||
|
||||
PR tree-optimization/42773
|
||||
|
|
|
|||
|
|
@ -0,0 +1,68 @@
|
|||
/* { dg-do run } */
|
||||
/* { dg-options "-O1 -ftree-vectorize" } */
|
||||
|
||||
unsigned int v1[] __attribute__ ((aligned(16))) =
|
||||
{
|
||||
0x80000000, 1, 0xa0000000, 2,
|
||||
3, 0xd0000000, 0xf0000000, 0xe0000000
|
||||
};
|
||||
unsigned int v2[] __attribute__ ((aligned(16))) =
|
||||
{
|
||||
4, 0xb0000000, 5, 0xc0000000,
|
||||
0xd0000000, 6, 7, 8
|
||||
};
|
||||
|
||||
unsigned int max[] =
|
||||
{
|
||||
0x80000000, 0xb0000000, 0xa0000000, 0xc0000000,
|
||||
0xd0000000, 0xd0000000, 0xf0000000, 0xe0000000
|
||||
};
|
||||
|
||||
unsigned int min[] =
|
||||
{
|
||||
4, 1, 5, 2,
|
||||
3, 6, 7, 8
|
||||
};
|
||||
|
||||
unsigned int res[8] __attribute__ ((aligned(16)));
|
||||
|
||||
extern void abort (void);
|
||||
|
||||
void
|
||||
find_max (void)
|
||||
{
|
||||
int i;
|
||||
|
||||
for (i = 0; i < 8; i++)
|
||||
res[i] = v1[i] < v2[i] ? v2[i] : v1[i];
|
||||
}
|
||||
|
||||
void
|
||||
find_min (void)
|
||||
{
|
||||
int i;
|
||||
|
||||
for (i = 0; i < 8; i++)
|
||||
res[i] = v1[i] > v2[i] ? v2[i] : v1[i];
|
||||
}
|
||||
|
||||
int main (void)
|
||||
{
|
||||
int i;
|
||||
int err = 0;
|
||||
|
||||
find_max ();
|
||||
for (i = 0; i < 8; i++)
|
||||
if (res[i] != max[i])
|
||||
err++;
|
||||
|
||||
find_min ();
|
||||
for (i = 0; i < 8; i++)
|
||||
if (res[i] != min[i])
|
||||
err++;
|
||||
|
||||
if (err)
|
||||
abort ();
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
|
@ -0,0 +1,68 @@
|
|||
/* { dg-do run } */
|
||||
/* { dg-options "-O1 -ftree-vectorize" } */
|
||||
|
||||
unsigned short v1[] __attribute__ ((aligned(16))) =
|
||||
{
|
||||
0x8000, 0x9000, 1, 10, 0xa000, 0xb000, 2, 20,
|
||||
3, 30, 0xd000, 0xe000, 0xf000, 0xe000, 25, 30
|
||||
};
|
||||
unsigned short v2[] __attribute__ ((aligned(16))) =
|
||||
{
|
||||
4, 40, 0xb000, 0x8000, 5, 50, 0xc000, 0xf000,
|
||||
0xd000, 0xa000, 6, 65, 7, 75, 0xe000, 0xc000
|
||||
};
|
||||
|
||||
unsigned short max[] =
|
||||
{
|
||||
0x8000, 0x9000, 0xb000, 0x8000, 0xa000, 0xb000, 0xc000, 0xf000,
|
||||
0xd000, 0xa000, 0xd000, 0xe000, 0xf000, 0xe000, 0xe000, 0xc000
|
||||
};
|
||||
|
||||
unsigned short min[] =
|
||||
{
|
||||
4, 40, 1, 10, 5, 50, 2, 20,
|
||||
3, 30, 6, 65, 7, 75, 25, 30
|
||||
};
|
||||
|
||||
unsigned short res[16] __attribute__ ((aligned(16)));
|
||||
|
||||
extern void abort (void);
|
||||
|
||||
void
|
||||
find_max (void)
|
||||
{
|
||||
int i;
|
||||
|
||||
for (i = 0; i < 16; i++)
|
||||
res[i] = v1[i] < v2[i] ? v2[i] : v1[i];
|
||||
}
|
||||
|
||||
void
|
||||
find_min (void)
|
||||
{
|
||||
int i;
|
||||
|
||||
for (i = 0; i < 16; i++)
|
||||
res[i] = v1[i] > v2[i] ? v2[i] : v1[i];
|
||||
}
|
||||
|
||||
int main (void)
|
||||
{
|
||||
int i;
|
||||
int err = 0;
|
||||
|
||||
find_max ();
|
||||
for (i = 0; i < 16; i++)
|
||||
if (res[i] != max[i])
|
||||
err++;
|
||||
|
||||
find_min ();
|
||||
for (i = 0; i < 16; i++)
|
||||
if (res[i] != min[i])
|
||||
err++;
|
||||
|
||||
if (err)
|
||||
abort ();
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
|
@ -0,0 +1,76 @@
|
|||
/* { dg-do run } */
|
||||
/* { dg-options "-O1 -ftree-vectorize" } */
|
||||
|
||||
unsigned char v1[] __attribute__ ((aligned(16))) =
|
||||
{
|
||||
0x80, 0xd0, 0x90, 0xa0, 1, 15, 10, 15,
|
||||
0xa0, 0xc0, 0xb0, 0xf0, 2, 25, 20, 35,
|
||||
3, 34, 30, 36, 0xd0, 0x80, 0xe0, 0xb0,
|
||||
0xf0, 0xe0, 0xe0, 0x80, 25, 34, 30, 40
|
||||
};
|
||||
unsigned char v2[] __attribute__ ((aligned(16))) =
|
||||
{
|
||||
4, 44, 40, 48, 0xb0, 0x80, 0x80, 0x90,
|
||||
5, 55, 50, 51, 0xc0, 0xb0, 0xf0, 0xd0,
|
||||
0xd0, 0x80, 0xa0, 0xf0, 6, 61, 65, 68,
|
||||
7, 76, 75, 81, 0xe0, 0xf0, 0xc0, 0x90
|
||||
};
|
||||
|
||||
unsigned char max[] =
|
||||
{
|
||||
0x80, 0xd0, 0x90, 0xa0, 0xb0, 0x80, 0x80, 0x90,
|
||||
0xa0, 0xc0, 0xb0, 0xf0, 0xc0, 0xb0, 0xf0, 0xd0,
|
||||
0xd0, 0x80, 0xa0, 0xf0, 0xd0, 0x80, 0xe0, 0xb0,
|
||||
0xf0, 0xe0, 0xe0, 0x80, 0xe0, 0xf0, 0xc0, 0x90
|
||||
};
|
||||
|
||||
unsigned char min[] =
|
||||
{
|
||||
4, 44, 40, 48, 1, 15, 10, 15,
|
||||
5, 55, 50, 51, 2, 25, 20, 35,
|
||||
3, 34, 30, 36, 6, 61, 65, 68,
|
||||
7, 76, 75, 81, 25, 34, 30, 40
|
||||
};
|
||||
|
||||
unsigned char res[32] __attribute__ ((aligned(16)));
|
||||
|
||||
extern void abort (void);
|
||||
|
||||
void
|
||||
find_max (void)
|
||||
{
|
||||
int i;
|
||||
|
||||
for (i = 0; i < 32; i++)
|
||||
res[i] = v1[i] < v2[i] ? v2[i] : v1[i];
|
||||
}
|
||||
|
||||
void
|
||||
find_min (void)
|
||||
{
|
||||
int i;
|
||||
|
||||
for (i = 0; i < 32; i++)
|
||||
res[i] = v1[i] > v2[i] ? v2[i] : v1[i];
|
||||
}
|
||||
|
||||
int main (void)
|
||||
{
|
||||
int i;
|
||||
int err = 0;
|
||||
|
||||
find_max ();
|
||||
for (i = 0; i < 32; i++)
|
||||
if (res[i] != max[i])
|
||||
err++;
|
||||
|
||||
find_min ();
|
||||
for (i = 0; i < 32; i++)
|
||||
if (res[i] != min[i])
|
||||
err++;
|
||||
|
||||
if (err)
|
||||
abort ();
|
||||
|
||||
return 0;
|
||||
}
|
||||
Loading…
Reference in New Issue