procyberian/gcc
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

re PR rtl-optimization/67609 (Generates wrong code for SSE2 _mm_load_pd)

Browse source 
2015-10-20  Vladimir Makarov  <vmakarov@redhat.com>

	PR rtl-optimization/67609
	* lra-splill.c (lra_final_code_change): Don't remove all
	sub-registers.

2015-10-20  Vladimir Makarov  <vmakarov@redhat.com>

	PR rtl-optimization/67609
	* gcc.target/i386/pr67609.c: New.

From-SVN: r229087
This commit is contained in:
Vladimir Makarov 2015-10-20 16:26:05 +00:00 • committed by Vladimir Makarov
parent bfb792b604
commit 8cda1c3eb2
4 changed files with 63 additions and 8 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

6
gcc/ChangeLog
View file

@ -1,3 +1,9 @@
2015-10-20 Vladimir Makarov <vmakarov@redhat.com>
PR rtl-optimization/67609
* lra-splill.c (lra_final_code_change): Don't remove all
sub-registers.
2015-10-20 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
* simplify-rtx.c (simplify_binary_operation): If either operand was

46
gcc/lra-spills.c
View file

@ -727,14 +727,44 @@ lra_final_code_change (void)
lra_insn_recog_data_t id = lra_get_insn_recog_data (insn);
struct lra_static_insn_data *static_id = id->insn_static_data;
bool insn_change_p = false;
for (i = id->insn_static_data->n_operands - 1; i >= 0; i--)
if ((DEBUG_INSN_P (insn) || ! static_id->operand[i].is_operator)
&& alter_subregs (id->operand_loc[i], ! DEBUG_INSN_P (insn)))
{
lra_update_dup (id, i);
insn_change_p = true;
}
for (i = id->insn_static_data->n_operands - 1; i >= 0; i--)
{
if (! DEBUG_INSN_P (insn) && static_id->operand[i].is_operator)
continue;
rtx op = *id->operand_loc[i];
if (static_id->operand[i].type == OP_OUT
&& GET_CODE (op) == SUBREG && REG_P (SUBREG_REG (op))
&& ! LRA_SUBREG_P (op))
{
hard_regno = REGNO (SUBREG_REG (op));
/* We can not always remove sub-registers of
hard-registers as we may lose information that
only a part of registers is changed and
subsequent optimizations may do wrong
transformations (e.g. dead code eliminations).
We can not also keep all sub-registers as the
subsequent optimizations can not handle all such
cases. Here is a compromise which works. */
if ((GET_MODE_SIZE (GET_MODE (op))
< GET_MODE_SIZE (GET_MODE (SUBREG_REG (op))))
&& (hard_regno_nregs[hard_regno][GET_MODE (SUBREG_REG (op))]
== hard_regno_nregs[hard_regno][GET_MODE (op)])
#ifdef STACK_REGS
&& (hard_regno < FIRST_STACK_REG
|| hard_regno > LAST_STACK_REG)
#endif
)
continue;
}
if (alter_subregs (id->operand_loc[i], ! DEBUG_INSN_P (insn)))
{
lra_update_dup (id, i);
insn_change_p = true;
}
}
if (insn_change_p)
lra_update_operator_dups (id);
}

5
gcc/testsuite/ChangeLog
View file

@ -1,3 +1,8 @@
2015-10-20 Vladimir Makarov <vmakarov@redhat.com>
PR rtl-optimization/67609
* gcc.target/i386/pr67609.c: New.
2015-10-20 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
* gcc.target/aarch64/fmul_fcvt_1.c: Add multiply-by-32 cases.

14
gcc/testsuite/gcc.target/i386/pr67609.c Normal file
View file

@ -0,0 +1,14 @@
/* { dg-do compile } */
/* { dg-options "-O2 -msse2" } */
/* { dg-require-effective-target lp64 } */
/* { dg-final { scan-assembler "movdqa" } } */
#include <emmintrin.h>
__m128d reg;
void set_lower(double b)
{
double v[2];
_mm_store_pd(v, reg);
v[0] = b;
reg = _mm_load_pd(v);
}