This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
RE: PATCH: testsuite/37074: gcc.dg/torture/stackalign/builtin-apply-4.c failed with SSE2
- From: "Ye, Joey" <joey dot ye at intel dot com>
- To: "H.J. Lu" <hjl dot tools at gmail dot com>
- Cc: "Uros Bizjak" <ubizjak at gmail dot com>, "Andrew Pinski" <pinskia at gmail dot com>, <gcc-patches at gcc dot gnu dot org>
- Date: Wed, 13 Aug 2008 11:16:46 +0800
- Subject: RE: PATCH: testsuite/37074: gcc.dg/torture/stackalign/builtin-apply-4.c failed with SSE2
HJ wrote:
>On Mon, Aug 11, 2008 at 11:37 PM, Ye, Joey <joey.ye@intel.com> wrote:
>> Em, it won't violate the rules as long as EMMS is executed before
>> function entry and exit, right?
>>
>
> Can you give it a try?
Following patch fixes builtin-apply-4.c. But the changes in i386.md is
ugly and inefficient, which insert EMMS after every mmx movd insn. Any
suggestion?
- Joey
Index: gcc/builtins.c
===================================================================
--- gcc/builtins.c (revision 138935)
+++ gcc/builtins.c (working copy)
@@ -1589,6 +1589,7 @@ expand_builtin_return (rtx result)
enum machine_mode mode;
rtx reg;
rtx call_fusage = 0;
+ int size_array[FIRST_PSEUDO_REGISTER];
result = convert_memory_address (Pmode, result);
@@ -1604,7 +1605,7 @@ expand_builtin_return (rtx result)
}
#endif
- /* Restore the return value and note that each value is used. */
+ /* Calculate offset of return value and stores in array for later
use. */
size = 0;
for (regno = 0; regno < FIRST_PSEUDO_REGISTER; regno++)
if ((mode = apply_result_mode[regno]) != VOIDmode)
@@ -1612,14 +1613,25 @@ expand_builtin_return (rtx result)
align = GET_MODE_ALIGNMENT (mode) / BITS_PER_UNIT;
if (size % align != 0)
size = CEIL (size, align) * align;
+ size_array[regno] = size;
+ size += GET_MODE_SIZE (mode);
+ }
+ else
+ size_array[regno] = -1;
+
+ /* Restore the return value and note that each value is used.
+ Insns are emitted in the reversed sequence to solve MMX messing up
+ FPU problem in IA32 backend. */
+ for (regno = FIRST_PSEUDO_REGISTER - 1; regno >=0; regno--)
+ if ((mode = apply_result_mode[regno]) != VOIDmode)
+ {
reg = gen_rtx_REG (mode, INCOMING_REGNO (regno));
- emit_move_insn (reg, adjust_address (result, mode, size));
+ emit_move_insn (reg, adjust_address (result, mode,
size_array[regno]));
push_to_sequence (call_fusage);
emit_use (reg);
call_fusage = get_insns ();
end_sequence ();
- size += GET_MODE_SIZE (mode);
}
/* Put the USE insns before the return. */
Index: gcc/config/i386/i386.md
===================================================================
--- gcc/config/i386/i386.md (revision 138935)
+++ gcc/config/i386/i386.md (working copy)
@@ -2223,8 +2223,8 @@
#
#
pxor\t%0, %0
- movq\t{%1, %0|%0, %1}
- movq\t{%1, %0|%0, %1}
+ movq\t{%1, %0|%0, %1}\n\temms
+ movq\t{%1, %0|%0, %1}\n\temms
pxor\t%0, %0
movq\t{%1, %0|%0, %1}
movdqa\t{%1, %0|%0, %1}