------- Comment #3 from ubizjak at gmail dot com 2010-01-08 20:47 -------
(In reply to comment #0)
> The attached reduced test case derived from ac.f90 in the polyhedron benchmark
> suite gives and ICE when compiled with gfortran 4.5 using -O3 -mfma4. This ICE
> showed up in gcc 4.5 rev 154969 with the following check in:
>
> 2009-12-02 Richard Henderson <r...@redhat.com>
>
> * config/i386/i386.c (ix86_fixup_binary_operands): For FMA4, force
> all operands into registers.
Do we actually need this fixup after all input operands of FMA4 insn were
changed to "nonimmediate operand" in
2009-12-03 Sebastian Pop <sebastian....@amd.com>
Richard Henderson <r...@redhat.com>
* config/i386/i386-protos.h (ix86_fma4_valid_op_p): Remove.
* config/i386/i386.c (ix86_fma4_valid_op_p): Remove.
* config/i386/i386.md: Do not use ix86_fma4_valid_op_p.
* config/i386/sse.md (fma4_*): Remove alternative with operand 1
matching a memory access. Do not use ix86_fma4_valid_op_p.
(xop_*): Same.
Do not use ix86_fma4_valid_op_p in FMA4 and XOP splitters.
and
2009-12-07 Sebastian Pop <sebastian....@amd.com>
* config/i386/sse.md: Remove all FMA4 splitters.
Allow the second operand of FMA4 insns to be a nonimmediate.
Fix comments punctuation.
I'd say that after these commits, 2009-12-02 fixup is not needed anymore and
should be removed.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42664
