Hello,
In late 2008 there was a post about a wrong-code bug on hardwarebug.org
(http://hardwarebug.org/2008/11/28/codesourcery-fails-again/#more-83). There
was this test case, and it was not vectorized correctly.
-----------------------
extern unsigned char dst[512] __attribute__((aligned(8)));
extern unsigned char src[512] __attribute__((aligned(8)));
void array_shift(void)
{
int i;
for (i = 0; i < 512; i++)
dst[i] = src[i] >> 7;
}
-------------------
With GCC trunk of today, vectorization doesn't happen at all:
t.c:8: note: type of def: 3.
t.c:8: note: vect_is_simple_use: operand 7
t.c:8: note: vector/scalar shift/rotate found.
t.c:8: note: not worthwhile without SIMD support.
t.c:8: note: not vectorized: relevant stmt not supported: D.1963_4 = D.1962_3
>> 7;
t.c:8: note: bad operation or unsupported loop bound.
t.c:5: note: vectorized 0 loops in function.
Compiler is trunk r156595 for arm-elf.
Compiler options: "-mcpu=cortex-a8 -mfpu=neon -O3 -fdump-tree-vect-details"
--
Summary: Missed vectorization on ARM NEON
Product: gcc
Version: 4.5.0
Status: UNCONFIRMED
Keywords: missed-optimization
Severity: normal
Priority: P3
Component: tree-optimization
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: steven at gcc dot gnu dot org
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43001
