http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48781
Summary: gcc generate movdqa instructions on unaligned memory
address when using -mtune=native -march=native
Product: gcc
Version: 4.4.4
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
AssignedTo: unassig...@gcc.gnu.org
ReportedBy: tanzhan...@gmail.com
GCC generated the movdaq instructions (requires 16-byte aligned memory address)
on the following simple code.
__uint128_t *mem_pool;
void test(unsigned int addr, unsigned int* data) {
*(__uint128_t *)data = mem_pool[addr];
}
Both 'mem_pool' and 'data' are not 128-bit aligned.
0000000000000000 <_Z4testjPj>:
0: 89 ff mov %edi,%edi
2: 48 8b 05 00 00 00 00 mov 0(%rip),%rax # 9 <_Z4testjPj+0x9>
9: 48 c1 e7 04 shl $0x4,%rdi
d: 66 0f 6f 04 07 movdqa (%rdi,%rax,1),%xmm0
12: 66 0f 7f 06 movdqa %xmm0,(%rsi)
16: c3 retq
I compiled this code with the following command:
g++44 -c test.cpp -O3 -mtune=native -march=native -o test.o
My target machine is Intel Xeon 5150. It works if don't use -mtune and -march.
The gcc is shipped with RHEL/Centos 5.6.
Using built-in specs.
Target: x86_64-redhat-linux6E
Configured with: ../configure --prefix=/usr --mandir=/usr/share/man
--infodir=/usr/share/info --with-bugurl=http://bugzilla.redhat.com/bugzilla
--enable-bootstrap --enable-shared --enable-threads=posix
--enable-checking=release --with-system-zlib --enable-__cxa_atexit
--disable-libunwind-exceptions --disable-gnu-unique-object
--enable-languages=c,c++,fortran --disable-libgcj
--with-mpfr=/builddir/build/BUILD/gcc-4.4.4-20100726/obj-x86_64-redhat-linux6E/mpfr-install/
--with-ppl=/builddir/build/BUILD/gcc-4.4.4-20100726/obj-x86_64-redhat-linux6E/ppl-install
--with-cloog=/builddir/build/BUILD/gcc-4.4.4-20100726/obj-x86_64-redhat-linux6E/cloog-install
--with-tune=generic --with-arch_32=i586 --build=x86_64-redhat-linux6E
Thread model: posix
gcc version 4.4.4 20100726 (Red Hat 4.4.4-13) (GCC)
Btw, if this is not correct, what is the right attribute to decorate a
uint128_t pointer in order to make use of 16-bit aligned SSE2 instructions? The
__atrribute__((align(x))) only works for variables, but not the value pointed
by pointers?
