On Fri, 28 Jun 2013, Jakub Jelinek wrote:
Hi!
On Thu, Jun 27, 2013 at 01:59:37PM +0200, Marc Glisse wrote:
--- testsuite/g++.dg/ext/pr57509.C (revision 0)
+++ testsuite/g++.dg/ext/pr57509.C (revision 0)
@@ -0,0 +1,16 @@
+/* { dg-do compile } */
+/* { dg-options "-std=c++11" } */
+
+template <bool> struct enable_if {};
+template <> struct enable_if<true> {typedef void type;};
+template <class T> void f (T& v) { v = __builtin_shuffle (v, v); }
+template <class T> void g (T) {}
+template <class T> auto g (T x) -> typename
enable_if<sizeof(__builtin_shuffle(x,x))!=2>::type {}
+typedef int v4i __attribute__((vector_size(4*sizeof(int))));
+typedef float v4f __attribute__((vector_size(4*sizeof(float))));
+int main(){
+ v4i a = {1,2,3,0};
+ f(a);
+ v4f b = {1,2,3,0};
+ g(b);
+}
Note this testcase fails on i686-linux:
/usr/src/gcc/gcc/testsuite/g++.dg/ext/pr57509.C: In function 'int main()':
/usr/src/gcc/gcc/testsuite/g++.dg/ext/pr57509.C:15:7: warning: SSE vector
argument without SSE enabled changes the ABI [enabled by default]
/usr/src/gcc/gcc/testsuite/g++.dg/ext/pr57509.C:15:7: warning: SSE vector
argument without SSE enabled changes the ABI [enabled by default]
/usr/src/gcc/gcc/testsuite/g++.dg/ext/pr57509.C:15:7: note: The ABI for passing
parameters with 16-byte alignment has changed in GCC 4.6
Gah, I know about this warning (that's why I pass vectors by pointer in
most testcases) but somehow keep forgetting it :-(
The note is actually pruned, but the warnings aren't. -Wno-abi -Wno-psabi
doesn't help,
I think it is a bug that the warning doesn't have an associated -Wxxx.
Should it use -Wabi, -Wpsabi, or some new flag like -Wunsupported-vector?
but -w does, so I'd suggest just to add -w to dg-options.
It seems better to do the following, so we still test for extra warnings:
-template <class T> void g (T) {}
-template <class T> auto g (T x) -> typename
enable_if<sizeof(__builtin_shuffle(x,x))!=2>::type {}
+template <class T> void g (T const&) {}
+template <class T> auto g (T const& x) -> typename
enable_if<sizeof(__builtin_shuffle(x,x))!=2>::type {}
Ok?
--
Marc Glisse
