Tell the compiler that v4f is aligned

kleisauke · kleisauke · commit e7faebf6af6b · 2021-03-20T11:33:37.000+01:00
It's guaranteed that this is now aligned on a 16-byte boundary.
diff --git a/configure.ac b/configure.ac
@@ -272,7 +272,7 @@ AM_GLIB_GNU_GETTEXT
 #   [ax_gcc_version_option=yes],
 #   [ax_gcc_version_option=no]
 # )
-AC_MSG_CHECKING([for gcc version])
+AC_MSG_CHECKING([for $CC version])
 GCC_VERSION=""
 version=$($CC -dumpversion)
 if test $? = 0; then 
@@ -326,7 +326,7 @@ AC_TYPE_SIZE_T
 
 # g++/gcc 4.x and 5.x have rather broken vector support ... 5.4.1 seems to
 # work, but 5.4.0 fails to even compile
-AC_MSG_CHECKING([for gcc with working vector support])
+AC_MSG_CHECKING([for $CC with working vector support])
 if test x"$GCC_VERSION_MAJOR" != x"4" -a x"$GCC_VERSION_MAJOR" != x"5"; then
   AC_MSG_RESULT([yes])
 else
@@ -339,7 +339,7 @@ if test x"$ax_cv_have_var_attribute_vector_size" = x"yes"; then
   AC_MSG_CHECKING([for C++ vector shuffle])
   AC_LANG_PUSH([C++])
   AC_TRY_COMPILE([
-    typedef float v4f __attribute__((vector_size(4 * sizeof(float))));
+    typedef float v4f __attribute__((vector_size(4 * sizeof(float)),aligned(16)));
   ],[
     v4f f; f[3] = 99;
   ],[
@@ -362,7 +362,7 @@ if test x"$have_vector_shuffle" = x"yes"; then
   AC_MSG_CHECKING([for C++ vector arithmetic])
   AC_LANG_PUSH([C++])
   AC_TRY_COMPILE([
-    typedef float v4f __attribute__((vector_size(4 * sizeof(float))));
+    typedef float v4f __attribute__((vector_size(4 * sizeof(float)),aligned(16)));
   ],[
     v4f f = {1, 2, 3, 4}; f *= 12.0; 
     v4f g = {5, 6, 7, 8}; f = g > 0 ? g : -1 * g;
@@ -382,7 +382,7 @@ if test x"$have_vector_arith" = x"yes"; then
   AC_MSG_CHECKING([for C++ signed constants in vector templates])
   AC_LANG_PUSH([C++])
   AC_TRY_COMPILE([
-    typedef float v4f __attribute__((vector_size(4 * sizeof(float))));
+    typedef float v4f __attribute__((vector_size(4 * sizeof(float)),aligned(16)));
     template <typename T>
     static void
     h( v4f B )
diff --git a/libvips/conversion/composite.cpp b/libvips/conversion/composite.cpp
@@ -85,7 +85,7 @@
 #ifdef HAVE_VECTOR_ARITH
 /* A vector of four floats.
  */
-typedef float v4f __attribute__((vector_size(4 * sizeof(float))));
+typedef float v4f __attribute__((vector_size(4 * sizeof(float)),aligned(16)));
 #endif /*HAVE_VECTOR_ARITH*/
 
 typedef struct _VipsCompositeBase {
