Revert "mmx: Use MMX2 intrinsics from xmmintrin.h directly."

This reverts commit 7de61d8d14e84623b6fa46506eb74f938287f536.

Newer versions of gcc allow inclusion of xmmintrin.h without -msse, but
still won't allow usage of the intrinsics.

Bugzilla: https://bugs.gentoo.org/show_bug.cgi?id=564024

2 files changed, 71 insertions, 8 deletions

diff --git a/configure.ac b/configure.ac
index 8370844..2178126 100644
--- a/configure.ac
+++ b/configure.ac
@@ -347,14 +347,21 @@ AC_COMPILE_IFELSE([AC_LANG_SOURCE([[
 #error "Need GCC >= 3.4 for MMX intrinsics"
 #endif
 #include <mmintrin.h>
-#include <xmmintrin.h>
 int main () {
     __m64 v = _mm_cvtsi32_si64 (1);
     __m64 w;
 
-    /* Test some intrinsics from xmmintrin.h */
-    w = _mm_shuffle_pi16(v, 5);
-    w = _mm_mulhi_pu16(w, w);
+    /* Some versions of clang will choke on K */
+    asm ("pshufw %2, %1, %0\n\t"
+        : "=y" (w)
+        : "y" (v), "K" (5)
+    );
+
+    /* Some versions of clang will choke on this */
+    asm ("pmulhuw %1, %0\n\t"
+	: "+y" (w)
+	: "y" (v)
+    );
 
     return _mm_cvtsi64_si32 (v);
 }]])], have_mmx_intrinsics=yes)
diff --git a/pixman/pixman-mmx.c b/pixman/pixman-mmx.c
index 88c3a39..05c48a4 100644
--- a/pixman/pixman-mmx.c
+++ b/pixman/pixman-mmx.c
@@ -40,9 +40,6 @@
 #else
 #include <mmintrin.h>
 #endif
-#ifdef USE_X86_MMX
-#include <xmmintrin.h>
-#endif
 #include "pixman-private.h"
 #include "pixman-combine32.h"
 #include "pixman-inlines.h"
@@ -62,7 +59,66 @@ _mm_empty (void)
 }
 #endif
 
-#ifndef _MM_SHUFFLE
+#ifdef USE_X86_MMX
+# if (defined(__SUNPRO_C) || defined(_MSC_VER) || defined(_WIN64))
+#  include <xmmintrin.h>
+# else
+/* We have to compile with -msse to use xmmintrin.h, but that causes SSE
+ * instructions to be generated that we don't want. Just duplicate the
+ * functions we want to use.  */
+extern __inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_movemask_pi8 (__m64 __A)
+{
+    int ret;
+
+    asm ("pmovmskb %1, %0\n\t"
+	: "=r" (ret)
+	: "y" (__A)
+    );
+
+    return ret;
+}
+
+extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_mulhi_pu16 (__m64 __A, __m64 __B)
+{
+    asm ("pmulhuw %1, %0\n\t"
+	: "+y" (__A)
+	: "y" (__B)
+    );
+    return __A;
+}
+
+#  ifdef __OPTIMIZE__
+extern __inline __m64 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_shuffle_pi16 (__m64 __A, int8_t const __N)
+{
+    __m64 ret;
+
+    asm ("pshufw %2, %1, %0\n\t"
+	: "=y" (ret)
+	: "y" (__A), "K" (__N)
+    );
+
+    return ret;
+}
+#  else
+#   define _mm_shuffle_pi16(A, N)					\
+    ({									\
+	__m64 ret;							\
+									\
+	asm ("pshufw %2, %1, %0\n\t"					\
+	     : "=y" (ret)						\
+	     : "y" (A), "K" ((const int8_t)N)				\
+	);								\
+									\
+	ret;								\
+    })
+#  endif
+# endif
+#endif
+
+#ifndef _MSC_VER
 #define _MM_SHUFFLE(fp3,fp2,fp1,fp0) \
  (((fp3) << 6) | ((fp2) << 4) | ((fp1) << 2) | (fp0))
 #endif