x86: Add asm for mbtree fixed point conversion

[x264] / common / cpu.h

diff --git a/common/cpu.h b/common/cpu.h
index 6a669735f3e72ef23172fc275b327e01b0f758e6..138141946fc4ff6929a5aa7eed29bb95643478be 100644 (file)
--- a/common/cpu.h
+++ b/common/cpu.h
@@ -1,10 +1,9 @@
 /*****************************************************************************
- * cpu.h: h264 encoder library
+ * cpu.h: cpu detection
  *****************************************************************************
- * Copyright (C) 2003 Laurent Aimar
- * $Id: cpu.h,v 1.1 2004/06/03 19:27:06 fenrir Exp $
+ * Copyright (C) 2004-2016 x264 project
  *
- * Authors: Laurent Aimar <fenrir@via.ecp.fr>
+ * Authors: Loren Merritt <lorenm@u.washington.edu>
  *
  * This program is free software; you can redistribute it and/or modify
  * it under the terms of the GNU General Public License as published by
@@ -18,7 +17,10 @@
  *
  * You should have received a copy of the GNU General Public License
  * along with this program; if not, write to the Free Software
- * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA  02111, USA.
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
+ *
+ * This program is also available under a commercial proprietary license.
+ * For more information, contact us at licensing@x264.com.
  *****************************************************************************/
 
 #ifndef X264_CPU_H
@@ -26,25 +28,46 @@
 
 uint32_t x264_cpu_detect( void );
 int      x264_cpu_num_processors( void );
-void     x264_emms( void );
+void     x264_cpu_emms( void );
+void     x264_cpu_sfence( void );
+#if HAVE_MMX
+/* There is no way to forbid the compiler from using float instructions
+ * before the emms so miscompilation could theoretically occur in the
+ * unlikely event that the compiler reorders emms and float instructions. */
+#if HAVE_X86_INLINE_ASM
+/* Clobbering memory makes the compiler less likely to reorder code. */
+#define x264_emms() asm volatile( "emms":::"memory","st","st(1)","st(2)", \
+                                  "st(3)","st(4)","st(5)","st(6)","st(7)" )
+#else
+#define x264_emms() x264_cpu_emms()
+#endif
+#else
+#define x264_emms()
+#endif
+#define x264_sfence x264_cpu_sfence
 
-/* kluge:
+/* kludge:
  * gcc can't give variables any greater alignment than the stack frame has.
- * We need 16 byte alignment for SSE2, so here we make sure that the stack is
- * aligned to 16 bytes.
+ * We need 32 byte alignment for AVX2, so here we make sure that the stack is
+ * aligned to 32 bytes.
  * gcc 4.2 introduced __attribute__((force_align_arg_pointer)) to fix this
  * problem, but I don't want to require such a new version.
- * This applies only to x86_32, since other architectures that need alignment
- * also have ABIs that ensure aligned stack. */
-#if defined(ARCH_X86) && defined(HAVE_MMX)
-void x264_stack_align( void (*func)(x264_t*), x264_t *arg );
+ * aligning to 32 bytes only works if the compiler supports keeping that
+ * alignment between functions (osdep.h handles manual alignment of arrays
+ * if it doesn't).
+ */
+#if (ARCH_X86 || STACK_ALIGNMENT > 16) && HAVE_MMX
+intptr_t x264_stack_align( void (*func)(), ... );
+#define x264_stack_align(func,...) x264_stack_align((void (*)())func, __VA_ARGS__)
 #else
-#define x264_stack_align(func,arg) func(arg)
+#define x264_stack_align(func,...) func(__VA_ARGS__)
 #endif
 
-extern const struct {
-    const char name[8];
-    int flags;
-} x264_cpu_names[];
+typedef struct
+{
+    const char name[16];
+    uint32_t flags;
+} x264_cpu_name_t;
+extern const x264_cpu_name_t x264_cpu_names[];
 
 #endif