workaround gcc's inability to align variables on the stack.

this crash was introduced in r642, but only because previous versions didn't use sse2 on the stack.


git-svn-id: svn://svn.videolan.org/x264/trunk@646 df754926-b1dd-0310-bc7b-ec298dee348c

common/cpu.h

@@ -30,4 +30,18 @@ int      x264_cpu_num_processors( void );
 /* probably MMX(EXT) centric but .... */
 void     x264_cpu_restore( uint32_t cpu );
 
+/* kluge:
+ * 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.
+ * 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. */
+#ifdef ARCH_X86
+void x264_stack_align( void (*func)(x264_t*), x264_t *arg );
+#else
+#define x264_stack_align(func,arg) func(arg)
+#endif
+
 #endif

common/i386/cpu-a.asm

@@ -101,3 +101,19 @@ cglobal x264_emms
     emms
     ret
 
+;-----------------------------------------------------------------------------
+; void x264_stack_align( void (*func)(void*), void *arg );
+;-----------------------------------------------------------------------------
+cglobal x264_stack_align
+    push ebp
+    mov  ebp, esp
+    sub  esp, 4
+    and  esp, ~15
+    mov  ecx, [ebp+8]
+    mov  edx, [ebp+12]
+    mov  [esp], edx
+    call ecx
+    mov  esp, ebp
+    pop  ebp
+    ret
+

encoder/encoder.c

@@ -1023,7 +1023,7 @@ static inline void x264_slice_init( x264_t *h, int i_nal_type, int i_global_qp )
     x264_macroblock_slice_init( h );
 }
 
-static int x264_slice_write( x264_t *h )
+static void x264_slice_write( x264_t *h )
 {
     int i_skip;
     int mb_xy;
@@ -1183,8 +1183,6 @@ static int x264_slice_write( x264_t *h )
                               - h->stat.frame.i_itex_bits
                               - h->stat.frame.i_ptex_bits
                               - h->stat.frame.i_hdr_bits;
-
-    return 0;
 }
 
 static void x264_thread_sync_context( x264_t *dst, x264_t *src )
@@ -1223,7 +1221,7 @@ static int x264_slices_write( x264_t *h )
         x264_visualize_init( h );
 #endif
 
-    x264_slice_write( h );
+    x264_stack_align( x264_slice_write, h );
     i_frame_size = h->out.nal[h->out.i_nal-1].i_payload;
     x264_fdec_filter_row( h, h->sps->i_mb_height );