x86: xvid_idct: port MMX iDCT to yasm

Also reduce the table duplication with SSE2 code, remove duplicated
macro parameters.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

libavcodec/x86/Makefile

@@ -73,7 +73,6 @@ MMX-OBJS-$(CONFIG_FDCTDSP)             += x86/fdct.o
 MMX-OBJS-$(CONFIG_IDCTDSP)             += x86/simple_idct.o
 
 # decoders/encoders
-MMX-OBJS-$(CONFIG_MPEG4_DECODER)       += x86/xvididct_mmx.o
 MMX-OBJS-$(CONFIG_SNOW_DECODER)        += x86/snowdsp.o
 MMX-OBJS-$(CONFIG_SNOW_ENCODER)        += x86/snowdsp.o
 MMX-OBJS-$(CONFIG_VC1_DECODER)         += x86/vc1dsp_mmx.o

libavcodec/x86/dct-test.c

@@ -60,11 +60,9 @@ static const struct algo idct_tab_arch[] = {
 #if HAVE_MMX_INLINE
     { "SIMPLE-MMX",  ff_simple_idct_mmx,  FF_IDCT_PERM_SIMPLE, AV_CPU_FLAG_MMX },
 #endif
-#if CONFIG_MPEG4_DECODER
-#if HAVE_MMX_INLINE
+#if CONFIG_MPEG4_DECODER && HAVE_YASM
+#if ARCH_X86_32
     { "XVID-MMX",    ff_xvid_idct_mmx,    FF_IDCT_PERM_NONE,   AV_CPU_FLAG_MMX,    1 },
-#endif
-#if HAVE_MMXEXT_INLINE
     { "XVID-MMXEXT", ff_xvid_idct_mmxext, FF_IDCT_PERM_NONE,   AV_CPU_FLAG_MMXEXT, 1 },
 #endif
 #if HAVE_SSE2_EXTERNAL
@@ -73,7 +71,7 @@ static const struct algo idct_tab_arch[] = {
     { "PR-SSE2",     ff_prores_idct_put_10_sse2_wrap, FF_IDCT_PERM_TRANSPOSE, AV_CPU_FLAG_SSE2, 1 },
 #endif
 #endif
-#endif /* CONFIG_MPEG4_DECODER */
+#endif /* CONFIG_MPEG4_DECODER && HAVE_YASM */
     { 0 }
 };
 

libavcodec/x86/xvididct_init.c

@@ -38,6 +38,32 @@ static void xvid_idct_sse2_add(uint8_t *dest, int line_size, short *block)
     ff_add_pixels_clamped(block, dest, line_size);
 }
 
+#if ARCH_X86_32
+static void xvid_idct_mmx_put(uint8_t *dest, int line_size, short *block)
+{
+    ff_xvid_idct_mmx(block);
+    ff_put_pixels_clamped(block, dest, line_size);
+}
+
+static void xvid_idct_mmx_add(uint8_t *dest, int line_size, short *block)
+{
+    ff_xvid_idct_mmx(block);
+    ff_add_pixels_clamped(block, dest, line_size);
+}
+
+static void xvid_idct_mmxext_put(uint8_t *dest, int line_size, short *block)
+{
+    ff_xvid_idct_mmxext(block);
+    ff_put_pixels_clamped(block, dest, line_size);
+}
+
+static void xvid_idct_mmxext_add(uint8_t *dest, int line_size, short *block)
+{
+    ff_xvid_idct_mmxext(block);
+    ff_add_pixels_clamped(block, dest, line_size);
+}
+#endif
+
 av_cold void ff_xvid_idct_init_x86(IDCTDSPContext *c, AVCodecContext *avctx,
                                    unsigned high_bit_depth)
 {
@@ -48,19 +74,21 @@ av_cold void ff_xvid_idct_init_x86(IDCTDSPContext *c, AVCodecContext *avctx,
           avctx->idct_algo == FF_IDCT_XVID))
         return;
 
-    if (INLINE_MMX(cpu_flags)) {
-        c->idct_put  = ff_xvid_idct_mmx_put;
-        c->idct_add  = ff_xvid_idct_mmx_add;
+#if ARCH_X86_32
+    if (EXTERNAL_MMX(cpu_flags)) {
+        c->idct_put  = xvid_idct_mmx_put;
+        c->idct_add  = xvid_idct_mmx_add;
         c->idct      = ff_xvid_idct_mmx;
         c->perm_type = FF_IDCT_PERM_NONE;
     }
 
-    if (INLINE_MMXEXT(cpu_flags)) {
-        c->idct_put  = ff_xvid_idct_mmxext_put;
-        c->idct_add  = ff_xvid_idct_mmxext_add;
+    if (EXTERNAL_MMXEXT(cpu_flags)) {
+        c->idct_put  = xvid_idct_mmxext_put;
+        c->idct_add  = xvid_idct_mmxext_add;
         c->idct      = ff_xvid_idct_mmxext;
         c->perm_type = FF_IDCT_PERM_NONE;
     }
+#endif
 
     if (EXTERNAL_SSE2(cpu_flags)) {
         c->idct_put  = xvid_idct_sse2_put;