chromas: more SSE2/MMX fixes, added I420_RGBA conversion

diff --git a/modules/video_chroma/i420_rgb.c b/modules/video_chroma/i420_rgb.c
index ca772275e246fdcb1bcf258de8c41a6e6db7012c..e92035040913fa7531a1f4b4363e431fe8fb70ad 100644 (file)
--- a/modules/video_chroma/i420_rgb.c
+++ b/modules/video_chroma/i420_rgb.c
@@ -161,8 +161,7 @@ static int Activate( vlc_object_t *p_this )
                     {
                         /* R8G8B8A8 pixel format */
                         msg_Dbg(p_this, "RGB pixel format is R8G8B8A8");
-                        //p_vout->chroma.pf_convert = E_(I420_B8G8R8A8);
-                        return -1;
+                        p_vout->chroma.pf_convert = E_(I420_R8G8B8A8);
                     }
                     else if( p_vout->output.i_rmask == 0x0000ff00
                           && p_vout->output.i_gmask == 0x00ff0000

diff --git a/modules/video_chroma/i420_rgb.h b/modules/video_chroma/i420_rgb.h
index 1ba1c0de6fa311c2722c46ff6689a64a58a6b82e..5573ccee51b6194271c520726b283b94bd0f6f64 100644 (file)
--- a/modules/video_chroma/i420_rgb.h
+++ b/modules/video_chroma/i420_rgb.h
@@ -64,6 +64,7 @@ void E_(I420_RGB32)        ( vout_thread_t *, picture_t *, picture_t * );
 void E_(I420_R5G5B5)       ( vout_thread_t *, picture_t *, picture_t * );
 void E_(I420_R5G6B5)       ( vout_thread_t *, picture_t *, picture_t * );
 void E_(I420_A8R8G8B8)     ( vout_thread_t *, picture_t *, picture_t * );
+void E_(I420_R8G8B8A8)     ( vout_thread_t *, picture_t *, picture_t * );
 void E_(I420_B8G8R8A8)     ( vout_thread_t *, picture_t *, picture_t * );
 void E_(I420_A8B8G8R8)     ( vout_thread_t *, picture_t *, picture_t * );
 #endif

diff --git a/modules/video_chroma/i420_rgb16.c b/modules/video_chroma/i420_rgb16.c
index fc622de24f596b49aad13e99cd14e61d30137618..9f21869c73999b88d2e151cb1552615cf5e7b90f 100644 (file)
--- a/modules/video_chroma/i420_rgb16.c
+++ b/modules/video_chroma/i420_rgb16.c
@@ -1140,6 +1140,245 @@ void E_(I420_A8R8G8B8)( vout_thread_t *p_vout, picture_t *p_src,
 #endif
 }
 
+void E_(I420_R8G8B8A8)( vout_thread_t *p_vout, picture_t *p_src,
+                                            picture_t *p_dest )
+{
+    /* We got this one from the old arguments */
+    uint32_t *p_pic = (uint32_t*)p_dest->p->p_pixels;
+    uint8_t  *p_y   = p_src->Y_PIXELS;
+    uint8_t  *p_u   = p_src->U_PIXELS;
+    uint8_t  *p_v   = p_src->V_PIXELS;
+
+    vlc_bool_t  b_hscale;                         /* horizontal scaling type */
+    unsigned int i_vscale;                          /* vertical scaling type */
+    unsigned int i_x, i_y;                /* horizontal and vertical indexes */
+
+    int         i_right_margin;
+    int         i_rewind;
+    int         i_scale_count;                       /* scale modulo counter */
+    int         i_chroma_width = p_vout->render.i_width / 2; /* chroma width */
+    uint32_t *  p_pic_start;       /* beginning of the current line for copy */
+    /* Conversion buffer pointer */
+    uint32_t *  p_buffer_start = (uint32_t*)p_vout->chroma.p_sys->p_buffer;
+    uint32_t *  p_buffer;
+
+    /* Offset array pointer */
+    int *       p_offset_start = p_vout->chroma.p_sys->p_offset;
+    int *       p_offset;
+
+    const int i_source_margin = p_src->p[0].i_pitch
+                                 - p_src->p[0].i_visible_pitch;
+    const int i_source_margin_c = p_src->p[1].i_pitch
+                                 - p_src->p[1].i_visible_pitch;
+
+    i_right_margin = p_dest->p->i_pitch - p_dest->p->i_visible_pitch;
+
+    /* Rule: when a picture of size (x1,y1) with aspect ratio r1 is rendered
+     * on a picture of size (x2,y2) with aspect ratio r2, if x1 grows to x1'
+     * then y1 grows to y1' = x1' * y2/x2 * r2/r1 */
+    SetOffset( p_vout->render.i_width, p_vout->render.i_height,
+               p_vout->output.i_width, p_vout->output.i_height,
+               &b_hscale, &i_vscale, p_offset_start );
+
+    /*
+     * Perform conversion
+     */
+    i_scale_count = ( i_vscale == 1 ) ?
+                    p_vout->output.i_height : p_vout->render.i_height;
+
+#if defined (MODULE_NAME_IS_i420_rgb_sse2)
+
+    if( p_vout->render.i_width & 15 )
+    {
+        i_rewind = 16 - ( p_vout->render.i_width & 15 );
+    }
+    else
+    {
+        i_rewind = 0;
+    }
+
+    /*
+    ** SSE2 128 bits fetch/store instructions are faster 
+    ** if memory access is 16 bytes aligned
+    */
+
+    p_buffer = b_hscale ? p_buffer_start : p_pic;
+    if( 0 == (15 & (p_src->p[Y_PLANE].i_pitch|
+                    p_dest->p->i_pitch|
+                    ((int)p_y)|
+                    ((int)p_buffer))) )
+    {
+        /* use faster SSE2 aligned fetch and store */
+        for( i_y = 0; i_y < p_vout->render.i_height; i_y++ )
+        {
+            p_pic_start = p_pic;
+
+            for ( i_x = p_vout->render.i_width / 16; i_x--; )
+            {
+                SSE2_CALL (
+                    SSE2_INIT_32_ALIGNED
+                    SSE2_YUV_MUL
+                    SSE2_YUV_ADD
+                    SSE2_UNPACK_32_RGBA_ALIGNED
+                );
+                p_y += 16;
+                p_u += 8;
+                p_v += 8;
+                p_buffer += 16;
+            }
+
+            /* Here we do some unaligned reads and duplicate conversions, but
+             * at least we have all the pixels */
+            if( i_rewind )
+            {
+                p_y -= i_rewind;
+                p_u -= i_rewind >> 1;
+                p_v -= i_rewind >> 1;
+                p_buffer -= i_rewind;
+                SSE2_CALL (
+                    SSE2_INIT_32_UNALIGNED
+                    SSE2_YUV_MUL
+                    SSE2_YUV_ADD
+                    SSE2_UNPACK_32_RGBA_UNALIGNED
+                );
+                p_y += 16;
+                p_u += 4;
+                p_v += 4;
+            }
+            SCALE_WIDTH;
+            SCALE_HEIGHT( 420, 4 );
+
+            p_y += i_source_margin;
+            if( i_y % 2 )
+            {
+                p_u += i_source_margin_c;
+                p_v += i_source_margin_c;
+            }
+            p_buffer = b_hscale ? p_buffer_start : p_pic;
+        }
+    }
+    else
+    {
+        /* use slower SSE2 unaligned fetch and store */
+        for( i_y = 0; i_y < p_vout->render.i_height; i_y++ )
+        {
+            p_pic_start = p_pic;
+            p_buffer = b_hscale ? p_buffer_start : p_pic;
+
+            for ( i_x = p_vout->render.i_width / 16; i_x--; )
+            {
+                SSE2_CALL (
+                    SSE2_INIT_32_UNALIGNED
+                    SSE2_YUV_MUL
+                    SSE2_YUV_ADD
+                    SSE2_UNPACK_32_RGBA_UNALIGNED
+                );
+                p_y += 16;
+                p_u += 8;
+                p_v += 8;
+                p_buffer += 16;
+            }
+
+            /* Here we do some unaligned reads and duplicate conversions, but
+             * at least we have all the pixels */
+            if( i_rewind )
+            {
+                p_y -= i_rewind;
+                p_u -= i_rewind >> 1;
+                p_v -= i_rewind >> 1;
+                p_buffer -= i_rewind;
+                SSE2_CALL (
+                    SSE2_INIT_32_UNALIGNED
+                    SSE2_YUV_MUL
+                    SSE2_YUV_ADD
+                    SSE2_UNPACK_32_RGBA_UNALIGNED
+                );
+                p_y += 16;
+                p_u += 8;
+                p_v += 8;
+            }
+            SCALE_WIDTH;
+            SCALE_HEIGHT( 420, 4 );
+
+            p_y += i_source_margin;
+            if( i_y % 2 )
+            {
+                p_u += i_source_margin_c;
+                p_v += i_source_margin_c;
+            }
+            p_buffer = b_hscale ? p_buffer_start : p_pic;
+        }
+    }
+
+    /* make sure all SSE2 stores are visible thereafter */
+    SSE2_END;
+
+#else // defined (MODULE_NAME_IS_i420_rgb_mmx)
+
+    if( p_vout->render.i_width & 7 )
+    {
+        i_rewind = 8 - ( p_vout->render.i_width & 7 );
+    }
+    else
+    {
+        i_rewind = 0;
+    }
+
+    for( i_y = 0; i_y < p_vout->render.i_height; i_y++ )
+    {
+        p_pic_start = p_pic;
+        p_buffer = b_hscale ? p_buffer_start : p_pic;
+
+        for ( i_x = p_vout->render.i_width / 8; i_x--; )
+        {
+            MMX_CALL (
+                MMX_INIT_32
+                MMX_YUV_MUL
+                MMX_YUV_ADD
+                MMX_UNPACK_32_RGBA
+            );
+            p_y += 8;
+            p_u += 4;
+            p_v += 4;
+            p_buffer += 8;
+        }
+
+        /* Here we do some unaligned reads and duplicate conversions, but
+         * at least we have all the pixels */
+        if( i_rewind )
+        {
+            p_y -= i_rewind;
+            p_u -= i_rewind >> 1;
+            p_v -= i_rewind >> 1;
+            p_buffer -= i_rewind;
+            MMX_CALL (
+                MMX_INIT_32
+                MMX_YUV_MUL
+                MMX_YUV_ADD
+                MMX_UNPACK_32_RGBA
+            );
+            p_y += 8;
+            p_u += 4;
+            p_v += 4;
+            p_buffer += 8;
+        }
+        SCALE_WIDTH;
+        SCALE_HEIGHT( 420, 4 );
+
+        p_y += i_source_margin;
+        if( i_y % 2 )
+        {
+            p_u += i_source_margin_c;
+            p_v += i_source_margin_c;
+        }
+    }
+
+    /* re-enable FPU registers */
+    MMX_END;
+
+#endif
+}
+
 void E_(I420_B8G8R8A8)( vout_thread_t *p_vout, picture_t *p_src,
                                             picture_t *p_dest )
 {

diff --git a/modules/video_chroma/i420_rgb_mmx.h b/modules/video_chroma/i420_rgb_mmx.h
index 3200a1f33434c31a48bfc3b70503eb5e2b3fa79b..3ff27cb19283dea5ccf8080408375942fa3545a9 100644 (file)
--- a/modules/video_chroma/i420_rgb_mmx.h
+++ b/modules/video_chroma/i420_rgb_mmx.h
@@ -300,6 +300,26 @@ punpckhwd %%mm1, %%mm0  #                 00 R7 B7 G7 00 R6 B6 G6           \n\
 movq      %%mm0, 24(%3) # Store ARGB7 ARGB6                                 \n\
 "
 
+#define MMX_UNPACK_32_RGBA "                                                \n\
+pxor      %%mm3, %%mm3  # zero mm3                                          \n\
+movq      %%mm2, %%mm4  #                 G7 G6 G5 G4 G3 G2 G1 G0           \n\
+punpcklbw %%mm1, %%mm4  #                 R3 G3 R2 G2 R1 G1 R0 G0           \n\
+punpcklbw %%mm0, %%mm3  #                 B3 00 B2 00 B1 00 B0 00           \n\
+movq      %%mm3, %%mm5  #                 R3 00 R2 00 R1 00 R0 00           \n\
+punpcklwd %%mm4, %%mm3  #                 R1 G1 B1 00 R0 G0 B0 00           \n\
+movq      %%mm3, (%3)   # Store RGBA1 RGBA0                                 \n\
+punpckhwd %%mm4, %%mm5  #                 R3 G3 B3 00 R2 G2 B2 00           \n\
+movq      %%mm5, 8(%3)  # Store RGBA3 RGBA2                                 \n\
+pxor      %%mm6, %%mm6  # zero mm6                                          \n\
+punpckhbw %%mm1, %%mm2  #                 R7 G7 R6 G6 R5 G5 R4 G4           \n\
+punpckhbw %%mm0, %%mm6  #                 B7 00 B6 00 B5 00 B4 00           \n\
+movq      %%mm6, %%mm0  #                 B7 00 B6 00 B5 00 B4 00           \n\
+punpcklwd %%mm2, %%mm6  #                 R5 G5 B5 00 R4 G4 B4 00           \n\
+movq      %%mm6, 16(%3) # Store RGBA5 RGBA4                                 \n\
+punpckhwd %%mm2, %%mm0  #                 R7 G7 B7 00 R6 G6 B6 00           \n\
+movq      %%mm0, 24(%3) # Store RGBA7 RGBA6                                 \n\
+"
+
 #define MMX_UNPACK_32_BGRA "                                                \n\
 pxor      %%mm3, %%mm3  # zero mm3                                          \n\
 movq      %%mm2, %%mm4  #                 G7 G6 G5 G4 G3 G2 G1 G0           \n\
@@ -356,15 +376,15 @@ movq      %%mm2, 24(%3) # Store ABGR7 ABGR6                                 \n\
 #define MMX_END _mm_empty()
     
 #define MMX_INIT_16                     \
-    mm0 = _mm_cvtsi32_si64((int)*p_u);  \
-    mm1 = _mm_cvtsi32_si64((int)*p_v);  \
+    mm0 = _mm_cvtsi32_si64(*(int*)p_u); \
+    mm1 = _mm_cvtsi32_si64(*(int*)p_v); \
     mm4 = _mm_setzero_si64();           \
-    mm6 = (__m64)*(uint64_t *)p_y
+    mm6 = (__m64)*(uint64_t *)p_y;
 
 #define MMX_INIT_32                     \
-    mm0 = _mm_cvtsi32_si64((int)*p_u);  \
+    mm0 = _mm_cvtsi32_si64(*(int*)p_u); \
     *(uint16_t *)p_buffer = 0;          \
-    mm1 = _mm_cvtsi32_si64((int)*p_v);  \
+    mm1 = _mm_cvtsi32_si64(*(int*)p_v); \
     mm4 = _mm_setzero_si64();           \
     mm6 = (__m64)*(uint64_t *)p_y;
 
@@ -483,6 +503,25 @@ movq      %%mm2, 24(%3) # Store ABGR7 ABGR6                                 \n\
     mm0 = _mm_unpackhi_pi16(mm0, mm1);          \
     *(uint64_t *)(p_buffer + 6) = (uint64_t)mm0;
 
+#define MMX_UNPACK_32_RGBA                      \
+    mm3 = _mm_setzero_si64();                   \
+    mm4 = mm2;                                  \
+    mm4 = _mm_unpacklo_pi8(mm4, mm1);           \
+    mm3 = _mm_unpacklo_pi8(mm3, mm0);           \
+    mm5 = mm3;                                  \
+    mm3 = _mm_unpacklo_pi16(mm3, mm4);          \
+    *(uint64_t *)p_buffer = (uint64_t)mm3;      \
+    mm5 = _mm_unpackhi_pi16(mm5, mm4);          \
+    *(uint64_t *)(p_buffer + 2) = (uint64_t)mm5;\
+    mm6 = _mm_setzero_si64();                   \
+    mm2 = _mm_unpackhi_pi8(mm2, mm1);           \
+    mm6 = _mm_unpackhi_pi8(mm6, mm0);           \
+    mm0 = mm6;                                  \
+    mm6 = _mm_unpacklo_pi16(mm6, mm2);          \
+    *(uint64_t *)(p_buffer + 4) = (uint64_t)mm6;\
+    mm0 = _mm_unpackhi_pi16(mm0, mm2);          \
+    *(uint64_t *)(p_buffer + 6) = (uint64_t)mm0;
+
 #define MMX_UNPACK_32_BGRA                      \
     mm3 = _mm_setzero_si64();                   \
     mm4 = mm2;                                  \
@@ -503,7 +542,23 @@ movq      %%mm2, 24(%3) # Store ABGR7 ABGR6                                 \n\
     *(uint64_t *)(p_buffer + 6) = (uint64_t)mm0;
 
 #define MMX_UNPACK_32_ABGR                      \
-    ;
+    mm3 = _mm_setzero_si64();                   \
+    mm4 = mm1;                                  \
+    mm4 = _mm_unpacklo_pi8(mm4, mm2);           \
+    mm5 = mm0;                                  \
+    mm5 = _mm_unpacklo_pi8(mm5, mm3);           \
+    mm6 = mm4;                                  \
+    mm4 = _mm_unpacklo_pi16(mm4, mm5);          \
+    *(uint64_t *)p_buffer = (uint64_t)mm4;      \
+    mm6 = _mm_unpackhi_pi16(mm6, mm5);          \
+    *(uint64_t *)(p_buffer + 2) = (uint64_t)mm6;\
+    mm1 = _mm_unpackhi_pi8(mm1, mm2);           \
+    mm0 = _mm_unpackhi_pi8(mm0, mm3);           \
+    mm2 = mm1;                                  \
+    mm1 = _mm_unpacklo_pi16(mm1, mm0);          \
+    *(uint64_t *)(p_buffer + 4) = (uint64_t)mm1;\
+    mm2 = _mm_unpackhi_pi16(mm2, mm0);          \
+    *(uint64_t *)(p_buffer + 6) = (uint64_t)mm2;
 
 #endif
 
@@ -795,6 +850,46 @@ punpckhwd %%xmm1, %%xmm0  #               00 R7 B7 G7 00 R6 B6 G6           \n\
 movdqu    %%xmm0, 48(%3)  # Store ARGB15 ARGB14 ARGB13 ARGB12               \n\
 "
 
+#define SSE2_UNPACK_32_RGBA_ALIGNED "                                       \n\
+pxor      %%xmm3, %%xmm3  # zero mm3                                        \n\
+movdqa    %%xmm2, %%xmm4  #                 G7 G6 G5 G4 G3 G2 G1 G0         \n\
+punpcklbw %%xmm1, %%xmm4  #                 R3 G3 R2 G2 R1 G1 R0 G0         \n\
+punpcklbw %%xmm0, %%xmm3  #                 B3 00 B2 00 B1 00 B0 00         \n\
+movdqa    %%xmm3, %%xmm5  #                 R3 00 R2 00 R1 00 R0 00         \n\
+punpcklwd %%xmm4, %%xmm3  #                 R1 G1 B1 00 R0 B0 G0 00         \n\
+movntdq   %%xmm3, (%3)    # Store RGBA3 RGBA2 RGBA1 RGBA0                   \n\
+punpckhwd %%xmm4, %%xmm5  #                 R3 G3 B3 00 R2 G2 B2 00         \n\
+movntdq   %%xmm5, 16(%3)  # Store RGBA7 RGBA6 RGBA5 RGBA4                   \n\
+pxor      %%xmm6, %%xmm6  # zero mm6                                        \n\
+punpckhbw %%xmm1, %%xmm2  #                 R7 G7 R6 G6 R5 G5 R4 G4         \n\
+punpckhbw %%xmm0, %%xmm6  #                 B7 00 B6 00 B5 00 B4 00         \n\
+movdqa    %%xmm6, %%xmm0  #                 B7 00 B6 00 B5 00 B4 00         \n\
+punpcklwd %%xmm2, %%xmm6  #                 R5 G5 B5 00 R4 G4 B4 00         \n\
+movntdq   %%xmm6, 32(%3)  # Store BGRA11 BGRA10 BGRA9 RGBA8                 \n\
+punpckhwd %%xmm2, %%xmm0  #                 R7 G7 B7 00 R6 G6 B6 00         \n\
+movntdq   %%xmm0, 48(%3)  # Store RGBA15 RGBA14 RGBA13 RGBA12               \n\
+"
+
+#define SSE2_UNPACK_32_RGBA_UNALIGNED "                                     \n\
+pxor      %%xmm3, %%xmm3  # zero mm3                                        \n\
+movdqa    %%xmm2, %%xmm4  #                 G7 G6 G5 G4 G3 G2 G1 G0         \n\
+punpcklbw %%xmm1, %%xmm4  #                 R3 G3 R2 G2 R1 G1 R0 G0         \n\
+punpcklbw %%xmm0, %%xmm3  #                 B3 00 B2 00 B1 00 B0 00         \n\
+movdqa    %%xmm3, %%xmm5  #                 R3 00 R2 00 R1 00 R0 00         \n\
+punpcklwd %%xmm4, %%xmm3  #                 R1 G1 B1 00 R0 B0 G0 00         \n\
+movdqu    %%xmm3, (%3)    # Store RGBA3 RGBA2 RGBA1 RGBA0                   \n\
+punpckhwd %%xmm4, %%xmm5  #                 R3 G3 B3 00 R2 G2 B2 00         \n\
+movdqu    %%xmm5, 16(%3)  # Store RGBA7 RGBA6 RGBA5 RGBA4                   \n\
+pxor      %%xmm6, %%xmm6  # zero mm6                                        \n\
+punpckhbw %%xmm1, %%xmm2  #                 R7 G7 R6 G6 R5 G5 R4 G4         \n\
+punpckhbw %%xmm0, %%xmm6  #                 B7 00 B6 00 B5 00 B4 00         \n\
+movdqa    %%xmm6, %%xmm0  #                 B7 00 B6 00 B5 00 B4 00         \n\
+punpcklwd %%xmm2, %%xmm6  #                 R5 G5 B5 00 R4 G4 B4 00         \n\
+movdqu    %%xmm6, 32(%3)  # Store RGBA11 RGBA10 RGBA9 RGBA8                 \n\
+punpckhwd %%xmm2, %%xmm0  #                 R7 G7 B7 00 R6 G6 B6 00         \n\
+movdqu    %%xmm0, 48(%3)  # Store RGBA15 RGBA14 RGBA13 RGBA12               \n\
+"
+
 #define SSE2_UNPACK_32_BGRA_ALIGNED "                                       \n\
 pxor      %%xmm3, %%xmm3  # zero mm3                                        \n\
 movdqa    %%xmm2, %%xmm4  #                 G7 G6 G5 G4 G3 G2 G1 G0         \n\
@@ -881,11 +976,11 @@ movdqu    %%xmm2, 48(%3)  # Store ABGR15 ABGR14 ABGR13 ABGR12               \n\
 
 #include <emmintrin.h>
 
-#define SSE2_CALL(SSE2_INSTRUCTIONS)            \
-    do {                                        \
-        __m128i xmm0, xmm1, xmm2, xmm3,         \
-                xmm4, xmm5, xmm6, xmm7;         \
-        SSE2_INSTRUCTIONS                       \
+#define SSE2_CALL(SSE2_INSTRUCTIONS)        \
+    do {                                    \
+        __m128i xmm0, xmm1, xmm2, xmm3,     \
+                xmm4, xmm5, xmm6, xmm7;     \
+        SSE2_INSTRUCTIONS                   \
     } while(0)
 
 #define SSE2_END  _mm_sfence()
@@ -971,179 +1066,249 @@ movdqu    %%xmm2, 48(%3)  # Store ABGR15 ABGR14 ABGR13 ABGR12               \n\
     xmm1 = _mm_unpacklo_epi8(xmm1, xmm4);   \
     xmm2 = _mm_unpacklo_epi8(xmm2, xmm5);
 
-#define SSE2_UNPACK_15_ALIGNED              \
-    xmm5 = _mm_set1_epi32(0xf8f8f8f8UL);    \
-    xmm0 = _mm_and_si128(xmm0, xmm5);       \
-    xmm0 = _mm_srli_epi16(xmm0, 3);         \
-    xmm2 = _mm_and_si128(xmm2, xmm5);       \
-    xmm1 = _mm_and_si128(xmm1, xmm5);       \
-    xmm1 = _mm_srli_epi16(xmm1, 1);         \
-    xmm4 = _mm_setzero_si128();             \
-    xmm5 = xmm0;                            \
-    xmm7 = xmm2;                            \
+#define SSE2_UNPACK_15_ALIGNED                      \
+    xmm5 = _mm_set1_epi32(0xf8f8f8f8UL);            \
+    xmm0 = _mm_and_si128(xmm0, xmm5);               \
+    xmm0 = _mm_srli_epi16(xmm0, 3);                 \
+    xmm2 = _mm_and_si128(xmm2, xmm5);               \
+    xmm1 = _mm_and_si128(xmm1, xmm5);               \
+    xmm1 = _mm_srli_epi16(xmm1, 1);                 \
+    xmm4 = _mm_setzero_si128();                     \
+    xmm5 = xmm0;                                    \
+    xmm7 = xmm2;                                    \
     \
-    xmm2 = _mm_unpacklo_epi8(xmm2, xmm4);   \
-    xmm0 = _mm_unpacklo_epi8(xmm0, xmm1);   \
-    xmm2 = _mm_slli_epi16(xmm2, 2);         \
-    xmm0 = _mm_or_si128(xmm0, xmm2);        \
-    _mm_stream_si128((__m128i*)p_buffer, xmm0); \
+    xmm2 = _mm_unpacklo_epi8(xmm2, xmm4);           \
+    xmm0 = _mm_unpacklo_epi8(xmm0, xmm1);           \
+    xmm2 = _mm_slli_epi16(xmm2, 2);                 \
+    xmm0 = _mm_or_si128(xmm0, xmm2);                \
+    _mm_stream_si128((__m128i*)p_buffer, xmm0);     \
     \
-    xmm7 = _mm_unpackhi_epi8(xmm7, xmm4);   \
-    xmm5 = _mm_unpackhi_epi8(xmm5, xmm1);   \
-    xmm7 = _mm_slli_epi16(xmm7, 2);         \
-    xmm5 = _mm_or_si128(xmm5, xmm7);        \
+    xmm7 = _mm_unpackhi_epi8(xmm7, xmm4);           \
+    xmm5 = _mm_unpackhi_epi8(xmm5, xmm1);           \
+    xmm7 = _mm_slli_epi16(xmm7, 2);                 \
+    xmm5 = _mm_or_si128(xmm5, xmm7);                \
     _mm_stream_si128((__m128i*)(p_buffer+8), xmm5);
 
-#define SSE2_UNPACK_15_UNALIGNED            \
-    xmm5 = _mm_set1_epi32(0xf8f8f8f8UL);    \
-    xmm0 = _mm_and_si128(xmm0, xmm5);       \
-    xmm0 = _mm_srli_epi16(xmm0, 3);         \
-    xmm2 = _mm_and_si128(xmm2, xmm5);       \
-    xmm1 = _mm_and_si128(xmm1, xmm5);       \
-    xmm1 = _mm_srli_epi16(xmm1, 1);         \
-    xmm4 = _mm_setzero_si128();             \
-    xmm5 = xmm0;                            \
-    xmm7 = xmm2;                            \
+#define SSE2_UNPACK_15_UNALIGNED                    \
+    xmm5 = _mm_set1_epi32(0xf8f8f8f8UL);            \
+    xmm0 = _mm_and_si128(xmm0, xmm5);               \
+    xmm0 = _mm_srli_epi16(xmm0, 3);                 \
+    xmm2 = _mm_and_si128(xmm2, xmm5);               \
+    xmm1 = _mm_and_si128(xmm1, xmm5);               \
+    xmm1 = _mm_srli_epi16(xmm1, 1);                 \
+    xmm4 = _mm_setzero_si128();                     \
+    xmm5 = xmm0;                                    \
+    xmm7 = xmm2;                                    \
     \
-    xmm2 = _mm_unpacklo_epi8(xmm2, xmm4);   \
-    xmm0 = _mm_unpacklo_epi8(xmm0, xmm1);   \
-    xmm2 = _mm_slli_epi16(xmm2, 2);         \
-    xmm0 = _mm_or_si128(xmm0, xmm2);        \
-    _mm_storeu_si128((__m128i*)p_buffer, xmm0); \
+    xmm2 = _mm_unpacklo_epi8(xmm2, xmm4);           \
+    xmm0 = _mm_unpacklo_epi8(xmm0, xmm1);           \
+    xmm2 = _mm_slli_epi16(xmm2, 2);                 \
+    xmm0 = _mm_or_si128(xmm0, xmm2);                \
+    _mm_storeu_si128((__m128i*)p_buffer, xmm0);     \
     \
-    xmm7 = _mm_unpackhi_epi8(xmm7, xmm4);   \
-    xmm5 = _mm_unpackhi_epi8(xmm5, xmm1);   \
-    xmm7 = _mm_slli_epi16(xmm7, 2);         \
-    xmm5 = _mm_or_si128(xmm5, xmm7);        \
+    xmm7 = _mm_unpackhi_epi8(xmm7, xmm4);           \
+    xmm5 = _mm_unpackhi_epi8(xmm5, xmm1);           \
+    xmm7 = _mm_slli_epi16(xmm7, 2);                 \
+    xmm5 = _mm_or_si128(xmm5, xmm7);                \
     _mm_storeu_si128((__m128i*)(p_buffer+16), xmm5);
 
-#define SSE2_UNPACK_16_ALIGNED              \
-    xmm5 = _mm_set1_epi32(0xf8f8f8f8UL);    \
-    xmm0 = _mm_and_si128(xmm0, xmm5);       \
-    xmm1 = _mm_and_si128(xmm1, xmm5);       \
-    xmm5 = _mm_set1_epi32(0xfcfcfcfcUL);    \
-    xmm2 = _mm_and_si128(xmm2, xmm5);       \
-    xmm0 = _mm_srli_epi16(xmm0, 3);         \
-    xmm4 = _mm_setzero_si128();             \
-    xmm5 = xmm0;                            \
-    xmm7 = xmm2;                            \
+#define SSE2_UNPACK_16_ALIGNED                      \
+    xmm5 = _mm_set1_epi32(0xf8f8f8f8UL);            \
+    xmm0 = _mm_and_si128(xmm0, xmm5);               \
+    xmm1 = _mm_and_si128(xmm1, xmm5);               \
+    xmm5 = _mm_set1_epi32(0xfcfcfcfcUL);            \
+    xmm2 = _mm_and_si128(xmm2, xmm5);               \
+    xmm0 = _mm_srli_epi16(xmm0, 3);                 \
+    xmm4 = _mm_setzero_si128();                     \
+    xmm5 = xmm0;                                    \
+    xmm7 = xmm2;                                    \
     \
-    xmm2 = _mm_unpacklo_epi8(xmm2, xmm4);   \
-    xmm0 = _mm_unpacklo_epi8(xmm0, xmm1);   \
-    xmm2 = _mm_slli_epi16(xmm2, 3);         \
-    xmm0 = _mm_or_si128(xmm0, xmm2);        \
-    _mm_stream_si128((__m128i*)p_buffer, xmm0); \
+    xmm2 = _mm_unpacklo_epi8(xmm2, xmm4);           \
+    xmm0 = _mm_unpacklo_epi8(xmm0, xmm1);           \
+    xmm2 = _mm_slli_epi16(xmm2, 3);                 \
+    xmm0 = _mm_or_si128(xmm0, xmm2);                \
+    _mm_stream_si128((__m128i*)p_buffer, xmm0);     \
     \
-    xmm7 = _mm_unpackhi_epi8(xmm7, xmm4);   \
-    xmm5 = _mm_unpackhi_epi8(xmm5, xmm1);   \
-    xmm7 = _mm_slli_epi16(xmm7, 3);         \
-    xmm5 = _mm_or_si128(xmm5, xmm7);        \
+    xmm7 = _mm_unpackhi_epi8(xmm7, xmm4);           \
+    xmm5 = _mm_unpackhi_epi8(xmm5, xmm1);           \
+    xmm7 = _mm_slli_epi16(xmm7, 3);                 \
+    xmm5 = _mm_or_si128(xmm5, xmm7);                \
     _mm_stream_si128((__m128i*)(p_buffer+8), xmm5);
 
-#define SSE2_UNPACK_16_UNALIGNED            \
-    xmm5 = _mm_set1_epi32(0xf8f8f8f8UL);    \
-    xmm0 = _mm_and_si128(xmm0, xmm5);       \
-    xmm1 = _mm_and_si128(xmm1, xmm5);       \
-    xmm5 = _mm_set1_epi32(0xfcfcfcfcUL);    \
-    xmm2 = _mm_and_si128(xmm2, xmm5);       \
-    xmm0 = _mm_srli_epi16(xmm0, 3);         \
-    xmm4 = _mm_setzero_si128();             \
-    xmm5 = xmm0;                            \
-    xmm7 = xmm2;                            \
+#define SSE2_UNPACK_16_UNALIGNED                    \
+    xmm5 = _mm_set1_epi32(0xf8f8f8f8UL);            \
+    xmm0 = _mm_and_si128(xmm0, xmm5);               \
+    xmm1 = _mm_and_si128(xmm1, xmm5);               \
+    xmm5 = _mm_set1_epi32(0xfcfcfcfcUL);            \
+    xmm2 = _mm_and_si128(xmm2, xmm5);               \
+    xmm0 = _mm_srli_epi16(xmm0, 3);                 \
+    xmm4 = _mm_setzero_si128();                     \
+    xmm5 = xmm0;                                    \
+    xmm7 = xmm2;                                    \
     \
-    xmm2 = _mm_unpacklo_epi8(xmm2, xmm4);   \
-    xmm0 = _mm_unpacklo_epi8(xmm0, xmm1);   \
-    xmm2 = _mm_slli_epi16(xmm2, 3);         \
-    xmm0 = _mm_or_si128(xmm0, xmm2);        \
-    _mm_storeu_si128((__m128i*)p_buffer, xmm0); \
+    xmm2 = _mm_unpacklo_epi8(xmm2, xmm4);           \
+    xmm0 = _mm_unpacklo_epi8(xmm0, xmm1);           \
+    xmm2 = _mm_slli_epi16(xmm2, 3);                 \
+    xmm0 = _mm_or_si128(xmm0, xmm2);                \
+    _mm_storeu_si128((__m128i*)p_buffer, xmm0);     \
     \
-    xmm7 = _mm_unpackhi_epi8(xmm7, xmm4);   \
-    xmm5 = _mm_unpackhi_epi8(xmm5, xmm1);   \
-    xmm7 = _mm_slli_epi16(xmm7, 3);         \
-    xmm5 = _mm_or_si128(xmm5, xmm7);        \
+    xmm7 = _mm_unpackhi_epi8(xmm7, xmm4);           \
+    xmm5 = _mm_unpackhi_epi8(xmm5, xmm1);           \
+    xmm7 = _mm_slli_epi16(xmm7, 3);                 \
+    xmm5 = _mm_or_si128(xmm5, xmm7);                \
     _mm_storeu_si128((__m128i*)(p_buffer+8), xmm5);
 
-#define SSE2_UNPACK_32_ARGB_ALIGNED         \
-    xmm3 = _mm_setzero_si128();             \
-    xmm4 = xmm0;                            \
-    xmm4 = _mm_unpacklo_epi8(xmm4, xmm2);   \
-    xmm5 = xmm1;                            \
-    xmm5 = _mm_unpacklo_epi8(xmm5, xmm3);   \
-    xmm6 = xmm4;                            \
-    xmm4 = _mm_unpacklo_epi16(xmm4, xmm5);  \
-    _mm_stream_si128((__m128i*)(p_buffer), xmm4); \
-    xmm6 = _mm_unpackhi_epi16(xmm6, xmm5);  \
+#define SSE2_UNPACK_32_ARGB_ALIGNED                 \
+    xmm3 = _mm_setzero_si128();                     \
+    xmm4 = xmm0;                                    \
+    xmm4 = _mm_unpacklo_epi8(xmm4, xmm2);           \
+    xmm5 = xmm1;                                    \
+    xmm5 = _mm_unpacklo_epi8(xmm5, xmm3);           \
+    xmm6 = xmm4;                                    \
+    xmm4 = _mm_unpacklo_epi16(xmm4, xmm5);          \
+    _mm_stream_si128((__m128i*)(p_buffer), xmm4);   \
+    xmm6 = _mm_unpackhi_epi16(xmm6, xmm5);          \
     _mm_stream_si128((__m128i*)(p_buffer+4), xmm6); \
-    xmm0 = _mm_unpackhi_epi8(xmm0, xmm2);   \
-    xmm1 = _mm_unpackhi_epi8(xmm1, xmm3);   \
-    xmm5 = xmm0;                            \
-    xmm5 = _mm_unpacklo_epi16(xmm5, xmm1);  \
+    xmm0 = _mm_unpackhi_epi8(xmm0, xmm2);           \
+    xmm1 = _mm_unpackhi_epi8(xmm1, xmm3);           \
+    xmm5 = xmm0;                                    \
+    xmm5 = _mm_unpacklo_epi16(xmm5, xmm1);          \
     _mm_stream_si128((__m128i*)(p_buffer+8), xmm5); \
-    xmm0 = _mm_unpackhi_epi16(xmm0, xmm1);  \
+    xmm0 = _mm_unpackhi_epi16(xmm0, xmm1);          \
     _mm_stream_si128((__m128i*)(p_buffer+12), xmm0);
 
-#define SSE2_UNPACK_32_ARGB_UNALIGNED       \
-    xmm3 = _mm_setzero_si128();             \
-    xmm4 = xmm0;                            \
-    xmm4 = _mm_unpacklo_epi8(xmm4, xmm2);   \
-    xmm5 = xmm1;                            \
-    xmm5 = _mm_unpacklo_epi8(xmm5, xmm3);   \
-    xmm6 = xmm4;                            \
-    xmm4 = _mm_unpacklo_epi16(xmm4, xmm5);  \
-    _mm_storeu_si128((__m128i*)(p_buffer), xmm4); \
-    xmm6 = _mm_unpackhi_epi16(xmm6, xmm5);  \
+#define SSE2_UNPACK_32_ARGB_UNALIGNED               \
+    xmm3 = _mm_setzero_si128();                     \
+    xmm4 = xmm0;                                    \
+    xmm4 = _mm_unpacklo_epi8(xmm4, xmm2);           \
+    xmm5 = xmm1;                                    \
+    xmm5 = _mm_unpacklo_epi8(xmm5, xmm3);           \
+    xmm6 = xmm4;                                    \
+    xmm4 = _mm_unpacklo_epi16(xmm4, xmm5);          \
+    _mm_storeu_si128((__m128i*)(p_buffer), xmm4);   \
+    xmm6 = _mm_unpackhi_epi16(xmm6, xmm5);          \
     _mm_storeu_si128((__m128i*)(p_buffer+4), xmm6); \
-    xmm0 = _mm_unpackhi_epi8(xmm0, xmm2);   \
-    xmm1 = _mm_unpackhi_epi8(xmm1, xmm3);   \
-    xmm5 = xmm0;                            \
-    xmm5 = _mm_unpacklo_epi16(xmm5, xmm1);  \
+    xmm0 = _mm_unpackhi_epi8(xmm0, xmm2);           \
+    xmm1 = _mm_unpackhi_epi8(xmm1, xmm3);           \
+    xmm5 = xmm0;                                    \
+    xmm5 = _mm_unpacklo_epi16(xmm5, xmm1);          \
     _mm_storeu_si128((__m128i*)(p_buffer+8), xmm5); \
-    xmm0 = _mm_unpackhi_epi16(xmm0, xmm1);  \
+    xmm0 = _mm_unpackhi_epi16(xmm0, xmm1);          \
     _mm_storeu_si128((__m128i*)(p_buffer+12), xmm0);
 
-#define SSE2_UNPACK_32_BGRA_ALIGNED         \
-    xmm3 = _mm_setzero_si128();             \
-    xmm4 = xmm2;                            \
-    xmm4 = _mm_unpacklo_epi8(xmm4, xmm0);   \
-    xmm3 = _mm_unpacklo_epi8(xmm3, xmm1);   \
-    xmm5 = xmm3;                            \
-    xmm3 = _mm_unpacklo_epi16(xmm3, xmm4);  \
-    _mm_stream_si128((__m128i*)(p_buffer), xmm3); \
-    xmm5 = _mm_unpackhi_epi16(xmm5, xmm4);  \
+#define SSE2_UNPACK_32_RGBA_ALIGNED                 \
+    xmm3 = _mm_setzero_si128();                     \
+    xmm4 = xmm2;                                    \
+    xmm4 = _mm_unpacklo_epi8(xmm4, xmm1);           \
+    xmm3 = _mm_unpacklo_epi8(xmm3, xmm0);           \
+    xmm5 = xmm3;                                    \
+    xmm3 = _mm_unpacklo_epi16(xmm3, xmm4);          \
+    _mm_stream_si128((__m128i*)(p_buffer), xmm3);   \
+    xmm5 = _mm_unpackhi_epi16(xmm5, xmm4);          \
     _mm_stream_si128((__m128i*)(p_buffer+4), xmm5); \
-    xmm6 = _mm_setzero_si128();             \
-    xmm2 = _mm_unpackhi_epi8(xmm2, xmm0);   \
-    xmm6 = _mm_unpackhi_epi8(xmm6, xmm1);   \
-    xmm0 = xmm6;                            \
-    xmm6 = _mm_unpacklo_epi16(xmm6, xmm2);  \
+    xmm6 = _mm_setzero_si128();                     \
+    xmm2 = _mm_unpackhi_epi8(xmm2, xmm1);           \
+    xmm6 = _mm_unpackhi_epi8(xmm6, xmm0);           \
+    xmm0 = xmm6;                                    \
+    xmm6 = _mm_unpacklo_epi16(xmm6, xmm2);          \
     _mm_stream_si128((__m128i*)(p_buffer+8), xmm6); \
-    xmm0 = _mm_unpackhi_epi16(xmm0, xmm2);  \
+    xmm0 = _mm_unpackhi_epi16(xmm0, xmm2);          \
     _mm_stream_si128((__m128i*)(p_buffer+12), xmm0);
 
-#define SSE2_UNPACK_32_BGRA_UNALIGNED       \
-    xmm3 = _mm_setzero_si128();             \
-    xmm4 = xmm2;                            \
-    xmm4 = _mm_unpacklo_epi8(xmm4, xmm0);   \
-    xmm3 = _mm_unpacklo_epi8(xmm3, xmm1);   \
-    xmm5 = xmm3;                            \
-    xmm3 = _mm_unpacklo_epi16(xmm3, xmm4);  \
-    _mm_storeu_si128((__m128i*)(p_buffer), xmm3); \
-    xmm5 = _mm_unpackhi_epi16(xmm5, xmm4);  \
+#define SSE2_UNPACK_32_RGBA_UNALIGNED               \
+    xmm3 = _mm_setzero_si128();                     \
+    xmm4 = xmm2;                                    \
+    xmm4 = _mm_unpacklo_epi8(xmm4, xmm1);           \
+    xmm3 = _mm_unpacklo_epi8(xmm3, xmm0);           \
+    xmm5 = xmm3;                                    \
+    xmm3 = _mm_unpacklo_epi16(xmm3, xmm4);          \
+    _mm_storeu_si128((__m128i*)(p_buffer), xmm3);   \
+    xmm5 = _mm_unpackhi_epi16(xmm5, xmm4);          \
     _mm_storeu_si128((__m128i*)(p_buffer+4), xmm5); \
-    xmm6 = _mm_setzero_si128();             \
-    xmm2 = _mm_unpackhi_epi8(xmm2, xmm0);   \
-    xmm6 = _mm_unpackhi_epi8(xmm6, xmm1);   \
-    xmm0 = xmm6;                            \
-    xmm6 = _mm_unpacklo_epi16(xmm6, xmm2);  \
+    xmm6 = _mm_setzero_si128();                     \
+    xmm2 = _mm_unpackhi_epi8(xmm2, xmm1);           \
+    xmm6 = _mm_unpackhi_epi8(xmm6, xmm0);           \
+    xmm0 = xmm6;                                    \
+    xmm6 = _mm_unpacklo_epi16(xmm6, xmm2);          \
     _mm_storeu_si128((__m128i*)(p_buffer+8), xmm6); \
-    xmm0 = _mm_unpackhi_epi16(xmm0, xmm2);  \
+    xmm0 = _mm_unpackhi_epi16(xmm0, xmm2);          \
     _mm_storeu_si128((__m128i*)(p_buffer+12), xmm0);
 
-#define SSE2_UNPACK_32_ABGR_ALIGNED         \
-    ;
+#define SSE2_UNPACK_32_BGRA_ALIGNED                 \
+    xmm3 = _mm_setzero_si128();                     \
+    xmm4 = xmm2;                                    \
+    xmm4 = _mm_unpacklo_epi8(xmm4, xmm0);           \
+    xmm3 = _mm_unpacklo_epi8(xmm3, xmm1);           \
+    xmm5 = xmm3;                                    \
+    xmm3 = _mm_unpacklo_epi16(xmm3, xmm4);          \
+    _mm_stream_si128((__m128i*)(p_buffer), xmm3);   \
+    xmm5 = _mm_unpackhi_epi16(xmm5, xmm4);          \
+    _mm_stream_si128((__m128i*)(p_buffer+4), xmm5); \
+    xmm6 = _mm_setzero_si128();                     \
+    xmm2 = _mm_unpackhi_epi8(xmm2, xmm0);           \
+    xmm6 = _mm_unpackhi_epi8(xmm6, xmm1);           \
+    xmm0 = xmm6;                                    \
+    xmm6 = _mm_unpacklo_epi16(xmm6, xmm2);          \
+    _mm_stream_si128((__m128i*)(p_buffer+8), xmm6); \
+    xmm0 = _mm_unpackhi_epi16(xmm0, xmm2);          \
+    _mm_stream_si128((__m128i*)(p_buffer+12), xmm0);
+
+#define SSE2_UNPACK_32_BGRA_UNALIGNED               \
+    xmm3 = _mm_setzero_si128();                     \
+    xmm4 = xmm2;                                    \
+    xmm4 = _mm_unpacklo_epi8(xmm4, xmm0);           \
+    xmm3 = _mm_unpacklo_epi8(xmm3, xmm1);           \
+    xmm5 = xmm3;                                    \
+    xmm3 = _mm_unpacklo_epi16(xmm3, xmm4);          \
+    _mm_storeu_si128((__m128i*)(p_buffer), xmm3);   \
+    xmm5 = _mm_unpackhi_epi16(xmm5, xmm4);          \
+    _mm_storeu_si128((__m128i*)(p_buffer+4), xmm5); \
+    xmm6 = _mm_setzero_si128();                     \
+    xmm2 = _mm_unpackhi_epi8(xmm2, xmm0);           \
+    xmm6 = _mm_unpackhi_epi8(xmm6, xmm1);           \
+    xmm0 = xmm6;                                    \
+    xmm6 = _mm_unpacklo_epi16(xmm6, xmm2);          \
+    _mm_storeu_si128((__m128i*)(p_buffer+8), xmm6); \
+    xmm0 = _mm_unpackhi_epi16(xmm0, xmm2);          \
+    _mm_storeu_si128((__m128i*)(p_buffer+12), xmm0);
 
-#define SSE2_UNPACK_32_ABGR_UNALIGNED       \
-    ;
+#define SSE2_UNPACK_32_ABGR_ALIGNED                 \
+    xmm3 = _mm_setzero_si128();                     \
+    xmm4 = xmm1;                                    \
+    xmm4 = _mm_unpacklo_epi8(xmm4, xmm2);           \
+    xmm5 = xmm0;                                    \
+    xmm5 = _mm_unpacklo_epi8(xmm5, xmm3);           \
+    xmm6 = xmm4;                                    \
+    xmm4 = _mm_unpacklo_epi16(xmm4, xmm5);          \
+    _mm_stream_si128((__m128i*)(p_buffer), xmm4);   \
+    xmm6 = _mm_unpackhi_epi16(xmm6, xmm5);          \
+    _mm_stream_si128((__m128i*)(p_buffer+4), xmm6); \
+    xmm1 = _mm_unpackhi_epi8(xmm1, xmm2);           \
+    xmm0 = _mm_unpackhi_epi8(xmm0, xmm3);           \
+    xmm2 = xmm1;                                    \
+    xmm1 = _mm_unpacklo_epi16(xmm1, xmm0);          \
+    _mm_stream_si128((__m128i*)(p_buffer+8), xmm1); \
+    xmm2 = _mm_unpackhi_epi16(xmm2, xmm0);          \
+    _mm_stream_si128((__m128i*)(p_buffer+12), xmm2);
+
+#define SSE2_UNPACK_32_ABGR_UNALIGNED               \
+    xmm3 = _mm_setzero_si128();                     \
+    xmm4 = xmm1;                                    \
+    xmm4 = _mm_unpacklo_epi8(xmm4, xmm2);           \
+    xmm5 = xmm0;                                    \
+    xmm5 = _mm_unpacklo_epi8(xmm5, xmm3);           \
+    xmm6 = xmm4;                                    \
+    xmm4 = _mm_unpacklo_epi16(xmm4, xmm5);          \
+    _mm_storeu_si128((__m128i*)(p_buffer), xmm4);   \
+    xmm6 = _mm_unpackhi_epi16(xmm6, xmm5);          \
+    _mm_storeu_si128((__m128i*)(p_buffer+4), xmm6); \
+    xmm1 = _mm_unpackhi_epi8(xmm1, xmm2);           \
+    xmm0 = _mm_unpackhi_epi8(xmm0, xmm3);           \
+    xmm2 = xmm1;                                    \
+    xmm1 = _mm_unpacklo_epi16(xmm1, xmm0);          \
+    _mm_storeu_si128((__m128i*)(p_buffer+8), xmm1); \
+    xmm2 = _mm_unpackhi_epi16(xmm2, xmm0);          \
+    _mm_storeu_si128((__m128i*)(p_buffer+12), xmm2);
 
 #endif
 

diff --git a/modules/video_chroma/i420_yuy2.h b/modules/video_chroma/i420_yuy2.h
index a881c6e5a88df9be81e3fc8b349ce03afe571a2c..1f35a3061f19fc3407dafc8e8922fadcffa323a3 100644 (file)
--- a/modules/video_chroma/i420_yuy2.h
+++ b/modules/video_chroma/i420_yuy2.h
@@ -138,56 +138,56 @@ movq      %%mm1, (%1)   # Store YUYV                                      \n\
 #define MMX_END _mm_empty()
     
 #define MMX_YUV420_YUYV                     \
-    mm1 = _mm_cvtsi32_si64((int)*p_u);      \
-    mm2 = _mm_cvtsi32_si64((int)*p_v);      \
+    mm1 = _mm_cvtsi32_si64(*(int*)p_u);     \
+    mm2 = _mm_cvtsi32_si64(*(int*)p_v);     \
     mm0 = (__m64)*(uint64_t*)p_y1;          \
     mm3 = (__m64)*(uint64_t*)p_y2;          \
     mm1 = _mm_unpacklo_pi8(mm1, mm2);       \
     mm2 = mm0;                              \
     mm2 = _mm_unpacklo_pi8(mm2, mm1);       \
-    *(uin64_t)p_line1 = (uint64)mm2;        \
+    *(uint64_t*)p_line1 = (uint64_t)mm2;    \
     mm0 = _mm_unpackhi_pi8(mm0, mm1);       \
-    *(uin64_t)(p_line1 + 4) = (uint64)mm0;  \
+    *(uint64_t*)(p_line1+8) = (uint64_t)mm0;\
     mm4 = mm3;                              \
     mm4 = _mm_unpacklo_pi8(mm4, mm1);       \
-    *(uin64_t)p_line2 = (uint64)mm4;        \
+    *(uint64_t*)p_line2 = (uint64_t)mm4;    \
     mm3 = _mm_unpackhi_pi8(mm3, mm1);       \
-    *(uin64_t)(p_line2 + 4) = (uint64)mm4;
+    *(uint64_t*)(p_line2+8) = (uint64_t)mm3;
 
 #define MMX_YUV420_YVYU                     \
-    mm2 = _mm_cvtsi32_si64((int)*p_u);      \
-    mm1 = _mm_cvtsi32_si64((int)*p_v);      \
+    mm2 = _mm_cvtsi32_si64(*(int*)p_u);     \
+    mm1 = _mm_cvtsi32_si64(*(int*)p_v);     \
     mm0 = (__m64)*(uint64_t*)p_y1;          \
     mm3 = (__m64)*(uint64_t*)p_y2;          \
     mm1 = _mm_unpacklo_pi8(mm1, mm2);       \
     mm2 = mm0;                              \
     mm2 = _mm_unpacklo_pi8(mm2, mm1);       \
-    *(uin64_t)p_line1 = (uint64)mm2;        \
+    *(uint64_t*)p_line1 = (uint64_t)mm2;    \
     mm0 = _mm_unpackhi_pi8(mm0, mm1);       \
-    *(uin64_t)(p_line1 + 4) = (uint64)mm0;  \
+    *(uint64_t*)(p_line1+8) = (uint64_t)mm0;\
     mm4 = mm3;                              \
     mm4 = _mm_unpacklo_pi8(mm4, mm1);       \
-    *(uin64_t)p_line2 = (uint64)mm4;        \
+    *(uint64_t*)p_line2 = (uint64_t)mm4;    \
     mm3 = _mm_unpackhi_pi8(mm3, mm1);       \
-    *(uin64_t)(p_line2 + 4) = (uint64)mm4;
+    *(uint64_t*)(p_line2+8) = (uint64_t)mm3;
 
 #define MMX_YUV420_UYVY                     \
-    mm1 = _mm_cvtsi32_si64((int)*p_u);      \
-    mm2 = _mm_cvtsi32_si64((int)*p_v);      \
+    mm1 = _mm_cvtsi32_si64(*(int*)p_u);     \
+    mm2 = _mm_cvtsi32_si64(*(int*)p_v);     \
     mm0 = (__m64)*(uint64_t*)p_y1;          \
     mm3 = (__m64)*(uint64_t*)p_y2;          \
     mm1 = _mm_unpacklo_pi8(mm1, mm2);       \
     mm2 = mm1;                              \
     mm2 = _mm_unpacklo_pi8(mm2, mm0);       \
-    *(uin64_t)p_line1 = (uint64)mm2;        \
+    *(uint64_t*)p_line1 = (uint64_t)mm2;    \
     mm2 = mm1;                              \
     mm2 = _mm_unpackhi_pi8(mm2, mm0);       \
-    *(uin64_t)(p_line1 + 4) = (uint64)mm2;  \
+    *(uint64_t*)(p_line1+8) = (uint64_t)mm2;\
     mm4 = mm1;                              \
     mm4 = _mm_unpacklo_pi8(mm4, mm3);       \
-    *(uin64_t)p_line2 = (uint64)mm4;        \
+    *(uint64_t*)p_line2 = (uint64_t)mm4;    \
     mm1 = _mm_unpackhi_pi8(mm1, mm3);       \
-    *(uin64_t)(p_line2 + 4) = (uint64)mm1;
+    *(uint64_t*)(p_line2+8) = (uint64_t)mm1;
 
 #endif