vp3: DC-only IDCT

2-4% faster overall decode

Originally committed as revision 22896 to svn://svn.ffmpeg.org/ffmpeg/trunk

3 files changed, 45 insertions, 0 deletions

diff --git a/libavcodec/x86/dsputil_mmx.c b/libavcodec/x86/dsputil_mmx.c
index 31387ebb30..cc2f881303 100644
--- a/libavcodec/x86/dsputil_mmx.c
+++ b/libavcodec/x86/dsputil_mmx.c
@@ -2653,6 +2653,9 @@ void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
                     c->vp3_h_loop_filter= ff_vp3_h_loop_filter_mmx2;
                 }
             }
+            if (CONFIG_VP3_DECODER) {
+                c->vp3_idct_dc_add = ff_vp3_idct_dc_add_mmx2;
+            }
 
 #define SET_QPEL_FUNCS(PFX, IDX, SIZE, CPU) \
             c->PFX ## _pixels_tab[IDX][ 0] = PFX ## SIZE ## _mc00_ ## CPU; \
diff --git a/libavcodec/x86/vp3dsp_mmx.c b/libavcodec/x86/vp3dsp_mmx.c
index fead8e8cef..309dd4aa5d 100644
--- a/libavcodec/x86/vp3dsp_mmx.c
+++ b/libavcodec/x86/vp3dsp_mmx.c
@@ -395,3 +395,44 @@ void ff_vp3_idct_add_mmx(uint8_t *dest, int line_size, DCTELEM *block)
     ff_vp3_idct_mmx(block);
     add_pixels_clamped_mmx(block, dest, line_size);
 }
+
+void ff_vp3_idct_dc_add_mmx2(uint8_t *dest, int linesize, const DCTELEM *block)
+{
+    int dc = block[0];
+    dc = (46341*dc)>>16;
+    dc = (46341*dc + (8<<16))>>20;
+
+    __asm__ volatile(
+        "movd          %3, %%mm0 \n\t"
+        "pshufw $0, %%mm0, %%mm0 \n\t"
+        "pxor       %%mm1, %%mm1 \n\t"
+        "psubw      %%mm0, %%mm1 \n\t"
+        "packuswb   %%mm0, %%mm0 \n\t"
+        "packuswb   %%mm1, %%mm1 \n\t"
+
+#define DC_ADD \
+        "movq        (%0), %%mm2 \n\t" \
+        "movq     (%0,%1), %%mm3 \n\t" \
+        "paddusb    %%mm0, %%mm2 \n\t" \
+        "movq   (%0,%1,2), %%mm4 \n\t" \
+        "paddusb    %%mm0, %%mm3 \n\t" \
+        "movq     (%0,%2), %%mm5 \n\t" \
+        "paddusb    %%mm0, %%mm4 \n\t" \
+        "paddusb    %%mm0, %%mm5 \n\t" \
+        "psubusb    %%mm1, %%mm2 \n\t" \
+        "psubusb    %%mm1, %%mm3 \n\t" \
+        "movq       %%mm2, (%0)  \n\t" \
+        "psubusb    %%mm1, %%mm4 \n\t" \
+        "movq       %%mm3, (%0,%1) \n\t" \
+        "psubusb    %%mm1, %%mm5 \n\t" \
+        "movq       %%mm4, (%0,%1,2) \n\t" \
+        "movq       %%mm5, (%0,%2) \n\t"
+
+        DC_ADD
+        "lea    (%0,%1,4), %0 \n\t"
+        DC_ADD
+
+        : "+r"(dest)
+        : "r"((x86_reg)linesize), "r"((x86_reg)3*linesize), "r"(dc)
+    );
+}
diff --git a/libavcodec/x86/vp3dsp_mmx.h b/libavcodec/x86/vp3dsp_mmx.h
index e565a33023..e0ebf0b0f4 100644
--- a/libavcodec/x86/vp3dsp_mmx.h
+++ b/libavcodec/x86/vp3dsp_mmx.h
@@ -28,6 +28,7 @@
 void ff_vp3_idct_mmx(int16_t *data);
 void ff_vp3_idct_put_mmx(uint8_t *dest, int line_size, DCTELEM *block);
 void ff_vp3_idct_add_mmx(uint8_t *dest, int line_size, DCTELEM *block);
+void ff_vp3_idct_dc_add_mmx2(uint8_t *dest, int line_size, const DCTELEM *block);
 
 void ff_vp3_v_loop_filter_mmx2(uint8_t *src, int stride, int *bounding_values);
 void ff_vp3_h_loop_filter_mmx2(uint8_t *src, int stride, int *bounding_values);