new ffmpeg files for build 4707

CVS patchset: 6252
CVS date: 2004/03/14 21:07:01

3 files changed, 1505 insertions, 0 deletions

diff --git a/src/libffmpeg/libavcodec/i386/vp3dsp_mmx.c b/src/libffmpeg/libavcodec/i386/vp3dsp_mmx.c
new file mode 100644
index 000000000..59020466f
--- /dev/null
+++ b/src/libffmpeg/libavcodec/i386/vp3dsp_mmx.c
@@ -0,0 +1,652 @@
+/*
+ * Copyright (C) 2004 the ffmpeg project
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+/**
+ * @file vp3dsp_mmx.c
+ * MMX-optimized functions cribbed from the original VP3 source code.
+ */
+
+#include "../dsputil.h"
+#include "mmx.h"
+
+#define IdctAdjustBeforeShift 8
+
+/* (12 * 4) 2-byte memory locations ( = 96 bytes total)
+ * idct_constants[0..15] = Mask table (M(I))
+ * idct_constants[16..43] = Cosine table (C(I))
+ * idct_constants[44..47] = 8
+ */
+static uint16_t idct_constants[(4 + 7 + 1) * 4];
+static uint16_t idct_cosine_table[7] = {
+    64277, 60547, 54491, 46341, 36410, 25080, 12785
+};
+
+#define r0 mm0
+#define r1 mm1
+#define r2 mm2
+#define r3 mm3
+#define r4 mm4
+#define r5 mm5
+#define r6 mm6
+#define r7 mm7
+
+/* from original comments: The Macro does IDct on 4 1-D Dcts */
+#define BeginIDCT() \
+    movq_m2r(*I(3), r2); \
+    movq_m2r(*C(3), r6); \
+    movq_r2r(r2, r4); \
+    movq_m2r(*J(5), r7); \
+    pmulhw_r2r(r6, r4); \
+    movq_m2r(*C(5), r1); \
+    pmulhw_r2r(r7, r6); \
+    movq_r2r(r1, r5); \
+    pmulhw_r2r(r2, r1); \
+    movq_m2r(*I(1), r3); \
+    pmulhw_r2r(r7, r5); \
+    movq_m2r(*C(1), r0); \
+    paddw_r2r(r2, r4); \
+    paddw_r2r(r7, r6); \
+    paddw_r2r(r1, r2); \
+    movq_m2r(*J(7), r1); \
+    paddw_r2r(r5, r7); \
+    movq_r2r(r0, r5); \
+    pmulhw_r2r(r3, r0); \
+    paddsw_r2r(r7, r4); \
+    pmulhw_r2r(r1, r5); \
+    movq_m2r(*C(7), r7); \
+    psubsw_r2r(r2, r6); \
+    paddw_r2r(r3, r0); \
+    pmulhw_r2r(r7, r3); \
+    movq_m2r(*I(2), r2); \
+    pmulhw_r2r(r1, r7); \
+    paddw_r2r(r1, r5); \
+    movq_r2r(r2, r1); \
+    pmulhw_m2r(*C(2), r2); \
+    psubsw_r2r(r5, r3); \
+    movq_m2r(*J(6), r5); \
+    paddsw_r2r(r7, r0); \
+    movq_r2r(r5, r7); \
+    psubsw_r2r(r4, r0); \
+    pmulhw_m2r(*C(2), r5); \
+    paddw_r2r(r1, r2); \
+    pmulhw_m2r(*C(6), r1); \
+    paddsw_r2r(r4, r4); \
+    paddsw_r2r(r0, r4); \
+    psubsw_r2r(r6, r3); \
+    paddw_r2r(r7, r5); \
+    paddsw_r2r(r6, r6); \
+    pmulhw_m2r(*C(6), r7); \
+    paddsw_r2r(r3, r6); \
+    movq_r2m(r4, *I(1)); \
+    psubsw_r2r(r5, r1); \
+    movq_m2r(*C(4), r4); \
+    movq_r2r(r3, r5); \
+    pmulhw_r2r(r4, r3); \
+    paddsw_r2r(r2, r7); \
+    movq_r2m(r6, *I(2)); \
+    movq_r2r(r0, r2); \
+    movq_m2r(*I(0), r6); \
+    pmulhw_r2r(r4, r0); \
+    paddw_r2r(r3, r5); \
+    movq_m2r(*J(4), r3); \
+    psubsw_r2r(r1, r5); \
+    paddw_r2r(r0, r2); \
+    psubsw_r2r(r3, r6); \
+    movq_r2r(r6, r0); \
+    pmulhw_r2r(r4, r6); \
+    paddsw_r2r(r3, r3); \
+    paddsw_r2r(r1, r1); \
+    paddsw_r2r(r0, r3); \
+    paddsw_r2r(r5, r1); \
+    pmulhw_r2r(r3, r4); \
+    paddsw_r2r(r0, r6); \
+    psubsw_r2r(r2, r6); \
+    paddsw_r2r(r2, r2); \
+    movq_m2r(*I(1), r0); \
+    paddsw_r2r(r6, r2); \
+    paddw_r2r(r3, r4); \
+    psubsw_r2r(r1, r2);
+
+/* RowIDCT gets ready to transpose */
+#define RowIDCT() \
+    \
+    BeginIDCT() \
+    \
+    movq_m2r(*I(2), r3); \
+    psubsw_r2r(r7, r4); \
+    paddsw_r2r(r1, r1); \
+    paddsw_r2r(r7, r7); \
+    paddsw_r2r(r2, r1); \
+    paddsw_r2r(r4, r7); \
+    psubsw_r2r(r3, r4); \
+    psubsw_r2r(r5, r6); \
+    paddsw_r2r(r5, r5); \
+    paddsw_r2r(r4, r3); \
+    paddsw_r2r(r6, r5); \
+    psubsw_r2r(r0, r7); \
+    paddsw_r2r(r0, r0); \
+    movq_r2m(r1, *I(1)); \
+    paddsw_r2r(r7, r0);
+
+/* Column IDCT normalizes and stores final results */
+#define ColumnIDCT() \
+    \
+    BeginIDCT() \
+    \
+    paddsw_m2r(*Eight, r2); \
+    paddsw_r2r(r1, r1); \
+    paddsw_r2r(r2, r1); \
+    psraw_i2r(4, r2); \
+    psubsw_r2r(r7, r4); \
+    psraw_i2r(4, r1); \
+    movq_m2r(*I(2), r3); \
+    paddsw_r2r(r7, r7); \
+    movq_r2m(r2, *I(2)); \
+    paddsw_r2r(r4, r7); \
+    movq_r2m(r1, *I(1)); \
+    psubsw_r2r(r3, r4); \
+    paddsw_m2r(*Eight, r4); \
+    paddsw_r2r(r3, r3); \
+    paddsw_r2r(r4, r3); \
+    psraw_i2r(4, r4); \
+    psubsw_r2r(r5, r6); \
+    psraw_i2r(4, r3); \
+    paddsw_m2r(*Eight, r6); \
+    paddsw_r2r(r5, r5); \
+    paddsw_r2r(r6, r5); \
+    psraw_i2r(4, r6); \
+    movq_r2m(r4, *J(4)); \
+    psraw_i2r(4, r5); \
+    movq_r2m(r3, *I(3)); \
+    psubsw_r2r(r0, r7); \
+    paddsw_m2r(*Eight, r7); \
+    paddsw_r2r(r0, r0); \
+    paddsw_r2r(r7, r0); \
+    psraw_i2r(4, r7); \
+    movq_r2m(r6, *J(6)); \
+    psraw_i2r(4, r0); \
+    movq_r2m(r5, *J(5)); \
+    movq_r2m(r7, *J(7)); \
+    movq_r2m(r0, *I(0));
+
+
+/* Following macro does two 4x4 transposes in place.
+
+  At entry (we assume):
+
+        r0 = a3 a2 a1 a0
+        I(1) = b3 b2 b1 b0
+        r2 = c3 c2 c1 c0
+        r3 = d3 d2 d1 d0
+
+        r4 = e3 e2 e1 e0
+        r5 = f3 f2 f1 f0
+        r6 = g3 g2 g1 g0
+        r7 = h3 h2 h1 h0
+
+   At exit, we have:
+
+        I(0) = d0 c0 b0 a0
+        I(1) = d1 c1 b1 a1
+        I(2) = d2 c2 b2 a2
+        I(3) = d3 c3 b3 a3
+
+        J(4) = h0 g0 f0 e0
+        J(5) = h1 g1 f1 e1
+        J(6) = h2 g2 f2 e2
+        J(7) = h3 g3 f3 e3
+
+   I(0) I(1) I(2) I(3)  is the transpose of r0 I(1) r2 r3.
+   J(4) J(5) J(6) J(7)  is the transpose of r4 r5 r6 r7.
+
+   Since r1 is free at entry, we calculate the Js first. */
+
+#define Transpose() \
+    movq_r2r(r4, r1); \
+    punpcklwd_r2r(r5, r4); \
+    movq_r2m(r0, *I(0)); \
+    punpckhwd_r2r(r5, r1); \
+    movq_r2r(r6, r0); \
+    punpcklwd_r2r(r7, r6); \
+    movq_r2r(r4, r5); \
+    punpckldq_r2r(r6, r4); \
+    punpckhdq_r2r(r6, r5); \
+    movq_r2r(r1, r6); \
+    movq_r2m(r4, *J(4)); \
+    punpckhwd_r2r(r7, r0); \
+    movq_r2m(r5, *J(5)); \
+    punpckhdq_r2r(r0, r6); \
+    movq_m2r(*I(0), r4); \
+    punpckldq_r2r(r0, r1); \
+    movq_m2r(*I(1), r5); \
+    movq_r2r(r4, r0); \
+    movq_r2m(r6, *J(7)); \
+    punpcklwd_r2r(r5, r0); \
+    movq_r2m(r1, *J(6)); \
+    punpckhwd_r2r(r5, r4); \
+    movq_r2r(r2, r5); \
+    punpcklwd_r2r(r3, r2); \
+    movq_r2r(r0, r1); \
+    punpckldq_r2r(r2, r0); \
+    punpckhdq_r2r(r2, r1); \
+    movq_r2r(r4, r2); \
+    movq_r2m(r0, *I(0)); \
+    punpckhwd_r2r(r3, r5); \
+    movq_r2m(r1, *I(1)); \
+    punpckhdq_r2r(r5, r4); \
+    punpckldq_r2r(r5, r2); \
+    movq_r2m(r4, *I(3)); \
+    movq_r2m(r2, *I(2));
+
+
+void vp3_dsp_init_mmx(void)
+{
+    int j = 16;
+    uint16_t *p;
+
+    do {
+        idct_constants[--j] = 0;
+    } while (j);
+
+    idct_constants[0]  = idct_constants[5] = 
+    idct_constants[10] = idct_constants[15] = 65535;
+
+    j = 1;
+    do {
+        p = idct_constants + ((j + 3) << 2);
+        p[0] = p[1] = p[2] = p[3] = idct_cosine_table[j - 1];
+    } while (++j <= 7);
+
+    idct_constants[44] = idct_constants[45] = 
+    idct_constants[46] = idct_constants[47] = IdctAdjustBeforeShift;
+}
+
+static void vp3_idct_mmx(int16_t *input_data, int16_t *dequant_matrix,
+     int16_t *output_data)
+{
+    /* eax = quantized input
+     * ebx = dequantizer matrix
+     * ecx = IDCT constants
+     *  M(I) = ecx + MaskOffset(0) + I * 8
+     *  C(I) = ecx + CosineOffset(32) + (I-1) * 8
+     * edx = output
+     * r0..r7 = mm0..mm7
+     */
+
+#define M(x) (idct_constants + x * 4)
+#define C(x) (idct_constants + 16 + (x - 1) * 4)
+#define Eight (idct_constants + 44)
+
+    movq_m2r(*input_data, r0);
+    pmullw_m2r(*dequant_matrix, r0);
+    movq_m2r(*(input_data + 8), r1);
+    pmullw_m2r(*(dequant_matrix + 8), r1);
+    movq_m2r(*M(0), r2);
+    movq_r2r(r0, r3);
+    movq_m2r(*(input_data + 4), r4);
+    psrlq_i2r(16, r0);
+    pmullw_m2r(*(dequant_matrix + 4), r4);
+    pand_r2r(r2, r3);
+    movq_r2r(r0, r5);
+    movq_r2r(r1, r6);
+    pand_r2r(r2, r5);
+    psllq_i2r(32, r6);
+    movq_m2r(*M(3), r7);
+    pxor_r2r(r5, r0);
+    pand_r2r(r6, r7);
+    por_r2r(r3, r0);
+    pxor_r2r(r7, r6);
+    por_r2r(r7, r0);
+    movq_m2r(*M(3), r7);
+    movq_r2r(r4, r3);
+    movq_r2m(r0, *output_data);
+
+    pand_r2r(r2, r3);
+    movq_m2r(*(input_data + 16), r0);
+    psllq_i2r(16, r3);
+    pmullw_m2r(*(dequant_matrix + 16), r0);
+    pand_r2r(r1, r7);
+    por_r2r(r3, r5);
+    por_r2r(r6, r7);
+    movq_m2r(*(input_data + 12), r3);
+    por_r2r(r5, r7);
+    pmullw_m2r(*(dequant_matrix + 12), r3);
+    psrlq_i2r(16, r4);
+    movq_r2m(r7, *(output_data + 8));
+
+    movq_r2r(r4, r5);
+    movq_r2r(r0, r7);
+    psrlq_i2r(16, r4);
+    psrlq_i2r(48, r7);
+    movq_r2r(r2, r6);
+    pand_r2r(r2, r5);
+    pand_r2r(r4, r6);
+    movq_r2m(r7, *(output_data + 40));
+
+    pxor_r2r(r6, r4);
+    psrlq_i2r(32, r1);
+    por_r2r(r5, r4);
+    movq_m2r(*M(3), r7);
+    pand_r2r(r2, r1);
+    movq_m2r(*(input_data + 24), r5);
+    psllq_i2r(16, r0);
+    pmullw_m2r(*(dequant_matrix + 24), r5);
+    pand_r2r(r0, r7);
+    movq_r2m(r1, *(output_data + 32));
+
+    por_r2r(r4, r7);
+    movq_r2r(r3, r4);
+    pand_r2r(r2, r3);
+    movq_m2r(*M(2), r1);
+    psllq_i2r(32, r3);
+    por_r2r(r3, r7);
+    movq_r2r(r5, r3);
+    psllq_i2r(48, r3);
+    pand_r2r(r0, r1);
+    movq_r2m(r7, *(output_data + 16));
+
+    por_r2r(r3, r6);
+    movq_m2r(*M(1), r7);
+    por_r2r(r1, r6);
+    movq_m2r(*(input_data + 28), r1);
+    pand_r2r(r4, r7);
+    pmullw_m2r(*(dequant_matrix + 28), r1);
+    por_r2r(r6, r7);
+    pand_m2r(*M(1), r0);
+    psrlq_i2r(32, r4);
+    movq_r2m(r7, *(output_data + 24));
+
+    movq_r2r(r4, r6);
+    movq_m2r(*M(3), r7);
+    pand_r2r(r2, r4);
+    movq_m2r(*M(1), r3);
+    pand_r2r(r1, r7);
+    pand_r2r(r5, r3);
+    por_r2r(r4, r0);
+    psllq_i2r(16, r3);
+    por_r2r(r0, r7);
+    movq_m2r(*M(2), r4);
+    por_r2r(r3, r7);
+    movq_m2r(*(input_data + 40), r0);
+    movq_r2r(r4, r3);
+    pmullw_m2r(*(dequant_matrix + 40), r0);
+    pand_r2r(r5, r4);
+    movq_r2m(r7, *(output_data + 4));
+
+    por_r2r(r4, r6);
+    movq_r2r(r3, r4);
+    psrlq_i2r(16, r6);
+    movq_r2r(r0, r7);
+    pand_r2r(r1, r4);
+    psllq_i2r(48, r7);
+    por_r2r(r4, r6);
+    movq_m2r(*(input_data + 44), r4);
+    por_r2r(r6, r7);
+    pmullw_m2r(*(dequant_matrix + 44), r4);
+    psrlq_i2r(16, r3);
+    movq_r2m(r7, *(output_data + 12));
+
+    pand_r2r(r1, r3);
+    psrlq_i2r(48, r5);
+    pand_r2r(r2, r1);
+    movq_m2r(*(input_data + 52), r6);
+    por_r2r(r3, r5);
+    pmullw_m2r(*(input_data + 52), r6);
+    psrlq_i2r(16, r0);
+    movq_r2r(r4, r7);
+    movq_r2r(r2, r3);
+    psllq_i2r(48, r7);
+    pand_r2r(r0, r3);
+    pxor_r2r(r3, r0);
+    psllq_i2r(32, r3);
+    por_r2r(r5, r7);
+    movq_r2r(r6, r5);
+    pand_m2r(*M(1), r6);
+    por_r2r(r3, r7);
+    psllq_i2r(32, r6);
+    por_r2r(r1, r0);
+    movq_r2m(r7, *(output_data + 20));
+
+    por_r2r(r6, r0);
+    movq_m2r(*(input_data + 60), r7);
+    movq_r2r(r5, r6);
+    pmullw_m2r(*(input_data + 60), r7);
+    psrlq_i2r(32, r5);
+    pand_r2r(r2, r6);
+    movq_r2r(r5, r1);
+    movq_r2m(r0, *(output_data + 28));
+
+    pand_r2r(r2, r1);
+    movq_m2r(*(input_data + 56), r0);
+    movq_r2r(r7, r3);
+    pmullw_m2r(*(dequant_matrix + 56), r0);
+    psllq_i2r(16, r3);
+    pand_m2r(*M(3), r7);
+    pxor_r2r(r1, r5);
+    por_r2r(r5, r6);
+    movq_r2r(r3, r5);
+    pand_m2r(*M(3), r5);
+    por_r2r(r1, r7);
+    movq_m2r(*(input_data + 48), r1);
+    pxor_r2r(r5, r3);
+    pmullw_m2r(*(dequant_matrix + 48), r1);
+    por_r2r(r3, r7);
+    por_r2r(r5, r6);
+    movq_r2r(r0, r5);
+    movq_r2m(r7, *(output_data + 60));
+
+    psrlq_i2r(16, r5);
+    pand_m2r(*M(2), r5);
+    movq_r2r(r0, r7);
+    por_r2r(r5, r6);
+    pand_r2r(r2, r0);
+    pxor_r2r(r0, r7);
+    psllq_i2r(32, r0);
+    movq_r2m(r6, *(output_data + 52));
+
+    psrlq_i2r(16, r4);
+    movq_m2r(*(input_data + 36), r5);
+    psllq_i2r(16, r7);
+    pmullw_m2r(*(dequant_matrix + 36), r5);
+    movq_r2r(r7, r6);
+    movq_m2r(*M(2), r3);
+    psllq_i2r(16, r6);
+    pand_m2r(*M(3), r7);
+    pand_r2r(r1, r3);
+    por_r2r(r0, r7);
+    movq_r2r(r1, r0);
+    pand_m2r(*M(3), r1);
+    por_r2r(r3, r6);
+    movq_r2r(r4, r3);
+    psrlq_i2r(32, r1);
+    pand_r2r(r2, r3);
+    por_r2r(r1, r7);
+    por_r2r(r3, r7);
+    movq_r2r(r4, r3);
+    pand_m2r(*M(1), r3);
+    movq_r2r(r5, r1);
+    movq_r2m(r7, *(output_data + 44));
+
+    psrlq_i2r(48, r5);
+    movq_m2r(*(input_data + 32), r7);
+    por_r2r(r3, r6);
+    pmullw_m2r(*(dequant_matrix + 32), r7);
+    por_r2r(r5, r6);
+    pand_m2r(*M(2), r4);
+    psllq_i2r(32, r0);
+    movq_r2m(r6, *(output_data + 36));
+
+    movq_r2r(r0, r6);
+    pand_m2r(*M(3), r0);
+    psllq_i2r(16, r6);
+    movq_m2r(*(input_data + 20), r5);
+    movq_r2r(r1, r3);
+    pmullw_m2r(*(dequant_matrix + 40), r5);
+    psrlq_i2r(16, r1);
+    pand_m2r(*M(1), r1);
+    por_r2r(r4, r0);
+    pand_r2r(r7, r2);
+    por_r2r(r1, r0);
+    por_r2r(r2, r0);
+    psllq_i2r(16, r3);
+    movq_r2r(r3, r4);
+    movq_r2r(r5, r2);
+    movq_r2m(r0, *(output_data + 56));
+
+    psrlq_i2r(48, r2);
+    pand_m2r(*M(2), r4);
+    por_r2r(r2, r6);
+    movq_m2r(*M(1), r2);
+    por_r2r(r4, r6);
+    pand_r2r(r7, r2);
+    psllq_i2r(32, r3);
+    por_m2r(*(output_data + 40), r3);
+
+    por_r2r(r2, r6);
+    movq_m2r(*M(3), r2);
+    psllq_i2r(16, r5);
+    movq_r2m(r6, *(output_data + 48));
+
+    pand_r2r(r5, r2);
+    movq_m2r(*M(2), r6);
+    pxor_r2r(r2, r5);
+    pand_r2r(r7, r6);
+    psrlq_i2r(32, r2);
+    pand_m2r(*M(3), r7);
+    por_r2r(r2, r3);
+    por_m2r(*(output_data + 32), r7);
+
+    por_r2r(r3, r6);
+    por_r2r(r5, r7);
+    movq_r2m(r6, *(output_data + 40));
+    movq_r2m(r7, *(output_data + 32));
+
+
+#undef M
+
+    /* at this point, function has completed dequantization + dezigzag + 
+     * partial transposition; now do the idct itself */
+
+#define I(K) (output_data + K * 8)
+#define J(K) (output_data + ((K - 4) * 8) + 4)
+
+    RowIDCT();
+    Transpose();
+
+#undef I
+#undef J
+#define I(K) (output_data + (K * 8) + 32)
+#define J(K) (output_data + ((K - 4) * 8) + 36)
+
+    RowIDCT();
+    Transpose();
+
+#undef I
+#undef J
+#define I(K) (output_data + K * 8)
+#define J(K) (output_data + K * 8)
+
+    ColumnIDCT();
+
+#undef I
+#undef J
+#define I(K) (output_data + (K * 8) + 4)
+#define J(K) (output_data + (K * 8) + 4)
+
+    ColumnIDCT();
+
+#undef I
+#undef J
+
+}
+
+void vp3_idct_put_mmx(int16_t *input_data, int16_t *dequant_matrix,
+    int coeff_count, uint8_t *dest, int stride)
+{
+    int16_t transformed_data[64];
+    int16_t *op;
+    int i, j;
+    uint8_t vector128[8] = { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };
+
+    vp3_idct_mmx(input_data, dequant_matrix, transformed_data);
+
+    /* place in final output */
+    op = transformed_data;
+    movq_m2r(*vector128, mm0);
+    for (i = 0; i < 8; i++) {
+#if 1
+        for (j = 0; j < 8; j++) {
+            if (*op < -128)
+                *dest = 0;
+            else if (*op > 127)
+                *dest = 255;
+            else
+                *dest = (uint8_t)(*op + 128);
+            op++;
+            dest++;
+        }
+        dest += (stride - 8);
+#else
+/* prototype optimization */
+        pxor_r2r(mm1, mm1);
+        packsswb_m2r(*(op + 4), mm1);
+        movq_r2r(mm1, mm2);
+        psrlq_i2r(32, mm2);
+        packsswb_m2r(*(op + 0), mm1);
+        op += 8;
+        por_r2r(mm2, mm1);
+        paddb_r2r(mm0, mm1);
+        movq_r2m(mm1, *dest);
+        dest += stride;
+#endif
+    }
+
+    /* be a good MMX citizen */
+    emms();
+}
+
+void vp3_idct_add_mmx(int16_t *input_data, int16_t *dequant_matrix,
+    int coeff_count, uint8_t *dest, int stride)
+{
+    int16_t transformed_data[64];
+    int16_t *op;
+    int i, j;
+    int16_t sample;
+
+    vp3_idct_mmx(input_data, dequant_matrix, transformed_data);
+
+    /* place in final output */
+    op = transformed_data;
+    for (i = 0; i < 8; i++) {
+        for (j = 0; j < 8; j++) {
+            sample = *dest + *op;
+            if (sample < 0)
+                *dest = 0;
+            else if (sample > 255)
+                *dest = 255;
+            else
+                *dest = (uint8_t)(sample & 0xFF);
+            op++;
+            dest++;
+        }
+        dest += (stride - 8);
+    }
+
+    /* be a good MMX citizen */
+    emms();
+}
diff --git a/src/libffmpeg/libavcodec/qtrle.c b/src/libffmpeg/libavcodec/qtrle.c
new file mode 100644
index 000000000..55fa98663
--- /dev/null
+++ b/src/libffmpeg/libavcodec/qtrle.c
@@ -0,0 +1,535 @@
+/*
+ * Quicktime Animation (RLE) Video Decoder
+ * Copyright (C) 2004 the ffmpeg project
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ *
+ */
+
+/**
+ * @file qtrle.c
+ * QT RLE Video Decoder by Mike Melanson (melanson@pcisys.net)
+ * For more information about the QT RLE format, visit:
+ *   http://www.pcisys.net/~melanson/codecs/
+ *
+ * The QT RLE decoder has seven modes of operation:
+ * 1, 2, 4, 8, 16, 24, and 32 bits per pixel. For modes 1, 2, 4, and 8
+ * the decoder outputs PAL8 colorspace data. 16-bit data yields RGB555
+ * data. 24-bit data is RGB24 and 32-bit data is RGBA32.
+ */
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <unistd.h>
+
+#include "common.h"
+#include "avcodec.h"
+#include "dsputil.h"
+
+typedef struct QtrleContext {
+
+    AVCodecContext *avctx;
+    DSPContext dsp;
+    AVFrame frame;
+
+    unsigned char *buf;
+    int size;
+
+} QtrleContext;
+
+#define CHECK_STREAM_PTR(n) \
+  if ((stream_ptr + n) > s->size) { \
+    av_log (s->avctx, AV_LOG_INFO, "Problem: stream_ptr out of bounds (%d >= %d)\n", \
+      stream_ptr + n, s->size); \
+    return; \
+  }
+
+#define CHECK_PIXEL_PTR(n) \
+  if (pixel_ptr + n > pixel_limit) { \
+    av_log (s->avctx, AV_LOG_INFO, "Problem: pixel_ptr >= pixel_limit (%d >= %d)\n", \
+      pixel_ptr + n, pixel_limit); \
+    return; \
+  } \
+
+static void qtrle_decode_1bpp(QtrleContext *s)
+{
+}
+
+static void qtrle_decode_2bpp(QtrleContext *s)
+{
+}
+
+static void qtrle_decode_4bpp(QtrleContext *s)
+{
+}
+
+static void qtrle_decode_8bpp(QtrleContext *s)
+{
+    int stream_ptr;
+    int header;
+    int start_line;
+    int lines_to_change;
+    int rle_code;
+    int row_ptr, pixel_ptr;
+    int row_inc = s->frame.linesize[0];
+    unsigned char pi1, pi2, pi3, pi4;  /* 4 palette indices */
+    unsigned char *rgb = s->frame.data[0];
+    int pixel_limit = s->frame.linesize[0] * s->avctx->height;
+
+    /* check if this frame is even supposed to change */
+    if (s->size < 8)
+        return;
+
+    /* start after the chunk size */
+    stream_ptr = 4;
+
+    /* fetch the header */
+    CHECK_STREAM_PTR(2);
+    header = BE_16(&s->buf[stream_ptr]);
+    stream_ptr += 2;
+
+    /* if a header is present, fetch additional decoding parameters */
+    if (header & 0x0008) {
+        CHECK_STREAM_PTR(8);
+        start_line = BE_16(&s->buf[stream_ptr]);
+        stream_ptr += 4;
+        lines_to_change = BE_16(&s->buf[stream_ptr]);
+        stream_ptr += 4;
+    } else {
+        start_line = 0;
+        lines_to_change = s->avctx->height;
+    }
+
+    row_ptr = row_inc * start_line;
+    while (lines_to_change--) {
+        CHECK_STREAM_PTR(2);
+        pixel_ptr = row_ptr + (4 * (s->buf[stream_ptr++] - 1));
+
+        while ((rle_code = (signed char)s->buf[stream_ptr++]) != -1) {
+            if (rle_code == 0) {
+                /* there's another skip code in the stream */
+                CHECK_STREAM_PTR(1);
+                pixel_ptr += (4 * (s->buf[stream_ptr++] - 1));
+            } else if (rle_code < 0) {
+                /* decode the run length code */
+                rle_code = -rle_code;
+                /* get the next 4 bytes from the stream, treat them as palette
+                 * indices, and output them rle_code times */
+                CHECK_STREAM_PTR(4);
+                pi1 = s->buf[stream_ptr++];
+                pi2 = s->buf[stream_ptr++];
+                pi3 = s->buf[stream_ptr++];
+                pi4 = s->buf[stream_ptr++];
+
+                CHECK_PIXEL_PTR(rle_code * 4);
+
+                while (rle_code--) {
+                    rgb[pixel_ptr++] = pi1;
+                    rgb[pixel_ptr++] = pi2;
+                    rgb[pixel_ptr++] = pi3;
+                    rgb[pixel_ptr++] = pi4;
+                }
+            } else {
+                /* copy the same pixel directly to output 4 times */
+                rle_code *= 4;
+                CHECK_STREAM_PTR(rle_code);
+                CHECK_PIXEL_PTR(rle_code);
+
+                while (rle_code--) {
+                    rgb[pixel_ptr++] = s->buf[stream_ptr++];
+                }
+            }
+        }
+        row_ptr += row_inc;
+    }
+}
+
+static void qtrle_decode_16bpp(QtrleContext *s)
+{
+    int stream_ptr;
+    int header;
+    int start_line;
+    int lines_to_change;
+    signed char rle_code;
+    int row_ptr, pixel_ptr;
+    int row_inc = s->frame.linesize[0];
+    unsigned short rgb16;
+    unsigned char *rgb = s->frame.data[0];
+    int pixel_limit = s->frame.linesize[0] * s->avctx->height;
+
+    /* check if this frame is even supposed to change */
+    if (s->size < 8)
+        return;
+
+    /* start after the chunk size */
+    stream_ptr = 4;
+
+    /* fetch the header */
+    CHECK_STREAM_PTR(2);
+    header = BE_16(&s->buf[stream_ptr]);
+    stream_ptr += 2;
+
+    /* if a header is present, fetch additional decoding parameters */
+    if (header & 0x0008) {
+        CHECK_STREAM_PTR(8);
+        start_line = BE_16(&s->buf[stream_ptr]);
+        stream_ptr += 4;
+        lines_to_change = BE_16(&s->buf[stream_ptr]);
+        stream_ptr += 4;
+    } else {
+        start_line = 0;
+        lines_to_change = s->avctx->height;
+    }
+
+    row_ptr = row_inc * start_line;
+    while (lines_to_change--) {
+        CHECK_STREAM_PTR(2);
+        pixel_ptr = row_ptr + (s->buf[stream_ptr++] - 1) * 2;
+
+        while ((rle_code = (signed char)s->buf[stream_ptr++]) != -1) {
+            if (rle_code == 0) {
+                /* there's another skip code in the stream */
+                CHECK_STREAM_PTR(1);
+                pixel_ptr += (s->buf[stream_ptr++] - 1) * 2;
+            } else if (rle_code < 0) {
+                /* decode the run length code */
+                rle_code = -rle_code;
+                CHECK_STREAM_PTR(2);
+                rgb16 = BE_16(&s->buf[stream_ptr]);
+                stream_ptr += 2;
+
+                CHECK_PIXEL_PTR(rle_code * 2);
+
+                while (rle_code--) {
+                    *(unsigned short *)(&rgb[pixel_ptr]) = rgb16;
+                    pixel_ptr += 2;
+                }
+            } else {
+                CHECK_STREAM_PTR(rle_code * 2);
+                CHECK_PIXEL_PTR(rle_code * 2);
+
+                /* copy pixels directly to output */
+                while (rle_code--) {
+                    rgb16 = BE_16(&s->buf[stream_ptr]);
+                    stream_ptr += 2;
+                    *(unsigned short *)(&rgb[pixel_ptr]) = rgb16;
+                    pixel_ptr += 2;
+                }
+            }
+        }
+        row_ptr += row_inc;
+    }
+}
+
+static void qtrle_decode_24bpp(QtrleContext *s)
+{
+    int stream_ptr;
+    int header;
+    int start_line;
+    int lines_to_change;
+    signed char rle_code;
+    int row_ptr, pixel_ptr;
+    int row_inc = s->frame.linesize[0];
+    unsigned char r, g, b;
+    unsigned char *rgb = s->frame.data[0];
+    int pixel_limit = s->frame.linesize[0] * s->avctx->height;
+
+    /* check if this frame is even supposed to change */
+    if (s->size < 8)
+        return;
+
+    /* start after the chunk size */
+    stream_ptr = 4;
+
+    /* fetch the header */
+    CHECK_STREAM_PTR(2);
+    header = BE_16(&s->buf[stream_ptr]);
+    stream_ptr += 2;
+
+    /* if a header is present, fetch additional decoding parameters */
+    if (header & 0x0008) {
+        CHECK_STREAM_PTR(8);
+        start_line = BE_16(&s->buf[stream_ptr]);
+        stream_ptr += 4;
+        lines_to_change = BE_16(&s->buf[stream_ptr]);
+        stream_ptr += 4;
+    } else {
+        start_line = 0;
+        lines_to_change = s->avctx->height;
+    }
+
+    row_ptr = row_inc * start_line;
+    while (lines_to_change--) {
+        CHECK_STREAM_PTR(2);
+        pixel_ptr = row_ptr + (s->buf[stream_ptr++] - 1) * 3;
+
+        while ((rle_code = (signed char)s->buf[stream_ptr++]) != -1) {
+            if (rle_code == 0) {
+                /* there's another skip code in the stream */
+                CHECK_STREAM_PTR(1);
+                pixel_ptr += (s->buf[stream_ptr++] - 1) * 3;
+            } else if (rle_code < 0) {
+                /* decode the run length code */
+                rle_code = -rle_code;
+                CHECK_STREAM_PTR(3);
+                r = s->buf[stream_ptr++];
+                g = s->buf[stream_ptr++];
+                b = s->buf[stream_ptr++];
+
+                CHECK_PIXEL_PTR(rle_code * 3);
+
+                while (rle_code--) {
+                    rgb[pixel_ptr++] = r;
+                    rgb[pixel_ptr++] = g;
+                    rgb[pixel_ptr++] = b;
+                }
+            } else {
+                CHECK_STREAM_PTR(rle_code * 3);
+                CHECK_PIXEL_PTR(rle_code * 3);
+
+                /* copy pixels directly to output */
+                while (rle_code--) {
+                    rgb[pixel_ptr++] = s->buf[stream_ptr++];
+                    rgb[pixel_ptr++] = s->buf[stream_ptr++];
+                    rgb[pixel_ptr++] = s->buf[stream_ptr++];
+                }
+            }
+        }
+        row_ptr += row_inc;
+    }
+}
+
+static void qtrle_decode_32bpp(QtrleContext *s)
+{
+    int stream_ptr;
+    int header;
+    int start_line;
+    int lines_to_change;
+    signed char rle_code;
+    int row_ptr, pixel_ptr;
+    int row_inc = s->frame.linesize[0];
+    unsigned char r, g, b;
+    unsigned int argb;
+    unsigned char *rgb = s->frame.data[0];
+    int pixel_limit = s->frame.linesize[0] * s->avctx->height;
+
+    /* check if this frame is even supposed to change */
+    if (s->size < 8)
+        return;
+
+    /* start after the chunk size */
+    stream_ptr = 4;
+
+    /* fetch the header */
+    CHECK_STREAM_PTR(2);
+    header = BE_16(&s->buf[stream_ptr]);
+    stream_ptr += 2;
+
+    /* if a header is present, fetch additional decoding parameters */
+    if (header & 0x0008) {
+        CHECK_STREAM_PTR(8);
+        start_line = BE_16(&s->buf[stream_ptr]);
+        stream_ptr += 4;
+        lines_to_change = BE_16(&s->buf[stream_ptr]);
+        stream_ptr += 4;
+    } else {
+        start_line = 0;
+        lines_to_change = s->avctx->height;
+    }
+
+    row_ptr = row_inc * start_line;
+    while (lines_to_change--) {
+        CHECK_STREAM_PTR(2);
+        pixel_ptr = row_ptr + (s->buf[stream_ptr++] - 1) * 4;
+
+        while ((rle_code = (signed char)s->buf[stream_ptr++]) != -1) {
+            if (rle_code == 0) {
+                /* there's another skip code in the stream */
+                CHECK_STREAM_PTR(1);
+                pixel_ptr += (s->buf[stream_ptr++] - 1) * 4;
+            } else if (rle_code < 0) {
+                /* decode the run length code */
+                rle_code = -rle_code;
+                CHECK_STREAM_PTR(4);
+                stream_ptr++;  /* skip the alpha (?) byte */
+                r = s->buf[stream_ptr++];
+                g = s->buf[stream_ptr++];
+                b = s->buf[stream_ptr++];
+                argb = (r << 16) | (g << 8) | (b << 0);
+
+                CHECK_PIXEL_PTR(rle_code * 4);
+
+                while (rle_code--) {
+                    *(unsigned int *)(&rgb[pixel_ptr]) = argb;
+                    pixel_ptr += 4;
+                }
+            } else {
+                CHECK_STREAM_PTR(rle_code * 4);
+                CHECK_PIXEL_PTR(rle_code * 4);
+
+                /* copy pixels directly to output */
+                while (rle_code--) {
+                    stream_ptr++;  /* skip the alpha (?) byte */
+                    r = s->buf[stream_ptr++];
+                    g = s->buf[stream_ptr++];
+                    b = s->buf[stream_ptr++];
+                    argb = (r << 16) | (g << 8) | (b << 0);
+                    *(unsigned int *)(&rgb[pixel_ptr]) = argb;
+                    pixel_ptr += 4;
+                }
+            }
+        }
+        row_ptr += row_inc;
+    }
+}
+
+static int qtrle_decode_init(AVCodecContext *avctx)
+{
+    QtrleContext *s = (QtrleContext *)avctx->priv_data;
+
+    s->avctx = avctx;
+    switch (avctx->bits_per_sample) {
+    case 1:
+    case 2:
+    case 4:
+    case 8:
+    case 33:
+    case 34:
+    case 36:
+    case 40:
+        avctx->pix_fmt = PIX_FMT_PAL8;
+        break;
+
+    case 16:
+        avctx->pix_fmt = PIX_FMT_RGB555;
+        break;
+
+    case 24:
+        avctx->pix_fmt = PIX_FMT_RGB24;
+        break;
+
+    case 32:
+        avctx->pix_fmt = PIX_FMT_RGBA32;
+        break;
+
+    default:
+        av_log (avctx, AV_LOG_ERROR, "Unsupported colorspace: %d bits/sample?\n",
+            avctx->bits_per_sample);
+        break;
+    }
+    avctx->has_b_frames = 0;
+    dsputil_init(&s->dsp, avctx);
+
+    s->frame.data[0] = NULL;
+
+    return 0;
+}
+
+static int qtrle_decode_frame(AVCodecContext *avctx,
+                              void *data, int *data_size,
+                              uint8_t *buf, int buf_size)
+{
+    QtrleContext *s = (QtrleContext *)avctx->priv_data;
+
+    /* no supplementary picture */
+    if (buf_size == 0)
+        return 0;
+
+    s->buf = buf;
+    s->size = buf_size;
+
+    s->frame.reference = 1;
+    s->frame.buffer_hints = FF_BUFFER_HINTS_VALID | FF_BUFFER_HINTS_PRESERVE |
+                            FF_BUFFER_HINTS_REUSABLE | FF_BUFFER_HINTS_READABLE;
+    if (avctx->reget_buffer(avctx, &s->frame)) {
+        av_log (s->avctx, AV_LOG_ERROR, "reget_buffer() failed\n");
+        return -1;
+    }
+
+    switch (avctx->bits_per_sample) {
+    case 1:
+    case 33:
+        qtrle_decode_1bpp(s);
+        break;
+
+    case 2:
+    case 34:
+        qtrle_decode_2bpp(s);
+        break;
+
+    case 4:
+    case 36:
+        qtrle_decode_4bpp(s);
+        break;
+
+    case 8:
+    case 40:
+        qtrle_decode_8bpp(s);
+        /* make the palette available on the way out */
+        memcpy(s->frame.data[1], s->avctx->palctrl->palette, AVPALETTE_SIZE);
+        if (s->avctx->palctrl->palette_changed) {
+            s->frame.palette_has_changed = 1;
+            s->avctx->palctrl->palette_changed = 0;
+        }
+        break;
+
+    case 16:
+        qtrle_decode_16bpp(s);
+        break;
+
+    case 24:
+        qtrle_decode_24bpp(s);
+        break;
+
+    case 32:
+        qtrle_decode_32bpp(s);
+        break;
+
+    default:
+        av_log (s->avctx, AV_LOG_ERROR, "Unsupported colorspace: %d bits/sample?\n",
+            avctx->bits_per_sample);
+        break;
+    }
+
+    *data_size = sizeof(AVFrame);
+    *(AVFrame*)data = s->frame;
+
+    /* always report that the buffer was completely consumed */
+    return buf_size;
+}
+
+static int qtrle_decode_end(AVCodecContext *avctx)
+{
+    QtrleContext *s = (QtrleContext *)avctx->priv_data;
+
+    if (s->frame.data[0])
+        avctx->release_buffer(avctx, &s->frame);
+
+    return 0;
+}
+
+AVCodec qtrle_decoder = {
+    "qtrle",
+    CODEC_TYPE_VIDEO,
+    CODEC_ID_QTRLE,
+    sizeof(QtrleContext),
+    qtrle_decode_init,
+    NULL,
+    qtrle_decode_end,
+    qtrle_decode_frame,
+    CODEC_CAP_DR1,
+};
+
diff --git a/src/libffmpeg/libavcodec/vp3dsp.c b/src/libffmpeg/libavcodec/vp3dsp.c
new file mode 100644
index 000000000..ec62d9456
--- /dev/null
+++ b/src/libffmpeg/libavcodec/vp3dsp.c
@@ -0,0 +1,318 @@
+/*
+ * Copyright (C) 2004 the ffmpeg project
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+/**
+ * @file vp3dsp.c
+ * Standard C DSP-oriented functions cribbed from the original VP3 
+ * source code.
+ */
+
+#include "common.h"
+#include "avcodec.h"
+#include "vp3data.h"
+
+#define IdctAdjustBeforeShift 8
+#define xC1S7 64277
+#define xC2S6 60547
+#define xC3S5 54491
+#define xC4S4 46341
+#define xC5S3 36410
+#define xC6S2 25080
+#define xC7S1 12785
+
+void vp3_dsp_init_c(void)
+{
+    /* nop */
+}
+
+static void vp3_idct_c(int32_t *dequantized_data, int16_t *output_data)
+{
+    int32_t *ip = dequantized_data;
+    int16_t *op = output_data;
+
+    int32_t A_, B_, C_, D_, _Ad, _Bd, _Cd, _Dd, E_, F_, G_, H_;
+    int32_t _Ed, _Gd, _Add, _Bdd, _Fd, _Hd;
+    int32_t t1, t2;
+
+    int i;
+
+    /* Inverse DCT on the rows now */
+    for (i = 0; i < 8; i++) {
+        /* Check for non-zero values */
+        if ( ip[0] | ip[1] | ip[2] | ip[3] | ip[4] | ip[5] | ip[6] | ip[7] ) {
+            t1 = (int32_t)(xC1S7 * ip[1]);
+            t2 = (int32_t)(xC7S1 * ip[7]);
+            t1 >>= 16;
+            t2 >>= 16;
+            A_ = t1 + t2;
+
+            t1 = (int32_t)(xC7S1 * ip[1]);
+            t2 = (int32_t)(xC1S7 * ip[7]);
+            t1 >>= 16;
+            t2 >>= 16;
+            B_ = t1 - t2;
+
+            t1 = (int32_t)(xC3S5 * ip[3]);
+            t2 = (int32_t)(xC5S3 * ip[5]);
+            t1 >>= 16;
+            t2 >>= 16;
+            C_ = t1 + t2;
+
+            t1 = (int32_t)(xC3S5 * ip[5]);
+            t2 = (int32_t)(xC5S3 * ip[3]);
+            t1 >>= 16;
+            t2 >>= 16;
+            D_ = t1 - t2;
+
+
+            t1 = (int32_t)(xC4S4 * (A_ - C_));
+            t1 >>= 16;
+            _Ad = t1;
+
+            t1 = (int32_t)(xC4S4 * (B_ - D_));
+            t1 >>= 16;
+            _Bd = t1;
+
+
+            _Cd = A_ + C_;
+            _Dd = B_ + D_;
+
+            t1 = (int32_t)(xC4S4 * (ip[0] + ip[4]));
+            t1 >>= 16;
+            E_ = t1;
+
+            t1 = (int32_t)(xC4S4 * (ip[0] - ip[4]));
+            t1 >>= 16;
+            F_ = t1;
+
+            t1 = (int32_t)(xC2S6 * ip[2]);
+            t2 = (int32_t)(xC6S2 * ip[6]);
+            t1 >>= 16;
+            t2 >>= 16;
+            G_ = t1 + t2;
+
+            t1 = (int32_t)(xC6S2 * ip[2]);
+            t2 = (int32_t)(xC2S6 * ip[6]);
+            t1 >>= 16;
+            t2 >>= 16;
+            H_ = t1 - t2;
+
+
+            _Ed = E_ - G_;
+            _Gd = E_ + G_;
+
+            _Add = F_ + _Ad;
+            _Bdd = _Bd - H_;
+
+            _Fd = F_ - _Ad;
+            _Hd = _Bd + H_;
+
+            /*  Final sequence of operations over-write original inputs. */
+            ip[0] = (int16_t)((_Gd + _Cd )   >> 0);
+            ip[7] = (int16_t)((_Gd - _Cd )   >> 0);
+
+            ip[1] = (int16_t)((_Add + _Hd )  >> 0);
+            ip[2] = (int16_t)((_Add - _Hd )  >> 0);
+
+            ip[3] = (int16_t)((_Ed + _Dd )   >> 0);
+            ip[4] = (int16_t)((_Ed - _Dd )   >> 0);
+
+            ip[5] = (int16_t)((_Fd + _Bdd )  >> 0);
+            ip[6] = (int16_t)((_Fd - _Bdd )  >> 0);
+
+        }
+
+        ip += 8;            /* next row */
+    }
+
+    ip = dequantized_data;
+
+    for ( i = 0; i < 8; i++) {
+        /* Check for non-zero values (bitwise or faster than ||) */
+        if ( ip[0 * 8] | ip[1 * 8] | ip[2 * 8] | ip[3 * 8] |
+             ip[4 * 8] | ip[5 * 8] | ip[6 * 8] | ip[7 * 8] ) {
+
+            t1 = (int32_t)(xC1S7 * ip[1*8]);
+            t2 = (int32_t)(xC7S1 * ip[7*8]);
+            t1 >>= 16;
+            t2 >>= 16;
+            A_ = t1 + t2;
+
+            t1 = (int32_t)(xC7S1 * ip[1*8]);
+            t2 = (int32_t)(xC1S7 * ip[7*8]);
+            t1 >>= 16;
+            t2 >>= 16;
+            B_ = t1 - t2;
+
+            t1 = (int32_t)(xC3S5 * ip[3*8]);
+            t2 = (int32_t)(xC5S3 * ip[5*8]);
+            t1 >>= 16;
+            t2 >>= 16;
+            C_ = t1 + t2;
+
+            t1 = (int32_t)(xC3S5 * ip[5*8]);
+            t2 = (int32_t)(xC5S3 * ip[3*8]);
+            t1 >>= 16;
+            t2 >>= 16;
+            D_ = t1 - t2;
+
+
+            t1 = (int32_t)(xC4S4 * (A_ - C_));
+            t1 >>= 16;
+            _Ad = t1;
+
+            t1 = (int32_t)(xC4S4 * (B_ - D_));
+            t1 >>= 16;
+            _Bd = t1;
+
+
+            _Cd = A_ + C_;
+            _Dd = B_ + D_;
+
+            t1 = (int32_t)(xC4S4 * (ip[0*8] + ip[4*8]));
+            t1 >>= 16;
+            E_ = t1;
+
+            t1 = (int32_t)(xC4S4 * (ip[0*8] - ip[4*8]));
+            t1 >>= 16;
+            F_ = t1;
+
+            t1 = (int32_t)(xC2S6 * ip[2*8]);
+            t2 = (int32_t)(xC6S2 * ip[6*8]);
+            t1 >>= 16;
+            t2 >>= 16;
+            G_ = t1 + t2;
+
+            t1 = (int32_t)(xC6S2 * ip[2*8]);
+            t2 = (int32_t)(xC2S6 * ip[6*8]);
+            t1 >>= 16;
+            t2 >>= 16;
+            H_ = t1 - t2;
+
+
+            _Ed = E_ - G_;
+            _Gd = E_ + G_;
+
+            _Add = F_ + _Ad;
+            _Bdd = _Bd - H_;
+
+            _Fd = F_ - _Ad;
+            _Hd = _Bd + H_;
+
+            _Gd += IdctAdjustBeforeShift;
+            _Add += IdctAdjustBeforeShift;
+            _Ed += IdctAdjustBeforeShift;
+            _Fd += IdctAdjustBeforeShift;
+
+            /* Final sequence of operations over-write original inputs. */
+            op[0*8] = (int16_t)((_Gd + _Cd )   >> 4);
+            op[7*8] = (int16_t)((_Gd - _Cd )   >> 4);
+
+            op[1*8] = (int16_t)((_Add + _Hd )  >> 4);
+            op[2*8] = (int16_t)((_Add - _Hd )  >> 4);
+
+            op[3*8] = (int16_t)((_Ed + _Dd )   >> 4);
+            op[4*8] = (int16_t)((_Ed - _Dd )   >> 4);
+
+            op[5*8] = (int16_t)((_Fd + _Bdd )  >> 4);
+            op[6*8] = (int16_t)((_Fd - _Bdd )  >> 4);
+
+        } else {
+
+            op[0*8] = 0;
+            op[7*8] = 0;
+            op[1*8] = 0;
+            op[2*8] = 0;
+            op[3*8] = 0;
+            op[4*8] = 0;
+            op[5*8] = 0;
+            op[6*8] = 0;
+        }
+
+        ip++;            /* next column */
+        op++;
+    }
+}
+
+void vp3_idct_put_c(int16_t *input_data, int16_t *dequant_matrix,
+    int coeff_count, uint8_t *dest, int stride)
+{
+    int32_t dequantized_data[64];
+    int16_t transformed_data[64];
+    int16_t *op;
+    int i, j;
+
+    /* de-zigzag and dequantize */
+    for (i = 0; i < coeff_count; i++) {
+        j = dezigzag_index[i];
+        dequantized_data[j] = dequant_matrix[i] * input_data[i];
+    }
+
+    vp3_idct_c(dequantized_data, transformed_data);
+
+    /* place in final output */
+    op = transformed_data;
+    for (i = 0; i < 8; i++) {
+        for (j = 0; j < 8; j++) {
+            if (*op < -128)
+                *dest = 0;
+            else if (*op > 127)
+                *dest = 255;
+            else
+                *dest = (uint8_t)(*op + 128);
+            op++;
+            dest++;
+        }
+        dest += (stride - 8);
+    }
+}
+
+void vp3_idct_add_c(int16_t *input_data, int16_t *dequant_matrix,
+    int coeff_count, uint8_t *dest, int stride)
+{
+    int32_t dequantized_data[64];
+    int16_t transformed_data[64];
+    int16_t *op;
+    int i, j;
+    int16_t sample;
+
+    /* de-zigzag and dequantize */
+    for (i = 0; i < coeff_count; i++) {
+        j = dezigzag_index[i];
+        dequantized_data[j] = dequant_matrix[i] * input_data[i];
+    }
+
+    vp3_idct_c(dequantized_data, transformed_data);
+
+    /* place in final output */
+    op = transformed_data;
+    for (i = 0; i < 8; i++) {
+        for (j = 0; j < 8; j++) {
+            sample = *dest + *op;
+            if (sample < 0)
+                *dest = 0;
+            else if (sample > 255)
+                *dest = 255;
+            else
+                *dest = (uint8_t)(sample & 0xFF);
+            op++;
+            dest++;
+        }
+        dest += (stride - 8);
+    }
+}