indeo4: add missing Haar and slanted transforms

That involves fixing INV_HAAR4() macro and changing it to work with different
input and output like INV_HAAR8() instead of in-place transform.

3 files changed, 294 insertions, 13 deletions

diff --git a/libavcodec/indeo4.c b/libavcodec/indeo4.c
index 8820a2c813..eb34688775 100644
--- a/libavcodec/indeo4.c
+++ b/libavcodec/indeo4.c
@@ -56,8 +56,8 @@ static const struct {
     int             is_2d_trans;
 } transforms[18] = {
     { ff_ivi_inverse_haar_8x8,  ff_ivi_dc_haar_2d,       1 },
-    { NULL, NULL, 0 }, /* inverse Haar 8x1 */
-    { NULL, NULL, 0 }, /* inverse Haar 1x8 */
+    { ff_ivi_row_haar8,         ff_ivi_dc_haar_2d,       0 },
+    { ff_ivi_col_haar8,         ff_ivi_dc_haar_2d,       0 },
     { ff_ivi_put_pixels_8x8,    ff_ivi_put_dc_pixel_8x8, 1 },
     { ff_ivi_inverse_slant_8x8, ff_ivi_dc_slant_2d,      1 },
     { ff_ivi_row_slant8,        ff_ivi_dc_row_slant,     1 },
@@ -65,13 +65,13 @@ static const struct {
     { NULL, NULL, 0 }, /* inverse DCT 8x8 */
     { NULL, NULL, 0 }, /* inverse DCT 8x1 */
     { NULL, NULL, 0 }, /* inverse DCT 1x8 */
-    { NULL, NULL, 0 }, /* inverse Haar 4x4 */
+    { ff_ivi_inverse_haar_4x4,  ff_ivi_dc_haar_2d,       1 },
     { ff_ivi_inverse_slant_4x4, ff_ivi_dc_slant_2d,      1 },
     { NULL, NULL, 0 }, /* no transform 4x4 */
-    { NULL, NULL, 0 }, /* inverse Haar 1x4 */
-    { NULL, NULL, 0 }, /* inverse Haar 4x1 */
-    { NULL, NULL, 0 }, /* inverse slant 1x4 */
-    { NULL, NULL, 0 }, /* inverse slant 4x1 */
+    { ff_ivi_row_haar4,         ff_ivi_dc_haar_2d,       0 },
+    { ff_ivi_col_haar4,         ff_ivi_dc_haar_2d,       0 },
+    { ff_ivi_row_slant4,        ff_ivi_dc_row_slant,     0 },
+    { ff_ivi_col_slant4,        ff_ivi_dc_col_slant,     0 },
     { NULL, NULL, 0 }, /* inverse DCT 4x4 */
 };
 
diff --git a/libavcodec/ivi_dsp.c b/libavcodec/ivi_dsp.c
index f5efa6521d..bd1f523f3e 100644
--- a/libavcodec/ivi_dsp.c
+++ b/libavcodec/ivi_dsp.c
@@ -248,12 +248,14 @@ void ff_ivi_recompose_haar(const IVIPlaneDesc *plane, uint8_t *dst,
     d8 = COMPENSATE(t8); }
 
 /** inverse 4-point Haar transform */
-#define INV_HAAR4(s1, s3, s5, s7) {\
-    HAAR_BFLY(s1, s5);  HAAR_BFLY(s1, s3);  HAAR_BFLY(s5, s7);\
-    s1 = COMPENSATE(s1);\
-    s3 = COMPENSATE(s3);\
-    s5 = COMPENSATE(s5);\
-    s7 = COMPENSATE(s7); }
+#define INV_HAAR4(s1, s3, s5, s7, d1, d2, d3, d4, t0, t1, t2, t3, t4) {\
+    IVI_HAAR_BFLY(s1, s3, t0, t1, t4);\
+    IVI_HAAR_BFLY(t0, s5, t2, t3, t4);\
+    d1 = COMPENSATE(t2);\
+    d2 = COMPENSATE(t3);\
+    IVI_HAAR_BFLY(t1, s7, t2, t3, t4);\
+    d3 = COMPENSATE(t2);\
+    d4 = COMPENSATE(t3); }
 
 void ff_ivi_inverse_haar_8x8(const int32_t *in, int16_t *out, uint32_t pitch,
                              const uint8_t *flags)
@@ -310,6 +312,153 @@ void ff_ivi_inverse_haar_8x8(const int32_t *in, int16_t *out, uint32_t pitch,
 #undef  COMPENSATE
 }
 
+void ff_ivi_row_haar8(const int32_t *in, int16_t *out, uint32_t pitch,
+                      const uint8_t *flags)
+{
+    int     i;
+    int     t0, t1, t2, t3, t4, t5, t6, t7, t8;
+
+    /* apply the InvHaar8 to all rows */
+#define COMPENSATE(x) (x)
+    for (i = 0; i < 8; i++) {
+        if (   !in[0] && !in[1] && !in[2] && !in[3]
+            && !in[4] && !in[5] && !in[6] && !in[7]) {
+            memset(out, 0, 8 * sizeof(out[0]));
+        } else {
+            INV_HAAR8(in[0],  in[1],  in[2],  in[3],
+                      in[4],  in[5],  in[6],  in[7],
+                      out[0], out[1], out[2], out[3],
+                      out[4], out[5], out[6], out[7],
+                      t0, t1, t2, t3, t4, t5, t6, t7, t8);
+        }
+        in  += 8;
+        out += pitch;
+    }
+#undef  COMPENSATE
+}
+
+void ff_ivi_col_haar8(const int32_t *in, int16_t *out, uint32_t pitch,
+                      const uint8_t *flags)
+{
+    int     i;
+    int     t0, t1, t2, t3, t4, t5, t6, t7, t8;
+
+    /* apply the InvHaar8 to all columns */
+#define COMPENSATE(x) (x)
+    for (i = 0; i < 8; i++) {
+        if (flags[i]) {
+            INV_HAAR8(in[ 0], in[ 8], in[16], in[24],
+                      in[32], in[40], in[48], in[56],
+                      out[0 * pitch], out[1 * pitch],
+                      out[2 * pitch], out[3 * pitch],
+                      out[4 * pitch], out[5 * pitch],
+                      out[6 * pitch], out[7 * pitch],
+                      t0, t1, t2, t3, t4, t5, t6, t7, t8);
+        } else
+            out[0 * pitch] = out[1 * pitch] =
+            out[2 * pitch] = out[3 * pitch] =
+            out[4 * pitch] = out[5 * pitch] =
+            out[6 * pitch] = out[7 * pitch] = 0;
+
+        in++;
+        out++;
+    }
+#undef  COMPENSATE
+}
+
+void ff_ivi_inverse_haar_4x4(const int32_t *in, int16_t *out, uint32_t pitch,
+                             const uint8_t *flags)
+{
+    int     i, shift, sp1, sp2;
+    const int32_t *src;
+    int32_t *dst;
+    int     tmp[16];
+    int     t0, t1, t2, t3, t4;
+
+    /* apply the InvHaar4 to all columns */
+#define COMPENSATE(x) (x)
+    src = in;
+    dst = tmp;
+    for (i = 0; i < 4; i++) {
+        if (flags[i]) {
+            /* pre-scaling */
+            shift = !(i & 2);
+            sp1 = src[0] << shift;
+            sp2 = src[4] << shift;
+            INV_HAAR4(   sp1,    sp2, src[8], src[12],
+                      dst[0], dst[4], dst[8], dst[12],
+                      t0, t1, t2, t3, t4);
+        } else
+            dst[0] = dst[4] = dst[8] = dst[12] = 0;
+
+        src++;
+        dst++;
+    }
+#undef  COMPENSATE
+
+    /* apply the InvHaar8 to all rows */
+#define COMPENSATE(x) (x)
+    src = tmp;
+    for (i = 0; i < 4; i++) {
+        if (!src[0] && !src[1] && !src[2] && !src[3]) {
+            memset(out, 0, 4 * sizeof(out[0]));
+        } else {
+            INV_HAAR4(src[0], src[1], src[2], src[3],
+                      out[0], out[1], out[2], out[3],
+                      t0, t1, t2, t3, t4);
+        }
+        src += 4;
+        out += pitch;
+    }
+#undef  COMPENSATE
+}
+
+void ff_ivi_row_haar4(const int32_t *in, int16_t *out, uint32_t pitch,
+                      const uint8_t *flags)
+{
+    int     i;
+    int     t0, t1, t2, t3, t4;
+
+    /* apply the InvHaar4 to all rows */
+#define COMPENSATE(x) (x)
+    for (i = 0; i < 4; i++) {
+        if (!in[0] && !in[1] && !in[2] && !in[3]) {
+            memset(out, 0, 4 * sizeof(out[0]));
+        } else {
+            INV_HAAR4(in[0], in[1], in[2], in[3],
+                      out[0], out[1], out[2], out[3],
+                      t0, t1, t2, t3, t4);
+        }
+        in  += 4;
+        out += pitch;
+    }
+#undef  COMPENSATE
+}
+
+void ff_ivi_col_haar4(const int32_t *in, int16_t *out, uint32_t pitch,
+                      const uint8_t *flags)
+{
+    int     i;
+    int     t0, t1, t2, t3, t4;
+
+    /* apply the InvHaar8 to all columns */
+#define COMPENSATE(x) (x)
+    for (i = 0; i < 4; i++) {
+        if (flags[i]) {
+            INV_HAAR4(in[0], in[4], in[8], in[12],
+                      out[0 * pitch], out[1 * pitch],
+                      out[2 * pitch], out[3 * pitch],
+                      t0, t1, t2, t3, t4);
+        } else
+            out[0 * pitch] = out[1 * pitch] =
+            out[2 * pitch] = out[3 * pitch] = 0;
+
+        in++;
+        out++;
+    }
+#undef  COMPENSATE
+}
+
 void ff_ivi_dc_haar_2d(const int32_t *in, int16_t *out, uint32_t pitch,
                        int blk_size)
 {
@@ -546,6 +695,49 @@ void ff_ivi_dc_col_slant(const int32_t *in, int16_t *out, uint32_t pitch, int bl
     }
 }
 
+void ff_ivi_row_slant4(const int32_t *in, int16_t *out, uint32_t pitch, const uint8_t *flags)
+{
+    int     i;
+    int     t0, t1, t2, t3, t4;
+
+#define COMPENSATE(x) ((x + 1)>>1)
+    for (i = 0; i < 4; i++) {
+        if (!in[0] && !in[1] && !in[2] && !in[3]) {
+            memset(out, 0, 4*sizeof(out[0]));
+        } else {
+            IVI_INV_SLANT4( in[0],  in[1],  in[2],  in[3],
+                           out[0], out[1], out[2], out[3],
+                           t0, t1, t2, t3, t4);
+        }
+        in  += 4;
+        out += pitch;
+    }
+#undef COMPENSATE
+}
+
+void ff_ivi_col_slant4(const int32_t *in, int16_t *out, uint32_t pitch, const uint8_t *flags)
+{
+    int     i, row2;
+    int     t0, t1, t2, t3, t4;
+
+    row2 = pitch << 1;
+
+#define COMPENSATE(x) ((x + 1)>>1)
+    for (i = 0; i < 4; i++) {
+        if (flags[i]) {
+            IVI_INV_SLANT4(in[0], in[4], in[8], in[12],
+                           out[0], out[pitch], out[row2], out[row2 + pitch],
+                           t0, t1, t2, t3, t4);
+        } else {
+            out[0] = out[pitch] = out[row2] = out[row2 + pitch] = 0;
+        }
+
+        in++;
+        out++;
+    }
+#undef COMPENSATE
+}
+
 void ff_ivi_put_pixels_8x8(const int32_t *in, int16_t *out, uint32_t pitch,
                            const uint8_t *flags)
 {
diff --git a/libavcodec/ivi_dsp.h b/libavcodec/ivi_dsp.h
index c46f8b2d8d..31d37e3f9f 100644
--- a/libavcodec/ivi_dsp.h
+++ b/libavcodec/ivi_dsp.h
@@ -66,6 +66,71 @@ void ff_ivi_inverse_haar_8x8(const int32_t *in, int16_t *out, uint32_t pitch,
                              const uint8_t *flags);
 
 /**
+ *  one-dimensional inverse 8-point Haar transform on rows for Indeo 4
+ *
+ *  @param[in]  in        pointer to the vector of transform coefficients
+ *  @param[out] out       pointer to the output buffer (frame)
+ *  @param[in]  pitch     pitch to move to the next y line
+ *  @param[in]  flags     pointer to the array of column flags:
+ *                        != 0 - non_empty column, 0 - empty one
+ *                        (this array must be filled by caller)
+ */
+void ff_ivi_row_haar8(const int32_t *in, int16_t *out, uint32_t pitch,
+                      const uint8_t *flags);
+
+/**
+ *  one-dimensional inverse 8-point Haar transform on columns for Indeo 4
+ *
+ *  @param[in]  in        pointer to the vector of transform coefficients
+ *  @param[out] out       pointer to the output buffer (frame)
+ *  @param[in]  pitch     pitch to move to the next y line
+ *  @param[in]  flags     pointer to the array of column flags:
+ *                        != 0 - non_empty column, 0 - empty one
+ *                        (this array must be filled by caller)
+ */
+void ff_ivi_col_haar8(const int32_t *in, int16_t *out, uint32_t pitch,
+                      const uint8_t *flags);
+
+/**
+ *  two-dimensional inverse Haar 4x4 transform for Indeo 4
+ *
+ *  @param[in]  in        pointer to the vector of transform coefficients
+ *  @param[out] out       pointer to the output buffer (frame)
+ *  @param[in]  pitch     pitch to move to the next y line
+ *  @param[in]  flags     pointer to the array of column flags:
+ *                        != 0 - non_empty column, 0 - empty one
+ *                        (this array must be filled by caller)
+ */
+void ff_ivi_inverse_haar_4x4(const int32_t *in, int16_t *out, uint32_t pitch,
+                             const uint8_t *flags);
+
+/**
+ *  one-dimensional inverse 4-point Haar transform on rows for Indeo 4
+ *
+ *  @param[in]  in        pointer to the vector of transform coefficients
+ *  @param[out] out       pointer to the output buffer (frame)
+ *  @param[in]  pitch     pitch to move to the next y line
+ *  @param[in]  flags     pointer to the array of column flags:
+ *                        != 0 - non_empty column, 0 - empty one
+ *                        (this array must be filled by caller)
+ */
+void ff_ivi_row_haar4(const int32_t *in, int16_t *out, uint32_t pitch,
+                      const uint8_t *flags);
+
+/**
+ *  one-dimensional inverse 4-point Haar transform on columns for Indeo 4
+ *
+ *  @param[in]  in        pointer to the vector of transform coefficients
+ *  @param[out] out       pointer to the output buffer (frame)
+ *  @param[in]  pitch     pitch to move to the next y line
+ *  @param[in]  flags     pointer to the array of column flags:
+ *                        != 0 - non_empty column, 0 - empty one
+ *                        (this array must be filled by caller)
+ */
+void ff_ivi_col_haar4(const int32_t *in, int16_t *out, uint32_t pitch,
+                      const uint8_t *flags);
+
+/**
  *  DC-only two-dimensional inverse Haar transform for Indeo 4.
  *  Performing the inverse transform in this case is equivalent to
  *  spreading DC_coeff >> 3 over the whole block.
@@ -142,6 +207,30 @@ void ff_ivi_col_slant8(const int32_t *in, int16_t *out, uint32_t pitch,
                        const uint8_t *flags);
 
 /**
+ *  inverse 1D row slant transform
+ *
+ *  @param[in]    in      pointer to the vector of transform coefficients
+ *  @param[out]   out     pointer to the output buffer (frame)
+ *  @param[in]    pitch   pitch to move to the next y line
+ *  @param[in]    flags   pointer to the array of column flags (unused here)
+ */
+void ff_ivi_row_slant4(const int32_t *in, int16_t *out, uint32_t pitch,
+                       const uint8_t *flags);
+
+/**
+ *  inverse 1D column slant transform
+ *
+ *  @param[in]    in      pointer to the vector of transform coefficients
+ *  @param[out]   out     pointer to the output buffer (frame)
+ *  @param[in]    pitch   pitch to move to the next y line
+ *  @param[in]    flags   pointer to the array of column flags:
+ *                        != 0 - non_empty column, 0 - empty one
+ *                        (this array must be filled by caller)
+ */
+void ff_ivi_col_slant4(const int32_t *in, int16_t *out, uint32_t pitch,
+                       const uint8_t *flags);
+
+/**
  *  DC-only inverse row slant transform
  */
 void ff_ivi_dc_row_slant(const int32_t *in, int16_t *out, uint32_t pitch, int blk_size);