* liboil/liboilclasses.h:

* liboil/liboilfuncs-04.h:
* liboil/liboilfuncs.h:
* liboil/liboiltrampolines.c:
* liboil/ref/sad8x8.c:
Add 12x12 and 16x16 classes.

1 files changed, 144 insertions, 0 deletions

diff --git a/liboil/ref/sad8x8.c b/liboil/ref/sad8x8.c
index 2b97529..bcc1041 100644
--- a/liboil/ref/sad8x8.c
+++ b/liboil/ref/sad8x8.c
@@ -89,6 +89,64 @@ OIL_DEFINE_CLASS (sad8x8_f64_2,
 OIL_DEFINE_CLASS (sad8x8_8xn_u8,
     "uint32_t *d_n, uint8_t *s1_8x8, int ss1, uint8_t *s2_8xnp7, int ss2, int n");
 
+/**
+ * oil_sad12x12_u8:
+ * @d_1:
+ * @s1_12x12:
+ * @ss1:
+ * @s2_12x12:
+ * @ss2:
+ *
+ * Calculates the sum of absolute differences between elements in @s1_12x12
+ * and @s2_12x12, and places the result in @d_1.
+ */
+OIL_DEFINE_CLASS (sad12x12_u8,
+    "uint32_t *d_1, uint8_t *s1_12x12, int ss1, uint8_t *s2_12x12, int ss2");
+/**
+ * oil_sad12x12_12xn_u8:
+ * @d_1:
+ * @s1_12x12:
+ * @ss1:
+ * @s2_12xnp11:
+ * @ss2:
+ * @n:
+ *
+ * Calculates the sum of absolute differences between elements in @s1_12x12
+ * and the 12x12 block of array elements starting at row i in @s2_12x12, and
+ * places the result in @d_n.
+ */
+OIL_DEFINE_CLASS (sad12x12_12xn_u8,
+    "uint32_t *d_n, uint8_t *s1_12x12, int ss1, uint8_t *s2_12xnp11, int ss2, int n");
+/**
+ * oil_sad16x16_u8:
+ * @d_1:
+ * @s1_16x16:
+ * @ss1:
+ * @s2_16x16:
+ * @ss2:
+ *
+ * Calculates the sum of absolute differences between elements in @s1_16x16
+ * and @s2_16x16, and places the result in @d_1.
+ */
+OIL_DEFINE_CLASS (sad16x16_u8,
+    "uint32_t *d_1, uint8_t *s1_16x16, int ss1, uint8_t *s2_16x16, int ss2");
+/**
+ * oil_sad16x16_16xn_u8:
+ * @d_1:
+ * @s1_16x16:
+ * @ss1:
+ * @s2_16xnp15:
+ * @ss2:
+ * @n:
+ *
+ * Calculates the sum of absolute differences between elements in @s1_16x16
+ * and the 16x16 block of array elements starting at row i in @s2_16x16, and
+ * places the result in @d_n.
+ */
+OIL_DEFINE_CLASS (sad16x16_16xn_u8,
+    "uint32_t *d_n, uint8_t *s1_16x16, int ss1, uint8_t *s2_16xnp15, int ss2, int n");
+
+
 static void
 sad8x8_f64_2_ref (double *dest, double *src1, int sstr1, double *src2,
     int sstr2)
@@ -170,3 +228,89 @@ sad8x8_8xn_u8_ref (uint32_t * dest, uint8_t * src1, int sstr1, uint8_t * src2,
 }
 OIL_DEFINE_IMPL_REF (sad8x8_8xn_u8_ref, sad8x8_8xn_u8);
 
+
+static void
+sad12x12_u8_ref (uint32_t * dest, uint8_t * src1, int sstr1, uint8_t * src2,
+    int sstr2)
+{
+  int i, j;
+  int d;
+  uint32_t sum;
+
+  sum = 0;
+  for (i = 0; i < 12; i++) {
+    for (j = 0; j < 12; j++) {
+      d = ((int) OIL_GET (src1, sstr1 * i + j * sizeof (uint8_t), uint8_t)) -
+          ((int) OIL_GET (src2, sstr2 * i + j * sizeof (uint8_t), uint8_t));
+      sum += (d < 0) ? -d : d;
+    }
+  }
+  *dest = sum;
+}
+OIL_DEFINE_IMPL_REF (sad12x12_u8_ref, sad12x12_u8);
+
+static void
+sad12x12_12xn_u8_ref (uint32_t * dest, uint8_t * src1, int sstr1, uint8_t * src2,
+    int sstr2, int n)
+{
+  int i, j, k;
+  int d;
+  uint32_t sum;
+
+  for (i=0;i<n;i++){
+    sum = 0;
+    for (j = 0; j < 12; j++) {
+      for (k = 0; k < 12; k++) {
+        d = ((int) OIL_GET (src1, sstr1 * j + k * sizeof (uint8_t), uint8_t)) -
+            ((int) OIL_GET (src2, sstr2 * (i+j) + k * sizeof (uint8_t), uint8_t));
+        sum += (d < 0) ? -d : d;
+      }
+    }
+    dest[i] = sum;
+  }
+}
+OIL_DEFINE_IMPL_REF (sad12x12_12xn_u8_ref, sad12x12_12xn_u8);
+
+
+static void
+sad16x16_u8_ref (uint32_t * dest, uint8_t * src1, int sstr1, uint8_t * src2,
+    int sstr2)
+{
+  int i, j;
+  int d;
+  uint32_t sum;
+
+  sum = 0;
+  for (i = 0; i < 16; i++) {
+    for (j = 0; j < 16; j++) {
+      d = ((int) OIL_GET (src1, sstr1 * i + j * sizeof (uint8_t), uint8_t)) -
+          ((int) OIL_GET (src2, sstr2 * i + j * sizeof (uint8_t), uint8_t));
+      sum += (d < 0) ? -d : d;
+    }
+  }
+  *dest = sum;
+}
+OIL_DEFINE_IMPL_REF (sad16x16_u8_ref, sad16x16_u8);
+
+static void
+sad16x16_16xn_u8_ref (uint32_t * dest, uint8_t * src1, int sstr1, uint8_t * src2,
+    int sstr2, int n)
+{
+  int i, j, k;
+  int d;
+  uint32_t sum;
+
+  for (i=0;i<n;i++){
+    sum = 0;
+    for (j = 0; j < 16; j++) {
+      for (k = 0; k < 16; k++) {
+        d = ((int) OIL_GET (src1, sstr1 * j + k * sizeof (uint8_t), uint8_t)) -
+            ((int) OIL_GET (src2, sstr2 * (i+j) + k * sizeof (uint8_t), uint8_t));
+        sum += (d < 0) ? -d : d;
+      }
+    }
+    dest[i] = sum;
+  }
+}
+OIL_DEFINE_IMPL_REF (sad16x16_16xn_u8_ref, sad16x16_16xn_u8);
+