Add neon optimize vp9_short_idct10_16x16_add.

vp9_short_idct10_16x16_add is used to handle the block that only have valid data
at top left 4x4 block. All the other datas are 0. So we could cut many
unnecessary calculations in order to save instructions.

Change-Id: I6e30a3fee1ece5af7f258532416d0bfddd1143f0

vp9/common/arm/neon/vp9_idct16x16_neon.c

@@ -20,6 +20,15 @@ extern void vp9_short_idct16x16_add_neon_pass2(int16_t *src,
                                                int16_t skip_adding,
                                                uint8_t *dest,
                                                int dest_stride);
+extern void vp9_short_idct10_16x16_add_neon_pass1(int16_t *input,
+                                               int16_t *output,
+                                               int output_stride);
+extern void vp9_short_idct10_16x16_add_neon_pass2(int16_t *src,
+                                               int16_t *output,
+                                               int16_t *pass1Output,
+                                               int16_t skip_adding,
+                                               uint8_t *dest,
+                                               int dest_stride);
 extern void save_registers();
 extern void restore_registers();
 
@@ -97,3 +106,64 @@ void vp9_short_idct16x16_add_neon(int16_t *input,
 
   return;
 }
+
+void vp9_short_idct10_16x16_add_neon(int16_t *input,
+                                  uint8_t *dest, int dest_stride) {
+  int16_t pass1_output[16*16] = {0};
+  int16_t row_idct_output[16*16] = {0};
+
+  // save d8-d15 register values.
+  save_registers();
+
+  /* Parallel idct on the upper 8 rows */
+  // First pass processes even elements 0, 2, 4, 6, 8, 10, 12, 14 and save the
+  // stage 6 result in pass1_output.
+  vp9_short_idct10_16x16_add_neon_pass1(input, pass1_output, 8);
+
+  // Second pass processes odd elements 1, 3, 5, 7, 9, 11, 13, 15 and combines
+  // with result in pass1(pass1_output) to calculate final result in stage 7
+  // which will be saved into row_idct_output.
+  vp9_short_idct10_16x16_add_neon_pass2(input+1,
+                                        row_idct_output,
+                                        pass1_output,
+                                        0,
+                                        dest,
+                                        dest_stride);
+
+  /* Skip Parallel idct on the lower 8 rows as they are all 0s */
+
+  /* Parallel idct on the left 8 columns */
+  // First pass processes even elements 0, 2, 4, 6, 8, 10, 12, 14 and save the
+  // stage 6 result in pass1_output.
+  vp9_short_idct16x16_add_neon_pass1(row_idct_output, pass1_output, 8);
+
+  // Second pass processes odd elements 1, 3, 5, 7, 9, 11, 13, 15 and combines
+  // with result in pass1(pass1_output) to calculate final result in stage 7.
+  // Then add the result to the destination data.
+  vp9_short_idct16x16_add_neon_pass2(row_idct_output+1,
+                                     row_idct_output,
+                                     pass1_output,
+                                     1,
+                                     dest,
+                                     dest_stride);
+
+  /* Parallel idct on the right 8 columns */
+  // First pass processes even elements 0, 2, 4, 6, 8, 10, 12, 14 and save the
+  // stage 6 result in pass1_output.
+  vp9_short_idct16x16_add_neon_pass1(row_idct_output+8*16, pass1_output, 8);
+
+  // Second pass processes odd elements 1, 3, 5, 7, 9, 11, 13, 15 and combines
+  // with result in pass1(pass1_output) to calculate final result in stage 7.
+  // Then add the result to the destination data.
+  vp9_short_idct16x16_add_neon_pass2(row_idct_output+8*16+1,
+                                     row_idct_output+8,
+                                     pass1_output,
+                                     1,
+                                     dest+8,
+                                     dest_stride);
+
+  // restore d8-d15 register values.
+  restore_registers();
+
+  return;
+}

vp9/common/vp9_rtcd_defs.sh

@@ -316,7 +316,7 @@ prototype void vp9_short_idct16x16_add "int16_t *input, uint8_t *dest, int dest_
 specialize vp9_short_idct16x16_add sse2 neon
 
 prototype void vp9_short_idct10_16x16_add "int16_t *input, uint8_t *dest, int dest_stride"
-specialize vp9_short_idct10_16x16_add sse2
+specialize vp9_short_idct10_16x16_add sse2 neon
 
 prototype void vp9_short_idct32x32_add "int16_t *input, uint8_t *dest, int dest_stride"
 specialize vp9_short_idct32x32_add sse2