VP9: Add ssse3 version of vpx_idct32x32_135_add()

Change-Id: I9a780131efaad28cf1ad233ae64c5c319a329727

1 files changed, 27 insertions, 0 deletions

diff --git a/vpx_dsp/inv_txfm.c b/vpx_dsp/inv_txfm.c
index 5f3cfddbd..a0f59bf75 100644
--- a/vpx_dsp/inv_txfm.c
+++ b/vpx_dsp/inv_txfm.c
@@ -1194,6 +1194,33 @@ void vpx_idct32x32_1024_add_c(const tran_low_t *input, uint8_t *dest,
   }
 }
 
+void vpx_idct32x32_135_add_c(const tran_low_t *input, uint8_t *dest,
+                             int stride) {
+  tran_low_t out[32 * 32] = {0};
+  tran_low_t *outptr = out;
+  int i, j;
+  tran_low_t temp_in[32], temp_out[32];
+
+  // Rows
+  // only upper-left 16x16 has non-zero coeff
+  for (i = 0; i < 16; ++i) {
+    idct32_c(input, outptr);
+    input += 32;
+    outptr += 32;
+  }
+
+  // Columns
+  for (i = 0; i < 32; ++i) {
+    for (j = 0; j < 32; ++j)
+      temp_in[j] = out[j * 32 + i];
+    idct32_c(temp_in, temp_out);
+    for (j = 0; j < 32; ++j) {
+      dest[j * stride + i] = clip_pixel_add(dest[j * stride + i],
+                                            ROUND_POWER_OF_TWO(temp_out[j], 6));
+    }
+  }
+}
+
 void vpx_idct32x32_34_add_c(const tran_low_t *input, uint8_t *dest,
                             int stride) {
   tran_low_t out[32 * 32] = {0};