Swap new size 8 and 16 fwd/inv transforms for ADST This is to make them similar to the ones in VP9. Change-Id: Iaebf625f2dce4f159b8a8615f37003d773ee6450

commit: 4d5cf537ff2f5935cc3b1fa852fb0fbd90adf4c5 [log] [tgz]
author: Urvang Joshi <urvang@google.com> Wed Dec 20 16:54:10 2017 -0800
committer: Urvang Joshi <urvang@google.com> Thu Dec 21 21:26:35 2017 +0000
tree: ae8194b3d502c1630964d78fd961f52b52c686b9
parent: eb2fd5c58a73b473bcd7f9a7cbc7a48a4f8aa141 [diff]
diff --git a/av1/common/av1_inv_txfm1d.c b/av1/common/av1_inv_txfm1d.c
index d48ca6f..f19a39a 100644
--- a/av1/common/av1_inv_txfm1d.c
+++ b/av1/common/av1_inv_txfm1d.c

@@ -812,16 +812,15 @@
 
   // stage 1;
   stage++;
-  assert(output != input);
   bf1 = output;
-  bf1[0] = input[0];
-  bf1[1] = -input[7];
-  bf1[2] = -input[3];
-  bf1[3] = input[4];
-  bf1[4] = -input[1];
-  bf1[5] = input[6];
-  bf1[6] = input[2];
-  bf1[7] = -input[5];
+  bf1[0] = input[7];
+  bf1[1] = input[0];
+  bf1[2] = input[5];
+  bf1[3] = input[2];
+  bf1[4] = input[3];
+  bf1[5] = input[4];
+  bf1[6] = input[1];
+  bf1[7] = input[6];
   range_check(stage, input, bf1, size, stage_range[stage]);
 
   // stage 2
@@ -829,28 +828,28 @@
   cospi = cospi_arr(cos_bit[stage]);
   bf0 = output;
   bf1 = step;
-  bf1[0] = bf0[0];
-  bf1[1] = bf0[1];
-  bf1[2] = half_btf(cospi[32], bf0[2], cospi[32], bf0[3], cos_bit[stage]);
-  bf1[3] = half_btf(cospi[32], bf0[2], -cospi[32], bf0[3], cos_bit[stage]);
-  bf1[4] = bf0[4];
-  bf1[5] = bf0[5];
-  bf1[6] = half_btf(cospi[32], bf0[6], cospi[32], bf0[7], cos_bit[stage]);
-  bf1[7] = half_btf(cospi[32], bf0[6], -cospi[32], bf0[7], cos_bit[stage]);
+  bf1[0] = half_btf(cospi[4], bf0[0], cospi[60], bf0[1], cos_bit[stage]);
+  bf1[1] = half_btf(-cospi[4], bf0[1], cospi[60], bf0[0], cos_bit[stage]);
+  bf1[2] = half_btf(cospi[20], bf0[2], cospi[44], bf0[3], cos_bit[stage]);
+  bf1[3] = half_btf(-cospi[20], bf0[3], cospi[44], bf0[2], cos_bit[stage]);
+  bf1[4] = half_btf(cospi[36], bf0[4], cospi[28], bf0[5], cos_bit[stage]);
+  bf1[5] = half_btf(-cospi[36], bf0[5], cospi[28], bf0[4], cos_bit[stage]);
+  bf1[6] = half_btf(cospi[52], bf0[6], cospi[12], bf0[7], cos_bit[stage]);
+  bf1[7] = half_btf(-cospi[52], bf0[7], cospi[12], bf0[6], cos_bit[stage]);
   range_check(stage, input, bf1, size, stage_range[stage]);
 
   // stage 3
   stage++;
   bf0 = step;
   bf1 = output;
-  bf1[0] = bf0[0] + bf0[2];
-  bf1[1] = bf0[1] + bf0[3];
-  bf1[2] = bf0[0] - bf0[2];
-  bf1[3] = bf0[1] - bf0[3];
-  bf1[4] = bf0[4] + bf0[6];
-  bf1[5] = bf0[5] + bf0[7];
-  bf1[6] = bf0[4] - bf0[6];
-  bf1[7] = bf0[5] - bf0[7];
+  bf1[0] = bf0[0] + bf0[4];
+  bf1[1] = bf0[1] + bf0[5];
+  bf1[2] = bf0[2] + bf0[6];
+  bf1[3] = bf0[3] + bf0[7];
+  bf1[4] = -bf0[4] + bf0[0];
+  bf1[5] = -bf0[5] + bf0[1];
+  bf1[6] = -bf0[6] + bf0[2];
+  bf1[7] = -bf0[7] + bf0[3];
   range_check(stage, input, bf1, size, stage_range[stage]);
 
   // stage 4
@@ -863,23 +862,23 @@
   bf1[2] = bf0[2];
   bf1[3] = bf0[3];
   bf1[4] = half_btf(cospi[16], bf0[4], cospi[48], bf0[5], cos_bit[stage]);
-  bf1[5] = half_btf(cospi[48], bf0[4], -cospi[16], bf0[5], cos_bit[stage]);
+  bf1[5] = half_btf(-cospi[16], bf0[5], cospi[48], bf0[4], cos_bit[stage]);
   bf1[6] = half_btf(-cospi[48], bf0[6], cospi[16], bf0[7], cos_bit[stage]);
-  bf1[7] = half_btf(cospi[16], bf0[6], cospi[48], bf0[7], cos_bit[stage]);
+  bf1[7] = half_btf(cospi[48], bf0[7], cospi[16], bf0[6], cos_bit[stage]);
   range_check(stage, input, bf1, size, stage_range[stage]);
 
   // stage 5
   stage++;
   bf0 = step;
   bf1 = output;
-  bf1[0] = bf0[0] + bf0[4];
-  bf1[1] = bf0[1] + bf0[5];
-  bf1[2] = bf0[2] + bf0[6];
-  bf1[3] = bf0[3] + bf0[7];
-  bf1[4] = bf0[0] - bf0[4];
-  bf1[5] = bf0[1] - bf0[5];
-  bf1[6] = bf0[2] - bf0[6];
-  bf1[7] = bf0[3] - bf0[7];
+  bf1[0] = bf0[0] + bf0[2];
+  bf1[1] = bf0[1] + bf0[3];
+  bf1[2] = -bf0[2] + bf0[0];
+  bf1[3] = -bf0[3] + bf0[1];
+  bf1[4] = bf0[4] + bf0[6];
+  bf1[5] = bf0[5] + bf0[7];
+  bf1[6] = -bf0[6] + bf0[4];
+  bf1[7] = -bf0[7] + bf0[5];
   range_check(stage, input, bf1, size, stage_range[stage]);
 
   // stage 6
@@ -887,28 +886,28 @@
   cospi = cospi_arr(cos_bit[stage]);
   bf0 = output;
   bf1 = step;
-  bf1[0] = half_btf(cospi[4], bf0[0], cospi[60], bf0[1], cos_bit[stage]);
-  bf1[1] = half_btf(cospi[60], bf0[0], -cospi[4], bf0[1], cos_bit[stage]);
-  bf1[2] = half_btf(cospi[20], bf0[2], cospi[44], bf0[3], cos_bit[stage]);
-  bf1[3] = half_btf(cospi[44], bf0[2], -cospi[20], bf0[3], cos_bit[stage]);
-  bf1[4] = half_btf(cospi[36], bf0[4], cospi[28], bf0[5], cos_bit[stage]);
-  bf1[5] = half_btf(cospi[28], bf0[4], -cospi[36], bf0[5], cos_bit[stage]);
-  bf1[6] = half_btf(cospi[52], bf0[6], cospi[12], bf0[7], cos_bit[stage]);
-  bf1[7] = half_btf(cospi[12], bf0[6], -cospi[52], bf0[7], cos_bit[stage]);
+  bf1[0] = bf0[0];
+  bf1[1] = bf0[1];
+  bf1[2] = half_btf(cospi[32], bf0[2], cospi[32], bf0[3], cos_bit[stage]);
+  bf1[3] = half_btf(-cospi[32], bf0[3], cospi[32], bf0[2], cos_bit[stage]);
+  bf1[4] = bf0[4];
+  bf1[5] = bf0[5];
+  bf1[6] = half_btf(cospi[32], bf0[6], cospi[32], bf0[7], cos_bit[stage]);
+  bf1[7] = half_btf(-cospi[32], bf0[7], cospi[32], bf0[6], cos_bit[stage]);
   range_check(stage, input, bf1, size, stage_range[stage]);
 
   // stage 7
   stage++;
   bf0 = step;
   bf1 = output;
-  bf1[0] = bf0[1];
-  bf1[1] = bf0[6];
-  bf1[2] = bf0[3];
-  bf1[3] = bf0[4];
-  bf1[4] = bf0[5];
-  bf1[5] = bf0[2];
-  bf1[6] = bf0[7];
-  bf1[7] = bf0[0];
+  bf1[0] = bf0[0];
+  bf1[1] = -bf0[4];
+  bf1[2] = bf0[6];
+  bf1[3] = -bf0[2];
+  bf1[4] = bf0[3];
+  bf1[5] = -bf0[7];
+  bf1[6] = bf0[5];
+  bf1[7] = -bf0[1];
   range_check(stage, input, bf1, size, stage_range[stage]);
 }
 
@@ -926,24 +925,23 @@
 
   // stage 1;
   stage++;
-  assert(output != input);
   bf1 = output;
-  bf1[0] = input[0];
-  bf1[1] = -input[15];
-  bf1[2] = -input[7];
-  bf1[3] = input[8];
-  bf1[4] = -input[3];
-  bf1[5] = input[12];
-  bf1[6] = input[4];
-  bf1[7] = -input[11];
-  bf1[8] = -input[1];
-  bf1[9] = input[14];
-  bf1[10] = input[6];
-  bf1[11] = -input[9];
-  bf1[12] = input[2];
-  bf1[13] = -input[13];
-  bf1[14] = -input[5];
-  bf1[15] = input[10];
+  bf1[0] = input[15];
+  bf1[1] = input[0];
+  bf1[2] = input[13];
+  bf1[3] = input[2];
+  bf1[4] = input[11];
+  bf1[5] = input[4];
+  bf1[6] = input[9];
+  bf1[7] = input[6];
+  bf1[8] = input[7];
+  bf1[9] = input[8];
+  bf1[10] = input[5];
+  bf1[11] = input[10];
+  bf1[12] = input[3];
+  bf1[13] = input[12];
+  bf1[14] = input[1];
+  bf1[15] = input[14];
   range_check(stage, input, bf1, size, stage_range[stage]);
 
   // stage 2
@@ -951,44 +949,44 @@
   cospi = cospi_arr(cos_bit[stage]);
   bf0 = output;
   bf1 = step;
-  bf1[0] = bf0[0];
-  bf1[1] = bf0[1];
-  bf1[2] = half_btf(cospi[32], bf0[2], cospi[32], bf0[3], cos_bit[stage]);
-  bf1[3] = half_btf(cospi[32], bf0[2], -cospi[32], bf0[3], cos_bit[stage]);
-  bf1[4] = bf0[4];
-  bf1[5] = bf0[5];
-  bf1[6] = half_btf(cospi[32], bf0[6], cospi[32], bf0[7], cos_bit[stage]);
-  bf1[7] = half_btf(cospi[32], bf0[6], -cospi[32], bf0[7], cos_bit[stage]);
-  bf1[8] = bf0[8];
-  bf1[9] = bf0[9];
-  bf1[10] = half_btf(cospi[32], bf0[10], cospi[32], bf0[11], cos_bit[stage]);
-  bf1[11] = half_btf(cospi[32], bf0[10], -cospi[32], bf0[11], cos_bit[stage]);
-  bf1[12] = bf0[12];
-  bf1[13] = bf0[13];
-  bf1[14] = half_btf(cospi[32], bf0[14], cospi[32], bf0[15], cos_bit[stage]);
-  bf1[15] = half_btf(cospi[32], bf0[14], -cospi[32], bf0[15], cos_bit[stage]);
+  bf1[0] = half_btf(cospi[2], bf0[0], cospi[62], bf0[1], cos_bit[stage]);
+  bf1[1] = half_btf(-cospi[2], bf0[1], cospi[62], bf0[0], cos_bit[stage]);
+  bf1[2] = half_btf(cospi[10], bf0[2], cospi[54], bf0[3], cos_bit[stage]);
+  bf1[3] = half_btf(-cospi[10], bf0[3], cospi[54], bf0[2], cos_bit[stage]);
+  bf1[4] = half_btf(cospi[18], bf0[4], cospi[46], bf0[5], cos_bit[stage]);
+  bf1[5] = half_btf(-cospi[18], bf0[5], cospi[46], bf0[4], cos_bit[stage]);
+  bf1[6] = half_btf(cospi[26], bf0[6], cospi[38], bf0[7], cos_bit[stage]);
+  bf1[7] = half_btf(-cospi[26], bf0[7], cospi[38], bf0[6], cos_bit[stage]);
+  bf1[8] = half_btf(cospi[34], bf0[8], cospi[30], bf0[9], cos_bit[stage]);
+  bf1[9] = half_btf(-cospi[34], bf0[9], cospi[30], bf0[8], cos_bit[stage]);
+  bf1[10] = half_btf(cospi[42], bf0[10], cospi[22], bf0[11], cos_bit[stage]);
+  bf1[11] = half_btf(-cospi[42], bf0[11], cospi[22], bf0[10], cos_bit[stage]);
+  bf1[12] = half_btf(cospi[50], bf0[12], cospi[14], bf0[13], cos_bit[stage]);
+  bf1[13] = half_btf(-cospi[50], bf0[13], cospi[14], bf0[12], cos_bit[stage]);
+  bf1[14] = half_btf(cospi[58], bf0[14], cospi[6], bf0[15], cos_bit[stage]);
+  bf1[15] = half_btf(-cospi[58], bf0[15], cospi[6], bf0[14], cos_bit[stage]);
   range_check(stage, input, bf1, size, stage_range[stage]);
 
   // stage 3
   stage++;
   bf0 = step;
   bf1 = output;
-  bf1[0] = bf0[0] + bf0[2];
-  bf1[1] = bf0[1] + bf0[3];
-  bf1[2] = bf0[0] - bf0[2];
-  bf1[3] = bf0[1] - bf0[3];
-  bf1[4] = bf0[4] + bf0[6];
-  bf1[5] = bf0[5] + bf0[7];
-  bf1[6] = bf0[4] - bf0[6];
-  bf1[7] = bf0[5] - bf0[7];
-  bf1[8] = bf0[8] + bf0[10];
-  bf1[9] = bf0[9] + bf0[11];
-  bf1[10] = bf0[8] - bf0[10];
-  bf1[11] = bf0[9] - bf0[11];
-  bf1[12] = bf0[12] + bf0[14];
-  bf1[13] = bf0[13] + bf0[15];
-  bf1[14] = bf0[12] - bf0[14];
-  bf1[15] = bf0[13] - bf0[15];
+  bf1[0] = bf0[0] + bf0[8];
+  bf1[1] = bf0[1] + bf0[9];
+  bf1[2] = bf0[2] + bf0[10];
+  bf1[3] = bf0[3] + bf0[11];
+  bf1[4] = bf0[4] + bf0[12];
+  bf1[5] = bf0[5] + bf0[13];
+  bf1[6] = bf0[6] + bf0[14];
+  bf1[7] = bf0[7] + bf0[15];
+  bf1[8] = -bf0[8] + bf0[0];
+  bf1[9] = -bf0[9] + bf0[1];
+  bf1[10] = -bf0[10] + bf0[2];
+  bf1[11] = -bf0[11] + bf0[3];
+  bf1[12] = -bf0[12] + bf0[4];
+  bf1[13] = -bf0[13] + bf0[5];
+  bf1[14] = -bf0[14] + bf0[6];
+  bf1[15] = -bf0[15] + bf0[7];
   range_check(stage, input, bf1, size, stage_range[stage]);
 
   // stage 4
@@ -1000,85 +998,85 @@
   bf1[1] = bf0[1];
   bf1[2] = bf0[2];
   bf1[3] = bf0[3];
-  bf1[4] = half_btf(cospi[16], bf0[4], cospi[48], bf0[5], cos_bit[stage]);
-  bf1[5] = half_btf(cospi[48], bf0[4], -cospi[16], bf0[5], cos_bit[stage]);
-  bf1[6] = half_btf(-cospi[48], bf0[6], cospi[16], bf0[7], cos_bit[stage]);
-  bf1[7] = half_btf(cospi[16], bf0[6], cospi[48], bf0[7], cos_bit[stage]);
-  bf1[8] = bf0[8];
-  bf1[9] = bf0[9];
-  bf1[10] = bf0[10];
-  bf1[11] = bf0[11];
-  bf1[12] = half_btf(cospi[16], bf0[12], cospi[48], bf0[13], cos_bit[stage]);
-  bf1[13] = half_btf(cospi[48], bf0[12], -cospi[16], bf0[13], cos_bit[stage]);
-  bf1[14] = half_btf(-cospi[48], bf0[14], cospi[16], bf0[15], cos_bit[stage]);
-  bf1[15] = half_btf(cospi[16], bf0[14], cospi[48], bf0[15], cos_bit[stage]);
-  range_check(stage, input, bf1, size, stage_range[stage]);
-
-  // stage 5
-  stage++;
-  bf0 = step;
-  bf1 = output;
-  bf1[0] = bf0[0] + bf0[4];
-  bf1[1] = bf0[1] + bf0[5];
-  bf1[2] = bf0[2] + bf0[6];
-  bf1[3] = bf0[3] + bf0[7];
-  bf1[4] = bf0[0] - bf0[4];
-  bf1[5] = bf0[1] - bf0[5];
-  bf1[6] = bf0[2] - bf0[6];
-  bf1[7] = bf0[3] - bf0[7];
-  bf1[8] = bf0[8] + bf0[12];
-  bf1[9] = bf0[9] + bf0[13];
-  bf1[10] = bf0[10] + bf0[14];
-  bf1[11] = bf0[11] + bf0[15];
-  bf1[12] = bf0[8] - bf0[12];
-  bf1[13] = bf0[9] - bf0[13];
-  bf1[14] = bf0[10] - bf0[14];
-  bf1[15] = bf0[11] - bf0[15];
-  range_check(stage, input, bf1, size, stage_range[stage]);
-
-  // stage 6
-  stage++;
-  cospi = cospi_arr(cos_bit[stage]);
-  bf0 = output;
-  bf1 = step;
-  bf1[0] = bf0[0];
-  bf1[1] = bf0[1];
-  bf1[2] = bf0[2];
-  bf1[3] = bf0[3];
   bf1[4] = bf0[4];
   bf1[5] = bf0[5];
   bf1[6] = bf0[6];
   bf1[7] = bf0[7];
   bf1[8] = half_btf(cospi[8], bf0[8], cospi[56], bf0[9], cos_bit[stage]);
-  bf1[9] = half_btf(cospi[56], bf0[8], -cospi[8], bf0[9], cos_bit[stage]);
+  bf1[9] = half_btf(-cospi[8], bf0[9], cospi[56], bf0[8], cos_bit[stage]);
   bf1[10] = half_btf(cospi[40], bf0[10], cospi[24], bf0[11], cos_bit[stage]);
-  bf1[11] = half_btf(cospi[24], bf0[10], -cospi[40], bf0[11], cos_bit[stage]);
+  bf1[11] = half_btf(-cospi[40], bf0[11], cospi[24], bf0[10], cos_bit[stage]);
   bf1[12] = half_btf(-cospi[56], bf0[12], cospi[8], bf0[13], cos_bit[stage]);
-  bf1[13] = half_btf(cospi[8], bf0[12], cospi[56], bf0[13], cos_bit[stage]);
+  bf1[13] = half_btf(cospi[56], bf0[13], cospi[8], bf0[12], cos_bit[stage]);
   bf1[14] = half_btf(-cospi[24], bf0[14], cospi[40], bf0[15], cos_bit[stage]);
-  bf1[15] = half_btf(cospi[40], bf0[14], cospi[24], bf0[15], cos_bit[stage]);
+  bf1[15] = half_btf(cospi[24], bf0[15], cospi[40], bf0[14], cos_bit[stage]);
+  range_check(stage, input, bf1, size, stage_range[stage]);
+
+  // stage 5
+  stage++;
+  bf0 = step;
+  bf1 = output;
+  bf1[0] = bf0[0] + bf0[4];
+  bf1[1] = bf0[1] + bf0[5];
+  bf1[2] = bf0[2] + bf0[6];
+  bf1[3] = bf0[3] + bf0[7];
+  bf1[4] = -bf0[4] + bf0[0];
+  bf1[5] = -bf0[5] + bf0[1];
+  bf1[6] = -bf0[6] + bf0[2];
+  bf1[7] = -bf0[7] + bf0[3];
+  bf1[8] = bf0[8] + bf0[12];
+  bf1[9] = bf0[9] + bf0[13];
+  bf1[10] = bf0[10] + bf0[14];
+  bf1[11] = bf0[11] + bf0[15];
+  bf1[12] = -bf0[12] + bf0[8];
+  bf1[13] = -bf0[13] + bf0[9];
+  bf1[14] = -bf0[14] + bf0[10];
+  bf1[15] = -bf0[15] + bf0[11];
+  range_check(stage, input, bf1, size, stage_range[stage]);
+
+  // stage 6
+  stage++;
+  cospi = cospi_arr(cos_bit[stage]);
+  bf0 = output;
+  bf1 = step;
+  bf1[0] = bf0[0];
+  bf1[1] = bf0[1];
+  bf1[2] = bf0[2];
+  bf1[3] = bf0[3];
+  bf1[4] = half_btf(cospi[16], bf0[4], cospi[48], bf0[5], cos_bit[stage]);
+  bf1[5] = half_btf(-cospi[16], bf0[5], cospi[48], bf0[4], cos_bit[stage]);
+  bf1[6] = half_btf(-cospi[48], bf0[6], cospi[16], bf0[7], cos_bit[stage]);
+  bf1[7] = half_btf(cospi[48], bf0[7], cospi[16], bf0[6], cos_bit[stage]);
+  bf1[8] = bf0[8];
+  bf1[9] = bf0[9];
+  bf1[10] = bf0[10];
+  bf1[11] = bf0[11];
+  bf1[12] = half_btf(cospi[16], bf0[12], cospi[48], bf0[13], cos_bit[stage]);
+  bf1[13] = half_btf(-cospi[16], bf0[13], cospi[48], bf0[12], cos_bit[stage]);
+  bf1[14] = half_btf(-cospi[48], bf0[14], cospi[16], bf0[15], cos_bit[stage]);
+  bf1[15] = half_btf(cospi[48], bf0[15], cospi[16], bf0[14], cos_bit[stage]);
   range_check(stage, input, bf1, size, stage_range[stage]);
 
   // stage 7
   stage++;
   bf0 = step;
   bf1 = output;
-  bf1[0] = bf0[0] + bf0[8];
-  bf1[1] = bf0[1] + bf0[9];
-  bf1[2] = bf0[2] + bf0[10];
-  bf1[3] = bf0[3] + bf0[11];
-  bf1[4] = bf0[4] + bf0[12];
-  bf1[5] = bf0[5] + bf0[13];
-  bf1[6] = bf0[6] + bf0[14];
-  bf1[7] = bf0[7] + bf0[15];
-  bf1[8] = bf0[0] - bf0[8];
-  bf1[9] = bf0[1] - bf0[9];
-  bf1[10] = bf0[2] - bf0[10];
-  bf1[11] = bf0[3] - bf0[11];
-  bf1[12] = bf0[4] - bf0[12];
-  bf1[13] = bf0[5] - bf0[13];
-  bf1[14] = bf0[6] - bf0[14];
-  bf1[15] = bf0[7] - bf0[15];
+  bf1[0] = bf0[0] + bf0[2];
+  bf1[1] = bf0[1] + bf0[3];
+  bf1[2] = -bf0[2] + bf0[0];
+  bf1[3] = -bf0[3] + bf0[1];
+  bf1[4] = bf0[4] + bf0[6];
+  bf1[5] = bf0[5] + bf0[7];
+  bf1[6] = -bf0[6] + bf0[4];
+  bf1[7] = -bf0[7] + bf0[5];
+  bf1[8] = bf0[8] + bf0[10];
+  bf1[9] = bf0[9] + bf0[11];
+  bf1[10] = -bf0[10] + bf0[8];
+  bf1[11] = -bf0[11] + bf0[9];
+  bf1[12] = bf0[12] + bf0[14];
+  bf1[13] = bf0[13] + bf0[15];
+  bf1[14] = -bf0[14] + bf0[12];
+  bf1[15] = -bf0[15] + bf0[13];
   range_check(stage, input, bf1, size, stage_range[stage]);
 
   // stage 8
@@ -1086,44 +1084,44 @@
   cospi = cospi_arr(cos_bit[stage]);
   bf0 = output;
   bf1 = step;
-  bf1[0] = half_btf(cospi[2], bf0[0], cospi[62], bf0[1], cos_bit[stage]);
-  bf1[1] = half_btf(cospi[62], bf0[0], -cospi[2], bf0[1], cos_bit[stage]);
-  bf1[2] = half_btf(cospi[10], bf0[2], cospi[54], bf0[3], cos_bit[stage]);
-  bf1[3] = half_btf(cospi[54], bf0[2], -cospi[10], bf0[3], cos_bit[stage]);
-  bf1[4] = half_btf(cospi[18], bf0[4], cospi[46], bf0[5], cos_bit[stage]);
-  bf1[5] = half_btf(cospi[46], bf0[4], -cospi[18], bf0[5], cos_bit[stage]);
-  bf1[6] = half_btf(cospi[26], bf0[6], cospi[38], bf0[7], cos_bit[stage]);
-  bf1[7] = half_btf(cospi[38], bf0[6], -cospi[26], bf0[7], cos_bit[stage]);
-  bf1[8] = half_btf(cospi[34], bf0[8], cospi[30], bf0[9], cos_bit[stage]);
-  bf1[9] = half_btf(cospi[30], bf0[8], -cospi[34], bf0[9], cos_bit[stage]);
-  bf1[10] = half_btf(cospi[42], bf0[10], cospi[22], bf0[11], cos_bit[stage]);
-  bf1[11] = half_btf(cospi[22], bf0[10], -cospi[42], bf0[11], cos_bit[stage]);
-  bf1[12] = half_btf(cospi[50], bf0[12], cospi[14], bf0[13], cos_bit[stage]);
-  bf1[13] = half_btf(cospi[14], bf0[12], -cospi[50], bf0[13], cos_bit[stage]);
-  bf1[14] = half_btf(cospi[58], bf0[14], cospi[6], bf0[15], cos_bit[stage]);
-  bf1[15] = half_btf(cospi[6], bf0[14], -cospi[58], bf0[15], cos_bit[stage]);
+  bf1[0] = bf0[0];
+  bf1[1] = bf0[1];
+  bf1[2] = half_btf(cospi[32], bf0[2], cospi[32], bf0[3], cos_bit[stage]);
+  bf1[3] = half_btf(-cospi[32], bf0[3], cospi[32], bf0[2], cos_bit[stage]);
+  bf1[4] = bf0[4];
+  bf1[5] = bf0[5];
+  bf1[6] = half_btf(cospi[32], bf0[6], cospi[32], bf0[7], cos_bit[stage]);
+  bf1[7] = half_btf(-cospi[32], bf0[7], cospi[32], bf0[6], cos_bit[stage]);
+  bf1[8] = bf0[8];
+  bf1[9] = bf0[9];
+  bf1[10] = half_btf(cospi[32], bf0[10], cospi[32], bf0[11], cos_bit[stage]);
+  bf1[11] = half_btf(-cospi[32], bf0[11], cospi[32], bf0[10], cos_bit[stage]);
+  bf1[12] = bf0[12];
+  bf1[13] = bf0[13];
+  bf1[14] = half_btf(cospi[32], bf0[14], cospi[32], bf0[15], cos_bit[stage]);
+  bf1[15] = half_btf(-cospi[32], bf0[15], cospi[32], bf0[14], cos_bit[stage]);
   range_check(stage, input, bf1, size, stage_range[stage]);
 
   // stage 9
   stage++;
   bf0 = step;
   bf1 = output;
-  bf1[0] = bf0[1];
-  bf1[1] = bf0[14];
-  bf1[2] = bf0[3];
-  bf1[3] = bf0[12];
-  bf1[4] = bf0[5];
-  bf1[5] = bf0[10];
-  bf1[6] = bf0[7];
-  bf1[7] = bf0[8];
-  bf1[8] = bf0[9];
-  bf1[9] = bf0[6];
-  bf1[10] = bf0[11];
-  bf1[11] = bf0[4];
-  bf1[12] = bf0[13];
-  bf1[13] = bf0[2];
-  bf1[14] = bf0[15];
-  bf1[15] = bf0[0];
+  bf1[0] = bf0[0];
+  bf1[1] = -bf0[8];
+  bf1[2] = bf0[12];
+  bf1[3] = -bf0[4];
+  bf1[4] = bf0[6];
+  bf1[5] = -bf0[14];
+  bf1[6] = bf0[10];
+  bf1[7] = -bf0[2];
+  bf1[8] = bf0[3];
+  bf1[9] = -bf0[11];
+  bf1[10] = bf0[15];
+  bf1[11] = -bf0[7];
+  bf1[12] = bf0[5];
+  bf1[13] = -bf0[13];
+  bf1[14] = bf0[9];
+  bf1[15] = -bf0[1];
   range_check(stage, input, bf1, size, stage_range[stage]);
 }
 

diff --git a/av1/common/x86/highbd_inv_txfm_sse4.c b/av1/common/x86/highbd_inv_txfm_sse4.c
index 95297e5..b21c7db 100644
--- a/av1/common/x86/highbd_inv_txfm_sse4.c
+++ b/av1/common/x86/highbd_inv_txfm_sse4.c

@@ -448,183 +448,305 @@
 
 static void iadst8x8_sse4_1(__m128i *in, __m128i *out, int bit) {
   const int32_t *cospi = cospi_arr(bit);
-  const __m128i cospi32 = _mm_set1_epi32(cospi[32]);
-  const __m128i cospi16 = _mm_set1_epi32(cospi[16]);
-  const __m128i cospim16 = _mm_set1_epi32(-cospi[16]);
-  const __m128i cospi48 = _mm_set1_epi32(cospi[48]);
-  const __m128i cospim48 = _mm_set1_epi32(-cospi[48]);
   const __m128i cospi4 = _mm_set1_epi32(cospi[4]);
-  const __m128i cospim4 = _mm_set1_epi32(-cospi[4]);
   const __m128i cospi60 = _mm_set1_epi32(cospi[60]);
   const __m128i cospi20 = _mm_set1_epi32(cospi[20]);
-  const __m128i cospim20 = _mm_set1_epi32(-cospi[20]);
   const __m128i cospi44 = _mm_set1_epi32(cospi[44]);
-  const __m128i cospi28 = _mm_set1_epi32(cospi[28]);
   const __m128i cospi36 = _mm_set1_epi32(cospi[36]);
-  const __m128i cospim36 = _mm_set1_epi32(-cospi[36]);
+  const __m128i cospi28 = _mm_set1_epi32(cospi[28]);
   const __m128i cospi52 = _mm_set1_epi32(cospi[52]);
-  const __m128i cospim52 = _mm_set1_epi32(-cospi[52]);
   const __m128i cospi12 = _mm_set1_epi32(cospi[12]);
+  const __m128i cospi16 = _mm_set1_epi32(cospi[16]);
+  const __m128i cospi48 = _mm_set1_epi32(cospi[48]);
+  const __m128i cospim48 = _mm_set1_epi32(-cospi[48]);
+  const __m128i cospi32 = _mm_set1_epi32(cospi[32]);
   const __m128i rnding = _mm_set1_epi32(1 << (bit - 1));
-  const __m128i zero = _mm_setzero_si128();
-  __m128i u0, u1, u2, u3, u4, u5, u6, u7;
-  __m128i v0, v1, v2, v3, v4, v5, v6, v7;
-  __m128i x, y;
-  int col;
+  const __m128i kZero = _mm_setzero_si128();
+  __m128i u[8], v[8], x;
 
-  // Note:
-  //  Even column: 0, 2, ..., 14
-  //  Odd column: 1, 3, ..., 15
-  //  one even column plus one odd column constructs one row (8 coeffs)
-  //  total we have 8 rows (8x8).
-  for (col = 0; col < 2; ++col) {
-    // stage 0
-    // stage 1
-    u0 = in[2 * 0 + col];
-    u1 = _mm_sub_epi32(zero, in[2 * 7 + col]);
-    u2 = _mm_sub_epi32(zero, in[2 * 3 + col]);
-    u3 = in[2 * 4 + col];
-    u4 = _mm_sub_epi32(zero, in[2 * 1 + col]);
-    u5 = in[2 * 6 + col];
-    u6 = in[2 * 2 + col];
-    u7 = _mm_sub_epi32(zero, in[2 * 5 + col]);
+  // Even 8 points: 0, 2, ..., 14
+  // stage 0
+  // stage 1
+  // stage 2
+  // (1)
+  u[0] = _mm_mullo_epi32(in[14], cospi4);
+  x = _mm_mullo_epi32(in[0], cospi60);
+  u[0] = _mm_add_epi32(u[0], x);
+  u[0] = _mm_add_epi32(u[0], rnding);
+  u[0] = _mm_srai_epi32(u[0], bit);
 
-    // stage 2
-    v0 = u0;
-    v1 = u1;
+  u[1] = _mm_mullo_epi32(in[14], cospi60);
+  x = _mm_mullo_epi32(in[0], cospi4);
+  u[1] = _mm_sub_epi32(u[1], x);
+  u[1] = _mm_add_epi32(u[1], rnding);
+  u[1] = _mm_srai_epi32(u[1], bit);
 
-    x = _mm_mullo_epi32(u2, cospi32);
-    y = _mm_mullo_epi32(u3, cospi32);
-    v2 = _mm_add_epi32(x, y);
-    v2 = _mm_add_epi32(v2, rnding);
-    v2 = _mm_srai_epi32(v2, bit);
+  // (2)
+  u[2] = _mm_mullo_epi32(in[10], cospi20);
+  x = _mm_mullo_epi32(in[4], cospi44);
+  u[2] = _mm_add_epi32(u[2], x);
+  u[2] = _mm_add_epi32(u[2], rnding);
+  u[2] = _mm_srai_epi32(u[2], bit);
 
-    v3 = _mm_sub_epi32(x, y);
-    v3 = _mm_add_epi32(v3, rnding);
-    v3 = _mm_srai_epi32(v3, bit);
+  u[3] = _mm_mullo_epi32(in[10], cospi44);
+  x = _mm_mullo_epi32(in[4], cospi20);
+  u[3] = _mm_sub_epi32(u[3], x);
+  u[3] = _mm_add_epi32(u[3], rnding);
+  u[3] = _mm_srai_epi32(u[3], bit);
 
-    v4 = u4;
-    v5 = u5;
+  // (3)
+  u[4] = _mm_mullo_epi32(in[6], cospi36);
+  x = _mm_mullo_epi32(in[8], cospi28);
+  u[4] = _mm_add_epi32(u[4], x);
+  u[4] = _mm_add_epi32(u[4], rnding);
+  u[4] = _mm_srai_epi32(u[4], bit);
 
-    x = _mm_mullo_epi32(u6, cospi32);
-    y = _mm_mullo_epi32(u7, cospi32);
-    v6 = _mm_add_epi32(x, y);
-    v6 = _mm_add_epi32(v6, rnding);
-    v6 = _mm_srai_epi32(v6, bit);
+  u[5] = _mm_mullo_epi32(in[6], cospi28);
+  x = _mm_mullo_epi32(in[8], cospi36);
+  u[5] = _mm_sub_epi32(u[5], x);
+  u[5] = _mm_add_epi32(u[5], rnding);
+  u[5] = _mm_srai_epi32(u[5], bit);
 
-    v7 = _mm_sub_epi32(x, y);
-    v7 = _mm_add_epi32(v7, rnding);
-    v7 = _mm_srai_epi32(v7, bit);
+  // (4)
+  u[6] = _mm_mullo_epi32(in[2], cospi52);
+  x = _mm_mullo_epi32(in[12], cospi12);
+  u[6] = _mm_add_epi32(u[6], x);
+  u[6] = _mm_add_epi32(u[6], rnding);
+  u[6] = _mm_srai_epi32(u[6], bit);
 
-    // stage 3
-    u0 = _mm_add_epi32(v0, v2);
-    u1 = _mm_add_epi32(v1, v3);
-    u2 = _mm_sub_epi32(v0, v2);
-    u3 = _mm_sub_epi32(v1, v3);
-    u4 = _mm_add_epi32(v4, v6);
-    u5 = _mm_add_epi32(v5, v7);
-    u6 = _mm_sub_epi32(v4, v6);
-    u7 = _mm_sub_epi32(v5, v7);
+  u[7] = _mm_mullo_epi32(in[2], cospi12);
+  x = _mm_mullo_epi32(in[12], cospi52);
+  u[7] = _mm_sub_epi32(u[7], x);
+  u[7] = _mm_add_epi32(u[7], rnding);
+  u[7] = _mm_srai_epi32(u[7], bit);
 
-    // stage 4
-    v0 = u0;
-    v1 = u1;
-    v2 = u2;
-    v3 = u3;
+  // stage 3
+  v[0] = _mm_add_epi32(u[0], u[4]);
+  v[4] = _mm_sub_epi32(u[0], u[4]);
+  v[1] = _mm_add_epi32(u[1], u[5]);
+  v[5] = _mm_sub_epi32(u[1], u[5]);
+  v[2] = _mm_add_epi32(u[2], u[6]);
+  v[6] = _mm_sub_epi32(u[2], u[6]);
+  v[3] = _mm_add_epi32(u[3], u[7]);
+  v[7] = _mm_sub_epi32(u[3], u[7]);
 
-    x = _mm_mullo_epi32(u4, cospi16);
-    y = _mm_mullo_epi32(u5, cospi48);
-    v4 = _mm_add_epi32(x, y);
-    v4 = _mm_add_epi32(v4, rnding);
-    v4 = _mm_srai_epi32(v4, bit);
+  // stage 4
+  u[0] = v[0];
+  u[1] = v[1];
+  u[2] = v[2];
+  u[3] = v[3];
 
-    x = _mm_mullo_epi32(u4, cospi48);
-    y = _mm_mullo_epi32(u5, cospim16);
-    v5 = _mm_add_epi32(x, y);
-    v5 = _mm_add_epi32(v5, rnding);
-    v5 = _mm_srai_epi32(v5, bit);
+  u[4] = _mm_mullo_epi32(v[4], cospi16);
+  x = _mm_mullo_epi32(v[5], cospi48);
+  u[4] = _mm_add_epi32(u[4], x);
+  u[4] = _mm_add_epi32(u[4], rnding);
+  u[4] = _mm_srai_epi32(u[4], bit);
 
-    x = _mm_mullo_epi32(u6, cospim48);
-    y = _mm_mullo_epi32(u7, cospi16);
-    v6 = _mm_add_epi32(x, y);
-    v6 = _mm_add_epi32(v6, rnding);
-    v6 = _mm_srai_epi32(v6, bit);
+  u[5] = _mm_mullo_epi32(v[4], cospi48);
+  x = _mm_mullo_epi32(v[5], cospi16);
+  u[5] = _mm_sub_epi32(u[5], x);
+  u[5] = _mm_add_epi32(u[5], rnding);
+  u[5] = _mm_srai_epi32(u[5], bit);
 
-    x = _mm_mullo_epi32(u6, cospi16);
-    y = _mm_mullo_epi32(u7, cospi48);
-    v7 = _mm_add_epi32(x, y);
-    v7 = _mm_add_epi32(v7, rnding);
-    v7 = _mm_srai_epi32(v7, bit);
+  u[6] = _mm_mullo_epi32(v[6], cospim48);
+  x = _mm_mullo_epi32(v[7], cospi16);
+  u[6] = _mm_add_epi32(u[6], x);
+  u[6] = _mm_add_epi32(u[6], rnding);
+  u[6] = _mm_srai_epi32(u[6], bit);
 
-    // stage 5
-    u0 = _mm_add_epi32(v0, v4);
-    u1 = _mm_add_epi32(v1, v5);
-    u2 = _mm_add_epi32(v2, v6);
-    u3 = _mm_add_epi32(v3, v7);
-    u4 = _mm_sub_epi32(v0, v4);
-    u5 = _mm_sub_epi32(v1, v5);
-    u6 = _mm_sub_epi32(v2, v6);
-    u7 = _mm_sub_epi32(v3, v7);
+  u[7] = _mm_mullo_epi32(v[6], cospi16);
+  x = _mm_mullo_epi32(v[7], cospim48);
+  u[7] = _mm_sub_epi32(u[7], x);
+  u[7] = _mm_add_epi32(u[7], rnding);
+  u[7] = _mm_srai_epi32(u[7], bit);
 
-    // stage 6
-    x = _mm_mullo_epi32(u0, cospi4);
-    y = _mm_mullo_epi32(u1, cospi60);
-    v0 = _mm_add_epi32(x, y);
-    v0 = _mm_add_epi32(v0, rnding);
-    v0 = _mm_srai_epi32(v0, bit);
+  // stage 5
+  v[0] = _mm_add_epi32(u[0], u[2]);
+  v[2] = _mm_sub_epi32(u[0], u[2]);
+  v[1] = _mm_add_epi32(u[1], u[3]);
+  v[3] = _mm_sub_epi32(u[1], u[3]);
+  v[4] = _mm_add_epi32(u[4], u[6]);
+  v[6] = _mm_sub_epi32(u[4], u[6]);
+  v[5] = _mm_add_epi32(u[5], u[7]);
+  v[7] = _mm_sub_epi32(u[5], u[7]);
 
-    x = _mm_mullo_epi32(u0, cospi60);
-    y = _mm_mullo_epi32(u1, cospim4);
-    v1 = _mm_add_epi32(x, y);
-    v1 = _mm_add_epi32(v1, rnding);
-    v1 = _mm_srai_epi32(v1, bit);
+  // stage 6
+  u[0] = v[0];
+  u[1] = v[1];
+  u[4] = v[4];
+  u[5] = v[5];
 
-    x = _mm_mullo_epi32(u2, cospi20);
-    y = _mm_mullo_epi32(u3, cospi44);
-    v2 = _mm_add_epi32(x, y);
-    v2 = _mm_add_epi32(v2, rnding);
-    v2 = _mm_srai_epi32(v2, bit);
+  v[0] = _mm_mullo_epi32(v[2], cospi32);
+  x = _mm_mullo_epi32(v[3], cospi32);
+  u[2] = _mm_add_epi32(v[0], x);
+  u[2] = _mm_add_epi32(u[2], rnding);
+  u[2] = _mm_srai_epi32(u[2], bit);
 
-    x = _mm_mullo_epi32(u2, cospi44);
-    y = _mm_mullo_epi32(u3, cospim20);
-    v3 = _mm_add_epi32(x, y);
-    v3 = _mm_add_epi32(v3, rnding);
-    v3 = _mm_srai_epi32(v3, bit);
+  u[3] = _mm_sub_epi32(v[0], x);
+  u[3] = _mm_add_epi32(u[3], rnding);
+  u[3] = _mm_srai_epi32(u[3], bit);
 
-    x = _mm_mullo_epi32(u4, cospi36);
-    y = _mm_mullo_epi32(u5, cospi28);
-    v4 = _mm_add_epi32(x, y);
-    v4 = _mm_add_epi32(v4, rnding);
-    v4 = _mm_srai_epi32(v4, bit);
+  v[0] = _mm_mullo_epi32(v[6], cospi32);
+  x = _mm_mullo_epi32(v[7], cospi32);
+  u[6] = _mm_add_epi32(v[0], x);
+  u[6] = _mm_add_epi32(u[6], rnding);
+  u[6] = _mm_srai_epi32(u[6], bit);
 
-    x = _mm_mullo_epi32(u4, cospi28);
-    y = _mm_mullo_epi32(u5, cospim36);
-    v5 = _mm_add_epi32(x, y);
-    v5 = _mm_add_epi32(v5, rnding);
-    v5 = _mm_srai_epi32(v5, bit);
+  u[7] = _mm_sub_epi32(v[0], x);
+  u[7] = _mm_add_epi32(u[7], rnding);
+  u[7] = _mm_srai_epi32(u[7], bit);
 
-    x = _mm_mullo_epi32(u6, cospi52);
-    y = _mm_mullo_epi32(u7, cospi12);
-    v6 = _mm_add_epi32(x, y);
-    v6 = _mm_add_epi32(v6, rnding);
-    v6 = _mm_srai_epi32(v6, bit);
+  // stage 7
+  out[0] = u[0];
+  out[2] = _mm_sub_epi32(kZero, u[4]);
+  out[4] = u[6];
+  out[6] = _mm_sub_epi32(kZero, u[2]);
+  out[8] = u[3];
+  out[10] = _mm_sub_epi32(kZero, u[7]);
+  out[12] = u[5];
+  out[14] = _mm_sub_epi32(kZero, u[1]);
 
-    x = _mm_mullo_epi32(u6, cospi12);
-    y = _mm_mullo_epi32(u7, cospim52);
-    v7 = _mm_add_epi32(x, y);
-    v7 = _mm_add_epi32(v7, rnding);
-    v7 = _mm_srai_epi32(v7, bit);
+  // Odd 8 points: 1, 3, ..., 15
+  // stage 0
+  // stage 1
+  // stage 2
+  // (1)
+  u[0] = _mm_mullo_epi32(in[15], cospi4);
+  x = _mm_mullo_epi32(in[1], cospi60);
+  u[0] = _mm_add_epi32(u[0], x);
+  u[0] = _mm_add_epi32(u[0], rnding);
+  u[0] = _mm_srai_epi32(u[0], bit);
 
-    // stage 7
-    out[2 * 0 + col] = v1;
-    out[2 * 1 + col] = v6;
-    out[2 * 2 + col] = v3;
-    out[2 * 3 + col] = v4;
-    out[2 * 4 + col] = v5;
-    out[2 * 5 + col] = v2;
-    out[2 * 6 + col] = v7;
-    out[2 * 7 + col] = v0;
-  }
+  u[1] = _mm_mullo_epi32(in[15], cospi60);
+  x = _mm_mullo_epi32(in[1], cospi4);
+  u[1] = _mm_sub_epi32(u[1], x);
+  u[1] = _mm_add_epi32(u[1], rnding);
+  u[1] = _mm_srai_epi32(u[1], bit);
+
+  // (2)
+  u[2] = _mm_mullo_epi32(in[11], cospi20);
+  x = _mm_mullo_epi32(in[5], cospi44);
+  u[2] = _mm_add_epi32(u[2], x);
+  u[2] = _mm_add_epi32(u[2], rnding);
+  u[2] = _mm_srai_epi32(u[2], bit);
+
+  u[3] = _mm_mullo_epi32(in[11], cospi44);
+  x = _mm_mullo_epi32(in[5], cospi20);
+  u[3] = _mm_sub_epi32(u[3], x);
+  u[3] = _mm_add_epi32(u[3], rnding);
+  u[3] = _mm_srai_epi32(u[3], bit);
+
+  // (3)
+  u[4] = _mm_mullo_epi32(in[7], cospi36);
+  x = _mm_mullo_epi32(in[9], cospi28);
+  u[4] = _mm_add_epi32(u[4], x);
+  u[4] = _mm_add_epi32(u[4], rnding);
+  u[4] = _mm_srai_epi32(u[4], bit);
+
+  u[5] = _mm_mullo_epi32(in[7], cospi28);
+  x = _mm_mullo_epi32(in[9], cospi36);
+  u[5] = _mm_sub_epi32(u[5], x);
+  u[5] = _mm_add_epi32(u[5], rnding);
+  u[5] = _mm_srai_epi32(u[5], bit);
+
+  // (4)
+  u[6] = _mm_mullo_epi32(in[3], cospi52);
+  x = _mm_mullo_epi32(in[13], cospi12);
+  u[6] = _mm_add_epi32(u[6], x);
+  u[6] = _mm_add_epi32(u[6], rnding);
+  u[6] = _mm_srai_epi32(u[6], bit);
+
+  u[7] = _mm_mullo_epi32(in[3], cospi12);
+  x = _mm_mullo_epi32(in[13], cospi52);
+  u[7] = _mm_sub_epi32(u[7], x);
+  u[7] = _mm_add_epi32(u[7], rnding);
+  u[7] = _mm_srai_epi32(u[7], bit);
+
+  // stage 3
+  v[0] = _mm_add_epi32(u[0], u[4]);
+  v[4] = _mm_sub_epi32(u[0], u[4]);
+  v[1] = _mm_add_epi32(u[1], u[5]);
+  v[5] = _mm_sub_epi32(u[1], u[5]);
+  v[2] = _mm_add_epi32(u[2], u[6]);
+  v[6] = _mm_sub_epi32(u[2], u[6]);
+  v[3] = _mm_add_epi32(u[3], u[7]);
+  v[7] = _mm_sub_epi32(u[3], u[7]);
+
+  // stage 4
+  u[0] = v[0];
+  u[1] = v[1];
+  u[2] = v[2];
+  u[3] = v[3];
+
+  u[4] = _mm_mullo_epi32(v[4], cospi16);
+  x = _mm_mullo_epi32(v[5], cospi48);
+  u[4] = _mm_add_epi32(u[4], x);
+  u[4] = _mm_add_epi32(u[4], rnding);
+  u[4] = _mm_srai_epi32(u[4], bit);
+
+  u[5] = _mm_mullo_epi32(v[4], cospi48);
+  x = _mm_mullo_epi32(v[5], cospi16);
+  u[5] = _mm_sub_epi32(u[5], x);
+  u[5] = _mm_add_epi32(u[5], rnding);
+  u[5] = _mm_srai_epi32(u[5], bit);
+
+  u[6] = _mm_mullo_epi32(v[6], cospim48);
+  x = _mm_mullo_epi32(v[7], cospi16);
+  u[6] = _mm_add_epi32(u[6], x);
+  u[6] = _mm_add_epi32(u[6], rnding);
+  u[6] = _mm_srai_epi32(u[6], bit);
+
+  u[7] = _mm_mullo_epi32(v[6], cospi16);
+  x = _mm_mullo_epi32(v[7], cospim48);
+  u[7] = _mm_sub_epi32(u[7], x);
+  u[7] = _mm_add_epi32(u[7], rnding);
+  u[7] = _mm_srai_epi32(u[7], bit);
+
+  // stage 5
+  v[0] = _mm_add_epi32(u[0], u[2]);
+  v[2] = _mm_sub_epi32(u[0], u[2]);
+  v[1] = _mm_add_epi32(u[1], u[3]);
+  v[3] = _mm_sub_epi32(u[1], u[3]);
+  v[4] = _mm_add_epi32(u[4], u[6]);
+  v[6] = _mm_sub_epi32(u[4], u[6]);
+  v[5] = _mm_add_epi32(u[5], u[7]);
+  v[7] = _mm_sub_epi32(u[5], u[7]);
+
+  // stage 6
+  u[0] = v[0];
+  u[1] = v[1];
+  u[4] = v[4];
+  u[5] = v[5];
+
+  v[0] = _mm_mullo_epi32(v[2], cospi32);
+  x = _mm_mullo_epi32(v[3], cospi32);
+  u[2] = _mm_add_epi32(v[0], x);
+  u[2] = _mm_add_epi32(u[2], rnding);
+  u[2] = _mm_srai_epi32(u[2], bit);
+
+  u[3] = _mm_sub_epi32(v[0], x);
+  u[3] = _mm_add_epi32(u[3], rnding);
+  u[3] = _mm_srai_epi32(u[3], bit);
+
+  v[0] = _mm_mullo_epi32(v[6], cospi32);
+  x = _mm_mullo_epi32(v[7], cospi32);
+  u[6] = _mm_add_epi32(v[0], x);
+  u[6] = _mm_add_epi32(u[6], rnding);
+  u[6] = _mm_srai_epi32(u[6], bit);
+
+  u[7] = _mm_sub_epi32(v[0], x);
+  u[7] = _mm_add_epi32(u[7], rnding);
+  u[7] = _mm_srai_epi32(u[7], bit);
+
+  // stage 7
+  out[1] = u[0];
+  out[3] = _mm_sub_epi32(kZero, u[4]);
+  out[5] = u[6];
+  out[7] = _mm_sub_epi32(kZero, u[2]);
+  out[9] = u[3];
+  out[11] = _mm_sub_epi32(kZero, u[7]);
+  out[13] = u[5];
+  out[15] = _mm_sub_epi32(kZero, u[1]);
 }
 
 static void round_shift_8x8(__m128i *in, int shift) {
@@ -1060,247 +1182,379 @@
 
 static void iadst16x16_sse4_1(__m128i *in, __m128i *out, int bit) {
   const int32_t *cospi = cospi_arr(bit);
-  const __m128i cospi32 = _mm_set1_epi32(cospi[32]);
-  const __m128i cospi48 = _mm_set1_epi32(cospi[48]);
-  const __m128i cospi16 = _mm_set1_epi32(cospi[16]);
-  const __m128i cospim16 = _mm_set1_epi32(-cospi[16]);
-  const __m128i cospim48 = _mm_set1_epi32(-cospi[48]);
-  const __m128i cospi8 = _mm_set1_epi32(cospi[8]);
-  const __m128i cospi56 = _mm_set1_epi32(cospi[56]);
-  const __m128i cospim56 = _mm_set1_epi32(-cospi[56]);
-  const __m128i cospim8 = _mm_set1_epi32(-cospi[8]);
-  const __m128i cospi24 = _mm_set1_epi32(cospi[24]);
-  const __m128i cospim24 = _mm_set1_epi32(-cospi[24]);
-  const __m128i cospim40 = _mm_set1_epi32(-cospi[40]);
-  const __m128i cospi40 = _mm_set1_epi32(cospi[40]);
   const __m128i cospi2 = _mm_set1_epi32(cospi[2]);
   const __m128i cospi62 = _mm_set1_epi32(cospi[62]);
-  const __m128i cospim2 = _mm_set1_epi32(-cospi[2]);
   const __m128i cospi10 = _mm_set1_epi32(cospi[10]);
   const __m128i cospi54 = _mm_set1_epi32(cospi[54]);
-  const __m128i cospim10 = _mm_set1_epi32(-cospi[10]);
   const __m128i cospi18 = _mm_set1_epi32(cospi[18]);
   const __m128i cospi46 = _mm_set1_epi32(cospi[46]);
-  const __m128i cospim18 = _mm_set1_epi32(-cospi[18]);
   const __m128i cospi26 = _mm_set1_epi32(cospi[26]);
   const __m128i cospi38 = _mm_set1_epi32(cospi[38]);
-  const __m128i cospim26 = _mm_set1_epi32(-cospi[26]);
   const __m128i cospi34 = _mm_set1_epi32(cospi[34]);
   const __m128i cospi30 = _mm_set1_epi32(cospi[30]);
-  const __m128i cospim34 = _mm_set1_epi32(-cospi[34]);
   const __m128i cospi42 = _mm_set1_epi32(cospi[42]);
   const __m128i cospi22 = _mm_set1_epi32(cospi[22]);
-  const __m128i cospim42 = _mm_set1_epi32(-cospi[42]);
   const __m128i cospi50 = _mm_set1_epi32(cospi[50]);
   const __m128i cospi14 = _mm_set1_epi32(cospi[14]);
-  const __m128i cospim50 = _mm_set1_epi32(-cospi[50]);
   const __m128i cospi58 = _mm_set1_epi32(cospi[58]);
   const __m128i cospi6 = _mm_set1_epi32(cospi[6]);
-  const __m128i cospim58 = _mm_set1_epi32(-cospi[58]);
+  const __m128i cospi8 = _mm_set1_epi32(cospi[8]);
+  const __m128i cospi56 = _mm_set1_epi32(cospi[56]);
+  const __m128i cospi40 = _mm_set1_epi32(cospi[40]);
+  const __m128i cospi24 = _mm_set1_epi32(cospi[24]);
+  const __m128i cospim56 = _mm_set1_epi32(-cospi[56]);
+  const __m128i cospim24 = _mm_set1_epi32(-cospi[24]);
+  const __m128i cospi48 = _mm_set1_epi32(cospi[48]);
+  const __m128i cospi16 = _mm_set1_epi32(cospi[16]);
+  const __m128i cospim48 = _mm_set1_epi32(-cospi[48]);
+  const __m128i cospi32 = _mm_set1_epi32(cospi[32]);
   const __m128i rnding = _mm_set1_epi32(1 << (bit - 1));
-  const __m128i zero = _mm_setzero_si128();
-
   __m128i u[16], v[16], x, y;
+  const int col_num = 4;
   int col;
 
-  for (col = 0; col < 4; ++col) {
+  // Calculate the column 0, 1, 2, 3
+  for (col = 0; col < col_num; ++col) {
     // stage 0
     // stage 1
-    u[0] = in[0 * 4 + col];
-    u[1] = _mm_sub_epi32(zero, in[15 * 4 + col]);
-    u[2] = _mm_sub_epi32(zero, in[7 * 4 + col]);
-    u[3] = in[8 * 4 + col];
-    u[4] = _mm_sub_epi32(zero, in[3 * 4 + col]);
-    u[5] = in[12 * 4 + col];
-    u[6] = in[4 * 4 + col];
-    u[7] = _mm_sub_epi32(zero, in[11 * 4 + col]);
-    u[8] = _mm_sub_epi32(zero, in[1 * 4 + col]);
-    u[9] = in[14 * 4 + col];
-    u[10] = in[6 * 4 + col];
-    u[11] = _mm_sub_epi32(zero, in[9 * 4 + col]);
-    u[12] = in[2 * 4 + col];
-    u[13] = _mm_sub_epi32(zero, in[13 * 4 + col]);
-    u[14] = _mm_sub_epi32(zero, in[5 * 4 + col]);
-    u[15] = in[10 * 4 + col];
-
     // stage 2
-    v[0] = u[0];
-    v[1] = u[1];
+    v[0] = _mm_mullo_epi32(in[15 * col_num + col], cospi2);
+    x = _mm_mullo_epi32(in[0 * col_num + col], cospi62);
+    v[0] = _mm_add_epi32(v[0], x);
+    v[0] = _mm_add_epi32(v[0], rnding);
+    v[0] = _mm_srai_epi32(v[0], bit);
 
-    x = _mm_mullo_epi32(u[2], cospi32);
-    y = _mm_mullo_epi32(u[3], cospi32);
-    v[2] = _mm_add_epi32(x, y);
+    v[1] = _mm_mullo_epi32(in[15 * col_num + col], cospi62);
+    x = _mm_mullo_epi32(in[0 * col_num + col], cospi2);
+    v[1] = _mm_sub_epi32(v[1], x);
+    v[1] = _mm_add_epi32(v[1], rnding);
+    v[1] = _mm_srai_epi32(v[1], bit);
+
+    v[2] = _mm_mullo_epi32(in[13 * col_num + col], cospi10);
+    x = _mm_mullo_epi32(in[2 * col_num + col], cospi54);
+    v[2] = _mm_add_epi32(v[2], x);
     v[2] = _mm_add_epi32(v[2], rnding);
     v[2] = _mm_srai_epi32(v[2], bit);
 
-    v[3] = _mm_sub_epi32(x, y);
+    v[3] = _mm_mullo_epi32(in[13 * col_num + col], cospi54);
+    x = _mm_mullo_epi32(in[2 * col_num + col], cospi10);
+    v[3] = _mm_sub_epi32(v[3], x);
     v[3] = _mm_add_epi32(v[3], rnding);
     v[3] = _mm_srai_epi32(v[3], bit);
 
-    v[4] = u[4];
-    v[5] = u[5];
+    v[4] = _mm_mullo_epi32(in[11 * col_num + col], cospi18);
+    x = _mm_mullo_epi32(in[4 * col_num + col], cospi46);
+    v[4] = _mm_add_epi32(v[4], x);
+    v[4] = _mm_add_epi32(v[4], rnding);
+    v[4] = _mm_srai_epi32(v[4], bit);
 
-    x = _mm_mullo_epi32(u[6], cospi32);
-    y = _mm_mullo_epi32(u[7], cospi32);
-    v[6] = _mm_add_epi32(x, y);
+    v[5] = _mm_mullo_epi32(in[11 * col_num + col], cospi46);
+    x = _mm_mullo_epi32(in[4 * col_num + col], cospi18);
+    v[5] = _mm_sub_epi32(v[5], x);
+    v[5] = _mm_add_epi32(v[5], rnding);
+    v[5] = _mm_srai_epi32(v[5], bit);
+
+    v[6] = _mm_mullo_epi32(in[9 * col_num + col], cospi26);
+    x = _mm_mullo_epi32(in[6 * col_num + col], cospi38);
+    v[6] = _mm_add_epi32(v[6], x);
     v[6] = _mm_add_epi32(v[6], rnding);
     v[6] = _mm_srai_epi32(v[6], bit);
 
-    v[7] = _mm_sub_epi32(x, y);
+    v[7] = _mm_mullo_epi32(in[9 * col_num + col], cospi38);
+    x = _mm_mullo_epi32(in[6 * col_num + col], cospi26);
+    v[7] = _mm_sub_epi32(v[7], x);
     v[7] = _mm_add_epi32(v[7], rnding);
     v[7] = _mm_srai_epi32(v[7], bit);
 
-    v[8] = u[8];
-    v[9] = u[9];
+    v[8] = _mm_mullo_epi32(in[7 * col_num + col], cospi34);
+    x = _mm_mullo_epi32(in[8 * col_num + col], cospi30);
+    v[8] = _mm_add_epi32(v[8], x);
+    v[8] = _mm_add_epi32(v[8], rnding);
+    v[8] = _mm_srai_epi32(v[8], bit);
 
-    x = _mm_mullo_epi32(u[10], cospi32);
-    y = _mm_mullo_epi32(u[11], cospi32);
-    v[10] = _mm_add_epi32(x, y);
+    v[9] = _mm_mullo_epi32(in[7 * col_num + col], cospi30);
+    x = _mm_mullo_epi32(in[8 * col_num + col], cospi34);
+    v[9] = _mm_sub_epi32(v[9], x);
+    v[9] = _mm_add_epi32(v[9], rnding);
+    v[9] = _mm_srai_epi32(v[9], bit);
+
+    v[10] = _mm_mullo_epi32(in[5 * col_num + col], cospi42);
+    x = _mm_mullo_epi32(in[10 * col_num + col], cospi22);
+    v[10] = _mm_add_epi32(v[10], x);
     v[10] = _mm_add_epi32(v[10], rnding);
     v[10] = _mm_srai_epi32(v[10], bit);
 
-    v[11] = _mm_sub_epi32(x, y);
+    v[11] = _mm_mullo_epi32(in[5 * col_num + col], cospi22);
+    x = _mm_mullo_epi32(in[10 * col_num + col], cospi42);
+    v[11] = _mm_sub_epi32(v[11], x);
     v[11] = _mm_add_epi32(v[11], rnding);
     v[11] = _mm_srai_epi32(v[11], bit);
 
-    v[12] = u[12];
-    v[13] = u[13];
+    v[12] = _mm_mullo_epi32(in[3 * col_num + col], cospi50);
+    x = _mm_mullo_epi32(in[12 * col_num + col], cospi14);
+    v[12] = _mm_add_epi32(v[12], x);
+    v[12] = _mm_add_epi32(v[12], rnding);
+    v[12] = _mm_srai_epi32(v[12], bit);
 
-    x = _mm_mullo_epi32(u[14], cospi32);
-    y = _mm_mullo_epi32(u[15], cospi32);
-    v[14] = _mm_add_epi32(x, y);
+    v[13] = _mm_mullo_epi32(in[3 * col_num + col], cospi14);
+    x = _mm_mullo_epi32(in[12 * col_num + col], cospi50);
+    v[13] = _mm_sub_epi32(v[13], x);
+    v[13] = _mm_add_epi32(v[13], rnding);
+    v[13] = _mm_srai_epi32(v[13], bit);
+
+    v[14] = _mm_mullo_epi32(in[1 * col_num + col], cospi58);
+    x = _mm_mullo_epi32(in[14 * col_num + col], cospi6);
+    v[14] = _mm_add_epi32(v[14], x);
     v[14] = _mm_add_epi32(v[14], rnding);
     v[14] = _mm_srai_epi32(v[14], bit);
 
-    v[15] = _mm_sub_epi32(x, y);
+    v[15] = _mm_mullo_epi32(in[1 * col_num + col], cospi6);
+    x = _mm_mullo_epi32(in[14 * col_num + col], cospi58);
+    v[15] = _mm_sub_epi32(v[15], x);
     v[15] = _mm_add_epi32(v[15], rnding);
     v[15] = _mm_srai_epi32(v[15], bit);
 
     // stage 3
-    u[0] = _mm_add_epi32(v[0], v[2]);
-    u[1] = _mm_add_epi32(v[1], v[3]);
-    u[2] = _mm_sub_epi32(v[0], v[2]);
-    u[3] = _mm_sub_epi32(v[1], v[3]);
-    u[4] = _mm_add_epi32(v[4], v[6]);
-    u[5] = _mm_add_epi32(v[5], v[7]);
-    u[6] = _mm_sub_epi32(v[4], v[6]);
-    u[7] = _mm_sub_epi32(v[5], v[7]);
-    u[8] = _mm_add_epi32(v[8], v[10]);
-    u[9] = _mm_add_epi32(v[9], v[11]);
-    u[10] = _mm_sub_epi32(v[8], v[10]);
-    u[11] = _mm_sub_epi32(v[9], v[11]);
-    u[12] = _mm_add_epi32(v[12], v[14]);
-    u[13] = _mm_add_epi32(v[13], v[15]);
-    u[14] = _mm_sub_epi32(v[12], v[14]);
-    u[15] = _mm_sub_epi32(v[13], v[15]);
+    u[0] = _mm_add_epi32(v[0], v[8]);
+    u[8] = _mm_sub_epi32(v[0], v[8]);
+    u[1] = _mm_add_epi32(v[1], v[9]);
+    u[9] = _mm_sub_epi32(v[1], v[9]);
+    u[2] = _mm_add_epi32(v[2], v[10]);
+    u[10] = _mm_sub_epi32(v[2], v[10]);
+    u[3] = _mm_add_epi32(v[3], v[11]);
+    u[11] = _mm_sub_epi32(v[3], v[11]);
+    u[4] = _mm_add_epi32(v[4], v[12]);
+    u[12] = _mm_sub_epi32(v[4], v[12]);
+    u[5] = _mm_add_epi32(v[5], v[13]);
+    u[13] = _mm_sub_epi32(v[5], v[13]);
+    u[6] = _mm_add_epi32(v[6], v[14]);
+    u[14] = _mm_sub_epi32(v[6], v[14]);
+    u[7] = _mm_add_epi32(v[7], v[15]);
+    u[15] = _mm_sub_epi32(v[7], v[15]);
 
     // stage 4
     v[0] = u[0];
     v[1] = u[1];
     v[2] = u[2];
     v[3] = u[3];
-    v[4] = half_btf_sse4_1(&cospi16, &u[4], &cospi48, &u[5], &rnding, bit);
-    v[5] = half_btf_sse4_1(&cospi48, &u[4], &cospim16, &u[5], &rnding, bit);
-    v[6] = half_btf_sse4_1(&cospim48, &u[6], &cospi16, &u[7], &rnding, bit);
-    v[7] = half_btf_sse4_1(&cospi16, &u[6], &cospi48, &u[7], &rnding, bit);
-    v[8] = u[8];
-    v[9] = u[9];
-    v[10] = u[10];
-    v[11] = u[11];
-    v[12] = half_btf_sse4_1(&cospi16, &u[12], &cospi48, &u[13], &rnding, bit);
-    v[13] = half_btf_sse4_1(&cospi48, &u[12], &cospim16, &u[13], &rnding, bit);
-    v[14] = half_btf_sse4_1(&cospim48, &u[14], &cospi16, &u[15], &rnding, bit);
-    v[15] = half_btf_sse4_1(&cospi16, &u[14], &cospi48, &u[15], &rnding, bit);
-
-    // stage 5
-    u[0] = _mm_add_epi32(v[0], v[4]);
-    u[1] = _mm_add_epi32(v[1], v[5]);
-    u[2] = _mm_add_epi32(v[2], v[6]);
-    u[3] = _mm_add_epi32(v[3], v[7]);
-    u[4] = _mm_sub_epi32(v[0], v[4]);
-    u[5] = _mm_sub_epi32(v[1], v[5]);
-    u[6] = _mm_sub_epi32(v[2], v[6]);
-    u[7] = _mm_sub_epi32(v[3], v[7]);
-    u[8] = _mm_add_epi32(v[8], v[12]);
-    u[9] = _mm_add_epi32(v[9], v[13]);
-    u[10] = _mm_add_epi32(v[10], v[14]);
-    u[11] = _mm_add_epi32(v[11], v[15]);
-    u[12] = _mm_sub_epi32(v[8], v[12]);
-    u[13] = _mm_sub_epi32(v[9], v[13]);
-    u[14] = _mm_sub_epi32(v[10], v[14]);
-    u[15] = _mm_sub_epi32(v[11], v[15]);
-
-    // stage 6
-    v[0] = u[0];
-    v[1] = u[1];
-    v[2] = u[2];
-    v[3] = u[3];
     v[4] = u[4];
     v[5] = u[5];
     v[6] = u[6];
     v[7] = u[7];
-    v[8] = half_btf_sse4_1(&cospi8, &u[8], &cospi56, &u[9], &rnding, bit);
-    v[9] = half_btf_sse4_1(&cospi56, &u[8], &cospim8, &u[9], &rnding, bit);
-    v[10] = half_btf_sse4_1(&cospi40, &u[10], &cospi24, &u[11], &rnding, bit);
-    v[11] = half_btf_sse4_1(&cospi24, &u[10], &cospim40, &u[11], &rnding, bit);
-    v[12] = half_btf_sse4_1(&cospim56, &u[12], &cospi8, &u[13], &rnding, bit);
-    v[13] = half_btf_sse4_1(&cospi8, &u[12], &cospi56, &u[13], &rnding, bit);
-    v[14] = half_btf_sse4_1(&cospim24, &u[14], &cospi40, &u[15], &rnding, bit);
-    v[15] = half_btf_sse4_1(&cospi40, &u[14], &cospi24, &u[15], &rnding, bit);
+
+    v[8] = _mm_mullo_epi32(u[8], cospi8);
+    x = _mm_mullo_epi32(u[9], cospi56);
+    v[8] = _mm_add_epi32(v[8], x);
+    v[8] = _mm_add_epi32(v[8], rnding);
+    v[8] = _mm_srai_epi32(v[8], bit);
+
+    v[9] = _mm_mullo_epi32(u[8], cospi56);
+    x = _mm_mullo_epi32(u[9], cospi8);
+    v[9] = _mm_sub_epi32(v[9], x);
+    v[9] = _mm_add_epi32(v[9], rnding);
+    v[9] = _mm_srai_epi32(v[9], bit);
+
+    v[10] = _mm_mullo_epi32(u[10], cospi40);
+    x = _mm_mullo_epi32(u[11], cospi24);
+    v[10] = _mm_add_epi32(v[10], x);
+    v[10] = _mm_add_epi32(v[10], rnding);
+    v[10] = _mm_srai_epi32(v[10], bit);
+
+    v[11] = _mm_mullo_epi32(u[10], cospi24);
+    x = _mm_mullo_epi32(u[11], cospi40);
+    v[11] = _mm_sub_epi32(v[11], x);
+    v[11] = _mm_add_epi32(v[11], rnding);
+    v[11] = _mm_srai_epi32(v[11], bit);
+
+    v[12] = _mm_mullo_epi32(u[12], cospim56);
+    x = _mm_mullo_epi32(u[13], cospi8);
+    v[12] = _mm_add_epi32(v[12], x);
+    v[12] = _mm_add_epi32(v[12], rnding);
+    v[12] = _mm_srai_epi32(v[12], bit);
+
+    v[13] = _mm_mullo_epi32(u[12], cospi8);
+    x = _mm_mullo_epi32(u[13], cospim56);
+    v[13] = _mm_sub_epi32(v[13], x);
+    v[13] = _mm_add_epi32(v[13], rnding);
+    v[13] = _mm_srai_epi32(v[13], bit);
+
+    v[14] = _mm_mullo_epi32(u[14], cospim24);
+    x = _mm_mullo_epi32(u[15], cospi40);
+    v[14] = _mm_add_epi32(v[14], x);
+    v[14] = _mm_add_epi32(v[14], rnding);
+    v[14] = _mm_srai_epi32(v[14], bit);
+
+    v[15] = _mm_mullo_epi32(u[14], cospi40);
+    x = _mm_mullo_epi32(u[15], cospim24);
+    v[15] = _mm_sub_epi32(v[15], x);
+    v[15] = _mm_add_epi32(v[15], rnding);
+    v[15] = _mm_srai_epi32(v[15], bit);
+
+    // stage 5
+    u[0] = _mm_add_epi32(v[0], v[4]);
+    u[4] = _mm_sub_epi32(v[0], v[4]);
+    u[1] = _mm_add_epi32(v[1], v[5]);
+    u[5] = _mm_sub_epi32(v[1], v[5]);
+    u[2] = _mm_add_epi32(v[2], v[6]);
+    u[6] = _mm_sub_epi32(v[2], v[6]);
+    u[3] = _mm_add_epi32(v[3], v[7]);
+    u[7] = _mm_sub_epi32(v[3], v[7]);
+    u[8] = _mm_add_epi32(v[8], v[12]);
+    u[12] = _mm_sub_epi32(v[8], v[12]);
+    u[9] = _mm_add_epi32(v[9], v[13]);
+    u[13] = _mm_sub_epi32(v[9], v[13]);
+    u[10] = _mm_add_epi32(v[10], v[14]);
+    u[14] = _mm_sub_epi32(v[10], v[14]);
+    u[11] = _mm_add_epi32(v[11], v[15]);
+    u[15] = _mm_sub_epi32(v[11], v[15]);
+
+    // stage 6
+    v[0] = u[0];
+    v[1] = u[1];
+    v[2] = u[2];
+    v[3] = u[3];
+
+    v[4] = _mm_mullo_epi32(u[4], cospi16);
+    x = _mm_mullo_epi32(u[5], cospi48);
+    v[4] = _mm_add_epi32(v[4], x);
+    v[4] = _mm_add_epi32(v[4], rnding);
+    v[4] = _mm_srai_epi32(v[4], bit);
+
+    v[5] = _mm_mullo_epi32(u[4], cospi48);
+    x = _mm_mullo_epi32(u[5], cospi16);
+    v[5] = _mm_sub_epi32(v[5], x);
+    v[5] = _mm_add_epi32(v[5], rnding);
+    v[5] = _mm_srai_epi32(v[5], bit);
+
+    v[6] = _mm_mullo_epi32(u[6], cospim48);
+    x = _mm_mullo_epi32(u[7], cospi16);
+    v[6] = _mm_add_epi32(v[6], x);
+    v[6] = _mm_add_epi32(v[6], rnding);
+    v[6] = _mm_srai_epi32(v[6], bit);
+
+    v[7] = _mm_mullo_epi32(u[6], cospi16);
+    x = _mm_mullo_epi32(u[7], cospim48);
+    v[7] = _mm_sub_epi32(v[7], x);
+    v[7] = _mm_add_epi32(v[7], rnding);
+    v[7] = _mm_srai_epi32(v[7], bit);
+
+    v[8] = u[8];
+    v[9] = u[9];
+    v[10] = u[10];
+    v[11] = u[11];
+
+    v[12] = _mm_mullo_epi32(u[12], cospi16);
+    x = _mm_mullo_epi32(u[13], cospi48);
+    v[12] = _mm_add_epi32(v[12], x);
+    v[12] = _mm_add_epi32(v[12], rnding);
+    v[12] = _mm_srai_epi32(v[12], bit);
+
+    v[13] = _mm_mullo_epi32(u[12], cospi48);
+    x = _mm_mullo_epi32(u[13], cospi16);
+    v[13] = _mm_sub_epi32(v[13], x);
+    v[13] = _mm_add_epi32(v[13], rnding);
+    v[13] = _mm_srai_epi32(v[13], bit);
+
+    v[14] = _mm_mullo_epi32(u[14], cospim48);
+    x = _mm_mullo_epi32(u[15], cospi16);
+    v[14] = _mm_add_epi32(v[14], x);
+    v[14] = _mm_add_epi32(v[14], rnding);
+    v[14] = _mm_srai_epi32(v[14], bit);
+
+    v[15] = _mm_mullo_epi32(u[14], cospi16);
+    x = _mm_mullo_epi32(u[15], cospim48);
+    v[15] = _mm_sub_epi32(v[15], x);
+    v[15] = _mm_add_epi32(v[15], rnding);
+    v[15] = _mm_srai_epi32(v[15], bit);
 
     // stage 7
-    u[0] = _mm_add_epi32(v[0], v[8]);
-    u[1] = _mm_add_epi32(v[1], v[9]);
-    u[2] = _mm_add_epi32(v[2], v[10]);
-    u[3] = _mm_add_epi32(v[3], v[11]);
-    u[4] = _mm_add_epi32(v[4], v[12]);
-    u[5] = _mm_add_epi32(v[5], v[13]);
-    u[6] = _mm_add_epi32(v[6], v[14]);
-    u[7] = _mm_add_epi32(v[7], v[15]);
-    u[8] = _mm_sub_epi32(v[0], v[8]);
-    u[9] = _mm_sub_epi32(v[1], v[9]);
-    u[10] = _mm_sub_epi32(v[2], v[10]);
-    u[11] = _mm_sub_epi32(v[3], v[11]);
-    u[12] = _mm_sub_epi32(v[4], v[12]);
-    u[13] = _mm_sub_epi32(v[5], v[13]);
-    u[14] = _mm_sub_epi32(v[6], v[14]);
-    u[15] = _mm_sub_epi32(v[7], v[15]);
+    u[0] = _mm_add_epi32(v[0], v[2]);
+    u[2] = _mm_sub_epi32(v[0], v[2]);
+    u[1] = _mm_add_epi32(v[1], v[3]);
+    u[3] = _mm_sub_epi32(v[1], v[3]);
+    u[4] = _mm_add_epi32(v[4], v[6]);
+    u[6] = _mm_sub_epi32(v[4], v[6]);
+    u[5] = _mm_add_epi32(v[5], v[7]);
+    u[7] = _mm_sub_epi32(v[5], v[7]);
+    u[8] = _mm_add_epi32(v[8], v[10]);
+    u[10] = _mm_sub_epi32(v[8], v[10]);
+    u[9] = _mm_add_epi32(v[9], v[11]);
+    u[11] = _mm_sub_epi32(v[9], v[11]);
+    u[12] = _mm_add_epi32(v[12], v[14]);
+    u[14] = _mm_sub_epi32(v[12], v[14]);
+    u[13] = _mm_add_epi32(v[13], v[15]);
+    u[15] = _mm_sub_epi32(v[13], v[15]);
 
     // stage 8
-    v[0] = half_btf_sse4_1(&cospi2, &u[0], &cospi62, &u[1], &rnding, bit);
-    v[1] = half_btf_sse4_1(&cospi62, &u[0], &cospim2, &u[1], &rnding, bit);
-    v[2] = half_btf_sse4_1(&cospi10, &u[2], &cospi54, &u[3], &rnding, bit);
-    v[3] = half_btf_sse4_1(&cospi54, &u[2], &cospim10, &u[3], &rnding, bit);
-    v[4] = half_btf_sse4_1(&cospi18, &u[4], &cospi46, &u[5], &rnding, bit);
-    v[5] = half_btf_sse4_1(&cospi46, &u[4], &cospim18, &u[5], &rnding, bit);
-    v[6] = half_btf_sse4_1(&cospi26, &u[6], &cospi38, &u[7], &rnding, bit);
-    v[7] = half_btf_sse4_1(&cospi38, &u[6], &cospim26, &u[7], &rnding, bit);
-    v[8] = half_btf_sse4_1(&cospi34, &u[8], &cospi30, &u[9], &rnding, bit);
-    v[9] = half_btf_sse4_1(&cospi30, &u[8], &cospim34, &u[9], &rnding, bit);
-    v[10] = half_btf_sse4_1(&cospi42, &u[10], &cospi22, &u[11], &rnding, bit);
-    v[11] = half_btf_sse4_1(&cospi22, &u[10], &cospim42, &u[11], &rnding, bit);
-    v[12] = half_btf_sse4_1(&cospi50, &u[12], &cospi14, &u[13], &rnding, bit);
-    v[13] = half_btf_sse4_1(&cospi14, &u[12], &cospim50, &u[13], &rnding, bit);
-    v[14] = half_btf_sse4_1(&cospi58, &u[14], &cospi6, &u[15], &rnding, bit);
-    v[15] = half_btf_sse4_1(&cospi6, &u[14], &cospim58, &u[15], &rnding, bit);
+    v[0] = u[0];
+    v[1] = u[1];
+
+    y = _mm_mullo_epi32(u[2], cospi32);
+    x = _mm_mullo_epi32(u[3], cospi32);
+    v[2] = _mm_add_epi32(y, x);
+    v[2] = _mm_add_epi32(v[2], rnding);
+    v[2] = _mm_srai_epi32(v[2], bit);
+
+    v[3] = _mm_sub_epi32(y, x);
+    v[3] = _mm_add_epi32(v[3], rnding);
+    v[3] = _mm_srai_epi32(v[3], bit);
+
+    v[4] = u[4];
+    v[5] = u[5];
+
+    y = _mm_mullo_epi32(u[6], cospi32);
+    x = _mm_mullo_epi32(u[7], cospi32);
+    v[6] = _mm_add_epi32(y, x);
+    v[6] = _mm_add_epi32(v[6], rnding);
+    v[6] = _mm_srai_epi32(v[6], bit);
+
+    v[7] = _mm_sub_epi32(y, x);
+    v[7] = _mm_add_epi32(v[7], rnding);
+    v[7] = _mm_srai_epi32(v[7], bit);
+
+    v[8] = u[8];
+    v[9] = u[9];
+
+    y = _mm_mullo_epi32(u[10], cospi32);
+    x = _mm_mullo_epi32(u[11], cospi32);
+    v[10] = _mm_add_epi32(y, x);
+    v[10] = _mm_add_epi32(v[10], rnding);
+    v[10] = _mm_srai_epi32(v[10], bit);
+
+    v[11] = _mm_sub_epi32(y, x);
+    v[11] = _mm_add_epi32(v[11], rnding);
+    v[11] = _mm_srai_epi32(v[11], bit);
+
+    v[12] = u[12];
+    v[13] = u[13];
+
+    y = _mm_mullo_epi32(u[14], cospi32);
+    x = _mm_mullo_epi32(u[15], cospi32);
+    v[14] = _mm_add_epi32(y, x);
+    v[14] = _mm_add_epi32(v[14], rnding);
+    v[14] = _mm_srai_epi32(v[14], bit);
+
+    v[15] = _mm_sub_epi32(y, x);
+    v[15] = _mm_add_epi32(v[15], rnding);
+    v[15] = _mm_srai_epi32(v[15], bit);
 
     // stage 9
-    out[0 * 4 + col] = v[1];
-    out[1 * 4 + col] = v[14];
-    out[2 * 4 + col] = v[3];
-    out[3 * 4 + col] = v[12];
-    out[4 * 4 + col] = v[5];
-    out[5 * 4 + col] = v[10];
-    out[6 * 4 + col] = v[7];
-    out[7 * 4 + col] = v[8];
-    out[8 * 4 + col] = v[9];
-    out[9 * 4 + col] = v[6];
-    out[10 * 4 + col] = v[11];
-    out[11 * 4 + col] = v[4];
-    out[12 * 4 + col] = v[13];
-    out[13 * 4 + col] = v[2];
-    out[14 * 4 + col] = v[15];
-    out[15 * 4 + col] = v[0];
+    out[0 * col_num + col] = v[0];
+    out[1 * col_num + col] = _mm_sub_epi32(_mm_set1_epi32(0), v[8]);
+    out[2 * col_num + col] = v[12];
+    out[3 * col_num + col] = _mm_sub_epi32(_mm_set1_epi32(0), v[4]);
+    out[4 * col_num + col] = v[6];
+    out[5 * col_num + col] = _mm_sub_epi32(_mm_set1_epi32(0), v[14]);
+    out[6 * col_num + col] = v[10];
+    out[7 * col_num + col] = _mm_sub_epi32(_mm_set1_epi32(0), v[2]);
+    out[8 * col_num + col] = v[3];
+    out[9 * col_num + col] = _mm_sub_epi32(_mm_set1_epi32(0), v[11]);
+    out[10 * col_num + col] = v[15];
+    out[11 * col_num + col] = _mm_sub_epi32(_mm_set1_epi32(0), v[7]);
+    out[12 * col_num + col] = v[5];
+    out[13 * col_num + col] = _mm_sub_epi32(_mm_set1_epi32(0), v[13]);
+    out[14 * col_num + col] = v[9];
+    out[15 * col_num + col] = _mm_sub_epi32(_mm_set1_epi32(0), v[1]);
   }
 }
 

diff --git a/av1/encoder/av1_fwd_txfm1d.c b/av1/encoder/av1_fwd_txfm1d.c
index 64d0316..2e66dac 100644
--- a/av1/encoder/av1_fwd_txfm1d.c
+++ b/av1/encoder/av1_fwd_txfm1d.c

@@ -30,7 +30,6 @@
   }
 #endif
 
-// TODO(angiebird): Make 1-d txfm functions static
 void av1_fdct4_new(const int32_t *input, int32_t *output, const int8_t *cos_bit,
                    const int8_t *stage_range) {
   const int32_t size = 4;
@@ -769,15 +768,16 @@
 
   // stage 1;
   stage++;
+  assert(output != input);
   bf1 = output;
-  bf1[0] = input[7];
-  bf1[1] = input[0];
-  bf1[2] = input[5];
-  bf1[3] = input[2];
-  bf1[4] = input[3];
-  bf1[5] = input[4];
-  bf1[6] = input[1];
-  bf1[7] = input[6];
+  bf1[0] = input[0];
+  bf1[1] = -input[7];
+  bf1[2] = -input[3];
+  bf1[3] = input[4];
+  bf1[4] = -input[1];
+  bf1[5] = input[6];
+  bf1[6] = input[2];
+  bf1[7] = -input[5];
   range_check(stage, input, bf1, size, stage_range[stage]);
 
   // stage 2
@@ -785,28 +785,28 @@
   cospi = cospi_arr(cos_bit[stage]);
   bf0 = output;
   bf1 = step;
-  bf1[0] = half_btf(cospi[4], bf0[0], cospi[60], bf0[1], cos_bit[stage]);
-  bf1[1] = half_btf(-cospi[4], bf0[1], cospi[60], bf0[0], cos_bit[stage]);
-  bf1[2] = half_btf(cospi[20], bf0[2], cospi[44], bf0[3], cos_bit[stage]);
-  bf1[3] = half_btf(-cospi[20], bf0[3], cospi[44], bf0[2], cos_bit[stage]);
-  bf1[4] = half_btf(cospi[36], bf0[4], cospi[28], bf0[5], cos_bit[stage]);
-  bf1[5] = half_btf(-cospi[36], bf0[5], cospi[28], bf0[4], cos_bit[stage]);
-  bf1[6] = half_btf(cospi[52], bf0[6], cospi[12], bf0[7], cos_bit[stage]);
-  bf1[7] = half_btf(-cospi[52], bf0[7], cospi[12], bf0[6], cos_bit[stage]);
+  bf1[0] = bf0[0];
+  bf1[1] = bf0[1];
+  bf1[2] = half_btf(cospi[32], bf0[2], cospi[32], bf0[3], cos_bit[stage]);
+  bf1[3] = half_btf(cospi[32], bf0[2], -cospi[32], bf0[3], cos_bit[stage]);
+  bf1[4] = bf0[4];
+  bf1[5] = bf0[5];
+  bf1[6] = half_btf(cospi[32], bf0[6], cospi[32], bf0[7], cos_bit[stage]);
+  bf1[7] = half_btf(cospi[32], bf0[6], -cospi[32], bf0[7], cos_bit[stage]);
   range_check(stage, input, bf1, size, stage_range[stage]);
 
   // stage 3
   stage++;
   bf0 = step;
   bf1 = output;
-  bf1[0] = bf0[0] + bf0[4];
-  bf1[1] = bf0[1] + bf0[5];
-  bf1[2] = bf0[2] + bf0[6];
-  bf1[3] = bf0[3] + bf0[7];
-  bf1[4] = -bf0[4] + bf0[0];
-  bf1[5] = -bf0[5] + bf0[1];
-  bf1[6] = -bf0[6] + bf0[2];
-  bf1[7] = -bf0[7] + bf0[3];
+  bf1[0] = bf0[0] + bf0[2];
+  bf1[1] = bf0[1] + bf0[3];
+  bf1[2] = bf0[0] - bf0[2];
+  bf1[3] = bf0[1] - bf0[3];
+  bf1[4] = bf0[4] + bf0[6];
+  bf1[5] = bf0[5] + bf0[7];
+  bf1[6] = bf0[4] - bf0[6];
+  bf1[7] = bf0[5] - bf0[7];
   range_check(stage, input, bf1, size, stage_range[stage]);
 
   // stage 4
@@ -819,23 +819,23 @@
   bf1[2] = bf0[2];
   bf1[3] = bf0[3];
   bf1[4] = half_btf(cospi[16], bf0[4], cospi[48], bf0[5], cos_bit[stage]);
-  bf1[5] = half_btf(-cospi[16], bf0[5], cospi[48], bf0[4], cos_bit[stage]);
+  bf1[5] = half_btf(cospi[48], bf0[4], -cospi[16], bf0[5], cos_bit[stage]);
   bf1[6] = half_btf(-cospi[48], bf0[6], cospi[16], bf0[7], cos_bit[stage]);
-  bf1[7] = half_btf(cospi[48], bf0[7], cospi[16], bf0[6], cos_bit[stage]);
+  bf1[7] = half_btf(cospi[16], bf0[6], cospi[48], bf0[7], cos_bit[stage]);
   range_check(stage, input, bf1, size, stage_range[stage]);
 
   // stage 5
   stage++;
   bf0 = step;
   bf1 = output;
-  bf1[0] = bf0[0] + bf0[2];
-  bf1[1] = bf0[1] + bf0[3];
-  bf1[2] = -bf0[2] + bf0[0];
-  bf1[3] = -bf0[3] + bf0[1];
-  bf1[4] = bf0[4] + bf0[6];
-  bf1[5] = bf0[5] + bf0[7];
-  bf1[6] = -bf0[6] + bf0[4];
-  bf1[7] = -bf0[7] + bf0[5];
+  bf1[0] = bf0[0] + bf0[4];
+  bf1[1] = bf0[1] + bf0[5];
+  bf1[2] = bf0[2] + bf0[6];
+  bf1[3] = bf0[3] + bf0[7];
+  bf1[4] = bf0[0] - bf0[4];
+  bf1[5] = bf0[1] - bf0[5];
+  bf1[6] = bf0[2] - bf0[6];
+  bf1[7] = bf0[3] - bf0[7];
   range_check(stage, input, bf1, size, stage_range[stage]);
 
   // stage 6
@@ -843,28 +843,28 @@
   cospi = cospi_arr(cos_bit[stage]);
   bf0 = output;
   bf1 = step;
-  bf1[0] = bf0[0];
-  bf1[1] = bf0[1];
-  bf1[2] = half_btf(cospi[32], bf0[2], cospi[32], bf0[3], cos_bit[stage]);
-  bf1[3] = half_btf(-cospi[32], bf0[3], cospi[32], bf0[2], cos_bit[stage]);
-  bf1[4] = bf0[4];
-  bf1[5] = bf0[5];
-  bf1[6] = half_btf(cospi[32], bf0[6], cospi[32], bf0[7], cos_bit[stage]);
-  bf1[7] = half_btf(-cospi[32], bf0[7], cospi[32], bf0[6], cos_bit[stage]);
+  bf1[0] = half_btf(cospi[4], bf0[0], cospi[60], bf0[1], cos_bit[stage]);
+  bf1[1] = half_btf(cospi[60], bf0[0], -cospi[4], bf0[1], cos_bit[stage]);
+  bf1[2] = half_btf(cospi[20], bf0[2], cospi[44], bf0[3], cos_bit[stage]);
+  bf1[3] = half_btf(cospi[44], bf0[2], -cospi[20], bf0[3], cos_bit[stage]);
+  bf1[4] = half_btf(cospi[36], bf0[4], cospi[28], bf0[5], cos_bit[stage]);
+  bf1[5] = half_btf(cospi[28], bf0[4], -cospi[36], bf0[5], cos_bit[stage]);
+  bf1[6] = half_btf(cospi[52], bf0[6], cospi[12], bf0[7], cos_bit[stage]);
+  bf1[7] = half_btf(cospi[12], bf0[6], -cospi[52], bf0[7], cos_bit[stage]);
   range_check(stage, input, bf1, size, stage_range[stage]);
 
   // stage 7
   stage++;
   bf0 = step;
   bf1 = output;
-  bf1[0] = bf0[0];
-  bf1[1] = -bf0[4];
-  bf1[2] = bf0[6];
-  bf1[3] = -bf0[2];
-  bf1[4] = bf0[3];
-  bf1[5] = -bf0[7];
-  bf1[6] = bf0[5];
-  bf1[7] = -bf0[1];
+  bf1[0] = bf0[1];
+  bf1[1] = bf0[6];
+  bf1[2] = bf0[3];
+  bf1[3] = bf0[4];
+  bf1[4] = bf0[5];
+  bf1[5] = bf0[2];
+  bf1[6] = bf0[7];
+  bf1[7] = bf0[0];
   range_check(stage, input, bf1, size, stage_range[stage]);
 }
 
@@ -882,23 +882,24 @@
 
   // stage 1;
   stage++;
+  assert(output != input);
   bf1 = output;
-  bf1[0] = input[15];
-  bf1[1] = input[0];
-  bf1[2] = input[13];
-  bf1[3] = input[2];
-  bf1[4] = input[11];
-  bf1[5] = input[4];
-  bf1[6] = input[9];
-  bf1[7] = input[6];
-  bf1[8] = input[7];
-  bf1[9] = input[8];
-  bf1[10] = input[5];
-  bf1[11] = input[10];
-  bf1[12] = input[3];
-  bf1[13] = input[12];
-  bf1[14] = input[1];
-  bf1[15] = input[14];
+  bf1[0] = input[0];
+  bf1[1] = -input[15];
+  bf1[2] = -input[7];
+  bf1[3] = input[8];
+  bf1[4] = -input[3];
+  bf1[5] = input[12];
+  bf1[6] = input[4];
+  bf1[7] = -input[11];
+  bf1[8] = -input[1];
+  bf1[9] = input[14];
+  bf1[10] = input[6];
+  bf1[11] = -input[9];
+  bf1[12] = input[2];
+  bf1[13] = -input[13];
+  bf1[14] = -input[5];
+  bf1[15] = input[10];
   range_check(stage, input, bf1, size, stage_range[stage]);
 
   // stage 2
@@ -906,44 +907,44 @@
   cospi = cospi_arr(cos_bit[stage]);
   bf0 = output;
   bf1 = step;
-  bf1[0] = half_btf(cospi[2], bf0[0], cospi[62], bf0[1], cos_bit[stage]);
-  bf1[1] = half_btf(-cospi[2], bf0[1], cospi[62], bf0[0], cos_bit[stage]);
-  bf1[2] = half_btf(cospi[10], bf0[2], cospi[54], bf0[3], cos_bit[stage]);
-  bf1[3] = half_btf(-cospi[10], bf0[3], cospi[54], bf0[2], cos_bit[stage]);
-  bf1[4] = half_btf(cospi[18], bf0[4], cospi[46], bf0[5], cos_bit[stage]);
-  bf1[5] = half_btf(-cospi[18], bf0[5], cospi[46], bf0[4], cos_bit[stage]);
-  bf1[6] = half_btf(cospi[26], bf0[6], cospi[38], bf0[7], cos_bit[stage]);
-  bf1[7] = half_btf(-cospi[26], bf0[7], cospi[38], bf0[6], cos_bit[stage]);
-  bf1[8] = half_btf(cospi[34], bf0[8], cospi[30], bf0[9], cos_bit[stage]);
-  bf1[9] = half_btf(-cospi[34], bf0[9], cospi[30], bf0[8], cos_bit[stage]);
-  bf1[10] = half_btf(cospi[42], bf0[10], cospi[22], bf0[11], cos_bit[stage]);
-  bf1[11] = half_btf(-cospi[42], bf0[11], cospi[22], bf0[10], cos_bit[stage]);
-  bf1[12] = half_btf(cospi[50], bf0[12], cospi[14], bf0[13], cos_bit[stage]);
-  bf1[13] = half_btf(-cospi[50], bf0[13], cospi[14], bf0[12], cos_bit[stage]);
-  bf1[14] = half_btf(cospi[58], bf0[14], cospi[6], bf0[15], cos_bit[stage]);
-  bf1[15] = half_btf(-cospi[58], bf0[15], cospi[6], bf0[14], cos_bit[stage]);
+  bf1[0] = bf0[0];
+  bf1[1] = bf0[1];
+  bf1[2] = half_btf(cospi[32], bf0[2], cospi[32], bf0[3], cos_bit[stage]);
+  bf1[3] = half_btf(cospi[32], bf0[2], -cospi[32], bf0[3], cos_bit[stage]);
+  bf1[4] = bf0[4];
+  bf1[5] = bf0[5];
+  bf1[6] = half_btf(cospi[32], bf0[6], cospi[32], bf0[7], cos_bit[stage]);
+  bf1[7] = half_btf(cospi[32], bf0[6], -cospi[32], bf0[7], cos_bit[stage]);
+  bf1[8] = bf0[8];
+  bf1[9] = bf0[9];
+  bf1[10] = half_btf(cospi[32], bf0[10], cospi[32], bf0[11], cos_bit[stage]);
+  bf1[11] = half_btf(cospi[32], bf0[10], -cospi[32], bf0[11], cos_bit[stage]);
+  bf1[12] = bf0[12];
+  bf1[13] = bf0[13];
+  bf1[14] = half_btf(cospi[32], bf0[14], cospi[32], bf0[15], cos_bit[stage]);
+  bf1[15] = half_btf(cospi[32], bf0[14], -cospi[32], bf0[15], cos_bit[stage]);
   range_check(stage, input, bf1, size, stage_range[stage]);
 
   // stage 3
   stage++;
   bf0 = step;
   bf1 = output;
-  bf1[0] = bf0[0] + bf0[8];
-  bf1[1] = bf0[1] + bf0[9];
-  bf1[2] = bf0[2] + bf0[10];
-  bf1[3] = bf0[3] + bf0[11];
-  bf1[4] = bf0[4] + bf0[12];
-  bf1[5] = bf0[5] + bf0[13];
-  bf1[6] = bf0[6] + bf0[14];
-  bf1[7] = bf0[7] + bf0[15];
-  bf1[8] = -bf0[8] + bf0[0];
-  bf1[9] = -bf0[9] + bf0[1];
-  bf1[10] = -bf0[10] + bf0[2];
-  bf1[11] = -bf0[11] + bf0[3];
-  bf1[12] = -bf0[12] + bf0[4];
-  bf1[13] = -bf0[13] + bf0[5];
-  bf1[14] = -bf0[14] + bf0[6];
-  bf1[15] = -bf0[15] + bf0[7];
+  bf1[0] = bf0[0] + bf0[2];
+  bf1[1] = bf0[1] + bf0[3];
+  bf1[2] = bf0[0] - bf0[2];
+  bf1[3] = bf0[1] - bf0[3];
+  bf1[4] = bf0[4] + bf0[6];
+  bf1[5] = bf0[5] + bf0[7];
+  bf1[6] = bf0[4] - bf0[6];
+  bf1[7] = bf0[5] - bf0[7];
+  bf1[8] = bf0[8] + bf0[10];
+  bf1[9] = bf0[9] + bf0[11];
+  bf1[10] = bf0[8] - bf0[10];
+  bf1[11] = bf0[9] - bf0[11];
+  bf1[12] = bf0[12] + bf0[14];
+  bf1[13] = bf0[13] + bf0[15];
+  bf1[14] = bf0[12] - bf0[14];
+  bf1[15] = bf0[13] - bf0[15];
   range_check(stage, input, bf1, size, stage_range[stage]);
 
   // stage 4
@@ -955,85 +956,85 @@
   bf1[1] = bf0[1];
   bf1[2] = bf0[2];
   bf1[3] = bf0[3];
+  bf1[4] = half_btf(cospi[16], bf0[4], cospi[48], bf0[5], cos_bit[stage]);
+  bf1[5] = half_btf(cospi[48], bf0[4], -cospi[16], bf0[5], cos_bit[stage]);
+  bf1[6] = half_btf(-cospi[48], bf0[6], cospi[16], bf0[7], cos_bit[stage]);
+  bf1[7] = half_btf(cospi[16], bf0[6], cospi[48], bf0[7], cos_bit[stage]);
+  bf1[8] = bf0[8];
+  bf1[9] = bf0[9];
+  bf1[10] = bf0[10];
+  bf1[11] = bf0[11];
+  bf1[12] = half_btf(cospi[16], bf0[12], cospi[48], bf0[13], cos_bit[stage]);
+  bf1[13] = half_btf(cospi[48], bf0[12], -cospi[16], bf0[13], cos_bit[stage]);
+  bf1[14] = half_btf(-cospi[48], bf0[14], cospi[16], bf0[15], cos_bit[stage]);
+  bf1[15] = half_btf(cospi[16], bf0[14], cospi[48], bf0[15], cos_bit[stage]);
+  range_check(stage, input, bf1, size, stage_range[stage]);
+
+  // stage 5
+  stage++;
+  bf0 = step;
+  bf1 = output;
+  bf1[0] = bf0[0] + bf0[4];
+  bf1[1] = bf0[1] + bf0[5];
+  bf1[2] = bf0[2] + bf0[6];
+  bf1[3] = bf0[3] + bf0[7];
+  bf1[4] = bf0[0] - bf0[4];
+  bf1[5] = bf0[1] - bf0[5];
+  bf1[6] = bf0[2] - bf0[6];
+  bf1[7] = bf0[3] - bf0[7];
+  bf1[8] = bf0[8] + bf0[12];
+  bf1[9] = bf0[9] + bf0[13];
+  bf1[10] = bf0[10] + bf0[14];
+  bf1[11] = bf0[11] + bf0[15];
+  bf1[12] = bf0[8] - bf0[12];
+  bf1[13] = bf0[9] - bf0[13];
+  bf1[14] = bf0[10] - bf0[14];
+  bf1[15] = bf0[11] - bf0[15];
+  range_check(stage, input, bf1, size, stage_range[stage]);
+
+  // stage 6
+  stage++;
+  cospi = cospi_arr(cos_bit[stage]);
+  bf0 = output;
+  bf1 = step;
+  bf1[0] = bf0[0];
+  bf1[1] = bf0[1];
+  bf1[2] = bf0[2];
+  bf1[3] = bf0[3];
   bf1[4] = bf0[4];
   bf1[5] = bf0[5];
   bf1[6] = bf0[6];
   bf1[7] = bf0[7];
   bf1[8] = half_btf(cospi[8], bf0[8], cospi[56], bf0[9], cos_bit[stage]);
-  bf1[9] = half_btf(-cospi[8], bf0[9], cospi[56], bf0[8], cos_bit[stage]);
+  bf1[9] = half_btf(cospi[56], bf0[8], -cospi[8], bf0[9], cos_bit[stage]);
   bf1[10] = half_btf(cospi[40], bf0[10], cospi[24], bf0[11], cos_bit[stage]);
-  bf1[11] = half_btf(-cospi[40], bf0[11], cospi[24], bf0[10], cos_bit[stage]);
+  bf1[11] = half_btf(cospi[24], bf0[10], -cospi[40], bf0[11], cos_bit[stage]);
   bf1[12] = half_btf(-cospi[56], bf0[12], cospi[8], bf0[13], cos_bit[stage]);
-  bf1[13] = half_btf(cospi[56], bf0[13], cospi[8], bf0[12], cos_bit[stage]);
+  bf1[13] = half_btf(cospi[8], bf0[12], cospi[56], bf0[13], cos_bit[stage]);
   bf1[14] = half_btf(-cospi[24], bf0[14], cospi[40], bf0[15], cos_bit[stage]);
-  bf1[15] = half_btf(cospi[24], bf0[15], cospi[40], bf0[14], cos_bit[stage]);
-  range_check(stage, input, bf1, size, stage_range[stage]);
-
-  // stage 5
-  stage++;
-  bf0 = step;
-  bf1 = output;
-  bf1[0] = bf0[0] + bf0[4];
-  bf1[1] = bf0[1] + bf0[5];
-  bf1[2] = bf0[2] + bf0[6];
-  bf1[3] = bf0[3] + bf0[7];
-  bf1[4] = -bf0[4] + bf0[0];
-  bf1[5] = -bf0[5] + bf0[1];
-  bf1[6] = -bf0[6] + bf0[2];
-  bf1[7] = -bf0[7] + bf0[3];
-  bf1[8] = bf0[8] + bf0[12];
-  bf1[9] = bf0[9] + bf0[13];
-  bf1[10] = bf0[10] + bf0[14];
-  bf1[11] = bf0[11] + bf0[15];
-  bf1[12] = -bf0[12] + bf0[8];
-  bf1[13] = -bf0[13] + bf0[9];
-  bf1[14] = -bf0[14] + bf0[10];
-  bf1[15] = -bf0[15] + bf0[11];
-  range_check(stage, input, bf1, size, stage_range[stage]);
-
-  // stage 6
-  stage++;
-  cospi = cospi_arr(cos_bit[stage]);
-  bf0 = output;
-  bf1 = step;
-  bf1[0] = bf0[0];
-  bf1[1] = bf0[1];
-  bf1[2] = bf0[2];
-  bf1[3] = bf0[3];
-  bf1[4] = half_btf(cospi[16], bf0[4], cospi[48], bf0[5], cos_bit[stage]);
-  bf1[5] = half_btf(-cospi[16], bf0[5], cospi[48], bf0[4], cos_bit[stage]);
-  bf1[6] = half_btf(-cospi[48], bf0[6], cospi[16], bf0[7], cos_bit[stage]);
-  bf1[7] = half_btf(cospi[48], bf0[7], cospi[16], bf0[6], cos_bit[stage]);
-  bf1[8] = bf0[8];
-  bf1[9] = bf0[9];
-  bf1[10] = bf0[10];
-  bf1[11] = bf0[11];
-  bf1[12] = half_btf(cospi[16], bf0[12], cospi[48], bf0[13], cos_bit[stage]);
-  bf1[13] = half_btf(-cospi[16], bf0[13], cospi[48], bf0[12], cos_bit[stage]);
-  bf1[14] = half_btf(-cospi[48], bf0[14], cospi[16], bf0[15], cos_bit[stage]);
-  bf1[15] = half_btf(cospi[48], bf0[15], cospi[16], bf0[14], cos_bit[stage]);
+  bf1[15] = half_btf(cospi[40], bf0[14], cospi[24], bf0[15], cos_bit[stage]);
   range_check(stage, input, bf1, size, stage_range[stage]);
 
   // stage 7
   stage++;
   bf0 = step;
   bf1 = output;
-  bf1[0] = bf0[0] + bf0[2];
-  bf1[1] = bf0[1] + bf0[3];
-  bf1[2] = -bf0[2] + bf0[0];
-  bf1[3] = -bf0[3] + bf0[1];
-  bf1[4] = bf0[4] + bf0[6];
-  bf1[5] = bf0[5] + bf0[7];
-  bf1[6] = -bf0[6] + bf0[4];
-  bf1[7] = -bf0[7] + bf0[5];
-  bf1[8] = bf0[8] + bf0[10];
-  bf1[9] = bf0[9] + bf0[11];
-  bf1[10] = -bf0[10] + bf0[8];
-  bf1[11] = -bf0[11] + bf0[9];
-  bf1[12] = bf0[12] + bf0[14];
-  bf1[13] = bf0[13] + bf0[15];
-  bf1[14] = -bf0[14] + bf0[12];
-  bf1[15] = -bf0[15] + bf0[13];
+  bf1[0] = bf0[0] + bf0[8];
+  bf1[1] = bf0[1] + bf0[9];
+  bf1[2] = bf0[2] + bf0[10];
+  bf1[3] = bf0[3] + bf0[11];
+  bf1[4] = bf0[4] + bf0[12];
+  bf1[5] = bf0[5] + bf0[13];
+  bf1[6] = bf0[6] + bf0[14];
+  bf1[7] = bf0[7] + bf0[15];
+  bf1[8] = bf0[0] - bf0[8];
+  bf1[9] = bf0[1] - bf0[9];
+  bf1[10] = bf0[2] - bf0[10];
+  bf1[11] = bf0[3] - bf0[11];
+  bf1[12] = bf0[4] - bf0[12];
+  bf1[13] = bf0[5] - bf0[13];
+  bf1[14] = bf0[6] - bf0[14];
+  bf1[15] = bf0[7] - bf0[15];
   range_check(stage, input, bf1, size, stage_range[stage]);
 
   // stage 8
@@ -1041,44 +1042,44 @@
   cospi = cospi_arr(cos_bit[stage]);
   bf0 = output;
   bf1 = step;
-  bf1[0] = bf0[0];
-  bf1[1] = bf0[1];
-  bf1[2] = half_btf(cospi[32], bf0[2], cospi[32], bf0[3], cos_bit[stage]);
-  bf1[3] = half_btf(-cospi[32], bf0[3], cospi[32], bf0[2], cos_bit[stage]);
-  bf1[4] = bf0[4];
-  bf1[5] = bf0[5];
-  bf1[6] = half_btf(cospi[32], bf0[6], cospi[32], bf0[7], cos_bit[stage]);
-  bf1[7] = half_btf(-cospi[32], bf0[7], cospi[32], bf0[6], cos_bit[stage]);
-  bf1[8] = bf0[8];
-  bf1[9] = bf0[9];
-  bf1[10] = half_btf(cospi[32], bf0[10], cospi[32], bf0[11], cos_bit[stage]);
-  bf1[11] = half_btf(-cospi[32], bf0[11], cospi[32], bf0[10], cos_bit[stage]);
-  bf1[12] = bf0[12];
-  bf1[13] = bf0[13];
-  bf1[14] = half_btf(cospi[32], bf0[14], cospi[32], bf0[15], cos_bit[stage]);
-  bf1[15] = half_btf(-cospi[32], bf0[15], cospi[32], bf0[14], cos_bit[stage]);
+  bf1[0] = half_btf(cospi[2], bf0[0], cospi[62], bf0[1], cos_bit[stage]);
+  bf1[1] = half_btf(cospi[62], bf0[0], -cospi[2], bf0[1], cos_bit[stage]);
+  bf1[2] = half_btf(cospi[10], bf0[2], cospi[54], bf0[3], cos_bit[stage]);
+  bf1[3] = half_btf(cospi[54], bf0[2], -cospi[10], bf0[3], cos_bit[stage]);
+  bf1[4] = half_btf(cospi[18], bf0[4], cospi[46], bf0[5], cos_bit[stage]);
+  bf1[5] = half_btf(cospi[46], bf0[4], -cospi[18], bf0[5], cos_bit[stage]);
+  bf1[6] = half_btf(cospi[26], bf0[6], cospi[38], bf0[7], cos_bit[stage]);
+  bf1[7] = half_btf(cospi[38], bf0[6], -cospi[26], bf0[7], cos_bit[stage]);
+  bf1[8] = half_btf(cospi[34], bf0[8], cospi[30], bf0[9], cos_bit[stage]);
+  bf1[9] = half_btf(cospi[30], bf0[8], -cospi[34], bf0[9], cos_bit[stage]);
+  bf1[10] = half_btf(cospi[42], bf0[10], cospi[22], bf0[11], cos_bit[stage]);
+  bf1[11] = half_btf(cospi[22], bf0[10], -cospi[42], bf0[11], cos_bit[stage]);
+  bf1[12] = half_btf(cospi[50], bf0[12], cospi[14], bf0[13], cos_bit[stage]);
+  bf1[13] = half_btf(cospi[14], bf0[12], -cospi[50], bf0[13], cos_bit[stage]);
+  bf1[14] = half_btf(cospi[58], bf0[14], cospi[6], bf0[15], cos_bit[stage]);
+  bf1[15] = half_btf(cospi[6], bf0[14], -cospi[58], bf0[15], cos_bit[stage]);
   range_check(stage, input, bf1, size, stage_range[stage]);
 
   // stage 9
   stage++;
   bf0 = step;
   bf1 = output;
-  bf1[0] = bf0[0];
-  bf1[1] = -bf0[8];
-  bf1[2] = bf0[12];
-  bf1[3] = -bf0[4];
-  bf1[4] = bf0[6];
-  bf1[5] = -bf0[14];
-  bf1[6] = bf0[10];
-  bf1[7] = -bf0[2];
-  bf1[8] = bf0[3];
-  bf1[9] = -bf0[11];
-  bf1[10] = bf0[15];
-  bf1[11] = -bf0[7];
-  bf1[12] = bf0[5];
-  bf1[13] = -bf0[13];
-  bf1[14] = bf0[9];
-  bf1[15] = -bf0[1];
+  bf1[0] = bf0[1];
+  bf1[1] = bf0[14];
+  bf1[2] = bf0[3];
+  bf1[3] = bf0[12];
+  bf1[4] = bf0[5];
+  bf1[5] = bf0[10];
+  bf1[6] = bf0[7];
+  bf1[7] = bf0[8];
+  bf1[8] = bf0[9];
+  bf1[9] = bf0[6];
+  bf1[10] = bf0[11];
+  bf1[11] = bf0[4];
+  bf1[12] = bf0[13];
+  bf1[13] = bf0[2];
+  bf1[14] = bf0[15];
+  bf1[15] = bf0[0];
   range_check(stage, input, bf1, size, stage_range[stage]);
 }
 

diff --git a/av1/encoder/x86/highbd_fwd_txfm_sse4.c b/av1/encoder/x86/highbd_fwd_txfm_sse4.c
index 9fc900d..d772761 100644
--- a/av1/encoder/x86/highbd_fwd_txfm_sse4.c
+++ b/av1/encoder/x86/highbd_fwd_txfm_sse4.c

@@ -622,305 +622,183 @@
 
 static void fadst8x8_sse4_1(__m128i *in, __m128i *out, int bit) {
   const int32_t *cospi = cospi_arr(bit);
-  const __m128i cospi4 = _mm_set1_epi32(cospi[4]);
-  const __m128i cospi60 = _mm_set1_epi32(cospi[60]);
-  const __m128i cospi20 = _mm_set1_epi32(cospi[20]);
-  const __m128i cospi44 = _mm_set1_epi32(cospi[44]);
-  const __m128i cospi36 = _mm_set1_epi32(cospi[36]);
-  const __m128i cospi28 = _mm_set1_epi32(cospi[28]);
-  const __m128i cospi52 = _mm_set1_epi32(cospi[52]);
-  const __m128i cospi12 = _mm_set1_epi32(cospi[12]);
+  const __m128i cospi32 = _mm_set1_epi32(cospi[32]);
   const __m128i cospi16 = _mm_set1_epi32(cospi[16]);
+  const __m128i cospim16 = _mm_set1_epi32(-cospi[16]);
   const __m128i cospi48 = _mm_set1_epi32(cospi[48]);
   const __m128i cospim48 = _mm_set1_epi32(-cospi[48]);
-  const __m128i cospi32 = _mm_set1_epi32(cospi[32]);
+  const __m128i cospi4 = _mm_set1_epi32(cospi[4]);
+  const __m128i cospim4 = _mm_set1_epi32(-cospi[4]);
+  const __m128i cospi60 = _mm_set1_epi32(cospi[60]);
+  const __m128i cospi20 = _mm_set1_epi32(cospi[20]);
+  const __m128i cospim20 = _mm_set1_epi32(-cospi[20]);
+  const __m128i cospi44 = _mm_set1_epi32(cospi[44]);
+  const __m128i cospi28 = _mm_set1_epi32(cospi[28]);
+  const __m128i cospi36 = _mm_set1_epi32(cospi[36]);
+  const __m128i cospim36 = _mm_set1_epi32(-cospi[36]);
+  const __m128i cospi52 = _mm_set1_epi32(cospi[52]);
+  const __m128i cospim52 = _mm_set1_epi32(-cospi[52]);
+  const __m128i cospi12 = _mm_set1_epi32(cospi[12]);
   const __m128i rnding = _mm_set1_epi32(1 << (bit - 1));
-  const __m128i kZero = _mm_setzero_si128();
-  __m128i u[8], v[8], x;
+  const __m128i zero = _mm_setzero_si128();
+  __m128i u0, u1, u2, u3, u4, u5, u6, u7;
+  __m128i v0, v1, v2, v3, v4, v5, v6, v7;
+  __m128i x, y;
+  int col;
 
-  // Even 8 points: 0, 2, ..., 14
-  // stage 0
-  // stage 1
-  // stage 2
-  // (1)
-  u[0] = _mm_mullo_epi32(in[14], cospi4);
-  x = _mm_mullo_epi32(in[0], cospi60);
-  u[0] = _mm_add_epi32(u[0], x);
-  u[0] = _mm_add_epi32(u[0], rnding);
-  u[0] = _mm_srai_epi32(u[0], bit);
+  // Note:
+  //  Even column: 0, 2, ..., 14
+  //  Odd column: 1, 3, ..., 15
+  //  one even column plus one odd column constructs one row (8 coeffs)
+  //  total we have 8 rows (8x8).
+  for (col = 0; col < 2; ++col) {
+    // stage 0
+    // stage 1
+    u0 = in[2 * 0 + col];
+    u1 = _mm_sub_epi32(zero, in[2 * 7 + col]);
+    u2 = _mm_sub_epi32(zero, in[2 * 3 + col]);
+    u3 = in[2 * 4 + col];
+    u4 = _mm_sub_epi32(zero, in[2 * 1 + col]);
+    u5 = in[2 * 6 + col];
+    u6 = in[2 * 2 + col];
+    u7 = _mm_sub_epi32(zero, in[2 * 5 + col]);
 
-  u[1] = _mm_mullo_epi32(in[14], cospi60);
-  x = _mm_mullo_epi32(in[0], cospi4);
-  u[1] = _mm_sub_epi32(u[1], x);
-  u[1] = _mm_add_epi32(u[1], rnding);
-  u[1] = _mm_srai_epi32(u[1], bit);
+    // stage 2
+    v0 = u0;
+    v1 = u1;
 
-  // (2)
-  u[2] = _mm_mullo_epi32(in[10], cospi20);
-  x = _mm_mullo_epi32(in[4], cospi44);
-  u[2] = _mm_add_epi32(u[2], x);
-  u[2] = _mm_add_epi32(u[2], rnding);
-  u[2] = _mm_srai_epi32(u[2], bit);
+    x = _mm_mullo_epi32(u2, cospi32);
+    y = _mm_mullo_epi32(u3, cospi32);
+    v2 = _mm_add_epi32(x, y);
+    v2 = _mm_add_epi32(v2, rnding);
+    v2 = _mm_srai_epi32(v2, bit);
 
-  u[3] = _mm_mullo_epi32(in[10], cospi44);
-  x = _mm_mullo_epi32(in[4], cospi20);
-  u[3] = _mm_sub_epi32(u[3], x);
-  u[3] = _mm_add_epi32(u[3], rnding);
-  u[3] = _mm_srai_epi32(u[3], bit);
+    v3 = _mm_sub_epi32(x, y);
+    v3 = _mm_add_epi32(v3, rnding);
+    v3 = _mm_srai_epi32(v3, bit);
 
-  // (3)
-  u[4] = _mm_mullo_epi32(in[6], cospi36);
-  x = _mm_mullo_epi32(in[8], cospi28);
-  u[4] = _mm_add_epi32(u[4], x);
-  u[4] = _mm_add_epi32(u[4], rnding);
-  u[4] = _mm_srai_epi32(u[4], bit);
+    v4 = u4;
+    v5 = u5;
 
-  u[5] = _mm_mullo_epi32(in[6], cospi28);
-  x = _mm_mullo_epi32(in[8], cospi36);
-  u[5] = _mm_sub_epi32(u[5], x);
-  u[5] = _mm_add_epi32(u[5], rnding);
-  u[5] = _mm_srai_epi32(u[5], bit);
+    x = _mm_mullo_epi32(u6, cospi32);
+    y = _mm_mullo_epi32(u7, cospi32);
+    v6 = _mm_add_epi32(x, y);
+    v6 = _mm_add_epi32(v6, rnding);
+    v6 = _mm_srai_epi32(v6, bit);
 
-  // (4)
-  u[6] = _mm_mullo_epi32(in[2], cospi52);
-  x = _mm_mullo_epi32(in[12], cospi12);
-  u[6] = _mm_add_epi32(u[6], x);
-  u[6] = _mm_add_epi32(u[6], rnding);
-  u[6] = _mm_srai_epi32(u[6], bit);
+    v7 = _mm_sub_epi32(x, y);
+    v7 = _mm_add_epi32(v7, rnding);
+    v7 = _mm_srai_epi32(v7, bit);
 
-  u[7] = _mm_mullo_epi32(in[2], cospi12);
-  x = _mm_mullo_epi32(in[12], cospi52);
-  u[7] = _mm_sub_epi32(u[7], x);
-  u[7] = _mm_add_epi32(u[7], rnding);
-  u[7] = _mm_srai_epi32(u[7], bit);
+    // stage 3
+    u0 = _mm_add_epi32(v0, v2);
+    u1 = _mm_add_epi32(v1, v3);
+    u2 = _mm_sub_epi32(v0, v2);
+    u3 = _mm_sub_epi32(v1, v3);
+    u4 = _mm_add_epi32(v4, v6);
+    u5 = _mm_add_epi32(v5, v7);
+    u6 = _mm_sub_epi32(v4, v6);
+    u7 = _mm_sub_epi32(v5, v7);
 
-  // stage 3
-  v[0] = _mm_add_epi32(u[0], u[4]);
-  v[4] = _mm_sub_epi32(u[0], u[4]);
-  v[1] = _mm_add_epi32(u[1], u[5]);
-  v[5] = _mm_sub_epi32(u[1], u[5]);
-  v[2] = _mm_add_epi32(u[2], u[6]);
-  v[6] = _mm_sub_epi32(u[2], u[6]);
-  v[3] = _mm_add_epi32(u[3], u[7]);
-  v[7] = _mm_sub_epi32(u[3], u[7]);
+    // stage 4
+    v0 = u0;
+    v1 = u1;
+    v2 = u2;
+    v3 = u3;
 
-  // stage 4
-  u[0] = v[0];
-  u[1] = v[1];
-  u[2] = v[2];
-  u[3] = v[3];
+    x = _mm_mullo_epi32(u4, cospi16);
+    y = _mm_mullo_epi32(u5, cospi48);
+    v4 = _mm_add_epi32(x, y);
+    v4 = _mm_add_epi32(v4, rnding);
+    v4 = _mm_srai_epi32(v4, bit);
 
-  u[4] = _mm_mullo_epi32(v[4], cospi16);
-  x = _mm_mullo_epi32(v[5], cospi48);
-  u[4] = _mm_add_epi32(u[4], x);
-  u[4] = _mm_add_epi32(u[4], rnding);
-  u[4] = _mm_srai_epi32(u[4], bit);
+    x = _mm_mullo_epi32(u4, cospi48);
+    y = _mm_mullo_epi32(u5, cospim16);
+    v5 = _mm_add_epi32(x, y);
+    v5 = _mm_add_epi32(v5, rnding);
+    v5 = _mm_srai_epi32(v5, bit);
 
-  u[5] = _mm_mullo_epi32(v[4], cospi48);
-  x = _mm_mullo_epi32(v[5], cospi16);
-  u[5] = _mm_sub_epi32(u[5], x);
-  u[5] = _mm_add_epi32(u[5], rnding);
-  u[5] = _mm_srai_epi32(u[5], bit);
+    x = _mm_mullo_epi32(u6, cospim48);
+    y = _mm_mullo_epi32(u7, cospi16);
+    v6 = _mm_add_epi32(x, y);
+    v6 = _mm_add_epi32(v6, rnding);
+    v6 = _mm_srai_epi32(v6, bit);
 
-  u[6] = _mm_mullo_epi32(v[6], cospim48);
-  x = _mm_mullo_epi32(v[7], cospi16);
-  u[6] = _mm_add_epi32(u[6], x);
-  u[6] = _mm_add_epi32(u[6], rnding);
-  u[6] = _mm_srai_epi32(u[6], bit);
+    x = _mm_mullo_epi32(u6, cospi16);
+    y = _mm_mullo_epi32(u7, cospi48);
+    v7 = _mm_add_epi32(x, y);
+    v7 = _mm_add_epi32(v7, rnding);
+    v7 = _mm_srai_epi32(v7, bit);
 
-  u[7] = _mm_mullo_epi32(v[6], cospi16);
-  x = _mm_mullo_epi32(v[7], cospim48);
-  u[7] = _mm_sub_epi32(u[7], x);
-  u[7] = _mm_add_epi32(u[7], rnding);
-  u[7] = _mm_srai_epi32(u[7], bit);
+    // stage 5
+    u0 = _mm_add_epi32(v0, v4);
+    u1 = _mm_add_epi32(v1, v5);
+    u2 = _mm_add_epi32(v2, v6);
+    u3 = _mm_add_epi32(v3, v7);
+    u4 = _mm_sub_epi32(v0, v4);
+    u5 = _mm_sub_epi32(v1, v5);
+    u6 = _mm_sub_epi32(v2, v6);
+    u7 = _mm_sub_epi32(v3, v7);
 
-  // stage 5
-  v[0] = _mm_add_epi32(u[0], u[2]);
-  v[2] = _mm_sub_epi32(u[0], u[2]);
-  v[1] = _mm_add_epi32(u[1], u[3]);
-  v[3] = _mm_sub_epi32(u[1], u[3]);
-  v[4] = _mm_add_epi32(u[4], u[6]);
-  v[6] = _mm_sub_epi32(u[4], u[6]);
-  v[5] = _mm_add_epi32(u[5], u[7]);
-  v[7] = _mm_sub_epi32(u[5], u[7]);
+    // stage 6
+    x = _mm_mullo_epi32(u0, cospi4);
+    y = _mm_mullo_epi32(u1, cospi60);
+    v0 = _mm_add_epi32(x, y);
+    v0 = _mm_add_epi32(v0, rnding);
+    v0 = _mm_srai_epi32(v0, bit);
 
-  // stage 6
-  u[0] = v[0];
-  u[1] = v[1];
-  u[4] = v[4];
-  u[5] = v[5];
+    x = _mm_mullo_epi32(u0, cospi60);
+    y = _mm_mullo_epi32(u1, cospim4);
+    v1 = _mm_add_epi32(x, y);
+    v1 = _mm_add_epi32(v1, rnding);
+    v1 = _mm_srai_epi32(v1, bit);
 
-  v[0] = _mm_mullo_epi32(v[2], cospi32);
-  x = _mm_mullo_epi32(v[3], cospi32);
-  u[2] = _mm_add_epi32(v[0], x);
-  u[2] = _mm_add_epi32(u[2], rnding);
-  u[2] = _mm_srai_epi32(u[2], bit);
+    x = _mm_mullo_epi32(u2, cospi20);
+    y = _mm_mullo_epi32(u3, cospi44);
+    v2 = _mm_add_epi32(x, y);
+    v2 = _mm_add_epi32(v2, rnding);
+    v2 = _mm_srai_epi32(v2, bit);
 
-  u[3] = _mm_sub_epi32(v[0], x);
-  u[3] = _mm_add_epi32(u[3], rnding);
-  u[3] = _mm_srai_epi32(u[3], bit);
+    x = _mm_mullo_epi32(u2, cospi44);
+    y = _mm_mullo_epi32(u3, cospim20);
+    v3 = _mm_add_epi32(x, y);
+    v3 = _mm_add_epi32(v3, rnding);
+    v3 = _mm_srai_epi32(v3, bit);
 
-  v[0] = _mm_mullo_epi32(v[6], cospi32);
-  x = _mm_mullo_epi32(v[7], cospi32);
-  u[6] = _mm_add_epi32(v[0], x);
-  u[6] = _mm_add_epi32(u[6], rnding);
-  u[6] = _mm_srai_epi32(u[6], bit);
+    x = _mm_mullo_epi32(u4, cospi36);
+    y = _mm_mullo_epi32(u5, cospi28);
+    v4 = _mm_add_epi32(x, y);
+    v4 = _mm_add_epi32(v4, rnding);
+    v4 = _mm_srai_epi32(v4, bit);
 
-  u[7] = _mm_sub_epi32(v[0], x);
-  u[7] = _mm_add_epi32(u[7], rnding);
-  u[7] = _mm_srai_epi32(u[7], bit);
+    x = _mm_mullo_epi32(u4, cospi28);
+    y = _mm_mullo_epi32(u5, cospim36);
+    v5 = _mm_add_epi32(x, y);
+    v5 = _mm_add_epi32(v5, rnding);
+    v5 = _mm_srai_epi32(v5, bit);
 
-  // stage 7
-  out[0] = u[0];
-  out[2] = _mm_sub_epi32(kZero, u[4]);
-  out[4] = u[6];
-  out[6] = _mm_sub_epi32(kZero, u[2]);
-  out[8] = u[3];
-  out[10] = _mm_sub_epi32(kZero, u[7]);
-  out[12] = u[5];
-  out[14] = _mm_sub_epi32(kZero, u[1]);
+    x = _mm_mullo_epi32(u6, cospi52);
+    y = _mm_mullo_epi32(u7, cospi12);
+    v6 = _mm_add_epi32(x, y);
+    v6 = _mm_add_epi32(v6, rnding);
+    v6 = _mm_srai_epi32(v6, bit);
 
-  // Odd 8 points: 1, 3, ..., 15
-  // stage 0
-  // stage 1
-  // stage 2
-  // (1)
-  u[0] = _mm_mullo_epi32(in[15], cospi4);
-  x = _mm_mullo_epi32(in[1], cospi60);
-  u[0] = _mm_add_epi32(u[0], x);
-  u[0] = _mm_add_epi32(u[0], rnding);
-  u[0] = _mm_srai_epi32(u[0], bit);
+    x = _mm_mullo_epi32(u6, cospi12);
+    y = _mm_mullo_epi32(u7, cospim52);
+    v7 = _mm_add_epi32(x, y);
+    v7 = _mm_add_epi32(v7, rnding);
+    v7 = _mm_srai_epi32(v7, bit);
 
-  u[1] = _mm_mullo_epi32(in[15], cospi60);
-  x = _mm_mullo_epi32(in[1], cospi4);
-  u[1] = _mm_sub_epi32(u[1], x);
-  u[1] = _mm_add_epi32(u[1], rnding);
-  u[1] = _mm_srai_epi32(u[1], bit);
-
-  // (2)
-  u[2] = _mm_mullo_epi32(in[11], cospi20);
-  x = _mm_mullo_epi32(in[5], cospi44);
-  u[2] = _mm_add_epi32(u[2], x);
-  u[2] = _mm_add_epi32(u[2], rnding);
-  u[2] = _mm_srai_epi32(u[2], bit);
-
-  u[3] = _mm_mullo_epi32(in[11], cospi44);
-  x = _mm_mullo_epi32(in[5], cospi20);
-  u[3] = _mm_sub_epi32(u[3], x);
-  u[3] = _mm_add_epi32(u[3], rnding);
-  u[3] = _mm_srai_epi32(u[3], bit);
-
-  // (3)
-  u[4] = _mm_mullo_epi32(in[7], cospi36);
-  x = _mm_mullo_epi32(in[9], cospi28);
-  u[4] = _mm_add_epi32(u[4], x);
-  u[4] = _mm_add_epi32(u[4], rnding);
-  u[4] = _mm_srai_epi32(u[4], bit);
-
-  u[5] = _mm_mullo_epi32(in[7], cospi28);
-  x = _mm_mullo_epi32(in[9], cospi36);
-  u[5] = _mm_sub_epi32(u[5], x);
-  u[5] = _mm_add_epi32(u[5], rnding);
-  u[5] = _mm_srai_epi32(u[5], bit);
-
-  // (4)
-  u[6] = _mm_mullo_epi32(in[3], cospi52);
-  x = _mm_mullo_epi32(in[13], cospi12);
-  u[6] = _mm_add_epi32(u[6], x);
-  u[6] = _mm_add_epi32(u[6], rnding);
-  u[6] = _mm_srai_epi32(u[6], bit);
-
-  u[7] = _mm_mullo_epi32(in[3], cospi12);
-  x = _mm_mullo_epi32(in[13], cospi52);
-  u[7] = _mm_sub_epi32(u[7], x);
-  u[7] = _mm_add_epi32(u[7], rnding);
-  u[7] = _mm_srai_epi32(u[7], bit);
-
-  // stage 3
-  v[0] = _mm_add_epi32(u[0], u[4]);
-  v[4] = _mm_sub_epi32(u[0], u[4]);
-  v[1] = _mm_add_epi32(u[1], u[5]);
-  v[5] = _mm_sub_epi32(u[1], u[5]);
-  v[2] = _mm_add_epi32(u[2], u[6]);
-  v[6] = _mm_sub_epi32(u[2], u[6]);
-  v[3] = _mm_add_epi32(u[3], u[7]);
-  v[7] = _mm_sub_epi32(u[3], u[7]);
-
-  // stage 4
-  u[0] = v[0];
-  u[1] = v[1];
-  u[2] = v[2];
-  u[3] = v[3];
-
-  u[4] = _mm_mullo_epi32(v[4], cospi16);
-  x = _mm_mullo_epi32(v[5], cospi48);
-  u[4] = _mm_add_epi32(u[4], x);
-  u[4] = _mm_add_epi32(u[4], rnding);
-  u[4] = _mm_srai_epi32(u[4], bit);
-
-  u[5] = _mm_mullo_epi32(v[4], cospi48);
-  x = _mm_mullo_epi32(v[5], cospi16);
-  u[5] = _mm_sub_epi32(u[5], x);
-  u[5] = _mm_add_epi32(u[5], rnding);
-  u[5] = _mm_srai_epi32(u[5], bit);
-
-  u[6] = _mm_mullo_epi32(v[6], cospim48);
-  x = _mm_mullo_epi32(v[7], cospi16);
-  u[6] = _mm_add_epi32(u[6], x);
-  u[6] = _mm_add_epi32(u[6], rnding);
-  u[6] = _mm_srai_epi32(u[6], bit);
-
-  u[7] = _mm_mullo_epi32(v[6], cospi16);
-  x = _mm_mullo_epi32(v[7], cospim48);
-  u[7] = _mm_sub_epi32(u[7], x);
-  u[7] = _mm_add_epi32(u[7], rnding);
-  u[7] = _mm_srai_epi32(u[7], bit);
-
-  // stage 5
-  v[0] = _mm_add_epi32(u[0], u[2]);
-  v[2] = _mm_sub_epi32(u[0], u[2]);
-  v[1] = _mm_add_epi32(u[1], u[3]);
-  v[3] = _mm_sub_epi32(u[1], u[3]);
-  v[4] = _mm_add_epi32(u[4], u[6]);
-  v[6] = _mm_sub_epi32(u[4], u[6]);
-  v[5] = _mm_add_epi32(u[5], u[7]);
-  v[7] = _mm_sub_epi32(u[5], u[7]);
-
-  // stage 6
-  u[0] = v[0];
-  u[1] = v[1];
-  u[4] = v[4];
-  u[5] = v[5];
-
-  v[0] = _mm_mullo_epi32(v[2], cospi32);
-  x = _mm_mullo_epi32(v[3], cospi32);
-  u[2] = _mm_add_epi32(v[0], x);
-  u[2] = _mm_add_epi32(u[2], rnding);
-  u[2] = _mm_srai_epi32(u[2], bit);
-
-  u[3] = _mm_sub_epi32(v[0], x);
-  u[3] = _mm_add_epi32(u[3], rnding);
-  u[3] = _mm_srai_epi32(u[3], bit);
-
-  v[0] = _mm_mullo_epi32(v[6], cospi32);
-  x = _mm_mullo_epi32(v[7], cospi32);
-  u[6] = _mm_add_epi32(v[0], x);
-  u[6] = _mm_add_epi32(u[6], rnding);
-  u[6] = _mm_srai_epi32(u[6], bit);
-
-  u[7] = _mm_sub_epi32(v[0], x);
-  u[7] = _mm_add_epi32(u[7], rnding);
-  u[7] = _mm_srai_epi32(u[7], bit);
-
-  // stage 7
-  out[1] = u[0];
-  out[3] = _mm_sub_epi32(kZero, u[4]);
-  out[5] = u[6];
-  out[7] = _mm_sub_epi32(kZero, u[2]);
-  out[9] = u[3];
-  out[11] = _mm_sub_epi32(kZero, u[7]);
-  out[13] = u[5];
-  out[15] = _mm_sub_epi32(kZero, u[1]);
+    // stage 7
+    out[2 * 0 + col] = v1;
+    out[2 * 1 + col] = v6;
+    out[2 * 2 + col] = v3;
+    out[2 * 3 + col] = v4;
+    out[2 * 4 + col] = v5;
+    out[2 * 5 + col] = v2;
+    out[2 * 6 + col] = v7;
+    out[2 * 7 + col] = v0;
+  }
 }
 
 void av1_fwd_txfm2d_8x8_sse4_1(const int16_t *input, int32_t *coeff, int stride,
@@ -1398,379 +1276,247 @@
 
 static void fadst16x16_sse4_1(__m128i *in, __m128i *out, int bit) {
   const int32_t *cospi = cospi_arr(bit);
-  const __m128i cospi2 = _mm_set1_epi32(cospi[2]);
-  const __m128i cospi62 = _mm_set1_epi32(cospi[62]);
-  const __m128i cospi10 = _mm_set1_epi32(cospi[10]);
-  const __m128i cospi54 = _mm_set1_epi32(cospi[54]);
-  const __m128i cospi18 = _mm_set1_epi32(cospi[18]);
-  const __m128i cospi46 = _mm_set1_epi32(cospi[46]);
-  const __m128i cospi26 = _mm_set1_epi32(cospi[26]);
-  const __m128i cospi38 = _mm_set1_epi32(cospi[38]);
-  const __m128i cospi34 = _mm_set1_epi32(cospi[34]);
-  const __m128i cospi30 = _mm_set1_epi32(cospi[30]);
-  const __m128i cospi42 = _mm_set1_epi32(cospi[42]);
-  const __m128i cospi22 = _mm_set1_epi32(cospi[22]);
-  const __m128i cospi50 = _mm_set1_epi32(cospi[50]);
-  const __m128i cospi14 = _mm_set1_epi32(cospi[14]);
-  const __m128i cospi58 = _mm_set1_epi32(cospi[58]);
-  const __m128i cospi6 = _mm_set1_epi32(cospi[6]);
-  const __m128i cospi8 = _mm_set1_epi32(cospi[8]);
-  const __m128i cospi56 = _mm_set1_epi32(cospi[56]);
-  const __m128i cospi40 = _mm_set1_epi32(cospi[40]);
-  const __m128i cospi24 = _mm_set1_epi32(cospi[24]);
-  const __m128i cospim56 = _mm_set1_epi32(-cospi[56]);
-  const __m128i cospim24 = _mm_set1_epi32(-cospi[24]);
+  const __m128i cospi32 = _mm_set1_epi32(cospi[32]);
   const __m128i cospi48 = _mm_set1_epi32(cospi[48]);
   const __m128i cospi16 = _mm_set1_epi32(cospi[16]);
+  const __m128i cospim16 = _mm_set1_epi32(-cospi[16]);
   const __m128i cospim48 = _mm_set1_epi32(-cospi[48]);
-  const __m128i cospi32 = _mm_set1_epi32(cospi[32]);
+  const __m128i cospi8 = _mm_set1_epi32(cospi[8]);
+  const __m128i cospi56 = _mm_set1_epi32(cospi[56]);
+  const __m128i cospim56 = _mm_set1_epi32(-cospi[56]);
+  const __m128i cospim8 = _mm_set1_epi32(-cospi[8]);
+  const __m128i cospi24 = _mm_set1_epi32(cospi[24]);
+  const __m128i cospim24 = _mm_set1_epi32(-cospi[24]);
+  const __m128i cospim40 = _mm_set1_epi32(-cospi[40]);
+  const __m128i cospi40 = _mm_set1_epi32(cospi[40]);
+  const __m128i cospi2 = _mm_set1_epi32(cospi[2]);
+  const __m128i cospi62 = _mm_set1_epi32(cospi[62]);
+  const __m128i cospim2 = _mm_set1_epi32(-cospi[2]);
+  const __m128i cospi10 = _mm_set1_epi32(cospi[10]);
+  const __m128i cospi54 = _mm_set1_epi32(cospi[54]);
+  const __m128i cospim10 = _mm_set1_epi32(-cospi[10]);
+  const __m128i cospi18 = _mm_set1_epi32(cospi[18]);
+  const __m128i cospi46 = _mm_set1_epi32(cospi[46]);
+  const __m128i cospim18 = _mm_set1_epi32(-cospi[18]);
+  const __m128i cospi26 = _mm_set1_epi32(cospi[26]);
+  const __m128i cospi38 = _mm_set1_epi32(cospi[38]);
+  const __m128i cospim26 = _mm_set1_epi32(-cospi[26]);
+  const __m128i cospi34 = _mm_set1_epi32(cospi[34]);
+  const __m128i cospi30 = _mm_set1_epi32(cospi[30]);
+  const __m128i cospim34 = _mm_set1_epi32(-cospi[34]);
+  const __m128i cospi42 = _mm_set1_epi32(cospi[42]);
+  const __m128i cospi22 = _mm_set1_epi32(cospi[22]);
+  const __m128i cospim42 = _mm_set1_epi32(-cospi[42]);
+  const __m128i cospi50 = _mm_set1_epi32(cospi[50]);
+  const __m128i cospi14 = _mm_set1_epi32(cospi[14]);
+  const __m128i cospim50 = _mm_set1_epi32(-cospi[50]);
+  const __m128i cospi58 = _mm_set1_epi32(cospi[58]);
+  const __m128i cospi6 = _mm_set1_epi32(cospi[6]);
+  const __m128i cospim58 = _mm_set1_epi32(-cospi[58]);
   const __m128i rnding = _mm_set1_epi32(1 << (bit - 1));
+  const __m128i zero = _mm_setzero_si128();
+
   __m128i u[16], v[16], x, y;
-  const int col_num = 4;
   int col;
 
-  // Calculate the column 0, 1, 2, 3
-  for (col = 0; col < col_num; ++col) {
+  for (col = 0; col < 4; ++col) {
     // stage 0
     // stage 1
+    u[0] = in[0 * 4 + col];
+    u[1] = _mm_sub_epi32(zero, in[15 * 4 + col]);
+    u[2] = _mm_sub_epi32(zero, in[7 * 4 + col]);
+    u[3] = in[8 * 4 + col];
+    u[4] = _mm_sub_epi32(zero, in[3 * 4 + col]);
+    u[5] = in[12 * 4 + col];
+    u[6] = in[4 * 4 + col];
+    u[7] = _mm_sub_epi32(zero, in[11 * 4 + col]);
+    u[8] = _mm_sub_epi32(zero, in[1 * 4 + col]);
+    u[9] = in[14 * 4 + col];
+    u[10] = in[6 * 4 + col];
+    u[11] = _mm_sub_epi32(zero, in[9 * 4 + col]);
+    u[12] = in[2 * 4 + col];
+    u[13] = _mm_sub_epi32(zero, in[13 * 4 + col]);
+    u[14] = _mm_sub_epi32(zero, in[5 * 4 + col]);
+    u[15] = in[10 * 4 + col];
+
     // stage 2
-    v[0] = _mm_mullo_epi32(in[15 * col_num + col], cospi2);
-    x = _mm_mullo_epi32(in[0 * col_num + col], cospi62);
-    v[0] = _mm_add_epi32(v[0], x);
-    v[0] = _mm_add_epi32(v[0], rnding);
-    v[0] = _mm_srai_epi32(v[0], bit);
+    v[0] = u[0];
+    v[1] = u[1];
 
-    v[1] = _mm_mullo_epi32(in[15 * col_num + col], cospi62);
-    x = _mm_mullo_epi32(in[0 * col_num + col], cospi2);
-    v[1] = _mm_sub_epi32(v[1], x);
-    v[1] = _mm_add_epi32(v[1], rnding);
-    v[1] = _mm_srai_epi32(v[1], bit);
-
-    v[2] = _mm_mullo_epi32(in[13 * col_num + col], cospi10);
-    x = _mm_mullo_epi32(in[2 * col_num + col], cospi54);
-    v[2] = _mm_add_epi32(v[2], x);
+    x = _mm_mullo_epi32(u[2], cospi32);
+    y = _mm_mullo_epi32(u[3], cospi32);
+    v[2] = _mm_add_epi32(x, y);
     v[2] = _mm_add_epi32(v[2], rnding);
     v[2] = _mm_srai_epi32(v[2], bit);
 
-    v[3] = _mm_mullo_epi32(in[13 * col_num + col], cospi54);
-    x = _mm_mullo_epi32(in[2 * col_num + col], cospi10);
-    v[3] = _mm_sub_epi32(v[3], x);
+    v[3] = _mm_sub_epi32(x, y);
     v[3] = _mm_add_epi32(v[3], rnding);
     v[3] = _mm_srai_epi32(v[3], bit);
 
-    v[4] = _mm_mullo_epi32(in[11 * col_num + col], cospi18);
-    x = _mm_mullo_epi32(in[4 * col_num + col], cospi46);
-    v[4] = _mm_add_epi32(v[4], x);
-    v[4] = _mm_add_epi32(v[4], rnding);
-    v[4] = _mm_srai_epi32(v[4], bit);
+    v[4] = u[4];
+    v[5] = u[5];
 
-    v[5] = _mm_mullo_epi32(in[11 * col_num + col], cospi46);
-    x = _mm_mullo_epi32(in[4 * col_num + col], cospi18);
-    v[5] = _mm_sub_epi32(v[5], x);
-    v[5] = _mm_add_epi32(v[5], rnding);
-    v[5] = _mm_srai_epi32(v[5], bit);
-
-    v[6] = _mm_mullo_epi32(in[9 * col_num + col], cospi26);
-    x = _mm_mullo_epi32(in[6 * col_num + col], cospi38);
-    v[6] = _mm_add_epi32(v[6], x);
+    x = _mm_mullo_epi32(u[6], cospi32);
+    y = _mm_mullo_epi32(u[7], cospi32);
+    v[6] = _mm_add_epi32(x, y);
     v[6] = _mm_add_epi32(v[6], rnding);
     v[6] = _mm_srai_epi32(v[6], bit);
 
-    v[7] = _mm_mullo_epi32(in[9 * col_num + col], cospi38);
-    x = _mm_mullo_epi32(in[6 * col_num + col], cospi26);
-    v[7] = _mm_sub_epi32(v[7], x);
+    v[7] = _mm_sub_epi32(x, y);
     v[7] = _mm_add_epi32(v[7], rnding);
     v[7] = _mm_srai_epi32(v[7], bit);
 
-    v[8] = _mm_mullo_epi32(in[7 * col_num + col], cospi34);
-    x = _mm_mullo_epi32(in[8 * col_num + col], cospi30);
-    v[8] = _mm_add_epi32(v[8], x);
-    v[8] = _mm_add_epi32(v[8], rnding);
-    v[8] = _mm_srai_epi32(v[8], bit);
+    v[8] = u[8];
+    v[9] = u[9];
 
-    v[9] = _mm_mullo_epi32(in[7 * col_num + col], cospi30);
-    x = _mm_mullo_epi32(in[8 * col_num + col], cospi34);
-    v[9] = _mm_sub_epi32(v[9], x);
-    v[9] = _mm_add_epi32(v[9], rnding);
-    v[9] = _mm_srai_epi32(v[9], bit);
-
-    v[10] = _mm_mullo_epi32(in[5 * col_num + col], cospi42);
-    x = _mm_mullo_epi32(in[10 * col_num + col], cospi22);
-    v[10] = _mm_add_epi32(v[10], x);
+    x = _mm_mullo_epi32(u[10], cospi32);
+    y = _mm_mullo_epi32(u[11], cospi32);
+    v[10] = _mm_add_epi32(x, y);
     v[10] = _mm_add_epi32(v[10], rnding);
     v[10] = _mm_srai_epi32(v[10], bit);
 
-    v[11] = _mm_mullo_epi32(in[5 * col_num + col], cospi22);
-    x = _mm_mullo_epi32(in[10 * col_num + col], cospi42);
-    v[11] = _mm_sub_epi32(v[11], x);
+    v[11] = _mm_sub_epi32(x, y);
     v[11] = _mm_add_epi32(v[11], rnding);
     v[11] = _mm_srai_epi32(v[11], bit);
 
-    v[12] = _mm_mullo_epi32(in[3 * col_num + col], cospi50);
-    x = _mm_mullo_epi32(in[12 * col_num + col], cospi14);
-    v[12] = _mm_add_epi32(v[12], x);
-    v[12] = _mm_add_epi32(v[12], rnding);
-    v[12] = _mm_srai_epi32(v[12], bit);
+    v[12] = u[12];
+    v[13] = u[13];
 
-    v[13] = _mm_mullo_epi32(in[3 * col_num + col], cospi14);
-    x = _mm_mullo_epi32(in[12 * col_num + col], cospi50);
-    v[13] = _mm_sub_epi32(v[13], x);
-    v[13] = _mm_add_epi32(v[13], rnding);
-    v[13] = _mm_srai_epi32(v[13], bit);
-
-    v[14] = _mm_mullo_epi32(in[1 * col_num + col], cospi58);
-    x = _mm_mullo_epi32(in[14 * col_num + col], cospi6);
-    v[14] = _mm_add_epi32(v[14], x);
+    x = _mm_mullo_epi32(u[14], cospi32);
+    y = _mm_mullo_epi32(u[15], cospi32);
+    v[14] = _mm_add_epi32(x, y);
     v[14] = _mm_add_epi32(v[14], rnding);
     v[14] = _mm_srai_epi32(v[14], bit);
 
-    v[15] = _mm_mullo_epi32(in[1 * col_num + col], cospi6);
-    x = _mm_mullo_epi32(in[14 * col_num + col], cospi58);
-    v[15] = _mm_sub_epi32(v[15], x);
+    v[15] = _mm_sub_epi32(x, y);
     v[15] = _mm_add_epi32(v[15], rnding);
     v[15] = _mm_srai_epi32(v[15], bit);
 
     // stage 3
-    u[0] = _mm_add_epi32(v[0], v[8]);
-    u[8] = _mm_sub_epi32(v[0], v[8]);
-    u[1] = _mm_add_epi32(v[1], v[9]);
-    u[9] = _mm_sub_epi32(v[1], v[9]);
-    u[2] = _mm_add_epi32(v[2], v[10]);
-    u[10] = _mm_sub_epi32(v[2], v[10]);
-    u[3] = _mm_add_epi32(v[3], v[11]);
-    u[11] = _mm_sub_epi32(v[3], v[11]);
-    u[4] = _mm_add_epi32(v[4], v[12]);
-    u[12] = _mm_sub_epi32(v[4], v[12]);
-    u[5] = _mm_add_epi32(v[5], v[13]);
-    u[13] = _mm_sub_epi32(v[5], v[13]);
-    u[6] = _mm_add_epi32(v[6], v[14]);
-    u[14] = _mm_sub_epi32(v[6], v[14]);
-    u[7] = _mm_add_epi32(v[7], v[15]);
-    u[15] = _mm_sub_epi32(v[7], v[15]);
+    u[0] = _mm_add_epi32(v[0], v[2]);
+    u[1] = _mm_add_epi32(v[1], v[3]);
+    u[2] = _mm_sub_epi32(v[0], v[2]);
+    u[3] = _mm_sub_epi32(v[1], v[3]);
+    u[4] = _mm_add_epi32(v[4], v[6]);
+    u[5] = _mm_add_epi32(v[5], v[7]);
+    u[6] = _mm_sub_epi32(v[4], v[6]);
+    u[7] = _mm_sub_epi32(v[5], v[7]);
+    u[8] = _mm_add_epi32(v[8], v[10]);
+    u[9] = _mm_add_epi32(v[9], v[11]);
+    u[10] = _mm_sub_epi32(v[8], v[10]);
+    u[11] = _mm_sub_epi32(v[9], v[11]);
+    u[12] = _mm_add_epi32(v[12], v[14]);
+    u[13] = _mm_add_epi32(v[13], v[15]);
+    u[14] = _mm_sub_epi32(v[12], v[14]);
+    u[15] = _mm_sub_epi32(v[13], v[15]);
 
     // stage 4
     v[0] = u[0];
     v[1] = u[1];
     v[2] = u[2];
     v[3] = u[3];
+    v[4] = half_btf_sse4_1(&cospi16, &u[4], &cospi48, &u[5], &rnding, bit);
+    v[5] = half_btf_sse4_1(&cospi48, &u[4], &cospim16, &u[5], &rnding, bit);
+    v[6] = half_btf_sse4_1(&cospim48, &u[6], &cospi16, &u[7], &rnding, bit);
+    v[7] = half_btf_sse4_1(&cospi16, &u[6], &cospi48, &u[7], &rnding, bit);
+    v[8] = u[8];
+    v[9] = u[9];
+    v[10] = u[10];
+    v[11] = u[11];
+    v[12] = half_btf_sse4_1(&cospi16, &u[12], &cospi48, &u[13], &rnding, bit);
+    v[13] = half_btf_sse4_1(&cospi48, &u[12], &cospim16, &u[13], &rnding, bit);
+    v[14] = half_btf_sse4_1(&cospim48, &u[14], &cospi16, &u[15], &rnding, bit);
+    v[15] = half_btf_sse4_1(&cospi16, &u[14], &cospi48, &u[15], &rnding, bit);
+
+    // stage 5
+    u[0] = _mm_add_epi32(v[0], v[4]);
+    u[1] = _mm_add_epi32(v[1], v[5]);
+    u[2] = _mm_add_epi32(v[2], v[6]);
+    u[3] = _mm_add_epi32(v[3], v[7]);
+    u[4] = _mm_sub_epi32(v[0], v[4]);
+    u[5] = _mm_sub_epi32(v[1], v[5]);
+    u[6] = _mm_sub_epi32(v[2], v[6]);
+    u[7] = _mm_sub_epi32(v[3], v[7]);
+    u[8] = _mm_add_epi32(v[8], v[12]);
+    u[9] = _mm_add_epi32(v[9], v[13]);
+    u[10] = _mm_add_epi32(v[10], v[14]);
+    u[11] = _mm_add_epi32(v[11], v[15]);
+    u[12] = _mm_sub_epi32(v[8], v[12]);
+    u[13] = _mm_sub_epi32(v[9], v[13]);
+    u[14] = _mm_sub_epi32(v[10], v[14]);
+    u[15] = _mm_sub_epi32(v[11], v[15]);
+
+    // stage 6
+    v[0] = u[0];
+    v[1] = u[1];
+    v[2] = u[2];
+    v[3] = u[3];
     v[4] = u[4];
     v[5] = u[5];
     v[6] = u[6];
     v[7] = u[7];
-
-    v[8] = _mm_mullo_epi32(u[8], cospi8);
-    x = _mm_mullo_epi32(u[9], cospi56);
-    v[8] = _mm_add_epi32(v[8], x);
-    v[8] = _mm_add_epi32(v[8], rnding);
-    v[8] = _mm_srai_epi32(v[8], bit);
-
-    v[9] = _mm_mullo_epi32(u[8], cospi56);
-    x = _mm_mullo_epi32(u[9], cospi8);
-    v[9] = _mm_sub_epi32(v[9], x);
-    v[9] = _mm_add_epi32(v[9], rnding);
-    v[9] = _mm_srai_epi32(v[9], bit);
-
-    v[10] = _mm_mullo_epi32(u[10], cospi40);
-    x = _mm_mullo_epi32(u[11], cospi24);
-    v[10] = _mm_add_epi32(v[10], x);
-    v[10] = _mm_add_epi32(v[10], rnding);
-    v[10] = _mm_srai_epi32(v[10], bit);
-
-    v[11] = _mm_mullo_epi32(u[10], cospi24);
-    x = _mm_mullo_epi32(u[11], cospi40);
-    v[11] = _mm_sub_epi32(v[11], x);
-    v[11] = _mm_add_epi32(v[11], rnding);
-    v[11] = _mm_srai_epi32(v[11], bit);
-
-    v[12] = _mm_mullo_epi32(u[12], cospim56);
-    x = _mm_mullo_epi32(u[13], cospi8);
-    v[12] = _mm_add_epi32(v[12], x);
-    v[12] = _mm_add_epi32(v[12], rnding);
-    v[12] = _mm_srai_epi32(v[12], bit);
-
-    v[13] = _mm_mullo_epi32(u[12], cospi8);
-    x = _mm_mullo_epi32(u[13], cospim56);
-    v[13] = _mm_sub_epi32(v[13], x);
-    v[13] = _mm_add_epi32(v[13], rnding);
-    v[13] = _mm_srai_epi32(v[13], bit);
-
-    v[14] = _mm_mullo_epi32(u[14], cospim24);
-    x = _mm_mullo_epi32(u[15], cospi40);
-    v[14] = _mm_add_epi32(v[14], x);
-    v[14] = _mm_add_epi32(v[14], rnding);
-    v[14] = _mm_srai_epi32(v[14], bit);
-
-    v[15] = _mm_mullo_epi32(u[14], cospi40);
-    x = _mm_mullo_epi32(u[15], cospim24);
-    v[15] = _mm_sub_epi32(v[15], x);
-    v[15] = _mm_add_epi32(v[15], rnding);
-    v[15] = _mm_srai_epi32(v[15], bit);
-
-    // stage 5
-    u[0] = _mm_add_epi32(v[0], v[4]);
-    u[4] = _mm_sub_epi32(v[0], v[4]);
-    u[1] = _mm_add_epi32(v[1], v[5]);
-    u[5] = _mm_sub_epi32(v[1], v[5]);
-    u[2] = _mm_add_epi32(v[2], v[6]);
-    u[6] = _mm_sub_epi32(v[2], v[6]);
-    u[3] = _mm_add_epi32(v[3], v[7]);
-    u[7] = _mm_sub_epi32(v[3], v[7]);
-    u[8] = _mm_add_epi32(v[8], v[12]);
-    u[12] = _mm_sub_epi32(v[8], v[12]);
-    u[9] = _mm_add_epi32(v[9], v[13]);
-    u[13] = _mm_sub_epi32(v[9], v[13]);
-    u[10] = _mm_add_epi32(v[10], v[14]);
-    u[14] = _mm_sub_epi32(v[10], v[14]);
-    u[11] = _mm_add_epi32(v[11], v[15]);
-    u[15] = _mm_sub_epi32(v[11], v[15]);
-
-    // stage 6
-    v[0] = u[0];
-    v[1] = u[1];
-    v[2] = u[2];
-    v[3] = u[3];
-
-    v[4] = _mm_mullo_epi32(u[4], cospi16);
-    x = _mm_mullo_epi32(u[5], cospi48);
-    v[4] = _mm_add_epi32(v[4], x);
-    v[4] = _mm_add_epi32(v[4], rnding);
-    v[4] = _mm_srai_epi32(v[4], bit);
-
-    v[5] = _mm_mullo_epi32(u[4], cospi48);
-    x = _mm_mullo_epi32(u[5], cospi16);
-    v[5] = _mm_sub_epi32(v[5], x);
-    v[5] = _mm_add_epi32(v[5], rnding);
-    v[5] = _mm_srai_epi32(v[5], bit);
-
-    v[6] = _mm_mullo_epi32(u[6], cospim48);
-    x = _mm_mullo_epi32(u[7], cospi16);
-    v[6] = _mm_add_epi32(v[6], x);
-    v[6] = _mm_add_epi32(v[6], rnding);
-    v[6] = _mm_srai_epi32(v[6], bit);
-
-    v[7] = _mm_mullo_epi32(u[6], cospi16);
-    x = _mm_mullo_epi32(u[7], cospim48);
-    v[7] = _mm_sub_epi32(v[7], x);
-    v[7] = _mm_add_epi32(v[7], rnding);
-    v[7] = _mm_srai_epi32(v[7], bit);
-
-    v[8] = u[8];
-    v[9] = u[9];
-    v[10] = u[10];
-    v[11] = u[11];
-
-    v[12] = _mm_mullo_epi32(u[12], cospi16);
-    x = _mm_mullo_epi32(u[13], cospi48);
-    v[12] = _mm_add_epi32(v[12], x);
-    v[12] = _mm_add_epi32(v[12], rnding);
-    v[12] = _mm_srai_epi32(v[12], bit);
-
-    v[13] = _mm_mullo_epi32(u[12], cospi48);
-    x = _mm_mullo_epi32(u[13], cospi16);
-    v[13] = _mm_sub_epi32(v[13], x);
-    v[13] = _mm_add_epi32(v[13], rnding);
-    v[13] = _mm_srai_epi32(v[13], bit);
-
-    v[14] = _mm_mullo_epi32(u[14], cospim48);
-    x = _mm_mullo_epi32(u[15], cospi16);
-    v[14] = _mm_add_epi32(v[14], x);
-    v[14] = _mm_add_epi32(v[14], rnding);
-    v[14] = _mm_srai_epi32(v[14], bit);
-
-    v[15] = _mm_mullo_epi32(u[14], cospi16);
-    x = _mm_mullo_epi32(u[15], cospim48);
-    v[15] = _mm_sub_epi32(v[15], x);
-    v[15] = _mm_add_epi32(v[15], rnding);
-    v[15] = _mm_srai_epi32(v[15], bit);
+    v[8] = half_btf_sse4_1(&cospi8, &u[8], &cospi56, &u[9], &rnding, bit);
+    v[9] = half_btf_sse4_1(&cospi56, &u[8], &cospim8, &u[9], &rnding, bit);
+    v[10] = half_btf_sse4_1(&cospi40, &u[10], &cospi24, &u[11], &rnding, bit);
+    v[11] = half_btf_sse4_1(&cospi24, &u[10], &cospim40, &u[11], &rnding, bit);
+    v[12] = half_btf_sse4_1(&cospim56, &u[12], &cospi8, &u[13], &rnding, bit);
+    v[13] = half_btf_sse4_1(&cospi8, &u[12], &cospi56, &u[13], &rnding, bit);
+    v[14] = half_btf_sse4_1(&cospim24, &u[14], &cospi40, &u[15], &rnding, bit);
+    v[15] = half_btf_sse4_1(&cospi40, &u[14], &cospi24, &u[15], &rnding, bit);
 
     // stage 7
-    u[0] = _mm_add_epi32(v[0], v[2]);
-    u[2] = _mm_sub_epi32(v[0], v[2]);
-    u[1] = _mm_add_epi32(v[1], v[3]);
-    u[3] = _mm_sub_epi32(v[1], v[3]);
-    u[4] = _mm_add_epi32(v[4], v[6]);
-    u[6] = _mm_sub_epi32(v[4], v[6]);
-    u[5] = _mm_add_epi32(v[5], v[7]);
-    u[7] = _mm_sub_epi32(v[5], v[7]);
-    u[8] = _mm_add_epi32(v[8], v[10]);
-    u[10] = _mm_sub_epi32(v[8], v[10]);
-    u[9] = _mm_add_epi32(v[9], v[11]);
-    u[11] = _mm_sub_epi32(v[9], v[11]);
-    u[12] = _mm_add_epi32(v[12], v[14]);
-    u[14] = _mm_sub_epi32(v[12], v[14]);
-    u[13] = _mm_add_epi32(v[13], v[15]);
-    u[15] = _mm_sub_epi32(v[13], v[15]);
+    u[0] = _mm_add_epi32(v[0], v[8]);
+    u[1] = _mm_add_epi32(v[1], v[9]);
+    u[2] = _mm_add_epi32(v[2], v[10]);
+    u[3] = _mm_add_epi32(v[3], v[11]);
+    u[4] = _mm_add_epi32(v[4], v[12]);
+    u[5] = _mm_add_epi32(v[5], v[13]);
+    u[6] = _mm_add_epi32(v[6], v[14]);
+    u[7] = _mm_add_epi32(v[7], v[15]);
+    u[8] = _mm_sub_epi32(v[0], v[8]);
+    u[9] = _mm_sub_epi32(v[1], v[9]);
+    u[10] = _mm_sub_epi32(v[2], v[10]);
+    u[11] = _mm_sub_epi32(v[3], v[11]);
+    u[12] = _mm_sub_epi32(v[4], v[12]);
+    u[13] = _mm_sub_epi32(v[5], v[13]);
+    u[14] = _mm_sub_epi32(v[6], v[14]);
+    u[15] = _mm_sub_epi32(v[7], v[15]);
 
     // stage 8
-    v[0] = u[0];
-    v[1] = u[1];
-
-    y = _mm_mullo_epi32(u[2], cospi32);
-    x = _mm_mullo_epi32(u[3], cospi32);
-    v[2] = _mm_add_epi32(y, x);
-    v[2] = _mm_add_epi32(v[2], rnding);
-    v[2] = _mm_srai_epi32(v[2], bit);
-
-    v[3] = _mm_sub_epi32(y, x);
-    v[3] = _mm_add_epi32(v[3], rnding);
-    v[3] = _mm_srai_epi32(v[3], bit);
-
-    v[4] = u[4];
-    v[5] = u[5];
-
-    y = _mm_mullo_epi32(u[6], cospi32);
-    x = _mm_mullo_epi32(u[7], cospi32);
-    v[6] = _mm_add_epi32(y, x);
-    v[6] = _mm_add_epi32(v[6], rnding);
-    v[6] = _mm_srai_epi32(v[6], bit);
-
-    v[7] = _mm_sub_epi32(y, x);
-    v[7] = _mm_add_epi32(v[7], rnding);
-    v[7] = _mm_srai_epi32(v[7], bit);
-
-    v[8] = u[8];
-    v[9] = u[9];
-
-    y = _mm_mullo_epi32(u[10], cospi32);
-    x = _mm_mullo_epi32(u[11], cospi32);
-    v[10] = _mm_add_epi32(y, x);
-    v[10] = _mm_add_epi32(v[10], rnding);
-    v[10] = _mm_srai_epi32(v[10], bit);
-
-    v[11] = _mm_sub_epi32(y, x);
-    v[11] = _mm_add_epi32(v[11], rnding);
-    v[11] = _mm_srai_epi32(v[11], bit);
-
-    v[12] = u[12];
-    v[13] = u[13];
-
-    y = _mm_mullo_epi32(u[14], cospi32);
-    x = _mm_mullo_epi32(u[15], cospi32);
-    v[14] = _mm_add_epi32(y, x);
-    v[14] = _mm_add_epi32(v[14], rnding);
-    v[14] = _mm_srai_epi32(v[14], bit);
-
-    v[15] = _mm_sub_epi32(y, x);
-    v[15] = _mm_add_epi32(v[15], rnding);
-    v[15] = _mm_srai_epi32(v[15], bit);
+    v[0] = half_btf_sse4_1(&cospi2, &u[0], &cospi62, &u[1], &rnding, bit);
+    v[1] = half_btf_sse4_1(&cospi62, &u[0], &cospim2, &u[1], &rnding, bit);
+    v[2] = half_btf_sse4_1(&cospi10, &u[2], &cospi54, &u[3], &rnding, bit);
+    v[3] = half_btf_sse4_1(&cospi54, &u[2], &cospim10, &u[3], &rnding, bit);
+    v[4] = half_btf_sse4_1(&cospi18, &u[4], &cospi46, &u[5], &rnding, bit);
+    v[5] = half_btf_sse4_1(&cospi46, &u[4], &cospim18, &u[5], &rnding, bit);
+    v[6] = half_btf_sse4_1(&cospi26, &u[6], &cospi38, &u[7], &rnding, bit);
+    v[7] = half_btf_sse4_1(&cospi38, &u[6], &cospim26, &u[7], &rnding, bit);
+    v[8] = half_btf_sse4_1(&cospi34, &u[8], &cospi30, &u[9], &rnding, bit);
+    v[9] = half_btf_sse4_1(&cospi30, &u[8], &cospim34, &u[9], &rnding, bit);
+    v[10] = half_btf_sse4_1(&cospi42, &u[10], &cospi22, &u[11], &rnding, bit);
+    v[11] = half_btf_sse4_1(&cospi22, &u[10], &cospim42, &u[11], &rnding, bit);
+    v[12] = half_btf_sse4_1(&cospi50, &u[12], &cospi14, &u[13], &rnding, bit);
+    v[13] = half_btf_sse4_1(&cospi14, &u[12], &cospim50, &u[13], &rnding, bit);
+    v[14] = half_btf_sse4_1(&cospi58, &u[14], &cospi6, &u[15], &rnding, bit);
+    v[15] = half_btf_sse4_1(&cospi6, &u[14], &cospim58, &u[15], &rnding, bit);
 
     // stage 9
-    out[0 * col_num + col] = v[0];
-    out[1 * col_num + col] = _mm_sub_epi32(_mm_set1_epi32(0), v[8]);
-    out[2 * col_num + col] = v[12];
-    out[3 * col_num + col] = _mm_sub_epi32(_mm_set1_epi32(0), v[4]);
-    out[4 * col_num + col] = v[6];
-    out[5 * col_num + col] = _mm_sub_epi32(_mm_set1_epi32(0), v[14]);
-    out[6 * col_num + col] = v[10];
-    out[7 * col_num + col] = _mm_sub_epi32(_mm_set1_epi32(0), v[2]);
-    out[8 * col_num + col] = v[3];
-    out[9 * col_num + col] = _mm_sub_epi32(_mm_set1_epi32(0), v[11]);
-    out[10 * col_num + col] = v[15];
-    out[11 * col_num + col] = _mm_sub_epi32(_mm_set1_epi32(0), v[7]);
-    out[12 * col_num + col] = v[5];
-    out[13 * col_num + col] = _mm_sub_epi32(_mm_set1_epi32(0), v[13]);
-    out[14 * col_num + col] = v[9];
-    out[15 * col_num + col] = _mm_sub_epi32(_mm_set1_epi32(0), v[1]);
+    out[0 * 4 + col] = v[1];
+    out[1 * 4 + col] = v[14];
+    out[2 * 4 + col] = v[3];
+    out[3 * 4 + col] = v[12];
+    out[4 * 4 + col] = v[5];
+    out[5 * 4 + col] = v[10];
+    out[6 * 4 + col] = v[7];
+    out[7 * 4 + col] = v[8];
+    out[8 * 4 + col] = v[9];
+    out[9 * 4 + col] = v[6];
+    out[10 * 4 + col] = v[11];
+    out[11 * 4 + col] = v[4];
+    out[12 * 4 + col] = v[13];
+    out[13 * 4 + col] = v[2];
+    out[14 * 4 + col] = v[15];
+    out[15 * 4 + col] = v[0];
   }
 }
 

diff --git a/test/av1_fwd_txfm2d_test.cc b/test/av1_fwd_txfm2d_test.cc
index 3f568e9..b25e981 100644
--- a/test/av1_fwd_txfm2d_test.cc
+++ b/test/av1_fwd_txfm2d_test.cc

@@ -147,7 +147,7 @@
     param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_4X8, 2.5, 0.4));
     param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_8X4, 2.5, 0.4));
     param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_8X16, 6.5, 1));
-    param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_16X8, 6, 1));
+    param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_16X8, 6.5, 1));
     param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_16X32, 46, 7));
     param_list.push_back(AV1FwdTxfm2dParam(tx_type, TX_32X16, 30, 7));
 

diff --git a/test/av1_inv_txfm2d_test.cc b/test/av1_inv_txfm2d_test.cc
index 953ec68..ca52cbd 100644
--- a/test/av1_inv_txfm2d_test.cc
+++ b/test/av1_inv_txfm2d_test.cc

@@ -146,12 +146,12 @@
   for (int t = 0; t <= FLIPADST_ADST; ++t) {
     const TX_TYPE tx_type = static_cast<TX_TYPE>(t);
     param_list.push_back(AV1InvTxfm2dParam(tx_type, TX_4X4, 2, 0.002));
-    param_list.push_back(AV1InvTxfm2dParam(tx_type, TX_8X8, 2, 0.02));
+    param_list.push_back(AV1InvTxfm2dParam(tx_type, TX_8X8, 2, 0.025));
     param_list.push_back(AV1InvTxfm2dParam(tx_type, TX_16X16, 2, 0.04));
     param_list.push_back(AV1InvTxfm2dParam(tx_type, TX_32X32, 4, 0.4));
 
-    param_list.push_back(AV1InvTxfm2dParam(tx_type, TX_4X8, 2, 0.012));
-    param_list.push_back(AV1InvTxfm2dParam(tx_type, TX_8X4, 2, 0.012));
+    param_list.push_back(AV1InvTxfm2dParam(tx_type, TX_4X8, 2, 0.016));
+    param_list.push_back(AV1InvTxfm2dParam(tx_type, TX_8X4, 2, 0.016));
     param_list.push_back(AV1InvTxfm2dParam(tx_type, TX_8X16, 2, 0.033));
     param_list.push_back(AV1InvTxfm2dParam(tx_type, TX_16X8, 2, 0.033));
     param_list.push_back(AV1InvTxfm2dParam(tx_type, TX_16X32, 2, 0.4));

diff --git a/test/transform_test_base.h b/test/transform_test_base.h
index 21441be..3db0c69 100644
--- a/test/transform_test_base.h
+++ b/test/transform_test_base.h

@@ -255,7 +255,7 @@
       int row_length = FindRowLength();
       // The minimum quant value is 4.
       for (int j = 0; j < num_coeffs_; ++j) {
-        EXPECT_EQ(output_block[j], output_ref_block[j])
+        ASSERT_EQ(output_block[j], output_ref_block[j])
             << "Not bit-exact at test index: " << i << ", "
             << "j = " << j << std::endl;
         EXPECT_GE(row_length * kDctMaxValue << (bit_depth_ - 8),
@@ -323,7 +323,7 @@
         const int diff = dst[j] - src[j];
 #endif
         const uint32_t error = diff * diff;
-        EXPECT_GE(static_cast<uint32_t>(limit), error)
+        ASSERT_GE(static_cast<uint32_t>(limit), error)
             << "Error: 4x4 IDCT has error " << error << " at index " << j;
       }
     }
commit	4d5cf537ff2f5935cc3b1fa852fb0fbd90adf4c5	[log] [tgz]
author	Urvang Joshi <urvang@google.com>	Wed Dec 20 16:54:10 2017 -0800
committer	Urvang Joshi <urvang@google.com>	Thu Dec 21 21:26:35 2017 +0000
tree	ae8194b3d502c1630964d78fd961f52b52c686b9
parent	eb2fd5c58a73b473bcd7f9a7cbc7a48a4f8aa141 [diff]