fix signed integer overflow of idct

Exposed by fuzz test in high bitdepth.
The bug is introduced in commit 64653fa.

BUG=webm:1466

Change-Id: Idd77d5c6a60efb9241471611ce1aba0646cb6ff5

diff --git a/test/invalid_file_test.cc b/test/invalid_file_test.cc
index 8f7cd10..79220b0 100644 (file)
--- a/test/invalid_file_test.cc
+++ b/test/invalid_file_test.cc
@@ -134,6 +134,9 @@ const DecodeParam kVP9InvalidFileTests[] = {
   { 1, "invalid-vp90-02-v2.webm" },
 #if CONFIG_VP9_HIGHBITDEPTH
   { 1, "invalid-vp90-2-00-quantizer-00.webm.ivf.s5861_r01-05_b6-.v2.ivf" },
+  { 1,
+    "invalid-vp90-2-21-resize_inter_320x180_5_3-4.webm.ivf.s45551_r01-05_b6-."
+    "ivf" },
 #endif
   { 1, "invalid-vp90-03-v3.webm" },
   { 1, "invalid-vp90-2-00-quantizer-11.webm.ivf.s52984_r01-05_b6-.ivf" },

diff --git a/test/test-data.mk b/test/test-data.mk
index 1656372..67a2568 100644 (file)
--- a/test/test-data.mk
+++ b/test/test-data.mk
@@ -774,6 +774,8 @@ LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-12-droppable_1.ivf.s367
 LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-12-droppable_1.ivf.s3676_r01-05_b6-.ivf.res
 LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-12-droppable_1.ivf.s73804_r01-05_b6-.ivf
 LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-12-droppable_1.ivf.s73804_r01-05_b6-.ivf.res
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-21-resize_inter_320x180_5_3-4.webm.ivf.s45551_r01-05_b6-.ivf
+LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-21-resize_inter_320x180_5_3-4.webm.ivf.s45551_r01-05_b6-.ivf.res
 LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp91-2-mixedrefcsp-444to420.ivf
 LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp91-2-mixedrefcsp-444to420.ivf.res
 LIBVPX_TEST_DATA-$(CONFIG_VP9_DECODER) += invalid-vp90-2-07-frame_parallel-1.webm

diff --git a/test/test-data.sha1 b/test/test-data.sha1
index 7948f9e..57d6502 100644 (file)
--- a/test/test-data.sha1
+++ b/test/test-data.sha1
@@ -6,6 +6,8 @@ b87815bf86020c592ccc7a846ba2e28ec8043902 *hantro_odd.yuv
 456d1493e52d32a5c30edf44a27debc1fa6b253a *invalid-vp90-2-00-quantizer-11.webm.ivf.s52984_r01-05_b6-.ivf.res
 c123d1f9f02fb4143abb5e271916e3a3080de8f6 *invalid-vp90-2-00-quantizer-11.webm.ivf.s52984_r01-05_b6-z.ivf
 456d1493e52d32a5c30edf44a27debc1fa6b253a *invalid-vp90-2-00-quantizer-11.webm.ivf.s52984_r01-05_b6-z.ivf.res
+efafb92b7567bc04c3f1432ea6c268c1c31affd5 *invalid-vp90-2-21-resize_inter_320x180_5_3-4.webm.ivf.s45551_r01-05_b6-.ivf
+5d9474c0309b7ca09a182d888f73b37a8fe1362c *invalid-vp90-2-21-resize_inter_320x180_5_3-4.webm.ivf.s45551_r01-05_b6-.ivf.res
 fe346136b9b8c1e6f6084cc106485706915795e4 *invalid-vp90-01-v3.webm
 5d9474c0309b7ca09a182d888f73b37a8fe1362c *invalid-vp90-01-v3.webm.res
 d78e2fceba5ac942246503ec8366f879c4775ca5 *invalid-vp90-02-v2.webm

diff --git a/vpx_dsp/arm/idct16x16_1_add_neon.c b/vpx_dsp/arm/idct16x16_1_add_neon.c
index 968bc5c..bf5192a 100644 (file)
--- a/vpx_dsp/arm/idct16x16_1_add_neon.c
+++ b/vpx_dsp/arm/idct16x16_1_add_neon.c
@@ -32,7 +32,8 @@ static INLINE void idct16x16_1_add_neg_kernel(uint8_t **dest, const int stride,
 
 void vpx_idct16x16_1_add_neon(const tran_low_t *input, uint8_t *dest,
                               int stride) {
-  const int16_t out0 = WRAPLOW(dct_const_round_shift(input[0] * cospi_16_64));
+  const int16_t out0 =
+      WRAPLOW(dct_const_round_shift((int16_t)input[0] * cospi_16_64));
   const int16_t out1 = WRAPLOW(dct_const_round_shift(out0 * cospi_16_64));
   const int16_t a1 = ROUND_POWER_OF_TWO(out1, 6);
 

diff --git a/vpx_dsp/arm/idct32x32_1_add_neon.c b/vpx_dsp/arm/idct32x32_1_add_neon.c
index 604d82a..8920b93 100644 (file)
--- a/vpx_dsp/arm/idct32x32_1_add_neon.c
+++ b/vpx_dsp/arm/idct32x32_1_add_neon.c
@@ -39,7 +39,8 @@ static INLINE void idct32x32_1_add_neg_kernel(uint8_t **dest, const int stride,
 void vpx_idct32x32_1_add_neon(const tran_low_t *input, uint8_t *dest,
                               int stride) {
   int i;
-  const int16_t out0 = WRAPLOW(dct_const_round_shift(input[0] * cospi_16_64));
+  const int16_t out0 =
+      WRAPLOW(dct_const_round_shift((int16_t)input[0] * cospi_16_64));
   const int16_t out1 = WRAPLOW(dct_const_round_shift(out0 * cospi_16_64));
   const int16_t a1 = ROUND_POWER_OF_TWO(out1, 6);
 

diff --git a/vpx_dsp/arm/idct4x4_1_add_neon.c b/vpx_dsp/arm/idct4x4_1_add_neon.c
index 21d21b0..a14b895 100644 (file)
--- a/vpx_dsp/arm/idct4x4_1_add_neon.c
+++ b/vpx_dsp/arm/idct4x4_1_add_neon.c
@@ -32,7 +32,8 @@ static INLINE void idct4x4_1_add_kernel(uint8_t **dest, const int stride,
 
 void vpx_idct4x4_1_add_neon(const tran_low_t *input, uint8_t *dest,
                             int stride) {
-  const int16_t out0 = WRAPLOW(dct_const_round_shift(input[0] * cospi_16_64));
+  const int16_t out0 =
+      WRAPLOW(dct_const_round_shift((int16_t)input[0] * cospi_16_64));
   const int16_t out1 = WRAPLOW(dct_const_round_shift(out0 * cospi_16_64));
   const int16_t a1 = ROUND_POWER_OF_TWO(out1, 4);
   const int16x8_t dc = vdupq_n_s16(a1);

diff --git a/vpx_dsp/arm/idct8x8_1_add_neon.c b/vpx_dsp/arm/idct8x8_1_add_neon.c
index 7bcce91..ce9b459 100644 (file)
--- a/vpx_dsp/arm/idct8x8_1_add_neon.c
+++ b/vpx_dsp/arm/idct8x8_1_add_neon.c
@@ -36,7 +36,8 @@ static INLINE void idct8x8_1_add_neg_kernel(uint8_t **dest, const int stride,
 
 void vpx_idct8x8_1_add_neon(const tran_low_t *input, uint8_t *dest,
                             int stride) {
-  const int16_t out0 = WRAPLOW(dct_const_round_shift(input[0] * cospi_16_64));
+  const int16_t out0 =
+      WRAPLOW(dct_const_round_shift((int16_t)input[0] * cospi_16_64));
   const int16_t out1 = WRAPLOW(dct_const_round_shift(out0 * cospi_16_64));
   const int16_t a1 = ROUND_POWER_OF_TWO(out1, 5);
 

diff --git a/vpx_dsp/inv_txfm.c b/vpx_dsp/inv_txfm.c
index 2320261..0194aa1 100644 (file)
--- a/vpx_dsp/inv_txfm.c
+++ b/vpx_dsp/inv_txfm.c
@@ -131,16 +131,16 @@ void iadst4_c(const tran_low_t *input, tran_low_t *output) {
 }
 
 void idct4_c(const tran_low_t *input, tran_low_t *output) {
-  tran_low_t step[4];
+  int16_t step[4];
   tran_high_t temp1, temp2;
 
   // stage 1
-  temp1 = (input[0] + input[2]) * cospi_16_64;
-  temp2 = (input[0] - input[2]) * cospi_16_64;
+  temp1 = ((int16_t)input[0] + (int16_t)input[2]) * cospi_16_64;
+  temp2 = ((int16_t)input[0] - (int16_t)input[2]) * cospi_16_64;
   step[0] = WRAPLOW(dct_const_round_shift(temp1));
   step[1] = WRAPLOW(dct_const_round_shift(temp2));
-  temp1 = input[1] * cospi_24_64 - input[3] * cospi_8_64;
-  temp2 = input[1] * cospi_8_64 + input[3] * cospi_24_64;
+  temp1 = (int16_t)input[1] * cospi_24_64 - (int16_t)input[3] * cospi_8_64;
+  temp2 = (int16_t)input[1] * cospi_8_64 + (int16_t)input[3] * cospi_24_64;
   step[2] = WRAPLOW(dct_const_round_shift(temp1));
   step[3] = WRAPLOW(dct_const_round_shift(temp2));
 
@@ -178,7 +178,8 @@ void vpx_idct4x4_16_add_c(const tran_low_t *input, uint8_t *dest, int stride) {
 void vpx_idct4x4_1_add_c(const tran_low_t *input, uint8_t *dest, int stride) {
   int i;
   tran_high_t a1;
-  tran_low_t out = WRAPLOW(dct_const_round_shift(input[0] * cospi_16_64));
+  tran_low_t out =
+      WRAPLOW(dct_const_round_shift((int16_t)input[0] * cospi_16_64));
 
   out = WRAPLOW(dct_const_round_shift(out * cospi_16_64));
   a1 = ROUND_POWER_OF_TWO(out, 4);
@@ -268,20 +269,20 @@ void iadst8_c(const tran_low_t *input, tran_low_t *output) {
 }
 
 void idct8_c(const tran_low_t *input, tran_low_t *output) {
-  tran_low_t step1[8], step2[8];
+  int16_t step1[8], step2[8];
   tran_high_t temp1, temp2;
 
   // stage 1
-  step1[0] = input[0];
-  step1[2] = input[4];
-  step1[1] = input[2];
-  step1[3] = input[6];
-  temp1 = input[1] * cospi_28_64 - input[7] * cospi_4_64;
-  temp2 = input[1] * cospi_4_64 + input[7] * cospi_28_64;
+  step1[0] = (int16_t)input[0];
+  step1[2] = (int16_t)input[4];
+  step1[1] = (int16_t)input[2];
+  step1[3] = (int16_t)input[6];
+  temp1 = (int16_t)input[1] * cospi_28_64 - (int16_t)input[7] * cospi_4_64;
+  temp2 = (int16_t)input[1] * cospi_4_64 + (int16_t)input[7] * cospi_28_64;
   step1[4] = WRAPLOW(dct_const_round_shift(temp1));
   step1[7] = WRAPLOW(dct_const_round_shift(temp2));
-  temp1 = input[5] * cospi_12_64 - input[3] * cospi_20_64;
-  temp2 = input[5] * cospi_20_64 + input[3] * cospi_12_64;
+  temp1 = (int16_t)input[5] * cospi_12_64 - (int16_t)input[3] * cospi_20_64;
+  temp2 = (int16_t)input[5] * cospi_20_64 + (int16_t)input[3] * cospi_12_64;
   step1[5] = WRAPLOW(dct_const_round_shift(temp1));
   step1[6] = WRAPLOW(dct_const_round_shift(temp2));
 
@@ -374,7 +375,8 @@ void vpx_idct8x8_12_add_c(const tran_low_t *input, uint8_t *dest, int stride) {
 void vpx_idct8x8_1_add_c(const tran_low_t *input, uint8_t *dest, int stride) {
   int i, j;
   tran_high_t a1;
-  tran_low_t out = WRAPLOW(dct_const_round_shift(input[0] * cospi_16_64));
+  tran_low_t out =
+      WRAPLOW(dct_const_round_shift((int16_t)input[0] * cospi_16_64));
 
   out = WRAPLOW(dct_const_round_shift(out * cospi_16_64));
   a1 = ROUND_POWER_OF_TWO(out, 5);
@@ -553,26 +555,26 @@ void iadst16_c(const tran_low_t *input, tran_low_t *output) {
 }
 
 void idct16_c(const tran_low_t *input, tran_low_t *output) {
-  tran_low_t step1[16], step2[16];
+  int16_t step1[16], step2[16];
   tran_high_t temp1, temp2;
 
   // stage 1
-  step1[0] = input[0 / 2];
-  step1[1] = input[16 / 2];
-  step1[2] = input[8 / 2];
-  step1[3] = input[24 / 2];
-  step1[4] = input[4 / 2];
-  step1[5] = input[20 / 2];
-  step1[6] = input[12 / 2];
-  step1[7] = input[28 / 2];
-  step1[8] = input[2 / 2];
-  step1[9] = input[18 / 2];
-  step1[10] = input[10 / 2];
-  step1[11] = input[26 / 2];
-  step1[12] = input[6 / 2];
-  step1[13] = input[22 / 2];
-  step1[14] = input[14 / 2];
-  step1[15] = input[30 / 2];
+  step1[0] = (int16_t)input[0 / 2];
+  step1[1] = (int16_t)input[16 / 2];
+  step1[2] = (int16_t)input[8 / 2];
+  step1[3] = (int16_t)input[24 / 2];
+  step1[4] = (int16_t)input[4 / 2];
+  step1[5] = (int16_t)input[20 / 2];
+  step1[6] = (int16_t)input[12 / 2];
+  step1[7] = (int16_t)input[28 / 2];
+  step1[8] = (int16_t)input[2 / 2];
+  step1[9] = (int16_t)input[18 / 2];
+  step1[10] = (int16_t)input[10 / 2];
+  step1[11] = (int16_t)input[26 / 2];
+  step1[12] = (int16_t)input[6 / 2];
+  step1[13] = (int16_t)input[22 / 2];
+  step1[14] = (int16_t)input[14 / 2];
+  step1[15] = (int16_t)input[30 / 2];
 
   // stage 2
   step2[0] = step1[0];
@@ -797,7 +799,8 @@ void vpx_idct16x16_10_add_c(const tran_low_t *input, uint8_t *dest,
 void vpx_idct16x16_1_add_c(const tran_low_t *input, uint8_t *dest, int stride) {
   int i, j;
   tran_high_t a1;
-  tran_low_t out = WRAPLOW(dct_const_round_shift(input[0] * cospi_16_64));
+  tran_low_t out =
+      WRAPLOW(dct_const_round_shift((int16_t)input[0] * cospi_16_64));
 
   out = WRAPLOW(dct_const_round_shift(out * cospi_16_64));
   a1 = ROUND_POWER_OF_TWO(out, 6);
@@ -808,64 +811,64 @@ void vpx_idct16x16_1_add_c(const tran_low_t *input, uint8_t *dest, int stride) {
 }
 
 void idct32_c(const tran_low_t *input, tran_low_t *output) {
-  tran_low_t step1[32], step2[32];
+  int16_t step1[32], step2[32];
   tran_high_t temp1, temp2;
 
   // stage 1
-  step1[0] = input[0];
-  step1[1] = input[16];
-  step1[2] = input[8];
-  step1[3] = input[24];
-  step1[4] = input[4];
-  step1[5] = input[20];
-  step1[6] = input[12];
-  step1[7] = input[28];
-  step1[8] = input[2];
-  step1[9] = input[18];
-  step1[10] = input[10];
-  step1[11] = input[26];
-  step1[12] = input[6];
-  step1[13] = input[22];
-  step1[14] = input[14];
-  step1[15] = input[30];
-
-  temp1 = input[1] * cospi_31_64 - input[31] * cospi_1_64;
-  temp2 = input[1] * cospi_1_64 + input[31] * cospi_31_64;
+  step1[0] = (int16_t)input[0];
+  step1[1] = (int16_t)input[16];
+  step1[2] = (int16_t)input[8];
+  step1[3] = (int16_t)input[24];
+  step1[4] = (int16_t)input[4];
+  step1[5] = (int16_t)input[20];
+  step1[6] = (int16_t)input[12];
+  step1[7] = (int16_t)input[28];
+  step1[8] = (int16_t)input[2];
+  step1[9] = (int16_t)input[18];
+  step1[10] = (int16_t)input[10];
+  step1[11] = (int16_t)input[26];
+  step1[12] = (int16_t)input[6];
+  step1[13] = (int16_t)input[22];
+  step1[14] = (int16_t)input[14];
+  step1[15] = (int16_t)input[30];
+
+  temp1 = (int16_t)input[1] * cospi_31_64 - (int16_t)input[31] * cospi_1_64;
+  temp2 = (int16_t)input[1] * cospi_1_64 + (int16_t)input[31] * cospi_31_64;
   step1[16] = WRAPLOW(dct_const_round_shift(temp1));
   step1[31] = WRAPLOW(dct_const_round_shift(temp2));
 
-  temp1 = input[17] * cospi_15_64 - input[15] * cospi_17_64;
-  temp2 = input[17] * cospi_17_64 + input[15] * cospi_15_64;
+  temp1 = (int16_t)input[17] * cospi_15_64 - (int16_t)input[15] * cospi_17_64;
+  temp2 = (int16_t)input[17] * cospi_17_64 + (int16_t)input[15] * cospi_15_64;
   step1[17] = WRAPLOW(dct_const_round_shift(temp1));
   step1[30] = WRAPLOW(dct_const_round_shift(temp2));
 
-  temp1 = input[9] * cospi_23_64 - input[23] * cospi_9_64;
-  temp2 = input[9] * cospi_9_64 + input[23] * cospi_23_64;
+  temp1 = (int16_t)input[9] * cospi_23_64 - (int16_t)input[23] * cospi_9_64;
+  temp2 = (int16_t)input[9] * cospi_9_64 + (int16_t)input[23] * cospi_23_64;
   step1[18] = WRAPLOW(dct_const_round_shift(temp1));
   step1[29] = WRAPLOW(dct_const_round_shift(temp2));
 
-  temp1 = input[25] * cospi_7_64 - input[7] * cospi_25_64;
-  temp2 = input[25] * cospi_25_64 + input[7] * cospi_7_64;
+  temp1 = (int16_t)input[25] * cospi_7_64 - (int16_t)input[7] * cospi_25_64;
+  temp2 = (int16_t)input[25] * cospi_25_64 + (int16_t)input[7] * cospi_7_64;
   step1[19] = WRAPLOW(dct_const_round_shift(temp1));
   step1[28] = WRAPLOW(dct_const_round_shift(temp2));
 
-  temp1 = input[5] * cospi_27_64 - input[27] * cospi_5_64;
-  temp2 = input[5] * cospi_5_64 + input[27] * cospi_27_64;
+  temp1 = (int16_t)input[5] * cospi_27_64 - (int16_t)input[27] * cospi_5_64;
+  temp2 = (int16_t)input[5] * cospi_5_64 + (int16_t)input[27] * cospi_27_64;
   step1[20] = WRAPLOW(dct_const_round_shift(temp1));
   step1[27] = WRAPLOW(dct_const_round_shift(temp2));
 
-  temp1 = input[21] * cospi_11_64 - input[11] * cospi_21_64;
-  temp2 = input[21] * cospi_21_64 + input[11] * cospi_11_64;
+  temp1 = (int16_t)input[21] * cospi_11_64 - (int16_t)input[11] * cospi_21_64;
+  temp2 = (int16_t)input[21] * cospi_21_64 + (int16_t)input[11] * cospi_11_64;
   step1[21] = WRAPLOW(dct_const_round_shift(temp1));
   step1[26] = WRAPLOW(dct_const_round_shift(temp2));
 
-  temp1 = input[13] * cospi_19_64 - input[19] * cospi_13_64;
-  temp2 = input[13] * cospi_13_64 + input[19] * cospi_19_64;
+  temp1 = (int16_t)input[13] * cospi_19_64 - (int16_t)input[19] * cospi_13_64;
+  temp2 = (int16_t)input[13] * cospi_13_64 + (int16_t)input[19] * cospi_19_64;
   step1[22] = WRAPLOW(dct_const_round_shift(temp1));
   step1[25] = WRAPLOW(dct_const_round_shift(temp2));
 
-  temp1 = input[29] * cospi_3_64 - input[3] * cospi_29_64;
-  temp2 = input[29] * cospi_29_64 + input[3] * cospi_3_64;
+  temp1 = (int16_t)input[29] * cospi_3_64 - (int16_t)input[3] * cospi_29_64;
+  temp2 = (int16_t)input[29] * cospi_29_64 + (int16_t)input[3] * cospi_3_64;
   step1[23] = WRAPLOW(dct_const_round_shift(temp1));
   step1[24] = WRAPLOW(dct_const_round_shift(temp2));
 
@@ -1260,7 +1263,8 @@ void vpx_idct32x32_34_add_c(const tran_low_t *input, uint8_t *dest,
 void vpx_idct32x32_1_add_c(const tran_low_t *input, uint8_t *dest, int stride) {
   int i, j;
   tran_high_t a1;
-  tran_low_t out = WRAPLOW(dct_const_round_shift(input[0] * cospi_16_64));
+  tran_low_t out =
+      WRAPLOW(dct_const_round_shift((int16_t)input[0] * cospi_16_64));
 
   out = WRAPLOW(dct_const_round_shift(out * cospi_16_64));
   a1 = ROUND_POWER_OF_TWO(out, 6);

diff --git a/vpx_dsp/x86/inv_txfm_sse2.c b/vpx_dsp/x86/inv_txfm_sse2.c
index 64fe50b..f6e56b6 100644 (file)
--- a/vpx_dsp/x86/inv_txfm_sse2.c
+++ b/vpx_dsp/x86/inv_txfm_sse2.c
@@ -51,7 +51,7 @@ void vpx_idct4x4_1_add_sse2(const tran_low_t *input, uint8_t *dest,
   int a;
   __m128i dc_value, d[2];
 
-  a = (int)dct_const_round_shift(input[0] * cospi_16_64);
+  a = (int)dct_const_round_shift((int16_t)input[0] * cospi_16_64);
   a = (int)dct_const_round_shift(a * cospi_16_64);
   a = ROUND_POWER_OF_TWO(a, 4);
 
@@ -210,7 +210,8 @@ void vpx_idct8x8_1_add_sse2(const tran_low_t *input, uint8_t *dest,
                             int stride) {
   __m128i dc_value;
   tran_high_t a1;
-  tran_low_t out = WRAPLOW(dct_const_round_shift(input[0] * cospi_16_64));
+  tran_low_t out =
+      WRAPLOW(dct_const_round_shift((int16_t)input[0] * cospi_16_64));
 
   out = WRAPLOW(dct_const_round_shift(out * cospi_16_64));
   a1 = ROUND_POWER_OF_TWO(out, 5);
@@ -547,7 +548,8 @@ void vpx_idct16x16_1_add_sse2(const tran_low_t *input, uint8_t *dest,
   __m128i dc_value;
   int i;
   tran_high_t a1;
-  tran_low_t out = WRAPLOW(dct_const_round_shift(input[0] * cospi_16_64));
+  tran_low_t out =
+      WRAPLOW(dct_const_round_shift((int16_t)input[0] * cospi_16_64));
 
   out = WRAPLOW(dct_const_round_shift(out * cospi_16_64));
   a1 = ROUND_POWER_OF_TWO(out, 6);
@@ -1334,7 +1336,8 @@ void vpx_idct32x32_1_add_sse2(const tran_low_t *input, uint8_t *dest,
   __m128i dc_value;
   int j;
   tran_high_t a1;
-  tran_low_t out = WRAPLOW(dct_const_round_shift(input[0] * cospi_16_64));
+  tran_low_t out =
+      WRAPLOW(dct_const_round_shift((int16_t)input[0] * cospi_16_64));
 
   out = WRAPLOW(dct_const_round_shift(out * cospi_16_64));
   a1 = ROUND_POWER_OF_TWO(out, 6);