Update aom to slightly newer commit ID

4 files changed, 415 insertions, 251 deletions

diff --git a/third_party/aom/av1/encoder/x86/av1_quantize_sse2.c b/third_party/aom/av1/encoder/x86/av1_quantize_sse2.c
index f9c95b6cb..190317389 100644
--- a/third_party/aom/av1/encoder/x86/av1_quantize_sse2.c
+++ b/third_party/aom/av1/encoder/x86/av1_quantize_sse2.c
@@ -15,13 +15,65 @@
 #include "./av1_rtcd.h"
 #include "aom/aom_integer.h"
 
-void av1_quantize_fp_sse2(const int16_t *coeff_ptr, intptr_t n_coeffs,
+static INLINE void read_coeff(const tran_low_t *coeff, intptr_t offset,
+                              __m128i *c0, __m128i *c1) {
+  const tran_low_t *addr = coeff + offset;
+#if CONFIG_HIGHBITDEPTH
+  const __m128i x0 = _mm_load_si128((const __m128i *)addr);
+  const __m128i x1 = _mm_load_si128((const __m128i *)addr + 1);
+  const __m128i x2 = _mm_load_si128((const __m128i *)addr + 2);
+  const __m128i x3 = _mm_load_si128((const __m128i *)addr + 3);
+  *c0 = _mm_packs_epi32(x0, x1);
+  *c1 = _mm_packs_epi32(x2, x3);
+#else
+  *c0 = _mm_load_si128((const __m128i *)addr);
+  *c1 = _mm_load_si128((const __m128i *)addr + 1);
+#endif
+}
+
+static INLINE void write_qcoeff(const __m128i *qc0, const __m128i *qc1,
+                                tran_low_t *qcoeff, intptr_t offset) {
+  tran_low_t *addr = qcoeff + offset;
+#if CONFIG_HIGHBITDEPTH
+  const __m128i zero = _mm_setzero_si128();
+  __m128i sign_bits = _mm_cmplt_epi16(*qc0, zero);
+  __m128i y0 = _mm_unpacklo_epi16(*qc0, sign_bits);
+  __m128i y1 = _mm_unpackhi_epi16(*qc0, sign_bits);
+  _mm_store_si128((__m128i *)addr, y0);
+  _mm_store_si128((__m128i *)addr + 1, y1);
+
+  sign_bits = _mm_cmplt_epi16(*qc1, zero);
+  y0 = _mm_unpacklo_epi16(*qc1, sign_bits);
+  y1 = _mm_unpackhi_epi16(*qc1, sign_bits);
+  _mm_store_si128((__m128i *)addr + 2, y0);
+  _mm_store_si128((__m128i *)addr + 3, y1);
+#else
+  _mm_store_si128((__m128i *)addr, *qc0);
+  _mm_store_si128((__m128i *)addr + 1, *qc1);
+#endif
+}
+
+static INLINE void write_zero(tran_low_t *qcoeff, intptr_t offset) {
+  const __m128i zero = _mm_setzero_si128();
+  tran_low_t *addr = qcoeff + offset;
+#if CONFIG_HIGHBITDEPTH
+  _mm_store_si128((__m128i *)addr, zero);
+  _mm_store_si128((__m128i *)addr + 1, zero);
+  _mm_store_si128((__m128i *)addr + 2, zero);
+  _mm_store_si128((__m128i *)addr + 3, zero);
+#else
+  _mm_store_si128((__m128i *)addr, zero);
+  _mm_store_si128((__m128i *)addr + 1, zero);
+#endif
+}
+
+void av1_quantize_fp_sse2(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
                           int skip_block, const int16_t *zbin_ptr,
                           const int16_t *round_ptr, const int16_t *quant_ptr,
-                          const int16_t *quant_shift_ptr, int16_t *qcoeff_ptr,
-                          int16_t *dqcoeff_ptr, const int16_t *dequant_ptr,
-                          uint16_t *eob_ptr, const int16_t *scan_ptr,
-                          const int16_t *iscan_ptr) {
+                          const int16_t *quant_shift_ptr,
+                          tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
+                          const int16_t *dequant_ptr, uint16_t *eob_ptr,
+                          const int16_t *scan_ptr, const int16_t *iscan_ptr) {
   __m128i zero;
   __m128i thr;
   int16_t nzflag;
@@ -54,8 +106,7 @@ void av1_quantize_fp_sse2(const int16_t *coeff_ptr, intptr_t n_coeffs,
         __m128i qcoeff0, qcoeff1;
         __m128i qtmp0, qtmp1;
         // Do DC and first 15 AC
-        coeff0 = _mm_load_si128((const __m128i *)(coeff_ptr + n_coeffs));
-        coeff1 = _mm_load_si128((const __m128i *)(coeff_ptr + n_coeffs) + 1);
+        read_coeff(coeff_ptr, n_coeffs, &coeff0, &coeff1);
 
         // Poor man's sign extract
         coeff0_sign = _mm_srai_epi16(coeff0, 15);
@@ -78,15 +129,13 @@ void av1_quantize_fp_sse2(const int16_t *coeff_ptr, intptr_t n_coeffs,
         qcoeff0 = _mm_sub_epi16(qcoeff0, coeff0_sign);
         qcoeff1 = _mm_sub_epi16(qcoeff1, coeff1_sign);
 
-        _mm_store_si128((__m128i *)(qcoeff_ptr + n_coeffs), qcoeff0);
-        _mm_store_si128((__m128i *)(qcoeff_ptr + n_coeffs) + 1, qcoeff1);
+        write_qcoeff(&qcoeff0, &qcoeff1, qcoeff_ptr, n_coeffs);
 
         coeff0 = _mm_mullo_epi16(qcoeff0, dequant);
         dequant = _mm_unpackhi_epi64(dequant, dequant);
         coeff1 = _mm_mullo_epi16(qcoeff1, dequant);
 
-        _mm_store_si128((__m128i *)(dqcoeff_ptr + n_coeffs), coeff0);
-        _mm_store_si128((__m128i *)(dqcoeff_ptr + n_coeffs) + 1, coeff1);
+        write_qcoeff(&coeff0, &coeff1, dqcoeff_ptr, n_coeffs);
       }
 
       {
@@ -121,8 +170,7 @@ void av1_quantize_fp_sse2(const int16_t *coeff_ptr, intptr_t n_coeffs,
         __m128i qcoeff0, qcoeff1;
         __m128i qtmp0, qtmp1;
 
-        coeff0 = _mm_load_si128((const __m128i *)(coeff_ptr + n_coeffs));
-        coeff1 = _mm_load_si128((const __m128i *)(coeff_ptr + n_coeffs) + 1);
+        read_coeff(coeff_ptr, n_coeffs, &coeff0, &coeff1);
 
         // Poor man's sign extract
         coeff0_sign = _mm_srai_epi16(coeff0, 15);
@@ -147,20 +195,15 @@ void av1_quantize_fp_sse2(const int16_t *coeff_ptr, intptr_t n_coeffs,
           qcoeff0 = _mm_sub_epi16(qcoeff0, coeff0_sign);
           qcoeff1 = _mm_sub_epi16(qcoeff1, coeff1_sign);
 
-          _mm_store_si128((__m128i *)(qcoeff_ptr + n_coeffs), qcoeff0);
-          _mm_store_si128((__m128i *)(qcoeff_ptr + n_coeffs) + 1, qcoeff1);
+          write_qcoeff(&qcoeff0, &qcoeff1, qcoeff_ptr, n_coeffs);
 
           coeff0 = _mm_mullo_epi16(qcoeff0, dequant);
           coeff1 = _mm_mullo_epi16(qcoeff1, dequant);
 
-          _mm_store_si128((__m128i *)(dqcoeff_ptr + n_coeffs), coeff0);
-          _mm_store_si128((__m128i *)(dqcoeff_ptr + n_coeffs) + 1, coeff1);
+          write_qcoeff(&coeff0, &coeff1, dqcoeff_ptr, n_coeffs);
         } else {
-          _mm_store_si128((__m128i *)(qcoeff_ptr + n_coeffs), zero);
-          _mm_store_si128((__m128i *)(qcoeff_ptr + n_coeffs) + 1, zero);
-
-          _mm_store_si128((__m128i *)(dqcoeff_ptr + n_coeffs), zero);
-          _mm_store_si128((__m128i *)(dqcoeff_ptr + n_coeffs) + 1, zero);
+          write_zero(qcoeff_ptr, n_coeffs);
+          write_zero(dqcoeff_ptr, n_coeffs);
         }
       }
 
@@ -200,10 +243,8 @@ void av1_quantize_fp_sse2(const int16_t *coeff_ptr, intptr_t n_coeffs,
     }
   } else {
     do {
-      _mm_store_si128((__m128i *)(dqcoeff_ptr + n_coeffs), zero);
-      _mm_store_si128((__m128i *)(dqcoeff_ptr + n_coeffs) + 1, zero);
-      _mm_store_si128((__m128i *)(qcoeff_ptr + n_coeffs), zero);
-      _mm_store_si128((__m128i *)(qcoeff_ptr + n_coeffs) + 1, zero);
+      write_zero(dqcoeff_ptr, n_coeffs);
+      write_zero(qcoeff_ptr, n_coeffs);
       n_coeffs += 8 * 2;
     } while (n_coeffs < 0);
     *eob_ptr = 0;
diff --git a/third_party/aom/av1/encoder/x86/corner_match_sse4.c b/third_party/aom/av1/encoder/x86/corner_match_sse4.c
new file mode 100644
index 000000000..179da0d28
--- /dev/null
+++ b/third_party/aom/av1/encoder/x86/corner_match_sse4.c
@@ -0,0 +1,91 @@
+#include <stdlib.h>
+#include <memory.h>
+#include <math.h>
+#include <assert.h>
+
+#include <smmintrin.h>
+
+#include "./av1_rtcd.h"
+#include "aom_ports/mem.h"
+#include "av1/encoder/corner_match.h"
+
+DECLARE_ALIGNED(16, static const uint8_t, byte_mask[16]) = {
+  255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 0, 0, 0
+};
+#if MATCH_SZ != 13
+#error "Need to change byte_mask in corner_match_sse4.c if MATCH_SZ != 13"
+#endif
+
+/* Compute corr(im1, im2) * MATCH_SZ * stddev(im1), where the
+   correlation/standard deviation are taken over MATCH_SZ by MATCH_SZ windows
+   of each image, centered at (x1, y1) and (x2, y2) respectively.
+*/
+double compute_cross_correlation_sse4_1(unsigned char *im1, int stride1, int x1,
+                                        int y1, unsigned char *im2, int stride2,
+                                        int x2, int y2) {
+  int i;
+  // 2 16-bit partial sums in lanes 0, 4 (== 2 32-bit partial sums in lanes 0,
+  // 2)
+  __m128i sum1_vec = _mm_setzero_si128();
+  __m128i sum2_vec = _mm_setzero_si128();
+  // 4 32-bit partial sums of squares
+  __m128i sumsq2_vec = _mm_setzero_si128();
+  __m128i cross_vec = _mm_setzero_si128();
+
+  const __m128i mask = _mm_load_si128((__m128i *)byte_mask);
+  const __m128i zero = _mm_setzero_si128();
+
+  im1 += (y1 - MATCH_SZ_BY2) * stride1 + (x1 - MATCH_SZ_BY2);
+  im2 += (y2 - MATCH_SZ_BY2) * stride2 + (x2 - MATCH_SZ_BY2);
+
+  for (i = 0; i < MATCH_SZ; ++i) {
+    const __m128i v1 =
+        _mm_and_si128(_mm_loadu_si128((__m128i *)&im1[i * stride1]), mask);
+    const __m128i v2 =
+        _mm_and_si128(_mm_loadu_si128((__m128i *)&im2[i * stride2]), mask);
+
+    // Using the 'sad' intrinsic here is a bit faster than adding
+    // v1_l + v1_r and v2_l + v2_r, plus it avoids the need for a 16->32 bit
+    // conversion step later, for a net speedup of ~10%
+    sum1_vec = _mm_add_epi16(sum1_vec, _mm_sad_epu8(v1, zero));
+    sum2_vec = _mm_add_epi16(sum2_vec, _mm_sad_epu8(v2, zero));
+
+    const __m128i v1_l = _mm_cvtepu8_epi16(v1);
+    const __m128i v1_r = _mm_cvtepu8_epi16(_mm_srli_si128(v1, 8));
+    const __m128i v2_l = _mm_cvtepu8_epi16(v2);
+    const __m128i v2_r = _mm_cvtepu8_epi16(_mm_srli_si128(v2, 8));
+
+    sumsq2_vec = _mm_add_epi32(
+        sumsq2_vec,
+        _mm_add_epi32(_mm_madd_epi16(v2_l, v2_l), _mm_madd_epi16(v2_r, v2_r)));
+    cross_vec = _mm_add_epi32(
+        cross_vec,
+        _mm_add_epi32(_mm_madd_epi16(v1_l, v2_l), _mm_madd_epi16(v1_r, v2_r)));
+  }
+
+  // Now we can treat the four registers (sum1_vec, sum2_vec, sumsq2_vec,
+  // cross_vec)
+  // as holding 4 32-bit elements each, which we want to sum horizontally.
+  // We do this by transposing and then summing vertically.
+  __m128i tmp_0 = _mm_unpacklo_epi32(sum1_vec, sum2_vec);
+  __m128i tmp_1 = _mm_unpackhi_epi32(sum1_vec, sum2_vec);
+  __m128i tmp_2 = _mm_unpacklo_epi32(sumsq2_vec, cross_vec);
+  __m128i tmp_3 = _mm_unpackhi_epi32(sumsq2_vec, cross_vec);
+
+  __m128i tmp_4 = _mm_unpacklo_epi64(tmp_0, tmp_2);
+  __m128i tmp_5 = _mm_unpackhi_epi64(tmp_0, tmp_2);
+  __m128i tmp_6 = _mm_unpacklo_epi64(tmp_1, tmp_3);
+  __m128i tmp_7 = _mm_unpackhi_epi64(tmp_1, tmp_3);
+
+  __m128i res =
+      _mm_add_epi32(_mm_add_epi32(tmp_4, tmp_5), _mm_add_epi32(tmp_6, tmp_7));
+
+  int sum1 = _mm_extract_epi32(res, 0);
+  int sum2 = _mm_extract_epi32(res, 1);
+  int sumsq2 = _mm_extract_epi32(res, 2);
+  int cross = _mm_extract_epi32(res, 3);
+
+  int var2 = sumsq2 * MATCH_SZ_SQ - sum2 * sum2;
+  int cov = cross * MATCH_SZ_SQ - sum1 * sum2;
+  return cov / sqrt((double)var2);
+}
diff --git a/third_party/aom/av1/encoder/x86/highbd_fwd_txfm_sse4.c b/third_party/aom/av1/encoder/x86/highbd_fwd_txfm_sse4.c
index f201a29aa..b56eed518 100644
--- a/third_party/aom/av1/encoder/x86/highbd_fwd_txfm_sse4.c
+++ b/third_party/aom/av1/encoder/x86/highbd_fwd_txfm_sse4.c
@@ -13,7 +13,7 @@
 
 #include "./av1_rtcd.h"
 #include "./aom_config.h"
-#include "av1/common/av1_fwd_txfm2d_cfg.h"
+#include "av1/common/av1_fwd_txfm1d_cfg.h"
 #include "av1/common/av1_txfm.h"
 #include "av1/common/x86/highbd_txfm_utility_sse4.h"
 #include "aom_dsp/txfm_common.h"
@@ -58,7 +58,7 @@ static INLINE void load_buffer_4x4(const int16_t *input, __m128i *in,
 // shift[1] is used in txfm_func_col()
 // shift[2] is used in txfm_func_row()
 static void fdct4x4_sse4_1(__m128i *in, int bit) {
-  const int32_t *cospi = cospi_arr[bit - cos_bit_min];
+  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]);
@@ -133,7 +133,7 @@ void av1_highbd_fht4x4_sse4_1(const int16_t *input, tran_low_t *output,
 }
 
 static void fadst4x4_sse4_1(__m128i *in, int bit) {
-  const int32_t *cospi = cospi_arr[bit - cos_bit_min];
+  const int32_t *cospi = cospi_arr(bit);
   const __m128i cospi8 = _mm_set1_epi32(cospi[8]);
   const __m128i cospi56 = _mm_set1_epi32(cospi[56]);
   const __m128i cospi40 = _mm_set1_epi32(cospi[40]);
@@ -209,71 +209,81 @@ static void fadst4x4_sse4_1(__m128i *in, int bit) {
 void av1_fwd_txfm2d_4x4_sse4_1(const int16_t *input, int32_t *coeff,
                                int input_stride, int tx_type, int bd) {
   __m128i in[4];
-  const TXFM_2D_CFG *cfg = NULL;
+  const TXFM_1D_CFG *row_cfg = NULL;
+  const TXFM_1D_CFG *col_cfg = NULL;
 
   switch (tx_type) {
     case DCT_DCT:
-      cfg = &fwd_txfm_2d_cfg_dct_dct_4;
-      load_buffer_4x4(input, in, input_stride, 0, 0, cfg->shift[0]);
-      fdct4x4_sse4_1(in, cfg->cos_bit_col[2]);
-      fdct4x4_sse4_1(in, cfg->cos_bit_row[2]);
+      row_cfg = &fwd_txfm_1d_row_cfg_dct_4;
+      col_cfg = &fwd_txfm_1d_col_cfg_dct_4;
+      load_buffer_4x4(input, in, input_stride, 0, 0, row_cfg->shift[0]);
+      fdct4x4_sse4_1(in, col_cfg->cos_bit[2]);
+      fdct4x4_sse4_1(in, row_cfg->cos_bit[2]);
       write_buffer_4x4(in, coeff);
       break;
     case ADST_DCT:
-      cfg = &fwd_txfm_2d_cfg_adst_dct_4;
-      load_buffer_4x4(input, in, input_stride, 0, 0, cfg->shift[0]);
-      fadst4x4_sse4_1(in, cfg->cos_bit_col[2]);
-      fdct4x4_sse4_1(in, cfg->cos_bit_row[2]);
+      row_cfg = &fwd_txfm_1d_row_cfg_dct_4;
+      col_cfg = &fwd_txfm_1d_col_cfg_adst_4;
+      load_buffer_4x4(input, in, input_stride, 0, 0, row_cfg->shift[0]);
+      fadst4x4_sse4_1(in, col_cfg->cos_bit[2]);
+      fdct4x4_sse4_1(in, row_cfg->cos_bit[2]);
       write_buffer_4x4(in, coeff);
       break;
     case DCT_ADST:
-      cfg = &fwd_txfm_2d_cfg_dct_adst_4;
-      load_buffer_4x4(input, in, input_stride, 0, 0, cfg->shift[0]);
-      fdct4x4_sse4_1(in, cfg->cos_bit_col[2]);
-      fadst4x4_sse4_1(in, cfg->cos_bit_row[2]);
+      row_cfg = &fwd_txfm_1d_row_cfg_adst_4;
+      col_cfg = &fwd_txfm_1d_col_cfg_dct_4;
+      load_buffer_4x4(input, in, input_stride, 0, 0, row_cfg->shift[0]);
+      fdct4x4_sse4_1(in, col_cfg->cos_bit[2]);
+      fadst4x4_sse4_1(in, row_cfg->cos_bit[2]);
       write_buffer_4x4(in, coeff);
       break;
     case ADST_ADST:
-      cfg = &fwd_txfm_2d_cfg_adst_adst_4;
-      load_buffer_4x4(input, in, input_stride, 0, 0, cfg->shift[0]);
-      fadst4x4_sse4_1(in, cfg->cos_bit_col[2]);
-      fadst4x4_sse4_1(in, cfg->cos_bit_row[2]);
+      row_cfg = &fwd_txfm_1d_row_cfg_adst_4;
+      col_cfg = &fwd_txfm_1d_col_cfg_adst_4;
+      load_buffer_4x4(input, in, input_stride, 0, 0, row_cfg->shift[0]);
+      fadst4x4_sse4_1(in, col_cfg->cos_bit[2]);
+      fadst4x4_sse4_1(in, row_cfg->cos_bit[2]);
       write_buffer_4x4(in, coeff);
       break;
 #if CONFIG_EXT_TX
     case FLIPADST_DCT:
-      cfg = &fwd_txfm_2d_cfg_adst_dct_4;
-      load_buffer_4x4(input, in, input_stride, 1, 0, cfg->shift[0]);
-      fadst4x4_sse4_1(in, cfg->cos_bit_col[2]);
-      fdct4x4_sse4_1(in, cfg->cos_bit_row[2]);
+      row_cfg = &fwd_txfm_1d_row_cfg_dct_4;
+      col_cfg = &fwd_txfm_1d_col_cfg_adst_4;
+      load_buffer_4x4(input, in, input_stride, 1, 0, row_cfg->shift[0]);
+      fadst4x4_sse4_1(in, col_cfg->cos_bit[2]);
+      fdct4x4_sse4_1(in, row_cfg->cos_bit[2]);
       write_buffer_4x4(in, coeff);
       break;
     case DCT_FLIPADST:
-      cfg = &fwd_txfm_2d_cfg_dct_adst_4;
-      load_buffer_4x4(input, in, input_stride, 0, 1, cfg->shift[0]);
-      fdct4x4_sse4_1(in, cfg->cos_bit_col[2]);
-      fadst4x4_sse4_1(in, cfg->cos_bit_row[2]);
+      row_cfg = &fwd_txfm_1d_row_cfg_adst_4;
+      col_cfg = &fwd_txfm_1d_col_cfg_dct_4;
+      load_buffer_4x4(input, in, input_stride, 0, 1, row_cfg->shift[0]);
+      fdct4x4_sse4_1(in, col_cfg->cos_bit[2]);
+      fadst4x4_sse4_1(in, row_cfg->cos_bit[2]);
       write_buffer_4x4(in, coeff);
       break;
     case FLIPADST_FLIPADST:
-      cfg = &fwd_txfm_2d_cfg_adst_adst_4;
-      load_buffer_4x4(input, in, input_stride, 1, 1, cfg->shift[0]);
-      fadst4x4_sse4_1(in, cfg->cos_bit_col[2]);
-      fadst4x4_sse4_1(in, cfg->cos_bit_row[2]);
+      row_cfg = &fwd_txfm_1d_row_cfg_adst_4;
+      col_cfg = &fwd_txfm_1d_col_cfg_adst_4;
+      load_buffer_4x4(input, in, input_stride, 1, 1, row_cfg->shift[0]);
+      fadst4x4_sse4_1(in, col_cfg->cos_bit[2]);
+      fadst4x4_sse4_1(in, row_cfg->cos_bit[2]);
       write_buffer_4x4(in, coeff);
       break;
     case ADST_FLIPADST:
-      cfg = &fwd_txfm_2d_cfg_adst_adst_4;
-      load_buffer_4x4(input, in, input_stride, 0, 1, cfg->shift[0]);
-      fadst4x4_sse4_1(in, cfg->cos_bit_col[2]);
-      fadst4x4_sse4_1(in, cfg->cos_bit_row[2]);
+      row_cfg = &fwd_txfm_1d_row_cfg_adst_4;
+      col_cfg = &fwd_txfm_1d_col_cfg_adst_4;
+      load_buffer_4x4(input, in, input_stride, 0, 1, row_cfg->shift[0]);
+      fadst4x4_sse4_1(in, col_cfg->cos_bit[2]);
+      fadst4x4_sse4_1(in, row_cfg->cos_bit[2]);
       write_buffer_4x4(in, coeff);
       break;
     case FLIPADST_ADST:
-      cfg = &fwd_txfm_2d_cfg_adst_adst_4;
-      load_buffer_4x4(input, in, input_stride, 1, 0, cfg->shift[0]);
-      fadst4x4_sse4_1(in, cfg->cos_bit_col[2]);
-      fadst4x4_sse4_1(in, cfg->cos_bit_row[2]);
+      row_cfg = &fwd_txfm_1d_row_cfg_adst_4;
+      col_cfg = &fwd_txfm_1d_col_cfg_adst_4;
+      load_buffer_4x4(input, in, input_stride, 1, 0, row_cfg->shift[0]);
+      fadst4x4_sse4_1(in, col_cfg->cos_bit[2]);
+      fadst4x4_sse4_1(in, row_cfg->cos_bit[2]);
       write_buffer_4x4(in, coeff);
       break;
 #endif
@@ -429,7 +439,7 @@ static INLINE void write_buffer_8x8(const __m128i *res, tran_low_t *output) {
 }
 
 static void fdct8x8_sse4_1(__m128i *in, __m128i *out, int bit) {
-  const int32_t *cospi = cospi_arr[bit - cos_bit_min];
+  const int32_t *cospi = cospi_arr(bit);
   const __m128i cospi32 = _mm_set1_epi32(cospi[32]);
   const __m128i cospim32 = _mm_set1_epi32(-cospi[32]);
   const __m128i cospi48 = _mm_set1_epi32(cospi[48]);
@@ -625,7 +635,7 @@ static void fdct8x8_sse4_1(__m128i *in, __m128i *out, int bit) {
 }
 
 static void fadst8x8_sse4_1(__m128i *in, __m128i *out, int bit) {
-  const int32_t *cospi = cospi_arr[bit - cos_bit_min];
+  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]);
@@ -930,97 +940,107 @@ static void fadst8x8_sse4_1(__m128i *in, __m128i *out, int bit) {
 void av1_fwd_txfm2d_8x8_sse4_1(const int16_t *input, int32_t *coeff, int stride,
                                int tx_type, int bd) {
   __m128i in[16], out[16];
-  const TXFM_2D_CFG *cfg = NULL;
+  const TXFM_1D_CFG *row_cfg = NULL;
+  const TXFM_1D_CFG *col_cfg = NULL;
 
   switch (tx_type) {
     case DCT_DCT:
-      cfg = &fwd_txfm_2d_cfg_dct_dct_8;
-      load_buffer_8x8(input, in, stride, 0, 0, cfg->shift[0]);
-      fdct8x8_sse4_1(in, out, cfg->cos_bit_col[2]);
-      col_txfm_8x8_rounding(out, -cfg->shift[1]);
+      row_cfg = &fwd_txfm_1d_row_cfg_dct_8;
+      col_cfg = &fwd_txfm_1d_col_cfg_dct_8;
+      load_buffer_8x8(input, in, stride, 0, 0, row_cfg->shift[0]);
+      fdct8x8_sse4_1(in, out, col_cfg->cos_bit[2]);
+      col_txfm_8x8_rounding(out, -row_cfg->shift[1]);
       transpose_8x8(out, in);
-      fdct8x8_sse4_1(in, out, cfg->cos_bit_row[2]);
+      fdct8x8_sse4_1(in, out, row_cfg->cos_bit[2]);
       transpose_8x8(out, in);
       write_buffer_8x8(in, coeff);
       break;
     case ADST_DCT:
-      cfg = &fwd_txfm_2d_cfg_adst_dct_8;
-      load_buffer_8x8(input, in, stride, 0, 0, cfg->shift[0]);
-      fadst8x8_sse4_1(in, out, cfg->cos_bit_col[2]);
-      col_txfm_8x8_rounding(out, -cfg->shift[1]);
+      row_cfg = &fwd_txfm_1d_row_cfg_dct_8;
+      col_cfg = &fwd_txfm_1d_col_cfg_adst_8;
+      load_buffer_8x8(input, in, stride, 0, 0, row_cfg->shift[0]);
+      fadst8x8_sse4_1(in, out, col_cfg->cos_bit[2]);
+      col_txfm_8x8_rounding(out, -row_cfg->shift[1]);
       transpose_8x8(out, in);
-      fdct8x8_sse4_1(in, out, cfg->cos_bit_row[2]);
+      fdct8x8_sse4_1(in, out, row_cfg->cos_bit[2]);
       transpose_8x8(out, in);
       write_buffer_8x8(in, coeff);
       break;
     case DCT_ADST:
-      cfg = &fwd_txfm_2d_cfg_dct_adst_8;
-      load_buffer_8x8(input, in, stride, 0, 0, cfg->shift[0]);
-      fdct8x8_sse4_1(in, out, cfg->cos_bit_col[2]);
-      col_txfm_8x8_rounding(out, -cfg->shift[1]);
+      row_cfg = &fwd_txfm_1d_row_cfg_adst_8;
+      col_cfg = &fwd_txfm_1d_col_cfg_dct_8;
+      load_buffer_8x8(input, in, stride, 0, 0, row_cfg->shift[0]);
+      fdct8x8_sse4_1(in, out, col_cfg->cos_bit[2]);
+      col_txfm_8x8_rounding(out, -row_cfg->shift[1]);
       transpose_8x8(out, in);
-      fadst8x8_sse4_1(in, out, cfg->cos_bit_row[2]);
+      fadst8x8_sse4_1(in, out, row_cfg->cos_bit[2]);
       transpose_8x8(out, in);
       write_buffer_8x8(in, coeff);
       break;
     case ADST_ADST:
-      cfg = &fwd_txfm_2d_cfg_adst_adst_8;
-      load_buffer_8x8(input, in, stride, 0, 0, cfg->shift[0]);
-      fadst8x8_sse4_1(in, out, cfg->cos_bit_col[2]);
-      col_txfm_8x8_rounding(out, -cfg->shift[1]);
+      row_cfg = &fwd_txfm_1d_row_cfg_adst_8;
+      col_cfg = &fwd_txfm_1d_col_cfg_adst_8;
+      load_buffer_8x8(input, in, stride, 0, 0, row_cfg->shift[0]);
+      fadst8x8_sse4_1(in, out, col_cfg->cos_bit[2]);
+      col_txfm_8x8_rounding(out, -row_cfg->shift[1]);
       transpose_8x8(out, in);
-      fadst8x8_sse4_1(in, out, cfg->cos_bit_row[2]);
+      fadst8x8_sse4_1(in, out, row_cfg->cos_bit[2]);
       transpose_8x8(out, in);
       write_buffer_8x8(in, coeff);
       break;
 #if CONFIG_EXT_TX
     case FLIPADST_DCT:
-      cfg = &fwd_txfm_2d_cfg_adst_dct_8;
-      load_buffer_8x8(input, in, stride, 1, 0, cfg->shift[0]);
-      fadst8x8_sse4_1(in, out, cfg->cos_bit_col[2]);
-      col_txfm_8x8_rounding(out, -cfg->shift[1]);
+      row_cfg = &fwd_txfm_1d_row_cfg_dct_8;
+      col_cfg = &fwd_txfm_1d_col_cfg_adst_8;
+      load_buffer_8x8(input, in, stride, 1, 0, row_cfg->shift[0]);
+      fadst8x8_sse4_1(in, out, col_cfg->cos_bit[2]);
+      col_txfm_8x8_rounding(out, -row_cfg->shift[1]);
       transpose_8x8(out, in);
-      fdct8x8_sse4_1(in, out, cfg->cos_bit_row[2]);
+      fdct8x8_sse4_1(in, out, row_cfg->cos_bit[2]);
       transpose_8x8(out, in);
       write_buffer_8x8(in, coeff);
       break;
     case DCT_FLIPADST:
-      cfg = &fwd_txfm_2d_cfg_dct_adst_8;
-      load_buffer_8x8(input, in, stride, 0, 1, cfg->shift[0]);
-      fdct8x8_sse4_1(in, out, cfg->cos_bit_col[2]);
-      col_txfm_8x8_rounding(out, -cfg->shift[1]);
+      row_cfg = &fwd_txfm_1d_row_cfg_adst_8;
+      col_cfg = &fwd_txfm_1d_col_cfg_dct_8;
+      load_buffer_8x8(input, in, stride, 0, 1, row_cfg->shift[0]);
+      fdct8x8_sse4_1(in, out, col_cfg->cos_bit[2]);
+      col_txfm_8x8_rounding(out, -row_cfg->shift[1]);
       transpose_8x8(out, in);
-      fadst8x8_sse4_1(in, out, cfg->cos_bit_row[2]);
+      fadst8x8_sse4_1(in, out, row_cfg->cos_bit[2]);
       transpose_8x8(out, in);
       write_buffer_8x8(in, coeff);
       break;
     case FLIPADST_FLIPADST:
-      cfg = &fwd_txfm_2d_cfg_adst_adst_8;
-      load_buffer_8x8(input, in, stride, 1, 1, cfg->shift[0]);
-      fadst8x8_sse4_1(in, out, cfg->cos_bit_col[2]);
-      col_txfm_8x8_rounding(out, -cfg->shift[1]);
+      row_cfg = &fwd_txfm_1d_row_cfg_adst_8;
+      col_cfg = &fwd_txfm_1d_col_cfg_adst_8;
+      load_buffer_8x8(input, in, stride, 1, 1, row_cfg->shift[0]);
+      fadst8x8_sse4_1(in, out, col_cfg->cos_bit[2]);
+      col_txfm_8x8_rounding(out, -row_cfg->shift[1]);
       transpose_8x8(out, in);
-      fadst8x8_sse4_1(in, out, cfg->cos_bit_row[2]);
+      fadst8x8_sse4_1(in, out, row_cfg->cos_bit[2]);
       transpose_8x8(out, in);
       write_buffer_8x8(in, coeff);
       break;
     case ADST_FLIPADST:
-      cfg = &fwd_txfm_2d_cfg_adst_adst_8;
-      load_buffer_8x8(input, in, stride, 0, 1, cfg->shift[0]);
-      fadst8x8_sse4_1(in, out, cfg->cos_bit_col[2]);
-      col_txfm_8x8_rounding(out, -cfg->shift[1]);
+      row_cfg = &fwd_txfm_1d_row_cfg_adst_8;
+      col_cfg = &fwd_txfm_1d_col_cfg_adst_8;
+      load_buffer_8x8(input, in, stride, 0, 1, row_cfg->shift[0]);
+      fadst8x8_sse4_1(in, out, col_cfg->cos_bit[2]);
+      col_txfm_8x8_rounding(out, -row_cfg->shift[1]);
       transpose_8x8(out, in);
-      fadst8x8_sse4_1(in, out, cfg->cos_bit_row[2]);
+      fadst8x8_sse4_1(in, out, row_cfg->cos_bit[2]);
       transpose_8x8(out, in);
       write_buffer_8x8(in, coeff);
       break;
     case FLIPADST_ADST:
-      cfg = &fwd_txfm_2d_cfg_adst_adst_8;
-      load_buffer_8x8(input, in, stride, 1, 0, cfg->shift[0]);
-      fadst8x8_sse4_1(in, out, cfg->cos_bit_col[2]);
-      col_txfm_8x8_rounding(out, -cfg->shift[1]);
+      row_cfg = &fwd_txfm_1d_row_cfg_adst_8;
+      col_cfg = &fwd_txfm_1d_col_cfg_adst_8;
+      load_buffer_8x8(input, in, stride, 1, 0, row_cfg->shift[0]);
+      fadst8x8_sse4_1(in, out, col_cfg->cos_bit[2]);
+      col_txfm_8x8_rounding(out, -row_cfg->shift[1]);
       transpose_8x8(out, in);
-      fadst8x8_sse4_1(in, out, cfg->cos_bit_row[2]);
+      fadst8x8_sse4_1(in, out, row_cfg->cos_bit[2]);
       transpose_8x8(out, in);
       write_buffer_8x8(in, coeff);
       break;
@@ -1107,7 +1127,7 @@ static INLINE void load_buffer_16x16(const int16_t *input, __m128i *out,
 }
 
 static void fdct16x16_sse4_1(__m128i *in, __m128i *out, int bit) {
-  const int32_t *cospi = cospi_arr[bit - cos_bit_min];
+  const int32_t *cospi = cospi_arr(bit);
   const __m128i cospi32 = _mm_set1_epi32(cospi[32]);
   const __m128i cospim32 = _mm_set1_epi32(-cospi[32]);
   const __m128i cospi48 = _mm_set1_epi32(cospi[48]);
@@ -1393,7 +1413,7 @@ static void fdct16x16_sse4_1(__m128i *in, __m128i *out, int bit) {
 }
 
 static void fadst16x16_sse4_1(__m128i *in, __m128i *out, int bit) {
-  const int32_t *cospi = cospi_arr[bit - cos_bit_min];
+  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]);
@@ -1794,97 +1814,107 @@ static void write_buffer_16x16(const __m128i *in, tran_low_t *output) {
 void av1_fwd_txfm2d_16x16_sse4_1(const int16_t *input, int32_t *coeff,
                                  int stride, int tx_type, int bd) {
   __m128i in[64], out[64];
-  const TXFM_2D_CFG *cfg = NULL;
+  const TXFM_1D_CFG *row_cfg = NULL;
+  const TXFM_1D_CFG *col_cfg = NULL;
 
   switch (tx_type) {
     case DCT_DCT:
-      cfg = &fwd_txfm_2d_cfg_dct_dct_16;
-      load_buffer_16x16(input, in, stride, 0, 0, cfg->shift[0]);
-      fdct16x16_sse4_1(in, out, cfg->cos_bit_col[0]);
-      col_txfm_16x16_rounding(out, -cfg->shift[1]);
+      row_cfg = &fwd_txfm_1d_row_cfg_dct_16;
+      col_cfg = &fwd_txfm_1d_col_cfg_dct_16;
+      load_buffer_16x16(input, in, stride, 0, 0, row_cfg->shift[0]);
+      fdct16x16_sse4_1(in, out, col_cfg->cos_bit[0]);
+      col_txfm_16x16_rounding(out, -row_cfg->shift[1]);
       transpose_16x16(out, in);
-      fdct16x16_sse4_1(in, out, cfg->cos_bit_row[0]);
+      fdct16x16_sse4_1(in, out, row_cfg->cos_bit[0]);
       transpose_16x16(out, in);
       write_buffer_16x16(in, coeff);
       break;
     case ADST_DCT:
-      cfg = &fwd_txfm_2d_cfg_adst_dct_16;
-      load_buffer_16x16(input, in, stride, 0, 0, cfg->shift[0]);
-      fadst16x16_sse4_1(in, out, cfg->cos_bit_col[0]);
-      col_txfm_16x16_rounding(out, -cfg->shift[1]);
+      row_cfg = &fwd_txfm_1d_row_cfg_dct_16;
+      col_cfg = &fwd_txfm_1d_col_cfg_adst_16;
+      load_buffer_16x16(input, in, stride, 0, 0, row_cfg->shift[0]);
+      fadst16x16_sse4_1(in, out, col_cfg->cos_bit[0]);
+      col_txfm_16x16_rounding(out, -row_cfg->shift[1]);
       transpose_16x16(out, in);
-      fdct16x16_sse4_1(in, out, cfg->cos_bit_row[0]);
+      fdct16x16_sse4_1(in, out, row_cfg->cos_bit[0]);
       transpose_16x16(out, in);
       write_buffer_16x16(in, coeff);
       break;
     case DCT_ADST:
-      cfg = &fwd_txfm_2d_cfg_dct_adst_16;
-      load_buffer_16x16(input, in, stride, 0, 0, cfg->shift[0]);
-      fdct16x16_sse4_1(in, out, cfg->cos_bit_col[0]);
-      col_txfm_16x16_rounding(out, -cfg->shift[1]);
+      row_cfg = &fwd_txfm_1d_row_cfg_adst_16;
+      col_cfg = &fwd_txfm_1d_col_cfg_dct_16;
+      load_buffer_16x16(input, in, stride, 0, 0, row_cfg->shift[0]);
+      fdct16x16_sse4_1(in, out, col_cfg->cos_bit[0]);
+      col_txfm_16x16_rounding(out, -row_cfg->shift[1]);
       transpose_16x16(out, in);
-      fadst16x16_sse4_1(in, out, cfg->cos_bit_row[0]);
+      fadst16x16_sse4_1(in, out, row_cfg->cos_bit[0]);
       transpose_16x16(out, in);
       write_buffer_16x16(in, coeff);
       break;
     case ADST_ADST:
-      cfg = &fwd_txfm_2d_cfg_adst_adst_16;
-      load_buffer_16x16(input, in, stride, 0, 0, cfg->shift[0]);
-      fadst16x16_sse4_1(in, out, cfg->cos_bit_col[0]);
-      col_txfm_16x16_rounding(out, -cfg->shift[1]);
+      row_cfg = &fwd_txfm_1d_row_cfg_adst_16;
+      col_cfg = &fwd_txfm_1d_col_cfg_adst_16;
+      load_buffer_16x16(input, in, stride, 0, 0, row_cfg->shift[0]);
+      fadst16x16_sse4_1(in, out, col_cfg->cos_bit[0]);
+      col_txfm_16x16_rounding(out, -row_cfg->shift[1]);
       transpose_16x16(out, in);
-      fadst16x16_sse4_1(in, out, cfg->cos_bit_row[0]);
+      fadst16x16_sse4_1(in, out, row_cfg->cos_bit[0]);
       transpose_16x16(out, in);
       write_buffer_16x16(in, coeff);
       break;
 #if CONFIG_EXT_TX
     case FLIPADST_DCT:
-      cfg = &fwd_txfm_2d_cfg_adst_dct_16;
-      load_buffer_16x16(input, in, stride, 1, 0, cfg->shift[0]);
-      fadst16x16_sse4_1(in, out, cfg->cos_bit_col[0]);
-      col_txfm_16x16_rounding(out, -cfg->shift[1]);
+      row_cfg = &fwd_txfm_1d_row_cfg_dct_16;
+      col_cfg = &fwd_txfm_1d_col_cfg_adst_16;
+      load_buffer_16x16(input, in, stride, 1, 0, row_cfg->shift[0]);
+      fadst16x16_sse4_1(in, out, col_cfg->cos_bit[0]);
+      col_txfm_16x16_rounding(out, -row_cfg->shift[1]);
       transpose_16x16(out, in);
-      fdct16x16_sse4_1(in, out, cfg->cos_bit_row[0]);
+      fdct16x16_sse4_1(in, out, row_cfg->cos_bit[0]);
       transpose_16x16(out, in);
       write_buffer_16x16(in, coeff);
       break;
     case DCT_FLIPADST:
-      cfg = &fwd_txfm_2d_cfg_dct_adst_16;
-      load_buffer_16x16(input, in, stride, 0, 1, cfg->shift[0]);
-      fdct16x16_sse4_1(in, out, cfg->cos_bit_col[0]);
-      col_txfm_16x16_rounding(out, -cfg->shift[1]);
+      row_cfg = &fwd_txfm_1d_row_cfg_adst_16;
+      col_cfg = &fwd_txfm_1d_col_cfg_dct_16;
+      load_buffer_16x16(input, in, stride, 0, 1, row_cfg->shift[0]);
+      fdct16x16_sse4_1(in, out, col_cfg->cos_bit[0]);
+      col_txfm_16x16_rounding(out, -row_cfg->shift[1]);
       transpose_16x16(out, in);
-      fadst16x16_sse4_1(in, out, cfg->cos_bit_row[0]);
+      fadst16x16_sse4_1(in, out, row_cfg->cos_bit[0]);
       transpose_16x16(out, in);
       write_buffer_16x16(in, coeff);
       break;
     case FLIPADST_FLIPADST:
-      cfg = &fwd_txfm_2d_cfg_adst_adst_16;
-      load_buffer_16x16(input, in, stride, 1, 1, cfg->shift[0]);
-      fadst16x16_sse4_1(in, out, cfg->cos_bit_col[0]);
-      col_txfm_16x16_rounding(out, -cfg->shift[1]);
+      row_cfg = &fwd_txfm_1d_row_cfg_adst_16;
+      col_cfg = &fwd_txfm_1d_col_cfg_adst_16;
+      load_buffer_16x16(input, in, stride, 1, 1, row_cfg->shift[0]);
+      fadst16x16_sse4_1(in, out, col_cfg->cos_bit[0]);
+      col_txfm_16x16_rounding(out, -row_cfg->shift[1]);
       transpose_16x16(out, in);
-      fadst16x16_sse4_1(in, out, cfg->cos_bit_row[0]);
+      fadst16x16_sse4_1(in, out, row_cfg->cos_bit[0]);
       transpose_16x16(out, in);
       write_buffer_16x16(in, coeff);
       break;
     case ADST_FLIPADST:
-      cfg = &fwd_txfm_2d_cfg_adst_adst_16;
-      load_buffer_16x16(input, in, stride, 0, 1, cfg->shift[0]);
-      fadst16x16_sse4_1(in, out, cfg->cos_bit_col[0]);
-      col_txfm_16x16_rounding(out, -cfg->shift[1]);
+      row_cfg = &fwd_txfm_1d_row_cfg_adst_16;
+      col_cfg = &fwd_txfm_1d_col_cfg_adst_16;
+      load_buffer_16x16(input, in, stride, 0, 1, row_cfg->shift[0]);
+      fadst16x16_sse4_1(in, out, col_cfg->cos_bit[0]);
+      col_txfm_16x16_rounding(out, -row_cfg->shift[1]);
       transpose_16x16(out, in);
-      fadst16x16_sse4_1(in, out, cfg->cos_bit_row[0]);
+      fadst16x16_sse4_1(in, out, row_cfg->cos_bit[0]);
       transpose_16x16(out, in);
       write_buffer_16x16(in, coeff);
       break;
     case FLIPADST_ADST:
-      cfg = &fwd_txfm_2d_cfg_adst_adst_16;
-      load_buffer_16x16(input, in, stride, 1, 0, cfg->shift[0]);
-      fadst16x16_sse4_1(in, out, cfg->cos_bit_col[0]);
-      col_txfm_16x16_rounding(out, -cfg->shift[1]);
+      row_cfg = &fwd_txfm_1d_row_cfg_adst_16;
+      col_cfg = &fwd_txfm_1d_col_cfg_adst_16;
+      load_buffer_16x16(input, in, stride, 1, 0, row_cfg->shift[0]);
+      fadst16x16_sse4_1(in, out, col_cfg->cos_bit[0]);
+      col_txfm_16x16_rounding(out, -row_cfg->shift[1]);
       transpose_16x16(out, in);
-      fadst16x16_sse4_1(in, out, cfg->cos_bit_row[0]);
+      fadst16x16_sse4_1(in, out, row_cfg->cos_bit[0]);
       transpose_16x16(out, in);
       write_buffer_16x16(in, coeff);
       break;
diff --git a/third_party/aom/av1/encoder/x86/hybrid_fwd_txfm_avx2.c b/third_party/aom/av1/encoder/x86/hybrid_fwd_txfm_avx2.c
index 198e4e4c4..8495ad1aa 100644
--- a/third_party/aom/av1/encoder/x86/hybrid_fwd_txfm_avx2.c
+++ b/third_party/aom/av1/encoder/x86/hybrid_fwd_txfm_avx2.c
@@ -269,8 +269,8 @@ static void fdct16_avx2(__m256i *in) {
   x0 = _mm256_unpacklo_epi16(v0, v1);
   x1 = _mm256_unpackhi_epi16(v0, v1);
 
-  t0 = butter_fly(x0, x1, cospi_p16_p16);
-  t1 = butter_fly(x0, x1, cospi_p16_m16);
+  t0 = butter_fly(&x0, &x1, &cospi_p16_p16);
+  t1 = butter_fly(&x0, &x1, &cospi_p16_m16);
 
   // 4, 12
   v0 = _mm256_sub_epi16(s1, s2);
@@ -279,8 +279,8 @@ static void fdct16_avx2(__m256i *in) {
   x0 = _mm256_unpacklo_epi16(v0, v1);
   x1 = _mm256_unpackhi_epi16(v0, v1);
 
-  t2 = butter_fly(x0, x1, cospi_p24_p08);
-  t3 = butter_fly(x0, x1, cospi_m08_p24);
+  t2 = butter_fly(&x0, &x1, &cospi_p24_p08);
+  t3 = butter_fly(&x0, &x1, &cospi_m08_p24);
 
   // 2, 6, 10, 14
   s0 = _mm256_sub_epi16(u3, u4);
@@ -294,8 +294,8 @@ static void fdct16_avx2(__m256i *in) {
   x0 = _mm256_unpacklo_epi16(s2, s1);
   x1 = _mm256_unpackhi_epi16(s2, s1);
 
-  v2 = butter_fly(x0, x1, cospi_p16_p16);  // output[5]
-  v1 = butter_fly(x0, x1, cospi_p16_m16);  // output[6]
+  v2 = butter_fly(&x0, &x1, &cospi_p16_p16);  // output[5]
+  v1 = butter_fly(&x0, &x1, &cospi_p16_m16);  // output[6]
 
   s0 = _mm256_add_epi16(v0, v1);  // step[4]
   s1 = _mm256_sub_epi16(v0, v1);  // step[5]
@@ -306,14 +306,14 @@ static void fdct16_avx2(__m256i *in) {
   x0 = _mm256_unpacklo_epi16(s0, s3);
   x1 = _mm256_unpackhi_epi16(s0, s3);
 
-  t4 = butter_fly(x0, x1, cospi_p28_p04);
-  t5 = butter_fly(x0, x1, cospi_m04_p28);
+  t4 = butter_fly(&x0, &x1, &cospi_p28_p04);
+  t5 = butter_fly(&x0, &x1, &cospi_m04_p28);
 
   // 10, 6
   x0 = _mm256_unpacklo_epi16(s1, s2);
   x1 = _mm256_unpackhi_epi16(s1, s2);
-  t6 = butter_fly(x0, x1, cospi_p12_p20);
-  t7 = butter_fly(x0, x1, cospi_m20_p12);
+  t6 = butter_fly(&x0, &x1, &cospi_p12_p20);
+  t7 = butter_fly(&x0, &x1, &cospi_m20_p12);
 
   // 1, 3, 5, 7, 9, 11, 13, 15
   s0 = _mm256_sub_epi16(in[7], in[8]);  // step[8]
@@ -337,14 +337,14 @@ static void fdct16_avx2(__m256i *in) {
   x0 = _mm256_unpacklo_epi16(u5, u2);
   x1 = _mm256_unpackhi_epi16(u5, u2);
 
-  s2 = butter_fly(x0, x1, cospi_p16_p16);  // step[13]
-  s5 = butter_fly(x0, x1, cospi_p16_m16);  // step[10]
+  s2 = butter_fly(&x0, &x1, &cospi_p16_p16);  // step[13]
+  s5 = butter_fly(&x0, &x1, &cospi_p16_m16);  // step[10]
 
   x0 = _mm256_unpacklo_epi16(u4, u3);
   x1 = _mm256_unpackhi_epi16(u4, u3);
 
-  s3 = butter_fly(x0, x1, cospi_p16_p16);  // step[12]
-  s4 = butter_fly(x0, x1, cospi_p16_m16);  // step[11]
+  s3 = butter_fly(&x0, &x1, &cospi_p16_p16);  // step[12]
+  s4 = butter_fly(&x0, &x1, &cospi_p16_m16);  // step[11]
 
   u0 = _mm256_add_epi16(s0, s4);  // output[8]
   u1 = _mm256_add_epi16(s1, s5);
@@ -364,14 +364,14 @@ static void fdct16_avx2(__m256i *in) {
   x0 = _mm256_unpacklo_epi16(u1, u6);
   x1 = _mm256_unpackhi_epi16(u1, u6);
 
-  s1 = butter_fly(x0, x1, cospi_m08_p24);
-  s6 = butter_fly(x0, x1, cospi_p24_p08);
+  s1 = butter_fly(&x0, &x1, &cospi_m08_p24);
+  s6 = butter_fly(&x0, &x1, &cospi_p24_p08);
 
   x0 = _mm256_unpacklo_epi16(u2, u5);
   x1 = _mm256_unpackhi_epi16(u2, u5);
 
-  s2 = butter_fly(x0, x1, cospi_m24_m08);
-  s5 = butter_fly(x0, x1, cospi_m08_p24);
+  s2 = butter_fly(&x0, &x1, &cospi_m24_m08);
+  s5 = butter_fly(&x0, &x1, &cospi_m08_p24);
 
   // stage 5
   u0 = _mm256_add_epi16(s0, s1);
@@ -386,23 +386,23 @@ static void fdct16_avx2(__m256i *in) {
   // stage 6
   x0 = _mm256_unpacklo_epi16(u0, u7);
   x1 = _mm256_unpackhi_epi16(u0, u7);
-  in[1] = butter_fly(x0, x1, cospi_p30_p02);
-  in[15] = butter_fly(x0, x1, cospi_m02_p30);
+  in[1] = butter_fly(&x0, &x1, &cospi_p30_p02);
+  in[15] = butter_fly(&x0, &x1, &cospi_m02_p30);
 
   x0 = _mm256_unpacklo_epi16(u1, u6);
   x1 = _mm256_unpackhi_epi16(u1, u6);
-  in[9] = butter_fly(x0, x1, cospi_p14_p18);
-  in[7] = butter_fly(x0, x1, cospi_m18_p14);
+  in[9] = butter_fly(&x0, &x1, &cospi_p14_p18);
+  in[7] = butter_fly(&x0, &x1, &cospi_m18_p14);
 
   x0 = _mm256_unpacklo_epi16(u2, u5);
   x1 = _mm256_unpackhi_epi16(u2, u5);
-  in[5] = butter_fly(x0, x1, cospi_p22_p10);
-  in[11] = butter_fly(x0, x1, cospi_m10_p22);
+  in[5] = butter_fly(&x0, &x1, &cospi_p22_p10);
+  in[11] = butter_fly(&x0, &x1, &cospi_m10_p22);
 
   x0 = _mm256_unpacklo_epi16(u3, u4);
   x1 = _mm256_unpackhi_epi16(u3, u4);
-  in[13] = butter_fly(x0, x1, cospi_p06_p26);
-  in[3] = butter_fly(x0, x1, cospi_m26_p06);
+  in[13] = butter_fly(&x0, &x1, &cospi_p06_p26);
+  in[3] = butter_fly(&x0, &x1, &cospi_m26_p06);
 }
 
 void fadst16_avx2(__m256i *in) {
@@ -953,7 +953,9 @@ void fadst16_avx2(__m256i *in) {
 }
 
 #if CONFIG_EXT_TX
-static void fidtx16_avx2(__m256i *in) { txfm_scaling16_avx2(Sqrt2, in); }
+static void fidtx16_avx2(__m256i *in) {
+  txfm_scaling16_avx2((int16_t)Sqrt2, in);
+}
 #endif
 
 void av1_fht16x16_avx2(const int16_t *input, tran_low_t *output, int stride,
@@ -964,28 +966,28 @@ void av1_fht16x16_avx2(const int16_t *input, tran_low_t *output, int stride,
     case DCT_DCT:
       load_buffer_16x16(input, stride, 0, 0, in);
       fdct16_avx2(in);
-      mm256_transpose_16x16(in);
+      mm256_transpose_16x16(in, in);
       right_shift_16x16(in);
       fdct16_avx2(in);
       break;
     case ADST_DCT:
       load_buffer_16x16(input, stride, 0, 0, in);
       fadst16_avx2(in);
-      mm256_transpose_16x16(in);
+      mm256_transpose_16x16(in, in);
       right_shift_16x16(in);
       fdct16_avx2(in);
       break;
     case DCT_ADST:
       load_buffer_16x16(input, stride, 0, 0, in);
       fdct16_avx2(in);
-      mm256_transpose_16x16(in);
+      mm256_transpose_16x16(in, in);
       right_shift_16x16(in);
       fadst16_avx2(in);
       break;
     case ADST_ADST:
       load_buffer_16x16(input, stride, 0, 0, in);
       fadst16_avx2(in);
-      mm256_transpose_16x16(in);
+      mm256_transpose_16x16(in, in);
       right_shift_16x16(in);
       fadst16_avx2(in);
       break;
@@ -993,91 +995,91 @@ void av1_fht16x16_avx2(const int16_t *input, tran_low_t *output, int stride,
     case FLIPADST_DCT:
       load_buffer_16x16(input, stride, 1, 0, in);
       fadst16_avx2(in);
-      mm256_transpose_16x16(in);
+      mm256_transpose_16x16(in, in);
       right_shift_16x16(in);
       fdct16_avx2(in);
       break;
     case DCT_FLIPADST:
       load_buffer_16x16(input, stride, 0, 1, in);
       fdct16_avx2(in);
-      mm256_transpose_16x16(in);
+      mm256_transpose_16x16(in, in);
       right_shift_16x16(in);
       fadst16_avx2(in);
       break;
     case FLIPADST_FLIPADST:
       load_buffer_16x16(input, stride, 1, 1, in);
       fadst16_avx2(in);
-      mm256_transpose_16x16(in);
+      mm256_transpose_16x16(in, in);
       right_shift_16x16(in);
       fadst16_avx2(in);
       break;
     case ADST_FLIPADST:
       load_buffer_16x16(input, stride, 0, 1, in);
       fadst16_avx2(in);
-      mm256_transpose_16x16(in);
+      mm256_transpose_16x16(in, in);
       right_shift_16x16(in);
       fadst16_avx2(in);
       break;
     case FLIPADST_ADST:
       load_buffer_16x16(input, stride, 1, 0, in);
       fadst16_avx2(in);
-      mm256_transpose_16x16(in);
+      mm256_transpose_16x16(in, in);
       right_shift_16x16(in);
       fadst16_avx2(in);
       break;
     case IDTX:
       load_buffer_16x16(input, stride, 0, 0, in);
       fidtx16_avx2(in);
-      mm256_transpose_16x16(in);
+      mm256_transpose_16x16(in, in);
       right_shift_16x16(in);
       fidtx16_avx2(in);
       break;
     case V_DCT:
       load_buffer_16x16(input, stride, 0, 0, in);
       fdct16_avx2(in);
-      mm256_transpose_16x16(in);
+      mm256_transpose_16x16(in, in);
       right_shift_16x16(in);
       fidtx16_avx2(in);
       break;
     case H_DCT:
       load_buffer_16x16(input, stride, 0, 0, in);
       fidtx16_avx2(in);
-      mm256_transpose_16x16(in);
+      mm256_transpose_16x16(in, in);
       right_shift_16x16(in);
       fdct16_avx2(in);
       break;
     case V_ADST:
       load_buffer_16x16(input, stride, 0, 0, in);
       fadst16_avx2(in);
-      mm256_transpose_16x16(in);
+      mm256_transpose_16x16(in, in);
       right_shift_16x16(in);
       fidtx16_avx2(in);
       break;
     case H_ADST:
       load_buffer_16x16(input, stride, 0, 0, in);
       fidtx16_avx2(in);
-      mm256_transpose_16x16(in);
+      mm256_transpose_16x16(in, in);
       right_shift_16x16(in);
       fadst16_avx2(in);
       break;
     case V_FLIPADST:
       load_buffer_16x16(input, stride, 1, 0, in);
       fadst16_avx2(in);
-      mm256_transpose_16x16(in);
+      mm256_transpose_16x16(in, in);
       right_shift_16x16(in);
       fidtx16_avx2(in);
       break;
     case H_FLIPADST:
       load_buffer_16x16(input, stride, 0, 1, in);
       fidtx16_avx2(in);
-      mm256_transpose_16x16(in);
+      mm256_transpose_16x16(in, in);
       right_shift_16x16(in);
       fadst16_avx2(in);
       break;
 #endif  // CONFIG_EXT_TX
     default: assert(0); break;
   }
-  mm256_transpose_16x16(in);
+  mm256_transpose_16x16(in, in);
   write_buffer_16x16(in, output);
   _mm256_zeroupper();
 }
@@ -1110,10 +1112,10 @@ static void mm256_vectors_swap(__m256i *a0, __m256i *a1, const int size) {
 }
 
 static void mm256_transpose_32x32(__m256i *in0, __m256i *in1) {
-  mm256_transpose_16x16(in0);
-  mm256_transpose_16x16(&in0[16]);
-  mm256_transpose_16x16(in1);
-  mm256_transpose_16x16(&in1[16]);
+  mm256_transpose_16x16(in0, in0);
+  mm256_transpose_16x16(&in0[16], &in0[16]);
+  mm256_transpose_16x16(in1, in1);
+  mm256_transpose_16x16(&in1[16], &in1[16]);
   mm256_vectors_swap(&in0[16], in1, 16);
 }
 
@@ -1247,23 +1249,23 @@ static void fdct16_odd_avx2(__m256i *in) {
 
   u0 = _mm256_unpacklo_epi16(in[4], in[11]);
   u1 = _mm256_unpackhi_epi16(in[4], in[11]);
-  y4 = butter_fly(u0, u1, cospi_m16_p16);
-  y11 = butter_fly(u0, u1, cospi_p16_p16);
+  y4 = butter_fly(&u0, &u1, &cospi_m16_p16);
+  y11 = butter_fly(&u0, &u1, &cospi_p16_p16);
 
   u0 = _mm256_unpacklo_epi16(in[5], in[10]);
   u1 = _mm256_unpackhi_epi16(in[5], in[10]);
-  y5 = butter_fly(u0, u1, cospi_m16_p16);
-  y10 = butter_fly(u0, u1, cospi_p16_p16);
+  y5 = butter_fly(&u0, &u1, &cospi_m16_p16);
+  y10 = butter_fly(&u0, &u1, &cospi_p16_p16);
 
   u0 = _mm256_unpacklo_epi16(in[6], in[9]);
   u1 = _mm256_unpackhi_epi16(in[6], in[9]);
-  y6 = butter_fly(u0, u1, cospi_m16_p16);
-  y9 = butter_fly(u0, u1, cospi_p16_p16);
+  y6 = butter_fly(&u0, &u1, &cospi_m16_p16);
+  y9 = butter_fly(&u0, &u1, &cospi_p16_p16);
 
   u0 = _mm256_unpacklo_epi16(in[7], in[8]);
   u1 = _mm256_unpackhi_epi16(in[7], in[8]);
-  y7 = butter_fly(u0, u1, cospi_m16_p16);
-  y8 = butter_fly(u0, u1, cospi_p16_p16);
+  y7 = butter_fly(&u0, &u1, &cospi_m16_p16);
+  y8 = butter_fly(&u0, &u1, &cospi_p16_p16);
 
   y12 = in[12];
   y13 = in[13];
@@ -1300,23 +1302,23 @@ static void fdct16_odd_avx2(__m256i *in) {
 
   u0 = _mm256_unpacklo_epi16(x2, x13);
   u1 = _mm256_unpackhi_epi16(x2, x13);
-  y2 = butter_fly(u0, u1, cospi_m08_p24);
-  y13 = butter_fly(u0, u1, cospi_p24_p08);
+  y2 = butter_fly(&u0, &u1, &cospi_m08_p24);
+  y13 = butter_fly(&u0, &u1, &cospi_p24_p08);
 
   u0 = _mm256_unpacklo_epi16(x3, x12);
   u1 = _mm256_unpackhi_epi16(x3, x12);
-  y3 = butter_fly(u0, u1, cospi_m08_p24);
-  y12 = butter_fly(u0, u1, cospi_p24_p08);
+  y3 = butter_fly(&u0, &u1, &cospi_m08_p24);
+  y12 = butter_fly(&u0, &u1, &cospi_p24_p08);
 
   u0 = _mm256_unpacklo_epi16(x4, x11);
   u1 = _mm256_unpackhi_epi16(x4, x11);
-  y4 = butter_fly(u0, u1, cospi_m24_m08);
-  y11 = butter_fly(u0, u1, cospi_m08_p24);
+  y4 = butter_fly(&u0, &u1, &cospi_m24_m08);
+  y11 = butter_fly(&u0, &u1, &cospi_m08_p24);
 
   u0 = _mm256_unpacklo_epi16(x5, x10);
   u1 = _mm256_unpackhi_epi16(x5, x10);
-  y5 = butter_fly(u0, u1, cospi_m24_m08);
-  y10 = butter_fly(u0, u1, cospi_m08_p24);
+  y5 = butter_fly(&u0, &u1, &cospi_m24_m08);
+  y10 = butter_fly(&u0, &u1, &cospi_m08_p24);
 
   // stage 5
   x0 = _mm256_add_epi16(y0, y3);
@@ -1349,23 +1351,23 @@ static void fdct16_odd_avx2(__m256i *in) {
 
   u0 = _mm256_unpacklo_epi16(x1, x14);
   u1 = _mm256_unpackhi_epi16(x1, x14);
-  y1 = butter_fly(u0, u1, cospi_m04_p28);
-  y14 = butter_fly(u0, u1, cospi_p28_p04);
+  y1 = butter_fly(&u0, &u1, &cospi_m04_p28);
+  y14 = butter_fly(&u0, &u1, &cospi_p28_p04);
 
   u0 = _mm256_unpacklo_epi16(x2, x13);
   u1 = _mm256_unpackhi_epi16(x2, x13);
-  y2 = butter_fly(u0, u1, cospi_m28_m04);
-  y13 = butter_fly(u0, u1, cospi_m04_p28);
+  y2 = butter_fly(&u0, &u1, &cospi_m28_m04);
+  y13 = butter_fly(&u0, &u1, &cospi_m04_p28);
 
   u0 = _mm256_unpacklo_epi16(x5, x10);
   u1 = _mm256_unpackhi_epi16(x5, x10);
-  y5 = butter_fly(u0, u1, cospi_m20_p12);
-  y10 = butter_fly(u0, u1, cospi_p12_p20);
+  y5 = butter_fly(&u0, &u1, &cospi_m20_p12);
+  y10 = butter_fly(&u0, &u1, &cospi_p12_p20);
 
   u0 = _mm256_unpacklo_epi16(x6, x9);
   u1 = _mm256_unpackhi_epi16(x6, x9);
-  y6 = butter_fly(u0, u1, cospi_m12_m20);
-  y9 = butter_fly(u0, u1, cospi_m20_p12);
+  y6 = butter_fly(&u0, &u1, &cospi_m12_m20);
+  y9 = butter_fly(&u0, &u1, &cospi_m20_p12);
 
   // stage 7
   x0 = _mm256_add_epi16(y0, y1);
@@ -1389,43 +1391,43 @@ static void fdct16_odd_avx2(__m256i *in) {
   // stage 8
   u0 = _mm256_unpacklo_epi16(x0, x15);
   u1 = _mm256_unpackhi_epi16(x0, x15);
-  in[0] = butter_fly(u0, u1, cospi_p31_p01);
-  in[15] = butter_fly(u0, u1, cospi_m01_p31);
+  in[0] = butter_fly(&u0, &u1, &cospi_p31_p01);
+  in[15] = butter_fly(&u0, &u1, &cospi_m01_p31);
 
   u0 = _mm256_unpacklo_epi16(x1, x14);
   u1 = _mm256_unpackhi_epi16(x1, x14);
-  in[1] = butter_fly(u0, u1, cospi_p15_p17);
-  in[14] = butter_fly(u0, u1, cospi_m17_p15);
+  in[1] = butter_fly(&u0, &u1, &cospi_p15_p17);
+  in[14] = butter_fly(&u0, &u1, &cospi_m17_p15);
 
   u0 = _mm256_unpacklo_epi16(x2, x13);
   u1 = _mm256_unpackhi_epi16(x2, x13);
-  in[2] = butter_fly(u0, u1, cospi_p23_p09);
-  in[13] = butter_fly(u0, u1, cospi_m09_p23);
+  in[2] = butter_fly(&u0, &u1, &cospi_p23_p09);
+  in[13] = butter_fly(&u0, &u1, &cospi_m09_p23);
 
   u0 = _mm256_unpacklo_epi16(x3, x12);
   u1 = _mm256_unpackhi_epi16(x3, x12);
-  in[3] = butter_fly(u0, u1, cospi_p07_p25);
-  in[12] = butter_fly(u0, u1, cospi_m25_p07);
+  in[3] = butter_fly(&u0, &u1, &cospi_p07_p25);
+  in[12] = butter_fly(&u0, &u1, &cospi_m25_p07);
 
   u0 = _mm256_unpacklo_epi16(x4, x11);
   u1 = _mm256_unpackhi_epi16(x4, x11);
-  in[4] = butter_fly(u0, u1, cospi_p27_p05);
-  in[11] = butter_fly(u0, u1, cospi_m05_p27);
+  in[4] = butter_fly(&u0, &u1, &cospi_p27_p05);
+  in[11] = butter_fly(&u0, &u1, &cospi_m05_p27);
 
   u0 = _mm256_unpacklo_epi16(x5, x10);
   u1 = _mm256_unpackhi_epi16(x5, x10);
-  in[5] = butter_fly(u0, u1, cospi_p11_p21);
-  in[10] = butter_fly(u0, u1, cospi_m21_p11);
+  in[5] = butter_fly(&u0, &u1, &cospi_p11_p21);
+  in[10] = butter_fly(&u0, &u1, &cospi_m21_p11);
 
   u0 = _mm256_unpacklo_epi16(x6, x9);
   u1 = _mm256_unpackhi_epi16(x6, x9);
-  in[6] = butter_fly(u0, u1, cospi_p19_p13);
-  in[9] = butter_fly(u0, u1, cospi_m13_p19);
+  in[6] = butter_fly(&u0, &u1, &cospi_p19_p13);
+  in[9] = butter_fly(&u0, &u1, &cospi_m13_p19);
 
   u0 = _mm256_unpacklo_epi16(x7, x8);
   u1 = _mm256_unpackhi_epi16(x7, x8);
-  in[7] = butter_fly(u0, u1, cospi_p03_p29);
-  in[8] = butter_fly(u0, u1, cospi_m29_p03);
+  in[7] = butter_fly(&u0, &u1, &cospi_p03_p29);
+  in[8] = butter_fly(&u0, &u1, &cospi_m29_p03);
 }
 
 static void fdct32_avx2(__m256i *in0, __m256i *in1) {
@@ -1464,7 +1466,7 @@ static INLINE void write_buffer_32x32(const __m256i *in0, const __m256i *in1,
 static void fhalfright32_16col_avx2(__m256i *in) {
   int i = 0;
   const __m256i zero = _mm256_setzero_si256();
-  const __m256i sqrt2 = _mm256_set1_epi16(Sqrt2);
+  const __m256i sqrt2 = _mm256_set1_epi16((int16_t)Sqrt2);
   const __m256i dct_rounding = _mm256_set1_epi32(DCT_CONST_ROUNDING);
   __m256i x0, x1;