shithub: libvpx

Download patch

ref: 6248f0c91fdfcbd88323030b4abe55e2f6c31025
parent: 3b85a5beb7fedba88540d74fd7fe8f89b2f578cb
author: Linfeng Zhang <linfengz@google.com>
date: Tue Jan 23 11:03:50 EST 2018

cosmetic: clean idct neon functions

Change-Id: I9c7c52567850aded0437b13ba1260e94441bc49d

--- a/vpx_dsp/arm/highbd_idct32x32_1024_add_neon.c
+++ b/vpx_dsp/arm/highbd_idct32x32_1024_add_neon.c
@@ -124,83 +124,77 @@
                                vrshrn_n_s64(q[3].val[1], DCT_CONST_BITS));
 }
 
-static INLINE void load_s32x4q_dual(
-    const int32_t *in, int32x4x2_t *const s0, int32x4x2_t *const s1,
-    int32x4x2_t *const s2, int32x4x2_t *const s3, int32x4x2_t *const s4,
-    int32x4x2_t *const s5, int32x4x2_t *const s6, int32x4x2_t *const s7) {
-  s0->val[0] = vld1q_s32(in);
-  s0->val[1] = vld1q_s32(in + 4);
+static INLINE void load_s32x4q_dual(const int32_t *in, int32x4x2_t *const s) {
+  s[0].val[0] = vld1q_s32(in);
+  s[0].val[1] = vld1q_s32(in + 4);
   in += 32;
-  s1->val[0] = vld1q_s32(in);
-  s1->val[1] = vld1q_s32(in + 4);
+  s[1].val[0] = vld1q_s32(in);
+  s[1].val[1] = vld1q_s32(in + 4);
   in += 32;
-  s2->val[0] = vld1q_s32(in);
-  s2->val[1] = vld1q_s32(in + 4);
+  s[2].val[0] = vld1q_s32(in);
+  s[2].val[1] = vld1q_s32(in + 4);
   in += 32;
-  s3->val[0] = vld1q_s32(in);
-  s3->val[1] = vld1q_s32(in + 4);
+  s[3].val[0] = vld1q_s32(in);
+  s[3].val[1] = vld1q_s32(in + 4);
   in += 32;
-  s4->val[0] = vld1q_s32(in);
-  s4->val[1] = vld1q_s32(in + 4);
+  s[4].val[0] = vld1q_s32(in);
+  s[4].val[1] = vld1q_s32(in + 4);
   in += 32;
-  s5->val[0] = vld1q_s32(in);
-  s5->val[1] = vld1q_s32(in + 4);
+  s[5].val[0] = vld1q_s32(in);
+  s[5].val[1] = vld1q_s32(in + 4);
   in += 32;
-  s6->val[0] = vld1q_s32(in);
-  s6->val[1] = vld1q_s32(in + 4);
+  s[6].val[0] = vld1q_s32(in);
+  s[6].val[1] = vld1q_s32(in + 4);
   in += 32;
-  s7->val[0] = vld1q_s32(in);
-  s7->val[1] = vld1q_s32(in + 4);
+  s[7].val[0] = vld1q_s32(in);
+  s[7].val[1] = vld1q_s32(in + 4);
 }
 
-static INLINE void transpose_and_store_s32_8x8(int32x4x2_t a0, int32x4x2_t a1,
-                                               int32x4x2_t a2, int32x4x2_t a3,
-                                               int32x4x2_t a4, int32x4x2_t a5,
-                                               int32x4x2_t a6, int32x4x2_t a7,
+static INLINE void transpose_and_store_s32_8x8(int32x4x2_t *const a,
                                                int32_t **out) {
-  transpose_s32_8x8(&a0, &a1, &a2, &a3, &a4, &a5, &a6, &a7);
+  transpose_s32_8x8(&a[0], &a[1], &a[2], &a[3], &a[4], &a[5], &a[6], &a[7]);
 
-  vst1q_s32(*out, a0.val[0]);
+  vst1q_s32(*out, a[0].val[0]);
   *out += 4;
-  vst1q_s32(*out, a0.val[1]);
+  vst1q_s32(*out, a[0].val[1]);
   *out += 4;
-  vst1q_s32(*out, a1.val[0]);
+  vst1q_s32(*out, a[1].val[0]);
   *out += 4;
-  vst1q_s32(*out, a1.val[1]);
+  vst1q_s32(*out, a[1].val[1]);
   *out += 4;
-  vst1q_s32(*out, a2.val[0]);
+  vst1q_s32(*out, a[2].val[0]);
   *out += 4;
-  vst1q_s32(*out, a2.val[1]);
+  vst1q_s32(*out, a[2].val[1]);
   *out += 4;
-  vst1q_s32(*out, a3.val[0]);
+  vst1q_s32(*out, a[3].val[0]);
   *out += 4;
-  vst1q_s32(*out, a3.val[1]);
+  vst1q_s32(*out, a[3].val[1]);
   *out += 4;
-  vst1q_s32(*out, a4.val[0]);
+  vst1q_s32(*out, a[4].val[0]);
   *out += 4;
-  vst1q_s32(*out, a4.val[1]);
+  vst1q_s32(*out, a[4].val[1]);
   *out += 4;
-  vst1q_s32(*out, a5.val[0]);
+  vst1q_s32(*out, a[5].val[0]);
   *out += 4;
-  vst1q_s32(*out, a5.val[1]);
+  vst1q_s32(*out, a[5].val[1]);
   *out += 4;
-  vst1q_s32(*out, a6.val[0]);
+  vst1q_s32(*out, a[6].val[0]);
   *out += 4;
-  vst1q_s32(*out, a6.val[1]);
+  vst1q_s32(*out, a[6].val[1]);
   *out += 4;
-  vst1q_s32(*out, a7.val[0]);
+  vst1q_s32(*out, a[7].val[0]);
   *out += 4;
-  vst1q_s32(*out, a7.val[1]);
+  vst1q_s32(*out, a[7].val[1]);
   *out += 4;
 }
 
 static INLINE void idct32_transpose_pair(const int32_t *input, int32_t *t_buf) {
   int i;
-  int32x4x2_t s0, s1, s2, s3, s4, s5, s6, s7;
+  int32x4x2_t s[8];
 
   for (i = 0; i < 4; i++, input += 8) {
-    load_s32x4q_dual(input, &s0, &s1, &s2, &s3, &s4, &s5, &s6, &s7);
-    transpose_and_store_s32_8x8(s0, s1, s2, s3, s4, s5, s6, s7, &t_buf);
+    load_s32x4q_dual(input, s);
+    transpose_and_store_s32_8x8(s, &t_buf);
   }
 }
 
--- a/vpx_dsp/arm/highbd_idct4x4_add_neon.c
+++ b/vpx_dsp/arm/highbd_idct4x4_add_neon.c
@@ -66,108 +66,104 @@
 }
 
 static INLINE void idct4x4_16_kernel_bd10(const int32x4_t cospis,
-                                          int32x4_t *const a0,
-                                          int32x4_t *const a1,
-                                          int32x4_t *const a2,
-                                          int32x4_t *const a3) {
+                                          int32x4_t *const a) {
   int32x4_t b0, b1, b2, b3;
 
-  transpose_s32_4x4(a0, a1, a2, a3);
-  b0 = vaddq_s32(*a0, *a2);
-  b1 = vsubq_s32(*a0, *a2);
+  transpose_s32_4x4(&a[0], &a[1], &a[2], &a[3]);
+  b0 = vaddq_s32(a[0], a[2]);
+  b1 = vsubq_s32(a[0], a[2]);
   b0 = vmulq_lane_s32(b0, vget_high_s32(cospis), 0);
   b1 = vmulq_lane_s32(b1, vget_high_s32(cospis), 0);
-  b2 = vmulq_lane_s32(*a1, vget_high_s32(cospis), 1);
-  b3 = vmulq_lane_s32(*a1, vget_low_s32(cospis), 1);
-  b2 = vmlsq_lane_s32(b2, *a3, vget_low_s32(cospis), 1);
-  b3 = vmlaq_lane_s32(b3, *a3, vget_high_s32(cospis), 1);
+  b2 = vmulq_lane_s32(a[1], vget_high_s32(cospis), 1);
+  b3 = vmulq_lane_s32(a[1], vget_low_s32(cospis), 1);
+  b2 = vmlsq_lane_s32(b2, a[3], vget_low_s32(cospis), 1);
+  b3 = vmlaq_lane_s32(b3, a[3], vget_high_s32(cospis), 1);
   b0 = vrshrq_n_s32(b0, DCT_CONST_BITS);
   b1 = vrshrq_n_s32(b1, DCT_CONST_BITS);
   b2 = vrshrq_n_s32(b2, DCT_CONST_BITS);
   b3 = vrshrq_n_s32(b3, DCT_CONST_BITS);
-  *a0 = vaddq_s32(b0, b3);
-  *a1 = vaddq_s32(b1, b2);
-  *a2 = vsubq_s32(b1, b2);
-  *a3 = vsubq_s32(b0, b3);
+  a[0] = vaddq_s32(b0, b3);
+  a[1] = vaddq_s32(b1, b2);
+  a[2] = vsubq_s32(b1, b2);
+  a[3] = vsubq_s32(b0, b3);
 }
 
 static INLINE void idct4x4_16_kernel_bd12(const int32x4_t cospis,
-                                          int32x4_t *const a0,
-                                          int32x4_t *const a1,
-                                          int32x4_t *const a2,
-                                          int32x4_t *const a3) {
+                                          int32x4_t *const a) {
   int32x4_t b0, b1, b2, b3;
-  int64x2_t c0, c1, c2, c3, c4, c5, c6, c7, c8, c9, c10, c11;
+  int64x2_t c[12];
 
-  transpose_s32_4x4(a0, a1, a2, a3);
-  b0 = vaddq_s32(*a0, *a2);
-  b1 = vsubq_s32(*a0, *a2);
-  c0 = vmull_lane_s32(vget_low_s32(b0), vget_high_s32(cospis), 0);
-  c1 = vmull_lane_s32(vget_high_s32(b0), vget_high_s32(cospis), 0);
-  c2 = vmull_lane_s32(vget_low_s32(b1), vget_high_s32(cospis), 0);
-  c3 = vmull_lane_s32(vget_high_s32(b1), vget_high_s32(cospis), 0);
-  c4 = vmull_lane_s32(vget_low_s32(*a1), vget_high_s32(cospis), 1);
-  c5 = vmull_lane_s32(vget_high_s32(*a1), vget_high_s32(cospis), 1);
-  c6 = vmull_lane_s32(vget_low_s32(*a1), vget_low_s32(cospis), 1);
-  c7 = vmull_lane_s32(vget_high_s32(*a1), vget_low_s32(cospis), 1);
-  c8 = vmull_lane_s32(vget_low_s32(*a3), vget_low_s32(cospis), 1);
-  c9 = vmull_lane_s32(vget_high_s32(*a3), vget_low_s32(cospis), 1);
-  c10 = vmull_lane_s32(vget_low_s32(*a3), vget_high_s32(cospis), 1);
-  c11 = vmull_lane_s32(vget_high_s32(*a3), vget_high_s32(cospis), 1);
-  c4 = vsubq_s64(c4, c8);
-  c5 = vsubq_s64(c5, c9);
-  c6 = vaddq_s64(c6, c10);
-  c7 = vaddq_s64(c7, c11);
-  b0 = vcombine_s32(vrshrn_n_s64(c0, DCT_CONST_BITS),
-                    vrshrn_n_s64(c1, DCT_CONST_BITS));
-  b1 = vcombine_s32(vrshrn_n_s64(c2, DCT_CONST_BITS),
-                    vrshrn_n_s64(c3, DCT_CONST_BITS));
-  b2 = vcombine_s32(vrshrn_n_s64(c4, DCT_CONST_BITS),
-                    vrshrn_n_s64(c5, DCT_CONST_BITS));
-  b3 = vcombine_s32(vrshrn_n_s64(c6, DCT_CONST_BITS),
-                    vrshrn_n_s64(c7, DCT_CONST_BITS));
-  *a0 = vaddq_s32(b0, b3);
-  *a1 = vaddq_s32(b1, b2);
-  *a2 = vsubq_s32(b1, b2);
-  *a3 = vsubq_s32(b0, b3);
+  transpose_s32_4x4(&a[0], &a[1], &a[2], &a[3]);
+  b0 = vaddq_s32(a[0], a[2]);
+  b1 = vsubq_s32(a[0], a[2]);
+  c[0] = vmull_lane_s32(vget_low_s32(b0), vget_high_s32(cospis), 0);
+  c[1] = vmull_lane_s32(vget_high_s32(b0), vget_high_s32(cospis), 0);
+  c[2] = vmull_lane_s32(vget_low_s32(b1), vget_high_s32(cospis), 0);
+  c[3] = vmull_lane_s32(vget_high_s32(b1), vget_high_s32(cospis), 0);
+  c[4] = vmull_lane_s32(vget_low_s32(a[1]), vget_high_s32(cospis), 1);
+  c[5] = vmull_lane_s32(vget_high_s32(a[1]), vget_high_s32(cospis), 1);
+  c[6] = vmull_lane_s32(vget_low_s32(a[1]), vget_low_s32(cospis), 1);
+  c[7] = vmull_lane_s32(vget_high_s32(a[1]), vget_low_s32(cospis), 1);
+  c[8] = vmull_lane_s32(vget_low_s32(a[3]), vget_low_s32(cospis), 1);
+  c[9] = vmull_lane_s32(vget_high_s32(a[3]), vget_low_s32(cospis), 1);
+  c[10] = vmull_lane_s32(vget_low_s32(a[3]), vget_high_s32(cospis), 1);
+  c[11] = vmull_lane_s32(vget_high_s32(a[3]), vget_high_s32(cospis), 1);
+  c[4] = vsubq_s64(c[4], c[8]);
+  c[5] = vsubq_s64(c[5], c[9]);
+  c[6] = vaddq_s64(c[6], c[10]);
+  c[7] = vaddq_s64(c[7], c[11]);
+  b0 = vcombine_s32(vrshrn_n_s64(c[0], DCT_CONST_BITS),
+                    vrshrn_n_s64(c[1], DCT_CONST_BITS));
+  b1 = vcombine_s32(vrshrn_n_s64(c[2], DCT_CONST_BITS),
+                    vrshrn_n_s64(c[3], DCT_CONST_BITS));
+  b2 = vcombine_s32(vrshrn_n_s64(c[4], DCT_CONST_BITS),
+                    vrshrn_n_s64(c[5], DCT_CONST_BITS));
+  b3 = vcombine_s32(vrshrn_n_s64(c[6], DCT_CONST_BITS),
+                    vrshrn_n_s64(c[7], DCT_CONST_BITS));
+  a[0] = vaddq_s32(b0, b3);
+  a[1] = vaddq_s32(b1, b2);
+  a[2] = vsubq_s32(b1, b2);
+  a[3] = vsubq_s32(b0, b3);
 }
 
 void vpx_highbd_idct4x4_16_add_neon(const tran_low_t *input, uint16_t *dest,
                                     int stride, int bd) {
   const int16x8_t max = vdupq_n_s16((1 << bd) - 1);
-  int32x4_t c0 = vld1q_s32(input);
-  int32x4_t c1 = vld1q_s32(input + 4);
-  int32x4_t c2 = vld1q_s32(input + 8);
-  int32x4_t c3 = vld1q_s32(input + 12);
-  int16x8_t a0, a1;
+  int16x8_t a[2];
+  int32x4_t c[4];
 
+  c[0] = vld1q_s32(input);
+  c[1] = vld1q_s32(input + 4);
+  c[2] = vld1q_s32(input + 8);
+  c[3] = vld1q_s32(input + 12);
+
   if (bd == 8) {
     const int16x4_t cospis = vld1_s16(kCospi);
 
     // Rows
-    a0 = vcombine_s16(vmovn_s32(c0), vmovn_s32(c1));
-    a1 = vcombine_s16(vmovn_s32(c2), vmovn_s32(c3));
-    idct4x4_16_kernel_bd8(cospis, &a0, &a1);
+    a[0] = vcombine_s16(vmovn_s32(c[0]), vmovn_s32(c[1]));
+    a[1] = vcombine_s16(vmovn_s32(c[2]), vmovn_s32(c[3]));
+    idct4x4_16_kernel_bd8(cospis, a);
 
     // Columns
-    a1 = vcombine_s16(vget_high_s16(a1), vget_low_s16(a1));
-    idct4x4_16_kernel_bd8(cospis, &a0, &a1);
-    a0 = vrshrq_n_s16(a0, 4);
-    a1 = vrshrq_n_s16(a1, 4);
+    a[1] = vcombine_s16(vget_high_s16(a[1]), vget_low_s16(a[1]));
+    idct4x4_16_kernel_bd8(cospis, a);
+    a[0] = vrshrq_n_s16(a[0], 4);
+    a[1] = vrshrq_n_s16(a[1], 4);
   } else {
     const int32x4_t cospis = vld1q_s32(kCospi32);
 
     if (bd == 10) {
-      idct4x4_16_kernel_bd10(cospis, &c0, &c1, &c2, &c3);
-      idct4x4_16_kernel_bd10(cospis, &c0, &c1, &c2, &c3);
+      idct4x4_16_kernel_bd10(cospis, c);
+      idct4x4_16_kernel_bd10(cospis, c);
     } else {
-      idct4x4_16_kernel_bd12(cospis, &c0, &c1, &c2, &c3);
-      idct4x4_16_kernel_bd12(cospis, &c0, &c1, &c2, &c3);
+      idct4x4_16_kernel_bd12(cospis, c);
+      idct4x4_16_kernel_bd12(cospis, c);
     }
-    a0 = vcombine_s16(vqrshrn_n_s32(c0, 4), vqrshrn_n_s32(c1, 4));
-    a1 = vcombine_s16(vqrshrn_n_s32(c3, 4), vqrshrn_n_s32(c2, 4));
+    a[0] = vcombine_s16(vqrshrn_n_s32(c[0], 4), vqrshrn_n_s32(c[1], 4));
+    a[1] = vcombine_s16(vqrshrn_n_s32(c[3], 4), vqrshrn_n_s32(c[2], 4));
   }
 
-  highbd_idct4x4_1_add_kernel1(&dest, stride, a0, max);
-  highbd_idct4x4_1_add_kernel2(&dest, stride, a1, max);
+  highbd_idct4x4_1_add_kernel1(&dest, stride, a[0], max);
+  highbd_idct4x4_1_add_kernel2(&dest, stride, a[1], max);
 }
--- a/vpx_dsp/arm/highbd_idct8x8_add_neon.c
+++ b/vpx_dsp/arm/highbd_idct8x8_add_neon.c
@@ -127,7 +127,7 @@
     int32x4_t *const io1, int32x4_t *const io2, int32x4_t *const io3,
     int32x4_t *const io4, int32x4_t *const io5, int32x4_t *const io6,
     int32x4_t *const io7) {
-  int32x2_t input_1l, input_1h, input_3l, input_3h;
+  int32x2_t input1l, input1h, input3l, input3h;
   int32x2_t step1l[2], step1h[2];
   int32x4_t step1[8], step2[8];
   int64x2_t t64[8];
@@ -136,23 +136,23 @@
   transpose_s32_4x4(io0, io1, io2, io3);
 
   // stage 1
-  input_1l = vget_low_s32(*io1);
-  input_1h = vget_high_s32(*io1);
-  input_3l = vget_low_s32(*io3);
-  input_3h = vget_high_s32(*io3);
+  input1l = vget_low_s32(*io1);
+  input1h = vget_high_s32(*io1);
+  input3l = vget_low_s32(*io3);
+  input3h = vget_high_s32(*io3);
   step1l[0] = vget_low_s32(*io0);
   step1h[0] = vget_high_s32(*io0);
   step1l[1] = vget_low_s32(*io2);
   step1h[1] = vget_high_s32(*io2);
 
-  t64[0] = vmull_lane_s32(input_1l, vget_high_s32(cospis1), 1);
-  t64[1] = vmull_lane_s32(input_1h, vget_high_s32(cospis1), 1);
-  t64[2] = vmull_lane_s32(input_3l, vget_high_s32(cospis1), 0);
-  t64[3] = vmull_lane_s32(input_3h, vget_high_s32(cospis1), 0);
-  t64[4] = vmull_lane_s32(input_3l, vget_low_s32(cospis1), 1);
-  t64[5] = vmull_lane_s32(input_3h, vget_low_s32(cospis1), 1);
-  t64[6] = vmull_lane_s32(input_1l, vget_low_s32(cospis1), 0);
-  t64[7] = vmull_lane_s32(input_1h, vget_low_s32(cospis1), 0);
+  t64[0] = vmull_lane_s32(input1l, vget_high_s32(cospis1), 1);
+  t64[1] = vmull_lane_s32(input1h, vget_high_s32(cospis1), 1);
+  t64[2] = vmull_lane_s32(input3l, vget_high_s32(cospis1), 0);
+  t64[3] = vmull_lane_s32(input3h, vget_high_s32(cospis1), 0);
+  t64[4] = vmull_lane_s32(input3l, vget_low_s32(cospis1), 1);
+  t64[5] = vmull_lane_s32(input3h, vget_low_s32(cospis1), 1);
+  t64[6] = vmull_lane_s32(input1l, vget_low_s32(cospis1), 0);
+  t64[7] = vmull_lane_s32(input1h, vget_low_s32(cospis1), 0);
   t32[0] = vrshrn_n_s64(t64[0], DCT_CONST_BITS);
   t32[1] = vrshrn_n_s64(t64[1], DCT_CONST_BITS);
   t32[2] = vrshrn_n_s64(t64[2], DCT_CONST_BITS);
@@ -222,9 +222,7 @@
   *io7 = vsubq_s32(step1[0], step2[7]);
 }
 
-static INLINE void highbd_add8x8(int16x8_t a0, int16x8_t a1, int16x8_t a2,
-                                 int16x8_t a3, int16x8_t a4, int16x8_t a5,
-                                 int16x8_t a6, int16x8_t a7, uint16_t *dest,
+static INLINE void highbd_add8x8(int16x8_t *const a, uint16_t *dest,
                                  const int stride, const int bd) {
   const int16x8_t max = vdupq_n_s16((1 << bd) - 1);
   const uint16_t *dst = dest;
@@ -248,14 +246,14 @@
   dst += stride;
   d7 = vld1q_u16(dst);
 
-  d0_s16 = vqaddq_s16(a0, vreinterpretq_s16_u16(d0));
-  d1_s16 = vqaddq_s16(a1, vreinterpretq_s16_u16(d1));
-  d2_s16 = vqaddq_s16(a2, vreinterpretq_s16_u16(d2));
-  d3_s16 = vqaddq_s16(a3, vreinterpretq_s16_u16(d3));
-  d4_s16 = vqaddq_s16(a4, vreinterpretq_s16_u16(d4));
-  d5_s16 = vqaddq_s16(a5, vreinterpretq_s16_u16(d5));
-  d6_s16 = vqaddq_s16(a6, vreinterpretq_s16_u16(d6));
-  d7_s16 = vqaddq_s16(a7, vreinterpretq_s16_u16(d7));
+  d0_s16 = vqaddq_s16(a[0], vreinterpretq_s16_u16(d0));
+  d1_s16 = vqaddq_s16(a[1], vreinterpretq_s16_u16(d1));
+  d2_s16 = vqaddq_s16(a[2], vreinterpretq_s16_u16(d2));
+  d3_s16 = vqaddq_s16(a[3], vreinterpretq_s16_u16(d3));
+  d4_s16 = vqaddq_s16(a[4], vreinterpretq_s16_u16(d4));
+  d5_s16 = vqaddq_s16(a[5], vreinterpretq_s16_u16(d5));
+  d6_s16 = vqaddq_s16(a[6], vreinterpretq_s16_u16(d6));
+  d7_s16 = vqaddq_s16(a[7], vreinterpretq_s16_u16(d7));
 
   d0_s16 = vminq_s16(d0_s16, max);
   d1_s16 = vminq_s16(d1_s16, max);
@@ -293,12 +291,14 @@
 
 void vpx_highbd_idct8x8_12_add_neon(const tran_low_t *input, uint16_t *dest,
                                     int stride, int bd) {
-  int32x4_t a0 = vld1q_s32(input);
-  int32x4_t a1 = vld1q_s32(input + 8);
-  int32x4_t a2 = vld1q_s32(input + 16);
-  int32x4_t a3 = vld1q_s32(input + 24);
-  int16x8_t c0, c1, c2, c3, c4, c5, c6, c7;
+  int32x4_t a[16];
+  int16x8_t c[8];
 
+  a[0] = vld1q_s32(input);
+  a[1] = vld1q_s32(input + 8);
+  a[2] = vld1q_s32(input + 16);
+  a[3] = vld1q_s32(input + 24);
+
   if (bd == 8) {
     const int16x8_t cospis = vld1q_s16(kCospi);
     const int16x8_t cospisd = vaddq_s16(cospis, cospis);
@@ -305,54 +305,52 @@
     const int16x4_t cospis0 = vget_low_s16(cospis);     // cospi 0, 8, 16, 24
     const int16x4_t cospisd0 = vget_low_s16(cospisd);   // doubled 0, 8, 16, 24
     const int16x4_t cospisd1 = vget_high_s16(cospisd);  // doubled 4, 12, 20, 28
-    int16x4_t b0 = vmovn_s32(a0);
-    int16x4_t b1 = vmovn_s32(a1);
-    int16x4_t b2 = vmovn_s32(a2);
-    int16x4_t b3 = vmovn_s32(a3);
-    int16x4_t b4, b5, b6, b7;
+    int16x4_t b[8];
 
-    idct8x8_12_pass1_bd8(cospis0, cospisd0, cospisd1, &b0, &b1, &b2, &b3, &b4,
-                         &b5, &b6, &b7);
-    idct8x8_12_pass2_bd8(cospis0, cospisd0, cospisd1, b0, b1, b2, b3, b4, b5,
-                         b6, b7, &c0, &c1, &c2, &c3, &c4, &c5, &c6, &c7);
-    c0 = vrshrq_n_s16(c0, 5);
-    c1 = vrshrq_n_s16(c1, 5);
-    c2 = vrshrq_n_s16(c2, 5);
-    c3 = vrshrq_n_s16(c3, 5);
-    c4 = vrshrq_n_s16(c4, 5);
-    c5 = vrshrq_n_s16(c5, 5);
-    c6 = vrshrq_n_s16(c6, 5);
-    c7 = vrshrq_n_s16(c7, 5);
+    b[0] = vmovn_s32(a[0]);
+    b[1] = vmovn_s32(a[1]);
+    b[2] = vmovn_s32(a[2]);
+    b[3] = vmovn_s32(a[3]);
+
+    idct8x8_12_pass1_bd8(cospis0, cospisd0, cospisd1, b);
+    idct8x8_12_pass2_bd8(cospis0, cospisd0, cospisd1, b, c);
+    c[0] = vrshrq_n_s16(c[0], 5);
+    c[1] = vrshrq_n_s16(c[1], 5);
+    c[2] = vrshrq_n_s16(c[2], 5);
+    c[3] = vrshrq_n_s16(c[3], 5);
+    c[4] = vrshrq_n_s16(c[4], 5);
+    c[5] = vrshrq_n_s16(c[5], 5);
+    c[6] = vrshrq_n_s16(c[6], 5);
+    c[7] = vrshrq_n_s16(c[7], 5);
   } else {
     const int32x4_t cospis0 = vld1q_s32(kCospi32);      // cospi 0, 8, 16, 24
     const int32x4_t cospis1 = vld1q_s32(kCospi32 + 4);  // cospi 4, 12, 20, 28
-    int32x4_t a4, a5, a6, a7, a8, a9, a10, a11, a12, a13, a14, a15;
 
     if (bd == 10) {
-      idct8x8_12_half1d_bd10(cospis0, cospis1, &a0, &a1, &a2, &a3, &a4, &a5,
-                             &a6, &a7);
-      idct8x8_12_half1d_bd10(cospis0, cospis1, &a0, &a1, &a2, &a3, &a8, &a9,
-                             &a10, &a11);
-      idct8x8_12_half1d_bd10(cospis0, cospis1, &a4, &a5, &a6, &a7, &a12, &a13,
-                             &a14, &a15);
+      idct8x8_12_half1d_bd10(cospis0, cospis1, &a[0], &a[1], &a[2], &a[3],
+                             &a[4], &a[5], &a[6], &a[7]);
+      idct8x8_12_half1d_bd10(cospis0, cospis1, &a[0], &a[1], &a[2], &a[3],
+                             &a[8], &a[9], &a[10], &a[11]);
+      idct8x8_12_half1d_bd10(cospis0, cospis1, &a[4], &a[5], &a[6], &a[7],
+                             &a[12], &a[13], &a[14], &a[15]);
     } else {
-      idct8x8_12_half1d_bd12(cospis0, cospis1, &a0, &a1, &a2, &a3, &a4, &a5,
-                             &a6, &a7);
-      idct8x8_12_half1d_bd12(cospis0, cospis1, &a0, &a1, &a2, &a3, &a8, &a9,
-                             &a10, &a11);
-      idct8x8_12_half1d_bd12(cospis0, cospis1, &a4, &a5, &a6, &a7, &a12, &a13,
-                             &a14, &a15);
+      idct8x8_12_half1d_bd12(cospis0, cospis1, &a[0], &a[1], &a[2], &a[3],
+                             &a[4], &a[5], &a[6], &a[7]);
+      idct8x8_12_half1d_bd12(cospis0, cospis1, &a[0], &a[1], &a[2], &a[3],
+                             &a[8], &a[9], &a[10], &a[11]);
+      idct8x8_12_half1d_bd12(cospis0, cospis1, &a[4], &a[5], &a[6], &a[7],
+                             &a[12], &a[13], &a[14], &a[15]);
     }
-    c0 = vcombine_s16(vrshrn_n_s32(a0, 5), vrshrn_n_s32(a4, 5));
-    c1 = vcombine_s16(vrshrn_n_s32(a1, 5), vrshrn_n_s32(a5, 5));
-    c2 = vcombine_s16(vrshrn_n_s32(a2, 5), vrshrn_n_s32(a6, 5));
-    c3 = vcombine_s16(vrshrn_n_s32(a3, 5), vrshrn_n_s32(a7, 5));
-    c4 = vcombine_s16(vrshrn_n_s32(a8, 5), vrshrn_n_s32(a12, 5));
-    c5 = vcombine_s16(vrshrn_n_s32(a9, 5), vrshrn_n_s32(a13, 5));
-    c6 = vcombine_s16(vrshrn_n_s32(a10, 5), vrshrn_n_s32(a14, 5));
-    c7 = vcombine_s16(vrshrn_n_s32(a11, 5), vrshrn_n_s32(a15, 5));
+    c[0] = vcombine_s16(vrshrn_n_s32(a[0], 5), vrshrn_n_s32(a[4], 5));
+    c[1] = vcombine_s16(vrshrn_n_s32(a[1], 5), vrshrn_n_s32(a[5], 5));
+    c[2] = vcombine_s16(vrshrn_n_s32(a[2], 5), vrshrn_n_s32(a[6], 5));
+    c[3] = vcombine_s16(vrshrn_n_s32(a[3], 5), vrshrn_n_s32(a[7], 5));
+    c[4] = vcombine_s16(vrshrn_n_s32(a[8], 5), vrshrn_n_s32(a[12], 5));
+    c[5] = vcombine_s16(vrshrn_n_s32(a[9], 5), vrshrn_n_s32(a[13], 5));
+    c[6] = vcombine_s16(vrshrn_n_s32(a[10], 5), vrshrn_n_s32(a[14], 5));
+    c[7] = vcombine_s16(vrshrn_n_s32(a[11], 5), vrshrn_n_s32(a[15], 5));
   }
-  highbd_add8x8(c0, c1, c2, c3, c4, c5, c6, c7, dest, stride, bd);
+  highbd_add8x8(c, dest, stride, bd);
 }
 
 static INLINE void idct8x8_64_half1d_bd10(
@@ -428,8 +426,8 @@
     int32x4_t *const io1, int32x4_t *const io2, int32x4_t *const io3,
     int32x4_t *const io4, int32x4_t *const io5, int32x4_t *const io6,
     int32x4_t *const io7) {
-  int32x2_t input_1l, input_1h, input_3l, input_3h, input_5l, input_5h,
-      input_7l, input_7h;
+  int32x2_t input1l, input1h, input3l, input3h, input5l, input5h, input7l,
+      input7h;
   int32x2_t step1l[4], step1h[4];
   int32x4_t step1[8], step2[8];
   int64x2_t t64[8];
@@ -438,14 +436,14 @@
   transpose_s32_8x4(io0, io1, io2, io3, io4, io5, io6, io7);
 
   // stage 1
-  input_1l = vget_low_s32(*io1);
-  input_1h = vget_high_s32(*io1);
-  input_3l = vget_low_s32(*io3);
-  input_3h = vget_high_s32(*io3);
-  input_5l = vget_low_s32(*io5);
-  input_5h = vget_high_s32(*io5);
-  input_7l = vget_low_s32(*io7);
-  input_7h = vget_high_s32(*io7);
+  input1l = vget_low_s32(*io1);
+  input1h = vget_high_s32(*io1);
+  input3l = vget_low_s32(*io3);
+  input3h = vget_high_s32(*io3);
+  input5l = vget_low_s32(*io5);
+  input5h = vget_high_s32(*io5);
+  input7l = vget_low_s32(*io7);
+  input7h = vget_high_s32(*io7);
   step1l[0] = vget_low_s32(*io0);
   step1h[0] = vget_high_s32(*io0);
   step1l[1] = vget_low_s32(*io2);
@@ -455,22 +453,22 @@
   step1l[3] = vget_low_s32(*io6);
   step1h[3] = vget_high_s32(*io6);
 
-  t64[0] = vmull_lane_s32(input_1l, vget_high_s32(cospis1), 1);
-  t64[1] = vmull_lane_s32(input_1h, vget_high_s32(cospis1), 1);
-  t64[2] = vmull_lane_s32(input_3l, vget_high_s32(cospis1), 0);
-  t64[3] = vmull_lane_s32(input_3h, vget_high_s32(cospis1), 0);
-  t64[4] = vmull_lane_s32(input_3l, vget_low_s32(cospis1), 1);
-  t64[5] = vmull_lane_s32(input_3h, vget_low_s32(cospis1), 1);
-  t64[6] = vmull_lane_s32(input_1l, vget_low_s32(cospis1), 0);
-  t64[7] = vmull_lane_s32(input_1h, vget_low_s32(cospis1), 0);
-  t64[0] = vmlsl_lane_s32(t64[0], input_7l, vget_low_s32(cospis1), 0);
-  t64[1] = vmlsl_lane_s32(t64[1], input_7h, vget_low_s32(cospis1), 0);
-  t64[2] = vmlal_lane_s32(t64[2], input_5l, vget_low_s32(cospis1), 1);
-  t64[3] = vmlal_lane_s32(t64[3], input_5h, vget_low_s32(cospis1), 1);
-  t64[4] = vmlsl_lane_s32(t64[4], input_5l, vget_high_s32(cospis1), 0);
-  t64[5] = vmlsl_lane_s32(t64[5], input_5h, vget_high_s32(cospis1), 0);
-  t64[6] = vmlal_lane_s32(t64[6], input_7l, vget_high_s32(cospis1), 1);
-  t64[7] = vmlal_lane_s32(t64[7], input_7h, vget_high_s32(cospis1), 1);
+  t64[0] = vmull_lane_s32(input1l, vget_high_s32(cospis1), 1);
+  t64[1] = vmull_lane_s32(input1h, vget_high_s32(cospis1), 1);
+  t64[2] = vmull_lane_s32(input3l, vget_high_s32(cospis1), 0);
+  t64[3] = vmull_lane_s32(input3h, vget_high_s32(cospis1), 0);
+  t64[4] = vmull_lane_s32(input3l, vget_low_s32(cospis1), 1);
+  t64[5] = vmull_lane_s32(input3h, vget_low_s32(cospis1), 1);
+  t64[6] = vmull_lane_s32(input1l, vget_low_s32(cospis1), 0);
+  t64[7] = vmull_lane_s32(input1h, vget_low_s32(cospis1), 0);
+  t64[0] = vmlsl_lane_s32(t64[0], input7l, vget_low_s32(cospis1), 0);
+  t64[1] = vmlsl_lane_s32(t64[1], input7h, vget_low_s32(cospis1), 0);
+  t64[2] = vmlal_lane_s32(t64[2], input5l, vget_low_s32(cospis1), 1);
+  t64[3] = vmlal_lane_s32(t64[3], input5h, vget_low_s32(cospis1), 1);
+  t64[4] = vmlsl_lane_s32(t64[4], input5l, vget_high_s32(cospis1), 0);
+  t64[5] = vmlsl_lane_s32(t64[5], input5h, vget_high_s32(cospis1), 0);
+  t64[6] = vmlal_lane_s32(t64[6], input7l, vget_high_s32(cospis1), 1);
+  t64[7] = vmlal_lane_s32(t64[7], input7h, vget_high_s32(cospis1), 1);
   t32[0] = vrshrn_n_s64(t64[0], DCT_CONST_BITS);
   t32[1] = vrshrn_n_s64(t64[1], DCT_CONST_BITS);
   t32[2] = vrshrn_n_s64(t64[2], DCT_CONST_BITS);
@@ -553,79 +551,83 @@
 
 void vpx_highbd_idct8x8_64_add_neon(const tran_low_t *input, uint16_t *dest,
                                     int stride, int bd) {
-  int32x4_t a0 = vld1q_s32(input);
-  int32x4_t a1 = vld1q_s32(input + 4);
-  int32x4_t a2 = vld1q_s32(input + 8);
-  int32x4_t a3 = vld1q_s32(input + 12);
-  int32x4_t a4 = vld1q_s32(input + 16);
-  int32x4_t a5 = vld1q_s32(input + 20);
-  int32x4_t a6 = vld1q_s32(input + 24);
-  int32x4_t a7 = vld1q_s32(input + 28);
-  int32x4_t a8 = vld1q_s32(input + 32);
-  int32x4_t a9 = vld1q_s32(input + 36);
-  int32x4_t a10 = vld1q_s32(input + 40);
-  int32x4_t a11 = vld1q_s32(input + 44);
-  int32x4_t a12 = vld1q_s32(input + 48);
-  int32x4_t a13 = vld1q_s32(input + 52);
-  int32x4_t a14 = vld1q_s32(input + 56);
-  int32x4_t a15 = vld1q_s32(input + 60);
-  int16x8_t c0, c1, c2, c3, c4, c5, c6, c7;
+  int32x4_t a[16];
+  int16x8_t c[8];
 
+  a[0] = vld1q_s32(input);
+  a[1] = vld1q_s32(input + 4);
+  a[2] = vld1q_s32(input + 8);
+  a[3] = vld1q_s32(input + 12);
+  a[4] = vld1q_s32(input + 16);
+  a[5] = vld1q_s32(input + 20);
+  a[6] = vld1q_s32(input + 24);
+  a[7] = vld1q_s32(input + 28);
+  a[8] = vld1q_s32(input + 32);
+  a[9] = vld1q_s32(input + 36);
+  a[10] = vld1q_s32(input + 40);
+  a[11] = vld1q_s32(input + 44);
+  a[12] = vld1q_s32(input + 48);
+  a[13] = vld1q_s32(input + 52);
+  a[14] = vld1q_s32(input + 56);
+  a[15] = vld1q_s32(input + 60);
+
   if (bd == 8) {
     const int16x8_t cospis = vld1q_s16(kCospi);
     const int16x4_t cospis0 = vget_low_s16(cospis);   // cospi 0, 8, 16, 24
     const int16x4_t cospis1 = vget_high_s16(cospis);  // cospi 4, 12, 20, 28
-    int16x8_t b0 = vcombine_s16(vmovn_s32(a0), vmovn_s32(a1));
-    int16x8_t b1 = vcombine_s16(vmovn_s32(a2), vmovn_s32(a3));
-    int16x8_t b2 = vcombine_s16(vmovn_s32(a4), vmovn_s32(a5));
-    int16x8_t b3 = vcombine_s16(vmovn_s32(a6), vmovn_s32(a7));
-    int16x8_t b4 = vcombine_s16(vmovn_s32(a8), vmovn_s32(a9));
-    int16x8_t b5 = vcombine_s16(vmovn_s32(a10), vmovn_s32(a11));
-    int16x8_t b6 = vcombine_s16(vmovn_s32(a12), vmovn_s32(a13));
-    int16x8_t b7 = vcombine_s16(vmovn_s32(a14), vmovn_s32(a15));
+    int16x8_t b[8];
 
-    idct8x8_64_1d_bd8(cospis0, cospis1, &b0, &b1, &b2, &b3, &b4, &b5, &b6, &b7);
-    idct8x8_64_1d_bd8(cospis0, cospis1, &b0, &b1, &b2, &b3, &b4, &b5, &b6, &b7);
+    b[0] = vcombine_s16(vmovn_s32(a[0]), vmovn_s32(a[1]));
+    b[1] = vcombine_s16(vmovn_s32(a[2]), vmovn_s32(a[3]));
+    b[2] = vcombine_s16(vmovn_s32(a[4]), vmovn_s32(a[5]));
+    b[3] = vcombine_s16(vmovn_s32(a[6]), vmovn_s32(a[7]));
+    b[4] = vcombine_s16(vmovn_s32(a[8]), vmovn_s32(a[9]));
+    b[5] = vcombine_s16(vmovn_s32(a[10]), vmovn_s32(a[11]));
+    b[6] = vcombine_s16(vmovn_s32(a[12]), vmovn_s32(a[13]));
+    b[7] = vcombine_s16(vmovn_s32(a[14]), vmovn_s32(a[15]));
 
-    c0 = vrshrq_n_s16(b0, 5);
-    c1 = vrshrq_n_s16(b1, 5);
-    c2 = vrshrq_n_s16(b2, 5);
-    c3 = vrshrq_n_s16(b3, 5);
-    c4 = vrshrq_n_s16(b4, 5);
-    c5 = vrshrq_n_s16(b5, 5);
-    c6 = vrshrq_n_s16(b6, 5);
-    c7 = vrshrq_n_s16(b7, 5);
+    idct8x8_64_1d_bd8(cospis0, cospis1, b);
+    idct8x8_64_1d_bd8(cospis0, cospis1, b);
+
+    c[0] = vrshrq_n_s16(b[0], 5);
+    c[1] = vrshrq_n_s16(b[1], 5);
+    c[2] = vrshrq_n_s16(b[2], 5);
+    c[3] = vrshrq_n_s16(b[3], 5);
+    c[4] = vrshrq_n_s16(b[4], 5);
+    c[5] = vrshrq_n_s16(b[5], 5);
+    c[6] = vrshrq_n_s16(b[6], 5);
+    c[7] = vrshrq_n_s16(b[7], 5);
   } else {
     const int32x4_t cospis0 = vld1q_s32(kCospi32);      // cospi 0, 8, 16, 24
     const int32x4_t cospis1 = vld1q_s32(kCospi32 + 4);  // cospi 4, 12, 20, 28
 
     if (bd == 10) {
-      idct8x8_64_half1d_bd10(cospis0, cospis1, &a0, &a1, &a2, &a3, &a4, &a5,
-                             &a6, &a7);
-      idct8x8_64_half1d_bd10(cospis0, cospis1, &a8, &a9, &a10, &a11, &a12, &a13,
-                             &a14, &a15);
-      idct8x8_64_half1d_bd10(cospis0, cospis1, &a0, &a8, &a1, &a9, &a2, &a10,
-                             &a3, &a11);
-      idct8x8_64_half1d_bd10(cospis0, cospis1, &a4, &a12, &a5, &a13, &a6, &a14,
-                             &a7, &a15);
+      idct8x8_64_half1d_bd10(cospis0, cospis1, &a[0], &a[1], &a[2], &a[3],
+                             &a[4], &a[5], &a[6], &a[7]);
+      idct8x8_64_half1d_bd10(cospis0, cospis1, &a[8], &a[9], &a[10], &a[11],
+                             &a[12], &a[13], &a[14], &a[15]);
+      idct8x8_64_half1d_bd10(cospis0, cospis1, &a[0], &a[8], &a[1], &a[9],
+                             &a[2], &a[10], &a[3], &a[11]);
+      idct8x8_64_half1d_bd10(cospis0, cospis1, &a[4], &a[12], &a[5], &a[13],
+                             &a[6], &a[14], &a[7], &a[15]);
     } else {
-      idct8x8_64_half1d_bd12(cospis0, cospis1, &a0, &a1, &a2, &a3, &a4, &a5,
-                             &a6, &a7);
-      idct8x8_64_half1d_bd12(cospis0, cospis1, &a8, &a9, &a10, &a11, &a12, &a13,
-                             &a14, &a15);
-      idct8x8_64_half1d_bd12(cospis0, cospis1, &a0, &a8, &a1, &a9, &a2, &a10,
-                             &a3, &a11);
-      idct8x8_64_half1d_bd12(cospis0, cospis1, &a4, &a12, &a5, &a13, &a6, &a14,
-                             &a7, &a15);
+      idct8x8_64_half1d_bd12(cospis0, cospis1, &a[0], &a[1], &a[2], &a[3],
+                             &a[4], &a[5], &a[6], &a[7]);
+      idct8x8_64_half1d_bd12(cospis0, cospis1, &a[8], &a[9], &a[10], &a[11],
+                             &a[12], &a[13], &a[14], &a[15]);
+      idct8x8_64_half1d_bd12(cospis0, cospis1, &a[0], &a[8], &a[1], &a[9],
+                             &a[2], &a[10], &a[3], &a[11]);
+      idct8x8_64_half1d_bd12(cospis0, cospis1, &a[4], &a[12], &a[5], &a[13],
+                             &a[6], &a[14], &a[7], &a[15]);
     }
-    c0 = vcombine_s16(vrshrn_n_s32(a0, 5), vrshrn_n_s32(a4, 5));
-    c1 = vcombine_s16(vrshrn_n_s32(a8, 5), vrshrn_n_s32(a12, 5));
-    c2 = vcombine_s16(vrshrn_n_s32(a1, 5), vrshrn_n_s32(a5, 5));
-    c3 = vcombine_s16(vrshrn_n_s32(a9, 5), vrshrn_n_s32(a13, 5));
-    c4 = vcombine_s16(vrshrn_n_s32(a2, 5), vrshrn_n_s32(a6, 5));
-    c5 = vcombine_s16(vrshrn_n_s32(a10, 5), vrshrn_n_s32(a14, 5));
-    c6 = vcombine_s16(vrshrn_n_s32(a3, 5), vrshrn_n_s32(a7, 5));
-    c7 = vcombine_s16(vrshrn_n_s32(a11, 5), vrshrn_n_s32(a15, 5));
+    c[0] = vcombine_s16(vrshrn_n_s32(a[0], 5), vrshrn_n_s32(a[4], 5));
+    c[1] = vcombine_s16(vrshrn_n_s32(a[8], 5), vrshrn_n_s32(a[12], 5));
+    c[2] = vcombine_s16(vrshrn_n_s32(a[1], 5), vrshrn_n_s32(a[5], 5));
+    c[3] = vcombine_s16(vrshrn_n_s32(a[9], 5), vrshrn_n_s32(a[13], 5));
+    c[4] = vcombine_s16(vrshrn_n_s32(a[2], 5), vrshrn_n_s32(a[6], 5));
+    c[5] = vcombine_s16(vrshrn_n_s32(a[10], 5), vrshrn_n_s32(a[14], 5));
+    c[6] = vcombine_s16(vrshrn_n_s32(a[3], 5), vrshrn_n_s32(a[7], 5));
+    c[7] = vcombine_s16(vrshrn_n_s32(a[11], 5), vrshrn_n_s32(a[15], 5));
   }
-  highbd_add8x8(c0, c1, c2, c3, c4, c5, c6, c7, dest, stride, bd);
+  highbd_add8x8(c, dest, stride, bd);
 }
--- a/vpx_dsp/arm/idct32x32_135_add_neon.c
+++ b/vpx_dsp/arm/idct32x32_135_add_neon.c
@@ -650,14 +650,10 @@
     highbd_add_and_store_bd8(out, output, stride);
   } else {
     uint8_t *const outputT = (uint8_t *)output;
-    add_and_store_u8_s16(out[0], out[1], out[2], out[3], out[4], out[5], out[6],
-                         out[7], outputT, stride);
-    add_and_store_u8_s16(out[8], out[9], out[10], out[11], out[12], out[13],
-                         out[14], out[15], outputT + (8 * stride), stride);
-    add_and_store_u8_s16(out[16], out[17], out[18], out[19], out[20], out[21],
-                         out[22], out[23], outputT + (16 * stride), stride);
-    add_and_store_u8_s16(out[24], out[25], out[26], out[27], out[28], out[29],
-                         out[30], out[31], outputT + (24 * stride), stride);
+    add_and_store_u8_s16(out + 0, outputT, stride);
+    add_and_store_u8_s16(out + 8, outputT + (8 * stride), stride);
+    add_and_store_u8_s16(out + 16, outputT + (16 * stride), stride);
+    add_and_store_u8_s16(out + 24, outputT + (24 * stride), stride);
   }
 }
 
--- a/vpx_dsp/arm/idct32x32_34_add_neon.c
+++ b/vpx_dsp/arm/idct32x32_34_add_neon.c
@@ -490,14 +490,10 @@
     highbd_add_and_store_bd8(out, output, stride);
   } else {
     uint8_t *const outputT = (uint8_t *)output;
-    add_and_store_u8_s16(out[0], out[1], out[2], out[3], out[4], out[5], out[6],
-                         out[7], outputT, stride);
-    add_and_store_u8_s16(out[8], out[9], out[10], out[11], out[12], out[13],
-                         out[14], out[15], outputT + (8 * stride), stride);
-    add_and_store_u8_s16(out[16], out[17], out[18], out[19], out[20], out[21],
-                         out[22], out[23], outputT + (16 * stride), stride);
-    add_and_store_u8_s16(out[24], out[25], out[26], out[27], out[28], out[29],
-                         out[30], out[31], outputT + (24 * stride), stride);
+    add_and_store_u8_s16(out + 0, outputT, stride);
+    add_and_store_u8_s16(out + 8, outputT + (8 * stride), stride);
+    add_and_store_u8_s16(out + 16, outputT + (16 * stride), stride);
+    add_and_store_u8_s16(out + 24, outputT + (24 * stride), stride);
   }
 }
 
--- a/vpx_dsp/arm/idct4x4_add_neon.c
+++ b/vpx_dsp/arm/idct4x4_add_neon.c
@@ -20,43 +20,41 @@
                              int stride) {
   const uint8_t *dst = dest;
   const int16x4_t cospis = vld1_s16(kCospi);
-  uint8x8_t dest01_u8;
-  uint32x2_t dest32_u32 = vdup_n_u32(0);
-  int16x8_t a0, a1;
-  uint8x8_t d01, d32;
-  uint16x8_t d01_u16, d32_u16;
+  uint32x2_t s32 = vdup_n_u32(0);
+  int16x8_t a[2];
+  uint8x8_t s, d[2];
+  uint16x8_t sum[2];
 
   assert(!((intptr_t)dest % sizeof(uint32_t)));
   assert(!(stride % sizeof(uint32_t)));
 
   // Rows
-  a0 = load_tran_low_to_s16q(input);
-  a1 = load_tran_low_to_s16q(input + 8);
-  idct4x4_16_kernel_bd8(cospis, &a0, &a1);
+  a[0] = load_tran_low_to_s16q(input);
+  a[1] = load_tran_low_to_s16q(input + 8);
+  idct4x4_16_kernel_bd8(cospis, a);
 
   // Columns
-  a1 = vcombine_s16(vget_high_s16(a1), vget_low_s16(a1));
-  idct4x4_16_kernel_bd8(cospis, &a0, &a1);
-  a0 = vrshrq_n_s16(a0, 4);
-  a1 = vrshrq_n_s16(a1, 4);
+  a[1] = vcombine_s16(vget_high_s16(a[1]), vget_low_s16(a[1]));
+  idct4x4_16_kernel_bd8(cospis, a);
+  a[0] = vrshrq_n_s16(a[0], 4);
+  a[1] = vrshrq_n_s16(a[1], 4);
 
-  dest01_u8 = load_u8(dst, stride);
+  s = load_u8(dst, stride);
   dst += 2 * stride;
   // The elements are loaded in reverse order.
-  dest32_u32 = vld1_lane_u32((const uint32_t *)dst, dest32_u32, 1);
+  s32 = vld1_lane_u32((const uint32_t *)dst, s32, 1);
   dst += stride;
-  dest32_u32 = vld1_lane_u32((const uint32_t *)dst, dest32_u32, 0);
+  s32 = vld1_lane_u32((const uint32_t *)dst, s32, 0);
 
-  d01_u16 = vaddw_u8(vreinterpretq_u16_s16(a0), dest01_u8);
-  d32_u16 =
-      vaddw_u8(vreinterpretq_u16_s16(a1), vreinterpret_u8_u32(dest32_u32));
-  d01 = vqmovun_s16(vreinterpretq_s16_u16(d01_u16));
-  d32 = vqmovun_s16(vreinterpretq_s16_u16(d32_u16));
+  sum[0] = vaddw_u8(vreinterpretq_u16_s16(a[0]), s);
+  sum[1] = vaddw_u8(vreinterpretq_u16_s16(a[1]), vreinterpret_u8_u32(s32));
+  d[0] = vqmovun_s16(vreinterpretq_s16_u16(sum[0]));
+  d[1] = vqmovun_s16(vreinterpretq_s16_u16(sum[1]));
 
-  store_u8(dest, stride, d01);
+  store_u8(dest, stride, d[0]);
   dest += 2 * stride;
   // The elements are stored in reverse order.
-  vst1_lane_u32((uint32_t *)dest, vreinterpret_u32_u8(d32), 1);
+  vst1_lane_u32((uint32_t *)dest, vreinterpret_u32_u8(d[1]), 1);
   dest += stride;
-  vst1_lane_u32((uint32_t *)dest, vreinterpret_u32_u8(d32), 0);
+  vst1_lane_u32((uint32_t *)dest, vreinterpret_u32_u8(d[1]), 0);
 }
--- a/vpx_dsp/arm/idct8x8_add_neon.c
+++ b/vpx_dsp/arm/idct8x8_add_neon.c
@@ -17,72 +17,26 @@
 #include "vpx_dsp/arm/transpose_neon.h"
 #include "vpx_dsp/txfm_common.h"
 
-static INLINE void add8x8(int16x8_t a0, int16x8_t a1, int16x8_t a2,
-                          int16x8_t a3, int16x8_t a4, int16x8_t a5,
-                          int16x8_t a6, int16x8_t a7, uint8_t *dest,
-                          const int stride) {
-  const uint8_t *dst = dest;
-  uint8x8_t d0, d1, d2, d3, d4, d5, d6, d7;
-  uint16x8_t d0_u16, d1_u16, d2_u16, d3_u16, d4_u16, d5_u16, d6_u16, d7_u16;
+static INLINE void idct8x8_add8x1(const int16x8_t a, uint8_t **const dest,
+                                  const int stride) {
+  const uint8x8_t s = vld1_u8(*dest);
+  const int16x8_t res = vrshrq_n_s16(a, 5);
+  const uint16x8_t q = vaddw_u8(vreinterpretq_u16_s16(res), s);
+  const uint8x8_t d = vqmovun_s16(vreinterpretq_s16_u16(q));
+  vst1_u8(*dest, d);
+  *dest += stride;
+}
 
-  a0 = vrshrq_n_s16(a0, 5);
-  a1 = vrshrq_n_s16(a1, 5);
-  a2 = vrshrq_n_s16(a2, 5);
-  a3 = vrshrq_n_s16(a3, 5);
-  a4 = vrshrq_n_s16(a4, 5);
-  a5 = vrshrq_n_s16(a5, 5);
-  a6 = vrshrq_n_s16(a6, 5);
-  a7 = vrshrq_n_s16(a7, 5);
-
-  d0 = vld1_u8(dst);
-  dst += stride;
-  d1 = vld1_u8(dst);
-  dst += stride;
-  d2 = vld1_u8(dst);
-  dst += stride;
-  d3 = vld1_u8(dst);
-  dst += stride;
-  d4 = vld1_u8(dst);
-  dst += stride;
-  d5 = vld1_u8(dst);
-  dst += stride;
-  d6 = vld1_u8(dst);
-  dst += stride;
-  d7 = vld1_u8(dst);
-
-  d0_u16 = vaddw_u8(vreinterpretq_u16_s16(a0), d0);
-  d1_u16 = vaddw_u8(vreinterpretq_u16_s16(a1), d1);
-  d2_u16 = vaddw_u8(vreinterpretq_u16_s16(a2), d2);
-  d3_u16 = vaddw_u8(vreinterpretq_u16_s16(a3), d3);
-  d4_u16 = vaddw_u8(vreinterpretq_u16_s16(a4), d4);
-  d5_u16 = vaddw_u8(vreinterpretq_u16_s16(a5), d5);
-  d6_u16 = vaddw_u8(vreinterpretq_u16_s16(a6), d6);
-  d7_u16 = vaddw_u8(vreinterpretq_u16_s16(a7), d7);
-
-  d0 = vqmovun_s16(vreinterpretq_s16_u16(d0_u16));
-  d1 = vqmovun_s16(vreinterpretq_s16_u16(d1_u16));
-  d2 = vqmovun_s16(vreinterpretq_s16_u16(d2_u16));
-  d3 = vqmovun_s16(vreinterpretq_s16_u16(d3_u16));
-  d4 = vqmovun_s16(vreinterpretq_s16_u16(d4_u16));
-  d5 = vqmovun_s16(vreinterpretq_s16_u16(d5_u16));
-  d6 = vqmovun_s16(vreinterpretq_s16_u16(d6_u16));
-  d7 = vqmovun_s16(vreinterpretq_s16_u16(d7_u16));
-
-  vst1_u8(dest, d0);
-  dest += stride;
-  vst1_u8(dest, d1);
-  dest += stride;
-  vst1_u8(dest, d2);
-  dest += stride;
-  vst1_u8(dest, d3);
-  dest += stride;
-  vst1_u8(dest, d4);
-  dest += stride;
-  vst1_u8(dest, d5);
-  dest += stride;
-  vst1_u8(dest, d6);
-  dest += stride;
-  vst1_u8(dest, d7);
+static INLINE void add8x8(int16x8_t *const out, uint8_t *dest,
+                          const int stride) {
+  idct8x8_add8x1(out[0], &dest, stride);
+  idct8x8_add8x1(out[1], &dest, stride);
+  idct8x8_add8x1(out[2], &dest, stride);
+  idct8x8_add8x1(out[3], &dest, stride);
+  idct8x8_add8x1(out[4], &dest, stride);
+  idct8x8_add8x1(out[5], &dest, stride);
+  idct8x8_add8x1(out[6], &dest, stride);
+  idct8x8_add8x1(out[7], &dest, stride);
 }
 
 void vpx_idct8x8_64_add_neon(const tran_low_t *input, uint8_t *dest,
@@ -90,18 +44,20 @@
   const int16x8_t cospis = vld1q_s16(kCospi);
   const int16x4_t cospis0 = vget_low_s16(cospis);   // cospi 0, 8, 16, 24
   const int16x4_t cospis1 = vget_high_s16(cospis);  // cospi 4, 12, 20, 28
-  int16x8_t a0 = load_tran_low_to_s16q(input);
-  int16x8_t a1 = load_tran_low_to_s16q(input + 8);
-  int16x8_t a2 = load_tran_low_to_s16q(input + 16);
-  int16x8_t a3 = load_tran_low_to_s16q(input + 24);
-  int16x8_t a4 = load_tran_low_to_s16q(input + 32);
-  int16x8_t a5 = load_tran_low_to_s16q(input + 40);
-  int16x8_t a6 = load_tran_low_to_s16q(input + 48);
-  int16x8_t a7 = load_tran_low_to_s16q(input + 56);
+  int16x8_t a[8];
 
-  idct8x8_64_1d_bd8(cospis0, cospis1, &a0, &a1, &a2, &a3, &a4, &a5, &a6, &a7);
-  idct8x8_64_1d_bd8(cospis0, cospis1, &a0, &a1, &a2, &a3, &a4, &a5, &a6, &a7);
-  add8x8(a0, a1, a2, a3, a4, a5, a6, a7, dest, stride);
+  a[0] = load_tran_low_to_s16q(input);
+  a[1] = load_tran_low_to_s16q(input + 8);
+  a[2] = load_tran_low_to_s16q(input + 16);
+  a[3] = load_tran_low_to_s16q(input + 24);
+  a[4] = load_tran_low_to_s16q(input + 32);
+  a[5] = load_tran_low_to_s16q(input + 40);
+  a[6] = load_tran_low_to_s16q(input + 48);
+  a[7] = load_tran_low_to_s16q(input + 56);
+
+  idct8x8_64_1d_bd8(cospis0, cospis1, a);
+  idct8x8_64_1d_bd8(cospis0, cospis1, a);
+  add8x8(a, dest, stride);
 }
 
 void vpx_idct8x8_12_add_neon(const tran_low_t *input, uint8_t *dest,
@@ -111,17 +67,15 @@
   const int16x4_t cospis0 = vget_low_s16(cospis);     // cospi 0, 8, 16, 24
   const int16x4_t cospisd0 = vget_low_s16(cospisd);   // doubled 0, 8, 16, 24
   const int16x4_t cospisd1 = vget_high_s16(cospisd);  // doubled 4, 12, 20, 28
-  int16x4_t a0, a1, a2, a3, a4, a5, a6, a7;
-  int16x8_t b0, b1, b2, b3, b4, b5, b6, b7;
+  int16x4_t a[8];
+  int16x8_t b[8];
 
-  a0 = load_tran_low_to_s16d(input);
-  a1 = load_tran_low_to_s16d(input + 8);
-  a2 = load_tran_low_to_s16d(input + 16);
-  a3 = load_tran_low_to_s16d(input + 24);
+  a[0] = load_tran_low_to_s16d(input);
+  a[1] = load_tran_low_to_s16d(input + 8);
+  a[2] = load_tran_low_to_s16d(input + 16);
+  a[3] = load_tran_low_to_s16d(input + 24);
 
-  idct8x8_12_pass1_bd8(cospis0, cospisd0, cospisd1, &a0, &a1, &a2, &a3, &a4,
-                       &a5, &a6, &a7);
-  idct8x8_12_pass2_bd8(cospis0, cospisd0, cospisd1, a0, a1, a2, a3, a4, a5, a6,
-                       a7, &b0, &b1, &b2, &b3, &b4, &b5, &b6, &b7);
-  add8x8(b0, b1, b2, b3, b4, b5, b6, b7, dest, stride);
+  idct8x8_12_pass1_bd8(cospis0, cospisd0, cospisd1, a);
+  idct8x8_12_pass2_bd8(cospis0, cospisd0, cospisd1, a, b);
+  add8x8(b, dest, stride);
 }
--- a/vpx_dsp/arm/idct_neon.h
+++ b/vpx_dsp/arm/idct_neon.h
@@ -217,64 +217,62 @@
 }
 
 // Shift the output down by 6 and add it to the destination buffer.
-static INLINE void add_and_store_u8_s16(const int16x8_t a0, const int16x8_t a1,
-                                        const int16x8_t a2, const int16x8_t a3,
-                                        const int16x8_t a4, const int16x8_t a5,
-                                        const int16x8_t a6, const int16x8_t a7,
-                                        uint8_t *b, const int b_stride) {
-  uint8x8_t b0, b1, b2, b3, b4, b5, b6, b7;
-  int16x8_t c0, c1, c2, c3, c4, c5, c6, c7;
-  b0 = vld1_u8(b);
-  b += b_stride;
-  b1 = vld1_u8(b);
-  b += b_stride;
-  b2 = vld1_u8(b);
-  b += b_stride;
-  b3 = vld1_u8(b);
-  b += b_stride;
-  b4 = vld1_u8(b);
-  b += b_stride;
-  b5 = vld1_u8(b);
-  b += b_stride;
-  b6 = vld1_u8(b);
-  b += b_stride;
-  b7 = vld1_u8(b);
-  b -= (7 * b_stride);
+static INLINE void add_and_store_u8_s16(const int16x8_t *const a, uint8_t *d,
+                                        const int stride) {
+  uint8x8_t b[8];
+  int16x8_t c[8];
 
+  b[0] = vld1_u8(d);
+  d += stride;
+  b[1] = vld1_u8(d);
+  d += stride;
+  b[2] = vld1_u8(d);
+  d += stride;
+  b[3] = vld1_u8(d);
+  d += stride;
+  b[4] = vld1_u8(d);
+  d += stride;
+  b[5] = vld1_u8(d);
+  d += stride;
+  b[6] = vld1_u8(d);
+  d += stride;
+  b[7] = vld1_u8(d);
+  d -= (7 * stride);
+
   // c = b + (a >> 6)
-  c0 = vrsraq_n_s16(vreinterpretq_s16_u16(vmovl_u8(b0)), a0, 6);
-  c1 = vrsraq_n_s16(vreinterpretq_s16_u16(vmovl_u8(b1)), a1, 6);
-  c2 = vrsraq_n_s16(vreinterpretq_s16_u16(vmovl_u8(b2)), a2, 6);
-  c3 = vrsraq_n_s16(vreinterpretq_s16_u16(vmovl_u8(b3)), a3, 6);
-  c4 = vrsraq_n_s16(vreinterpretq_s16_u16(vmovl_u8(b4)), a4, 6);
-  c5 = vrsraq_n_s16(vreinterpretq_s16_u16(vmovl_u8(b5)), a5, 6);
-  c6 = vrsraq_n_s16(vreinterpretq_s16_u16(vmovl_u8(b6)), a6, 6);
-  c7 = vrsraq_n_s16(vreinterpretq_s16_u16(vmovl_u8(b7)), a7, 6);
+  c[0] = vrsraq_n_s16(vreinterpretq_s16_u16(vmovl_u8(b[0])), a[0], 6);
+  c[1] = vrsraq_n_s16(vreinterpretq_s16_u16(vmovl_u8(b[1])), a[1], 6);
+  c[2] = vrsraq_n_s16(vreinterpretq_s16_u16(vmovl_u8(b[2])), a[2], 6);
+  c[3] = vrsraq_n_s16(vreinterpretq_s16_u16(vmovl_u8(b[3])), a[3], 6);
+  c[4] = vrsraq_n_s16(vreinterpretq_s16_u16(vmovl_u8(b[4])), a[4], 6);
+  c[5] = vrsraq_n_s16(vreinterpretq_s16_u16(vmovl_u8(b[5])), a[5], 6);
+  c[6] = vrsraq_n_s16(vreinterpretq_s16_u16(vmovl_u8(b[6])), a[6], 6);
+  c[7] = vrsraq_n_s16(vreinterpretq_s16_u16(vmovl_u8(b[7])), a[7], 6);
 
-  b0 = vqmovun_s16(c0);
-  b1 = vqmovun_s16(c1);
-  b2 = vqmovun_s16(c2);
-  b3 = vqmovun_s16(c3);
-  b4 = vqmovun_s16(c4);
-  b5 = vqmovun_s16(c5);
-  b6 = vqmovun_s16(c6);
-  b7 = vqmovun_s16(c7);
+  b[0] = vqmovun_s16(c[0]);
+  b[1] = vqmovun_s16(c[1]);
+  b[2] = vqmovun_s16(c[2]);
+  b[3] = vqmovun_s16(c[3]);
+  b[4] = vqmovun_s16(c[4]);
+  b[5] = vqmovun_s16(c[5]);
+  b[6] = vqmovun_s16(c[6]);
+  b[7] = vqmovun_s16(c[7]);
 
-  vst1_u8(b, b0);
-  b += b_stride;
-  vst1_u8(b, b1);
-  b += b_stride;
-  vst1_u8(b, b2);
-  b += b_stride;
-  vst1_u8(b, b3);
-  b += b_stride;
-  vst1_u8(b, b4);
-  b += b_stride;
-  vst1_u8(b, b5);
-  b += b_stride;
-  vst1_u8(b, b6);
-  b += b_stride;
-  vst1_u8(b, b7);
+  vst1_u8(d, b[0]);
+  d += stride;
+  vst1_u8(d, b[1]);
+  d += stride;
+  vst1_u8(d, b[2]);
+  d += stride;
+  vst1_u8(d, b[3]);
+  d += stride;
+  vst1_u8(d, b[4]);
+  d += stride;
+  vst1_u8(d, b[5]);
+  d += stride;
+  vst1_u8(d, b[6]);
+  d += stride;
+  vst1_u8(d, b[7]);
 }
 
 static INLINE uint8x16_t create_dcq(const int16_t dc) {
@@ -284,55 +282,53 @@
 }
 
 static INLINE void idct4x4_16_kernel_bd8(const int16x4_t cospis,
-                                         int16x8_t *const a0,
-                                         int16x8_t *const a1) {
-  int16x4_t b0, b1, b2, b3;
-  int32x4_t c0, c1, c2, c3;
-  int16x8_t d0, d1;
+                                         int16x8_t *const a) {
+  int16x4_t b[4];
+  int32x4_t c[4];
+  int16x8_t d[2];
 
-  transpose_s16_4x4q(a0, a1);
-  b0 = vget_low_s16(*a0);
-  b1 = vget_high_s16(*a0);
-  b2 = vget_low_s16(*a1);
-  b3 = vget_high_s16(*a1);
-  c0 = vmull_lane_s16(b0, cospis, 2);
-  c2 = vmull_lane_s16(b1, cospis, 2);
-  c1 = vsubq_s32(c0, c2);
-  c0 = vaddq_s32(c0, c2);
-  c2 = vmull_lane_s16(b2, cospis, 3);
-  c3 = vmull_lane_s16(b2, cospis, 1);
-  c2 = vmlsl_lane_s16(c2, b3, cospis, 1);
-  c3 = vmlal_lane_s16(c3, b3, cospis, 3);
-  b0 = vrshrn_n_s32(c0, DCT_CONST_BITS);
-  b1 = vrshrn_n_s32(c1, DCT_CONST_BITS);
-  b2 = vrshrn_n_s32(c2, DCT_CONST_BITS);
-  b3 = vrshrn_n_s32(c3, DCT_CONST_BITS);
-  d0 = vcombine_s16(b0, b1);
-  d1 = vcombine_s16(b3, b2);
-  *a0 = vaddq_s16(d0, d1);
-  *a1 = vsubq_s16(d0, d1);
+  transpose_s16_4x4q(&a[0], &a[1]);
+  b[0] = vget_low_s16(a[0]);
+  b[1] = vget_high_s16(a[0]);
+  b[2] = vget_low_s16(a[1]);
+  b[3] = vget_high_s16(a[1]);
+  c[0] = vmull_lane_s16(b[0], cospis, 2);
+  c[2] = vmull_lane_s16(b[1], cospis, 2);
+  c[1] = vsubq_s32(c[0], c[2]);
+  c[0] = vaddq_s32(c[0], c[2]);
+  c[2] = vmull_lane_s16(b[2], cospis, 3);
+  c[3] = vmull_lane_s16(b[2], cospis, 1);
+  c[2] = vmlsl_lane_s16(c[2], b[3], cospis, 1);
+  c[3] = vmlal_lane_s16(c[3], b[3], cospis, 3);
+  b[0] = vrshrn_n_s32(c[0], DCT_CONST_BITS);
+  b[1] = vrshrn_n_s32(c[1], DCT_CONST_BITS);
+  b[2] = vrshrn_n_s32(c[2], DCT_CONST_BITS);
+  b[3] = vrshrn_n_s32(c[3], DCT_CONST_BITS);
+  d[0] = vcombine_s16(b[0], b[1]);
+  d[1] = vcombine_s16(b[3], b[2]);
+  a[0] = vaddq_s16(d[0], d[1]);
+  a[1] = vsubq_s16(d[0], d[1]);
 }
 
-static INLINE void idct8x8_12_pass1_bd8(
-    const int16x4_t cospis0, const int16x4_t cospisd0, const int16x4_t cospisd1,
-    int16x4_t *const io0, int16x4_t *const io1, int16x4_t *const io2,
-    int16x4_t *const io3, int16x4_t *const io4, int16x4_t *const io5,
-    int16x4_t *const io6, int16x4_t *const io7) {
+static INLINE void idct8x8_12_pass1_bd8(const int16x4_t cospis0,
+                                        const int16x4_t cospisd0,
+                                        const int16x4_t cospisd1,
+                                        int16x4_t *const io) {
   int16x4_t step1[8], step2[8];
   int32x4_t t32[2];
 
-  transpose_s16_4x4d(io0, io1, io2, io3);
+  transpose_s16_4x4d(&io[0], &io[1], &io[2], &io[3]);
 
   // stage 1
-  step1[4] = vqrdmulh_lane_s16(*io1, cospisd1, 3);
-  step1[5] = vqrdmulh_lane_s16(*io3, cospisd1, 2);
-  step1[6] = vqrdmulh_lane_s16(*io3, cospisd1, 1);
-  step1[7] = vqrdmulh_lane_s16(*io1, cospisd1, 0);
+  step1[4] = vqrdmulh_lane_s16(io[1], cospisd1, 3);
+  step1[5] = vqrdmulh_lane_s16(io[3], cospisd1, 2);
+  step1[6] = vqrdmulh_lane_s16(io[3], cospisd1, 1);
+  step1[7] = vqrdmulh_lane_s16(io[1], cospisd1, 0);
 
   // stage 2
-  step2[1] = vqrdmulh_lane_s16(*io0, cospisd0, 2);
-  step2[2] = vqrdmulh_lane_s16(*io2, cospisd0, 3);
-  step2[3] = vqrdmulh_lane_s16(*io2, cospisd0, 1);
+  step2[1] = vqrdmulh_lane_s16(io[0], cospisd0, 2);
+  step2[2] = vqrdmulh_lane_s16(io[2], cospisd0, 3);
+  step2[3] = vqrdmulh_lane_s16(io[2], cospisd0, 1);
 
   step2[4] = vadd_s16(step1[4], step1[5]);
   step2[5] = vsub_s16(step1[4], step1[5]);
@@ -352,32 +348,28 @@
   step1[6] = vrshrn_n_s32(t32[1], DCT_CONST_BITS);
 
   // stage 4
-  *io0 = vadd_s16(step1[0], step2[7]);
-  *io1 = vadd_s16(step1[1], step1[6]);
-  *io2 = vadd_s16(step1[2], step1[5]);
-  *io3 = vadd_s16(step1[3], step2[4]);
-  *io4 = vsub_s16(step1[3], step2[4]);
-  *io5 = vsub_s16(step1[2], step1[5]);
-  *io6 = vsub_s16(step1[1], step1[6]);
-  *io7 = vsub_s16(step1[0], step2[7]);
+  io[0] = vadd_s16(step1[0], step2[7]);
+  io[1] = vadd_s16(step1[1], step1[6]);
+  io[2] = vadd_s16(step1[2], step1[5]);
+  io[3] = vadd_s16(step1[3], step2[4]);
+  io[4] = vsub_s16(step1[3], step2[4]);
+  io[5] = vsub_s16(step1[2], step1[5]);
+  io[6] = vsub_s16(step1[1], step1[6]);
+  io[7] = vsub_s16(step1[0], step2[7]);
 }
 
-static INLINE void idct8x8_12_pass2_bd8(
-    const int16x4_t cospis0, const int16x4_t cospisd0, const int16x4_t cospisd1,
-    const int16x4_t input0, const int16x4_t input1, const int16x4_t input2,
-    const int16x4_t input3, const int16x4_t input4, const int16x4_t input5,
-    const int16x4_t input6, const int16x4_t input7, int16x8_t *const output0,
-    int16x8_t *const output1, int16x8_t *const output2,
-    int16x8_t *const output3, int16x8_t *const output4,
-    int16x8_t *const output5, int16x8_t *const output6,
-    int16x8_t *const output7) {
+static INLINE void idct8x8_12_pass2_bd8(const int16x4_t cospis0,
+                                        const int16x4_t cospisd0,
+                                        const int16x4_t cospisd1,
+                                        const int16x4_t *const input,
+                                        int16x8_t *const output) {
   int16x8_t in[4];
   int16x8_t step1[8], step2[8];
   int32x4_t t32[8];
   int16x4_t t16[8];
 
-  transpose_s16_4x8(input0, input1, input2, input3, input4, input5, input6,
-                    input7, &in[0], &in[1], &in[2], &in[3]);
+  transpose_s16_4x8(input[0], input[1], input[2], input[3], input[4], input[5],
+                    input[6], input[7], &in[0], &in[1], &in[2], &in[3]);
 
   // stage 1
   step1[4] = vqrdmulhq_lane_s16(in[1], cospisd1, 3);
@@ -415,66 +407,63 @@
   step1[6] = vcombine_s16(t16[2], t16[3]);
 
   // stage 4
-  *output0 = vaddq_s16(step1[0], step2[7]);
-  *output1 = vaddq_s16(step1[1], step1[6]);
-  *output2 = vaddq_s16(step1[2], step1[5]);
-  *output3 = vaddq_s16(step1[3], step2[4]);
-  *output4 = vsubq_s16(step1[3], step2[4]);
-  *output5 = vsubq_s16(step1[2], step1[5]);
-  *output6 = vsubq_s16(step1[1], step1[6]);
-  *output7 = vsubq_s16(step1[0], step2[7]);
+  output[0] = vaddq_s16(step1[0], step2[7]);
+  output[1] = vaddq_s16(step1[1], step1[6]);
+  output[2] = vaddq_s16(step1[2], step1[5]);
+  output[3] = vaddq_s16(step1[3], step2[4]);
+  output[4] = vsubq_s16(step1[3], step2[4]);
+  output[5] = vsubq_s16(step1[2], step1[5]);
+  output[6] = vsubq_s16(step1[1], step1[6]);
+  output[7] = vsubq_s16(step1[0], step2[7]);
 }
 
 static INLINE void idct8x8_64_1d_bd8(const int16x4_t cospis0,
                                      const int16x4_t cospis1,
-                                     int16x8_t *const io0, int16x8_t *const io1,
-                                     int16x8_t *const io2, int16x8_t *const io3,
-                                     int16x8_t *const io4, int16x8_t *const io5,
-                                     int16x8_t *const io6,
-                                     int16x8_t *const io7) {
-  int16x4_t input_1l, input_1h, input_3l, input_3h, input_5l, input_5h,
-      input_7l, input_7h;
+                                     int16x8_t *const io) {
+  int16x4_t input1l, input1h, input3l, input3h, input5l, input5h, input7l,
+      input7h;
   int16x4_t step1l[4], step1h[4];
   int16x8_t step1[8], step2[8];
   int32x4_t t32[8];
   int16x4_t t16[8];
 
-  transpose_s16_8x8(io0, io1, io2, io3, io4, io5, io6, io7);
+  transpose_s16_8x8(&io[0], &io[1], &io[2], &io[3], &io[4], &io[5], &io[6],
+                    &io[7]);
 
   // stage 1
-  input_1l = vget_low_s16(*io1);
-  input_1h = vget_high_s16(*io1);
-  input_3l = vget_low_s16(*io3);
-  input_3h = vget_high_s16(*io3);
-  input_5l = vget_low_s16(*io5);
-  input_5h = vget_high_s16(*io5);
-  input_7l = vget_low_s16(*io7);
-  input_7h = vget_high_s16(*io7);
-  step1l[0] = vget_low_s16(*io0);
-  step1h[0] = vget_high_s16(*io0);
-  step1l[1] = vget_low_s16(*io2);
-  step1h[1] = vget_high_s16(*io2);
-  step1l[2] = vget_low_s16(*io4);
-  step1h[2] = vget_high_s16(*io4);
-  step1l[3] = vget_low_s16(*io6);
-  step1h[3] = vget_high_s16(*io6);
+  input1l = vget_low_s16(io[1]);
+  input1h = vget_high_s16(io[1]);
+  input3l = vget_low_s16(io[3]);
+  input3h = vget_high_s16(io[3]);
+  input5l = vget_low_s16(io[5]);
+  input5h = vget_high_s16(io[5]);
+  input7l = vget_low_s16(io[7]);
+  input7h = vget_high_s16(io[7]);
+  step1l[0] = vget_low_s16(io[0]);
+  step1h[0] = vget_high_s16(io[0]);
+  step1l[1] = vget_low_s16(io[2]);
+  step1h[1] = vget_high_s16(io[2]);
+  step1l[2] = vget_low_s16(io[4]);
+  step1h[2] = vget_high_s16(io[4]);
+  step1l[3] = vget_low_s16(io[6]);
+  step1h[3] = vget_high_s16(io[6]);
 
-  t32[0] = vmull_lane_s16(input_1l, cospis1, 3);
-  t32[1] = vmull_lane_s16(input_1h, cospis1, 3);
-  t32[2] = vmull_lane_s16(input_3l, cospis1, 2);
-  t32[3] = vmull_lane_s16(input_3h, cospis1, 2);
-  t32[4] = vmull_lane_s16(input_3l, cospis1, 1);
-  t32[5] = vmull_lane_s16(input_3h, cospis1, 1);
-  t32[6] = vmull_lane_s16(input_1l, cospis1, 0);
-  t32[7] = vmull_lane_s16(input_1h, cospis1, 0);
-  t32[0] = vmlsl_lane_s16(t32[0], input_7l, cospis1, 0);
-  t32[1] = vmlsl_lane_s16(t32[1], input_7h, cospis1, 0);
-  t32[2] = vmlal_lane_s16(t32[2], input_5l, cospis1, 1);
-  t32[3] = vmlal_lane_s16(t32[3], input_5h, cospis1, 1);
-  t32[4] = vmlsl_lane_s16(t32[4], input_5l, cospis1, 2);
-  t32[5] = vmlsl_lane_s16(t32[5], input_5h, cospis1, 2);
-  t32[6] = vmlal_lane_s16(t32[6], input_7l, cospis1, 3);
-  t32[7] = vmlal_lane_s16(t32[7], input_7h, cospis1, 3);
+  t32[0] = vmull_lane_s16(input1l, cospis1, 3);
+  t32[1] = vmull_lane_s16(input1h, cospis1, 3);
+  t32[2] = vmull_lane_s16(input3l, cospis1, 2);
+  t32[3] = vmull_lane_s16(input3h, cospis1, 2);
+  t32[4] = vmull_lane_s16(input3l, cospis1, 1);
+  t32[5] = vmull_lane_s16(input3h, cospis1, 1);
+  t32[6] = vmull_lane_s16(input1l, cospis1, 0);
+  t32[7] = vmull_lane_s16(input1h, cospis1, 0);
+  t32[0] = vmlsl_lane_s16(t32[0], input7l, cospis1, 0);
+  t32[1] = vmlsl_lane_s16(t32[1], input7h, cospis1, 0);
+  t32[2] = vmlal_lane_s16(t32[2], input5l, cospis1, 1);
+  t32[3] = vmlal_lane_s16(t32[3], input5h, cospis1, 1);
+  t32[4] = vmlsl_lane_s16(t32[4], input5l, cospis1, 2);
+  t32[5] = vmlsl_lane_s16(t32[5], input5h, cospis1, 2);
+  t32[6] = vmlal_lane_s16(t32[6], input7l, cospis1, 3);
+  t32[7] = vmlal_lane_s16(t32[7], input7h, cospis1, 3);
   t16[0] = vrshrn_n_s32(t32[0], DCT_CONST_BITS);
   t16[1] = vrshrn_n_s32(t32[1], DCT_CONST_BITS);
   t16[2] = vrshrn_n_s32(t32[2], DCT_CONST_BITS);
@@ -541,14 +530,14 @@
   step1[6] = vcombine_s16(t16[2], t16[3]);
 
   // stage 4
-  *io0 = vaddq_s16(step1[0], step2[7]);
-  *io1 = vaddq_s16(step1[1], step1[6]);
-  *io2 = vaddq_s16(step1[2], step1[5]);
-  *io3 = vaddq_s16(step1[3], step2[4]);
-  *io4 = vsubq_s16(step1[3], step2[4]);
-  *io5 = vsubq_s16(step1[2], step1[5]);
-  *io6 = vsubq_s16(step1[1], step1[6]);
-  *io7 = vsubq_s16(step1[0], step2[7]);
+  io[0] = vaddq_s16(step1[0], step2[7]);
+  io[1] = vaddq_s16(step1[1], step1[6]);
+  io[2] = vaddq_s16(step1[2], step1[5]);
+  io[3] = vaddq_s16(step1[3], step2[4]);
+  io[4] = vsubq_s16(step1[3], step2[4]);
+  io[5] = vsubq_s16(step1[2], step1[5]);
+  io[6] = vsubq_s16(step1[1], step1[6]);
+  io[7] = vsubq_s16(step1[0], step2[7]);
 }
 
 static INLINE void idct16x16_add_wrap_low_8x2(const int32x4_t *const t32,
@@ -786,73 +775,72 @@
   vst1q_s16(output, out[15]);
 }
 
-static INLINE void idct16x16_add8x1(int16x8_t res, uint8_t **dest,
+static INLINE void idct16x16_add8x1(const int16x8_t a, uint8_t **const dest,
                                     const int stride) {
-  uint8x8_t d = vld1_u8(*dest);
-  uint16x8_t q;
-
-  res = vrshrq_n_s16(res, 6);
-  q = vaddw_u8(vreinterpretq_u16_s16(res), d);
-  d = vqmovun_s16(vreinterpretq_s16_u16(q));
+  const uint8x8_t s = vld1_u8(*dest);
+  const int16x8_t res = vrshrq_n_s16(a, 6);
+  const uint16x8_t q = vaddw_u8(vreinterpretq_u16_s16(res), s);
+  const uint8x8_t d = vqmovun_s16(vreinterpretq_s16_u16(q));
   vst1_u8(*dest, d);
   *dest += stride;
 }
 
-static INLINE void highbd_idct16x16_add8x1(int16x8_t res, const int16x8_t max,
-                                           uint16_t **dest, const int stride) {
-  uint16x8_t d = vld1q_u16(*dest);
-
-  res = vqaddq_s16(res, vreinterpretq_s16_u16(d));
-  res = vminq_s16(res, max);
-  d = vqshluq_n_s16(res, 0);
+static INLINE void highbd_idct16x16_add8x1(const int16x8_t a,
+                                           const int16x8_t max,
+                                           uint16_t **const dest,
+                                           const int stride) {
+  const uint16x8_t s = vld1q_u16(*dest);
+  const int16x8_t res0 = vqaddq_s16(a, vreinterpretq_s16_u16(s));
+  const int16x8_t res1 = vminq_s16(res0, max);
+  const uint16x8_t d = vqshluq_n_s16(res1, 0);
   vst1q_u16(*dest, d);
   *dest += stride;
 }
 
-static INLINE void highbd_idct16x16_add8x1_bd8(int16x8_t res, uint16_t **dest,
+static INLINE void highbd_idct16x16_add8x1_bd8(const int16x8_t a,
+                                               uint16_t **const dest,
                                                const int stride) {
-  uint16x8_t d = vld1q_u16(*dest);
-
-  res = vrsraq_n_s16(vreinterpretq_s16_u16(d), res, 6);
-  d = vmovl_u8(vqmovun_s16(res));
+  const uint16x8_t s = vld1q_u16(*dest);
+  const int16x8_t res = vrsraq_n_s16(vreinterpretq_s16_u16(s), a, 6);
+  const uint16x8_t d = vmovl_u8(vqmovun_s16(res));
   vst1q_u16(*dest, d);
   *dest += stride;
 }
 
 static INLINE void highbd_add_and_store_bd8(const int16x8_t *const a,
-                                            uint16_t *out, const int b_stride) {
-  highbd_idct16x16_add8x1_bd8(a[0], &out, b_stride);
-  highbd_idct16x16_add8x1_bd8(a[1], &out, b_stride);
-  highbd_idct16x16_add8x1_bd8(a[2], &out, b_stride);
-  highbd_idct16x16_add8x1_bd8(a[3], &out, b_stride);
-  highbd_idct16x16_add8x1_bd8(a[4], &out, b_stride);
-  highbd_idct16x16_add8x1_bd8(a[5], &out, b_stride);
-  highbd_idct16x16_add8x1_bd8(a[6], &out, b_stride);
-  highbd_idct16x16_add8x1_bd8(a[7], &out, b_stride);
-  highbd_idct16x16_add8x1_bd8(a[8], &out, b_stride);
-  highbd_idct16x16_add8x1_bd8(a[9], &out, b_stride);
-  highbd_idct16x16_add8x1_bd8(a[10], &out, b_stride);
-  highbd_idct16x16_add8x1_bd8(a[11], &out, b_stride);
-  highbd_idct16x16_add8x1_bd8(a[12], &out, b_stride);
-  highbd_idct16x16_add8x1_bd8(a[13], &out, b_stride);
-  highbd_idct16x16_add8x1_bd8(a[14], &out, b_stride);
-  highbd_idct16x16_add8x1_bd8(a[15], &out, b_stride);
-  highbd_idct16x16_add8x1_bd8(a[16], &out, b_stride);
-  highbd_idct16x16_add8x1_bd8(a[17], &out, b_stride);
-  highbd_idct16x16_add8x1_bd8(a[18], &out, b_stride);
-  highbd_idct16x16_add8x1_bd8(a[19], &out, b_stride);
-  highbd_idct16x16_add8x1_bd8(a[20], &out, b_stride);
-  highbd_idct16x16_add8x1_bd8(a[21], &out, b_stride);
-  highbd_idct16x16_add8x1_bd8(a[22], &out, b_stride);
-  highbd_idct16x16_add8x1_bd8(a[23], &out, b_stride);
-  highbd_idct16x16_add8x1_bd8(a[24], &out, b_stride);
-  highbd_idct16x16_add8x1_bd8(a[25], &out, b_stride);
-  highbd_idct16x16_add8x1_bd8(a[26], &out, b_stride);
-  highbd_idct16x16_add8x1_bd8(a[27], &out, b_stride);
-  highbd_idct16x16_add8x1_bd8(a[28], &out, b_stride);
-  highbd_idct16x16_add8x1_bd8(a[29], &out, b_stride);
-  highbd_idct16x16_add8x1_bd8(a[30], &out, b_stride);
-  highbd_idct16x16_add8x1_bd8(a[31], &out, b_stride);
+                                            uint16_t *out, const int stride) {
+  highbd_idct16x16_add8x1_bd8(a[0], &out, stride);
+  highbd_idct16x16_add8x1_bd8(a[1], &out, stride);
+  highbd_idct16x16_add8x1_bd8(a[2], &out, stride);
+  highbd_idct16x16_add8x1_bd8(a[3], &out, stride);
+  highbd_idct16x16_add8x1_bd8(a[4], &out, stride);
+  highbd_idct16x16_add8x1_bd8(a[5], &out, stride);
+  highbd_idct16x16_add8x1_bd8(a[6], &out, stride);
+  highbd_idct16x16_add8x1_bd8(a[7], &out, stride);
+  highbd_idct16x16_add8x1_bd8(a[8], &out, stride);
+  highbd_idct16x16_add8x1_bd8(a[9], &out, stride);
+  highbd_idct16x16_add8x1_bd8(a[10], &out, stride);
+  highbd_idct16x16_add8x1_bd8(a[11], &out, stride);
+  highbd_idct16x16_add8x1_bd8(a[12], &out, stride);
+  highbd_idct16x16_add8x1_bd8(a[13], &out, stride);
+  highbd_idct16x16_add8x1_bd8(a[14], &out, stride);
+  highbd_idct16x16_add8x1_bd8(a[15], &out, stride);
+  highbd_idct16x16_add8x1_bd8(a[16], &out, stride);
+  highbd_idct16x16_add8x1_bd8(a[17], &out, stride);
+  highbd_idct16x16_add8x1_bd8(a[18], &out, stride);
+  highbd_idct16x16_add8x1_bd8(a[19], &out, stride);
+  highbd_idct16x16_add8x1_bd8(a[20], &out, stride);
+  highbd_idct16x16_add8x1_bd8(a[21], &out, stride);
+  highbd_idct16x16_add8x1_bd8(a[22], &out, stride);
+  highbd_idct16x16_add8x1_bd8(a[23], &out, stride);
+  highbd_idct16x16_add8x1_bd8(a[24], &out, stride);
+  highbd_idct16x16_add8x1_bd8(a[25], &out, stride);
+  highbd_idct16x16_add8x1_bd8(a[26], &out, stride);
+  highbd_idct16x16_add8x1_bd8(a[27], &out, stride);
+  highbd_idct16x16_add8x1_bd8(a[28], &out, stride);
+  highbd_idct16x16_add8x1_bd8(a[29], &out, stride);
+  highbd_idct16x16_add8x1_bd8(a[30], &out, stride);
+  highbd_idct16x16_add8x1_bd8(a[31], &out, stride);
 }
 
 static INLINE void highbd_idct16x16_add_store(const int32x4x2_t *const out,