shithub: dav1d

Download patch

ref: 32e265a86e535b5fad47bcac9b54f83e1e5eab33
parent: 96da9cc232b4b238bcd215025d075039a5b0a86f
author: Martin Storsjö <[email protected]>
date: Mon Feb 10 05:02:53 EST 2020

arm: looprestoration: Add 8bpc to existing function names, add HIGHBD_*_SUFFIX

Don't add it to dav1d_sgr_calc_ab1/2_neon and box3/5_v, as the same
concrete function implementations can be shared for both 8 and 16 bpc
for those functions.

--- a/src/arm/32/looprestoration.S
+++ b/src/arm/32/looprestoration.S
@@ -687,12 +687,12 @@
 
 #define SUM_STRIDE (384+16)
 
-// void dav1d_sgr_box3_h_neon(int32_t *sumsq, int16_t *sum,
-//                            const pixel (*left)[4],
-//                            const pixel *src, const ptrdiff_t stride,
-//                            const int w, const int h,
-//                            const enum LrEdgeFlags edges);
-function sgr_box3_h_neon, export=1
+// void dav1d_sgr_box3_h_8bpc_neon(int32_t *sumsq, int16_t *sum,
+//                                 const pixel (*left)[4],
+//                                 const pixel *src, const ptrdiff_t stride,
+//                                 const int w, const int h,
+//                                 const enum LrEdgeFlags edges);
+function sgr_box3_h_8bpc_neon, export=1
         push            {r4-r11,lr}
         vpush           {q4-q7}
         ldrd            r4,  r5,  [sp, #100]
@@ -961,12 +961,12 @@
 .purgem add3
 endfunc
 
-// void dav1d_sgr_box5_h_neon(int32_t *sumsq, int16_t *sum,
-//                            const pixel (*left)[4],
-//                            const pixel *src, const ptrdiff_t stride,
-//                            const int w, const int h,
-//                            const enum LrEdgeFlags edges);
-function sgr_box5_h_neon, export=1
+// void dav1d_sgr_box5_h_8bpc_neon(int32_t *sumsq, int16_t *sum,
+//                                 const pixel (*left)[4],
+//                                 const pixel *src, const ptrdiff_t stride,
+//                                 const int w, const int h,
+//                                 const enum LrEdgeFlags edges);
+function sgr_box5_h_8bpc_neon, export=1
         push            {r4-r11,lr}
         vpush           {q4-q7}
         ldrd            r4,  r5,  [sp, #100]
@@ -1661,11 +1661,11 @@
 
 #define FILTER_OUT_STRIDE 384
 
-// void dav1d_sgr_finish_filter1_neon(int16_t *tmp,
-//                                    const pixel *src, const ptrdiff_t stride,
-//                                    const int32_t *a, const int16_t *b,
-//                                    const int w, const int h);
-function sgr_finish_filter1_neon, export=1
+// void dav1d_sgr_finish_filter1_8bpc_neon(int16_t *tmp,
+//                                         const pixel *src, const ptrdiff_t stride,
+//                                         const int32_t *a, const int16_t *b,
+//                                         const int w, const int h);
+function sgr_finish_filter1_8bpc_neon, export=1
         push            {r4-r11,lr}
         vpush           {q4-q7}
         ldrd            r4,  r5,  [sp, #100]
@@ -1765,11 +1765,11 @@
         pop             {r4-r11,pc}
 endfunc
 
-// void dav1d_sgr_finish_filter2_neon(int16_t *tmp,
-//                                    const pixel *src, const ptrdiff_t stride,
-//                                    const int32_t *a, const int16_t *b,
-//                                    const int w, const int h);
-function sgr_finish_filter2_neon, export=1
+// void dav1d_sgr_finish_filter2_8bpc_neon(int16_t *tmp,
+//                                         const pixel *src, const ptrdiff_t stride,
+//                                         const int32_t *a, const int16_t *b,
+//                                         const int w, const int h);
+function sgr_finish_filter2_8bpc_neon, export=1
         push            {r4-r11,lr}
         vpush           {q4-q7}
         ldrd            r4,  r5,  [sp, #100]
@@ -1925,11 +1925,11 @@
         pop             {r4-r11,pc}
 endfunc
 
-// void dav1d_sgr_weighted1_neon(pixel *dst, const ptrdiff_t dst_stride,
-//                               const pixel *src, const ptrdiff_t src_stride,
-//                               const int16_t *t1, const int w, const int h,
-//                               const int wt);
-function sgr_weighted1_neon, export=1
+// void dav1d_sgr_weighted1_8bpc_neon(pixel *dst, const ptrdiff_t dst_stride,
+//                                    const pixel *src, const ptrdiff_t src_stride,
+//                                    const int16_t *t1, const int w, const int h,
+//                                    const int wt);
+function sgr_weighted1_8bpc_neon, export=1
         push            {r4-r9,lr}
         ldrd            r4,  r5,  [sp, #28]
         ldrd            r6,  r7,  [sp, #36]
@@ -2009,12 +2009,12 @@
         pop             {r4-r9,pc}
 endfunc
 
-// void dav1d_sgr_weighted2_neon(pixel *dst, const ptrdiff_t stride,
-//                               const pixel *src, const ptrdiff_t src_stride,
-//                               const int16_t *t1, const int16_t *t2,
-//                               const int w, const int h,
-//                               const int16_t wt[2]);
-function sgr_weighted2_neon, export=1
+// void dav1d_sgr_weighted2_8bpc_neon(pixel *dst, const ptrdiff_t stride,
+//                                    const pixel *src, const ptrdiff_t src_stride,
+//                                    const int16_t *t1, const int16_t *t2,
+//                                    const int w, const int h,
+//                                    const int16_t wt[2]);
+function sgr_weighted2_8bpc_neon, export=1
         push            {r4-r11,lr}
         ldrd            r4,  r5,  [sp, #36]
         ldrd            r6,  r7,  [sp, #44]
--- a/src/arm/64/looprestoration.S
+++ b/src/arm/64/looprestoration.S
@@ -617,12 +617,12 @@
 
 #define SUM_STRIDE (384+16)
 
-// void dav1d_sgr_box3_h_neon(int32_t *sumsq, int16_t *sum,
-//                            const pixel (*left)[4],
-//                            const pixel *src, const ptrdiff_t stride,
-//                            const int w, const int h,
-//                            const enum LrEdgeFlags edges);
-function sgr_box3_h_neon, export=1
+// void dav1d_sgr_box3_h_8bpc_neon(int32_t *sumsq, int16_t *sum,
+//                                 const pixel (*left)[4],
+//                                 const pixel *src, const ptrdiff_t stride,
+//                                 const int w, const int h,
+//                                 const enum LrEdgeFlags edges);
+function sgr_box3_h_8bpc_neon, export=1
         add             w5,  w5,  #2 // w += 2
 
         // Set up pointers for reading/writing alternate rows
@@ -879,12 +879,12 @@
 .purgem add3
 endfunc
 
-// void dav1d_sgr_box5_h_neon(int32_t *sumsq, int16_t *sum,
-//                            const pixel (*left)[4],
-//                            const pixel *src, const ptrdiff_t stride,
-//                            const int w, const int h,
-//                            const enum LrEdgeFlags edges);
-function sgr_box5_h_neon, export=1
+// void dav1d_sgr_box5_h_8bpc_neon(int32_t *sumsq, int16_t *sum,
+//                                 const pixel (*left)[4],
+//                                 const pixel *src, const ptrdiff_t stride,
+//                                 const int w, const int h,
+//                                 const enum LrEdgeFlags edges);
+function sgr_box5_h_8bpc_neon, export=1
         add             w5,  w5,  #2 // w += 2
 
         // Set up pointers for reading/writing alternate rows
@@ -1540,11 +1540,11 @@
 
 #define FILTER_OUT_STRIDE 384
 
-// void dav1d_sgr_finish_filter1_neon(int16_t *tmp,
-//                                    const pixel *src, const ptrdiff_t stride,
-//                                    const int32_t *a, const int16_t *b,
-//                                    const int w, const int h);
-function sgr_finish_filter1_neon, export=1
+// void dav1d_sgr_finish_filter1_8bpc_neon(int16_t *tmp,
+//                                         const pixel *src, const ptrdiff_t stride,
+//                                         const int32_t *a, const int16_t *b,
+//                                         const int w, const int h);
+function sgr_finish_filter1_8bpc_neon, export=1
         sub             x7,  x3,  #(4*SUM_STRIDE)
         add             x8,  x3,  #(4*SUM_STRIDE)
         sub             x9,  x4,  #(2*SUM_STRIDE)
@@ -1657,11 +1657,11 @@
         ret
 endfunc
 
-// void dav1d_sgr_finish_filter2_neon(int16_t *tmp,
-//                                    const pixel *src, const ptrdiff_t stride,
-//                                    const int32_t *a, const int16_t *b,
-//                                    const int w, const int h);
-function sgr_finish_filter2_neon, export=1
+// void dav1d_sgr_finish_filter2_8bpc_neon(int16_t *tmp,
+//                                         const pixel *src, const ptrdiff_t stride,
+//                                         const int32_t *a, const int16_t *b,
+//                                         const int w, const int h);
+function sgr_finish_filter2_8bpc_neon, export=1
         add             x7,  x3,  #(4*(SUM_STRIDE))
         sub             x3,  x3,  #(4*(SUM_STRIDE))
         add             x8,  x4,  #(2*(SUM_STRIDE))
@@ -1807,11 +1807,11 @@
         ret
 endfunc
 
-// void dav1d_sgr_weighted1_neon(pixel *dst, const ptrdiff_t dst_stride,
-//                               const pixel *src, const ptrdiff_t src_stride,
-//                               const int16_t *t1, const int w, const int h,
-//                               const int wt);
-function sgr_weighted1_neon, export=1
+// void dav1d_sgr_weighted1_8bpc_neon(pixel *dst, const ptrdiff_t dst_stride,
+//                                    const pixel *src, const ptrdiff_t src_stride,
+//                                    const int16_t *t1, const int w, const int h,
+//                                    const int wt);
+function sgr_weighted1_8bpc_neon, export=1
         dup             v31.8h, w7
         cmp             x6,  #2
         add             x9,  x0,  x1
@@ -1887,12 +1887,12 @@
         ret
 endfunc
 
-// void dav1d_sgr_weighted2_neon(pixel *dst, const ptrdiff_t stride,
-//                               const pixel *src, const ptrdiff_t src_stride,
-//                               const int16_t *t1, const int16_t *t2,
-//                               const int w, const int h,
-//                               const int16_t wt[2]);
-function sgr_weighted2_neon, export=1
+// void dav1d_sgr_weighted2_8bpc_neon(pixel *dst, const ptrdiff_t stride,
+//                                    const pixel *src, const ptrdiff_t src_stride,
+//                                    const int16_t *t1, const int16_t *t2,
+//                                    const int w, const int h,
+//                                    const int16_t wt[2]);
+function sgr_weighted2_8bpc_neon, export=1
         ldr             x8,  [sp]
         cmp             x7,  #2
         add             x10, x0,  x1
--- a/src/arm/looprestoration_init_tmpl.c
+++ b/src/arm/looprestoration_init_tmpl.c
@@ -107,20 +107,20 @@
 #endif
 
 #if BITDEPTH == 8
-void dav1d_sgr_box3_h_neon(int32_t *sumsq, int16_t *sum,
-                           const pixel (*left)[4],
-                           const pixel *src, const ptrdiff_t stride,
-                           const int w, const int h,
-                           const enum LrEdgeFlags edges);
+void BF(dav1d_sgr_box3_h, neon)(int32_t *sumsq, int16_t *sum,
+                                const pixel (*left)[4],
+                                const pixel *src, const ptrdiff_t stride,
+                                const int w, const int h,
+                                const enum LrEdgeFlags edges);
 void dav1d_sgr_box3_v_neon(int32_t *sumsq, int16_t *sum,
                            const int w, const int h,
                            const enum LrEdgeFlags edges);
 void dav1d_sgr_calc_ab1_neon(int32_t *a, int16_t *b,
                              const int w, const int h, const int strength);
-void dav1d_sgr_finish_filter1_neon(int16_t *tmp,
-                                   const pixel *src, const ptrdiff_t stride,
-                                   const int32_t *a, const int16_t *b,
-                                   const int w, const int h);
+void BF(dav1d_sgr_finish_filter1, neon)(int16_t *tmp,
+                                        const pixel *src, const ptrdiff_t stride,
+                                        const int32_t *a, const int16_t *b,
+                                        const int w, const int h);
 
 /* filter with a 3x3 box (radius=1) */
 static void dav1d_sgr_filter1_neon(int16_t *tmp,
@@ -128,7 +128,8 @@
                                    const pixel (*left)[4],
                                    const pixel *lpf, const ptrdiff_t lpf_stride,
                                    const int w, const int h, const int strength,
-                                   const enum LrEdgeFlags edges)
+                                   const enum LrEdgeFlags edges
+                                   HIGHBD_DECL_SUFFIX)
 {
     ALIGN_STK_16(int32_t, sumsq_mem, (384 + 16) * 68 + 8,);
     int32_t *const sumsq = &sumsq_mem[(384 + 16) * 2 + 8], *const a = sumsq;
@@ -135,35 +136,35 @@
     ALIGN_STK_16(int16_t, sum_mem, (384 + 16) * 68 + 16,);
     int16_t *const sum = &sum_mem[(384 + 16) * 2 + 16], *const b = sum;
 
-    dav1d_sgr_box3_h_neon(sumsq, sum, left, src, stride, w, h, edges);
+    BF(dav1d_sgr_box3_h, neon)(sumsq, sum, left, src, stride, w, h, edges);
     if (edges & LR_HAVE_TOP)
-        dav1d_sgr_box3_h_neon(&sumsq[-2 * (384 + 16)], &sum[-2 * (384 + 16)],
-                              NULL, lpf, lpf_stride, w, 2, edges);
+        BF(dav1d_sgr_box3_h, neon)(&sumsq[-2 * (384 + 16)], &sum[-2 * (384 + 16)],
+                                   NULL, lpf, lpf_stride, w, 2, edges);
 
     if (edges & LR_HAVE_BOTTOM)
-        dav1d_sgr_box3_h_neon(&sumsq[h * (384 + 16)], &sum[h * (384 + 16)],
-                              NULL, lpf + 6 * PXSTRIDE(lpf_stride),
-                              lpf_stride, w, 2, edges);
+        BF(dav1d_sgr_box3_h, neon)(&sumsq[h * (384 + 16)], &sum[h * (384 + 16)],
+                                   NULL, lpf + 6 * PXSTRIDE(lpf_stride),
+                                   lpf_stride, w, 2, edges);
 
     dav1d_sgr_box3_v_neon(sumsq, sum, w, h, edges);
     dav1d_sgr_calc_ab1_neon(a, b, w, h, strength);
-    dav1d_sgr_finish_filter1_neon(tmp, src, stride, a, b, w, h);
+    BF(dav1d_sgr_finish_filter1, neon)(tmp, src, stride, a, b, w, h);
 }
 
-void dav1d_sgr_box5_h_neon(int32_t *sumsq, int16_t *sum,
-                           const pixel (*left)[4],
-                           const pixel *src, const ptrdiff_t stride,
-                           const int w, const int h,
-                           const enum LrEdgeFlags edges);
+void BF(dav1d_sgr_box5_h, neon)(int32_t *sumsq, int16_t *sum,
+                                const pixel (*left)[4],
+                                const pixel *src, const ptrdiff_t stride,
+                                const int w, const int h,
+                                const enum LrEdgeFlags edges);
 void dav1d_sgr_box5_v_neon(int32_t *sumsq, int16_t *sum,
                            const int w, const int h,
                            const enum LrEdgeFlags edges);
 void dav1d_sgr_calc_ab2_neon(int32_t *a, int16_t *b,
                              const int w, const int h, const int strength);
-void dav1d_sgr_finish_filter2_neon(int16_t *tmp,
-                                   const pixel *src, const ptrdiff_t stride,
-                                   const int32_t *a, const int16_t *b,
-                                   const int w, const int h);
+void BF(dav1d_sgr_finish_filter2, neon)(int16_t *tmp,
+                                        const pixel *src, const ptrdiff_t stride,
+                                        const int32_t *a, const int16_t *b,
+                                        const int w, const int h);
 
 /* filter with a 5x5 box (radius=2) */
 static void dav1d_sgr_filter2_neon(int16_t *tmp,
@@ -171,7 +172,8 @@
                                    const pixel (*left)[4],
                                    const pixel *lpf, const ptrdiff_t lpf_stride,
                                    const int w, const int h, const int strength,
-                                   const enum LrEdgeFlags edges)
+                                   const enum LrEdgeFlags edges
+                                   HIGHBD_DECL_SUFFIX)
 {
     ALIGN_STK_16(int32_t, sumsq_mem, (384 + 16) * 68 + 8,);
     int32_t *const sumsq = &sumsq_mem[(384 + 16) * 2 + 8], *const a = sumsq;
@@ -178,52 +180,57 @@
     ALIGN_STK_16(int16_t, sum_mem, (384 + 16) * 68 + 16,);
     int16_t *const sum = &sum_mem[(384 + 16) * 2 + 16], *const b = sum;
 
-    dav1d_sgr_box5_h_neon(sumsq, sum, left, src, stride, w, h, edges);
+    BF(dav1d_sgr_box5_h, neon)(sumsq, sum, left, src, stride, w, h, edges);
     if (edges & LR_HAVE_TOP)
-        dav1d_sgr_box5_h_neon(&sumsq[-2 * (384 + 16)], &sum[-2 * (384 + 16)],
-                              NULL, lpf, lpf_stride, w, 2, edges);
+        BF(dav1d_sgr_box5_h, neon)(&sumsq[-2 * (384 + 16)], &sum[-2 * (384 + 16)],
+                                   NULL, lpf, lpf_stride, w, 2, edges);
 
     if (edges & LR_HAVE_BOTTOM)
-        dav1d_sgr_box5_h_neon(&sumsq[h * (384 + 16)], &sum[h * (384 + 16)],
-                              NULL, lpf + 6 * PXSTRIDE(lpf_stride),
-                              lpf_stride, w, 2, edges);
+        BF(dav1d_sgr_box5_h, neon)(&sumsq[h * (384 + 16)], &sum[h * (384 + 16)],
+                                   NULL, lpf + 6 * PXSTRIDE(lpf_stride),
+                                   lpf_stride, w, 2, edges);
 
     dav1d_sgr_box5_v_neon(sumsq, sum, w, h, edges);
     dav1d_sgr_calc_ab2_neon(a, b, w, h, strength);
-    dav1d_sgr_finish_filter2_neon(tmp, src, stride, a, b, w, h);
+    BF(dav1d_sgr_finish_filter2, neon)(tmp, src, stride, a, b, w, h);
 }
 
-void dav1d_sgr_weighted1_neon(pixel *dst, const ptrdiff_t dst_stride,
-                              const pixel *src, const ptrdiff_t src_stride,
-                              const int16_t *t1, const int w, const int h,
-                              const int wt);
-void dav1d_sgr_weighted2_neon(pixel *dst, const ptrdiff_t dst_stride,
-                              const pixel *src, const ptrdiff_t src_stride,
-                              const int16_t *t1, const int16_t *t2,
-                              const int w, const int h,
-                              const int16_t wt[2]);
+void BF(dav1d_sgr_weighted1, neon)(pixel *dst, const ptrdiff_t dst_stride,
+                                   const pixel *src, const ptrdiff_t src_stride,
+                                   const int16_t *t1, const int w, const int h,
+                                   const int wt HIGHBD_DECL_SUFFIX);
+void BF(dav1d_sgr_weighted2, neon)(pixel *dst, const ptrdiff_t dst_stride,
+                                   const pixel *src, const ptrdiff_t src_stride,
+                                   const int16_t *t1, const int16_t *t2,
+                                   const int w, const int h,
+                                   const int16_t wt[2] HIGHBD_DECL_SUFFIX);
 
 static void sgr_filter_neon(pixel *const dst, const ptrdiff_t dst_stride,
                              const pixel (*const left)[4],
                              const pixel *lpf, const ptrdiff_t lpf_stride,
                              const int w, const int h, const int sgr_idx,
-                             const int16_t sgr_wt[7], const enum LrEdgeFlags edges)
+                             const int16_t sgr_wt[7], const enum LrEdgeFlags edges
+                             HIGHBD_DECL_SUFFIX)
 {
     if (!dav1d_sgr_params[sgr_idx][0]) {
         ALIGN_STK_16(int16_t, tmp, 64 * 384,);
         dav1d_sgr_filter1_neon(tmp, dst, dst_stride, left, lpf, lpf_stride,
-                               w, h, dav1d_sgr_params[sgr_idx][3], edges);
+                               w, h, dav1d_sgr_params[sgr_idx][3], edges
+                               HIGHBD_TAIL_SUFFIX);
         if (w >= 8)
-            dav1d_sgr_weighted1_neon(dst, dst_stride, dst, dst_stride,
-                                     tmp, w & ~7, h, (1 << 7) - sgr_wt[1]);
+            BF(dav1d_sgr_weighted1, neon)(dst, dst_stride, dst, dst_stride,
+                                          tmp, w & ~7, h, (1 << 7) - sgr_wt[1]
+                                          HIGHBD_TAIL_SUFFIX);
         if (w & 7) {
             // For uneven widths, do a full 8 pixel wide filtering into a temp
             // buffer and copy out the narrow slice of pixels separately into
             // dest.
             ALIGN_STK_16(pixel, stripe, 64 * 8,);
-            dav1d_sgr_weighted1_neon(stripe, w & 7, dst + (w & ~7), dst_stride,
-                                     tmp + (w & ~7), w & 7, h,
-                                     (1 << 7) - sgr_wt[1]);
+            BF(dav1d_sgr_weighted1, neon)(stripe, (w & 7) * sizeof(pixel),
+                                          dst + (w & ~7), dst_stride,
+                                          tmp + (w & ~7), w & 7, h,
+                                          (1 << 7) - sgr_wt[1]
+                                          HIGHBD_TAIL_SUFFIX);
             BF(dav1d_copy_narrow, neon)(dst + (w & ~7), dst_stride, stripe,
                                         w & 7, h);
         }
@@ -230,17 +237,21 @@
     } else if (!dav1d_sgr_params[sgr_idx][1]) {
         ALIGN_STK_16(int16_t, tmp, 64 * 384,);
         dav1d_sgr_filter2_neon(tmp, dst, dst_stride, left, lpf, lpf_stride,
-                               w, h, dav1d_sgr_params[sgr_idx][2], edges);
+                               w, h, dav1d_sgr_params[sgr_idx][2], edges
+                               HIGHBD_TAIL_SUFFIX);
         if (w >= 8)
-            dav1d_sgr_weighted1_neon(dst, dst_stride, dst, dst_stride,
-                                     tmp, w & ~7, h, sgr_wt[0]);
+            BF(dav1d_sgr_weighted1, neon)(dst, dst_stride, dst, dst_stride,
+                                          tmp, w & ~7, h, sgr_wt[0]
+                                          HIGHBD_TAIL_SUFFIX);
         if (w & 7) {
             // For uneven widths, do a full 8 pixel wide filtering into a temp
             // buffer and copy out the narrow slice of pixels separately into
             // dest.
             ALIGN_STK_16(pixel, stripe, 64 * 8,);
-            dav1d_sgr_weighted1_neon(stripe, w & 7, dst + (w & ~7), dst_stride,
-                                     tmp + (w & ~7), w & 7, h, sgr_wt[0]);
+            BF(dav1d_sgr_weighted1, neon)(stripe, (w & 7) * sizeof(pixel),
+                                          dst + (w & ~7), dst_stride,
+                                          tmp + (w & ~7), w & 7, h, sgr_wt[0]
+                                          HIGHBD_TAIL_SUFFIX);
             BF(dav1d_copy_narrow, neon)(dst + (w & ~7), dst_stride, stripe,
                                         w & 7, h);
         }
@@ -248,21 +259,25 @@
         ALIGN_STK_16(int16_t, tmp1, 64 * 384,);
         ALIGN_STK_16(int16_t, tmp2, 64 * 384,);
         dav1d_sgr_filter2_neon(tmp1, dst, dst_stride, left, lpf, lpf_stride,
-                               w, h, dav1d_sgr_params[sgr_idx][2], edges);
+                               w, h, dav1d_sgr_params[sgr_idx][2], edges
+                               HIGHBD_TAIL_SUFFIX);
         dav1d_sgr_filter1_neon(tmp2, dst, dst_stride, left, lpf, lpf_stride,
-                               w, h, dav1d_sgr_params[sgr_idx][3], edges);
+                               w, h, dav1d_sgr_params[sgr_idx][3], edges
+                               HIGHBD_TAIL_SUFFIX);
         const int16_t wt[2] = { sgr_wt[0], 128 - sgr_wt[0] - sgr_wt[1] };
         if (w >= 8)
-            dav1d_sgr_weighted2_neon(dst, dst_stride, dst, dst_stride,
-                                     tmp1, tmp2, w & ~7, h, wt);
+            BF(dav1d_sgr_weighted2, neon)(dst, dst_stride, dst, dst_stride,
+                                          tmp1, tmp2, w & ~7, h, wt
+                                          HIGHBD_TAIL_SUFFIX);
         if (w & 7) {
             // For uneven widths, do a full 8 pixel wide filtering into a temp
             // buffer and copy out the narrow slice of pixels separately into
             // dest.
             ALIGN_STK_16(pixel, stripe, 64 * 8,);
-            dav1d_sgr_weighted2_neon(stripe, w & 7, dst + (w & ~7), dst_stride,
-                                     tmp1 + (w & ~7), tmp2 + (w & ~7),
-                                     w & 7, h, wt);
+            BF(dav1d_sgr_weighted2, neon)(stripe, (w & 7) * sizeof(pixel),
+                                          dst + (w & ~7), dst_stride,
+                                          tmp1 + (w & ~7), tmp2 + (w & ~7),
+                                          w & 7, h, wt HIGHBD_TAIL_SUFFIX);
             BF(dav1d_copy_narrow, neon)(dst + (w & ~7), dst_stride, stripe,
                                         w & 7, h);
         }