aarch64: vp9itxfm: Reorder the idct coefficients for better pairing

All elements are used pairwise, except for the first one.
Previously, the 16th element was unused. Move the unused element
to the second slot, to make the later element pairs not split
across registers.

This simplifies loading only parts of the coefficients,
reducing the difference to the 16 bpp version.

Signed-off-by: Martin Storsjö <martin@martin.st>

1 files changed, 62 insertions, 62 deletions

diff --git a/libavcodec/aarch64/vp9itxfm_neon.S b/libavcodec/aarch64/vp9itxfm_neon.S
index b6c2575236..d4fc2163aa 100644
--- a/libavcodec/aarch64/vp9itxfm_neon.S
+++ b/libavcodec/aarch64/vp9itxfm_neon.S
@@ -22,7 +22,7 @@
 #include "neon.S"
 
 const itxfm4_coeffs, align=4
-        .short  11585, 6270, 15137, 0
+        .short  11585, 0, 6270, 15137
 iadst4_coeffs:
         .short  5283, 15212, 9929, 13377
 endconst
@@ -30,8 +30,8 @@ endconst
 const iadst8_coeffs, align=4
         .short  16305, 1606, 14449, 7723, 10394, 12665, 4756, 15679
 idct_coeffs:
-        .short  11585, 6270, 15137, 3196, 16069, 13623, 9102, 1606
-        .short  16305, 12665, 10394, 7723, 14449, 15679, 4756, 0
+        .short  11585, 0, 6270, 15137, 3196, 16069, 13623, 9102
+        .short  1606, 16305, 12665, 10394, 7723, 14449, 15679, 4756
         .short  804, 16364, 12140, 11003, 7005, 14811, 15426, 5520
         .short  3981, 15893, 14053, 8423, 9760, 13160, 16207, 2404
 endconst
@@ -192,14 +192,14 @@ endconst
 .endm
 
 .macro idct4 c0, c1, c2, c3
-        smull           v22.4s,    \c1\().4h, v0.h[2]
-        smull           v20.4s,    \c1\().4h, v0.h[1]
+        smull           v22.4s,    \c1\().4h, v0.h[3]
+        smull           v20.4s,    \c1\().4h, v0.h[2]
         add             v16.4h,    \c0\().4h, \c2\().4h
         sub             v17.4h,    \c0\().4h, \c2\().4h
-        smlal           v22.4s,    \c3\().4h, v0.h[1]
+        smlal           v22.4s,    \c3\().4h, v0.h[2]
         smull           v18.4s,    v16.4h,    v0.h[0]
         smull           v19.4s,    v17.4h,    v0.h[0]
-        smlsl           v20.4s,    \c3\().4h, v0.h[2]
+        smlsl           v20.4s,    \c3\().4h, v0.h[3]
         rshrn           v22.4h,    v22.4s,    #14
         rshrn           v18.4h,    v18.4s,    #14
         rshrn           v19.4h,    v19.4s,    #14
@@ -326,9 +326,9 @@ itxfm_func4x4 iwht,  iwht
 
 .macro idct8
         dmbutterfly0    v16, v20, v16, v20, v2, v3, v4, v5, v6, v7 // v16 = t0a, v20 = t1a
-        dmbutterfly     v18, v22, v0.h[1], v0.h[2], v2, v3, v4, v5 // v18 = t2a, v22 = t3a
-        dmbutterfly     v17, v23, v0.h[3], v0.h[4], v2, v3, v4, v5 // v17 = t4a, v23 = t7a
-        dmbutterfly     v21, v19, v0.h[5], v0.h[6], v2, v3, v4, v5 // v21 = t5a, v19 = t6a
+        dmbutterfly     v18, v22, v0.h[2], v0.h[3], v2, v3, v4, v5 // v18 = t2a, v22 = t3a
+        dmbutterfly     v17, v23, v0.h[4], v0.h[5], v2, v3, v4, v5 // v17 = t4a, v23 = t7a
+        dmbutterfly     v21, v19, v0.h[6], v0.h[7], v2, v3, v4, v5 // v21 = t5a, v19 = t6a
 
         butterfly_8h    v24, v25, v16, v22 // v24 = t0, v25 = t3
         butterfly_8h    v28, v29, v17, v21 // v28 = t4, v29 = t5a
@@ -361,8 +361,8 @@ itxfm_func4x4 iwht,  iwht
         dmbutterfly0    v19, v20, v6, v7, v24, v26, v27, v28, v29, v30   // v19 = -out[3], v20 = out[4]
         neg             v19.8h,   v19.8h  // v19 = out[3]
 
-        dmbutterfly_l   v26, v27, v28, v29, v5,  v3,  v0.h[1], v0.h[2]   // v26,v27 = t5a, v28,v29 = t4a
-        dmbutterfly_l   v2,  v3,  v4,  v5,  v31, v25, v0.h[2], v0.h[1]   // v2,v3   = t6a, v4,v5   = t7a
+        dmbutterfly_l   v26, v27, v28, v29, v5,  v3,  v0.h[2], v0.h[3]   // v26,v27 = t5a, v28,v29 = t4a
+        dmbutterfly_l   v2,  v3,  v4,  v5,  v31, v25, v0.h[3], v0.h[2]   // v2,v3   = t6a, v4,v5   = t7a
 
         dbutterfly_n    v17, v30, v28, v29, v2,  v3,  v6,  v7,  v24, v25 // v17 = -out[1], v30 = t6
         dbutterfly_n    v22, v31, v26, v27, v4,  v5,  v6,  v7,  v24, v25 // v22 = out[6],  v31 = t7
@@ -543,13 +543,13 @@ endfunc
 
 function idct16
         dmbutterfly0    v16, v24, v16, v24, v2, v3, v4, v5, v6, v7 // v16 = t0a,  v24 = t1a
-        dmbutterfly     v20, v28, v0.h[1], v0.h[2], v2, v3, v4, v5 // v20 = t2a,  v28 = t3a
-        dmbutterfly     v18, v30, v0.h[3], v0.h[4], v2, v3, v4, v5 // v18 = t4a,  v30 = t7a
-        dmbutterfly     v26, v22, v0.h[5], v0.h[6], v2, v3, v4, v5 // v26 = t5a,  v22 = t6a
-        dmbutterfly     v17, v31, v0.h[7], v1.h[0], v2, v3, v4, v5 // v17 = t8a,  v31 = t15a
-        dmbutterfly     v25, v23, v1.h[1], v1.h[2], v2, v3, v4, v5 // v25 = t9a,  v23 = t14a
-        dmbutterfly     v21, v27, v1.h[3], v1.h[4], v2, v3, v4, v5 // v21 = t10a, v27 = t13a
-        dmbutterfly     v29, v19, v1.h[5], v1.h[6], v2, v3, v4, v5 // v29 = t11a, v19 = t12a
+        dmbutterfly     v20, v28, v0.h[2], v0.h[3], v2, v3, v4, v5 // v20 = t2a,  v28 = t3a
+        dmbutterfly     v18, v30, v0.h[4], v0.h[5], v2, v3, v4, v5 // v18 = t4a,  v30 = t7a
+        dmbutterfly     v26, v22, v0.h[6], v0.h[7], v2, v3, v4, v5 // v26 = t5a,  v22 = t6a
+        dmbutterfly     v17, v31, v1.h[0], v1.h[1], v2, v3, v4, v5 // v17 = t8a,  v31 = t15a
+        dmbutterfly     v25, v23, v1.h[2], v1.h[3], v2, v3, v4, v5 // v25 = t9a,  v23 = t14a
+        dmbutterfly     v21, v27, v1.h[4], v1.h[5], v2, v3, v4, v5 // v21 = t10a, v27 = t13a
+        dmbutterfly     v29, v19, v1.h[6], v1.h[7], v2, v3, v4, v5 // v29 = t11a, v19 = t12a
 
         butterfly_8h    v4,  v28, v16, v28               // v4  = t0,   v28 = t3
         butterfly_8h    v5,  v20, v24, v20               // v5  = t1,   v20 = t2
@@ -561,20 +561,20 @@ function idct16
         butterfly_8h    v29, v23, v31, v23               // v29 = t15,  v23 = t14
 
         dmbutterfly0    v22, v26, v22, v26, v2, v3, v18, v19, v30, v31        // v22 = t6a,  v26 = t5a
-        dmbutterfly     v23, v25, v0.h[1], v0.h[2], v18, v19, v30, v31        // v23 = t9a,  v25 = t14a
-        dmbutterfly     v27, v21, v0.h[1], v0.h[2], v18, v19, v30, v31, neg=1 // v27 = t13a, v21 = t10a
+        dmbutterfly     v23, v25, v0.h[2], v0.h[3], v18, v19, v30, v31        // v23 = t9a,  v25 = t14a
+        dmbutterfly     v27, v21, v0.h[2], v0.h[3], v18, v19, v30, v31, neg=1 // v27 = t13a, v21 = t10a
         idct16_end
 endfunc
 
 function idct16_half
         dmbutterfly0_h  v16, v24, v16, v24, v2, v3, v4, v5, v6, v7 // v16 = t0a,  v24 = t1a
-        dmbutterfly_h1  v20, v28, v0.h[1], v0.h[2], v2, v3, v4, v5 // v20 = t2a,  v28 = t3a
-        dmbutterfly_h1  v18, v30, v0.h[3], v0.h[4], v2, v3, v4, v5 // v18 = t4a,  v30 = t7a
-        dmbutterfly_h2  v26, v22, v0.h[5], v0.h[6], v2, v3, v4, v5 // v26 = t5a,  v22 = t6a
-        dmbutterfly_h1  v17, v31, v0.h[7], v1.h[0], v2, v3, v4, v5 // v17 = t8a,  v31 = t15a
-        dmbutterfly_h2  v25, v23, v1.h[1], v1.h[2], v2, v3, v4, v5 // v25 = t9a,  v23 = t14a
-        dmbutterfly_h1  v21, v27, v1.h[3], v1.h[4], v2, v3, v4, v5 // v21 = t10a, v27 = t13a
-        dmbutterfly_h2  v29, v19, v1.h[5], v1.h[6], v2, v3, v4, v5 // v29 = t11a, v19 = t12a
+        dmbutterfly_h1  v20, v28, v0.h[2], v0.h[3], v2, v3, v4, v5 // v20 = t2a,  v28 = t3a
+        dmbutterfly_h1  v18, v30, v0.h[4], v0.h[5], v2, v3, v4, v5 // v18 = t4a,  v30 = t7a
+        dmbutterfly_h2  v26, v22, v0.h[6], v0.h[7], v2, v3, v4, v5 // v26 = t5a,  v22 = t6a
+        dmbutterfly_h1  v17, v31, v1.h[0], v1.h[1], v2, v3, v4, v5 // v17 = t8a,  v31 = t15a
+        dmbutterfly_h2  v25, v23, v1.h[2], v1.h[3], v2, v3, v4, v5 // v25 = t9a,  v23 = t14a
+        dmbutterfly_h1  v21, v27, v1.h[4], v1.h[5], v2, v3, v4, v5 // v21 = t10a, v27 = t13a
+        dmbutterfly_h2  v29, v19, v1.h[6], v1.h[7], v2, v3, v4, v5 // v29 = t11a, v19 = t12a
 
         butterfly_8h    v4,  v28, v16, v28               // v4  = t0,   v28 = t3
         butterfly_8h    v5,  v20, v24, v20               // v5  = t1,   v20 = t2
@@ -586,20 +586,20 @@ function idct16_half
         butterfly_8h    v29, v23, v31, v23               // v29 = t15,  v23 = t14
 
         dmbutterfly0    v22, v26, v22, v26, v2, v3, v18, v19, v30, v31        // v22 = t6a,  v26 = t5a
-        dmbutterfly     v23, v25, v0.h[1], v0.h[2], v18, v19, v30, v31        // v23 = t9a,  v25 = t14a
-        dmbutterfly     v27, v21, v0.h[1], v0.h[2], v18, v19, v30, v31, neg=1 // v27 = t13a, v21 = t10a
+        dmbutterfly     v23, v25, v0.h[2], v0.h[3], v18, v19, v30, v31        // v23 = t9a,  v25 = t14a
+        dmbutterfly     v27, v21, v0.h[2], v0.h[3], v18, v19, v30, v31, neg=1 // v27 = t13a, v21 = t10a
         idct16_end
 endfunc
 
 function idct16_quarter
-        dsmull_h        v24, v25, v19, v1.h[6]
-        dsmull_h        v4,  v5,  v17, v0.h[7]
-        dsmull_h        v7,  v6,  v18, v0.h[4]
-        dsmull_h        v30, v31, v18, v0.h[3]
+        dsmull_h        v24, v25, v19, v1.h[7]
+        dsmull_h        v4,  v5,  v17, v1.h[0]
+        dsmull_h        v7,  v6,  v18, v0.h[5]
+        dsmull_h        v30, v31, v18, v0.h[4]
         neg             v24.4s,  v24.4s
         neg             v25.4s,  v25.4s
-        dsmull_h        v29, v28, v17, v1.h[0]
-        dsmull_h        v26, v27, v19, v1.h[5]
+        dsmull_h        v29, v28, v17, v1.h[1]
+        dsmull_h        v26, v27, v19, v1.h[6]
         dsmull_h        v22, v23, v16, v0.h[0]
         drshrn_h        v24, v24, v25, #14
         drshrn_h        v16, v4,  v5,  #14
@@ -609,8 +609,8 @@ function idct16_quarter
         drshrn_h        v17, v26, v27, #14
         drshrn_h        v28, v22, v23, #14
 
-        dmbutterfly_l   v20, v21, v22, v23, v17, v24, v0.h[1], v0.h[2]
-        dmbutterfly_l   v18, v19, v30, v31, v29, v16, v0.h[1], v0.h[2]
+        dmbutterfly_l   v20, v21, v22, v23, v17, v24, v0.h[2], v0.h[3]
+        dmbutterfly_l   v18, v19, v30, v31, v29, v16, v0.h[2], v0.h[3]
         neg             v22.4s,  v22.4s
         neg             v23.4s,  v23.4s
         drshrn_h        v27, v20, v21, #14
@@ -646,16 +646,16 @@ function iadst16
         dmbutterfly_l   v10, v11, v8,  v9,  v17, v30, v1.h[7], v1.h[6]   // v10,v11 = t15,  v8,v9   = t14
         ld1             {v0.8h}, [x10]
         dbutterfly_n    v22, v30, v6,  v7,  v10, v11, v12, v13, v10, v11 // v22     = t7a,  v30     = t15a
-        dmbutterfly_l   v14, v15, v12, v13, v23, v24, v0.h[3], v0.h[4]   // v14,v15 = t9,   v12,v13 = t8
+        dmbutterfly_l   v14, v15, v12, v13, v23, v24, v0.h[4], v0.h[5]   // v14,v15 = t9,   v12,v13 = t8
         dbutterfly_n    v25, v17, v4,  v5,  v8,  v9,  v6,  v7,  v8,  v9  // v25     = t6a,  v17     = t14a
 
-        dmbutterfly_l   v4,  v5,  v6,  v7,  v28, v19, v0.h[4], v0.h[3]   // v4,v5   = t12,  v6,v7   = t13
+        dmbutterfly_l   v4,  v5,  v6,  v7,  v28, v19, v0.h[5], v0.h[4]   // v4,v5   = t12,  v6,v7   = t13
         dbutterfly_n    v23, v19, v12, v13, v4,  v5,  v8,  v9,  v4,  v5  // v23     = t8a,  v19     = t12a
-        dmbutterfly_l   v10, v11, v8,  v9,  v21, v26, v0.h[5], v0.h[6]   // v10,v11 = t11,  v8,v9   = t10
+        dmbutterfly_l   v10, v11, v8,  v9,  v21, v26, v0.h[6], v0.h[7]   // v10,v11 = t11,  v8,v9   = t10
         butterfly_8h_r  v4,  v27, v16, v27               // v4  = t4,   v27 = t0
         dbutterfly_n    v24, v28, v14, v15, v6,  v7,  v12, v13, v6,  v7  // v24     = t9a,  v28     = t13a
 
-        dmbutterfly_l   v12, v13, v14, v15, v30, v17, v0.h[6], v0.h[5]   // v12,v13 = t14,  v14,v15 = t15
+        dmbutterfly_l   v12, v13, v14, v15, v30, v17, v0.h[7], v0.h[6]   // v12,v13 = t14,  v14,v15 = t15
         butterfly_8h_r  v5,  v20, v31, v20               // v5  = t5, v20 = t1
         dbutterfly_n    v21, v17, v8,  v9,  v12, v13, v6,  v7,  v12, v13 // v21     = t10a, v17     = t14a
         dbutterfly_n    v26, v30, v10, v11, v14, v15, v8,  v9,  v14, v15 // v26     = t11a, v30     = t15a
@@ -663,15 +663,15 @@ function iadst16
         butterfly_8h_r  v6,  v25, v18, v25               // v6  = t6, v25 = t2
         butterfly_8h_r  v7,  v22, v29, v22               // v7  = t7, v22 = t3
 
-        dmbutterfly_l   v10, v11, v8,  v9,  v19, v28, v0.h[1], v0.h[2]   // v10,v11 = t13,  v8,v9   = t12
-        dmbutterfly_l   v12, v13, v14, v15, v30, v17, v0.h[2], v0.h[1]   // v12,v13 = t14,  v14,v15 = t15
+        dmbutterfly_l   v10, v11, v8,  v9,  v19, v28, v0.h[2], v0.h[3]   // v10,v11 = t13,  v8,v9   = t12
+        dmbutterfly_l   v12, v13, v14, v15, v30, v17, v0.h[3], v0.h[2]   // v12,v13 = t14,  v14,v15 = t15
 
         dbutterfly_n    v18, v30, v8,  v9,  v12, v13, v16, v17, v12, v13 // v18   = out[2], v30     = t14a
         dbutterfly_n    v29, v17, v10, v11, v14, v15, v12, v13, v14, v15 // v29 = -out[13], v17     = t15a
         neg             v29.8h, v29.8h                   // v29 = out[13]
 
-        dmbutterfly_l   v10, v11, v8,  v9,  v4,  v5,  v0.h[1], v0.h[2]   // v10,v11 = t5a,  v8,v9   = t4a
-        dmbutterfly_l   v12, v13, v14, v15, v7,  v6,  v0.h[2], v0.h[1]   // v12,v13 = t6a,  v14,v15 = t7a
+        dmbutterfly_l   v10, v11, v8,  v9,  v4,  v5,  v0.h[2], v0.h[3]   // v10,v11 = t5a,  v8,v9   = t4a
+        dmbutterfly_l   v12, v13, v14, v15, v7,  v6,  v0.h[3], v0.h[2]   // v12,v13 = t6a,  v14,v15 = t7a
 
         butterfly_8h    v2,  v6,  v27, v25               // v2 = out[0], v6 = t2a
         butterfly_8h    v3,  v7,  v23, v21               // v3 =-out[1], v7 = t10
@@ -1101,10 +1101,10 @@ endfunc
         butterfly_8h    v7,  v3,  v29, v31 // v7  = t31a, v3  = t28a
         butterfly_8h    v22, v27, v24, v27 // v22 = t30,  v27 = t29
 
-        dmbutterfly     v27, v20, v0.h[1], v0.h[2], v24, v25, v30, v31        // v27 = t18a, v20 = t29a
-        dmbutterfly     v3,  v5,  v0.h[1], v0.h[2], v24, v25, v30, v31        // v3  = t19,  v5  = t28
-        dmbutterfly     v28, v6,  v0.h[1], v0.h[2], v24, v25, v30, v31, neg=1 // v28 = t27,  v6  = t20
-        dmbutterfly     v26, v21, v0.h[1], v0.h[2], v24, v25, v30, v31, neg=1 // v26 = t26a, v21 = t21a
+        dmbutterfly     v27, v20, v0.h[2], v0.h[3], v24, v25, v30, v31        // v27 = t18a, v20 = t29a
+        dmbutterfly     v3,  v5,  v0.h[2], v0.h[3], v24, v25, v30, v31        // v3  = t19,  v5  = t28
+        dmbutterfly     v28, v6,  v0.h[2], v0.h[3], v24, v25, v30, v31, neg=1 // v28 = t27,  v6  = t20
+        dmbutterfly     v26, v21, v0.h[2], v0.h[3], v24, v25, v30, v31, neg=1 // v26 = t26a, v21 = t21a
 
         butterfly_8h    v31, v24, v7,  v4  // v31 = t31,  v24 = t24
         butterfly_8h    v30, v25, v22, v23 // v30 = t30a, v25 = t25a
@@ -1141,10 +1141,10 @@ function idct32_odd
         butterfly_8h    v29, v23, v31, v23 // v29 = t31, v23 = t30
         butterfly_8h    v31, v27, v19, v27 // v31 = t28, v27 = t29
 
-        dmbutterfly     v23, v24, v0.h[3], v0.h[4], v16, v17, v18, v19        // v23 = t17a, v24 = t30a
-        dmbutterfly     v27, v20, v0.h[3], v0.h[4], v16, v17, v18, v19, neg=1 // v27 = t29a, v20 = t18a
-        dmbutterfly     v21, v26, v0.h[5], v0.h[6], v16, v17, v18, v19        // v21 = t21a, v26 = t26a
-        dmbutterfly     v25, v22, v0.h[5], v0.h[6], v16, v17, v18, v19, neg=1 // v25 = t25a, v22 = t22a
+        dmbutterfly     v23, v24, v0.h[4], v0.h[5], v16, v17, v18, v19        // v23 = t17a, v24 = t30a
+        dmbutterfly     v27, v20, v0.h[4], v0.h[5], v16, v17, v18, v19, neg=1 // v27 = t29a, v20 = t18a
+        dmbutterfly     v21, v26, v0.h[6], v0.h[7], v16, v17, v18, v19        // v21 = t21a, v26 = t26a
+        dmbutterfly     v25, v22, v0.h[6], v0.h[7], v16, v17, v18, v19, neg=1 // v25 = t25a, v22 = t22a
         idct32_end
 endfunc
 
@@ -1167,10 +1167,10 @@ function idct32_odd_half
         butterfly_8h    v29, v23, v31, v23 // v29 = t31, v23 = t30
         butterfly_8h    v31, v27, v19, v27 // v31 = t28, v27 = t29
 
-        dmbutterfly     v23, v24, v0.h[3], v0.h[4], v16, v17, v18, v19        // v23 = t17a, v24 = t30a
-        dmbutterfly     v27, v20, v0.h[3], v0.h[4], v16, v17, v18, v19, neg=1 // v27 = t29a, v20 = t18a
-        dmbutterfly     v21, v26, v0.h[5], v0.h[6], v16, v17, v18, v19        // v21 = t21a, v26 = t26a
-        dmbutterfly     v25, v22, v0.h[5], v0.h[6], v16, v17, v18, v19, neg=1 // v25 = t25a, v22 = t22a
+        dmbutterfly     v23, v24, v0.h[4], v0.h[5], v16, v17, v18, v19        // v23 = t17a, v24 = t30a
+        dmbutterfly     v27, v20, v0.h[4], v0.h[5], v16, v17, v18, v19, neg=1 // v27 = t29a, v20 = t18a
+        dmbutterfly     v21, v26, v0.h[6], v0.h[7], v16, v17, v18, v19        // v21 = t21a, v26 = t26a
+        dmbutterfly     v25, v22, v0.h[6], v0.h[7], v16, v17, v18, v19, neg=1 // v25 = t25a, v22 = t22a
         idct32_end
 endfunc
 
@@ -1198,18 +1198,18 @@ function idct32_odd_quarter
         drshrn_h        v6,  v20, v21, #14
         drshrn_h        v30, v24, v25, #14
 
-        dmbutterfly_l   v16, v17, v18, v19, v29, v4,  v0.h[3], v0.h[4]
-        dmbutterfly_l   v27, v26, v20, v21, v31, v5,  v0.h[3], v0.h[4]
+        dmbutterfly_l   v16, v17, v18, v19, v29, v4,  v0.h[4], v0.h[5]
+        dmbutterfly_l   v27, v26, v20, v21, v31, v5,  v0.h[4], v0.h[5]
         drshrn_h        v23, v16, v17, #14
         drshrn_h        v24, v18, v19, #14
         neg             v20.4s, v20.4s
         neg             v21.4s, v21.4s
         drshrn_h        v27, v27, v26, #14
         drshrn_h        v20, v20, v21, #14
-        dmbutterfly_l   v16, v17, v18, v19, v30, v6,  v0.h[5], v0.h[6]
+        dmbutterfly_l   v16, v17, v18, v19, v30, v6,  v0.h[6], v0.h[7]
         drshrn_h        v21, v16, v17, #14
         drshrn_h        v26, v18, v19, #14
-        dmbutterfly_l   v16, v17, v18, v19, v28, v7,  v0.h[5], v0.h[6]
+        dmbutterfly_l   v16, v17, v18, v19, v28, v7,  v0.h[6], v0.h[7]
         drshrn_h        v25, v16, v17, #14
         neg             v18.4s, v18.4s
         neg             v19.4s, v19.4s