hal: vsx: further optimize v_signmask

Use the quadword bit permutation instruction to creatively move
the sign bits to create the mask. Note that values above 127 will
result in 0.

modules/core/include/opencv2/core/hal/intrin_vsx.hpp

@@ -845,36 +845,24 @@ inline v_uint64x2 v_popcount(const v_int64x2& a)
 /** Mask **/
 inline int v_signmask(const v_uint8x16& a)
 {
-    vec_uchar16 sv  = vec_sr(a.val, vec_uchar16_sp(7));
-    static const vec_uchar16 slm = {0, 1, 2, 3, 4, 5, 6, 7, 0, 1, 2, 3, 4, 5, 6, 7};
-    sv = vec_sl(sv, slm);
-    vec_uint4 sv4 = vec_sum4s(sv, vec_uint4_z);
-    static const vec_uint4 slm4 = {0, 0, 8, 8};
-    sv4 = vec_sl(sv4, slm4);
-    return vec_extract(vec_sums((vec_int4) sv4, vec_int4_z), 3);
+    static const vec_uchar16 qperm = {120, 112, 104, 96, 88, 80, 72, 64, 56, 48, 40, 32, 24, 16, 8, 0};
+    return vec_extract((vec_int4)vec_vbpermq(v_reinterpret_as_u8(a).val, qperm), 2);
 }
 inline int v_signmask(const v_int8x16& a)
 { return v_signmask(v_reinterpret_as_u8(a)); }
 
 inline int v_signmask(const v_int16x8& a)
 {
-    static const vec_ushort8 slm = {0, 1, 2, 3, 4, 5, 6, 7};
-    vec_short8 sv = vec_sr(a.val, vec_ushort8_sp(15));
-    sv = vec_sl(sv, slm);
-    vec_int4 svi = vec_int4_z;
-    svi = vec_sums(vec_sum4s(sv, svi), svi);
-    return vec_extract(svi, 3);
+    static const vec_uchar16 qperm = {112, 96, 80, 64, 48, 32, 16, 0, 128, 128, 128, 128, 128, 128, 128, 128};
+    return vec_extract((vec_int4)vec_vbpermq(v_reinterpret_as_u8(a).val, qperm), 2);
 }
 inline int v_signmask(const v_uint16x8& a)
 { return v_signmask(v_reinterpret_as_s16(a)); }
 
 inline int v_signmask(const v_int32x4& a)
 {
-    static const vec_uint4 slm = {0, 1, 2, 3};
-    vec_int4 sv = vec_sr(a.val, vec_uint4_sp(31));
-    sv = vec_sl(sv, slm);
-    sv = vec_sums(sv, vec_int4_z);
-    return vec_extract(sv, 3);
+    static const vec_uchar16 qperm = {96, 64, 32, 0, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128};
+    return vec_extract((vec_int4)vec_vbpermq(v_reinterpret_as_u8(a).val, qperm), 2);
 }
 inline int v_signmask(const v_uint32x4& a)
 { return v_signmask(v_reinterpret_as_s32(a)); }