RubyGems - carray - Versions diffs - 3.0.5 → 3.0.6 - Mend

carray 3.0.5 → 3.0.6

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (119) hide show

checksums.yaml +4 -4
data/CHANGELOG.md +233 -0
data/README.md +1 -1
data/carray.gemspec +3 -2
data/ext/ca_axis_dispatch.c +6 -0
data/ext/ca_axis_group.c +1 -1
data/ext/ca_categorical_iterator.c +7 -7
data/ext/ca_kernel_iterator.c +116 -231
data/ext/ca_kernel_iterator.h +410 -254
data/ext/ca_obj_bincmp.c +9 -1
data/ext/ca_obj_binop.c +10 -1
data/ext/ca_obj_bitfield.c +2 -0
data/ext/ca_obj_byte_swap.c +2 -0
data/ext/ca_obj_face.c +2 -0
data/ext/ca_obj_fake.c +3 -1
data/ext/ca_obj_grid.c +167 -12
data/ext/ca_obj_meld.c +69 -113
data/ext/ca_obj_moncmp.c +9 -1
data/ext/ca_obj_monop.c +40 -2
data/ext/ca_obj_object.c +54 -20
data/ext/ca_obj_reduce.c +15 -1
data/ext/ca_obj_refer.c +65 -20
data/ext/ca_obj_remap.c +69 -10
data/ext/ca_obj_roll.c +1 -0
data/ext/ca_obj_select.c +4 -1
data/ext/ca_obj_select_axis.c +1 -0
data/ext/ca_obj_shift.c +1 -0
data/ext/ca_obj_stack.c +310 -111
data/ext/ca_obj_stride.c +131 -45
data/ext/ca_obj_string.c +18 -2
data/ext/ca_obj_tile.c +1 -0
data/ext/ca_obj_triop.c +9 -1
data/ext/ca_obj_window.c +1 -0
data/ext/ca_op_ipower.c +8 -8
data/ext/caframe_csv_reader.c +1 -1
data/ext/carray.h +2 -1
data/ext/carray_access.c +27 -64
data/ext/carray_address_basis.c +134 -0
data/ext/carray_attribute.c +8 -0
data/ext/carray_broadcast.c +3 -4
data/ext/carray_build_flags.h +1 -1
data/ext/carray_cast.c +8 -11
data/ext/carray_conversion.c +9 -5
data/ext/carray_core.c +811 -72
data/ext/carray_element.c +13 -9
data/ext/carray_histogram.c +25 -3
data/ext/carray_internal.h +147 -12
data/ext/carray_kernels_bincmp.c +2007 -3
data/ext/carray_kernels_binop.c +25365 -5199
data/ext/carray_kernels_init.c +8 -8
data/ext/carray_kernels_monop.c +10696 -3678
data/ext/carray_kernels_reduce_aggregate.c +4429 -702
data/ext/carray_kernels_reduce_boolean.c +1533 -15
data/ext/carray_kernels_reduce_cumulative.c +1808 -248
data/ext/carray_kernels_reduce_extreme.c +4508 -511
data/ext/carray_kernels_reduce_variance.c +2284 -152
data/ext/carray_kernels_scan.c +240 -240
data/ext/carray_kernels_search.c +4211 -264
data/ext/carray_kernels_sort.c +367 -450
data/ext/carray_lazy.c +47 -0
data/ext/carray_loop.c +1 -1
data/ext/carray_mask.c +130 -139
data/ext/carray_median_percentile.c +4 -4
data/ext/carray_memory_view.c +13 -7
data/ext/carray_operator.c +147 -44
data/ext/carray_order.c +3 -2
data/ext/carray_random.c +4 -1
data/ext/carray_slab.c +97 -21
data/ext/carray_sort.c +1 -1
data/ext/carray_test.c +74 -19
data/ext/carray_time_text.c +2 -2
data/ext/extconf.rb +11 -0
data/ext/mkkernel.rb +346 -94
data/ext/ruby_carray.c +4 -1
data/ext/version.h +4 -4
data/lib/carray/attribute.rb +25 -1
data/lib/carray/autoload_carray.rb +3 -3
data/lib/carray/autoload_method_extension.rb +5 -3
data/lib/carray/axis_group.rb +13 -7
data/lib/carray/basics.rb +1 -1
data/lib/carray/block_iterator.rb +17 -3
data/lib/carray/categorical.rb +6 -8
data/lib/carray/categorical_iterator.rb +2 -2
data/lib/carray/complex.rb +2 -2
data/lib/carray/conditional.rb +3 -2
data/lib/carray/frame/column_verbs.rb +45 -14
data/lib/carray/frame/convert.rb +19 -0
data/lib/carray/frame/csv_parser.rb +32 -32
data/lib/carray/frame/describe.rb +4 -2
data/lib/carray/frame/frame.rb +399 -78
data/lib/carray/frame/group.rb +679 -53
data/lib/carray/frame/io.rb +209 -71
data/lib/carray/frame/join.rb +146 -37
data/lib/carray/frame/records.rb +3 -3
data/lib/carray/frame/reshape.rb +89 -11
data/lib/carray/frame/sort.rb +57 -9
data/lib/carray/frame/verbs.rb +384 -49
data/lib/carray/frame/windows.rb +44 -6
data/lib/carray/inspect.rb +7 -2
data/lib/carray/iterator.rb +32 -0
data/lib/carray/lazy.rb +17 -8
data/lib/carray/mask_gap_fill.rb +37 -17
data/lib/carray/meld_reduce.rb +1 -1
data/lib/carray/methods/choose.rb +3 -2
data/lib/carray/methods/meshgrid.rb +3 -1
data/lib/carray/runtime.rb +47 -8
data/lib/carray/segment_iterator.rb +23 -8
data/lib/carray/serialize.rb +34 -17
data/lib/carray/stack.rb +18 -6
data/lib/carray/time.rb +5 -0
data/lib/carray/window_iterator.rb +16 -4
data/yard-stubs/ca_obj_array.rb +26 -1
data/yard-stubs/carray_cast.rb +8 -5
data/yard-stubs/carray_conversion.rb +8 -2
data/yard-stubs/carray_mask.rb +20 -1
data/yard-stubs/carray_math.rb +31 -16
data/yard-stubs/carray_memory_view.rb +7 -0
data/yard-stubs/carray_test.rb +7 -4
metadata +18 -4

data/ext/carray_kernels_bincmp.c CHANGED Viewed

@@ -75,6 +75,34 @@ ca_bincmp_feq_float32_t (ca_size_t n, boolean8_t *m,
      }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_float32_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      {
+      ca_u_float32_t f1a = fabs((float64_t) *p1);
+      ca_u_float32_t f2a = fabs((float64_t) *p2);
+      ca_u_float32_t fmax = (f1a > f2a) ? f1a : f2a;
+      (*p3) = ( fabs(((float64_t) *p1)-( (float64_t) *p2)) <= fmax * FLT_EPSILON ) ? 1 : 0;
+     }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_float32_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      {
+      ca_u_float32_t f1a = fabs((float64_t) *p1);
+      ca_u_float32_t f2a = fabs((float64_t) *p2);
+      ca_u_float32_t fmax = (f1a > f2a) ? f1a : f2a;
+      (*p3) = ( fabs(((float64_t) *p1)-( (float64_t) *p2)) <= fmax * FLT_EPSILON ) ? 1 : 0;
+     }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -120,6 +148,34 @@ ca_bincmp_feq_float64_t (ca_size_t n, boolean8_t *m,
      }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_float64_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      {
+      ca_u_float64_t f1a = fabs((float64_t) *p1);
+      ca_u_float64_t f2a = fabs((float64_t) *p2);
+      ca_u_float64_t fmax = (f1a > f2a) ? f1a : f2a;
+      (*p3) = ( fabs(((float64_t) *p1)-( (float64_t) *p2)) <= fmax * DBL_EPSILON ) ? 1 : 0;
+     }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_float64_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      {
+      ca_u_float64_t f1a = fabs((float64_t) *p1);
+      ca_u_float64_t f2a = fabs((float64_t) *p2);
+      ca_u_float64_t fmax = (f1a > f2a) ? f1a : f2a;
+      (*p3) = ( fabs(((float64_t) *p1)-( (float64_t) *p2)) <= fmax * DBL_EPSILON ) ? 1 : 0;
+     }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -226,6 +282,24 @@ ca_bincmp_eq_boolean8_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) == (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    boolean8_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) == (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    boolean8_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) == (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -261,6 +335,24 @@ ca_bincmp_eq_int8_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) == (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    int8_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) == (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    int8_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) == (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -296,6 +388,24 @@ ca_bincmp_eq_uint8_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) == (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    uint8_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) == (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    uint8_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) == (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -331,6 +441,24 @@ ca_bincmp_eq_int16_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) == (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_int16_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) == (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_int16_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) == (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -366,6 +494,24 @@ ca_bincmp_eq_uint16_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) == (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_uint16_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) == (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_uint16_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) == (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -401,6 +547,24 @@ ca_bincmp_eq_int32_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) == (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_int32_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) == (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_int32_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) == (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -436,6 +600,24 @@ ca_bincmp_eq_uint32_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) == (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_uint32_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) == (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_uint32_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) == (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -471,6 +653,24 @@ ca_bincmp_eq_int64_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) == (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_int64_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) == (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_int64_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) == (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -506,6 +706,24 @@ ca_bincmp_eq_uint64_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) == (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_uint64_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) == (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_uint64_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) == (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -541,6 +759,24 @@ ca_bincmp_eq_float32_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) == (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_float32_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) == (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_float32_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) == (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -576,6 +812,24 @@ ca_bincmp_eq_float64_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) == (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_float64_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) == (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_float64_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) == (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -611,6 +865,24 @@ ca_bincmp_eq_VALUE (ca_size_t n, boolean8_t *m,
         { (*p3) = RTEST(rb_funcall((*p1), rb_intern("=="), 1, (*p2))) ? 1 : 0; }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    VALUE sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = RTEST(rb_funcall((*p1), rb_intern("=="), 1, (*p2))) ? 1 : 0; }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    VALUE sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = RTEST(rb_funcall((*p1), rb_intern("=="), 1, (*p2))) ? 1 : 0; }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -646,6 +918,24 @@ ca_bincmp_eq_cmplx64_t (ca_size_t n, boolean8_t *m,
         { (*p3) = (*p1) == (*p2); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_cmplx64_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = (*p1) == (*p2); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_cmplx64_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = (*p1) == (*p2); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -681,6 +971,24 @@ ca_bincmp_eq_cmplx128_t (ca_size_t n, boolean8_t *m,
         { (*p3) = (*p1) == (*p2); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_cmplx128_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = (*p1) == (*p2); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_cmplx128_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = (*p1) == (*p2); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -782,6 +1090,24 @@ ca_bincmp_ne_boolean8_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) != (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    boolean8_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) != (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    boolean8_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) != (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -817,6 +1143,24 @@ ca_bincmp_ne_int8_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) != (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    int8_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) != (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    int8_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) != (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -852,6 +1196,24 @@ ca_bincmp_ne_uint8_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) != (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    uint8_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) != (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    uint8_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) != (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -887,6 +1249,24 @@ ca_bincmp_ne_int16_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) != (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_int16_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) != (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_int16_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) != (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -922,6 +1302,24 @@ ca_bincmp_ne_uint16_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) != (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_uint16_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) != (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_uint16_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) != (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -957,6 +1355,24 @@ ca_bincmp_ne_int32_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) != (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_int32_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) != (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_int32_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) != (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -992,6 +1408,24 @@ ca_bincmp_ne_uint32_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) != (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_uint32_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) != (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_uint32_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) != (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -1027,6 +1461,24 @@ ca_bincmp_ne_int64_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) != (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_int64_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) != (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_int64_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) != (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -1062,6 +1514,24 @@ ca_bincmp_ne_uint64_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) != (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_uint64_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) != (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_uint64_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) != (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -1097,6 +1567,24 @@ ca_bincmp_ne_float32_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) != (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_float32_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) != (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_float32_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) != (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -1132,6 +1620,24 @@ ca_bincmp_ne_float64_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) != (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_float64_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) != (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_float64_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) != (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -1167,6 +1673,24 @@ ca_bincmp_ne_VALUE (ca_size_t n, boolean8_t *m,
         { (*p3) = RTEST(rb_funcall((*p1), rb_intern("=="), 1, (*p2))) ? 0 : 1; }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    VALUE sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = RTEST(rb_funcall((*p1), rb_intern("=="), 1, (*p2))) ? 0 : 1; }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    VALUE sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = RTEST(rb_funcall((*p1), rb_intern("=="), 1, (*p2))) ? 0 : 1; }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -1202,6 +1726,24 @@ ca_bincmp_ne_cmplx64_t (ca_size_t n, boolean8_t *m,
         { (*p3) = (*p1) != (*p2); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_cmplx64_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = (*p1) != (*p2); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_cmplx64_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = (*p1) != (*p2); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -1237,6 +1779,24 @@ ca_bincmp_ne_cmplx128_t (ca_size_t n, boolean8_t *m,
         { (*p3) = (*p1) != (*p2); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_cmplx128_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = (*p1) != (*p2); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_cmplx128_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = (*p1) != (*p2); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -1344,6 +1904,24 @@ ca_bincmp_gt_boolean8_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) > (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    boolean8_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) > (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    boolean8_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) > (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -1379,6 +1957,24 @@ ca_bincmp_gt_int8_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) > (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    int8_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) > (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    int8_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) > (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -1414,6 +2010,24 @@ ca_bincmp_gt_uint8_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) > (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    uint8_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) > (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    uint8_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) > (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -1449,6 +2063,24 @@ ca_bincmp_gt_int16_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) > (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_int16_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) > (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_int16_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) > (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -1484,6 +2116,24 @@ ca_bincmp_gt_uint16_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) > (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_uint16_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) > (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_uint16_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) > (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -1519,6 +2169,24 @@ ca_bincmp_gt_int32_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) > (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_int32_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) > (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_int32_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) > (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -1554,6 +2222,24 @@ ca_bincmp_gt_uint32_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) > (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_uint32_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) > (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_uint32_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) > (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -1589,6 +2275,24 @@ ca_bincmp_gt_int64_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) > (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_int64_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) > (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_int64_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) > (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -1624,6 +2328,24 @@ ca_bincmp_gt_uint64_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) > (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_uint64_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) > (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_uint64_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) > (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -1659,10 +2381,28 @@ ca_bincmp_gt_float32_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) > (*p2) ); }
       }
     }
-  } else {
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_float32_t sv = *q2;
+    p2 = &sv;
     for (k=0; k<n; k++) {
-      p1 = q1 + k*i1;
-      p2 = q2 + k*i2;
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) > (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_float32_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) > (*p2) ); }
+    }
+  } else {
+    for (k=0; k<n; k++) {
+      p1 = q1 + k*i1;
+      p2 = q2 + k*i2;
       p3 = q3 + k*i3;
       { (*p3) = ( (*p1) > (*p2) ); }
     }
@@ -1694,6 +2434,24 @@ ca_bincmp_gt_float64_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) > (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_float64_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) > (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_float64_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) > (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -1729,6 +2487,24 @@ ca_bincmp_gt_VALUE (ca_size_t n, boolean8_t *m,
         { (*p3) = RTEST(rb_funcall((*p1), rb_intern(">"), 1, (*p2))) ? 1 : 0; }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    VALUE sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = RTEST(rb_funcall((*p1), rb_intern(">"), 1, (*p2))) ? 1 : 0; }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    VALUE sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = RTEST(rb_funcall((*p1), rb_intern(">"), 1, (*p2))) ? 1 : 0; }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -1836,6 +2612,24 @@ ca_bincmp_lt_boolean8_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) < (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    boolean8_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) < (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    boolean8_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) < (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -1871,6 +2665,24 @@ ca_bincmp_lt_int8_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) < (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    int8_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) < (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    int8_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) < (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -1906,6 +2718,24 @@ ca_bincmp_lt_uint8_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) < (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    uint8_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) < (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    uint8_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) < (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -1941,6 +2771,24 @@ ca_bincmp_lt_int16_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) < (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_int16_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) < (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_int16_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) < (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -1976,6 +2824,24 @@ ca_bincmp_lt_uint16_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) < (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_uint16_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) < (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_uint16_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) < (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -2011,6 +2877,24 @@ ca_bincmp_lt_int32_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) < (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_int32_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) < (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_int32_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) < (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -2046,6 +2930,24 @@ ca_bincmp_lt_uint32_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) < (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_uint32_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) < (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_uint32_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) < (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -2081,6 +2983,24 @@ ca_bincmp_lt_int64_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) < (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_int64_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) < (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_int64_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) < (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -2116,6 +3036,24 @@ ca_bincmp_lt_uint64_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) < (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_uint64_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) < (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_uint64_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) < (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -2151,6 +3089,24 @@ ca_bincmp_lt_float32_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) < (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_float32_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) < (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_float32_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) < (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -2186,6 +3142,24 @@ ca_bincmp_lt_float64_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) < (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_float64_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) < (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_float64_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) < (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -2221,6 +3195,24 @@ ca_bincmp_lt_VALUE (ca_size_t n, boolean8_t *m,
         { (*p3) = RTEST(rb_funcall((*p1), rb_intern("<"), 1, (*p2))) ? 1 : 0; }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    VALUE sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = RTEST(rb_funcall((*p1), rb_intern("<"), 1, (*p2))) ? 1 : 0; }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    VALUE sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = RTEST(rb_funcall((*p1), rb_intern("<"), 1, (*p2))) ? 1 : 0; }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -2328,6 +3320,24 @@ ca_bincmp_ge_boolean8_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) >= (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    boolean8_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) >= (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    boolean8_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) >= (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -2363,6 +3373,24 @@ ca_bincmp_ge_int8_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) >= (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    int8_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) >= (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    int8_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) >= (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -2398,6 +3426,24 @@ ca_bincmp_ge_uint8_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) >= (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    uint8_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) >= (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    uint8_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) >= (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -2433,6 +3479,24 @@ ca_bincmp_ge_int16_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) >= (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_int16_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) >= (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_int16_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) >= (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -2468,6 +3532,24 @@ ca_bincmp_ge_uint16_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) >= (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_uint16_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) >= (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_uint16_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) >= (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -2503,6 +3585,24 @@ ca_bincmp_ge_int32_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) >= (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_int32_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) >= (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_int32_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) >= (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -2538,6 +3638,24 @@ ca_bincmp_ge_uint32_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) >= (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_uint32_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) >= (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_uint32_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) >= (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -2573,6 +3691,24 @@ ca_bincmp_ge_int64_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) >= (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_int64_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) >= (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_int64_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) >= (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -2608,6 +3744,24 @@ ca_bincmp_ge_uint64_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) >= (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_uint64_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) >= (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_uint64_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) >= (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -2643,6 +3797,24 @@ ca_bincmp_ge_float32_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) >= (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_float32_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) >= (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_float32_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) >= (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -2678,6 +3850,24 @@ ca_bincmp_ge_float64_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) >= (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_float64_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) >= (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_float64_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) >= (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -2713,6 +3903,24 @@ ca_bincmp_ge_VALUE (ca_size_t n, boolean8_t *m,
         { (*p3) = RTEST(rb_funcall((*p1), rb_intern(">="), 1, (*p2))) ? 1 : 0; }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    VALUE sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = RTEST(rb_funcall((*p1), rb_intern(">="), 1, (*p2))) ? 1 : 0; }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    VALUE sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = RTEST(rb_funcall((*p1), rb_intern(">="), 1, (*p2))) ? 1 : 0; }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -2820,6 +4028,24 @@ ca_bincmp_le_boolean8_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) <= (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    boolean8_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) <= (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    boolean8_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) <= (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -2855,6 +4081,24 @@ ca_bincmp_le_int8_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) <= (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    int8_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) <= (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    int8_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) <= (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -2890,6 +4134,24 @@ ca_bincmp_le_uint8_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) <= (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    uint8_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) <= (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    uint8_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) <= (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -2925,6 +4187,24 @@ ca_bincmp_le_int16_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) <= (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_int16_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) <= (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_int16_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) <= (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -2960,6 +4240,24 @@ ca_bincmp_le_uint16_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) <= (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_uint16_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) <= (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_uint16_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) <= (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -2995,6 +4293,24 @@ ca_bincmp_le_int32_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) <= (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_int32_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) <= (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_int32_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) <= (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -3030,6 +4346,24 @@ ca_bincmp_le_uint32_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) <= (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_uint32_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) <= (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_uint32_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) <= (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -3065,6 +4399,24 @@ ca_bincmp_le_int64_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) <= (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_int64_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) <= (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_int64_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) <= (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -3100,6 +4452,24 @@ ca_bincmp_le_uint64_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) <= (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_uint64_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) <= (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_uint64_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) <= (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -3135,6 +4505,24 @@ ca_bincmp_le_float32_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) <= (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_float32_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) <= (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_float32_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) <= (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -3170,6 +4558,24 @@ ca_bincmp_le_float64_t (ca_size_t n, boolean8_t *m,
         { (*p3) = ( (*p1) <= (*p2) ); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_float64_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) <= (*p2) ); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_float64_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = ( (*p1) <= (*p2) ); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -3205,6 +4611,24 @@ ca_bincmp_le_VALUE (ca_size_t n, boolean8_t *m,
         { (*p3) = RTEST(rb_funcall((*p1), rb_intern("<="), 1, (*p2))) ? 1 : 0; }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    VALUE sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = RTEST(rb_funcall((*p1), rb_intern("<="), 1, (*p2))) ? 1 : 0; }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    VALUE sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = RTEST(rb_funcall((*p1), rb_intern("<="), 1, (*p2))) ? 1 : 0; }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -3310,6 +4734,24 @@ ca_bincmp_match_VALUE (ca_size_t n, boolean8_t *m,
         { (*p3) = RTEST(rb_funcall((*p1), rb_intern("=~"), 1, (*p2))); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    VALUE sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = RTEST(rb_funcall((*p1), rb_intern("=~"), 1, (*p2))); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    VALUE sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = RTEST(rb_funcall((*p1), rb_intern("=~"), 1, (*p2))); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -3376,6 +4818,24 @@ ca_bincmp_is_kind_of_VALUE (ca_size_t n, boolean8_t *m,
         { (*p3) = RTEST(rb_obj_is_kind_of((*p1), (*p2))); }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    VALUE sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = RTEST(rb_obj_is_kind_of((*p1), (*p2))); }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    VALUE sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = RTEST(rb_obj_is_kind_of((*p1), (*p2))); }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -3442,6 +4902,24 @@ ca_bincmp_is_close_int8_t (ca_size_t n, boolean8_t *m,
         { (*p3) = (fabs((double)(*p1) - (double)(*p2)) <= tol) ? 1 : 0; }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    int8_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = (fabs((double)(*p1) - (double)(*p2)) <= tol) ? 1 : 0; }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    int8_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = (fabs((double)(*p1) - (double)(*p2)) <= tol) ? 1 : 0; }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -3477,6 +4955,24 @@ ca_bincmp_is_close_uint8_t (ca_size_t n, boolean8_t *m,
         { (*p3) = (fabs((double)(*p1) - (double)(*p2)) <= tol) ? 1 : 0; }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    uint8_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = (fabs((double)(*p1) - (double)(*p2)) <= tol) ? 1 : 0; }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    uint8_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = (fabs((double)(*p1) - (double)(*p2)) <= tol) ? 1 : 0; }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -3512,6 +5008,24 @@ ca_bincmp_is_close_int16_t (ca_size_t n, boolean8_t *m,
         { (*p3) = (fabs((double)(*p1) - (double)(*p2)) <= tol) ? 1 : 0; }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_int16_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = (fabs((double)(*p1) - (double)(*p2)) <= tol) ? 1 : 0; }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_int16_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = (fabs((double)(*p1) - (double)(*p2)) <= tol) ? 1 : 0; }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -3547,6 +5061,24 @@ ca_bincmp_is_close_uint16_t (ca_size_t n, boolean8_t *m,
         { (*p3) = (fabs((double)(*p1) - (double)(*p2)) <= tol) ? 1 : 0; }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_uint16_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = (fabs((double)(*p1) - (double)(*p2)) <= tol) ? 1 : 0; }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_uint16_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = (fabs((double)(*p1) - (double)(*p2)) <= tol) ? 1 : 0; }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -3582,6 +5114,24 @@ ca_bincmp_is_close_int32_t (ca_size_t n, boolean8_t *m,
         { (*p3) = (fabs((double)(*p1) - (double)(*p2)) <= tol) ? 1 : 0; }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_int32_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = (fabs((double)(*p1) - (double)(*p2)) <= tol) ? 1 : 0; }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_int32_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = (fabs((double)(*p1) - (double)(*p2)) <= tol) ? 1 : 0; }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -3617,6 +5167,24 @@ ca_bincmp_is_close_uint32_t (ca_size_t n, boolean8_t *m,
         { (*p3) = (fabs((double)(*p1) - (double)(*p2)) <= tol) ? 1 : 0; }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_uint32_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = (fabs((double)(*p1) - (double)(*p2)) <= tol) ? 1 : 0; }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_uint32_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = (fabs((double)(*p1) - (double)(*p2)) <= tol) ? 1 : 0; }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -3652,6 +5220,24 @@ ca_bincmp_is_close_int64_t (ca_size_t n, boolean8_t *m,
         { (*p3) = (fabs((double)(*p1) - (double)(*p2)) <= tol) ? 1 : 0; }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_int64_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = (fabs((double)(*p1) - (double)(*p2)) <= tol) ? 1 : 0; }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_int64_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = (fabs((double)(*p1) - (double)(*p2)) <= tol) ? 1 : 0; }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -3687,6 +5273,24 @@ ca_bincmp_is_close_uint64_t (ca_size_t n, boolean8_t *m,
         { (*p3) = (fabs((double)(*p1) - (double)(*p2)) <= tol) ? 1 : 0; }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_uint64_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = (fabs((double)(*p1) - (double)(*p2)) <= tol) ? 1 : 0; }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_uint64_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = (fabs((double)(*p1) - (double)(*p2)) <= tol) ? 1 : 0; }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -3722,6 +5326,24 @@ ca_bincmp_is_close_float32_t (ca_size_t n, boolean8_t *m,
         { (*p3) = (fabs((*p1) - (*p2)) <= tol) ? 1 : 0; }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_float32_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = (fabs((*p1) - (*p2)) <= tol) ? 1 : 0; }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_float32_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = (fabs((*p1) - (*p2)) <= tol) ? 1 : 0; }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -3757,6 +5379,24 @@ ca_bincmp_is_close_float64_t (ca_size_t n, boolean8_t *m,
         { (*p3) = (fabs((*p1) - (*p2)) <= tol) ? 1 : 0; }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_float64_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = (fabs((*p1) - (*p2)) <= tol) ? 1 : 0; }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_float64_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = (fabs((*p1) - (*p2)) <= tol) ? 1 : 0; }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -3792,6 +5432,24 @@ ca_bincmp_is_close_cmplx64_t (ca_size_t n, boolean8_t *m,
         { (*p3) = (cabs((*p1) - (*p2)) <= tol) ? 1 : 0; }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_cmplx64_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = (cabs((*p1) - (*p2)) <= tol) ? 1 : 0; }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_cmplx64_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = (cabs((*p1) - (*p2)) <= tol) ? 1 : 0; }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -3827,6 +5485,24 @@ ca_bincmp_is_close_cmplx128_t (ca_size_t n, boolean8_t *m,
         { (*p3) = (cabs((*p1) - (*p2)) <= tol) ? 1 : 0; }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_cmplx128_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      { (*p3) = (cabs((*p1) - (*p2)) <= tol) ? 1 : 0; }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_cmplx128_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      { (*p3) = (cabs((*p1) - (*p2)) <= tol) ? 1 : 0; }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -3899,6 +5575,34 @@ ca_bincmp_is_equiv_int8_t (ca_size_t n, boolean8_t *m,
      }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    int8_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      {
+      double a = (double)(*p1), b = (double)(*p2);
+      double aa = fabs(a), ab = fabs(b);
+      double m = (aa > ab) ? aa : ab;
+      (*p3) = ( (*p1) == (*p2) || (m > 0 && fabs(a - b) / m <= tol) ) ? 1 : 0;
+     }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    int8_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      {
+      double a = (double)(*p1), b = (double)(*p2);
+      double aa = fabs(a), ab = fabs(b);
+      double m = (aa > ab) ? aa : ab;
+      (*p3) = ( (*p1) == (*p2) || (m > 0 && fabs(a - b) / m <= tol) ) ? 1 : 0;
+     }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -3944,6 +5648,34 @@ ca_bincmp_is_equiv_uint8_t (ca_size_t n, boolean8_t *m,
      }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    uint8_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      {
+      double a = (double)(*p1), b = (double)(*p2);
+      double aa = fabs(a), ab = fabs(b);
+      double m = (aa > ab) ? aa : ab;
+      (*p3) = ( (*p1) == (*p2) || (m > 0 && fabs(a - b) / m <= tol) ) ? 1 : 0;
+     }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    uint8_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      {
+      double a = (double)(*p1), b = (double)(*p2);
+      double aa = fabs(a), ab = fabs(b);
+      double m = (aa > ab) ? aa : ab;
+      (*p3) = ( (*p1) == (*p2) || (m > 0 && fabs(a - b) / m <= tol) ) ? 1 : 0;
+     }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -3989,6 +5721,34 @@ ca_bincmp_is_equiv_int16_t (ca_size_t n, boolean8_t *m,
      }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_int16_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      {
+      double a = (double)(*p1), b = (double)(*p2);
+      double aa = fabs(a), ab = fabs(b);
+      double m = (aa > ab) ? aa : ab;
+      (*p3) = ( (*p1) == (*p2) || (m > 0 && fabs(a - b) / m <= tol) ) ? 1 : 0;
+     }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_int16_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      {
+      double a = (double)(*p1), b = (double)(*p2);
+      double aa = fabs(a), ab = fabs(b);
+      double m = (aa > ab) ? aa : ab;
+      (*p3) = ( (*p1) == (*p2) || (m > 0 && fabs(a - b) / m <= tol) ) ? 1 : 0;
+     }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -4034,6 +5794,34 @@ ca_bincmp_is_equiv_uint16_t (ca_size_t n, boolean8_t *m,
      }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_uint16_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      {
+      double a = (double)(*p1), b = (double)(*p2);
+      double aa = fabs(a), ab = fabs(b);
+      double m = (aa > ab) ? aa : ab;
+      (*p3) = ( (*p1) == (*p2) || (m > 0 && fabs(a - b) / m <= tol) ) ? 1 : 0;
+     }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_uint16_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      {
+      double a = (double)(*p1), b = (double)(*p2);
+      double aa = fabs(a), ab = fabs(b);
+      double m = (aa > ab) ? aa : ab;
+      (*p3) = ( (*p1) == (*p2) || (m > 0 && fabs(a - b) / m <= tol) ) ? 1 : 0;
+     }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -4079,6 +5867,34 @@ ca_bincmp_is_equiv_int32_t (ca_size_t n, boolean8_t *m,
      }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_int32_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      {
+      double a = (double)(*p1), b = (double)(*p2);
+      double aa = fabs(a), ab = fabs(b);
+      double m = (aa > ab) ? aa : ab;
+      (*p3) = ( (*p1) == (*p2) || (m > 0 && fabs(a - b) / m <= tol) ) ? 1 : 0;
+     }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_int32_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      {
+      double a = (double)(*p1), b = (double)(*p2);
+      double aa = fabs(a), ab = fabs(b);
+      double m = (aa > ab) ? aa : ab;
+      (*p3) = ( (*p1) == (*p2) || (m > 0 && fabs(a - b) / m <= tol) ) ? 1 : 0;
+     }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -4124,6 +5940,34 @@ ca_bincmp_is_equiv_uint32_t (ca_size_t n, boolean8_t *m,
      }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_uint32_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      {
+      double a = (double)(*p1), b = (double)(*p2);
+      double aa = fabs(a), ab = fabs(b);
+      double m = (aa > ab) ? aa : ab;
+      (*p3) = ( (*p1) == (*p2) || (m > 0 && fabs(a - b) / m <= tol) ) ? 1 : 0;
+     }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_uint32_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      {
+      double a = (double)(*p1), b = (double)(*p2);
+      double aa = fabs(a), ab = fabs(b);
+      double m = (aa > ab) ? aa : ab;
+      (*p3) = ( (*p1) == (*p2) || (m > 0 && fabs(a - b) / m <= tol) ) ? 1 : 0;
+     }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -4169,6 +6013,34 @@ ca_bincmp_is_equiv_int64_t (ca_size_t n, boolean8_t *m,
      }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_int64_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      {
+      double a = (double)(*p1), b = (double)(*p2);
+      double aa = fabs(a), ab = fabs(b);
+      double m = (aa > ab) ? aa : ab;
+      (*p3) = ( (*p1) == (*p2) || (m > 0 && fabs(a - b) / m <= tol) ) ? 1 : 0;
+     }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_int64_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      {
+      double a = (double)(*p1), b = (double)(*p2);
+      double aa = fabs(a), ab = fabs(b);
+      double m = (aa > ab) ? aa : ab;
+      (*p3) = ( (*p1) == (*p2) || (m > 0 && fabs(a - b) / m <= tol) ) ? 1 : 0;
+     }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -4214,6 +6086,34 @@ ca_bincmp_is_equiv_uint64_t (ca_size_t n, boolean8_t *m,
      }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_uint64_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      {
+      double a = (double)(*p1), b = (double)(*p2);
+      double aa = fabs(a), ab = fabs(b);
+      double m = (aa > ab) ? aa : ab;
+      (*p3) = ( (*p1) == (*p2) || (m > 0 && fabs(a - b) / m <= tol) ) ? 1 : 0;
+     }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_uint64_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      {
+      double a = (double)(*p1), b = (double)(*p2);
+      double aa = fabs(a), ab = fabs(b);
+      double m = (aa > ab) ? aa : ab;
+      (*p3) = ( (*p1) == (*p2) || (m > 0 && fabs(a - b) / m <= tol) ) ? 1 : 0;
+     }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -4258,6 +6158,32 @@ ca_bincmp_is_equiv_float32_t (ca_size_t n, boolean8_t *m,
      }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_float32_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      {
+      ca_u_float32_t aa = fabs((*p1)), ab = fabs((*p2));
+      ca_u_float32_t m = (aa > ab) ? aa : ab;
+      (*p3) = ( (*p1) == (*p2) || (m > 0 && fabs((*p1) - (*p2)) / m <= tol) ) ? 1 : 0;
+     }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_float32_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      {
+      ca_u_float32_t aa = fabs((*p1)), ab = fabs((*p2));
+      ca_u_float32_t m = (aa > ab) ? aa : ab;
+      (*p3) = ( (*p1) == (*p2) || (m > 0 && fabs((*p1) - (*p2)) / m <= tol) ) ? 1 : 0;
+     }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -4301,6 +6227,32 @@ ca_bincmp_is_equiv_float64_t (ca_size_t n, boolean8_t *m,
      }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_float64_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      {
+      ca_u_float64_t aa = fabs((*p1)), ab = fabs((*p2));
+      ca_u_float64_t m = (aa > ab) ? aa : ab;
+      (*p3) = ( (*p1) == (*p2) || (m > 0 && fabs((*p1) - (*p2)) / m <= tol) ) ? 1 : 0;
+     }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_float64_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      {
+      ca_u_float64_t aa = fabs((*p1)), ab = fabs((*p2));
+      ca_u_float64_t m = (aa > ab) ? aa : ab;
+      (*p3) = ( (*p1) == (*p2) || (m > 0 && fabs((*p1) - (*p2)) / m <= tol) ) ? 1 : 0;
+     }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -4344,6 +6296,32 @@ ca_bincmp_is_equiv_cmplx64_t (ca_size_t n, boolean8_t *m,
      }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_cmplx64_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      {
+      double aa = cabs((*p1)), ab = cabs((*p2));
+      double m = (aa > ab) ? aa : ab;
+      (*p3) = ( m > 0 && cabs((*p1) - (*p2)) / m <= tol ) ? 1 : 0;
+     }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_cmplx64_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      {
+      double aa = cabs((*p1)), ab = cabs((*p2));
+      double m = (aa > ab) ? aa : ab;
+      (*p3) = ( m > 0 && cabs((*p1) - (*p2)) / m <= tol ) ? 1 : 0;
+     }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;
@@ -4387,6 +6365,32 @@ ca_bincmp_is_equiv_cmplx128_t (ca_size_t n, boolean8_t *m,
      }
       }
     }
+  } else if ( i3 == 1 && i1 == 1 && i2 == 0 ) {
+    /* scalar on the right: read it once, so the boolean store
+       cannot be taken to change it and the loop vectorizes */
+    ca_u_cmplx128_t sv = *q2;
+    p2 = &sv;
+    for (k=0; k<n; k++) {
+      p1 = q1 + k;
+      p3 = q3 + k;
+      {
+      double aa = cabs((*p1)), ab = cabs((*p2));
+      double m = (aa > ab) ? aa : ab;
+      (*p3) = ( m > 0 && cabs((*p1) - (*p2)) / m <= tol ) ? 1 : 0;
+     }
+    }
+  } else if ( i3 == 1 && i1 == 0 && i2 == 1 ) {
+    ca_u_cmplx128_t sv = *q1;
+    p1 = &sv;
+    for (k=0; k<n; k++) {
+      p2 = q2 + k;
+      p3 = q3 + k;
+      {
+      double aa = cabs((*p1)), ab = cabs((*p2));
+      double m = (aa > ab) ? aa : ab;
+      (*p3) = ( m > 0 && cabs((*p1) - (*p2)) / m <= tol ) ? 1 : 0;
+     }
+    }
   } else {
     for (k=0; k<n; k++) {
       p1 = q1 + k*i1;