RubyGems - carray - Versions diffs - 3.0.4 → 3.0.6 - Mend

carray 3.0.4 → 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 (155) hide show

checksums.yaml +4 -4
data/CHANGELOG.md +438 -0
data/README.md +1 -1
data/carray.gemspec +3 -2
data/ext/ca_array_pool.c +25 -1
data/ext/ca_axis_dispatch.c +19 -20
data/ext/ca_axis_group.c +1 -1
data/ext/ca_categorical_iterator.c +34 -27
data/ext/ca_compare.c +4 -4
data/ext/ca_composite_dispatch.c +9 -9
data/ext/ca_kernel_iterator.c +122 -237
data/ext/ca_kernel_iterator.h +522 -337
data/ext/ca_obj_array.c +69 -19
data/ext/ca_obj_bincmp.c +10 -1
data/ext/ca_obj_binop.c +11 -1
data/ext/ca_obj_bitarray.c +5 -1
data/ext/ca_obj_bitfield.c +60 -43
data/ext/ca_obj_byte_swap.c +4 -1
data/ext/ca_obj_const_string.c +15 -3
data/ext/ca_obj_face.c +44 -1
data/ext/ca_obj_fake.c +3 -1
data/ext/ca_obj_field.c +13 -3
data/ext/ca_obj_grid.c +219 -17
data/ext/ca_obj_meld.c +69 -113
data/ext/ca_obj_moncmp.c +9 -1
data/ext/ca_obj_monop.c +86 -12
data/ext/ca_obj_object.c +118 -32
data/ext/ca_obj_record.c +3 -0
data/ext/ca_obj_reduce.c +19 -1
data/ext/ca_obj_refer.c +80 -26
data/ext/ca_obj_remap.c +69 -10
data/ext/ca_obj_roll.c +27 -6
data/ext/ca_obj_select.c +38 -6
data/ext/ca_obj_select_axis.c +1 -0
data/ext/ca_obj_shift.c +32 -7
data/ext/ca_obj_stack.c +311 -111
data/ext/ca_obj_stride.c +159 -64
data/ext/ca_obj_string.c +27 -2
data/ext/ca_obj_tile.c +28 -7
data/ext/ca_obj_transpose.c +8 -4
data/ext/ca_obj_triop.c +12 -1
data/ext/ca_obj_window.c +80 -50
data/ext/ca_op_ipower.c +12 -12
data/ext/ca_sort_kernels.h +6 -3
data/ext/caframe_csv_reader.c +1 -1
data/ext/carray.h +35 -1
data/ext/carray_access.c +43 -67
data/ext/carray_address_basis.c +134 -0
data/ext/carray_attribute.c +32 -0
data/ext/carray_broadcast.c +3 -4
data/ext/carray_build_flags.h +1 -1
data/ext/carray_cast.c +251 -93
data/ext/carray_cast_func.rb +56 -26
data/ext/carray_conversion.c +11 -7
data/ext/carray_core.c +814 -72
data/ext/carray_count.c +6 -0
data/ext/carray_element.c +22 -13
data/ext/carray_factorize.c +20 -20
data/ext/carray_generate.c +8 -2
data/ext/carray_histogram.c +26 -4
data/ext/carray_index_classifier.c +4 -4
data/ext/carray_internal.h +159 -12
data/ext/carray_kernels_bincmp.c +2174 -170
data/ext/carray_kernels_binop.c +25653 -5487
data/ext/carray_kernels_init.c +8 -8
data/ext/carray_kernels_moncmp.c +96 -96
data/ext/carray_kernels_monop.c +12253 -5235
data/ext/carray_kernels_reduce_aggregate.c +6150 -2423
data/ext/carray_kernels_reduce_boolean.c +1533 -15
data/ext/carray_kernels_reduce_cumulative.c +2442 -882
data/ext/carray_kernels_reduce_extreme.c +5658 -1661
data/ext/carray_kernels_reduce_variance.c +2534 -402
data/ext/carray_kernels_scan.c +368 -368
data/ext/carray_kernels_search.c +4551 -604
data/ext/carray_kernels_sort.c +659 -751
data/ext/carray_kernels_triop.c +114 -114
data/ext/carray_key_windows.c +4 -0
data/ext/carray_lazy.c +47 -0
data/ext/carray_loop.c +1 -1
data/ext/carray_mask.c +175 -161
data/ext/carray_median_percentile.c +4 -4
data/ext/carray_memory_view.c +15 -9
data/ext/carray_operator.c +152 -49
data/ext/carray_order.c +230 -10
data/ext/carray_random.c +22 -19
data/ext/carray_scatter.c +24 -7
data/ext/carray_slab.c +128 -29
data/ext/carray_sort.c +224 -16
data/ext/carray_sort_kernel.c +5 -7
data/ext/carray_struct.c +35 -43
data/ext/carray_test.c +74 -19
data/ext/carray_time_text.c +6 -4
data/ext/carray_utils.c +63 -21
data/ext/extconf.rb +11 -0
data/ext/mkkernel.rb +546 -199
data/ext/ruby_carray.c +4 -1
data/ext/version.h +4 -4
data/lib/carray/attribute.rb +42 -5
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 +28 -8
data/lib/carray/block_iterator.rb +17 -3
data/lib/carray/categorical.rb +35 -23
data/lib/carray/categorical_iterator.rb +8 -8
data/lib/carray/complex.rb +2 -2
data/lib/carray/conditional.rb +3 -2
data/lib/carray/const_string.rb +10 -0
data/lib/carray/data_type_extension.rb +8 -1
data/lib/carray/fixlen_string.rb +41 -2
data/lib/carray/frame/column_verbs.rb +48 -16
data/lib/carray/frame/concat.rb +17 -3
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 +405 -85
data/lib/carray/frame/group.rb +777 -64
data/lib/carray/frame/io.rb +212 -72
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/fuse_source.rb +31 -2
data/lib/carray/inspect.rb +7 -2
data/lib/carray/iterator.rb +32 -0
data/lib/carray/lazy.rb +52 -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/is_in.rb +8 -1
data/lib/carray/methods/meshgrid.rb +3 -1
data/lib/carray/runtime.rb +47 -8
data/lib/carray/segment_iterator.rb +35 -16
data/lib/carray/serialize.rb +53 -19
data/lib/carray/stack.rb +33 -18
data/lib/carray/string.rb +8 -0
data/lib/carray/string_operation_extension.rb +13 -5
data/lib/carray/struct_builder.rb +5 -3
data/lib/carray/time.rb +112 -68
data/lib/carray/window_iterator.rb +16 -4
data/yard-stubs/ca_obj_array.rb +26 -1
data/yard-stubs/ca_obj_transpose.rb +5 -4
data/yard-stubs/carray_attribute.rb +3 -1
data/yard-stubs/carray_cast.rb +10 -5
data/yard-stubs/carray_conversion.rb +13 -3
data/yard-stubs/carray_element.rb +7 -6
data/yard-stubs/carray_mask.rb +28 -4
data/yard-stubs/carray_math.rb +45 -20
data/yard-stubs/carray_memory_view.rb +7 -0
data/yard-stubs/carray_order.rb +100 -4
data/yard-stubs/carray_scatter.rb +7 -5
data/yard-stubs/carray_test.rb +7 -4
metadata +18 -4

data/ext/carray_kernels_bincmp.c CHANGED Viewed

@@ -52,8 +52,8 @@ ca_bincmp_feq_float32_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  float *q1 = (float *) ptr1, *q2 = (float *) ptr2;
-  float *p1 = q1, *p2 = q2;
+  ca_u_float32_t *q1 = (ca_u_float32_t *) ptr1, *q2 = (ca_u_float32_t *) ptr2;
+  ca_u_float32_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -68,22 +68,50 @@ ca_bincmp_feq_float32_t (ca_size_t n, boolean8_t *m,
         p2 = q2 + k*i2;
         p3 = q3 + k*i3;
         {
-      float f1a = fabs((float64_t) *p1);
-      float f2a = fabs((float64_t) *p2);
-      float fmax = (f1a > f2a) ? f1a : f2a;
+      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 == 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;
       p2 = q2 + k*i2;
       p3 = q3 + k*i3;
       {
-      float f1a = fabs((float64_t) *p1);
-      float f2a = fabs((float64_t) *p2);
-      float fmax = (f1a > f2a) ? f1a : f2a;
+      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;
      }
     }
@@ -97,8 +125,8 @@ ca_bincmp_feq_float64_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  double *q1 = (double *) ptr1, *q2 = (double *) ptr2;
-  double *p1 = q1, *p2 = q2;
+  ca_u_float64_t *q1 = (ca_u_float64_t *) ptr1, *q2 = (ca_u_float64_t *) ptr2;
+  ca_u_float64_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -113,22 +141,50 @@ ca_bincmp_feq_float64_t (ca_size_t n, boolean8_t *m,
         p2 = q2 + k*i2;
         p3 = q3 + k*i3;
         {
-      double f1a = fabs((float64_t) *p1);
-      double f2a = fabs((float64_t) *p2);
-      double fmax = (f1a > f2a) ? f1a : f2a;
+      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 == 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;
       p2 = q2 + k*i2;
       p3 = q3 + k*i3;
       {
-      double f1a = fabs((float64_t) *p1);
-      double f2a = fabs((float64_t) *p2);
-      double fmax = (f1a > f2a) ? f1a : f2a;
+      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;
      }
     }
@@ -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;
@@ -313,8 +423,8 @@ ca_bincmp_eq_int16_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  int16_t *q1 = (int16_t *) ptr1, *q2 = (int16_t *) ptr2;
-  int16_t *p1 = q1, *p2 = q2;
+  ca_u_int16_t *q1 = (ca_u_int16_t *) ptr1, *q2 = (ca_u_int16_t *) ptr2;
+  ca_u_int16_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -348,8 +476,8 @@ ca_bincmp_eq_uint16_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  uint16_t *q1 = (uint16_t *) ptr1, *q2 = (uint16_t *) ptr2;
-  uint16_t *p1 = q1, *p2 = q2;
+  ca_u_uint16_t *q1 = (ca_u_uint16_t *) ptr1, *q2 = (ca_u_uint16_t *) ptr2;
+  ca_u_uint16_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -383,8 +529,8 @@ ca_bincmp_eq_int32_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  int32_t *q1 = (int32_t *) ptr1, *q2 = (int32_t *) ptr2;
-  int32_t *p1 = q1, *p2 = q2;
+  ca_u_int32_t *q1 = (ca_u_int32_t *) ptr1, *q2 = (ca_u_int32_t *) ptr2;
+  ca_u_int32_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -418,8 +582,8 @@ ca_bincmp_eq_uint32_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  uint32_t *q1 = (uint32_t *) ptr1, *q2 = (uint32_t *) ptr2;
-  uint32_t *p1 = q1, *p2 = q2;
+  ca_u_uint32_t *q1 = (ca_u_uint32_t *) ptr1, *q2 = (ca_u_uint32_t *) ptr2;
+  ca_u_uint32_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -453,8 +635,8 @@ ca_bincmp_eq_int64_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  int64_t *q1 = (int64_t *) ptr1, *q2 = (int64_t *) ptr2;
-  int64_t *p1 = q1, *p2 = q2;
+  ca_u_int64_t *q1 = (ca_u_int64_t *) ptr1, *q2 = (ca_u_int64_t *) ptr2;
+  ca_u_int64_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -488,8 +688,8 @@ ca_bincmp_eq_uint64_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  uint64_t *q1 = (uint64_t *) ptr1, *q2 = (uint64_t *) ptr2;
-  uint64_t *p1 = q1, *p2 = q2;
+  ca_u_uint64_t *q1 = (ca_u_uint64_t *) ptr1, *q2 = (ca_u_uint64_t *) ptr2;
+  ca_u_uint64_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -523,8 +741,8 @@ ca_bincmp_eq_float32_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  float *q1 = (float *) ptr1, *q2 = (float *) ptr2;
-  float *p1 = q1, *p2 = q2;
+  ca_u_float32_t *q1 = (ca_u_float32_t *) ptr1, *q2 = (ca_u_float32_t *) ptr2;
+  ca_u_float32_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -558,8 +794,8 @@ ca_bincmp_eq_float64_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  double *q1 = (double *) ptr1, *q2 = (double *) ptr2;
-  double *p1 = q1, *p2 = q2;
+  ca_u_float64_t *q1 = (ca_u_float64_t *) ptr1, *q2 = (ca_u_float64_t *) ptr2;
+  ca_u_float64_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -628,8 +900,8 @@ ca_bincmp_eq_cmplx64_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  cmplx64_t *q1 = (cmplx64_t *) ptr1, *q2 = (cmplx64_t *) ptr2;
-  cmplx64_t *p1 = q1, *p2 = q2;
+  ca_u_cmplx64_t *q1 = (ca_u_cmplx64_t *) ptr1, *q2 = (ca_u_cmplx64_t *) ptr2;
+  ca_u_cmplx64_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -663,8 +953,8 @@ ca_bincmp_eq_cmplx128_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  cmplx128_t *q1 = (cmplx128_t *) ptr1, *q2 = (cmplx128_t *) ptr2;
-  cmplx128_t *p1 = q1, *p2 = q2;
+  ca_u_cmplx128_t *q1 = (ca_u_cmplx128_t *) ptr1, *q2 = (ca_u_cmplx128_t *) ptr2;
+  ca_u_cmplx128_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -869,8 +1231,8 @@ ca_bincmp_ne_int16_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  int16_t *q1 = (int16_t *) ptr1, *q2 = (int16_t *) ptr2;
-  int16_t *p1 = q1, *p2 = q2;
+  ca_u_int16_t *q1 = (ca_u_int16_t *) ptr1, *q2 = (ca_u_int16_t *) ptr2;
+  ca_u_int16_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -904,8 +1284,8 @@ ca_bincmp_ne_uint16_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  uint16_t *q1 = (uint16_t *) ptr1, *q2 = (uint16_t *) ptr2;
-  uint16_t *p1 = q1, *p2 = q2;
+  ca_u_uint16_t *q1 = (ca_u_uint16_t *) ptr1, *q2 = (ca_u_uint16_t *) ptr2;
+  ca_u_uint16_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -939,8 +1337,8 @@ ca_bincmp_ne_int32_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  int32_t *q1 = (int32_t *) ptr1, *q2 = (int32_t *) ptr2;
-  int32_t *p1 = q1, *p2 = q2;
+  ca_u_int32_t *q1 = (ca_u_int32_t *) ptr1, *q2 = (ca_u_int32_t *) ptr2;
+  ca_u_int32_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -974,8 +1390,8 @@ ca_bincmp_ne_uint32_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  uint32_t *q1 = (uint32_t *) ptr1, *q2 = (uint32_t *) ptr2;
-  uint32_t *p1 = q1, *p2 = q2;
+  ca_u_uint32_t *q1 = (ca_u_uint32_t *) ptr1, *q2 = (ca_u_uint32_t *) ptr2;
+  ca_u_uint32_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -1009,8 +1443,8 @@ ca_bincmp_ne_int64_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  int64_t *q1 = (int64_t *) ptr1, *q2 = (int64_t *) ptr2;
-  int64_t *p1 = q1, *p2 = q2;
+  ca_u_int64_t *q1 = (ca_u_int64_t *) ptr1, *q2 = (ca_u_int64_t *) ptr2;
+  ca_u_int64_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -1044,8 +1496,8 @@ ca_bincmp_ne_uint64_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  uint64_t *q1 = (uint64_t *) ptr1, *q2 = (uint64_t *) ptr2;
-  uint64_t *p1 = q1, *p2 = q2;
+  ca_u_uint64_t *q1 = (ca_u_uint64_t *) ptr1, *q2 = (ca_u_uint64_t *) ptr2;
+  ca_u_uint64_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -1079,8 +1549,8 @@ ca_bincmp_ne_float32_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  float *q1 = (float *) ptr1, *q2 = (float *) ptr2;
-  float *p1 = q1, *p2 = q2;
+  ca_u_float32_t *q1 = (ca_u_float32_t *) ptr1, *q2 = (ca_u_float32_t *) ptr2;
+  ca_u_float32_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -1114,8 +1602,8 @@ ca_bincmp_ne_float64_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  double *q1 = (double *) ptr1, *q2 = (double *) ptr2;
-  double *p1 = q1, *p2 = q2;
+  ca_u_float64_t *q1 = (ca_u_float64_t *) ptr1, *q2 = (ca_u_float64_t *) ptr2;
+  ca_u_float64_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -1184,8 +1708,8 @@ ca_bincmp_ne_cmplx64_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  cmplx64_t *q1 = (cmplx64_t *) ptr1, *q2 = (cmplx64_t *) ptr2;
-  cmplx64_t *p1 = q1, *p2 = q2;
+  ca_u_cmplx64_t *q1 = (ca_u_cmplx64_t *) ptr1, *q2 = (ca_u_cmplx64_t *) ptr2;
+  ca_u_cmplx64_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -1219,8 +1761,8 @@ ca_bincmp_ne_cmplx128_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  cmplx128_t *q1 = (cmplx128_t *) ptr1, *q2 = (cmplx128_t *) ptr2;
-  cmplx128_t *p1 = q1, *p2 = q2;
+  ca_u_cmplx128_t *q1 = (ca_u_cmplx128_t *) ptr1, *q2 = (ca_u_cmplx128_t *) ptr2;
+  ca_u_cmplx128_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -1431,8 +2045,8 @@ ca_bincmp_gt_int16_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  int16_t *q1 = (int16_t *) ptr1, *q2 = (int16_t *) ptr2;
-  int16_t *p1 = q1, *p2 = q2;
+  ca_u_int16_t *q1 = (ca_u_int16_t *) ptr1, *q2 = (ca_u_int16_t *) ptr2;
+  ca_u_int16_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -1466,8 +2098,8 @@ ca_bincmp_gt_uint16_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  uint16_t *q1 = (uint16_t *) ptr1, *q2 = (uint16_t *) ptr2;
-  uint16_t *p1 = q1, *p2 = q2;
+  ca_u_uint16_t *q1 = (ca_u_uint16_t *) ptr1, *q2 = (ca_u_uint16_t *) ptr2;
+  ca_u_uint16_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -1501,8 +2151,8 @@ ca_bincmp_gt_int32_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  int32_t *q1 = (int32_t *) ptr1, *q2 = (int32_t *) ptr2;
-  int32_t *p1 = q1, *p2 = q2;
+  ca_u_int32_t *q1 = (ca_u_int32_t *) ptr1, *q2 = (ca_u_int32_t *) ptr2;
+  ca_u_int32_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -1536,8 +2204,8 @@ ca_bincmp_gt_uint32_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  uint32_t *q1 = (uint32_t *) ptr1, *q2 = (uint32_t *) ptr2;
-  uint32_t *p1 = q1, *p2 = q2;
+  ca_u_uint32_t *q1 = (ca_u_uint32_t *) ptr1, *q2 = (ca_u_uint32_t *) ptr2;
+  ca_u_uint32_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -1571,8 +2257,8 @@ ca_bincmp_gt_int64_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  int64_t *q1 = (int64_t *) ptr1, *q2 = (int64_t *) ptr2;
-  int64_t *p1 = q1, *p2 = q2;
+  ca_u_int64_t *q1 = (ca_u_int64_t *) ptr1, *q2 = (ca_u_int64_t *) ptr2;
+  ca_u_int64_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -1606,8 +2310,8 @@ ca_bincmp_gt_uint64_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  uint64_t *q1 = (uint64_t *) ptr1, *q2 = (uint64_t *) ptr2;
-  uint64_t *p1 = q1, *p2 = q2;
+  ca_u_uint64_t *q1 = (ca_u_uint64_t *) ptr1, *q2 = (ca_u_uint64_t *) ptr2;
+  ca_u_uint64_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -1641,8 +2363,8 @@ ca_bincmp_gt_float32_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  float *q1 = (float *) ptr1, *q2 = (float *) ptr2;
-  float *p1 = q1, *p2 = q2;
+  ca_u_float32_t *q1 = (ca_u_float32_t *) ptr1, *q2 = (ca_u_float32_t *) ptr2;
+  ca_u_float32_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -1659,6 +2381,24 @@ ca_bincmp_gt_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;
@@ -1676,8 +2416,8 @@ ca_bincmp_gt_float64_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  double *q1 = (double *) ptr1, *q2 = (double *) ptr2;
-  double *p1 = q1, *p2 = q2;
+  ca_u_float64_t *q1 = (ca_u_float64_t *) ptr1, *q2 = (ca_u_float64_t *) ptr2;
+  ca_u_float64_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -1729,9 +2487,27 @@ ca_bincmp_gt_VALUE (ca_size_t n, boolean8_t *m,
         { (*p3) = RTEST(rb_funcall((*p1), rb_intern(">"), 1, (*p2))) ? 1 : 0; }
       }
     }
-  } 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 */
+    VALUE sv = *q2;
+    p2 = &sv;
     for (k=0; k<n; k++) {
-      p1 = q1 + k*i1;
+      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;
       p2 = q2 + k*i2;
       p3 = q3 + k*i3;
       { (*p3) = RTEST(rb_funcall((*p1), rb_intern(">"), 1, (*p2))) ? 1 : 0; }
@@ -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;
@@ -1923,8 +2753,8 @@ ca_bincmp_lt_int16_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  int16_t *q1 = (int16_t *) ptr1, *q2 = (int16_t *) ptr2;
-  int16_t *p1 = q1, *p2 = q2;
+  ca_u_int16_t *q1 = (ca_u_int16_t *) ptr1, *q2 = (ca_u_int16_t *) ptr2;
+  ca_u_int16_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -1958,8 +2806,8 @@ ca_bincmp_lt_uint16_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  uint16_t *q1 = (uint16_t *) ptr1, *q2 = (uint16_t *) ptr2;
-  uint16_t *p1 = q1, *p2 = q2;
+  ca_u_uint16_t *q1 = (ca_u_uint16_t *) ptr1, *q2 = (ca_u_uint16_t *) ptr2;
+  ca_u_uint16_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -1993,8 +2859,8 @@ ca_bincmp_lt_int32_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  int32_t *q1 = (int32_t *) ptr1, *q2 = (int32_t *) ptr2;
-  int32_t *p1 = q1, *p2 = q2;
+  ca_u_int32_t *q1 = (ca_u_int32_t *) ptr1, *q2 = (ca_u_int32_t *) ptr2;
+  ca_u_int32_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -2028,8 +2912,8 @@ ca_bincmp_lt_uint32_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  uint32_t *q1 = (uint32_t *) ptr1, *q2 = (uint32_t *) ptr2;
-  uint32_t *p1 = q1, *p2 = q2;
+  ca_u_uint32_t *q1 = (ca_u_uint32_t *) ptr1, *q2 = (ca_u_uint32_t *) ptr2;
+  ca_u_uint32_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -2063,8 +2965,8 @@ ca_bincmp_lt_int64_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  int64_t *q1 = (int64_t *) ptr1, *q2 = (int64_t *) ptr2;
-  int64_t *p1 = q1, *p2 = q2;
+  ca_u_int64_t *q1 = (ca_u_int64_t *) ptr1, *q2 = (ca_u_int64_t *) ptr2;
+  ca_u_int64_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -2098,8 +3018,8 @@ ca_bincmp_lt_uint64_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  uint64_t *q1 = (uint64_t *) ptr1, *q2 = (uint64_t *) ptr2;
-  uint64_t *p1 = q1, *p2 = q2;
+  ca_u_uint64_t *q1 = (ca_u_uint64_t *) ptr1, *q2 = (ca_u_uint64_t *) ptr2;
+  ca_u_uint64_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -2133,8 +3071,8 @@ ca_bincmp_lt_float32_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  float *q1 = (float *) ptr1, *q2 = (float *) ptr2;
-  float *p1 = q1, *p2 = q2;
+  ca_u_float32_t *q1 = (ca_u_float32_t *) ptr1, *q2 = (ca_u_float32_t *) ptr2;
+  ca_u_float32_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -2168,8 +3124,8 @@ ca_bincmp_lt_float64_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  double *q1 = (double *) ptr1, *q2 = (double *) ptr2;
-  double *p1 = q1, *p2 = q2;
+  ca_u_float64_t *q1 = (ca_u_float64_t *) ptr1, *q2 = (ca_u_float64_t *) ptr2;
+  ca_u_float64_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -2415,8 +3461,8 @@ ca_bincmp_ge_int16_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  int16_t *q1 = (int16_t *) ptr1, *q2 = (int16_t *) ptr2;
-  int16_t *p1 = q1, *p2 = q2;
+  ca_u_int16_t *q1 = (ca_u_int16_t *) ptr1, *q2 = (ca_u_int16_t *) ptr2;
+  ca_u_int16_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -2450,8 +3514,8 @@ ca_bincmp_ge_uint16_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  uint16_t *q1 = (uint16_t *) ptr1, *q2 = (uint16_t *) ptr2;
-  uint16_t *p1 = q1, *p2 = q2;
+  ca_u_uint16_t *q1 = (ca_u_uint16_t *) ptr1, *q2 = (ca_u_uint16_t *) ptr2;
+  ca_u_uint16_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -2485,8 +3567,8 @@ ca_bincmp_ge_int32_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  int32_t *q1 = (int32_t *) ptr1, *q2 = (int32_t *) ptr2;
-  int32_t *p1 = q1, *p2 = q2;
+  ca_u_int32_t *q1 = (ca_u_int32_t *) ptr1, *q2 = (ca_u_int32_t *) ptr2;
+  ca_u_int32_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -2520,8 +3620,8 @@ ca_bincmp_ge_uint32_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  uint32_t *q1 = (uint32_t *) ptr1, *q2 = (uint32_t *) ptr2;
-  uint32_t *p1 = q1, *p2 = q2;
+  ca_u_uint32_t *q1 = (ca_u_uint32_t *) ptr1, *q2 = (ca_u_uint32_t *) ptr2;
+  ca_u_uint32_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -2555,8 +3673,8 @@ ca_bincmp_ge_int64_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  int64_t *q1 = (int64_t *) ptr1, *q2 = (int64_t *) ptr2;
-  int64_t *p1 = q1, *p2 = q2;
+  ca_u_int64_t *q1 = (ca_u_int64_t *) ptr1, *q2 = (ca_u_int64_t *) ptr2;
+  ca_u_int64_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -2590,8 +3726,8 @@ ca_bincmp_ge_uint64_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  uint64_t *q1 = (uint64_t *) ptr1, *q2 = (uint64_t *) ptr2;
-  uint64_t *p1 = q1, *p2 = q2;
+  ca_u_uint64_t *q1 = (ca_u_uint64_t *) ptr1, *q2 = (ca_u_uint64_t *) ptr2;
+  ca_u_uint64_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -2625,8 +3779,8 @@ ca_bincmp_ge_float32_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  float *q1 = (float *) ptr1, *q2 = (float *) ptr2;
-  float *p1 = q1, *p2 = q2;
+  ca_u_float32_t *q1 = (ca_u_float32_t *) ptr1, *q2 = (ca_u_float32_t *) ptr2;
+  ca_u_float32_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -2660,8 +3832,8 @@ ca_bincmp_ge_float64_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  double *q1 = (double *) ptr1, *q2 = (double *) ptr2;
-  double *p1 = q1, *p2 = q2;
+  ca_u_float64_t *q1 = (ca_u_float64_t *) ptr1, *q2 = (ca_u_float64_t *) ptr2;
+  ca_u_float64_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -2907,8 +4169,8 @@ ca_bincmp_le_int16_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  int16_t *q1 = (int16_t *) ptr1, *q2 = (int16_t *) ptr2;
-  int16_t *p1 = q1, *p2 = q2;
+  ca_u_int16_t *q1 = (ca_u_int16_t *) ptr1, *q2 = (ca_u_int16_t *) ptr2;
+  ca_u_int16_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -2942,8 +4222,8 @@ ca_bincmp_le_uint16_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  uint16_t *q1 = (uint16_t *) ptr1, *q2 = (uint16_t *) ptr2;
-  uint16_t *p1 = q1, *p2 = q2;
+  ca_u_uint16_t *q1 = (ca_u_uint16_t *) ptr1, *q2 = (ca_u_uint16_t *) ptr2;
+  ca_u_uint16_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -2977,8 +4275,8 @@ ca_bincmp_le_int32_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  int32_t *q1 = (int32_t *) ptr1, *q2 = (int32_t *) ptr2;
-  int32_t *p1 = q1, *p2 = q2;
+  ca_u_int32_t *q1 = (ca_u_int32_t *) ptr1, *q2 = (ca_u_int32_t *) ptr2;
+  ca_u_int32_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -3012,8 +4328,8 @@ ca_bincmp_le_uint32_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  uint32_t *q1 = (uint32_t *) ptr1, *q2 = (uint32_t *) ptr2;
-  uint32_t *p1 = q1, *p2 = q2;
+  ca_u_uint32_t *q1 = (ca_u_uint32_t *) ptr1, *q2 = (ca_u_uint32_t *) ptr2;
+  ca_u_uint32_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -3047,8 +4381,8 @@ ca_bincmp_le_int64_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  int64_t *q1 = (int64_t *) ptr1, *q2 = (int64_t *) ptr2;
-  int64_t *p1 = q1, *p2 = q2;
+  ca_u_int64_t *q1 = (ca_u_int64_t *) ptr1, *q2 = (ca_u_int64_t *) ptr2;
+  ca_u_int64_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -3082,8 +4434,8 @@ ca_bincmp_le_uint64_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  uint64_t *q1 = (uint64_t *) ptr1, *q2 = (uint64_t *) ptr2;
-  uint64_t *p1 = q1, *p2 = q2;
+  ca_u_uint64_t *q1 = (ca_u_uint64_t *) ptr1, *q2 = (ca_u_uint64_t *) ptr2;
+  ca_u_uint64_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -3117,8 +4487,8 @@ ca_bincmp_le_float32_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  float *q1 = (float *) ptr1, *q2 = (float *) ptr2;
-  float *p1 = q1, *p2 = q2;
+  ca_u_float32_t *q1 = (ca_u_float32_t *) ptr1, *q2 = (ca_u_float32_t *) ptr2;
+  ca_u_float32_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -3152,8 +4540,8 @@ ca_bincmp_le_float64_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  double *q1 = (double *) ptr1, *q2 = (double *) ptr2;
-  double *p1 = q1, *p2 = q2;
+  ca_u_float64_t *q1 = (ca_u_float64_t *) ptr1, *q2 = (ca_u_float64_t *) ptr2;
+  ca_u_float64_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -3494,8 +4990,8 @@ ca_bincmp_is_close_int16_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  int16_t *q1 = (int16_t *) ptr1, *q2 = (int16_t *) ptr2;
-  int16_t *p1 = q1, *p2 = q2;
+  ca_u_int16_t *q1 = (ca_u_int16_t *) ptr1, *q2 = (ca_u_int16_t *) ptr2;
+  ca_u_int16_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -3529,8 +5043,8 @@ ca_bincmp_is_close_uint16_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  uint16_t *q1 = (uint16_t *) ptr1, *q2 = (uint16_t *) ptr2;
-  uint16_t *p1 = q1, *p2 = q2;
+  ca_u_uint16_t *q1 = (ca_u_uint16_t *) ptr1, *q2 = (ca_u_uint16_t *) ptr2;
+  ca_u_uint16_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -3564,8 +5096,8 @@ ca_bincmp_is_close_int32_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  int32_t *q1 = (int32_t *) ptr1, *q2 = (int32_t *) ptr2;
-  int32_t *p1 = q1, *p2 = q2;
+  ca_u_int32_t *q1 = (ca_u_int32_t *) ptr1, *q2 = (ca_u_int32_t *) ptr2;
+  ca_u_int32_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -3599,8 +5149,8 @@ ca_bincmp_is_close_uint32_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  uint32_t *q1 = (uint32_t *) ptr1, *q2 = (uint32_t *) ptr2;
-  uint32_t *p1 = q1, *p2 = q2;
+  ca_u_uint32_t *q1 = (ca_u_uint32_t *) ptr1, *q2 = (ca_u_uint32_t *) ptr2;
+  ca_u_uint32_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -3634,8 +5202,8 @@ ca_bincmp_is_close_int64_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  int64_t *q1 = (int64_t *) ptr1, *q2 = (int64_t *) ptr2;
-  int64_t *p1 = q1, *p2 = q2;
+  ca_u_int64_t *q1 = (ca_u_int64_t *) ptr1, *q2 = (ca_u_int64_t *) ptr2;
+  ca_u_int64_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -3669,8 +5255,8 @@ ca_bincmp_is_close_uint64_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  uint64_t *q1 = (uint64_t *) ptr1, *q2 = (uint64_t *) ptr2;
-  uint64_t *p1 = q1, *p2 = q2;
+  ca_u_uint64_t *q1 = (ca_u_uint64_t *) ptr1, *q2 = (ca_u_uint64_t *) ptr2;
+  ca_u_uint64_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -3704,8 +5308,8 @@ ca_bincmp_is_close_float32_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  float *q1 = (float *) ptr1, *q2 = (float *) ptr2;
-  float *p1 = q1, *p2 = q2;
+  ca_u_float32_t *q1 = (ca_u_float32_t *) ptr1, *q2 = (ca_u_float32_t *) ptr2;
+  ca_u_float32_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -3739,8 +5361,8 @@ ca_bincmp_is_close_float64_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  double *q1 = (double *) ptr1, *q2 = (double *) ptr2;
-  double *p1 = q1, *p2 = q2;
+  ca_u_float64_t *q1 = (ca_u_float64_t *) ptr1, *q2 = (ca_u_float64_t *) ptr2;
+  ca_u_float64_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -3774,8 +5414,8 @@ ca_bincmp_is_close_cmplx64_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  cmplx64_t *q1 = (cmplx64_t *) ptr1, *q2 = (cmplx64_t *) ptr2;
-  cmplx64_t *p1 = q1, *p2 = q2;
+  ca_u_cmplx64_t *q1 = (ca_u_cmplx64_t *) ptr1, *q2 = (ca_u_cmplx64_t *) ptr2;
+  ca_u_cmplx64_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -3809,8 +5467,8 @@ ca_bincmp_is_close_cmplx128_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  cmplx128_t *q1 = (cmplx128_t *) ptr1, *q2 = (cmplx128_t *) ptr2;
-  cmplx128_t *p1 = q1, *p2 = q2;
+  ca_u_cmplx128_t *q1 = (ca_u_cmplx128_t *) ptr1, *q2 = (ca_u_cmplx128_t *) ptr2;
+  ca_u_cmplx128_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -3966,8 +5698,8 @@ ca_bincmp_is_equiv_int16_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  int16_t *q1 = (int16_t *) ptr1, *q2 = (int16_t *) ptr2;
-  int16_t *p1 = q1, *p2 = q2;
+  ca_u_int16_t *q1 = (ca_u_int16_t *) ptr1, *q2 = (ca_u_int16_t *) ptr2;
+  ca_u_int16_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -4011,8 +5771,8 @@ ca_bincmp_is_equiv_uint16_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  uint16_t *q1 = (uint16_t *) ptr1, *q2 = (uint16_t *) ptr2;
-  uint16_t *p1 = q1, *p2 = q2;
+  ca_u_uint16_t *q1 = (ca_u_uint16_t *) ptr1, *q2 = (ca_u_uint16_t *) ptr2;
+  ca_u_uint16_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -4056,8 +5844,8 @@ ca_bincmp_is_equiv_int32_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  int32_t *q1 = (int32_t *) ptr1, *q2 = (int32_t *) ptr2;
-  int32_t *p1 = q1, *p2 = q2;
+  ca_u_int32_t *q1 = (ca_u_int32_t *) ptr1, *q2 = (ca_u_int32_t *) ptr2;
+  ca_u_int32_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -4101,8 +5917,8 @@ ca_bincmp_is_equiv_uint32_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  uint32_t *q1 = (uint32_t *) ptr1, *q2 = (uint32_t *) ptr2;
-  uint32_t *p1 = q1, *p2 = q2;
+  ca_u_uint32_t *q1 = (ca_u_uint32_t *) ptr1, *q2 = (ca_u_uint32_t *) ptr2;
+  ca_u_uint32_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -4146,8 +5990,8 @@ ca_bincmp_is_equiv_int64_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  int64_t *q1 = (int64_t *) ptr1, *q2 = (int64_t *) ptr2;
-  int64_t *p1 = q1, *p2 = q2;
+  ca_u_int64_t *q1 = (ca_u_int64_t *) ptr1, *q2 = (ca_u_int64_t *) ptr2;
+  ca_u_int64_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -4191,8 +6063,8 @@ ca_bincmp_is_equiv_uint64_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  uint64_t *q1 = (uint64_t *) ptr1, *q2 = (uint64_t *) ptr2;
-  uint64_t *p1 = q1, *p2 = q2;
+  ca_u_uint64_t *q1 = (ca_u_uint64_t *) ptr1, *q2 = (ca_u_uint64_t *) ptr2;
+  ca_u_uint64_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -4236,8 +6136,8 @@ ca_bincmp_is_equiv_float32_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  float *q1 = (float *) ptr1, *q2 = (float *) ptr2;
-  float *p1 = q1, *p2 = q2;
+  ca_u_float32_t *q1 = (ca_u_float32_t *) ptr1, *q2 = (ca_u_float32_t *) ptr2;
+  ca_u_float32_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -4252,20 +6152,46 @@ ca_bincmp_is_equiv_float32_t (ca_size_t n, boolean8_t *m,
         p2 = q2 + k*i2;
         p3 = q3 + k*i3;
         {
-      float aa = fabs((*p1)), ab = fabs((*p2));
-      float m = (aa > ab) ? aa : ab;
+      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 == 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;
       p2 = q2 + k*i2;
       p3 = q3 + k*i3;
       {
-      float aa = fabs((*p1)), ab = fabs((*p2));
-      float m = (aa > ab) ? aa : ab;
+      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;
      }
     }
@@ -4279,8 +6205,8 @@ ca_bincmp_is_equiv_float64_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  double *q1 = (double *) ptr1, *q2 = (double *) ptr2;
-  double *p1 = q1, *p2 = q2;
+  ca_u_float64_t *q1 = (ca_u_float64_t *) ptr1, *q2 = (ca_u_float64_t *) ptr2;
+  ca_u_float64_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -4295,20 +6221,46 @@ ca_bincmp_is_equiv_float64_t (ca_size_t n, boolean8_t *m,
         p2 = q2 + k*i2;
         p3 = q3 + k*i3;
         {
-      double aa = fabs((*p1)), ab = fabs((*p2));
-      double m = (aa > ab) ? aa : ab;
+      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 == 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;
       p2 = q2 + k*i2;
       p3 = q3 + k*i3;
       {
-      double aa = fabs((*p1)), ab = fabs((*p2));
-      double m = (aa > ab) ? aa : ab;
+      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;
      }
     }
@@ -4322,8 +6274,8 @@ ca_bincmp_is_equiv_cmplx64_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  cmplx64_t *q1 = (cmplx64_t *) ptr1, *q2 = (cmplx64_t *) ptr2;
-  cmplx64_t *p1 = q1, *p2 = q2;
+  ca_u_cmplx64_t *q1 = (ca_u_cmplx64_t *) ptr1, *q2 = (ca_u_cmplx64_t *) ptr2;
+  ca_u_cmplx64_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;
@@ -4365,8 +6343,8 @@ ca_bincmp_is_equiv_cmplx128_t (ca_size_t n, boolean8_t *m,
                                   char *ptr3, ca_size_t b3, ca_size_t i3,
                                   double tol)
 {
-  cmplx128_t *q1 = (cmplx128_t *) ptr1, *q2 = (cmplx128_t *) ptr2;
-  cmplx128_t *p1 = q1, *p2 = q2;
+  ca_u_cmplx128_t *q1 = (ca_u_cmplx128_t *) ptr1, *q2 = (ca_u_cmplx128_t *) ptr2;
+  ca_u_cmplx128_t *p1 = q1, *p2 = q2;
   boolean8_t *q3 = (boolean8_t *) ptr3;
   boolean8_t *p3 = q3;
   ca_size_t k;
@@ -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;