RubyGems - carray - Versions diffs - 3.0.4 → 3.0.5 - Mend

carray 3.0.4 → 3.0.5

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 (111) hide show

checksums.yaml +4 -4
data/CHANGELOG.md +205 -0
data/carray.gemspec +1 -1
data/ext/ca_array_pool.c +25 -1
data/ext/ca_axis_dispatch.c +13 -20
data/ext/ca_categorical_iterator.c +27 -20
data/ext/ca_compare.c +4 -4
data/ext/ca_composite_dispatch.c +9 -9
data/ext/ca_kernel_iterator.c +6 -6
data/ext/ca_kernel_iterator.h +112 -83
data/ext/ca_obj_array.c +69 -19
data/ext/ca_obj_bincmp.c +1 -0
data/ext/ca_obj_binop.c +1 -0
data/ext/ca_obj_bitarray.c +5 -1
data/ext/ca_obj_bitfield.c +58 -43
data/ext/ca_obj_byte_swap.c +2 -1
data/ext/ca_obj_const_string.c +15 -3
data/ext/ca_obj_face.c +42 -1
data/ext/ca_obj_field.c +13 -3
data/ext/ca_obj_grid.c +52 -5
data/ext/ca_obj_monop.c +46 -10
data/ext/ca_obj_object.c +64 -12
data/ext/ca_obj_record.c +3 -0
data/ext/ca_obj_reduce.c +4 -0
data/ext/ca_obj_refer.c +18 -9
data/ext/ca_obj_roll.c +26 -6
data/ext/ca_obj_select.c +34 -5
data/ext/ca_obj_shift.c +31 -7
data/ext/ca_obj_stack.c +1 -0
data/ext/ca_obj_stride.c +28 -19
data/ext/ca_obj_string.c +11 -2
data/ext/ca_obj_tile.c +27 -7
data/ext/ca_obj_transpose.c +8 -4
data/ext/ca_obj_triop.c +3 -0
data/ext/ca_obj_window.c +79 -50
data/ext/ca_op_ipower.c +4 -4
data/ext/ca_sort_kernels.h +6 -3
data/ext/carray.h +33 -0
data/ext/carray_access.c +16 -3
data/ext/carray_attribute.c +24 -0
data/ext/carray_cast.c +243 -82
data/ext/carray_cast_func.rb +56 -26
data/ext/carray_conversion.c +2 -2
data/ext/carray_core.c +3 -0
data/ext/carray_count.c +6 -0
data/ext/carray_element.c +9 -4
data/ext/carray_factorize.c +20 -20
data/ext/carray_generate.c +8 -2
data/ext/carray_histogram.c +1 -1
data/ext/carray_index_classifier.c +4 -4
data/ext/carray_internal.h +12 -0
data/ext/carray_kernels_bincmp.c +168 -168
data/ext/carray_kernels_binop.c +900 -900
data/ext/carray_kernels_init.c +4 -4
data/ext/carray_kernels_moncmp.c +96 -96
data/ext/carray_kernels_monop.c +1984 -1984
data/ext/carray_kernels_reduce_aggregate.c +1733 -1733
data/ext/carray_kernels_reduce_cumulative.c +642 -642
data/ext/carray_kernels_reduce_extreme.c +1150 -1150
data/ext/carray_kernels_reduce_variance.c +258 -258
data/ext/carray_kernels_scan.c +128 -128
data/ext/carray_kernels_search.c +340 -340
data/ext/carray_kernels_sort.c +295 -304
data/ext/carray_kernels_triop.c +114 -114
data/ext/carray_key_windows.c +4 -0
data/ext/carray_mask.c +45 -22
data/ext/carray_memory_view.c +2 -2
data/ext/carray_operator.c +5 -5
data/ext/carray_order.c +227 -8
data/ext/carray_random.c +18 -18
data/ext/carray_scatter.c +24 -7
data/ext/carray_slab.c +32 -9
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_time_text.c +4 -2
data/ext/carray_utils.c +63 -21
data/ext/mkkernel.rb +210 -115
data/ext/version.h +4 -4
data/lib/carray/attribute.rb +17 -4
data/lib/carray/basics.rb +27 -7
data/lib/carray/categorical.rb +30 -16
data/lib/carray/categorical_iterator.rb +6 -6
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 +4 -3
data/lib/carray/frame/concat.rb +17 -3
data/lib/carray/frame/frame.rb +6 -7
data/lib/carray/frame/group.rb +100 -13
data/lib/carray/frame/io.rb +3 -1
data/lib/carray/fuse_source.rb +31 -2
data/lib/carray/lazy.rb +35 -0
data/lib/carray/methods/is_in.rb +8 -1
data/lib/carray/segment_iterator.rb +12 -8
data/lib/carray/serialize.rb +19 -2
data/lib/carray/stack.rb +15 -12
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 +107 -68
data/yard-stubs/ca_obj_transpose.rb +5 -4
data/yard-stubs/carray_attribute.rb +3 -1
data/yard-stubs/carray_cast.rb +2 -0
data/yard-stubs/carray_conversion.rb +5 -1
data/yard-stubs/carray_element.rb +7 -6
data/yard-stubs/carray_mask.rb +8 -3
data/yard-stubs/carray_math.rb +14 -4
data/yard-stubs/carray_order.rb +100 -4
data/yard-stubs/carray_scatter.rb +7 -5
metadata +1 -1

data/ext/carray_kernels_sort.c CHANGED Viewed

@@ -101,13 +101,13 @@ sort_index_ki_native_i8 (VALUE self, CArray *ca, int axis, int do_stable, int ma
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(int8_t *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = k;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(int8_t *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = k;
           n_masked++;
         }
@@ -115,13 +115,13 @@ sort_index_ki_native_i8 (VALUE self, CArray *ca, int axis, int do_stable, int ma
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(int8_t *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = k;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(int8_t *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = k;
           n_valid++;
         }
@@ -129,7 +129,7 @@ sort_index_ki_native_i8 (VALUE self, CArray *ca, int axis, int do_stable, int ma
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(int8_t *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = k;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -216,13 +216,13 @@ sort_index_ki_native_u8 (VALUE self, CArray *ca, int axis, int do_stable, int ma
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(uint8_t *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = k;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(uint8_t *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = k;
           n_masked++;
         }
@@ -230,13 +230,13 @@ sort_index_ki_native_u8 (VALUE self, CArray *ca, int axis, int do_stable, int ma
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(uint8_t *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = k;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(uint8_t *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = k;
           n_valid++;
         }
@@ -244,7 +244,7 @@ sort_index_ki_native_u8 (VALUE self, CArray *ca, int axis, int do_stable, int ma
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(uint8_t *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = k;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -276,7 +276,7 @@ sort_index_ki_native_u8 (VALUE self, CArray *ca, int axis, int do_stable, int ma
 }
 typedef struct {
-  int16_t v;
+  ca_u_int16_t v;
   ca_size_t i;
 } sort_index_pair_i16;
@@ -331,13 +331,13 @@ sort_index_ki_native_i16 (VALUE self, CArray *ca, int axis, int do_stable, int m
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(int16_t *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = k;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(int16_t *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = k;
           n_masked++;
         }
@@ -345,13 +345,13 @@ sort_index_ki_native_i16 (VALUE self, CArray *ca, int axis, int do_stable, int m
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(int16_t *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = k;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(int16_t *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = k;
           n_valid++;
         }
@@ -359,7 +359,7 @@ sort_index_ki_native_i16 (VALUE self, CArray *ca, int axis, int do_stable, int m
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(int16_t *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = k;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -391,7 +391,7 @@ sort_index_ki_native_i16 (VALUE self, CArray *ca, int axis, int do_stable, int m
 }
 typedef struct {
-  uint16_t v;
+  ca_u_uint16_t v;
   ca_size_t i;
 } sort_index_pair_u16;
@@ -446,13 +446,13 @@ sort_index_ki_native_u16 (VALUE self, CArray *ca, int axis, int do_stable, int m
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(uint16_t *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = k;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(uint16_t *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = k;
           n_masked++;
         }
@@ -460,13 +460,13 @@ sort_index_ki_native_u16 (VALUE self, CArray *ca, int axis, int do_stable, int m
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(uint16_t *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = k;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(uint16_t *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = k;
           n_valid++;
         }
@@ -474,7 +474,7 @@ sort_index_ki_native_u16 (VALUE self, CArray *ca, int axis, int do_stable, int m
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(uint16_t *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = k;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -506,7 +506,7 @@ sort_index_ki_native_u16 (VALUE self, CArray *ca, int axis, int do_stable, int m
 }
 typedef struct {
-  int32_t v;
+  ca_u_int32_t v;
   ca_size_t i;
 } sort_index_pair_i32;
@@ -561,13 +561,13 @@ sort_index_ki_native_i32 (VALUE self, CArray *ca, int axis, int do_stable, int m
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(int32_t *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = k;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(int32_t *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = k;
           n_masked++;
         }
@@ -575,13 +575,13 @@ sort_index_ki_native_i32 (VALUE self, CArray *ca, int axis, int do_stable, int m
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(int32_t *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = k;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(int32_t *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = k;
           n_valid++;
         }
@@ -589,7 +589,7 @@ sort_index_ki_native_i32 (VALUE self, CArray *ca, int axis, int do_stable, int m
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(int32_t *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = k;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -621,7 +621,7 @@ sort_index_ki_native_i32 (VALUE self, CArray *ca, int axis, int do_stable, int m
 }
 typedef struct {
-  uint32_t v;
+  ca_u_uint32_t v;
   ca_size_t i;
 } sort_index_pair_u32;
@@ -676,13 +676,13 @@ sort_index_ki_native_u32 (VALUE self, CArray *ca, int axis, int do_stable, int m
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(uint32_t *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = k;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(uint32_t *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = k;
           n_masked++;
         }
@@ -690,13 +690,13 @@ sort_index_ki_native_u32 (VALUE self, CArray *ca, int axis, int do_stable, int m
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(uint32_t *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = k;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(uint32_t *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = k;
           n_valid++;
         }
@@ -704,7 +704,7 @@ sort_index_ki_native_u32 (VALUE self, CArray *ca, int axis, int do_stable, int m
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(uint32_t *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = k;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -736,7 +736,7 @@ sort_index_ki_native_u32 (VALUE self, CArray *ca, int axis, int do_stable, int m
 }
 typedef struct {
-  int64_t v;
+  ca_u_int64_t v;
   ca_size_t i;
 } sort_index_pair_i64;
@@ -791,13 +791,13 @@ sort_index_ki_native_i64 (VALUE self, CArray *ca, int axis, int do_stable, int m
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(int64_t *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = k;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(int64_t *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = k;
           n_masked++;
         }
@@ -805,13 +805,13 @@ sort_index_ki_native_i64 (VALUE self, CArray *ca, int axis, int do_stable, int m
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(int64_t *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = k;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(int64_t *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = k;
           n_valid++;
         }
@@ -819,7 +819,7 @@ sort_index_ki_native_i64 (VALUE self, CArray *ca, int axis, int do_stable, int m
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(int64_t *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = k;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -851,7 +851,7 @@ sort_index_ki_native_i64 (VALUE self, CArray *ca, int axis, int do_stable, int m
 }
 typedef struct {
-  uint64_t v;
+  ca_u_uint64_t v;
   ca_size_t i;
 } sort_index_pair_u64;
@@ -906,13 +906,13 @@ sort_index_ki_native_u64 (VALUE self, CArray *ca, int axis, int do_stable, int m
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(uint64_t *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = k;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(uint64_t *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = k;
           n_masked++;
         }
@@ -920,13 +920,13 @@ sort_index_ki_native_u64 (VALUE self, CArray *ca, int axis, int do_stable, int m
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(uint64_t *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = k;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(uint64_t *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = k;
           n_valid++;
         }
@@ -934,7 +934,7 @@ sort_index_ki_native_u64 (VALUE self, CArray *ca, int axis, int do_stable, int m
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(uint64_t *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = k;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -966,7 +966,7 @@ sort_index_ki_native_u64 (VALUE self, CArray *ca, int axis, int do_stable, int m
 }
 typedef struct {
-  float v;
+  ca_u_float32_t v;
   ca_size_t i;
 } sort_index_pair_f32;
@@ -1021,13 +1021,13 @@ sort_index_ki_native_f32 (VALUE self, CArray *ca, int axis, int do_stable, int m
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(float *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = k;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(float *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = k;
           n_masked++;
         }
@@ -1035,13 +1035,13 @@ sort_index_ki_native_f32 (VALUE self, CArray *ca, int axis, int do_stable, int m
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(float *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = k;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(float *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = k;
           n_valid++;
         }
@@ -1049,7 +1049,7 @@ sort_index_ki_native_f32 (VALUE self, CArray *ca, int axis, int do_stable, int m
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(float *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = k;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -1085,7 +1085,7 @@ sort_index_ki_native_f32 (VALUE self, CArray *ca, int axis, int do_stable, int m
 }
 typedef struct {
-  double v;
+  ca_u_float64_t v;
   ca_size_t i;
 } sort_index_pair_f64;
@@ -1140,13 +1140,13 @@ sort_index_ki_native_f64 (VALUE self, CArray *ca, int axis, int do_stable, int m
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(double *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = k;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(double *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = k;
           n_masked++;
         }
@@ -1154,13 +1154,13 @@ sort_index_ki_native_f64 (VALUE self, CArray *ca, int axis, int do_stable, int m
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(double *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = k;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(double *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = k;
           n_valid++;
         }
@@ -1168,7 +1168,7 @@ sort_index_ki_native_f64 (VALUE self, CArray *ca, int axis, int do_stable, int m
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(double *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = k;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -1259,13 +1259,13 @@ sort_index_ki_native_bool (VALUE self, CArray *ca, int axis, int do_stable, int
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(boolean8_t *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = k;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(boolean8_t *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = k;
           n_masked++;
         }
@@ -1273,13 +1273,13 @@ sort_index_ki_native_bool (VALUE self, CArray *ca, int axis, int do_stable, int
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(boolean8_t *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = k;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(boolean8_t *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = k;
           n_valid++;
         }
@@ -1287,7 +1287,7 @@ sort_index_ki_native_bool (VALUE self, CArray *ca, int axis, int do_stable, int
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(boolean8_t *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = k;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -1408,13 +1408,13 @@ sort_index_ki_native_object_body (VALUE arg)
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(VALUE *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = k;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(VALUE *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = k;
           n_masked++;
         }
@@ -1422,13 +1422,13 @@ sort_index_ki_native_object_body (VALUE arg)
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(VALUE *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = k;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(VALUE *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = k;
           n_valid++;
         }
@@ -1436,7 +1436,7 @@ sort_index_ki_native_object_body (VALUE arg)
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(VALUE *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = k;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -1528,10 +1528,7 @@ sort_index_ki_native_fixlen (VALUE self, CArray *ca, int axis, int do_stable, in
   ca_size_t fiber_n = st_in.slab_elements;
   volatile VALUE hbuf = 0, haux = 0;
   sort_index_pair_fixlen *buf = ALLOCV_N(sort_index_pair_fixlen, hbuf, fiber_n);
-  sort_index_pair_fixlen *aux = NULL;
-  if ( do_stable ) {
-    aux = ALLOCV_N(sort_index_pair_fixlen, haux, fiber_n);
-  }
+  (void) haux;
   char       *pi, *po;
   boolean8_t *mi, *mo;
@@ -1778,13 +1775,13 @@ sort_addr_ki_native_i8 (VALUE self, CArray *ca, int axis, int do_stable, int mas
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(int8_t *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(int8_t *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
@@ -1792,13 +1789,13 @@ sort_addr_ki_native_i8 (VALUE self, CArray *ca, int axis, int do_stable, int mas
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(int8_t *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(int8_t *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
@@ -1806,7 +1803,7 @@ sort_addr_ki_native_i8 (VALUE self, CArray *ca, int axis, int do_stable, int mas
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(int8_t *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = outer_off + k * axis_vstride;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -1925,13 +1922,13 @@ sort_addr_ki_native_u8 (VALUE self, CArray *ca, int axis, int do_stable, int mas
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(uint8_t *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(uint8_t *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
@@ -1939,13 +1936,13 @@ sort_addr_ki_native_u8 (VALUE self, CArray *ca, int axis, int do_stable, int mas
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(uint8_t *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(uint8_t *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
@@ -1953,7 +1950,7 @@ sort_addr_ki_native_u8 (VALUE self, CArray *ca, int axis, int do_stable, int mas
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(uint8_t *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = outer_off + k * axis_vstride;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -1990,7 +1987,7 @@ sort_addr_ki_native_u8 (VALUE self, CArray *ca, int axis, int do_stable, int mas
 }
 typedef struct {
-  int16_t v;
+  ca_u_int16_t v;
   ca_size_t i;
 } sort_addr_pair_i16;
@@ -2072,13 +2069,13 @@ sort_addr_ki_native_i16 (VALUE self, CArray *ca, int axis, int do_stable, int ma
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(int16_t *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(int16_t *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
@@ -2086,13 +2083,13 @@ sort_addr_ki_native_i16 (VALUE self, CArray *ca, int axis, int do_stable, int ma
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(int16_t *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(int16_t *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
@@ -2100,7 +2097,7 @@ sort_addr_ki_native_i16 (VALUE self, CArray *ca, int axis, int do_stable, int ma
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(int16_t *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = outer_off + k * axis_vstride;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -2137,7 +2134,7 @@ sort_addr_ki_native_i16 (VALUE self, CArray *ca, int axis, int do_stable, int ma
 }
 typedef struct {
-  uint16_t v;
+  ca_u_uint16_t v;
   ca_size_t i;
 } sort_addr_pair_u16;
@@ -2219,13 +2216,13 @@ sort_addr_ki_native_u16 (VALUE self, CArray *ca, int axis, int do_stable, int ma
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(uint16_t *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(uint16_t *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
@@ -2233,13 +2230,13 @@ sort_addr_ki_native_u16 (VALUE self, CArray *ca, int axis, int do_stable, int ma
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(uint16_t *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(uint16_t *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
@@ -2247,7 +2244,7 @@ sort_addr_ki_native_u16 (VALUE self, CArray *ca, int axis, int do_stable, int ma
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(uint16_t *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = outer_off + k * axis_vstride;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -2284,7 +2281,7 @@ sort_addr_ki_native_u16 (VALUE self, CArray *ca, int axis, int do_stable, int ma
 }
 typedef struct {
-  int32_t v;
+  ca_u_int32_t v;
   ca_size_t i;
 } sort_addr_pair_i32;
@@ -2366,13 +2363,13 @@ sort_addr_ki_native_i32 (VALUE self, CArray *ca, int axis, int do_stable, int ma
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(int32_t *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(int32_t *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
@@ -2380,13 +2377,13 @@ sort_addr_ki_native_i32 (VALUE self, CArray *ca, int axis, int do_stable, int ma
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(int32_t *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(int32_t *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
@@ -2394,7 +2391,7 @@ sort_addr_ki_native_i32 (VALUE self, CArray *ca, int axis, int do_stable, int ma
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(int32_t *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = outer_off + k * axis_vstride;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -2431,7 +2428,7 @@ sort_addr_ki_native_i32 (VALUE self, CArray *ca, int axis, int do_stable, int ma
 }
 typedef struct {
-  uint32_t v;
+  ca_u_uint32_t v;
   ca_size_t i;
 } sort_addr_pair_u32;
@@ -2513,13 +2510,13 @@ sort_addr_ki_native_u32 (VALUE self, CArray *ca, int axis, int do_stable, int ma
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(uint32_t *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(uint32_t *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
@@ -2527,13 +2524,13 @@ sort_addr_ki_native_u32 (VALUE self, CArray *ca, int axis, int do_stable, int ma
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(uint32_t *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(uint32_t *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
@@ -2541,7 +2538,7 @@ sort_addr_ki_native_u32 (VALUE self, CArray *ca, int axis, int do_stable, int ma
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(uint32_t *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = outer_off + k * axis_vstride;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -2578,7 +2575,7 @@ sort_addr_ki_native_u32 (VALUE self, CArray *ca, int axis, int do_stable, int ma
 }
 typedef struct {
-  int64_t v;
+  ca_u_int64_t v;
   ca_size_t i;
 } sort_addr_pair_i64;
@@ -2660,13 +2657,13 @@ sort_addr_ki_native_i64 (VALUE self, CArray *ca, int axis, int do_stable, int ma
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(int64_t *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(int64_t *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
@@ -2674,13 +2671,13 @@ sort_addr_ki_native_i64 (VALUE self, CArray *ca, int axis, int do_stable, int ma
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(int64_t *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(int64_t *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
@@ -2688,7 +2685,7 @@ sort_addr_ki_native_i64 (VALUE self, CArray *ca, int axis, int do_stable, int ma
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(int64_t *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = outer_off + k * axis_vstride;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -2725,7 +2722,7 @@ sort_addr_ki_native_i64 (VALUE self, CArray *ca, int axis, int do_stable, int ma
 }
 typedef struct {
-  uint64_t v;
+  ca_u_uint64_t v;
   ca_size_t i;
 } sort_addr_pair_u64;
@@ -2807,13 +2804,13 @@ sort_addr_ki_native_u64 (VALUE self, CArray *ca, int axis, int do_stable, int ma
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(uint64_t *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(uint64_t *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
@@ -2821,13 +2818,13 @@ sort_addr_ki_native_u64 (VALUE self, CArray *ca, int axis, int do_stable, int ma
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(uint64_t *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(uint64_t *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
@@ -2835,7 +2832,7 @@ sort_addr_ki_native_u64 (VALUE self, CArray *ca, int axis, int do_stable, int ma
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(uint64_t *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = outer_off + k * axis_vstride;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -2872,7 +2869,7 @@ sort_addr_ki_native_u64 (VALUE self, CArray *ca, int axis, int do_stable, int ma
 }
 typedef struct {
-  float v;
+  ca_u_float32_t v;
   ca_size_t i;
 } sort_addr_pair_f32;
@@ -2954,13 +2951,13 @@ sort_addr_ki_native_f32 (VALUE self, CArray *ca, int axis, int do_stable, int ma
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(float *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(float *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
@@ -2968,13 +2965,13 @@ sort_addr_ki_native_f32 (VALUE self, CArray *ca, int axis, int do_stable, int ma
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(float *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(float *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
@@ -2982,7 +2979,7 @@ sort_addr_ki_native_f32 (VALUE self, CArray *ca, int axis, int do_stable, int ma
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(float *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = outer_off + k * axis_vstride;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -3023,7 +3020,7 @@ sort_addr_ki_native_f32 (VALUE self, CArray *ca, int axis, int do_stable, int ma
 }
 typedef struct {
-  double v;
+  ca_u_float64_t v;
   ca_size_t i;
 } sort_addr_pair_f64;
@@ -3105,13 +3102,13 @@ sort_addr_ki_native_f64 (VALUE self, CArray *ca, int axis, int do_stable, int ma
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(double *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(double *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
@@ -3119,13 +3116,13 @@ sort_addr_ki_native_f64 (VALUE self, CArray *ca, int axis, int do_stable, int ma
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(double *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(double *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
@@ -3133,7 +3130,7 @@ sort_addr_ki_native_f64 (VALUE self, CArray *ca, int axis, int do_stable, int ma
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(double *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = outer_off + k * axis_vstride;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -3256,13 +3253,13 @@ sort_addr_ki_native_bool (VALUE self, CArray *ca, int axis, int do_stable, int m
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(boolean8_t *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(boolean8_t *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
@@ -3270,13 +3267,13 @@ sort_addr_ki_native_bool (VALUE self, CArray *ca, int axis, int do_stable, int m
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(boolean8_t *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(boolean8_t *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
@@ -3284,7 +3281,7 @@ sort_addr_ki_native_bool (VALUE self, CArray *ca, int axis, int do_stable, int m
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(boolean8_t *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = outer_off + k * axis_vstride;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -3437,13 +3434,13 @@ sort_addr_ki_native_object_body (VALUE arg)
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(VALUE *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(VALUE *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
@@ -3451,13 +3448,13 @@ sort_addr_ki_native_object_body (VALUE arg)
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(VALUE *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(VALUE *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
@@ -3465,7 +3462,7 @@ sort_addr_ki_native_object_body (VALUE arg)
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(VALUE *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = outer_off + k * axis_vstride;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -3562,10 +3559,7 @@ sort_addr_ki_native_fixlen (VALUE self, CArray *ca, int axis, int do_stable, int
   ca_size_t fiber_n = st_in.slab_elements;
   volatile VALUE hbuf = 0, haux = 0;
   sort_addr_pair_fixlen *buf = ALLOCV_N(sort_addr_pair_fixlen, hbuf, fiber_n);
-  sort_addr_pair_fixlen *aux = NULL;
-  if ( do_stable ) {
-    aux = ALLOCV_N(sort_addr_pair_fixlen, haux, fiber_n);
-  }
+  (void) haux;
   /* :view_flat semantics: kernel computes view-flat address
      (= row-major position in ca's shape, 0..elements-1) for
@@ -3852,13 +3846,13 @@ partition_index_ki_native_i8 (VALUE self, CArray *ca, int axis, ca_size_t kth, i
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(int8_t *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = k;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(int8_t *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = k;
           n_masked++;
         }
@@ -3866,13 +3860,13 @@ partition_index_ki_native_i8 (VALUE self, CArray *ca, int axis, ca_size_t kth, i
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(int8_t *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = k;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(int8_t *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = k;
           n_valid++;
         }
@@ -3880,7 +3874,7 @@ partition_index_ki_native_i8 (VALUE self, CArray *ca, int axis, ca_size_t kth, i
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(int8_t *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = k;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -4031,13 +4025,13 @@ partition_index_ki_native_u8 (VALUE self, CArray *ca, int axis, ca_size_t kth, i
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(uint8_t *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = k;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(uint8_t *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = k;
           n_masked++;
         }
@@ -4045,13 +4039,13 @@ partition_index_ki_native_u8 (VALUE self, CArray *ca, int axis, ca_size_t kth, i
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(uint8_t *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = k;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(uint8_t *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = k;
           n_valid++;
         }
@@ -4059,7 +4053,7 @@ partition_index_ki_native_u8 (VALUE self, CArray *ca, int axis, ca_size_t kth, i
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(uint8_t *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = k;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -4090,7 +4084,7 @@ partition_index_ki_native_u8 (VALUE self, CArray *ca, int axis, ca_size_t kth, i
 }
 typedef struct {
-  int16_t v;
+  ca_u_int16_t v;
   ca_size_t i;
 } partition_index_pair_i16;
@@ -4210,13 +4204,13 @@ partition_index_ki_native_i16 (VALUE self, CArray *ca, int axis, ca_size_t kth,
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(int16_t *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = k;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(int16_t *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = k;
           n_masked++;
         }
@@ -4224,13 +4218,13 @@ partition_index_ki_native_i16 (VALUE self, CArray *ca, int axis, ca_size_t kth,
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(int16_t *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = k;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(int16_t *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = k;
           n_valid++;
         }
@@ -4238,7 +4232,7 @@ partition_index_ki_native_i16 (VALUE self, CArray *ca, int axis, ca_size_t kth,
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(int16_t *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = k;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -4269,7 +4263,7 @@ partition_index_ki_native_i16 (VALUE self, CArray *ca, int axis, ca_size_t kth,
 }
 typedef struct {
-  uint16_t v;
+  ca_u_uint16_t v;
   ca_size_t i;
 } partition_index_pair_u16;
@@ -4389,13 +4383,13 @@ partition_index_ki_native_u16 (VALUE self, CArray *ca, int axis, ca_size_t kth,
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(uint16_t *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = k;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(uint16_t *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = k;
           n_masked++;
         }
@@ -4403,13 +4397,13 @@ partition_index_ki_native_u16 (VALUE self, CArray *ca, int axis, ca_size_t kth,
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(uint16_t *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = k;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(uint16_t *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = k;
           n_valid++;
         }
@@ -4417,7 +4411,7 @@ partition_index_ki_native_u16 (VALUE self, CArray *ca, int axis, ca_size_t kth,
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(uint16_t *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = k;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -4448,7 +4442,7 @@ partition_index_ki_native_u16 (VALUE self, CArray *ca, int axis, ca_size_t kth,
 }
 typedef struct {
-  int32_t v;
+  ca_u_int32_t v;
   ca_size_t i;
 } partition_index_pair_i32;
@@ -4568,13 +4562,13 @@ partition_index_ki_native_i32 (VALUE self, CArray *ca, int axis, ca_size_t kth,
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(int32_t *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = k;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(int32_t *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = k;
           n_masked++;
         }
@@ -4582,13 +4576,13 @@ partition_index_ki_native_i32 (VALUE self, CArray *ca, int axis, ca_size_t kth,
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(int32_t *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = k;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(int32_t *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = k;
           n_valid++;
         }
@@ -4596,7 +4590,7 @@ partition_index_ki_native_i32 (VALUE self, CArray *ca, int axis, ca_size_t kth,
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(int32_t *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = k;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -4627,7 +4621,7 @@ partition_index_ki_native_i32 (VALUE self, CArray *ca, int axis, ca_size_t kth,
 }
 typedef struct {
-  uint32_t v;
+  ca_u_uint32_t v;
   ca_size_t i;
 } partition_index_pair_u32;
@@ -4747,13 +4741,13 @@ partition_index_ki_native_u32 (VALUE self, CArray *ca, int axis, ca_size_t kth,
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(uint32_t *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = k;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(uint32_t *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = k;
           n_masked++;
         }
@@ -4761,13 +4755,13 @@ partition_index_ki_native_u32 (VALUE self, CArray *ca, int axis, ca_size_t kth,
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(uint32_t *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = k;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(uint32_t *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = k;
           n_valid++;
         }
@@ -4775,7 +4769,7 @@ partition_index_ki_native_u32 (VALUE self, CArray *ca, int axis, ca_size_t kth,
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(uint32_t *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = k;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -4806,7 +4800,7 @@ partition_index_ki_native_u32 (VALUE self, CArray *ca, int axis, ca_size_t kth,
 }
 typedef struct {
-  int64_t v;
+  ca_u_int64_t v;
   ca_size_t i;
 } partition_index_pair_i64;
@@ -4926,13 +4920,13 @@ partition_index_ki_native_i64 (VALUE self, CArray *ca, int axis, ca_size_t kth,
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(int64_t *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = k;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(int64_t *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = k;
           n_masked++;
         }
@@ -4940,13 +4934,13 @@ partition_index_ki_native_i64 (VALUE self, CArray *ca, int axis, ca_size_t kth,
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(int64_t *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = k;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(int64_t *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = k;
           n_valid++;
         }
@@ -4954,7 +4948,7 @@ partition_index_ki_native_i64 (VALUE self, CArray *ca, int axis, ca_size_t kth,
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(int64_t *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = k;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -4985,7 +4979,7 @@ partition_index_ki_native_i64 (VALUE self, CArray *ca, int axis, ca_size_t kth,
 }
 typedef struct {
-  uint64_t v;
+  ca_u_uint64_t v;
   ca_size_t i;
 } partition_index_pair_u64;
@@ -5105,13 +5099,13 @@ partition_index_ki_native_u64 (VALUE self, CArray *ca, int axis, ca_size_t kth,
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(uint64_t *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = k;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(uint64_t *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = k;
           n_masked++;
         }
@@ -5119,13 +5113,13 @@ partition_index_ki_native_u64 (VALUE self, CArray *ca, int axis, ca_size_t kth,
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(uint64_t *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = k;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(uint64_t *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = k;
           n_valid++;
         }
@@ -5133,7 +5127,7 @@ partition_index_ki_native_u64 (VALUE self, CArray *ca, int axis, ca_size_t kth,
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(uint64_t *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = k;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -5164,7 +5158,7 @@ partition_index_ki_native_u64 (VALUE self, CArray *ca, int axis, ca_size_t kth,
 }
 typedef struct {
-  float v;
+  ca_u_float32_t v;
   ca_size_t i;
 } partition_index_pair_f32;
@@ -5284,13 +5278,13 @@ partition_index_ki_native_f32 (VALUE self, CArray *ca, int axis, ca_size_t kth,
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(float *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = k;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(float *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = k;
           n_masked++;
         }
@@ -5298,13 +5292,13 @@ partition_index_ki_native_f32 (VALUE self, CArray *ca, int axis, ca_size_t kth,
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(float *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = k;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(float *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = k;
           n_valid++;
         }
@@ -5312,7 +5306,7 @@ partition_index_ki_native_f32 (VALUE self, CArray *ca, int axis, ca_size_t kth,
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(float *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = k;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -5343,7 +5337,7 @@ partition_index_ki_native_f32 (VALUE self, CArray *ca, int axis, ca_size_t kth,
 }
 typedef struct {
-  double v;
+  ca_u_float64_t v;
   ca_size_t i;
 } partition_index_pair_f64;
@@ -5463,13 +5457,13 @@ partition_index_ki_native_f64 (VALUE self, CArray *ca, int axis, ca_size_t kth,
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(double *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = k;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(double *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = k;
           n_masked++;
         }
@@ -5477,13 +5471,13 @@ partition_index_ki_native_f64 (VALUE self, CArray *ca, int axis, ca_size_t kth,
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(double *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = k;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(double *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = k;
           n_valid++;
         }
@@ -5491,7 +5485,7 @@ partition_index_ki_native_f64 (VALUE self, CArray *ca, int axis, ca_size_t kth,
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(double *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = k;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -5642,13 +5636,13 @@ partition_index_ki_native_bool (VALUE self, CArray *ca, int axis, ca_size_t kth,
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(boolean8_t *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = k;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(boolean8_t *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = k;
           n_masked++;
         }
@@ -5656,13 +5650,13 @@ partition_index_ki_native_bool (VALUE self, CArray *ca, int axis, ca_size_t kth,
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(boolean8_t *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = k;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(boolean8_t *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = k;
           n_valid++;
         }
@@ -5670,7 +5664,7 @@ partition_index_ki_native_bool (VALUE self, CArray *ca, int axis, ca_size_t kth,
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(boolean8_t *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = k;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -5845,13 +5839,13 @@ partition_index_ki_native_object_body (VALUE arg)
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(VALUE *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = k;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(VALUE *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = k;
           n_masked++;
         }
@@ -5859,13 +5853,13 @@ partition_index_ki_native_object_body (VALUE arg)
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(VALUE *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = k;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(VALUE *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = k;
           n_valid++;
         }
@@ -5873,7 +5867,7 @@ partition_index_ki_native_object_body (VALUE arg)
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(VALUE *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = k;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -6322,13 +6316,13 @@ partition_addr_ki_native_i8 (VALUE self, CArray *ca, int axis, ca_size_t kth, in
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(int8_t *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(int8_t *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
@@ -6336,13 +6330,13 @@ partition_addr_ki_native_i8 (VALUE self, CArray *ca, int axis, ca_size_t kth, in
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(int8_t *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(int8_t *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
@@ -6350,7 +6344,7 @@ partition_addr_ki_native_i8 (VALUE self, CArray *ca, int axis, ca_size_t kth, in
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(int8_t *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = outer_off + k * axis_vstride;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -6533,13 +6527,13 @@ partition_addr_ki_native_u8 (VALUE self, CArray *ca, int axis, ca_size_t kth, in
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(uint8_t *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(uint8_t *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
@@ -6547,13 +6541,13 @@ partition_addr_ki_native_u8 (VALUE self, CArray *ca, int axis, ca_size_t kth, in
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(uint8_t *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(uint8_t *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
@@ -6561,7 +6555,7 @@ partition_addr_ki_native_u8 (VALUE self, CArray *ca, int axis, ca_size_t kth, in
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(uint8_t *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = outer_off + k * axis_vstride;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -6597,7 +6591,7 @@ partition_addr_ki_native_u8 (VALUE self, CArray *ca, int axis, ca_size_t kth, in
 }
 typedef struct {
-  int16_t v;
+  ca_u_int16_t v;
   ca_size_t i;
 } partition_addr_pair_i16;
@@ -6744,13 +6738,13 @@ partition_addr_ki_native_i16 (VALUE self, CArray *ca, int axis, ca_size_t kth, i
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(int16_t *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(int16_t *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
@@ -6758,13 +6752,13 @@ partition_addr_ki_native_i16 (VALUE self, CArray *ca, int axis, ca_size_t kth, i
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(int16_t *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(int16_t *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
@@ -6772,7 +6766,7 @@ partition_addr_ki_native_i16 (VALUE self, CArray *ca, int axis, ca_size_t kth, i
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(int16_t *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = outer_off + k * axis_vstride;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -6808,7 +6802,7 @@ partition_addr_ki_native_i16 (VALUE self, CArray *ca, int axis, ca_size_t kth, i
 }
 typedef struct {
-  uint16_t v;
+  ca_u_uint16_t v;
   ca_size_t i;
 } partition_addr_pair_u16;
@@ -6955,13 +6949,13 @@ partition_addr_ki_native_u16 (VALUE self, CArray *ca, int axis, ca_size_t kth, i
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(uint16_t *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(uint16_t *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
@@ -6969,13 +6963,13 @@ partition_addr_ki_native_u16 (VALUE self, CArray *ca, int axis, ca_size_t kth, i
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(uint16_t *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(uint16_t *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
@@ -6983,7 +6977,7 @@ partition_addr_ki_native_u16 (VALUE self, CArray *ca, int axis, ca_size_t kth, i
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(uint16_t *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = outer_off + k * axis_vstride;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -7019,7 +7013,7 @@ partition_addr_ki_native_u16 (VALUE self, CArray *ca, int axis, ca_size_t kth, i
 }
 typedef struct {
-  int32_t v;
+  ca_u_int32_t v;
   ca_size_t i;
 } partition_addr_pair_i32;
@@ -7166,13 +7160,13 @@ partition_addr_ki_native_i32 (VALUE self, CArray *ca, int axis, ca_size_t kth, i
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(int32_t *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(int32_t *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
@@ -7180,13 +7174,13 @@ partition_addr_ki_native_i32 (VALUE self, CArray *ca, int axis, ca_size_t kth, i
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(int32_t *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(int32_t *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
@@ -7194,7 +7188,7 @@ partition_addr_ki_native_i32 (VALUE self, CArray *ca, int axis, ca_size_t kth, i
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(int32_t *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = outer_off + k * axis_vstride;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -7230,7 +7224,7 @@ partition_addr_ki_native_i32 (VALUE self, CArray *ca, int axis, ca_size_t kth, i
 }
 typedef struct {
-  uint32_t v;
+  ca_u_uint32_t v;
   ca_size_t i;
 } partition_addr_pair_u32;
@@ -7377,13 +7371,13 @@ partition_addr_ki_native_u32 (VALUE self, CArray *ca, int axis, ca_size_t kth, i
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(uint32_t *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(uint32_t *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
@@ -7391,13 +7385,13 @@ partition_addr_ki_native_u32 (VALUE self, CArray *ca, int axis, ca_size_t kth, i
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(uint32_t *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(uint32_t *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
@@ -7405,7 +7399,7 @@ partition_addr_ki_native_u32 (VALUE self, CArray *ca, int axis, ca_size_t kth, i
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(uint32_t *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = outer_off + k * axis_vstride;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -7441,7 +7435,7 @@ partition_addr_ki_native_u32 (VALUE self, CArray *ca, int axis, ca_size_t kth, i
 }
 typedef struct {
-  int64_t v;
+  ca_u_int64_t v;
   ca_size_t i;
 } partition_addr_pair_i64;
@@ -7588,13 +7582,13 @@ partition_addr_ki_native_i64 (VALUE self, CArray *ca, int axis, ca_size_t kth, i
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(int64_t *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(int64_t *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
@@ -7602,13 +7596,13 @@ partition_addr_ki_native_i64 (VALUE self, CArray *ca, int axis, ca_size_t kth, i
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(int64_t *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(int64_t *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
@@ -7616,7 +7610,7 @@ partition_addr_ki_native_i64 (VALUE self, CArray *ca, int axis, ca_size_t kth, i
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(int64_t *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = outer_off + k * axis_vstride;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -7652,7 +7646,7 @@ partition_addr_ki_native_i64 (VALUE self, CArray *ca, int axis, ca_size_t kth, i
 }
 typedef struct {
-  uint64_t v;
+  ca_u_uint64_t v;
   ca_size_t i;
 } partition_addr_pair_u64;
@@ -7799,13 +7793,13 @@ partition_addr_ki_native_u64 (VALUE self, CArray *ca, int axis, ca_size_t kth, i
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(uint64_t *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(uint64_t *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
@@ -7813,13 +7807,13 @@ partition_addr_ki_native_u64 (VALUE self, CArray *ca, int axis, ca_size_t kth, i
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(uint64_t *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(uint64_t *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
@@ -7827,7 +7821,7 @@ partition_addr_ki_native_u64 (VALUE self, CArray *ca, int axis, ca_size_t kth, i
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(uint64_t *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = outer_off + k * axis_vstride;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -7863,7 +7857,7 @@ partition_addr_ki_native_u64 (VALUE self, CArray *ca, int axis, ca_size_t kth, i
 }
 typedef struct {
-  float v;
+  ca_u_float32_t v;
   ca_size_t i;
 } partition_addr_pair_f32;
@@ -8010,13 +8004,13 @@ partition_addr_ki_native_f32 (VALUE self, CArray *ca, int axis, ca_size_t kth, i
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(float *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(float *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
@@ -8024,13 +8018,13 @@ partition_addr_ki_native_f32 (VALUE self, CArray *ca, int axis, ca_size_t kth, i
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(float *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(float *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
@@ -8038,7 +8032,7 @@ partition_addr_ki_native_f32 (VALUE self, CArray *ca, int axis, ca_size_t kth, i
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(float *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = outer_off + k * axis_vstride;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -8074,7 +8068,7 @@ partition_addr_ki_native_f32 (VALUE self, CArray *ca, int axis, ca_size_t kth, i
 }
 typedef struct {
-  double v;
+  ca_u_float64_t v;
   ca_size_t i;
 } partition_addr_pair_f64;
@@ -8221,13 +8215,13 @@ partition_addr_ki_native_f64 (VALUE self, CArray *ca, int axis, ca_size_t kth, i
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(double *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(double *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
@@ -8235,13 +8229,13 @@ partition_addr_ki_native_f64 (VALUE self, CArray *ca, int axis, ca_size_t kth, i
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(double *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(double *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
@@ -8249,7 +8243,7 @@ partition_addr_ki_native_f64 (VALUE self, CArray *ca, int axis, ca_size_t kth, i
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(double *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = outer_off + k * axis_vstride;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -8432,13 +8426,13 @@ partition_addr_ki_native_bool (VALUE self, CArray *ca, int axis, ca_size_t kth,
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(boolean8_t *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(boolean8_t *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
@@ -8446,13 +8440,13 @@ partition_addr_ki_native_bool (VALUE self, CArray *ca, int axis, ca_size_t kth,
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(boolean8_t *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(boolean8_t *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
@@ -8460,7 +8454,7 @@ partition_addr_ki_native_bool (VALUE self, CArray *ca, int axis, ca_size_t kth,
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(boolean8_t *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = outer_off + k * axis_vstride;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -8667,13 +8661,13 @@ partition_addr_ki_native_object_body (VALUE arg)
       if ( masked_last ) {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_valid].v = *(VALUE *) (pi + k * istride);
+          memcpy(&buf[n_valid].v, pi + k * istride, sizeof(buf[n_valid].v));
           buf[n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_valid + n_masked].v = *(VALUE *) (pi + k * istride);
+          memcpy(&buf[n_valid + n_masked].v, pi + k * istride, sizeof(buf[n_valid + n_masked].v));
           buf[n_valid + n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
@@ -8681,13 +8675,13 @@ partition_addr_ki_native_object_body (VALUE arg)
       } else {
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( ! mi[k * mstride] ) continue;
-          buf[n_masked].v = *(VALUE *) (pi + k * istride);
+          memcpy(&buf[n_masked].v, pi + k * istride, sizeof(buf[n_masked].v));
           buf[n_masked].i = outer_off + k * axis_vstride;
           n_masked++;
         }
         for ( ca_size_t k = 0; k < fiber_n; k++ ) {
           if ( mi[k * mstride] ) continue;
-          buf[n_masked + n_valid].v = *(VALUE *) (pi + k * istride);
+          memcpy(&buf[n_masked + n_valid].v, pi + k * istride, sizeof(buf[n_masked + n_valid].v));
           buf[n_masked + n_valid].i = outer_off + k * axis_vstride;
           n_valid++;
         }
@@ -8695,7 +8689,7 @@ partition_addr_ki_native_object_body (VALUE arg)
       }
     } else {
       for ( ca_size_t k = 0; k < fiber_n; k++ ) {
-          buf[k].v = *(VALUE *) (pi + k * istride);
+          memcpy(&buf[k].v, pi + k * istride, sizeof(buf[k].v));
         buf[k].i = outer_off + k * axis_vstride;
       }
       sort_lo = 0; sort_n = fiber_n;
@@ -9068,7 +9062,7 @@ rank_index_ki_native_i8 (VALUE self, CArray *ca, int axis, int do_stable, int de
     ca_size_t k_unmasked = 0;
     for ( ca_size_t k = 0; k < fiber_n; k++ ) {
       if ( mi && mi[k * mstride] ) continue;
-      buf[k_unmasked].v = *(int8_t *) (pi + k * istride);
+      memcpy(&buf[k_unmasked].v, pi + k * istride, sizeof(buf[k_unmasked].v));
       buf[k_unmasked].i = k;
       k_unmasked++;
     }
@@ -9199,7 +9193,7 @@ rank_index_ki_native_u8 (VALUE self, CArray *ca, int axis, int do_stable, int de
     ca_size_t k_unmasked = 0;
     for ( ca_size_t k = 0; k < fiber_n; k++ ) {
       if ( mi && mi[k * mstride] ) continue;
-      buf[k_unmasked].v = *(uint8_t *) (pi + k * istride);
+      memcpy(&buf[k_unmasked].v, pi + k * istride, sizeof(buf[k_unmasked].v));
       buf[k_unmasked].i = k;
       k_unmasked++;
     }
@@ -9274,7 +9268,7 @@ rank_index_ki_native_u8 (VALUE self, CArray *ca, int axis, int do_stable, int de
 }
 typedef struct {
-  int16_t v;
+  ca_u_int16_t v;
   ca_size_t i;
 } rank_index_pair_i16;
@@ -9330,7 +9324,7 @@ rank_index_ki_native_i16 (VALUE self, CArray *ca, int axis, int do_stable, int d
     ca_size_t k_unmasked = 0;
     for ( ca_size_t k = 0; k < fiber_n; k++ ) {
       if ( mi && mi[k * mstride] ) continue;
-      buf[k_unmasked].v = *(int16_t *) (pi + k * istride);
+      memcpy(&buf[k_unmasked].v, pi + k * istride, sizeof(buf[k_unmasked].v));
       buf[k_unmasked].i = k;
       k_unmasked++;
     }
@@ -9405,7 +9399,7 @@ rank_index_ki_native_i16 (VALUE self, CArray *ca, int axis, int do_stable, int d
 }
 typedef struct {
-  uint16_t v;
+  ca_u_uint16_t v;
   ca_size_t i;
 } rank_index_pair_u16;
@@ -9461,7 +9455,7 @@ rank_index_ki_native_u16 (VALUE self, CArray *ca, int axis, int do_stable, int d
     ca_size_t k_unmasked = 0;
     for ( ca_size_t k = 0; k < fiber_n; k++ ) {
       if ( mi && mi[k * mstride] ) continue;
-      buf[k_unmasked].v = *(uint16_t *) (pi + k * istride);
+      memcpy(&buf[k_unmasked].v, pi + k * istride, sizeof(buf[k_unmasked].v));
       buf[k_unmasked].i = k;
       k_unmasked++;
     }
@@ -9536,7 +9530,7 @@ rank_index_ki_native_u16 (VALUE self, CArray *ca, int axis, int do_stable, int d
 }
 typedef struct {
-  int32_t v;
+  ca_u_int32_t v;
   ca_size_t i;
 } rank_index_pair_i32;
@@ -9592,7 +9586,7 @@ rank_index_ki_native_i32 (VALUE self, CArray *ca, int axis, int do_stable, int d
     ca_size_t k_unmasked = 0;
     for ( ca_size_t k = 0; k < fiber_n; k++ ) {
       if ( mi && mi[k * mstride] ) continue;
-      buf[k_unmasked].v = *(int32_t *) (pi + k * istride);
+      memcpy(&buf[k_unmasked].v, pi + k * istride, sizeof(buf[k_unmasked].v));
       buf[k_unmasked].i = k;
       k_unmasked++;
     }
@@ -9667,7 +9661,7 @@ rank_index_ki_native_i32 (VALUE self, CArray *ca, int axis, int do_stable, int d
 }
 typedef struct {
-  uint32_t v;
+  ca_u_uint32_t v;
   ca_size_t i;
 } rank_index_pair_u32;
@@ -9723,7 +9717,7 @@ rank_index_ki_native_u32 (VALUE self, CArray *ca, int axis, int do_stable, int d
     ca_size_t k_unmasked = 0;
     for ( ca_size_t k = 0; k < fiber_n; k++ ) {
       if ( mi && mi[k * mstride] ) continue;
-      buf[k_unmasked].v = *(uint32_t *) (pi + k * istride);
+      memcpy(&buf[k_unmasked].v, pi + k * istride, sizeof(buf[k_unmasked].v));
       buf[k_unmasked].i = k;
       k_unmasked++;
     }
@@ -9798,7 +9792,7 @@ rank_index_ki_native_u32 (VALUE self, CArray *ca, int axis, int do_stable, int d
 }
 typedef struct {
-  int64_t v;
+  ca_u_int64_t v;
   ca_size_t i;
 } rank_index_pair_i64;
@@ -9854,7 +9848,7 @@ rank_index_ki_native_i64 (VALUE self, CArray *ca, int axis, int do_stable, int d
     ca_size_t k_unmasked = 0;
     for ( ca_size_t k = 0; k < fiber_n; k++ ) {
       if ( mi && mi[k * mstride] ) continue;
-      buf[k_unmasked].v = *(int64_t *) (pi + k * istride);
+      memcpy(&buf[k_unmasked].v, pi + k * istride, sizeof(buf[k_unmasked].v));
       buf[k_unmasked].i = k;
       k_unmasked++;
     }
@@ -9929,7 +9923,7 @@ rank_index_ki_native_i64 (VALUE self, CArray *ca, int axis, int do_stable, int d
 }
 typedef struct {
-  uint64_t v;
+  ca_u_uint64_t v;
   ca_size_t i;
 } rank_index_pair_u64;
@@ -9985,7 +9979,7 @@ rank_index_ki_native_u64 (VALUE self, CArray *ca, int axis, int do_stable, int d
     ca_size_t k_unmasked = 0;
     for ( ca_size_t k = 0; k < fiber_n; k++ ) {
       if ( mi && mi[k * mstride] ) continue;
-      buf[k_unmasked].v = *(uint64_t *) (pi + k * istride);
+      memcpy(&buf[k_unmasked].v, pi + k * istride, sizeof(buf[k_unmasked].v));
       buf[k_unmasked].i = k;
       k_unmasked++;
     }
@@ -10060,7 +10054,7 @@ rank_index_ki_native_u64 (VALUE self, CArray *ca, int axis, int do_stable, int d
 }
 typedef struct {
-  float v;
+  ca_u_float32_t v;
   ca_size_t i;
 } rank_index_pair_f32;
@@ -10116,7 +10110,7 @@ rank_index_ki_native_f32 (VALUE self, CArray *ca, int axis, int do_stable, int d
     ca_size_t k_unmasked = 0;
     for ( ca_size_t k = 0; k < fiber_n; k++ ) {
       if ( mi && mi[k * mstride] ) continue;
-      buf[k_unmasked].v = *(float *) (pi + k * istride);
+      memcpy(&buf[k_unmasked].v, pi + k * istride, sizeof(buf[k_unmasked].v));
       buf[k_unmasked].i = k;
       k_unmasked++;
     }
@@ -10197,7 +10191,7 @@ rank_index_ki_native_f32 (VALUE self, CArray *ca, int axis, int do_stable, int d
 }
 typedef struct {
-  double v;
+  ca_u_float64_t v;
   ca_size_t i;
 } rank_index_pair_f64;
@@ -10253,7 +10247,7 @@ rank_index_ki_native_f64 (VALUE self, CArray *ca, int axis, int do_stable, int d
     ca_size_t k_unmasked = 0;
     for ( ca_size_t k = 0; k < fiber_n; k++ ) {
       if ( mi && mi[k * mstride] ) continue;
-      buf[k_unmasked].v = *(double *) (pi + k * istride);
+      memcpy(&buf[k_unmasked].v, pi + k * istride, sizeof(buf[k_unmasked].v));
       buf[k_unmasked].i = k;
       k_unmasked++;
     }
@@ -10390,7 +10384,7 @@ rank_index_ki_native_bool (VALUE self, CArray *ca, int axis, int do_stable, int
     ca_size_t k_unmasked = 0;
     for ( ca_size_t k = 0; k < fiber_n; k++ ) {
       if ( mi && mi[k * mstride] ) continue;
-      buf[k_unmasked].v = *(boolean8_t *) (pi + k * istride);
+      memcpy(&buf[k_unmasked].v, pi + k * istride, sizeof(buf[k_unmasked].v));
       buf[k_unmasked].i = k;
       k_unmasked++;
     }
@@ -10555,7 +10549,7 @@ rank_index_ki_native_object_body (VALUE arg)
     ca_size_t k_unmasked = 0;
     for ( ca_size_t k = 0; k < fiber_n; k++ ) {
       if ( mi && mi[k * mstride] ) continue;
-      buf[k_unmasked].v = *(VALUE *) (pi + k * istride);
+      memcpy(&buf[k_unmasked].v, pi + k * istride, sizeof(buf[k_unmasked].v));
       buf[k_unmasked].i = k;
       k_unmasked++;
     }
@@ -10697,10 +10691,7 @@ rank_index_ki_native_fixlen (VALUE self, CArray *ca, int axis, int do_stable, in
   ca_size_t fiber_n = st_in.slab_elements;
   volatile VALUE hbuf = 0, haux = 0;
   rank_index_pair_fixlen *buf = ALLOCV_N(rank_index_pair_fixlen, hbuf, fiber_n);
-  rank_index_pair_fixlen *aux = NULL;
-  if ( do_stable ) {
-    aux = ALLOCV_N(rank_index_pair_fixlen, haux, fiber_n);
-  }
+  (void) haux;
   char       *pi, *po;
   boolean8_t *mi, *mo;