numo-narray-alt 0.9.10 → 0.9.12

data/ext/numo/narray/src/t_uint16.c

@@ -43,6 +43,50 @@ static ID id_to_a;
 VALUE cT;
 extern VALUE cRT;
 
+#include "mh/coerce_cast.h"
+#include "mh/to_a.h"
+#include "mh/fill.h"
+#include "mh/format.h"
+#include "mh/format_to_a.h"
+#include "mh/inspect.h"
+#include "mh/op/add.h"
+#include "mh/op/sub.h"
+#include "mh/op/mul.h"
+#include "mh/op/div.h"
+#include "mh/op/mod.h"
+#include "mh/divmod.h"
+#include "mh/comp/eq.h"
+#include "mh/comp/ne.h"
+#include "mh/comp/gt.h"
+#include "mh/comp/ge.h"
+#include "mh/comp/lt.h"
+#include "mh/comp/le.h"
+#include "mh/bit/and.h"
+#include "mh/bit/or.h"
+#include "mh/bit/xor.h"
+#include "mh/bit/not.h"
+#include "mh/bit/left_shift.h"
+#include "mh/bit/right_shift.h"
+#include "mh/clip.h"
+#include "mh/sum.h"
+#include "mh/prod.h"
+#include "mh/min.h"
+#include "mh/max.h"
+#include "mh/ptp.h"
+#include "mh/max_index.h"
+#include "mh/min_index.h"
+#include "mh/argmax.h"
+#include "mh/argmin.h"
+#include "mh/maximum.h"
+#include "mh/minimum.h"
+#include "mh/minmax.h"
+#include "mh/bincount.h"
+#include "mh/cumsum.h"
+#include "mh/cumprod.h"
+#include "mh/mulsum.h"
+#include "mh/seq.h"
+#include "mh/eye.h"
+#include "mh/rand.h"
 #include "mh/mean.h"
 #include "mh/var.h"
 #include "mh/stddev.h"
@@ -50,6 +94,50 @@ extern VALUE cRT;
 
 typedef u_int16_t uint16; // Type aliases for shorter notation
                           // following the codebase naming convention.
+DEF_NARRAY_COERCE_CAST_METHOD_FUNC(uint16)
+DEF_NARRAY_TO_A_METHOD_FUNC(uint16)
+DEF_NARRAY_FILL_METHOD_FUNC(uint16)
+DEF_NARRAY_FORMAT_METHOD_FUNC(uint16)
+DEF_NARRAY_FORMAT_TO_A_METHOD_FUNC(uint16)
+DEF_NARRAY_INSPECT_METHOD_FUNC(uint16)
+DEF_NARRAY_ADD_METHOD_FUNC(uint16, numo_cUInt16)
+DEF_NARRAY_SUB_METHOD_FUNC(uint16, numo_cUInt16)
+DEF_NARRAY_MUL_METHOD_FUNC(uint16, numo_cUInt16)
+DEF_NARRAY_INT_DIV_METHOD_FUNC(uint16, numo_cUInt16)
+DEF_NARRAY_INT_MOD_METHOD_FUNC(uint16, numo_cUInt16)
+DEF_NARRAY_INT_DIVMOD_METHOD_FUNC(uint16, numo_cUInt16)
+DEF_NARRAY_EQ_METHOD_FUNC(uint16, numo_cUInt16)
+DEF_NARRAY_NE_METHOD_FUNC(uint16, numo_cUInt16)
+DEF_NARRAY_GT_METHOD_FUNC(uint16, numo_cUInt16)
+DEF_NARRAY_GE_METHOD_FUNC(uint16, numo_cUInt16)
+DEF_NARRAY_LT_METHOD_FUNC(uint16, numo_cUInt16)
+DEF_NARRAY_LE_METHOD_FUNC(uint16, numo_cUInt16)
+DEF_NARRAY_INT_BIT_AND_METHOD_FUNC(uint16, numo_cUInt16)
+DEF_NARRAY_INT_BIT_OR_METHOD_FUNC(uint16, numo_cUInt16)
+DEF_NARRAY_INT_BIT_XOR_METHOD_FUNC(uint16, numo_cUInt16)
+DEF_NARRAY_INT_BIT_NOT_METHOD_FUNC(uint16, numo_cUInt16)
+DEF_NARRAY_INT_LEFT_SHIFT_METHOD_FUNC(uint16, numo_cUInt16)
+DEF_NARRAY_INT_RIGHT_SHIFT_METHOD_FUNC(uint16, numo_cUInt16)
+DEF_NARRAY_CLIP_METHOD_FUNC(uint16, numo_cUInt16)
+DEF_NARRAY_INT_SUM_METHOD_FUNC(uint16, numo_cUInt16, u_int64_t, numo_cUInt64)
+DEF_NARRAY_INT_PROD_METHOD_FUNC(uint16, numo_cUInt16, u_int64_t, numo_cUInt64)
+DEF_NARRAY_INT_MIN_METHOD_FUNC(uint16, numo_cUInt16)
+DEF_NARRAY_INT_MAX_METHOD_FUNC(uint16, numo_cUInt16)
+DEF_NARRAY_INT_PTP_METHOD_FUNC(uint16, numo_cUInt16)
+DEF_NARRAY_INT_MAX_INDEX_METHOD_FUNC(uint16)
+DEF_NARRAY_INT_MIN_INDEX_METHOD_FUNC(uint16)
+DEF_NARRAY_INT_ARGMAX_METHOD_FUNC(uint16)
+DEF_NARRAY_INT_ARGMIN_METHOD_FUNC(uint16)
+DEF_NARRAY_INT_MAXIMUM_METHOD_FUNC(uint16, numo_cUInt16)
+DEF_NARRAY_INT_MINIMUM_METHOD_FUNC(uint16, numo_cUInt16)
+DEF_NARRAY_INT_MINMAX_METHOD_FUNC(uint16, numo_cUInt16)
+DEF_NARRAY_UINT_BINCOUNT_METHOD_FUNC(uint16, numo_cUInt16)
+DEF_NARRAY_INT_CUMSUM_METHOD_FUNC(uint16, numo_cUInt16)
+DEF_NARRAY_INT_CUMPROD_METHOD_FUNC(uint16, numo_cUInt16)
+DEF_NARRAY_INT_MULSUM_METHOD_FUNC(uint16, numo_cUInt16)
+DEF_NARRAY_INT_SEQ_METHOD_FUNC(uint16)
+DEF_NARRAY_EYE_METHOD_FUNC(uint16)
+DEF_NARRAY_INT_RAND_METHOD_FUNC(uint16)
 DEF_NARRAY_INT_MEAN_METHOD_FUNC(uint16, numo_cUInt16)
 DEF_NARRAY_INT_VAR_METHOD_FUNC(uint16, numo_cUInt16)
 DEF_NARRAY_INT_STDDEV_METHOD_FUNC(uint16, numo_cUInt16)
@@ -1214,204 +1302,6 @@ static VALUE uint16_aset(int argc, VALUE* argv, VALUE self) {
   return argv[argc];
 }
 
-/*
-  return NArray with cast to the type of self.
-  @overload coerce_cast(type)
-    @return [nil]
-*/
-static VALUE uint16_coerce_cast(VALUE self, VALUE type) {
-  return Qnil;
-}
-
-static void iter_uint16_to_a(na_loop_t* const lp) {
-  size_t i, s1;
-  char* p1;
-  size_t* idx1;
-  dtype x;
-  volatile VALUE a, y;
-
-  INIT_COUNTER(lp, i);
-  INIT_PTR_IDX(lp, 0, p1, s1, idx1);
-  a = rb_ary_new2(i);
-  rb_ary_push(lp->args[1].value, a);
-  if (idx1) {
-    for (; i--;) {
-      GET_DATA_INDEX(p1, idx1, dtype, x);
-      y = m_data_to_num(x);
-      rb_ary_push(a, y);
-    }
-  } else {
-    for (; i--;) {
-      GET_DATA_STRIDE(p1, s1, dtype, x);
-      y = m_data_to_num(x);
-      rb_ary_push(a, y);
-    }
-  }
-}
-
-/*
-  Convert self to Array.
-  @overload to_a
-    @return [Array]
-*/
-static VALUE uint16_to_a(VALUE self) {
-  ndfunc_arg_in_t ain[3] = { { Qnil, 0 }, { sym_loop_opt }, { sym_option } };
-  ndfunc_arg_out_t aout[1] = { { rb_cArray, 0 } }; // dummy?
-  ndfunc_t ndf = { iter_uint16_to_a, FULL_LOOP_NIP, 3, 1, ain, aout };
-  return na_ndloop_cast_narray_to_rarray(&ndf, self, Qnil);
-}
-
-static void iter_uint16_fill(na_loop_t* const lp) {
-  size_t i;
-  char* p1;
-  ssize_t s1;
-  size_t* idx1;
-  VALUE x = lp->option;
-  dtype y;
-  INIT_COUNTER(lp, i);
-  INIT_PTR_IDX(lp, 0, p1, s1, idx1);
-  y = m_num_to_data(x);
-  if (idx1) {
-    for (; i--;) {
-      SET_DATA_INDEX(p1, idx1, dtype, y);
-    }
-  } else {
-    for (; i--;) {
-      SET_DATA_STRIDE(p1, s1, dtype, y);
-    }
-  }
-}
-
-/*
-  Fill elements with other.
-  @overload fill other
-    @param [Numeric] other
-    @return [Numo::UInt16] self.
-*/
-static VALUE uint16_fill(VALUE self, VALUE val) {
-  ndfunc_arg_in_t ain[2] = { { OVERWRITE, 0 }, { sym_option } };
-  ndfunc_t ndf = { iter_uint16_fill, FULL_LOOP, 2, 0, ain, 0 };
-
-  na_ndloop(&ndf, 2, self, val);
-  return self;
-}
-
-static VALUE format_uint16(VALUE fmt, dtype* x) {
-  // fix-me
-  char s[48];
-  int n;
-
-  if (NIL_P(fmt)) {
-    n = m_sprintf(s, *x);
-    return rb_str_new(s, n);
-  }
-  return rb_funcall(fmt, '%', 1, m_data_to_num(*x));
-}
-
-static void iter_uint16_format(na_loop_t* const lp) {
-  size_t i;
-  char *p1, *p2;
-  ssize_t s1, s2;
-  size_t* idx1;
-  dtype* x;
-  VALUE y;
-  VALUE fmt = lp->option;
-  INIT_COUNTER(lp, i);
-  INIT_PTR_IDX(lp, 0, p1, s1, idx1);
-  INIT_PTR(lp, 1, p2, s2);
-  if (idx1) {
-    for (; i--;) {
-      x = (dtype*)(p1 + *idx1);
-      idx1++;
-      y = format_uint16(fmt, x);
-      SET_DATA_STRIDE(p2, s2, VALUE, y);
-    }
-  } else {
-    for (; i--;) {
-      x = (dtype*)p1;
-      p1 += s1;
-      y = format_uint16(fmt, x);
-      SET_DATA_STRIDE(p2, s2, VALUE, y);
-    }
-  }
-}
-
-/*
-  Format elements into strings.
-  @overload format format
-    @param [String] format
-    @return [Numo::RObject] array of formatted strings.
-*/
-static VALUE uint16_format(int argc, VALUE* argv, VALUE self) {
-  VALUE fmt = Qnil;
-
-  ndfunc_arg_in_t ain[2] = { { Qnil, 0 }, { sym_option } };
-  ndfunc_arg_out_t aout[1] = { { numo_cRObject, 0 } };
-  ndfunc_t ndf = { iter_uint16_format, FULL_LOOP_NIP, 2, 1, ain, aout };
-
-  rb_scan_args(argc, argv, "01", &fmt);
-  return na_ndloop(&ndf, 2, self, fmt);
-}
-
-static void iter_uint16_format_to_a(na_loop_t* const lp) {
-  size_t i;
-  char* p1;
-  ssize_t s1;
-  size_t* idx1;
-  dtype* x;
-  VALUE y;
-  volatile VALUE a;
-  VALUE fmt = lp->option;
-  INIT_COUNTER(lp, i);
-  INIT_PTR_IDX(lp, 0, p1, s1, idx1);
-  a = rb_ary_new2(i);
-  rb_ary_push(lp->args[1].value, a);
-  if (idx1) {
-    for (; i--;) {
-      x = (dtype*)(p1 + *idx1);
-      idx1++;
-      y = format_uint16(fmt, x);
-      rb_ary_push(a, y);
-    }
-  } else {
-    for (; i--;) {
-      x = (dtype*)p1;
-      p1 += s1;
-      y = format_uint16(fmt, x);
-      rb_ary_push(a, y);
-    }
-  }
-}
-
-/*
-  Format elements into strings.
-  @overload format_to_a format
-    @param [String] format
-    @return [Array] array of formatted strings.
-*/
-static VALUE uint16_format_to_a(int argc, VALUE* argv, VALUE self) {
-  VALUE fmt = Qnil;
-  ndfunc_arg_in_t ain[3] = { { Qnil, 0 }, { sym_loop_opt }, { sym_option } };
-  ndfunc_arg_out_t aout[1] = { { rb_cArray, 0 } }; // dummy?
-  ndfunc_t ndf = { iter_uint16_format_to_a, FULL_LOOP_NIP, 3, 1, ain, aout };
-
-  rb_scan_args(argc, argv, "01", &fmt);
-  return na_ndloop_cast_narray_to_rarray(&ndf, self, fmt);
-}
-
-static VALUE iter_uint16_inspect(char* ptr, size_t pos, VALUE fmt) {
-  return format_uint16(fmt, (dtype*)(ptr + pos));
-}
-
-/*
-  Returns a string containing a human-readable representation of NArray.
-  @overload inspect
-    @return [String]
-*/
-static VALUE uint16_inspect(VALUE ary) {
-  return na_ndloop_inspect(ary, iter_uint16_inspect, Qnil);
-}
-
 static void iter_uint16_each(na_loop_t* const lp) {
   size_t i, s1;
   char* p1;
@@ -1738,3354 +1628,372 @@ static VALUE uint16_abs(VALUE self) {
   return na_ndloop(&ndf, 1, self);
 }
 
-#define check_intdivzero(y)                                                                    \
-  {}
-
-static void iter_uint16_add(na_loop_t* const lp) {
-  size_t i = 0;
-  size_t n;
+static void iter_uint16_pow(na_loop_t* const lp) {
+  size_t i;
   char *p1, *p2, *p3;
   ssize_t s1, s2, s3;
-
-  INIT_COUNTER(lp, n);
+  dtype x, y;
+  INIT_COUNTER(lp, i);
   INIT_PTR(lp, 0, p1, s1);
   INIT_PTR(lp, 1, p2, s2);
   INIT_PTR(lp, 2, p3, s3);
-
-  //
-  if (is_aligned(p1, sizeof(dtype)) && is_aligned(p2, sizeof(dtype)) &&
-      is_aligned(p3, sizeof(dtype))) {
-
-    if (s1 == sizeof(dtype) && s2 == sizeof(dtype) && s3 == sizeof(dtype)) {
-      if (p1 == p3) { // inplace case
-        for (; i < n; i++) {
-          check_intdivzero(((dtype*)p2)[i]);
-          ((dtype*)p1)[i] = m_add(((dtype*)p1)[i], ((dtype*)p2)[i]);
-        }
-      } else {
-        for (; i < n; i++) {
-          check_intdivzero(((dtype*)p2)[i]);
-          ((dtype*)p3)[i] = m_add(((dtype*)p1)[i], ((dtype*)p2)[i]);
-        }
-      }
-      return;
-    }
-
-    if (is_aligned_step(s1, sizeof(dtype)) && is_aligned_step(s2, sizeof(dtype)) &&
-        is_aligned_step(s3, sizeof(dtype))) {
-      //
-
-      if (s2 == 0) { // Broadcasting from scalar value.
-        check_intdivzero(*(dtype*)p2);
-        if (s1 == sizeof(dtype) && s3 == sizeof(dtype)) {
-          if (p1 == p3) { // inplace case
-            for (; i < n; i++) {
-              ((dtype*)p1)[i] = m_add(((dtype*)p1)[i], *(dtype*)p2);
-            }
-          } else {
-            for (; i < n; i++) {
-              ((dtype*)p3)[i] = m_add(((dtype*)p1)[i], *(dtype*)p2);
-            }
-          }
-        } else {
-          for (i = 0; i < n; i++) {
-            *(dtype*)p3 = m_add(*(dtype*)p1, *(dtype*)p2);
-            p1 += s1;
-            p3 += s3;
-          }
-        }
-      } else {
-        if (p1 == p3) { // inplace case
-          for (i = 0; i < n; i++) {
-            check_intdivzero(*(dtype*)p2);
-            *(dtype*)p1 = m_add(*(dtype*)p1, *(dtype*)p2);
-            p1 += s1;
-            p2 += s2;
-          }
-        } else {
-          for (i = 0; i < n; i++) {
-            check_intdivzero(*(dtype*)p2);
-            *(dtype*)p3 = m_add(*(dtype*)p1, *(dtype*)p2);
-            p1 += s1;
-            p2 += s2;
-            p3 += s3;
-          }
-        }
-      }
-
-      return;
-      //
-    }
-  }
-  for (i = 0; i < n; i++) {
-    dtype x, y, z;
+  for (; i--;) {
     GET_DATA_STRIDE(p1, s1, dtype, x);
     GET_DATA_STRIDE(p2, s2, dtype, y);
-    check_intdivzero(y);
-    z = m_add(x, y);
-    SET_DATA_STRIDE(p3, s3, dtype, z);
+    x = m_pow(x, y);
+    SET_DATA_STRIDE(p3, s3, dtype, x);
+  }
+}
+
+static void iter_uint16_pow_int32(na_loop_t* const lp) {
+  size_t i;
+  char *p1, *p2, *p3;
+  ssize_t s1, s2, s3;
+  dtype x;
+  int32_t y;
+  INIT_COUNTER(lp, i);
+  INIT_PTR(lp, 0, p1, s1);
+  INIT_PTR(lp, 1, p2, s2);
+  INIT_PTR(lp, 2, p3, s3);
+  for (; i--;) {
+    GET_DATA_STRIDE(p1, s1, dtype, x);
+    GET_DATA_STRIDE(p2, s2, int32_t, y);
+    x = m_pow_int(x, y);
+    SET_DATA_STRIDE(p3, s3, dtype, x);
   }
-  //
 }
-#undef check_intdivzero
 
-static VALUE uint16_add_self(VALUE self, VALUE other) {
+static VALUE uint16_pow_self(VALUE self, VALUE other) {
   ndfunc_arg_in_t ain[2] = { { cT, 0 }, { cT, 0 } };
+  ndfunc_arg_in_t ain_i[2] = { { cT, 0 }, { numo_cInt32, 0 } };
   ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_uint16_add, STRIDE_LOOP, 2, 1, ain, aout };
+  ndfunc_t ndf = { iter_uint16_pow, STRIDE_LOOP, 2, 1, ain, aout };
+  ndfunc_t ndf_i = { iter_uint16_pow_int32, STRIDE_LOOP, 2, 1, ain_i, aout };
 
-  return na_ndloop(&ndf, 2, self, other);
+  // fixme : use na.integer?
+  if (FIXNUM_P(other) || rb_obj_is_kind_of(other, numo_cInt32)) {
+    return na_ndloop(&ndf_i, 2, self, other);
+  } else {
+    return na_ndloop(&ndf, 2, self, other);
+  }
 }
 
 /*
-  Binary add.
-  @overload + other
+  Binary power.
+  @overload ** other
     @param [Numo::NArray,Numeric] other
-    @return [Numo::NArray] self + other
+    @return [Numo::NArray] self to the other-th power.
 */
-static VALUE uint16_add(VALUE self, VALUE other) {
+static VALUE uint16_pow(VALUE self, VALUE other) {
 
   VALUE klass, v;
-
   klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
   if (klass == cT) {
-    return uint16_add_self(self, other);
+    return uint16_pow_self(self, other);
   } else {
     v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, '+', 1, other);
+    return rb_funcall(v, id_pow, 1, other);
   }
 }
 
-#define check_intdivzero(y)                                                                    \
-  {}
-
-static void iter_uint16_sub(na_loop_t* const lp) {
-  size_t i = 0;
-  size_t n;
-  char *p1, *p2, *p3;
-  ssize_t s1, s2, s3;
+static void iter_uint16_minus(na_loop_t* const lp) {
+  size_t i, n;
+  char *p1, *p2;
+  ssize_t s1, s2;
+  size_t *idx1, *idx2;
+  dtype x;
 
   INIT_COUNTER(lp, n);
-  INIT_PTR(lp, 0, p1, s1);
-  INIT_PTR(lp, 1, p2, s2);
-  INIT_PTR(lp, 2, p3, s3);
-
-  //
-  if (is_aligned(p1, sizeof(dtype)) && is_aligned(p2, sizeof(dtype)) &&
-      is_aligned(p3, sizeof(dtype))) {
+  INIT_PTR_IDX(lp, 0, p1, s1, idx1);
+  INIT_PTR_IDX(lp, 1, p2, s2, idx2);
 
-    if (s1 == sizeof(dtype) && s2 == sizeof(dtype) && s3 == sizeof(dtype)) {
-      if (p1 == p3) { // inplace case
-        for (; i < n; i++) {
-          check_intdivzero(((dtype*)p2)[i]);
-          ((dtype*)p1)[i] = m_sub(((dtype*)p1)[i], ((dtype*)p2)[i]);
-        }
-      } else {
-        for (; i < n; i++) {
-          check_intdivzero(((dtype*)p2)[i]);
-          ((dtype*)p3)[i] = m_sub(((dtype*)p1)[i], ((dtype*)p2)[i]);
-        }
+  if (idx1) {
+    if (idx2) {
+      for (i = 0; i < n; i++) {
+        GET_DATA_INDEX(p1, idx1, dtype, x);
+        x = m_minus(x);
+        SET_DATA_INDEX(p2, idx2, dtype, x);
+      }
+    } else {
+      for (i = 0; i < n; i++) {
+        GET_DATA_INDEX(p1, idx1, dtype, x);
+        x = m_minus(x);
+        SET_DATA_STRIDE(p2, s2, dtype, x);
       }
-      return;
     }
-
-    if (is_aligned_step(s1, sizeof(dtype)) && is_aligned_step(s2, sizeof(dtype)) &&
-        is_aligned_step(s3, sizeof(dtype))) {
+  } else {
+    if (idx2) {
+      for (i = 0; i < n; i++) {
+        GET_DATA_STRIDE(p1, s1, dtype, x);
+        x = m_minus(x);
+        SET_DATA_INDEX(p2, idx2, dtype, x);
+      }
+    } else {
       //
-
-      if (s2 == 0) { // Broadcasting from scalar value.
-        check_intdivzero(*(dtype*)p2);
-        if (s1 == sizeof(dtype) && s3 == sizeof(dtype)) {
-          if (p1 == p3) { // inplace case
-            for (; i < n; i++) {
-              ((dtype*)p1)[i] = m_sub(((dtype*)p1)[i], *(dtype*)p2);
-            }
-          } else {
-            for (; i < n; i++) {
-              ((dtype*)p3)[i] = m_sub(((dtype*)p1)[i], *(dtype*)p2);
-            }
-          }
-        } else {
+      if (is_aligned(p1, sizeof(dtype)) && is_aligned(p2, sizeof(dtype))) {
+        if (s1 == sizeof(dtype) && s2 == sizeof(dtype)) {
           for (i = 0; i < n; i++) {
-            *(dtype*)p3 = m_sub(*(dtype*)p1, *(dtype*)p2);
-            p1 += s1;
-            p3 += s3;
+            ((dtype*)p2)[i] = m_minus(((dtype*)p1)[i]);
           }
+          return;
         }
-      } else {
-        if (p1 == p3) { // inplace case
-          for (i = 0; i < n; i++) {
-            check_intdivzero(*(dtype*)p2);
-            *(dtype*)p1 = m_sub(*(dtype*)p1, *(dtype*)p2);
-            p1 += s1;
-            p2 += s2;
-          }
-        } else {
+        if (is_aligned_step(s1, sizeof(dtype)) && is_aligned_step(s2, sizeof(dtype))) {
+          //
           for (i = 0; i < n; i++) {
-            check_intdivzero(*(dtype*)p2);
-            *(dtype*)p3 = m_sub(*(dtype*)p1, *(dtype*)p2);
+            *(dtype*)p2 = m_minus(*(dtype*)p1);
             p1 += s1;
             p2 += s2;
-            p3 += s3;
           }
+          return;
+          //
         }
       }
-
-      return;
+      for (i = 0; i < n; i++) {
+        GET_DATA_STRIDE(p1, s1, dtype, x);
+        x = m_minus(x);
+        SET_DATA_STRIDE(p2, s2, dtype, x);
+      }
       //
     }
   }
-  for (i = 0; i < n; i++) {
-    dtype x, y, z;
-    GET_DATA_STRIDE(p1, s1, dtype, x);
-    GET_DATA_STRIDE(p2, s2, dtype, y);
-    check_intdivzero(y);
-    z = m_sub(x, y);
-    SET_DATA_STRIDE(p3, s3, dtype, z);
-  }
-  //
 }
-#undef check_intdivzero
 
-static VALUE uint16_sub_self(VALUE self, VALUE other) {
-  ndfunc_arg_in_t ain[2] = { { cT, 0 }, { cT, 0 } };
+/*
+  Unary minus.
+  @overload -@
+    @return [Numo::UInt16] minus of self.
+*/
+static VALUE uint16_minus(VALUE self) {
+  ndfunc_arg_in_t ain[1] = { { cT, 0 } };
   ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_uint16_sub, STRIDE_LOOP, 2, 1, ain, aout };
+  ndfunc_t ndf = { iter_uint16_minus, FULL_LOOP, 1, 1, ain, aout };
 
-  return na_ndloop(&ndf, 2, self, other);
+  return na_ndloop(&ndf, 1, self);
 }
 
-/*
-  Binary sub.
-  @overload - other
-    @param [Numo::NArray,Numeric] other
-    @return [Numo::NArray] self - other
-*/
-static VALUE uint16_sub(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return uint16_sub_self(self, other);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, '-', 1, other);
-  }
-}
-
-#define check_intdivzero(y)                                                                    \
-  {}
-
-static void iter_uint16_mul(na_loop_t* const lp) {
-  size_t i = 0;
-  size_t n;
-  char *p1, *p2, *p3;
-  ssize_t s1, s2, s3;
+static void iter_uint16_reciprocal(na_loop_t* const lp) {
+  size_t i, n;
+  char *p1, *p2;
+  ssize_t s1, s2;
+  size_t *idx1, *idx2;
+  dtype x;
 
   INIT_COUNTER(lp, n);
-  INIT_PTR(lp, 0, p1, s1);
-  INIT_PTR(lp, 1, p2, s2);
-  INIT_PTR(lp, 2, p3, s3);
-
-  //
-  if (is_aligned(p1, sizeof(dtype)) && is_aligned(p2, sizeof(dtype)) &&
-      is_aligned(p3, sizeof(dtype))) {
+  INIT_PTR_IDX(lp, 0, p1, s1, idx1);
+  INIT_PTR_IDX(lp, 1, p2, s2, idx2);
 
-    if (s1 == sizeof(dtype) && s2 == sizeof(dtype) && s3 == sizeof(dtype)) {
-      if (p1 == p3) { // inplace case
-        for (; i < n; i++) {
-          check_intdivzero(((dtype*)p2)[i]);
-          ((dtype*)p1)[i] = m_mul(((dtype*)p1)[i], ((dtype*)p2)[i]);
-        }
-      } else {
-        for (; i < n; i++) {
-          check_intdivzero(((dtype*)p2)[i]);
-          ((dtype*)p3)[i] = m_mul(((dtype*)p1)[i], ((dtype*)p2)[i]);
-        }
+  if (idx1) {
+    if (idx2) {
+      for (i = 0; i < n; i++) {
+        GET_DATA_INDEX(p1, idx1, dtype, x);
+        x = m_reciprocal(x);
+        SET_DATA_INDEX(p2, idx2, dtype, x);
       }
-      return;
-    }
-
-    if (is_aligned_step(s1, sizeof(dtype)) && is_aligned_step(s2, sizeof(dtype)) &&
-        is_aligned_step(s3, sizeof(dtype))) {
-      //
-
-      if (s2 == 0) { // Broadcasting from scalar value.
-        check_intdivzero(*(dtype*)p2);
-        if (s1 == sizeof(dtype) && s3 == sizeof(dtype)) {
-          if (p1 == p3) { // inplace case
-            for (; i < n; i++) {
-              ((dtype*)p1)[i] = m_mul(((dtype*)p1)[i], *(dtype*)p2);
-            }
-          } else {
-            for (; i < n; i++) {
-              ((dtype*)p3)[i] = m_mul(((dtype*)p1)[i], *(dtype*)p2);
-            }
-          }
-        } else {
-          for (i = 0; i < n; i++) {
-            *(dtype*)p3 = m_mul(*(dtype*)p1, *(dtype*)p2);
-            p1 += s1;
-            p3 += s3;
-          }
-        }
-      } else {
-        if (p1 == p3) { // inplace case
-          for (i = 0; i < n; i++) {
-            check_intdivzero(*(dtype*)p2);
-            *(dtype*)p1 = m_mul(*(dtype*)p1, *(dtype*)p2);
-            p1 += s1;
-            p2 += s2;
-          }
-        } else {
-          for (i = 0; i < n; i++) {
-            check_intdivzero(*(dtype*)p2);
-            *(dtype*)p3 = m_mul(*(dtype*)p1, *(dtype*)p2);
-            p1 += s1;
-            p2 += s2;
-            p3 += s3;
-          }
-        }
+    } else {
+      for (i = 0; i < n; i++) {
+        GET_DATA_INDEX(p1, idx1, dtype, x);
+        x = m_reciprocal(x);
+        SET_DATA_STRIDE(p2, s2, dtype, x);
       }
-
-      return;
-      //
     }
-  }
-  for (i = 0; i < n; i++) {
-    dtype x, y, z;
-    GET_DATA_STRIDE(p1, s1, dtype, x);
-    GET_DATA_STRIDE(p2, s2, dtype, y);
-    check_intdivzero(y);
-    z = m_mul(x, y);
-    SET_DATA_STRIDE(p3, s3, dtype, z);
-  }
-  //
-}
-#undef check_intdivzero
-
-static VALUE uint16_mul_self(VALUE self, VALUE other) {
-  ndfunc_arg_in_t ain[2] = { { cT, 0 }, { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_uint16_mul, STRIDE_LOOP, 2, 1, ain, aout };
-
-  return na_ndloop(&ndf, 2, self, other);
-}
-
-/*
-  Binary mul.
-  @overload * other
-    @param [Numo::NArray,Numeric] other
-    @return [Numo::NArray] self * other
-*/
-static VALUE uint16_mul(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return uint16_mul_self(self, other);
   } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, '*', 1, other);
-  }
-}
-
-#define check_intdivzero(y)                                                                    \
-  if ((y) == 0) {                                                                              \
-    lp->err_type = rb_eZeroDivError;                                                           \
-    return;                                                                                    \
-  }
-
-static void iter_uint16_div(na_loop_t* const lp) {
-  size_t i = 0;
-  size_t n;
-  char *p1, *p2, *p3;
-  ssize_t s1, s2, s3;
-
-  INIT_COUNTER(lp, n);
-  INIT_PTR(lp, 0, p1, s1);
-  INIT_PTR(lp, 1, p2, s2);
-  INIT_PTR(lp, 2, p3, s3);
-
-  //
-  if (is_aligned(p1, sizeof(dtype)) && is_aligned(p2, sizeof(dtype)) &&
-      is_aligned(p3, sizeof(dtype))) {
-
-    if (s1 == sizeof(dtype) && s2 == sizeof(dtype) && s3 == sizeof(dtype)) {
-      if (p1 == p3) { // inplace case
-        for (; i < n; i++) {
-          check_intdivzero(((dtype*)p2)[i]);
-          ((dtype*)p1)[i] = m_div(((dtype*)p1)[i], ((dtype*)p2)[i]);
-        }
-      } else {
-        for (; i < n; i++) {
-          check_intdivzero(((dtype*)p2)[i]);
-          ((dtype*)p3)[i] = m_div(((dtype*)p1)[i], ((dtype*)p2)[i]);
-        }
+    if (idx2) {
+      for (i = 0; i < n; i++) {
+        GET_DATA_STRIDE(p1, s1, dtype, x);
+        x = m_reciprocal(x);
+        SET_DATA_INDEX(p2, idx2, dtype, x);
       }
-      return;
-    }
-
-    if (is_aligned_step(s1, sizeof(dtype)) && is_aligned_step(s2, sizeof(dtype)) &&
-        is_aligned_step(s3, sizeof(dtype))) {
+    } else {
       //
-
-      if (s2 == 0) { // Broadcasting from scalar value.
-        check_intdivzero(*(dtype*)p2);
-        if (s1 == sizeof(dtype) && s3 == sizeof(dtype)) {
-          if (p1 == p3) { // inplace case
-            for (; i < n; i++) {
-              ((dtype*)p1)[i] = m_div(((dtype*)p1)[i], *(dtype*)p2);
-            }
-          } else {
-            for (; i < n; i++) {
-              ((dtype*)p3)[i] = m_div(((dtype*)p1)[i], *(dtype*)p2);
-            }
-          }
-        } else {
+      if (is_aligned(p1, sizeof(dtype)) && is_aligned(p2, sizeof(dtype))) {
+        if (s1 == sizeof(dtype) && s2 == sizeof(dtype)) {
           for (i = 0; i < n; i++) {
-            *(dtype*)p3 = m_div(*(dtype*)p1, *(dtype*)p2);
-            p1 += s1;
-            p3 += s3;
+            ((dtype*)p2)[i] = m_reciprocal(((dtype*)p1)[i]);
           }
+          return;
         }
-      } else {
-        if (p1 == p3) { // inplace case
-          for (i = 0; i < n; i++) {
-            check_intdivzero(*(dtype*)p2);
-            *(dtype*)p1 = m_div(*(dtype*)p1, *(dtype*)p2);
-            p1 += s1;
-            p2 += s2;
-          }
-        } else {
+        if (is_aligned_step(s1, sizeof(dtype)) && is_aligned_step(s2, sizeof(dtype))) {
+          //
           for (i = 0; i < n; i++) {
-            check_intdivzero(*(dtype*)p2);
-            *(dtype*)p3 = m_div(*(dtype*)p1, *(dtype*)p2);
+            *(dtype*)p2 = m_reciprocal(*(dtype*)p1);
             p1 += s1;
             p2 += s2;
-            p3 += s3;
           }
+          return;
+          //
         }
       }
-
-      return;
-      //
-    }
-  }
-  for (i = 0; i < n; i++) {
-    dtype x, y, z;
-    GET_DATA_STRIDE(p1, s1, dtype, x);
-    GET_DATA_STRIDE(p2, s2, dtype, y);
-    check_intdivzero(y);
-    z = m_div(x, y);
-    SET_DATA_STRIDE(p3, s3, dtype, z);
-  }
-  //
-}
-#undef check_intdivzero
-
-static VALUE uint16_div_self(VALUE self, VALUE other) {
-  ndfunc_arg_in_t ain[2] = { { cT, 0 }, { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_uint16_div, STRIDE_LOOP, 2, 1, ain, aout };
-
-  return na_ndloop(&ndf, 2, self, other);
-}
-
-/*
-  Binary div.
-  @overload / other
-    @param [Numo::NArray,Numeric] other
-    @return [Numo::NArray] self / other
-*/
-static VALUE uint16_div(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return uint16_div_self(self, other);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, '/', 1, other);
-  }
-}
-
-#define check_intdivzero(y)                                                                    \
-  if ((y) == 0) {                                                                              \
-    lp->err_type = rb_eZeroDivError;                                                           \
-    return;                                                                                    \
-  }
-
-static void iter_uint16_mod(na_loop_t* const lp) {
-  size_t i = 0;
-  size_t n;
-  char *p1, *p2, *p3;
-  ssize_t s1, s2, s3;
-
-  INIT_COUNTER(lp, n);
-  INIT_PTR(lp, 0, p1, s1);
-  INIT_PTR(lp, 1, p2, s2);
-  INIT_PTR(lp, 2, p3, s3);
-
-  //
-  if (is_aligned(p1, sizeof(dtype)) && is_aligned(p2, sizeof(dtype)) &&
-      is_aligned(p3, sizeof(dtype))) {
-
-    if (s1 == sizeof(dtype) && s2 == sizeof(dtype) && s3 == sizeof(dtype)) {
-      if (p1 == p3) { // inplace case
-        for (; i < n; i++) {
-          check_intdivzero(((dtype*)p2)[i]);
-          ((dtype*)p1)[i] = m_mod(((dtype*)p1)[i], ((dtype*)p2)[i]);
-        }
-      } else {
-        for (; i < n; i++) {
-          check_intdivzero(((dtype*)p2)[i]);
-          ((dtype*)p3)[i] = m_mod(((dtype*)p1)[i], ((dtype*)p2)[i]);
-        }
-      }
-      return;
-    }
-
-    if (is_aligned_step(s1, sizeof(dtype)) && is_aligned_step(s2, sizeof(dtype)) &&
-        is_aligned_step(s3, sizeof(dtype))) {
-      //
-
-      if (s2 == 0) { // Broadcasting from scalar value.
-        check_intdivzero(*(dtype*)p2);
-        if (s1 == sizeof(dtype) && s3 == sizeof(dtype)) {
-          if (p1 == p3) { // inplace case
-            for (; i < n; i++) {
-              ((dtype*)p1)[i] = m_mod(((dtype*)p1)[i], *(dtype*)p2);
-            }
-          } else {
-            for (; i < n; i++) {
-              ((dtype*)p3)[i] = m_mod(((dtype*)p1)[i], *(dtype*)p2);
-            }
-          }
-        } else {
-          for (i = 0; i < n; i++) {
-            *(dtype*)p3 = m_mod(*(dtype*)p1, *(dtype*)p2);
-            p1 += s1;
-            p3 += s3;
-          }
-        }
-      } else {
-        if (p1 == p3) { // inplace case
-          for (i = 0; i < n; i++) {
-            check_intdivzero(*(dtype*)p2);
-            *(dtype*)p1 = m_mod(*(dtype*)p1, *(dtype*)p2);
-            p1 += s1;
-            p2 += s2;
-          }
-        } else {
-          for (i = 0; i < n; i++) {
-            check_intdivzero(*(dtype*)p2);
-            *(dtype*)p3 = m_mod(*(dtype*)p1, *(dtype*)p2);
-            p1 += s1;
-            p2 += s2;
-            p3 += s3;
-          }
-        }
+      for (i = 0; i < n; i++) {
+        GET_DATA_STRIDE(p1, s1, dtype, x);
+        x = m_reciprocal(x);
+        SET_DATA_STRIDE(p2, s2, dtype, x);
       }
-
-      return;
       //
     }
   }
-  for (i = 0; i < n; i++) {
-    dtype x, y, z;
-    GET_DATA_STRIDE(p1, s1, dtype, x);
-    GET_DATA_STRIDE(p2, s2, dtype, y);
-    check_intdivzero(y);
-    z = m_mod(x, y);
-    SET_DATA_STRIDE(p3, s3, dtype, z);
-  }
-  //
-}
-#undef check_intdivzero
-
-static VALUE uint16_mod_self(VALUE self, VALUE other) {
-  ndfunc_arg_in_t ain[2] = { { cT, 0 }, { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_uint16_mod, STRIDE_LOOP, 2, 1, ain, aout };
-
-  return na_ndloop(&ndf, 2, self, other);
-}
-
-/*
-  Binary mod.
-  @overload % other
-    @param [Numo::NArray,Numeric] other
-    @return [Numo::NArray] self % other
-*/
-static VALUE uint16_mod(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return uint16_mod_self(self, other);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, '%', 1, other);
-  }
-}
-
-static void iter_uint16_divmod(na_loop_t* const lp) {
-  size_t i, n;
-  char *p1, *p2, *p3, *p4;
-  ssize_t s1, s2, s3, s4;
-  dtype x, y, a, b;
-  INIT_COUNTER(lp, n);
-  INIT_PTR(lp, 0, p1, s1);
-  INIT_PTR(lp, 1, p2, s2);
-  INIT_PTR(lp, 2, p3, s3);
-  INIT_PTR(lp, 3, p4, s4);
-  for (i = n; i--;) {
-    GET_DATA_STRIDE(p1, s1, dtype, x);
-    GET_DATA_STRIDE(p2, s2, dtype, y);
-    if (y == 0) {
-      lp->err_type = rb_eZeroDivError;
-      return;
-    }
-    m_divmod(x, y, a, b);
-    SET_DATA_STRIDE(p3, s3, dtype, a);
-    SET_DATA_STRIDE(p4, s4, dtype, b);
-  }
-}
-
-static VALUE uint16_divmod_self(VALUE self, VALUE other) {
-  ndfunc_arg_in_t ain[2] = { { cT, 0 }, { cT, 0 } };
-  ndfunc_arg_out_t aout[2] = { { cT, 0 }, { cT, 0 } };
-  ndfunc_t ndf = { iter_uint16_divmod, STRIDE_LOOP, 2, 2, ain, aout };
-
-  return na_ndloop(&ndf, 2, self, other);
 }
 
 /*
-  Binary divmod.
-  @overload divmod other
-    @param [Numo::NArray,Numeric] other
-    @return [Numo::NArray] divmod of self and other.
-*/
-static VALUE uint16_divmod(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return uint16_divmod_self(self, other);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, id_divmod, 1, other);
-  }
-}
-
-static void iter_uint16_pow(na_loop_t* const lp) {
-  size_t i;
-  char *p1, *p2, *p3;
-  ssize_t s1, s2, s3;
-  dtype x, y;
-  INIT_COUNTER(lp, i);
-  INIT_PTR(lp, 0, p1, s1);
-  INIT_PTR(lp, 1, p2, s2);
-  INIT_PTR(lp, 2, p3, s3);
-  for (; i--;) {
-    GET_DATA_STRIDE(p1, s1, dtype, x);
-    GET_DATA_STRIDE(p2, s2, dtype, y);
-    x = m_pow(x, y);
-    SET_DATA_STRIDE(p3, s3, dtype, x);
-  }
-}
-
-static void iter_uint16_pow_int32(na_loop_t* const lp) {
-  size_t i;
-  char *p1, *p2, *p3;
-  ssize_t s1, s2, s3;
-  dtype x;
-  int32_t y;
-  INIT_COUNTER(lp, i);
-  INIT_PTR(lp, 0, p1, s1);
-  INIT_PTR(lp, 1, p2, s2);
-  INIT_PTR(lp, 2, p3, s3);
-  for (; i--;) {
-    GET_DATA_STRIDE(p1, s1, dtype, x);
-    GET_DATA_STRIDE(p2, s2, int32_t, y);
-    x = m_pow_int(x, y);
-    SET_DATA_STRIDE(p3, s3, dtype, x);
-  }
-}
-
-static VALUE uint16_pow_self(VALUE self, VALUE other) {
-  ndfunc_arg_in_t ain[2] = { { cT, 0 }, { cT, 0 } };
-  ndfunc_arg_in_t ain_i[2] = { { cT, 0 }, { numo_cInt32, 0 } };
-  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_uint16_pow, STRIDE_LOOP, 2, 1, ain, aout };
-  ndfunc_t ndf_i = { iter_uint16_pow_int32, STRIDE_LOOP, 2, 1, ain_i, aout };
-
-  // fixme : use na.integer?
-  if (FIXNUM_P(other) || rb_obj_is_kind_of(other, numo_cInt32)) {
-    return na_ndloop(&ndf_i, 2, self, other);
-  } else {
-    return na_ndloop(&ndf, 2, self, other);
-  }
-}
-
-/*
-  Binary power.
-  @overload ** other
-    @param [Numo::NArray,Numeric] other
-    @return [Numo::NArray] self to the other-th power.
-*/
-static VALUE uint16_pow(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return uint16_pow_self(self, other);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, id_pow, 1, other);
-  }
-}
-
-static void iter_uint16_minus(na_loop_t* const lp) {
-  size_t i, n;
-  char *p1, *p2;
-  ssize_t s1, s2;
-  size_t *idx1, *idx2;
-  dtype x;
-
-  INIT_COUNTER(lp, n);
-  INIT_PTR_IDX(lp, 0, p1, s1, idx1);
-  INIT_PTR_IDX(lp, 1, p2, s2, idx2);
-
-  if (idx1) {
-    if (idx2) {
-      for (i = 0; i < n; i++) {
-        GET_DATA_INDEX(p1, idx1, dtype, x);
-        x = m_minus(x);
-        SET_DATA_INDEX(p2, idx2, dtype, x);
-      }
-    } else {
-      for (i = 0; i < n; i++) {
-        GET_DATA_INDEX(p1, idx1, dtype, x);
-        x = m_minus(x);
-        SET_DATA_STRIDE(p2, s2, dtype, x);
-      }
-    }
-  } else {
-    if (idx2) {
-      for (i = 0; i < n; i++) {
-        GET_DATA_STRIDE(p1, s1, dtype, x);
-        x = m_minus(x);
-        SET_DATA_INDEX(p2, idx2, dtype, x);
-      }
-    } else {
-      //
-      if (is_aligned(p1, sizeof(dtype)) && is_aligned(p2, sizeof(dtype))) {
-        if (s1 == sizeof(dtype) && s2 == sizeof(dtype)) {
-          for (i = 0; i < n; i++) {
-            ((dtype*)p2)[i] = m_minus(((dtype*)p1)[i]);
-          }
-          return;
-        }
-        if (is_aligned_step(s1, sizeof(dtype)) && is_aligned_step(s2, sizeof(dtype))) {
-          //
-          for (i = 0; i < n; i++) {
-            *(dtype*)p2 = m_minus(*(dtype*)p1);
-            p1 += s1;
-            p2 += s2;
-          }
-          return;
-          //
-        }
-      }
-      for (i = 0; i < n; i++) {
-        GET_DATA_STRIDE(p1, s1, dtype, x);
-        x = m_minus(x);
-        SET_DATA_STRIDE(p2, s2, dtype, x);
-      }
-      //
-    }
-  }
-}
-
-/*
-  Unary minus.
-  @overload -@
-    @return [Numo::UInt16] minus of self.
-*/
-static VALUE uint16_minus(VALUE self) {
-  ndfunc_arg_in_t ain[1] = { { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_uint16_minus, FULL_LOOP, 1, 1, ain, aout };
-
-  return na_ndloop(&ndf, 1, self);
-}
-
-static void iter_uint16_reciprocal(na_loop_t* const lp) {
-  size_t i, n;
-  char *p1, *p2;
-  ssize_t s1, s2;
-  size_t *idx1, *idx2;
-  dtype x;
-
-  INIT_COUNTER(lp, n);
-  INIT_PTR_IDX(lp, 0, p1, s1, idx1);
-  INIT_PTR_IDX(lp, 1, p2, s2, idx2);
-
-  if (idx1) {
-    if (idx2) {
-      for (i = 0; i < n; i++) {
-        GET_DATA_INDEX(p1, idx1, dtype, x);
-        x = m_reciprocal(x);
-        SET_DATA_INDEX(p2, idx2, dtype, x);
-      }
-    } else {
-      for (i = 0; i < n; i++) {
-        GET_DATA_INDEX(p1, idx1, dtype, x);
-        x = m_reciprocal(x);
-        SET_DATA_STRIDE(p2, s2, dtype, x);
-      }
-    }
-  } else {
-    if (idx2) {
-      for (i = 0; i < n; i++) {
-        GET_DATA_STRIDE(p1, s1, dtype, x);
-        x = m_reciprocal(x);
-        SET_DATA_INDEX(p2, idx2, dtype, x);
-      }
-    } else {
-      //
-      if (is_aligned(p1, sizeof(dtype)) && is_aligned(p2, sizeof(dtype))) {
-        if (s1 == sizeof(dtype) && s2 == sizeof(dtype)) {
-          for (i = 0; i < n; i++) {
-            ((dtype*)p2)[i] = m_reciprocal(((dtype*)p1)[i]);
-          }
-          return;
-        }
-        if (is_aligned_step(s1, sizeof(dtype)) && is_aligned_step(s2, sizeof(dtype))) {
-          //
-          for (i = 0; i < n; i++) {
-            *(dtype*)p2 = m_reciprocal(*(dtype*)p1);
-            p1 += s1;
-            p2 += s2;
-          }
-          return;
-          //
-        }
-      }
-      for (i = 0; i < n; i++) {
-        GET_DATA_STRIDE(p1, s1, dtype, x);
-        x = m_reciprocal(x);
-        SET_DATA_STRIDE(p2, s2, dtype, x);
-      }
-      //
-    }
-  }
-}
-
-/*
-  Unary reciprocal.
-  @overload reciprocal
-    @return [Numo::UInt16] reciprocal of self.
+  Unary reciprocal.
+  @overload reciprocal
+    @return [Numo::UInt16] reciprocal of self.
 */
 static VALUE uint16_reciprocal(VALUE self) {
   ndfunc_arg_in_t ain[1] = { { cT, 0 } };
   ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_uint16_reciprocal, FULL_LOOP, 1, 1, ain, aout };
-
-  return na_ndloop(&ndf, 1, self);
-}
-
-static void iter_uint16_sign(na_loop_t* const lp) {
-  size_t i, n;
-  char *p1, *p2;
-  ssize_t s1, s2;
-  size_t *idx1, *idx2;
-  dtype x;
-
-  INIT_COUNTER(lp, n);
-  INIT_PTR_IDX(lp, 0, p1, s1, idx1);
-  INIT_PTR_IDX(lp, 1, p2, s2, idx2);
-
-  if (idx1) {
-    if (idx2) {
-      for (i = 0; i < n; i++) {
-        GET_DATA_INDEX(p1, idx1, dtype, x);
-        x = m_sign(x);
-        SET_DATA_INDEX(p2, idx2, dtype, x);
-      }
-    } else {
-      for (i = 0; i < n; i++) {
-        GET_DATA_INDEX(p1, idx1, dtype, x);
-        x = m_sign(x);
-        SET_DATA_STRIDE(p2, s2, dtype, x);
-      }
-    }
-  } else {
-    if (idx2) {
-      for (i = 0; i < n; i++) {
-        GET_DATA_STRIDE(p1, s1, dtype, x);
-        x = m_sign(x);
-        SET_DATA_INDEX(p2, idx2, dtype, x);
-      }
-    } else {
-      //
-      if (is_aligned(p1, sizeof(dtype)) && is_aligned(p2, sizeof(dtype))) {
-        if (s1 == sizeof(dtype) && s2 == sizeof(dtype)) {
-          for (i = 0; i < n; i++) {
-            ((dtype*)p2)[i] = m_sign(((dtype*)p1)[i]);
-          }
-          return;
-        }
-        if (is_aligned_step(s1, sizeof(dtype)) && is_aligned_step(s2, sizeof(dtype))) {
-          //
-          for (i = 0; i < n; i++) {
-            *(dtype*)p2 = m_sign(*(dtype*)p1);
-            p1 += s1;
-            p2 += s2;
-          }
-          return;
-          //
-        }
-      }
-      for (i = 0; i < n; i++) {
-        GET_DATA_STRIDE(p1, s1, dtype, x);
-        x = m_sign(x);
-        SET_DATA_STRIDE(p2, s2, dtype, x);
-      }
-      //
-    }
-  }
-}
-
-/*
-  Unary sign.
-  @overload sign
-    @return [Numo::UInt16] sign of self.
-*/
-static VALUE uint16_sign(VALUE self) {
-  ndfunc_arg_in_t ain[1] = { { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_uint16_sign, FULL_LOOP, 1, 1, ain, aout };
-
-  return na_ndloop(&ndf, 1, self);
-}
-
-static void iter_uint16_square(na_loop_t* const lp) {
-  size_t i, n;
-  char *p1, *p2;
-  ssize_t s1, s2;
-  size_t *idx1, *idx2;
-  dtype x;
-
-  INIT_COUNTER(lp, n);
-  INIT_PTR_IDX(lp, 0, p1, s1, idx1);
-  INIT_PTR_IDX(lp, 1, p2, s2, idx2);
-
-  if (idx1) {
-    if (idx2) {
-      for (i = 0; i < n; i++) {
-        GET_DATA_INDEX(p1, idx1, dtype, x);
-        x = m_square(x);
-        SET_DATA_INDEX(p2, idx2, dtype, x);
-      }
-    } else {
-      for (i = 0; i < n; i++) {
-        GET_DATA_INDEX(p1, idx1, dtype, x);
-        x = m_square(x);
-        SET_DATA_STRIDE(p2, s2, dtype, x);
-      }
-    }
-  } else {
-    if (idx2) {
-      for (i = 0; i < n; i++) {
-        GET_DATA_STRIDE(p1, s1, dtype, x);
-        x = m_square(x);
-        SET_DATA_INDEX(p2, idx2, dtype, x);
-      }
-    } else {
-      //
-      if (is_aligned(p1, sizeof(dtype)) && is_aligned(p2, sizeof(dtype))) {
-        if (s1 == sizeof(dtype) && s2 == sizeof(dtype)) {
-          for (i = 0; i < n; i++) {
-            ((dtype*)p2)[i] = m_square(((dtype*)p1)[i]);
-          }
-          return;
-        }
-        if (is_aligned_step(s1, sizeof(dtype)) && is_aligned_step(s2, sizeof(dtype))) {
-          //
-          for (i = 0; i < n; i++) {
-            *(dtype*)p2 = m_square(*(dtype*)p1);
-            p1 += s1;
-            p2 += s2;
-          }
-          return;
-          //
-        }
-      }
-      for (i = 0; i < n; i++) {
-        GET_DATA_STRIDE(p1, s1, dtype, x);
-        x = m_square(x);
-        SET_DATA_STRIDE(p2, s2, dtype, x);
-      }
-      //
-    }
-  }
-}
-
-/*
-  Unary square.
-  @overload square
-    @return [Numo::UInt16] square of self.
-*/
-static VALUE uint16_square(VALUE self) {
-  ndfunc_arg_in_t ain[1] = { { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_uint16_square, FULL_LOOP, 1, 1, ain, aout };
-
-  return na_ndloop(&ndf, 1, self);
-}
-
-static void iter_uint16_eq(na_loop_t* const lp) {
-  size_t i;
-  char *p1, *p2;
-  BIT_DIGIT* a3;
-  size_t p3;
-  ssize_t s1, s2, s3;
-  dtype x, y;
-  BIT_DIGIT b;
-  INIT_COUNTER(lp, i);
-  INIT_PTR(lp, 0, p1, s1);
-  INIT_PTR(lp, 1, p2, s2);
-  INIT_PTR_BIT(lp, 2, a3, p3, s3);
-  for (; i--;) {
-    GET_DATA_STRIDE(p1, s1, dtype, x);
-    GET_DATA_STRIDE(p2, s2, dtype, y);
-    b = (m_eq(x, y)) ? 1 : 0;
-    STORE_BIT(a3, p3, b);
-    p3 += s3;
-  }
-}
-
-static VALUE uint16_eq_self(VALUE self, VALUE other) {
-  ndfunc_arg_in_t ain[2] = { { cT, 0 }, { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { numo_cBit, 0 } };
-  ndfunc_t ndf = { iter_uint16_eq, STRIDE_LOOP, 2, 1, ain, aout };
-
-  return na_ndloop(&ndf, 2, self, other);
-}
-
-/*
-  Comparison eq other.
-  @overload eq other
-    @param [Numo::NArray,Numeric] other
-    @return [Numo::Bit] result of self eq other.
-*/
-static VALUE uint16_eq(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return uint16_eq_self(self, other);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, id_eq, 1, other);
-  }
-}
-
-static void iter_uint16_ne(na_loop_t* const lp) {
-  size_t i;
-  char *p1, *p2;
-  BIT_DIGIT* a3;
-  size_t p3;
-  ssize_t s1, s2, s3;
-  dtype x, y;
-  BIT_DIGIT b;
-  INIT_COUNTER(lp, i);
-  INIT_PTR(lp, 0, p1, s1);
-  INIT_PTR(lp, 1, p2, s2);
-  INIT_PTR_BIT(lp, 2, a3, p3, s3);
-  for (; i--;) {
-    GET_DATA_STRIDE(p1, s1, dtype, x);
-    GET_DATA_STRIDE(p2, s2, dtype, y);
-    b = (m_ne(x, y)) ? 1 : 0;
-    STORE_BIT(a3, p3, b);
-    p3 += s3;
-  }
-}
-
-static VALUE uint16_ne_self(VALUE self, VALUE other) {
-  ndfunc_arg_in_t ain[2] = { { cT, 0 }, { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { numo_cBit, 0 } };
-  ndfunc_t ndf = { iter_uint16_ne, STRIDE_LOOP, 2, 1, ain, aout };
-
-  return na_ndloop(&ndf, 2, self, other);
-}
-
-/*
-  Comparison ne other.
-  @overload ne other
-    @param [Numo::NArray,Numeric] other
-    @return [Numo::Bit] result of self ne other.
-*/
-static VALUE uint16_ne(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return uint16_ne_self(self, other);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, id_ne, 1, other);
-  }
-}
-
-#define check_intdivzero(y)                                                                    \
-  {}
-
-static void iter_uint16_bit_and(na_loop_t* const lp) {
-  size_t i = 0;
-  size_t n;
-  char *p1, *p2, *p3;
-  ssize_t s1, s2, s3;
-
-  INIT_COUNTER(lp, n);
-  INIT_PTR(lp, 0, p1, s1);
-  INIT_PTR(lp, 1, p2, s2);
-  INIT_PTR(lp, 2, p3, s3);
-
-  //
-  if (is_aligned(p1, sizeof(dtype)) && is_aligned(p2, sizeof(dtype)) &&
-      is_aligned(p3, sizeof(dtype))) {
-
-    if (s1 == sizeof(dtype) && s2 == sizeof(dtype) && s3 == sizeof(dtype)) {
-      if (p1 == p3) { // inplace case
-        for (; i < n; i++) {
-          check_intdivzero(((dtype*)p2)[i]);
-          ((dtype*)p1)[i] = m_bit_and(((dtype*)p1)[i], ((dtype*)p2)[i]);
-        }
-      } else {
-        for (; i < n; i++) {
-          check_intdivzero(((dtype*)p2)[i]);
-          ((dtype*)p3)[i] = m_bit_and(((dtype*)p1)[i], ((dtype*)p2)[i]);
-        }
-      }
-      return;
-    }
-
-    if (is_aligned_step(s1, sizeof(dtype)) && is_aligned_step(s2, sizeof(dtype)) &&
-        is_aligned_step(s3, sizeof(dtype))) {
-      //
-
-      if (s2 == 0) { // Broadcasting from scalar value.
-        check_intdivzero(*(dtype*)p2);
-        if (s1 == sizeof(dtype) && s3 == sizeof(dtype)) {
-          if (p1 == p3) { // inplace case
-            for (; i < n; i++) {
-              ((dtype*)p1)[i] = m_bit_and(((dtype*)p1)[i], *(dtype*)p2);
-            }
-          } else {
-            for (; i < n; i++) {
-              ((dtype*)p3)[i] = m_bit_and(((dtype*)p1)[i], *(dtype*)p2);
-            }
-          }
-        } else {
-          for (i = 0; i < n; i++) {
-            *(dtype*)p3 = m_bit_and(*(dtype*)p1, *(dtype*)p2);
-            p1 += s1;
-            p3 += s3;
-          }
-        }
-      } else {
-        if (p1 == p3) { // inplace case
-          for (i = 0; i < n; i++) {
-            check_intdivzero(*(dtype*)p2);
-            *(dtype*)p1 = m_bit_and(*(dtype*)p1, *(dtype*)p2);
-            p1 += s1;
-            p2 += s2;
-          }
-        } else {
-          for (i = 0; i < n; i++) {
-            check_intdivzero(*(dtype*)p2);
-            *(dtype*)p3 = m_bit_and(*(dtype*)p1, *(dtype*)p2);
-            p1 += s1;
-            p2 += s2;
-            p3 += s3;
-          }
-        }
-      }
-
-      return;
-      //
-    }
-  }
-  for (i = 0; i < n; i++) {
-    dtype x, y, z;
-    GET_DATA_STRIDE(p1, s1, dtype, x);
-    GET_DATA_STRIDE(p2, s2, dtype, y);
-    check_intdivzero(y);
-    z = m_bit_and(x, y);
-    SET_DATA_STRIDE(p3, s3, dtype, z);
-  }
-  //
-}
-#undef check_intdivzero
-
-static VALUE uint16_bit_and_self(VALUE self, VALUE other) {
-  ndfunc_arg_in_t ain[2] = { { cT, 0 }, { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_uint16_bit_and, STRIDE_LOOP, 2, 1, ain, aout };
-
-  return na_ndloop(&ndf, 2, self, other);
-}
-
-/*
-  Binary bit_and.
-  @overload & other
-    @param [Numo::NArray,Numeric] other
-    @return [Numo::NArray] self & other
-*/
-static VALUE uint16_bit_and(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return uint16_bit_and_self(self, other);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, '&', 1, other);
-  }
-}
-
-#define check_intdivzero(y)                                                                    \
-  {}
-
-static void iter_uint16_bit_or(na_loop_t* const lp) {
-  size_t i = 0;
-  size_t n;
-  char *p1, *p2, *p3;
-  ssize_t s1, s2, s3;
-
-  INIT_COUNTER(lp, n);
-  INIT_PTR(lp, 0, p1, s1);
-  INIT_PTR(lp, 1, p2, s2);
-  INIT_PTR(lp, 2, p3, s3);
-
-  //
-  if (is_aligned(p1, sizeof(dtype)) && is_aligned(p2, sizeof(dtype)) &&
-      is_aligned(p3, sizeof(dtype))) {
-
-    if (s1 == sizeof(dtype) && s2 == sizeof(dtype) && s3 == sizeof(dtype)) {
-      if (p1 == p3) { // inplace case
-        for (; i < n; i++) {
-          check_intdivzero(((dtype*)p2)[i]);
-          ((dtype*)p1)[i] = m_bit_or(((dtype*)p1)[i], ((dtype*)p2)[i]);
-        }
-      } else {
-        for (; i < n; i++) {
-          check_intdivzero(((dtype*)p2)[i]);
-          ((dtype*)p3)[i] = m_bit_or(((dtype*)p1)[i], ((dtype*)p2)[i]);
-        }
-      }
-      return;
-    }
-
-    if (is_aligned_step(s1, sizeof(dtype)) && is_aligned_step(s2, sizeof(dtype)) &&
-        is_aligned_step(s3, sizeof(dtype))) {
-      //
-
-      if (s2 == 0) { // Broadcasting from scalar value.
-        check_intdivzero(*(dtype*)p2);
-        if (s1 == sizeof(dtype) && s3 == sizeof(dtype)) {
-          if (p1 == p3) { // inplace case
-            for (; i < n; i++) {
-              ((dtype*)p1)[i] = m_bit_or(((dtype*)p1)[i], *(dtype*)p2);
-            }
-          } else {
-            for (; i < n; i++) {
-              ((dtype*)p3)[i] = m_bit_or(((dtype*)p1)[i], *(dtype*)p2);
-            }
-          }
-        } else {
-          for (i = 0; i < n; i++) {
-            *(dtype*)p3 = m_bit_or(*(dtype*)p1, *(dtype*)p2);
-            p1 += s1;
-            p3 += s3;
-          }
-        }
-      } else {
-        if (p1 == p3) { // inplace case
-          for (i = 0; i < n; i++) {
-            check_intdivzero(*(dtype*)p2);
-            *(dtype*)p1 = m_bit_or(*(dtype*)p1, *(dtype*)p2);
-            p1 += s1;
-            p2 += s2;
-          }
-        } else {
-          for (i = 0; i < n; i++) {
-            check_intdivzero(*(dtype*)p2);
-            *(dtype*)p3 = m_bit_or(*(dtype*)p1, *(dtype*)p2);
-            p1 += s1;
-            p2 += s2;
-            p3 += s3;
-          }
-        }
-      }
-
-      return;
-      //
-    }
-  }
-  for (i = 0; i < n; i++) {
-    dtype x, y, z;
-    GET_DATA_STRIDE(p1, s1, dtype, x);
-    GET_DATA_STRIDE(p2, s2, dtype, y);
-    check_intdivzero(y);
-    z = m_bit_or(x, y);
-    SET_DATA_STRIDE(p3, s3, dtype, z);
-  }
-  //
-}
-#undef check_intdivzero
-
-static VALUE uint16_bit_or_self(VALUE self, VALUE other) {
-  ndfunc_arg_in_t ain[2] = { { cT, 0 }, { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_uint16_bit_or, STRIDE_LOOP, 2, 1, ain, aout };
-
-  return na_ndloop(&ndf, 2, self, other);
-}
-
-/*
-  Binary bit_or.
-  @overload | other
-    @param [Numo::NArray,Numeric] other
-    @return [Numo::NArray] self | other
-*/
-static VALUE uint16_bit_or(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return uint16_bit_or_self(self, other);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, '|', 1, other);
-  }
-}
-
-#define check_intdivzero(y)                                                                    \
-  {}
-
-static void iter_uint16_bit_xor(na_loop_t* const lp) {
-  size_t i = 0;
-  size_t n;
-  char *p1, *p2, *p3;
-  ssize_t s1, s2, s3;
-
-  INIT_COUNTER(lp, n);
-  INIT_PTR(lp, 0, p1, s1);
-  INIT_PTR(lp, 1, p2, s2);
-  INIT_PTR(lp, 2, p3, s3);
-
-  //
-  if (is_aligned(p1, sizeof(dtype)) && is_aligned(p2, sizeof(dtype)) &&
-      is_aligned(p3, sizeof(dtype))) {
-
-    if (s1 == sizeof(dtype) && s2 == sizeof(dtype) && s3 == sizeof(dtype)) {
-      if (p1 == p3) { // inplace case
-        for (; i < n; i++) {
-          check_intdivzero(((dtype*)p2)[i]);
-          ((dtype*)p1)[i] = m_bit_xor(((dtype*)p1)[i], ((dtype*)p2)[i]);
-        }
-      } else {
-        for (; i < n; i++) {
-          check_intdivzero(((dtype*)p2)[i]);
-          ((dtype*)p3)[i] = m_bit_xor(((dtype*)p1)[i], ((dtype*)p2)[i]);
-        }
-      }
-      return;
-    }
-
-    if (is_aligned_step(s1, sizeof(dtype)) && is_aligned_step(s2, sizeof(dtype)) &&
-        is_aligned_step(s3, sizeof(dtype))) {
-      //
-
-      if (s2 == 0) { // Broadcasting from scalar value.
-        check_intdivzero(*(dtype*)p2);
-        if (s1 == sizeof(dtype) && s3 == sizeof(dtype)) {
-          if (p1 == p3) { // inplace case
-            for (; i < n; i++) {
-              ((dtype*)p1)[i] = m_bit_xor(((dtype*)p1)[i], *(dtype*)p2);
-            }
-          } else {
-            for (; i < n; i++) {
-              ((dtype*)p3)[i] = m_bit_xor(((dtype*)p1)[i], *(dtype*)p2);
-            }
-          }
-        } else {
-          for (i = 0; i < n; i++) {
-            *(dtype*)p3 = m_bit_xor(*(dtype*)p1, *(dtype*)p2);
-            p1 += s1;
-            p3 += s3;
-          }
-        }
-      } else {
-        if (p1 == p3) { // inplace case
-          for (i = 0; i < n; i++) {
-            check_intdivzero(*(dtype*)p2);
-            *(dtype*)p1 = m_bit_xor(*(dtype*)p1, *(dtype*)p2);
-            p1 += s1;
-            p2 += s2;
-          }
-        } else {
-          for (i = 0; i < n; i++) {
-            check_intdivzero(*(dtype*)p2);
-            *(dtype*)p3 = m_bit_xor(*(dtype*)p1, *(dtype*)p2);
-            p1 += s1;
-            p2 += s2;
-            p3 += s3;
-          }
-        }
-      }
-
-      return;
-      //
-    }
-  }
-  for (i = 0; i < n; i++) {
-    dtype x, y, z;
-    GET_DATA_STRIDE(p1, s1, dtype, x);
-    GET_DATA_STRIDE(p2, s2, dtype, y);
-    check_intdivzero(y);
-    z = m_bit_xor(x, y);
-    SET_DATA_STRIDE(p3, s3, dtype, z);
-  }
-  //
-}
-#undef check_intdivzero
-
-static VALUE uint16_bit_xor_self(VALUE self, VALUE other) {
-  ndfunc_arg_in_t ain[2] = { { cT, 0 }, { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_uint16_bit_xor, STRIDE_LOOP, 2, 1, ain, aout };
-
-  return na_ndloop(&ndf, 2, self, other);
-}
-
-/*
-  Binary bit_xor.
-  @overload ^ other
-    @param [Numo::NArray,Numeric] other
-    @return [Numo::NArray] self ^ other
-*/
-static VALUE uint16_bit_xor(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return uint16_bit_xor_self(self, other);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, '^', 1, other);
-  }
-}
-
-static void iter_uint16_bit_not(na_loop_t* const lp) {
-  size_t i, n;
-  char *p1, *p2;
-  ssize_t s1, s2;
-  size_t *idx1, *idx2;
-  dtype x;
-
-  INIT_COUNTER(lp, n);
-  INIT_PTR_IDX(lp, 0, p1, s1, idx1);
-  INIT_PTR_IDX(lp, 1, p2, s2, idx2);
-
-  if (idx1) {
-    if (idx2) {
-      for (i = 0; i < n; i++) {
-        GET_DATA_INDEX(p1, idx1, dtype, x);
-        x = m_bit_not(x);
-        SET_DATA_INDEX(p2, idx2, dtype, x);
-      }
-    } else {
-      for (i = 0; i < n; i++) {
-        GET_DATA_INDEX(p1, idx1, dtype, x);
-        x = m_bit_not(x);
-        SET_DATA_STRIDE(p2, s2, dtype, x);
-      }
-    }
-  } else {
-    if (idx2) {
-      for (i = 0; i < n; i++) {
-        GET_DATA_STRIDE(p1, s1, dtype, x);
-        x = m_bit_not(x);
-        SET_DATA_INDEX(p2, idx2, dtype, x);
-      }
-    } else {
-      //
-      if (is_aligned(p1, sizeof(dtype)) && is_aligned(p2, sizeof(dtype))) {
-        if (s1 == sizeof(dtype) && s2 == sizeof(dtype)) {
-          for (i = 0; i < n; i++) {
-            ((dtype*)p2)[i] = m_bit_not(((dtype*)p1)[i]);
-          }
-          return;
-        }
-        if (is_aligned_step(s1, sizeof(dtype)) && is_aligned_step(s2, sizeof(dtype))) {
-          //
-          for (i = 0; i < n; i++) {
-            *(dtype*)p2 = m_bit_not(*(dtype*)p1);
-            p1 += s1;
-            p2 += s2;
-          }
-          return;
-          //
-        }
-      }
-      for (i = 0; i < n; i++) {
-        GET_DATA_STRIDE(p1, s1, dtype, x);
-        x = m_bit_not(x);
-        SET_DATA_STRIDE(p2, s2, dtype, x);
-      }
-      //
-    }
-  }
-}
-
-/*
-  Unary bit_not.
-  @overload ~
-    @return [Numo::UInt16] bit_not of self.
-*/
-static VALUE uint16_bit_not(VALUE self) {
-  ndfunc_arg_in_t ain[1] = { { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_uint16_bit_not, FULL_LOOP, 1, 1, ain, aout };
-
-  return na_ndloop(&ndf, 1, self);
-}
-
-#define check_intdivzero(y)                                                                    \
-  {}
-
-static void iter_uint16_left_shift(na_loop_t* const lp) {
-  size_t i = 0;
-  size_t n;
-  char *p1, *p2, *p3;
-  ssize_t s1, s2, s3;
-
-  INIT_COUNTER(lp, n);
-  INIT_PTR(lp, 0, p1, s1);
-  INIT_PTR(lp, 1, p2, s2);
-  INIT_PTR(lp, 2, p3, s3);
-
-  //
-  if (is_aligned(p1, sizeof(dtype)) && is_aligned(p2, sizeof(dtype)) &&
-      is_aligned(p3, sizeof(dtype))) {
-
-    if (s1 == sizeof(dtype) && s2 == sizeof(dtype) && s3 == sizeof(dtype)) {
-      if (p1 == p3) { // inplace case
-        for (; i < n; i++) {
-          check_intdivzero(((dtype*)p2)[i]);
-          ((dtype*)p1)[i] = m_left_shift(((dtype*)p1)[i], ((dtype*)p2)[i]);
-        }
-      } else {
-        for (; i < n; i++) {
-          check_intdivzero(((dtype*)p2)[i]);
-          ((dtype*)p3)[i] = m_left_shift(((dtype*)p1)[i], ((dtype*)p2)[i]);
-        }
-      }
-      return;
-    }
-
-    if (is_aligned_step(s1, sizeof(dtype)) && is_aligned_step(s2, sizeof(dtype)) &&
-        is_aligned_step(s3, sizeof(dtype))) {
-      //
-
-      if (s2 == 0) { // Broadcasting from scalar value.
-        check_intdivzero(*(dtype*)p2);
-        if (s1 == sizeof(dtype) && s3 == sizeof(dtype)) {
-          if (p1 == p3) { // inplace case
-            for (; i < n; i++) {
-              ((dtype*)p1)[i] = m_left_shift(((dtype*)p1)[i], *(dtype*)p2);
-            }
-          } else {
-            for (; i < n; i++) {
-              ((dtype*)p3)[i] = m_left_shift(((dtype*)p1)[i], *(dtype*)p2);
-            }
-          }
-        } else {
-          for (i = 0; i < n; i++) {
-            *(dtype*)p3 = m_left_shift(*(dtype*)p1, *(dtype*)p2);
-            p1 += s1;
-            p3 += s3;
-          }
-        }
-      } else {
-        if (p1 == p3) { // inplace case
-          for (i = 0; i < n; i++) {
-            check_intdivzero(*(dtype*)p2);
-            *(dtype*)p1 = m_left_shift(*(dtype*)p1, *(dtype*)p2);
-            p1 += s1;
-            p2 += s2;
-          }
-        } else {
-          for (i = 0; i < n; i++) {
-            check_intdivzero(*(dtype*)p2);
-            *(dtype*)p3 = m_left_shift(*(dtype*)p1, *(dtype*)p2);
-            p1 += s1;
-            p2 += s2;
-            p3 += s3;
-          }
-        }
-      }
-
-      return;
-      //
-    }
-  }
-  for (i = 0; i < n; i++) {
-    dtype x, y, z;
-    GET_DATA_STRIDE(p1, s1, dtype, x);
-    GET_DATA_STRIDE(p2, s2, dtype, y);
-    check_intdivzero(y);
-    z = m_left_shift(x, y);
-    SET_DATA_STRIDE(p3, s3, dtype, z);
-  }
-  //
-}
-#undef check_intdivzero
-
-static VALUE uint16_left_shift_self(VALUE self, VALUE other) {
-  ndfunc_arg_in_t ain[2] = { { cT, 0 }, { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_uint16_left_shift, STRIDE_LOOP, 2, 1, ain, aout };
-
-  return na_ndloop(&ndf, 2, self, other);
-}
-
-/*
-  Binary left_shift.
-  @overload << other
-    @param [Numo::NArray,Numeric] other
-    @return [Numo::NArray] self << other
-*/
-static VALUE uint16_left_shift(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return uint16_left_shift_self(self, other);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, id_left_shift, 1, other);
-  }
-}
-
-#define check_intdivzero(y)                                                                    \
-  {}
-
-static void iter_uint16_right_shift(na_loop_t* const lp) {
-  size_t i = 0;
-  size_t n;
-  char *p1, *p2, *p3;
-  ssize_t s1, s2, s3;
-
-  INIT_COUNTER(lp, n);
-  INIT_PTR(lp, 0, p1, s1);
-  INIT_PTR(lp, 1, p2, s2);
-  INIT_PTR(lp, 2, p3, s3);
-
-  //
-  if (is_aligned(p1, sizeof(dtype)) && is_aligned(p2, sizeof(dtype)) &&
-      is_aligned(p3, sizeof(dtype))) {
-
-    if (s1 == sizeof(dtype) && s2 == sizeof(dtype) && s3 == sizeof(dtype)) {
-      if (p1 == p3) { // inplace case
-        for (; i < n; i++) {
-          check_intdivzero(((dtype*)p2)[i]);
-          ((dtype*)p1)[i] = m_right_shift(((dtype*)p1)[i], ((dtype*)p2)[i]);
-        }
-      } else {
-        for (; i < n; i++) {
-          check_intdivzero(((dtype*)p2)[i]);
-          ((dtype*)p3)[i] = m_right_shift(((dtype*)p1)[i], ((dtype*)p2)[i]);
-        }
-      }
-      return;
-    }
-
-    if (is_aligned_step(s1, sizeof(dtype)) && is_aligned_step(s2, sizeof(dtype)) &&
-        is_aligned_step(s3, sizeof(dtype))) {
-      //
-
-      if (s2 == 0) { // Broadcasting from scalar value.
-        check_intdivzero(*(dtype*)p2);
-        if (s1 == sizeof(dtype) && s3 == sizeof(dtype)) {
-          if (p1 == p3) { // inplace case
-            for (; i < n; i++) {
-              ((dtype*)p1)[i] = m_right_shift(((dtype*)p1)[i], *(dtype*)p2);
-            }
-          } else {
-            for (; i < n; i++) {
-              ((dtype*)p3)[i] = m_right_shift(((dtype*)p1)[i], *(dtype*)p2);
-            }
-          }
-        } else {
-          for (i = 0; i < n; i++) {
-            *(dtype*)p3 = m_right_shift(*(dtype*)p1, *(dtype*)p2);
-            p1 += s1;
-            p3 += s3;
-          }
-        }
-      } else {
-        if (p1 == p3) { // inplace case
-          for (i = 0; i < n; i++) {
-            check_intdivzero(*(dtype*)p2);
-            *(dtype*)p1 = m_right_shift(*(dtype*)p1, *(dtype*)p2);
-            p1 += s1;
-            p2 += s2;
-          }
-        } else {
-          for (i = 0; i < n; i++) {
-            check_intdivzero(*(dtype*)p2);
-            *(dtype*)p3 = m_right_shift(*(dtype*)p1, *(dtype*)p2);
-            p1 += s1;
-            p2 += s2;
-            p3 += s3;
-          }
-        }
-      }
-
-      return;
-      //
-    }
-  }
-  for (i = 0; i < n; i++) {
-    dtype x, y, z;
-    GET_DATA_STRIDE(p1, s1, dtype, x);
-    GET_DATA_STRIDE(p2, s2, dtype, y);
-    check_intdivzero(y);
-    z = m_right_shift(x, y);
-    SET_DATA_STRIDE(p3, s3, dtype, z);
-  }
-  //
-}
-#undef check_intdivzero
-
-static VALUE uint16_right_shift_self(VALUE self, VALUE other) {
-  ndfunc_arg_in_t ain[2] = { { cT, 0 }, { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_uint16_right_shift, STRIDE_LOOP, 2, 1, ain, aout };
-
-  return na_ndloop(&ndf, 2, self, other);
-}
-
-/*
-  Binary right_shift.
-  @overload >> other
-    @param [Numo::NArray,Numeric] other
-    @return [Numo::NArray] self >> other
-*/
-static VALUE uint16_right_shift(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return uint16_right_shift_self(self, other);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, id_right_shift, 1, other);
-  }
-}
-
-static void iter_uint16_gt(na_loop_t* const lp) {
-  size_t i;
-  char *p1, *p2;
-  BIT_DIGIT* a3;
-  size_t p3;
-  ssize_t s1, s2, s3;
-  dtype x, y;
-  BIT_DIGIT b;
-  INIT_COUNTER(lp, i);
-  INIT_PTR(lp, 0, p1, s1);
-  INIT_PTR(lp, 1, p2, s2);
-  INIT_PTR_BIT(lp, 2, a3, p3, s3);
-  for (; i--;) {
-    GET_DATA_STRIDE(p1, s1, dtype, x);
-    GET_DATA_STRIDE(p2, s2, dtype, y);
-    b = (m_gt(x, y)) ? 1 : 0;
-    STORE_BIT(a3, p3, b);
-    p3 += s3;
-  }
-}
-
-static VALUE uint16_gt_self(VALUE self, VALUE other) {
-  ndfunc_arg_in_t ain[2] = { { cT, 0 }, { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { numo_cBit, 0 } };
-  ndfunc_t ndf = { iter_uint16_gt, STRIDE_LOOP, 2, 1, ain, aout };
-
-  return na_ndloop(&ndf, 2, self, other);
-}
-
-/*
-  Comparison gt other.
-  @overload gt other
-    @param [Numo::NArray,Numeric] other
-    @return [Numo::Bit] result of self gt other.
-*/
-static VALUE uint16_gt(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return uint16_gt_self(self, other);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, id_gt, 1, other);
-  }
-}
-
-static void iter_uint16_ge(na_loop_t* const lp) {
-  size_t i;
-  char *p1, *p2;
-  BIT_DIGIT* a3;
-  size_t p3;
-  ssize_t s1, s2, s3;
-  dtype x, y;
-  BIT_DIGIT b;
-  INIT_COUNTER(lp, i);
-  INIT_PTR(lp, 0, p1, s1);
-  INIT_PTR(lp, 1, p2, s2);
-  INIT_PTR_BIT(lp, 2, a3, p3, s3);
-  for (; i--;) {
-    GET_DATA_STRIDE(p1, s1, dtype, x);
-    GET_DATA_STRIDE(p2, s2, dtype, y);
-    b = (m_ge(x, y)) ? 1 : 0;
-    STORE_BIT(a3, p3, b);
-    p3 += s3;
-  }
-}
-
-static VALUE uint16_ge_self(VALUE self, VALUE other) {
-  ndfunc_arg_in_t ain[2] = { { cT, 0 }, { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { numo_cBit, 0 } };
-  ndfunc_t ndf = { iter_uint16_ge, STRIDE_LOOP, 2, 1, ain, aout };
-
-  return na_ndloop(&ndf, 2, self, other);
-}
-
-/*
-  Comparison ge other.
-  @overload ge other
-    @param [Numo::NArray,Numeric] other
-    @return [Numo::Bit] result of self ge other.
-*/
-static VALUE uint16_ge(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return uint16_ge_self(self, other);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, id_ge, 1, other);
-  }
-}
-
-static void iter_uint16_lt(na_loop_t* const lp) {
-  size_t i;
-  char *p1, *p2;
-  BIT_DIGIT* a3;
-  size_t p3;
-  ssize_t s1, s2, s3;
-  dtype x, y;
-  BIT_DIGIT b;
-  INIT_COUNTER(lp, i);
-  INIT_PTR(lp, 0, p1, s1);
-  INIT_PTR(lp, 1, p2, s2);
-  INIT_PTR_BIT(lp, 2, a3, p3, s3);
-  for (; i--;) {
-    GET_DATA_STRIDE(p1, s1, dtype, x);
-    GET_DATA_STRIDE(p2, s2, dtype, y);
-    b = (m_lt(x, y)) ? 1 : 0;
-    STORE_BIT(a3, p3, b);
-    p3 += s3;
-  }
-}
-
-static VALUE uint16_lt_self(VALUE self, VALUE other) {
-  ndfunc_arg_in_t ain[2] = { { cT, 0 }, { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { numo_cBit, 0 } };
-  ndfunc_t ndf = { iter_uint16_lt, STRIDE_LOOP, 2, 1, ain, aout };
-
-  return na_ndloop(&ndf, 2, self, other);
-}
-
-/*
-  Comparison lt other.
-  @overload lt other
-    @param [Numo::NArray,Numeric] other
-    @return [Numo::Bit] result of self lt other.
-*/
-static VALUE uint16_lt(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return uint16_lt_self(self, other);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, id_lt, 1, other);
-  }
-}
-
-static void iter_uint16_le(na_loop_t* const lp) {
-  size_t i;
-  char *p1, *p2;
-  BIT_DIGIT* a3;
-  size_t p3;
-  ssize_t s1, s2, s3;
-  dtype x, y;
-  BIT_DIGIT b;
-  INIT_COUNTER(lp, i);
-  INIT_PTR(lp, 0, p1, s1);
-  INIT_PTR(lp, 1, p2, s2);
-  INIT_PTR_BIT(lp, 2, a3, p3, s3);
-  for (; i--;) {
-    GET_DATA_STRIDE(p1, s1, dtype, x);
-    GET_DATA_STRIDE(p2, s2, dtype, y);
-    b = (m_le(x, y)) ? 1 : 0;
-    STORE_BIT(a3, p3, b);
-    p3 += s3;
-  }
-}
-
-static VALUE uint16_le_self(VALUE self, VALUE other) {
-  ndfunc_arg_in_t ain[2] = { { cT, 0 }, { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { numo_cBit, 0 } };
-  ndfunc_t ndf = { iter_uint16_le, STRIDE_LOOP, 2, 1, ain, aout };
-
-  return na_ndloop(&ndf, 2, self, other);
-}
-
-/*
-  Comparison le other.
-  @overload le other
-    @param [Numo::NArray,Numeric] other
-    @return [Numo::Bit] result of self le other.
-*/
-static VALUE uint16_le(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return uint16_le_self(self, other);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, id_le, 1, other);
-  }
-}
-
-static void iter_uint16_clip(na_loop_t* const lp) {
-  size_t i;
-  char *p1, *p2, *p3, *p4;
-  ssize_t s1, s2, s3, s4;
-  dtype x, min, max;
-  INIT_COUNTER(lp, i);
-  INIT_PTR(lp, 0, p1, s1);
-  INIT_PTR(lp, 1, p2, s2);
-  INIT_PTR(lp, 2, p3, s3);
-  INIT_PTR(lp, 3, p4, s4);
-  for (; i--;) {
-    GET_DATA_STRIDE(p1, s1, dtype, x);
-    GET_DATA_STRIDE(p2, s2, dtype, min);
-    GET_DATA_STRIDE(p3, s3, dtype, max);
-    if (m_gt(min, max)) {
-      rb_raise(nary_eOperationError, "min is greater than max");
-    }
-    if (m_lt(x, min)) {
-      x = min;
-    }
-    if (m_gt(x, max)) {
-      x = max;
-    }
-    SET_DATA_STRIDE(p4, s4, dtype, x);
-  }
-}
-
-static void iter_uint16_clip_min(na_loop_t* const lp) {
-  size_t i;
-  char *p1, *p2, *p3;
-  ssize_t s1, s2, s3;
-  dtype x, min;
-  INIT_COUNTER(lp, i);
-  INIT_PTR(lp, 0, p1, s1);
-  INIT_PTR(lp, 1, p2, s2);
-  INIT_PTR(lp, 2, p3, s3);
-  for (; i--;) {
-    GET_DATA_STRIDE(p1, s1, dtype, x);
-    GET_DATA_STRIDE(p2, s2, dtype, min);
-    if (m_lt(x, min)) {
-      x = min;
-    }
-    SET_DATA_STRIDE(p3, s3, dtype, x);
-  }
-}
-
-static void iter_uint16_clip_max(na_loop_t* const lp) {
-  size_t i;
-  char *p1, *p2, *p3;
-  ssize_t s1, s2, s3;
-  dtype x, max;
-  INIT_COUNTER(lp, i);
-  INIT_PTR(lp, 0, p1, s1);
-  INIT_PTR(lp, 1, p2, s2);
-  INIT_PTR(lp, 2, p3, s3);
-  for (; i--;) {
-    GET_DATA_STRIDE(p1, s1, dtype, x);
-    GET_DATA_STRIDE(p2, s2, dtype, max);
-    if (m_gt(x, max)) {
-      x = max;
-    }
-    SET_DATA_STRIDE(p3, s3, dtype, x);
-  }
-}
-
-/*
-  Clip array elements by [min,max].
-  If either of min or max is nil, one side is clipped.
-  @overload clip(min,max)
-    @param [Numo::NArray,Numeric] min
-    @param [Numo::NArray,Numeric] max
-    @return [Numo::NArray] result of clip.
-
-  @example
-      a = Numo::Int32.new(10).seq
-      # => Numo::Int32#shape=[10]
-      # [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]
-
-      a.clip(1,8)
-      # => Numo::Int32#shape=[10]
-      # [1, 1, 2, 3, 4, 5, 6, 7, 8, 8]
-
-      a.inplace.clip(3,6)
-      a
-      # => Numo::Int32#shape=[10]
-      # [3, 3, 3, 3, 4, 5, 6, 6, 6, 6]
-
-      b = Numo::Int32.new(10).seq
-      # => Numo::Int32#shape=[10]
-      # [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]
-
-      b.clip([3,4,1,1,1,4,4,4,4,4], 8)
-      # => Numo::Int32#shape=[10]
-      # [3, 4, 2, 3, 4, 5, 6, 7, 8, 8]
-*/
-static VALUE uint16_clip(VALUE self, VALUE min, VALUE max) {
-  ndfunc_arg_in_t ain[3] = { { Qnil, 0 }, { cT, 0 }, { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf_min = { iter_uint16_clip_min, STRIDE_LOOP, 2, 1, ain, aout };
-  ndfunc_t ndf_max = { iter_uint16_clip_max, STRIDE_LOOP, 2, 1, ain, aout };
-  ndfunc_t ndf_both = { iter_uint16_clip, STRIDE_LOOP, 3, 1, ain, aout };
-
-  if (RTEST(min)) {
-    if (RTEST(max)) {
-      return na_ndloop(&ndf_both, 3, self, min, max);
-    } else {
-      return na_ndloop(&ndf_min, 2, self, min);
-    }
-  } else {
-    if (RTEST(max)) {
-      return na_ndloop(&ndf_max, 2, self, max);
-    }
-  }
-  rb_raise(rb_eArgError, "min and max are not given");
-  return Qnil;
-}
-
-static void iter_uint16_sum(na_loop_t* const lp) {
-  size_t n;
-  char *p1, *p2;
-  ssize_t s1;
-
-  INIT_COUNTER(lp, n);
-  INIT_PTR(lp, 0, p1, s1);
-  p2 = lp->args[1].ptr + lp->args[1].iter[0].pos;
-
-  *(u_int64_t*)p2 = f_sum(n, p1, s1);
-}
-
-/*
-  sum of self.
-  @overload sum(axis:nil, keepdims:false)
-    @param [Numeric,Array,Range] axis  Performs sum along the axis.
-    @param [TrueClass] keepdims  If true, the reduced axes are left in the result array as
-    dimensions with size one.
-    @return [Numo::UInt16] returns result of sum.
-*/
-static VALUE uint16_sum(int argc, VALUE* argv, VALUE self) {
-  VALUE v, reduce;
-  ndfunc_arg_in_t ain[2] = { { cT, 0 }, { sym_reduce, 0 } };
-  ndfunc_arg_out_t aout[1] = { { numo_cUInt64, 0 } };
-  ndfunc_t ndf = { iter_uint16_sum, STRIDE_LOOP_NIP | NDF_FLAT_REDUCE, 2, 1, ain, aout };
-
-  reduce = na_reduce_dimension(argc, argv, 1, &self, &ndf, 0);
-
-  v = na_ndloop(&ndf, 2, self, reduce);
-
-  return rb_funcall(v, rb_intern("extract"), 0);
-}
-
-static void iter_uint16_prod(na_loop_t* const lp) {
-  size_t n;
-  char *p1, *p2;
-  ssize_t s1;
-
-  INIT_COUNTER(lp, n);
-  INIT_PTR(lp, 0, p1, s1);
-  p2 = lp->args[1].ptr + lp->args[1].iter[0].pos;
-
-  *(u_int64_t*)p2 = f_prod(n, p1, s1);
-}
-
-/*
-  prod of self.
-  @overload prod(axis:nil, keepdims:false)
-    @param [Numeric,Array,Range] axis  Performs prod along the axis.
-    @param [TrueClass] keepdims  If true, the reduced axes are left in the result array as
-    dimensions with size one.
-    @return [Numo::UInt16] returns result of prod.
-*/
-static VALUE uint16_prod(int argc, VALUE* argv, VALUE self) {
-  VALUE v, reduce;
-  ndfunc_arg_in_t ain[2] = { { cT, 0 }, { sym_reduce, 0 } };
-  ndfunc_arg_out_t aout[1] = { { numo_cUInt64, 0 } };
-  ndfunc_t ndf = { iter_uint16_prod, STRIDE_LOOP_NIP | NDF_FLAT_REDUCE, 2, 1, ain, aout };
-
-  reduce = na_reduce_dimension(argc, argv, 1, &self, &ndf, 0);
-
-  v = na_ndloop(&ndf, 2, self, reduce);
-
-  return rb_funcall(v, rb_intern("extract"), 0);
-}
-
-static void iter_uint16_min(na_loop_t* const lp) {
-  size_t n;
-  char *p1, *p2;
-  ssize_t s1;
-
-  INIT_COUNTER(lp, n);
-  INIT_PTR(lp, 0, p1, s1);
-  p2 = lp->args[1].ptr + lp->args[1].iter[0].pos;
-
-  *(dtype*)p2 = f_min(n, p1, s1);
-}
-
-/*
-  min of self.
-  @overload min(axis:nil, keepdims:false)
-    @param [Numeric,Array,Range] axis  Performs min along the axis.
-    @param [TrueClass] keepdims  If true, the reduced axes are left in the result array as
-    dimensions with size one.
-    @return [Numo::UInt16] returns result of min.
-*/
-static VALUE uint16_min(int argc, VALUE* argv, VALUE self) {
-  VALUE v, reduce;
-  ndfunc_arg_in_t ain[2] = { { cT, 0 }, { sym_reduce, 0 } };
-  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_uint16_min, STRIDE_LOOP_NIP | NDF_FLAT_REDUCE, 2, 1, ain, aout };
-
-  reduce = na_reduce_dimension(argc, argv, 1, &self, &ndf, 0);
-
-  v = na_ndloop(&ndf, 2, self, reduce);
-
-  return uint16_extract(v);
-}
-
-static void iter_uint16_max(na_loop_t* const lp) {
-  size_t n;
-  char *p1, *p2;
-  ssize_t s1;
-
-  INIT_COUNTER(lp, n);
-  INIT_PTR(lp, 0, p1, s1);
-  p2 = lp->args[1].ptr + lp->args[1].iter[0].pos;
-
-  *(dtype*)p2 = f_max(n, p1, s1);
-}
-
-/*
-  max of self.
-  @overload max(axis:nil, keepdims:false)
-    @param [Numeric,Array,Range] axis  Performs max along the axis.
-    @param [TrueClass] keepdims  If true, the reduced axes are left in the result array as
-    dimensions with size one.
-    @return [Numo::UInt16] returns result of max.
-*/
-static VALUE uint16_max(int argc, VALUE* argv, VALUE self) {
-  VALUE v, reduce;
-  ndfunc_arg_in_t ain[2] = { { cT, 0 }, { sym_reduce, 0 } };
-  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_uint16_max, STRIDE_LOOP_NIP | NDF_FLAT_REDUCE, 2, 1, ain, aout };
-
-  reduce = na_reduce_dimension(argc, argv, 1, &self, &ndf, 0);
-
-  v = na_ndloop(&ndf, 2, self, reduce);
-
-  return uint16_extract(v);
-}
-
-static void iter_uint16_ptp(na_loop_t* const lp) {
-  size_t n;
-  char *p1, *p2;
-  ssize_t s1;
-
-  INIT_COUNTER(lp, n);
-  INIT_PTR(lp, 0, p1, s1);
-  p2 = lp->args[1].ptr + lp->args[1].iter[0].pos;
-
-  *(dtype*)p2 = f_ptp(n, p1, s1);
-}
-
-/*
-  ptp of self.
-  @overload ptp(axis:nil, keepdims:false)
-    @param [Numeric,Array,Range] axis  Performs ptp along the axis.
-    @param [TrueClass] keepdims  If true, the reduced axes are left in the result array as
-    dimensions with size one.
-    @return [Numo::UInt16] returns result of ptp.
-*/
-static VALUE uint16_ptp(int argc, VALUE* argv, VALUE self) {
-  VALUE v, reduce;
-  ndfunc_arg_in_t ain[2] = { { cT, 0 }, { sym_reduce, 0 } };
-  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_uint16_ptp, STRIDE_LOOP_NIP | NDF_FLAT_REDUCE, 2, 1, ain, aout };
-
-  reduce = na_reduce_dimension(argc, argv, 1, &self, &ndf, 0);
-
-  v = na_ndloop(&ndf, 2, self, reduce);
-
-  return uint16_extract(v);
-}
-
-#define idx_t int64_t
-static void iter_uint16_max_index_index64(na_loop_t* const lp) {
-  size_t n, idx;
-  char *d_ptr, *i_ptr, *o_ptr;
-  ssize_t d_step, i_step;
-
-  INIT_COUNTER(lp, n);
-  INIT_PTR(lp, 0, d_ptr, d_step);
-
-  idx = f_max_index(n, d_ptr, d_step);
-
-  INIT_PTR(lp, 1, i_ptr, i_step);
-  o_ptr = NDL_PTR(lp, 2);
-  *(idx_t*)o_ptr = *(idx_t*)(i_ptr + i_step * idx);
-}
-#undef idx_t
-
-#define idx_t int32_t
-static void iter_uint16_max_index_index32(na_loop_t* const lp) {
-  size_t n, idx;
-  char *d_ptr, *i_ptr, *o_ptr;
-  ssize_t d_step, i_step;
-
-  INIT_COUNTER(lp, n);
-  INIT_PTR(lp, 0, d_ptr, d_step);
-
-  idx = f_max_index(n, d_ptr, d_step);
-
-  INIT_PTR(lp, 1, i_ptr, i_step);
-  o_ptr = NDL_PTR(lp, 2);
-  *(idx_t*)o_ptr = *(idx_t*)(i_ptr + i_step * idx);
-}
-#undef idx_t
-
-/*
-  Index of the maximum value.
-  @overload max_index(axis:nil)
-    @param [Numeric,Array,Range] axis  Finds maximum values along the axis and returns **flat
-    1-d indices**.
-    @return [Integer,Numo::Int] returns result indices.
-  @see #argmax
-  @see #max
-
-  @example
-      a = Numo::NArray[3,4,1,2]
-      a.max_index  #=> 1
-
-      b = Numo::NArray[[3,4,1],[2,0,5]]
-      b.max_index             #=> 5
-      b.max_index(axis:1)     #=> [1, 5]
-      b.max_index(axis:0)     #=> [0, 1, 5]
-      b[b.max_index(axis:0)]  #=> [3, 4, 5]
- */
-static VALUE uint16_max_index(int argc, VALUE* argv, VALUE self) {
-  narray_t* na;
-  VALUE idx, reduce;
-  ndfunc_arg_in_t ain[3] = { { Qnil, 0 }, { Qnil, 0 }, { sym_reduce, 0 } };
-  ndfunc_arg_out_t aout[1] = { { 0, 0, 0 } };
-  ndfunc_t ndf = { 0, STRIDE_LOOP_NIP | NDF_FLAT_REDUCE | NDF_EXTRACT, 3, 1, ain, aout };
-
-  GetNArray(self, na);
-  if (na->ndim == 0) {
-    return INT2FIX(0);
-  }
-  if (na->size > (~(u_int32_t)0)) {
-    aout[0].type = numo_cInt64;
-    idx = nary_new(numo_cInt64, na->ndim, na->shape);
-    ndf.func = iter_uint16_max_index_index64;
-
-    reduce = na_reduce_dimension(argc, argv, 1, &self, &ndf, 0);
-
-  } else {
-    aout[0].type = numo_cInt32;
-    idx = nary_new(numo_cInt32, na->ndim, na->shape);
-    ndf.func = iter_uint16_max_index_index32;
-
-    reduce = na_reduce_dimension(argc, argv, 1, &self, &ndf, 0);
-  }
-  rb_funcall(idx, rb_intern("seq"), 0);
-
-  return na_ndloop(&ndf, 3, self, idx, reduce);
-}
-
-#define idx_t int64_t
-static void iter_uint16_min_index_index64(na_loop_t* const lp) {
-  size_t n, idx;
-  char *d_ptr, *i_ptr, *o_ptr;
-  ssize_t d_step, i_step;
-
-  INIT_COUNTER(lp, n);
-  INIT_PTR(lp, 0, d_ptr, d_step);
-
-  idx = f_min_index(n, d_ptr, d_step);
-
-  INIT_PTR(lp, 1, i_ptr, i_step);
-  o_ptr = NDL_PTR(lp, 2);
-  *(idx_t*)o_ptr = *(idx_t*)(i_ptr + i_step * idx);
-}
-#undef idx_t
-
-#define idx_t int32_t
-static void iter_uint16_min_index_index32(na_loop_t* const lp) {
-  size_t n, idx;
-  char *d_ptr, *i_ptr, *o_ptr;
-  ssize_t d_step, i_step;
-
-  INIT_COUNTER(lp, n);
-  INIT_PTR(lp, 0, d_ptr, d_step);
-
-  idx = f_min_index(n, d_ptr, d_step);
-
-  INIT_PTR(lp, 1, i_ptr, i_step);
-  o_ptr = NDL_PTR(lp, 2);
-  *(idx_t*)o_ptr = *(idx_t*)(i_ptr + i_step * idx);
-}
-#undef idx_t
-
-/*
-  Index of the minimum value.
-  @overload min_index(axis:nil)
-    @param [Numeric,Array,Range] axis  Finds minimum values along the axis and returns **flat
-    1-d indices**.
-    @return [Integer,Numo::Int] returns result indices.
-  @see #argmin
-  @see #min
-
-  @example
-      a = Numo::NArray[3,4,1,2]
-      a.min_index  #=> 2
-
-      b = Numo::NArray[[3,4,1],[2,0,5]]
-      b.min_index             #=> 4
-      b.min_index(axis:1)     #=> [2, 4]
-      b.min_index(axis:0)     #=> [3, 4, 2]
-      b[b.min_index(axis:0)]  #=> [2, 0, 1]
- */
-static VALUE uint16_min_index(int argc, VALUE* argv, VALUE self) {
-  narray_t* na;
-  VALUE idx, reduce;
-  ndfunc_arg_in_t ain[3] = { { Qnil, 0 }, { Qnil, 0 }, { sym_reduce, 0 } };
-  ndfunc_arg_out_t aout[1] = { { 0, 0, 0 } };
-  ndfunc_t ndf = { 0, STRIDE_LOOP_NIP | NDF_FLAT_REDUCE | NDF_EXTRACT, 3, 1, ain, aout };
-
-  GetNArray(self, na);
-  if (na->ndim == 0) {
-    return INT2FIX(0);
-  }
-  if (na->size > (~(u_int32_t)0)) {
-    aout[0].type = numo_cInt64;
-    idx = nary_new(numo_cInt64, na->ndim, na->shape);
-    ndf.func = iter_uint16_min_index_index64;
-
-    reduce = na_reduce_dimension(argc, argv, 1, &self, &ndf, 0);
-
-  } else {
-    aout[0].type = numo_cInt32;
-    idx = nary_new(numo_cInt32, na->ndim, na->shape);
-    ndf.func = iter_uint16_min_index_index32;
-
-    reduce = na_reduce_dimension(argc, argv, 1, &self, &ndf, 0);
-  }
-  rb_funcall(idx, rb_intern("seq"), 0);
-
-  return na_ndloop(&ndf, 3, self, idx, reduce);
-}
-
-#define idx_t int64_t
-static void iter_uint16_argmax_arg64(na_loop_t* const lp) {
-  size_t n, idx;
-  char *d_ptr, *o_ptr;
-  ssize_t d_step;
-
-  INIT_COUNTER(lp, n);
-  INIT_PTR(lp, 0, d_ptr, d_step);
-
-  idx = f_max_index(n, d_ptr, d_step);
-
-  o_ptr = NDL_PTR(lp, 1);
-  *(idx_t*)o_ptr = (idx_t)idx;
-}
-#undef idx_t
-
-#define idx_t int32_t
-static void iter_uint16_argmax_arg32(na_loop_t* const lp) {
-  size_t n, idx;
-  char *d_ptr, *o_ptr;
-  ssize_t d_step;
-
-  INIT_COUNTER(lp, n);
-  INIT_PTR(lp, 0, d_ptr, d_step);
-
-  idx = f_max_index(n, d_ptr, d_step);
-
-  o_ptr = NDL_PTR(lp, 1);
-  *(idx_t*)o_ptr = (idx_t)idx;
-}
-#undef idx_t
-
-/*
-  Index of the maximum value.
-  @overload argmax(axis:nil)
-    @param [Numeric,Array,Range] axis  Finds maximum values along the axis and returns **indices
-    along the axis**.
-    @return [Integer,Numo::Int] returns the result indices.
-  @see #max_index
-  @see #max
-
-  @example
-      a = Numo::NArray[3,4,1,2]
-      a.argmax  #=> 1
-
-      b = Numo::NArray[[3,4,1],[2,0,5]]
-      b.argmax                       #=> 5
-      b.argmax(axis:1)               #=> [1, 2]
-      b.argmax(axis:0)               #=> [0, 0, 1]
-      b.at(b.argmax(axis:0), 0..-1)  #=> [3, 4, 5]
- */
-static VALUE uint16_argmax(int argc, VALUE* argv, VALUE self) {
-  narray_t* na;
-  VALUE reduce;
-  ndfunc_arg_in_t ain[2] = { { Qnil, 0 }, { sym_reduce, 0 } };
-  ndfunc_arg_out_t aout[1] = { { 0, 0, 0 } };
-  ndfunc_t ndf = { 0, STRIDE_LOOP_NIP | NDF_FLAT_REDUCE | NDF_EXTRACT, 2, 1, ain, aout };
-
-  GetNArray(self, na);
-  if (na->ndim == 0) {
-    return INT2FIX(0);
-  }
-  if (na->size > (~(u_int32_t)0)) {
-    aout[0].type = numo_cInt64;
-    ndf.func = iter_uint16_argmax_arg64;
-
-    reduce = na_reduce_dimension(argc, argv, 1, &self, &ndf, 0);
-
-  } else {
-    aout[0].type = numo_cInt32;
-    ndf.func = iter_uint16_argmax_arg32;
-
-    reduce = na_reduce_dimension(argc, argv, 1, &self, &ndf, 0);
-  }
-
-  return na_ndloop(&ndf, 2, self, reduce);
-}
-
-#define idx_t int64_t
-static void iter_uint16_argmin_arg64(na_loop_t* const lp) {
-  size_t n, idx;
-  char *d_ptr, *o_ptr;
-  ssize_t d_step;
-
-  INIT_COUNTER(lp, n);
-  INIT_PTR(lp, 0, d_ptr, d_step);
-
-  idx = f_min_index(n, d_ptr, d_step);
-
-  o_ptr = NDL_PTR(lp, 1);
-  *(idx_t*)o_ptr = (idx_t)idx;
-}
-#undef idx_t
-
-#define idx_t int32_t
-static void iter_uint16_argmin_arg32(na_loop_t* const lp) {
-  size_t n, idx;
-  char *d_ptr, *o_ptr;
-  ssize_t d_step;
-
-  INIT_COUNTER(lp, n);
-  INIT_PTR(lp, 0, d_ptr, d_step);
-
-  idx = f_min_index(n, d_ptr, d_step);
-
-  o_ptr = NDL_PTR(lp, 1);
-  *(idx_t*)o_ptr = (idx_t)idx;
-}
-#undef idx_t
-
-/*
-  Index of the minimum value.
-  @overload argmin(axis:nil)
-    @param [Numeric,Array,Range] axis  Finds minimum values along the axis and returns **indices
-    along the axis**.
-    @return [Integer,Numo::Int] returns the result indices.
-  @see #min_index
-  @see #min
-
-  @example
-      a = Numo::NArray[3,4,1,2]
-      a.argmin  #=> 2
-
-      b = Numo::NArray[[3,4,1],[2,0,5]]
-      b.argmin                       #=> 4
-      b.argmin(axis:1)               #=> [2, 1]
-      b.argmin(axis:0)               #=> [1, 1, 0]
-      b.at(b.argmin(axis:0), 0..-1)  #=> [2, 0, 1]
- */
-static VALUE uint16_argmin(int argc, VALUE* argv, VALUE self) {
-  narray_t* na;
-  VALUE reduce;
-  ndfunc_arg_in_t ain[2] = { { Qnil, 0 }, { sym_reduce, 0 } };
-  ndfunc_arg_out_t aout[1] = { { 0, 0, 0 } };
-  ndfunc_t ndf = { 0, STRIDE_LOOP_NIP | NDF_FLAT_REDUCE | NDF_EXTRACT, 2, 1, ain, aout };
-
-  GetNArray(self, na);
-  if (na->ndim == 0) {
-    return INT2FIX(0);
-  }
-  if (na->size > (~(u_int32_t)0)) {
-    aout[0].type = numo_cInt64;
-    ndf.func = iter_uint16_argmin_arg64;
-
-    reduce = na_reduce_dimension(argc, argv, 1, &self, &ndf, 0);
-
-  } else {
-    aout[0].type = numo_cInt32;
-    ndf.func = iter_uint16_argmin_arg32;
-
-    reduce = na_reduce_dimension(argc, argv, 1, &self, &ndf, 0);
-  }
-
-  return na_ndloop(&ndf, 2, self, reduce);
-}
-
-static void iter_uint16_minmax(na_loop_t* const lp) {
-  size_t n;
-  char* p1;
-  ssize_t s1;
-  dtype xmin, xmax;
-
-  INIT_COUNTER(lp, n);
-  INIT_PTR(lp, 0, p1, s1);
-
-  f_minmax(n, p1, s1, &xmin, &xmax);
-
-  *(dtype*)(lp->args[1].ptr + lp->args[1].iter[0].pos) = xmin;
-  *(dtype*)(lp->args[2].ptr + lp->args[2].iter[0].pos) = xmax;
-}
-
-/*
-  minmax of self.
-  @overload minmax(axis:nil, keepdims:false)
-    @param [Numeric,Array,Range] axis  Finds min-max along the axis.
-    @param [TrueClass] keepdims (keyword) If true, the reduced axes are left in the result array
-    as dimensions with size one.
-    @return [Numo::UInt16,Numo::UInt16] min and max of self.
-*/
-static VALUE uint16_minmax(int argc, VALUE* argv, VALUE self) {
-  VALUE reduce;
-  ndfunc_arg_in_t ain[2] = { { cT, 0 }, { sym_reduce, 0 } };
-  ndfunc_arg_out_t aout[2] = { { cT, 0 }, { cT, 0 } };
-  ndfunc_t ndf = {
-    iter_uint16_minmax, STRIDE_LOOP_NIP | NDF_FLAT_REDUCE | NDF_EXTRACT, 2, 2, ain, aout
-  };
-
-  reduce = na_reduce_dimension(argc, argv, 1, &self, &ndf, 0);
-
-  return na_ndloop(&ndf, 2, self, reduce);
-}
-
-static void iter_uint16_s_maximum(na_loop_t* const lp) {
-  size_t i, n;
-  char *p1, *p2, *p3;
-  ssize_t s1, s2, s3;
-
-  INIT_COUNTER(lp, n);
-  INIT_PTR(lp, 0, p1, s1);
-  INIT_PTR(lp, 1, p2, s2);
-  INIT_PTR(lp, 2, p3, s3);
-
-  for (i = 0; i < n; i++) {
-    dtype x, y, z;
-    GET_DATA_STRIDE(p1, s1, dtype, x);
-    GET_DATA_STRIDE(p2, s2, dtype, y);
-    GET_DATA(p3, dtype, z);
-    z = f_maximum(x, y);
-    SET_DATA_STRIDE(p3, s3, dtype, z);
-  }
-}
-
-static VALUE uint16_s_maximum(int argc, VALUE* argv, VALUE mod) {
-  VALUE a1 = Qnil;
-  VALUE a2 = Qnil;
-  ndfunc_arg_in_t ain[2] = { { cT, 0 }, { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_uint16_s_maximum, STRIDE_LOOP_NIP, 2, 1, ain, aout };
-
-  rb_scan_args(argc, argv, "20", &a1, &a2);
-
-  return na_ndloop(&ndf, 2, a1, a2);
-}
-
-static void iter_uint16_s_minimum(na_loop_t* const lp) {
-  size_t i, n;
-  char *p1, *p2, *p3;
-  ssize_t s1, s2, s3;
-
-  INIT_COUNTER(lp, n);
-  INIT_PTR(lp, 0, p1, s1);
-  INIT_PTR(lp, 1, p2, s2);
-  INIT_PTR(lp, 2, p3, s3);
-
-  for (i = 0; i < n; i++) {
-    dtype x, y, z;
-    GET_DATA_STRIDE(p1, s1, dtype, x);
-    GET_DATA_STRIDE(p2, s2, dtype, y);
-    GET_DATA(p3, dtype, z);
-    z = f_minimum(x, y);
-    SET_DATA_STRIDE(p3, s3, dtype, z);
-  }
-}
-
-static VALUE uint16_s_minimum(int argc, VALUE* argv, VALUE mod) {
-  VALUE a1 = Qnil;
-  VALUE a2 = Qnil;
-  ndfunc_arg_in_t ain[2] = { { cT, 0 }, { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_uint16_s_minimum, STRIDE_LOOP_NIP, 2, 1, ain, aout };
-
-  rb_scan_args(argc, argv, "20", &a1, &a2);
-
-  return na_ndloop(&ndf, 2, a1, a2);
-}
-
-// ------- Integer count without weights -------
-
-static void iter_uint16_bincount_32(na_loop_t* const lp) {
-  size_t i, x, n;
-  char *p1, *p2;
-  ssize_t s1, s2;
-  size_t* idx1;
-
-  INIT_PTR_IDX(lp, 0, p1, s1, idx1);
-  INIT_PTR(lp, 1, p2, s2);
-  i = lp->args[0].shape[0];
-  n = lp->args[1].shape[0];
-
-  // initialize
-  for (x = 0; x < n; x++) {
-    *(u_int32_t*)(p2 + s2 * x) = 0;
-  }
-
-  if (idx1) {
-    for (; i--;) {
-      GET_DATA_INDEX(p1, idx1, dtype, x);
-      (*(u_int32_t*)(p2 + s2 * x))++;
-    }
-  } else {
-    for (; i--;) {
-      GET_DATA_STRIDE(p1, s1, dtype, x);
-      (*(u_int32_t*)(p2 + s2 * x))++;
-    }
-  }
-}
-
-static VALUE uint16_bincount_32(VALUE self, size_t length) {
-  size_t shape_out[1] = { length };
-  ndfunc_arg_in_t ain[1] = { { cT, 1 } };
-  ndfunc_arg_out_t aout[1] = { { numo_cUInt32, 1, shape_out } };
-  ndfunc_t ndf = {
-    iter_uint16_bincount_32, NO_LOOP | NDF_STRIDE_LOOP | NDF_INDEX_LOOP, 1, 1, ain, aout
-  };
-
-  return na_ndloop(&ndf, 1, self);
-}
-
-static void iter_uint16_bincount_64(na_loop_t* const lp) {
-  size_t i, x, n;
-  char *p1, *p2;
-  ssize_t s1, s2;
-  size_t* idx1;
-
-  INIT_PTR_IDX(lp, 0, p1, s1, idx1);
-  INIT_PTR(lp, 1, p2, s2);
-  i = lp->args[0].shape[0];
-  n = lp->args[1].shape[0];
-
-  // initialize
-  for (x = 0; x < n; x++) {
-    *(u_int64_t*)(p2 + s2 * x) = 0;
-  }
-
-  if (idx1) {
-    for (; i--;) {
-      GET_DATA_INDEX(p1, idx1, dtype, x);
-      (*(u_int64_t*)(p2 + s2 * x))++;
-    }
-  } else {
-    for (; i--;) {
-      GET_DATA_STRIDE(p1, s1, dtype, x);
-      (*(u_int64_t*)(p2 + s2 * x))++;
-    }
-  }
-}
-
-static VALUE uint16_bincount_64(VALUE self, size_t length) {
-  size_t shape_out[1] = { length };
-  ndfunc_arg_in_t ain[1] = { { cT, 1 } };
-  ndfunc_arg_out_t aout[1] = { { numo_cUInt64, 1, shape_out } };
-  ndfunc_t ndf = {
-    iter_uint16_bincount_64, NO_LOOP | NDF_STRIDE_LOOP | NDF_INDEX_LOOP, 1, 1, ain, aout
-  };
-
-  return na_ndloop(&ndf, 1, self);
-}
-// ------- end of Integer count without weights -------
-
-// ------- Float count with weights -------
-
-static void iter_uint16_bincount_sf(na_loop_t* const lp) {
-  float w;
-  size_t i, x, n, m;
-  char *p1, *p2, *p3;
-  ssize_t s1, s2, s3;
-
-  INIT_PTR(lp, 0, p1, s1);
-  INIT_PTR(lp, 1, p2, s2);
-  INIT_PTR(lp, 2, p3, s3);
-  i = lp->args[0].shape[0];
-  m = lp->args[1].shape[0];
-  n = lp->args[2].shape[0];
-
-  if (i != m) {
-    rb_raise(nary_eShapeError, "size mismatch along last axis between self and weight");
-  }
-
-  // initialize
-  for (x = 0; x < n; x++) {
-    *(float*)(p3 + s3 * x) = 0;
-  }
-  for (; i--;) {
-    GET_DATA_STRIDE(p1, s1, dtype, x);
-    GET_DATA_STRIDE(p2, s2, float, w);
-    (*(float*)(p3 + s3 * x)) += w;
-  }
-}
-
-static VALUE uint16_bincount_sf(VALUE self, VALUE weight, size_t length) {
-  size_t shape_out[1] = { length };
-  ndfunc_arg_in_t ain[2] = { { cT, 1 }, { numo_cSFloat, 1 } };
-  ndfunc_arg_out_t aout[1] = { { numo_cSFloat, 1, shape_out } };
-  ndfunc_t ndf = { iter_uint16_bincount_sf, NO_LOOP | NDF_STRIDE_LOOP, 2, 1, ain, aout };
-
-  return na_ndloop(&ndf, 2, self, weight);
-}
-
-static void iter_uint16_bincount_df(na_loop_t* const lp) {
-  double w;
-  size_t i, x, n, m;
-  char *p1, *p2, *p3;
-  ssize_t s1, s2, s3;
-
-  INIT_PTR(lp, 0, p1, s1);
-  INIT_PTR(lp, 1, p2, s2);
-  INIT_PTR(lp, 2, p3, s3);
-  i = lp->args[0].shape[0];
-  m = lp->args[1].shape[0];
-  n = lp->args[2].shape[0];
-
-  if (i != m) {
-    rb_raise(nary_eShapeError, "size mismatch along last axis between self and weight");
-  }
-
-  // initialize
-  for (x = 0; x < n; x++) {
-    *(double*)(p3 + s3 * x) = 0;
-  }
-  for (; i--;) {
-    GET_DATA_STRIDE(p1, s1, dtype, x);
-    GET_DATA_STRIDE(p2, s2, double, w);
-    (*(double*)(p3 + s3 * x)) += w;
-  }
-}
-
-static VALUE uint16_bincount_df(VALUE self, VALUE weight, size_t length) {
-  size_t shape_out[1] = { length };
-  ndfunc_arg_in_t ain[2] = { { cT, 1 }, { numo_cDFloat, 1 } };
-  ndfunc_arg_out_t aout[1] = { { numo_cDFloat, 1, shape_out } };
-  ndfunc_t ndf = { iter_uint16_bincount_df, NO_LOOP | NDF_STRIDE_LOOP, 2, 1, ain, aout };
-
-  return na_ndloop(&ndf, 2, self, weight);
-}
-// ------- end of Float count with weights -------
-
-/*
-  Count the number of occurrences of each non-negative integer value.
-  Only Integer-types has this method.
-
-  @overload bincount([weight], minlength:nil)
-    @param [SFloat or DFloat or Array] weight (optional) Array of
-      float values. Its size along last axis should be same as that of self.
-    @param [Integer] minlength (keyword, optional) Minimum size along
-      last axis for the output array.
-    @return [UInt32 or UInt64 or SFloat or DFloat]
-      Returns Float NArray if weight array is supplied,
-      otherwise returns UInt32 or UInt64 depending on the size along last axis.
-  @example
-    Numo::Int32[0..4].bincount
-    # => Numo::UInt32#shape=[5]
-    # [1, 1, 1, 1, 1]
-
-    Numo::Int32[0, 1, 1, 3, 2, 1, 7].bincount
-    # => Numo::UInt32#shape=[8]
-    # [1, 3, 1, 1, 0, 0, 0, 1]
-
-    x = Numo::Int32[0, 1, 1, 3, 2, 1, 7, 23]
-    x.bincount.size == x.max+1
-    # => true
-
-    w = Numo::DFloat[0.3, 0.5, 0.2, 0.7, 1.0, -0.6]
-    x = Numo::Int32[0, 1, 1, 2, 2, 2]
-    x.bincount(w)
-    # => Numo::DFloat#shape=[3]
-    # [0.3, 0.7, 1.1]
-
-*/
-static VALUE uint16_bincount(int argc, VALUE* argv, VALUE self) {
-  VALUE weight = Qnil, kw = Qnil;
-  VALUE opts[1] = { Qundef };
-  VALUE v, wclass;
-  ID table[1] = { id_minlength };
-  size_t length, minlength;
-
-  rb_scan_args(argc, argv, "01:", &weight, &kw);
-  rb_get_kwargs(kw, table, 0, 1, opts);
-
-  v = uint16_max(0, 0, self);
-
-  length = NUM2SIZET(v) + 1;
-
-  if (opts[0] != Qundef) {
-    minlength = NUM2SIZET(opts[0]);
-    if (minlength > length) {
-      length = minlength;
-    }
-  }
-
-  if (NIL_P(weight)) {
-    if (length > 4294967295ul) {
-      return uint16_bincount_64(self, length);
-    } else {
-      return uint16_bincount_32(self, length);
-    }
-  } else {
-    wclass = rb_obj_class(weight);
-    if (wclass == numo_cSFloat) {
-      return uint16_bincount_sf(self, weight, length);
-    } else {
-      return uint16_bincount_df(self, weight, length);
-    }
-  }
-}
-
-static void iter_uint16_cumsum(na_loop_t* const lp) {
-  size_t i;
-  char *p1, *p2;
-  ssize_t s1, s2;
-  dtype x, y;
-
-  INIT_COUNTER(lp, i);
-  INIT_PTR(lp, 0, p1, s1);
-  INIT_PTR(lp, 1, p2, s2);
-
-  GET_DATA_STRIDE(p1, s1, dtype, x);
-  SET_DATA_STRIDE(p2, s2, dtype, x);
-  for (i--; i--;) {
-    GET_DATA_STRIDE(p1, s1, dtype, y);
-    m_cumsum(x, y);
-    SET_DATA_STRIDE(p2, s2, dtype, x);
-  }
-}
-
-/*
-  cumsum of self.
-  @overload cumsum(axis:nil, nan:false)
-    @param [Numeric,Array,Range] axis  Performs cumsum along the axis.
-    @param [TrueClass] nan  If true, apply NaN-aware algorithm (avoid NaN if exists).
-    @return [Numo::UInt16] cumsum of self.
-*/
-static VALUE uint16_cumsum(int argc, VALUE* argv, VALUE self) {
-  VALUE reduce;
-  ndfunc_arg_in_t ain[2] = { { cT, 0 }, { sym_reduce, 0 } };
-  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = {
-    iter_uint16_cumsum, STRIDE_LOOP | NDF_FLAT_REDUCE | NDF_CUM, 2, 1, ain, aout
-  };
-
-  reduce = na_reduce_dimension(argc, argv, 1, &self, &ndf, 0);
-
-  return na_ndloop(&ndf, 2, self, reduce);
-}
-
-static void iter_uint16_cumprod(na_loop_t* const lp) {
-  size_t i;
-  char *p1, *p2;
-  ssize_t s1, s2;
-  dtype x, y;
-
-  INIT_COUNTER(lp, i);
-  INIT_PTR(lp, 0, p1, s1);
-  INIT_PTR(lp, 1, p2, s2);
-
-  GET_DATA_STRIDE(p1, s1, dtype, x);
-  SET_DATA_STRIDE(p2, s2, dtype, x);
-  for (i--; i--;) {
-    GET_DATA_STRIDE(p1, s1, dtype, y);
-    m_cumprod(x, y);
-    SET_DATA_STRIDE(p2, s2, dtype, x);
-  }
-}
-
-/*
-  cumprod of self.
-  @overload cumprod(axis:nil, nan:false)
-    @param [Numeric,Array,Range] axis  Performs cumprod along the axis.
-    @param [TrueClass] nan  If true, apply NaN-aware algorithm (avoid NaN if exists).
-    @return [Numo::UInt16] cumprod of self.
-*/
-static VALUE uint16_cumprod(int argc, VALUE* argv, VALUE self) {
-  VALUE reduce;
-  ndfunc_arg_in_t ain[2] = { { cT, 0 }, { sym_reduce, 0 } };
-  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = {
-    iter_uint16_cumprod, STRIDE_LOOP | NDF_FLAT_REDUCE | NDF_CUM, 2, 1, ain, aout
-  };
-
-  reduce = na_reduce_dimension(argc, argv, 1, &self, &ndf, 0);
-
-  return na_ndloop(&ndf, 2, self, reduce);
-}
-
-//
-static void iter_uint16_mulsum(na_loop_t* const lp) {
-  size_t i, n;
-  char *p1, *p2, *p3;
-  ssize_t s1, s2, s3;
-
-  INIT_COUNTER(lp, n);
-  INIT_PTR(lp, 0, p1, s1);
-  INIT_PTR(lp, 1, p2, s2);
-  INIT_PTR(lp, 2, p3, s3);
-
-  if (s3 == 0) {
-    dtype z;
-    // Reduce loop
-    GET_DATA(p3, dtype, z);
-    for (i = 0; i < n; i++) {
-      dtype x, y;
-      GET_DATA_STRIDE(p1, s1, dtype, x);
-      GET_DATA_STRIDE(p2, s2, dtype, y);
-      m_mulsum(x, y, z);
-    }
-    SET_DATA(p3, dtype, z);
-    return;
-  } else {
-    for (i = 0; i < n; i++) {
-      dtype x, y, z;
-      GET_DATA_STRIDE(p1, s1, dtype, x);
-      GET_DATA_STRIDE(p2, s2, dtype, y);
-      GET_DATA(p3, dtype, z);
-      m_mulsum(x, y, z);
-      SET_DATA_STRIDE(p3, s3, dtype, z);
-    }
-  }
-}
-//
-
-static VALUE uint16_mulsum_self(int argc, VALUE* argv, VALUE self) {
-  VALUE v, reduce;
-  VALUE naryv[2];
-  ndfunc_arg_in_t ain[4] = { { cT, 0 }, { cT, 0 }, { sym_reduce, 0 }, { sym_init, 0 } };
-  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_uint16_mulsum, STRIDE_LOOP_NIP, 4, 1, ain, aout };
-
-  if (argc < 1) {
-    rb_raise(rb_eArgError, "wrong number of arguments (%d for >=1)", argc);
-  }
-  // should fix below: [self.ndim,other.ndim].max or?
-  naryv[0] = self;
-  naryv[1] = argv[0];
-  //
-  reduce = na_reduce_dimension(argc - 1, argv + 1, 2, naryv, &ndf, 0);
-  //
-
-  v = na_ndloop(&ndf, 4, self, argv[0], reduce, m_mulsum_init);
-  return uint16_extract(v);
-}
-
-/*
-  Binary mulsum.
-
-  @overload mulsum(other, axis:nil, keepdims:false)
-    @param [Numo::NArray,Numeric] other
-    @param [Numeric,Array,Range] axis  Performs mulsum along the axis.
-    @param [TrueClass] keepdims (keyword) If true, the reduced axes are left in the result array
-    as dimensions with size one.
-    @return [Numo::NArray] mulsum of self and other.
-*/
-static VALUE uint16_mulsum(int argc, VALUE* argv, VALUE self) {
-  //
-  VALUE klass, v;
-  //
-  if (argc < 1) {
-    rb_raise(rb_eArgError, "wrong number of arguments (%d for >=1)", argc);
-  }
-  //
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(argv[0]));
-  if (klass == cT) {
-    return uint16_mulsum_self(argc, argv, self);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    //
-    return rb_funcallv_kw(v, rb_intern("mulsum"), argc, argv, RB_PASS_CALLED_KEYWORDS);
-    //
-  }
-  //
-}
-
-typedef double seq_data_t;
-
-typedef double seq_count_t;
-
-typedef struct {
-  seq_data_t beg;
-  seq_data_t step;
-  seq_count_t count;
-} seq_opt_t;
-
-static void iter_uint16_seq(na_loop_t* const lp) {
-  size_t i;
-  char* p1;
-  ssize_t s1;
-  size_t* idx1;
-  dtype x;
-  seq_data_t beg, step;
-  seq_count_t c;
-  seq_opt_t* g;
-
-  INIT_COUNTER(lp, i);
-  INIT_PTR_IDX(lp, 0, p1, s1, idx1);
-  g = (seq_opt_t*)(lp->opt_ptr);
-  beg = g->beg;
-  step = g->step;
-  c = g->count;
-  if (idx1) {
-    for (; i--;) {
-      x = f_seq(beg, step, c++);
-      *(dtype*)(p1 + *idx1) = x;
-      idx1++;
-    }
-  } else {
-    for (; i--;) {
-      x = f_seq(beg, step, c++);
-      *(dtype*)(p1) = x;
-      p1 += s1;
-    }
-  }
-  g->count = c;
-}
-
-/*
-  Set linear sequence of numbers to self. The sequence is obtained from
-     beg+i*step
-  where i is 1-dimensional index.
-  @overload seq([beg,[step]])
-    @param [Numeric] beg  beginning of sequence. (default=0)
-    @param [Numeric] step  step of sequence. (default=1)
-    @return [Numo::UInt16] self.
-  @example
-    Numo::DFloat.new(6).seq(1,-0.2)
-    # => Numo::DFloat#shape=[6]
-    # [1, 0.8, 0.6, 0.4, 0.2, 0]
-
-    Numo::DComplex.new(6).seq(1,-0.2+0.2i)
-    # => Numo::DComplex#shape=[6]
-    # [1+0i, 0.8+0.2i, 0.6+0.4i, 0.4+0.6i, 0.2+0.8i, 0+1i]
-*/
-static VALUE uint16_seq(int argc, VALUE* argv, VALUE self) {
-  seq_opt_t* g;
-  VALUE vbeg = Qnil, vstep = Qnil;
-  ndfunc_arg_in_t ain[1] = { { OVERWRITE, 0 } };
-  ndfunc_t ndf = { iter_uint16_seq, FULL_LOOP, 1, 0, ain, 0 };
-
-  g = ALLOCA_N(seq_opt_t, 1);
-  g->beg = m_zero;
-  g->step = m_one;
-  g->count = 0;
-  rb_scan_args(argc, argv, "02", &vbeg, &vstep);
-  if (vbeg != Qnil) {
-    g->beg = NUM2DBL(vbeg);
-  }
-  if (vstep != Qnil) {
-    g->step = NUM2DBL(vstep);
-  }
-
-  na_ndloop3(&ndf, g, 1, self);
-  return self;
-}
-
-static void iter_uint16_eye(na_loop_t* const lp) {
-  size_t n0, n1;
-  size_t i0, i1;
-  ssize_t s0, s1;
-  char *p0, *p1;
-  char* g;
-  ssize_t kofs;
-  dtype data;
-
-  g = (char*)(lp->opt_ptr);
-  kofs = *(ssize_t*)g;
-  data = *(dtype*)(g + sizeof(ssize_t));
-
-  n0 = lp->args[0].shape[0];
-  n1 = lp->args[0].shape[1];
-  s0 = lp->args[0].iter[0].step;
-  s1 = lp->args[0].iter[1].step;
-  p0 = NDL_PTR(lp, 0);
-
-  for (i0 = 0; i0 < n0; i0++) {
-    p1 = p0;
-    for (i1 = 0; i1 < n1; i1++) {
-      *(dtype*)p1 = (i0 + kofs == i1) ? data : m_zero;
-      p1 += s1;
-    }
-    p0 += s0;
-  }
-}
-
-/*
-  Eye: Set a value to diagonal components, set 0 to non-diagonal components.
-  @overload eye([element,offset])
-    @param [Numeric] element  Diagonal element to be stored. Default is 1.
-    @param [Integer] offset Diagonal offset from the main diagonal.  The
-        default is 0. k>0 for diagonals above the main diagonal, and k<0
-        for diagonals below the main diagonal.
-    @return [Numo::UInt16] eye of self.
-*/
-static VALUE uint16_eye(int argc, VALUE* argv, VALUE self) {
-  ndfunc_arg_in_t ain[1] = { { OVERWRITE, 2 } };
-  ndfunc_t ndf = { iter_uint16_eye, NO_LOOP, 1, 0, ain, 0 };
-  ssize_t kofs;
-  dtype data;
-  char* g;
-  int nd;
-  narray_t* na;
-
-  // check arguments
-  if (argc > 2) {
-    rb_raise(rb_eArgError, "too many arguments (%d for 0..2)", argc);
-  } else if (argc == 2) {
-    data = m_num_to_data(argv[0]);
-    kofs = NUM2SSIZET(argv[1]);
-  } else if (argc == 1) {
-    data = m_num_to_data(argv[0]);
-    kofs = 0;
-  } else {
-    data = m_one;
-    kofs = 0;
-  }
-
-  GetNArray(self, na);
-  nd = na->ndim;
-  if (nd < 2) {
-    rb_raise(nary_eDimensionError, "less than 2-d array");
-  }
-
-  // Diagonal offset from the main diagonal.
-  if (kofs >= 0) {
-    if ((size_t)(kofs) >= na->shape[nd - 1]) {
-      rb_raise(
-        rb_eArgError,
-        "invalid diagonal offset(%" SZF "d) for "
-        "last dimension size(%" SZF "d)",
-        kofs, na->shape[nd - 1]
-      );
-    }
-  } else {
-    if ((size_t)(-kofs) >= na->shape[nd - 2]) {
-      rb_raise(
-        rb_eArgError,
-        "invalid diagonal offset(%" SZF "d) for "
-        "last-1 dimension size(%" SZF "d)",
-        kofs, na->shape[nd - 2]
-      );
-    }
-  }
-
-  g = ALLOCA_N(char, sizeof(ssize_t) + sizeof(dtype));
-  *(ssize_t*)g = kofs;
-  *(dtype*)(g + sizeof(ssize_t)) = data;
+  ndfunc_t ndf = { iter_uint16_reciprocal, FULL_LOOP, 1, 1, ain, aout };
 
-  na_ndloop3(&ndf, g, 1, self);
-  return self;
+  return na_ndloop(&ndf, 1, self);
 }
 
-#define HWID (sizeof(dtype) * 4)
-
-static int msb_pos(uint32_t a) {
-  int width = HWID;
-  int pos = 0;
-  uint32_t mask = (((dtype)1 << HWID) - 1) << HWID;
+static void iter_uint16_sign(na_loop_t* const lp) {
+  size_t i, n;
+  char *p1, *p2;
+  ssize_t s1, s2;
+  size_t *idx1, *idx2;
+  dtype x;
 
-  if (a == 0) {
-    return -1;
-  }
+  INIT_COUNTER(lp, n);
+  INIT_PTR_IDX(lp, 0, p1, s1, idx1);
+  INIT_PTR_IDX(lp, 1, p2, s2, idx2);
 
-  while (width) {
-    if (a & mask) {
-      pos += width;
+  if (idx1) {
+    if (idx2) {
+      for (i = 0; i < n; i++) {
+        GET_DATA_INDEX(p1, idx1, dtype, x);
+        x = m_sign(x);
+        SET_DATA_INDEX(p2, idx2, dtype, x);
+      }
+    } else {
+      for (i = 0; i < n; i++) {
+        GET_DATA_INDEX(p1, idx1, dtype, x);
+        x = m_sign(x);
+        SET_DATA_STRIDE(p2, s2, dtype, x);
+      }
+    }
+  } else {
+    if (idx2) {
+      for (i = 0; i < n; i++) {
+        GET_DATA_STRIDE(p1, s1, dtype, x);
+        x = m_sign(x);
+        SET_DATA_INDEX(p2, idx2, dtype, x);
+      }
     } else {
-      mask >>= width;
+      //
+      if (is_aligned(p1, sizeof(dtype)) && is_aligned(p2, sizeof(dtype))) {
+        if (s1 == sizeof(dtype) && s2 == sizeof(dtype)) {
+          for (i = 0; i < n; i++) {
+            ((dtype*)p2)[i] = m_sign(((dtype*)p1)[i]);
+          }
+          return;
+        }
+        if (is_aligned_step(s1, sizeof(dtype)) && is_aligned_step(s2, sizeof(dtype))) {
+          //
+          for (i = 0; i < n; i++) {
+            *(dtype*)p2 = m_sign(*(dtype*)p1);
+            p1 += s1;
+            p2 += s2;
+          }
+          return;
+          //
+        }
+      }
+      for (i = 0; i < n; i++) {
+        GET_DATA_STRIDE(p1, s1, dtype, x);
+        x = m_sign(x);
+        SET_DATA_STRIDE(p2, s2, dtype, x);
+      }
+      //
     }
-    width >>= 1;
-    mask &= mask << width;
   }
-  return pos;
 }
 
-/* generates a random number on [0,max) */
-inline static dtype m_rand(uint32_t max, int shift) {
-  uint32_t x;
-  do {
-    x = gen_rand32();
-    x >>= shift;
-  } while (x >= max);
-  return x;
-}
+/*
+  Unary sign.
+  @overload sign
+    @return [Numo::UInt16] sign of self.
+*/
+static VALUE uint16_sign(VALUE self) {
+  ndfunc_arg_in_t ain[1] = { { cT, 0 } };
+  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
+  ndfunc_t ndf = { iter_uint16_sign, FULL_LOOP, 1, 1, ain, aout };
 
-typedef struct {
-  dtype low;
-  uint32_t max;
-} rand_opt_t;
+  return na_ndloop(&ndf, 1, self);
+}
 
-static void iter_uint16_rand(na_loop_t* const lp) {
-  size_t i;
-  char* p1;
-  ssize_t s1;
-  size_t* idx1;
+static void iter_uint16_square(na_loop_t* const lp) {
+  size_t i, n;
+  char *p1, *p2;
+  ssize_t s1, s2;
+  size_t *idx1, *idx2;
   dtype x;
-  rand_opt_t* g;
-  dtype low;
-  uint32_t max;
-  int shift;
 
-  INIT_COUNTER(lp, i);
+  INIT_COUNTER(lp, n);
   INIT_PTR_IDX(lp, 0, p1, s1, idx1);
-  g = (rand_opt_t*)(lp->opt_ptr);
-  low = g->low;
-  max = g->max;
-  shift = 31 - msb_pos(max);
+  INIT_PTR_IDX(lp, 1, p2, s2, idx2);
 
   if (idx1) {
-    for (; i--;) {
-      x = m_add(m_rand(max, shift), low);
-      SET_DATA_INDEX(p1, idx1, dtype, x);
+    if (idx2) {
+      for (i = 0; i < n; i++) {
+        GET_DATA_INDEX(p1, idx1, dtype, x);
+        x = m_square(x);
+        SET_DATA_INDEX(p2, idx2, dtype, x);
+      }
+    } else {
+      for (i = 0; i < n; i++) {
+        GET_DATA_INDEX(p1, idx1, dtype, x);
+        x = m_square(x);
+        SET_DATA_STRIDE(p2, s2, dtype, x);
+      }
     }
   } else {
-    for (; i--;) {
-      x = m_add(m_rand(max, shift), low);
-      SET_DATA_STRIDE(p1, s1, dtype, x);
+    if (idx2) {
+      for (i = 0; i < n; i++) {
+        GET_DATA_STRIDE(p1, s1, dtype, x);
+        x = m_square(x);
+        SET_DATA_INDEX(p2, idx2, dtype, x);
+      }
+    } else {
+      //
+      if (is_aligned(p1, sizeof(dtype)) && is_aligned(p2, sizeof(dtype))) {
+        if (s1 == sizeof(dtype) && s2 == sizeof(dtype)) {
+          for (i = 0; i < n; i++) {
+            ((dtype*)p2)[i] = m_square(((dtype*)p1)[i]);
+          }
+          return;
+        }
+        if (is_aligned_step(s1, sizeof(dtype)) && is_aligned_step(s2, sizeof(dtype))) {
+          //
+          for (i = 0; i < n; i++) {
+            *(dtype*)p2 = m_square(*(dtype*)p1);
+            p1 += s1;
+            p2 += s2;
+          }
+          return;
+          //
+        }
+      }
+      for (i = 0; i < n; i++) {
+        GET_DATA_STRIDE(p1, s1, dtype, x);
+        x = m_square(x);
+        SET_DATA_STRIDE(p2, s2, dtype, x);
+      }
+      //
     }
   }
 }
 
 /*
-  Generate uniformly distributed random numbers on self narray.
-  @overload rand([[low],high])
-    @param [Numeric] low  lower inclusive boundary of random numbers. (default=0)
-    @param [Numeric] high  upper exclusive boundary of random numbers. (default=1 or 1+1i for
-    complex types)
-    @return [Numo::UInt16] self.
-  @example
-    Numo::DFloat.new(6).rand
-    # => Numo::DFloat#shape=[6]
-    # [0.0617545, 0.373067, 0.794815, 0.201042, 0.116041, 0.344032]
-
-    Numo::DComplex.new(6).rand(5+5i)
-    # => Numo::DComplex#shape=[6]
-    # [2.69974+3.68908i, 0.825443+0.254414i, 0.540323+0.34354i, 4.52061+2.39322i, ...]
-
-    Numo::Int32.new(6).rand(2,5)
-    # => Numo::Int32#shape=[6]
-    # [4, 3, 3, 2, 4, 2]
+  Unary square.
+  @overload square
+    @return [Numo::UInt16] square of self.
 */
-static VALUE uint16_rand(int argc, VALUE* argv, VALUE self) {
-  rand_opt_t g;
-  VALUE v1 = Qnil, v2 = Qnil;
-  dtype high;
-  ndfunc_arg_in_t ain[1] = { { OVERWRITE, 0 } };
-  ndfunc_t ndf = { iter_uint16_rand, FULL_LOOP, 1, 0, ain, 0 };
-
-  rb_scan_args(argc, argv, "11", &v1, &v2);
-  if (v2 == Qnil) {
-    g.low = m_zero;
-    g.max = high = m_num_to_data(v1);
-
-  } else {
-    g.low = m_num_to_data(v1);
-    high = m_num_to_data(v2);
-    g.max = m_sub(high, g.low);
-  }
-
-  if (high <= g.low) {
-    rb_raise(rb_eArgError, "high must be larger than low");
-  }
+static VALUE uint16_square(VALUE self) {
+  ndfunc_arg_in_t ain[1] = { { cT, 0 } };
+  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
+  ndfunc_t ndf = { iter_uint16_square, FULL_LOOP, 1, 1, ain, aout };
 
-  na_ndloop3(&ndf, &g, 1, self);
-  return self;
+  return na_ndloop(&ndf, 1, self);
 }
 
 static void iter_uint16_poly(na_loop_t* const lp) {
@@ -5711,22 +2619,91 @@ void Init_numo_uint16(void) {
   rb_define_singleton_method(cT, "cast", uint16_s_cast, 1);
   rb_define_method(cT, "[]", uint16_aref, -1);
   rb_define_method(cT, "[]=", uint16_aset, -1);
+  /**
+   * return NArray with cast to the type of self.
+   * @overload coerce_cast(type)
+   *   @return [nil]
+   */
   rb_define_method(cT, "coerce_cast", uint16_coerce_cast, 1);
+  /**
+   * Convert self to Array.
+   * @overload to_a
+   *   @return [Array]
+   */
   rb_define_method(cT, "to_a", uint16_to_a, 0);
+  /**
+   * Fill elements with other.
+   * @overload fill other
+   *   @param [Numeric] other
+   *   @return [Numo::UInt16] self.
+   */
   rb_define_method(cT, "fill", uint16_fill, 1);
+  /**
+   * Format elements into strings.
+   * @overload format format
+   *   @param [String] format
+   *   @return [Numo::RObject] array of formatted strings.
+   */
   rb_define_method(cT, "format", uint16_format, -1);
+  /**
+   * Format elements into strings.
+   * @overload format_to_a format
+   *   @param [String] format
+   *   @return [Array] array of formatted strings.
+   */
   rb_define_method(cT, "format_to_a", uint16_format_to_a, -1);
+  /**
+   * Returns a string containing a human-readable representation of NArray.
+   * @overload inspect
+   *   @return [String]
+   */
   rb_define_method(cT, "inspect", uint16_inspect, 0);
   rb_define_method(cT, "each", uint16_each, 0);
   rb_define_method(cT, "map", uint16_map, 0);
   rb_define_method(cT, "each_with_index", uint16_each_with_index, 0);
   rb_define_method(cT, "map_with_index", uint16_map_with_index, 0);
   rb_define_method(cT, "abs", uint16_abs, 0);
+  /**
+   * Binary add.
+   * @overload + other
+   *   @param [Numo::NArray,Numeric] other
+   *   @return [Numo::NArray] self + other
+   */
   rb_define_method(cT, "+", uint16_add, 1);
+  /**
+   * Binary sub.
+   * @overload - other
+   *   @param [Numo::NArray,Numeric] other
+   *   @return [Numo::NArray] self - other
+   */
   rb_define_method(cT, "-", uint16_sub, 1);
+  /**
+   * Binary mul.
+   * @overload * other
+   *   @param [Numo::NArray,Numeric] other
+   *   @return [Numo::NArray] self * other
+   */
   rb_define_method(cT, "*", uint16_mul, 1);
+  /**
+   * Binary div.
+   * @overload / other
+   *   @param [Numo::NArray,Numeric] other
+   *   @return [Numo::NArray] self / other
+   */
   rb_define_method(cT, "/", uint16_div, 1);
+  /**
+   * Binary mod.
+   * @overload % other
+   *   @param [Numo::NArray,Numeric] other
+   *   @return [Numo::NArray] self % other
+   */
   rb_define_method(cT, "%", uint16_mod, 1);
+  /**
+   * Binary divmod.
+   * @overload divmod other
+   *   @param [Numo::NArray,Numeric] other
+   *   @return [Numo::NArray] divmod of self and other.
+   */
   rb_define_method(cT, "divmod", uint16_divmod, 1);
   rb_define_method(cT, "**", uint16_pow, 1);
   rb_define_alias(cT, "pow", "**");
@@ -5737,39 +2714,264 @@ void Init_numo_uint16(void) {
   rb_define_alias(cT, "conj", "view");
   rb_define_alias(cT, "im", "view");
   rb_define_alias(cT, "conjugate", "conj");
+  /**
+   * Comparison eq other.
+   * @overload eq other
+   *   @param [Numo::NArray,Numeric] other
+   *   @return [Numo::Bit] result of self eq other.
+   */
   rb_define_method(cT, "eq", uint16_eq, 1);
+  /**
+   * Comparison ne other.
+   * @overload ne other
+   *   @param [Numo::NArray,Numeric] other
+   *   @return [Numo::Bit] result of self ne other.
+   */
   rb_define_method(cT, "ne", uint16_ne, 1);
   rb_define_alias(cT, "nearly_eq", "eq");
   rb_define_alias(cT, "close_to", "nearly_eq");
+  /**
+   * Binary bit_and.
+   * @overload & other
+   *   @param [Numo::NArray,Numeric] other
+   *   @return [Numo::NArray] self & other
+   */
   rb_define_method(cT, "&", uint16_bit_and, 1);
+  /**
+   * Binary bit_or.
+   * @overload | other
+   *   @param [Numo::NArray,Numeric] other
+   *   @return [Numo::NArray] self | other
+   */
   rb_define_method(cT, "|", uint16_bit_or, 1);
+  /**
+   * Binary bit_xor.
+   * @overload ^ other
+   *   @param [Numo::NArray,Numeric] other
+   *   @return [Numo::NArray] self ^ other
+   */
   rb_define_method(cT, "^", uint16_bit_xor, 1);
+  /**
+   * Unary bit_not.
+   * @overload ~
+   *   @return [Numo::UInt16] bit_not of self.
+   */
   rb_define_method(cT, "~", uint16_bit_not, 0);
+  /**
+   * Binary left_shift.
+   * @overload << other
+   *   @param [Numo::NArray,Numeric] other
+   *   @return [Numo::NArray] self << other
+   */
   rb_define_method(cT, "<<", uint16_left_shift, 1);
+  /**
+   * Binary right_shift.
+   * @overload >> other
+   *   @param [Numo::NArray,Numeric] other
+   *   @return [Numo::NArray] self >> other
+   */
   rb_define_method(cT, ">>", uint16_right_shift, 1);
   rb_define_alias(cT, "floor", "view");
   rb_define_alias(cT, "round", "view");
   rb_define_alias(cT, "ceil", "view");
   rb_define_alias(cT, "trunc", "view");
   rb_define_alias(cT, "rint", "view");
+  /**
+   * Comparison gt other.
+   * @overload gt other
+   *   @param [Numo::NArray,Numeric] other
+   *   @return [Numo::Bit] result of self gt other.
+   */
   rb_define_method(cT, "gt", uint16_gt, 1);
+  /**
+   * Comparison ge other.
+   * @overload ge other
+   *   @param [Numo::NArray,Numeric] other
+   *   @return [Numo::Bit] result of self ge other.
+   */
   rb_define_method(cT, "ge", uint16_ge, 1);
+  /**
+   * Comparison lt other.
+   * @overload lt other
+   *   @param [Numo::NArray,Numeric] other
+   *   @return [Numo::Bit] result of self lt other.
+   */
   rb_define_method(cT, "lt", uint16_lt, 1);
+  /**
+   * Comparison le other.
+   * @overload le other
+   *   @param [Numo::NArray,Numeric] other
+   *   @return [Numo::Bit] result of self le other.
+   */
   rb_define_method(cT, "le", uint16_le, 1);
   rb_define_alias(cT, ">", "gt");
   rb_define_alias(cT, ">=", "ge");
   rb_define_alias(cT, "<", "lt");
   rb_define_alias(cT, "<=", "le");
+  /**
+   * Clip array elements by [min,max].
+   * If either of min or max is nil, one side is clipped.
+   * @overload clip(min,max)
+   *   @param [Numo::NArray,Numeric] min
+   *   @param [Numo::NArray,Numeric] max
+   *   @return [Numo::NArray] result of clip.
+   *
+   * @example
+   *     a = Numo::Int32.new(10).seq
+   *     # => Numo::Int32#shape=[10]
+   *     # [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]
+   *
+   *     a.clip(1,8)
+   *     # => Numo::Int32#shape=[10]
+   *     # [1, 1, 2, 3, 4, 5, 6, 7, 8, 8]
+   *
+   *     a.inplace.clip(3,6)
+   *     a
+   *     # => Numo::Int32#shape=[10]
+   *     # [3, 3, 3, 3, 4, 5, 6, 6, 6, 6]
+   *
+   *     b = Numo::Int32.new(10).seq
+   *     # => Numo::Int32#shape=[10]
+   *     # [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]
+   *
+   *     b.clip([3,4,1,1,1,4,4,4,4,4], 8)
+   *     # => Numo::Int32#shape=[10]
+   *     # [3, 4, 2, 3, 4, 5, 6, 7, 8, 8]
+   */
   rb_define_method(cT, "clip", uint16_clip, 2);
+  /**
+   * sum of self.
+   * @overload sum(axis:nil, keepdims:false)
+   *   @param [Numeric,Array,Range] axis  Performs sum along the axis.
+   *   @param [TrueClass] keepdims  If true, the reduced axes are left in the result array as
+   *     dimensions with size one.
+   *   @return [Numo::UInt64] returns result of sum.
+   */
   rb_define_method(cT, "sum", uint16_sum, -1);
+  /**
+   * prod of self.
+   * @overload prod(axis:nil, keepdims:false)
+   *   @param [Numeric,Array,Range] axis  Performs prod along the axis.
+   *   @param [TrueClass] keepdims  If true, the reduced axes are left in the result array as
+   *     dimensions with size one.
+   *   @return [Numo::UInt64] returns result of prod.
+   */
   rb_define_method(cT, "prod", uint16_prod, -1);
+  /**
+   * min of self.
+   * @overload min(axis:nil, keepdims:false)
+   *   @param [Numeric,Array,Range] axis  Performs min along the axis.
+   *   @param [TrueClass] keepdims  If true, the reduced axes are left in the result array as
+   *     dimensions with size one.
+   *   @return [Numo::UInt16] returns result of min.
+   */
   rb_define_method(cT, "min", uint16_min, -1);
+  /**
+   * max of self.
+   * @overload max(axis:nil, keepdims:false)
+   *   @param [Numeric,Array,Range] axis  Performs max along the axis.
+   *   @param [TrueClass] keepdims  If true, the reduced axes are left in the result array as
+   *     dimensions with size one.
+   *   @return [Numo::UInt16] returns result of max.
+   */
   rb_define_method(cT, "max", uint16_max, -1);
+  /**
+   * ptp of self.
+   * @overload ptp(axis:nil, keepdims:false)
+   *   @param [Numeric,Array,Range] axis  Performs ptp along the axis.
+   *   @param [TrueClass] keepdims  If true, the reduced axes are left in the result array as
+   *     dimensions with size one.
+   *   @return [Numo::UInt16] returns result of ptp.
+   */
   rb_define_method(cT, "ptp", uint16_ptp, -1);
+  /**
+   * Index of the maximum value.
+   * @overload max_index(axis:nil)
+   *   @param [Numeric,Array,Range] axis  Finds maximum values along the axis and
+   *     returns **flat 1-d indices**.
+   *   @return [Integer,Numo::Int] returns result indices.
+   * @see #argmax
+   * @see #max
+   *
+   * @example
+   *     a = Numo::NArray[3,4,1,2]
+   *     a.max_index  #=> 1
+   *
+   *     b = Numo::NArray[[3,4,1],[2,0,5]]
+   *     b.max_index             #=> 5
+   *     b.max_index(axis:1)     #=> [1, 5]
+   *     b.max_index(axis:0)     #=> [0, 1, 5]
+   *     b[b.max_index(axis:0)]  #=> [3, 4, 5]
+   */
   rb_define_method(cT, "max_index", uint16_max_index, -1);
+  /**
+   * Index of the minimum value.
+   * @overload min_index(axis:nil)
+   *   @param [Numeric,Array,Range] axis  Finds minimum values along the axis and
+   *     returns **flat 1-d indices**.
+   *   @return [Integer,Numo::Int] returns result indices.
+   * @see #argmin
+   * @see #min
+   *
+   * @example
+   *     a = Numo::NArray[3,4,1,2]
+   *     a.min_index  #=> 2
+   *
+   *     b = Numo::NArray[[3,4,1],[2,0,5]]
+   *     b.min_index             #=> 4
+   *     b.min_index(axis:1)     #=> [2, 4]
+   *     b.min_index(axis:0)     #=> [3, 4, 2]
+   *     b[b.min_index(axis:0)]  #=> [2, 0, 1]
+   */
   rb_define_method(cT, "min_index", uint16_min_index, -1);
+  /**
+   * Index of the maximum value.
+   * @overload argmax(axis:nil)
+   *   @param [Numeric,Array,Range] axis  Finds maximum values along the axis and
+   *     returns **indices along the axis**.
+   *   @return [Integer,Numo::Int] returns the result indices.
+   * @see #max_index
+   * @see #max
+   *
+   * @example
+   *     a = Numo::NArray[3,4,1,2]
+   *     a.argmax  #=> 1
+   *
+   *     b = Numo::NArray[[3,4,1],[2,0,5]]
+   *     b.argmax                       #=> 5
+   *     b.argmax(axis:1)               #=> [1, 2]
+   *     b.argmax(axis:0)               #=> [0, 0, 1]
+   *     b.at(b.argmax(axis:0), 0..-1)  #=> [3, 4, 5]
+   */
   rb_define_method(cT, "argmax", uint16_argmax, -1);
+  /**
+   * Index of the minimum value.
+   * @overload argmin(axis:nil)
+   *   @param [Numeric,Array,Range] axis  Finds minimum values along the axis and
+   *     returns **indices along the axis**.
+   *   @return [Integer,Numo::Int] returns the result indices.
+   * @see #min_index
+   * @see #min
+   *
+   * @example
+   *     a = Numo::NArray[3,4,1,2]
+   *     a.argmin  #=> 2
+   *
+   *     b = Numo::NArray[[3,4,1],[2,0,5]]
+   *     b.argmin                       #=> 4
+   *     b.argmin(axis:1)               #=> [2, 1]
+   *     b.argmin(axis:0)               #=> [1, 1, 0]
+   *     b.at(b.argmin(axis:0), 0..-1)  #=> [2, 0, 1]
+   */
   rb_define_method(cT, "argmin", uint16_argmin, -1);
+  /**
+   * minmax of self.
+   * @overload minmax(axis:nil, keepdims:false)
+   *   @param [Numeric,Array,Range] axis  Finds min-max along the axis.
+   *   @param [TrueClass] keepdims (keyword) If true, the reduced axes are left in
+   *     the result array as dimensions with size one.
+   *   @return [Numo::UInt16,Numo::UInt16] min and max of self.
+   */
   rb_define_method(cT, "minmax", uint16_minmax, -1);
   /**
    * Element-wise maximum of two arrays.
@@ -5785,13 +2987,114 @@ void Init_numo_uint16(void) {
    *   @return [Numo::UInt16]
    */
   rb_define_module_function(cT, "minimum", uint16_s_minimum, -1);
+  /**
+   * Count the number of occurrences of each non-negative integer value.
+   * Only Integer-types has this method.
+   *
+   * @overload bincount([weight], minlength:nil)
+   *   @param [SFloat or DFloat or Array] weight (optional) Array of
+   *     float values. Its size along last axis should be same as that of self.
+   *   @param [Integer] minlength (keyword, optional) Minimum size along
+   *     last axis for the output array.
+   *   @return [UInt32 or UInt64 or SFloat or DFloat]
+   *     Returns Float NArray if weight array is supplied,
+   *     otherwise returns UInt32 or UInt64 depending on the size along last axis.
+   * @example
+   *   Numo::Int32[0..4].bincount
+   *   # => Numo::UInt32#shape=[5]
+   *   # [1, 1, 1, 1, 1]
+   *
+   *   Numo::Int32[0, 1, 1, 3, 2, 1, 7].bincount
+   *   # => Numo::UInt32#shape=[8]
+   *   # [1, 3, 1, 1, 0, 0, 0, 1]
+   *
+   *   x = Numo::Int32[0, 1, 1, 3, 2, 1, 7, 23]
+   *   x.bincount.size == x.max+1
+   *   # => true
+   *
+   *   w = Numo::DFloat[0.3, 0.5, 0.2, 0.7, 1.0, -0.6]
+   *   x = Numo::Int32[0, 1, 1, 2, 2, 2]
+   *   x.bincount(w)
+   *   # => Numo::DFloat#shape=[3]
+   *   # [0.3, 0.7, 1.1]
+   */
   rb_define_method(cT, "bincount", uint16_bincount, -1);
+  /**
+   * cumsum of self.
+   * @overload cumsum(axis:nil, nan:false)
+   *   @param [Numeric,Array,Range] axis  Performs cumsum along the axis.
+   *   @param [TrueClass] nan  If true, apply NaN-aware algorithm (avoid NaN if exists).
+   *   @return [Numo::UInt16] cumsum of self.
+   */
   rb_define_method(cT, "cumsum", uint16_cumsum, -1);
+  /**
+   * cumprod of self.
+   * @overload cumprod(axis:nil, nan:false)
+   *   @param [Numeric,Array,Range] axis  Performs cumprod along the axis.
+   *   @param [TrueClass] nan  If true, apply NaN-aware algorithm (avoid NaN if exists).
+   *   @return [Numo::UInt16] cumprod of self.
+   */
   rb_define_method(cT, "cumprod", uint16_cumprod, -1);
+  /**
+   * Binary mulsum.
+   *
+   * @overload mulsum(other, axis:nil, keepdims:false)
+   *   @param [Numo::NArray,Numeric] other
+   *   @param [Numeric,Array,Range] axis  Performs mulsum along the axis.
+   *   @param [TrueClass] keepdims (keyword) If true, the reduced axes are left in
+   *     the result array as dimensions with size one.
+   *   @return [Numo::NArray] mulsum of self and other.
+   */
   rb_define_method(cT, "mulsum", uint16_mulsum, -1);
+  /**
+   * Set linear sequence of numbers to self. The sequence is obtained from
+   *    beg+i*step
+   * where i is 1-dimensional index.
+   * @overload seq([beg,[step]])
+   *   @param [Numeric] beg  beginning of sequence. (default=0)
+   *   @param [Numeric] step  step of sequence. (default=1)
+   *   @return [Numo::UInt16] self.
+   * @example
+   *   Numo::DFloat.new(6).seq(1,-0.2)
+   *   # => Numo::DFloat#shape=[6]
+   *   # [1, 0.8, 0.6, 0.4, 0.2, 0]
+   *
+   *   Numo::DComplex.new(6).seq(1,-0.2+0.2i)
+   *   # => Numo::DComplex#shape=[6]
+   *   # [1+0i, 0.8+0.2i, 0.6+0.4i, 0.4+0.6i, 0.2+0.8i, 0+1i]
+   */
   rb_define_method(cT, "seq", uint16_seq, -1);
+  /**
+   * Eye: Set a value to diagonal components, set 0 to non-diagonal components.
+   * @overload eye([element,offset])
+   *   @param [Numeric] element  Diagonal element to be stored. Default is 1.
+   *   @param [Integer] offset Diagonal offset from the main diagonal.  The
+   *       default is 0. k>0 for diagonals above the main diagonal, and k<0
+   *       for diagonals below the main diagonal.
+   *   @return [Numo::UInt16] eye of self.
+   */
   rb_define_method(cT, "eye", uint16_eye, -1);
   rb_define_alias(cT, "indgen", "seq");
+  /**
+   * Generate uniformly distributed random numbers on self narray.
+   * @overload rand([[low],high])
+   *   @param [Numeric] low  lower inclusive boundary of random numbers. (default=0)
+   *   @param [Numeric] high  upper exclusive boundary of random numbers.
+   *     (default=1 or 1+1i for complex types)
+   *   @return [Numo::UInt16] self.
+   * @example
+   *   Numo::DFloat.new(6).rand
+   *   # => Numo::DFloat#shape=[6]
+   *   # [0.0617545, 0.373067, 0.794815, 0.201042, 0.116041, 0.344032]
+   *
+   *   Numo::DComplex.new(6).rand(5+5i)
+   *   # => Numo::DComplex#shape=[6]
+   *   # [2.69974+3.68908i, 0.825443+0.254414i, 0.540323+0.34354i, 4.52061+2.39322i, ...]
+   *
+   *   Numo::Int32.new(6).rand(2,5)
+   *   # => Numo::Int32#shape=[6]
+   *   # [4, 3, 3, 2, 4, 2]
+   */
   rb_define_method(cT, "rand", uint16_rand, -1);
   rb_define_method(cT, "poly", uint16_poly, -2);