numo-narray-alt 0.9.11 → 0.9.13

data/ext/numo/narray/src/t_int32.c

@@ -43,6 +43,35 @@ 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/pow.h"
+#include "mh/minus.h"
+#include "mh/reciprocal.h"
+#include "mh/sign.h"
+#include "mh/square.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"
@@ -56,6 +85,7 @@ extern VALUE cRT;
 #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"
@@ -70,6 +100,35 @@ extern VALUE cRT;
 
 typedef int32_t int32; // Type aliases for shorter notation
                        // following the codebase naming convention.
+DEF_NARRAY_COERCE_CAST_METHOD_FUNC(int32)
+DEF_NARRAY_TO_A_METHOD_FUNC(int32)
+DEF_NARRAY_FILL_METHOD_FUNC(int32)
+DEF_NARRAY_FORMAT_METHOD_FUNC(int32)
+DEF_NARRAY_FORMAT_TO_A_METHOD_FUNC(int32)
+DEF_NARRAY_INSPECT_METHOD_FUNC(int32)
+DEF_NARRAY_ADD_METHOD_FUNC(int32, numo_cInt32)
+DEF_NARRAY_SUB_METHOD_FUNC(int32, numo_cInt32)
+DEF_NARRAY_MUL_METHOD_FUNC(int32, numo_cInt32)
+DEF_NARRAY_INT_DIV_METHOD_FUNC(int32, numo_cInt32)
+DEF_NARRAY_INT_MOD_METHOD_FUNC(int32, numo_cInt32)
+DEF_NARRAY_INT_DIVMOD_METHOD_FUNC(int32, numo_cInt32)
+DEF_NARRAY_POW_METHOD_FUNC(int32, numo_cInt32)
+DEF_NARRAY_MINUS_METHOD_FUNC(int32, numo_cInt32)
+DEF_NARRAY_RECIPROCAL_METHOD_FUNC(int32, numo_cInt32)
+DEF_NARRAY_SIGN_METHOD_FUNC(int32, numo_cInt32)
+DEF_NARRAY_SQUARE_METHOD_FUNC(int32, numo_cInt32)
+DEF_NARRAY_EQ_METHOD_FUNC(int32, numo_cInt32)
+DEF_NARRAY_NE_METHOD_FUNC(int32, numo_cInt32)
+DEF_NARRAY_GT_METHOD_FUNC(int32, numo_cInt32)
+DEF_NARRAY_GE_METHOD_FUNC(int32, numo_cInt32)
+DEF_NARRAY_LT_METHOD_FUNC(int32, numo_cInt32)
+DEF_NARRAY_LE_METHOD_FUNC(int32, numo_cInt32)
+DEF_NARRAY_INT_BIT_AND_METHOD_FUNC(int32, numo_cInt32)
+DEF_NARRAY_INT_BIT_OR_METHOD_FUNC(int32, numo_cInt32)
+DEF_NARRAY_INT_BIT_XOR_METHOD_FUNC(int32, numo_cInt32)
+DEF_NARRAY_INT_BIT_NOT_METHOD_FUNC(int32, numo_cInt32)
+DEF_NARRAY_INT_LEFT_SHIFT_METHOD_FUNC(int32, numo_cInt32)
+DEF_NARRAY_INT_RIGHT_SHIFT_METHOD_FUNC(int32, numo_cInt32)
 DEF_NARRAY_CLIP_METHOD_FUNC(int32, numo_cInt32)
 DEF_NARRAY_INT_SUM_METHOD_FUNC(int32, numo_cInt32, int64_t, numo_cInt64)
 DEF_NARRAY_INT_PROD_METHOD_FUNC(int32, numo_cInt32, int64_t, numo_cInt64)
@@ -83,6 +142,7 @@ DEF_NARRAY_INT_ARGMIN_METHOD_FUNC(int32)
 DEF_NARRAY_INT_MAXIMUM_METHOD_FUNC(int32, numo_cInt32)
 DEF_NARRAY_INT_MINIMUM_METHOD_FUNC(int32, numo_cInt32)
 DEF_NARRAY_INT_MINMAX_METHOD_FUNC(int32, numo_cInt32)
+DEF_NARRAY_INT_BINCOUNT_METHOD_FUNC(int32, numo_cInt32)
 DEF_NARRAY_INT_CUMSUM_METHOD_FUNC(int32, numo_cInt32)
 DEF_NARRAY_INT_CUMPROD_METHOD_FUNC(int32, numo_cInt32)
 DEF_NARRAY_INT_MULSUM_METHOD_FUNC(int32, numo_cInt32)
@@ -1253,204 +1313,6 @@ static VALUE int32_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 int32_coerce_cast(VALUE self, VALUE type) {
-  return Qnil;
-}
-
-static void iter_int32_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 int32_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_int32_to_a, FULL_LOOP_NIP, 3, 1, ain, aout };
-  return na_ndloop_cast_narray_to_rarray(&ndf, self, Qnil);
-}
-
-static void iter_int32_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::Int32] self.
-*/
-static VALUE int32_fill(VALUE self, VALUE val) {
-  ndfunc_arg_in_t ain[2] = { { OVERWRITE, 0 }, { sym_option } };
-  ndfunc_t ndf = { iter_int32_fill, FULL_LOOP, 2, 0, ain, 0 };
-
-  na_ndloop(&ndf, 2, self, val);
-  return self;
-}
-
-static VALUE format_int32(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_int32_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_int32(fmt, x);
-      SET_DATA_STRIDE(p2, s2, VALUE, y);
-    }
-  } else {
-    for (; i--;) {
-      x = (dtype*)p1;
-      p1 += s1;
-      y = format_int32(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 int32_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_int32_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_int32_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_int32(fmt, x);
-      rb_ary_push(a, y);
-    }
-  } else {
-    for (; i--;) {
-      x = (dtype*)p1;
-      p1 += s1;
-      y = format_int32(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 int32_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_int32_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_int32_inspect(char* ptr, size_t pos, VALUE fmt) {
-  return format_int32(fmt, (dtype*)(ptr + pos));
-}
-
-/*
-  Returns a string containing a human-readable representation of NArray.
-  @overload inspect
-    @return [String]
-*/
-static VALUE int32_inspect(VALUE ary) {
-  return na_ndloop_inspect(ary, iter_int32_inspect, Qnil);
-}
-
 static void iter_int32_each(na_loop_t* const lp) {
   size_t i, s1;
   char* p1;
@@ -1588,2388 +1450,195 @@ static void iter_int32_each_with_index(na_loop_t* const lp) {
   int nd, md;
 
   c = (size_t*)(lp->opt_ptr);
-  nd = lp->ndim;
-  if (nd > 0) {
-    nd--;
-  }
-  md = nd + 2;
-  a = ALLOCA_N(VALUE, md);
-
-  INIT_COUNTER(lp, i);
-  INIT_PTR_IDX(lp, 0, p1, s1, idx1);
-  c[nd] = 0;
-  if (idx1) {
-    for (; i--;) {
-      GET_DATA_INDEX(p1, idx1, dtype, x);
-      yield_each_with_index(x, c, a, nd, md);
-      c[nd]++;
-    }
-  } else {
-    for (; i--;) {
-      GET_DATA_STRIDE(p1, s1, dtype, x);
-      yield_each_with_index(x, c, a, nd, md);
-      c[nd]++;
-    }
-  }
-}
-
-/*
-  Invokes the given block once for each element of self,
-  passing that element and indices along each axis as parameters.
-  @overload each_with_index
-    For a block `{|x,i,j,...| ... }`,
-    @yieldparam [Numeric] x  an element
-    @yieldparam [Integer] i,j,...  multitimensional indices
-    @return [Numo::NArray] self
-  @see #each
-  @see #map_with_index
-*/
-static VALUE int32_each_with_index(VALUE self) {
-  ndfunc_arg_in_t ain[1] = { { Qnil, 0 } };
-  ndfunc_t ndf = { iter_int32_each_with_index, FULL_LOOP_NIP, 1, 0, ain, 0 };
-
-  na_ndloop_with_index(&ndf, 1, self);
-  return self;
-}
-
-static inline dtype yield_map_with_index(dtype x, size_t* c, VALUE* a, int nd, int md) {
-  int j;
-  VALUE y;
-
-  a[0] = m_data_to_num(x);
-  for (j = 0; j <= nd; j++) {
-    a[j + 1] = SIZET2NUM(c[j]);
-  }
-  y = rb_yield(rb_ary_new4(md, a));
-  return m_num_to_data(y);
-}
-
-static void iter_int32_map_with_index(na_loop_t* const lp) {
-  size_t i;
-  char *p1, *p2;
-  ssize_t s1, s2;
-  size_t *idx1, *idx2;
-  dtype x;
-  VALUE* a;
-  size_t* c;
-  int nd, md;
-
-  c = (size_t*)(lp->opt_ptr);
-  nd = lp->ndim;
-  if (nd > 0) {
-    nd--;
-  }
-  md = nd + 2;
-  a = ALLOCA_N(VALUE, md);
-
-  INIT_COUNTER(lp, i);
-  INIT_PTR_IDX(lp, 0, p1, s1, idx1);
-  INIT_PTR_IDX(lp, 1, p2, s2, idx2);
-
-  c[nd] = 0;
-  if (idx1) {
-    if (idx2) {
-      for (; i--;) {
-        GET_DATA_INDEX(p1, idx1, dtype, x);
-        x = yield_map_with_index(x, c, a, nd, md);
-        SET_DATA_INDEX(p2, idx2, dtype, x);
-        c[nd]++;
-      }
-    } else {
-      for (; i--;) {
-        GET_DATA_INDEX(p1, idx1, dtype, x);
-        x = yield_map_with_index(x, c, a, nd, md);
-        SET_DATA_STRIDE(p2, s2, dtype, x);
-        c[nd]++;
-      }
-    }
-  } else {
-    if (idx2) {
-      for (; i--;) {
-        GET_DATA_STRIDE(p1, s1, dtype, x);
-        x = yield_map_with_index(x, c, a, nd, md);
-        SET_DATA_INDEX(p2, idx2, dtype, x);
-        c[nd]++;
-      }
-    } else {
-      for (; i--;) {
-        GET_DATA_STRIDE(p1, s1, dtype, x);
-        x = yield_map_with_index(x, c, a, nd, md);
-        SET_DATA_STRIDE(p2, s2, dtype, x);
-        c[nd]++;
-      }
-    }
-  }
-}
-
-/*
-  Invokes the given block once for each element of self,
-  passing that element and indices along each axis as parameters.
-  Creates a new NArray containing the values returned by the block.
-  Inplace option is allowed, i.e., `nary.inplace.map` overwrites `nary`.
-  @overload map_with_index
-    For a block `{|x,i,j,...| ... }`,
-    @yieldparam [Numeric] x  an element
-    @yieldparam [Integer] i,j,...  multitimensional indices
-    @return [Numo::NArray] mapped array
-  @see #map
-  @see #each_with_index
-*/
-static VALUE int32_map_with_index(VALUE self) {
-  ndfunc_arg_in_t ain[1] = { { Qnil, 0 } };
-  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_int32_map_with_index, FULL_LOOP, 1, 1, ain, aout };
-
-  return na_ndloop_with_index(&ndf, 1, self);
-}
-
-static void iter_int32_abs(na_loop_t* const lp) {
-  size_t i;
-  char *p1, *p2;
-  ssize_t s1, s2;
-  size_t *idx1, *idx2;
-  dtype x;
-  rtype y;
-  INIT_COUNTER(lp, i);
-  INIT_PTR_IDX(lp, 0, p1, s1, idx1);
-  INIT_PTR_IDX(lp, 1, p2, s2, idx2);
-  if (idx1) {
-    if (idx2) {
-      for (; i--;) {
-        GET_DATA_INDEX(p1, idx1, dtype, x);
-        y = m_abs(x);
-        SET_DATA_INDEX(p2, idx2, rtype, y);
-      }
-    } else {
-      for (; i--;) {
-        GET_DATA_INDEX(p1, idx1, dtype, x);
-        y = m_abs(x);
-        SET_DATA_STRIDE(p2, s2, rtype, y);
-      }
-    }
-  } else {
-    if (idx2) {
-      for (; i--;) {
-        GET_DATA_STRIDE(p1, s1, dtype, x);
-        y = m_abs(x);
-        SET_DATA_INDEX(p2, idx2, rtype, y);
-      }
-    } else {
-      for (; i--;) {
-        GET_DATA_STRIDE(p1, s1, dtype, x);
-        y = m_abs(x);
-        SET_DATA_STRIDE(p2, s2, rtype, y);
-      }
-    }
-  }
-}
-
-/*
-  abs of self.
-  @overload abs
-    @return [Numo::Int32] abs of self.
-*/
-static VALUE int32_abs(VALUE self) {
-  ndfunc_arg_in_t ain[1] = { { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { cRT, 0 } };
-  ndfunc_t ndf = { iter_int32_abs, FULL_LOOP, 1, 1, ain, aout };
-
-  return na_ndloop(&ndf, 1, self);
-}
-
-#define check_intdivzero(y)                                                                    \
-  {}
-
-static void iter_int32_add(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_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;
-    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);
-  }
-  //
-}
-#undef check_intdivzero
-
-static VALUE int32_add_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_int32_add, STRIDE_LOOP, 2, 1, ain, aout };
-
-  return na_ndloop(&ndf, 2, self, other);
-}
-
-/*
-  Binary add.
-  @overload + other
-    @param [Numo::NArray,Numeric] other
-    @return [Numo::NArray] self + other
-*/
-static VALUE int32_add(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return int32_add_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_int32_sub(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_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]);
-        }
-      }
-      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_sub(((dtype*)p1)[i], *(dtype*)p2);
-            }
-          } else {
-            for (; i < n; i++) {
-              ((dtype*)p3)[i] = m_sub(((dtype*)p1)[i], *(dtype*)p2);
-            }
-          }
-        } else {
-          for (i = 0; i < n; i++) {
-            *(dtype*)p3 = m_sub(*(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_sub(*(dtype*)p1, *(dtype*)p2);
-            p1 += s1;
-            p2 += s2;
-          }
-        } else {
-          for (i = 0; i < n; i++) {
-            check_intdivzero(*(dtype*)p2);
-            *(dtype*)p3 = m_sub(*(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_sub(x, y);
-    SET_DATA_STRIDE(p3, s3, dtype, z);
-  }
-  //
-}
-#undef check_intdivzero
-
-static VALUE int32_sub_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_int32_sub, STRIDE_LOOP, 2, 1, ain, aout };
-
-  return na_ndloop(&ndf, 2, self, other);
-}
-
-/*
-  Binary sub.
-  @overload - other
-    @param [Numo::NArray,Numeric] other
-    @return [Numo::NArray] self - other
-*/
-static VALUE int32_sub(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return int32_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_int32_mul(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_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]);
-        }
-      }
-      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;
-          }
-        }
-      }
-
-      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 int32_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_int32_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 int32_mul(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return int32_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_int32_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]);
-        }
-      }
-      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_div(((dtype*)p1)[i], *(dtype*)p2);
-            }
-          } else {
-            for (; i < n; i++) {
-              ((dtype*)p3)[i] = m_div(((dtype*)p1)[i], *(dtype*)p2);
-            }
-          }
-        } else {
-          for (i = 0; i < n; i++) {
-            *(dtype*)p3 = m_div(*(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_div(*(dtype*)p1, *(dtype*)p2);
-            p1 += s1;
-            p2 += s2;
-          }
-        } else {
-          for (i = 0; i < n; i++) {
-            check_intdivzero(*(dtype*)p2);
-            *(dtype*)p3 = m_div(*(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_div(x, y);
-    SET_DATA_STRIDE(p3, s3, dtype, z);
-  }
-  //
-}
-#undef check_intdivzero
-
-static VALUE int32_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_int32_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 int32_div(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return int32_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_int32_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;
-          }
-        }
-      }
-
-      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 int32_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_int32_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 int32_mod(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return int32_mod_self(self, other);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, '%', 1, other);
-  }
-}
-
-static void iter_int32_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 int32_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_int32_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 int32_divmod(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return int32_divmod_self(self, other);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, id_divmod, 1, other);
-  }
-}
-
-static void iter_int32_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_int32_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 int32_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_int32_pow, STRIDE_LOOP, 2, 1, ain, aout };
-  ndfunc_t ndf_i = { iter_int32_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 int32_pow(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return int32_pow_self(self, other);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, id_pow, 1, other);
-  }
-}
-
-static void iter_int32_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::Int32] minus of self.
-*/
-static VALUE int32_minus(VALUE self) {
-  ndfunc_arg_in_t ain[1] = { { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_int32_minus, FULL_LOOP, 1, 1, ain, aout };
-
-  return na_ndloop(&ndf, 1, self);
-}
-
-static void iter_int32_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::Int32] reciprocal of self.
-*/
-static VALUE int32_reciprocal(VALUE self) {
-  ndfunc_arg_in_t ain[1] = { { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_int32_reciprocal, FULL_LOOP, 1, 1, ain, aout };
-
-  return na_ndloop(&ndf, 1, self);
-}
-
-static void iter_int32_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::Int32] sign of self.
-*/
-static VALUE int32_sign(VALUE self) {
-  ndfunc_arg_in_t ain[1] = { { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_int32_sign, FULL_LOOP, 1, 1, ain, aout };
-
-  return na_ndloop(&ndf, 1, self);
-}
-
-static void iter_int32_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::Int32] square of self.
-*/
-static VALUE int32_square(VALUE self) {
-  ndfunc_arg_in_t ain[1] = { { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_int32_square, FULL_LOOP, 1, 1, ain, aout };
-
-  return na_ndloop(&ndf, 1, self);
-}
-
-static void iter_int32_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 int32_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_int32_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 int32_eq(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return int32_eq_self(self, other);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, id_eq, 1, other);
-  }
-}
-
-static void iter_int32_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 int32_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_int32_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 int32_ne(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return int32_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_int32_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 int32_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_int32_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 int32_bit_and(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return int32_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_int32_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 int32_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_int32_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 int32_bit_or(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return int32_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_int32_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 int32_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_int32_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 int32_bit_xor(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return int32_bit_xor_self(self, other);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, '^', 1, other);
-  }
-}
-
-static void iter_int32_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::Int32] bit_not of self.
-*/
-static VALUE int32_bit_not(VALUE self) {
-  ndfunc_arg_in_t ain[1] = { { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_int32_bit_not, FULL_LOOP, 1, 1, ain, aout };
-
-  return na_ndloop(&ndf, 1, self);
-}
-
-#define check_intdivzero(y)                                                                    \
-  {}
-
-static void iter_int32_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 int32_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_int32_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 int32_left_shift(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return int32_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_int32_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 int32_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_int32_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 int32_right_shift(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return int32_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_int32_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 int32_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_int32_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 int32_gt(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return int32_gt_self(self, other);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, id_gt, 1, other);
-  }
-}
-
-static void iter_int32_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 int32_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_int32_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 int32_ge(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return int32_ge_self(self, other);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, id_ge, 1, other);
-  }
-}
-
-static void iter_int32_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 int32_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_int32_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 int32_lt(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return int32_lt_self(self, other);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, id_lt, 1, other);
-  }
-}
-
-static void iter_int32_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 int32_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_int32_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 int32_le(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return int32_le_self(self, other);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, id_le, 1, other);
-  }
-}
-
-// ------- Integer count without weights -------
-
-static void iter_int32_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 int32_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_int32_bincount_32, NO_LOOP | NDF_STRIDE_LOOP | NDF_INDEX_LOOP, 1, 1, ain, aout
-  };
-
-  return na_ndloop(&ndf, 1, self);
-}
-
-static void iter_int32_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;
+  nd = lp->ndim;
+  if (nd > 0) {
+    nd--;
   }
+  md = nd + 2;
+  a = ALLOCA_N(VALUE, md);
 
+  INIT_COUNTER(lp, i);
+  INIT_PTR_IDX(lp, 0, p1, s1, idx1);
+  c[nd] = 0;
   if (idx1) {
     for (; i--;) {
       GET_DATA_INDEX(p1, idx1, dtype, x);
-      (*(u_int64_t*)(p2 + s2 * x))++;
+      yield_each_with_index(x, c, a, nd, md);
+      c[nd]++;
     }
   } else {
     for (; i--;) {
       GET_DATA_STRIDE(p1, s1, dtype, x);
-      (*(u_int64_t*)(p2 + s2 * x))++;
+      yield_each_with_index(x, c, a, nd, md);
+      c[nd]++;
     }
   }
 }
 
-static VALUE int32_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_int32_bincount_64, NO_LOOP | NDF_STRIDE_LOOP | NDF_INDEX_LOOP, 1, 1, ain, aout
-  };
+/*
+  Invokes the given block once for each element of self,
+  passing that element and indices along each axis as parameters.
+  @overload each_with_index
+    For a block `{|x,i,j,...| ... }`,
+    @yieldparam [Numeric] x  an element
+    @yieldparam [Integer] i,j,...  multitimensional indices
+    @return [Numo::NArray] self
+  @see #each
+  @see #map_with_index
+*/
+static VALUE int32_each_with_index(VALUE self) {
+  ndfunc_arg_in_t ain[1] = { { Qnil, 0 } };
+  ndfunc_t ndf = { iter_int32_each_with_index, FULL_LOOP_NIP, 1, 0, ain, 0 };
 
-  return na_ndloop(&ndf, 1, self);
+  na_ndloop_with_index(&ndf, 1, self);
+  return self;
 }
-// ------- end of Integer count without weights -------
-
-// ------- Float count with weights -------
-
-static void iter_int32_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");
-  }
+static inline dtype yield_map_with_index(dtype x, size_t* c, VALUE* a, int nd, int md) {
+  int j;
+  VALUE y;
 
-  // 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;
+  a[0] = m_data_to_num(x);
+  for (j = 0; j <= nd; j++) {
+    a[j + 1] = SIZET2NUM(c[j]);
   }
+  y = rb_yield(rb_ary_new4(md, a));
+  return m_num_to_data(y);
 }
 
-static VALUE int32_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_int32_bincount_sf, NO_LOOP | NDF_STRIDE_LOOP, 2, 1, ain, aout };
-
-  return na_ndloop(&ndf, 2, self, weight);
-}
-
-static void iter_int32_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");
-  }
+static void iter_int32_map_with_index(na_loop_t* const lp) {
+  size_t i;
+  char *p1, *p2;
+  ssize_t s1, s2;
+  size_t *idx1, *idx2;
+  dtype x;
+  VALUE* a;
+  size_t* c;
+  int nd, md;
 
-  // 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;
+  c = (size_t*)(lp->opt_ptr);
+  nd = lp->ndim;
+  if (nd > 0) {
+    nd--;
   }
-}
+  md = nd + 2;
+  a = ALLOCA_N(VALUE, md);
 
-static VALUE int32_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_int32_bincount_df, NO_LOOP | NDF_STRIDE_LOOP, 2, 1, ain, aout };
+  INIT_COUNTER(lp, i);
+  INIT_PTR_IDX(lp, 0, p1, s1, idx1);
+  INIT_PTR_IDX(lp, 1, p2, s2, idx2);
 
-  return na_ndloop(&ndf, 2, self, weight);
+  c[nd] = 0;
+  if (idx1) {
+    if (idx2) {
+      for (; i--;) {
+        GET_DATA_INDEX(p1, idx1, dtype, x);
+        x = yield_map_with_index(x, c, a, nd, md);
+        SET_DATA_INDEX(p2, idx2, dtype, x);
+        c[nd]++;
+      }
+    } else {
+      for (; i--;) {
+        GET_DATA_INDEX(p1, idx1, dtype, x);
+        x = yield_map_with_index(x, c, a, nd, md);
+        SET_DATA_STRIDE(p2, s2, dtype, x);
+        c[nd]++;
+      }
+    }
+  } else {
+    if (idx2) {
+      for (; i--;) {
+        GET_DATA_STRIDE(p1, s1, dtype, x);
+        x = yield_map_with_index(x, c, a, nd, md);
+        SET_DATA_INDEX(p2, idx2, dtype, x);
+        c[nd]++;
+      }
+    } else {
+      for (; i--;) {
+        GET_DATA_STRIDE(p1, s1, dtype, x);
+        x = yield_map_with_index(x, c, a, nd, md);
+        SET_DATA_STRIDE(p2, s2, dtype, x);
+        c[nd]++;
+      }
+    }
+  }
 }
-// ------- 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]
-
+  Invokes the given block once for each element of self,
+  passing that element and indices along each axis as parameters.
+  Creates a new NArray containing the values returned by the block.
+  Inplace option is allowed, i.e., `nary.inplace.map` overwrites `nary`.
+  @overload map_with_index
+    For a block `{|x,i,j,...| ... }`,
+    @yieldparam [Numeric] x  an element
+    @yieldparam [Integer] i,j,...  multitimensional indices
+    @return [Numo::NArray] mapped array
+  @see #map
+  @see #each_with_index
 */
-static VALUE int32_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 = int32_minmax(0, 0, self);
-  if (m_num_to_data(RARRAY_AREF(v, 0)) < 0) {
-    rb_raise(rb_eArgError, "array items must be non-netagive");
-  }
-  v = RARRAY_AREF(v, 1);
-
-  length = NUM2SIZET(v) + 1;
+static VALUE int32_map_with_index(VALUE self) {
+  ndfunc_arg_in_t ain[1] = { { Qnil, 0 } };
+  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
+  ndfunc_t ndf = { iter_int32_map_with_index, FULL_LOOP, 1, 1, ain, aout };
 
-  if (opts[0] != Qundef) {
-    minlength = NUM2SIZET(opts[0]);
-    if (minlength > length) {
-      length = minlength;
-    }
-  }
+  return na_ndloop_with_index(&ndf, 1, self);
+}
 
-  if (NIL_P(weight)) {
-    if (length > 4294967295ul) {
-      return int32_bincount_64(self, length);
+static void iter_int32_abs(na_loop_t* const lp) {
+  size_t i;
+  char *p1, *p2;
+  ssize_t s1, s2;
+  size_t *idx1, *idx2;
+  dtype x;
+  rtype y;
+  INIT_COUNTER(lp, i);
+  INIT_PTR_IDX(lp, 0, p1, s1, idx1);
+  INIT_PTR_IDX(lp, 1, p2, s2, idx2);
+  if (idx1) {
+    if (idx2) {
+      for (; i--;) {
+        GET_DATA_INDEX(p1, idx1, dtype, x);
+        y = m_abs(x);
+        SET_DATA_INDEX(p2, idx2, rtype, y);
+      }
     } else {
-      return int32_bincount_32(self, length);
+      for (; i--;) {
+        GET_DATA_INDEX(p1, idx1, dtype, x);
+        y = m_abs(x);
+        SET_DATA_STRIDE(p2, s2, rtype, y);
+      }
     }
   } else {
-    wclass = rb_obj_class(weight);
-    if (wclass == numo_cSFloat) {
-      return int32_bincount_sf(self, weight, length);
+    if (idx2) {
+      for (; i--;) {
+        GET_DATA_STRIDE(p1, s1, dtype, x);
+        y = m_abs(x);
+        SET_DATA_INDEX(p2, idx2, rtype, y);
+      }
     } else {
-      return int32_bincount_df(self, weight, length);
+      for (; i--;) {
+        GET_DATA_STRIDE(p1, s1, dtype, x);
+        y = m_abs(x);
+        SET_DATA_STRIDE(p2, s2, rtype, y);
+      }
     }
   }
 }
 
+/*
+  abs of self.
+  @overload abs
+    @return [Numo::Int32] abs of self.
+*/
+static VALUE int32_abs(VALUE self) {
+  ndfunc_arg_in_t ain[1] = { { cT, 0 } };
+  ndfunc_arg_out_t aout[1] = { { cRT, 0 } };
+  ndfunc_t ndf = { iter_int32_abs, FULL_LOOP, 1, 1, ain, aout };
+
+  return na_ndloop(&ndf, 1, self);
+}
+
 static void iter_int32_poly(na_loop_t* const lp) {
   size_t i;
   dtype x, y, a;
@@ -4593,50 +2262,216 @@ void Init_numo_int32(void) {
   rb_define_singleton_method(cT, "cast", int32_s_cast, 1);
   rb_define_method(cT, "[]", int32_aref, -1);
   rb_define_method(cT, "[]=", int32_aset, -1);
+  /**
+   * return NArray with cast to the type of self.
+   * @overload coerce_cast(type)
+   *   @return [nil]
+   */
   rb_define_method(cT, "coerce_cast", int32_coerce_cast, 1);
+  /**
+   * Convert self to Array.
+   * @overload to_a
+   *   @return [Array]
+   */
   rb_define_method(cT, "to_a", int32_to_a, 0);
+  /**
+   * Fill elements with other.
+   * @overload fill other
+   *   @param [Numeric] other
+   *   @return [Numo::Int32] self.
+   */
   rb_define_method(cT, "fill", int32_fill, 1);
+  /**
+   * Format elements into strings.
+   * @overload format format
+   *   @param [String] format
+   *   @return [Numo::RObject] array of formatted strings.
+   */
   rb_define_method(cT, "format", int32_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", int32_format_to_a, -1);
+  /**
+   * Returns a string containing a human-readable representation of NArray.
+   * @overload inspect
+   *   @return [String]
+   */
   rb_define_method(cT, "inspect", int32_inspect, 0);
   rb_define_method(cT, "each", int32_each, 0);
   rb_define_method(cT, "map", int32_map, 0);
   rb_define_method(cT, "each_with_index", int32_each_with_index, 0);
   rb_define_method(cT, "map_with_index", int32_map_with_index, 0);
   rb_define_method(cT, "abs", int32_abs, 0);
+  /**
+   * Binary add.
+   * @overload + other
+   *   @param [Numo::NArray,Numeric] other
+   *   @return [Numo::NArray] self + other
+   */
   rb_define_method(cT, "+", int32_add, 1);
+  /**
+   * Binary sub.
+   * @overload - other
+   *   @param [Numo::NArray,Numeric] other
+   *   @return [Numo::NArray] self - other
+   */
   rb_define_method(cT, "-", int32_sub, 1);
+  /**
+   * Binary mul.
+   * @overload * other
+   *   @param [Numo::NArray,Numeric] other
+   *   @return [Numo::NArray] self * other
+   */
   rb_define_method(cT, "*", int32_mul, 1);
+  /**
+   * Binary div.
+   * @overload / other
+   *   @param [Numo::NArray,Numeric] other
+   *   @return [Numo::NArray] self / other
+   */
   rb_define_method(cT, "/", int32_div, 1);
+  /**
+   * Binary mod.
+   * @overload % other
+   *   @param [Numo::NArray,Numeric] other
+   *   @return [Numo::NArray] self % other
+   */
   rb_define_method(cT, "%", int32_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", int32_divmod, 1);
+  /**
+   * Binary power.
+   * @overload ** other
+   *   @param [Numo::NArray,Numeric] other
+   *   @return [Numo::NArray] self to the other-th power.
+   */
   rb_define_method(cT, "**", int32_pow, 1);
   rb_define_alias(cT, "pow", "**");
+  /**
+   * Unary minus.
+   * @overload -@
+   *   @return [Numo::Int32] minus of self.
+   */
   rb_define_method(cT, "-@", int32_minus, 0);
+  /**
+   * Unary reciprocal.
+   * @overload reciprocal
+   *   @return [Numo::Int32] reciprocal of self.
+   */
   rb_define_method(cT, "reciprocal", int32_reciprocal, 0);
+  /**
+   * Unary sign.
+   * @overload sign
+   *   @return [Numo::Int32] sign of self.
+   */
   rb_define_method(cT, "sign", int32_sign, 0);
+  /**
+   * Unary square.
+   * @overload square
+   *   @return [Numo::Int32] square of self.
+   */
   rb_define_method(cT, "square", int32_square, 0);
   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", int32_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", int32_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, "&", int32_bit_and, 1);
+  /**
+   * Binary bit_or.
+   * @overload | other
+   *   @param [Numo::NArray,Numeric] other
+   *   @return [Numo::NArray] self | other
+   */
   rb_define_method(cT, "|", int32_bit_or, 1);
+  /**
+   * Binary bit_xor.
+   * @overload ^ other
+   *   @param [Numo::NArray,Numeric] other
+   *   @return [Numo::NArray] self ^ other
+   */
   rb_define_method(cT, "^", int32_bit_xor, 1);
+  /**
+   * Unary bit_not.
+   * @overload ~
+   *   @return [Numo::Int32] bit_not of self.
+   */
   rb_define_method(cT, "~", int32_bit_not, 0);
+  /**
+   * Binary left_shift.
+   * @overload << other
+   *   @param [Numo::NArray,Numeric] other
+   *   @return [Numo::NArray] self << other
+   */
   rb_define_method(cT, "<<", int32_left_shift, 1);
+  /**
+   * Binary right_shift.
+   * @overload >> other
+   *   @param [Numo::NArray,Numeric] other
+   *   @return [Numo::NArray] self >> other
+   */
   rb_define_method(cT, ">>", int32_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", int32_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", int32_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", int32_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", int32_le, 1);
   rb_define_alias(cT, ">", "gt");
   rb_define_alias(cT, ">=", "ge");
@@ -4821,6 +2656,37 @@ void Init_numo_int32(void) {
    *   @return [Numo::Int32]
    */
   rb_define_module_function(cT, "minimum", int32_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", int32_bincount, -1);
   /**
    * cumsum of self.