numo-narray-alt 0.9.11 → 0.9.12

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

@@ -42,7 +42,36 @@ 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/round/floor.h"
+#include "mh/round/round.h"
+#include "mh/round/ceil.h"
+#include "mh/round/trunc.h"
+#include "mh/round/rint.h"
+#include "mh/comp/eq.h"
+#include "mh/comp/ne.h"
+#include "mh/comp/nearly_eq.h"
+#include "mh/comp/gt.h"
+#include "mh/comp/ge.h"
+#include "mh/comp/lt.h"
+#include "mh/comp/le.h"
 #include "mh/clip.h"
+#include "mh/isnan.h"
+#include "mh/isinf.h"
+#include "mh/isposinf.h"
+#include "mh/isneginf.h"
+#include "mh/isfinite.h"
 #include "mh/sum.h"
 #include "mh/prod.h"
 #include "mh/mean.h"
@@ -98,7 +127,43 @@ extern VALUE cRT;
 
 typedef float sfloat; // Type aliases for shorter notation
                       // following the codebase naming convention.
+DEF_NARRAY_COERCE_CAST_METHOD_FUNC(sfloat)
+DEF_NARRAY_TO_A_METHOD_FUNC(sfloat)
+DEF_NARRAY_FILL_METHOD_FUNC(sfloat)
+DEF_NARRAY_FORMAT_METHOD_FUNC(sfloat)
+DEF_NARRAY_FORMAT_TO_A_METHOD_FUNC(sfloat)
+DEF_NARRAY_INSPECT_METHOD_FUNC(sfloat)
+#ifdef __SSE2__
+DEF_NARRAY_SFLT_ADD_SSE2_METHOD_FUNC()
+DEF_NARRAY_SFLT_SUB_SSE2_METHOD_FUNC()
+DEF_NARRAY_SFLT_MUL_SSE2_METHOD_FUNC()
+DEF_NARRAY_SFLT_DIV_SSE2_METHOD_FUNC()
+#else
+DEF_NARRAY_ADD_METHOD_FUNC(sfloat, numo_cSFloat)
+DEF_NARRAY_SUB_METHOD_FUNC(sfloat, numo_cSFloat)
+DEF_NARRAY_MUL_METHOD_FUNC(sfloat, numo_cSFloat)
+DEF_NARRAY_FLT_DIV_METHOD_FUNC(sfloat, numo_cSFloat)
+#endif
+DEF_NARRAY_FLT_MOD_METHOD_FUNC(sfloat, numo_cSFloat)
+DEF_NARRAY_FLT_DIVMOD_METHOD_FUNC(sfloat, numo_cSFloat)
+DEF_NARRAY_FLT_FLOOR_METHOD_FUNC(sfloat, numo_cSFloat)
+DEF_NARRAY_FLT_ROUND_METHOD_FUNC(sfloat, numo_cSFloat)
+DEF_NARRAY_FLT_CEIL_METHOD_FUNC(sfloat, numo_cSFloat)
+DEF_NARRAY_FLT_TRUNC_METHOD_FUNC(sfloat, numo_cSFloat)
+DEF_NARRAY_FLT_RINT_METHOD_FUNC(sfloat, numo_cSFloat)
+DEF_NARRAY_EQ_METHOD_FUNC(sfloat, numo_cSFloat)
+DEF_NARRAY_NE_METHOD_FUNC(sfloat, numo_cSFloat)
+DEF_NARRAY_NEARLY_EQ_METHOD_FUNC(sfloat, numo_cSFloat)
+DEF_NARRAY_GT_METHOD_FUNC(sfloat, numo_cSFloat)
+DEF_NARRAY_GE_METHOD_FUNC(sfloat, numo_cSFloat)
+DEF_NARRAY_LT_METHOD_FUNC(sfloat, numo_cSFloat)
+DEF_NARRAY_LE_METHOD_FUNC(sfloat, numo_cSFloat)
 DEF_NARRAY_CLIP_METHOD_FUNC(sfloat, numo_cSFloat)
+DEF_NARRAY_FLT_ISNAN_METHOD_FUNC(sfloat, numo_cSFloat)
+DEF_NARRAY_FLT_ISINF_METHOD_FUNC(sfloat, numo_cSFloat)
+DEF_NARRAY_FLT_ISPOSINF_METHOD_FUNC(sfloat, numo_cSFloat)
+DEF_NARRAY_FLT_ISNEGINF_METHOD_FUNC(sfloat, numo_cSFloat)
+DEF_NARRAY_FLT_ISFINITE_METHOD_FUNC(sfloat, numo_cSFloat)
 DEF_NARRAY_FLT_SUM_METHOD_FUNC(sfloat, numo_cSFloat)
 DEF_NARRAY_FLT_PROD_METHOD_FUNC(sfloat, numo_cSFloat)
 DEF_NARRAY_FLT_MEAN_METHOD_FUNC(sfloat, numo_cSFloat, float, numo_cSFloat)
@@ -1275,171 +1340,6 @@ static VALUE sfloat_aset(int argc, VALUE* argv, VALUE self) {
   return argv[argc];
 }
 
-static VALUE sfloat_coerce_cast(VALUE self, VALUE type) {
-  return Qnil;
-}
-
-static void iter_sfloat_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);
-    }
-  }
-}
-
-static VALUE sfloat_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_sfloat_to_a, FULL_LOOP_NIP, 3, 1, ain, aout };
-  return na_ndloop_cast_narray_to_rarray(&ndf, self, Qnil);
-}
-
-static void iter_sfloat_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);
-    }
-  }
-}
-
-static VALUE sfloat_fill(VALUE self, VALUE val) {
-  ndfunc_arg_in_t ain[2] = { { OVERWRITE, 0 }, { sym_option } };
-  ndfunc_t ndf = { iter_sfloat_fill, FULL_LOOP, 2, 0, ain, 0 };
-
-  na_ndloop(&ndf, 2, self, val);
-  return self;
-}
-
-static VALUE format_sfloat(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_sfloat_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_sfloat(fmt, x);
-      SET_DATA_STRIDE(p2, s2, VALUE, y);
-    }
-  } else {
-    for (; i--;) {
-      x = (dtype*)p1;
-      p1 += s1;
-      y = format_sfloat(fmt, x);
-      SET_DATA_STRIDE(p2, s2, VALUE, y);
-    }
-  }
-}
-
-static VALUE sfloat_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_sfloat_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_sfloat_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_sfloat(fmt, x);
-      rb_ary_push(a, y);
-    }
-  } else {
-    for (; i--;) {
-      x = (dtype*)p1;
-      p1 += s1;
-      y = format_sfloat(fmt, x);
-      rb_ary_push(a, y);
-    }
-  }
-}
-
-static VALUE sfloat_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_sfloat_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_sfloat_inspect(char* ptr, size_t pos, VALUE fmt) {
-  return format_sfloat(fmt, (dtype*)(ptr + pos));
-}
-
-static VALUE sfloat_inspect(VALUE ary) {
-  return na_ndloop_inspect(ary, iter_sfloat_inspect, Qnil);
-}
-
 static void iter_sfloat_each(na_loop_t* const lp) {
   size_t i, s1;
   char* p1;
@@ -1722,2354 +1622,461 @@ static VALUE sfloat_abs(VALUE self) {
   return na_ndloop(&ndf, 1, self);
 }
 
-#define check_intdivzero(y)                                                                    \
-  {}
-
-static void iter_sfloat_add(na_loop_t* const lp) {
-  size_t i = 0;
-  size_t n;
+static void iter_sfloat_pow(na_loop_t* const lp) {
+  size_t i;
   char *p1, *p2, *p3;
   ssize_t s1, s2, s3;
-
-#ifdef __SSE2__
-  size_t cnt;
-  size_t cnt_simd_loop = -1;
-
-  __m128 a;
-  __m128 b;
-
-  size_t num_pack; // Number of elements packed for SIMD.
-  num_pack = SIMD_ALIGNMENT_SIZE / sizeof(dtype);
-#endif
-  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)) {
-#ifdef __SSE2__
-      // Check number of elements. & Check same alignment.
-      if ((n >= num_pack) &&
-          is_same_aligned3(
-            &((dtype*)p1)[i], &((dtype*)p2)[i], &((dtype*)p3)[i], SIMD_ALIGNMENT_SIZE
-          )) {
-        // Calculate up to the position just before the start of SIMD computation.
-        cnt = get_count_of_elements_not_aligned_to_simd_size(
-          &((dtype*)p1)[i], SIMD_ALIGNMENT_SIZE, sizeof(dtype)
-        );
-#endif
-        if (p1 == p3) { // inplace case
-#ifdef __SSE2__
-          for (; i < cnt; i++) {
-#else
-        for (; i < n; i++) {
-          check_intdivzero(((dtype*)p2)[i]);
-#endif
-            ((dtype*)p1)[i] = m_add(((dtype*)p1)[i], ((dtype*)p2)[i]);
-          }
-        } else {
-#ifdef __SSE2__
-          for (; i < cnt; i++) {
-#else
-        for (; i < n; i++) {
-          check_intdivzero(((dtype*)p2)[i]);
-#endif
-            ((dtype*)p3)[i] = m_add(((dtype*)p1)[i], ((dtype*)p2)[i]);
-          }
-        }
-
-#ifdef __SSE2__
-        // Get the count of SIMD computation loops.
-        cnt_simd_loop = (n - i) % num_pack;
-
-        // SIMD computation.
-        if (p1 == p3) { // inplace case
-          for (; i < n - cnt_simd_loop; i += num_pack) {
-            a = _mm_load_ps(&((dtype*)p1)[i]);
-            b = _mm_load_ps(&((dtype*)p2)[i]);
-            a = _mm_add_ps(a, b);
-            _mm_store_ps(&((dtype*)p1)[i], a);
-          }
-        } else {
-          for (; i < n - cnt_simd_loop; i += num_pack) {
-            a = _mm_load_ps(&((dtype*)p1)[i]);
-            b = _mm_load_ps(&((dtype*)p2)[i]);
-            a = _mm_add_ps(a, b);
-            _mm_stream_ps(&((dtype*)p3)[i], a);
-          }
-        }
-      }
-
-      // Compute the remainder of the SIMD operation.
-      if (cnt_simd_loop != 0) {
-        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]);
-          }
-        }
-      }
-#endif
-      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)) {
-#ifdef __SSE2__
-          // Broadcast a scalar value and use it for SIMD computation.
-          b = _mm_load1_ps(&((dtype*)p2)[0]);
-
-          // Check number of elements. & Check same alignment.
-          if ((n >= num_pack) &&
-              is_same_aligned2(&((dtype*)p1)[i], &((dtype*)p3)[i], SIMD_ALIGNMENT_SIZE)) {
-            // Calculate up to the position just before the start of SIMD computation.
-            cnt = get_count_of_elements_not_aligned_to_simd_size(
-              &((dtype*)p1)[i], SIMD_ALIGNMENT_SIZE, sizeof(dtype)
-            );
-#endif
-            if (p1 == p3) { // inplace case
-#ifdef __SSE2__
-              for (; i < cnt; i++) {
-#else
-            for (; i < n; i++) {
-#endif
-                ((dtype*)p1)[i] = m_add(((dtype*)p1)[i], *(dtype*)p2);
-              }
-            } else {
-#ifdef __SSE2__
-              for (; i < cnt; i++) {
-#else
-            for (; i < n; i++) {
-#endif
-                ((dtype*)p3)[i] = m_add(((dtype*)p1)[i], *(dtype*)p2);
-              }
-            }
-
-#ifdef __SSE2__
-            // Get the count of SIMD computation loops.
-            cnt_simd_loop = (n - i) % num_pack;
-
-            // SIMD computation.
-            if (p1 == p3) { // inplace case
-              for (; i < n - cnt_simd_loop; i += num_pack) {
-                a = _mm_load_ps(&((dtype*)p1)[i]);
-                a = _mm_add_ps(a, b);
-                _mm_store_ps(&((dtype*)p1)[i], a);
-              }
-            } else {
-              for (; i < n - cnt_simd_loop; i += num_pack) {
-                a = _mm_load_ps(&((dtype*)p1)[i]);
-                a = _mm_add_ps(a, b);
-                _mm_stream_ps(&((dtype*)p3)[i], a);
-              }
-            }
-          }
-
-          // Compute the remainder of the SIMD operation.
-          if (cnt_simd_loop != 0) {
-            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);
-              }
-            }
-          }
-#endif
-        } 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_sfloat_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 sfloat_add_self(VALUE self, VALUE other) {
+static VALUE sfloat_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_sfloat_add, STRIDE_LOOP, 2, 1, ain, aout };
+  ndfunc_t ndf = { iter_sfloat_pow, STRIDE_LOOP, 2, 1, ain, aout };
+  ndfunc_t ndf_i = { iter_sfloat_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);
+  }
 }
 
-static VALUE sfloat_add(VALUE self, VALUE other) {
+static VALUE sfloat_pow(VALUE self, VALUE other) {
 
   VALUE klass, v;
-
   klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
   if (klass == cT) {
-    return sfloat_add_self(self, other);
+    return sfloat_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_sfloat_sub(na_loop_t* const lp) {
-  size_t i = 0;
-  size_t n;
-  char *p1, *p2, *p3;
-  ssize_t s1, s2, s3;
-
-#ifdef __SSE2__
-  size_t cnt;
-  size_t cnt_simd_loop = -1;
-
-  __m128 a;
-  __m128 b;
+static void iter_sfloat_minus(na_loop_t* const lp) {
+  size_t i, n;
+  char *p1, *p2;
+  ssize_t s1, s2;
+  size_t *idx1, *idx2;
+  dtype x;
 
-  size_t num_pack; // Number of elements packed for SIMD.
-  num_pack = SIMD_ALIGNMENT_SIZE / sizeof(dtype);
-#endif
   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)) {
-#ifdef __SSE2__
-      // Check number of elements. & Check same alignment.
-      if ((n >= num_pack) &&
-          is_same_aligned3(
-            &((dtype*)p1)[i], &((dtype*)p2)[i], &((dtype*)p3)[i], SIMD_ALIGNMENT_SIZE
-          )) {
-        // Calculate up to the position just before the start of SIMD computation.
-        cnt = get_count_of_elements_not_aligned_to_simd_size(
-          &((dtype*)p1)[i], SIMD_ALIGNMENT_SIZE, sizeof(dtype)
-        );
-#endif
-        if (p1 == p3) { // inplace case
-#ifdef __SSE2__
-          for (; i < cnt; i++) {
-#else
-        for (; i < n; i++) {
-          check_intdivzero(((dtype*)p2)[i]);
-#endif
-            ((dtype*)p1)[i] = m_sub(((dtype*)p1)[i], ((dtype*)p2)[i]);
-          }
-        } else {
-#ifdef __SSE2__
-          for (; i < cnt; i++) {
-#else
-        for (; i < n; i++) {
-          check_intdivzero(((dtype*)p2)[i]);
-#endif
-            ((dtype*)p3)[i] = m_sub(((dtype*)p1)[i], ((dtype*)p2)[i]);
-          }
-        }
-
-#ifdef __SSE2__
-        // Get the count of SIMD computation loops.
-        cnt_simd_loop = (n - i) % num_pack;
+  INIT_PTR_IDX(lp, 0, p1, s1, idx1);
+  INIT_PTR_IDX(lp, 1, p2, s2, idx2);
 
-        // SIMD computation.
-        if (p1 == p3) { // inplace case
-          for (; i < n - cnt_simd_loop; i += num_pack) {
-            a = _mm_load_ps(&((dtype*)p1)[i]);
-            b = _mm_load_ps(&((dtype*)p2)[i]);
-            a = _mm_sub_ps(a, b);
-            _mm_store_ps(&((dtype*)p1)[i], a);
-          }
-        } else {
-          for (; i < n - cnt_simd_loop; i += num_pack) {
-            a = _mm_load_ps(&((dtype*)p1)[i]);
-            b = _mm_load_ps(&((dtype*)p2)[i]);
-            a = _mm_sub_ps(a, b);
-            _mm_stream_ps(&((dtype*)p3)[i], a);
-          }
-        }
+  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);
       }
-
-      // Compute the remainder of the SIMD operation.
-      if (cnt_simd_loop != 0) {
-        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]);
-          }
-        }
+    } 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);
       }
-#endif
-      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)) {
-#ifdef __SSE2__
-          // Broadcast a scalar value and use it for SIMD computation.
-          b = _mm_load1_ps(&((dtype*)p2)[0]);
-
-          // Check number of elements. & Check same alignment.
-          if ((n >= num_pack) &&
-              is_same_aligned2(&((dtype*)p1)[i], &((dtype*)p3)[i], SIMD_ALIGNMENT_SIZE)) {
-            // Calculate up to the position just before the start of SIMD computation.
-            cnt = get_count_of_elements_not_aligned_to_simd_size(
-              &((dtype*)p1)[i], SIMD_ALIGNMENT_SIZE, sizeof(dtype)
-            );
-#endif
-            if (p1 == p3) { // inplace case
-#ifdef __SSE2__
-              for (; i < cnt; i++) {
-#else
-            for (; i < n; i++) {
-#endif
-                ((dtype*)p1)[i] = m_sub(((dtype*)p1)[i], *(dtype*)p2);
-              }
-            } else {
-#ifdef __SSE2__
-              for (; i < cnt; i++) {
-#else
-            for (; i < n; i++) {
-#endif
-                ((dtype*)p3)[i] = m_sub(((dtype*)p1)[i], *(dtype*)p2);
-              }
-            }
-
-#ifdef __SSE2__
-            // Get the count of SIMD computation loops.
-            cnt_simd_loop = (n - i) % num_pack;
-
-            // SIMD computation.
-            if (p1 == p3) { // inplace case
-              for (; i < n - cnt_simd_loop; i += num_pack) {
-                a = _mm_load_ps(&((dtype*)p1)[i]);
-                a = _mm_sub_ps(a, b);
-                _mm_store_ps(&((dtype*)p1)[i], a);
-              }
-            } else {
-              for (; i < n - cnt_simd_loop; i += num_pack) {
-                a = _mm_load_ps(&((dtype*)p1)[i]);
-                a = _mm_sub_ps(a, b);
-                _mm_stream_ps(&((dtype*)p3)[i], a);
-              }
-            }
-          }
-
-          // Compute the remainder of the SIMD operation.
-          if (cnt_simd_loop != 0) {
-            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);
-              }
-            }
-          }
-#endif
-        } 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 sfloat_sub_self(VALUE self, VALUE other) {
-  ndfunc_arg_in_t ain[2] = { { cT, 0 }, { cT, 0 } };
+static VALUE sfloat_minus(VALUE self) {
+  ndfunc_arg_in_t ain[1] = { { cT, 0 } };
   ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_sfloat_sub, STRIDE_LOOP, 2, 1, ain, aout };
-
-  return na_ndloop(&ndf, 2, self, other);
-}
-
-static VALUE sfloat_sub(VALUE self, VALUE other) {
-
-  VALUE klass, v;
+  ndfunc_t ndf = { iter_sfloat_minus, FULL_LOOP, 1, 1, ain, aout };
 
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return sfloat_sub_self(self, other);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, '-', 1, other);
-  }
+  return na_ndloop(&ndf, 1, self);
 }
 
-#define check_intdivzero(y)                                                                    \
-  {}
-
-static void iter_sfloat_mul(na_loop_t* const lp) {
-  size_t i = 0;
-  size_t n;
-  char *p1, *p2, *p3;
-  ssize_t s1, s2, s3;
-
-#ifdef __SSE2__
-  size_t cnt;
-  size_t cnt_simd_loop = -1;
-
-  __m128 a;
-  __m128 b;
+static void iter_sfloat_reciprocal(na_loop_t* const lp) {
+  size_t i, n;
+  char *p1, *p2;
+  ssize_t s1, s2;
+  size_t *idx1, *idx2;
+  dtype x;
 
-  size_t num_pack; // Number of elements packed for SIMD.
-  num_pack = SIMD_ALIGNMENT_SIZE / sizeof(dtype);
-#endif
   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)) {
-#ifdef __SSE2__
-      // Check number of elements. & Check same alignment.
-      if ((n >= num_pack) &&
-          is_same_aligned3(
-            &((dtype*)p1)[i], &((dtype*)p2)[i], &((dtype*)p3)[i], SIMD_ALIGNMENT_SIZE
-          )) {
-        // Calculate up to the position just before the start of SIMD computation.
-        cnt = get_count_of_elements_not_aligned_to_simd_size(
-          &((dtype*)p1)[i], SIMD_ALIGNMENT_SIZE, sizeof(dtype)
-        );
-#endif
-        if (p1 == p3) { // inplace case
-#ifdef __SSE2__
-          for (; i < cnt; i++) {
-#else
-        for (; i < n; i++) {
-          check_intdivzero(((dtype*)p2)[i]);
-#endif
-            ((dtype*)p1)[i] = m_mul(((dtype*)p1)[i], ((dtype*)p2)[i]);
-          }
-        } else {
-#ifdef __SSE2__
-          for (; i < cnt; i++) {
-#else
-        for (; i < n; i++) {
-          check_intdivzero(((dtype*)p2)[i]);
-#endif
-            ((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);
+      }
+    } 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;
         }
-
-#ifdef __SSE2__
-        // Get the count of SIMD computation loops.
-        cnt_simd_loop = (n - i) % num_pack;
-
-        // SIMD computation.
-        if (p1 == p3) { // inplace case
-          for (; i < n - cnt_simd_loop; i += num_pack) {
-            a = _mm_load_ps(&((dtype*)p1)[i]);
-            b = _mm_load_ps(&((dtype*)p2)[i]);
-            a = _mm_mul_ps(a, b);
-            _mm_store_ps(&((dtype*)p1)[i], a);
-          }
-        } else {
-          for (; i < n - cnt_simd_loop; i += num_pack) {
-            a = _mm_load_ps(&((dtype*)p1)[i]);
-            b = _mm_load_ps(&((dtype*)p2)[i]);
-            a = _mm_mul_ps(a, b);
-            _mm_stream_ps(&((dtype*)p3)[i], a);
+        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;
+          //
         }
       }
-
-      // Compute the remainder of the SIMD operation.
-      if (cnt_simd_loop != 0) {
-        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]);
-          }
-        }
+      for (i = 0; i < n; i++) {
+        GET_DATA_STRIDE(p1, s1, dtype, x);
+        x = m_reciprocal(x);
+        SET_DATA_STRIDE(p2, s2, dtype, x);
       }
-#endif
-      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)) {
-#ifdef __SSE2__
-          // Broadcast a scalar value and use it for SIMD computation.
-          b = _mm_load1_ps(&((dtype*)p2)[0]);
-
-          // Check number of elements. & Check same alignment.
-          if ((n >= num_pack) &&
-              is_same_aligned2(&((dtype*)p1)[i], &((dtype*)p3)[i], SIMD_ALIGNMENT_SIZE)) {
-            // Calculate up to the position just before the start of SIMD computation.
-            cnt = get_count_of_elements_not_aligned_to_simd_size(
-              &((dtype*)p1)[i], SIMD_ALIGNMENT_SIZE, sizeof(dtype)
-            );
-#endif
-            if (p1 == p3) { // inplace case
-#ifdef __SSE2__
-              for (; i < cnt; i++) {
-#else
-            for (; i < n; i++) {
-#endif
-                ((dtype*)p1)[i] = m_mul(((dtype*)p1)[i], *(dtype*)p2);
-              }
-            } else {
-#ifdef __SSE2__
-              for (; i < cnt; i++) {
-#else
-            for (; i < n; i++) {
-#endif
-                ((dtype*)p3)[i] = m_mul(((dtype*)p1)[i], *(dtype*)p2);
-              }
-            }
-
-#ifdef __SSE2__
-            // Get the count of SIMD computation loops.
-            cnt_simd_loop = (n - i) % num_pack;
-
-            // SIMD computation.
-            if (p1 == p3) { // inplace case
-              for (; i < n - cnt_simd_loop; i += num_pack) {
-                a = _mm_load_ps(&((dtype*)p1)[i]);
-                a = _mm_mul_ps(a, b);
-                _mm_store_ps(&((dtype*)p1)[i], a);
-              }
-            } else {
-              for (; i < n - cnt_simd_loop; i += num_pack) {
-                a = _mm_load_ps(&((dtype*)p1)[i]);
-                a = _mm_mul_ps(a, b);
-                _mm_stream_ps(&((dtype*)p3)[i], a);
-              }
-            }
-          }
-
-          // Compute the remainder of the SIMD operation.
-          if (cnt_simd_loop != 0) {
-            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);
-              }
-            }
-          }
-#endif
-        } 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 sfloat_mul_self(VALUE self, VALUE other) {
-  ndfunc_arg_in_t ain[2] = { { cT, 0 }, { cT, 0 } };
+static VALUE sfloat_reciprocal(VALUE self) {
+  ndfunc_arg_in_t ain[1] = { { cT, 0 } };
   ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_sfloat_mul, STRIDE_LOOP, 2, 1, ain, aout };
-
-  return na_ndloop(&ndf, 2, self, other);
-}
-
-static VALUE sfloat_mul(VALUE self, VALUE other) {
-
-  VALUE klass, v;
+  ndfunc_t ndf = { iter_sfloat_reciprocal, FULL_LOOP, 1, 1, ain, aout };
 
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return sfloat_mul_self(self, other);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, '*', 1, other);
-  }
+  return na_ndloop(&ndf, 1, self);
 }
 
-#define check_intdivzero(y)                                                                    \
-  {}
-
-static void iter_sfloat_div(na_loop_t* const lp) {
-  size_t i = 0;
-  size_t n;
-  char *p1, *p2, *p3;
-  ssize_t s1, s2, s3;
-
-#ifdef __SSE2__
-  size_t cnt;
-  size_t cnt_simd_loop = -1;
-
-  __m128 a;
-  __m128 b;
+static void iter_sfloat_sign(na_loop_t* const lp) {
+  size_t i, n;
+  char *p1, *p2;
+  ssize_t s1, s2;
+  size_t *idx1, *idx2;
+  dtype x;
 
-  size_t num_pack; // Number of elements packed for SIMD.
-  num_pack = SIMD_ALIGNMENT_SIZE / sizeof(dtype);
-#endif
   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)) {
-#ifdef __SSE2__
-      // Check number of elements. & Check same alignment.
-      if ((n >= num_pack) &&
-          is_same_aligned3(
-            &((dtype*)p1)[i], &((dtype*)p2)[i], &((dtype*)p3)[i], SIMD_ALIGNMENT_SIZE
-          )) {
-        // Calculate up to the position just before the start of SIMD computation.
-        cnt = get_count_of_elements_not_aligned_to_simd_size(
-          &((dtype*)p1)[i], SIMD_ALIGNMENT_SIZE, sizeof(dtype)
-        );
-#endif
-        if (p1 == p3) { // inplace case
-#ifdef __SSE2__
-          for (; i < cnt; i++) {
-#else
-        for (; i < n; i++) {
-          check_intdivzero(((dtype*)p2)[i]);
-#endif
-            ((dtype*)p1)[i] = m_div(((dtype*)p1)[i], ((dtype*)p2)[i]);
-          }
-        } else {
-#ifdef __SSE2__
-          for (; i < cnt; i++) {
-#else
-        for (; i < n; i++) {
-          check_intdivzero(((dtype*)p2)[i]);
-#endif
-            ((dtype*)p3)[i] = m_div(((dtype*)p1)[i], ((dtype*)p2)[i]);
-          }
-        }
-
-#ifdef __SSE2__
-        // Get the count of SIMD computation loops.
-        cnt_simd_loop = (n - i) % num_pack;
+  INIT_PTR_IDX(lp, 0, p1, s1, idx1);
+  INIT_PTR_IDX(lp, 1, p2, s2, idx2);
 
-        // SIMD computation.
-        if (p1 == p3) { // inplace case
-          for (; i < n - cnt_simd_loop; i += num_pack) {
-            a = _mm_load_ps(&((dtype*)p1)[i]);
-            b = _mm_load_ps(&((dtype*)p2)[i]);
-            a = _mm_div_ps(a, b);
-            _mm_store_ps(&((dtype*)p1)[i], a);
-          }
-        } else {
-          for (; i < n - cnt_simd_loop; i += num_pack) {
-            a = _mm_load_ps(&((dtype*)p1)[i]);
-            b = _mm_load_ps(&((dtype*)p2)[i]);
-            a = _mm_div_ps(a, b);
-            _mm_stream_ps(&((dtype*)p3)[i], a);
-          }
-        }
+  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);
       }
-
-      // Compute the remainder of the SIMD operation.
-      if (cnt_simd_loop != 0) {
-        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]);
-          }
-        }
+    } 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);
       }
-#endif
-      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_sign(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)) {
-#ifdef __SSE2__
-          // Broadcast a scalar value and use it for SIMD computation.
-          b = _mm_load1_ps(&((dtype*)p2)[0]);
-
-          // Check number of elements. & Check same alignment.
-          if ((n >= num_pack) &&
-              is_same_aligned2(&((dtype*)p1)[i], &((dtype*)p3)[i], SIMD_ALIGNMENT_SIZE)) {
-            // Calculate up to the position just before the start of SIMD computation.
-            cnt = get_count_of_elements_not_aligned_to_simd_size(
-              &((dtype*)p1)[i], SIMD_ALIGNMENT_SIZE, sizeof(dtype)
-            );
-#endif
-            if (p1 == p3) { // inplace case
-#ifdef __SSE2__
-              for (; i < cnt; i++) {
-#else
-            for (; i < n; i++) {
-#endif
-                ((dtype*)p1)[i] = m_div(((dtype*)p1)[i], *(dtype*)p2);
-              }
-            } else {
-#ifdef __SSE2__
-              for (; i < cnt; i++) {
-#else
-            for (; i < n; i++) {
-#endif
-                ((dtype*)p3)[i] = m_div(((dtype*)p1)[i], *(dtype*)p2);
-              }
-            }
-
-#ifdef __SSE2__
-            // Get the count of SIMD computation loops.
-            cnt_simd_loop = (n - i) % num_pack;
-
-            // SIMD computation.
-            if (p1 == p3) { // inplace case
-              for (; i < n - cnt_simd_loop; i += num_pack) {
-                a = _mm_load_ps(&((dtype*)p1)[i]);
-                a = _mm_div_ps(a, b);
-                _mm_store_ps(&((dtype*)p1)[i], a);
-              }
-            } else {
-              for (; i < n - cnt_simd_loop; i += num_pack) {
-                a = _mm_load_ps(&((dtype*)p1)[i]);
-                a = _mm_div_ps(a, b);
-                _mm_stream_ps(&((dtype*)p3)[i], a);
-              }
-            }
-          }
-
-          // Compute the remainder of the SIMD operation.
-          if (cnt_simd_loop != 0) {
-            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);
-              }
-            }
-          }
-#endif
-        } 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_sign(((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_sign(*(dtype*)p1);
             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 sfloat_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_sfloat_div, STRIDE_LOOP, 2, 1, ain, aout };
-
-  return na_ndloop(&ndf, 2, self, other);
-}
-
-static VALUE sfloat_div(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return sfloat_div_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_sfloat_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 sfloat_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_sfloat_mod, STRIDE_LOOP, 2, 1, ain, aout };
-
-  return na_ndloop(&ndf, 2, self, other);
-}
-
-static VALUE sfloat_mod(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return sfloat_mod_self(self, other);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, '%', 1, other);
-  }
-}
-
-static void iter_sfloat_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);
-    m_divmod(x, y, a, b);
-    SET_DATA_STRIDE(p3, s3, dtype, a);
-    SET_DATA_STRIDE(p4, s4, dtype, b);
-  }
-}
-
-static VALUE sfloat_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_sfloat_divmod, STRIDE_LOOP, 2, 2, ain, aout };
-
-  return na_ndloop(&ndf, 2, self, other);
-}
-
-static VALUE sfloat_divmod(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return sfloat_divmod_self(self, other);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, id_divmod, 1, other);
-  }
-}
-
-static void iter_sfloat_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_sfloat_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 sfloat_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_sfloat_pow, STRIDE_LOOP, 2, 1, ain, aout };
-  ndfunc_t ndf_i = { iter_sfloat_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);
-  }
-}
-
-static VALUE sfloat_pow(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return sfloat_pow_self(self, other);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, id_pow, 1, other);
-  }
-}
-
-static void iter_sfloat_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);
-      }
-      //
-    }
-  }
-}
-
-static VALUE sfloat_minus(VALUE self) {
-  ndfunc_arg_in_t ain[1] = { { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_sfloat_minus, FULL_LOOP, 1, 1, ain, aout };
-
-  return na_ndloop(&ndf, 1, self);
-}
-
-static void iter_sfloat_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);
-      }
-      //
-    }
-  }
-}
-
-static VALUE sfloat_reciprocal(VALUE self) {
-  ndfunc_arg_in_t ain[1] = { { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_sfloat_reciprocal, FULL_LOOP, 1, 1, ain, aout };
-
-  return na_ndloop(&ndf, 1, self);
-}
-
-static void iter_sfloat_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);
-      }
-      //
-    }
-  }
-}
-
-static VALUE sfloat_sign(VALUE self) {
-  ndfunc_arg_in_t ain[1] = { { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_sfloat_sign, FULL_LOOP, 1, 1, ain, aout };
-
-  return na_ndloop(&ndf, 1, self);
-}
-
-static void iter_sfloat_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);
-      }
-      //
-    }
-  }
-}
-
-static VALUE sfloat_square(VALUE self) {
-  ndfunc_arg_in_t ain[1] = { { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_sfloat_square, FULL_LOOP, 1, 1, ain, aout };
-
-  return na_ndloop(&ndf, 1, self);
-}
-
-static void iter_sfloat_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 sfloat_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_sfloat_eq, STRIDE_LOOP, 2, 1, ain, aout };
-
-  return na_ndloop(&ndf, 2, self, other);
-}
-
-static VALUE sfloat_eq(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return sfloat_eq_self(self, other);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, id_eq, 1, other);
-  }
-}
-
-static void iter_sfloat_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 sfloat_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_sfloat_ne, STRIDE_LOOP, 2, 1, ain, aout };
-
-  return na_ndloop(&ndf, 2, self, other);
-}
-
-static VALUE sfloat_ne(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return sfloat_ne_self(self, other);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, id_ne, 1, other);
-  }
-}
-
-static void iter_sfloat_nearly_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_nearly_eq(x, y)) ? 1 : 0;
-    STORE_BIT(a3, p3, b);
-    p3 += s3;
-  }
-}
-
-static VALUE sfloat_nearly_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_sfloat_nearly_eq, STRIDE_LOOP, 2, 1, ain, aout };
-
-  return na_ndloop(&ndf, 2, self, other);
-}
-
-static VALUE sfloat_nearly_eq(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return sfloat_nearly_eq_self(self, other);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, id_nearly_eq, 1, other);
-  }
-}
-
-static void iter_sfloat_floor(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_floor(x);
-        SET_DATA_INDEX(p2, idx2, dtype, x);
-      }
-    } else {
-      for (i = 0; i < n; i++) {
-        GET_DATA_INDEX(p1, idx1, dtype, x);
-        x = m_floor(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_floor(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_floor(((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_floor(*(dtype*)p1);
-            p1 += s1;
-            p2 += s2;
-          }
-          return;
-          //
-        }
-      }
-      for (i = 0; i < n; i++) {
-        GET_DATA_STRIDE(p1, s1, dtype, x);
-        x = m_floor(x);
-        SET_DATA_STRIDE(p2, s2, dtype, x);
-      }
-      //
-    }
-  }
-}
-
-static VALUE sfloat_floor(VALUE self) {
-  ndfunc_arg_in_t ain[1] = { { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_sfloat_floor, FULL_LOOP, 1, 1, ain, aout };
-
-  return na_ndloop(&ndf, 1, self);
-}
-
-static void iter_sfloat_round(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_round(x);
-        SET_DATA_INDEX(p2, idx2, dtype, x);
-      }
-    } else {
-      for (i = 0; i < n; i++) {
-        GET_DATA_INDEX(p1, idx1, dtype, x);
-        x = m_round(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_round(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_round(((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_round(*(dtype*)p1);
-            p1 += s1;
-            p2 += s2;
-          }
-          return;
-          //
-        }
-      }
-      for (i = 0; i < n; i++) {
-        GET_DATA_STRIDE(p1, s1, dtype, x);
-        x = m_round(x);
-        SET_DATA_STRIDE(p2, s2, dtype, x);
-      }
-      //
-    }
-  }
-}
-
-static VALUE sfloat_round(VALUE self) {
-  ndfunc_arg_in_t ain[1] = { { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_sfloat_round, FULL_LOOP, 1, 1, ain, aout };
-
-  return na_ndloop(&ndf, 1, self);
-}
-
-static void iter_sfloat_ceil(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_ceil(x);
-        SET_DATA_INDEX(p2, idx2, dtype, x);
-      }
-    } else {
-      for (i = 0; i < n; i++) {
-        GET_DATA_INDEX(p1, idx1, dtype, x);
-        x = m_ceil(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_ceil(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_ceil(((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_ceil(*(dtype*)p1);
-            p1 += s1;
-            p2 += s2;
-          }
-          return;
-          //
-        }
-      }
-      for (i = 0; i < n; i++) {
-        GET_DATA_STRIDE(p1, s1, dtype, x);
-        x = m_ceil(x);
-        SET_DATA_STRIDE(p2, s2, dtype, x);
-      }
-      //
-    }
-  }
-}
-
-static VALUE sfloat_ceil(VALUE self) {
-  ndfunc_arg_in_t ain[1] = { { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_sfloat_ceil, FULL_LOOP, 1, 1, ain, aout };
-
-  return na_ndloop(&ndf, 1, self);
-}
-
-static void iter_sfloat_trunc(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_trunc(x);
-        SET_DATA_INDEX(p2, idx2, dtype, x);
-      }
-    } else {
-      for (i = 0; i < n; i++) {
-        GET_DATA_INDEX(p1, idx1, dtype, x);
-        x = m_trunc(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_trunc(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_trunc(((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_trunc(*(dtype*)p1);
-            p1 += s1;
-            p2 += s2;
-          }
-          return;
-          //
-        }
-      }
-      for (i = 0; i < n; i++) {
-        GET_DATA_STRIDE(p1, s1, dtype, x);
-        x = m_trunc(x);
-        SET_DATA_STRIDE(p2, s2, dtype, x);
-      }
-      //
-    }
-  }
-}
-
-static VALUE sfloat_trunc(VALUE self) {
-  ndfunc_arg_in_t ain[1] = { { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_sfloat_trunc, FULL_LOOP, 1, 1, ain, aout };
-
-  return na_ndloop(&ndf, 1, self);
-}
-
-static void iter_sfloat_rint(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_rint(x);
-        SET_DATA_INDEX(p2, idx2, dtype, x);
-      }
-    } else {
-      for (i = 0; i < n; i++) {
-        GET_DATA_INDEX(p1, idx1, dtype, x);
-        x = m_rint(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_rint(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_rint(((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_rint(*(dtype*)p1);
-            p1 += s1;
-            p2 += s2;
-          }
-          return;
-          //
-        }
-      }
-      for (i = 0; i < n; i++) {
-        GET_DATA_STRIDE(p1, s1, dtype, x);
-        x = m_rint(x);
-        SET_DATA_STRIDE(p2, s2, dtype, x);
-      }
-      //
-    }
-  }
-}
-
-static VALUE sfloat_rint(VALUE self) {
-  ndfunc_arg_in_t ain[1] = { { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
-  ndfunc_t ndf = { iter_sfloat_rint, FULL_LOOP, 1, 1, ain, aout };
-
-  return na_ndloop(&ndf, 1, self);
-}
-
-#define check_intdivzero(y)                                                                    \
-  {}
-
-static void iter_sfloat_copysign(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_copysign(((dtype*)p1)[i], ((dtype*)p2)[i]);
-        }
-      } else {
-        for (; i < n; i++) {
-          check_intdivzero(((dtype*)p2)[i]);
-          ((dtype*)p3)[i] = m_copysign(((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_copysign(((dtype*)p1)[i], *(dtype*)p2);
-            }
-          } else {
-            for (; i < n; i++) {
-              ((dtype*)p3)[i] = m_copysign(((dtype*)p1)[i], *(dtype*)p2);
-            }
-          }
-        } else {
-          for (i = 0; i < n; i++) {
-            *(dtype*)p3 = m_copysign(*(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_copysign(*(dtype*)p1, *(dtype*)p2);
-            p1 += s1;
-            p2 += s2;
-          }
-        } else {
-          for (i = 0; i < n; i++) {
-            check_intdivzero(*(dtype*)p2);
-            *(dtype*)p3 = m_copysign(*(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_copysign(x, y);
-    SET_DATA_STRIDE(p3, s3, dtype, z);
-  }
-  //
-}
-#undef check_intdivzero
-
-static VALUE sfloat_copysign_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_sfloat_copysign, STRIDE_LOOP, 2, 1, ain, aout };
-
-  return na_ndloop(&ndf, 2, self, other);
-}
-
-static VALUE sfloat_copysign(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return sfloat_copysign_self(self, other);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, id_copysign, 1, other);
-  }
-}
-
-static void iter_sfloat_signbit(na_loop_t* const lp) {
-  size_t i;
-  char* p1;
-  BIT_DIGIT* a2;
-  size_t p2;
-  ssize_t s1, s2;
-  size_t* idx1;
-  dtype x;
-  BIT_DIGIT b;
-  INIT_COUNTER(lp, i);
-  INIT_PTR_IDX(lp, 0, p1, s1, idx1);
-  INIT_PTR_BIT(lp, 1, a2, p2, s2);
-  if (idx1) {
-    for (; i--;) {
-      GET_DATA_INDEX(p1, idx1, dtype, x);
-      b = (m_signbit(x)) ? 1 : 0;
-      STORE_BIT(a2, p2, b);
-      p2 += s2;
-    }
-  } else {
-    for (; i--;) {
-      GET_DATA_STRIDE(p1, s1, dtype, x);
-      b = (m_signbit(x)) ? 1 : 0;
-      STORE_BIT(a2, p2, b);
-      p2 += s2;
-    }
-  }
-}
-
-static VALUE sfloat_signbit(VALUE self) {
-  ndfunc_arg_in_t ain[1] = { { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { numo_cBit, 0 } };
-  ndfunc_t ndf = { iter_sfloat_signbit, FULL_LOOP, 1, 1, ain, aout };
-
-  return na_ndloop(&ndf, 1, self);
-}
-
-static void iter_sfloat_modf(na_loop_t* const lp) {
-  size_t i;
-  char *p1, *p2, *p3;
-  ssize_t s1, s2, s3;
-  dtype x, y, z;
-  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);
-    m_modf(x, y, z);
-    SET_DATA_STRIDE(p2, s2, dtype, y);
-    SET_DATA_STRIDE(p3, s3, dtype, z);
-  }
-}
-
-static VALUE sfloat_modf(VALUE self) {
-  ndfunc_arg_in_t ain[1] = { { cT, 0 } };
-  ndfunc_arg_out_t aout[2] = { { cT, 0 }, { cT, 0 } };
-  ndfunc_t ndf = { iter_sfloat_modf, STRIDE_LOOP, 1, 2, ain, aout };
-
-  return na_ndloop(&ndf, 1, self);
-}
-
-static void iter_sfloat_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 sfloat_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_sfloat_gt, STRIDE_LOOP, 2, 1, ain, aout };
-
-  return na_ndloop(&ndf, 2, self, other);
-}
-
-static VALUE sfloat_gt(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return sfloat_gt_self(self, other);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, id_gt, 1, other);
-  }
-}
-
-static void iter_sfloat_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 sfloat_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_sfloat_ge, STRIDE_LOOP, 2, 1, ain, aout };
-
-  return na_ndloop(&ndf, 2, self, other);
-}
-
-static VALUE sfloat_ge(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return sfloat_ge_self(self, other);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, id_ge, 1, other);
-  }
-}
-
-static void iter_sfloat_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 sfloat_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_sfloat_lt, STRIDE_LOOP, 2, 1, ain, aout };
-
-  return na_ndloop(&ndf, 2, self, other);
-}
-
-static VALUE sfloat_lt(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return sfloat_lt_self(self, other);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, id_lt, 1, other);
-  }
-}
-
-static void iter_sfloat_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 sfloat_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_sfloat_le, STRIDE_LOOP, 2, 1, ain, aout };
-
-  return na_ndloop(&ndf, 2, self, other);
-}
-
-static VALUE sfloat_le(VALUE self, VALUE other) {
-
-  VALUE klass, v;
-  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
-  if (klass == cT) {
-    return sfloat_le_self(self, other);
-  } else {
-    v = rb_funcall(klass, id_cast, 1, self);
-    return rb_funcall(v, id_le, 1, other);
-  }
-}
-
-static void iter_sfloat_isnan(na_loop_t* const lp) {
-  size_t i;
-  char* p1;
-  BIT_DIGIT* a2;
-  size_t p2;
-  ssize_t s1, s2;
-  size_t* idx1;
-  dtype x;
-  BIT_DIGIT b;
-  INIT_COUNTER(lp, i);
-  INIT_PTR_IDX(lp, 0, p1, s1, idx1);
-  INIT_PTR_BIT(lp, 1, a2, p2, s2);
-  if (idx1) {
-    for (; i--;) {
-      GET_DATA_INDEX(p1, idx1, dtype, x);
-      b = (m_isnan(x)) ? 1 : 0;
-      STORE_BIT(a2, p2, b);
-      p2 += s2;
-    }
-  } else {
-    for (; i--;) {
-      GET_DATA_STRIDE(p1, s1, dtype, x);
-      b = (m_isnan(x)) ? 1 : 0;
-      STORE_BIT(a2, p2, b);
-      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);
+      }
+      //
     }
   }
 }
 
-static VALUE sfloat_isnan(VALUE self) {
+static VALUE sfloat_sign(VALUE self) {
   ndfunc_arg_in_t ain[1] = { { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { numo_cBit, 0 } };
-  ndfunc_t ndf = { iter_sfloat_isnan, FULL_LOOP, 1, 1, ain, aout };
+  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
+  ndfunc_t ndf = { iter_sfloat_sign, FULL_LOOP, 1, 1, ain, aout };
 
   return na_ndloop(&ndf, 1, self);
 }
 
-static void iter_sfloat_isinf(na_loop_t* const lp) {
-  size_t i;
-  char* p1;
-  BIT_DIGIT* a2;
-  size_t p2;
+static void iter_sfloat_square(na_loop_t* const lp) {
+  size_t i, n;
+  char *p1, *p2;
   ssize_t s1, s2;
-  size_t* idx1;
+  size_t *idx1, *idx2;
   dtype x;
-  BIT_DIGIT b;
-  INIT_COUNTER(lp, i);
+
+  INIT_COUNTER(lp, n);
   INIT_PTR_IDX(lp, 0, p1, s1, idx1);
-  INIT_PTR_BIT(lp, 1, a2, p2, s2);
+  INIT_PTR_IDX(lp, 1, p2, s2, idx2);
+
   if (idx1) {
-    for (; i--;) {
-      GET_DATA_INDEX(p1, idx1, dtype, x);
-      b = (m_isinf(x)) ? 1 : 0;
-      STORE_BIT(a2, p2, b);
-      p2 += s2;
+    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--;) {
-      GET_DATA_STRIDE(p1, s1, dtype, x);
-      b = (m_isinf(x)) ? 1 : 0;
-      STORE_BIT(a2, p2, b);
-      p2 += s2;
+    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);
+      }
+      //
     }
   }
 }
 
-static VALUE sfloat_isinf(VALUE self) {
+static VALUE sfloat_square(VALUE self) {
   ndfunc_arg_in_t ain[1] = { { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { numo_cBit, 0 } };
-  ndfunc_t ndf = { iter_sfloat_isinf, FULL_LOOP, 1, 1, ain, aout };
+  ndfunc_arg_out_t aout[1] = { { cT, 0 } };
+  ndfunc_t ndf = { iter_sfloat_square, FULL_LOOP, 1, 1, ain, aout };
 
   return na_ndloop(&ndf, 1, self);
 }
 
-static void iter_sfloat_isposinf(na_loop_t* const lp) {
-  size_t i;
-  char* p1;
-  BIT_DIGIT* a2;
-  size_t p2;
-  ssize_t s1, s2;
-  size_t* idx1;
-  dtype x;
-  BIT_DIGIT b;
-  INIT_COUNTER(lp, i);
-  INIT_PTR_IDX(lp, 0, p1, s1, idx1);
-  INIT_PTR_BIT(lp, 1, a2, p2, s2);
-  if (idx1) {
-    for (; i--;) {
-      GET_DATA_INDEX(p1, idx1, dtype, x);
-      b = (m_isposinf(x)) ? 1 : 0;
-      STORE_BIT(a2, p2, b);
-      p2 += s2;
+#define check_intdivzero(y)                                                                    \
+  {}
+
+static void iter_sfloat_copysign(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_copysign(((dtype*)p1)[i], ((dtype*)p2)[i]);
+        }
+      } else {
+        for (; i < n; i++) {
+          check_intdivzero(((dtype*)p2)[i]);
+          ((dtype*)p3)[i] = m_copysign(((dtype*)p1)[i], ((dtype*)p2)[i]);
+        }
+      }
+      return;
     }
-  } else {
-    for (; i--;) {
-      GET_DATA_STRIDE(p1, s1, dtype, x);
-      b = (m_isposinf(x)) ? 1 : 0;
-      STORE_BIT(a2, p2, b);
-      p2 += s2;
+
+    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_copysign(((dtype*)p1)[i], *(dtype*)p2);
+            }
+          } else {
+            for (; i < n; i++) {
+              ((dtype*)p3)[i] = m_copysign(((dtype*)p1)[i], *(dtype*)p2);
+            }
+          }
+        } else {
+          for (i = 0; i < n; i++) {
+            *(dtype*)p3 = m_copysign(*(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_copysign(*(dtype*)p1, *(dtype*)p2);
+            p1 += s1;
+            p2 += s2;
+          }
+        } else {
+          for (i = 0; i < n; i++) {
+            check_intdivzero(*(dtype*)p2);
+            *(dtype*)p3 = m_copysign(*(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_copysign(x, y);
+    SET_DATA_STRIDE(p3, s3, dtype, z);
+  }
+  //
 }
+#undef check_intdivzero
 
-static VALUE sfloat_isposinf(VALUE self) {
-  ndfunc_arg_in_t ain[1] = { { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { numo_cBit, 0 } };
-  ndfunc_t ndf = { iter_sfloat_isposinf, FULL_LOOP, 1, 1, ain, aout };
+static VALUE sfloat_copysign_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_sfloat_copysign, STRIDE_LOOP, 2, 1, ain, aout };
 
-  return na_ndloop(&ndf, 1, self);
+  return na_ndloop(&ndf, 2, self, other);
+}
+
+static VALUE sfloat_copysign(VALUE self, VALUE other) {
+
+  VALUE klass, v;
+
+  klass = na_upcast(rb_obj_class(self), rb_obj_class(other));
+  if (klass == cT) {
+    return sfloat_copysign_self(self, other);
+  } else {
+    v = rb_funcall(klass, id_cast, 1, self);
+    return rb_funcall(v, id_copysign, 1, other);
+  }
 }
 
-static void iter_sfloat_isneginf(na_loop_t* const lp) {
+static void iter_sfloat_signbit(na_loop_t* const lp) {
   size_t i;
   char* p1;
   BIT_DIGIT* a2;
@@ -4084,61 +2091,49 @@ static void iter_sfloat_isneginf(na_loop_t* const lp) {
   if (idx1) {
     for (; i--;) {
       GET_DATA_INDEX(p1, idx1, dtype, x);
-      b = (m_isneginf(x)) ? 1 : 0;
+      b = (m_signbit(x)) ? 1 : 0;
       STORE_BIT(a2, p2, b);
       p2 += s2;
     }
   } else {
     for (; i--;) {
       GET_DATA_STRIDE(p1, s1, dtype, x);
-      b = (m_isneginf(x)) ? 1 : 0;
+      b = (m_signbit(x)) ? 1 : 0;
       STORE_BIT(a2, p2, b);
       p2 += s2;
     }
   }
 }
 
-static VALUE sfloat_isneginf(VALUE self) {
+static VALUE sfloat_signbit(VALUE self) {
   ndfunc_arg_in_t ain[1] = { { cT, 0 } };
   ndfunc_arg_out_t aout[1] = { { numo_cBit, 0 } };
-  ndfunc_t ndf = { iter_sfloat_isneginf, FULL_LOOP, 1, 1, ain, aout };
+  ndfunc_t ndf = { iter_sfloat_signbit, FULL_LOOP, 1, 1, ain, aout };
 
   return na_ndloop(&ndf, 1, self);
 }
 
-static void iter_sfloat_isfinite(na_loop_t* const lp) {
+static void iter_sfloat_modf(na_loop_t* const lp) {
   size_t i;
-  char* p1;
-  BIT_DIGIT* a2;
-  size_t p2;
-  ssize_t s1, s2;
-  size_t* idx1;
-  dtype x;
-  BIT_DIGIT b;
+  char *p1, *p2, *p3;
+  ssize_t s1, s2, s3;
+  dtype x, y, z;
   INIT_COUNTER(lp, i);
-  INIT_PTR_IDX(lp, 0, p1, s1, idx1);
-  INIT_PTR_BIT(lp, 1, a2, p2, s2);
-  if (idx1) {
-    for (; i--;) {
-      GET_DATA_INDEX(p1, idx1, dtype, x);
-      b = (m_isfinite(x)) ? 1 : 0;
-      STORE_BIT(a2, p2, b);
-      p2 += s2;
-    }
-  } else {
-    for (; i--;) {
-      GET_DATA_STRIDE(p1, s1, dtype, x);
-      b = (m_isfinite(x)) ? 1 : 0;
-      STORE_BIT(a2, p2, b);
-      p2 += s2;
-    }
+  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);
+    m_modf(x, y, z);
+    SET_DATA_STRIDE(p2, s2, dtype, y);
+    SET_DATA_STRIDE(p3, s3, dtype, z);
   }
 }
 
-static VALUE sfloat_isfinite(VALUE self) {
+static VALUE sfloat_modf(VALUE self) {
   ndfunc_arg_in_t ain[1] = { { cT, 0 } };
-  ndfunc_arg_out_t aout[1] = { { numo_cBit, 0 } };
-  ndfunc_t ndf = { iter_sfloat_isfinite, FULL_LOOP, 1, 1, ain, aout };
+  ndfunc_arg_out_t aout[2] = { { cT, 0 }, { cT, 0 } };
+  ndfunc_t ndf = { iter_sfloat_modf, STRIDE_LOOP, 1, 2, ain, aout };
 
   return na_ndloop(&ndf, 1, self);
 }