cumo 0.4.3 → 0.5.1

data/ext/cumo/narray/gen/tmpl/logseq.c

@@ -62,7 +62,7 @@ static void
 
 /*
   Set logarithmic sequence of numbers to self. The sequence is obtained from
-     base**(beg+i*step)
+     `base**(beg+i*step)`
   where i is 1-dimensional index.
   Applicable classes: DFloat, SFloat, DComplex, SCopmplex.
 
@@ -74,11 +74,12 @@ static void
 
   @example
     Cumo::DFloat.new(5).logseq(4,-1,2)
-    => Cumo::DFloat#shape=[5]
-      [16, 8, 4, 2, 1]
+    # => Cumo::DFloat#shape=[5]
+    #   [16, 8, 4, 2, 1]
+
     Cumo::DComplex.new(5).logseq(0,1i*Math::PI/3,Math::E)
-    => Cumo::DComplex#shape=[5]
-      [1+7.26156e-310i, 0.5+0.866025i, -0.5+0.866025i, -1+1.22465e-16i, ...]
+    # => Cumo::DComplex#shape=[5]
+    #   [1+7.26156e-310i, 0.5+0.866025i, -0.5+0.866025i, -1+1.22465e-16i, ...]
 */
 static VALUE
 <%=c_func(-1)%>(int argc, VALUE *args, VALUE self)

data/ext/cumo/narray/gen/tmpl/map_with_index.c

@@ -78,14 +78,13 @@ static void
   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 <%=name%>
-
-  For a block {|x,i,j,...| ... }
-  @yield [x,i,j,...]  x is an element, i,j,... are multidimensional indices.
-
+  For a block `{|x,i,j,...| ... }`,
+  @yieldparam [Numeric] x  an element
+  @yieldparam [Integer] i,j,...  multitimensional indices
   @return [Cumo::NArray] mapped array
-
+  @see #map
+  @see #each_with_index
 */
 static VALUE
 <%=c_func(0)%>(VALUE self)

data/ext/cumo/narray/gen/tmpl/median.c

@@ -40,8 +40,8 @@ static void
 <% else %>
   @overload <%=name%>(axis:nil, keepdims:false)
 <% end %>
-  @param [Numeric,Array,Range] axis (keyword) Affected dimensions.
-  @param [TrueClass] keepdims (keyword) If true, the reduced axes are left in the result array as dimensions with size one.
+  @param [Numeric,Array,Range] axis  Finds <%=name%> along the axis.
+  @param [TrueClass] keepdims  If true, the reduced axes are left in the result array as dimensions with size one.
   @return [Cumo::<%=class_name%>] returns <%=name%> of self.
 */
 

data/ext/cumo/narray/gen/tmpl/minmax.c

@@ -26,7 +26,7 @@ static void
 <% else %>
   @overload <%=name%>(axis:nil, keepdims:false)
 <% end %>
-  @param [Numeric,Array,Range] axis (keyword) Affected dimensions.
+  @param [Numeric,Array,Range] axis  Finds min-max along the axis.
   @param [TrueClass] keepdims (keyword) If true, the reduced axes are left in the result array as dimensions with size one.
   @return [Cumo::<%=class_name%>,Cumo::<%=class_name%>] min and max of self.
 */

data/ext/cumo/narray/gen/tmpl/poly.c

@@ -18,10 +18,10 @@ static void
 }
 
 /*
-  Polynomial.: a0 + a1*x + a2*x**2 + a3*x**3 + ... + an*x**n
-  @overload <%=name%> a0, a1, ...
-  @param [Cumo::NArray,Numeric] a0
-  @param [Cumo::NArray,Numeric] a1 , ...
+  Calculate polynomial.
+    `x.poly(a0,a1,a2,...,an) = a0 + a1*x + a2*x**2 + ... + an*x**n`
+  @overload <%=name%> a0, a1, ..., an
+  @param [Cumo::NArray,Numeric] a0,a1,...,an
   @return [Cumo::<%=class_name%>]
 */
 static VALUE

data/ext/cumo/narray/gen/tmpl/pooling_backward.c

@@ -126,7 +126,7 @@ POOLING_BACKAWARD_ERROR:
 }
 
 #else // CUDNN_FOUND
-VALUE cumo_cuda_eCUDNNError;
+#include "cumo/cuda/cudnn.h"
 
 static VALUE
 <%=c_func(-1)%>(int argc, VALUE argv[], VALUE self)

data/ext/cumo/narray/gen/tmpl/pooling_forward.c

@@ -126,7 +126,7 @@ POLLING_FORWARD_ERROR:
 }
 
 #else // CUDNN_FOUND
-VALUE cumo_cuda_eCUDNNError;
+#include "cumo/cuda/cudnn.h"
 
 static VALUE
 <%=c_func(-1)%>(int argc, VALUE argv[], VALUE self)

data/ext/cumo/narray/gen/tmpl/qsort.c

@@ -76,11 +76,7 @@
         (es) % sizeof(long) ? 2 : (es) == sizeof(long)? 0 : 1;
 
 static inline void
-swapfunc(a, b, n, swaptype)
-     char       *a,
-     *b;
-     size_t          n;
-     int                     swaptype;
+swapfunc(char *a, char *b, size_t n, int swaptype)
 {
     if (swaptype <= 1)
         swapcode(long, a, b, n);

data/ext/cumo/narray/gen/tmpl/rand.c

@@ -115,14 +115,16 @@ static void
   @return [Cumo::<%=class_name%>] self.
   @example
     Cumo::DFloat.new(6).rand
-    => Cumo::DFloat#shape=[6]
-       [0.0617545, 0.373067, 0.794815, 0.201042, 0.116041, 0.344032]
+    # => Cumo::DFloat#shape=[6]
+    #    [0.0617545, 0.373067, 0.794815, 0.201042, 0.116041, 0.344032]
+
     Cumo::DComplex.new(6).rand(5+5i)
-    => Cumo::DComplex#shape=[6]
-       [2.69974+3.68908i, 0.825443+0.254414i, 0.540323+0.34354i, 4.52061+2.39322i, ...]
+    # => Cumo::DComplex#shape=[6]
+    #    [2.69974+3.68908i, 0.825443+0.254414i, 0.540323+0.34354i, 4.52061+2.39322i, ...]
+
     Cumo::Int32.new(6).rand(2,5)
-    => Cumo::Int32#shape=[6]
-       [4, 3, 3, 2, 4, 2]
+    # => Cumo::Int32#shape=[6]
+    #    [4, 3, 3, 2, 4, 2]
 */
 static VALUE
 <%=c_func(-1)%>(int argc, VALUE *args, VALUE self)

data/ext/cumo/narray/gen/tmpl/rand_norm.c

@@ -77,24 +77,26 @@ static void
   @return [Cumo::<%=class_name%>] self.
   @example
     Cumo::DFloat.new(5,5).rand_norm
-    => Cumo::DFloat#shape=[5,5]
-       [[-0.581255, -0.168354, 0.586895, -0.595142, -0.802802],
-        [-0.326106, 0.282922, 1.68427, 0.918499, -0.0485384],
-        [-0.464453, -0.992194, 0.413794, -0.60717, -0.699695],
-        [-1.64168, 0.48676, -0.875871, -1.43275, 0.812172],
-        [-0.209975, -0.103612, -0.878617, -1.42495, 1.0968]]
+    # => Cumo::DFloat#shape=[5,5]
+    #    [[-0.581255, -0.168354, 0.586895, -0.595142, -0.802802],
+    #     [-0.326106, 0.282922, 1.68427, 0.918499, -0.0485384],
+    #     [-0.464453, -0.992194, 0.413794, -0.60717, -0.699695],
+    #     [-1.64168, 0.48676, -0.875871, -1.43275, 0.812172],
+    #     [-0.209975, -0.103612, -0.878617, -1.42495, 1.0968]]
+
     Cumo::DFloat.new(5,5).rand_norm(10,0.1)
-    => Cumo::DFloat#shape=[5,5]
-       [[9.9019, 9.90339, 10.0826, 9.98384, 9.72861],
-        [9.81507, 10.0272, 9.91445, 10.0568, 9.88923],
-        [10.0234, 9.97874, 9.96011, 9.9006, 9.99964],
-        [10.0186, 9.94598, 9.92236, 9.99811, 9.97003],
-        [9.79266, 9.95044, 9.95212, 9.93692, 10.2027]]
+    # => Cumo::DFloat#shape=[5,5]
+    #    [[9.9019, 9.90339, 10.0826, 9.98384, 9.72861],
+    #     [9.81507, 10.0272, 9.91445, 10.0568, 9.88923],
+    #     [10.0234, 9.97874, 9.96011, 9.9006, 9.99964],
+    #     [10.0186, 9.94598, 9.92236, 9.99811, 9.97003],
+    #     [9.79266, 9.95044, 9.95212, 9.93692, 10.2027]]
+
     Cumo::DComplex.new(3,3).rand_norm(5+5i)
-    => Cumo::DComplex#shape=[3,3]
-       [[5.84303+4.40052i, 4.00984+6.08982i, 5.10979+5.13215i],
-        [4.26477+3.99655i, 4.90052+5.00763i, 4.46607+2.3444i],
-        [4.5528+7.11003i, 5.62117+6.69094i, 5.05443+5.35133i]]
+    # => Cumo::DComplex#shape=[3,3]
+    #    [[5.84303+4.40052i, 4.00984+6.08982i, 5.10979+5.13215i],
+    #     [4.26477+3.99655i, 4.90052+5.00763i, 4.46607+2.3444i],
+    #     [4.5528+7.11003i, 5.62117+6.69094i, 5.05443+5.35133i]]
 */
 static VALUE
 <%=c_func(-1)%>(int argc, VALUE *args, VALUE self)

data/ext/cumo/narray/gen/tmpl/seq.c

@@ -80,11 +80,12 @@ static void
   @return [Cumo::<%=class_name%>] self.
   @example
     Cumo::DFloat.new(6).seq(1,-0.2)
-    => Cumo::DFloat#shape=[6]
-       [1, 0.8, 0.6, 0.4, 0.2, 0]
+    # => Cumo::DFloat#shape=[6]
+    #    [1, 0.8, 0.6, 0.4, 0.2, 0]
+
     Cumo::DComplex.new(6).seq(1,-0.2+0.2i)
-    => Cumo::DComplex#shape=[6]
-       [1+0i, 0.8+0.2i, 0.6+0.4i, 0.4+0.6i, 0.2+0.8i, 0+1i]
+    # => Cumo::DComplex#shape=[6]
+    #    [1+0i, 0.8+0.2i, 0.6+0.4i, 0.4+0.6i, 0.2+0.8i, 0+1i]
 */
 static VALUE
 <%=c_func(-1)%>(int argc, VALUE *args, VALUE self)

data/ext/cumo/narray/gen/tmpl/sort.c

@@ -22,17 +22,17 @@ static void
 <% else %>
   @overload <%=name%>(axis:nil)
 <% end %>
-  @param [Numeric,Array,Range] axis  Affected dimensions.
+  @param [Numeric,Array,Range] axis  Performs <%=name%> along the axis.
   @return [Cumo::<%=class_name%>] returns result of <%=name%>.
   @example
-      Cumo::DFloat[3,4,1,2].sort => Cumo::DFloat[1,2,3,4]
+      Cumo::DFloat[3,4,1,2].sort # => Cumo::DFloat[1,2,3,4]
 */
 static VALUE
 <%=c_func(-1)%>(int argc, VALUE *argv, VALUE self)
 {
     VALUE reduce;
     cumo_ndfunc_arg_in_t ain[2] = {{CUMO_OVERWRITE,0},{cumo_sym_reduce,0}};
-    cumo_ndfunc_t ndf = {0, CUMO_STRIDE_LOOP|CUMO_NDF_FLAT_REDUCE, 2,0, ain,0};
+    cumo_ndfunc_t ndf = {0, CUMO_NDF_HAS_LOOP|CUMO_NDF_FLAT_REDUCE, 2,0, ain,0};
 
     if (!CUMO_TEST_INPLACE(self)) {
         self = cumo_na_copy(self);

data/ext/cumo/narray/gen/tmpl/sort_index.c

@@ -52,10 +52,10 @@ static void
 <% else %>
   @overload <%=name%>(axis:nil)
 <% end %>
-  @param [Numeric,Array,Range] axis  Affected dimensions.
+  @param [Numeric,Array,Range] axis  Performs <%=name%> along the axis.
   @return [Integer,Cumo::Int] returns result index of <%=name%>.
   @example
-      Cumo::NArray[3,4,1,2].sort_index => Cumo::Int32[2,3,0,1]
+      Cumo::NArray[3,4,1,2].sort_index # => Cumo::Int32[2,3,0,1]
 */
 static VALUE
 <%=c_func(-1)%>(int argc, VALUE *argv, VALUE self)

data/ext/cumo/narray/gen/tmpl_bit/aref.c

@@ -2,42 +2,36 @@ static VALUE
 <%=c_func(-1)%>_cpu(int argc, VALUE *argv, VALUE self);
 
 /*
-  Array element referenece or slice view.
+  Array indexing.
   @overload [](dim0,...,dimL)
-  @param [Numeric,Range,etc] dim0,...,dimL  Multi-dimensional Index.
-  @return [Numeric,NArray::<%=class_name%>] Element object or NArray view.
+  @param [Numeric,Range,Array,Cumo::Bit,Cumo::Int32,Cumo::Int64] dim0,...,dimL  Multi-dimensional Index.
+  @return [Cumo::Bit,Numeric] Element value or NArray view.
 
-  --- Returns the element at +dim0+, +dim1+, ... are Numeric indices
-  for each dimension, or returns a NArray View as a sliced subarray if
-  +dim0+, +dim1+, ... includes other than Numeric index, e.g., Range
-  or Array or true.
+  --- Returns an element at `dim0`, `dim1`, ... are Numeric indices for each dimension, or returns a NArray View as a sliced array if `dim0`, `dim1`, ... includes other than Numeric index, e.g., Range or Array or true.
+
+  @see #[]=
 
   @example
-      a = Cumo::DFloat.new(4,5).seq
-      => Cumo::DFloat#shape=[4,5]
-      [[0, 1, 2, 3, 4],
-       [5, 6, 7, 8, 9],
-       [10, 11, 12, 13, 14],
-       [15, 16, 17, 18, 19]]
-
-      a[1,1]
-      => 6.0
-
-      a[1..3,1]
-      => Cumo::DFloat#shape=[3]
-      [6, 11, 16]
-
-      a[1,[1,3,4]]
-      => Cumo::DFloat#shape=[3]
-      [6, 8, 9]
-
-      a[true,2].fill(99)
-      a
-      => Cumo::DFloat#shape=[4,5]
-      [[0, 1, 99, 3, 4],
-       [5, 6, 99, 8, 9],
-       [10, 11, 99, 13, 14],
-       [15, 16, 99, 18, 19]]
+      a = Cumo::Int32.new(3,4).seq
+      # => Cumo::Int32#shape=[3,4]
+      # [[0, 1, 2, 3],
+      #  [4, 5, 6, 7],
+      #  [8, 9, 10, 11]]
+
+      b = (a%2).eq(0)
+      # => Cumo::Bit#shape=[3,4]
+      # [[1, 0, 1, 0],
+      #  [1, 0, 1, 0],
+      #  [1, 0, 1, 0]]
+
+      b[true,(0..-1)%2]
+      # => Cumo::Bit(view)#shape=[3,2]
+      # [[1, 1],
+      #  [1, 1],
+      #  [1, 1]]
+
+      b[1,1]
+      # => 0
  */
 static VALUE
 <%=c_func(-1)%>(int argc, VALUE *argv, VALUE self)

data/ext/cumo/narray/gen/tmpl_bit/aset.c

@@ -1,41 +1,29 @@
 /*
-  Array element(s) set.
+  Array element(s) assignment.
   @overload []=(dim0,..,dimL,val)
-  @param [Numeric,Range,etc] dim0,..,dimL  Multi-dimensional Index.
+  @param [Numeric,Range,Array,Cumo::Bit,Cumo::Int32,Cumo::Int64] dim0,..,dimL  Multi-dimensional Index.
   @param [Numeric,Cumo::NArray,etc] val  Value(s) to be set to self.
-  @return [Numeric] returns val (last argument).
+  @return [Numeric] returns `val` (last argument).
 
-  --- Replace element(s) at +dim0+, +dim1+, ... (index/range/array/true
-  for each dimention). Broadcasting mechanism is applied.
+  Replaces element(s) at `dim0`, `dim1`, ... . Broadcasting mechanism is applied.
 
-  @example
-      a = Cumo::DFloat.new(3,4).seq
-      => Cumo::DFloat#shape=[3,4]
-      [[0, 1, 2, 3],
-       [4, 5, 6, 7],
-       [8, 9, 10, 11]]
-
-      a[1,2]=99
-      a
-      => Cumo::DFloat#shape=[3,4]
-      [[0, 1, 2, 3],
-       [4, 5, 99, 7],
-       [8, 9, 10, 11]]
+  @see #[]
 
-      a[1,[0,2]] = [101,102]
-      a
-      => Cumo::DFloat#shape=[3,4]
-      [[0, 1, 2, 3],
-       [101, 5, 102, 7],
-       [8, 9, 10, 11]]
+  @example
+      a = Cumo::Bit.new(4,5).fill(0)
+      # => Cumo::Bit#shape=[4,5]
+      # [[0, 0, 0, 0, 0],
+      #  [0, 0, 0, 0, 0],
+      #  [0, 0, 0, 0, 0],
+      #  [0, 0, 0, 0, 0]]
 
-      a[1,true]=99
+      a[(0..-1)%2,(1..-1)%2] = 1
       a
-      => Cumo::DFloat#shape=[3,4]
-      [[0, 1, 2, 3],
-       [99, 99, 99, 99],
-       [8, 9, 10, 11]]
-
+      # => Cumo::Bit#shape=[4,5]
+      # [[0, 1, 0, 1, 0],
+      #  [0, 0, 0, 0, 0],
+      #  [0, 1, 0, 1, 0],
+      #  [0, 0, 0, 0, 0]]
 */
 static VALUE
 <%=c_func(-1)%>(int argc, VALUE *argv, VALUE self)

data/ext/cumo/narray/gen/tmpl_bit/binary.c

@@ -25,10 +25,8 @@ static void
             CUMO_STORE_BIT_STEP(a3, p3, s3, idx3, x);
         }
     } else {
-        o1 =  p1 % CUMO_NB;
-        o1 -= p3;
-        o2 =  p2 % CUMO_NB;
-        o2 -= p3;
+        o1 =  p1-p3;
+        o2 =  p2-p3;
         l1 =  CUMO_NB+o1;
         r1 =  CUMO_NB-o1;
         l2 =  CUMO_NB+o2;
@@ -58,23 +56,53 @@ static void
             }
         } else {
             for (; n>=CUMO_NB; n-=CUMO_NB) {
-                x = *a1>>o1;
-                if (o1<0)  x |= *(a1-1)>>l1;
-                if (o1>0)  x |= *(a1+1)<<r1;
+                if (o1==0) {
+                    x = *a1;
+                } else if (o1>0) {
+                    x = *a1>>o1  | *(a1+1)<<r1;
+                } else {
+                    x = *a1<<-o1 | *(a1-1)>>l1;
+                }
                 a1++;
-                y = *a2>>o2;
-                if (o2<0)  y |= *(a2-1)>>l2;
-                if (o2>0)  y |= *(a2+1)<<r2;
+                if (o2==0) {
+                    y = *a2;
+                } else if (o2>0) {
+                    y = *a2>>o2  | *(a2+1)<<r2;
+                } else {
+                    y = *a2<<-o2 | *(a2-1)>>l2;
+                }
                 a2++;
                 x = m_<%=name%>(x,y);
                 *(a3++) = x;
             }
         }
         if (n>0) {
-            x = *a1>>o1;
-            if (o1<0)  x |= *(a1-1)>>l1;
-            y = *a2>>o2;
-            if (o2<0)  y |= *(a2-1)>>l2;
+            if (o1==0) {
+                x = *a1;
+            } else if (o1>0) {
+                x = *a1>>o1;
+                if ((int)n>r1) {
+                    x |= *(a1+1)<<r1;
+                }
+            } else {
+                x = *(a1-1)>>l1;
+                if ((int)n>-o1) {
+                    x |= *a1<<-o1;
+                }
+            }
+            if (o2==0) {
+                y = *a2;
+            } else if (o2>0) {
+                y = *a2>>o2;
+                if ((int)n>r2) {
+                    y |= *(a2+1)<<r2;
+                }
+            } else {
+                y = *(a2-1)>>l2;
+                if ((int)n>-o2) {
+                    y |= *a2<<-o2;
+                }
+            }
             x = m_<%=name%>(x,y);
             *a3 = (x & CUMO_SLB(n)) | (*a3 & CUMO_BALL<<n);
         }

data/ext/cumo/narray/gen/tmpl_bit/bit_count.c

@@ -53,10 +53,15 @@ static VALUE
         return <%=c_func(-1)%>_cpu(argc, argv, self);
     } else {
         VALUE v, reduce;
+        cumo_narray_t *na;
         cumo_ndfunc_arg_in_t ain[3] = {{cT,0},{cumo_sym_reduce,0},{cumo_sym_init,0}};
         cumo_ndfunc_arg_out_t aout[1] = {{cumo_cUInt64,0}};
         cumo_ndfunc_t ndf = { <%=c_iter%>, CUMO_FULL_LOOP_NIP, 3, 1, ain, aout };
 
+        CumoGetNArray(self,na);
+        if (CUMO_NA_SIZE(na)==0) {
+            return INT2FIX(0);
+        }
         reduce = cumo_na_reduce_dimension(argc, argv, 1, &self, &ndf, 0);
         v = cumo_na_ndloop(&ndf, 3, self, reduce, INT2FIX(0));
         return v;

data/ext/cumo/narray/gen/tmpl_bit/bit_reduce.c

@@ -111,10 +111,15 @@ static VALUE
 <%=c_func(-1)%>(int argc, VALUE *argv, VALUE self)
 {
     VALUE v, reduce;
+    cumo_narray_t *na;
     cumo_ndfunc_arg_in_t ain[3] = {{cT,0},{cumo_sym_reduce,0},{cumo_sym_init,0}};
     cumo_ndfunc_arg_out_t aout[1] = {{cumo_cBit,0}};
     cumo_ndfunc_t ndf = {<%=c_iter%>, CUMO_FULL_LOOP_NIP, 3,1, ain,aout};
 
+    CumoGetNArray(self,na);
+    if (CUMO_NA_SIZE(na)==0) {
+        return INT2FIX(0);
+    }
     reduce = cumo_na_reduce_dimension(argc, argv, 1, &self, &ndf, 0);
     v = cumo_na_ndloop(&ndf, 3, self, reduce, INT2FIX(<%=init_bit%>));
     if (argc > 0) {

data/ext/cumo/narray/gen/tmpl_bit/mask.c

@@ -78,6 +78,10 @@ static void
 #define cIndex cumo_cInt32
 #endif
 
+static void shape_error(void) {
+    rb_raise(cumo_na_eShapeError,"mask and masked arrays must have the same shape");
+}
+
 /*
   Return subarray of argument masked with self bit array.
   @overload <%=op_map%>(array)
@@ -87,17 +91,33 @@ static void
 static VALUE
 <%=c_func(1)%>(VALUE mask, VALUE val)
 {
-    volatile VALUE idx_1, view;
+    int i;
+    VALUE idx_1, view;
     cumo_narray_data_t *nidx;
-    cumo_narray_view_t *nv;
-    cumo_narray_t      *na;
-    cumo_narray_view_t *na1;
+    cumo_narray_view_t *nv, *nv_val;
+    cumo_narray_t      *na, *na_mask;
     cumo_stridx_t stridx0;
     size_t n_1;
     where_opt_t g;
     cumo_ndfunc_arg_in_t ain[2] = {{cT,0},{Qnil,0}};
     cumo_ndfunc_t ndf = {<%=c_iter%>, CUMO_FULL_LOOP, 2, 0, ain, 0};
 
+    // cast val to NArray
+    if (!rb_obj_is_kind_of(val, cumo_cNArray)) {
+        val = rb_funcall(cumo_cNArray, cumo_id_cast, 1, val);
+    }
+    // shapes of mask and val must be same
+    CumoGetNArray(val, na);
+    CumoGetNArray(mask, na_mask);
+    if (na_mask->ndim != na->ndim) {
+        shape_error();
+    }
+    for (i=0; i<na->ndim; i++) {
+        if (na_mask->shape[i] != na->shape[i]) {
+            shape_error();
+        }
+    }
+
     // TODO(sonots): bit_count_true synchronizes with CPU. Avoid.
     n_1 = NUM2SIZET(<%=find_tmpl("count_true_cpu").c_func%>(0, NULL, mask));
     idx_1 = cumo_na_new(cIndex, 1, &n_1);
@@ -114,19 +134,19 @@ static VALUE
     CumoGetNArrayData(idx_1,nidx);
     CUMO_SDX_SET_INDEX(stridx0,(size_t*)nidx->ptr);
     nidx->ptr = NULL;
+    RB_GC_GUARD(idx_1);
 
     nv->stridx = ALLOC_N(cumo_stridx_t,1);
     nv->stridx[0] = stridx0;
     nv->offset = 0;
 
-    CumoGetNArray(val, na);
     switch(CUMO_NA_TYPE(na)) {
     case CUMO_NARRAY_DATA_T:
         nv->data = val;
         break;
     case CUMO_NARRAY_VIEW_T:
-        CumoGetNArrayView(val, na1);
-        nv->data = na1->data;
+        CumoGetNArrayView(val, nv_val);
+        nv->data = nv_val->data;
         break;
     default:
         rb_raise(rb_eRuntimeError,"invalid CUMO_NA_TYPE: %d",CUMO_NA_TYPE(na));

data/ext/cumo/narray/gen/tmpl_bit/store_bit.c

@@ -22,8 +22,7 @@ static void
             CUMO_STORE_BIT_STEP(a3, p3, s3, idx3, x);
         }
     } else {
-        o1 =  p1 % CUMO_NB;
-        o1 -= p3;
+        o1 =  p1-p3;
         l1 =  CUMO_NB+o1;
         r1 =  CUMO_NB-o1;
         if (p3>0 || n<CUMO_NB) {
@@ -44,16 +43,31 @@ static void
             }
         } else {
             for (; n>=CUMO_NB; n-=CUMO_NB) {
-                x = *a1>>o1;
-                if (o1<0)  x |= *(a1-1)>>l1;
-                if (o1>0)  x |= *(a1+1)<<r1;
+                if (o1==0) {
+                    x = *a1;
+                } else if (o1>0) {
+                    x = *a1>>o1  | *(a1+1)<<r1;
+                } else {
+                    x = *a1<<-o1 | *(a1-1)>>l1;
+                }
                 a1++;
                 *(a3++) = x;
             }
         }
         if (n>0) {
-            x = *a1>>o1;
-            if (o1<0)  x |= *(a1-1)>>l1;
+            if (o1==0) {
+                x = *a1;
+            } else if (o1>0) {
+                x = *a1>>o1;
+                if ((int)n>r1) {
+                    x |= *(a1+1)<<r1;
+                }
+            } else {
+                x = *(a1-1)>>l1;
+                if ((int)n>-o1) {
+                    x |= *a1<<-o1;
+                }
+            }
             *a3 = (x & CUMO_SLB(n)) | (*a3 & CUMO_BALL<<n);
         }
     }

data/ext/cumo/narray/gen/tmpl_bit/unary.c

@@ -24,8 +24,7 @@ static void
             CUMO_STORE_BIT_STEP(a3, p3, s3, idx3, y);
         }
     } else {
-        o1 =  p1 % CUMO_NB;
-        o1 -= p3;
+        o1 =  p1-p3;
         l1 =  CUMO_NB+o1;
         r1 =  CUMO_NB-o1;
         if (p3>0 || n<CUMO_NB) {
@@ -48,17 +47,32 @@ static void
             }
         } else {
             for (; n>=CUMO_NB; n-=CUMO_NB) {
-                x = *a1>>o1;
-                if (o1<0)  x |= *(a1-1)>>l1;
-                if (o1>0)  x |= *(a1+1)<<r1;
+                if (o1==0) {
+                    x = *a1;
+                } else if (o1>0) {
+                    x = *a1>>o1  | *(a1+1)<<r1;
+                } else {
+                    x = *a1<<-o1 | *(a1-1)>>l1;
+                }
                 a1++;
                 y = m_<%=name%>(x);
                 *(a3++) = y;
             }
         }
         if (n>0) {
-            x = *a1>>o1;
-            if (o1<0)  x |= *(a1-1)>>l1;
+            if (o1==0) {
+                x = *a1;
+            } else if (o1>0) {
+                x = *a1>>o1;
+                if ((int)n>r1) {
+                    x |= *(a1+1)<<r1;
+                }
+            } else {
+                x = *(a1-1)>>l1;
+                if ((int)n>-o1) {
+                    x |= *a1<<-o1;
+                }
+            }
             y = m_<%=name%>(x);
             *a3 = (y & CUMO_SLB(n)) | (*a3 & CUMO_BALL<<n);
         }