cumo 0.1.0

data/ext/cumo/narray/gen/tmpl/unary_kernel.cu

@@ -0,0 +1,72 @@
+<% if type_name == 'robject' || name == 'map' %>
+<% else %>
+__global__ void <%="cumo_#{c_iter}_index_index_kernel"%>(char *p1, char *p2, size_t *idx1, size_t *idx2, uint64_t n)
+{
+    for (uint64_t i = blockIdx.x * blockDim.x + threadIdx.x; i < n; i += blockDim.x * gridDim.x) {
+        *(dtype*)(p2 + idx2[i]) = m_<%=name%>(*(dtype*)(p1 + idx1[i]));
+    }
+}
+
+__global__ void <%="cumo_#{c_iter}_index_stride_kernel"%>(char *p1, char *p2, size_t *idx1, ssize_t s2, uint64_t n)
+{
+    for (uint64_t i = blockIdx.x * blockDim.x + threadIdx.x; i < n; i += blockDim.x * gridDim.x) {
+        *(dtype*)(p2 + (i * s2)) = m_<%=name%>(*(dtype*)(p1 + idx1[i]));
+    }
+}
+
+__global__ void <%="cumo_#{c_iter}_stride_index_kernel"%>(char *p1, char *p2, ssize_t s1, size_t *idx2, uint64_t n)
+{
+    for (uint64_t i = blockIdx.x * blockDim.x + threadIdx.x; i < n; i += blockDim.x * gridDim.x) {
+        *(dtype*)(p2 + idx2[i]) = m_<%=name%>(*(dtype*)(p1 + (i * s1)));
+    }
+}
+
+__global__ void <%="cumo_#{c_iter}_stride_stride_kernel"%>(char *p1, char *p2, ssize_t s1, ssize_t s2, uint64_t n)
+{
+    for (uint64_t i = blockIdx.x * blockDim.x + threadIdx.x; i < n; i += blockDim.x * gridDim.x) {
+        *(dtype*)(p2 + (i * s2)) = m_<%=name%>(*(dtype*)(p1 + (i * s1)));
+    }
+}
+
+__global__ void <%="cumo_#{c_iter}_contiguous_kernel"%>(char *p1, char *p2, uint64_t n)
+{
+    for (uint64_t i = blockIdx.x * blockDim.x + threadIdx.x; i < n; i += blockDim.x * gridDim.x) {
+        ((dtype*)p2)[i] = m_<%=name%>(((dtype*)p1)[i]);
+    }
+}
+
+void <%="cumo_#{c_iter}_index_index_kernel_launch"%>(char *p1, char *p2, size_t *idx1, size_t *idx2, uint64_t n)
+{
+    size_t gridDim = get_gridDim(n);
+    size_t blockDim = get_blockDim(n);
+    <%="cumo_#{c_iter}_index_index_kernel"%><<<gridDim, blockDim>>>(p1,p2,idx1,idx2,n);
+}
+
+void <%="cumo_#{c_iter}_index_stride_kernel_launch"%>(char *p1, char *p2, size_t *idx1, ssize_t s2, uint64_t n)
+{
+    size_t gridDim = get_gridDim(n);
+    size_t blockDim = get_blockDim(n);
+    <%="cumo_#{c_iter}_index_stride_kernel"%><<<gridDim, blockDim>>>(p1,p2,idx1,s2,n);
+}
+
+void <%="cumo_#{c_iter}_stride_index_kernel_launch"%>(char *p1, char *p2, ssize_t s1, size_t *idx2, uint64_t n)
+{
+    size_t gridDim = get_gridDim(n);
+    size_t blockDim = get_blockDim(n);
+    <%="cumo_#{c_iter}_stride_index_kernel"%><<<gridDim, blockDim>>>(p1,p2,s1,idx2,n);
+}
+
+void <%="cumo_#{c_iter}_stride_stride_kernel_launch"%>(char *p1, char *p2, ssize_t s1, ssize_t s2, uint64_t n)
+{
+    size_t gridDim = get_gridDim(n);
+    size_t blockDim = get_blockDim(n);
+    <%="cumo_#{c_iter}_stride_stride_kernel"%><<<gridDim, blockDim>>>(p1,p2,s1,s2,n);
+}
+
+void <%="cumo_#{c_iter}_contiguous_kernel_launch"%>(char *p1, char *p2, uint64_t n)
+{
+    size_t gridDim = get_gridDim(n);
+    size_t blockDim = get_blockDim(n);
+    <%="cumo_#{c_iter}_contiguous_kernel"%><<<gridDim, blockDim>>>(p1,p2,n);
+}
+<% end %>

data/ext/cumo/narray/gen/tmpl/unary_ret2.c

@@ -0,0 +1,34 @@
+static void
+<%=c_iter%>(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);
+    SHOW_SYNCHRONIZE_FIXME_WARNING_ONCE("<%=name%>", "<%=type_name%>");
+    for (; i--;) {
+        GET_DATA_STRIDE(p1,s1,dtype,x);
+        m_<%=name%>(x,y,z);
+        SET_DATA_STRIDE(p2,s2,dtype,y);
+        SET_DATA_STRIDE(p3,s3,dtype,z);
+    }
+}
+
+/*
+  <%=name%> of self.
+  @overload <%=name%>
+  @return [Cumo::<%=real_class_name%>] <%=name%> of self.
+*/
+static VALUE
+<%=c_func(0)%>(VALUE self)
+{
+    ndfunc_arg_in_t ain[1] = {{cT,0}};
+    ndfunc_arg_out_t aout[2] = {{cT,0},{cT,0}};
+    ndfunc_t ndf = {<%=c_iter%>, STRIDE_LOOP, 1,2, ain,aout};
+
+    return na_ndloop(&ndf, 1, self);
+}

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

@@ -0,0 +1,86 @@
+<% unless type_name == 'robject' %>
+void <%="cumo_#{c_iter}_index_index_kernel_launch"%>(char *p1, char *p2, size_t *idx1, size_t *idx2, uint64_t n);
+void <%="cumo_#{c_iter}_index_stride_kernel_launch"%>(char *p1, char *p2, size_t *idx1, ssize_t s2, uint64_t n);
+void <%="cumo_#{c_iter}_stride_index_kernel_launch"%>(char *p1, char *p2, ssize_t s1, size_t *idx2, uint64_t n);
+void <%="cumo_#{c_iter}_stride_stride_kernel_launch"%>(char *p1, char *p2, ssize_t s1, ssize_t s2, uint64_t n);
+<% end %>
+
+static void
+<%=c_iter%>(na_loop_t *const lp)
+{
+    size_t  i;
+    char   *p1, *p2;
+    ssize_t s1, s2;
+    size_t *idx1, *idx2;
+    INIT_COUNTER(lp, i);
+    INIT_PTR_IDX(lp, 0, p1, s1, idx1);
+    INIT_PTR_IDX(lp, 1, p2, s2, idx2);
+
+    <% if type_name == 'robject' %>
+    {
+        dtype x;
+        SHOW_SYNCHRONIZE_FIXME_WARNING_ONCE("<%=name%>", "<%=type_name%>");
+        if (idx1) {
+            if (idx2) {
+                for (; i--;) {
+                    GET_DATA_INDEX(p1,idx1,dtype,x);
+                    x = m_<%=name%>(x);
+                    SET_DATA_INDEX(p2,idx2,dtype,x);
+                }
+            } else {
+                for (; i--;) {
+                    GET_DATA_INDEX(p1,idx1,dtype,x);
+                    x = m_<%=name%>(x);
+                    SET_DATA_STRIDE(p2,s2,dtype,x);
+                }
+            }
+        } else {
+            if (idx2) {
+                for (; i--;) {
+                    GET_DATA_STRIDE(p1,s1,dtype,x);
+                    x = m_<%=name%>(x);
+                    SET_DATA_INDEX(p2,idx2,dtype,x);
+                }
+            } else {
+                for (; i--;) {
+                    GET_DATA_STRIDE(p1,s1,dtype,x);
+                    x = m_<%=name%>(x);
+                    SET_DATA_STRIDE(p2,s2,dtype,x);
+                }
+            }
+        }
+    }
+    <% else %>
+    {
+        if (idx1) {
+            if (idx2) {
+                <%="cumo_#{c_iter}_index_index_kernel_launch"%>(p1,p2,idx1,idx2,i);
+            } else {
+                <%="cumo_#{c_iter}_index_stride_kernel_launch"%>(p1,p2,idx1,s2,i);
+            }
+        } else {
+            if (idx2) {
+                <%="cumo_#{c_iter}_stride_index_kernel_launch"%>(p1,p2,s1,idx2,i);
+            } else {
+                <%="cumo_#{c_iter}_stride_stride_kernel_launch"%>(p1,p2,s1,s2,i);
+            }
+        }
+    }
+    <% end %>
+}
+
+/*
+  Calculate <%=name%>(x).
+  @overload <%=name%>(x)
+  @param [Cumo::NArray,Numeric] x  input value
+  @return [Cumo::<%=class_name%>] result of <%=name%>(x).
+*/
+static VALUE
+<%=c_func(1)%>(VALUE mod, VALUE a1)
+{
+    ndfunc_arg_in_t ain[1] = {{cT,0}};
+    ndfunc_arg_out_t aout[1] = {{cT,0}};
+    ndfunc_t ndf = { <%=c_iter%>, FULL_LOOP, 1, 1, ain, aout };
+
+    return na_ndloop(&ndf, 1, a1);
+}

data/ext/cumo/narray/gen/tmpl/unary_s_kernel.cu

@@ -0,0 +1,58 @@
+<% unless type_name == 'robject' %>
+__global__ void <%="cumo_#{c_iter}_index_index_kernel"%>(char *p1, char *p2, size_t *idx1, size_t *idx2, uint64_t n)
+{
+    for (uint64_t i = blockIdx.x * blockDim.x + threadIdx.x; i < n; i += blockDim.x * gridDim.x) {
+        *(dtype*)(p2 + idx2[i]) = m_<%=name%>(*(dtype*)(p1 + idx1[i]));
+    }
+}
+
+__global__ void <%="cumo_#{c_iter}_index_stride_kernel"%>(char *p1, char *p2, size_t *idx1, ssize_t s2, uint64_t n)
+{
+    for (uint64_t i = blockIdx.x * blockDim.x + threadIdx.x; i < n; i += blockDim.x * gridDim.x) {
+        *(dtype*)(p2 + (i * s2)) = m_<%=name%>(*(dtype*)(p1 + idx1[i]));
+    }
+}
+
+__global__ void <%="cumo_#{c_iter}_stride_index_kernel"%>(char *p1, char *p2, ssize_t s1, size_t *idx2, uint64_t n)
+{
+    for (uint64_t i = blockIdx.x * blockDim.x + threadIdx.x; i < n; i += blockDim.x * gridDim.x) {
+        *(dtype*)(p2 + idx2[i]) = m_<%=name%>(*(dtype*)(p1 + (i * s1)));
+    }
+}
+
+__global__ void <%="cumo_#{c_iter}_stride_stride_kernel"%>(char *p1, char *p2, ssize_t s1, ssize_t s2, uint64_t n)
+{
+    for (uint64_t i = blockIdx.x * blockDim.x + threadIdx.x; i < n; i += blockDim.x * gridDim.x) {
+        *(dtype*)(p2 + (i * s2)) = m_<%=name%>(*(dtype*)(p1 + (i * s1)));
+    }
+}
+
+void <%="cumo_#{c_iter}_index_index_kernel_launch"%>(char *p1, char *p2, size_t *idx1, size_t *idx2, uint64_t n)
+{
+    size_t gridDim = get_gridDim(n);
+    size_t blockDim = get_blockDim(n);
+    <%="cumo_#{c_iter}_index_index_kernel"%><<<gridDim, blockDim>>>(p1,p2,idx1,idx2,n);
+}
+
+void <%="cumo_#{c_iter}_index_stride_kernel_launch"%>(char *p1, char *p2, size_t *idx1, ssize_t s2, uint64_t n)
+{
+    size_t gridDim = get_gridDim(n);
+    size_t blockDim = get_blockDim(n);
+    <%="cumo_#{c_iter}_index_stride_kernel"%><<<gridDim, blockDim>>>(p1,p2,idx1,s2,n);
+}
+
+void <%="cumo_#{c_iter}_stride_index_kernel_launch"%>(char *p1, char *p2, ssize_t s1, size_t *idx2, uint64_t n)
+{
+    size_t gridDim = get_gridDim(n);
+    size_t blockDim = get_blockDim(n);
+    <%="cumo_#{c_iter}_stride_index_kernel"%><<<gridDim, blockDim>>>(p1,p2,s1,idx2,n);
+}
+
+void <%="cumo_#{c_iter}_stride_stride_kernel_launch"%>(char *p1, char *p2, ssize_t s1, ssize_t s2, uint64_t n)
+{
+    size_t gridDim = get_gridDim(n);
+    size_t blockDim = get_blockDim(n);
+    <%="cumo_#{c_iter}_stride_stride_kernel"%><<<gridDim, blockDim>>>(p1,p2,s1,s2,n);
+}
+
+<% end %>

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

@@ -0,0 +1,24 @@
+static VALUE
+<%=c_func(0)%>(VALUE self)
+{
+    narray_t *na;
+    char *ptr;
+
+    GetNArray(self,na);
+
+    switch(NA_TYPE(na)) {
+    case NARRAY_DATA_T:
+        ptr = NA_DATA_PTR(na);
+        if (na->size > 0 && ptr == NULL) {
+            ptr = cumo_cuda_runtime_malloc(((na->size-1)/8/sizeof(BIT_DIGIT)+1)*sizeof(BIT_DIGIT));
+            NA_DATA_PTR(na) = ptr;
+        }
+        break;
+    case NARRAY_VIEW_T:
+        rb_funcall(NA_VIEW_DATA(na), rb_intern("allocate"), 0);
+        break;
+    default:
+        rb_raise(rb_eRuntimeError,"invalid narray type");
+    }
+    return self;
+}

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

@@ -0,0 +1,54 @@
+static VALUE
+<%=c_func(-1)%>_cpu(int argc, VALUE *argv, VALUE self);
+
+/*
+  Array element referenece or slice view.
+  @overload [](dim0,...,dimL)
+  @param [Numeric,Range,etc] dim0,...,dimL  Multi-dimensional Index.
+  @return [Numeric,NArray::<%=class_name%>] Element object 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.
+
+  @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]]
+ */
+static VALUE
+<%=c_func(-1)%>(int argc, VALUE *argv, VALUE self)
+{
+    if (cumo_compatible_mode_enabled_p()) {
+        return <%=c_func(-1)%>_cpu(argc, argv, self);
+    } else {
+        int result_nd;
+        size_t pos;
+
+        result_nd = na_get_result_dimension(self, argc, argv, 1, &pos);
+        return na_aref_main(argc, argv, self, 0, result_nd, pos);
+    }
+}

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

@@ -0,0 +1,57 @@
+/*
+  Array element referenece or slice view.
+  @overload [](dim0,...,dimL)
+  @param [Numeric,Range,etc] dim0,...,dimL  Multi-dimensional Index.
+  @return [Numeric,NArray::<%=class_name%>] Element object 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.
+
+  @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]]
+ */
+static VALUE
+<%=c_func(-1)%>(int argc, VALUE *argv, VALUE self)
+{
+    int nd;
+    size_t pos;
+    char *ptr;
+    dtype x;
+
+    nd = na_get_result_dimension(self, argc, argv, 1, &pos);
+    if (nd) {
+        return na_aref_main(argc, argv, self, 0, nd, pos);
+    } else {
+        ptr = na_get_pointer_for_read(self);
+        SHOW_SYNCHRONIZE_WARNING_ONCE("<%=name%>", "<%=type_name%>");
+        cumo_cuda_runtime_check_status(cudaDeviceSynchronize());
+        LOAD_BIT(ptr,pos,x);
+        return m_data_to_num(x);
+    }
+}

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

@@ -0,0 +1,56 @@
+/*
+  Array element(s) set.
+  @overload []=(dim0,..,dimL,val)
+  @param [Numeric,Range,etc] dim0,..,dimL  Multi-dimensional Index.
+  @param [Numeric,Cumo::NArray,etc] val  Value(s) to be set to self.
+  @return [Numeric] returns val (last argument).
+
+  --- Replace element(s) at +dim0+, +dim1+, ... (index/range/array/true
+  for each dimention). 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]]
+
+      a[1,[0,2]] = [101,102]
+      a
+      => Cumo::DFloat#shape=[3,4]
+      [[0, 1, 2, 3],
+       [101, 5, 102, 7],
+       [8, 9, 10, 11]]
+
+      a[1,true]=99
+      a
+      => Cumo::DFloat#shape=[3,4]
+      [[0, 1, 2, 3],
+       [99, 99, 99, 99],
+       [8, 9, 10, 11]]
+
+*/
+static VALUE
+<%=c_func(-1)%>(int argc, VALUE *argv, VALUE self)
+{
+    int nd;
+    size_t pos;
+    VALUE a;
+
+    argc--;
+    if (argc==0) {
+        <%=c_func.sub(/_aset/,"_store")%>(self, argv[argc]);
+    } else {
+        nd = na_get_result_dimension(self, argc, argv, 1, &pos);
+        a = na_aref_main(argc, argv, self, 0, nd, pos);
+        <%=c_func.sub(/_aset/,"_store")%>(a, argv[argc]);
+    }
+    return argv[argc];
+}