RubyGems - carray - Versions diffs - 3.0.4 → 3.0.5 - Mend

carray 3.0.4 → 3.0.5

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (111) hide show

checksums.yaml +4 -4
data/CHANGELOG.md +205 -0
data/carray.gemspec +1 -1
data/ext/ca_array_pool.c +25 -1
data/ext/ca_axis_dispatch.c +13 -20
data/ext/ca_categorical_iterator.c +27 -20
data/ext/ca_compare.c +4 -4
data/ext/ca_composite_dispatch.c +9 -9
data/ext/ca_kernel_iterator.c +6 -6
data/ext/ca_kernel_iterator.h +112 -83
data/ext/ca_obj_array.c +69 -19
data/ext/ca_obj_bincmp.c +1 -0
data/ext/ca_obj_binop.c +1 -0
data/ext/ca_obj_bitarray.c +5 -1
data/ext/ca_obj_bitfield.c +58 -43
data/ext/ca_obj_byte_swap.c +2 -1
data/ext/ca_obj_const_string.c +15 -3
data/ext/ca_obj_face.c +42 -1
data/ext/ca_obj_field.c +13 -3
data/ext/ca_obj_grid.c +52 -5
data/ext/ca_obj_monop.c +46 -10
data/ext/ca_obj_object.c +64 -12
data/ext/ca_obj_record.c +3 -0
data/ext/ca_obj_reduce.c +4 -0
data/ext/ca_obj_refer.c +18 -9
data/ext/ca_obj_roll.c +26 -6
data/ext/ca_obj_select.c +34 -5
data/ext/ca_obj_shift.c +31 -7
data/ext/ca_obj_stack.c +1 -0
data/ext/ca_obj_stride.c +28 -19
data/ext/ca_obj_string.c +11 -2
data/ext/ca_obj_tile.c +27 -7
data/ext/ca_obj_transpose.c +8 -4
data/ext/ca_obj_triop.c +3 -0
data/ext/ca_obj_window.c +79 -50
data/ext/ca_op_ipower.c +4 -4
data/ext/ca_sort_kernels.h +6 -3
data/ext/carray.h +33 -0
data/ext/carray_access.c +16 -3
data/ext/carray_attribute.c +24 -0
data/ext/carray_cast.c +243 -82
data/ext/carray_cast_func.rb +56 -26
data/ext/carray_conversion.c +2 -2
data/ext/carray_core.c +3 -0
data/ext/carray_count.c +6 -0
data/ext/carray_element.c +9 -4
data/ext/carray_factorize.c +20 -20
data/ext/carray_generate.c +8 -2
data/ext/carray_histogram.c +1 -1
data/ext/carray_index_classifier.c +4 -4
data/ext/carray_internal.h +12 -0
data/ext/carray_kernels_bincmp.c +168 -168
data/ext/carray_kernels_binop.c +900 -900
data/ext/carray_kernels_init.c +4 -4
data/ext/carray_kernels_moncmp.c +96 -96
data/ext/carray_kernels_monop.c +1984 -1984
data/ext/carray_kernels_reduce_aggregate.c +1733 -1733
data/ext/carray_kernels_reduce_cumulative.c +642 -642
data/ext/carray_kernels_reduce_extreme.c +1150 -1150
data/ext/carray_kernels_reduce_variance.c +258 -258
data/ext/carray_kernels_scan.c +128 -128
data/ext/carray_kernels_search.c +340 -340
data/ext/carray_kernels_sort.c +295 -304
data/ext/carray_kernels_triop.c +114 -114
data/ext/carray_key_windows.c +4 -0
data/ext/carray_mask.c +45 -22
data/ext/carray_memory_view.c +2 -2
data/ext/carray_operator.c +5 -5
data/ext/carray_order.c +227 -8
data/ext/carray_random.c +18 -18
data/ext/carray_scatter.c +24 -7
data/ext/carray_slab.c +32 -9
data/ext/carray_sort.c +224 -16
data/ext/carray_sort_kernel.c +5 -7
data/ext/carray_struct.c +35 -43
data/ext/carray_time_text.c +4 -2
data/ext/carray_utils.c +63 -21
data/ext/mkkernel.rb +210 -115
data/ext/version.h +4 -4
data/lib/carray/attribute.rb +17 -4
data/lib/carray/basics.rb +27 -7
data/lib/carray/categorical.rb +30 -16
data/lib/carray/categorical_iterator.rb +6 -6
data/lib/carray/const_string.rb +10 -0
data/lib/carray/data_type_extension.rb +8 -1
data/lib/carray/fixlen_string.rb +41 -2
data/lib/carray/frame/column_verbs.rb +4 -3
data/lib/carray/frame/concat.rb +17 -3
data/lib/carray/frame/frame.rb +6 -7
data/lib/carray/frame/group.rb +100 -13
data/lib/carray/frame/io.rb +3 -1
data/lib/carray/fuse_source.rb +31 -2
data/lib/carray/lazy.rb +35 -0
data/lib/carray/methods/is_in.rb +8 -1
data/lib/carray/segment_iterator.rb +12 -8
data/lib/carray/serialize.rb +19 -2
data/lib/carray/stack.rb +15 -12
data/lib/carray/string.rb +8 -0
data/lib/carray/string_operation_extension.rb +13 -5
data/lib/carray/struct_builder.rb +5 -3
data/lib/carray/time.rb +107 -68
data/yard-stubs/ca_obj_transpose.rb +5 -4
data/yard-stubs/carray_attribute.rb +3 -1
data/yard-stubs/carray_cast.rb +2 -0
data/yard-stubs/carray_conversion.rb +5 -1
data/yard-stubs/carray_element.rb +7 -6
data/yard-stubs/carray_mask.rb +8 -3
data/yard-stubs/carray_math.rb +14 -4
data/yard-stubs/carray_order.rb +100 -4
data/yard-stubs/carray_scatter.rb +7 -5
metadata +1 -1

data/ext/carray_element.c CHANGED Viewed

@@ -20,17 +20,22 @@
 #include "ca_obj_face.h"
 #include "carray_internal.h"   /* ca_obj_extremum_replaces */
-/* An element index is an Integer, as it is for `[]`; a Float would be cut
-   to an integer by NUM2SIZE. */
+/* An element index is read as `[]` reads a key: an Integer, or a Float
+   with no fraction. */
 static inline ca_size_t
 elem_index_value (VALUE v)
 {
+  ca_size_t k;
   if ( FIXNUM_P(v) ) return (ca_size_t) FIX2LONG(v);
-  if ( ! RB_INTEGER_TYPE_P(v) ) {
+  switch ( ca_index_key_position(v, &k) ) {
+  case 1:
+    return k;
+  case 2:
+    return CA_LENGTH_MAX;   /* past int64: past every axis, refused as out of range */
+  default:
     rb_raise(rb_eIndexError, "object '%"PRIsVALUE"' is invalid for an index",
              rb_inspect(v));
   }
-  return NUM2SIZE(v);
 }
 /* T_ARRAY index parser.  ridx is assumed already known to be T_ARRAY.

data/ext/carray_factorize.c CHANGED Viewed

@@ -865,7 +865,7 @@ fz_factorize_appearance_body (VALUE arg)
      element to a 64-bit key; equality within one data type is preserved. */
   #define FZ_LOOP(T, WIDEN)                                                  \
     do {                                                                     \
-      const T  *ip = (const T *) p_in;                                       \
+      const CA_U(T)  *ip = (const CA_U(T) *) p_in;                                       \
       uint32_t *op = (uint32_t *) p_out;                                     \
       for ( ca_size_t i = 0; i < n; i++ ) {                                  \
         if ( m && m[i] ) { op[i] = 0xFFFFFFFFu; continue; }                  \
@@ -884,7 +884,7 @@ fz_factorize_appearance_body (VALUE arg)
      kept as-is), so masked source cells store the sentinel. */
   #define FZ_LOOP_FLOAT(T)                                     \
     do {                                                                     \
-      const T  *ip = (const T *) p_in;                                       \
+      const CA_U(T)  *ip = (const CA_U(T) *) p_in;                                       \
       uint32_t *op = (uint32_t *) p_out;                                     \
       for ( ca_size_t i = 0; i < n; i++ ) {                                  \
         if ( m && m[i] ) { op[i] = 0xFFFFFFFFu; continue; }                  \
@@ -1073,7 +1073,7 @@ fz_mask_duplicates_body (VALUE arg)
      interns first appearances, so is_new == 0 flags a duplicate. */
   #define MD_LOOP(T, WIDEN)                                                   \
     do {                                                                      \
-      const T    *ip = (const T *) p_in;                                      \
+      const CA_U(T)    *ip = (const CA_U(T) *) p_in;                                      \
       boolean8_t *op = (boolean8_t *) p_out;                                  \
       for ( ca_size_t i = 0; i < n; i++ ) {                                   \
         if ( m && m[i] ) { op[i] = 0; continue; }                            \
@@ -1091,7 +1091,7 @@ fz_mask_duplicates_body (VALUE arg)
      before the bitwise key. */
   #define MD_LOOP_FLOAT(T)                                      \
     do {                                                                      \
-      const T    *ip = (const T *) p_in;                                      \
+      const CA_U(T)    *ip = (const CA_U(T) *) p_in;                                      \
       boolean8_t *op = (boolean8_t *) p_out;                                  \
       for ( ca_size_t i = 0; i < n; i++ ) {                                   \
         if ( m && m[i] ) { op[i] = 0; continue; }                            \
@@ -1225,7 +1225,7 @@ fz_unique_flat_body (VALUE arg)
   #define UQ_LOOP(T, WIDEN)                                                   \
     do {                                                                      \
-      const T *ip = (const T *) p_in;                                         \
+      const CA_U(T) *ip = (const CA_U(T) *) p_in;                                         \
       for ( ca_size_t i = 0; i < n; i++ ) {                                   \
         if ( m && m[i] ) { continue; }                                        \
         uint64_t key = (uint64_t) (WIDEN ip[i]);                              \
@@ -1239,7 +1239,7 @@ fz_unique_flat_body (VALUE arg)
      pushed level is the raw first-seen element. */
   #define UQ_LOOP_FLOAT(T)                                      \
     do {                                                                      \
-      const T *ip = (const T *) p_in;                                         \
+      const CA_U(T) *ip = (const CA_U(T) *) p_in;                                         \
       for ( ca_size_t i = 0; i < n; i++ ) {                                   \
         if ( m && m[i] ) { continue; }                                        \
         T v = ip[i];                                                          \
@@ -1340,7 +1340,7 @@ fz_intern_all (ca_iter_state *st_v, fz_hash *h, CArray *cv, fz_levels *lv)
   #define FZ_IA(T, WIDEN)                                                      \
     do {                                                                       \
-      const T *ip = (const T *) p_v;                                           \
+      const CA_U(T) *ip = (const CA_U(T) *) p_v;                                           \
       for ( ca_size_t i = 0; i < nv; i++ ) {                                   \
         if ( mv && mv[i] ) { continue; }                                       \
         uint64_t key = (uint64_t) (WIDEN ip[i]);                             \
@@ -1352,7 +1352,7 @@ fz_intern_all (ca_iter_state *st_v, fz_hash *h, CArray *cv, fz_levels *lv)
   #define FZ_IA_FLOAT(T)                                         \
     do {                                                                       \
-      const T *ip = (const T *) p_v;                                           \
+      const CA_U(T) *ip = (const CA_U(T) *) p_v;                                           \
       for ( ca_size_t i = 0; i < nv; i++ ) {                                   \
         if ( mv && mv[i] ) { continue; }                                       \
         T v = ip[i];                                                           \
@@ -1493,7 +1493,7 @@ fz_is_in_body (VALUE arg)
     #define IN_PROBE(T, WIDEN)                                                  \
       do {                                                                      \
-        const T    *ip = (const T *) p_in;                                      \
+        const CA_U(T)    *ip = (const CA_U(T) *) p_in;                                      \
         boolean8_t *op = (boolean8_t *) p_out;                                  \
         for ( ca_size_t i = 0; i < n; i++ ) {                                   \
           if ( m && m[i] ) { op[i] = 0; continue; }                            \
@@ -1504,7 +1504,7 @@ fz_is_in_body (VALUE arg)
     #define IN_PROBE_FLOAT(T)                                     \
       do {                                                                      \
-        const T    *ip = (const T *) p_in;                                      \
+        const CA_U(T)    *ip = (const CA_U(T) *) p_in;                                      \
         boolean8_t *op = (boolean8_t *) p_out;                                  \
         for ( ca_size_t i = 0; i < n; i++ ) {                                   \
           if ( m && m[i] ) { op[i] = 0; continue; }                            \
@@ -1661,7 +1661,7 @@ fz_locate_addr_body (VALUE arg)
     #define LOC_BUILD(T, WIDEN)                                                 \
       do {                                                                      \
-        const T *ip = (const T *) p_r;                                          \
+        const CA_U(T) *ip = (const CA_U(T) *) p_r;                                          \
         for ( ca_size_t i = 0; i < nr; i++ ) {                                  \
           if ( mr && mr[i] ) { continue; }                                     \
           uint64_t key = (uint64_t) (WIDEN ip[i]);                            \
@@ -1674,7 +1674,7 @@ fz_locate_addr_body (VALUE arg)
     #define LOC_BUILD_FLOAT(T)                                    \
       do {                                                                      \
-        const T *ip = (const T *) p_r;                                          \
+        const CA_U(T) *ip = (const CA_U(T) *) p_r;                                          \
         for ( ca_size_t i = 0; i < nr; i++ ) {                                  \
           if ( mr && mr[i] ) { continue; }                                     \
           T v = ip[i];                                                          \
@@ -1751,7 +1751,7 @@ fz_locate_addr_body (VALUE arg)
     #define LOC_PROBE(T, WIDEN)                                                 \
       do {                                                                      \
-        const T *ip = (const T *) p_in;                                         \
+        const CA_U(T) *ip = (const CA_U(T) *) p_in;                                         \
         const int64_t *ab = (const int64_t *) addr.p;                           \
         for ( ca_size_t i = 0; i < n; i++ ) {                                   \
           ca_size_t o = base + i;                                               \
@@ -1765,7 +1765,7 @@ fz_locate_addr_body (VALUE arg)
     #define LOC_PROBE_FLOAT(T)                                    \
       do {                                                                      \
-        const T *ip = (const T *) p_in;                                         \
+        const CA_U(T) *ip = (const CA_U(T) *) p_in;                                         \
         const int64_t *ab = (const int64_t *) addr.p;                           \
         for ( ca_size_t i = 0; i < n; i++ ) {                                   \
           ca_size_t o = base + i;                                               \
@@ -1929,7 +1929,7 @@ fz_set_relation_body (VALUE arg)
   #define SR(T, WIDEN)                                                         \
     do {                                                                       \
-      const T *ip = (const T *) p;                                             \
+      const CA_U(T) *ip = (const CA_U(T) *) p;                                             \
       for ( ca_size_t i = 0; i < n; i++ ) {                                    \
         if ( m && m[i] ) { continue; }                                         \
         uint64_t key = (uint64_t) (WIDEN ip[i]);                             \
@@ -1943,7 +1943,7 @@ fz_set_relation_body (VALUE arg)
   #define SR_FLOAT(T)                                            \
     do {                                                                       \
-      const T *ip = (const T *) p;                                             \
+      const CA_U(T) *ip = (const CA_U(T) *) p;                                             \
       for ( ca_size_t i = 0; i < n; i++ ) {                                    \
         if ( m && m[i] ) { continue; }                                         \
         T v = ip[i];                                                           \
@@ -2188,7 +2188,7 @@ fz_value_counts_flat_body (VALUE arg)
   #define VC_LOOP(T, WIDEN)                                                   \
     do {                                                                      \
-      const T *ip = (const T *) p_in;                                         \
+      const CA_U(T) *ip = (const CA_U(T) *) p_in;                                         \
       for ( ca_size_t i = 0; i < n; i++ ) {                                   \
         if ( m && m[i] ) { continue; }                                        \
         uint64_t key = (uint64_t) (WIDEN ip[i]);                              \
@@ -2206,7 +2206,7 @@ fz_value_counts_flat_body (VALUE arg)
   #define VC_LOOP_FLOAT(T)                                      \
     do {                                                                      \
-      const T *ip = (const T *) p_in;                                         \
+      const CA_U(T) *ip = (const CA_U(T) *) p_in;                                         \
       for ( ca_size_t i = 0; i < n; i++ ) {                                   \
         if ( m && m[i] ) { continue; }                                        \
         T v = ip[i];                                                          \
@@ -2516,7 +2516,7 @@ fz_is_mode_body (VALUE arg)
   /* Integer: WIDEN gives the key directly.  Two passes over the fiber. */
   #define IM_BODY(T, WIDEN)                                                   \
     do {                                                                      \
-      const T    *ip = (const T *) p_in;                                      \
+      const CA_U(T)    *ip = (const CA_U(T) *) p_in;                                      \
       boolean8_t *op = (boolean8_t *) p_out;                                  \
       for ( ca_size_t i = 0; i < n; i++ ) {                                   \
         if ( m && m[i] ) { continue; }                                        \
@@ -2543,7 +2543,7 @@ fz_is_mode_body (VALUE arg)
   /* Float: NaN collapses to one canonical key; -0.0 / +0.0 share a key. */
   #define IM_BODY_FLOAT(T)                                      \
     do {                                                                      \
-      const T    *ip = (const T *) p_in;                                      \
+      const CA_U(T)    *ip = (const CA_U(T) *) p_in;                                      \
       boolean8_t *op = (boolean8_t *) p_out;                                  \
       for ( int pass = 0; pass < 2; pass++ ) {                                \
         int64_t mx = 0;                                                       \

data/ext/carray_generate.c CHANGED Viewed

@@ -117,7 +117,7 @@ ca_seq_is_real (VALUE v)
 /* Fill the whole array in row-major order with `step*i + offset`. */
 #define proc_seq_bang(type, from, to)       \
   {                                         \
-    type *p = (type *)ca->ptr;              \
+    CA_U(type) *p = (CA_U(type) *)ca->ptr;              \
     ca_size_t i;                              \
     if ( NIL_P(roffset) && NIL_P(rstep) ) { \
       for (i=0; i<ca->elements; i++) {    \
@@ -149,7 +149,7 @@ ca_seq_is_real (VALUE v)
    `outer_size` come from the enclosing scope (ca_seq_geometry). */
 #define proc_seq_bang_axis(type, from, to)                              \
   {                                                                     \
-    type *p = (type *)ca->ptr;                                          \
+    CA_U(type) *p = (CA_U(type) *)ca->ptr;                                          \
     ca_size_t o, k, in;                                                 \
     if ( ca_seq_is_real(rstep) || ca_seq_is_real(roffset) ) {           \
       double offset = (NIL_P(roffset)) ? 0 : NUM2DBL(roffset);          \
@@ -289,6 +289,9 @@ ca_seq_bang_object_body (VALUE arg)
     }
     else if ( ! NIL_P(rmethod) ) {                /* a.seq(obj, :method) */
       ID id_method = SYM2ID(rmethod);
+      if ( ca->elements == 0 ) {                  /* no first cell to write */
+        return Qnil;
+      }
       *p++ = ca_seq_cell(c, rval = roffset);
       for (i=1; i<ca->elements; i++) {
         *p++ = ca_seq_cell(c, rval = rb_funcall(rval, id_method, 0));
@@ -297,6 +300,9 @@ ca_seq_bang_object_body (VALUE arg)
     else {                                        /* a.seq(obj, step) */
       ID id_plus = rb_intern("+");
       rstep   = ( NIL_P(rstep) ) ? INT2NUM(1) : rstep;
+      if ( ca->elements == 0 ) {                  /* no first cell to write */
+        return Qnil;
+      }
       *p++ = ca_seq_cell(c, rval = roffset);
       for (i=1; i<ca->elements; i++) {
         *p++ = ca_seq_cell(c, rval = rb_funcall(rval, id_plus, 1, rstep));

data/ext/carray_histogram.c CHANGED Viewed

@@ -312,7 +312,7 @@ histogram_scatter_walk (VALUE arg)
       ca_size_t off = fiber_base;
       int       masked = 0;
       for ( kk = 0; kk < (ca_size_t) M; kk++ ) {
-        double v = *(const double *)(p + a * ss + kk * cs);
+        double v = *(const ca_u_float64_t *)(p + a * ss + kk * cs);
         if ( (m && m[a * sms + kk * cms]) || isnan(v) ) { masked = 1; break; }
         off += histbin_index(&axes[kk], v) * ext_stride[kk];
       }

data/ext/carray_index_classifier.c CHANGED Viewed

@@ -164,8 +164,8 @@ ca_classifier_key_too_large (VALUE v, const char *where)
    Returns 1 with the position in *out, 2 for an Integer past int64 (past
    every axis, which is no longer than int64), and 0 for anything else (a
    Float with a fraction or past 2**53, NaN, Inf, a Rational, a String). */
-static int
-ca_classifier_key_position (VALUE v, ca_size_t *out)
+int
+ca_index_key_position (VALUE v, ca_size_t *out)
 {
   if ( FIXNUM_P(v) ) {
     *out = (ca_size_t) FIX2LONG(v);
@@ -201,7 +201,7 @@ static ca_size_t
 ca_classifier_key_int_where (VALUE v, int axis, const char *where)
 {
   ca_size_t k;
-  switch ( ca_classifier_key_position(v, &k) ) {
+  switch ( ca_index_key_position(v, &k) ) {
   case 1:
     return k;
   case 2:
@@ -232,7 +232,7 @@ static ca_size_t
 ca_classifier_key_step (VALUE v, int axis)
 {
   ca_size_t k;
-  if ( ca_classifier_key_position(v, &k) == 2 ) {
+  if ( ca_index_key_position(v, &k) == 2 ) {
     return rb_big_sign(v) ? (ca_size_t) INT64_MAX : (ca_size_t) INT64_MIN;
   }
   return ca_classifier_key_int(v, axis);

data/ext/carray_internal.h CHANGED Viewed

@@ -393,6 +393,10 @@ VALUE    rb_ary_flatten_for_elements (VALUE ary, ca_size_t elements, void *ap);
 const char *ca_calling_method_name (void);
 VALUE       ca_answer_as_caller (VALUE (*body)(VALUE), VALUE arg);
 long        ca_integer_arg (VALUE v, const char *arg, const char *name);
+/* 1 with the position of an index key (an Integer, or a Float with no
+   fraction while exact), 2 for an Integer past int64, 0 otherwise. */
+int         ca_index_key_position (VALUE v, ca_size_t *out);
+void        ca_refuse_without_common_type (VALUE self, VALUE other);
 ca_size_t   ca_shape_elements (int8_t ndim, const ca_size_t *dim, ca_size_t bytes);
 void        ca_check_index_array (VALUE v);
 VALUE       ca_order_key (VALUE v, const char *name);
@@ -414,6 +418,13 @@ NORETURN(void ca_op_refuse_unimplemented (VALUE operand, int8_t data_type,
 void        ca_check_rank_fits (long ndim, const char *name);
 /* The checks of carray.h inside the library, each naming the method the
    user called; the public ones stay as they are for extensions. */
+/* A wrap of a caller's buffer lent to a CAObject hook (CA_FLAG_LENT), and
+   the mark that a view was taken of it while it was lent
+   (CA_FLAG_LENT_VIEWED).  rb_ca_set_parent sets the second; the hook's
+   caller reads both when it takes the buffer back (ca_obj_object.c). */
+#define CA_FLAG_LENT        2048
+#define CA_FLAG_LENT_VIEWED 4096
 #define CA_CHECK_DATA_TYPE_GIVEN(data_type) \
   do { \
     int ca_dt_ = (data_type); \
@@ -728,6 +739,7 @@ int         ca_op_raises_on_cell (int kind, int op_id, int8_t dt);
    gate first, and a string reference compares in its own type, padding
    included. */
 int8_t      ca_query_match_type (int8_t ref, VALUE query);
+VALUE       ca_fixlen_widen_for_query (VALUE self, VALUE query);
 static inline int
 ca_query_match_applies (CArray *ref, VALUE query)