carray 3.0.4 → 3.0.6

data/ext/carray_kernels_scan.c

@@ -49,7 +49,7 @@ static VALUE
 cumsum_ki_native_i8 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_FLOAT64),
-                                          INT2NUM(sizeof(double)));
+                                          INT2NUM(sizeof(ca_u_float64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -58,6 +58,7 @@ cumsum_ki_native_i8 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -73,12 +74,11 @@ cumsum_ki_native_i8 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(int8_t, double, st_in, pi, mi,
+    CA_SLAB_SCAN_T(int8_t, ca_u_float64_t, st_in, pi, mi,
                          st_out, po, 0, acc += v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -91,7 +91,7 @@ static VALUE
 cumsum_ki_native_u8 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_FLOAT64),
-                                          INT2NUM(sizeof(double)));
+                                          INT2NUM(sizeof(ca_u_float64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -100,6 +100,7 @@ cumsum_ki_native_u8 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -115,12 +116,11 @@ cumsum_ki_native_u8 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(uint8_t, double, st_in, pi, mi,
+    CA_SLAB_SCAN_T(uint8_t, ca_u_float64_t, st_in, pi, mi,
                          st_out, po, 0, acc += v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -133,7 +133,7 @@ static VALUE
 cumsum_ki_native_i16 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_FLOAT64),
-                                          INT2NUM(sizeof(double)));
+                                          INT2NUM(sizeof(ca_u_float64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -142,6 +142,7 @@ cumsum_ki_native_i16 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -157,12 +158,11 @@ cumsum_ki_native_i16 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(int16_t, double, st_in, pi, mi,
+    CA_SLAB_SCAN_T(ca_u_int16_t, ca_u_float64_t, st_in, pi, mi,
                          st_out, po, 0, acc += v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -175,7 +175,7 @@ static VALUE
 cumsum_ki_native_u16 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_FLOAT64),
-                                          INT2NUM(sizeof(double)));
+                                          INT2NUM(sizeof(ca_u_float64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -184,6 +184,7 @@ cumsum_ki_native_u16 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -199,12 +200,11 @@ cumsum_ki_native_u16 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(uint16_t, double, st_in, pi, mi,
+    CA_SLAB_SCAN_T(ca_u_uint16_t, ca_u_float64_t, st_in, pi, mi,
                          st_out, po, 0, acc += v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -217,7 +217,7 @@ static VALUE
 cumsum_ki_native_i32 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_FLOAT64),
-                                          INT2NUM(sizeof(double)));
+                                          INT2NUM(sizeof(ca_u_float64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -226,6 +226,7 @@ cumsum_ki_native_i32 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -241,12 +242,11 @@ cumsum_ki_native_i32 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(int32_t, double, st_in, pi, mi,
+    CA_SLAB_SCAN_T(ca_u_int32_t, ca_u_float64_t, st_in, pi, mi,
                          st_out, po, 0, acc += v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -259,7 +259,7 @@ static VALUE
 cumsum_ki_native_u32 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_FLOAT64),
-                                          INT2NUM(sizeof(double)));
+                                          INT2NUM(sizeof(ca_u_float64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -268,6 +268,7 @@ cumsum_ki_native_u32 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -283,12 +284,11 @@ cumsum_ki_native_u32 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(uint32_t, double, st_in, pi, mi,
+    CA_SLAB_SCAN_T(ca_u_uint32_t, ca_u_float64_t, st_in, pi, mi,
                          st_out, po, 0, acc += v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -301,7 +301,7 @@ static VALUE
 cumsum_ki_native_i64 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_FLOAT64),
-                                          INT2NUM(sizeof(double)));
+                                          INT2NUM(sizeof(ca_u_float64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -310,6 +310,7 @@ cumsum_ki_native_i64 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -325,12 +326,11 @@ cumsum_ki_native_i64 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(int64_t, double, st_in, pi, mi,
+    CA_SLAB_SCAN_T(ca_u_int64_t, ca_u_float64_t, st_in, pi, mi,
                          st_out, po, 0, acc += v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -343,7 +343,7 @@ static VALUE
 cumsum_ki_native_u64 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_FLOAT64),
-                                          INT2NUM(sizeof(double)));
+                                          INT2NUM(sizeof(ca_u_float64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -352,6 +352,7 @@ cumsum_ki_native_u64 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -367,12 +368,11 @@ cumsum_ki_native_u64 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(uint64_t, double, st_in, pi, mi,
+    CA_SLAB_SCAN_T(ca_u_uint64_t, ca_u_float64_t, st_in, pi, mi,
                          st_out, po, 0, acc += v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -385,7 +385,7 @@ static VALUE
 cumsum_ki_native_f32 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_FLOAT64),
-                                          INT2NUM(sizeof(double)));
+                                          INT2NUM(sizeof(ca_u_float64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -394,6 +394,7 @@ cumsum_ki_native_f32 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -409,12 +410,11 @@ cumsum_ki_native_f32 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(float, double, st_in, pi, mi,
+    CA_SLAB_SCAN_T(ca_u_float32_t, ca_u_float64_t, st_in, pi, mi,
                          st_out, po, 0, acc += v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -427,7 +427,7 @@ static VALUE
 cumsum_ki_native_f64 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_FLOAT64),
-                                          INT2NUM(sizeof(double)));
+                                          INT2NUM(sizeof(ca_u_float64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -436,6 +436,7 @@ cumsum_ki_native_f64 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -451,12 +452,11 @@ cumsum_ki_native_f64 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(double, double, st_in, pi, mi,
+    CA_SLAB_SCAN_T(ca_u_float64_t, ca_u_float64_t, st_in, pi, mi,
                          st_out, po, 0, acc += v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -469,7 +469,7 @@ static VALUE
 cumsum_ki_native_cmplx64 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_CMPLX128),
-                                          INT2NUM(sizeof(cmplx128_t)));
+                                          INT2NUM(sizeof(ca_u_cmplx128_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -478,6 +478,7 @@ cumsum_ki_native_cmplx64 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -493,12 +494,11 @@ cumsum_ki_native_cmplx64 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(cmplx64_t, cmplx128_t, st_in, pi, mi,
+    CA_SLAB_SCAN_T(ca_u_cmplx64_t, ca_u_cmplx128_t, st_in, pi, mi,
                          st_out, po, 0, acc += v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -511,7 +511,7 @@ static VALUE
 cumsum_ki_native_cmplx128 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_CMPLX128),
-                                          INT2NUM(sizeof(cmplx128_t)));
+                                          INT2NUM(sizeof(ca_u_cmplx128_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -520,6 +520,7 @@ cumsum_ki_native_cmplx128 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -535,12 +536,11 @@ cumsum_ki_native_cmplx128 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(cmplx128_t, cmplx128_t, st_in, pi, mi,
+    CA_SLAB_SCAN_T(ca_u_cmplx128_t, ca_u_cmplx128_t, st_in, pi, mi,
                          st_out, po, 0, acc += v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -553,7 +553,7 @@ static VALUE
 cumsum_ki_native_bool (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_UINT64),
-                                          INT2NUM(sizeof(uint64_t)));
+                                          INT2NUM(sizeof(ca_u_uint64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -562,6 +562,7 @@ cumsum_ki_native_bool (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -577,12 +578,11 @@ cumsum_ki_native_bool (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(boolean8_t, uint64_t, st_in, pi, mi,
+    CA_SLAB_SCAN_T(boolean8_t, ca_u_uint64_t, st_in, pi, mi,
                          st_out, po, 0, acc += v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -591,9 +591,9 @@ cumsum_ki_native_bool (VALUE self, CArray *ca, int axis)
   return vout;
 }
 
-/* The object lane calls Ruby for every cell, so a walk can raise part
-   way.  Its states live in the caller's frame, and ca_iter_ensure
-   finishes them however the body leaves. */
+/* The walks can raise part way (a Ruby call per cell in the object
+   lane, an allocation in any lane).  Their states live in the caller's
+   frame, and ca_iter_ensure finishes them however the body leaves. */
 typedef struct {
   ca_iter_state states[2];
   VALUE self;
@@ -623,6 +623,7 @@ cumsum_ki_native_object_body (VALUE arg)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
                              slab_axes, 1, 0);
@@ -637,7 +638,6 @@ cumsum_ki_native_object_body (VALUE arg)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
@@ -660,7 +660,7 @@ cumsum_ki_native_object (VALUE self, CArray *ca, int axis)
   fr.self = self;
   fr.ca = ca;
   fr.axis = axis;
-  return ca_iter_ensure(2, fr.states, cumsum_ki_native_object_body, (VALUE) &fr);
+  return ca_iter_ensure_computing(ca, 2, fr.states, cumsum_ki_native_object_body, (VALUE) &fr);
 }
 
 
@@ -747,6 +747,7 @@ cumsum_in_own_type_ki_native_i8 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -762,7 +763,6 @@ cumsum_in_own_type_ki_native_i8 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
@@ -789,6 +789,7 @@ cumsum_in_own_type_ki_native_u8 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -804,7 +805,6 @@ cumsum_in_own_type_ki_native_u8 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
@@ -822,7 +822,7 @@ static VALUE
 cumsum_in_own_type_ki_native_i16 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_INT16),
-                                          INT2NUM(sizeof(int16_t)));
+                                          INT2NUM(sizeof(ca_u_int16_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -831,6 +831,7 @@ cumsum_in_own_type_ki_native_i16 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -846,12 +847,11 @@ cumsum_in_own_type_ki_native_i16 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(int16_t, int16_t, st_in, pi, mi,
+    CA_SLAB_SCAN_T(ca_u_int16_t, ca_u_int16_t, st_in, pi, mi,
                          st_out, po, 0, acc += v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -864,7 +864,7 @@ static VALUE
 cumsum_in_own_type_ki_native_u16 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_UINT16),
-                                          INT2NUM(sizeof(uint16_t)));
+                                          INT2NUM(sizeof(ca_u_uint16_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -873,6 +873,7 @@ cumsum_in_own_type_ki_native_u16 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -888,12 +889,11 @@ cumsum_in_own_type_ki_native_u16 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(uint16_t, uint16_t, st_in, pi, mi,
+    CA_SLAB_SCAN_T(ca_u_uint16_t, ca_u_uint16_t, st_in, pi, mi,
                          st_out, po, 0, acc += v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -906,7 +906,7 @@ static VALUE
 cumsum_in_own_type_ki_native_i32 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_INT32),
-                                          INT2NUM(sizeof(int32_t)));
+                                          INT2NUM(sizeof(ca_u_int32_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -915,6 +915,7 @@ cumsum_in_own_type_ki_native_i32 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -930,12 +931,11 @@ cumsum_in_own_type_ki_native_i32 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(int32_t, int32_t, st_in, pi, mi,
+    CA_SLAB_SCAN_T(ca_u_int32_t, ca_u_int32_t, st_in, pi, mi,
                          st_out, po, 0, acc += v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -948,7 +948,7 @@ static VALUE
 cumsum_in_own_type_ki_native_u32 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_UINT32),
-                                          INT2NUM(sizeof(uint32_t)));
+                                          INT2NUM(sizeof(ca_u_uint32_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -957,6 +957,7 @@ cumsum_in_own_type_ki_native_u32 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -972,12 +973,11 @@ cumsum_in_own_type_ki_native_u32 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(uint32_t, uint32_t, st_in, pi, mi,
+    CA_SLAB_SCAN_T(ca_u_uint32_t, ca_u_uint32_t, st_in, pi, mi,
                          st_out, po, 0, acc += v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -990,7 +990,7 @@ static VALUE
 cumsum_in_own_type_ki_native_i64 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_INT64),
-                                          INT2NUM(sizeof(int64_t)));
+                                          INT2NUM(sizeof(ca_u_int64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -999,6 +999,7 @@ cumsum_in_own_type_ki_native_i64 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -1014,12 +1015,11 @@ cumsum_in_own_type_ki_native_i64 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(int64_t, int64_t, st_in, pi, mi,
+    CA_SLAB_SCAN_T(ca_u_int64_t, ca_u_int64_t, st_in, pi, mi,
                          st_out, po, 0, acc += v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -1032,7 +1032,7 @@ static VALUE
 cumsum_in_own_type_ki_native_u64 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_UINT64),
-                                          INT2NUM(sizeof(uint64_t)));
+                                          INT2NUM(sizeof(ca_u_uint64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -1041,6 +1041,7 @@ cumsum_in_own_type_ki_native_u64 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -1056,12 +1057,11 @@ cumsum_in_own_type_ki_native_u64 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(uint64_t, uint64_t, st_in, pi, mi,
+    CA_SLAB_SCAN_T(ca_u_uint64_t, ca_u_uint64_t, st_in, pi, mi,
                          st_out, po, 0, acc += v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -1132,7 +1132,7 @@ static VALUE
 cumprod_ki_native_i8 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_FLOAT64),
-                                          INT2NUM(sizeof(double)));
+                                          INT2NUM(sizeof(ca_u_float64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -1141,6 +1141,7 @@ cumprod_ki_native_i8 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -1156,12 +1157,11 @@ cumprod_ki_native_i8 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(int8_t, double, st_in, pi, mi,
+    CA_SLAB_SCAN_T(int8_t, ca_u_float64_t, st_in, pi, mi,
                          st_out, po, 1, acc *= v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -1174,7 +1174,7 @@ static VALUE
 cumprod_ki_native_u8 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_FLOAT64),
-                                          INT2NUM(sizeof(double)));
+                                          INT2NUM(sizeof(ca_u_float64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -1183,6 +1183,7 @@ cumprod_ki_native_u8 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -1198,12 +1199,11 @@ cumprod_ki_native_u8 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(uint8_t, double, st_in, pi, mi,
+    CA_SLAB_SCAN_T(uint8_t, ca_u_float64_t, st_in, pi, mi,
                          st_out, po, 1, acc *= v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -1216,7 +1216,7 @@ static VALUE
 cumprod_ki_native_i16 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_FLOAT64),
-                                          INT2NUM(sizeof(double)));
+                                          INT2NUM(sizeof(ca_u_float64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -1225,6 +1225,7 @@ cumprod_ki_native_i16 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -1240,12 +1241,11 @@ cumprod_ki_native_i16 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(int16_t, double, st_in, pi, mi,
+    CA_SLAB_SCAN_T(ca_u_int16_t, ca_u_float64_t, st_in, pi, mi,
                          st_out, po, 1, acc *= v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -1258,7 +1258,7 @@ static VALUE
 cumprod_ki_native_u16 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_FLOAT64),
-                                          INT2NUM(sizeof(double)));
+                                          INT2NUM(sizeof(ca_u_float64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -1267,6 +1267,7 @@ cumprod_ki_native_u16 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -1282,12 +1283,11 @@ cumprod_ki_native_u16 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(uint16_t, double, st_in, pi, mi,
+    CA_SLAB_SCAN_T(ca_u_uint16_t, ca_u_float64_t, st_in, pi, mi,
                          st_out, po, 1, acc *= v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -1300,7 +1300,7 @@ static VALUE
 cumprod_ki_native_i32 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_FLOAT64),
-                                          INT2NUM(sizeof(double)));
+                                          INT2NUM(sizeof(ca_u_float64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -1309,6 +1309,7 @@ cumprod_ki_native_i32 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -1324,12 +1325,11 @@ cumprod_ki_native_i32 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(int32_t, double, st_in, pi, mi,
+    CA_SLAB_SCAN_T(ca_u_int32_t, ca_u_float64_t, st_in, pi, mi,
                          st_out, po, 1, acc *= v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -1342,7 +1342,7 @@ static VALUE
 cumprod_ki_native_u32 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_FLOAT64),
-                                          INT2NUM(sizeof(double)));
+                                          INT2NUM(sizeof(ca_u_float64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -1351,6 +1351,7 @@ cumprod_ki_native_u32 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -1366,12 +1367,11 @@ cumprod_ki_native_u32 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(uint32_t, double, st_in, pi, mi,
+    CA_SLAB_SCAN_T(ca_u_uint32_t, ca_u_float64_t, st_in, pi, mi,
                          st_out, po, 1, acc *= v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -1384,7 +1384,7 @@ static VALUE
 cumprod_ki_native_i64 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_FLOAT64),
-                                          INT2NUM(sizeof(double)));
+                                          INT2NUM(sizeof(ca_u_float64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -1393,6 +1393,7 @@ cumprod_ki_native_i64 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -1408,12 +1409,11 @@ cumprod_ki_native_i64 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(int64_t, double, st_in, pi, mi,
+    CA_SLAB_SCAN_T(ca_u_int64_t, ca_u_float64_t, st_in, pi, mi,
                          st_out, po, 1, acc *= v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -1426,7 +1426,7 @@ static VALUE
 cumprod_ki_native_u64 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_FLOAT64),
-                                          INT2NUM(sizeof(double)));
+                                          INT2NUM(sizeof(ca_u_float64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -1435,6 +1435,7 @@ cumprod_ki_native_u64 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -1450,12 +1451,11 @@ cumprod_ki_native_u64 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(uint64_t, double, st_in, pi, mi,
+    CA_SLAB_SCAN_T(ca_u_uint64_t, ca_u_float64_t, st_in, pi, mi,
                          st_out, po, 1, acc *= v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -1468,7 +1468,7 @@ static VALUE
 cumprod_ki_native_f32 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_FLOAT64),
-                                          INT2NUM(sizeof(double)));
+                                          INT2NUM(sizeof(ca_u_float64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -1477,6 +1477,7 @@ cumprod_ki_native_f32 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -1492,12 +1493,11 @@ cumprod_ki_native_f32 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(float, double, st_in, pi, mi,
+    CA_SLAB_SCAN_T(ca_u_float32_t, ca_u_float64_t, st_in, pi, mi,
                          st_out, po, 1, acc *= v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -1510,7 +1510,7 @@ static VALUE
 cumprod_ki_native_f64 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_FLOAT64),
-                                          INT2NUM(sizeof(double)));
+                                          INT2NUM(sizeof(ca_u_float64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -1519,6 +1519,7 @@ cumprod_ki_native_f64 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -1534,12 +1535,11 @@ cumprod_ki_native_f64 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(double, double, st_in, pi, mi,
+    CA_SLAB_SCAN_T(ca_u_float64_t, ca_u_float64_t, st_in, pi, mi,
                          st_out, po, 1, acc *= v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -1552,7 +1552,7 @@ static VALUE
 cumprod_ki_native_cmplx64 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_CMPLX128),
-                                          INT2NUM(sizeof(cmplx128_t)));
+                                          INT2NUM(sizeof(ca_u_cmplx128_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -1561,6 +1561,7 @@ cumprod_ki_native_cmplx64 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -1576,12 +1577,11 @@ cumprod_ki_native_cmplx64 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(cmplx64_t, cmplx128_t, st_in, pi, mi,
+    CA_SLAB_SCAN_T(ca_u_cmplx64_t, ca_u_cmplx128_t, st_in, pi, mi,
                          st_out, po, 1, acc *= v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -1594,7 +1594,7 @@ static VALUE
 cumprod_ki_native_cmplx128 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_CMPLX128),
-                                          INT2NUM(sizeof(cmplx128_t)));
+                                          INT2NUM(sizeof(ca_u_cmplx128_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -1603,6 +1603,7 @@ cumprod_ki_native_cmplx128 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -1618,12 +1619,11 @@ cumprod_ki_native_cmplx128 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(cmplx128_t, cmplx128_t, st_in, pi, mi,
+    CA_SLAB_SCAN_T(ca_u_cmplx128_t, ca_u_cmplx128_t, st_in, pi, mi,
                          st_out, po, 1, acc *= v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -1636,7 +1636,7 @@ static VALUE
 cumprod_ki_native_bool (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_UINT64),
-                                          INT2NUM(sizeof(uint64_t)));
+                                          INT2NUM(sizeof(ca_u_uint64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -1645,6 +1645,7 @@ cumprod_ki_native_bool (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -1660,12 +1661,11 @@ cumprod_ki_native_bool (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(boolean8_t, uint64_t, st_in, pi, mi,
+    CA_SLAB_SCAN_T(boolean8_t, ca_u_uint64_t, st_in, pi, mi,
                          st_out, po, 1, acc *= v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -1674,9 +1674,9 @@ cumprod_ki_native_bool (VALUE self, CArray *ca, int axis)
   return vout;
 }
 
-/* The object lane calls Ruby for every cell, so a walk can raise part
-   way.  Its states live in the caller's frame, and ca_iter_ensure
-   finishes them however the body leaves. */
+/* The walks can raise part way (a Ruby call per cell in the object
+   lane, an allocation in any lane).  Their states live in the caller's
+   frame, and ca_iter_ensure finishes them however the body leaves. */
 typedef struct {
   ca_iter_state states[2];
   VALUE self;
@@ -1706,6 +1706,7 @@ cumprod_ki_native_object_body (VALUE arg)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
                              slab_axes, 1, 0);
@@ -1720,7 +1721,6 @@ cumprod_ki_native_object_body (VALUE arg)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
@@ -1743,7 +1743,7 @@ cumprod_ki_native_object (VALUE self, CArray *ca, int axis)
   fr.self = self;
   fr.ca = ca;
   fr.axis = axis;
-  return ca_iter_ensure(2, fr.states, cumprod_ki_native_object_body, (VALUE) &fr);
+  return ca_iter_ensure_computing(ca, 2, fr.states, cumprod_ki_native_object_body, (VALUE) &fr);
 }
 
 
@@ -1830,6 +1830,13 @@ cummax_ki_native_i8 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
+    boolean8_t *op_mask = NULL;
+    if ( ca_has_mask(ca) ) {
+      ca_create_mask(co);
+      op_mask = (boolean8_t *) co->mask->ptr;
+    }
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -1845,13 +1852,6 @@ cummax_ki_native_i8 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
-    boolean8_t *op_mask = NULL;
-    if ( ca_has_mask(ca) ) {
-      ca_create_mask(co);
-      op_mask = (boolean8_t *) co->mask->ptr;
-    }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
@@ -1881,6 +1881,13 @@ cummax_ki_native_u8 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
+    boolean8_t *op_mask = NULL;
+    if ( ca_has_mask(ca) ) {
+      ca_create_mask(co);
+      op_mask = (boolean8_t *) co->mask->ptr;
+    }
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -1896,13 +1903,6 @@ cummax_ki_native_u8 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
-    boolean8_t *op_mask = NULL;
-    if ( ca_has_mask(ca) ) {
-      ca_create_mask(co);
-      op_mask = (boolean8_t *) co->mask->ptr;
-    }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
@@ -1923,7 +1923,7 @@ static VALUE
 cummax_ki_native_i16 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_INT16),
-                                          INT2NUM(sizeof(int16_t)));
+                                          INT2NUM(sizeof(ca_u_int16_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -1932,6 +1932,13 @@ cummax_ki_native_i16 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
+    boolean8_t *op_mask = NULL;
+    if ( ca_has_mask(ca) ) {
+      ca_create_mask(co);
+      op_mask = (boolean8_t *) co->mask->ptr;
+    }
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -1947,13 +1954,6 @@ cummax_ki_native_i16 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
-    boolean8_t *op_mask = NULL;
-    if ( ca_has_mask(ca) ) {
-      ca_create_mask(co);
-      op_mask = (boolean8_t *) co->mask->ptr;
-    }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
@@ -1961,7 +1961,7 @@ cummax_ki_native_i16 (VALUE self, CArray *ca, int axis)
     boolean8_t *mo_fiber = op_mask
             ? op_mask + (ca_size_t)(po - st_out.alias_ptr) / (ca_size_t) co->bytes
             : NULL;
-          CA_SLAB_SCAN_T_GATED(int16_t, int16_t, st_in, pi, mi,
+          CA_SLAB_SCAN_T_GATED(ca_u_int16_t, ca_u_int16_t, st_in, pi, mi,
                          st_out, po, mo_fiber, INT16_MIN, if (v > acc) acc = v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -1974,7 +1974,7 @@ static VALUE
 cummax_ki_native_u16 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_UINT16),
-                                          INT2NUM(sizeof(uint16_t)));
+                                          INT2NUM(sizeof(ca_u_uint16_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -1983,6 +1983,13 @@ cummax_ki_native_u16 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
+    boolean8_t *op_mask = NULL;
+    if ( ca_has_mask(ca) ) {
+      ca_create_mask(co);
+      op_mask = (boolean8_t *) co->mask->ptr;
+    }
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -1998,13 +2005,6 @@ cummax_ki_native_u16 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
-    boolean8_t *op_mask = NULL;
-    if ( ca_has_mask(ca) ) {
-      ca_create_mask(co);
-      op_mask = (boolean8_t *) co->mask->ptr;
-    }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
@@ -2012,7 +2012,7 @@ cummax_ki_native_u16 (VALUE self, CArray *ca, int axis)
     boolean8_t *mo_fiber = op_mask
             ? op_mask + (ca_size_t)(po - st_out.alias_ptr) / (ca_size_t) co->bytes
             : NULL;
-          CA_SLAB_SCAN_T_GATED(uint16_t, uint16_t, st_in, pi, mi,
+          CA_SLAB_SCAN_T_GATED(ca_u_uint16_t, ca_u_uint16_t, st_in, pi, mi,
                          st_out, po, mo_fiber, 0, if (v > acc) acc = v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -2025,7 +2025,7 @@ static VALUE
 cummax_ki_native_i32 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_INT32),
-                                          INT2NUM(sizeof(int32_t)));
+                                          INT2NUM(sizeof(ca_u_int32_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -2034,6 +2034,13 @@ cummax_ki_native_i32 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
+    boolean8_t *op_mask = NULL;
+    if ( ca_has_mask(ca) ) {
+      ca_create_mask(co);
+      op_mask = (boolean8_t *) co->mask->ptr;
+    }
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -2049,13 +2056,6 @@ cummax_ki_native_i32 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
-    boolean8_t *op_mask = NULL;
-    if ( ca_has_mask(ca) ) {
-      ca_create_mask(co);
-      op_mask = (boolean8_t *) co->mask->ptr;
-    }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
@@ -2063,7 +2063,7 @@ cummax_ki_native_i32 (VALUE self, CArray *ca, int axis)
     boolean8_t *mo_fiber = op_mask
             ? op_mask + (ca_size_t)(po - st_out.alias_ptr) / (ca_size_t) co->bytes
             : NULL;
-          CA_SLAB_SCAN_T_GATED(int32_t, int32_t, st_in, pi, mi,
+          CA_SLAB_SCAN_T_GATED(ca_u_int32_t, ca_u_int32_t, st_in, pi, mi,
                          st_out, po, mo_fiber, INT32_MIN, if (v > acc) acc = v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -2076,7 +2076,7 @@ static VALUE
 cummax_ki_native_u32 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_UINT32),
-                                          INT2NUM(sizeof(uint32_t)));
+                                          INT2NUM(sizeof(ca_u_uint32_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -2085,6 +2085,13 @@ cummax_ki_native_u32 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
+    boolean8_t *op_mask = NULL;
+    if ( ca_has_mask(ca) ) {
+      ca_create_mask(co);
+      op_mask = (boolean8_t *) co->mask->ptr;
+    }
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -2100,13 +2107,6 @@ cummax_ki_native_u32 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
-    boolean8_t *op_mask = NULL;
-    if ( ca_has_mask(ca) ) {
-      ca_create_mask(co);
-      op_mask = (boolean8_t *) co->mask->ptr;
-    }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
@@ -2114,7 +2114,7 @@ cummax_ki_native_u32 (VALUE self, CArray *ca, int axis)
     boolean8_t *mo_fiber = op_mask
             ? op_mask + (ca_size_t)(po - st_out.alias_ptr) / (ca_size_t) co->bytes
             : NULL;
-          CA_SLAB_SCAN_T_GATED(uint32_t, uint32_t, st_in, pi, mi,
+          CA_SLAB_SCAN_T_GATED(ca_u_uint32_t, ca_u_uint32_t, st_in, pi, mi,
                          st_out, po, mo_fiber, 0, if (v > acc) acc = v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -2127,7 +2127,7 @@ static VALUE
 cummax_ki_native_i64 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_INT64),
-                                          INT2NUM(sizeof(int64_t)));
+                                          INT2NUM(sizeof(ca_u_int64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -2136,6 +2136,13 @@ cummax_ki_native_i64 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
+    boolean8_t *op_mask = NULL;
+    if ( ca_has_mask(ca) ) {
+      ca_create_mask(co);
+      op_mask = (boolean8_t *) co->mask->ptr;
+    }
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -2151,13 +2158,6 @@ cummax_ki_native_i64 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
-    boolean8_t *op_mask = NULL;
-    if ( ca_has_mask(ca) ) {
-      ca_create_mask(co);
-      op_mask = (boolean8_t *) co->mask->ptr;
-    }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
@@ -2165,7 +2165,7 @@ cummax_ki_native_i64 (VALUE self, CArray *ca, int axis)
     boolean8_t *mo_fiber = op_mask
             ? op_mask + (ca_size_t)(po - st_out.alias_ptr) / (ca_size_t) co->bytes
             : NULL;
-          CA_SLAB_SCAN_T_GATED(int64_t, int64_t, st_in, pi, mi,
+          CA_SLAB_SCAN_T_GATED(ca_u_int64_t, ca_u_int64_t, st_in, pi, mi,
                          st_out, po, mo_fiber, INT64_MIN, if (v > acc) acc = v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -2178,7 +2178,7 @@ static VALUE
 cummax_ki_native_u64 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_UINT64),
-                                          INT2NUM(sizeof(uint64_t)));
+                                          INT2NUM(sizeof(ca_u_uint64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -2187,6 +2187,13 @@ cummax_ki_native_u64 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
+    boolean8_t *op_mask = NULL;
+    if ( ca_has_mask(ca) ) {
+      ca_create_mask(co);
+      op_mask = (boolean8_t *) co->mask->ptr;
+    }
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -2202,13 +2209,6 @@ cummax_ki_native_u64 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
-    boolean8_t *op_mask = NULL;
-    if ( ca_has_mask(ca) ) {
-      ca_create_mask(co);
-      op_mask = (boolean8_t *) co->mask->ptr;
-    }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
@@ -2216,7 +2216,7 @@ cummax_ki_native_u64 (VALUE self, CArray *ca, int axis)
     boolean8_t *mo_fiber = op_mask
             ? op_mask + (ca_size_t)(po - st_out.alias_ptr) / (ca_size_t) co->bytes
             : NULL;
-          CA_SLAB_SCAN_T_GATED(uint64_t, uint64_t, st_in, pi, mi,
+          CA_SLAB_SCAN_T_GATED(ca_u_uint64_t, ca_u_uint64_t, st_in, pi, mi,
                          st_out, po, mo_fiber, 0, if (v > acc) acc = v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -2229,7 +2229,7 @@ static VALUE
 cummax_ki_native_f32 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_FLOAT32),
-                                          INT2NUM(sizeof(float)));
+                                          INT2NUM(sizeof(ca_u_float32_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -2238,6 +2238,13 @@ cummax_ki_native_f32 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
+    boolean8_t *op_mask = NULL;
+    if ( ca_has_mask(ca) ) {
+      ca_create_mask(co);
+      op_mask = (boolean8_t *) co->mask->ptr;
+    }
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -2253,13 +2260,6 @@ cummax_ki_native_f32 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
-    boolean8_t *op_mask = NULL;
-    if ( ca_has_mask(ca) ) {
-      ca_create_mask(co);
-      op_mask = (boolean8_t *) co->mask->ptr;
-    }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
@@ -2267,7 +2267,7 @@ cummax_ki_native_f32 (VALUE self, CArray *ca, int axis)
     boolean8_t *mo_fiber = op_mask
             ? op_mask + (ca_size_t)(po - st_out.alias_ptr) / (ca_size_t) co->bytes
             : NULL;
-          CA_SLAB_SCAN_T_GATED(float, float, st_in, pi, mi,
+          CA_SLAB_SCAN_T_GATED(ca_u_float32_t, ca_u_float32_t, st_in, pi, mi,
                          st_out, po, mo_fiber, NAN, acc = fmax(acc, v); r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -2280,7 +2280,7 @@ static VALUE
 cummax_ki_native_f64 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_FLOAT64),
-                                          INT2NUM(sizeof(double)));
+                                          INT2NUM(sizeof(ca_u_float64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -2289,6 +2289,13 @@ cummax_ki_native_f64 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
+    boolean8_t *op_mask = NULL;
+    if ( ca_has_mask(ca) ) {
+      ca_create_mask(co);
+      op_mask = (boolean8_t *) co->mask->ptr;
+    }
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -2304,13 +2311,6 @@ cummax_ki_native_f64 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
-    boolean8_t *op_mask = NULL;
-    if ( ca_has_mask(ca) ) {
-      ca_create_mask(co);
-      op_mask = (boolean8_t *) co->mask->ptr;
-    }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
@@ -2318,7 +2318,7 @@ cummax_ki_native_f64 (VALUE self, CArray *ca, int axis)
     boolean8_t *mo_fiber = op_mask
             ? op_mask + (ca_size_t)(po - st_out.alias_ptr) / (ca_size_t) co->bytes
             : NULL;
-          CA_SLAB_SCAN_T_GATED(double, double, st_in, pi, mi,
+          CA_SLAB_SCAN_T_GATED(ca_u_float64_t, ca_u_float64_t, st_in, pi, mi,
                          st_out, po, mo_fiber, NAN, acc = fmax(acc, v); r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -2331,7 +2331,7 @@ static VALUE
 cummax_ki_native_bool (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_UINT64),
-                                          INT2NUM(sizeof(uint64_t)));
+                                          INT2NUM(sizeof(ca_u_uint64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -2340,6 +2340,13 @@ cummax_ki_native_bool (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
+    boolean8_t *op_mask = NULL;
+    if ( ca_has_mask(ca) ) {
+      ca_create_mask(co);
+      op_mask = (boolean8_t *) co->mask->ptr;
+    }
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -2355,13 +2362,6 @@ cummax_ki_native_bool (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
-    boolean8_t *op_mask = NULL;
-    if ( ca_has_mask(ca) ) {
-      ca_create_mask(co);
-      op_mask = (boolean8_t *) co->mask->ptr;
-    }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
@@ -2369,7 +2369,7 @@ cummax_ki_native_bool (VALUE self, CArray *ca, int axis)
     boolean8_t *mo_fiber = op_mask
             ? op_mask + (ca_size_t)(po - st_out.alias_ptr) / (ca_size_t) co->bytes
             : NULL;
-          CA_SLAB_SCAN_T_GATED(boolean8_t, uint64_t, st_in, pi, mi,
+          CA_SLAB_SCAN_T_GATED(boolean8_t, ca_u_uint64_t, st_in, pi, mi,
                          st_out, po, mo_fiber, 0, if ((uint64_t) v > acc) acc = v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -2378,9 +2378,9 @@ cummax_ki_native_bool (VALUE self, CArray *ca, int axis)
   return vout;
 }
 
-/* The object lane calls Ruby for every cell, so a walk can raise part
-   way.  Its states live in the caller's frame, and ca_iter_ensure
-   finishes them however the body leaves. */
+/* The walks can raise part way (a Ruby call per cell in the object
+   lane, an allocation in any lane).  Their states live in the caller's
+   frame, and ca_iter_ensure finishes them however the body leaves. */
 typedef struct {
   ca_iter_state states[2];
   VALUE self;
@@ -2410,6 +2410,13 @@ cummax_ki_native_object_body (VALUE arg)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
+    boolean8_t *op_mask = NULL;
+    if ( ca_has_mask(ca) ) {
+      ca_create_mask(co);
+      op_mask = (boolean8_t *) co->mask->ptr;
+    }
+
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
                              slab_axes, 1, 0);
@@ -2424,13 +2431,6 @@ cummax_ki_native_object_body (VALUE arg)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
-    boolean8_t *op_mask = NULL;
-    if ( ca_has_mask(ca) ) {
-      ca_create_mask(co);
-      op_mask = (boolean8_t *) co->mask->ptr;
-    }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
@@ -2456,7 +2456,7 @@ cummax_ki_native_object (VALUE self, CArray *ca, int axis)
   fr.self = self;
   fr.ca = ca;
   fr.axis = axis;
-  return ca_iter_ensure(2, fr.states, cummax_ki_native_object_body, (VALUE) &fr);
+  return ca_iter_ensure_computing(ca, 2, fr.states, cummax_ki_native_object_body, (VALUE) &fr);
 }
 
 
@@ -2592,6 +2592,13 @@ cummin_ki_native_i8 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
+    boolean8_t *op_mask = NULL;
+    if ( ca_has_mask(ca) ) {
+      ca_create_mask(co);
+      op_mask = (boolean8_t *) co->mask->ptr;
+    }
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -2607,13 +2614,6 @@ cummin_ki_native_i8 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
-    boolean8_t *op_mask = NULL;
-    if ( ca_has_mask(ca) ) {
-      ca_create_mask(co);
-      op_mask = (boolean8_t *) co->mask->ptr;
-    }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
@@ -2643,6 +2643,13 @@ cummin_ki_native_u8 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
+    boolean8_t *op_mask = NULL;
+    if ( ca_has_mask(ca) ) {
+      ca_create_mask(co);
+      op_mask = (boolean8_t *) co->mask->ptr;
+    }
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -2658,13 +2665,6 @@ cummin_ki_native_u8 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
-    boolean8_t *op_mask = NULL;
-    if ( ca_has_mask(ca) ) {
-      ca_create_mask(co);
-      op_mask = (boolean8_t *) co->mask->ptr;
-    }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
@@ -2685,7 +2685,7 @@ static VALUE
 cummin_ki_native_i16 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_INT16),
-                                          INT2NUM(sizeof(int16_t)));
+                                          INT2NUM(sizeof(ca_u_int16_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -2694,6 +2694,13 @@ cummin_ki_native_i16 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
+    boolean8_t *op_mask = NULL;
+    if ( ca_has_mask(ca) ) {
+      ca_create_mask(co);
+      op_mask = (boolean8_t *) co->mask->ptr;
+    }
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -2709,13 +2716,6 @@ cummin_ki_native_i16 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
-    boolean8_t *op_mask = NULL;
-    if ( ca_has_mask(ca) ) {
-      ca_create_mask(co);
-      op_mask = (boolean8_t *) co->mask->ptr;
-    }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
@@ -2723,7 +2723,7 @@ cummin_ki_native_i16 (VALUE self, CArray *ca, int axis)
     boolean8_t *mo_fiber = op_mask
             ? op_mask + (ca_size_t)(po - st_out.alias_ptr) / (ca_size_t) co->bytes
             : NULL;
-          CA_SLAB_SCAN_T_GATED(int16_t, int16_t, st_in, pi, mi,
+          CA_SLAB_SCAN_T_GATED(ca_u_int16_t, ca_u_int16_t, st_in, pi, mi,
                          st_out, po, mo_fiber, INT16_MAX, if (v < acc) acc = v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -2736,7 +2736,7 @@ static VALUE
 cummin_ki_native_u16 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_UINT16),
-                                          INT2NUM(sizeof(uint16_t)));
+                                          INT2NUM(sizeof(ca_u_uint16_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -2745,6 +2745,13 @@ cummin_ki_native_u16 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
+    boolean8_t *op_mask = NULL;
+    if ( ca_has_mask(ca) ) {
+      ca_create_mask(co);
+      op_mask = (boolean8_t *) co->mask->ptr;
+    }
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -2760,13 +2767,6 @@ cummin_ki_native_u16 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
-    boolean8_t *op_mask = NULL;
-    if ( ca_has_mask(ca) ) {
-      ca_create_mask(co);
-      op_mask = (boolean8_t *) co->mask->ptr;
-    }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
@@ -2774,7 +2774,7 @@ cummin_ki_native_u16 (VALUE self, CArray *ca, int axis)
     boolean8_t *mo_fiber = op_mask
             ? op_mask + (ca_size_t)(po - st_out.alias_ptr) / (ca_size_t) co->bytes
             : NULL;
-          CA_SLAB_SCAN_T_GATED(uint16_t, uint16_t, st_in, pi, mi,
+          CA_SLAB_SCAN_T_GATED(ca_u_uint16_t, ca_u_uint16_t, st_in, pi, mi,
                          st_out, po, mo_fiber, UINT16_MAX, if (v < acc) acc = v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -2787,7 +2787,7 @@ static VALUE
 cummin_ki_native_i32 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_INT32),
-                                          INT2NUM(sizeof(int32_t)));
+                                          INT2NUM(sizeof(ca_u_int32_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -2796,6 +2796,13 @@ cummin_ki_native_i32 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
+    boolean8_t *op_mask = NULL;
+    if ( ca_has_mask(ca) ) {
+      ca_create_mask(co);
+      op_mask = (boolean8_t *) co->mask->ptr;
+    }
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -2811,13 +2818,6 @@ cummin_ki_native_i32 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
-    boolean8_t *op_mask = NULL;
-    if ( ca_has_mask(ca) ) {
-      ca_create_mask(co);
-      op_mask = (boolean8_t *) co->mask->ptr;
-    }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
@@ -2825,7 +2825,7 @@ cummin_ki_native_i32 (VALUE self, CArray *ca, int axis)
     boolean8_t *mo_fiber = op_mask
             ? op_mask + (ca_size_t)(po - st_out.alias_ptr) / (ca_size_t) co->bytes
             : NULL;
-          CA_SLAB_SCAN_T_GATED(int32_t, int32_t, st_in, pi, mi,
+          CA_SLAB_SCAN_T_GATED(ca_u_int32_t, ca_u_int32_t, st_in, pi, mi,
                          st_out, po, mo_fiber, INT32_MAX, if (v < acc) acc = v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -2838,7 +2838,7 @@ static VALUE
 cummin_ki_native_u32 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_UINT32),
-                                          INT2NUM(sizeof(uint32_t)));
+                                          INT2NUM(sizeof(ca_u_uint32_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -2847,6 +2847,13 @@ cummin_ki_native_u32 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
+    boolean8_t *op_mask = NULL;
+    if ( ca_has_mask(ca) ) {
+      ca_create_mask(co);
+      op_mask = (boolean8_t *) co->mask->ptr;
+    }
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -2862,13 +2869,6 @@ cummin_ki_native_u32 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
-    boolean8_t *op_mask = NULL;
-    if ( ca_has_mask(ca) ) {
-      ca_create_mask(co);
-      op_mask = (boolean8_t *) co->mask->ptr;
-    }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
@@ -2876,7 +2876,7 @@ cummin_ki_native_u32 (VALUE self, CArray *ca, int axis)
     boolean8_t *mo_fiber = op_mask
             ? op_mask + (ca_size_t)(po - st_out.alias_ptr) / (ca_size_t) co->bytes
             : NULL;
-          CA_SLAB_SCAN_T_GATED(uint32_t, uint32_t, st_in, pi, mi,
+          CA_SLAB_SCAN_T_GATED(ca_u_uint32_t, ca_u_uint32_t, st_in, pi, mi,
                          st_out, po, mo_fiber, UINT32_MAX, if (v < acc) acc = v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -2889,7 +2889,7 @@ static VALUE
 cummin_ki_native_i64 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_INT64),
-                                          INT2NUM(sizeof(int64_t)));
+                                          INT2NUM(sizeof(ca_u_int64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -2898,6 +2898,13 @@ cummin_ki_native_i64 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
+    boolean8_t *op_mask = NULL;
+    if ( ca_has_mask(ca) ) {
+      ca_create_mask(co);
+      op_mask = (boolean8_t *) co->mask->ptr;
+    }
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -2913,13 +2920,6 @@ cummin_ki_native_i64 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
-    boolean8_t *op_mask = NULL;
-    if ( ca_has_mask(ca) ) {
-      ca_create_mask(co);
-      op_mask = (boolean8_t *) co->mask->ptr;
-    }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
@@ -2927,7 +2927,7 @@ cummin_ki_native_i64 (VALUE self, CArray *ca, int axis)
     boolean8_t *mo_fiber = op_mask
             ? op_mask + (ca_size_t)(po - st_out.alias_ptr) / (ca_size_t) co->bytes
             : NULL;
-          CA_SLAB_SCAN_T_GATED(int64_t, int64_t, st_in, pi, mi,
+          CA_SLAB_SCAN_T_GATED(ca_u_int64_t, ca_u_int64_t, st_in, pi, mi,
                          st_out, po, mo_fiber, INT64_MAX, if (v < acc) acc = v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -2940,7 +2940,7 @@ static VALUE
 cummin_ki_native_u64 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_UINT64),
-                                          INT2NUM(sizeof(uint64_t)));
+                                          INT2NUM(sizeof(ca_u_uint64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -2949,6 +2949,13 @@ cummin_ki_native_u64 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
+    boolean8_t *op_mask = NULL;
+    if ( ca_has_mask(ca) ) {
+      ca_create_mask(co);
+      op_mask = (boolean8_t *) co->mask->ptr;
+    }
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -2964,13 +2971,6 @@ cummin_ki_native_u64 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
-    boolean8_t *op_mask = NULL;
-    if ( ca_has_mask(ca) ) {
-      ca_create_mask(co);
-      op_mask = (boolean8_t *) co->mask->ptr;
-    }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
@@ -2978,7 +2978,7 @@ cummin_ki_native_u64 (VALUE self, CArray *ca, int axis)
     boolean8_t *mo_fiber = op_mask
             ? op_mask + (ca_size_t)(po - st_out.alias_ptr) / (ca_size_t) co->bytes
             : NULL;
-          CA_SLAB_SCAN_T_GATED(uint64_t, uint64_t, st_in, pi, mi,
+          CA_SLAB_SCAN_T_GATED(ca_u_uint64_t, ca_u_uint64_t, st_in, pi, mi,
                          st_out, po, mo_fiber, UINT64_MAX, if (v < acc) acc = v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -2991,7 +2991,7 @@ static VALUE
 cummin_ki_native_f32 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_FLOAT32),
-                                          INT2NUM(sizeof(float)));
+                                          INT2NUM(sizeof(ca_u_float32_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -3000,6 +3000,13 @@ cummin_ki_native_f32 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
+    boolean8_t *op_mask = NULL;
+    if ( ca_has_mask(ca) ) {
+      ca_create_mask(co);
+      op_mask = (boolean8_t *) co->mask->ptr;
+    }
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -3015,13 +3022,6 @@ cummin_ki_native_f32 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
-    boolean8_t *op_mask = NULL;
-    if ( ca_has_mask(ca) ) {
-      ca_create_mask(co);
-      op_mask = (boolean8_t *) co->mask->ptr;
-    }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
@@ -3029,7 +3029,7 @@ cummin_ki_native_f32 (VALUE self, CArray *ca, int axis)
     boolean8_t *mo_fiber = op_mask
             ? op_mask + (ca_size_t)(po - st_out.alias_ptr) / (ca_size_t) co->bytes
             : NULL;
-          CA_SLAB_SCAN_T_GATED(float, float, st_in, pi, mi,
+          CA_SLAB_SCAN_T_GATED(ca_u_float32_t, ca_u_float32_t, st_in, pi, mi,
                          st_out, po, mo_fiber, NAN, acc = fmin(acc, v); r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -3042,7 +3042,7 @@ static VALUE
 cummin_ki_native_f64 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_FLOAT64),
-                                          INT2NUM(sizeof(double)));
+                                          INT2NUM(sizeof(ca_u_float64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -3051,6 +3051,13 @@ cummin_ki_native_f64 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
+    boolean8_t *op_mask = NULL;
+    if ( ca_has_mask(ca) ) {
+      ca_create_mask(co);
+      op_mask = (boolean8_t *) co->mask->ptr;
+    }
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -3066,13 +3073,6 @@ cummin_ki_native_f64 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
-    boolean8_t *op_mask = NULL;
-    if ( ca_has_mask(ca) ) {
-      ca_create_mask(co);
-      op_mask = (boolean8_t *) co->mask->ptr;
-    }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
@@ -3080,7 +3080,7 @@ cummin_ki_native_f64 (VALUE self, CArray *ca, int axis)
     boolean8_t *mo_fiber = op_mask
             ? op_mask + (ca_size_t)(po - st_out.alias_ptr) / (ca_size_t) co->bytes
             : NULL;
-          CA_SLAB_SCAN_T_GATED(double, double, st_in, pi, mi,
+          CA_SLAB_SCAN_T_GATED(ca_u_float64_t, ca_u_float64_t, st_in, pi, mi,
                          st_out, po, mo_fiber, NAN, acc = fmin(acc, v); r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -3093,7 +3093,7 @@ static VALUE
 cummin_ki_native_bool (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_UINT64),
-                                          INT2NUM(sizeof(uint64_t)));
+                                          INT2NUM(sizeof(ca_u_uint64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -3102,6 +3102,13 @@ cummin_ki_native_bool (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
+    boolean8_t *op_mask = NULL;
+    if ( ca_has_mask(ca) ) {
+      ca_create_mask(co);
+      op_mask = (boolean8_t *) co->mask->ptr;
+    }
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -3117,13 +3124,6 @@ cummin_ki_native_bool (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
-    boolean8_t *op_mask = NULL;
-    if ( ca_has_mask(ca) ) {
-      ca_create_mask(co);
-      op_mask = (boolean8_t *) co->mask->ptr;
-    }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
@@ -3131,7 +3131,7 @@ cummin_ki_native_bool (VALUE self, CArray *ca, int axis)
     boolean8_t *mo_fiber = op_mask
             ? op_mask + (ca_size_t)(po - st_out.alias_ptr) / (ca_size_t) co->bytes
             : NULL;
-          CA_SLAB_SCAN_T_GATED(boolean8_t, uint64_t, st_in, pi, mi,
+          CA_SLAB_SCAN_T_GATED(boolean8_t, ca_u_uint64_t, st_in, pi, mi,
                          st_out, po, mo_fiber, 1, if ((uint64_t) v < acc) acc = v; r = acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -3140,9 +3140,9 @@ cummin_ki_native_bool (VALUE self, CArray *ca, int axis)
   return vout;
 }
 
-/* The object lane calls Ruby for every cell, so a walk can raise part
-   way.  Its states live in the caller's frame, and ca_iter_ensure
-   finishes them however the body leaves. */
+/* The walks can raise part way (a Ruby call per cell in the object
+   lane, an allocation in any lane).  Their states live in the caller's
+   frame, and ca_iter_ensure finishes them however the body leaves. */
 typedef struct {
   ca_iter_state states[2];
   VALUE self;
@@ -3172,6 +3172,13 @@ cummin_ki_native_object_body (VALUE arg)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
+    boolean8_t *op_mask = NULL;
+    if ( ca_has_mask(ca) ) {
+      ca_create_mask(co);
+      op_mask = (boolean8_t *) co->mask->ptr;
+    }
+
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
                              slab_axes, 1, 0);
@@ -3186,13 +3193,6 @@ cummin_ki_native_object_body (VALUE arg)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
-    boolean8_t *op_mask = NULL;
-    if ( ca_has_mask(ca) ) {
-      ca_create_mask(co);
-      op_mask = (boolean8_t *) co->mask->ptr;
-    }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
@@ -3218,7 +3218,7 @@ cummin_ki_native_object (VALUE self, CArray *ca, int axis)
   fr.self = self;
   fr.ca = ca;
   fr.axis = axis;
-  return ca_iter_ensure(2, fr.states, cummin_ki_native_object_body, (VALUE) &fr);
+  return ca_iter_ensure_computing(ca, 2, fr.states, cummin_ki_native_object_body, (VALUE) &fr);
 }
 
 
@@ -3345,7 +3345,7 @@ static VALUE
 cumcount_ki_native_i8 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_INT64),
-                                          INT2NUM(sizeof(int64_t)));
+                                          INT2NUM(sizeof(ca_u_int64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -3354,6 +3354,7 @@ cumcount_ki_native_i8 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -3369,12 +3370,11 @@ cumcount_ki_native_i8 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(int8_t, int64_t, st_in, pi, mi,
+    CA_SLAB_SCAN_T(int8_t, ca_u_int64_t, st_in, pi, mi,
                          st_out, po, 0, (void) v; r = ++acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -3387,7 +3387,7 @@ static VALUE
 cumcount_ki_native_u8 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_INT64),
-                                          INT2NUM(sizeof(int64_t)));
+                                          INT2NUM(sizeof(ca_u_int64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -3396,6 +3396,7 @@ cumcount_ki_native_u8 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -3411,12 +3412,11 @@ cumcount_ki_native_u8 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(uint8_t, int64_t, st_in, pi, mi,
+    CA_SLAB_SCAN_T(uint8_t, ca_u_int64_t, st_in, pi, mi,
                          st_out, po, 0, (void) v; r = ++acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -3429,7 +3429,7 @@ static VALUE
 cumcount_ki_native_i16 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_INT64),
-                                          INT2NUM(sizeof(int64_t)));
+                                          INT2NUM(sizeof(ca_u_int64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -3438,6 +3438,7 @@ cumcount_ki_native_i16 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -3453,12 +3454,11 @@ cumcount_ki_native_i16 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(int16_t, int64_t, st_in, pi, mi,
+    CA_SLAB_SCAN_T(ca_u_int16_t, ca_u_int64_t, st_in, pi, mi,
                          st_out, po, 0, (void) v; r = ++acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -3471,7 +3471,7 @@ static VALUE
 cumcount_ki_native_u16 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_INT64),
-                                          INT2NUM(sizeof(int64_t)));
+                                          INT2NUM(sizeof(ca_u_int64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -3480,6 +3480,7 @@ cumcount_ki_native_u16 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -3495,12 +3496,11 @@ cumcount_ki_native_u16 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(uint16_t, int64_t, st_in, pi, mi,
+    CA_SLAB_SCAN_T(ca_u_uint16_t, ca_u_int64_t, st_in, pi, mi,
                          st_out, po, 0, (void) v; r = ++acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -3513,7 +3513,7 @@ static VALUE
 cumcount_ki_native_i32 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_INT64),
-                                          INT2NUM(sizeof(int64_t)));
+                                          INT2NUM(sizeof(ca_u_int64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -3522,6 +3522,7 @@ cumcount_ki_native_i32 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -3537,12 +3538,11 @@ cumcount_ki_native_i32 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(int32_t, int64_t, st_in, pi, mi,
+    CA_SLAB_SCAN_T(ca_u_int32_t, ca_u_int64_t, st_in, pi, mi,
                          st_out, po, 0, (void) v; r = ++acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -3555,7 +3555,7 @@ static VALUE
 cumcount_ki_native_u32 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_INT64),
-                                          INT2NUM(sizeof(int64_t)));
+                                          INT2NUM(sizeof(ca_u_int64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -3564,6 +3564,7 @@ cumcount_ki_native_u32 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -3579,12 +3580,11 @@ cumcount_ki_native_u32 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(uint32_t, int64_t, st_in, pi, mi,
+    CA_SLAB_SCAN_T(ca_u_uint32_t, ca_u_int64_t, st_in, pi, mi,
                          st_out, po, 0, (void) v; r = ++acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -3597,7 +3597,7 @@ static VALUE
 cumcount_ki_native_i64 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_INT64),
-                                          INT2NUM(sizeof(int64_t)));
+                                          INT2NUM(sizeof(ca_u_int64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -3606,6 +3606,7 @@ cumcount_ki_native_i64 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -3621,12 +3622,11 @@ cumcount_ki_native_i64 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(int64_t, int64_t, st_in, pi, mi,
+    CA_SLAB_SCAN_T(ca_u_int64_t, ca_u_int64_t, st_in, pi, mi,
                          st_out, po, 0, (void) v; r = ++acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -3639,7 +3639,7 @@ static VALUE
 cumcount_ki_native_u64 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_INT64),
-                                          INT2NUM(sizeof(int64_t)));
+                                          INT2NUM(sizeof(ca_u_int64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -3648,6 +3648,7 @@ cumcount_ki_native_u64 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -3663,12 +3664,11 @@ cumcount_ki_native_u64 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(uint64_t, int64_t, st_in, pi, mi,
+    CA_SLAB_SCAN_T(ca_u_uint64_t, ca_u_int64_t, st_in, pi, mi,
                          st_out, po, 0, (void) v; r = ++acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -3681,7 +3681,7 @@ static VALUE
 cumcount_ki_native_f32 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_INT64),
-                                          INT2NUM(sizeof(int64_t)));
+                                          INT2NUM(sizeof(ca_u_int64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -3690,6 +3690,7 @@ cumcount_ki_native_f32 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -3705,12 +3706,11 @@ cumcount_ki_native_f32 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(float, int64_t, st_in, pi, mi,
+    CA_SLAB_SCAN_T(ca_u_float32_t, ca_u_int64_t, st_in, pi, mi,
                          st_out, po, 0, (void) v; r = ++acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -3723,7 +3723,7 @@ static VALUE
 cumcount_ki_native_f64 (VALUE self, CArray *ca, int axis)
 {
   VALUE   vout = rb_ca_template_with_type(self, INT2NUM(CA_INT64),
-                                          INT2NUM(sizeof(int64_t)));
+                                          INT2NUM(sizeof(ca_u_int64_t)));
   CArray *co;
   GetCArray(vout, co);
 
@@ -3732,6 +3732,7 @@ cumcount_ki_native_f64 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -3747,12 +3748,11 @@ cumcount_ki_native_f64 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_T(double, int64_t, st_in, pi, mi,
+    CA_SLAB_SCAN_T(ca_u_float64_t, ca_u_int64_t, st_in, pi, mi,
                          st_out, po, 0, (void) v; r = ++acc);
     ca_iter_state_sync_slab(&st_out);
   }
@@ -3834,6 +3834,7 @@ uniq_scan_ki_native_i8 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -3849,7 +3850,6 @@ uniq_scan_ki_native_i8 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
@@ -3876,6 +3876,7 @@ uniq_scan_ki_native_u8 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -3891,7 +3892,6 @@ uniq_scan_ki_native_u8 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
@@ -3918,6 +3918,7 @@ uniq_scan_ki_native_i16 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -3933,12 +3934,11 @@ uniq_scan_ki_native_i16 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_TA(int16_t, boolean8_t, int16_t, st_in, pi, mi,
+    CA_SLAB_SCAN_TA(ca_u_int16_t, boolean8_t, ca_u_int16_t, st_in, pi, mi,
                          st_out, po, 0, if (first) { acc = v; r = 0; } else if (v == acc) { r = 1; } else { acc = v; r = 0; });
     ca_iter_state_sync_slab(&st_out);
   }
@@ -3960,6 +3960,7 @@ uniq_scan_ki_native_u16 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -3975,12 +3976,11 @@ uniq_scan_ki_native_u16 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_TA(uint16_t, boolean8_t, uint16_t, st_in, pi, mi,
+    CA_SLAB_SCAN_TA(ca_u_uint16_t, boolean8_t, ca_u_uint16_t, st_in, pi, mi,
                          st_out, po, 0, if (first) { acc = v; r = 0; } else if (v == acc) { r = 1; } else { acc = v; r = 0; });
     ca_iter_state_sync_slab(&st_out);
   }
@@ -4002,6 +4002,7 @@ uniq_scan_ki_native_i32 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -4017,12 +4018,11 @@ uniq_scan_ki_native_i32 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_TA(int32_t, boolean8_t, int32_t, st_in, pi, mi,
+    CA_SLAB_SCAN_TA(ca_u_int32_t, boolean8_t, ca_u_int32_t, st_in, pi, mi,
                          st_out, po, 0, if (first) { acc = v; r = 0; } else if (v == acc) { r = 1; } else { acc = v; r = 0; });
     ca_iter_state_sync_slab(&st_out);
   }
@@ -4044,6 +4044,7 @@ uniq_scan_ki_native_u32 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -4059,12 +4060,11 @@ uniq_scan_ki_native_u32 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_TA(uint32_t, boolean8_t, uint32_t, st_in, pi, mi,
+    CA_SLAB_SCAN_TA(ca_u_uint32_t, boolean8_t, ca_u_uint32_t, st_in, pi, mi,
                          st_out, po, 0, if (first) { acc = v; r = 0; } else if (v == acc) { r = 1; } else { acc = v; r = 0; });
     ca_iter_state_sync_slab(&st_out);
   }
@@ -4086,6 +4086,7 @@ uniq_scan_ki_native_i64 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -4101,12 +4102,11 @@ uniq_scan_ki_native_i64 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_TA(int64_t, boolean8_t, int64_t, st_in, pi, mi,
+    CA_SLAB_SCAN_TA(ca_u_int64_t, boolean8_t, ca_u_int64_t, st_in, pi, mi,
                          st_out, po, 0, if (first) { acc = v; r = 0; } else if (v == acc) { r = 1; } else { acc = v; r = 0; });
     ca_iter_state_sync_slab(&st_out);
   }
@@ -4128,6 +4128,7 @@ uniq_scan_ki_native_u64 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -4143,12 +4144,11 @@ uniq_scan_ki_native_u64 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_TA(uint64_t, boolean8_t, uint64_t, st_in, pi, mi,
+    CA_SLAB_SCAN_TA(ca_u_uint64_t, boolean8_t, ca_u_uint64_t, st_in, pi, mi,
                          st_out, po, 0, if (first) { acc = v; r = 0; } else if (v == acc) { r = 1; } else { acc = v; r = 0; });
     ca_iter_state_sync_slab(&st_out);
   }
@@ -4170,6 +4170,7 @@ uniq_scan_ki_native_f32 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -4185,12 +4186,11 @@ uniq_scan_ki_native_f32 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_TA(float, boolean8_t, float, st_in, pi, mi,
+    CA_SLAB_SCAN_TA(ca_u_float32_t, boolean8_t, ca_u_float32_t, st_in, pi, mi,
                          st_out, po, 0, if (first) { acc = v; r = 0; } else if (v == acc) { r = 1; } else { acc = v; r = 0; });
     ca_iter_state_sync_slab(&st_out);
   }
@@ -4212,6 +4212,7 @@ uniq_scan_ki_native_f64 (VALUE self, CArray *ca, int axis)
      INIT inside the macro. */
   int8_t slab_axes[CA_RANK_MAX];
   slab_axes[0] = (int8_t) axis;
+
   ca_iter_state st_in, st_out;
   int rc;
   rc = ca_iter_state_init_l2(&st_in,  ca, CA_SLAB_AXES,
@@ -4227,12 +4228,11 @@ uniq_scan_ki_native_f64 (VALUE self, CArray *ca, int axis)
     rb_raise(rb_eRuntimeError,
              "%s: output init failed rc=%d", ca_calling_method_name(), rc);
   }
-
   char       *pi, *po;
   boolean8_t *mi, *mo;
   while ( ca_iter_state_next_slab_axes(&st_in,  &pi, &mi) &&
           ca_iter_state_next_slab_axes(&st_out, &po, &mo) ) {
-    CA_SLAB_SCAN_TA(double, boolean8_t, double, st_in, pi, mi,
+    CA_SLAB_SCAN_TA(ca_u_float64_t, boolean8_t, ca_u_float64_t, st_in, pi, mi,
                          st_out, po, 0, if (first) { acc = v; r = 0; } else if (v == acc) { r = 1; } else { acc = v; r = 0; });
     ca_iter_state_sync_slab(&st_out);
   }