carray 2.0.1 → 3.0.1

data/ext/carray_kernels_triop.c

@@ -0,0 +1,1392 @@
+/* ---------------------------------------------------------------------------
+
+   carray_kernels.c -- GENERATED by ext/mkkernel.rb -- DO NOT EDIT
+
+   Kernel definitions live in mkkernel.rb (search for `MkKernel.reduce`).
+   Regenerate by running:
+
+     cd ext && ruby mkkernel.rb > carray_kernels.c
+
+   extconf.rb regenerates automatically when mkkernel.rb is touched.
+
+   --------------------------------------------------------------------------- */
+
+#include "carray.h"
+#include "ca_kernel_iterator.h"
+#include "ca_sort_kernels.h"   /* sort kernels: internal, not via the carray.h umbrella */
+#include "carray_internal.h"   /* ca_lazy_arena_*, ca_is_lazy_view (streaming reduce) */
+#include "ca_obj_face.h"       /* rb_ca_strip_face_value (Face ordering gate) */
+#include <math.h>
+#include <stdint.h>
+#include <stdlib.h>   /* qsort, mergesort (HAVE_MERGESORT) -- for :sort kind */
+
+/* MEMO_REDUCTION_FASTPATH_ENTITY_ONLY_GUARD Tier 2: CAStack identity is
+   probed by operation-table function pointer (obj_type is runtime-
+   assigned, so no CA_OBJ_STACK compile-time constant exists). */
+extern ca_operation_function_t ca_stack_func;
+/* BOOL2VAL: bool -> Ruby (Qtrue/Qfalse).  Used as ruby_scalar wrapper
+   for the all/any flat-reduction Ruby surface so `a.all` / `a.any`
+   return real true/false, not Integer 0/1.  (BOOL2OBJ already exists
+   but returns Integer per CArray's boolean-as-int convention.)  */
+#ifndef CARRAY_BOOL2VAL_DEFINED
+#define CARRAY_BOOL2VAL_DEFINED
+static inline VALUE BOOL2VAL (boolean8_t x) { return x ? Qtrue : Qfalse; }
+#endif
+
+#include "ca_op_powi.h"
+#include "ca_op_cmplx64.h"
+
+/* ===== triop fma ============================================ */
+
+static void
+ca_triop_fma_int8_t (ca_size_t n, boolean8_t *m,
+                                 char *ptr1, ca_size_t i1,
+                                 char *ptr2, ca_size_t i2,
+                                 char *ptr3, ca_size_t i3,
+                                 char *ptr4, ca_size_t i4)
+{
+  int8_t *q1 = (int8_t *) ptr1, *q2 = (int8_t *) ptr2,
+            *q3 = (int8_t *) ptr3, *q4 = (int8_t *) ptr4;
+  int8_t *p1 = q1, *p2 = q2, *p3 = q3, *p4 = q4;
+  ca_size_t k;
+  if ( m ) {
+    boolean8_t *pm;
+    for (k=0; k<n; k++) {
+      pm = m + k;
+      if ( ! *pm ) {
+        p1 = q1 + k*i1;
+        p2 = q2 + k*i2;
+        p3 = q3 + k*i3;
+        p4 = q4 + k*i4;
+        { (*p4) = (*p1) * (*p2) + (*p3); }
+      }
+    }
+  } else {
+    for (k=0; k<n; k++) {
+      p1 = q1 + k*i1;
+      p2 = q2 + k*i2;
+      p3 = q3 + k*i3;
+      p4 = q4 + k*i4;
+      { (*p4) = (*p1) * (*p2) + (*p3); }
+    }
+  }
+}
+
+static void
+ca_triop_fma_uint8_t (ca_size_t n, boolean8_t *m,
+                                 char *ptr1, ca_size_t i1,
+                                 char *ptr2, ca_size_t i2,
+                                 char *ptr3, ca_size_t i3,
+                                 char *ptr4, ca_size_t i4)
+{
+  uint8_t *q1 = (uint8_t *) ptr1, *q2 = (uint8_t *) ptr2,
+            *q3 = (uint8_t *) ptr3, *q4 = (uint8_t *) ptr4;
+  uint8_t *p1 = q1, *p2 = q2, *p3 = q3, *p4 = q4;
+  ca_size_t k;
+  if ( m ) {
+    boolean8_t *pm;
+    for (k=0; k<n; k++) {
+      pm = m + k;
+      if ( ! *pm ) {
+        p1 = q1 + k*i1;
+        p2 = q2 + k*i2;
+        p3 = q3 + k*i3;
+        p4 = q4 + k*i4;
+        { (*p4) = (*p1) * (*p2) + (*p3); }
+      }
+    }
+  } else {
+    for (k=0; k<n; k++) {
+      p1 = q1 + k*i1;
+      p2 = q2 + k*i2;
+      p3 = q3 + k*i3;
+      p4 = q4 + k*i4;
+      { (*p4) = (*p1) * (*p2) + (*p3); }
+    }
+  }
+}
+
+static void
+ca_triop_fma_int16_t (ca_size_t n, boolean8_t *m,
+                                 char *ptr1, ca_size_t i1,
+                                 char *ptr2, ca_size_t i2,
+                                 char *ptr3, ca_size_t i3,
+                                 char *ptr4, ca_size_t i4)
+{
+  int16_t *q1 = (int16_t *) ptr1, *q2 = (int16_t *) ptr2,
+            *q3 = (int16_t *) ptr3, *q4 = (int16_t *) ptr4;
+  int16_t *p1 = q1, *p2 = q2, *p3 = q3, *p4 = q4;
+  ca_size_t k;
+  if ( m ) {
+    boolean8_t *pm;
+    for (k=0; k<n; k++) {
+      pm = m + k;
+      if ( ! *pm ) {
+        p1 = q1 + k*i1;
+        p2 = q2 + k*i2;
+        p3 = q3 + k*i3;
+        p4 = q4 + k*i4;
+        { (*p4) = (*p1) * (*p2) + (*p3); }
+      }
+    }
+  } else {
+    for (k=0; k<n; k++) {
+      p1 = q1 + k*i1;
+      p2 = q2 + k*i2;
+      p3 = q3 + k*i3;
+      p4 = q4 + k*i4;
+      { (*p4) = (*p1) * (*p2) + (*p3); }
+    }
+  }
+}
+
+static void
+ca_triop_fma_uint16_t (ca_size_t n, boolean8_t *m,
+                                 char *ptr1, ca_size_t i1,
+                                 char *ptr2, ca_size_t i2,
+                                 char *ptr3, ca_size_t i3,
+                                 char *ptr4, ca_size_t i4)
+{
+  uint16_t *q1 = (uint16_t *) ptr1, *q2 = (uint16_t *) ptr2,
+            *q3 = (uint16_t *) ptr3, *q4 = (uint16_t *) ptr4;
+  uint16_t *p1 = q1, *p2 = q2, *p3 = q3, *p4 = q4;
+  ca_size_t k;
+  if ( m ) {
+    boolean8_t *pm;
+    for (k=0; k<n; k++) {
+      pm = m + k;
+      if ( ! *pm ) {
+        p1 = q1 + k*i1;
+        p2 = q2 + k*i2;
+        p3 = q3 + k*i3;
+        p4 = q4 + k*i4;
+        { (*p4) = (*p1) * (*p2) + (*p3); }
+      }
+    }
+  } else {
+    for (k=0; k<n; k++) {
+      p1 = q1 + k*i1;
+      p2 = q2 + k*i2;
+      p3 = q3 + k*i3;
+      p4 = q4 + k*i4;
+      { (*p4) = (*p1) * (*p2) + (*p3); }
+    }
+  }
+}
+
+static void
+ca_triop_fma_int32_t (ca_size_t n, boolean8_t *m,
+                                 char *ptr1, ca_size_t i1,
+                                 char *ptr2, ca_size_t i2,
+                                 char *ptr3, ca_size_t i3,
+                                 char *ptr4, ca_size_t i4)
+{
+  int32_t *q1 = (int32_t *) ptr1, *q2 = (int32_t *) ptr2,
+            *q3 = (int32_t *) ptr3, *q4 = (int32_t *) ptr4;
+  int32_t *p1 = q1, *p2 = q2, *p3 = q3, *p4 = q4;
+  ca_size_t k;
+  if ( m ) {
+    boolean8_t *pm;
+    for (k=0; k<n; k++) {
+      pm = m + k;
+      if ( ! *pm ) {
+        p1 = q1 + k*i1;
+        p2 = q2 + k*i2;
+        p3 = q3 + k*i3;
+        p4 = q4 + k*i4;
+        { (*p4) = (*p1) * (*p2) + (*p3); }
+      }
+    }
+  } else {
+    for (k=0; k<n; k++) {
+      p1 = q1 + k*i1;
+      p2 = q2 + k*i2;
+      p3 = q3 + k*i3;
+      p4 = q4 + k*i4;
+      { (*p4) = (*p1) * (*p2) + (*p3); }
+    }
+  }
+}
+
+static void
+ca_triop_fma_uint32_t (ca_size_t n, boolean8_t *m,
+                                 char *ptr1, ca_size_t i1,
+                                 char *ptr2, ca_size_t i2,
+                                 char *ptr3, ca_size_t i3,
+                                 char *ptr4, ca_size_t i4)
+{
+  uint32_t *q1 = (uint32_t *) ptr1, *q2 = (uint32_t *) ptr2,
+            *q3 = (uint32_t *) ptr3, *q4 = (uint32_t *) ptr4;
+  uint32_t *p1 = q1, *p2 = q2, *p3 = q3, *p4 = q4;
+  ca_size_t k;
+  if ( m ) {
+    boolean8_t *pm;
+    for (k=0; k<n; k++) {
+      pm = m + k;
+      if ( ! *pm ) {
+        p1 = q1 + k*i1;
+        p2 = q2 + k*i2;
+        p3 = q3 + k*i3;
+        p4 = q4 + k*i4;
+        { (*p4) = (*p1) * (*p2) + (*p3); }
+      }
+    }
+  } else {
+    for (k=0; k<n; k++) {
+      p1 = q1 + k*i1;
+      p2 = q2 + k*i2;
+      p3 = q3 + k*i3;
+      p4 = q4 + k*i4;
+      { (*p4) = (*p1) * (*p2) + (*p3); }
+    }
+  }
+}
+
+static void
+ca_triop_fma_int64_t (ca_size_t n, boolean8_t *m,
+                                 char *ptr1, ca_size_t i1,
+                                 char *ptr2, ca_size_t i2,
+                                 char *ptr3, ca_size_t i3,
+                                 char *ptr4, ca_size_t i4)
+{
+  int64_t *q1 = (int64_t *) ptr1, *q2 = (int64_t *) ptr2,
+            *q3 = (int64_t *) ptr3, *q4 = (int64_t *) ptr4;
+  int64_t *p1 = q1, *p2 = q2, *p3 = q3, *p4 = q4;
+  ca_size_t k;
+  if ( m ) {
+    boolean8_t *pm;
+    for (k=0; k<n; k++) {
+      pm = m + k;
+      if ( ! *pm ) {
+        p1 = q1 + k*i1;
+        p2 = q2 + k*i2;
+        p3 = q3 + k*i3;
+        p4 = q4 + k*i4;
+        { (*p4) = (*p1) * (*p2) + (*p3); }
+      }
+    }
+  } else {
+    for (k=0; k<n; k++) {
+      p1 = q1 + k*i1;
+      p2 = q2 + k*i2;
+      p3 = q3 + k*i3;
+      p4 = q4 + k*i4;
+      { (*p4) = (*p1) * (*p2) + (*p3); }
+    }
+  }
+}
+
+static void
+ca_triop_fma_uint64_t (ca_size_t n, boolean8_t *m,
+                                 char *ptr1, ca_size_t i1,
+                                 char *ptr2, ca_size_t i2,
+                                 char *ptr3, ca_size_t i3,
+                                 char *ptr4, ca_size_t i4)
+{
+  uint64_t *q1 = (uint64_t *) ptr1, *q2 = (uint64_t *) ptr2,
+            *q3 = (uint64_t *) ptr3, *q4 = (uint64_t *) ptr4;
+  uint64_t *p1 = q1, *p2 = q2, *p3 = q3, *p4 = q4;
+  ca_size_t k;
+  if ( m ) {
+    boolean8_t *pm;
+    for (k=0; k<n; k++) {
+      pm = m + k;
+      if ( ! *pm ) {
+        p1 = q1 + k*i1;
+        p2 = q2 + k*i2;
+        p3 = q3 + k*i3;
+        p4 = q4 + k*i4;
+        { (*p4) = (*p1) * (*p2) + (*p3); }
+      }
+    }
+  } else {
+    for (k=0; k<n; k++) {
+      p1 = q1 + k*i1;
+      p2 = q2 + k*i2;
+      p3 = q3 + k*i3;
+      p4 = q4 + k*i4;
+      { (*p4) = (*p1) * (*p2) + (*p3); }
+    }
+  }
+}
+
+static void
+ca_triop_fma_float32_t (ca_size_t n, boolean8_t *m,
+                                 char *ptr1, ca_size_t i1,
+                                 char *ptr2, ca_size_t i2,
+                                 char *ptr3, ca_size_t i3,
+                                 char *ptr4, ca_size_t i4)
+{
+  float *q1 = (float *) ptr1, *q2 = (float *) ptr2,
+            *q3 = (float *) ptr3, *q4 = (float *) ptr4;
+  float *p1 = q1, *p2 = q2, *p3 = q3, *p4 = q4;
+  ca_size_t k;
+  if ( m ) {
+    boolean8_t *pm;
+    for (k=0; k<n; k++) {
+      pm = m + k;
+      if ( ! *pm ) {
+        p1 = q1 + k*i1;
+        p2 = q2 + k*i2;
+        p3 = q3 + k*i3;
+        p4 = q4 + k*i4;
+        { (*p4) = fma((*p1), (*p2), (*p3)); }
+      }
+    }
+  } else {
+    for (k=0; k<n; k++) {
+      p1 = q1 + k*i1;
+      p2 = q2 + k*i2;
+      p3 = q3 + k*i3;
+      p4 = q4 + k*i4;
+      { (*p4) = fma((*p1), (*p2), (*p3)); }
+    }
+  }
+}
+
+static void
+ca_triop_fma_float64_t (ca_size_t n, boolean8_t *m,
+                                 char *ptr1, ca_size_t i1,
+                                 char *ptr2, ca_size_t i2,
+                                 char *ptr3, ca_size_t i3,
+                                 char *ptr4, ca_size_t i4)
+{
+  double *q1 = (double *) ptr1, *q2 = (double *) ptr2,
+            *q3 = (double *) ptr3, *q4 = (double *) ptr4;
+  double *p1 = q1, *p2 = q2, *p3 = q3, *p4 = q4;
+  ca_size_t k;
+  if ( m ) {
+    boolean8_t *pm;
+    for (k=0; k<n; k++) {
+      pm = m + k;
+      if ( ! *pm ) {
+        p1 = q1 + k*i1;
+        p2 = q2 + k*i2;
+        p3 = q3 + k*i3;
+        p4 = q4 + k*i4;
+        { (*p4) = fma((*p1), (*p2), (*p3)); }
+      }
+    }
+  } else {
+    for (k=0; k<n; k++) {
+      p1 = q1 + k*i1;
+      p2 = q2 + k*i2;
+      p3 = q3 + k*i3;
+      p4 = q4 + k*i4;
+      { (*p4) = fma((*p1), (*p2), (*p3)); }
+    }
+  }
+}
+
+static void
+ca_triop_fma_cmplx64_t (ca_size_t n, boolean8_t *m,
+                                 char *ptr1, ca_size_t i1,
+                                 char *ptr2, ca_size_t i2,
+                                 char *ptr3, ca_size_t i3,
+                                 char *ptr4, ca_size_t i4)
+{
+  cmplx64_t *q1 = (cmplx64_t *) ptr1, *q2 = (cmplx64_t *) ptr2,
+            *q3 = (cmplx64_t *) ptr3, *q4 = (cmplx64_t *) ptr4;
+  cmplx64_t *p1 = q1, *p2 = q2, *p3 = q3, *p4 = q4;
+  ca_size_t k;
+  if ( m ) {
+    boolean8_t *pm;
+    for (k=0; k<n; k++) {
+      pm = m + k;
+      if ( ! *pm ) {
+        p1 = q1 + k*i1;
+        p2 = q2 + k*i2;
+        p3 = q3 + k*i3;
+        p4 = q4 + k*i4;
+        { (*p4) = (*p1) * (*p2) + (*p3); }
+      }
+    }
+  } else {
+    for (k=0; k<n; k++) {
+      p1 = q1 + k*i1;
+      p2 = q2 + k*i2;
+      p3 = q3 + k*i3;
+      p4 = q4 + k*i4;
+      { (*p4) = (*p1) * (*p2) + (*p3); }
+    }
+  }
+}
+
+static void
+ca_triop_fma_cmplx128_t (ca_size_t n, boolean8_t *m,
+                                 char *ptr1, ca_size_t i1,
+                                 char *ptr2, ca_size_t i2,
+                                 char *ptr3, ca_size_t i3,
+                                 char *ptr4, ca_size_t i4)
+{
+  cmplx128_t *q1 = (cmplx128_t *) ptr1, *q2 = (cmplx128_t *) ptr2,
+            *q3 = (cmplx128_t *) ptr3, *q4 = (cmplx128_t *) ptr4;
+  cmplx128_t *p1 = q1, *p2 = q2, *p3 = q3, *p4 = q4;
+  ca_size_t k;
+  if ( m ) {
+    boolean8_t *pm;
+    for (k=0; k<n; k++) {
+      pm = m + k;
+      if ( ! *pm ) {
+        p1 = q1 + k*i1;
+        p2 = q2 + k*i2;
+        p3 = q3 + k*i3;
+        p4 = q4 + k*i4;
+        { (*p4) = (*p1) * (*p2) + (*p3); }
+      }
+    }
+  } else {
+    for (k=0; k<n; k++) {
+      p1 = q1 + k*i1;
+      p2 = q2 + k*i2;
+      p3 = q3 + k*i3;
+      p4 = q4 + k*i4;
+      { (*p4) = (*p1) * (*p2) + (*p3); }
+    }
+  }
+}
+
+static void
+ca_triop_fma_VALUE (ca_size_t n, boolean8_t *m,
+                                 char *ptr1, ca_size_t i1,
+                                 char *ptr2, ca_size_t i2,
+                                 char *ptr3, ca_size_t i3,
+                                 char *ptr4, ca_size_t i4)
+{
+  VALUE *q1 = (VALUE *) ptr1, *q2 = (VALUE *) ptr2,
+            *q3 = (VALUE *) ptr3, *q4 = (VALUE *) ptr4;
+  VALUE *p1 = q1, *p2 = q2, *p3 = q3, *p4 = q4;
+  ca_size_t k;
+  if ( m ) {
+    boolean8_t *pm;
+    for (k=0; k<n; k++) {
+      pm = m + k;
+      if ( ! *pm ) {
+        p1 = q1 + k*i1;
+        p2 = q2 + k*i2;
+        p3 = q3 + k*i3;
+        p4 = q4 + k*i4;
+        { (*p4) = rb_funcall(rb_funcall((*p1), rb_intern("*"), 1, (*p2)), rb_intern("+"), 1, (*p3)); }
+      }
+    }
+  } else {
+    for (k=0; k<n; k++) {
+      p1 = q1 + k*i1;
+      p2 = q2 + k*i2;
+      p3 = q3 + k*i3;
+      p4 = q4 + k*i4;
+      { (*p4) = rb_funcall(rb_funcall((*p1), rb_intern("*"), 1, (*p2)), rb_intern("+"), 1, (*p3)); }
+    }
+  }
+}
+
+ca_triop_func_t
+ca_triop_fma[CA_NTYPE] = {
+  ca_triop_not_implement,
+  ca_triop_not_implement,
+  ca_triop_fma_int8_t,
+  ca_triop_fma_uint8_t,
+  ca_triop_fma_int16_t,
+  ca_triop_fma_uint16_t,
+  ca_triop_fma_int32_t,
+  ca_triop_fma_uint32_t,
+  ca_triop_fma_int64_t,
+  ca_triop_fma_uint64_t,
+  ca_triop_fma_float32_t,
+  ca_triop_fma_float64_t,
+  ca_triop_not_implement,
+  ca_triop_fma_cmplx64_t,
+  ca_triop_fma_cmplx128_t,
+  ca_triop_not_implement,
+  ca_triop_fma_VALUE,
+};
+
+VALUE rb_ca_fma (VALUE self, VALUE other2, VALUE other3)
+{ return rb_ca_call_triop(self, other2, other3, ca_triop_fma); }
+
+VALUE rb_ca_fma_bang (VALUE self, VALUE other2, VALUE other3)
+{ return rb_ca_call_triop_bang(self, other2, other3, ca_triop_fma); }
+
+/* ===== triop fms ============================================ */
+
+static void
+ca_triop_fms_int8_t (ca_size_t n, boolean8_t *m,
+                                 char *ptr1, ca_size_t i1,
+                                 char *ptr2, ca_size_t i2,
+                                 char *ptr3, ca_size_t i3,
+                                 char *ptr4, ca_size_t i4)
+{
+  int8_t *q1 = (int8_t *) ptr1, *q2 = (int8_t *) ptr2,
+            *q3 = (int8_t *) ptr3, *q4 = (int8_t *) ptr4;
+  int8_t *p1 = q1, *p2 = q2, *p3 = q3, *p4 = q4;
+  ca_size_t k;
+  if ( m ) {
+    boolean8_t *pm;
+    for (k=0; k<n; k++) {
+      pm = m + k;
+      if ( ! *pm ) {
+        p1 = q1 + k*i1;
+        p2 = q2 + k*i2;
+        p3 = q3 + k*i3;
+        p4 = q4 + k*i4;
+        { (*p4) = (*p1) * (*p2) - (*p3); }
+      }
+    }
+  } else {
+    for (k=0; k<n; k++) {
+      p1 = q1 + k*i1;
+      p2 = q2 + k*i2;
+      p3 = q3 + k*i3;
+      p4 = q4 + k*i4;
+      { (*p4) = (*p1) * (*p2) - (*p3); }
+    }
+  }
+}
+
+static void
+ca_triop_fms_uint8_t (ca_size_t n, boolean8_t *m,
+                                 char *ptr1, ca_size_t i1,
+                                 char *ptr2, ca_size_t i2,
+                                 char *ptr3, ca_size_t i3,
+                                 char *ptr4, ca_size_t i4)
+{
+  uint8_t *q1 = (uint8_t *) ptr1, *q2 = (uint8_t *) ptr2,
+            *q3 = (uint8_t *) ptr3, *q4 = (uint8_t *) ptr4;
+  uint8_t *p1 = q1, *p2 = q2, *p3 = q3, *p4 = q4;
+  ca_size_t k;
+  if ( m ) {
+    boolean8_t *pm;
+    for (k=0; k<n; k++) {
+      pm = m + k;
+      if ( ! *pm ) {
+        p1 = q1 + k*i1;
+        p2 = q2 + k*i2;
+        p3 = q3 + k*i3;
+        p4 = q4 + k*i4;
+        { (*p4) = (*p1) * (*p2) - (*p3); }
+      }
+    }
+  } else {
+    for (k=0; k<n; k++) {
+      p1 = q1 + k*i1;
+      p2 = q2 + k*i2;
+      p3 = q3 + k*i3;
+      p4 = q4 + k*i4;
+      { (*p4) = (*p1) * (*p2) - (*p3); }
+    }
+  }
+}
+
+static void
+ca_triop_fms_int16_t (ca_size_t n, boolean8_t *m,
+                                 char *ptr1, ca_size_t i1,
+                                 char *ptr2, ca_size_t i2,
+                                 char *ptr3, ca_size_t i3,
+                                 char *ptr4, ca_size_t i4)
+{
+  int16_t *q1 = (int16_t *) ptr1, *q2 = (int16_t *) ptr2,
+            *q3 = (int16_t *) ptr3, *q4 = (int16_t *) ptr4;
+  int16_t *p1 = q1, *p2 = q2, *p3 = q3, *p4 = q4;
+  ca_size_t k;
+  if ( m ) {
+    boolean8_t *pm;
+    for (k=0; k<n; k++) {
+      pm = m + k;
+      if ( ! *pm ) {
+        p1 = q1 + k*i1;
+        p2 = q2 + k*i2;
+        p3 = q3 + k*i3;
+        p4 = q4 + k*i4;
+        { (*p4) = (*p1) * (*p2) - (*p3); }
+      }
+    }
+  } else {
+    for (k=0; k<n; k++) {
+      p1 = q1 + k*i1;
+      p2 = q2 + k*i2;
+      p3 = q3 + k*i3;
+      p4 = q4 + k*i4;
+      { (*p4) = (*p1) * (*p2) - (*p3); }
+    }
+  }
+}
+
+static void
+ca_triop_fms_uint16_t (ca_size_t n, boolean8_t *m,
+                                 char *ptr1, ca_size_t i1,
+                                 char *ptr2, ca_size_t i2,
+                                 char *ptr3, ca_size_t i3,
+                                 char *ptr4, ca_size_t i4)
+{
+  uint16_t *q1 = (uint16_t *) ptr1, *q2 = (uint16_t *) ptr2,
+            *q3 = (uint16_t *) ptr3, *q4 = (uint16_t *) ptr4;
+  uint16_t *p1 = q1, *p2 = q2, *p3 = q3, *p4 = q4;
+  ca_size_t k;
+  if ( m ) {
+    boolean8_t *pm;
+    for (k=0; k<n; k++) {
+      pm = m + k;
+      if ( ! *pm ) {
+        p1 = q1 + k*i1;
+        p2 = q2 + k*i2;
+        p3 = q3 + k*i3;
+        p4 = q4 + k*i4;
+        { (*p4) = (*p1) * (*p2) - (*p3); }
+      }
+    }
+  } else {
+    for (k=0; k<n; k++) {
+      p1 = q1 + k*i1;
+      p2 = q2 + k*i2;
+      p3 = q3 + k*i3;
+      p4 = q4 + k*i4;
+      { (*p4) = (*p1) * (*p2) - (*p3); }
+    }
+  }
+}
+
+static void
+ca_triop_fms_int32_t (ca_size_t n, boolean8_t *m,
+                                 char *ptr1, ca_size_t i1,
+                                 char *ptr2, ca_size_t i2,
+                                 char *ptr3, ca_size_t i3,
+                                 char *ptr4, ca_size_t i4)
+{
+  int32_t *q1 = (int32_t *) ptr1, *q2 = (int32_t *) ptr2,
+            *q3 = (int32_t *) ptr3, *q4 = (int32_t *) ptr4;
+  int32_t *p1 = q1, *p2 = q2, *p3 = q3, *p4 = q4;
+  ca_size_t k;
+  if ( m ) {
+    boolean8_t *pm;
+    for (k=0; k<n; k++) {
+      pm = m + k;
+      if ( ! *pm ) {
+        p1 = q1 + k*i1;
+        p2 = q2 + k*i2;
+        p3 = q3 + k*i3;
+        p4 = q4 + k*i4;
+        { (*p4) = (*p1) * (*p2) - (*p3); }
+      }
+    }
+  } else {
+    for (k=0; k<n; k++) {
+      p1 = q1 + k*i1;
+      p2 = q2 + k*i2;
+      p3 = q3 + k*i3;
+      p4 = q4 + k*i4;
+      { (*p4) = (*p1) * (*p2) - (*p3); }
+    }
+  }
+}
+
+static void
+ca_triop_fms_uint32_t (ca_size_t n, boolean8_t *m,
+                                 char *ptr1, ca_size_t i1,
+                                 char *ptr2, ca_size_t i2,
+                                 char *ptr3, ca_size_t i3,
+                                 char *ptr4, ca_size_t i4)
+{
+  uint32_t *q1 = (uint32_t *) ptr1, *q2 = (uint32_t *) ptr2,
+            *q3 = (uint32_t *) ptr3, *q4 = (uint32_t *) ptr4;
+  uint32_t *p1 = q1, *p2 = q2, *p3 = q3, *p4 = q4;
+  ca_size_t k;
+  if ( m ) {
+    boolean8_t *pm;
+    for (k=0; k<n; k++) {
+      pm = m + k;
+      if ( ! *pm ) {
+        p1 = q1 + k*i1;
+        p2 = q2 + k*i2;
+        p3 = q3 + k*i3;
+        p4 = q4 + k*i4;
+        { (*p4) = (*p1) * (*p2) - (*p3); }
+      }
+    }
+  } else {
+    for (k=0; k<n; k++) {
+      p1 = q1 + k*i1;
+      p2 = q2 + k*i2;
+      p3 = q3 + k*i3;
+      p4 = q4 + k*i4;
+      { (*p4) = (*p1) * (*p2) - (*p3); }
+    }
+  }
+}
+
+static void
+ca_triop_fms_int64_t (ca_size_t n, boolean8_t *m,
+                                 char *ptr1, ca_size_t i1,
+                                 char *ptr2, ca_size_t i2,
+                                 char *ptr3, ca_size_t i3,
+                                 char *ptr4, ca_size_t i4)
+{
+  int64_t *q1 = (int64_t *) ptr1, *q2 = (int64_t *) ptr2,
+            *q3 = (int64_t *) ptr3, *q4 = (int64_t *) ptr4;
+  int64_t *p1 = q1, *p2 = q2, *p3 = q3, *p4 = q4;
+  ca_size_t k;
+  if ( m ) {
+    boolean8_t *pm;
+    for (k=0; k<n; k++) {
+      pm = m + k;
+      if ( ! *pm ) {
+        p1 = q1 + k*i1;
+        p2 = q2 + k*i2;
+        p3 = q3 + k*i3;
+        p4 = q4 + k*i4;
+        { (*p4) = (*p1) * (*p2) - (*p3); }
+      }
+    }
+  } else {
+    for (k=0; k<n; k++) {
+      p1 = q1 + k*i1;
+      p2 = q2 + k*i2;
+      p3 = q3 + k*i3;
+      p4 = q4 + k*i4;
+      { (*p4) = (*p1) * (*p2) - (*p3); }
+    }
+  }
+}
+
+static void
+ca_triop_fms_uint64_t (ca_size_t n, boolean8_t *m,
+                                 char *ptr1, ca_size_t i1,
+                                 char *ptr2, ca_size_t i2,
+                                 char *ptr3, ca_size_t i3,
+                                 char *ptr4, ca_size_t i4)
+{
+  uint64_t *q1 = (uint64_t *) ptr1, *q2 = (uint64_t *) ptr2,
+            *q3 = (uint64_t *) ptr3, *q4 = (uint64_t *) ptr4;
+  uint64_t *p1 = q1, *p2 = q2, *p3 = q3, *p4 = q4;
+  ca_size_t k;
+  if ( m ) {
+    boolean8_t *pm;
+    for (k=0; k<n; k++) {
+      pm = m + k;
+      if ( ! *pm ) {
+        p1 = q1 + k*i1;
+        p2 = q2 + k*i2;
+        p3 = q3 + k*i3;
+        p4 = q4 + k*i4;
+        { (*p4) = (*p1) * (*p2) - (*p3); }
+      }
+    }
+  } else {
+    for (k=0; k<n; k++) {
+      p1 = q1 + k*i1;
+      p2 = q2 + k*i2;
+      p3 = q3 + k*i3;
+      p4 = q4 + k*i4;
+      { (*p4) = (*p1) * (*p2) - (*p3); }
+    }
+  }
+}
+
+static void
+ca_triop_fms_float32_t (ca_size_t n, boolean8_t *m,
+                                 char *ptr1, ca_size_t i1,
+                                 char *ptr2, ca_size_t i2,
+                                 char *ptr3, ca_size_t i3,
+                                 char *ptr4, ca_size_t i4)
+{
+  float *q1 = (float *) ptr1, *q2 = (float *) ptr2,
+            *q3 = (float *) ptr3, *q4 = (float *) ptr4;
+  float *p1 = q1, *p2 = q2, *p3 = q3, *p4 = q4;
+  ca_size_t k;
+  if ( m ) {
+    boolean8_t *pm;
+    for (k=0; k<n; k++) {
+      pm = m + k;
+      if ( ! *pm ) {
+        p1 = q1 + k*i1;
+        p2 = q2 + k*i2;
+        p3 = q3 + k*i3;
+        p4 = q4 + k*i4;
+        { (*p4) = fma((*p1), (*p2), -(*p3)); }
+      }
+    }
+  } else {
+    for (k=0; k<n; k++) {
+      p1 = q1 + k*i1;
+      p2 = q2 + k*i2;
+      p3 = q3 + k*i3;
+      p4 = q4 + k*i4;
+      { (*p4) = fma((*p1), (*p2), -(*p3)); }
+    }
+  }
+}
+
+static void
+ca_triop_fms_float64_t (ca_size_t n, boolean8_t *m,
+                                 char *ptr1, ca_size_t i1,
+                                 char *ptr2, ca_size_t i2,
+                                 char *ptr3, ca_size_t i3,
+                                 char *ptr4, ca_size_t i4)
+{
+  double *q1 = (double *) ptr1, *q2 = (double *) ptr2,
+            *q3 = (double *) ptr3, *q4 = (double *) ptr4;
+  double *p1 = q1, *p2 = q2, *p3 = q3, *p4 = q4;
+  ca_size_t k;
+  if ( m ) {
+    boolean8_t *pm;
+    for (k=0; k<n; k++) {
+      pm = m + k;
+      if ( ! *pm ) {
+        p1 = q1 + k*i1;
+        p2 = q2 + k*i2;
+        p3 = q3 + k*i3;
+        p4 = q4 + k*i4;
+        { (*p4) = fma((*p1), (*p2), -(*p3)); }
+      }
+    }
+  } else {
+    for (k=0; k<n; k++) {
+      p1 = q1 + k*i1;
+      p2 = q2 + k*i2;
+      p3 = q3 + k*i3;
+      p4 = q4 + k*i4;
+      { (*p4) = fma((*p1), (*p2), -(*p3)); }
+    }
+  }
+}
+
+static void
+ca_triop_fms_cmplx64_t (ca_size_t n, boolean8_t *m,
+                                 char *ptr1, ca_size_t i1,
+                                 char *ptr2, ca_size_t i2,
+                                 char *ptr3, ca_size_t i3,
+                                 char *ptr4, ca_size_t i4)
+{
+  cmplx64_t *q1 = (cmplx64_t *) ptr1, *q2 = (cmplx64_t *) ptr2,
+            *q3 = (cmplx64_t *) ptr3, *q4 = (cmplx64_t *) ptr4;
+  cmplx64_t *p1 = q1, *p2 = q2, *p3 = q3, *p4 = q4;
+  ca_size_t k;
+  if ( m ) {
+    boolean8_t *pm;
+    for (k=0; k<n; k++) {
+      pm = m + k;
+      if ( ! *pm ) {
+        p1 = q1 + k*i1;
+        p2 = q2 + k*i2;
+        p3 = q3 + k*i3;
+        p4 = q4 + k*i4;
+        { (*p4) = (*p1) * (*p2) - (*p3); }
+      }
+    }
+  } else {
+    for (k=0; k<n; k++) {
+      p1 = q1 + k*i1;
+      p2 = q2 + k*i2;
+      p3 = q3 + k*i3;
+      p4 = q4 + k*i4;
+      { (*p4) = (*p1) * (*p2) - (*p3); }
+    }
+  }
+}
+
+static void
+ca_triop_fms_cmplx128_t (ca_size_t n, boolean8_t *m,
+                                 char *ptr1, ca_size_t i1,
+                                 char *ptr2, ca_size_t i2,
+                                 char *ptr3, ca_size_t i3,
+                                 char *ptr4, ca_size_t i4)
+{
+  cmplx128_t *q1 = (cmplx128_t *) ptr1, *q2 = (cmplx128_t *) ptr2,
+            *q3 = (cmplx128_t *) ptr3, *q4 = (cmplx128_t *) ptr4;
+  cmplx128_t *p1 = q1, *p2 = q2, *p3 = q3, *p4 = q4;
+  ca_size_t k;
+  if ( m ) {
+    boolean8_t *pm;
+    for (k=0; k<n; k++) {
+      pm = m + k;
+      if ( ! *pm ) {
+        p1 = q1 + k*i1;
+        p2 = q2 + k*i2;
+        p3 = q3 + k*i3;
+        p4 = q4 + k*i4;
+        { (*p4) = (*p1) * (*p2) - (*p3); }
+      }
+    }
+  } else {
+    for (k=0; k<n; k++) {
+      p1 = q1 + k*i1;
+      p2 = q2 + k*i2;
+      p3 = q3 + k*i3;
+      p4 = q4 + k*i4;
+      { (*p4) = (*p1) * (*p2) - (*p3); }
+    }
+  }
+}
+
+static void
+ca_triop_fms_VALUE (ca_size_t n, boolean8_t *m,
+                                 char *ptr1, ca_size_t i1,
+                                 char *ptr2, ca_size_t i2,
+                                 char *ptr3, ca_size_t i3,
+                                 char *ptr4, ca_size_t i4)
+{
+  VALUE *q1 = (VALUE *) ptr1, *q2 = (VALUE *) ptr2,
+            *q3 = (VALUE *) ptr3, *q4 = (VALUE *) ptr4;
+  VALUE *p1 = q1, *p2 = q2, *p3 = q3, *p4 = q4;
+  ca_size_t k;
+  if ( m ) {
+    boolean8_t *pm;
+    for (k=0; k<n; k++) {
+      pm = m + k;
+      if ( ! *pm ) {
+        p1 = q1 + k*i1;
+        p2 = q2 + k*i2;
+        p3 = q3 + k*i3;
+        p4 = q4 + k*i4;
+        { (*p4) = rb_funcall(rb_funcall((*p1), rb_intern("*"), 1, (*p2)), rb_intern("-"), 1, (*p3)); }
+      }
+    }
+  } else {
+    for (k=0; k<n; k++) {
+      p1 = q1 + k*i1;
+      p2 = q2 + k*i2;
+      p3 = q3 + k*i3;
+      p4 = q4 + k*i4;
+      { (*p4) = rb_funcall(rb_funcall((*p1), rb_intern("*"), 1, (*p2)), rb_intern("-"), 1, (*p3)); }
+    }
+  }
+}
+
+ca_triop_func_t
+ca_triop_fms[CA_NTYPE] = {
+  ca_triop_not_implement,
+  ca_triop_not_implement,
+  ca_triop_fms_int8_t,
+  ca_triop_fms_uint8_t,
+  ca_triop_fms_int16_t,
+  ca_triop_fms_uint16_t,
+  ca_triop_fms_int32_t,
+  ca_triop_fms_uint32_t,
+  ca_triop_fms_int64_t,
+  ca_triop_fms_uint64_t,
+  ca_triop_fms_float32_t,
+  ca_triop_fms_float64_t,
+  ca_triop_not_implement,
+  ca_triop_fms_cmplx64_t,
+  ca_triop_fms_cmplx128_t,
+  ca_triop_not_implement,
+  ca_triop_fms_VALUE,
+};
+
+VALUE rb_ca_fms (VALUE self, VALUE other2, VALUE other3)
+{ return rb_ca_call_triop(self, other2, other3, ca_triop_fms); }
+
+VALUE rb_ca_fms_bang (VALUE self, VALUE other2, VALUE other3)
+{ return rb_ca_call_triop_bang(self, other2, other3, ca_triop_fms); }
+
+/* ===== triop clip ============================================ */
+
+static void
+ca_triop_clip_int8_t (ca_size_t n, boolean8_t *m,
+                                 char *ptr1, ca_size_t i1,
+                                 char *ptr2, ca_size_t i2,
+                                 char *ptr3, ca_size_t i3,
+                                 char *ptr4, ca_size_t i4)
+{
+  int8_t *q1 = (int8_t *) ptr1, *q2 = (int8_t *) ptr2,
+            *q3 = (int8_t *) ptr3, *q4 = (int8_t *) ptr4;
+  int8_t *p1 = q1, *p2 = q2, *p3 = q3, *p4 = q4;
+  ca_size_t k;
+  if ( m ) {
+    boolean8_t *pm;
+    for (k=0; k<n; k++) {
+      pm = m + k;
+      if ( ! *pm ) {
+        p1 = q1 + k*i1;
+        p2 = q2 + k*i2;
+        p3 = q3 + k*i3;
+        p4 = q4 + k*i4;
+        { (*p4) = ((*p1) < (*p2)) ? (*p2) : (((*p1) > (*p3)) ? (*p3) : (*p1)); }
+      }
+    }
+  } else {
+    for (k=0; k<n; k++) {
+      p1 = q1 + k*i1;
+      p2 = q2 + k*i2;
+      p3 = q3 + k*i3;
+      p4 = q4 + k*i4;
+      { (*p4) = ((*p1) < (*p2)) ? (*p2) : (((*p1) > (*p3)) ? (*p3) : (*p1)); }
+    }
+  }
+}
+
+static void
+ca_triop_clip_uint8_t (ca_size_t n, boolean8_t *m,
+                                 char *ptr1, ca_size_t i1,
+                                 char *ptr2, ca_size_t i2,
+                                 char *ptr3, ca_size_t i3,
+                                 char *ptr4, ca_size_t i4)
+{
+  uint8_t *q1 = (uint8_t *) ptr1, *q2 = (uint8_t *) ptr2,
+            *q3 = (uint8_t *) ptr3, *q4 = (uint8_t *) ptr4;
+  uint8_t *p1 = q1, *p2 = q2, *p3 = q3, *p4 = q4;
+  ca_size_t k;
+  if ( m ) {
+    boolean8_t *pm;
+    for (k=0; k<n; k++) {
+      pm = m + k;
+      if ( ! *pm ) {
+        p1 = q1 + k*i1;
+        p2 = q2 + k*i2;
+        p3 = q3 + k*i3;
+        p4 = q4 + k*i4;
+        { (*p4) = ((*p1) < (*p2)) ? (*p2) : (((*p1) > (*p3)) ? (*p3) : (*p1)); }
+      }
+    }
+  } else {
+    for (k=0; k<n; k++) {
+      p1 = q1 + k*i1;
+      p2 = q2 + k*i2;
+      p3 = q3 + k*i3;
+      p4 = q4 + k*i4;
+      { (*p4) = ((*p1) < (*p2)) ? (*p2) : (((*p1) > (*p3)) ? (*p3) : (*p1)); }
+    }
+  }
+}
+
+static void
+ca_triop_clip_int16_t (ca_size_t n, boolean8_t *m,
+                                 char *ptr1, ca_size_t i1,
+                                 char *ptr2, ca_size_t i2,
+                                 char *ptr3, ca_size_t i3,
+                                 char *ptr4, ca_size_t i4)
+{
+  int16_t *q1 = (int16_t *) ptr1, *q2 = (int16_t *) ptr2,
+            *q3 = (int16_t *) ptr3, *q4 = (int16_t *) ptr4;
+  int16_t *p1 = q1, *p2 = q2, *p3 = q3, *p4 = q4;
+  ca_size_t k;
+  if ( m ) {
+    boolean8_t *pm;
+    for (k=0; k<n; k++) {
+      pm = m + k;
+      if ( ! *pm ) {
+        p1 = q1 + k*i1;
+        p2 = q2 + k*i2;
+        p3 = q3 + k*i3;
+        p4 = q4 + k*i4;
+        { (*p4) = ((*p1) < (*p2)) ? (*p2) : (((*p1) > (*p3)) ? (*p3) : (*p1)); }
+      }
+    }
+  } else {
+    for (k=0; k<n; k++) {
+      p1 = q1 + k*i1;
+      p2 = q2 + k*i2;
+      p3 = q3 + k*i3;
+      p4 = q4 + k*i4;
+      { (*p4) = ((*p1) < (*p2)) ? (*p2) : (((*p1) > (*p3)) ? (*p3) : (*p1)); }
+    }
+  }
+}
+
+static void
+ca_triop_clip_uint16_t (ca_size_t n, boolean8_t *m,
+                                 char *ptr1, ca_size_t i1,
+                                 char *ptr2, ca_size_t i2,
+                                 char *ptr3, ca_size_t i3,
+                                 char *ptr4, ca_size_t i4)
+{
+  uint16_t *q1 = (uint16_t *) ptr1, *q2 = (uint16_t *) ptr2,
+            *q3 = (uint16_t *) ptr3, *q4 = (uint16_t *) ptr4;
+  uint16_t *p1 = q1, *p2 = q2, *p3 = q3, *p4 = q4;
+  ca_size_t k;
+  if ( m ) {
+    boolean8_t *pm;
+    for (k=0; k<n; k++) {
+      pm = m + k;
+      if ( ! *pm ) {
+        p1 = q1 + k*i1;
+        p2 = q2 + k*i2;
+        p3 = q3 + k*i3;
+        p4 = q4 + k*i4;
+        { (*p4) = ((*p1) < (*p2)) ? (*p2) : (((*p1) > (*p3)) ? (*p3) : (*p1)); }
+      }
+    }
+  } else {
+    for (k=0; k<n; k++) {
+      p1 = q1 + k*i1;
+      p2 = q2 + k*i2;
+      p3 = q3 + k*i3;
+      p4 = q4 + k*i4;
+      { (*p4) = ((*p1) < (*p2)) ? (*p2) : (((*p1) > (*p3)) ? (*p3) : (*p1)); }
+    }
+  }
+}
+
+static void
+ca_triop_clip_int32_t (ca_size_t n, boolean8_t *m,
+                                 char *ptr1, ca_size_t i1,
+                                 char *ptr2, ca_size_t i2,
+                                 char *ptr3, ca_size_t i3,
+                                 char *ptr4, ca_size_t i4)
+{
+  int32_t *q1 = (int32_t *) ptr1, *q2 = (int32_t *) ptr2,
+            *q3 = (int32_t *) ptr3, *q4 = (int32_t *) ptr4;
+  int32_t *p1 = q1, *p2 = q2, *p3 = q3, *p4 = q4;
+  ca_size_t k;
+  if ( m ) {
+    boolean8_t *pm;
+    for (k=0; k<n; k++) {
+      pm = m + k;
+      if ( ! *pm ) {
+        p1 = q1 + k*i1;
+        p2 = q2 + k*i2;
+        p3 = q3 + k*i3;
+        p4 = q4 + k*i4;
+        { (*p4) = ((*p1) < (*p2)) ? (*p2) : (((*p1) > (*p3)) ? (*p3) : (*p1)); }
+      }
+    }
+  } else {
+    for (k=0; k<n; k++) {
+      p1 = q1 + k*i1;
+      p2 = q2 + k*i2;
+      p3 = q3 + k*i3;
+      p4 = q4 + k*i4;
+      { (*p4) = ((*p1) < (*p2)) ? (*p2) : (((*p1) > (*p3)) ? (*p3) : (*p1)); }
+    }
+  }
+}
+
+static void
+ca_triop_clip_uint32_t (ca_size_t n, boolean8_t *m,
+                                 char *ptr1, ca_size_t i1,
+                                 char *ptr2, ca_size_t i2,
+                                 char *ptr3, ca_size_t i3,
+                                 char *ptr4, ca_size_t i4)
+{
+  uint32_t *q1 = (uint32_t *) ptr1, *q2 = (uint32_t *) ptr2,
+            *q3 = (uint32_t *) ptr3, *q4 = (uint32_t *) ptr4;
+  uint32_t *p1 = q1, *p2 = q2, *p3 = q3, *p4 = q4;
+  ca_size_t k;
+  if ( m ) {
+    boolean8_t *pm;
+    for (k=0; k<n; k++) {
+      pm = m + k;
+      if ( ! *pm ) {
+        p1 = q1 + k*i1;
+        p2 = q2 + k*i2;
+        p3 = q3 + k*i3;
+        p4 = q4 + k*i4;
+        { (*p4) = ((*p1) < (*p2)) ? (*p2) : (((*p1) > (*p3)) ? (*p3) : (*p1)); }
+      }
+    }
+  } else {
+    for (k=0; k<n; k++) {
+      p1 = q1 + k*i1;
+      p2 = q2 + k*i2;
+      p3 = q3 + k*i3;
+      p4 = q4 + k*i4;
+      { (*p4) = ((*p1) < (*p2)) ? (*p2) : (((*p1) > (*p3)) ? (*p3) : (*p1)); }
+    }
+  }
+}
+
+static void
+ca_triop_clip_int64_t (ca_size_t n, boolean8_t *m,
+                                 char *ptr1, ca_size_t i1,
+                                 char *ptr2, ca_size_t i2,
+                                 char *ptr3, ca_size_t i3,
+                                 char *ptr4, ca_size_t i4)
+{
+  int64_t *q1 = (int64_t *) ptr1, *q2 = (int64_t *) ptr2,
+            *q3 = (int64_t *) ptr3, *q4 = (int64_t *) ptr4;
+  int64_t *p1 = q1, *p2 = q2, *p3 = q3, *p4 = q4;
+  ca_size_t k;
+  if ( m ) {
+    boolean8_t *pm;
+    for (k=0; k<n; k++) {
+      pm = m + k;
+      if ( ! *pm ) {
+        p1 = q1 + k*i1;
+        p2 = q2 + k*i2;
+        p3 = q3 + k*i3;
+        p4 = q4 + k*i4;
+        { (*p4) = ((*p1) < (*p2)) ? (*p2) : (((*p1) > (*p3)) ? (*p3) : (*p1)); }
+      }
+    }
+  } else {
+    for (k=0; k<n; k++) {
+      p1 = q1 + k*i1;
+      p2 = q2 + k*i2;
+      p3 = q3 + k*i3;
+      p4 = q4 + k*i4;
+      { (*p4) = ((*p1) < (*p2)) ? (*p2) : (((*p1) > (*p3)) ? (*p3) : (*p1)); }
+    }
+  }
+}
+
+static void
+ca_triop_clip_uint64_t (ca_size_t n, boolean8_t *m,
+                                 char *ptr1, ca_size_t i1,
+                                 char *ptr2, ca_size_t i2,
+                                 char *ptr3, ca_size_t i3,
+                                 char *ptr4, ca_size_t i4)
+{
+  uint64_t *q1 = (uint64_t *) ptr1, *q2 = (uint64_t *) ptr2,
+            *q3 = (uint64_t *) ptr3, *q4 = (uint64_t *) ptr4;
+  uint64_t *p1 = q1, *p2 = q2, *p3 = q3, *p4 = q4;
+  ca_size_t k;
+  if ( m ) {
+    boolean8_t *pm;
+    for (k=0; k<n; k++) {
+      pm = m + k;
+      if ( ! *pm ) {
+        p1 = q1 + k*i1;
+        p2 = q2 + k*i2;
+        p3 = q3 + k*i3;
+        p4 = q4 + k*i4;
+        { (*p4) = ((*p1) < (*p2)) ? (*p2) : (((*p1) > (*p3)) ? (*p3) : (*p1)); }
+      }
+    }
+  } else {
+    for (k=0; k<n; k++) {
+      p1 = q1 + k*i1;
+      p2 = q2 + k*i2;
+      p3 = q3 + k*i3;
+      p4 = q4 + k*i4;
+      { (*p4) = ((*p1) < (*p2)) ? (*p2) : (((*p1) > (*p3)) ? (*p3) : (*p1)); }
+    }
+  }
+}
+
+static void
+ca_triop_clip_float32_t (ca_size_t n, boolean8_t *m,
+                                 char *ptr1, ca_size_t i1,
+                                 char *ptr2, ca_size_t i2,
+                                 char *ptr3, ca_size_t i3,
+                                 char *ptr4, ca_size_t i4)
+{
+  float *q1 = (float *) ptr1, *q2 = (float *) ptr2,
+            *q3 = (float *) ptr3, *q4 = (float *) ptr4;
+  float *p1 = q1, *p2 = q2, *p3 = q3, *p4 = q4;
+  ca_size_t k;
+  if ( m ) {
+    boolean8_t *pm;
+    for (k=0; k<n; k++) {
+      pm = m + k;
+      if ( ! *pm ) {
+        p1 = q1 + k*i1;
+        p2 = q2 + k*i2;
+        p3 = q3 + k*i3;
+        p4 = q4 + k*i4;
+        { (*p4) = isnan(*p1) ? (*p1) : fmin(fmax((*p1), (*p2)), (*p3)); }
+      }
+    }
+  } else {
+    for (k=0; k<n; k++) {
+      p1 = q1 + k*i1;
+      p2 = q2 + k*i2;
+      p3 = q3 + k*i3;
+      p4 = q4 + k*i4;
+      { (*p4) = isnan(*p1) ? (*p1) : fmin(fmax((*p1), (*p2)), (*p3)); }
+    }
+  }
+}
+
+static void
+ca_triop_clip_float64_t (ca_size_t n, boolean8_t *m,
+                                 char *ptr1, ca_size_t i1,
+                                 char *ptr2, ca_size_t i2,
+                                 char *ptr3, ca_size_t i3,
+                                 char *ptr4, ca_size_t i4)
+{
+  double *q1 = (double *) ptr1, *q2 = (double *) ptr2,
+            *q3 = (double *) ptr3, *q4 = (double *) ptr4;
+  double *p1 = q1, *p2 = q2, *p3 = q3, *p4 = q4;
+  ca_size_t k;
+  if ( m ) {
+    boolean8_t *pm;
+    for (k=0; k<n; k++) {
+      pm = m + k;
+      if ( ! *pm ) {
+        p1 = q1 + k*i1;
+        p2 = q2 + k*i2;
+        p3 = q3 + k*i3;
+        p4 = q4 + k*i4;
+        { (*p4) = isnan(*p1) ? (*p1) : fmin(fmax((*p1), (*p2)), (*p3)); }
+      }
+    }
+  } else {
+    for (k=0; k<n; k++) {
+      p1 = q1 + k*i1;
+      p2 = q2 + k*i2;
+      p3 = q3 + k*i3;
+      p4 = q4 + k*i4;
+      { (*p4) = isnan(*p1) ? (*p1) : fmin(fmax((*p1), (*p2)), (*p3)); }
+    }
+  }
+}
+
+static void
+ca_triop_clip_VALUE (ca_size_t n, boolean8_t *m,
+                                 char *ptr1, ca_size_t i1,
+                                 char *ptr2, ca_size_t i2,
+                                 char *ptr3, ca_size_t i3,
+                                 char *ptr4, ca_size_t i4)
+{
+  VALUE *q1 = (VALUE *) ptr1, *q2 = (VALUE *) ptr2,
+            *q3 = (VALUE *) ptr3, *q4 = (VALUE *) ptr4;
+  VALUE *p1 = q1, *p2 = q2, *p3 = q3, *p4 = q4;
+  ca_size_t k;
+  if ( m ) {
+    boolean8_t *pm;
+    for (k=0; k<n; k++) {
+      pm = m + k;
+      if ( ! *pm ) {
+        p1 = q1 + k*i1;
+        p2 = q2 + k*i2;
+        p3 = q3 + k*i3;
+        p4 = q4 + k*i4;
+        { { VALUE _v=(*p1); if (rb_funcall(_v, rb_intern("<"), 1, (*p2))) _v=(*p2); else if (rb_funcall(_v, rb_intern(">"), 1, (*p3))) _v=(*p3); (*p4) = _v; } }
+      }
+    }
+  } else {
+    for (k=0; k<n; k++) {
+      p1 = q1 + k*i1;
+      p2 = q2 + k*i2;
+      p3 = q3 + k*i3;
+      p4 = q4 + k*i4;
+      { { VALUE _v=(*p1); if (rb_funcall(_v, rb_intern("<"), 1, (*p2))) _v=(*p2); else if (rb_funcall(_v, rb_intern(">"), 1, (*p3))) _v=(*p3); (*p4) = _v; } }
+    }
+  }
+}
+
+ca_triop_func_t
+ca_triop_clip[CA_NTYPE] = {
+  ca_triop_not_implement,
+  ca_triop_not_implement,
+  ca_triop_clip_int8_t,
+  ca_triop_clip_uint8_t,
+  ca_triop_clip_int16_t,
+  ca_triop_clip_uint16_t,
+  ca_triop_clip_int32_t,
+  ca_triop_clip_uint32_t,
+  ca_triop_clip_int64_t,
+  ca_triop_clip_uint64_t,
+  ca_triop_clip_float32_t,
+  ca_triop_clip_float64_t,
+  ca_triop_not_implement,
+  ca_triop_not_implement,
+  ca_triop_not_implement,
+  ca_triop_not_implement,
+  ca_triop_clip_VALUE,
+};
+
+VALUE rb_ca_clip (VALUE self, VALUE other2, VALUE other3)
+{ return rb_ca_call_triop(self, other2, other3, ca_triop_clip); }
+
+VALUE rb_ca_clip_bang (VALUE self, VALUE other2, VALUE other3)
+{ return rb_ca_call_triop_bang(self, other2, other3, ca_triop_clip); }
+
+void
+Init_carray_kernels_triop (void)
+{
+  rb_define_method(rb_cCArray, "fma", rb_ca_fma, 2);
+  rb_define_method(rb_cCArray, "fma!", rb_ca_fma_bang, 2);
+  rb_define_method(rb_cCArray, "fms", rb_ca_fms, 2);
+  rb_define_method(rb_cCArray, "fms!", rb_ca_fms_bang, 2);
+  rb_define_method(rb_cCArray, "__clip_ki__", rb_ca_clip, 2);
+}