Add integrate() function to T-SIMD library

Moved out of experimental.H to ext.H and added tests
ti-uni-bielefeld · Oct 24, 2024 · 7f9bd18 · 7f9bd18
1 parent 8953c1e
commit 7f9bd18
Show file tree

Hide file tree

Showing 5 changed files with 110 additions and 49 deletions.
diff --git a/src/lib/tsimd/experimental.H b/src/lib/tsimd/experimental.H
@@ -36,57 +36,11 @@
 #include <cstddef>
 
 namespace simd {
+namespace experimental {
 
-// ========================================================================
-// integration of values in simd::Vec
-// ========================================================================
-
-// example:
-//        0    1    2    3    4    5    6    7
-// +      -    0    1    2    3    4    5    6    slle(1)
-//      --------------------------------------
-//        0  0-1  1-2  2-3  3-4  4-5  5-6  6-7
-// +      -    -    0  0-1  1-2  2-3  3-4  4-5    slle(2)
-//      --------------------------------------
-//        0  0-1  0-2  0-3  1-4  2-5  3-6  4-7
-// +      -    -    -    -    0  0-1  0-2  0-3    slle(4)
-//      --------------------------------------
-//        0  0-1  0-2  0-3  0-4  0-5  0-6  0-7
-//
-// problem: slle has immediate argument, T-SIMD: template parameter
-
-// primary template
-template <typename T, size_t SIMD_WIDTH, int SHIFT, int END_SHIFT>
-struct HInt
-{
-public:
-  static SIMD_INLINE simd::Vec<T, SIMD_WIDTH> integrate(
-    const simd::Vec<T, SIMD_WIDTH> &v)
-  {
-    return HInt<T, SIMD_WIDTH, 2 * SHIFT, END_SHIFT>::integrate(
-      simd::add(v, simd::slle<SHIFT>(v)));
-  }
-};
-
-// termination template
-template <typename T, size_t SIMD_WIDTH, int END_SHIFT>
-struct HInt<T, SIMD_WIDTH, END_SHIFT, END_SHIFT>
-{
-public:
-  static SIMD_INLINE simd::Vec<T, SIMD_WIDTH> integrate(
-    const simd::Vec<T, SIMD_WIDTH> &v)
-  {
-    return v;
-  }
-};
-
-template <typename T, size_t SIMD_WIDTH>
-static SIMD_INLINE simd::Vec<T, SIMD_WIDTH> integrate(
-  const simd::Vec<T, SIMD_WIDTH> &v)
-{
-  return HInt<T, SIMD_WIDTH, 1, simd::Vec<T, SIMD_WIDTH>::elems>::integrate(v);
-}
+// put experimental code here
 
+} // namespace experimental
 } // namespace simd
 
 #endif
diff --git a/src/lib/tsimd/ext.H b/src/lib/tsimd/ext.H
@@ -3268,6 +3268,72 @@ static SIMD_INLINE void transpose(
   // internal::ext::transpose2inplcLane(rows);
 }
 
+// ========================================================================
+// integration of values in simd::Vec
+// ========================================================================
+
+// example:
+//        0    1    2    3    4    5    6    7
+// +      -    0    1    2    3    4    5    6    slle(1)
+//      --------------------------------------
+//        0  0-1  1-2  2-3  3-4  4-5  5-6  6-7
+// +      -    -    0  0-1  1-2  2-3  3-4  4-5    slle(2)
+//      --------------------------------------
+//        0  0-1  0-2  0-3  1-4  2-5  3-6  4-7
+// +      -    -    -    -    0  0-1  0-2  0-3    slle(4)
+//      --------------------------------------
+//        0  0-1  0-2  0-3  0-4  0-5  0-6  0-7
+//
+// problem: slle has immediate argument, T-SIMD: template parameter
+
+namespace internal {
+namespace ext {
+// primary template
+template <typename T, size_t SIMD_WIDTH, int SHIFT, int END_SHIFT>
+struct HInt
+{
+public:
+  static SIMD_INLINE simd::Vec<T, SIMD_WIDTH> integrate(
+    const simd::Vec<T, SIMD_WIDTH> &v)
+  {
+    return HInt<T, SIMD_WIDTH, 2 * SHIFT, END_SHIFT>::integrate(
+      simd::add(v, simd::slle<SHIFT>(v)));
+  }
+};
+
+// termination template
+template <typename T, size_t SIMD_WIDTH, int END_SHIFT>
+struct HInt<T, SIMD_WIDTH, END_SHIFT, END_SHIFT>
+{
+public:
+  static SIMD_INLINE simd::Vec<T, SIMD_WIDTH> integrate(
+    const simd::Vec<T, SIMD_WIDTH> &v)
+  {
+    return v;
+  }
+};
+} // namespace ext
+} // namespace internal
+
+/**
+ * @ingroup group_horizontal
+ * @brief Integrates the values of a Vec.
+ *
+ * This function integrates the values of a Vec, i.e. it computes the
+ * prefix sum of the elements.
+ *
+ * @param[in] v Vec to integrate
+ *
+ * @return Vec with integrated values
+ */
+template <typename T, size_t SIMD_WIDTH>
+static SIMD_INLINE simd::Vec<T, SIMD_WIDTH> integrate(
+  const simd::Vec<T, SIMD_WIDTH> &v)
+{
+  return internal::ext::HInt<T, SIMD_WIDTH, 1,
+                             simd::Vec<T, SIMD_WIDTH>::elems>::integrate(v);
+}
+
 // ===========================================================================
 // setones: set all bits to 1
 // ===========================================================================

diff --git a/src/test/autotest/serial.H b/src/test/autotest/serial.H
@@ -2599,6 +2599,21 @@ static SIMD_INLINE uint64_t msb2int(const SerialVec<Double, SIMD_WIDTH> &a)
   return msb2int(reinterpret<uint64_t>(a));
 }
 
+// ---------------------------------------------------------------------------
+// integrate
+// ---------------------------------------------------------------------------
+template <typename T, size_t SIMD_WIDTH>
+static SIMD_INLINE SerialVec<T, SIMD_WIDTH> integrate(const SerialVec<T, SIMD_WIDTH> &a)
+{
+  SerialVec<T, SIMD_WIDTH> c;
+  T sum = 0;
+  for (size_t i = 0; i < c.elements; i++) {
+    sum += a[i];
+    c[i] = sum;
+  }
+  return c;
+}
+
 } // namespace auto_test
 } // namespace simd
 

diff --git a/src/test/autotest/test1.C b/src/test/autotest/test1.C
@@ -202,6 +202,12 @@ int main(int argc, char *argv[])
   TestAll<Unary, SW, Div2rd>::test(repeats1, pattern);
   TestFloat<Binary, SW, Sign>::test(repeats1, pattern);
   TestAll<Binary, SW, AbsDiff>::test(repeats1, pattern);
+  TestInt<Unary, SW, Integrate>::test(repeats1, pattern);
+
+  // TODO: integrate produces errors for floating point types because of
+  // different summation orders. How to fix this?
+  // TestFloat<Unary, SW, Integrate>::test(repeats1, pattern);
+
   // TODO: operators
   // TODO: Vecs
   printf("simdvecautotest1 complete\n");

diff --git a/src/test/autotest/wrappers.H b/src/test/autotest/wrappers.H
@@ -2572,6 +2572,26 @@ struct Iota
   }
 };
 
+template <typename T, size_t SIMD_WIDTH>
+struct Integrate
+{
+  static SIMD_INLINE std::string name()
+  {
+    return t2s<T, SIMD_WIDTH>("integrate");
+  }
+
+  static SIMD_INLINE void randomizeInput(SerialVec<T, SIMD_WIDTH> &vec)
+  {
+    vec.randomize();
+  }
+
+  template <template <typename, size_t> class VEC>
+  static SIMD_INLINE VEC<T, SIMD_WIDTH> apply(const VEC<T, SIMD_WIDTH> &a)
+  {
+    return integrate(a);
+  }
+};
+
 } // namespace auto_test
 } // namespace simd