Data update

Task/Dot-product/C/dot-product.c

@@ -1,28 +1,46 @@
+/*****************************************************************
+ * arbitrary vector length/type dot/scalar product               *
+ * as preprocessor macro with OpenMP SIMD and tree vectorize        *
+ * CFLAGS="-march=native -O3 -std=<c|gnu>23                      *
+ * -mfpmath=<your SIMD implementation>                           *
+ * -ftree-vectorize -fopensmp-simd"                              *
+ *****************************************************************/
 #include <stdio.h>
 #include <stdlib.h>
+#include <stdint.h>
+#include <stddef.h>
+#include <stdbool.h>
+#include <stdalign.h>
+#include <sys/param.h>
 
-int dot_product(int *, int *, size_t);
 
-int
-main(void)
+#define cnt(a) sizeof((a))/sizeof(typeof((a)[0]))
+#define dot(a,b,t,n,i) \
+({ \
+t dst = (t)i; \
+_Pragma("omp simd reduction(+:dst)") \
+for(size_t j = 0; j < MIN(MIN(cnt(a),cnt(b)),n); j++) \
+        dst += (t)((a)[j] * (b)[j]); \
+dst; \
+})
+
+/* default floating point scalar */
+typedef double flt;
+/* default dot products for length 3/4
+#define dot3(a,b) dot(a,b,flt,3,0)
+#define dot4(a,b) dot(a,b,flt,4,0)
+
+int main(int argc, char** argv)
 {
-        int a[3] = {1, 3, -5};
-        int b[3] = {4, -2, -1};
+        flt a[3] = { 1,  3, -5};
+        flt b[3] = { 4, -2, -1};
 
-        printf("%d\n", dot_product(a, b, sizeof(a) / sizeof(a[0])));
+        /**************************************************
+         * cast output to double and use "%+e" allowing   *
+         * homogeneous formatted indented output of any   *
+         * numerical scalar return type of dot.           *
+         **************************************************/
+        printf("%+e\n", (double)dot3(a, b));
 
         return EXIT_SUCCESS;
 }
-
-int
-dot_product(int *a, int *b, size_t n)
-{
-        int sum = 0;
-        size_t i;
-
-        for (i = 0; i < n; i++) {
-                sum += a[i] * b[i];
-        }
-
-        return sum;
-}