benchmarks: add Taylor series for pi

test/benchmarks/taylor_pi.c

@@ -0,0 +1,25 @@
+#define WASM_EXPORT __attribute__((visibility("default")))
+
+// Pretty bad Pi approximation using Taylor series.
+// It is set to single-precision to test single-precision instructions.
+static inline float _taylor_pi(unsigned n)
+{
+    float sum = 1.0f;
+    int sign = -1;
+
+    for (unsigned i = 1; i < n; i++) {
+       sum += sign / (2.0f * i + 1.0f);
+       sign = -sign;
+    }
+
+    // This is the result in float precision
+    sum = 4.0f * sum;
+
+    return sum;
+}
+
+WASM_EXPORT unsigned long long taylor_pi(unsigned n)
+{
+    // Display all 16 digits of float precision as a 64-bit integer
+    return _taylor_pi(n) * 10000000000000000ULL;
+}

test/benchmarks/taylor_pi.inputs

@@ -0,0 +1,7 @@
+pi_1000000_runs
+taylor_pi
+i:I
+1000000
+
+31415954898419712
+