Add docs and example

examples/perfbench/perfbench.cpp

@@ -69,6 +69,7 @@ static PerfTest tests[] = {
     { xyzSumAOS, "serial", ispc::xyzSumAOSStdlib, "ispc", "AOS vector element sum (stdlib swizzle)" },
     { xyzSumAOS, "serial", ispc::xyzSumAOSNoCoalesce, "ispc", "AOS vector element sum (no coalescing)" },
     { xyzSumSOA, "serial", ispc::xyzSumSOA, "ispc", "SOA vector element sum" },
+    { xyzSumSOA, "serial", (FuncType *) ispc::xyzSumVarying, "ispc", "Varying vector element sum" },
     { ispc::gathers, "gather", ispc::loads, "vector load", "Memory reads" },
     { ispc::scatters, "scatter", ispc::stores, "vector store", "Memory writes" },
 };

examples/perfbench/perfbench.ispc

@@ -104,6 +104,50 @@ export void xyzSumSOA(uniform float array[], uniform int count,
     result[2] = reduce_add(zsum);
 }
 
+export void xyzSumVarying(varying float array[], uniform int count,
+                          uniform float zeros[], uniform float result[]) {
+    float xsum = 0, ysum = 0, zsum = 0;
+    varying float * uniform ap = array;
+    assert(programCount <= 8);
+
+    for (uniform int i = 0; i < count/3; i += 8) {
+      if (programCount == 4) {
+        float x0 = ap[0];
+        float y0 = ap[2];
+        float z0 = ap[4];
+
+        xsum += x0;
+        ysum += y0;
+        zsum += z0;
+
+        float x1 = ap[1];
+        float y1 = ap[3];
+        float z1 = ap[5];
+
+        xsum += x1;
+        ysum += y1;
+        zsum += z1;
+
+        ap += 6;
+      }
+      else {
+        // programCount == 8
+        float x = ap[0];
+        float y = ap[1];
+        float z = ap[2];
+
+        xsum += x;
+        ysum += y;
+        zsum += z;
+
+        ap += 3;
+      }
+    }
+    result[0] = reduce_add(xsum);
+    result[1] = reduce_add(ysum);
+    result[2] = reduce_add(zsum);
+}
+
 export void gathers(uniform float array[], uniform int count,
                     uniform float zeros[], uniform float result[]) {
     float sum = 0;