[WIP] Adds validation of code asumption.

cms-sw · Oct 26, 2023 · a0cf55d · a0cf55d
1 parent d2c77bf
commit a0cf55d
Showing 1 changed file with 13 additions and 0 deletions.
diff --git a/HeterogeneousCore/AlpakaInterface/interface/prefixScan.h b/HeterogeneousCore/AlpakaInterface/interface/prefixScan.h
@@ -10,6 +10,18 @@
 namespace cms {
   namespace alpakatools {
 
+    template <typename T, typename = std::enable_if_t<std::is_integral_v<T>>>
+    constexpr bool isPowerOf2(T v) {
+      // returns true iif v has only one bit set.
+      while (v) {
+        if (v & 1)
+          return !(v >> 1);
+        else
+          v >>= 1;
+      }
+      return false;
+    }
+
     template <typename TAcc, typename T, typename = std::enable_if_t<alpaka::isAccelerator<TAcc>>>
     ALPAKA_FN_ACC ALPAKA_FN_INLINE void warpPrefixScan(const TAcc& acc, int32_t laneId, T const* ci, T* co, uint32_t i) {
       // ci and co may be the same
@@ -42,6 +54,7 @@ namespace cms {
         ALPAKA_ASSERT_OFFLOAD(size <= warpSize * warpSize);
         ALPAKA_ASSERT_OFFLOAD(0 == blockDimension % warpSize);
         auto first = blockThreadIdx;
+        ALPAKA_ASSERT_OFFLOAD(isPowerOf2(warpSize));
         auto laneId = blockThreadIdx & (warpSize - 1);
 
         for (auto i = first; i < size; i += blockDimension) {