Fix kernel argument indices bug

Author: fabiomestre

source/adapters/cuda/kernel.hpp

@@ -16,6 +16,7 @@
 #include <atomic>
 #include <cassert>
 #include <numeric>
+#include <algorithm>
 
 #include "program.hpp"
 
@@ -68,6 +69,8 @@ struct ur_kernel_handle_t_ {
     args_size_t ParamSizes;
     /// Byte offset into /p Storage allocation for each parameter.
     args_index_t Indices;
+    /// Largest argument index that has been added to this kernel so far.
+    size_t InsertPos = 0;
     /// Aligned size in bytes for each local memory parameter after padding has
     /// been added. Zero if the argument at the index isn't a local memory
     /// argument.
@@ -110,13 +113,13 @@ struct ur_kernel_handle_t_ {
         OriginalLocalMemSize.resize(Index + 1);
       }
       ParamSizes[Index] = Size;
-      // calculate the insertion point on the array
-      size_t InsertPos = std::accumulate(std::begin(ParamSizes),
-                                         std::begin(ParamSizes) + Index, 0);
-      // Update the stored value for the argument
+
       std::memcpy(&Storage[InsertPos], Arg, Size);
+
       Indices[Index] = &Storage[InsertPos];
       AlignedLocalMemSize[Index] = LocalSize;
+
+      InsertPos += Size;
     }
 
     /// Returns the padded size and offset of a local memory argument.
@@ -177,10 +180,7 @@ struct ur_kernel_handle_t_ {
         AlignedLocalMemSize[SuccIndex] = SuccAlignedLocalSize;
 
         // Store new offset into local data
-        const size_t InsertPos =
-            std::accumulate(std::begin(ParamSizes),
-                            std::begin(ParamSizes) + SuccIndex, size_t{0});
-        std::memcpy(&Storage[InsertPos], &SuccAlignedLocalOffset,
+        std::memcpy(Indices[SuccIndex], &SuccAlignedLocalOffset,
                     sizeof(size_t));
       }
     }