diff --git a/lib/gc/Transforms/GPU/GpuToGpuOcl.cpp b/lib/gc/Transforms/GPU/GpuToGpuOcl.cpp
index dfcd1dab..e8fa2440 100644
--- a/lib/gc/Transforms/GPU/GpuToGpuOcl.cpp
+++ b/lib/gc/Transforms/GPU/GpuToGpuOcl.cpp
@@ -415,11 +415,30 @@ struct ConvertLaunch final : ConvertOpPattern<gpu::LaunchFuncOp> {
     blockSize.emplace_back(gpuLaunch.getBlockSizeY());
     blockSize.emplace_back(gpuLaunch.getBlockSizeZ());
 
-    for (auto arg : adaptor.getKernelOperands()) {
+    for (auto arg : gpuLaunch.getKernelOperands()) {
       auto type = arg.getType();
-      // Assuming, that the value is either an integer or a float or a pointer.
-      // In the latter case, the size is 0 bytes.
-      auto size = type.isIntOrFloat() ? type.getIntOrFloatBitWidth() / 8 : 0;
+      size_t size;
+      if (isa<MemRefType>(type)) {
+        size = 0; // A special case for pointers
+      } else if (type.isIndex()) {
+        size = helper.idxType.getIntOrFloatBitWidth() / 8;
+      } else if (type.isIntOrFloat()) {
+        size = type.getIntOrFloatBitWidth() / 8;
+      } else if (auto vectorType = dyn_cast<VectorType>(type)) {
+        type = vectorType.getElementType();
+        if (type.isIntOrFloat()) {
+          size = type.getIntOrFloatBitWidth();
+        } else if (type.isIndex()) {
+          size = helper.idxType.getIntOrFloatBitWidth();
+        } else {
+          llvm::errs() << "Unsupported vector element type: " << type << "\n";
+          return false;
+        }
+        size *= vectorType.getNumElements() / 8;
+      } else {
+        llvm::errs() << "Unsupported type: " << type << "\n";
+        return false;
+      }
       argSize.emplace_back(helper.idxConstant(rewriter, loc, size));
     }