diff --git a/projects/clr/rocclr/device/pal/palkernel.cpp b/projects/clr/rocclr/device/pal/palkernel.cpp
index 84d4040fb1..e71c334718 100644
--- a/projects/clr/rocclr/device/pal/palkernel.cpp
+++ b/projects/clr/rocclr/device/pal/palkernel.cpp
@@ -552,6 +552,12 @@ bool LightningKernel::postLoad() {
   if (workGroupInfo_.size_ == 0) {
     return false;
   }
+  if ((workGroupInfo_.usedStackSize_ & 0x1) == 0x1) {
+    workGroupInfo_.scratchRegs_ =
+        std::max<uint32_t>(device().StackSize(), workGroupInfo_.scratchRegs_ * sizeof(uint32_t)) ;
+    workGroupInfo_.scratchRegs_ = amd::alignUp(workGroupInfo_.scratchRegs_, 16) / sizeof(uint32_t);
+    workGroupInfo_.privateMemSize_ = workGroupInfo_.scratchRegs_ * sizeof(uint32_t);
+  }
 
   // handle the printf metadata if any
   std::vector<std::string> printfStr;
diff --git a/projects/clr/rocclr/device/pal/palvirtual.cpp b/projects/clr/rocclr/device/pal/palvirtual.cpp
index e57de246b2..c7ae5f79e9 100644
--- a/projects/clr/rocclr/device/pal/palvirtual.cpp
+++ b/projects/clr/rocclr/device/pal/palvirtual.cpp
@@ -2678,13 +2678,6 @@ bool VirtualGPU::submitKernelInternal(const amd::NDRangeContainer& sizes, const
       dispatchParam.scratchSize = scratch->size_;
       dispatchParam.scratchOffset = scratch->offset_;
       dispatchParam.workitemPrivateSegmentSize = hsaKernel.spillSegSize();
-      if ((hsaKernel.workGroupInfo()->usedStackSize_ & 0x1) == 0x1) {
-        dispatchParam.workitemPrivateSegmentSize =
-            std::max<uint64_t>(dev().StackSize(), dispatchParam.workitemPrivateSegmentSize);
-        if (dispatchParam.workitemPrivateSegmentSize > 16 * Ki) {
-          dispatchParam.workitemPrivateSegmentSize = 16 * Ki;
-        }
-      }
     }
     dispatchParam.pCpuAqlCode = hsaKernel.cpuAqlCode();
     dispatchParam.hsaQueueVa = hsaQueueMem_->vmAddress();