Index: lib/CodeGen/PPCGCodeGeneration.cpp
===================================================================
--- lib/CodeGen/PPCGCodeGeneration.cpp
+++ lib/CodeGen/PPCGCodeGeneration.cpp
@@ -580,7 +580,7 @@
   /// Insert function calls to retrieve the SPIR group/local ids.
   ///
   /// @param The kernel to generate the function calls for.
-  void insertKernelCallsSPIR(ppcg_kernel *Kernel);
+  void insertKernelCallsSPIR(ppcg_kernel *Kernel, bool SizeTypeIs64bit);
 
   /// Setup the creation of functions referenced by the GPU kernel.
   ///
@@ -2120,7 +2120,7 @@
   }
 }
 
-void GPUNodeBuilder::insertKernelCallsSPIR(ppcg_kernel *Kernel) {
+void GPUNodeBuilder::insertKernelCallsSPIR(ppcg_kernel *Kernel, bool SizeTypeIs64bit) {
   const char *GroupName[3] = {"_Z12get_group_idj",
                               "_Z12get_group_idj",
                               "_Z12get_group_idj"};
@@ -2128,8 +2128,9 @@
   const char *LocalName[3] = {"_Z12get_local_idj",
                               "_Z12get_local_idj",
                               "_Z12get_local_idj"};
+  IntegerType *SizeType = SizeTypeIs64bit ? Builder.getInt64Ty() : Builder.getInt32Ty();
 
-  auto createFunc = [this](const char *Name, __isl_take isl_id *Id, int Dim) mutable {
+  auto createFunc = [this](const char *Name, __isl_take isl_id *Id, int Dim, IntegerType *SizeType) mutable {
     Module *M = Builder.GetInsertBlock()->getParent()->getParent();
     Function *FN = M->getFunction(Name);
 
@@ -2138,22 +2139,21 @@
       GlobalValue::LinkageTypes Linkage = Function::ExternalLinkage;
       std::vector<Type *> Args;
       Args.push_back(Builder.getInt32Ty());
-      FunctionType *Ty = FunctionType::get(Builder.getInt32Ty(), Args, false);
+      FunctionType *Ty = FunctionType::get(SizeType, Args, false);
       FN = Function::Create(Ty, Linkage, Name, M);
       FN->setCallingConv(CallingConv::SPIR_FUNC);
     }
 
     Value *Val = Builder.CreateCall(FN, {Builder.getInt32(Dim)});
-    Val = Builder.CreateIntCast(Val, Builder.getInt64Ty(), false, Name);
     IDToValue[Id] = Val;
     KernelIDs.insert(std::unique_ptr<isl_id, IslIdDeleter>(Id));
   };
 
   for (int i = 0; i < Kernel->n_grid; ++i)
-    createFunc(GroupName[i], isl_id_list_get_id(Kernel->block_ids, i), i);
+    createFunc(GroupName[i], isl_id_list_get_id(Kernel->block_ids, i), i, SizeType);
 
   for (int i = 0; i < Kernel->n_block; ++i)
-    createFunc(LocalName[i], isl_id_list_get_id(Kernel->thread_ids, i), i);
+    createFunc(LocalName[i], isl_id_list_get_id(Kernel->thread_ids, i), i, SizeType);
 }
 
 void GPUNodeBuilder::prepareKernelArguments(ppcg_kernel *Kernel, Function *FN) {
@@ -2333,8 +2333,10 @@
     insertKernelIntrinsics(Kernel);
     break;
   case GPUArch::SPIR32:
+    insertKernelCallsSPIR(Kernel, false);
+    break;
   case GPUArch::SPIR64:
-    insertKernelCallsSPIR(Kernel);
+    insertKernelCallsSPIR(Kernel, true);
     break;
   }
 }