[clang] [Clang][OpenCL][AMDGPU] Allow a kernel to call another kernel (PR #115821)

Aniket Lal via cfe-commits cfe-commits at lists.llvm.org
Wed Nov 27 01:10:17 PST 2024


================
@@ -127,7 +127,10 @@ static const BlockExpr *getBlockExpr(const Expr *E) {
 void CGOpenCLRuntime::recordBlockInfo(const BlockExpr *E,
                                       llvm::Function *InvokeF,
                                       llvm::Value *Block, llvm::Type *BlockTy) {
-  assert(!EnqueuedBlockMap.contains(E) && "Block expression emitted twice");
+
+  // FIXME: Since OpenCL Kernels are emitted twice (kernel version and stub
+  // version), its constituent BlockExpr will also be emitted twice.
+  // assert(!EnqueuedBlockMap.contains(E) && "Block expression emitted twice");
----------------
lalaniket8 wrote:

Added support to distinguish between BlockExpr emitted from Kernel and its device variant by adding a relevant flag in EnqueuedBlockInfo. Updated assert to look at this flag.

https://github.com/llvm/llvm-project/pull/115821


More information about the cfe-commits mailing list