[Mlir-commits] [mlir] [mlir][gpu] Introduce `gpu.dynamic_shared_memory` Op (PR #71546)

[Guray Ozen]

================
@@ -554,6 +555,104 @@ static IntegerAttr wrapNumericMemorySpace(MLIRContext *ctx, unsigned space) {
   return IntegerAttr::get(IntegerType::get(ctx, 64), space);
 }
 
+/// Generates a symbol with 0-sized array type for dynamic shared memory usage,
+/// or uses existing symbol.
+LLVM::GlobalOp
+getDynamicSharedMemorySymbol(ConversionPatternRewriter &rewriter,
+                             Operation *moduleOp, gpu::DynamicSharedMemoryOp op,
+                             const LLVMTypeConverter *typeConverter,
+                             MemRefType memrefType, unsigned alignmentBit) {
+  uint64_t alignmentByte = alignmentBit / memrefType.getElementTypeBitWidth();
+
+  FailureOr<unsigned> addressSpace =
+      typeConverter->getMemRefAddressSpace(memrefType);
+  if (failed(addressSpace)) {
+    op->emitError() << "conversion of memref memory space "
+                    << memrefType.getMemorySpace()
+                    << " to integer address space "
+                       "failed. Consider adding memory space conversions.";
+  }
+
+  // Step 1. Collect symbol names of LLVM::GlobalOp Ops. Also if any of
+  // LLVM::GlobalOp is suitable for shared memory, return it.
+  llvm::StringSet<> existingGlobalNames;
+  for (auto globalOp :
+       moduleOp->getRegion(0).front().template getOps<LLVM::GlobalOp>()) {
----------------
grypp wrote:

the moduleOp is actually `Operation*`, so I could not see `getBody`
```
Operation *moduleOp = funcOp->getParentWithTrait<OpTrait::SymbolTable>();
```

https://github.com/llvm/llvm-project/pull/71546