[Mlir-commits] [mlir] [MLIR][NVGPU] Add truncf and extf Ops (PR #199700)

[Guray Ozen]

================
@@ -699,6 +699,138 @@ LogicalResult RcpOp::verify() {
   return success();
 }
 
+//===----------------------------------------------------------------------===//
+// NVGPU_ConvertFPTruncOp
+//===----------------------------------------------------------------------===//
+
+static bool isShapedContainerType(Type t) {
+  return llvm::isa<VectorType, RankedTensorType, UnrankedTensorType>(t);
----------------
grypp wrote:

tensor dialect has different semantic than vector or memref. it's SSA value and not in memory. It needs to be materialized aka bufferized. 

Current pipelines are following. If you support tensor on gpu, this requires some pipeline change since currently nvgpu is lowered down in the stack.   
- tensor
- bufferize
- ... bunch oh others...
- nvgpu-to-nvvm
- convert-llvm

If we don't have direct user for tensor dialect, let's exclude it from the PR.  


https://github.com/llvm/llvm-project/pull/199700