[Triton-MLIR] Minor fixes to enable fused-softmax and layer-norm tutorials (#835)

lib/Conversion/TritonGPUToLLVM/PtxAsmFormat.cpp

@@ -125,7 +125,7 @@ std::string PTXBuilder::dump() const {
     lines.push_back(exec->dump());
   }
 
-  return strJoin(lines, "\r\n");
+  return strJoin(lines, "\n\t");
 }
 
 PTXInstrExecution &PTXInstrCommon::call(ArrayRef<Operand *> oprs) {

lib/Conversion/TritonGPUToLLVM/TritonGPUToLLVM.cpp

@@ -1051,7 +1051,7 @@ struct LoadOpConversion
       if (other) {
         for (size_t ii = 0; ii < nWords; ++ii) {
           PTXInstr &mov = *ptxBuilder.create<>("mov");
-          mov.o("u", width);
+          mov.o("u" + std::to_string(width));
 
           size_t size = width / valueElemNbits;