[Backend] Make ConvertTritonGPUToLLVMPass's tmaMetadata a member (#2271)

.. instead of an option.

This partially addresses https://github.com/openai/triton/issues/2265 to
no longer crash when printing a pass pipeline in textual form.

It is not a proper solution for the fact that pass results should be
stored in the IR and not in a pointer argument.

include/triton/Conversion/TritonGPUToLLVM/Passes.td

@@ -27,9 +27,6 @@ def ConvertTritonGPUToLLVM : Pass<"convert-triton-gpu-to-llvm", "mlir::ModuleOp"
         Option<"computeCapability", "compute-capability",
                "int32_t", /*default*/"80",
                "device compute capability">,
-        Option<"tmaMetadata", "tma-metadata",
-               "mlir::triton::gpu::TMAMetadataTy*", /*default*/"nullptr",
-               "tma metadata to the runtime">,
         Option<"target", "target", "enum Target", "mlir::triton::Target::Default",
                "compile for target compatible LLVM",
                "llvm::cl::values("

include/triton/Conversion/TritonGPUToLLVM/TritonGPUToLLVMPass.h

@@ -21,7 +21,8 @@ enum Target { NVVM, ROCDL, Default = NVVM };
 
 std::unique_ptr<OperationPass<ModuleOp>> createConvertTritonGPUToLLVMPass();
 std::unique_ptr<OperationPass<ModuleOp>>
-createConvertTritonGPUToLLVMPass(const ConvertTritonGPUToLLVMOptions &options);
+createConvertTritonGPUToLLVMPass(int32_t computeCapability, Target target,
+                                 mlir::triton::gpu::TMAMetadataTy *tmaMetadata);
 
 } // namespace triton