[llvm] [LLVM][NVPTX] Add movmatrix intrinsic and PTX instruction support (PR #190109)

[Alex MacLean via llvm-commits]

================
@@ -3936,6 +3936,34 @@ an event.
 For more information on the pmevent instructions, refer to the `PTX ISA
 <https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#miscellaneous-instructions-pmevent>`__.
 
+movmatrix Intrinsics
+--------------------
+
+'``llvm.nvvm.movmatrix.sync.aligned.m8n8.trans.b16``'
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+Syntax:
+"""""""
+
+.. code-block:: llvm
+
+  declare i32 @llvm.nvvm.movmatrix.sync.aligned.m8n8.trans.b16(i32 %src)
+
+Overview:
+"""""""""
+
+The '``@llvm.nvvm.movmatrix.sync.aligned.m8n8.trans.b16``' intrinsic generates
+the ``movmatrix.sync.aligned.m8n8.trans.b16`` PTX instruction, which performs
----------------
AlexMaclean wrote:

I think the intrinsic documentation should not be so focused on the PTX lowering, instead we should just define what the semantics are and perhaps reference the PTX briefly at the end. 


https://github.com/llvm/llvm-project/pull/190109