[llvm] [NVPTX] Improve copy avoidance during lowering. (PR #106423)

[Fei Peng via llvm-commits]

================
@@ -124,11 +124,11 @@
 //
 //    define void @foo(ptr byval(%struct.s) %input) {
 //      %input1 = addrspacecast ptr %input to ptr addrspace(101)
-//      ; the following intrinsic converts pointer to generic. We don't use an addrspacecast
-//      ; to prevent generic -> param -> generic from getting cancelled out
-//      %input1.gen = call ptr @llvm.nvvm.ptr.param.to.gen.p0.p101(ptr addrspace(101) %input1)
-//      %call = call i32 @escape(ptr %input1.gen)
-//      ret void
+//      ; the following intrinsic converts pointer to generic. We don't use an
+//      addrspacecast ; to prevent generic -> param -> generic from getting
+//      cancelled out %input1.gen = call ptr
+//      @llvm.nvvm.ptr.param.to.gen.p0.p101(ptr addrspace(101) %input1) %call =
+//      call i32 @escape(ptr %input1.gen) ret void
----------------
fiigii wrote:

comments are messed up by formating  

https://github.com/llvm/llvm-project/pull/106423