[IR] Set nocapture attribute for va_start, va_end, and va_copy

[eush]

Summary

This adds nocapture attribute to the definitions of intrinsics va_start,
va_end, and va_copy, which indicates that they do not make any copies of
pointers to va_list that outlive the call.

This allows TailCallElim to mark tail calls in variadic functions as such, and
targets to successfully eliminate these calls if it's otherwise possible.

Currently unmarked tail calls in variadic functions are not recognized because
TailCallElim conservatively extends the lifetime of va_list local to the end
of the function.

[RKSimon]

Sorry @eush, but I don't know enough about this to review.

[efriedma]

As far as I know, for all in-tree targets, this should be fine... and for all ABIs I know of in common use. I'm a little nervous that it will eventually break some obscure platform where a va_list can somehow point to itself, though. (At least, an implementation like that seems plausible.)

On a related note, I'm a little concerned that this missed optimization is actually covering up a bug: markTails() in TailRecursionElimination currently doesn't model the fact that on many targets, va_lists point to memory that's implicitly allocated on the stack, so it adds incorrect "tail" markings to certain calls. (This is technically orthogonal, but it's much more likely to come up with this patch because people usually declare va_list as a local variable.)

[ahatanak]

Is this patch still being reviewed?

docs/LangRef.rst

 '``llvm.va_start``' Intrinsic
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
 Syntax:
 """""""
 
 ::
 
-      declare void @llvm.va_start(i8* <arglist>)
+      declare void @llvm.va_start(i8* nocapture <arglist>)
 
 Overview:
 """""""""
 
 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
 subsequent use by ``va_arg``.
 
 Arguments:

 '``llvm.va_end``' Intrinsic
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
 Syntax:
 """""""
 
 ::
 
-      declare void @llvm.va_end(i8* <arglist>)
+      declare void @llvm.va_end(i8* nocapture <arglist>)
 
 Overview:
 """""""""
 
 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
 
 Arguments:

 '``llvm.va_copy``' Intrinsic
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
 Syntax:
 """""""
 
 ::
 
-      declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
+      declare void @llvm.va_copy(i8* nocapture <destarglist>, i8* nocapture <srcarglist>)
 
 Overview:
 """""""""
 
 The '``llvm.va_copy``' intrinsic copies the current argument position
 from the source argument list to the destination argument list.
 
 Arguments:

include/llvm/IR/Intrinsics.td

 class MSBuiltin<string name> {
   string MSBuiltinName = name;
 }
 
 
 //===--------------- Variable Argument Handling Intrinsics ----------------===//
 //
 
-def int_vastart : Intrinsic<[], [llvm_ptr_ty], [], "llvm.va_start">;
-def int_vacopy  : Intrinsic<[], [llvm_ptr_ty, llvm_ptr_ty], [],
-                            "llvm.va_copy">;
-def int_vaend   : Intrinsic<[], [llvm_ptr_ty], [], "llvm.va_end">;
+def int_vastart : Intrinsic<[], [llvm_ptr_ty], [NoCapture<0>], "llvm.va_start">;
+def int_vacopy : Intrinsic<[], [llvm_ptr_ty, llvm_ptr_ty],
+                           [NoCapture<0>, NoCapture<1>], "llvm.va_copy">;
+def int_vaend   : Intrinsic<[], [llvm_ptr_ty], [NoCapture<0>], "llvm.va_end">;
 
 //===------------------- Garbage Collection Intrinsics --------------------===//
 //
 def int_gcroot  : Intrinsic<[],
                             [llvm_ptrptr_ty, llvm_ptr_ty]>;
 def int_gcread  : Intrinsic<[llvm_ptr_ty],
                             [llvm_ptr_ty, llvm_ptrptr_ty],
                             [IntrReadMem, IntrArgMemOnly]>;

test/Transforms/TailCallElim/basic.ll

entry:
   %agg.tmp = alloca %struct.foo
   %0 = bitcast %struct.foo* %agg.tmp to i8*
   %1 = bitcast %struct.foo* %f to i8*
   call void @llvm.memcpy.p0i8.p0i8.i64(i8* %0, i8* %1, i64 40, i1 false)
   call void @bar(%struct.foo* byval %agg.tmp)
   ret void
 }
 
+; Argument pointer is not captured by va_start or va_end, and the call can be
+; marked tail.
+define void @test16(...) {
+; CHECK-LABEL: @test16
+; CHECK: tail call void @noarg()
+entry:
+  %ap = alloca i8
+  call void @llvm.va_start(i8* %ap)
+  call void @llvm.va_end(i8* %ap)
+  call void @noarg()
+  ret void
+}
+
 declare void @bar(%struct.foo* byval)
 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture writeonly, i8* nocapture readonly, i64, i1)
+declare void @llvm.va_start(i8* nocapture)
+declare void @llvm.va_end(i8* nocapture)