[PATCH] D22315: Teach fast isel about thiscall (and callee-pop) calls.

[Nico Weber via llvm-commits]

thakis added inline comments.

================
Comment at: test/CodeGen/X86/fast-isel-call.ll:59
@@ +58,3 @@
+; STDERR-NOT: FastISel missed call:   call x86_thiscallcc void @thiscallfun
+; CHECK: test5:
+%struct.S = type { i8 }
----------------
rnk wrote:
> We should check for correct code
Done. (e.g. win32_sret.ll already checked for that, but having more testing doesn't hurt of course.)

================
Comment at: test/CodeGen/X86/fast-isel-call.ll:64
@@ +63,3 @@
+  %s = alloca %struct.S, align 1
+  call x86_thiscallcc void @thiscallfun(%struct.S* %s, i32 43)
+  ret void
----------------
rnk wrote:
> If you add a store to an alloca after the call, it will ensure that the call sequence end stack adjustment isn't eliminated as dead code.
Does that happen in fast isel mode? As far as I can tell, it doesn't.


http://reviews.llvm.org/D22315