[X86] Assign i8/i16 to the stack for Win64 as well.

[mstorsjo]

Summary

This makes sure that a i8/i16 parameter after the 4th parameter gets assigned to the stack, instead of continuing matching in the generic CC_X86_64_C, after SVN r328386. (After that change, such a param would end up assigned to EDI.)

[rnk]

This works for 64-bit, but I think the 32-bit side needs fixing. I think I will try to incorporate this into a re-land of rL328386 later today. Sorry for the breakage. :(

[rnk]

Looks like 32-bit doesn't need to be fixed because of the way the delegation is arranged. In any case, I combined this with the re-land in D44846. PTAL

[mstorsjo]

Taken care of as part of D44846.

lib/Target/X86/X86CallingConv.td

def CC_X86_Win64_C : CallingConv<[
 
   // The first 4 FP/Vector arguments are passed in XMM registers.
   CCIfType<[f32, f64, v16i8, v8i16, v4i32, v2i64, v4f32, v2f64],
            CCAssignToRegWithShadow<[XMM0, XMM1, XMM2, XMM3],
                                    [RCX , RDX , R8  , R9  ]>>,
 
   // Integer/FP values get stored in stack slots that are 8 bytes in size and
   // 8-byte aligned if there are no more registers to hold them.
-  CCIfType<[i32, i64, f32, f64], CCAssignToStack<8, 8>>
+  CCIfType<[i8, i16, i32, i64, f32, f64], CCAssignToStack<8, 8>>
 ]>;
 
 def CC_X86_Win64_VectorCall : CallingConv<[
   CCCustom<"CC_X86_64_VectorCall">,
 
   // Delegate to fastcall to handle integer types.
   CCDelegateTo<CC_X86_Win64_C>
 ]>;

test/CodeGen/X86/win64-smallparams.ll

+; RUN: llc < %s -mtriple=x86_64-windows-msvc | FileCheck %s --check-prefix=CHECK
+; RUN: llc < %s -mtriple=x86_64-windows-gnu | FileCheck %s --check-prefix=CHECK
+
+define dso_local void @call() {
+entry:
+  tail call void @manyargs(i8 1, i16 2, i8 3, i16 4, i8 5, i16 6)
+  ret void
+}
+
+declare dso_local void @manyargs(i8, i16, i8, i16, i8, i16)
+
+; CHECK-LABEL: call:
+; CHECK-DAG: movw $6, 40(%rsp)
+; CHECK-DAG: movb $5, 32(%rsp)
+; CHECK-DAG: movb $1, %cl
+; CHECK-DAG: movw $2, %dx
+; CHECK-DAG: movb $3, %r8b
+; CHECK-DAG: movw $4, %r9w
+; CHECK: callq manyargs