// types are identical size, use a bitcast to convert (e.g. two differing
// vector types).
MVT RegVT = *PhysReg.second->vt_begin();
- if (RegVT.getSizeInBits() == OpInfo.ConstraintVT.getSizeInBits()) {
+ if (RegVT.getSizeInBits() == OpInfo.CallOperand.getValueSizeInBits()) {
OpInfo.CallOperand = DAG.getNode(ISD::BITCAST, DL,
RegVT, OpInfo.CallOperand);
OpInfo.ConstraintVT = RegVT;
--- /dev/null
+; RUN: not llc %s -verify-machineinstrs -mtriple=armv7-none-linux-gnu -mattr=+neon 2>&1 | FileCheck %s
+
+%struct.float4 = type { float, float, float, float }
+
+; CHECK: error: Don't know how to handle indirect register inputs yet for constraint 'w'
+define float @inline_func(float %f1, float %f2) #0 {
+ %c1 = alloca %struct.float4, align 4
+ %c2 = alloca %struct.float4, align 4
+ %c3 = alloca %struct.float4, align 4
+ call void asm sideeffect "vmul.f32 ${2:q}, ${0:q}, ${1:q}", "=*r,=*r,*w"(%struct.float4* %c1, %struct.float4* %c2, %struct.float4* %c3) #1, !srcloc !1
+ %x = getelementptr inbounds %struct.float4* %c3, i32 0, i32 0
+ %1 = load float* %x, align 4
+ ret float %1
+}
+
+!1 = metadata !{i32 271, i32 305}