[AArch64] Dont inline streaming fn into non-streaming caller

Without this change, the following test would fail to compile
with `-march=armv8-a+sme`:

  void func1(const svuint32_t *in, svuint32_t *out) {
    [&]() __arm_streaming { *out = *in; }();
  }

But in general, it's probably better never to inline
streaming functions into non-streaming functions, because
they will have been marked as 'streaming' for a reason
by the user.

Author: sdesmalen-arm

llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp

@@ -270,6 +270,13 @@ bool AArch64TTIImpl::areInlineCompatible(const Function *Caller,
                                          const Function *Callee) const {
   SMECallAttrs CallAttrs(*Caller, *Callee);
 
+  // Never inline a function explicitly marked as being streaming,
+  // into a non-streaming function. Assume it was marked as streaming
+  // for a reason.
+  if (CallAttrs.caller().hasNonStreamingInterfaceAndBody() &&
+      CallAttrs.callee().hasStreamingInterfaceOrBody())
+    return false;
+
   // When inlining, we should consider the body of the function, not the
   // interface.
   if (CallAttrs.callee().hasStreamingBody()) {

llvm/test/Transforms/Inline/AArch64/sme-pstatesm-attrs.ll

@@ -690,7 +690,7 @@ define void @simple_streaming_function(ptr %ptr) #0 "aarch64_pstate_sm_enabled"
 define void @non_streaming_caller_streaming_callee_dont_inline(ptr %ptr) #0 {
 ; CHECK-LABEL: define void @non_streaming_caller_streaming_callee_dont_inline
 ; CHECK-SAME: (ptr [[PTR:%.*]]) #[[ATTR1]] {
-; CHECK-NEXT:    store <vscale x 4 x i32> zeroinitializer, ptr [[PTR]], align 16
+; CHECK-NEXT:    call void @simple_streaming_function(ptr [[PTR]])
 ; CHECK-NEXT:    ret void
 ;
   call void @simple_streaming_function(ptr %ptr)
@@ -711,7 +711,7 @@ define void @simple_locally_streaming_function(ptr %ptr) #0 "aarch64_pstate_sm_b
 define void @non_streaming_caller_locally_streaming_callee_dont_inline(ptr %ptr) #0 {
 ; CHECK-LABEL: define void @non_streaming_caller_locally_streaming_callee_dont_inline
 ; CHECK-SAME: (ptr [[PTR:%.*]]) #[[ATTR1]] {
-; CHECK-NEXT:    store <vscale x 4 x i32> zeroinitializer, ptr [[PTR]], align 16
+; CHECK-NEXT:    call void @simple_locally_streaming_function(ptr [[PTR]])
 ; CHECK-NEXT:    ret void
 ;
   call void @simple_locally_streaming_function(ptr %ptr)