[Inliner] Propagate SROA analysis through invariant group intrinsics

[aeubanks]

Summary

SROA can handle invariant group intrinsics, let the inliner know that
for better heuristics when the intrinsics are present.

This fixes size issues in a couple files when turning on
-fstrict-vtable-pointers in Chrome.

[mtrofin]

The paranoid in me would argue for having a flag enabling this, should there be any regressions, but if we feel this is benign, ok.

[rnk]

lgtm

[aeubanks]

In D100249#2683418, @mtrofin wrote:

The paranoid in me would argue for having a flag enabling this, should there be any regressions, but if we feel this is benign, ok.

These intrinsics are only emitted under Clang's -fstrict-vtable-pointers flag (which I do hope to turn on by default at some point).

[aeubanks]

2>&1

llvm/lib/Analysis/InlineCost.cpp

case Intrinsic::memmove:
       return false;
     case Intrinsic::icall_branch_funnel:
     case Intrinsic::localescape:
       HasUninlineableIntrinsic = true;
       return false;
     case Intrinsic::vastart:
       InitsVargArgs = true;
       return false;
+    case Intrinsic::launder_invariant_group:
+    case Intrinsic::strip_invariant_group:
+      if (auto *SROAArg = getSROAArgForValueOrNull(II->getOperand(0)))
+        SROAArgValues[II] = SROAArg;
+      return true;
     }
   }
 
   if (F == Call.getFunction()) {
     // This flag will fully abort the analysis, so don't bother with anything
     // else.
     IsRecursiveCall = true;
     return false;

llvm/test/Transforms/Inline/invariant-group-sroa.ll

+; RUN: opt -passes='print<inline-cost>' -disable-output %s | FileCheck %s
+
+; SROA analysis should yield non-zero savings for allocas passed through invariant group intrinsics
+; CHECK: SROACostSavings: 10
+
+declare i8* @llvm.launder.invariant.group.p0i8(i8*)
+declare i8* @llvm.strip.invariant.group.p0i8(i8*)
+
+declare void @b()
+
+define i32 @f() {
+  %a = alloca i32
+  %r = call i32 @g(i32* %a)
+  ret i32 %r
+}
+
+define i32 @g(i32* %a) {
+  %a_i8 = bitcast i32* %a to i8*
+  %a_inv_i8 = call i8* @llvm.launder.invariant.group.p0i8(i8* %a_i8)
+  %a_inv = bitcast i8* %a_inv_i8 to i32*
+  %i1 = load i32, i32* %a_inv
+  %i2 = load i32, i32* %a_inv
+  %i3 = add i32 %i1, %i2
+  %t = call i8* @llvm.strip.invariant.group.p0i8(i8* %a_inv_i8)
+  ret i32 %i3
+}