Mark that SpeculativeExecution preserves Globals Alias Analysis.

[kristof.beyls]

Summary

A few benchmarks with lots of accesses to global variables in the hot
loops regressed a lot since r266399, which added the
SpeculativeExecution pass to the default pipeline. The problem is that
this pass doesn't mark Globals Alias Analysis as preserved. Globals
Alias Analysis is computed in a module pass, whereas
SpeculativeExecution is a function pass, and a lot of passes dependent
on the Globals Alias Analysis to optimize these benchmarks are also
function passes. As such, the Globals Alias Analysis information cannot
be recomputed between SpeculativeExecution and the following function
passes needing that information.

SpeculativeExecution doesn't invalidate Globals Alias Analysis, so mark
it as such to fix those performance regressions.

[jlebar]

Thank you for finding this!

[chandlerc]

LGTM (too)

lib/Transforms/Scalar/SpeculativeExecution.cpp

 // on other targets, it is a nop.
 //
 // This lets you include this pass unconditionally in the IR pass pipeline, but
 // only enable it for relevant targets.
 //
 //===----------------------------------------------------------------------===//
 
 #include "llvm/ADT/SmallSet.h"
+#include "llvm/Analysis/GlobalsModRef.h"
 #include "llvm/Analysis/TargetTransformInfo.h"
 #include "llvm/Analysis/ValueTracking.h"
 #include "llvm/IR/Instructions.h"
 #include "llvm/IR/Module.h"
 #include "llvm/IR/Operator.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 

 INITIALIZE_PASS_BEGIN(SpeculativeExecution, "speculative-execution",
                       "Speculatively execute instructions", false, false)
 INITIALIZE_PASS_DEPENDENCY(TargetTransformInfoWrapperPass)
 INITIALIZE_PASS_END(SpeculativeExecution, "speculative-execution",
                     "Speculatively execute instructions", false, false)
 
 void SpeculativeExecution::getAnalysisUsage(AnalysisUsage &AU) const {
   AU.addRequired<TargetTransformInfoWrapperPass>();
+  AU.addPreserved<GlobalsAAWrapperPass>();
 }
 
 bool SpeculativeExecution::runOnFunction(Function &F) {
   if (skipFunction(F))
     return false;
 
   TTI = &getAnalysis<TargetTransformInfoWrapperPass>().getTTI(F);
   if (OnlyIfDivergentTarget && !TTI->hasBranchDivergence()) {

test/Transforms/PhaseOrdering/globalaa-retained.ll

+; RUN: opt -O3 -S < %s | FileCheck %s
+target datalayout = "e-m:e-i64:64-i128:128-n32:64-S128"
+target triple = "aarch64"
+
+@v = internal unnamed_addr global i32 0, align 4
+@p = common global i32* null, align 8
+
+; Function Attrs: norecurse nounwind
+define void @f(i32 %n) {
+entry:
+  %0 = load i32, i32* @v, align 4
+  %inc = add nsw i32 %0, 1
+  store i32 %inc, i32* @v, align 4
+  %1 = load i32*, i32** @p, align 8
+  store i32 %n, i32* %1, align 4
+  %2 = load i32, i32* @v, align 4
+  %inc1 = add nsw i32 %2, 1
+  store i32 %inc1, i32* @v, align 4
+  ret void
+}
+
+; check variable v is loaded only once after optimization, which should be
+; prove that globalsAA survives until the optimization that can use it to
+; optimize away the duplicate load/stores on variable v.
+; CHECK:     load i32, i32* @v, align 4
+; CHECK-NOT: load i32, i32* @v, align 4