[CodeGen][TargetPassConfig] Add TargetTransformInfo pass correctly

[evgeny777]

Summary

When -start-after or -start-before is given tti pass is added with default TargetTransformInfo and LSR (probably other passes also) fails to make some optimizations. This also makes it impossible to properly analyze codegen pipeline, because in the example below foo.s and bar.s will almost always be different:

llc -O3 foo.ll -o foo.s

llc -O3 -stop-after=<pass> foo.ll -o foo.mir
llc -O3 -start-after=<pass> foo.mir -o bar.s

[ychen]

I also hit this recently. LGTM.

[MaskRay]

llvm/lib/CodeGen/TargetPassConfig.cpp

882

Thanks! In the case of -start-after, this line was a no-op..

How did you notice the problem?

[evgeny777]

How did you notice the problem?

I have to study compiler regressions every now and then and use -start-after / -run-pass to inspect / modify intermediate pipeline states.

llvm/lib/CodeGen/TargetPassConfig.cpp

bool TargetPassConfig::addCoreISelPasses() {
   return false;
 }
 
 bool TargetPassConfig::addISelPasses() {
   if (TM->useEmulatedTLS())
     addPass(createLowerEmuTLSPass());
 
   addPass(createPreISelIntrinsicLoweringPass());
-  addPass(createTargetTransformInfoWrapperPass(TM->getTargetIRAnalysis()));
+  PM->add(createTargetTransformInfoWrapperPass(TM->getTargetIRAnalysis()));
   addIRPasses();
   addCodeGenPrepare();
   addPassesToHandleExceptions();
   addISelPrepare();
 
   return addCoreISelPasses();
 }
 

llvm/test/CodeGen/AArch64/partial-pipeline-execution.ll

+; RUN: llc -O3 %s -o %t.s
+; RUN: llc -O3 -stop-after=atomic-expand %s -o %t.mir
+; RUN: llc -O3 -start-after=atomic-expand %s -o %t2.s
+
+; If we add tti pass correctly files should be identical
+; Otherwise LSR will use default TargetTransformInfo and
+; optimize the loop differently
+; RUN: cmp %t.s %t2.s
+
+; ModuleID = 'loop.c'
+source_filename = "loop.c"
+target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
+target triple = "aarch64-none-linux-gnu"
+
+@q = dso_local local_unnamed_addr global i32* null, align 8
+
+; Function Attrs: nofree norecurse nounwind
+define dso_local i32 @main(i32 %argc, i8** nocapture readnone %argv) local_unnamed_addr #0 {
+entry:
+  %cmp5 = icmp sgt i32 %argc, 0
+  br i1 %cmp5, label %for.body.lr.ph, label %for.cond.cleanup
+
+for.body.lr.ph:                                   ; preds = %entry
+  %0 = load i32*, i32** @q, align 8, !tbaa !2
+  %1 = zext i32 %argc to i64
+  %2 = add nsw i64 %1, -1
+  %3 = lshr i64 %2, 5
+  %4 = add nuw nsw i64 %3, 1
+  %min.iters.check = icmp eq i64 %3, 0
+  br i1 %min.iters.check, label %for.body.preheader, label %vector.ph
+
+for.body.preheader:                               ; preds = %middle.block, %for.body.lr.ph
+  %indvars.iv.ph = phi i64 [ 0, %for.body.lr.ph ], [ %ind.end, %middle.block ]
+  br label %for.body
+
+vector.ph:                                        ; preds = %for.body.lr.ph
+  %n.vec = and i64 %4, 1152921504606846974
+  %ind.end = shl i64 %n.vec, 5
+  br label %vector.body
+
+vector.body:                                      ; preds = %vector.body, %vector.ph
+  %index = phi i64 [ 0, %vector.ph ], [ %index.next, %vector.body ]
+  %offset.idx = shl i64 %index, 5
+  %induction7 = or i64 %offset.idx, 32
+  %5 = getelementptr inbounds i32, i32* %0, i64 %offset.idx
+  %6 = getelementptr inbounds i32, i32* %0, i64 %induction7
+  %7 = trunc i64 %offset.idx to i32
+  %8 = trunc i64 %induction7 to i32
+  store i32 %7, i32* %5, align 4, !tbaa !6
+  store i32 %8, i32* %6, align 4, !tbaa !6
+  %index.next = add i64 %index, 2
+  %9 = icmp eq i64 %index.next, %n.vec
+  br i1 %9, label %middle.block, label %vector.body, !llvm.loop !8
+
+middle.block:                                     ; preds = %vector.body
+  %cmp.n = icmp eq i64 %4, %n.vec
+  br i1 %cmp.n, label %for.cond.cleanup, label %for.body.preheader
+
+for.cond.cleanup:                                 ; preds = %for.body, %middle.block, %entry
+  ret i32 0
+
+for.body:                                         ; preds = %for.body.preheader, %for.body
+  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ %indvars.iv.ph, %for.body.preheader ]
+  %arrayidx = getelementptr inbounds i32, i32* %0, i64 %indvars.iv
+  %10 = trunc i64 %indvars.iv to i32
+  store i32 %10, i32* %arrayidx, align 4, !tbaa !6
+  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 32
+  %cmp = icmp ult i64 %indvars.iv.next, %1
+  br i1 %cmp, label %for.body, label %for.cond.cleanup, !llvm.loop !10
+}
+
+attributes #0 = { nofree norecurse nounwind "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "frame-pointer"="non-leaf" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="generic" "target-features"="+neon" "unsafe-fp-math"="false" "use-soft-float"="false" }
+
+!llvm.module.flags = !{!0}
+!llvm.ident = !{!1}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{!"clang version 11.0.0 (https://github.com/llvm/llvm-project.git d9943e7f0ce888733ee7ba91da432e5f01f7aa85)"}
+!2 = !{!3, !3, i64 0}
+!3 = !{!"any pointer", !4, i64 0}
+!4 = !{!"omnipotent char", !5, i64 0}
+!5 = !{!"Simple C/C++ TBAA"}
+!6 = !{!7, !7, i64 0}
+!7 = !{!"int", !4, i64 0}
+!8 = distinct !{!8, !9}
+!9 = !{!"llvm.loop.isvectorized", i32 1}
+!10 = distinct !{!10, !9}