[ThinLTO] Handle values exported via indirect reference from global

[tejohnson]

Summary

The function import logic during the thin link currently has handling
for the fact that we will currently import constant unnamed_addr global
variables in the backend. It will conservatively mark all values
referenced in the initializer lists of exported global variables as also
exported.

However, this only handled values referenced directly from the
initializer list of an exported global variable. If the value is itself
a constant unnamed_addr variable, we could end up exporting its
references as well. Therefore, do this by invoking exportGlobalInModule
recursively, instead of just adding the first level of initializer list
references to the ExportList directly.

Fixes PR31052.

[krasin]

Peter,

can you please take a look at this CL, if you have cycles? The CL does not look terribly complicated and it's the only known reason for the Chromium ThinLTO bot being broken:
https://build.chromium.org/p/chromium.fyi/builders/ThinLTO%20Linux%20ToT/builds/551

Thank you in advance!

[krasin]

Oh, I have just learned about the existence of https://reviews.llvm.org/D26880
Probably, my request to review the CL is outdated (I have not yet looked into the other CL to make a real guess)

[tejohnson]

I've updated D26880 to note that we should go with that approach. Will abandon this fix.

lib/Transforms/IPO/FunctionImport.cpp

static void exportGlobalInModule(const ModuleSummaryIndex &Index,
   // trigger importing  the initializer for `constant unnamed addr` globals that
   // are referenced. We conservatively export all the referenced symbols for
   // every global to workaround this, so that the ExportList is accurate.
   // FIXME: with a "isConstant" flag in the summary we could be more targetted.
   for (auto &Ref : GVS->refs()) {
     auto GUID = Ref.getGUID();
     auto *RefSummary = FindGlobalSummaryInModule(GUID);
     if (RefSummary)
-      // Found a ref in the current module, mark it as exported
-      ExportList.insert(GUID);
+      exportGlobalInModule(Index, ExportModulePath, GUID, ExportList);
   }
 }
 
 using EdgeInfo = std::pair<const FunctionSummary *, unsigned /* Threshold */>;
 
 /// Compute the list of functions to import for a given caller. Mark these
 /// imported functions and the symbols they reference in their source module as
 /// exported from their source module.

test/ThinLTO/X86/Inputs/global_const.ll

+; ModuleID = 'global_const_input.o'
+source_filename = "global_const_input.cc"
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+@.str = private unnamed_addr constant [4 x i8] c"%d\0A\00", align 1
+
+; Function Attrs: norecurse uwtable
+define i32 @main() local_unnamed_addr #0 {
+entry:
+  %call = tail call i32 ()* @_Z8GetCallNii(i32 1, i32 1)
+  %call1 = tail call i32 %call()
+  %call2 = tail call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([4 x i8], [4 x i8]* @.str, i64 0, i64 0), i32 %call1)
+  %call3 = tail call i32 ()* @_Z8GetCallNii(i32 2, i32 1)
+  %call4 = tail call i32 %call3()
+  %call5 = tail call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([4 x i8], [4 x i8]* @.str, i64 0, i64 0), i32 %call4)
+  ret i32 0
+}
+
+; Function Attrs: nounwind
+declare i32 @printf(i8* nocapture readonly, ...) local_unnamed_addr #1
+
+declare i32 ()* @_Z8GetCallNii(i32, i32) local_unnamed_addr #2

test/ThinLTO/X86/global_const.ll

+; Ensure we promote global variables referenced indirectly due to exported
+; unnamed_addr constant switch.table
+
+; Do setup work for all below tests: generate bitcode and combined index
+; RUN: opt -module-summary %s -o %t1.bc
+; RUN: opt -module-summary %p/Inputs/global_const.ll -o %t2.bc
+; RUN: llvm-lto -thinlto-action=thinlink -o %t3.bc %t1.bc %t2.bc
+
+; RUN: llvm-lto -thinlto-action=promote %t1.bc -thinlto-index=%t3.bc -o - | llvm-dis -o - | FileCheck %s --check-prefix=EXPORT
+
+; RUN: llvm-lto2 %t1.bc %t2.bc -o %t.o -save-temps \
+; RUN:     -r=%t1.bc,_Z8GetCallNii,pxl \
+; RUN:     -r=%t2.bc,main,pxl \
+; RUN:     -r=%t2.bc,printf,pxl \
+; RUN:     -r=%t2.bc,_Z8GetCallNii,
+; RUN: llvm-dis < %t.o.0.1.promote.bc | FileCheck  %s --check-prefix=EXPORT
+
+; EXPORT-DAG: define hidden i32 @_ZL5CallXv.llvm.0
+; EXPORT-DAG: define hidden i32 @_ZL5CallYv.llvm.0
+
+; ModuleID = 'global_const.o'
+source_filename = "global_const.cc"
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+@_ZL6Funcs0 = internal unnamed_addr constant [4 x i32 ()*] [i32 ()* @_ZL5CallXv, i32 ()* @_ZL5CallYv, i32 ()* @_ZL5CallYv, i32 ()* @_ZL5CallYv], align 16
+@_ZL6Funcs1 = internal unnamed_addr constant [4 x i32 ()*] [i32 ()* @_ZL5CallYv, i32 ()* @_ZL5CallXv, i32 ()* @_ZL5CallYv, i32 ()* @_ZL5CallYv], align 16
+@_ZL6Funcs2 = internal unnamed_addr constant [4 x i32 ()*] [i32 ()* @_ZL5CallYv, i32 ()* @_ZL5CallYv, i32 ()* @_ZL5CallXv, i32 ()* @_ZL5CallYv], align 16
[e-mail] = private unnamed_addr constant [3 x [4 x i32 ()*]*] [[4 x i32 ()*]* @_ZL6Funcs0, [4 x i32 ()*]* @_ZL6Funcs1, [4 x i32 ()*]* @_ZL6Funcs2]
+
+; Function Attrs: norecurse nounwind readonly uwtable
+define i32 ()* @_Z8GetCallNii(i32 %i, i32 %j) local_unnamed_addr #0 {
+entry:
+  %0 = icmp ult i32 %i, 3
+  br i1 %0, label %switch.lookup, label %return
+
+switch.lookup:                                    ; preds = %entry
+  %1 = sext i32 %i to i64
+  %switch.gep = getelementptr inbounds [3 x [4 x i32 ()*]*], [3 x [4 x i32 ()*]*]* @switch.table, i64 0, i64 %1
+  %switch.load = load [4 x i32 ()*]*, [4 x i32 ()*]** %switch.gep, align 8
+  %idxprom5 = sext i32 %j to i64
+  %arrayidx6 = getelementptr inbounds [4 x i32 ()*], [4 x i32 ()*]* %switch.load, i64 0, i64 %idxprom5
+  %2 = load i32 ()*, i32 ()** %arrayidx6, align 8
+  br label %return
+
+return:                                           ; preds = %switch.lookup, %entry
+  %retval.0 = phi i32 ()* [ null, %entry ], [ %2, %switch.lookup ]
+  ret i32 ()* %retval.0
+}
+
+; Function Attrs: norecurse nounwind readnone uwtable
+define internal i32 @_ZL5CallXv() #1 {
+entry:
+  ret i32 0
+}
+
+; Function Attrs: norecurse nounwind readnone uwtable
+define internal i32 @_ZL5CallYv() #1 {
+entry:
+  ret i32 1
+}