[ThinLTO] Pass down opt level to LTO backend and handle -O0 LTO in new PM

[tejohnson]

Summary

The opt level was not being passed down to the ThinLTO backend when
invoked via clang (for distributed ThinLTO).

This exposed an issue where the new PM was asserting if the Thin or
regular LTO backend pipelines were invoked with -O0 (not a new issue,
could be provoked by invoking in-process *LTO backends via linker using
new PM and -O0). Fix this similar to the old PM where -O0 only does the
necessary lowering of type metadata (WPD and LowerTypeTest passes) and
then quits, rather than asserting.

[mehdi_amini]

llvm/test/tools/gold/X86/opt-level.ll

53 ↗

(On Diff #196263)

This is intended? I'm surprised the two PMs don't have the same list of passes for a given opt level?

[tejohnson]

llvm/test/tools/gold/X86/opt-level.ll

53 ↗

(On Diff #196263)

I'm really not sure. I did compare the post-link LTO pipelines of both PMs at O0/O1/O2 and confirmed that the old PM is doing many more passes than the new PM at O1. Probably a question for @chandlerc ? In any case, I didn't want to address it here since it is orthogonal.

[tejohnson]

llvm/test/tools/gold/X86/opt-level.ll

53 ↗

(On Diff #196263)

Some more info:

This is the regular LTO post-link pipeline, ThinLTO post-link pipeline uses essentially the same as a normal opt pipeline so would be consistent at -O1.

The optimization missing from the new PM regular LTO post link pipeline that affects this test is SimplifyCFG. This and a few other optimizations are added in the old PM at O1 via PassManagerBuilder::addLateLTOOptimizationPasses. Note that PassManagerBuilder::addLTOOptimizationPasses does exit early at -O1 (similar to where we exit early in the new PM for the regular LTO post link compilation). But the "late" LTO optimization passes are added unconditionally above -O0. Is this correct/desired? There isn't an equivalent in the new PM.

[chandlerc]

llvm/test/tools/gold/X86/opt-level.ll

53 ↗

(On Diff #196263)

I don't think it was intentional, but I'm not sure I would directly replicate it if it requires adding an entirely new customization point. Is their some simpler way of getting equivalent results at O1?

[tejohnson]

llvm/test/tools/gold/X86/opt-level.ll

53 ↗

(On Diff #196263)

Yeah we can presumably just add those optimizations to the -O1 early exit path in PassBuilder::buildLTODefaultPipeline. I can send a patch (but would like to get this one in as it is a bugfix and orthogonal).

[xur]

LGTM. We need to Initialize the OptLevel no matter what.

[mehdi_amini]

llvm/test/tools/gold/X86/opt-level.ll

53 ↗

(On Diff #196263)

(I fully agree it is orthogonal, I just spotted this as a separate bug indeed, thanks for fixing!)

cfe/trunk/lib/CodeGen/BackendUtil.cpp

if (Error E = Conf.addSaveTemps(CGOpts.SaveTempsFilePrefix + ".",
       });
     }
   }
   Conf.CPU = TOpts.CPU;
   Conf.CodeModel = getCodeModel(CGOpts);
   Conf.MAttrs = TOpts.Features;
   Conf.RelocModel = CGOpts.RelocationModel;
   Conf.CGOptLevel = getCGOptLevel(CGOpts);
+  Conf.OptLevel = CGOpts.OptimizationLevel;
   initTargetOptions(Conf.Options, CGOpts, TOpts, LOpts, HeaderOpts);
   Conf.SampleProfile = std::move(SampleProfile);
 
   // Context sensitive profile.
   if (CGOpts.hasProfileCSIRInstr()) {
     Conf.RunCSIRInstr = true;
     Conf.CSIRProfile = std::move(CGOpts.InstrProfileOutput);
   } else if (CGOpts.hasProfileCSIRUse()) {

cfe/trunk/test/CodeGen/thinlto-debug-pm.c

-// Test to ensure -fdebug-pass-manager works when invoking the
-// ThinLTO backend path with the new PM.
+// Test to ensure the opt level is passed down to the ThinLTO backend.
 // REQUIRES: x86-registered-target
 // RUN: %clang_cc1 -o %t.o -flto=thin -fexperimental-new-pass-manager -triple x86_64-unknown-linux-gnu -emit-llvm-bc %s
 // RUN: llvm-lto -thinlto -o %t %t.o
-// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -emit-obj -O2 -o %t2.o -x ir %t.o -fthinlto-index=%t.thinlto.bc -fdebug-pass-manager -fexperimental-new-pass-manager 2>&1 | FileCheck %s
-// CHECK: Running pass:
+
+// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -emit-obj -O2 -o %t2.o -x ir %t.o -fthinlto-index=%t.thinlto.bc -fdebug-pass-manager -fexperimental-new-pass-manager 2>&1 | FileCheck %s --check-prefix=O2-NEWPM
+// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -emit-obj -O0 -o %t2.o -x ir %t.o -fthinlto-index=%t.thinlto.bc -fdebug-pass-manager -fexperimental-new-pass-manager 2>&1 | FileCheck %s --check-prefix=O0-NEWPM
+// O2-NEWPM: Running pass: LoopVectorizePass
+// O0-NEWPM-NOT: Running pass: LoopVectorizePass
+
+// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -emit-obj -O2 -o %t2.o -x ir %t.o -fthinlto-index=%t.thinlto.bc -mllvm -debug-pass=Structure 2>&1 | FileCheck %s --check-prefix=O2-OLDPM
+// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -emit-obj -O0 -o %t2.o -x ir %t.o -fthinlto-index=%t.thinlto.bc -mllvm -debug-pass=Structure 2>&1 | FileCheck %s --check-prefix=O0-OLDPM
+// O2-OLDPM: Loop Vectorization
+// O0-OLDPM-NOT: Loop Vectorization
 
 void foo() {
 }

cfe/trunk/test/CodeGen/thinlto-distributed-cfi-devirt.ll

 ; RUN: llvm-dis %t.o.thinlto.bc -o - | FileCheck %s --check-prefix=CHECK-DIS
 ; Round trip it through llvm-as
 ; RUN: llvm-dis %t.o.thinlto.bc -o - | llvm-as -o - | llvm-dis -o - | FileCheck %s --check-prefix=CHECK-DIS
 ; CHECK-DIS: ^0 = module: (path: "{{.*}}thinlto-distributed-cfi-devirt.ll.tmp.o", hash: ({{.*}}, {{.*}}, {{.*}}, {{.*}}, {{.*}}))
 ; CHECK-DIS: ^1 = gv: (guid: 8346051122425466633, summaries: (function: (module: ^0, flags: (linkage: external, notEligibleToImport: 0, live: 1, dsoLocal: 0), insts: 18, typeIdInfo: (typeTests: (^2), typeCheckedLoadVCalls: (vFuncId: (^2, offset: 8), vFuncId: (^2, offset: 0))))))
 ; CHECK-DIS: ^2 = typeid: (name: "_ZTS1A", summary: (typeTestRes: (kind: allOnes, sizeM1BitWidth: 7), wpdResolutions: ((offset: 0, wpdRes: (kind: branchFunnel)), (offset: 8, wpdRes: (kind: singleImpl, singleImplName: "_ZN1A1nEi"))))) ; guid = 7004155349499253778
 
 ; RUN: %clang_cc1 -triple x86_64-grtev4-linux-gnu \
-; RUN:   -emit-obj -fthinlto-index=%t.o.thinlto.bc \
+; RUN:   -emit-obj -fthinlto-index=%t.o.thinlto.bc -O2 \
 ; RUN:   -emit-llvm -o - -x ir %t.o | FileCheck %s --check-prefixes=CHECK-IR
 
 ; Check that backend does not fail generating native code.
 ; RUN: %clang_cc1 -triple x86_64-grtev4-linux-gnu \
-; RUN:   -emit-obj -fthinlto-index=%t.o.thinlto.bc \
+; RUN:   -emit-obj -fthinlto-index=%t.o.thinlto.bc -O2 \
 ; RUN:   -o %t.native.o -x ir %t.o
 
 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
 target triple = "x86_64-grtev4-linux-gnu"
 
 %struct.A = type { i32 (...)** }
 %struct.B = type { %struct.A }
 %struct.C = type { %struct.A }

llvm/trunk/lib/Passes/PassBuilder.cpp

if (ImportSummary) {
     //
     // For example, GVN may transform the pattern assume(type.test) appearing in
     // two basic blocks into assume(phi(type.test, type.test)), which would
     // transform a dependency on a WPD resolution into a dependency on a type
     // identifier resolution for CFI.
     //
     // Also, WPD has access to more precise information than ICP and can
     // devirtualize more effectively, so it should operate on the IR first.
+    //
+    // The WPD and LowerTypeTest passes need to run at -O0 to lower type
+    // metadata and intrinsics.
     MPM.addPass(WholeProgramDevirtPass(nullptr, ImportSummary));
     MPM.addPass(LowerTypeTestsPass(nullptr, ImportSummary));
   }
 
+  if (Level == O0)
+    return MPM;
+
   // Force any function attributes we want the rest of the pipeline to observe.
   MPM.addPass(ForceFunctionAttrsPass());
 
   // Add the core simplification pipeline.
   MPM.addPass(buildModuleSimplificationPipeline(Level, ThinLTOPhase::PostLink,
                                                 DebugLogging));
 
   // Now add the optimization pipeline.
PassBuilder::buildLTOPreLinkDefaultPipeline(OptimizationLevel Level,
   // FIXME: We should use a customized pre-link pipeline!
   return buildPerModuleDefaultPipeline(Level, DebugLogging,
                                        /* LTOPreLink */true);
 }
 
 ModulePassManager
 PassBuilder::buildLTODefaultPipeline(OptimizationLevel Level, bool DebugLogging,
                                      ModuleSummaryIndex *ExportSummary) {
-  assert(Level != O0 && "Must request optimizations for the default pipeline!");
   ModulePassManager MPM(DebugLogging);
 
+  if (Level == O0) {
+    // The WPD and LowerTypeTest passes need to run at -O0 to lower type
+    // metadata and intrinsics.
+    MPM.addPass(WholeProgramDevirtPass(ExportSummary, nullptr));
+    MPM.addPass(LowerTypeTestsPass(ExportSummary, nullptr));
+    return MPM;
+  }
+
   if (PGOOpt && PGOOpt->Action == PGOOptions::SampleUse) {
     // Load sample profile before running the LTO optimization pipeline.
     MPM.addPass(SampleProfileLoaderPass(PGOOpt->ProfileFile,
                                         PGOOpt->ProfileRemappingFile,
                                         false /* ThinLTOPhase::PreLink */));
     // Cache ProfileSummaryAnalysis once to avoid the potential need to insert
     // RequireAnalysisPass for PSI before subsequent non-module passes.
     MPM.addPass(RequireAnalysisPass<ProfileSummaryAnalysis, Module>());

llvm/trunk/test/tools/gold/X86/opt-level.ll

 ; RUN: llvm-as -o %t.bc %s
 ; RUN: %gold -plugin %llvmshlibdir/LLVMgold%shlibext -plugin-opt=save-temps \
 ; RUN:    -m elf_x86_64 \
 ; RUN:    -plugin-opt=O0 -r -o %t.o %t.bc
 ; RUN: llvm-dis < %t.o.0.4.opt.bc -o - | FileCheck --check-prefix=CHECK-O0 %s
 ; RUN: %gold -plugin %llvmshlibdir/LLVMgold%shlibext -plugin-opt=save-temps \
 ; RUN:    -m elf_x86_64 \
 ; RUN:    -plugin-opt=O1 -r -o %t.o %t.bc
-; RUN: llvm-dis < %t.o.0.4.opt.bc -o - | FileCheck --check-prefix=CHECK-O1 %s
+; RUN: llvm-dis < %t.o.0.4.opt.bc -o - | FileCheck --check-prefix=CHECK-O1 --check-prefix=CHECK-O1-OLDPM %s
 ; RUN: %gold -plugin %llvmshlibdir/LLVMgold%shlibext -plugin-opt=save-temps \
 ; RUN:    -m elf_x86_64 \
 ; RUN:    -plugin-opt=O2 -r -o %t.o %t.bc
 ; RUN: llvm-dis < %t.o.0.4.opt.bc -o - | FileCheck --check-prefix=CHECK-O2 %s
 
+; RUN: %gold -plugin %llvmshlibdir/LLVMgold%shlibext -plugin-opt=save-temps \
+; RUN:    -m elf_x86_64 --plugin-opt=new-pass-manager \
+; RUN:    -plugin-opt=O0 -r -o %t.o %t.bc
+; RUN: llvm-dis < %t.o.0.4.opt.bc -o - | FileCheck --check-prefix=CHECK-O0 %s
+; RUN: %gold -plugin %llvmshlibdir/LLVMgold%shlibext -plugin-opt=save-temps \
+; RUN:    -m elf_x86_64 --plugin-opt=new-pass-manager \
+; RUN:    -plugin-opt=O1 -r -o %t.o %t.bc
+; RUN: llvm-dis < %t.o.0.4.opt.bc -o - | FileCheck --check-prefix=CHECK-O1 --check-prefix=CHECK-O1-NEWPM %s
+; RUN: %gold -plugin %llvmshlibdir/LLVMgold%shlibext -plugin-opt=save-temps \
+; RUN:    -m elf_x86_64 --plugin-opt=new-pass-manager \
+; RUN:    -plugin-opt=O2 -r -o %t.o %t.bc
+; RUN: llvm-dis < %t.o.0.4.opt.bc -o - | FileCheck --check-prefix=CHECK-O2 %s
+
 ; CHECK-O0: define internal void @foo(
 ; CHECK-O1: define internal void @foo(
 ; CHECK-O2-NOT: define internal void @foo(
 
 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
 target triple = "x86_64-unknown-linux-gnu"
 
 define internal void @foo() {
   ret void
 }
 
 ; CHECK-O0: define internal i32 @bar(
 ; CHECK-O1: define internal i32 @bar(
 define internal i32 @bar(i1 %p) {
   br i1 %p, label %t, label %f
 
 t:
   br label %end
 
 f:
   br label %end
 
 end:
   ; CHECK-O0: phi
-  ; CHECK-O1: select
+  ; CHECK-O1-OLDPM: select
+  ; The new PM does not do as many optimizations at O1
+  ; CHECK-O1-NEWPM: phi
   %r = phi i32 [ 1, %t ], [ 2, %f ]
   ret i32 %r
 }
 
 define i1 @baz() {
   call void @foo()
   %c = call i32 @bar(i1 true)
   %p = call i1 @llvm.type.test(i8* undef, metadata !"typeid1")