[WPD]: Enable speculative devirtualizatoin. (llvm#159048)

This patch implements the speculative devirtualization feature in the
LLVM backend.
It handles the case of single implementation devirtualization where
there is a single possible callee of a virtual function.
- Add cl::opt 'devirtualize-speculatively' to enable it.
- Flag is disabled by default.
- It works regardless of the visibility of the object.
- Not enabled for LTO for now.

Author: hassnaaHamdi

llvm/lib/Transforms/IPO/WholeProgramDevirt.cpp

@@ -24,7 +24,8 @@
 //   returns 0, or a single vtable's function returns 1, replace each virtual
 //   call with a comparison of the vptr against that vtable's address.
 //
-// This pass is intended to be used during the regular and thin LTO pipelines:
+// This pass is intended to be used during the regular/thin and non-LTO
+// pipelines:
 //
 // During regular LTO, the pass determines the best optimization for each
 // virtual call and applies the resolutions directly to virtual calls that are
@@ -48,6 +49,14 @@
 //   is supported.
 // - Import phase: (same as with hybrid case above).
 //
+// During Speculative devirtualization mode -not restricted to LTO-:
+// - The pass applies speculative devirtualization without requiring any type of
+//   visibility.
+// - Skips other features like virtual constant propagation, uniform return
+//   value optimization, unique return value optimization and branch funnels as
+//   they need LTO.
+// - This mode is enabled via 'devirtualize-speculatively' flag.
+//
 //===----------------------------------------------------------------------===//
 
 #include "llvm/Transforms/IPO/WholeProgramDevirt.h"
@@ -61,7 +70,9 @@
 #include "llvm/Analysis/AssumptionCache.h"
 #include "llvm/Analysis/BasicAliasAnalysis.h"
 #include "llvm/Analysis/BlockFrequencyInfo.h"
+#include "llvm/Analysis/ModuleSummaryAnalysis.h"
 #include "llvm/Analysis/OptimizationRemarkEmitter.h"
+#include "llvm/Analysis/ProfileSummaryInfo.h"
 #include "llvm/Analysis/TypeMetadataUtils.h"
 #include "llvm/Bitcode/BitcodeReader.h"
 #include "llvm/Bitcode/BitcodeWriter.h"
@@ -145,6 +156,13 @@ static cl::opt<std::string> ClWriteSummary(
              "bitcode, otherwise YAML"),
     cl::Hidden);
 
+// TODO: This option eventually should support any public visibility vtables
+// with/out LTO.
+static cl::opt<bool> ClDevirtualizeSpeculatively(
+    "devirtualize-speculatively",
+    cl::desc("Enable speculative devirtualization optimization"),
+    cl::init(false));
+
 static cl::opt<unsigned>
     ClThreshold("wholeprogramdevirt-branch-funnel-threshold", cl::Hidden,
                 cl::init(10),
@@ -892,6 +910,8 @@ void llvm::updatePublicTypeTestCalls(Module &M,
       CI->eraseFromParent();
     }
   } else {
+    // TODO: Don't replace public type tests when speculative devirtualization
+    // gets enabled in LTO mode.
     auto *True = ConstantInt::getTrue(M.getContext());
     for (Use &U : make_early_inc_range(PublicTypeTestFunc->uses())) {
       auto *CI = cast<CallInst>(U.getUser());
@@ -1083,10 +1103,10 @@ bool DevirtModule::tryFindVirtualCallTargets(
     if (!TM.Bits->GV->isConstant())
       return false;
 
-    // We cannot perform whole program devirtualization analysis on a vtable
-    // with public LTO visibility.
-    if (TM.Bits->GV->getVCallVisibility() ==
-        GlobalObject::VCallVisibilityPublic)
+    // Without ClDevirtualizeSpeculatively, we cannot perform whole program
+    // devirtualization analysis on a vtable with public LTO visibility.
+    if (!ClDevirtualizeSpeculatively && TM.Bits->GV->getVCallVisibility() ==
+                                            GlobalObject::VCallVisibilityPublic)
       return false;
 
     Function *Fn = nullptr;
@@ -1105,6 +1125,12 @@ bool DevirtModule::tryFindVirtualCallTargets(
     if (Fn->getName() == "__cxa_pure_virtual")
       continue;
 
+    // In most cases empty functions will be overridden by the
+    // implementation of the derived class, so we can skip them.
+    if (ClDevirtualizeSpeculatively && Fn->getReturnType()->isVoidTy() &&
+        Fn->getInstructionCount() <= 1)
+      continue;
+
     // We can disregard unreachable functions as possible call targets, as
     // unreachable functions shouldn't be called.
     if (mustBeUnreachableFunction(Fn, ExportSummary))
@@ -1223,10 +1249,12 @@ void DevirtModule::applySingleImplDevirt(VTableSlotInfo &SlotInfo,
         CallTrap->setDebugLoc(CB.getDebugLoc());
       }
 
-      // If fallback checking is enabled, add support to compare the virtual
-      // function pointer to the devirtualized target. In case of a mismatch,
-      // fall back to indirect call.
-      if (DevirtCheckMode == WPDCheckMode::Fallback) {
+      // If fallback checking or speculative devirtualization are enabled,
+      // add support to compare the virtual function pointer to the
+      // devirtualized target. In case of a mismatch, fall back to indirect
+      // call.
+      if (DevirtCheckMode == WPDCheckMode::Fallback ||
+          ClDevirtualizeSpeculatively) {
         MDNode *Weights = MDBuilder(M.getContext()).createLikelyBranchWeights();
         // Version the indirect call site. If the called value is equal to the
         // given callee, 'NewInst' will be executed, otherwise the original call
@@ -2057,15 +2085,15 @@ void DevirtModule::scanTypeTestUsers(
     Function *TypeTestFunc,
     DenseMap<Metadata *, std::set<TypeMemberInfo>> &TypeIdMap) {
   // Find all virtual calls via a virtual table pointer %p under an assumption
-  // of the form llvm.assume(llvm.type.test(%p, %md)). This indicates that %p
-  // points to a member of the type identifier %md. Group calls by (type ID,
-  // offset) pair (effectively the identity of the virtual function) and store
-  // to CallSlots.
+  // of the form llvm.assume(llvm.type.test(%p, %md)) or
+  // llvm.assume(llvm.public.type.test(%p, %md)).
+  // This indicates that %p points to a member of the type identifier %md.
+  // Group calls by (type ID, offset) pair (effectively the identity of the
+  // virtual function) and store to CallSlots.
   for (Use &U : llvm::make_early_inc_range(TypeTestFunc->uses())) {
     auto *CI = dyn_cast<CallInst>(U.getUser());
     if (!CI)
       continue;
-
     // Search for virtual calls based on %p and add them to DevirtCalls.
     SmallVector<DevirtCallSite, 1> DevirtCalls;
     SmallVector<CallInst *, 1> Assumes;
@@ -2348,6 +2376,12 @@ bool DevirtModule::run() {
       (ImportSummary && ImportSummary->partiallySplitLTOUnits()))
     return false;
 
+  Function *PublicTypeTestFunc = nullptr;
+  // If we are in speculative devirtualization mode, we can work on the public
+  // type test intrinsics.
+  if (ClDevirtualizeSpeculatively)
+    PublicTypeTestFunc =
+        Intrinsic::getDeclarationIfExists(&M, Intrinsic::public_type_test);
   Function *TypeTestFunc =
       Intrinsic::getDeclarationIfExists(&M, Intrinsic::type_test);
   Function *TypeCheckedLoadFunc =
@@ -2361,8 +2395,9 @@ bool DevirtModule::run() {
   // module, this pass has nothing to do. But if we are exporting, we also need
   // to handle any users that appear only in the function summaries.
   if (!ExportSummary &&
-      (!TypeTestFunc || TypeTestFunc->use_empty() || !AssumeFunc ||
-       AssumeFunc->use_empty()) &&
+      (((!PublicTypeTestFunc || PublicTypeTestFunc->use_empty()) &&
+        (!TypeTestFunc || TypeTestFunc->use_empty())) ||
+       !AssumeFunc || AssumeFunc->use_empty()) &&
       (!TypeCheckedLoadFunc || TypeCheckedLoadFunc->use_empty()) &&
       (!TypeCheckedLoadRelativeFunc ||
        TypeCheckedLoadRelativeFunc->use_empty()))
@@ -2373,6 +2408,9 @@ bool DevirtModule::run() {
   DenseMap<Metadata *, std::set<TypeMemberInfo>> TypeIdMap;
   buildTypeIdentifierMap(Bits, TypeIdMap);
 
+  if (PublicTypeTestFunc && AssumeFunc)
+    scanTypeTestUsers(PublicTypeTestFunc, TypeIdMap);
+
   if (TypeTestFunc && AssumeFunc)
     scanTypeTestUsers(TypeTestFunc, TypeIdMap);
 
@@ -2472,8 +2510,12 @@ bool DevirtModule::run() {
                  .WPDRes[S.first.ByteOffset];
     if (tryFindVirtualCallTargets(TargetsForSlot, TypeMemberInfos,
                                   S.first.ByteOffset, ExportSummary)) {
-
-      if (!trySingleImplDevirt(ExportSummary, TargetsForSlot, S.second, Res)) {
+      bool SingleImplDevirt =
+          trySingleImplDevirt(ExportSummary, TargetsForSlot, S.second, Res);
+      // Out of speculative devirtualization mode, Try to apply virtual constant
+      // propagation or branch funneling.
+      // TODO: This should eventually be enabled for non-public type tests.
+      if (!SingleImplDevirt && !ClDevirtualizeSpeculatively) {
         DidVirtualConstProp |=
             tryVirtualConstProp(TargetsForSlot, S.second, Res, S.first);
 

llvm/test/Transforms/WholeProgramDevirt/speculative-devirt-single-impl.ll

@@ -0,0 +1,132 @@
+; -stats requires asserts
+; REQUIRES: asserts
+
+; Check that we can still devirtualize outside LTO mode when speculative devirtualization is enabled.
+; Check that we skip devirtualization for empty functions in speculative devirtualization mode
+
+; RUN: opt -S -passes=wholeprogramdevirt -devirtualize-speculatively \
+; RUN: -pass-remarks=wholeprogramdevirt -stats %s 2>&1 | FileCheck %s
+
+target datalayout = "e-p:64:64"
+target triple = "x86_64-unknown-linux-gnu"
+
+; CHECK: remark: devirt-single.cc:30:32: single-impl: devirtualized a call to vf
+; CHECK: remark: devirt-single.cc:41:32: single-impl: devirtualized a call to vf
+; CHECK: remark: devirt-single.cc:51:32: single-impl: devirtualized a call to vf
+; CHECK: remark: devirt-single.cc:13:0: devirtualized vf
+; CHECK-NOT: devirtualized
+
+@vt1 = constant [1 x ptr] [ptr @vf], !type !8
+@vt2 = constant [1 x ptr] [ptr @vf_empty], !type !12
+
+define i1 @vf(ptr %this) #0 !dbg !7 {
+  ret i1 true
+}
+
+; This should NOT be devirtualized because during non-lto empty functions
+; are skipped.
+define void @vf_empty(ptr %this) !dbg !11 {
+  ret void
+}
+
+; CHECK: define void @call
+define void @call(ptr %obj) #1 !dbg !5 {
+  %vtable = load ptr, ptr %obj
+  %p = call i1 @llvm.public.type.test(ptr %vtable, metadata !"typeid")
+  call void @llvm.assume(i1 %p)
+  %fptr = load ptr, ptr %vtable
+  ; CHECK: if.true.direct_targ:
+  ; CHECK:   call i1 @vf(
+  ; CHECK: if.false.orig_indirect:
+  ; CHECK:   call i1 %fptr(
+  call i1 %fptr(ptr %obj), !dbg !6
+  ret void
+}
+
+
+; CHECK: define void @call1
+define void @call1(ptr %obj) #1 !dbg !9 {
+  %vtable = load ptr, ptr %obj
+  %p = call i1 @llvm.type.test(ptr %vtable, metadata !"typeid1")
+  call void @llvm.assume(i1 %p)
+  %fptr = load ptr, ptr %vtable, align 8
+  ; CHECK: call i1 %fptr
+  %1 = call i1 %fptr(ptr %obj), !dbg !10
+  ret void
+}
+declare ptr @llvm.load.relative.i32(ptr, i32)
+
+@vt3 = private unnamed_addr constant [1 x i32] [
+  i32 trunc (i64 sub (i64 ptrtoint (ptr dso_local_equivalent @vf to i64), i64 ptrtoint (ptr @vt3 to i64)) to i32)
+], align 4, !type !15
+
+; CHECK: define void @call2
+define void @call2(ptr %obj) #1 !dbg !13 {
+  %vtable = load ptr, ptr %obj
+  %p = call i1 @llvm.type.test(ptr %vtable, metadata !"typeid2")
+  call void @llvm.assume(i1 %p)
+  %fptr = call ptr @llvm.load.relative.i32(ptr %vtable, i32 0)
+  ; CHECK: if.true.direct_targ:
+  ; CHECK:   call i1 @vf(
+  ; CHECK: if.false.orig_indirect:
+  ; CHECK:   call i1 %fptr(
+  call i1 %fptr(ptr %obj), !dbg !14
+  ret void
+}
+
+@_ZTV1A.local = private unnamed_addr constant { [3 x i32] } { [3 x i32] [
+  i32 0,  ; offset to top
+  i32 0,  ; rtti
+  i32 trunc (i64 sub (i64 ptrtoint (ptr dso_local_equivalent @vf to i64), i64 ptrtoint (ptr getelementptr inbounds ({ [3 x i32] }, ptr @_ZTV1A.local, i32 0, i32 0, i32 2) to i64)) to i32)  ; vf_emptyunc offset
+] }, align 4, !type !18
+
+; CHECK: define void @call3
+define void @call3(ptr %obj) #1 !dbg !16 {
+  %vtable = load ptr, ptr %obj
+  %p = call i1 @llvm.type.test(ptr %vtable, metadata !"typeid3")
+  call void @llvm.assume(i1 %p)
+  %fptr = call ptr @llvm.load.relative.i32(ptr %vtable, i32 8)
+  ; CHECK: if.true.direct_targ:
+  ; CHECK:   call i1 @vf(
+  ; CHECK: if.false.orig_indirect:
+  ; CHECK:   call i1 %fptr(
+  call i1 %fptr(ptr %obj), !dbg !17
+  ret void
+}
+
+
+declare i1 @llvm.type.test(ptr, metadata)
+declare i1 @llvm.public.type.test(ptr, metadata)
+declare void @llvm.assume(i1)
+
+!llvm.dbg.cu = !{!0}
+!llvm.module.flags = !{!2, !3}
+!llvm.ident = !{!4}
+
+!0 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus, file: !1, producer: "clang version 4.0.0 (trunk 278098)", isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug)
+!1 = !DIFile(filename: "devirt-single.cc", directory: ".")
+!2 = !{i32 2, !"Dwarf Version", i32 4}
+!3 = !{i32 2, !"Debug Info Version", i32 3}
+!4 = !{!"clang version 4.0.0 (trunk 278098)"}
+!5 = distinct !DISubprogram(name: "call", linkageName: "_Z4callPv", scope: !1, file: !1, line: 29, isLocal: false, isDefinition: true, scopeLine: 9, flags: DIFlagPrototyped, isOptimized: false, unit: !0)
+!6 = !DILocation(line: 30, column: 32, scope: !5)
+!7 = distinct !DISubprogram(name: "vf", linkageName: "_ZN3vt12vfEv", scope: !1, file: !1, line: 13, isLocal: false, isDefinition: true, scopeLine: 13, flags: DIFlagPrototyped, isOptimized: false, unit: !0)
+!8 = !{i32 0, !"typeid"}
+
+!9 = distinct !DISubprogram(name: "call1", linkageName: "_Z5call1Pv", scope: !1, file: !1, line: 31, isLocal: false, isDefinition: true, scopeLine: 9, flags: DIFlagPrototyped, isOptimized: false, unit: !0)
+!10 = !DILocation(line: 35, column: 32, scope: !9)
+!11 = distinct !DISubprogram(name: "vf_empty", linkageName: "_ZN3vt18vf_emptyEv", scope: !1, file: !1, line: 23, isLocal: false, isDefinition: true, scopeLine: 23, flags: DIFlagPrototyped, isOptimized: false, unit: !0)
+!12 = !{i32 0, !"typeid1"}
+
+!13 = distinct !DISubprogram(name: "call2", linkageName: "_Z5call2Pv", scope: !1, file: !1, line: 40, isLocal: false, isDefinition: true, scopeLine: 9, flags: DIFlagPrototyped, isOptimized: false, unit: !0)
+!14 = !DILocation(line: 41, column: 32, scope: !13)
+!15 = !{i32 0, !"typeid2"}
+
+!16 = distinct !DISubprogram(name: "call3", linkageName: "_Z5call3Pv", scope: !1, file: !1, line: 50, isLocal: false, isDefinition: true, scopeLine: 9, flags: DIFlagPrototyped, isOptimized: false, unit: !0)
+!17 = !DILocation(line: 51, column: 32, scope: !16)
+!18 = !{i32 0, !"typeid3"}
+
+
+
+; CHECK: 1 wholeprogramdevirt - Number of whole program devirtualization targets
+; CHECK: 3 wholeprogramdevirt - Number of single implementation devirtualizations

llvm/test/Transforms/WholeProgramDevirt/virtual-const-prop-check.ll

@@ -11,6 +11,9 @@
 ; Check wildcard
 ; RUN: opt -S -passes=wholeprogramdevirt -whole-program-visibility -pass-remarks=wholeprogramdevirt -wholeprogramdevirt-skip=vf?i1 %s 2>&1 | FileCheck %s --check-prefix=SKIP
 
+; Check that no stats are reported in speculative devirtualization mode as the virtual const prop is disabled.
+; RUN: opt -S -passes=wholeprogramdevirt -devirtualize-speculatively -stats %s 2>&1 | FileCheck %s --check-prefix=CHECK-SPECULATIVE-WPD
+
 target datalayout = "e-p:64:64"
 target triple = "x86_64-unknown-linux-gnu"
 
@@ -225,3 +228,7 @@ declare ptr @llvm.load.relative.i32(ptr, i32)
 ; CHECK: 2 wholeprogramdevirt - Number of unique return value optimizations
 ; CHECK: 2 wholeprogramdevirt - Number of virtual constant propagations
 ; CHECK: 2 wholeprogramdevirt - Number of 1 bit virtual constant propagations
+
+; CHECK-SPECULATIVE-WPD-NOT: 0 wholeprogramdevirt - Number of unique return value optimizations
+; CHECK-SPECULATIVE-WPD-NOT: 0 wholeprogramdevirt - Number of virtual constant propagations
+; CHECK-SPECULATIVE-WPD-NOT: 0 wholeprogramdevirt - Number of 1 bit virtual constant propagations