summaryrefslogtreecommitdiff
path: root/compiler/optimizing
diff options
context:
space:
mode:
Diffstat (limited to 'compiler/optimizing')
-rw-r--r--compiler/optimizing/bounds_check_elimination.cc33
-rw-r--r--compiler/optimizing/inliner.cc13
-rw-r--r--compiler/optimizing/intrinsics_arm.cc7
-rw-r--r--compiler/optimizing/intrinsics_arm64.cc7
-rw-r--r--compiler/optimizing/intrinsics_mips64.cc11
-rw-r--r--compiler/optimizing/intrinsics_x86.cc6
-rw-r--r--compiler/optimizing/intrinsics_x86_64.cc6
-rw-r--r--compiler/optimizing/nodes.cc61
-rw-r--r--compiler/optimizing/nodes.h1
9 files changed, 100 insertions, 45 deletions
diff --git a/compiler/optimizing/bounds_check_elimination.cc b/compiler/optimizing/bounds_check_elimination.cc
index a7a1c0f2c4..f2929bcc18 100644
--- a/compiler/optimizing/bounds_check_elimination.cc
+++ b/compiler/optimizing/bounds_check_elimination.cc
@@ -533,6 +533,8 @@ class BCEVisitor : public HGraphVisitor {
first_index_bounds_check_map_(
std::less<int>(),
graph->GetArena()->Adapter(kArenaAllocBoundsCheckElimination)),
+ dynamic_bce_standby_(
+ graph->GetArena()->Adapter(kArenaAllocBoundsCheckElimination)),
early_exit_loop_(
std::less<uint32_t>(),
graph->GetArena()->Adapter(kArenaAllocBoundsCheckElimination)),
@@ -553,6 +555,13 @@ class BCEVisitor : public HGraphVisitor {
}
void Finish() {
+ // Retry dynamic bce candidates on standby that are still in the graph.
+ for (HBoundsCheck* bounds_check : dynamic_bce_standby_) {
+ if (bounds_check->IsInBlock()) {
+ TryDynamicBCE(bounds_check);
+ }
+ }
+
// Preserve SSA structure which may have been broken by adding one or more
// new taken-test structures (see TransformLoopForDeoptimizationIfNeeded()).
InsertPhiNodes();
@@ -561,6 +570,7 @@ class BCEVisitor : public HGraphVisitor {
early_exit_loop_.clear();
taken_test_loop_.clear();
finite_loop_.clear();
+ dynamic_bce_standby_.clear();
}
private:
@@ -1301,7 +1311,7 @@ class BCEVisitor : public HGraphVisitor {
if (DynamicBCESeemsProfitable(loop, instruction->GetBlock()) &&
induction_range_.CanGenerateCode(
instruction, index, &needs_finite_test, &needs_taken_test) &&
- CanHandleInfiniteLoop(loop, index, needs_finite_test) &&
+ CanHandleInfiniteLoop(loop, instruction, index, needs_finite_test) &&
CanHandleLength(loop, length, needs_taken_test)) { // do this test last (may code gen)
HInstruction* lower = nullptr;
HInstruction* upper = nullptr;
@@ -1433,7 +1443,7 @@ class BCEVisitor : public HGraphVisitor {
* ensure the loop is finite.
*/
bool CanHandleInfiniteLoop(
- HLoopInformation* loop, HInstruction* index, bool needs_infinite_test) {
+ HLoopInformation* loop, HBoundsCheck* check, HInstruction* index, bool needs_infinite_test) {
if (needs_infinite_test) {
// If we already forced the loop to be finite, allow directly.
const uint32_t loop_id = loop->GetHeader()->GetBlockId();
@@ -1455,6 +1465,9 @@ class BCEVisitor : public HGraphVisitor {
}
}
}
+ // If bounds check made it this far, it is worthwhile to check later if
+ // the loop was forced finite by another candidate.
+ dynamic_bce_standby_.push_back(check);
return false;
}
return true;
@@ -1676,6 +1689,9 @@ class BCEVisitor : public HGraphVisitor {
// in a block that checks an index against that HArrayLength.
ArenaSafeMap<int, HBoundsCheck*> first_index_bounds_check_map_;
+ // Stand by list for dynamic bce.
+ ArenaVector<HBoundsCheck*> dynamic_bce_standby_;
+
// Early-exit loop bookkeeping.
ArenaSafeMap<uint32_t, bool> early_exit_loop_;
@@ -1711,21 +1727,18 @@ void BoundsCheckElimination::Run() {
// that value dominated by that instruction fits in that range. Range of that
// value can be narrowed further down in the dominator tree.
BCEVisitor visitor(graph_, side_effects_, induction_analysis_);
- HBasicBlock* last_visited_block = nullptr;
for (HReversePostOrderIterator it(*graph_); !it.Done(); it.Advance()) {
HBasicBlock* current = it.Current();
- if (current == last_visited_block) {
- // We may insert blocks into the reverse post order list when processing
- // a loop header. Don't process it again.
- DCHECK(current->IsLoopHeader());
- continue;
- }
if (visitor.IsAddedBlock(current)) {
// Skip added blocks. Their effects are already taken care of.
continue;
}
visitor.VisitBasicBlock(current);
- last_visited_block = current;
+ // Skip forward to the current block in case new basic blocks were inserted
+ // (which always appear earlier in reverse post order) to avoid visiting the
+ // same basic block twice.
+ for ( ; !it.Done() && it.Current() != current; it.Advance()) {
+ }
}
// Perform cleanup.
diff --git a/compiler/optimizing/inliner.cc b/compiler/optimizing/inliner.cc
index 3f67e481f9..3e3719e6ea 100644
--- a/compiler/optimizing/inliner.cc
+++ b/compiler/optimizing/inliner.cc
@@ -1010,6 +1010,8 @@ bool HInliner::TryBuildAndInlineHelper(HInvoke* invoke_instruction,
// at runtime, we change this call as if it was a virtual call.
invoke_type = kVirtual;
}
+
+ const int32_t caller_instruction_counter = graph_->GetCurrentInstructionId();
HGraph* callee_graph = new (graph_->GetArena()) HGraph(
graph_->GetArena(),
callee_dex_file,
@@ -1019,7 +1021,7 @@ bool HInliner::TryBuildAndInlineHelper(HInvoke* invoke_instruction,
invoke_type,
graph_->IsDebuggable(),
/* osr */ false,
- graph_->GetCurrentInstructionId());
+ caller_instruction_counter);
callee_graph->SetArtMethod(resolved_method);
OptimizingCompilerStats inline_stats;
@@ -1219,7 +1221,16 @@ bool HInliner::TryBuildAndInlineHelper(HInvoke* invoke_instruction,
}
number_of_inlined_instructions_ += number_of_instructions;
+ DCHECK_EQ(caller_instruction_counter, graph_->GetCurrentInstructionId())
+ << "No instructions can be added to the outer graph while inner graph is being built";
+
+ const int32_t callee_instruction_counter = callee_graph->GetCurrentInstructionId();
+ graph_->SetCurrentInstructionId(callee_instruction_counter);
*return_replacement = callee_graph->InlineInto(graph_, invoke_instruction);
+
+ DCHECK_EQ(callee_instruction_counter, callee_graph->GetCurrentInstructionId())
+ << "No instructions can be added to the inner graph during inlining into the outer graph";
+
return true;
}
diff --git a/compiler/optimizing/intrinsics_arm.cc b/compiler/optimizing/intrinsics_arm.cc
index 4ce919ee39..276085ef30 100644
--- a/compiler/optimizing/intrinsics_arm.cc
+++ b/compiler/optimizing/intrinsics_arm.cc
@@ -1151,6 +1151,7 @@ static void GenerateVisitStringIndexOf(HInvoke* invoke,
__ LoadFromOffset(kLoadWord, LR, TR,
QUICK_ENTRYPOINT_OFFSET(kArmWordSize, pIndexOf).Int32Value());
+ CheckEntrypointTypes<kQuickIndexOf, int32_t, void*, uint32_t, uint32_t>();
__ blx(LR);
if (slow_path != nullptr) {
@@ -1242,6 +1243,12 @@ void IntrinsicLocationsBuilderARM::VisitStringNewStringFromChars(HInvoke* invoke
void IntrinsicCodeGeneratorARM::VisitStringNewStringFromChars(HInvoke* invoke) {
ArmAssembler* assembler = GetAssembler();
+ // No need to emit code checking whether `locations->InAt(2)` is a null
+ // pointer, as callers of the native method
+ //
+ // java.lang.StringFactory.newStringFromChars(int offset, int charCount, char[] data)
+ //
+ // all include a null check on `data` before calling that method.
__ LoadFromOffset(
kLoadWord, LR, TR, QUICK_ENTRYPOINT_OFFSET(kArmWordSize, pAllocStringFromChars).Int32Value());
codegen_->RecordPcInfo(invoke, invoke->GetDexPc());
diff --git a/compiler/optimizing/intrinsics_arm64.cc b/compiler/optimizing/intrinsics_arm64.cc
index 4be1695a94..9f9d8c493f 100644
--- a/compiler/optimizing/intrinsics_arm64.cc
+++ b/compiler/optimizing/intrinsics_arm64.cc
@@ -1301,6 +1301,7 @@ static void GenerateVisitStringIndexOf(HInvoke* invoke,
}
__ Ldr(lr, MemOperand(tr, QUICK_ENTRYPOINT_OFFSET(kArm64WordSize, pIndexOf).Int32Value()));
+ CheckEntrypointTypes<kQuickIndexOf, int32_t, void*, uint32_t, uint32_t>();
__ Blr(lr);
if (slow_path != nullptr) {
@@ -1392,6 +1393,12 @@ void IntrinsicLocationsBuilderARM64::VisitStringNewStringFromChars(HInvoke* invo
void IntrinsicCodeGeneratorARM64::VisitStringNewStringFromChars(HInvoke* invoke) {
vixl::MacroAssembler* masm = GetVIXLAssembler();
+ // No need to emit code checking whether `locations->InAt(2)` is a null
+ // pointer, as callers of the native method
+ //
+ // java.lang.StringFactory.newStringFromChars(int offset, int charCount, char[] data)
+ //
+ // all include a null check on `data` before calling that method.
__ Ldr(lr,
MemOperand(tr, QUICK_ENTRYPOINT_OFFSET(kArm64WordSize, pAllocStringFromChars).Int32Value()));
codegen_->RecordPcInfo(invoke, invoke->GetDexPc());
diff --git a/compiler/optimizing/intrinsics_mips64.cc b/compiler/optimizing/intrinsics_mips64.cc
index ca2652b74a..a7a2560043 100644
--- a/compiler/optimizing/intrinsics_mips64.cc
+++ b/compiler/optimizing/intrinsics_mips64.cc
@@ -1636,6 +1636,7 @@ static void GenerateStringIndexOf(HInvoke* invoke,
TMP,
TR,
QUICK_ENTRYPOINT_OFFSET(kMips64DoublewordSize, pIndexOf).Int32Value());
+ CheckEntrypointTypes<kQuickIndexOf, int32_t, void*, uint32_t, uint32_t>();
__ Jalr(TMP);
__ Nop();
@@ -1685,7 +1686,7 @@ void IntrinsicCodeGeneratorMIPS64::VisitStringIndexOfAfter(HInvoke* invoke) {
invoke, GetAssembler(), codegen_, GetAllocator(), /* start_at_zero */ false);
}
-// java.lang.String.String(byte[] bytes)
+// java.lang.StringFactory.newStringFromBytes(byte[] data, int high, int offset, int byteCount)
void IntrinsicLocationsBuilderMIPS64::VisitStringNewStringFromBytes(HInvoke* invoke) {
LocationSummary* locations = new (arena_) LocationSummary(invoke,
LocationSummary::kCall,
@@ -1719,7 +1720,7 @@ void IntrinsicCodeGeneratorMIPS64::VisitStringNewStringFromBytes(HInvoke* invoke
__ Bind(slow_path->GetExitLabel());
}
-// java.lang.String.String(char[] value)
+// java.lang.StringFactory.newStringFromChars(int offset, int charCount, char[] data)
void IntrinsicLocationsBuilderMIPS64::VisitStringNewStringFromChars(HInvoke* invoke) {
LocationSummary* locations = new (arena_) LocationSummary(invoke,
LocationSummary::kCall,
@@ -1735,6 +1736,12 @@ void IntrinsicLocationsBuilderMIPS64::VisitStringNewStringFromChars(HInvoke* inv
void IntrinsicCodeGeneratorMIPS64::VisitStringNewStringFromChars(HInvoke* invoke) {
Mips64Assembler* assembler = GetAssembler();
+ // No need to emit code checking whether `locations->InAt(2)` is a null
+ // pointer, as callers of the native method
+ //
+ // java.lang.StringFactory.newStringFromChars(int offset, int charCount, char[] data)
+ //
+ // all include a null check on `data` before calling that method.
__ LoadFromOffset(kLoadDoubleword,
TMP,
TR,
diff --git a/compiler/optimizing/intrinsics_x86.cc b/compiler/optimizing/intrinsics_x86.cc
index 0df4553f56..bfa4e374f7 100644
--- a/compiler/optimizing/intrinsics_x86.cc
+++ b/compiler/optimizing/intrinsics_x86.cc
@@ -1564,6 +1564,12 @@ void IntrinsicLocationsBuilderX86::VisitStringNewStringFromChars(HInvoke* invoke
void IntrinsicCodeGeneratorX86::VisitStringNewStringFromChars(HInvoke* invoke) {
X86Assembler* assembler = GetAssembler();
+ // No need to emit code checking whether `locations->InAt(2)` is a null
+ // pointer, as callers of the native method
+ //
+ // java.lang.StringFactory.newStringFromChars(int offset, int charCount, char[] data)
+ //
+ // all include a null check on `data` before calling that method.
__ fs()->call(Address::Absolute(QUICK_ENTRYPOINT_OFFSET(kX86WordSize, pAllocStringFromChars)));
codegen_->RecordPcInfo(invoke, invoke->GetDexPc());
}
diff --git a/compiler/optimizing/intrinsics_x86_64.cc b/compiler/optimizing/intrinsics_x86_64.cc
index 2a9e684d11..15c399712d 100644
--- a/compiler/optimizing/intrinsics_x86_64.cc
+++ b/compiler/optimizing/intrinsics_x86_64.cc
@@ -1659,6 +1659,12 @@ void IntrinsicLocationsBuilderX86_64::VisitStringNewStringFromChars(HInvoke* inv
void IntrinsicCodeGeneratorX86_64::VisitStringNewStringFromChars(HInvoke* invoke) {
X86_64Assembler* assembler = GetAssembler();
+ // No need to emit code checking whether `locations->InAt(2)` is a null
+ // pointer, as callers of the native method
+ //
+ // java.lang.StringFactory.newStringFromChars(int offset, int charCount, char[] data)
+ //
+ // all include a null check on `data` before calling that method.
__ gs()->call(Address::Absolute(QUICK_ENTRYPOINT_OFFSET(kX86_64WordSize, pAllocStringFromChars),
/* no_rip */ true));
codegen_->RecordPcInfo(invoke, invoke->GetDexPc());
diff --git a/compiler/optimizing/nodes.cc b/compiler/optimizing/nodes.cc
index 87b9c022df..0e0b83e4b4 100644
--- a/compiler/optimizing/nodes.cc
+++ b/compiler/optimizing/nodes.cc
@@ -1976,34 +1976,6 @@ HInstruction* HGraph::InlineInto(HGraph* outer_graph, HInvoke* invoke) {
at->MergeWithInlined(first);
exit_block_->ReplaceWith(to);
- // Update all predecessors of the exit block (now the `to` block)
- // to not `HReturn` but `HGoto` instead.
- bool returns_void = to->GetPredecessors()[0]->GetLastInstruction()->IsReturnVoid();
- if (to->GetPredecessors().size() == 1) {
- HBasicBlock* predecessor = to->GetPredecessors()[0];
- HInstruction* last = predecessor->GetLastInstruction();
- if (!returns_void) {
- return_value = last->InputAt(0);
- }
- predecessor->AddInstruction(new (allocator) HGoto(last->GetDexPc()));
- predecessor->RemoveInstruction(last);
- } else {
- if (!returns_void) {
- // There will be multiple returns.
- return_value = new (allocator) HPhi(
- allocator, kNoRegNumber, 0, HPhi::ToPhiType(invoke->GetType()), to->GetDexPc());
- to->AddPhi(return_value->AsPhi());
- }
- for (HBasicBlock* predecessor : to->GetPredecessors()) {
- HInstruction* last = predecessor->GetLastInstruction();
- if (!returns_void) {
- return_value->AsPhi()->AddInput(last->InputAt(0));
- }
- predecessor->AddInstruction(new (allocator) HGoto(last->GetDexPc()));
- predecessor->RemoveInstruction(last);
- }
- }
-
// Update the meta information surrounding blocks:
// (1) the graph they are now in,
// (2) the reverse post order of that graph,
@@ -2048,11 +2020,36 @@ HInstruction* HGraph::InlineInto(HGraph* outer_graph, HInvoke* invoke) {
// Only `to` can become a back edge, as the inlined blocks
// are predecessors of `to`.
UpdateLoopAndTryInformationOfNewBlock(to, at, /* replace_if_back_edge */ true);
- }
- // Update the next instruction id of the outer graph, so that instructions
- // added later get bigger ids than those in the inner graph.
- outer_graph->SetCurrentInstructionId(GetNextInstructionId());
+ // Update all predecessors of the exit block (now the `to` block)
+ // to not `HReturn` but `HGoto` instead.
+ bool returns_void = to->GetPredecessors()[0]->GetLastInstruction()->IsReturnVoid();
+ if (to->GetPredecessors().size() == 1) {
+ HBasicBlock* predecessor = to->GetPredecessors()[0];
+ HInstruction* last = predecessor->GetLastInstruction();
+ if (!returns_void) {
+ return_value = last->InputAt(0);
+ }
+ predecessor->AddInstruction(new (allocator) HGoto(last->GetDexPc()));
+ predecessor->RemoveInstruction(last);
+ } else {
+ if (!returns_void) {
+ // There will be multiple returns.
+ return_value = new (allocator) HPhi(
+ allocator, kNoRegNumber, 0, HPhi::ToPhiType(invoke->GetType()), to->GetDexPc());
+ to->AddPhi(return_value->AsPhi());
+ }
+ for (HBasicBlock* predecessor : to->GetPredecessors()) {
+ HInstruction* last = predecessor->GetLastInstruction();
+ if (!returns_void) {
+ DCHECK(last->IsReturn());
+ return_value->AsPhi()->AddInput(last->InputAt(0));
+ }
+ predecessor->AddInstruction(new (allocator) HGoto(last->GetDexPc()));
+ predecessor->RemoveInstruction(last);
+ }
+ }
+ }
// Walk over the entry block and:
// - Move constants from the entry block to the outer_graph's entry block,
diff --git a/compiler/optimizing/nodes.h b/compiler/optimizing/nodes.h
index e3dbe16547..b684cc697f 100644
--- a/compiler/optimizing/nodes.h
+++ b/compiler/optimizing/nodes.h
@@ -387,6 +387,7 @@ class HGraph : public ArenaObject<kArenaAllocGraph> {
}
void SetCurrentInstructionId(int32_t id) {
+ DCHECK_GE(id, current_instruction_id_);
current_instruction_id_ = id;
}
generated by cgit v1.2.3-59-g8ed1b (git 2.39.0) at 2025-12-11 15:59:25 +0000