ART: Optimizing compiler intrinsics
Add intrinsics infrastructure to the optimizing compiler.
Add almost all intrinsics supported by Quick to the x86-64 backend.
Further intrinsics require more assembler support.
Change-Id: I48de9b44c82886bb298d16e74e12a9506b8e8807
diff --git a/compiler/optimizing/code_generator_x86_64.cc b/compiler/optimizing/code_generator_x86_64.cc
index 5fc24f7..3d7f122 100644
--- a/compiler/optimizing/code_generator_x86_64.cc
+++ b/compiler/optimizing/code_generator_x86_64.cc
@@ -18,6 +18,8 @@
#include "entrypoints/quick/quick_entrypoints.h"
#include "gc/accounting/card_table.h"
+#include "intrinsics.h"
+#include "intrinsics_x86_64.h"
#include "mirror/array-inl.h"
#include "mirror/art_method.h"
#include "mirror/class.h"
@@ -61,20 +63,6 @@
#define __ reinterpret_cast<X86_64Assembler*>(codegen->GetAssembler())->
-class SlowPathCodeX86_64 : public SlowPathCode {
- public:
- SlowPathCodeX86_64() : entry_label_(), exit_label_() {}
-
- Label* GetEntryLabel() { return &entry_label_; }
- Label* GetExitLabel() { return &exit_label_; }
-
- private:
- Label entry_label_;
- Label exit_label_;
-
- DISALLOW_COPY_AND_ASSIGN(SlowPathCodeX86_64);
-};
-
class NullCheckSlowPathX86_64 : public SlowPathCodeX86_64 {
public:
explicit NullCheckSlowPathX86_64(HNullCheck* instruction) : instruction_(instruction) {}
@@ -375,6 +363,31 @@
return kEqual;
}
+void CodeGeneratorX86_64::GenerateStaticOrDirectCall(HInvokeStaticOrDirect* invoke,
+ CpuRegister temp) {
+ // All registers are assumed to be correctly set up.
+
+ // TODO: Implement all kinds of calls:
+ // 1) boot -> boot
+ // 2) app -> boot
+ // 3) app -> app
+ //
+ // Currently we implement the app -> app logic, which looks up in the resolve cache.
+
+ // temp = method;
+ LoadCurrentMethod(temp);
+ // temp = temp->dex_cache_resolved_methods_;
+ __ movl(temp, Address(temp, mirror::ArtMethod::DexCacheResolvedMethodsOffset().SizeValue()));
+ // temp = temp[index_in_cache]
+ __ movl(temp, Address(temp, CodeGenerator::GetCacheOffset(invoke->GetDexMethodIndex())));
+ // (temp + offset_of_quick_compiled_code)()
+ __ call(Address(temp, mirror::ArtMethod::EntryPointFromQuickCompiledCodeOffset(
+ kX86_64WordSize).SizeValue()));
+
+ DCHECK(!IsLeafMethod());
+ RecordPcInfo(invoke, invoke->GetDexPc());
+}
+
void CodeGeneratorX86_64::DumpCoreRegister(std::ostream& stream, int reg) const {
stream << X86_64ManagedRegister::FromCpuRegister(Register(reg));
}
@@ -1123,30 +1136,31 @@
}
void LocationsBuilderX86_64::VisitInvokeStaticOrDirect(HInvokeStaticOrDirect* invoke) {
+ IntrinsicLocationsBuilderX86_64 intrinsic(GetGraph()->GetArena());
+ if (intrinsic.TryDispatch(invoke)) {
+ return;
+ }
+
HandleInvoke(invoke);
}
+static bool TryGenerateIntrinsicCode(HInvoke* invoke, CodeGeneratorX86_64* codegen) {
+ if (invoke->GetLocations()->Intrinsified()) {
+ IntrinsicCodeGeneratorX86_64 intrinsic(codegen);
+ intrinsic.Dispatch(invoke);
+ return true;
+ }
+ return false;
+}
+
void InstructionCodeGeneratorX86_64::VisitInvokeStaticOrDirect(HInvokeStaticOrDirect* invoke) {
- CpuRegister temp = invoke->GetLocations()->GetTemp(0).AsRegister<CpuRegister>();
- // TODO: Implement all kinds of calls:
- // 1) boot -> boot
- // 2) app -> boot
- // 3) app -> app
- //
- // Currently we implement the app -> app logic, which looks up in the resolve cache.
+ if (TryGenerateIntrinsicCode(invoke, codegen_)) {
+ return;
+ }
- // temp = method;
- codegen_->LoadCurrentMethod(temp);
- // temp = temp->dex_cache_resolved_methods_;
- __ movl(temp, Address(temp, mirror::ArtMethod::DexCacheResolvedMethodsOffset().SizeValue()));
- // temp = temp[index_in_cache]
- __ movl(temp, Address(temp, CodeGenerator::GetCacheOffset(invoke->GetIndexInDexCache())));
- // (temp + offset_of_quick_compiled_code)()
- __ call(Address(temp, mirror::ArtMethod::EntryPointFromQuickCompiledCodeOffset(
- kX86_64WordSize).SizeValue()));
-
- DCHECK(!codegen_->IsLeafMethod());
- codegen_->RecordPcInfo(invoke, invoke->GetDexPc());
+ codegen_->GenerateStaticOrDirectCall(
+ invoke,
+ invoke->GetLocations()->GetTemp(0).AsRegister<CpuRegister>());
}
void LocationsBuilderX86_64::HandleInvoke(HInvoke* invoke) {
@@ -1182,10 +1196,19 @@
}
void LocationsBuilderX86_64::VisitInvokeVirtual(HInvokeVirtual* invoke) {
+ IntrinsicLocationsBuilderX86_64 intrinsic(GetGraph()->GetArena());
+ if (intrinsic.TryDispatch(invoke)) {
+ return;
+ }
+
HandleInvoke(invoke);
}
void InstructionCodeGeneratorX86_64::VisitInvokeVirtual(HInvokeVirtual* invoke) {
+ if (TryGenerateIntrinsicCode(invoke, codegen_)) {
+ return;
+ }
+
CpuRegister temp = invoke->GetLocations()->GetTemp(0).AsRegister<CpuRegister>();
size_t method_offset = mirror::Class::EmbeddedVTableOffset().SizeValue() +
invoke->GetVTableIndex() * sizeof(mirror::Class::VTableEntry);