Add Math.multiplyHigh intrinsic

Test: ./art/test/testrunner/testrunner.py --target  --optimizing --64 -t 082-inline-execute
Test: ./art/test/testrunner/testrunner.py --host --optimizing --64 -t 082-inline-execute
Change-Id: I4b6cafa8b9e513eca7c5c139440024d87a7ef758

4 files changed, 40 insertions, 0 deletions

diff --git a/compiler/optimizing/intrinsics_arm64.cc b/compiler/optimizing/intrinsics_arm64.cc
index d58d8f4b0a..d0c64c2230 100644
--- a/compiler/optimizing/intrinsics_arm64.cc
+++ b/compiler/optimizing/intrinsics_arm64.cc
@@ -3856,6 +3856,23 @@ void IntrinsicCodeGeneratorARM64::VisitLongDivideUnsigned(HInvoke* invoke) {
   GenerateDivideUnsigned(invoke, codegen_);
 }
 
+void IntrinsicLocationsBuilderARM64::VisitMathMultiplyHigh(HInvoke* invoke) {
+  CreateIntIntToIntLocations(allocator_, invoke);
+}
+
+void IntrinsicCodeGeneratorARM64::VisitMathMultiplyHigh(HInvoke* invoke) {
+  LocationSummary* locations = invoke->GetLocations();
+  MacroAssembler* masm = codegen_->GetVIXLAssembler();
+  DataType::Type type = invoke->GetType();
+  DCHECK(type == DataType::Type::kInt64);
+
+  Register x = RegisterFrom(locations->InAt(0), type);
+  Register y = RegisterFrom(locations->InAt(1), type);
+  Register out = RegisterFrom(locations->Out(), type);
+
+  __ Smulh(out, x, y);
+}
+
 class VarHandleSlowPathARM64 : public IntrinsicSlowPathARM64 {
  public:
   VarHandleSlowPathARM64(HInvoke* invoke, std::memory_order order)
diff --git a/compiler/optimizing/intrinsics_arm_vixl.cc b/compiler/optimizing/intrinsics_arm_vixl.cc
index 42156d43c5..2a2f245025 100644
--- a/compiler/optimizing/intrinsics_arm_vixl.cc
+++ b/compiler/optimizing/intrinsics_arm_vixl.cc
@@ -5371,6 +5371,7 @@ UNIMPLEMENTED_INTRINSIC(ARMVIXL, FP16Greater)
 UNIMPLEMENTED_INTRINSIC(ARMVIXL, FP16GreaterEquals)
 UNIMPLEMENTED_INTRINSIC(ARMVIXL, FP16Less)
 UNIMPLEMENTED_INTRINSIC(ARMVIXL, FP16LessEquals)
+UNIMPLEMENTED_INTRINSIC(ARMVIXL, MathMultiplyHigh)
 
 UNIMPLEMENTED_INTRINSIC(ARMVIXL, StringStringIndexOf);
 UNIMPLEMENTED_INTRINSIC(ARMVIXL, StringStringIndexOfAfter);
diff --git a/compiler/optimizing/intrinsics_x86.cc b/compiler/optimizing/intrinsics_x86.cc
index a4f1bd1df2..b0c4b5736f 100644
--- a/compiler/optimizing/intrinsics_x86.cc
+++ b/compiler/optimizing/intrinsics_x86.cc
@@ -4613,6 +4613,7 @@ UNIMPLEMENTED_INTRINSIC(X86, FP16Greater)
 UNIMPLEMENTED_INTRINSIC(X86, FP16GreaterEquals)
 UNIMPLEMENTED_INTRINSIC(X86, FP16Less)
 UNIMPLEMENTED_INTRINSIC(X86, FP16LessEquals)
+UNIMPLEMENTED_INTRINSIC(X86, MathMultiplyHigh)
 
 UNIMPLEMENTED_INTRINSIC(X86, StringStringIndexOf);
 UNIMPLEMENTED_INTRINSIC(X86, StringStringIndexOfAfter);
diff --git a/compiler/optimizing/intrinsics_x86_64.cc b/compiler/optimizing/intrinsics_x86_64.cc
index 7d056f9c59..63511817c0 100644
--- a/compiler/optimizing/intrinsics_x86_64.cc
+++ b/compiler/optimizing/intrinsics_x86_64.cc
@@ -2830,6 +2830,27 @@ void IntrinsicCodeGeneratorX86_64::VisitIntegerDivideUnsigned(HInvoke* invoke) {
   __ Bind(slow_path->GetExitLabel());
 }
 
+void IntrinsicLocationsBuilderX86_64::VisitMathMultiplyHigh(HInvoke* invoke) {
+  LocationSummary* locations =
+      new (allocator_) LocationSummary(invoke, LocationSummary::kNoCall, kIntrinsified);
+  locations->SetInAt(0, Location::RegisterLocation(RAX));
+  locations->SetInAt(1, Location::RequiresRegister());
+  locations->SetOut(Location::RegisterLocation(RDX));
+  locations->AddTemp(Location::RegisterLocation(RAX));
+}
+
+void IntrinsicCodeGeneratorX86_64::VisitMathMultiplyHigh(HInvoke* invoke) {
+  X86_64Assembler* assembler = GetAssembler();
+  LocationSummary* locations = invoke->GetLocations();
+
+  CpuRegister y = locations->InAt(1).AsRegister<CpuRegister>();
+
+  DCHECK_EQ(locations->InAt(0).AsRegister<Register>(), RAX);
+  DCHECK_EQ(locations->Out().AsRegister<Register>(), RDX);
+
+  __ imulq(y);
+}
+
 
 UNIMPLEMENTED_INTRINSIC(X86_64, FloatIsInfinite)
 UNIMPLEMENTED_INTRINSIC(X86_64, DoubleIsInfinite)