8 files changed, 88 insertions, 0 deletions
diff --git a/compiler/optimizing/intrinsics.cc b/compiler/optimizing/intrinsics.cc
index c6da9a3f5e..5caf077858 100644
--- a/compiler/optimizing/intrinsics.cc
+++ b/compiler/optimizing/intrinsics.cc
@@ -176,6 +176,16 @@ static Intrinsics GetIntrinsic(InlineMethod method) {
       }
 
     // Misc data processing.
+    case kIntrinsicBitCount:
+      switch (GetType(method.d.data, true)) {
+        case Primitive::kPrimInt:
+          return Intrinsics::kIntegerBitCount;
+        case Primitive::kPrimLong:
+          return Intrinsics::kLongBitCount;
+        default:
+          LOG(FATAL) << "Unknown/unsupported op size " << method.d.data;
+          UNREACHABLE();
+      }
     case kIntrinsicNumberOfLeadingZeros:
       switch (GetType(method.d.data, true)) {
         case Primitive::kPrimInt:
diff --git a/compiler/optimizing/intrinsics_arm.cc b/compiler/optimizing/intrinsics_arm.cc
index b1fbf28204..e72f927e44 100644
--- a/compiler/optimizing/intrinsics_arm.cc
+++ b/compiler/optimizing/intrinsics_arm.cc
@@ -1577,10 +1577,12 @@ void IntrinsicLocationsBuilderARM::Visit ## Name(HInvoke* invoke ATTRIBUTE_UNUSE
 void IntrinsicCodeGeneratorARM::Visit ## Name(HInvoke* invoke ATTRIBUTE_UNUSED) {    \
 }
 
+UNIMPLEMENTED_INTRINSIC(IntegerBitCount)
 UNIMPLEMENTED_INTRINSIC(IntegerReverse)
 UNIMPLEMENTED_INTRINSIC(IntegerReverseBytes)
 UNIMPLEMENTED_INTRINSIC(IntegerRotateLeft)
 UNIMPLEMENTED_INTRINSIC(IntegerRotateRight)
+UNIMPLEMENTED_INTRINSIC(LongBitCount)
 UNIMPLEMENTED_INTRINSIC(LongReverse)
 UNIMPLEMENTED_INTRINSIC(LongReverseBytes)
 UNIMPLEMENTED_INTRINSIC(LongRotateLeft)
diff --git a/compiler/optimizing/intrinsics_arm64.cc b/compiler/optimizing/intrinsics_arm64.cc
index e857f6f2ec..8cf2d4f393 100644
--- a/compiler/optimizing/intrinsics_arm64.cc
+++ b/compiler/optimizing/intrinsics_arm64.cc
@@ -1464,8 +1464,10 @@ void IntrinsicLocationsBuilderARM64::Visit ## Name(HInvoke* invoke ATTRIBUTE_UNU
 void IntrinsicCodeGeneratorARM64::Visit ## Name(HInvoke* invoke ATTRIBUTE_UNUSED) {    \
 }
 
+UNIMPLEMENTED_INTRINSIC(IntegerBitCount)
 UNIMPLEMENTED_INTRINSIC(IntegerRotateLeft)
 UNIMPLEMENTED_INTRINSIC(IntegerRotateRight)
+UNIMPLEMENTED_INTRINSIC(LongBitCount)
 UNIMPLEMENTED_INTRINSIC(LongRotateLeft)
 UNIMPLEMENTED_INTRINSIC(LongRotateRight)
 UNIMPLEMENTED_INTRINSIC(SystemArrayCopyChar)
diff --git a/compiler/optimizing/intrinsics_list.h b/compiler/optimizing/intrinsics_list.h
index 2e87546282..ea380347da 100644
--- a/compiler/optimizing/intrinsics_list.h
+++ b/compiler/optimizing/intrinsics_list.h
@@ -28,12 +28,14 @@
   V(FloatIntBitsToFloat, kStatic, kNeedsEnvironmentOrCache, kNoSideEffects, kNoThrow) \
   V(IntegerReverse, kStatic, kNeedsEnvironmentOrCache, kNoSideEffects, kNoThrow) \
   V(IntegerReverseBytes, kStatic, kNeedsEnvironmentOrCache, kNoSideEffects, kNoThrow) \
+  V(IntegerBitCount, kStatic, kNeedsEnvironmentOrCache, kNoSideEffects, kNoThrow) \
   V(IntegerNumberOfLeadingZeros, kStatic, kNeedsEnvironmentOrCache, kNoSideEffects, kNoThrow) \
   V(IntegerNumberOfTrailingZeros, kStatic, kNeedsEnvironmentOrCache, kNoSideEffects, kNoThrow) \
   V(IntegerRotateRight, kStatic, kNeedsEnvironmentOrCache, kNoSideEffects, kNoThrow) \
   V(IntegerRotateLeft, kStatic, kNeedsEnvironmentOrCache, kNoSideEffects, kNoThrow) \
   V(LongReverse, kStatic, kNeedsEnvironmentOrCache, kNoSideEffects, kNoThrow) \
   V(LongReverseBytes, kStatic, kNeedsEnvironmentOrCache, kNoSideEffects, kNoThrow) \
+  V(LongBitCount, kStatic, kNeedsEnvironmentOrCache, kNoSideEffects, kNoThrow) \
   V(LongNumberOfLeadingZeros, kStatic, kNeedsEnvironmentOrCache, kNoSideEffects, kNoThrow) \
   V(LongNumberOfTrailingZeros, kStatic, kNeedsEnvironmentOrCache, kNoSideEffects, kNoThrow) \
   V(LongRotateRight, kStatic, kNeedsEnvironmentOrCache, kNoSideEffects, kNoThrow) \
diff --git a/compiler/optimizing/intrinsics_mips.cc b/compiler/optimizing/intrinsics_mips.cc
index bc126a2716..81112b1a34 100644
--- a/compiler/optimizing/intrinsics_mips.cc
+++ b/compiler/optimizing/intrinsics_mips.cc
@@ -935,6 +935,9 @@ void IntrinsicLocationsBuilderMIPS::Visit ## Name(HInvoke* invoke ATTRIBUTE_UNUS
 void IntrinsicCodeGeneratorMIPS::Visit ## Name(HInvoke* invoke ATTRIBUTE_UNUSED) {    \
 }
 
+UNIMPLEMENTED_INTRINSIC(IntegerBitCount)
+UNIMPLEMENTED_INTRINSIC(LongBitCount)
+
 UNIMPLEMENTED_INTRINSIC(MathAbsDouble)
 UNIMPLEMENTED_INTRINSIC(MathAbsFloat)
 UNIMPLEMENTED_INTRINSIC(MathAbsInt)
diff --git a/compiler/optimizing/intrinsics_mips64.cc b/compiler/optimizing/intrinsics_mips64.cc
index 8b45ea7c4f..ac969e39fa 100644
--- a/compiler/optimizing/intrinsics_mips64.cc
+++ b/compiler/optimizing/intrinsics_mips64.cc
@@ -1724,6 +1724,9 @@ void IntrinsicLocationsBuilderMIPS64::Visit ## Name(HInvoke* invoke ATTRIBUTE_UN
 void IntrinsicCodeGeneratorMIPS64::Visit ## Name(HInvoke* invoke ATTRIBUTE_UNUSED) {    \
 }
 
+UNIMPLEMENTED_INTRINSIC(IntegerBitCount)
+UNIMPLEMENTED_INTRINSIC(LongBitCount)
+
 UNIMPLEMENTED_INTRINSIC(MathRoundDouble)
 UNIMPLEMENTED_INTRINSIC(MathRoundFloat)
 
diff --git a/compiler/optimizing/intrinsics_x86.cc b/compiler/optimizing/intrinsics_x86.cc
index 677f2e9c81..4715bdca2e 100644
--- a/compiler/optimizing/intrinsics_x86.cc
+++ b/compiler/optimizing/intrinsics_x86.cc
@@ -2518,8 +2518,10 @@ void IntrinsicCodeGeneratorX86::Visit ## Name(HInvoke* invoke ATTRIBUTE_UNUSED)
 
 UNIMPLEMENTED_INTRINSIC(MathRoundDouble)
 UNIMPLEMENTED_INTRINSIC(ReferenceGetReferent)
+UNIMPLEMENTED_INTRINSIC(IntegerBitCount)
 UNIMPLEMENTED_INTRINSIC(IntegerRotateLeft)
 UNIMPLEMENTED_INTRINSIC(IntegerRotateRight)
+UNIMPLEMENTED_INTRINSIC(LongBitCount)
 UNIMPLEMENTED_INTRINSIC(LongRotateRight)
 UNIMPLEMENTED_INTRINSIC(LongRotateLeft)
 UNIMPLEMENTED_INTRINSIC(SystemArrayCopy)
diff --git a/compiler/optimizing/intrinsics_x86_64.cc b/compiler/optimizing/intrinsics_x86_64.cc
index 690cf3d413..23a628f243 100644
--- a/compiler/optimizing/intrinsics_x86_64.cc
+++ b/compiler/optimizing/intrinsics_x86_64.cc
@@ -2368,6 +2368,70 @@ void IntrinsicCodeGeneratorX86_64::VisitLongReverse(HInvoke* invoke) {
   SwapBits64(reg, temp1, temp2, 4, INT64_C(0x0f0f0f0f0f0f0f0f), assembler);
 }
 
+static void CreateBitCountLocations(
+    ArenaAllocator* arena, CodeGeneratorX86_64* codegen, HInvoke* invoke) {
+  if (!codegen->GetInstructionSetFeatures().HasPopCnt()) {
+    // Do nothing if there is no popcnt support. This results in generating
+    // a call for the intrinsic rather than direct code.
+    return;
+  }
+  LocationSummary* locations = new (arena) LocationSummary(invoke,
+                                                           LocationSummary::kNoCall,
+                                                           kIntrinsified);
+  locations->SetInAt(0, Location::Any());
+  locations->SetOut(Location::RequiresRegister());
+}
+
+static void GenBitCount(X86_64Assembler* assembler, HInvoke* invoke, bool is_long) {
+  LocationSummary* locations = invoke->GetLocations();
+  Location src = locations->InAt(0);
+  CpuRegister out = locations->Out().AsRegister<CpuRegister>();
+
+  if (invoke->InputAt(0)->IsConstant()) {
+    // Evaluate this at compile time.
+    int64_t value = Int64FromConstant(invoke->InputAt(0)->AsConstant());
+    value = is_long
+        ? POPCOUNT(static_cast<uint64_t>(value))
+        : POPCOUNT(static_cast<uint32_t>(value));
+    if (value == 0) {
+      __ xorl(out, out);
+    } else {
+      __ movl(out, Immediate(value));
+    }
+    return;
+  }
+
+  if (src.IsRegister()) {
+    if (is_long) {
+      __ popcntq(out, src.AsRegister<CpuRegister>());
+    } else {
+      __ popcntl(out, src.AsRegister<CpuRegister>());
+    }
+  } else if (is_long) {
+    DCHECK(src.IsDoubleStackSlot());
+    __ popcntq(out, Address(CpuRegister(RSP), src.GetStackIndex()));
+  } else {
+    DCHECK(src.IsStackSlot());
+    __ popcntl(out, Address(CpuRegister(RSP), src.GetStackIndex()));
+  }
+}
+
+void IntrinsicLocationsBuilderX86_64::VisitIntegerBitCount(HInvoke* invoke) {
+  CreateBitCountLocations(arena_, codegen_, invoke);
+}
+
+void IntrinsicCodeGeneratorX86_64::VisitIntegerBitCount(HInvoke* invoke) {
+  GenBitCount(GetAssembler(), invoke, /* is_long */ false);
+}
+
+void IntrinsicLocationsBuilderX86_64::VisitLongBitCount(HInvoke* invoke) {
+  CreateBitCountLocations(arena_, codegen_, invoke);
+}
+
+void IntrinsicCodeGeneratorX86_64::VisitLongBitCount(HInvoke* invoke) {
+  GenBitCount(GetAssembler(), invoke, /* is_long */ true);
+}
+
 static void CreateLeadingZeroLocations(ArenaAllocator* arena, HInvoke* invoke) {
   LocationSummary* locations = new (arena) LocationSummary(invoke,
                                                            LocationSummary::kNoCall,