ARM/ARM64: Implement numberOfLeadingZeros intrinsic.
Change-Id: I4042fb7a0b75140475dcfca23e8f79d310f5333b
diff --git a/compiler/dex/quick/dex_file_method_inliner.cc b/compiler/dex/quick/dex_file_method_inliner.cc
index 2568ee3..7fc6fa2 100644
--- a/compiler/dex/quick/dex_file_method_inliner.cc
+++ b/compiler/dex/quick/dex_file_method_inliner.cc
@@ -38,6 +38,7 @@
true, // kIntrinsicFloatCvt
true, // kIntrinsicReverseBits
true, // kIntrinsicReverseBytes
+ true, // kIntrinsicNumberOfLeadingZeros
true, // kIntrinsicAbsInt
true, // kIntrinsicAbsLong
true, // kIntrinsicAbsFloat
@@ -75,6 +76,8 @@
static_assert(kIntrinsicIsStatic[kIntrinsicFloatCvt], "FloatCvt must be static");
static_assert(kIntrinsicIsStatic[kIntrinsicReverseBits], "ReverseBits must be static");
static_assert(kIntrinsicIsStatic[kIntrinsicReverseBytes], "ReverseBytes must be static");
+static_assert(kIntrinsicIsStatic[kIntrinsicNumberOfLeadingZeros],
+ "NumberOfLeadingZeros must be static");
static_assert(kIntrinsicIsStatic[kIntrinsicAbsInt], "AbsInt must be static");
static_assert(kIntrinsicIsStatic[kIntrinsicAbsLong], "AbsLong must be static");
static_assert(kIntrinsicIsStatic[kIntrinsicAbsFloat], "AbsFloat must be static");
@@ -225,6 +228,7 @@
"putObjectVolatile", // kNameCachePutObjectVolatile
"putOrderedObject", // kNameCachePutOrderedObject
"arraycopy", // kNameCacheArrayCopy
+ "numberOfLeadingZeros", // kNameCacheNumberOfLeadingZeros
};
const DexFileMethodInliner::ProtoDef DexFileMethodInliner::kProtoCacheDefs[] = {
@@ -368,6 +372,9 @@
INTRINSIC(JavaLangInteger, Reverse, I_I, kIntrinsicReverseBits, k32),
INTRINSIC(JavaLangLong, Reverse, J_J, kIntrinsicReverseBits, k64),
+ INTRINSIC(JavaLangInteger, NumberOfLeadingZeros, I_I, kIntrinsicNumberOfLeadingZeros, k32),
+ INTRINSIC(JavaLangLong, NumberOfLeadingZeros, J_I, kIntrinsicNumberOfLeadingZeros, k64),
+
INTRINSIC(JavaLangMath, Abs, I_I, kIntrinsicAbsInt, 0),
INTRINSIC(JavaLangStrictMath, Abs, I_I, kIntrinsicAbsInt, 0),
INTRINSIC(JavaLangMath, Abs, J_J, kIntrinsicAbsLong, 0),
@@ -614,6 +621,8 @@
intrinsic.d.data & kIntrinsicFlagIsOrdered);
case kIntrinsicSystemArrayCopyCharArray:
return backend->GenInlinedArrayCopyCharArray(info);
+ case kIntrinsicNumberOfLeadingZeros:
+ return false; // not implemented in quick
default:
LOG(FATAL) << "Unexpected intrinsic opcode: " << intrinsic.opcode;
return false; // avoid warning "control reaches end of non-void function"
diff --git a/compiler/dex/quick/dex_file_method_inliner.h b/compiler/dex/quick/dex_file_method_inliner.h
index a8cb9f0..bcb9ee5 100644
--- a/compiler/dex/quick/dex_file_method_inliner.h
+++ b/compiler/dex/quick/dex_file_method_inliner.h
@@ -206,6 +206,7 @@
kNameCachePutObjectVolatile,
kNameCachePutOrderedObject,
kNameCacheArrayCopy,
+ kNameCacheNumberOfLeadingZeros,
kNameCacheLast
};
diff --git a/compiler/optimizing/intrinsics.cc b/compiler/optimizing/intrinsics.cc
index bc7da80..55e964e 100644
--- a/compiler/optimizing/intrinsics.cc
+++ b/compiler/optimizing/intrinsics.cc
@@ -103,6 +103,16 @@
LOG(FATAL) << "Unknown/unsupported op size " << method.d.data;
UNREACHABLE();
}
+ case kIntrinsicNumberOfLeadingZeros:
+ switch (GetType(method.d.data, true)) {
+ case Primitive::kPrimInt:
+ return Intrinsics::kIntegerNumberOfLeadingZeros;
+ case Primitive::kPrimLong:
+ return Intrinsics::kLongNumberOfLeadingZeros;
+ default:
+ LOG(FATAL) << "Unknown/unsupported op size " << method.d.data;
+ UNREACHABLE();
+ }
// Abs.
case kIntrinsicAbsDouble:
diff --git a/compiler/optimizing/intrinsics_arm.cc b/compiler/optimizing/intrinsics_arm.cc
index b4dbf75..a797654 100644
--- a/compiler/optimizing/intrinsics_arm.cc
+++ b/compiler/optimizing/intrinsics_arm.cc
@@ -224,6 +224,48 @@
locations->SetOut(Location::RequiresFpuRegister(), Location::kNoOutputOverlap);
}
+static void GenNumberOfLeadingZeros(LocationSummary* locations,
+ Primitive::Type type,
+ ArmAssembler* assembler) {
+ Location in = locations->InAt(0);
+ Register out = locations->Out().AsRegister<Register>();
+
+ DCHECK((type == Primitive::kPrimInt) || (type == Primitive::kPrimLong));
+
+ if (type == Primitive::kPrimLong) {
+ Register in_reg_lo = in.AsRegisterPairLow<Register>();
+ Register in_reg_hi = in.AsRegisterPairHigh<Register>();
+ Label end;
+ __ clz(out, in_reg_hi);
+ __ CompareAndBranchIfNonZero(in_reg_hi, &end);
+ __ clz(out, in_reg_lo);
+ __ AddConstant(out, 32);
+ __ Bind(&end);
+ } else {
+ __ clz(out, in.AsRegister<Register>());
+ }
+}
+
+void IntrinsicLocationsBuilderARM::VisitIntegerNumberOfLeadingZeros(HInvoke* invoke) {
+ CreateIntToIntLocations(arena_, invoke);
+}
+
+void IntrinsicCodeGeneratorARM::VisitIntegerNumberOfLeadingZeros(HInvoke* invoke) {
+ GenNumberOfLeadingZeros(invoke->GetLocations(), Primitive::kPrimInt, GetAssembler());
+}
+
+void IntrinsicLocationsBuilderARM::VisitLongNumberOfLeadingZeros(HInvoke* invoke) {
+ LocationSummary* locations = new (arena_) LocationSummary(invoke,
+ LocationSummary::kNoCall,
+ kIntrinsified);
+ locations->SetInAt(0, Location::RequiresRegister());
+ locations->SetOut(Location::RequiresRegister(), Location::kOutputOverlap);
+}
+
+void IntrinsicCodeGeneratorARM::VisitLongNumberOfLeadingZeros(HInvoke* invoke) {
+ GenNumberOfLeadingZeros(invoke->GetLocations(), Primitive::kPrimLong, GetAssembler());
+}
+
static void MathAbsFP(LocationSummary* locations, bool is64bit, ArmAssembler* assembler) {
Location in = locations->InAt(0);
Location out = locations->Out();
diff --git a/compiler/optimizing/intrinsics_arm64.cc b/compiler/optimizing/intrinsics_arm64.cc
index 78ac167..2c93fea 100644
--- a/compiler/optimizing/intrinsics_arm64.cc
+++ b/compiler/optimizing/intrinsics_arm64.cc
@@ -260,6 +260,33 @@
GenReverseBytes(invoke->GetLocations(), Primitive::kPrimShort, GetVIXLAssembler());
}
+static void GenNumberOfLeadingZeros(LocationSummary* locations,
+ Primitive::Type type,
+ vixl::MacroAssembler* masm) {
+ DCHECK(type == Primitive::kPrimInt || type == Primitive::kPrimLong);
+
+ Location in = locations->InAt(0);
+ Location out = locations->Out();
+
+ __ Clz(RegisterFrom(out, type), RegisterFrom(in, type));
+}
+
+void IntrinsicLocationsBuilderARM64::VisitIntegerNumberOfLeadingZeros(HInvoke* invoke) {
+ CreateIntToIntLocations(arena_, invoke);
+}
+
+void IntrinsicCodeGeneratorARM64::VisitIntegerNumberOfLeadingZeros(HInvoke* invoke) {
+ GenNumberOfLeadingZeros(invoke->GetLocations(), Primitive::kPrimInt, GetVIXLAssembler());
+}
+
+void IntrinsicLocationsBuilderARM64::VisitLongNumberOfLeadingZeros(HInvoke* invoke) {
+ CreateIntToIntLocations(arena_, invoke);
+}
+
+void IntrinsicCodeGeneratorARM64::VisitLongNumberOfLeadingZeros(HInvoke* invoke) {
+ GenNumberOfLeadingZeros(invoke->GetLocations(), Primitive::kPrimLong, GetVIXLAssembler());
+}
+
static void GenReverse(LocationSummary* locations,
Primitive::Type type,
vixl::MacroAssembler* masm) {
diff --git a/compiler/optimizing/intrinsics_list.h b/compiler/optimizing/intrinsics_list.h
index 2c9248f..d28c5a3 100644
--- a/compiler/optimizing/intrinsics_list.h
+++ b/compiler/optimizing/intrinsics_list.h
@@ -27,8 +27,10 @@
V(FloatIntBitsToFloat, kStatic) \
V(IntegerReverse, kStatic) \
V(IntegerReverseBytes, kStatic) \
+ V(IntegerNumberOfLeadingZeros, kStatic) \
V(LongReverse, kStatic) \
V(LongReverseBytes, kStatic) \
+ V(LongNumberOfLeadingZeros, kStatic) \
V(ShortReverseBytes, kStatic) \
V(MathAbsDouble, kStatic) \
V(MathAbsFloat, kStatic) \
diff --git a/compiler/optimizing/intrinsics_x86.cc b/compiler/optimizing/intrinsics_x86.cc
index 0d6ca09..993c005 100644
--- a/compiler/optimizing/intrinsics_x86.cc
+++ b/compiler/optimizing/intrinsics_x86.cc
@@ -1756,6 +1756,8 @@
UNIMPLEMENTED_INTRINSIC(StringGetCharsNoCheck)
UNIMPLEMENTED_INTRINSIC(SystemArrayCopyChar)
UNIMPLEMENTED_INTRINSIC(ReferenceGetReferent)
+UNIMPLEMENTED_INTRINSIC(IntegerNumberOfLeadingZeros)
+UNIMPLEMENTED_INTRINSIC(LongNumberOfLeadingZeros)
#undef UNIMPLEMENTED_INTRINSIC
diff --git a/compiler/optimizing/intrinsics_x86_64.cc b/compiler/optimizing/intrinsics_x86_64.cc
index ea342e9..8ab0b77 100644
--- a/compiler/optimizing/intrinsics_x86_64.cc
+++ b/compiler/optimizing/intrinsics_x86_64.cc
@@ -1615,6 +1615,8 @@
UNIMPLEMENTED_INTRINSIC(StringGetCharsNoCheck)
UNIMPLEMENTED_INTRINSIC(SystemArrayCopyChar)
UNIMPLEMENTED_INTRINSIC(ReferenceGetReferent)
+UNIMPLEMENTED_INTRINSIC(IntegerNumberOfLeadingZeros)
+UNIMPLEMENTED_INTRINSIC(LongNumberOfLeadingZeros)
#undef UNIMPLEMENTED_INTRINSIC
diff --git a/compiler/utils/arm/assembler_thumb2_test.cc b/compiler/utils/arm/assembler_thumb2_test.cc
index 004853f..84f5cb1 100644
--- a/compiler/utils/arm/assembler_thumb2_test.cc
+++ b/compiler/utils/arm/assembler_thumb2_test.cc
@@ -1011,4 +1011,12 @@
__ GetAdjustedPosition(label.Position()));
}
+TEST_F(AssemblerThumb2Test, Clz) {
+ __ clz(arm::R0, arm::R1);
+
+ const char* expected = "clz r0, r1\n";
+
+ DriverStr(expected, "clz");
+}
+
} // namespace art