Merge "Don't show sizes with sample paths."
diff --git a/benchmark/Android.bp b/benchmark/Android.bp
index e784508..d0dfec9 100644
--- a/benchmark/Android.bp
+++ b/benchmark/Android.bp
@@ -49,7 +49,7 @@
name: "libartbenchmark-micronative-host",
host_supported: true,
device_supported: false,
- defaults: ["art_defaults", "art_debug_defaults"],
+ defaults: ["art_debug_defaults", "art_defaults" ],
srcs: [
"jni_loader.cc",
"micro-native/micro_native.cc",
diff --git a/compiler/linker/arm64/relative_patcher_arm64.cc b/compiler/linker/arm64/relative_patcher_arm64.cc
index 6d2b243..117684a 100644
--- a/compiler/linker/arm64/relative_patcher_arm64.cc
+++ b/compiler/linker/arm64/relative_patcher_arm64.cc
@@ -54,11 +54,11 @@
inline bool IsAdrpPatch(const LinkerPatch& patch) {
switch (patch.GetType()) {
- case LinkerPatch::Type::kMethodRelative:
case LinkerPatch::Type::kCall:
case LinkerPatch::Type::kCallRelative:
case LinkerPatch::Type::kBakerReadBarrierBranch:
return false;
+ case LinkerPatch::Type::kMethodRelative:
case LinkerPatch::Type::kTypeRelative:
case LinkerPatch::Type::kTypeBssEntry:
case LinkerPatch::Type::kStringRelative:
@@ -567,10 +567,10 @@
return false;
}
- // And since LinkerPatch::Type::kStringRelative is using the result of the ADRP
- // for an ADD immediate, check for that as well. We generalize a bit to include
- // ADD/ADDS/SUB/SUBS immediate that either uses the ADRP destination or stores
- // the result to a different register.
+ // And since LinkerPatch::Type::k{Method,Type,String}Relative is using the result
+ // of the ADRP for an ADD immediate, check for that as well. We generalize a bit
+ // to include ADD/ADDS/SUB/SUBS immediate that either uses the ADRP destination
+ // or stores the result to a different register.
if ((next_insn & 0x1f000000) == 0x11000000 &&
((((next_insn >> 5) ^ adrp) & 0x1f) == 0 || ((next_insn ^ adrp) & 0x1f) != 0)) {
return false;
diff --git a/compiler/optimizing/code_generator_arm_vixl.cc b/compiler/optimizing/code_generator_arm_vixl.cc
index 34821f8..1f8e1ef 100644
--- a/compiler/optimizing/code_generator_arm_vixl.cc
+++ b/compiler/optimizing/code_generator_arm_vixl.cc
@@ -2139,7 +2139,8 @@
static void GenerateLongComparesAndJumps(HCondition* cond,
vixl32::Label* true_label,
vixl32::Label* false_label,
- CodeGeneratorARMVIXL* codegen) {
+ CodeGeneratorARMVIXL* codegen,
+ bool is_far_target = true) {
LocationSummary* locations = cond->GetLocations();
Location left = locations->InAt(0);
Location right = locations->InAt(1);
@@ -2190,12 +2191,12 @@
__ Cmp(left_high, val_high);
if (if_cond == kCondNE) {
- __ B(ARMCondition(true_high_cond), true_label);
+ __ B(ARMCondition(true_high_cond), true_label, is_far_target);
} else if (if_cond == kCondEQ) {
- __ B(ARMCondition(false_high_cond), false_label);
+ __ B(ARMCondition(false_high_cond), false_label, is_far_target);
} else {
- __ B(ARMCondition(true_high_cond), true_label);
- __ B(ARMCondition(false_high_cond), false_label);
+ __ B(ARMCondition(true_high_cond), true_label, is_far_target);
+ __ B(ARMCondition(false_high_cond), false_label, is_far_target);
}
// Must be equal high, so compare the lows.
__ Cmp(left_low, val_low);
@@ -2205,19 +2206,19 @@
__ Cmp(left_high, right_high);
if (if_cond == kCondNE) {
- __ B(ARMCondition(true_high_cond), true_label);
+ __ B(ARMCondition(true_high_cond), true_label, is_far_target);
} else if (if_cond == kCondEQ) {
- __ B(ARMCondition(false_high_cond), false_label);
+ __ B(ARMCondition(false_high_cond), false_label, is_far_target);
} else {
- __ B(ARMCondition(true_high_cond), true_label);
- __ B(ARMCondition(false_high_cond), false_label);
+ __ B(ARMCondition(true_high_cond), true_label, is_far_target);
+ __ B(ARMCondition(false_high_cond), false_label, is_far_target);
}
// Must be equal high, so compare the lows.
__ Cmp(left_low, right_low);
}
// The last comparison might be unsigned.
// TODO: optimize cases where this is always true/false
- __ B(final_condition, true_label);
+ __ B(final_condition, true_label, is_far_target);
}
static void GenerateConditionLong(HCondition* cond, CodeGeneratorARMVIXL* codegen) {
@@ -2292,7 +2293,7 @@
vixl32::Label* const final_label = codegen->GetFinalLabel(cond, &done_label);
vixl32::Label true_label, false_label;
- GenerateLongComparesAndJumps(cond, &true_label, &false_label, codegen);
+ GenerateLongComparesAndJumps(cond, &true_label, &false_label, codegen, /* is_far_target */ false);
// False case: result = 0.
__ Bind(&false_label);
@@ -2957,7 +2958,8 @@
void InstructionCodeGeneratorARMVIXL::GenerateCompareTestAndBranch(HCondition* condition,
vixl32::Label* true_target_in,
- vixl32::Label* false_target_in) {
+ vixl32::Label* false_target_in,
+ bool is_far_target) {
if (CanGenerateTest(condition, codegen_->GetAssembler())) {
vixl32::Label* non_fallthrough_target;
bool invert;
@@ -2973,7 +2975,7 @@
const auto cond = GenerateTest(condition, invert, codegen_);
- __ B(cond.first, non_fallthrough_target);
+ __ B(cond.first, non_fallthrough_target, is_far_target);
if (false_target_in != nullptr && false_target_in != non_fallthrough_target) {
__ B(false_target_in);
@@ -2989,7 +2991,7 @@
vixl32::Label* false_target = (false_target_in == nullptr) ? &fallthrough : false_target_in;
DCHECK_EQ(condition->InputAt(0)->GetType(), Primitive::kPrimLong);
- GenerateLongComparesAndJumps(condition, true_target, false_target, codegen_);
+ GenerateLongComparesAndJumps(condition, true_target, false_target, codegen_, is_far_target);
if (false_target != &fallthrough) {
__ B(false_target);
@@ -3057,7 +3059,7 @@
// the HCondition, generate the comparison directly.
Primitive::Type type = condition->InputAt(0)->GetType();
if (type == Primitive::kPrimLong || Primitive::IsFloatingPointType(type)) {
- GenerateCompareTestAndBranch(condition, true_target, false_target);
+ GenerateCompareTestAndBranch(condition, true_target, false_target, far_target);
return;
}
@@ -3076,14 +3078,14 @@
if (right.IsImmediate() && right.GetImmediate() == 0 && (arm_cond.Is(ne) || arm_cond.Is(eq))) {
if (arm_cond.Is(eq)) {
- __ CompareAndBranchIfZero(left, non_fallthrough_target);
+ __ CompareAndBranchIfZero(left, non_fallthrough_target, far_target);
} else {
DCHECK(arm_cond.Is(ne));
- __ CompareAndBranchIfNonZero(left, non_fallthrough_target);
+ __ CompareAndBranchIfNonZero(left, non_fallthrough_target, far_target);
}
} else {
__ Cmp(left, right);
- __ B(arm_cond, non_fallthrough_target);
+ __ B(arm_cond, non_fallthrough_target, far_target);
}
}
diff --git a/compiler/optimizing/code_generator_arm_vixl.h b/compiler/optimizing/code_generator_arm_vixl.h
index 91f7524..91e9a3ed 100644
--- a/compiler/optimizing/code_generator_arm_vixl.h
+++ b/compiler/optimizing/code_generator_arm_vixl.h
@@ -400,7 +400,8 @@
bool far_target = true);
void GenerateCompareTestAndBranch(HCondition* condition,
vixl::aarch32::Label* true_target,
- vixl::aarch32::Label* false_target);
+ vixl::aarch32::Label* false_target,
+ bool is_far_target = true);
void DivRemOneOrMinusOne(HBinaryOperation* instruction);
void DivRemByPowerOfTwo(HBinaryOperation* instruction);
void GenerateDivRemWithAnyConstant(HBinaryOperation* instruction);
diff --git a/compiler/optimizing/intrinsics_arm.cc b/compiler/optimizing/intrinsics_arm.cc
index e8a62aa..9803c9a 100644
--- a/compiler/optimizing/intrinsics_arm.cc
+++ b/compiler/optimizing/intrinsics_arm.cc
@@ -2758,12 +2758,15 @@
int32_t offset = Thread::InterruptedOffset<kArmPointerSize>().Int32Value();
__ LoadFromOffset(kLoadWord, out, TR, offset);
Label done;
- __ CompareAndBranchIfZero(out, &done);
+ Label* const final_label = codegen_->GetFinalLabel(invoke, &done);
+ __ CompareAndBranchIfZero(out, final_label);
__ dmb(ISH);
__ LoadImmediate(IP, 0);
__ StoreToOffset(kStoreWord, IP, TR, offset);
__ dmb(ISH);
- __ Bind(&done);
+ if (done.IsLinked()) {
+ __ Bind(&done);
+ }
}
UNIMPLEMENTED_INTRINSIC(ARM, MathMinDoubleDouble)
diff --git a/compiler/optimizing/intrinsics_arm_vixl.cc b/compiler/optimizing/intrinsics_arm_vixl.cc
index ce3ba52..1a33b0e 100644
--- a/compiler/optimizing/intrinsics_arm_vixl.cc
+++ b/compiler/optimizing/intrinsics_arm_vixl.cc
@@ -3127,7 +3127,7 @@
__ Add(out, in, -info.low);
__ Cmp(out, info.high - info.low + 1);
vixl32::Label allocate, done;
- __ B(hs, &allocate);
+ __ B(hs, &allocate, /* is_far_target */ false);
// If the value is within the bounds, load the j.l.Integer directly from the array.
uint32_t data_offset = mirror::Array::DataOffset(kHeapReferenceSize).Uint32Value();
uint32_t address = dchecked_integral_cast<uint32_t>(reinterpret_cast<uintptr_t>(info.cache));
@@ -3164,12 +3164,15 @@
UseScratchRegisterScope temps(assembler->GetVIXLAssembler());
vixl32::Register temp = temps.Acquire();
vixl32::Label done;
- __ CompareAndBranchIfZero(out, &done, /* far_target */ false);
+ vixl32::Label* const final_label = codegen_->GetFinalLabel(invoke, &done);
+ __ CompareAndBranchIfZero(out, final_label, /* far_target */ false);
__ Dmb(vixl32::ISH);
__ Mov(temp, 0);
assembler->StoreToOffset(kStoreWord, temp, tr, offset);
__ Dmb(vixl32::ISH);
- __ Bind(&done);
+ if (done.IsReferenced()) {
+ __ Bind(&done);
+ }
}
UNIMPLEMENTED_INTRINSIC(ARMVIXL, MathRoundDouble) // Could be done by changing rounding mode, maybe?
diff --git a/compiler/optimizing/loop_optimization.cc b/compiler/optimizing/loop_optimization.cc
index 963df5a..94787c9 100644
--- a/compiler/optimizing/loop_optimization.cc
+++ b/compiler/optimizing/loop_optimization.cc
@@ -173,6 +173,39 @@
return false;
}
+// Detect situations with same-extension narrower operands.
+// Returns true on success and sets is_unsigned accordingly.
+static bool IsNarrowerOperands(HInstruction* a,
+ HInstruction* b,
+ Primitive::Type type,
+ /*out*/ HInstruction** r,
+ /*out*/ HInstruction** s,
+ /*out*/ bool* is_unsigned) {
+ if (IsSignExtensionAndGet(a, type, r) && IsSignExtensionAndGet(b, type, s)) {
+ *is_unsigned = false;
+ return true;
+ } else if (IsZeroExtensionAndGet(a, type, r) && IsZeroExtensionAndGet(b, type, s)) {
+ *is_unsigned = true;
+ return true;
+ }
+ return false;
+}
+
+// As above, single operand.
+static bool IsNarrowerOperand(HInstruction* a,
+ Primitive::Type type,
+ /*out*/ HInstruction** r,
+ /*out*/ bool* is_unsigned) {
+ if (IsSignExtensionAndGet(a, type, r)) {
+ *is_unsigned = false;
+ return true;
+ } else if (IsZeroExtensionAndGet(a, type, r)) {
+ *is_unsigned = true;
+ return true;
+ }
+ return false;
+}
+
// Detect up to two instructions a and b, and an acccumulated constant c.
static bool IsAddConstHelper(HInstruction* instruction,
/*out*/ HInstruction** a,
@@ -756,7 +789,7 @@
return !IsUsedOutsideLoop(node->loop_info, instruction) && !instruction->DoesAnyWrite();
}
-// TODO: more operations and intrinsics, detect saturation arithmetic, etc.
+// TODO: saturation arithmetic.
bool HLoopOptimization::VectorizeUse(LoopNode* node,
HInstruction* instruction,
bool generate_code,
@@ -867,25 +900,38 @@
return true;
}
// Deal with vector restrictions.
+ HInstruction* opa = instruction->InputAt(0);
+ HInstruction* opb = instruction->InputAt(1);
+ HInstruction* r = opa;
+ bool is_unsigned = false;
if ((HasVectorRestrictions(restrictions, kNoShift)) ||
(instruction->IsShr() && HasVectorRestrictions(restrictions, kNoShr))) {
return false; // unsupported instruction
- } else if ((instruction->IsShr() || instruction->IsUShr()) &&
- HasVectorRestrictions(restrictions, kNoHiBits)) {
- return false; // hibits may impact lobits; TODO: we can do better!
+ } else if (HasVectorRestrictions(restrictions, kNoHiBits)) {
+ // Shifts right need extra care to account for higher order bits.
+ // TODO: less likely shr/unsigned and ushr/signed can by flipping signess.
+ if (instruction->IsShr() &&
+ (!IsNarrowerOperand(opa, type, &r, &is_unsigned) || is_unsigned)) {
+ return false; // reject, unless all operands are sign-extension narrower
+ } else if (instruction->IsUShr() &&
+ (!IsNarrowerOperand(opa, type, &r, &is_unsigned) || !is_unsigned)) {
+ return false; // reject, unless all operands are zero-extension narrower
+ }
}
// Accept shift operator for vectorizable/invariant operands.
// TODO: accept symbolic, albeit loop invariant shift factors.
- HInstruction* opa = instruction->InputAt(0);
- HInstruction* opb = instruction->InputAt(1);
+ DCHECK(r != nullptr);
+ if (generate_code && vector_mode_ != kVector) { // de-idiom
+ r = opa;
+ }
int64_t distance = 0;
- if (VectorizeUse(node, opa, generate_code, type, restrictions) &&
+ if (VectorizeUse(node, r, generate_code, type, restrictions) &&
IsInt64AndGet(opb, /*out*/ &distance)) {
// Restrict shift distance to packed data type width.
int64_t max_distance = Primitive::ComponentSize(type) * 8;
if (0 <= distance && distance < max_distance) {
if (generate_code) {
- GenerateVecOp(instruction, vector_map_->Get(opa), opb, type);
+ GenerateVecOp(instruction, vector_map_->Get(r), opb, type);
}
return true;
}
@@ -899,16 +945,23 @@
case Intrinsics::kMathAbsFloat:
case Intrinsics::kMathAbsDouble: {
// Deal with vector restrictions.
- if (HasVectorRestrictions(restrictions, kNoAbs) ||
- HasVectorRestrictions(restrictions, kNoHiBits)) {
- // TODO: we can do better for some hibits cases.
+ HInstruction* opa = instruction->InputAt(0);
+ HInstruction* r = opa;
+ bool is_unsigned = false;
+ if (HasVectorRestrictions(restrictions, kNoAbs)) {
return false;
+ } else if (HasVectorRestrictions(restrictions, kNoHiBits) &&
+ (!IsNarrowerOperand(opa, type, &r, &is_unsigned) || is_unsigned)) {
+ return false; // reject, unless operand is sign-extension narrower
}
// Accept ABS(x) for vectorizable operand.
- HInstruction* opa = instruction->InputAt(0);
- if (VectorizeUse(node, opa, generate_code, type, restrictions)) {
+ DCHECK(r != nullptr);
+ if (generate_code && vector_mode_ != kVector) { // de-idiom
+ r = opa;
+ }
+ if (VectorizeUse(node, r, generate_code, type, restrictions)) {
if (generate_code) {
- GenerateVecOp(instruction, vector_map_->Get(opa), nullptr, type);
+ GenerateVecOp(instruction, vector_map_->Get(r), nullptr, type);
}
return true;
}
@@ -923,18 +976,28 @@
case Intrinsics::kMathMaxFloatFloat:
case Intrinsics::kMathMaxDoubleDouble: {
// Deal with vector restrictions.
- if (HasVectorRestrictions(restrictions, kNoMinMax) ||
- HasVectorRestrictions(restrictions, kNoHiBits)) {
- // TODO: we can do better for some hibits cases.
- return false;
- }
- // Accept MIN/MAX(x, y) for vectorizable operands.
HInstruction* opa = instruction->InputAt(0);
HInstruction* opb = instruction->InputAt(1);
- if (VectorizeUse(node, opa, generate_code, type, restrictions) &&
- VectorizeUse(node, opb, generate_code, type, restrictions)) {
+ HInstruction* r = opa;
+ HInstruction* s = opb;
+ bool is_unsigned = false;
+ if (HasVectorRestrictions(restrictions, kNoMinMax)) {
+ return false;
+ } else if (HasVectorRestrictions(restrictions, kNoHiBits) &&
+ !IsNarrowerOperands(opa, opb, type, &r, &s, &is_unsigned)) {
+ return false; // reject, unless all operands are same-extension narrower
+ }
+ // Accept MIN/MAX(x, y) for vectorizable operands.
+ DCHECK(r != nullptr && s != nullptr);
+ if (generate_code && vector_mode_ != kVector) { // de-idiom
+ r = opa;
+ s = opb;
+ }
+ if (VectorizeUse(node, r, generate_code, type, restrictions) &&
+ VectorizeUse(node, s, generate_code, type, restrictions)) {
if (generate_code) {
- GenerateVecOp(instruction, vector_map_->Get(opa), vector_map_->Get(opb), type);
+ GenerateVecOp(
+ instruction, vector_map_->Get(r), vector_map_->Get(s), type, is_unsigned);
}
return true;
}
@@ -959,11 +1022,11 @@
switch (type) {
case Primitive::kPrimBoolean:
case Primitive::kPrimByte:
- *restrictions |= kNoDiv | kNoAbs;
+ *restrictions |= kNoDiv;
return TrySetVectorLength(16);
case Primitive::kPrimChar:
case Primitive::kPrimShort:
- *restrictions |= kNoDiv | kNoAbs;
+ *restrictions |= kNoDiv;
return TrySetVectorLength(8);
case Primitive::kPrimInt:
*restrictions |= kNoDiv;
@@ -1098,13 +1161,14 @@
void HLoopOptimization::GenerateVecOp(HInstruction* org,
HInstruction* opa,
HInstruction* opb,
- Primitive::Type type) {
+ Primitive::Type type,
+ bool is_unsigned) {
if (vector_mode_ == kSequential) {
- // Scalar code follows implicit integral promotion.
- if (type == Primitive::kPrimBoolean ||
- type == Primitive::kPrimByte ||
- type == Primitive::kPrimChar ||
- type == Primitive::kPrimShort) {
+ // Non-converting scalar code follows implicit integral promotion.
+ if (!org->IsTypeConversion() && (type == Primitive::kPrimBoolean ||
+ type == Primitive::kPrimByte ||
+ type == Primitive::kPrimChar ||
+ type == Primitive::kPrimShort)) {
type = Primitive::kPrimInt;
}
}
@@ -1185,7 +1249,6 @@
case Intrinsics::kMathMinLongLong:
case Intrinsics::kMathMinFloatFloat:
case Intrinsics::kMathMinDoubleDouble: {
- bool is_unsigned = false; // TODO: detect unsigned versions
vector = new (global_allocator_)
HVecMin(global_allocator_, opa, opb, type, vector_length_, is_unsigned);
break;
@@ -1194,7 +1257,6 @@
case Intrinsics::kMathMaxLongLong:
case Intrinsics::kMathMaxFloatFloat:
case Intrinsics::kMathMaxDoubleDouble: {
- bool is_unsigned = false; // TODO: detect unsigned versions
vector = new (global_allocator_)
HVecMax(global_allocator_, opa, opb, type, vector_length_, is_unsigned);
break;
@@ -1258,7 +1320,7 @@
Primitive::Type type,
uint64_t restrictions) {
// Test for top level arithmetic shift right x >> 1 or logical shift right x >>> 1
- // (note whether the sign bit in higher precision is shifted in has no effect
+ // (note whether the sign bit in wider precision is shifted in has no effect
// on the narrow precision computed by the idiom).
int64_t distance = 0;
if ((instruction->IsShr() ||
@@ -1269,6 +1331,7 @@
HInstruction* b = nullptr;
int64_t c = 0;
if (IsAddConst(instruction->InputAt(0), /*out*/ &a, /*out*/ &b, /*out*/ &c)) {
+ DCHECK(a != nullptr && b != nullptr);
// Accept c == 1 (rounded) or c == 0 (not rounded).
bool is_rounded = false;
if (c == 1) {
@@ -1280,11 +1343,7 @@
HInstruction* r = nullptr;
HInstruction* s = nullptr;
bool is_unsigned = false;
- if (IsZeroExtensionAndGet(a, type, &r) && IsZeroExtensionAndGet(b, type, &s)) {
- is_unsigned = true;
- } else if (IsSignExtensionAndGet(a, type, &r) && IsSignExtensionAndGet(b, type, &s)) {
- is_unsigned = false;
- } else {
+ if (!IsNarrowerOperands(a, b, type, &r, &s, &is_unsigned)) {
return false;
}
// Deal with vector restrictions.
@@ -1295,6 +1354,10 @@
// Accept recognized halving add for vectorizable operands. Vectorized code uses the
// shorthand idiomatic operation. Sequential code uses the original scalar expressions.
DCHECK(r != nullptr && s != nullptr);
+ if (generate_code && vector_mode_ != kVector) { // de-idiom
+ r = instruction->InputAt(0);
+ s = instruction->InputAt(1);
+ }
if (VectorizeUse(node, r, generate_code, type, restrictions) &&
VectorizeUse(node, s, generate_code, type, restrictions)) {
if (generate_code) {
@@ -1308,12 +1371,7 @@
is_unsigned,
is_rounded));
} else {
- VectorizeUse(node, instruction->InputAt(0), generate_code, type, restrictions);
- VectorizeUse(node, instruction->InputAt(1), generate_code, type, restrictions);
- GenerateVecOp(instruction,
- vector_map_->Get(instruction->InputAt(0)),
- vector_map_->Get(instruction->InputAt(1)),
- type);
+ GenerateVecOp(instruction, vector_map_->Get(r), vector_map_->Get(s), type);
}
}
return true;
diff --git a/compiler/optimizing/loop_optimization.h b/compiler/optimizing/loop_optimization.h
index 6d5978d..35298d4 100644
--- a/compiler/optimizing/loop_optimization.h
+++ b/compiler/optimizing/loop_optimization.h
@@ -137,7 +137,11 @@
HInstruction* opa,
HInstruction* opb,
Primitive::Type type);
- void GenerateVecOp(HInstruction* org, HInstruction* opa, HInstruction* opb, Primitive::Type type);
+ void GenerateVecOp(HInstruction* org,
+ HInstruction* opa,
+ HInstruction* opb,
+ Primitive::Type type,
+ bool is_unsigned = false);
// Vectorization idioms.
bool VectorizeHalvingAddIdiom(LoopNode* node,
diff --git a/compiler/optimizing/nodes.cc b/compiler/optimizing/nodes.cc
index bde7f2c..6899910 100644
--- a/compiler/optimizing/nodes.cc
+++ b/compiler/optimizing/nodes.cc
@@ -2642,7 +2642,7 @@
case HInvokeStaticOrDirect::MethodLoadKind::kBootImageLinkTimePcRelative:
return os << "BootImageLinkTimePcRelative";
case HInvokeStaticOrDirect::MethodLoadKind::kDirectAddress:
- return os << "Direct";
+ return os << "DirectAddress";
case HInvokeStaticOrDirect::MethodLoadKind::kDexCachePcRelative:
return os << "DexCachePcRelative";
case HInvokeStaticOrDirect::MethodLoadKind::kDexCacheViaMethod:
diff --git a/dexoptanalyzer/Android.bp b/dexoptanalyzer/Android.bp
index cf4c99e..da6663d 100644
--- a/dexoptanalyzer/Android.bp
+++ b/dexoptanalyzer/Android.bp
@@ -48,8 +48,8 @@
art_cc_binary {
name: "dexoptanalyzerd",
defaults: [
- "dexoptanalyzer-defaults",
"art_debug_defaults",
+ "dexoptanalyzer-defaults",
],
shared_libs: [
"libartd",
diff --git a/disassembler/Android.bp b/disassembler/Android.bp
index 8dfada2..086b8c7 100644
--- a/disassembler/Android.bp
+++ b/disassembler/Android.bp
@@ -47,8 +47,8 @@
art_cc_library {
name: "libartd-disassembler",
defaults: [
- "libart-disassembler-defaults",
"art_debug_defaults",
+ "libart-disassembler-defaults",
],
shared_libs: [
// For disassembler_arm*.
diff --git a/imgdiag/Android.bp b/imgdiag/Android.bp
index eaeb78e..9459bb5 100644
--- a/imgdiag/Android.bp
+++ b/imgdiag/Android.bp
@@ -64,8 +64,8 @@
art_cc_binary {
name: "imgdiagd",
defaults: [
- "imgdiag-defaults",
"art_debug_defaults",
+ "imgdiag-defaults",
],
shared_libs: [
"libartd",
diff --git a/patchoat/Android.bp b/patchoat/Android.bp
index a78f97d..d3bc2a7 100644
--- a/patchoat/Android.bp
+++ b/patchoat/Android.bp
@@ -40,8 +40,8 @@
art_cc_binary {
name: "patchoatd",
defaults: [
- "patchoat-defaults",
"art_debug_defaults",
+ "patchoat-defaults",
],
shared_libs: [
"libartd",
diff --git a/profman/Android.bp b/profman/Android.bp
index 2dcbaee..a327ef2 100644
--- a/profman/Android.bp
+++ b/profman/Android.bp
@@ -49,8 +49,8 @@
art_cc_binary {
name: "profmand",
defaults: [
- "profman-defaults",
"art_debug_defaults",
+ "profman-defaults",
],
shared_libs: [
"libartd",
diff --git a/runtime/base/logging.cc b/runtime/base/logging.cc
index 55b4306..553928d 100644
--- a/runtime/base/logging.cc
+++ b/runtime/base/logging.cc
@@ -112,7 +112,7 @@
if (priority == ANDROID_LOG_FATAL) {
// Allocate buffer for snprintf(buf, buf_size, "%s:%u] %s", file, line, message) below.
// If allocation fails, fall back to printing only the message.
- buf_size = strlen(file) + 1 /* ':' */ + std::numeric_limits<typeof(line)>::max_digits10 +
+ buf_size = strlen(file) + 1 /* ':' */ + std::numeric_limits<decltype(line)>::max_digits10 +
2 /* "] " */ + strlen(message) + 1 /* terminating 0 */;
buf = reinterpret_cast<char*>(malloc(buf_size));
}
diff --git a/runtime/base/safe_copy_test.cc b/runtime/base/safe_copy_test.cc
index 987895e..a9ec952 100644
--- a/runtime/base/safe_copy_test.cc
+++ b/runtime/base/safe_copy_test.cc
@@ -23,80 +23,86 @@
#include <sys/mman.h>
#include <sys/user.h>
+#include "globals.h"
+
namespace art {
#if defined(__linux__)
TEST(SafeCopyTest, smoke) {
+ DCHECK_EQ(kPageSize, static_cast<decltype(kPageSize)>(PAGE_SIZE));
+
// Map four pages, mark the second one as PROT_NONE, unmap the last one.
- void* map = mmap(nullptr, PAGE_SIZE * 4, PROT_READ | PROT_WRITE,
+ void* map = mmap(nullptr, kPageSize * 4, PROT_READ | PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
ASSERT_NE(MAP_FAILED, map);
char* page1 = static_cast<char*>(map);
- char* page2 = page1 + PAGE_SIZE;
- char* page3 = page2 + PAGE_SIZE;
- char* page4 = page3 + PAGE_SIZE;
- ASSERT_EQ(0, mprotect(page1 + PAGE_SIZE, PAGE_SIZE, PROT_NONE));
- ASSERT_EQ(0, munmap(page4, PAGE_SIZE));
+ char* page2 = page1 + kPageSize;
+ char* page3 = page2 + kPageSize;
+ char* page4 = page3 + kPageSize;
+ ASSERT_EQ(0, mprotect(page1 + kPageSize, kPageSize, PROT_NONE));
+ ASSERT_EQ(0, munmap(page4, kPageSize));
page1[0] = 'a';
- page1[PAGE_SIZE - 1] = 'z';
+ page1[kPageSize - 1] = 'z';
page3[0] = 'b';
- page3[PAGE_SIZE - 1] = 'y';
+ page3[kPageSize - 1] = 'y';
- char buf[PAGE_SIZE];
+ char buf[kPageSize];
// Completely valid read.
memset(buf, 0xCC, sizeof(buf));
- EXPECT_EQ(static_cast<ssize_t>(PAGE_SIZE), SafeCopy(buf, page1, PAGE_SIZE)) << strerror(errno);
- EXPECT_EQ(0, memcmp(buf, page1, PAGE_SIZE));
+ EXPECT_EQ(static_cast<ssize_t>(kPageSize), SafeCopy(buf, page1, kPageSize)) << strerror(errno);
+ EXPECT_EQ(0, memcmp(buf, page1, kPageSize));
// Reading into a guard page.
memset(buf, 0xCC, sizeof(buf));
- EXPECT_EQ(static_cast<ssize_t>(PAGE_SIZE - 1), SafeCopy(buf, page1 + 1, PAGE_SIZE));
- EXPECT_EQ(0, memcmp(buf, page1 + 1, PAGE_SIZE - 1));
+ EXPECT_EQ(static_cast<ssize_t>(kPageSize - 1), SafeCopy(buf, page1 + 1, kPageSize));
+ EXPECT_EQ(0, memcmp(buf, page1 + 1, kPageSize - 1));
// Reading from a guard page into a real page.
memset(buf, 0xCC, sizeof(buf));
- EXPECT_EQ(0, SafeCopy(buf, page2 + PAGE_SIZE - 1, PAGE_SIZE));
+ EXPECT_EQ(0, SafeCopy(buf, page2 + kPageSize - 1, kPageSize));
// Reading off of the end of a mapping.
memset(buf, 0xCC, sizeof(buf));
- EXPECT_EQ(static_cast<ssize_t>(PAGE_SIZE), SafeCopy(buf, page3, PAGE_SIZE * 2));
- EXPECT_EQ(0, memcmp(buf, page3, PAGE_SIZE));
+ EXPECT_EQ(static_cast<ssize_t>(kPageSize), SafeCopy(buf, page3, kPageSize * 2));
+ EXPECT_EQ(0, memcmp(buf, page3, kPageSize));
// Completely invalid.
- EXPECT_EQ(0, SafeCopy(buf, page1 + PAGE_SIZE, PAGE_SIZE));
+ EXPECT_EQ(0, SafeCopy(buf, page1 + kPageSize, kPageSize));
// Clean up.
- ASSERT_EQ(0, munmap(map, PAGE_SIZE * 3));
+ ASSERT_EQ(0, munmap(map, kPageSize * 3));
}
TEST(SafeCopyTest, alignment) {
+ DCHECK_EQ(kPageSize, static_cast<decltype(kPageSize)>(PAGE_SIZE));
+
// Copy the middle of a mapping to the end of another one.
- void* src_map = mmap(nullptr, PAGE_SIZE * 3, PROT_READ | PROT_WRITE,
+ void* src_map = mmap(nullptr, kPageSize * 3, PROT_READ | PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
ASSERT_NE(MAP_FAILED, src_map);
// Add a guard page to make sure we don't write past the end of the mapping.
- void* dst_map = mmap(nullptr, PAGE_SIZE * 4, PROT_READ | PROT_WRITE,
+ void* dst_map = mmap(nullptr, kPageSize * 4, PROT_READ | PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
ASSERT_NE(MAP_FAILED, dst_map);
char* src = static_cast<char*>(src_map);
char* dst = static_cast<char*>(dst_map);
- ASSERT_EQ(0, mprotect(dst + 3 * PAGE_SIZE, PAGE_SIZE, PROT_NONE));
+ ASSERT_EQ(0, mprotect(dst + 3 * kPageSize, kPageSize, PROT_NONE));
src[512] = 'a';
- src[PAGE_SIZE * 3 - 512 - 1] = 'z';
+ src[kPageSize * 3 - 512 - 1] = 'z';
- EXPECT_EQ(static_cast<ssize_t>(PAGE_SIZE * 3 - 1024),
- SafeCopy(dst + 1024, src + 512, PAGE_SIZE * 3 - 1024));
- EXPECT_EQ(0, memcmp(dst + 1024, src + 512, PAGE_SIZE * 3 - 1024));
+ EXPECT_EQ(static_cast<ssize_t>(kPageSize * 3 - 1024),
+ SafeCopy(dst + 1024, src + 512, kPageSize * 3 - 1024));
+ EXPECT_EQ(0, memcmp(dst + 1024, src + 512, kPageSize * 3 - 1024));
- ASSERT_EQ(0, munmap(src_map, PAGE_SIZE * 3));
- ASSERT_EQ(0, munmap(dst_map, PAGE_SIZE * 4));
+ ASSERT_EQ(0, munmap(src_map, kPageSize * 3));
+ ASSERT_EQ(0, munmap(dst_map, kPageSize * 4));
}
#endif // defined(__linux__)
diff --git a/runtime/entrypoints/quick/quick_entrypoints_list.h b/runtime/entrypoints/quick/quick_entrypoints_list.h
index e2d45ac..74e7c18 100644
--- a/runtime/entrypoints/quick/quick_entrypoints_list.h
+++ b/runtime/entrypoints/quick/quick_entrypoints_list.h
@@ -145,22 +145,22 @@
V(A64Load, int64_t, volatile const int64_t *) \
V(A64Store, void, volatile int64_t *, int64_t) \
\
- V(NewEmptyString, void) \
- V(NewStringFromBytes_B, void) \
- V(NewStringFromBytes_BI, void) \
- V(NewStringFromBytes_BII, void) \
- V(NewStringFromBytes_BIII, void) \
- V(NewStringFromBytes_BIIString, void) \
- V(NewStringFromBytes_BString, void) \
- V(NewStringFromBytes_BIICharset, void) \
- V(NewStringFromBytes_BCharset, void) \
- V(NewStringFromChars_C, void) \
- V(NewStringFromChars_CII, void) \
- V(NewStringFromChars_IIC, void) \
- V(NewStringFromCodePoints, void) \
- V(NewStringFromString, void) \
- V(NewStringFromStringBuffer, void) \
- V(NewStringFromStringBuilder, void) \
+ V(NewEmptyString, void, void) \
+ V(NewStringFromBytes_B, void, void) \
+ V(NewStringFromBytes_BI, void, void) \
+ V(NewStringFromBytes_BII, void, void) \
+ V(NewStringFromBytes_BIII, void, void) \
+ V(NewStringFromBytes_BIIString, void, void) \
+ V(NewStringFromBytes_BString, void, void) \
+ V(NewStringFromBytes_BIICharset, void, void) \
+ V(NewStringFromBytes_BCharset, void, void) \
+ V(NewStringFromChars_C, void, void) \
+ V(NewStringFromChars_CII, void, void) \
+ V(NewStringFromChars_IIC, void, void) \
+ V(NewStringFromCodePoints, void, void) \
+ V(NewStringFromString, void, void) \
+ V(NewStringFromStringBuffer, void, void) \
+ V(NewStringFromStringBuilder, void, void) \
\
V(ReadBarrierJni, void, mirror::CompressedReference<mirror::Object>*, Thread*) \
V(ReadBarrierMarkReg00, mirror::Object*, mirror::Object*) \
diff --git a/runtime/gc/heap.cc b/runtime/gc/heap.cc
index ef4fa28..df097a0 100644
--- a/runtime/gc/heap.cc
+++ b/runtime/gc/heap.cc
@@ -4004,7 +4004,8 @@
native_blocking_gcs_finished_++;
native_blocking_gc_cond_->Broadcast(self);
}
- } else if (new_value > NativeAllocationGcWatermark() && !IsGCRequestPending()) {
+ } else if (new_value > NativeAllocationGcWatermark() * HeapGrowthMultiplier() &&
+ !IsGCRequestPending()) {
// Trigger another GC because there have been enough native bytes
// allocated since the last GC.
if (IsGcConcurrent()) {
diff --git a/runtime/interpreter/interpreter_common.h b/runtime/interpreter/interpreter_common.h
index 2589ad0..fdc0505 100644
--- a/runtime/interpreter/interpreter_common.h
+++ b/runtime/interpreter/interpreter_common.h
@@ -140,12 +140,6 @@
result->SetJ(0);
return false;
} else {
- if (called_method->IsIntrinsic()) {
- if (MterpHandleIntrinsic(&shadow_frame, called_method, inst, inst_data,
- shadow_frame.GetResultRegister())) {
- return !self->IsExceptionPending();
- }
- }
jit::Jit* jit = Runtime::Current()->GetJit();
if (jit != nullptr) {
if (type == kVirtual) {
@@ -153,6 +147,12 @@
}
jit->AddSamples(self, sf_method, 1, /*with_backedges*/false);
}
+ if (called_method->IsIntrinsic()) {
+ if (MterpHandleIntrinsic(&shadow_frame, called_method, inst, inst_data,
+ shadow_frame.GetResultRegister())) {
+ return !self->IsExceptionPending();
+ }
+ }
return DoCall<false, false>(called_method, self, shadow_frame, inst, inst_data, result);
}
}
diff --git a/runtime/interpreter/mterp/mterp.cc b/runtime/interpreter/mterp/mterp.cc
index a53040c..5f94d04 100644
--- a/runtime/interpreter/mterp/mterp.cc
+++ b/runtime/interpreter/mterp/mterp.cc
@@ -276,6 +276,12 @@
vtable_idx, kRuntimePointerSize);
if ((called_method != nullptr) && called_method->IsIntrinsic()) {
if (MterpHandleIntrinsic(shadow_frame, called_method, inst, inst_data, result_register)) {
+ jit::Jit* jit = Runtime::Current()->GetJit();
+ if (jit != nullptr) {
+ jit->InvokeVirtualOrInterface(
+ receiver, shadow_frame->GetMethod(), shadow_frame->GetDexPC(), called_method);
+ jit->AddSamples(self, shadow_frame->GetMethod(), 1, /*with_backedges*/false);
+ }
return !self->IsExceptionPending();
}
}
diff --git a/runtime/jit/jit_code_cache.cc b/runtime/jit/jit_code_cache.cc
index 5232252..5ce5447 100644
--- a/runtime/jit/jit_code_cache.cc
+++ b/runtime/jit/jit_code_cache.cc
@@ -1269,6 +1269,7 @@
std::vector<ProfileMethodInfo>& methods) {
ScopedTrace trace(__FUNCTION__);
MutexLock mu(Thread::Current(), lock_);
+ uint16_t jit_compile_threshold = Runtime::Current()->GetJITOptions()->GetCompileThreshold();
for (const ProfilingInfo* info : profiling_infos_) {
ArtMethod* method = info->GetMethod();
const DexFile* dex_file = method->GetDexFile();
@@ -1277,6 +1278,16 @@
continue;
}
std::vector<ProfileMethodInfo::ProfileInlineCache> inline_caches;
+
+ // If the method didn't reach the compilation threshold don't save the inline caches.
+ // They might be incomplete and cause unnecessary deoptimizations.
+ // If the inline cache is empty the compiler will generate a regular invoke virtual/interface.
+ if (method->GetCounter() < jit_compile_threshold) {
+ methods.emplace_back(/*ProfileMethodInfo*/
+ dex_file, method->GetDexMethodIndex(), inline_caches);
+ continue;
+ }
+
for (size_t i = 0; i < info->number_of_inline_caches_; ++i) {
std::vector<ProfileMethodInfo::ProfileClassReference> profile_classes;
const InlineCache& cache = info->cache_[i];
diff --git a/runtime/mirror/object_test.cc b/runtime/mirror/object_test.cc
index d7527d5..6230ae9 100644
--- a/runtime/mirror/object_test.cc
+++ b/runtime/mirror/object_test.cc
@@ -582,7 +582,7 @@
// Primitive types are only assignable to themselves
const char* prims = "ZBCSIJFD";
- Class* prim_types[strlen(prims)];
+ std::vector<Class*> prim_types(strlen(prims));
for (size_t i = 0; i < strlen(prims); i++) {
prim_types[i] = class_linker_->FindPrimitiveClass(prims[i]);
}
diff --git a/runtime/runtime.h b/runtime/runtime.h
index 483d255..4e143e0 100644
--- a/runtime/runtime.h
+++ b/runtime/runtime.h
@@ -681,6 +681,14 @@
deoptimization_counts_[static_cast<size_t>(kind)]++;
}
+ uint32_t GetNumberOfDeoptimizations() const {
+ uint32_t result = 0;
+ for (size_t i = 0; i <= static_cast<size_t>(DeoptimizationKind::kLast); ++i) {
+ result += deoptimization_counts_[i];
+ }
+ return result;
+ }
+
private:
static void InitPlatformSignalHandlers();
diff --git a/test/623-checker-loop-regressions/src/Main.java b/test/623-checker-loop-regressions/src/Main.java
index 520e7c3..3a2145bf 100644
--- a/test/623-checker-loop-regressions/src/Main.java
+++ b/test/623-checker-loop-regressions/src/Main.java
@@ -351,6 +351,35 @@
}
}
+ /// CHECK-START: void Main.typeConv(byte[], byte[]) loop_optimization (before)
+ /// CHECK-DAG: <<One:i\d+>> IntConstant 1 loop:none
+ /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none
+ /// CHECK-DAG: <<Get:b\d+>> ArrayGet [{{l\d+}},<<Phi>>] loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<Add:i\d+>> Add [<<Get>>,<<One>>] loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<Cnv:b\d+>> TypeConversion [<<Add>>] loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: ArraySet [{{l\d+}},<<Phi>>,<<Cnv>>] loop:<<Loop>> outer_loop:none
+ //
+ /// CHECK-START-ARM64: void Main.typeConv(byte[], byte[]) loop_optimization (after)
+ /// CHECK-DAG: <<One:i\d+>> IntConstant 1 loop:none
+ /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<One>>] loop:none
+ /// CHECK-DAG: <<Phi1:i\d+>> Phi loop:<<Loop1:B\d+>> outer_loop:none
+ /// CHECK-DAG: <<Load:d\d+>> VecLoad [{{l\d+}},<<Phi1>>] loop:<<Loop1>> outer_loop:none
+ /// CHECK-DAG: <<Vadd:d\d+>> VecAdd [<<Load>>,<<Repl>>] loop:<<Loop1>> outer_loop:none
+ /// CHECK-DAG: VecStore [{{l\d+}},<<Phi1>>,<<Vadd>>] loop:<<Loop1>> outer_loop:none
+ /// CHECK-DAG: <<Phi2:i\d+>> Phi loop:<<Loop2:B\d+>> outer_loop:none
+ /// CHECK-DAG: <<Get:b\d+>> ArrayGet [{{l\d+}},<<Phi2>>] loop:<<Loop2>> outer_loop:none
+ /// CHECK-DAG: <<Add:i\d+>> Add [<<Get>>,<<One>>] loop:<<Loop2>> outer_loop:none
+ /// CHECK-DAG: <<Cnv:b\d+>> TypeConversion [<<Add>>] loop:<<Loop2>> outer_loop:none
+ /// CHECK-DAG: ArraySet [{{l\d+}},<<Phi2>>,<<Cnv>>] loop:<<Loop2>> outer_loop:none
+ //
+ // Scalar code in cleanup loop uses correct byte type on array get and type conversion.
+ private static void typeConv(byte[] a, byte[] b) {
+ int len = Math.min(a.length, b.length);
+ for (int i = 0; i < len; i++) {
+ a[i] = (byte) (b[i] + 1);
+ }
+ }
+
public static void main(String[] args) {
expectEquals(10, earlyExitFirst(-1));
for (int i = 0; i <= 10; i++) {
@@ -453,6 +482,17 @@
expectEquals(40, bt[i]);
}
+ byte[] b1 = new byte[259]; // few extra iterations
+ byte[] b2 = new byte[259];
+ for (int i = 0; i < 259; i++) {
+ b1[i] = 0;
+ b2[i] = (byte) i;
+ }
+ typeConv(b1, b2);
+ for (int i = 0; i < 259; i++) {
+ expectEquals((byte)(i + 1), b1[i]);
+ }
+
System.out.println("passed");
}
diff --git a/test/640-checker-byte-simd/src/Main.java b/test/640-checker-byte-simd/src/Main.java
index 10b20b8..21d71e8 100644
--- a/test/640-checker-byte-simd/src/Main.java
+++ b/test/640-checker-byte-simd/src/Main.java
@@ -135,8 +135,10 @@
/// CHECK-DAG: ArraySet loop:<<Loop>> outer_loop:none
//
/// CHECK-START-ARM64: void Main.sar2() loop_optimization (after)
- //
- // TODO: fill in when supported
+ /// CHECK-DAG: Phi loop:<<Loop:B\d+>> outer_loop:none
+ /// CHECK-DAG: VecLoad loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: VecShr loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: VecStore loop:<<Loop>> outer_loop:none
static void sar2() {
for (int i = 0; i < 128; i++)
a[i] >>= 2;
@@ -147,9 +149,9 @@
/// CHECK-DAG: ArrayGet loop:<<Loop>> outer_loop:none
/// CHECK-DAG: ArraySet loop:<<Loop>> outer_loop:none
//
- /// CHECK-START-ARM64: void Main.shr2() loop_optimization (after)
- //
- // TODO: fill in when supported
+ // TODO: would need signess flip.
+ /// CHECK-START: void Main.shr2() loop_optimization (after)
+ /// CHECK-NOT: VecUShr
static void shr2() {
for (int i = 0; i < 128; i++)
a[i] >>>= 2;
diff --git a/test/640-checker-char-simd/src/Main.java b/test/640-checker-char-simd/src/Main.java
index 0628b36..89d4b6b 100644
--- a/test/640-checker-char-simd/src/Main.java
+++ b/test/640-checker-char-simd/src/Main.java
@@ -134,9 +134,9 @@
/// CHECK-DAG: ArrayGet loop:<<Loop>> outer_loop:none
/// CHECK-DAG: ArraySet loop:<<Loop>> outer_loop:none
//
- /// CHECK-START-ARM64: void Main.sar2() loop_optimization (after)
- //
- // TODO: fill in when supported
+ // TODO: would need signess flip.
+ /// CHECK-START: void Main.sar2() loop_optimization (after)
+ /// CHECK-NOT: VecShr
static void sar2() {
for (int i = 0; i < 128; i++)
a[i] >>= 2;
@@ -148,8 +148,10 @@
/// CHECK-DAG: ArraySet loop:<<Loop>> outer_loop:none
//
/// CHECK-START-ARM64: void Main.shr2() loop_optimization (after)
- //
- // TODO: fill in when supported
+ /// CHECK-DAG: Phi loop:<<Loop:B\d+>> outer_loop:none
+ /// CHECK-DAG: VecLoad loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: VecUShr loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: VecStore loop:<<Loop>> outer_loop:none
static void shr2() {
for (int i = 0; i < 128; i++)
a[i] >>>= 2;
diff --git a/test/640-checker-double-simd/src/Main.java b/test/640-checker-double-simd/src/Main.java
index 0d4f87a..5709b5d 100644
--- a/test/640-checker-double-simd/src/Main.java
+++ b/test/640-checker-double-simd/src/Main.java
@@ -122,8 +122,10 @@
/// CHECK-DAG: ArraySet loop:<<Loop>> outer_loop:none
//
/// CHECK-START-ARM64: void Main.conv(long[]) loop_optimization (after)
+ /// CHECK-NOT: VecLoad
+ /// CHECK-NOT: VecStore
//
- // TODO: fill in when supported
+ // TODO: fill in when long2double is supported
static void conv(long[] b) {
for (int i = 0; i < 128; i++)
a[i] = b[i];
diff --git a/test/640-checker-int-simd/src/Main.java b/test/640-checker-int-simd/src/Main.java
index 97048eb..9ee553c 100644
--- a/test/640-checker-int-simd/src/Main.java
+++ b/test/640-checker-int-simd/src/Main.java
@@ -136,8 +136,10 @@
/// CHECK-DAG: ArraySet loop:<<Loop>> outer_loop:none
//
/// CHECK-START-ARM64: void Main.sar2() loop_optimization (after)
- //
- // TODO: fill in when supported
+ /// CHECK-DAG: Phi loop:<<Loop:B\d+>> outer_loop:none
+ /// CHECK-DAG: VecLoad loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: VecShr loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: VecStore loop:<<Loop>> outer_loop:none
static void sar2() {
for (int i = 0; i < 128; i++)
a[i] >>= 2;
@@ -149,8 +151,10 @@
/// CHECK-DAG: ArraySet loop:<<Loop>> outer_loop:none
//
/// CHECK-START-ARM64: void Main.shr2() loop_optimization (after)
- //
- // TODO: fill in when supported
+ /// CHECK-DAG: Phi loop:<<Loop:B\d+>> outer_loop:none
+ /// CHECK-DAG: VecLoad loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: VecUShr loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: VecStore loop:<<Loop>> outer_loop:none
static void shr2() {
for (int i = 0; i < 128; i++)
a[i] >>>= 2;
diff --git a/test/640-checker-long-simd/src/Main.java b/test/640-checker-long-simd/src/Main.java
index e42c716..8f6af9d 100644
--- a/test/640-checker-long-simd/src/Main.java
+++ b/test/640-checker-long-simd/src/Main.java
@@ -134,8 +134,10 @@
/// CHECK-DAG: ArraySet loop:<<Loop>> outer_loop:none
//
/// CHECK-START-ARM64: void Main.sar2() loop_optimization (after)
- //
- // TODO: fill in when supported
+ /// CHECK-DAG: Phi loop:<<Loop:B\d+>> outer_loop:none
+ /// CHECK-DAG: VecLoad loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: VecShr loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: VecStore loop:<<Loop>> outer_loop:none
static void sar2() {
for (int i = 0; i < 128; i++)
a[i] >>= 2;
@@ -147,8 +149,10 @@
/// CHECK-DAG: ArraySet loop:<<Loop>> outer_loop:none
//
/// CHECK-START-ARM64: void Main.shr2() loop_optimization (after)
- //
- // TODO: fill in when supported
+ /// CHECK-DAG: Phi loop:<<Loop:B\d+>> outer_loop:none
+ /// CHECK-DAG: VecLoad loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: VecUShr loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: VecStore loop:<<Loop>> outer_loop:none
static void shr2() {
for (int i = 0; i < 128; i++)
a[i] >>>= 2;
diff --git a/test/640-checker-short-simd/src/Main.java b/test/640-checker-short-simd/src/Main.java
index 241f8e6..f62c726 100644
--- a/test/640-checker-short-simd/src/Main.java
+++ b/test/640-checker-short-simd/src/Main.java
@@ -135,8 +135,10 @@
/// CHECK-DAG: ArraySet loop:<<Loop>> outer_loop:none
//
/// CHECK-START-ARM64: void Main.sar2() loop_optimization (after)
- //
- // TODO: fill in when supported
+ /// CHECK-DAG: Phi loop:<<Loop:B\d+>> outer_loop:none
+ /// CHECK-DAG: VecLoad loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: VecShr loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: VecStore loop:<<Loop>> outer_loop:none
static void sar2() {
for (int i = 0; i < 128; i++)
a[i] >>= 2;
@@ -147,9 +149,9 @@
/// CHECK-DAG: ArrayGet loop:<<Loop>> outer_loop:none
/// CHECK-DAG: ArraySet loop:<<Loop>> outer_loop:none
//
- /// CHECK-START-ARM64: void Main.shr2() loop_optimization (after)
- //
- // TODO: fill in when supported
+ // TODO: would need signess flip.
+ /// CHECK-START: void Main.shr2() loop_optimization (after)
+ /// CHECK-NOT: VecUShr
static void shr2() {
for (int i = 0; i < 128; i++)
a[i] >>>= 2;
diff --git a/test/645-checker-abs-simd/src/Main.java b/test/645-checker-abs-simd/src/Main.java
index 76850ab..5a63d9f 100644
--- a/test/645-checker-abs-simd/src/Main.java
+++ b/test/645-checker-abs-simd/src/Main.java
@@ -22,6 +22,67 @@
private static final int SPQUIET = 1 << 22;
private static final long DPQUIET = 1L << 51;
+ /// CHECK-START: void Main.doitByte(byte[]) loop_optimization (before)
+ /// CHECK-DAG: Phi loop:<<Loop:B\d+>> outer_loop:none
+ /// CHECK-DAG: ArrayGet loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: InvokeStaticOrDirect intrinsic:MathAbsInt loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: ArraySet loop:<<Loop>> outer_loop:none
+ //
+ /// CHECK-START-ARM64: void Main.doitByte(byte[]) loop_optimization (after)
+ /// CHECK-DAG: Phi loop:<<Loop1:B\d+>> outer_loop:none
+ /// CHECK-DAG: VecLoad loop:<<Loop1>> outer_loop:none
+ /// CHECK-DAG: VecAbs loop:<<Loop1>> outer_loop:none
+ /// CHECK-DAG: VecStore loop:<<Loop1>> outer_loop:none
+ /// CHECK-DAG: Phi loop:<<Loop2:B\d+>> outer_loop:none
+ /// CHECK-DAG: ArrayGet loop:<<Loop2>> outer_loop:none
+ /// CHECK-DAG: InvokeStaticOrDirect intrinsic:MathAbsInt loop:<<Loop2>> outer_loop:none
+ /// CHECK-DAG: ArraySet loop:<<Loop2>> outer_loop:none
+ //
+ /// CHECK-EVAL: "<<Loop1>>" != "<<Loop2>>"
+ private static void doitByte(byte[] x) {
+ for (int i = 0; i < x.length; i++) {
+ x[i] = (byte) Math.abs(x[i]);
+ }
+ }
+
+ /// CHECK-START: void Main.doitChar(char[]) loop_optimization (before)
+ /// CHECK-DAG: Phi loop:<<Loop:B\d+>> outer_loop:none
+ /// CHECK-DAG: ArrayGet loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: InvokeStaticOrDirect intrinsic:MathAbsInt loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: ArraySet loop:<<Loop>> outer_loop:none
+ //
+ /// CHECK-START: void Main.doitChar(char[]) loop_optimization (after)
+ /// CHECK-NOT: VecAbs
+ private static void doitChar(char[] x) {
+ // Basically a nop due to zero extension.
+ for (int i = 0; i < x.length; i++) {
+ x[i] = (char) Math.abs(x[i]);
+ }
+ }
+
+ /// CHECK-START: void Main.doitShort(short[]) loop_optimization (before)
+ /// CHECK-DAG: Phi loop:<<Loop:B\d+>> outer_loop:none
+ /// CHECK-DAG: ArrayGet loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: InvokeStaticOrDirect intrinsic:MathAbsInt loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: ArraySet loop:<<Loop>> outer_loop:none
+ //
+ /// CHECK-START-ARM64: void Main.doitShort(short[]) loop_optimization (after)
+ /// CHECK-DAG: Phi loop:<<Loop1:B\d+>> outer_loop:none
+ /// CHECK-DAG: VecLoad loop:<<Loop1>> outer_loop:none
+ /// CHECK-DAG: VecAbs loop:<<Loop1>> outer_loop:none
+ /// CHECK-DAG: VecStore loop:<<Loop1>> outer_loop:none
+ /// CHECK-DAG: Phi loop:<<Loop2:B\d+>> outer_loop:none
+ /// CHECK-DAG: ArrayGet loop:<<Loop2>> outer_loop:none
+ /// CHECK-DAG: InvokeStaticOrDirect intrinsic:MathAbsInt loop:<<Loop2>> outer_loop:none
+ /// CHECK-DAG: ArraySet loop:<<Loop2>> outer_loop:none
+ //
+ /// CHECK-EVAL: "<<Loop1>>" != "<<Loop2>>"
+ private static void doitShort(short[] x) {
+ for (int i = 0; i < x.length; i++) {
+ x[i] = (short) Math.abs(x[i]);
+ }
+ }
+
/// CHECK-START: void Main.doitInt(int[]) loop_optimization (before)
/// CHECK-DAG: Phi loop:<<Loop:B\d+>> outer_loop:none
/// CHECK-DAG: ArrayGet loop:<<Loop>> outer_loop:none
@@ -52,8 +113,16 @@
/// CHECK-DAG: ArraySet loop:<<Loop>> outer_loop:none
//
/// CHECK-START-ARM64: void Main.doitLong(long[]) loop_optimization (after)
+ /// CHECK-DAG: Phi loop:<<Loop1:B\d+>> outer_loop:none
+ /// CHECK-DAG: VecLoad loop:<<Loop1>> outer_loop:none
+ /// CHECK-DAG: VecAbs loop:<<Loop1>> outer_loop:none
+ /// CHECK-DAG: VecStore loop:<<Loop1>> outer_loop:none
+ /// CHECK-DAG: Phi loop:<<Loop2:B\d+>> outer_loop:none
+ /// CHECK-DAG: ArrayGet loop:<<Loop2>> outer_loop:none
+ /// CHECK-DAG: InvokeStaticOrDirect intrinsic:MathAbsLong loop:<<Loop2>> outer_loop:none
+ /// CHECK-DAG: ArraySet loop:<<Loop2>> outer_loop:none
//
- // TODO: Not supported yet.
+ /// CHECK-EVAL: "<<Loop1>>" != "<<Loop2>>"
private static void doitLong(long[] x) {
for (int i = 0; i < x.length; i++) {
x[i] = Math.abs(x[i]);
@@ -90,8 +159,16 @@
/// CHECK-DAG: ArraySet loop:<<Loop>> outer_loop:none
//
/// CHECK-START-ARM64: void Main.doitDouble(double[]) loop_optimization (after)
+ /// CHECK-DAG: Phi loop:<<Loop1:B\d+>> outer_loop:none
+ /// CHECK-DAG: VecLoad loop:<<Loop1>> outer_loop:none
+ /// CHECK-DAG: VecAbs loop:<<Loop1>> outer_loop:none
+ /// CHECK-DAG: VecStore loop:<<Loop1>> outer_loop:none
+ /// CHECK-DAG: Phi loop:<<Loop2:B\d+>> outer_loop:none
+ /// CHECK-DAG: ArrayGet loop:<<Loop2>> outer_loop:none
+ /// CHECK-DAG: InvokeStaticOrDirect intrinsic:MathAbsDouble loop:<<Loop2>> outer_loop:none
+ /// CHECK-DAG: ArraySet loop:<<Loop2>> outer_loop:none
//
- // TODO: Not supported yet.
+ /// CHECK-EVAL: "<<Loop1>>" != "<<Loop2>>"
private static void doitDouble(double[] x) {
for (int i = 0; i < x.length; i++) {
x[i] = Math.abs(x[i]);
@@ -99,6 +176,31 @@
}
public static void main(String[] args) {
+ // Bytes, chars, shorts.
+ byte[] xb = new byte[256];
+ for (int i = 0; i < 256; i++) {
+ xb[i] = (byte) i;
+ }
+ doitByte(xb);
+ for (int i = 0; i < 256; i++) {
+ expectEquals32((byte) Math.abs((byte) i), xb[i]);
+ }
+ char[] xc = new char[1024 * 64];
+ for (int i = 0; i < 1024 * 64; i++) {
+ xc[i] = (char) i;
+ }
+ doitChar(xc);
+ for (int i = 0; i < 1024 *64; i++) {
+ expectEquals32((char) Math.abs((char) i), xc[i]);
+ }
+ short[] xs = new short[1024 * 64];
+ for (int i = 0; i < 1024 * 64; i++) {
+ xs[i] = (short) i;
+ }
+ doitShort(xs);
+ for (int i = 0; i < 1024 * 64; i++) {
+ expectEquals32((short) Math.abs((short) i), xs[i]);
+ }
// Set up minint32, maxint32 and some others.
int[] xi = new int[8];
xi[0] = 0x80000000;
diff --git a/test/651-checker-byte-simd-minmax/src/Main.java b/test/651-checker-byte-simd-minmax/src/Main.java
index 8211ace..fe45807 100644
--- a/test/651-checker-byte-simd-minmax/src/Main.java
+++ b/test/651-checker-byte-simd-minmax/src/Main.java
@@ -27,9 +27,12 @@
/// CHECK-DAG: <<Cnv:b\d+>> TypeConversion [<<Min>>] loop:<<Loop>> outer_loop:none
/// CHECK-DAG: ArraySet [{{l\d+}},<<Phi>>,<<Cnv>>] loop:<<Loop>> outer_loop:none
//
- // TODO: narrow type vectorization.
- /// CHECK-START: void Main.doitMin(byte[], byte[], byte[]) loop_optimization (after)
- /// CHECK-NOT: VecMin
+ /// CHECK-START-ARM64: void Main.doitMin(byte[], byte[], byte[]) loop_optimization (after)
+ /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none
+ /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<Min:d\d+>> VecMin [<<Get1>>,<<Get2>>] unsigned:false loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: VecStore [{{l\d+}},<<Phi>>,<<Min>>] loop:<<Loop>> outer_loop:none
private static void doitMin(byte[] x, byte[] y, byte[] z) {
int min = Math.min(x.length, Math.min(y.length, z.length));
for (int i = 0; i < min; i++) {
@@ -37,6 +40,30 @@
}
}
+ /// CHECK-START-ARM64: void Main.doitMinUnsigned(byte[], byte[], byte[]) loop_optimization (before)
+ /// CHECK-DAG: <<I255:i\d+>> IntConstant 255 loop:none
+ /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none
+ /// CHECK-DAG: <<Get1:b\d+>> ArrayGet loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<Get2:b\d+>> ArrayGet loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<And1:i\d+>> And [<<Get1>>,<<I255>>] loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<And2:i\d+>> And [<<Get2>>,<<I255>>] loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<Min:i\d+>> InvokeStaticOrDirect [<<And1>>,<<And2>>] intrinsic:MathMinIntInt loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<Cnv:b\d+>> TypeConversion [<<Min>>] loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: ArraySet [{{l\d+}},<<Phi>>,<<Cnv>>] loop:<<Loop>> outer_loop:none
+ //
+ /// CHECK-START-ARM64: void Main.doitMinUnsigned(byte[], byte[], byte[]) loop_optimization (after)
+ /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none
+ /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<Min:d\d+>> VecMin [<<Get1>>,<<Get2>>] unsigned:true loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: VecStore [{{l\d+}},<<Phi>>,<<Min>>] loop:<<Loop>> outer_loop:none
+ private static void doitMinUnsigned(byte[] x, byte[] y, byte[] z) {
+ int min = Math.min(x.length, Math.min(y.length, z.length));
+ for (int i = 0; i < min; i++) {
+ x[i] = (byte) Math.min(y[i] & 0xff, z[i] & 0xff);
+ }
+ }
+
/// CHECK-START: void Main.doitMax(byte[], byte[], byte[]) loop_optimization (before)
/// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none
/// CHECK-DAG: <<Get1:b\d+>> ArrayGet loop:<<Loop>> outer_loop:none
@@ -45,9 +72,12 @@
/// CHECK-DAG: <<Cnv:b\d+>> TypeConversion [<<Max>>] loop:<<Loop>> outer_loop:none
/// CHECK-DAG: ArraySet [{{l\d+}},<<Phi>>,<<Cnv>>] loop:<<Loop>> outer_loop:none
//
- // TODO: narrow type vectorization.
- /// CHECK-START: void Main.doitMax(byte[], byte[], byte[]) loop_optimization (after)
- /// CHECK-NOT: VecMax
+ /// CHECK-START-ARM64: void Main.doitMax(byte[], byte[], byte[]) loop_optimization (after)
+ /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none
+ /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<Max:d\d+>> VecMax [<<Get1>>,<<Get2>>] unsigned:false loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: VecStore [{{l\d+}},<<Phi>>,<<Max>>] loop:<<Loop>> outer_loop:none
private static void doitMax(byte[] x, byte[] y, byte[] z) {
int min = Math.min(x.length, Math.min(y.length, z.length));
for (int i = 0; i < min; i++) {
@@ -55,6 +85,30 @@
}
}
+ /// CHECK-START-ARM64: void Main.doitMaxUnsigned(byte[], byte[], byte[]) loop_optimization (before)
+ /// CHECK-DAG: <<I255:i\d+>> IntConstant 255 loop:none
+ /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none
+ /// CHECK-DAG: <<Get1:b\d+>> ArrayGet loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<Get2:b\d+>> ArrayGet loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<And1:i\d+>> And [<<Get1>>,<<I255>>] loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<And2:i\d+>> And [<<Get2>>,<<I255>>] loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<Max:i\d+>> InvokeStaticOrDirect [<<And1>>,<<And2>>] intrinsic:MathMaxIntInt loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<Cnv:b\d+>> TypeConversion [<<Max>>] loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: ArraySet [{{l\d+}},<<Phi>>,<<Cnv>>] loop:<<Loop>> outer_loop:none
+ //
+ /// CHECK-START-ARM64: void Main.doitMaxUnsigned(byte[], byte[], byte[]) loop_optimization (after)
+ /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none
+ /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<Max:d\d+>> VecMax [<<Get1>>,<<Get2>>] unsigned:true loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: VecStore [{{l\d+}},<<Phi>>,<<Max>>] loop:<<Loop>> outer_loop:none
+ private static void doitMaxUnsigned(byte[] x, byte[] y, byte[] z) {
+ int min = Math.min(x.length, Math.min(y.length, z.length));
+ for (int i = 0; i < min; i++) {
+ x[i] = (byte) Math.max(y[i] & 0xff, z[i] & 0xff);
+ }
+ }
+
public static void main(String[] args) {
// Initialize cross-values for all possible values.
int total = 256 * 256;
@@ -77,11 +131,21 @@
byte expected = (byte) Math.min(y[i], z[i]);
expectEquals(expected, x[i]);
}
+ doitMinUnsigned(x, y, z);
+ for (int i = 0; i < total; i++) {
+ byte expected = (byte) Math.min(y[i] & 0xff, z[i] & 0xff);
+ expectEquals(expected, x[i]);
+ }
doitMax(x, y, z);
for (int i = 0; i < total; i++) {
byte expected = (byte) Math.max(y[i], z[i]);
expectEquals(expected, x[i]);
}
+ doitMaxUnsigned(x, y, z);
+ for (int i = 0; i < total; i++) {
+ byte expected = (byte) Math.max(y[i] & 0xff, z[i] & 0xff);
+ expectEquals(expected, x[i]);
+ }
System.out.println("passed");
}
diff --git a/test/651-checker-char-simd-minmax/src/Main.java b/test/651-checker-char-simd-minmax/src/Main.java
index 5ce7b94..e2998da 100644
--- a/test/651-checker-char-simd-minmax/src/Main.java
+++ b/test/651-checker-char-simd-minmax/src/Main.java
@@ -27,9 +27,12 @@
/// CHECK-DAG: <<Cnv:c\d+>> TypeConversion [<<Min>>] loop:<<Loop>> outer_loop:none
/// CHECK-DAG: ArraySet [{{l\d+}},<<Phi>>,<<Cnv>>] loop:<<Loop>> outer_loop:none
//
- // TODO: narrow type vectorization.
- /// CHECK-START: void Main.doitMin(char[], char[], char[]) loop_optimization (after)
- /// CHECK-NOT: VecMin
+ /// CHECK-START-ARM64: void Main.doitMin(char[], char[], char[]) loop_optimization (after)
+ /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none
+ /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<Min:d\d+>> VecMin [<<Get1>>,<<Get2>>] unsigned:true loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: VecStore [{{l\d+}},<<Phi>>,<<Min>>] loop:<<Loop>> outer_loop:none
private static void doitMin(char[] x, char[] y, char[] z) {
int min = Math.min(x.length, Math.min(y.length, z.length));
for (int i = 0; i < min; i++) {
@@ -45,9 +48,12 @@
/// CHECK-DAG: <<Cnv:c\d+>> TypeConversion [<<Max>>] loop:<<Loop>> outer_loop:none
/// CHECK-DAG: ArraySet [{{l\d+}},<<Phi>>,<<Cnv>>] loop:<<Loop>> outer_loop:none
//
- // TODO: narrow type vectorization.
- /// CHECK-START: void Main.doitMax(char[], char[], char[]) loop_optimization (after)
- /// CHECK-NOT: VecMax
+ /// CHECK-START-ARM64: void Main.doitMax(char[], char[], char[]) loop_optimization (after)
+ /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none
+ /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<Max:d\d+>> VecMax [<<Get1>>,<<Get2>>] unsigned:true loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: VecStore [{{l\d+}},<<Phi>>,<<Max>>] loop:<<Loop>> outer_loop:none
private static void doitMax(char[] x, char[] y, char[] z) {
int min = Math.min(x.length, Math.min(y.length, z.length));
for (int i = 0; i < min; i++) {
diff --git a/test/651-checker-double-simd-minmax/src/Main.java b/test/651-checker-double-simd-minmax/src/Main.java
index e1711ae..cf04f85 100644
--- a/test/651-checker-double-simd-minmax/src/Main.java
+++ b/test/651-checker-double-simd-minmax/src/Main.java
@@ -48,7 +48,7 @@
/// CHECK-DAG: <<Max:d\d+>> InvokeStaticOrDirect [<<Get1>>,<<Get2>>] intrinsic:MathMaxDoubleDouble loop:<<Loop>> outer_loop:none
/// CHECK-DAG: ArraySet [{{l\d+}},<<Phi>>,<<Max>>] loop:<<Loop>> outer_loop:none
//
- // TODO-x86: 0.0 vs -0.0?
+ // TODO x86: 0.0 vs -0.0?
//
/// CHECK-START-ARM64: void Main.doitMax(double[], double[], double[]) loop_optimization (after)
/// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none
diff --git a/test/651-checker-int-simd-minmax/src/Main.java b/test/651-checker-int-simd-minmax/src/Main.java
index 4e05a9d..6cee7b5 100644
--- a/test/651-checker-int-simd-minmax/src/Main.java
+++ b/test/651-checker-int-simd-minmax/src/Main.java
@@ -30,7 +30,7 @@
/// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none
/// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop>> outer_loop:none
/// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none
- /// CHECK-DAG: <<Min:d\d+>> VecMin [<<Get1>>,<<Get2>>] loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<Min:d\d+>> VecMin [<<Get1>>,<<Get2>>] unsigned:false loop:<<Loop>> outer_loop:none
/// CHECK-DAG: VecStore [{{l\d+}},<<Phi>>,<<Min>>] loop:<<Loop>> outer_loop:none
private static void doitMin(int[] x, int[] y, int[] z) {
int min = Math.min(x.length, Math.min(y.length, z.length));
@@ -50,7 +50,7 @@
/// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none
/// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop>> outer_loop:none
/// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none
- /// CHECK-DAG: <<Max:d\d+>> VecMax [<<Get1>>,<<Get2>>] loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<Max:d\d+>> VecMax [<<Get1>>,<<Get2>>] unsigned:false loop:<<Loop>> outer_loop:none
/// CHECK-DAG: VecStore [{{l\d+}},<<Phi>>,<<Max>>] loop:<<Loop>> outer_loop:none
private static void doitMax(int[] x, int[] y, int[] z) {
int min = Math.min(x.length, Math.min(y.length, z.length));
diff --git a/test/651-checker-short-simd-minmax/src/Main.java b/test/651-checker-short-simd-minmax/src/Main.java
index f34f526..09485a2 100644
--- a/test/651-checker-short-simd-minmax/src/Main.java
+++ b/test/651-checker-short-simd-minmax/src/Main.java
@@ -27,9 +27,12 @@
/// CHECK-DAG: <<Cnv:s\d+>> TypeConversion [<<Min>>] loop:<<Loop>> outer_loop:none
/// CHECK-DAG: ArraySet [{{l\d+}},<<Phi>>,<<Cnv>>] loop:<<Loop>> outer_loop:none
//
- // TODO: narrow type vectorization.
- /// CHECK-START: void Main.doitMin(short[], short[], short[]) loop_optimization (after)
- /// CHECK-NOT: VecMin
+ /// CHECK-START-ARM64: void Main.doitMin(short[], short[], short[]) loop_optimization (after)
+ /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none
+ /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<Min:d\d+>> VecMin [<<Get1>>,<<Get2>>] unsigned:false loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: VecStore [{{l\d+}},<<Phi>>,<<Min>>] loop:<<Loop>> outer_loop:none
private static void doitMin(short[] x, short[] y, short[] z) {
int min = Math.min(x.length, Math.min(y.length, z.length));
for (int i = 0; i < min; i++) {
@@ -37,6 +40,30 @@
}
}
+ /// CHECK-START-ARM64: void Main.doitMinUnsigned(short[], short[], short[]) loop_optimization (before)
+ /// CHECK-DAG: <<IMAX:i\d+>> IntConstant 65535 loop:none
+ /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none
+ /// CHECK-DAG: <<Get1:s\d+>> ArrayGet loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<Get2:s\d+>> ArrayGet loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<And1:i\d+>> And [<<Get1>>,<<IMAX>>] loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<And2:i\d+>> And [<<Get2>>,<<IMAX>>] loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<Min:i\d+>> InvokeStaticOrDirect [<<And1>>,<<And2>>] intrinsic:MathMinIntInt loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<Cnv:s\d+>> TypeConversion [<<Min>>] loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: ArraySet [{{l\d+}},<<Phi>>,<<Cnv>>] loop:<<Loop>> outer_loop:none
+ //
+ /// CHECK-START-ARM64: void Main.doitMinUnsigned(short[], short[], short[]) loop_optimization (after)
+ /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none
+ /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<Min:d\d+>> VecMin [<<Get1>>,<<Get2>>] unsigned:true loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: VecStore [{{l\d+}},<<Phi>>,<<Min>>] loop:<<Loop>> outer_loop:none
+ private static void doitMinUnsigned(short[] x, short[] y, short[] z) {
+ int min = Math.min(x.length, Math.min(y.length, z.length));
+ for (int i = 0; i < min; i++) {
+ x[i] = (short) Math.min(y[i] & 0xffff, z[i] & 0xffff);
+ }
+ }
+
/// CHECK-START: void Main.doitMax(short[], short[], short[]) loop_optimization (before)
/// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none
/// CHECK-DAG: <<Get1:s\d+>> ArrayGet loop:<<Loop>> outer_loop:none
@@ -45,9 +72,12 @@
/// CHECK-DAG: <<Cnv:s\d+>> TypeConversion [<<Max>>] loop:<<Loop>> outer_loop:none
/// CHECK-DAG: ArraySet [{{l\d+}},<<Phi>>,<<Cnv>>] loop:<<Loop>> outer_loop:none
//
- // TODO: narrow type vectorization.
- /// CHECK-START: void Main.doitMax(short[], short[], short[]) loop_optimization (after)
- /// CHECK-NOT: VecMax
+ /// CHECK-START-ARM64: void Main.doitMax(short[], short[], short[]) loop_optimization (after)
+ /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none
+ /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<Max:d\d+>> VecMax [<<Get1>>,<<Get2>>] unsigned:false loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: VecStore [{{l\d+}},<<Phi>>,<<Max>>] loop:<<Loop>> outer_loop:none
private static void doitMax(short[] x, short[] y, short[] z) {
int min = Math.min(x.length, Math.min(y.length, z.length));
for (int i = 0; i < min; i++) {
@@ -55,6 +85,30 @@
}
}
+ /// CHECK-START-ARM64: void Main.doitMaxUnsigned(short[], short[], short[]) loop_optimization (before)
+ /// CHECK-DAG: <<IMAX:i\d+>> IntConstant 65535 loop:none
+ /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none
+ /// CHECK-DAG: <<Get1:s\d+>> ArrayGet loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<Get2:s\d+>> ArrayGet loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<And1:i\d+>> And [<<Get1>>,<<IMAX>>] loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<And2:i\d+>> And [<<Get2>>,<<IMAX>>] loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<Max:i\d+>> InvokeStaticOrDirect [<<And1>>,<<And2>>] intrinsic:MathMaxIntInt loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<Cnv:s\d+>> TypeConversion [<<Max>>] loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: ArraySet [{{l\d+}},<<Phi>>,<<Cnv>>] loop:<<Loop>> outer_loop:none
+ //
+ /// CHECK-START-ARM64: void Main.doitMaxUnsigned(short[], short[], short[]) loop_optimization (after)
+ /// CHECK-DAG: <<Phi:i\d+>> Phi loop:<<Loop:B\d+>> outer_loop:none
+ /// CHECK-DAG: <<Get1:d\d+>> VecLoad loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<Get2:d\d+>> VecLoad loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: <<Max:d\d+>> VecMax [<<Get1>>,<<Get2>>] unsigned:true loop:<<Loop>> outer_loop:none
+ /// CHECK-DAG: VecStore [{{l\d+}},<<Phi>>,<<Max>>] loop:<<Loop>> outer_loop:none
+ private static void doitMaxUnsigned(short[] x, short[] y, short[] z) {
+ int min = Math.min(x.length, Math.min(y.length, z.length));
+ for (int i = 0; i < min; i++) {
+ x[i] = (short) Math.max(y[i] & 0xffff, z[i] & 0xffff);
+ }
+ }
+
public static void main(String[] args) {
short[] interesting = {
(short) 0x0000, (short) 0x0001, (short) 0x007f,
@@ -91,11 +145,21 @@
short expected = (short) Math.min(y[i], z[i]);
expectEquals(expected, x[i]);
}
+ doitMinUnsigned(x, y, z);
+ for (int i = 0; i < total; i++) {
+ short expected = (short) Math.min(y[i] & 0xffff, z[i] & 0xffff);
+ expectEquals(expected, x[i]);
+ }
doitMax(x, y, z);
for (int i = 0; i < total; i++) {
short expected = (short) Math.max(y[i], z[i]);
expectEquals(expected, x[i]);
}
+ doitMaxUnsigned(x, y, z);
+ for (int i = 0; i < total; i++) {
+ short expected = (short) Math.max(y[i] & 0xffff, z[i] & 0xffff);
+ expectEquals(expected, x[i]);
+ }
System.out.println("passed");
}
diff --git a/test/652-deopt-intrinsic/expected.txt b/test/652-deopt-intrinsic/expected.txt
new file mode 100644
index 0000000..6a5618e
--- /dev/null
+++ b/test/652-deopt-intrinsic/expected.txt
@@ -0,0 +1 @@
+JNI_OnLoad called
diff --git a/test/652-deopt-intrinsic/info.txt b/test/652-deopt-intrinsic/info.txt
new file mode 100644
index 0000000..58a90fa
--- /dev/null
+++ b/test/652-deopt-intrinsic/info.txt
@@ -0,0 +1,2 @@
+Regression test for the interpreter/JIT, where the interpreter used to not
+record inline caches when seeing an intrinsic.
diff --git a/test/652-deopt-intrinsic/src/Main.java b/test/652-deopt-intrinsic/src/Main.java
new file mode 100644
index 0000000..a82580c
--- /dev/null
+++ b/test/652-deopt-intrinsic/src/Main.java
@@ -0,0 +1,42 @@
+/*
+ * Copyright (C) 2017 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+public class Main {
+ public static void main(String[] args) {
+ System.loadLibrary(args[0]);
+ loop();
+ ensureJitCompiled(Main.class, "$noinline$doCall");
+ loop();
+ }
+
+ public static void loop() {
+ Main m = new Main();
+ for (int i = 0; i < 5000; i++) {
+ $noinline$doCall("foo");
+ $noinline$doCall(m);
+ if (numberOfDeoptimizations() != 0) {
+ throw new Error("Unexpected deoptimizations");
+ }
+ }
+ }
+
+ public static boolean $noinline$doCall(Object foo) {
+ return foo.equals(Main.class);
+ }
+
+ public static native int numberOfDeoptimizations();
+ public static native void ensureJitCompiled(Class<?> cls, String methodName);
+}
diff --git a/test/Android.bp b/test/Android.bp
index 1679669..599b011 100644
--- a/test/Android.bp
+++ b/test/Android.bp
@@ -51,9 +51,9 @@
// These really are gtests, but the gtest library comes from libart-gtest.so
gtest: false,
defaults: [
- "art_defaults",
- "art_debug_defaults",
"art_test_defaults",
+ "art_debug_defaults",
+ "art_defaults",
],
shared_libs: [
@@ -128,8 +128,8 @@
name: "libart-gtest-defaults",
host_supported: true,
defaults: [
- "art_defaults",
"art_debug_defaults",
+ "art_defaults",
],
shared_libs: [
"libartd",
@@ -202,8 +202,8 @@
cc_defaults {
name: "libartagent-defaults",
defaults: [
- "art_defaults",
"art_test_defaults",
+ "art_defaults",
],
shared_libs: [
"libbacktrace",
@@ -234,8 +234,8 @@
name: "libartagentd",
srcs: ["900-hello-plugin/load_unload.cc"],
defaults: [
- "libartagent-defaults",
"art_debug_defaults",
+ "libartagent-defaults",
],
shared_libs: ["libartd"],
}
@@ -313,8 +313,8 @@
art_cc_test_library {
name: "libtiagentd",
defaults: [
- "libtiagent-defaults",
"art_debug_defaults",
+ "libtiagent-defaults",
],
shared_libs: ["libartd"],
}
@@ -340,8 +340,8 @@
art_cc_test_library {
name: "libtistressd",
defaults: [
- "libtistress-defaults",
"art_debug_defaults",
+ "libtistress-defaults",
],
shared_libs: ["libartd"],
}
@@ -355,8 +355,8 @@
cc_defaults {
name: "libarttest-defaults",
defaults: [
- "art_defaults",
"art_test_defaults",
+ "art_defaults",
],
srcs: [
"common/runtime_state.cc",
@@ -421,8 +421,8 @@
art_cc_test_library {
name: "libarttestd",
defaults: [
- "libarttest-defaults",
"art_debug_defaults",
+ "libarttest-defaults",
],
shared_libs: ["libartd"],
}
@@ -431,9 +431,9 @@
name: "libnativebridgetest",
shared_libs: ["libart"],
defaults: [
- "art_defaults",
- "art_debug_defaults",
"art_test_defaults",
+ "art_debug_defaults",
+ "art_defaults",
],
srcs: ["115-native-bridge/nativebridge.cc"],
target: {
diff --git a/test/Android.run-test-jvmti-java-library.mk b/test/Android.run-test-jvmti-java-library.mk
index c480be5..da28b4c 100644
--- a/test/Android.run-test-jvmti-java-library.mk
+++ b/test/Android.run-test-jvmti-java-library.mk
@@ -151,4 +151,8 @@
$(eval $(call GEN_JVMTI_RUN_TEST_GENERATED_FILE,$(NR))))
LOCAL_JAVA_RESOURCE_FILES := $(JVMTI_RUN_TEST_GENERATED_FILES)
+# We only want to depend on libcore.
+LOCAL_NO_STANDARD_LIBRARIES := true
+LOCAL_JAVA_LIBRARIES := core-all
+
include $(BUILD_JAVA_LIBRARY)
diff --git a/test/common/runtime_state.cc b/test/common/runtime_state.cc
index b683a27..d2cfbff 100644
--- a/test/common/runtime_state.cc
+++ b/test/common/runtime_state.cc
@@ -238,4 +238,8 @@
return method->GetCounter();
}
+extern "C" JNIEXPORT int JNICALL Java_Main_numberOfDeoptimizations(JNIEnv*, jclass) {
+ return Runtime::Current()->GetNumberOfDeoptimizations();
+}
+
} // namespace art
diff --git a/test/knownfailures.json b/test/knownfailures.json
index 4b44df7..96c2967 100644
--- a/test/knownfailures.json
+++ b/test/knownfailures.json
@@ -651,8 +651,7 @@
"969-iface-super",
"981-dedup-original-dex",
"984-obsolete-invoke",
- "985-re-obsolete",
- "987-stack-trace-dumping"
+ "985-re-obsolete"
],
"description": "The tests above fail with --build-with-javac-dx.",
"env_vars": {"ANDROID_COMPILE_WITH_JACK": "false"},
