summaryrefslogtreecommitdiff
path: root/compiler/optimizing
diff options
context:
space:
mode:
Diffstat (limited to 'compiler/optimizing')
-rw-r--r--compiler/optimizing/inliner.cc17
-rw-r--r--compiler/optimizing/inliner.h3
-rw-r--r--compiler/optimizing/intrinsics_arm64.cc145
-rw-r--r--compiler/optimizing/intrinsics_arm_vixl.cc1
-rw-r--r--compiler/optimizing/intrinsics_mips.cc1
-rw-r--r--compiler/optimizing/intrinsics_mips64.cc1
-rw-r--r--compiler/optimizing/intrinsics_x86.cc1
-rw-r--r--compiler/optimizing/intrinsics_x86_64.cc1
-rw-r--r--compiler/optimizing/optimization.cc2
-rw-r--r--compiler/optimizing/optimization.h1
-rw-r--r--compiler/optimizing/optimizing_compiler.cc20
11 files changed, 168 insertions, 25 deletions
diff --git a/compiler/optimizing/inliner.cc b/compiler/optimizing/inliner.cc
index ec9322270a..417d794264 100644
--- a/compiler/optimizing/inliner.cc
+++ b/compiler/optimizing/inliner.cc
@@ -28,7 +28,6 @@
#include "dex/inline_method_analyser.h"
#include "dex/verification_results.h"
#include "dex/verified_method.h"
-#include "driver/compiler_driver-inl.h"
#include "driver/compiler_options.h"
#include "driver/dex_compilation_unit.h"
#include "instruction_simplifier.h"
@@ -408,7 +407,7 @@ ArtMethod* HInliner::TryCHADevirtualization(ArtMethod* resolved_method) {
return single_impl;
}
-static bool IsMethodUnverified(CompilerDriver* const compiler_driver, ArtMethod* method)
+static bool IsMethodUnverified(const CompilerOptions& compiler_options, ArtMethod* method)
REQUIRES_SHARED(Locks::mutator_lock_) {
if (!method->GetDeclaringClass()->IsVerified()) {
if (Runtime::Current()->UseJitCompilation()) {
@@ -417,8 +416,9 @@ static bool IsMethodUnverified(CompilerDriver* const compiler_driver, ArtMethod*
return true;
}
uint16_t class_def_idx = method->GetDeclaringClass()->GetDexClassDefIndex();
- if (!compiler_driver->IsMethodVerifiedWithoutFailures(
- method->GetDexMethodIndex(), class_def_idx, *method->GetDexFile())) {
+ if (!compiler_options.IsMethodVerifiedWithoutFailures(method->GetDexMethodIndex(),
+ class_def_idx,
+ *method->GetDexFile())) {
// Method has soft or hard failures, don't analyze.
return true;
}
@@ -426,11 +426,11 @@ static bool IsMethodUnverified(CompilerDriver* const compiler_driver, ArtMethod*
return false;
}
-static bool AlwaysThrows(CompilerDriver* const compiler_driver, ArtMethod* method)
+static bool AlwaysThrows(const CompilerOptions& compiler_options, ArtMethod* method)
REQUIRES_SHARED(Locks::mutator_lock_) {
DCHECK(method != nullptr);
// Skip non-compilable and unverified methods.
- if (!method->IsCompilable() || IsMethodUnverified(compiler_driver, method)) {
+ if (!method->IsCompilable() || IsMethodUnverified(compiler_options, method)) {
return false;
}
// Skip native methods, methods with try blocks, and methods that are too large.
@@ -518,7 +518,7 @@ bool HInliner::TryInline(HInvoke* invoke_instruction) {
MaybeRecordStat(stats_, MethodCompilationStat::kInlinedInvokeVirtualOrInterface);
}
}
- } else if (!cha_devirtualize && AlwaysThrows(compiler_driver_, actual_method)) {
+ } else if (!cha_devirtualize && AlwaysThrows(codegen_->GetCompilerOptions(), actual_method)) {
// Set always throws property for non-inlined method call with single target
// (unless it was obtained through CHA, because that would imply we have
// to add the CHA dependency, which seems not worth it).
@@ -1500,7 +1500,7 @@ bool HInliner::TryBuildAndInline(HInvoke* invoke_instruction,
return false;
}
- if (IsMethodUnverified(compiler_driver_, method)) {
+ if (IsMethodUnverified(codegen_->GetCompilerOptions(), method)) {
LOG_FAIL(stats_, MethodCompilationStat::kNotInlinedNotVerified)
<< "Method " << method->PrettyMethod()
<< " couldn't be verified, so it cannot be inlined";
@@ -2069,7 +2069,6 @@ void HInliner::RunOptimizations(HGraph* callee_graph,
codegen_,
outer_compilation_unit_,
dex_compilation_unit,
- compiler_driver_,
handles_,
inline_stats_,
total_number_of_dex_registers_ + accessor.RegistersSize(),
diff --git a/compiler/optimizing/inliner.h b/compiler/optimizing/inliner.h
index 6fd0c204b2..8ac2163a94 100644
--- a/compiler/optimizing/inliner.h
+++ b/compiler/optimizing/inliner.h
@@ -38,7 +38,6 @@ class HInliner : public HOptimization {
CodeGenerator* codegen,
const DexCompilationUnit& outer_compilation_unit,
const DexCompilationUnit& caller_compilation_unit,
- CompilerDriver* compiler_driver,
VariableSizedHandleScope* handles,
OptimizingCompilerStats* stats,
size_t total_number_of_dex_registers,
@@ -51,7 +50,6 @@ class HInliner : public HOptimization {
outer_compilation_unit_(outer_compilation_unit),
caller_compilation_unit_(caller_compilation_unit),
codegen_(codegen),
- compiler_driver_(compiler_driver),
total_number_of_dex_registers_(total_number_of_dex_registers),
total_number_of_instructions_(total_number_of_instructions),
parent_(parent),
@@ -280,7 +278,6 @@ class HInliner : public HOptimization {
const DexCompilationUnit& outer_compilation_unit_;
const DexCompilationUnit& caller_compilation_unit_;
CodeGenerator* const codegen_;
- CompilerDriver* const compiler_driver_;
const size_t total_number_of_dex_registers_;
size_t total_number_of_instructions_;
diff --git a/compiler/optimizing/intrinsics_arm64.cc b/compiler/optimizing/intrinsics_arm64.cc
index 6d04b0e9d9..1688ea7811 100644
--- a/compiler/optimizing/intrinsics_arm64.cc
+++ b/compiler/optimizing/intrinsics_arm64.cc
@@ -2950,6 +2950,151 @@ void IntrinsicCodeGeneratorARM64::VisitCRC32Update(HInvoke* invoke) {
__ Mvn(out, out);
}
+// The threshold for sizes of arrays to use the library provided implementation
+// of CRC32.updateBytes instead of the intrinsic.
+static constexpr int32_t kCRC32UpdateBytesThreshold = 64 * 1024;
+
+void IntrinsicLocationsBuilderARM64::VisitCRC32UpdateBytes(HInvoke* invoke) {
+ if (!codegen_->GetInstructionSetFeatures().HasCRC()) {
+ return;
+ }
+
+ LocationSummary* locations
+ = new (allocator_) LocationSummary(invoke,
+ LocationSummary::kCallOnSlowPath,
+ kIntrinsified);
+
+ locations->SetInAt(0, Location::RequiresRegister());
+ locations->SetInAt(1, Location::RequiresRegister());
+ locations->SetInAt(2, Location::RegisterOrConstant(invoke->InputAt(2)));
+ locations->SetInAt(3, Location::RequiresRegister());
+ locations->AddTemp(Location::RequiresRegister());
+ locations->SetOut(Location::RequiresRegister());
+}
+
+// Lower the invoke of CRC32.updateBytes(int crc, byte[] b, int off, int len)
+//
+// Note: The intrinsic is not used if len exceeds a threshold.
+void IntrinsicCodeGeneratorARM64::VisitCRC32UpdateBytes(HInvoke* invoke) {
+ DCHECK(codegen_->GetInstructionSetFeatures().HasCRC());
+
+ auto masm = GetVIXLAssembler();
+ auto locations = invoke->GetLocations();
+
+ auto slow_path =
+ new (codegen_->GetScopedAllocator()) IntrinsicSlowPathARM64(invoke);
+ codegen_->AddSlowPath(slow_path);
+
+ Register length = WRegisterFrom(locations->InAt(3));
+ __ Cmp(length, kCRC32UpdateBytesThreshold);
+ __ B(slow_path->GetEntryLabel(), hi);
+
+ const uint32_t array_data_offset =
+ mirror::Array::DataOffset(Primitive::kPrimByte).Uint32Value();
+ Register ptr = XRegisterFrom(locations->GetTemp(0));
+ Register array = XRegisterFrom(locations->InAt(1));
+ auto offset = locations->InAt(2);
+ if (offset.IsConstant()) {
+ int32_t offset_value = offset.GetConstant()->AsIntConstant()->GetValue();
+ __ Add(ptr, array, array_data_offset + offset_value);
+ } else {
+ __ Add(ptr, array, array_data_offset);
+ __ Add(ptr, ptr, XRegisterFrom(offset));
+ }
+
+ // The algorithm of CRC32 of bytes is:
+ // crc = ~crc
+ // process a few first bytes to make the array 8-byte aligned
+ // while array has 8 bytes do:
+ // crc = crc32_of_8bytes(crc, 8_bytes(array))
+ // if array has 4 bytes:
+ // crc = crc32_of_4bytes(crc, 4_bytes(array))
+ // if array has 2 bytes:
+ // crc = crc32_of_2bytes(crc, 2_bytes(array))
+ // if array has a byte:
+ // crc = crc32_of_byte(crc, 1_byte(array))
+ // crc = ~crc
+
+ vixl::aarch64::Label loop, done;
+ vixl::aarch64::Label process_4bytes, process_2bytes, process_1byte;
+ vixl::aarch64::Label aligned2, aligned4, aligned8;
+
+ // Use VIXL scratch registers as the VIXL macro assembler won't use them in
+ // instructions below.
+ UseScratchRegisterScope temps(masm);
+ Register len = temps.AcquireW();
+ Register array_elem = temps.AcquireW();
+
+ Register out = WRegisterFrom(locations->Out());
+ __ Mvn(out, WRegisterFrom(locations->InAt(0)));
+ __ Mov(len, length);
+
+ __ Tbz(ptr, 0, &aligned2);
+ __ Subs(len, len, 1);
+ __ B(&done, lo);
+ __ Ldrb(array_elem, MemOperand(ptr, 1, PostIndex));
+ __ Crc32b(out, out, array_elem);
+
+ __ Bind(&aligned2);
+ __ Tbz(ptr, 1, &aligned4);
+ __ Subs(len, len, 2);
+ __ B(&process_1byte, lo);
+ __ Ldrh(array_elem, MemOperand(ptr, 2, PostIndex));
+ __ Crc32h(out, out, array_elem);
+
+ __ Bind(&aligned4);
+ __ Tbz(ptr, 2, &aligned8);
+ __ Subs(len, len, 4);
+ __ B(&process_2bytes, lo);
+ __ Ldr(array_elem, MemOperand(ptr, 4, PostIndex));
+ __ Crc32w(out, out, array_elem);
+
+ __ Bind(&aligned8);
+ __ Subs(len, len, 8);
+ // If len < 8 go to process data by 4 bytes, 2 bytes and a byte.
+ __ B(&process_4bytes, lo);
+
+ // The main loop processing data by 8 bytes.
+ __ Bind(&loop);
+ __ Ldr(array_elem.X(), MemOperand(ptr, 8, PostIndex));
+ __ Subs(len, len, 8);
+ __ Crc32x(out, out, array_elem.X());
+ // if len >= 8, process the next 8 bytes.
+ __ B(&loop, hs);
+
+ // Process the data which is less than 8 bytes.
+ // The code generated below works with values of len
+ // which come in the range [-8, 0].
+ // The first three bits are used to detect whether 4 bytes or 2 bytes or
+ // a byte can be processed.
+ // The checking order is from bit 2 to bit 0:
+ // bit 2 is set: at least 4 bytes available
+ // bit 1 is set: at least 2 bytes available
+ // bit 0 is set: at least a byte available
+ __ Bind(&process_4bytes);
+ // Goto process_2bytes if less than four bytes available
+ __ Tbz(len, 2, &process_2bytes);
+ __ Ldr(array_elem, MemOperand(ptr, 4, PostIndex));
+ __ Crc32w(out, out, array_elem);
+
+ __ Bind(&process_2bytes);
+ // Goto process_1bytes if less than two bytes available
+ __ Tbz(len, 1, &process_1byte);
+ __ Ldrh(array_elem, MemOperand(ptr, 2, PostIndex));
+ __ Crc32h(out, out, array_elem);
+
+ __ Bind(&process_1byte);
+ // Goto done if no bytes available
+ __ Tbz(len, 0, &done);
+ __ Ldrb(array_elem, MemOperand(ptr));
+ __ Crc32b(out, out, array_elem);
+
+ __ Bind(&done);
+ __ Mvn(out, out);
+
+ __ Bind(slow_path->GetExitLabel());
+}
+
UNIMPLEMENTED_INTRINSIC(ARM64, ReferenceGetReferent)
UNIMPLEMENTED_INTRINSIC(ARM64, StringStringIndexOf);
diff --git a/compiler/optimizing/intrinsics_arm_vixl.cc b/compiler/optimizing/intrinsics_arm_vixl.cc
index 4d45a9991c..88f1457c20 100644
--- a/compiler/optimizing/intrinsics_arm_vixl.cc
+++ b/compiler/optimizing/intrinsics_arm_vixl.cc
@@ -3060,6 +3060,7 @@ UNIMPLEMENTED_INTRINSIC(ARMVIXL, UnsafeCASLong) // High register pressure.
UNIMPLEMENTED_INTRINSIC(ARMVIXL, SystemArrayCopyChar)
UNIMPLEMENTED_INTRINSIC(ARMVIXL, ReferenceGetReferent)
UNIMPLEMENTED_INTRINSIC(ARMVIXL, CRC32Update)
+UNIMPLEMENTED_INTRINSIC(ARMVIXL, CRC32UpdateBytes)
UNIMPLEMENTED_INTRINSIC(ARMVIXL, StringStringIndexOf);
UNIMPLEMENTED_INTRINSIC(ARMVIXL, StringStringIndexOfAfter);
diff --git a/compiler/optimizing/intrinsics_mips.cc b/compiler/optimizing/intrinsics_mips.cc
index 21fb7d7f1c..08ba0a0adf 100644
--- a/compiler/optimizing/intrinsics_mips.cc
+++ b/compiler/optimizing/intrinsics_mips.cc
@@ -2697,6 +2697,7 @@ UNIMPLEMENTED_INTRINSIC(MIPS, ReferenceGetReferent)
UNIMPLEMENTED_INTRINSIC(MIPS, SystemArrayCopy)
UNIMPLEMENTED_INTRINSIC(MIPS, CRC32Update)
+UNIMPLEMENTED_INTRINSIC(MIPS, CRC32UpdateBytes)
UNIMPLEMENTED_INTRINSIC(MIPS, StringStringIndexOf);
UNIMPLEMENTED_INTRINSIC(MIPS, StringStringIndexOfAfter);
diff --git a/compiler/optimizing/intrinsics_mips64.cc b/compiler/optimizing/intrinsics_mips64.cc
index 4b86f5d423..59d3ba2488 100644
--- a/compiler/optimizing/intrinsics_mips64.cc
+++ b/compiler/optimizing/intrinsics_mips64.cc
@@ -2347,6 +2347,7 @@ void IntrinsicCodeGeneratorMIPS64::VisitReachabilityFence(HInvoke* invoke ATTRIB
UNIMPLEMENTED_INTRINSIC(MIPS64, ReferenceGetReferent)
UNIMPLEMENTED_INTRINSIC(MIPS64, SystemArrayCopy)
UNIMPLEMENTED_INTRINSIC(MIPS64, CRC32Update)
+UNIMPLEMENTED_INTRINSIC(MIPS64, CRC32UpdateBytes)
UNIMPLEMENTED_INTRINSIC(MIPS64, StringStringIndexOf);
UNIMPLEMENTED_INTRINSIC(MIPS64, StringStringIndexOfAfter);
diff --git a/compiler/optimizing/intrinsics_x86.cc b/compiler/optimizing/intrinsics_x86.cc
index a73f4e8b94..1d94950e4d 100644
--- a/compiler/optimizing/intrinsics_x86.cc
+++ b/compiler/optimizing/intrinsics_x86.cc
@@ -3071,6 +3071,7 @@ UNIMPLEMENTED_INTRINSIC(X86, DoubleIsInfinite)
UNIMPLEMENTED_INTRINSIC(X86, IntegerHighestOneBit)
UNIMPLEMENTED_INTRINSIC(X86, LongHighestOneBit)
UNIMPLEMENTED_INTRINSIC(X86, CRC32Update)
+UNIMPLEMENTED_INTRINSIC(X86, CRC32UpdateBytes)
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 88c766fabc..4f0b61d88e 100644
--- a/compiler/optimizing/intrinsics_x86_64.cc
+++ b/compiler/optimizing/intrinsics_x86_64.cc
@@ -2738,6 +2738,7 @@ UNIMPLEMENTED_INTRINSIC(X86_64, ReferenceGetReferent)
UNIMPLEMENTED_INTRINSIC(X86_64, FloatIsInfinite)
UNIMPLEMENTED_INTRINSIC(X86_64, DoubleIsInfinite)
UNIMPLEMENTED_INTRINSIC(X86_64, CRC32Update)
+UNIMPLEMENTED_INTRINSIC(X86_64, CRC32UpdateBytes)
UNIMPLEMENTED_INTRINSIC(X86_64, StringStringIndexOf);
UNIMPLEMENTED_INTRINSIC(X86_64, StringStringIndexOfAfter);
diff --git a/compiler/optimizing/optimization.cc b/compiler/optimizing/optimization.cc
index 0f971e100e..b75afad4c8 100644
--- a/compiler/optimizing/optimization.cc
+++ b/compiler/optimizing/optimization.cc
@@ -181,7 +181,6 @@ ArenaVector<HOptimization*> ConstructOptimizations(
HGraph* graph,
OptimizingCompilerStats* stats,
CodeGenerator* codegen,
- CompilerDriver* driver,
const DexCompilationUnit& dex_compilation_unit,
VariableSizedHandleScope* handles) {
ArenaVector<HOptimization*> optimizations(allocator->Adapter());
@@ -258,7 +257,6 @@ ArenaVector<HOptimization*> ConstructOptimizations(
codegen,
dex_compilation_unit, // outer_compilation_unit
dex_compilation_unit, // outermost_compilation_unit
- driver,
handles,
stats,
accessor.RegistersSize(),
diff --git a/compiler/optimizing/optimization.h b/compiler/optimizing/optimization.h
index 490007d9d9..ce44b5f81a 100644
--- a/compiler/optimizing/optimization.h
+++ b/compiler/optimizing/optimization.h
@@ -147,7 +147,6 @@ ArenaVector<HOptimization*> ConstructOptimizations(
HGraph* graph,
OptimizingCompilerStats* stats,
CodeGenerator* codegen,
- CompilerDriver* driver,
const DexCompilationUnit& dex_compilation_unit,
VariableSizedHandleScope* handles);
diff --git a/compiler/optimizing/optimizing_compiler.cc b/compiler/optimizing/optimizing_compiler.cc
index 641368b87a..92aaa19121 100644
--- a/compiler/optimizing/optimizing_compiler.cc
+++ b/compiler/optimizing/optimizing_compiler.cc
@@ -321,7 +321,6 @@ class OptimizingCompiler final : public Compiler {
graph,
compilation_stats_.get(),
codegen,
- GetCompilerDriver(),
dex_compilation_unit,
handles);
DCHECK_EQ(length, optimizations.size());
@@ -962,9 +961,9 @@ CodeGenerator* OptimizingCompiler::TryCompileIntrinsic(
arena_stack,
dex_file,
method_idx,
- compiler_driver->GetCompilerOptions().GetInstructionSet(),
+ compiler_options.GetInstructionSet(),
kInvalidInvokeType,
- compiler_driver->GetCompilerOptions().GetDebuggable(),
+ compiler_options.GetDebuggable(),
/* osr */ false);
DCHECK(Runtime::Current()->IsAotCompiler());
@@ -1043,12 +1042,13 @@ CompiledMethod* OptimizingCompiler::Compile(const DexFile::CodeItem* code_item,
const DexFile& dex_file,
Handle<mirror::DexCache> dex_cache) const {
CompilerDriver* compiler_driver = GetCompilerDriver();
+ const CompilerOptions& compiler_options = compiler_driver->GetCompilerOptions();
CompiledMethod* compiled_method = nullptr;
Runtime* runtime = Runtime::Current();
DCHECK(runtime->IsAotCompiler());
- const VerifiedMethod* verified_method = compiler_driver->GetVerifiedMethod(&dex_file, method_idx);
+ const VerifiedMethod* verified_method = compiler_options.GetVerifiedMethod(&dex_file, method_idx);
DCHECK(!verified_method->HasRuntimeThrow());
- if (compiler_driver->IsMethodVerifiedWithoutFailures(method_idx, class_def_idx, dex_file) ||
+ if (compiler_options.IsMethodVerifiedWithoutFailures(method_idx, class_def_idx, dex_file) ||
verifier::CanCompilerHandleVerificationFailure(
verified_method->GetEncounteredVerificationFailures())) {
ArenaAllocator allocator(runtime->GetArenaPool());
@@ -1080,7 +1080,7 @@ CompiledMethod* OptimizingCompiler::Compile(const DexFile::CodeItem* code_item,
// Go to native so that we don't block GC during compilation.
ScopedThreadSuspension sts(soa.Self(), kNative);
if (method != nullptr && UNLIKELY(method->IsIntrinsic())) {
- DCHECK(compiler_driver->GetCompilerOptions().IsBootImage());
+ DCHECK(compiler_options.IsBootImage());
codegen.reset(
TryCompileIntrinsic(&allocator,
&arena_stack,
@@ -1099,7 +1099,7 @@ CompiledMethod* OptimizingCompiler::Compile(const DexFile::CodeItem* code_item,
&code_allocator,
dex_compilation_unit,
method,
- compiler_driver->GetCompilerOptions().IsBaseline(),
+ compiler_options.IsBaseline(),
/* osr= */ false,
&handles));
}
@@ -1128,7 +1128,7 @@ CompiledMethod* OptimizingCompiler::Compile(const DexFile::CodeItem* code_item,
}
} else {
MethodCompilationStat method_stat;
- if (compiler_driver->GetCompilerOptions().VerifyAtRuntime()) {
+ if (compiler_options.VerifyAtRuntime()) {
method_stat = MethodCompilationStat::kNotCompiledVerifyAtRuntime;
} else {
method_stat = MethodCompilationStat::kNotCompiledVerificationError;
@@ -1137,8 +1137,8 @@ CompiledMethod* OptimizingCompiler::Compile(const DexFile::CodeItem* code_item,
}
if (kIsDebugBuild &&
- compiler_driver->GetCompilerOptions().CompilingWithCoreImage() &&
- IsInstructionSetSupported(compiler_driver->GetCompilerOptions().GetInstructionSet())) {
+ compiler_options.CompilingWithCoreImage() &&
+ IsInstructionSetSupported(compiler_options.GetInstructionSet())) {
// For testing purposes, we put a special marker on method names
// that should be compiled with this compiler (when the
// instruction set is supported). This makes sure we're not
generated by cgit v1.2.3-59-g8ed1b (git 2.39.0) at 2025-10-20 17:59:20 +0000