Merge "Return correct JDWP type tag for array class"
diff --git a/Android.mk b/Android.mk
index b87f0d3..6c388e5 100644
--- a/Android.mk
+++ b/Android.mk
@@ -232,7 +232,7 @@
define declare-test-art-target-run-test
.PHONY: test-art-target-run-test-$(1)
test-art-target-run-test-$(1): test-art-target-sync $(DX) $(HOST_OUT_EXECUTABLES)/jasmin
- DX=$(abspath $(DX)) JASMIN=$(abspath $(HOST_OUT_EXECUTABLES)/jasmin) art/test/run-test $(1)
+ DX=$(abspath $(DX)) JASMIN=$(abspath $(HOST_OUT_EXECUTABLES)/jasmin) art/test/run-test $(DALVIKVM_FLAGS) $(1)
@echo test-art-target-run-test-$(1) PASSED
TEST_ART_TARGET_RUN_TEST_TARGETS += test-art-target-run-test-$(1)
diff --git a/build/Android.executable.mk b/build/Android.executable.mk
index ba54e04..551b03c 100644
--- a/build/Android.executable.mk
+++ b/build/Android.executable.mk
@@ -84,6 +84,7 @@
else
LOCAL_CFLAGS += $(ART_HOST_NON_DEBUG_CFLAGS)
endif
+ LOCAL_LDLIBS += -lpthread
endif
ifeq ($$(art_ndebug_or_debug),ndebug)
diff --git a/build/Android.gtest.mk b/build/Android.gtest.mk
index 62916e3..da0b500 100644
--- a/build/Android.gtest.mk
+++ b/build/Android.gtest.mk
@@ -176,6 +176,7 @@
# GCC host compiled tests fail with this linked, presumably due to destructors that run.
LOCAL_STATIC_LIBRARIES += libgtest_host
endif
+ LOCAL_LDLIBS += -lpthread -ldl
include $(BUILD_HOST_EXECUTABLE)
art_gtest_exe := $(HOST_OUT_EXECUTABLES)/$$(LOCAL_MODULE)
ART_HOST_GTEST_EXECUTABLES += $$(art_gtest_exe)
diff --git a/compiler/Android.mk b/compiler/Android.mk
index 4eb9ff5..b17cd52 100644
--- a/compiler/Android.mk
+++ b/compiler/Android.mk
@@ -69,6 +69,7 @@
jni/quick/arm64/calling_convention_arm64.cc \
jni/quick/mips/calling_convention_mips.cc \
jni/quick/x86/calling_convention_x86.cc \
+ jni/quick/x86_64/calling_convention_x86_64.cc \
jni/quick/calling_convention.cc \
jni/quick/jni_compiler.cc \
optimizing/builder.cc \
@@ -89,6 +90,8 @@
utils/mips/managed_register_mips.cc \
utils/x86/assembler_x86.cc \
utils/x86/managed_register_x86.cc \
+ utils/x86_64/assembler_x86_64.cc \
+ utils/x86_64/managed_register_x86_64.cc \
utils/scoped_arena_allocator.cc \
buffered_output_stream.cc \
compilers.cc \
diff --git a/compiler/compiled_method.cc b/compiler/compiled_method.cc
index 344f3ef..8e013c1 100644
--- a/compiler/compiled_method.cc
+++ b/compiler/compiled_method.cc
@@ -105,6 +105,7 @@
case kArm64:
case kMips:
case kX86:
+ case kX86_64:
return 0;
case kThumb2: {
// +1 to set the low-order bit so a BLX will switch to Thumb mode
@@ -123,6 +124,7 @@
case kArm64:
case kMips:
case kX86:
+ case kX86_64:
return code_pointer;
case kThumb2: {
uintptr_t address = reinterpret_cast<uintptr_t>(code_pointer);
diff --git a/compiler/dex/compiler_enums.h b/compiler/dex/compiler_enums.h
index eb4a336..964a222 100644
--- a/compiler/dex/compiler_enums.h
+++ b/compiler/dex/compiler_enums.h
@@ -417,7 +417,7 @@
enum FixupKind {
kFixupNone,
kFixupLabel, // For labels we just adjust the offset.
- kFixupLoad, // Mostly for imediates.
+ kFixupLoad, // Mostly for immediates.
kFixupVLoad, // FP load which *may* be pc-relative.
kFixupCBxZ, // Cbz, Cbnz.
kFixupPushPop, // Not really pc relative, but changes size based on args.
diff --git a/compiler/dex/frontend.cc b/compiler/dex/frontend.cc
index 64fa685..3f122de 100644
--- a/compiler/dex/frontend.cc
+++ b/compiler/dex/frontend.cc
@@ -287,9 +287,7 @@
}
cu.EndTiming();
- driver.GetTimingsLogger()->Start();
driver.GetTimingsLogger()->AddLogger(cu.timings);
- driver.GetTimingsLogger()->End();
return result;
}
diff --git a/compiler/dex/quick/arm/assemble_arm.cc b/compiler/dex/quick/arm/assemble_arm.cc
index caf84d9..7955d6c 100644
--- a/compiler/dex/quick/arm/assemble_arm.cc
+++ b/compiler/dex/quick/arm/assemble_arm.cc
@@ -1062,134 +1062,142 @@
*/
#define PADDING_MOV_R5_R5 0x1C2D
-void ArmMir2Lir::EncodeLIR(LIR* lir) {
- int opcode = lir->opcode;
- if (IsPseudoLirOp(opcode)) {
- if (UNLIKELY(opcode == kPseudoPseudoAlign4)) {
- // Note: size for this opcode will be either 0 or 2 depending on final alignment.
- lir->u.a.bytes[0] = (PADDING_MOV_R5_R5 & 0xff);
- lir->u.a.bytes[1] = ((PADDING_MOV_R5_R5 >> 8) & 0xff);
- lir->flags.size = (lir->offset & 0x2);
- }
- } else if (LIKELY(!lir->flags.is_nop)) {
- const ArmEncodingMap *encoder = &EncodingMap[lir->opcode];
- uint32_t bits = encoder->skeleton;
- for (int i = 0; i < 4; i++) {
- uint32_t operand;
- uint32_t value;
- operand = lir->operands[i];
- ArmEncodingKind kind = encoder->field_loc[i].kind;
- if (LIKELY(kind == kFmtBitBlt)) {
- value = (operand << encoder->field_loc[i].start) &
- ((1 << (encoder->field_loc[i].end + 1)) - 1);
- bits |= value;
- } else {
- switch (encoder->field_loc[i].kind) {
- case kFmtSkip:
- break; // Nothing to do, but continue to next.
- case kFmtUnused:
- i = 4; // Done, break out of the enclosing loop.
- break;
- case kFmtFPImm:
- value = ((operand & 0xF0) >> 4) << encoder->field_loc[i].end;
- value |= (operand & 0x0F) << encoder->field_loc[i].start;
- bits |= value;
- break;
- case kFmtBrOffset:
- value = ((operand & 0x80000) >> 19) << 26;
- value |= ((operand & 0x40000) >> 18) << 11;
- value |= ((operand & 0x20000) >> 17) << 13;
- value |= ((operand & 0x1f800) >> 11) << 16;
- value |= (operand & 0x007ff);
- bits |= value;
- break;
- case kFmtShift5:
- value = ((operand & 0x1c) >> 2) << 12;
- value |= (operand & 0x03) << 6;
- bits |= value;
- break;
- case kFmtShift:
- value = ((operand & 0x70) >> 4) << 12;
- value |= (operand & 0x0f) << 4;
- bits |= value;
- break;
- case kFmtBWidth:
- value = operand - 1;
- bits |= value;
- break;
- case kFmtLsb:
- value = ((operand & 0x1c) >> 2) << 12;
- value |= (operand & 0x03) << 6;
- bits |= value;
- break;
- case kFmtImm6:
- value = ((operand & 0x20) >> 5) << 9;
- value |= (operand & 0x1f) << 3;
- bits |= value;
- break;
- case kFmtDfp: {
- DCHECK(ARM_DOUBLEREG(operand));
- DCHECK_EQ((operand & 0x1), 0U);
- uint32_t reg_name = (operand & ARM_FP_REG_MASK) >> 1;
- /* Snag the 1-bit slice and position it */
- value = ((reg_name & 0x10) >> 4) << encoder->field_loc[i].end;
- /* Extract and position the 4-bit slice */
- value |= (reg_name & 0x0f) << encoder->field_loc[i].start;
- bits |= value;
- break;
+uint8_t* ArmMir2Lir::EncodeLIRs(uint8_t* write_pos, LIR* lir) {
+ for (; lir != NULL; lir = NEXT_LIR(lir)) {
+ if (!lir->flags.is_nop) {
+ int opcode = lir->opcode;
+ if (IsPseudoLirOp(opcode)) {
+ if (UNLIKELY(opcode == kPseudoPseudoAlign4)) {
+ // Note: size for this opcode will be either 0 or 2 depending on final alignment.
+ if (lir->offset & 0x2) {
+ write_pos[0] = (PADDING_MOV_R5_R5 & 0xff);
+ write_pos[1] = ((PADDING_MOV_R5_R5 >> 8) & 0xff);
+ write_pos += 2;
}
- case kFmtSfp:
- DCHECK(ARM_SINGLEREG(operand));
- /* Snag the 1-bit slice and position it */
- value = (operand & 0x1) << encoder->field_loc[i].end;
- /* Extract and position the 4-bit slice */
- value |= ((operand & 0x1e) >> 1) << encoder->field_loc[i].start;
+ }
+ } else if (LIKELY(!lir->flags.is_nop)) {
+ const ArmEncodingMap *encoder = &EncodingMap[lir->opcode];
+ uint32_t bits = encoder->skeleton;
+ for (int i = 0; i < 4; i++) {
+ uint32_t operand;
+ uint32_t value;
+ operand = lir->operands[i];
+ ArmEncodingKind kind = encoder->field_loc[i].kind;
+ if (LIKELY(kind == kFmtBitBlt)) {
+ value = (operand << encoder->field_loc[i].start) &
+ ((1 << (encoder->field_loc[i].end + 1)) - 1);
bits |= value;
- break;
- case kFmtImm12:
- case kFmtModImm:
- value = ((operand & 0x800) >> 11) << 26;
- value |= ((operand & 0x700) >> 8) << 12;
- value |= operand & 0x0ff;
- bits |= value;
- break;
- case kFmtImm16:
- value = ((operand & 0x0800) >> 11) << 26;
- value |= ((operand & 0xf000) >> 12) << 16;
- value |= ((operand & 0x0700) >> 8) << 12;
- value |= operand & 0x0ff;
- bits |= value;
- break;
- case kFmtOff24: {
- uint32_t signbit = (operand >> 31) & 0x1;
- uint32_t i1 = (operand >> 22) & 0x1;
- uint32_t i2 = (operand >> 21) & 0x1;
- uint32_t imm10 = (operand >> 11) & 0x03ff;
- uint32_t imm11 = operand & 0x07ff;
- uint32_t j1 = (i1 ^ signbit) ? 0 : 1;
- uint32_t j2 = (i2 ^ signbit) ? 0 : 1;
- value = (signbit << 26) | (j1 << 13) | (j2 << 11) | (imm10 << 16) |
- imm11;
- bits |= value;
+ } else {
+ switch (encoder->field_loc[i].kind) {
+ case kFmtSkip:
+ break; // Nothing to do, but continue to next.
+ case kFmtUnused:
+ i = 4; // Done, break out of the enclosing loop.
+ break;
+ case kFmtFPImm:
+ value = ((operand & 0xF0) >> 4) << encoder->field_loc[i].end;
+ value |= (operand & 0x0F) << encoder->field_loc[i].start;
+ bits |= value;
+ break;
+ case kFmtBrOffset:
+ value = ((operand & 0x80000) >> 19) << 26;
+ value |= ((operand & 0x40000) >> 18) << 11;
+ value |= ((operand & 0x20000) >> 17) << 13;
+ value |= ((operand & 0x1f800) >> 11) << 16;
+ value |= (operand & 0x007ff);
+ bits |= value;
+ break;
+ case kFmtShift5:
+ value = ((operand & 0x1c) >> 2) << 12;
+ value |= (operand & 0x03) << 6;
+ bits |= value;
+ break;
+ case kFmtShift:
+ value = ((operand & 0x70) >> 4) << 12;
+ value |= (operand & 0x0f) << 4;
+ bits |= value;
+ break;
+ case kFmtBWidth:
+ value = operand - 1;
+ bits |= value;
+ break;
+ case kFmtLsb:
+ value = ((operand & 0x1c) >> 2) << 12;
+ value |= (operand & 0x03) << 6;
+ bits |= value;
+ break;
+ case kFmtImm6:
+ value = ((operand & 0x20) >> 5) << 9;
+ value |= (operand & 0x1f) << 3;
+ bits |= value;
+ break;
+ case kFmtDfp: {
+ DCHECK(ARM_DOUBLEREG(operand));
+ DCHECK_EQ((operand & 0x1), 0U);
+ uint32_t reg_name = (operand & ARM_FP_REG_MASK) >> 1;
+ /* Snag the 1-bit slice and position it */
+ value = ((reg_name & 0x10) >> 4) << encoder->field_loc[i].end;
+ /* Extract and position the 4-bit slice */
+ value |= (reg_name & 0x0f) << encoder->field_loc[i].start;
+ bits |= value;
+ break;
+ }
+ case kFmtSfp:
+ DCHECK(ARM_SINGLEREG(operand));
+ /* Snag the 1-bit slice and position it */
+ value = (operand & 0x1) << encoder->field_loc[i].end;
+ /* Extract and position the 4-bit slice */
+ value |= ((operand & 0x1e) >> 1) << encoder->field_loc[i].start;
+ bits |= value;
+ break;
+ case kFmtImm12:
+ case kFmtModImm:
+ value = ((operand & 0x800) >> 11) << 26;
+ value |= ((operand & 0x700) >> 8) << 12;
+ value |= operand & 0x0ff;
+ bits |= value;
+ break;
+ case kFmtImm16:
+ value = ((operand & 0x0800) >> 11) << 26;
+ value |= ((operand & 0xf000) >> 12) << 16;
+ value |= ((operand & 0x0700) >> 8) << 12;
+ value |= operand & 0x0ff;
+ bits |= value;
+ break;
+ case kFmtOff24: {
+ uint32_t signbit = (operand >> 31) & 0x1;
+ uint32_t i1 = (operand >> 22) & 0x1;
+ uint32_t i2 = (operand >> 21) & 0x1;
+ uint32_t imm10 = (operand >> 11) & 0x03ff;
+ uint32_t imm11 = operand & 0x07ff;
+ uint32_t j1 = (i1 ^ signbit) ? 0 : 1;
+ uint32_t j2 = (i2 ^ signbit) ? 0 : 1;
+ value = (signbit << 26) | (j1 << 13) | (j2 << 11) | (imm10 << 16) |
+ imm11;
+ bits |= value;
+ }
+ break;
+ default:
+ LOG(FATAL) << "Bad fmt:" << encoder->field_loc[i].kind;
}
- break;
- default:
- LOG(FATAL) << "Bad fmt:" << encoder->field_loc[i].kind;
+ }
+ }
+ if (encoder->size == 4) {
+ write_pos[0] = ((bits >> 16) & 0xff);
+ write_pos[1] = ((bits >> 24) & 0xff);
+ write_pos[2] = (bits & 0xff);
+ write_pos[3] = ((bits >> 8) & 0xff);
+ write_pos += 4;
+ } else {
+ DCHECK_EQ(encoder->size, 2);
+ write_pos[0] = (bits & 0xff);
+ write_pos[1] = ((bits >> 8) & 0xff);
+ write_pos += 2;
}
}
}
- if (encoder->size == 4) {
- lir->u.a.bytes[0] = ((bits >> 16) & 0xff);
- lir->u.a.bytes[1] = ((bits >> 24) & 0xff);
- lir->u.a.bytes[2] = (bits & 0xff);
- lir->u.a.bytes[3] = ((bits >> 8) & 0xff);
- } else {
- DCHECK_EQ(encoder->size, 2);
- lir->u.a.bytes[0] = (bits & 0xff);
- lir->u.a.bytes[1] = ((bits >> 8) & 0xff);
- }
- lir->flags.size = encoder->size;
}
+ return write_pos;
}
// Assemble the LIR into binary instruction format.
@@ -1198,7 +1206,7 @@
LIR* prev_lir;
cu_->NewTimingSplit("Assemble");
int assembler_retries = 0;
- CodeOffset starting_offset = EncodeRange(first_lir_insn_, last_lir_insn_, 0);
+ CodeOffset starting_offset = LinkFixupInsns(first_lir_insn_, last_lir_insn_, 0);
data_offset_ = (starting_offset + 0x3) & ~0x3;
int32_t offset_adjustment;
AssignDataOffsets();
@@ -1304,8 +1312,6 @@
lir->operands[2] = 0;
lir->operands[1] = base_reg;
}
- // Must redo encoding here - won't ever revisit this node.
- EncodeLIR(lir);
prev_lir = new_adr; // Continue scan with new_adr;
lir = new_adr->u.a.pcrel_next;
res = kRetryAll;
@@ -1346,9 +1352,8 @@
/* operand[0] is src1 in both cb[n]z & CmpRI8 */
lir->operands[1] = 0;
lir->target = 0;
- EncodeLIR(lir); // NOTE: sets flags.size.
+ lir->flags.size = EncodingMap[lir->opcode].size;
// Add back the new size.
- DCHECK_EQ(lir->flags.size, static_cast<uint32_t>(EncodingMap[lir->opcode].size));
offset_adjustment += lir->flags.size;
// Set up the new following inst.
new_inst->offset = lir->offset + lir->flags.size;
@@ -1570,20 +1575,6 @@
default:
LOG(FATAL) << "Unexpected case " << lir->flags.fixup;
}
- /*
- * If one of the pc-relative instructions expanded we'll have
- * to make another pass. Don't bother to fully assemble the
- * instruction.
- */
- if (res == kSuccess) {
- EncodeLIR(lir);
- if (assembler_retries == 0) {
- // Go ahead and fix up the code buffer image.
- for (int i = 0; i < lir->flags.size; i++) {
- code_buffer_[lir->offset + i] = lir->u.a.bytes[i];
- }
- }
- }
prev_lir = lir;
lir = lir->u.a.pcrel_next;
}
@@ -1602,21 +1593,15 @@
}
}
- // Rebuild the CodeBuffer if we had to retry; otherwise it should be good as-is.
- if (assembler_retries != 0) {
- code_buffer_.clear();
- for (LIR* lir = first_lir_insn_; lir != NULL; lir = NEXT_LIR(lir)) {
- if (lir->flags.is_nop) {
- continue;
- } else {
- for (int i = 0; i < lir->flags.size; i++) {
- code_buffer_.push_back(lir->u.a.bytes[i]);
- }
- }
- }
- }
+ // Build the CodeBuffer.
+ DCHECK_LE(data_offset_, total_size_);
+ code_buffer_.reserve(total_size_);
+ code_buffer_.resize(starting_offset);
+ uint8_t* write_pos = &code_buffer_[0];
+ write_pos = EncodeLIRs(write_pos, first_lir_insn_);
+ DCHECK_EQ(static_cast<CodeOffset>(write_pos - &code_buffer_[0]), starting_offset);
- data_offset_ = (code_buffer_.size() + 0x3) & ~0x3;
+ DCHECK_EQ(data_offset_, (code_buffer_.size() + 0x3) & ~0x3);
// Install literals
InstallLiteralPools();
@@ -1641,19 +1626,11 @@
}
// Encode instruction bit pattern and assign offsets.
-uint32_t ArmMir2Lir::EncodeRange(LIR* head_lir, LIR* tail_lir, uint32_t offset) {
+uint32_t ArmMir2Lir::LinkFixupInsns(LIR* head_lir, LIR* tail_lir, uint32_t offset) {
LIR* end_lir = tail_lir->next;
- /*
- * A significant percentage of methods can be assembled in a single pass. We'll
- * go ahead and build the code image here, leaving holes for pc-relative fixup
- * codes. If the code size changes during that pass, we'll have to throw away
- * this work - but if not, we're ready to go.
- */
- code_buffer_.reserve(estimated_native_code_size_ + 256); // Add a little slop.
LIR* last_fixup = NULL;
for (LIR* lir = head_lir; lir != end_lir; lir = NEXT_LIR(lir)) {
- lir->offset = offset;
if (!lir->flags.is_nop) {
if (lir->flags.fixup != kFixupNone) {
if (!IsPseudoLirOp(lir->opcode)) {
@@ -1675,11 +1652,7 @@
last_fixup->u.a.pcrel_next = lir;
}
last_fixup = lir;
- } else {
- EncodeLIR(lir);
- }
- for (int i = 0; i < lir->flags.size; i++) {
- code_buffer_.push_back(lir->u.a.bytes[i]);
+ lir->offset = offset;
}
offset += lir->flags.size;
}
diff --git a/compiler/dex/quick/arm/call_arm.cc b/compiler/dex/quick/arm/call_arm.cc
index 175fc06..d6724f1 100644
--- a/compiler/dex/quick/arm/call_arm.cc
+++ b/compiler/dex/quick/arm/call_arm.cc
@@ -183,15 +183,18 @@
LockCallTemps(); // Prepare for explicit register usage
constexpr bool kArchVariantHasGoodBranchPredictor = false; // TODO: true if cortex-A15.
if (kArchVariantHasGoodBranchPredictor) {
- LIR* null_check_branch;
+ LIR* null_check_branch = nullptr;
if ((opt_flags & MIR_IGNORE_NULL_CHECK) && !(cu_->disable_opt & (1 << kNullCheckElimination))) {
null_check_branch = nullptr; // No null check.
} else {
// If the null-check fails its handled by the slow-path to reduce exception related meta-data.
- null_check_branch = OpCmpImmBranch(kCondEq, rs_r0, 0, NULL);
+ if (Runtime::Current()->ExplicitNullChecks()) {
+ null_check_branch = OpCmpImmBranch(kCondEq, rs_r0, 0, NULL);
+ }
}
LoadWordDisp(rs_rARM_SELF, Thread::ThinLockIdOffset().Int32Value(), rs_r2);
NewLIR3(kThumb2Ldrex, r1, r0, mirror::Object::MonitorOffset().Int32Value() >> 2);
+ MarkPossibleNullPointerException(opt_flags);
LIR* not_unlocked_branch = OpCmpImmBranch(kCondNe, rs_r1, 0, NULL);
NewLIR4(kThumb2Strex, r1, r2, r0, mirror::Object::MonitorOffset().Int32Value() >> 2);
LIR* lock_success_branch = OpCmpImmBranch(kCondEq, rs_r1, 0, NULL);
@@ -216,8 +219,8 @@
// Explicit null-check as slow-path is entered using an IT.
GenNullCheck(rs_r0, opt_flags);
LoadWordDisp(rs_rARM_SELF, Thread::ThinLockIdOffset().Int32Value(), rs_r2);
- MarkPossibleNullPointerException(opt_flags);
NewLIR3(kThumb2Ldrex, r1, r0, mirror::Object::MonitorOffset().Int32Value() >> 2);
+ MarkPossibleNullPointerException(opt_flags);
OpRegImm(kOpCmp, rs_r1, 0);
OpIT(kCondEq, "");
NewLIR4(kThumb2Strex/*eq*/, r1, r2, r0, mirror::Object::MonitorOffset().Int32Value() >> 2);
@@ -241,7 +244,7 @@
FlushAllRegs();
LoadValueDirectFixed(rl_src, rs_r0); // Get obj
LockCallTemps(); // Prepare for explicit register usage
- LIR* null_check_branch;
+ LIR* null_check_branch = nullptr;
LoadWordDisp(rs_rARM_SELF, Thread::ThinLockIdOffset().Int32Value(), rs_r2);
constexpr bool kArchVariantHasGoodBranchPredictor = false; // TODO: true if cortex-A15.
if (kArchVariantHasGoodBranchPredictor) {
@@ -249,9 +252,12 @@
null_check_branch = nullptr; // No null check.
} else {
// If the null-check fails its handled by the slow-path to reduce exception related meta-data.
- null_check_branch = OpCmpImmBranch(kCondEq, rs_r0, 0, NULL);
+ if (Runtime::Current()->ExplicitNullChecks()) {
+ null_check_branch = OpCmpImmBranch(kCondEq, rs_r0, 0, NULL);
+ }
}
LoadWordDisp(rs_r0, mirror::Object::MonitorOffset().Int32Value(), rs_r1);
+ MarkPossibleNullPointerException(opt_flags);
LoadConstantNoClobber(rs_r3, 0);
LIR* slow_unlock_branch = OpCmpBranch(kCondNe, rs_r1, rs_r2, NULL);
StoreWordDisp(rs_r0, mirror::Object::MonitorOffset().Int32Value(), rs_r3);
@@ -404,11 +410,17 @@
}
} else {
// Implicit stack overflow check.
- // Generate a load from [sp, #-framesize]. If this is in the stack
+ // Generate a load from [sp, #-overflowsize]. If this is in the stack
// redzone we will get a segmentation fault.
- OpRegImm(kOpSub, rs_rARM_SP, frame_size_without_spills);
- LoadWordDisp(rs_rARM_SP, 0, rs_rARM_LR);
+ //
+ // Caveat coder: if someone changes the kStackOverflowReservedBytes value
+ // we need to make sure that it's loadable in an immediate field of
+ // a sub instruction. Otherwise we will get a temp allocation and the
+ // code size will increase.
+ OpRegRegImm(kOpSub, rs_r12, rs_rARM_SP, Thread::kStackOverflowReservedBytes);
+ LoadWordDisp(rs_r12, 0, rs_r12);
MarkPossibleStackOverflowException();
+ OpRegImm(kOpSub, rs_rARM_SP, frame_size_without_spills);
}
} else {
OpRegImm(kOpSub, rs_rARM_SP, frame_size_without_spills);
diff --git a/compiler/dex/quick/arm/codegen_arm.h b/compiler/dex/quick/arm/codegen_arm.h
index 0f1e171..8bfdb6a 100644
--- a/compiler/dex/quick/arm/codegen_arm.h
+++ b/compiler/dex/quick/arm/codegen_arm.h
@@ -77,10 +77,10 @@
// Required for target - miscellaneous.
void AssembleLIR();
- uint32_t EncodeRange(LIR* head_lir, LIR* tail_lir, uint32_t starting_offset);
+ uint32_t LinkFixupInsns(LIR* head_lir, LIR* tail_lir, CodeOffset offset);
int AssignInsnOffsets();
void AssignOffsets();
- void EncodeLIR(LIR* lir);
+ static uint8_t* EncodeLIRs(uint8_t* write_pos, LIR* lir);
void DumpResourceMask(LIR* lir, uint64_t mask, const char* prefix);
void SetupTargetResourceMasks(LIR* lir, uint64_t flags);
const char* GetTargetInstFmt(int opcode);
diff --git a/compiler/dex/quick/arm/utility_arm.cc b/compiler/dex/quick/arm/utility_arm.cc
index 1ec0a2c..8df5b25 100644
--- a/compiler/dex/quick/arm/utility_arm.cc
+++ b/compiler/dex/quick/arm/utility_arm.cc
@@ -828,6 +828,7 @@
int encoded_disp = displacement;
bool already_generated = false;
int dest_low_reg = r_dest.IsPair() ? r_dest.GetLowReg() : r_dest.GetReg();
+ bool null_pointer_safepoint = false;
switch (size) {
case kDouble:
case kLong:
@@ -848,6 +849,7 @@
displacement >> 2);
} else {
load = LoadBaseDispBody(r_base, displacement, r_dest.GetLow(), kWord, s_reg);
+ null_pointer_safepoint = true;
LoadBaseDispBody(r_base, displacement + 4, r_dest.GetHigh(), kWord, INVALID_SREG);
}
already_generated = true;
@@ -939,6 +941,11 @@
// TODO: in future may need to differentiate Dalvik accesses w/ spills
if (r_base == rs_rARM_SP) {
AnnotateDalvikRegAccess(load, displacement >> 2, true /* is_load */, r_dest.Is64Bit());
+ } else {
+ // We might need to generate a safepoint if we have two store instructions (wide or double).
+ if (!Runtime::Current()->ExplicitNullChecks() && null_pointer_safepoint) {
+ MarkSafepointPC(load);
+ }
}
return load;
}
@@ -965,6 +972,7 @@
int encoded_disp = displacement;
bool already_generated = false;
int src_low_reg = r_src.IsPair() ? r_src.GetLowReg() : r_src.GetReg();
+ bool null_pointer_safepoint = false;
switch (size) {
case kLong:
case kDouble:
@@ -974,6 +982,7 @@
displacement >> 2);
} else {
store = StoreBaseDispBody(r_base, displacement, r_src.GetLow(), kWord);
+ null_pointer_safepoint = true;
StoreBaseDispBody(r_base, displacement + 4, r_src.GetHigh(), kWord);
}
already_generated = true;
@@ -1061,6 +1070,11 @@
// TODO: In future, may need to differentiate Dalvik & spill accesses
if (r_base == rs_rARM_SP) {
AnnotateDalvikRegAccess(store, displacement >> 2, false /* is_load */, r_src.Is64Bit());
+ } else {
+ // We might need to generate a safepoint if we have two store instructions (wide or double).
+ if (!Runtime::Current()->ExplicitNullChecks() && null_pointer_safepoint) {
+ MarkSafepointPC(store);
+ }
}
return store;
}
diff --git a/compiler/dex/quick/codegen_util.cc b/compiler/dex/quick/codegen_util.cc
index 717ad86..4c6c7a4 100644
--- a/compiler/dex/quick/codegen_util.cc
+++ b/compiler/dex/quick/codegen_util.cc
@@ -702,7 +702,8 @@
uint32_t native_offset = it.NativePcOffset();
uint32_t dex_pc = it.DexPc();
const uint8_t* references = dex_gc_map.FindBitMap(dex_pc, false);
- CHECK(references != NULL) << "Missing ref for dex pc 0x" << std::hex << dex_pc;
+ CHECK(references != NULL) << "Missing ref for dex pc 0x" << std::hex << dex_pc <<
+ ": " << PrettyMethod(cu_->method_idx, *cu_->dex_file);
native_gc_map_builder.AddEntry(native_offset, references);
}
}
diff --git a/compiler/dex/quick/gen_common.cc b/compiler/dex/quick/gen_common.cc
index 2afa5ca..866ce5f 100644
--- a/compiler/dex/quick/gen_common.cc
+++ b/compiler/dex/quick/gen_common.cc
@@ -74,14 +74,19 @@
/* Perform null-check on a register. */
LIR* Mir2Lir::GenNullCheck(RegStorage m_reg, int opt_flags) {
if (Runtime::Current()->ExplicitNullChecks()) {
- if (!(cu_->disable_opt & (1 << kNullCheckElimination)) && (opt_flags & MIR_IGNORE_NULL_CHECK)) {
- return NULL;
- }
- return GenImmedCheck(kCondEq, m_reg, 0, kThrowNullPointer);
+ return GenExplicitNullCheck(m_reg, opt_flags);
}
return nullptr;
}
+/* Perform an explicit null-check on a register. */
+LIR* Mir2Lir::GenExplicitNullCheck(RegStorage m_reg, int opt_flags) {
+ if (!(cu_->disable_opt & (1 << kNullCheckElimination)) && (opt_flags & MIR_IGNORE_NULL_CHECK)) {
+ return NULL;
+ }
+ return GenImmedCheck(kCondEq, m_reg, 0, kThrowNullPointer);
+}
+
void Mir2Lir::MarkPossibleNullPointerException(int opt_flags) {
if (!Runtime::Current()->ExplicitNullChecks()) {
if (!(cu_->disable_opt & (1 << kNullCheckElimination)) && (opt_flags & MIR_IGNORE_NULL_CHECK)) {
@@ -732,6 +737,7 @@
OpRegRegImm(kOpAdd, reg_ptr, rl_obj.reg, field_info.FieldOffset().Int32Value());
rl_result = EvalLoc(rl_dest, reg_class, true);
LoadBaseDispWide(reg_ptr, 0, rl_result.reg, INVALID_SREG);
+ MarkPossibleNullPointerException(opt_flags);
if (field_info.IsVolatile()) {
// Without context sensitive analysis, we must issue the most conservative barriers.
// In this case, either a load or store may follow so we issue both barriers.
diff --git a/compiler/dex/quick/gen_invoke.cc b/compiler/dex/quick/gen_invoke.cc
index a0242d5..7689b51 100644
--- a/compiler/dex/quick/gen_invoke.cc
+++ b/compiler/dex/quick/gen_invoke.cc
@@ -781,7 +781,17 @@
type, skip_this);
if (pcrLabel) {
- *pcrLabel = GenNullCheck(TargetReg(kArg1), info->opt_flags);
+ if (Runtime::Current()->ExplicitNullChecks()) {
+ *pcrLabel = GenExplicitNullCheck(TargetReg(kArg1), info->opt_flags);
+ } else {
+ *pcrLabel = nullptr;
+ // In lieu of generating a check for kArg1 being null, we need to
+ // perform a load when doing implicit checks.
+ RegStorage tmp = AllocTemp();
+ LoadWordDisp(TargetReg(kArg1), 0, tmp);
+ MarkPossibleNullPointerException(info->opt_flags);
+ FreeTemp(tmp);
+ }
}
return call_state;
}
@@ -987,7 +997,17 @@
call_state = next_call_insn(cu_, info, call_state, target_method, vtable_idx,
direct_code, direct_method, type);
if (pcrLabel) {
- *pcrLabel = GenNullCheck(TargetReg(kArg1), info->opt_flags);
+ if (Runtime::Current()->ExplicitNullChecks()) {
+ *pcrLabel = GenExplicitNullCheck(TargetReg(kArg1), info->opt_flags);
+ } else {
+ *pcrLabel = nullptr;
+ // In lieu of generating a check for kArg1 being null, we need to
+ // perform a load when doing implicit checks.
+ RegStorage tmp = AllocTemp();
+ LoadWordDisp(TargetReg(kArg1), 0, tmp);
+ MarkPossibleNullPointerException(info->opt_flags);
+ FreeTemp(tmp);
+ }
}
return call_state;
}
@@ -1299,7 +1319,7 @@
LoadValueDirectFixed(rl_start, reg_start);
}
RegStorage r_tgt = LoadHelper(QUICK_ENTRYPOINT_OFFSET(pIndexOf));
- GenNullCheck(reg_ptr, info->opt_flags);
+ GenExplicitNullCheck(reg_ptr, info->opt_flags);
LIR* high_code_point_branch =
rl_char.is_const ? nullptr : OpCmpImmBranch(kCondGt, reg_char, 0xFFFF, nullptr);
// NOTE: not a safepoint
@@ -1337,7 +1357,7 @@
LoadValueDirectFixed(rl_cmp, reg_cmp);
RegStorage r_tgt = (cu_->instruction_set != kX86) ?
LoadHelper(QUICK_ENTRYPOINT_OFFSET(pStringCompareTo)) : RegStorage::InvalidReg();
- GenNullCheck(reg_this, info->opt_flags);
+ GenExplicitNullCheck(reg_this, info->opt_flags);
info->opt_flags |= MIR_IGNORE_NULL_CHECK; // Record that we've null checked.
// TUNING: check if rl_cmp.s_reg_low is already null checked
LIR* cmp_null_check_branch = OpCmpImmBranch(kCondEq, reg_cmp, 0, nullptr);
diff --git a/compiler/dex/quick/mir_to_lir.cc b/compiler/dex/quick/mir_to_lir.cc
index e81a037..cd3dadb 100644
--- a/compiler/dex/quick/mir_to_lir.cc
+++ b/compiler/dex/quick/mir_to_lir.cc
@@ -446,6 +446,7 @@
GenNullCheck(rl_src[0].reg, opt_flags);
rl_result = EvalLoc(rl_dest, kCoreReg, true);
LoadWordDisp(rl_src[0].reg, len_offset, rl_result.reg);
+ MarkPossibleNullPointerException(opt_flags);
StoreValue(rl_dest, rl_result);
break;
diff --git a/compiler/dex/quick/mir_to_lir.h b/compiler/dex/quick/mir_to_lir.h
index bac35aa..10f431f 100644
--- a/compiler/dex/quick/mir_to_lir.h
+++ b/compiler/dex/quick/mir_to_lir.h
@@ -131,7 +131,6 @@
struct AssemblyInfo {
LIR* pcrel_next; // Chain of LIR nodes needing pc relative fixups.
- uint8_t bytes[16]; // Encoded instruction bytes.
};
struct LIR {
@@ -151,7 +150,7 @@
} flags;
union {
UseDefMasks m; // Use & Def masks used during optimization.
- AssemblyInfo a; // Instruction encoding used during assembly phase.
+ AssemblyInfo a; // Instruction info used during assembly phase.
} u;
int32_t operands[5]; // [0..4] = [dest, src1, src2, extra, extra2].
};
@@ -340,7 +339,7 @@
return code_buffer_.size() / sizeof(code_buffer_[0]);
}
- bool IsPseudoLirOp(int opcode) {
+ static bool IsPseudoLirOp(int opcode) {
return (opcode < 0);
}
@@ -565,6 +564,7 @@
void ForceImplicitNullCheck(RegStorage reg, int opt_flags);
LIR* GenImmedCheck(ConditionCode c_code, RegStorage reg, int imm_val, ThrowKind kind);
LIR* GenNullCheck(RegStorage m_reg, int opt_flags);
+ LIR* GenExplicitNullCheck(RegStorage m_reg, int opt_flags);
LIR* GenRegRegCheck(ConditionCode c_code, RegStorage reg1, RegStorage reg2, ThrowKind kind);
void GenCompareAndBranch(Instruction::Code opcode, RegLocation rl_src1,
RegLocation rl_src2, LIR* taken, LIR* fall_through);
diff --git a/compiler/jni/quick/arm/calling_convention_arm.cc b/compiler/jni/quick/arm/calling_convention_arm.cc
index 78f403c..28b438e 100644
--- a/compiler/jni/quick/arm/calling_convention_arm.cc
+++ b/compiler/jni/quick/arm/calling_convention_arm.cc
@@ -85,7 +85,7 @@
return result;
}
-const std::vector<ManagedRegister>& ArmManagedRuntimeCallingConvention::EntrySpills() {
+const ManagedRegisterEntrySpills& ArmManagedRuntimeCallingConvention::EntrySpills() {
// We spill the argument registers on ARM to free them up for scratch use, we then assume
// all arguments are on the stack.
if (entry_spills_.size() == 0) {
diff --git a/compiler/jni/quick/arm/calling_convention_arm.h b/compiler/jni/quick/arm/calling_convention_arm.h
index fc2d857..96bbb7e 100644
--- a/compiler/jni/quick/arm/calling_convention_arm.h
+++ b/compiler/jni/quick/arm/calling_convention_arm.h
@@ -36,10 +36,10 @@
bool IsCurrentParamOnStack() OVERRIDE;
ManagedRegister CurrentParamRegister() OVERRIDE;
FrameOffset CurrentParamStackOffset() OVERRIDE;
- const std::vector<ManagedRegister>& EntrySpills() OVERRIDE;
+ const ManagedRegisterEntrySpills& EntrySpills() OVERRIDE;
private:
- std::vector<ManagedRegister> entry_spills_;
+ ManagedRegisterEntrySpills entry_spills_;
DISALLOW_COPY_AND_ASSIGN(ArmManagedRuntimeCallingConvention);
};
diff --git a/compiler/jni/quick/arm64/calling_convention_arm64.cc b/compiler/jni/quick/arm64/calling_convention_arm64.cc
index c4d0d45..ff899b7 100644
--- a/compiler/jni/quick/arm64/calling_convention_arm64.cc
+++ b/compiler/jni/quick/arm64/calling_convention_arm64.cc
@@ -85,7 +85,7 @@
return result;
}
-const std::vector<ManagedRegister>& Arm64ManagedRuntimeCallingConvention::EntrySpills() {
+const ManagedRegisterEntrySpills& Arm64ManagedRuntimeCallingConvention::EntrySpills() {
// We spill the argument registers on ARM64 to free them up for scratch use, we then assume
// all arguments are on the stack.
if (entry_spills_.size() == 0) {
diff --git a/compiler/jni/quick/arm64/calling_convention_arm64.h b/compiler/jni/quick/arm64/calling_convention_arm64.h
index 2dcf1af..7e33830 100644
--- a/compiler/jni/quick/arm64/calling_convention_arm64.h
+++ b/compiler/jni/quick/arm64/calling_convention_arm64.h
@@ -36,10 +36,10 @@
bool IsCurrentParamOnStack() OVERRIDE;
ManagedRegister CurrentParamRegister() OVERRIDE;
FrameOffset CurrentParamStackOffset() OVERRIDE;
- const std::vector<ManagedRegister>& EntrySpills() OVERRIDE;
+ const ManagedRegisterEntrySpills& EntrySpills() OVERRIDE;
private:
- std::vector<ManagedRegister> entry_spills_;
+ ManagedRegisterEntrySpills entry_spills_;
DISALLOW_COPY_AND_ASSIGN(Arm64ManagedRuntimeCallingConvention);
};
diff --git a/compiler/jni/quick/calling_convention.cc b/compiler/jni/quick/calling_convention.cc
index 5856df4..043bcea 100644
--- a/compiler/jni/quick/calling_convention.cc
+++ b/compiler/jni/quick/calling_convention.cc
@@ -21,6 +21,7 @@
#include "jni/quick/arm64/calling_convention_arm64.h"
#include "jni/quick/mips/calling_convention_mips.h"
#include "jni/quick/x86/calling_convention_x86.h"
+#include "jni/quick/x86_64/calling_convention_x86_64.h"
#include "utils.h"
namespace art {
@@ -44,6 +45,8 @@
return new mips::MipsManagedRuntimeCallingConvention(is_static, is_synchronized, shorty);
case kX86:
return new x86::X86ManagedRuntimeCallingConvention(is_static, is_synchronized, shorty);
+ case kX86_64:
+ return new x86_64::X86_64ManagedRuntimeCallingConvention(is_static, is_synchronized, shorty);
default:
LOG(FATAL) << "Unknown InstructionSet: " << instruction_set;
return NULL;
@@ -61,6 +64,9 @@
itr_longs_and_doubles_++;
itr_slots_++;
}
+ if (IsParamAFloatOrDouble(itr_args_)) {
+ itr_float_and_doubles_++;
+ }
if (IsCurrentParamAReference()) {
itr_refs_++;
}
@@ -85,6 +91,10 @@
return IsParamAReference(itr_args_);
}
+bool ManagedRuntimeCallingConvention::IsCurrentParamAFloatOrDouble() {
+ return IsParamAFloatOrDouble(itr_args_);
+}
+
// JNI calling convention
JniCallingConvention* JniCallingConvention::Create(bool is_static, bool is_synchronized,
@@ -100,6 +110,8 @@
return new mips::MipsJniCallingConvention(is_static, is_synchronized, shorty);
case kX86:
return new x86::X86JniCallingConvention(is_static, is_synchronized, shorty);
+ case kX86_64:
+ return new x86_64::X86_64JniCallingConvention(is_static, is_synchronized, shorty);
default:
LOG(FATAL) << "Unknown InstructionSet: " << instruction_set;
return NULL;
@@ -111,9 +123,8 @@
}
FrameOffset JniCallingConvention::SavedLocalReferenceCookieOffset() const {
- size_t start_of_sirt = SirtNumRefsOffset().Int32Value() + kPointerSize;
- size_t references_size = kPointerSize * ReferenceCount(); // size excluding header
- return FrameOffset(start_of_sirt + references_size);
+ size_t references_size = kSirtPointerSize * ReferenceCount(); // size excluding header
+ return FrameOffset(SirtReferencesOffset().Int32Value() + references_size);
}
FrameOffset JniCallingConvention::ReturnValueSaveLocation() const {
@@ -139,6 +150,9 @@
itr_slots_++;
}
}
+ if (IsCurrentParamAFloatOrDouble()) {
+ itr_float_and_doubles_++;
+ }
if (IsCurrentParamAReference()) {
itr_refs_++;
}
@@ -159,14 +173,25 @@
}
}
+bool JniCallingConvention::IsCurrentParamAFloatOrDouble() {
+ switch (itr_args_) {
+ case kJniEnv:
+ return false; // JNIEnv*
+ case kObjectOrClass:
+ return false; // jobject or jclass
+ default: {
+ int arg_pos = itr_args_ - NumberOfExtraArgumentsForJni();
+ return IsParamAFloatOrDouble(arg_pos);
+ }
+ }
+}
+
// Return position of SIRT entry holding reference at the current iterator
// position
FrameOffset JniCallingConvention::CurrentParamSirtEntryOffset() {
CHECK(IsCurrentParamAReference());
CHECK_LT(SirtLinkOffset(), SirtNumRefsOffset());
- // Address of 1st SIRT entry
- int result = SirtNumRefsOffset().Int32Value() + kPointerSize;
- result += itr_refs_ * kPointerSize;
+ int result = SirtReferencesOffset().Int32Value() + itr_refs_ * kSirtPointerSize;
CHECK_GT(result, SirtNumRefsOffset().Int32Value());
return FrameOffset(result);
}
diff --git a/compiler/jni/quick/calling_convention.h b/compiler/jni/quick/calling_convention.h
index f2b7fd9..fe3d1cd 100644
--- a/compiler/jni/quick/calling_convention.h
+++ b/compiler/jni/quick/calling_convention.h
@@ -60,23 +60,35 @@
itr_args_ = 0;
itr_refs_ = 0;
itr_longs_and_doubles_ = 0;
+ itr_float_and_doubles_ = 0;
}
virtual ~CallingConvention() {}
protected:
CallingConvention(bool is_static, bool is_synchronized, const char* shorty)
- : displacement_(0), is_static_(is_static), is_synchronized_(is_synchronized),
+ : displacement_(0), kSirtPointerSize(sizeof(StackReference<mirror::Object>)), is_static_(is_static), is_synchronized_(is_synchronized),
shorty_(shorty) {
num_args_ = (is_static ? 0 : 1) + strlen(shorty) - 1;
num_ref_args_ = is_static ? 0 : 1; // The implicit this pointer.
+ num_float_or_double_args_ = 0;
num_long_or_double_args_ = 0;
for (size_t i = 1; i < strlen(shorty); i++) {
char ch = shorty_[i];
- if (ch == 'L') {
+ switch (ch) {
+ case 'L':
num_ref_args_++;
- } else if ((ch == 'D') || (ch == 'J')) {
+ break;
+ case 'J':
num_long_or_double_args_++;
+ break;
+ case 'D':
+ num_long_or_double_args_++;
+ num_float_or_double_args_++;
+ break;
+ case 'F':
+ num_float_or_double_args_++;
+ break;
}
}
}
@@ -97,6 +109,16 @@
char ch = shorty_[param];
return (ch == 'J' || ch == 'D');
}
+ bool IsParamAFloatOrDouble(unsigned int param) const {
+ DCHECK_LT(param, NumArgs());
+ if (IsStatic()) {
+ param++; // 0th argument must skip return value at start of the shorty
+ } else if (param == 0) {
+ return false; // this argument
+ }
+ char ch = shorty_[param];
+ return (ch == 'F' || ch == 'D');
+ }
bool IsParamAReference(unsigned int param) const {
DCHECK_LT(param, NumArgs());
if (IsStatic()) {
@@ -112,6 +134,9 @@
size_t NumLongOrDoubleArgs() const {
return num_long_or_double_args_;
}
+ size_t NumFloatOrDoubleArgs() const {
+ return num_float_or_double_args_;
+ }
size_t NumReferenceArgs() const {
return num_ref_args_;
}
@@ -141,8 +166,11 @@
unsigned int itr_args_;
// Number of longs and doubles seen along argument list
unsigned int itr_longs_and_doubles_;
+ // Number of float and doubles seen along argument list
+ unsigned int itr_float_and_doubles_;
// Space for frames below this on the stack
FrameOffset displacement_;
+ size_t kSirtPointerSize;
private:
const bool is_static_;
@@ -150,6 +178,7 @@
std::string shorty_;
size_t num_args_;
size_t num_ref_args_;
+ size_t num_float_or_double_args_;
size_t num_long_or_double_args_;
};
@@ -174,6 +203,7 @@
bool HasNext();
void Next();
bool IsCurrentParamAReference();
+ bool IsCurrentParamAFloatOrDouble();
bool IsCurrentArgExplicit(); // ie a non-implict argument such as this
bool IsCurrentArgPossiblyNull();
size_t CurrentParamSize();
@@ -185,7 +215,7 @@
virtual ~ManagedRuntimeCallingConvention() {}
// Registers to spill to caller's out registers on entry.
- virtual const std::vector<ManagedRegister>& EntrySpills() = 0;
+ virtual const ManagedRegisterEntrySpills& EntrySpills() = 0;
protected:
ManagedRuntimeCallingConvention(bool is_static, bool is_synchronized, const char* shorty)
@@ -241,6 +271,7 @@
bool HasNext();
virtual void Next();
bool IsCurrentParamAReference();
+ bool IsCurrentParamAFloatOrDouble();
size_t CurrentParamSize();
virtual bool IsCurrentParamInRegister() = 0;
virtual bool IsCurrentParamOnStack() = 0;
@@ -255,13 +286,21 @@
return FrameOffset(displacement_.Int32Value() +
kPointerSize); // above Method*
}
+
+ FrameOffset SirtLinkOffset() const {
+ return FrameOffset(SirtOffset().Int32Value() +
+ StackIndirectReferenceTable::LinkOffset());
+ }
+
FrameOffset SirtNumRefsOffset() const {
return FrameOffset(SirtOffset().Int32Value() +
StackIndirectReferenceTable::NumberOfReferencesOffset());
}
- FrameOffset SirtLinkOffset() const {
- return FrameOffset(SirtOffset().Int32Value() +
- StackIndirectReferenceTable::LinkOffset());
+
+ FrameOffset SirtReferencesOffset() const {
+ // The StackIndirectReferenceTable::number_of_references_ type is uint32_t
+ return FrameOffset(SirtNumRefsOffset().Int32Value() +
+ sizeof(uint32_t));
}
virtual ~JniCallingConvention() {}
diff --git a/compiler/jni/quick/jni_compiler.cc b/compiler/jni/quick/jni_compiler.cc
index 1c9aed8..c89bc40 100644
--- a/compiler/jni/quick/jni_compiler.cc
+++ b/compiler/jni/quick/jni_compiler.cc
@@ -271,7 +271,7 @@
mr_conv->InterproceduralScratchRegister());
// 10. Fix differences in result widths.
- if (instruction_set == kX86) {
+ if (instruction_set == kX86 || instruction_set == kX86_64) {
if (main_jni_conv->GetReturnType() == Primitive::kPrimByte ||
main_jni_conv->GetReturnType() == Primitive::kPrimShort) {
__ SignExtend(main_jni_conv->ReturnRegister(),
diff --git a/compiler/jni/quick/mips/calling_convention_mips.cc b/compiler/jni/quick/mips/calling_convention_mips.cc
index 0a48500..ea39d60 100644
--- a/compiler/jni/quick/mips/calling_convention_mips.cc
+++ b/compiler/jni/quick/mips/calling_convention_mips.cc
@@ -85,7 +85,7 @@
return result;
}
-const std::vector<ManagedRegister>& MipsManagedRuntimeCallingConvention::EntrySpills() {
+const ManagedRegisterEntrySpills& MipsManagedRuntimeCallingConvention::EntrySpills() {
// We spill the argument registers on MIPS to free them up for scratch use, we then assume
// all arguments are on the stack.
if (entry_spills_.size() == 0) {
diff --git a/compiler/jni/quick/mips/calling_convention_mips.h b/compiler/jni/quick/mips/calling_convention_mips.h
index 445f453..1a9053a 100644
--- a/compiler/jni/quick/mips/calling_convention_mips.h
+++ b/compiler/jni/quick/mips/calling_convention_mips.h
@@ -35,10 +35,10 @@
bool IsCurrentParamOnStack() OVERRIDE;
ManagedRegister CurrentParamRegister() OVERRIDE;
FrameOffset CurrentParamStackOffset() OVERRIDE;
- const std::vector<ManagedRegister>& EntrySpills() OVERRIDE;
+ const ManagedRegisterEntrySpills& EntrySpills() OVERRIDE;
private:
- std::vector<ManagedRegister> entry_spills_;
+ ManagedRegisterEntrySpills entry_spills_;
DISALLOW_COPY_AND_ASSIGN(MipsManagedRuntimeCallingConvention);
};
diff --git a/compiler/jni/quick/x86/calling_convention_x86.cc b/compiler/jni/quick/x86/calling_convention_x86.cc
index 8b5c86d..8d22fe6 100644
--- a/compiler/jni/quick/x86/calling_convention_x86.cc
+++ b/compiler/jni/quick/x86/calling_convention_x86.cc
@@ -90,7 +90,7 @@
(itr_slots_ * kPointerSize)); // offset into in args
}
-const std::vector<ManagedRegister>& X86ManagedRuntimeCallingConvention::EntrySpills() {
+const ManagedRegisterEntrySpills& X86ManagedRuntimeCallingConvention::EntrySpills() {
// We spill the argument registers on X86 to free them up for scratch use, we then assume
// all arguments are on the stack.
if (entry_spills_.size() == 0) {
diff --git a/compiler/jni/quick/x86/calling_convention_x86.h b/compiler/jni/quick/x86/calling_convention_x86.h
index e814c7e..2dab059 100644
--- a/compiler/jni/quick/x86/calling_convention_x86.h
+++ b/compiler/jni/quick/x86/calling_convention_x86.h
@@ -37,9 +37,9 @@
bool IsCurrentParamOnStack() OVERRIDE;
ManagedRegister CurrentParamRegister() OVERRIDE;
FrameOffset CurrentParamStackOffset() OVERRIDE;
- const std::vector<ManagedRegister>& EntrySpills() OVERRIDE;
+ const ManagedRegisterEntrySpills& EntrySpills() OVERRIDE;
private:
- std::vector<ManagedRegister> entry_spills_;
+ ManagedRegisterEntrySpills entry_spills_;
DISALLOW_COPY_AND_ASSIGN(X86ManagedRuntimeCallingConvention);
};
diff --git a/compiler/jni/quick/x86_64/calling_convention_x86_64.cc b/compiler/jni/quick/x86_64/calling_convention_x86_64.cc
new file mode 100644
index 0000000..8ebea46
--- /dev/null
+++ b/compiler/jni/quick/x86_64/calling_convention_x86_64.cc
@@ -0,0 +1,203 @@
+/*
+ * Copyright (C) 2014 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.
+ */
+
+#include "calling_convention_x86_64.h"
+
+#include "base/logging.h"
+#include "utils/x86_64/managed_register_x86_64.h"
+#include "utils.h"
+
+namespace art {
+namespace x86_64 {
+
+// Calling convention
+
+ManagedRegister X86_64ManagedRuntimeCallingConvention::InterproceduralScratchRegister() {
+ return X86_64ManagedRegister::FromCpuRegister(RAX);
+}
+
+ManagedRegister X86_64JniCallingConvention::InterproceduralScratchRegister() {
+ return X86_64ManagedRegister::FromCpuRegister(RAX);
+}
+
+ManagedRegister X86_64JniCallingConvention::ReturnScratchRegister() const {
+ return ManagedRegister::NoRegister(); // No free regs, so assembler uses push/pop
+}
+
+static ManagedRegister ReturnRegisterForShorty(const char* shorty, bool jni) {
+ if (shorty[0] == 'F' || shorty[0] == 'D') {
+ return X86_64ManagedRegister::FromXmmRegister(_XMM0);
+ } else if (shorty[0] == 'J') {
+ return X86_64ManagedRegister::FromCpuRegister(RAX);
+ } else if (shorty[0] == 'V') {
+ return ManagedRegister::NoRegister();
+ } else {
+ return X86_64ManagedRegister::FromCpuRegister(RAX);
+ }
+}
+
+ManagedRegister X86_64ManagedRuntimeCallingConvention::ReturnRegister() {
+ return ReturnRegisterForShorty(GetShorty(), false);
+}
+
+ManagedRegister X86_64JniCallingConvention::ReturnRegister() {
+ return ReturnRegisterForShorty(GetShorty(), true);
+}
+
+ManagedRegister X86_64JniCallingConvention::IntReturnRegister() {
+ return X86_64ManagedRegister::FromCpuRegister(RAX);
+}
+
+// Managed runtime calling convention
+
+ManagedRegister X86_64ManagedRuntimeCallingConvention::MethodRegister() {
+ return X86_64ManagedRegister::FromCpuRegister(RDI);
+}
+
+bool X86_64ManagedRuntimeCallingConvention::IsCurrentParamInRegister() {
+ return !IsCurrentParamOnStack();
+}
+
+bool X86_64ManagedRuntimeCallingConvention::IsCurrentParamOnStack() {
+ // We assume all parameters are on stack, args coming via registers are spilled as entry_spills
+ return true;
+}
+
+ManagedRegister X86_64ManagedRuntimeCallingConvention::CurrentParamRegister() {
+ ManagedRegister res = ManagedRegister::NoRegister();
+ if (!IsCurrentParamAFloatOrDouble()) {
+ switch (itr_args_ - itr_float_and_doubles_) {
+ case 0: res = X86_64ManagedRegister::FromCpuRegister(RSI); break;
+ case 1: res = X86_64ManagedRegister::FromCpuRegister(RDX); break;
+ case 2: res = X86_64ManagedRegister::FromCpuRegister(RCX); break;
+ case 3: res = X86_64ManagedRegister::FromCpuRegister(R8); break;
+ case 4: res = X86_64ManagedRegister::FromCpuRegister(R9); break;
+ }
+ } else if (itr_float_and_doubles_ < 8) {
+ // First eight float parameters are passed via XMM0..XMM7
+ res = X86_64ManagedRegister::FromXmmRegister(
+ static_cast<XmmRegister>(_XMM0 + itr_float_and_doubles_));
+ }
+ return res;
+}
+
+FrameOffset X86_64ManagedRuntimeCallingConvention::CurrentParamStackOffset() {
+ return FrameOffset(displacement_.Int32Value() + // displacement
+ kPointerSize + // Method*
+ (itr_slots_ * sizeof(uint32_t))); // offset into in args
+}
+
+const ManagedRegisterEntrySpills& X86_64ManagedRuntimeCallingConvention::EntrySpills() {
+ // We spill the argument registers on X86 to free them up for scratch use, we then assume
+ // all arguments are on the stack.
+ if (entry_spills_.size() == 0) {
+ ResetIterator(FrameOffset(0));
+ while (HasNext()) {
+ ManagedRegister in_reg = CurrentParamRegister();
+ if (!in_reg.IsNoRegister()) {
+ int32_t size = IsParamALongOrDouble(itr_args_)? 8 : 4;
+ int32_t spill_offset = CurrentParamStackOffset().Uint32Value();
+ ManagedRegisterSpill spill(in_reg, size, spill_offset);
+ entry_spills_.push_back(spill);
+ }
+ Next();
+ }
+ }
+ return entry_spills_;
+}
+
+// JNI calling convention
+
+X86_64JniCallingConvention::X86_64JniCallingConvention(bool is_static, bool is_synchronized,
+ const char* shorty)
+ : JniCallingConvention(is_static, is_synchronized, shorty) {
+ callee_save_regs_.push_back(X86_64ManagedRegister::FromCpuRegister(RBX));
+ callee_save_regs_.push_back(X86_64ManagedRegister::FromCpuRegister(RBP));
+ callee_save_regs_.push_back(X86_64ManagedRegister::FromCpuRegister(R12));
+ callee_save_regs_.push_back(X86_64ManagedRegister::FromCpuRegister(R13));
+ callee_save_regs_.push_back(X86_64ManagedRegister::FromCpuRegister(R14));
+ callee_save_regs_.push_back(X86_64ManagedRegister::FromCpuRegister(R15));
+}
+
+uint32_t X86_64JniCallingConvention::CoreSpillMask() const {
+ return 1 << RBX | 1 << RBP | 1 << R12 | 1 << R13 | 1 << R14 | 1 << R15 | 1 << R13 | 1 << kNumberOfCpuRegisters;
+}
+
+size_t X86_64JniCallingConvention::FrameSize() {
+ // Method*, return address and callee save area size, local reference segment state
+ size_t frame_data_size = (3 + CalleeSaveRegisters().size()) * kPointerSize;
+ // References plus link_ (pointer) and number_of_references_ (uint32_t) for SIRT header
+ size_t sirt_size = kPointerSize + sizeof(uint32_t) + ReferenceCount()*kSirtPointerSize;
+ // Plus return value spill area size
+ return RoundUp(frame_data_size + sirt_size + SizeOfReturnValue(), kStackAlignment);
+}
+
+size_t X86_64JniCallingConvention::OutArgSize() {
+ return RoundUp(NumberOfOutgoingStackArgs() * kPointerSize, kStackAlignment);
+}
+
+bool X86_64JniCallingConvention::IsCurrentParamInRegister() {
+ return !IsCurrentParamOnStack();
+}
+
+bool X86_64JniCallingConvention::IsCurrentParamOnStack() {
+ return CurrentParamRegister().IsNoRegister();
+}
+
+ManagedRegister X86_64JniCallingConvention::CurrentParamRegister() {
+ ManagedRegister res = ManagedRegister::NoRegister();
+ if (!IsCurrentParamAFloatOrDouble()) {
+ switch (itr_args_ - itr_float_and_doubles_) {
+ case 0: res = X86_64ManagedRegister::FromCpuRegister(RDI); break;
+ case 1: res = X86_64ManagedRegister::FromCpuRegister(RSI); break;
+ case 2: res = X86_64ManagedRegister::FromCpuRegister(RDX); break;
+ case 3: res = X86_64ManagedRegister::FromCpuRegister(RCX); break;
+ case 4: res = X86_64ManagedRegister::FromCpuRegister(R8); break;
+ case 5: res = X86_64ManagedRegister::FromCpuRegister(R9); break;
+ }
+ } else if (itr_float_and_doubles_ < 8) {
+ // First eight float parameters are passed via XMM0..XMM7
+ res = X86_64ManagedRegister::FromXmmRegister(
+ static_cast<XmmRegister>(_XMM0 + itr_float_and_doubles_));
+ }
+ return res;
+}
+
+FrameOffset X86_64JniCallingConvention::CurrentParamStackOffset() {
+ size_t offset = itr_args_
+ - std::min(8U, itr_float_and_doubles_) // Float arguments passed through Xmm0..Xmm7
+ - std::min(6U, itr_args_ - itr_float_and_doubles_); // Integer arguments passed through GPR
+ return FrameOffset(displacement_.Int32Value() - OutArgSize() + (offset * kPointerSize));
+}
+
+size_t X86_64JniCallingConvention::NumberOfOutgoingStackArgs() {
+ size_t static_args = IsStatic() ? 1 : 0; // count jclass
+ // regular argument parameters and this
+ size_t param_args = NumArgs() + NumLongOrDoubleArgs();
+ // count JNIEnv* and return pc (pushed after Method*)
+ size_t total_args = static_args + param_args + 2;
+
+ // Float arguments passed through Xmm0..Xmm7
+ // Other (integer) arguments passed through GPR (RDI, RSI, RDX, RCX, R8, R9)
+ size_t total_stack_args = total_args
+ - std::min(8U, static_cast<unsigned int>(NumFloatOrDoubleArgs()))
+ - std::min(6U, static_cast<unsigned int>(NumArgs() - NumFloatOrDoubleArgs()));
+
+ return total_stack_args;
+}
+
+} // namespace x86_64
+} // namespace art
diff --git a/compiler/jni/quick/x86_64/calling_convention_x86_64.h b/compiler/jni/quick/x86_64/calling_convention_x86_64.h
new file mode 100644
index 0000000..d7f7762
--- /dev/null
+++ b/compiler/jni/quick/x86_64/calling_convention_x86_64.h
@@ -0,0 +1,83 @@
+/*
+ * Copyright (C) 2014 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.
+ */
+
+#ifndef ART_COMPILER_JNI_QUICK_X86_64_CALLING_CONVENTION_X86_64_H_
+#define ART_COMPILER_JNI_QUICK_X86_64_CALLING_CONVENTION_X86_64_H_
+
+#include "jni/quick/calling_convention.h"
+
+namespace art {
+namespace x86_64 {
+
+class X86_64ManagedRuntimeCallingConvention FINAL : public ManagedRuntimeCallingConvention {
+ public:
+ explicit X86_64ManagedRuntimeCallingConvention(bool is_static, bool is_synchronized,
+ const char* shorty)
+ : ManagedRuntimeCallingConvention(is_static, is_synchronized, shorty) {}
+ ~X86_64ManagedRuntimeCallingConvention() OVERRIDE {}
+ // Calling convention
+ ManagedRegister ReturnRegister() OVERRIDE;
+ ManagedRegister InterproceduralScratchRegister() OVERRIDE;
+ // Managed runtime calling convention
+ ManagedRegister MethodRegister() OVERRIDE;
+ bool IsCurrentParamInRegister() OVERRIDE;
+ bool IsCurrentParamOnStack() OVERRIDE;
+ ManagedRegister CurrentParamRegister() OVERRIDE;
+ FrameOffset CurrentParamStackOffset() OVERRIDE;
+ const ManagedRegisterEntrySpills& EntrySpills() OVERRIDE;
+ private:
+ ManagedRegisterEntrySpills entry_spills_;
+ DISALLOW_COPY_AND_ASSIGN(X86_64ManagedRuntimeCallingConvention);
+};
+
+class X86_64JniCallingConvention FINAL : public JniCallingConvention {
+ public:
+ explicit X86_64JniCallingConvention(bool is_static, bool is_synchronized, const char* shorty);
+ ~X86_64JniCallingConvention() OVERRIDE {}
+ // Calling convention
+ ManagedRegister ReturnRegister() OVERRIDE;
+ ManagedRegister IntReturnRegister() OVERRIDE;
+ ManagedRegister InterproceduralScratchRegister() OVERRIDE;
+ // JNI calling convention
+ size_t FrameSize() OVERRIDE;
+ size_t OutArgSize() OVERRIDE;
+ const std::vector<ManagedRegister>& CalleeSaveRegisters() const OVERRIDE {
+ return callee_save_regs_;
+ }
+ ManagedRegister ReturnScratchRegister() const OVERRIDE;
+ uint32_t CoreSpillMask() const OVERRIDE;
+ uint32_t FpSpillMask() const OVERRIDE {
+ return 0;
+ }
+ bool IsCurrentParamInRegister() OVERRIDE;
+ bool IsCurrentParamOnStack() OVERRIDE;
+ ManagedRegister CurrentParamRegister() OVERRIDE;
+ FrameOffset CurrentParamStackOffset() OVERRIDE;
+
+ protected:
+ size_t NumberOfOutgoingStackArgs() OVERRIDE;
+
+ private:
+ // TODO: these values aren't unique and can be shared amongst instances
+ std::vector<ManagedRegister> callee_save_regs_;
+
+ DISALLOW_COPY_AND_ASSIGN(X86_64JniCallingConvention);
+};
+
+} // namespace x86_64
+} // namespace art
+
+#endif // ART_COMPILER_JNI_QUICK_X86_64_CALLING_CONVENTION_X86_64_H_
diff --git a/compiler/optimizing/builder.cc b/compiler/optimizing/builder.cc
index 39535e9..64ecdb5 100644
--- a/compiler/optimizing/builder.cc
+++ b/compiler/optimizing/builder.cc
@@ -16,10 +16,12 @@
*/
#include "dex_file.h"
+#include "dex_file-inl.h"
#include "dex_instruction.h"
#include "dex_instruction-inl.h"
#include "builder.h"
#include "nodes.h"
+#include "primitive.h"
namespace art {
@@ -192,6 +194,55 @@
break;
}
+ case Instruction::INVOKE_STATIC: {
+ uint32_t method_idx = instruction.VRegB_35c();
+ const DexFile::MethodId& method_id = dex_file_->GetMethodId(method_idx);
+ uint32_t return_type_idx = dex_file_->GetProtoId(method_id.proto_idx_).return_type_idx_;
+ const char* descriptor = dex_file_->StringByTypeIdx(return_type_idx);
+ const size_t number_of_arguments = instruction.VRegA_35c();
+ if (number_of_arguments != 0) {
+ return false;
+ }
+ if (Primitive::GetType(descriptor[0]) != Primitive::kPrimVoid) {
+ return false;
+ }
+ current_block_->AddInstruction(new (arena_) HInvokeStatic(
+ arena_, number_of_arguments, dex_offset, method_idx));
+ break;
+ }
+
+ case Instruction::ADD_INT: {
+ HInstruction* first = LoadLocal(instruction.VRegB());
+ HInstruction* second = LoadLocal(instruction.VRegC());
+ current_block_->AddInstruction(new (arena_) HAdd(Primitive::kPrimInt, first, second));
+ UpdateLocal(instruction.VRegA(), current_block_->GetLastInstruction());
+ break;
+ }
+
+ case Instruction::ADD_INT_2ADDR: {
+ HInstruction* first = LoadLocal(instruction.VRegA());
+ HInstruction* second = LoadLocal(instruction.VRegB());
+ current_block_->AddInstruction(new (arena_) HAdd(Primitive::kPrimInt, first, second));
+ UpdateLocal(instruction.VRegA(), current_block_->GetLastInstruction());
+ break;
+ }
+
+ case Instruction::ADD_INT_LIT16: {
+ HInstruction* first = LoadLocal(instruction.VRegB());
+ HInstruction* second = GetConstant(instruction.VRegC_22s());
+ current_block_->AddInstruction(new (arena_) HAdd(Primitive::kPrimInt, first, second));
+ UpdateLocal(instruction.VRegA(), current_block_->GetLastInstruction());
+ break;
+ }
+
+ case Instruction::ADD_INT_LIT8: {
+ HInstruction* first = LoadLocal(instruction.VRegB());
+ HInstruction* second = GetConstant(instruction.VRegC_22b());
+ current_block_->AddInstruction(new (arena_) HAdd(Primitive::kPrimInt, first, second));
+ UpdateLocal(instruction.VRegA(), current_block_->GetLastInstruction());
+ break;
+ }
+
case Instruction::NOP:
break;
diff --git a/compiler/optimizing/builder.h b/compiler/optimizing/builder.h
index fff83a1..46ca9aa 100644
--- a/compiler/optimizing/builder.h
+++ b/compiler/optimizing/builder.h
@@ -18,6 +18,7 @@
#define ART_COMPILER_OPTIMIZING_BUILDER_H_
#include "dex_file.h"
+#include "driver/dex_compilation_unit.h"
#include "utils/allocation.h"
#include "utils/growable_array.h"
@@ -33,7 +34,9 @@
class HGraphBuilder : public ValueObject {
public:
- explicit HGraphBuilder(ArenaAllocator* arena)
+ HGraphBuilder(ArenaAllocator* arena,
+ const DexCompilationUnit* dex_compilation_unit = nullptr,
+ const DexFile* dex_file = nullptr)
: arena_(arena),
branch_targets_(arena, 0),
locals_(arena, 0),
@@ -42,7 +45,9 @@
current_block_(nullptr),
graph_(nullptr),
constant0_(nullptr),
- constant1_(nullptr) { }
+ constant1_(nullptr),
+ dex_file_(dex_file),
+ dex_compilation_unit_(dex_compilation_unit) { }
HGraph* BuildGraph(const DexFile::CodeItem& code);
@@ -83,6 +88,9 @@
HIntConstant* constant0_;
HIntConstant* constant1_;
+ const DexFile* const dex_file_;
+ const DexCompilationUnit* const dex_compilation_unit_;
+
DISALLOW_COPY_AND_ASSIGN(HGraphBuilder);
};
diff --git a/compiler/optimizing/code_generator.cc b/compiler/optimizing/code_generator.cc
index bb6ac84..b86665b 100644
--- a/compiler/optimizing/code_generator.cc
+++ b/compiler/optimizing/code_generator.cc
@@ -21,8 +21,11 @@
#include "dex/verified_method.h"
#include "driver/dex_compilation_unit.h"
#include "gc_map_builder.h"
+#include "leb128.h"
+#include "mapping_table.h"
#include "utils/assembler.h"
#include "verifier/dex_gc_map.h"
+#include "vmap_table.h"
namespace art {
@@ -120,8 +123,95 @@
dex_compilation_unit.GetVerifiedMethod()->GetDexGcMap();
verifier::DexPcToReferenceMap dex_gc_map(&(gc_map_raw)[0]);
- GcMapBuilder builder(data, 0, 0, dex_gc_map.RegWidth());
+ uint32_t max_native_offset = 0;
+ for (size_t i = 0; i < pc_infos_.Size(); i++) {
+ uint32_t native_offset = pc_infos_.Get(i).native_pc;
+ if (native_offset > max_native_offset) {
+ max_native_offset = native_offset;
+ }
+ }
+
+ GcMapBuilder builder(data, pc_infos_.Size(), max_native_offset, dex_gc_map.RegWidth());
+ for (size_t i = 0; i < pc_infos_.Size(); i++) {
+ struct PcInfo pc_info = pc_infos_.Get(i);
+ uint32_t native_offset = pc_info.native_pc;
+ uint32_t dex_pc = pc_info.dex_pc;
+ const uint8_t* references = dex_gc_map.FindBitMap(dex_pc, false);
+ CHECK(references != NULL) << "Missing ref for dex pc 0x" << std::hex << dex_pc;
+ builder.AddEntry(native_offset, references);
+ }
}
+void CodeGenerator::BuildMappingTable(std::vector<uint8_t>* data) const {
+ uint32_t pc2dex_data_size = 0u;
+ uint32_t pc2dex_entries = pc_infos_.Size();
+ uint32_t pc2dex_offset = 0u;
+ int32_t pc2dex_dalvik_offset = 0;
+ uint32_t dex2pc_data_size = 0u;
+ uint32_t dex2pc_entries = 0u;
+
+ // We currently only have pc2dex entries.
+ for (size_t i = 0; i < pc2dex_entries; i++) {
+ struct PcInfo pc_info = pc_infos_.Get(i);
+ pc2dex_data_size += UnsignedLeb128Size(pc_info.native_pc - pc2dex_offset);
+ pc2dex_data_size += SignedLeb128Size(pc_info.dex_pc - pc2dex_dalvik_offset);
+ pc2dex_offset = pc_info.native_pc;
+ pc2dex_dalvik_offset = pc_info.dex_pc;
+ }
+
+ uint32_t total_entries = pc2dex_entries + dex2pc_entries;
+ uint32_t hdr_data_size = UnsignedLeb128Size(total_entries) + UnsignedLeb128Size(pc2dex_entries);
+ uint32_t data_size = hdr_data_size + pc2dex_data_size + dex2pc_data_size;
+ data->resize(data_size);
+
+ uint8_t* data_ptr = &(*data)[0];
+ uint8_t* write_pos = data_ptr;
+ write_pos = EncodeUnsignedLeb128(write_pos, total_entries);
+ write_pos = EncodeUnsignedLeb128(write_pos, pc2dex_entries);
+ DCHECK_EQ(static_cast<size_t>(write_pos - data_ptr), hdr_data_size);
+ uint8_t* write_pos2 = write_pos + pc2dex_data_size;
+
+ pc2dex_offset = 0u;
+ pc2dex_dalvik_offset = 0u;
+ for (size_t i = 0; i < pc2dex_entries; i++) {
+ struct PcInfo pc_info = pc_infos_.Get(i);
+ DCHECK(pc2dex_offset <= pc_info.native_pc);
+ write_pos = EncodeUnsignedLeb128(write_pos, pc_info.native_pc - pc2dex_offset);
+ write_pos = EncodeSignedLeb128(write_pos, pc_info.dex_pc - pc2dex_dalvik_offset);
+ pc2dex_offset = pc_info.native_pc;
+ pc2dex_dalvik_offset = pc_info.dex_pc;
+ }
+ DCHECK_EQ(static_cast<size_t>(write_pos - data_ptr), hdr_data_size + pc2dex_data_size);
+ DCHECK_EQ(static_cast<size_t>(write_pos2 - data_ptr), data_size);
+
+ if (kIsDebugBuild) {
+ // Verify the encoded table holds the expected data.
+ MappingTable table(data_ptr);
+ CHECK_EQ(table.TotalSize(), total_entries);
+ CHECK_EQ(table.PcToDexSize(), pc2dex_entries);
+ auto it = table.PcToDexBegin();
+ auto it2 = table.DexToPcBegin();
+ for (size_t i = 0; i < pc2dex_entries; i++) {
+ struct PcInfo pc_info = pc_infos_.Get(i);
+ CHECK_EQ(pc_info.native_pc, it.NativePcOffset());
+ CHECK_EQ(pc_info.dex_pc, it.DexPc());
+ ++it;
+ }
+ CHECK(it == table.PcToDexEnd());
+ CHECK(it2 == table.DexToPcEnd());
+ }
+}
+
+void CodeGenerator::BuildVMapTable(std::vector<uint8_t>* data) const {
+ Leb128EncodingVector vmap_encoder;
+ size_t size = 1 + 1 /* marker */ + 0;
+ vmap_encoder.Reserve(size + 1u); // All values are likely to be one byte in ULEB128 (<128).
+ vmap_encoder.PushBackUnsigned(size);
+ // We're currently always saving the frame pointer, so set it in the table as a temporary.
+ vmap_encoder.PushBackUnsigned(kVRegTempBaseReg + VmapTable::kEntryAdjustment);
+ vmap_encoder.PushBackUnsigned(VmapTable::kAdjustedFpMarker);
+
+ *data = vmap_encoder.GetData();
+}
} // namespace art
diff --git a/compiler/optimizing/code_generator.h b/compiler/optimizing/code_generator.h
index 63f8cbf..24dcab6 100644
--- a/compiler/optimizing/code_generator.h
+++ b/compiler/optimizing/code_generator.h
@@ -38,6 +38,11 @@
DISALLOW_COPY_AND_ASSIGN(CodeAllocator);
};
+struct PcInfo {
+ uint32_t dex_pc;
+ uintptr_t native_pc;
+};
+
/**
* A Location is an abstraction over the potential location
* of an instruction. It could be in register or stack.
@@ -81,7 +86,8 @@
class LocationSummary : public ArenaObject {
public:
explicit LocationSummary(HInstruction* instruction)
- : inputs(instruction->GetBlock()->GetGraph()->GetArena(), instruction->InputCount()) {
+ : inputs(instruction->GetBlock()->GetGraph()->GetArena(), instruction->InputCount()),
+ temps(instruction->GetBlock()->GetGraph()->GetArena(), 0) {
inputs.SetSize(instruction->InputCount());
for (int i = 0; i < instruction->InputCount(); i++) {
inputs.Put(i, Location());
@@ -100,10 +106,19 @@
output = Location(location);
}
+ void AddTemp(Location location) {
+ temps.Add(location);
+ }
+
+ Location GetTemp(uint32_t at) const {
+ return temps.Get(at);
+ }
+
Location Out() const { return output; }
private:
GrowableArray<Location> inputs;
+ GrowableArray<Location> temps;
Location output;
DISALLOW_COPY_AND_ASSIGN(LocationSummary);
@@ -134,9 +149,17 @@
uint32_t GetFrameSize() const { return frame_size_; }
void SetFrameSize(uint32_t size) { frame_size_ = size; }
+ uint32_t GetCoreSpillMask() const { return core_spill_mask_; }
- void BuildMappingTable(std::vector<uint8_t>* vector) const { }
- void BuildVMapTable(std::vector<uint8_t>* vector) const { }
+ void RecordPcInfo(uint32_t dex_pc) {
+ struct PcInfo pc_info;
+ pc_info.dex_pc = dex_pc;
+ pc_info.native_pc = GetAssembler()->CodeSize();
+ pc_infos_.Add(pc_info);
+ }
+
+ void BuildMappingTable(std::vector<uint8_t>* vector) const;
+ void BuildVMapTable(std::vector<uint8_t>* vector) const;
void BuildNativeGCMap(
std::vector<uint8_t>* vector, const DexCompilationUnit& dex_compilation_unit) const;
@@ -144,23 +167,26 @@
explicit CodeGenerator(HGraph* graph)
: frame_size_(0),
graph_(graph),
- block_labels_(graph->GetArena(), 0) {
+ block_labels_(graph->GetArena(), 0),
+ pc_infos_(graph->GetArena(), 32) {
block_labels_.SetSize(graph->GetBlocks()->Size());
}
~CodeGenerator() { }
+ // Frame size required for this method.
+ uint32_t frame_size_;
+ uint32_t core_spill_mask_;
+
private:
void InitLocations(HInstruction* instruction);
void CompileBlock(HBasicBlock* block);
void CompileEntryBlock();
- // Frame size required for this method.
- uint32_t frame_size_;
-
HGraph* const graph_;
// Labels for each block that will be compiled.
GrowableArray<Label> block_labels_;
+ GrowableArray<PcInfo> pc_infos_;
DISALLOW_COPY_AND_ASSIGN(CodeGenerator);
};
diff --git a/compiler/optimizing/code_generator_arm.cc b/compiler/optimizing/code_generator_arm.cc
index 04bdc34..68c997b 100644
--- a/compiler/optimizing/code_generator_arm.cc
+++ b/compiler/optimizing/code_generator_arm.cc
@@ -18,17 +18,27 @@
#include "utils/assembler.h"
#include "utils/arm/assembler_arm.h"
+#include "mirror/array.h"
+#include "mirror/art_method.h"
+
#define __ reinterpret_cast<ArmAssembler*>(GetAssembler())->
namespace art {
namespace arm {
void CodeGeneratorARM::GenerateFrameEntry() {
+ core_spill_mask_ |= (1 << LR);
+ // We're currently always using FP, which is callee-saved in Quick.
+ core_spill_mask_ |= (1 << FP);
+
__ PushList((1 << FP) | (1 << LR));
__ mov(FP, ShifterOperand(SP));
- if (GetFrameSize() != 0) {
- __ AddConstant(SP, -GetFrameSize());
- }
+
+ // Add the current ART method to the frame size, the return pc, and FP.
+ SetFrameSize(RoundUp(GetFrameSize() + 3 * kWordSize, kStackAlignment));
+ // PC and FP have already been pushed on the stack.
+ __ AddConstant(SP, -(GetFrameSize() - 2 * kWordSize));
+ __ str(R0, Address(SP, 0));
}
void CodeGeneratorARM::GenerateFrameExit() {
@@ -173,5 +183,71 @@
codegen_->GenerateFrameExit();
}
+void LocationsBuilderARM::VisitInvokeStatic(HInvokeStatic* invoke) {
+ LocationSummary* locations = new (GetGraph()->GetArena()) LocationSummary(invoke);
+ CHECK_EQ(invoke->InputCount(), 0);
+ locations->AddTemp(Location(R0));
+ invoke->SetLocations(locations);
+}
+
+void InstructionCodeGeneratorARM::LoadCurrentMethod(Register reg) {
+ __ ldr(reg, Address(SP, 0));
+}
+
+void InstructionCodeGeneratorARM::VisitInvokeStatic(HInvokeStatic* invoke) {
+ Register temp = invoke->GetLocations()->GetTemp(0).reg<Register>();
+ size_t index_in_cache = mirror::Array::DataOffset(sizeof(mirror::Object*)).Int32Value() +
+ invoke->GetIndexInDexCache() * kWordSize;
+
+ // TODO: Implement all kinds of calls:
+ // 1) boot -> boot
+ // 2) app -> boot
+ // 3) app -> app
+ //
+ // Currently we implement the app -> app logic, which looks up in the resolve cache.
+
+ // temp = method;
+ LoadCurrentMethod(temp);
+ // temp = temp->dex_cache_resolved_methods_;
+ __ ldr(temp, Address(temp, mirror::ArtMethod::DexCacheResolvedMethodsOffset().Int32Value()));
+ // temp = temp[index_in_cache]
+ __ ldr(temp, Address(temp, index_in_cache));
+ // LR = temp[offset_of_quick_compiled_code]
+ __ ldr(LR, Address(temp,
+ mirror::ArtMethod::EntryPointFromQuickCompiledCodeOffset().Int32Value()));
+ // LR()
+ __ blx(LR);
+
+ codegen_->RecordPcInfo(invoke->GetDexPc());
+}
+
+void LocationsBuilderARM::VisitAdd(HAdd* add) {
+ LocationSummary* locations = new (GetGraph()->GetArena()) LocationSummary(add);
+ switch (add->GetResultType()) {
+ case Primitive::kPrimInt: {
+ locations->SetInAt(0, Location(R0));
+ locations->SetInAt(1, Location(R1));
+ locations->SetOut(Location(R0));
+ break;
+ }
+ default:
+ LOG(FATAL) << "Unimplemented";
+ }
+ add->SetLocations(locations);
+}
+
+void InstructionCodeGeneratorARM::VisitAdd(HAdd* add) {
+ LocationSummary* locations = add->GetLocations();
+ switch (add->GetResultType()) {
+ case Primitive::kPrimInt:
+ __ add(locations->Out().reg<Register>(),
+ locations->InAt(0).reg<Register>(),
+ ShifterOperand(locations->InAt(1).reg<Register>()));
+ break;
+ default:
+ LOG(FATAL) << "Unimplemented";
+ }
+}
+
} // namespace arm
} // namespace art
diff --git a/compiler/optimizing/code_generator_arm.h b/compiler/optimizing/code_generator_arm.h
index 52a7bf4..7a2835d 100644
--- a/compiler/optimizing/code_generator_arm.h
+++ b/compiler/optimizing/code_generator_arm.h
@@ -58,6 +58,7 @@
#undef DECLARE_VISIT_INSTRUCTION
Assembler* GetAssembler() const { return assembler_; }
+ void LoadCurrentMethod(Register reg);
private:
Assembler* const assembler_;
diff --git a/compiler/optimizing/code_generator_x86.cc b/compiler/optimizing/code_generator_x86.cc
index c4bda56..1764486 100644
--- a/compiler/optimizing/code_generator_x86.cc
+++ b/compiler/optimizing/code_generator_x86.cc
@@ -18,18 +18,28 @@
#include "utils/assembler.h"
#include "utils/x86/assembler_x86.h"
+#include "mirror/array.h"
+#include "mirror/art_method.h"
+
#define __ reinterpret_cast<X86Assembler*>(GetAssembler())->
namespace art {
namespace x86 {
void CodeGeneratorX86::GenerateFrameEntry() {
+ // Create a fake register to mimic Quick.
+ static const int kFakeReturnRegister = 8;
+ core_spill_mask_ |= (1 << kFakeReturnRegister);
+ // We're currently always using EBP, which is callee-saved in Quick.
+ core_spill_mask_ |= (1 << EBP);
+
__ pushl(EBP);
__ movl(EBP, ESP);
-
- if (GetFrameSize() != 0) {
- __ subl(ESP, Immediate(GetFrameSize()));
- }
+ // Add the current ART method to the frame size, the return pc, and EBP.
+ SetFrameSize(RoundUp(GetFrameSize() + 3 * kWordSize, kStackAlignment));
+ // The PC and EBP have already been pushed on the stack.
+ __ subl(ESP, Immediate(GetFrameSize() - 2 * kWordSize));
+ __ movl(Address(ESP, 0), EAX);
}
void CodeGeneratorX86::GenerateFrameExit() {
@@ -45,6 +55,10 @@
__ pushl(location.reg<Register>());
}
+void InstructionCodeGeneratorX86::LoadCurrentMethod(Register reg) {
+ __ movl(reg, Address(ESP, 0));
+}
+
void CodeGeneratorX86::Move(HInstruction* instruction, Location location) {
HIntConstant* constant = instruction->AsIntConstant();
if (constant != nullptr) {
@@ -110,7 +124,8 @@
static int32_t GetStackSlot(HLocal* local) {
// We are currently using EBP to access locals, so the offset must be negative.
- return (local->GetRegNumber() + 1) * -kWordSize;
+ // +1 for going backwards, +1 for the method pointer.
+ return (local->GetRegNumber() + 2) * -kWordSize;
}
void InstructionCodeGeneratorX86::VisitLoadLocal(HLoadLocal* load) {
@@ -172,5 +187,63 @@
__ ret();
}
+void LocationsBuilderX86::VisitInvokeStatic(HInvokeStatic* invoke) {
+ LocationSummary* locations = new (GetGraph()->GetArena()) LocationSummary(invoke);
+ CHECK_EQ(invoke->InputCount(), 0);
+ locations->AddTemp(Location(EAX));
+ invoke->SetLocations(locations);
+}
+
+void InstructionCodeGeneratorX86::VisitInvokeStatic(HInvokeStatic* invoke) {
+ Register temp = invoke->GetLocations()->GetTemp(0).reg<Register>();
+ size_t index_in_cache = mirror::Array::DataOffset(sizeof(mirror::Object*)).Int32Value() +
+ invoke->GetIndexInDexCache() * kWordSize;
+
+ // TODO: Implement all kinds of calls:
+ // 1) boot -> boot
+ // 2) app -> boot
+ // 3) app -> app
+ //
+ // Currently we implement the app -> app logic, which looks up in the resolve cache.
+
+ // temp = method;
+ LoadCurrentMethod(temp);
+ // temp = temp->dex_cache_resolved_methods_;
+ __ movl(temp, Address(temp, mirror::ArtMethod::DexCacheResolvedMethodsOffset().Int32Value()));
+ // temp = temp[index_in_cache]
+ __ movl(temp, Address(temp, index_in_cache));
+ // (temp + offset_of_quick_compiled_code)()
+ __ call(Address(temp, mirror::ArtMethod::EntryPointFromQuickCompiledCodeOffset().Int32Value()));
+
+ codegen_->RecordPcInfo(invoke->GetDexPc());
+}
+
+void LocationsBuilderX86::VisitAdd(HAdd* add) {
+ LocationSummary* locations = new (GetGraph()->GetArena()) LocationSummary(add);
+ switch (add->GetResultType()) {
+ case Primitive::kPrimInt: {
+ locations->SetInAt(0, Location(EAX));
+ locations->SetInAt(1, Location(ECX));
+ locations->SetOut(Location(EAX));
+ break;
+ }
+ default:
+ LOG(FATAL) << "Unimplemented";
+ }
+ add->SetLocations(locations);
+}
+
+void InstructionCodeGeneratorX86::VisitAdd(HAdd* add) {
+ LocationSummary* locations = add->GetLocations();
+ switch (add->GetResultType()) {
+ case Primitive::kPrimInt:
+ DCHECK_EQ(locations->InAt(0).reg<Register>(), locations->Out().reg<Register>());
+ __ addl(locations->InAt(0).reg<Register>(), locations->InAt(1).reg<Register>());
+ break;
+ default:
+ LOG(FATAL) << "Unimplemented";
+ }
+}
+
} // namespace x86
} // namespace art
diff --git a/compiler/optimizing/code_generator_x86.h b/compiler/optimizing/code_generator_x86.h
index ad2a061..505237b 100644
--- a/compiler/optimizing/code_generator_x86.h
+++ b/compiler/optimizing/code_generator_x86.h
@@ -57,6 +57,8 @@
#undef DECLARE_VISIT_INSTRUCTION
+ void LoadCurrentMethod(Register reg);
+
Assembler* GetAssembler() const { return assembler_; }
private:
diff --git a/compiler/optimizing/codegen_test.cc b/compiler/optimizing/codegen_test.cc
index ff743d8..d40990e 100644
--- a/compiler/optimizing/codegen_test.cc
+++ b/compiler/optimizing/codegen_test.cc
@@ -196,4 +196,42 @@
TestCode(data, true, 0);
}
+TEST(CodegenTest, ReturnAdd1) {
+ const uint16_t data[] = TWO_REGISTERS_CODE_ITEM(
+ Instruction::CONST_4 | 3 << 12 | 0,
+ Instruction::CONST_4 | 4 << 12 | 1 << 8,
+ Instruction::ADD_INT, 1 << 8 | 0,
+ Instruction::RETURN);
+
+ TestCode(data, true, 7);
+}
+
+TEST(CodegenTest, ReturnAdd2) {
+ const uint16_t data[] = TWO_REGISTERS_CODE_ITEM(
+ Instruction::CONST_4 | 3 << 12 | 0,
+ Instruction::CONST_4 | 4 << 12 | 1 << 8,
+ Instruction::ADD_INT_2ADDR | 1 << 12,
+ Instruction::RETURN);
+
+ TestCode(data, true, 7);
+}
+
+TEST(CodegenTest, ReturnAdd3) {
+ const uint16_t data[] = ONE_REGISTER_CODE_ITEM(
+ Instruction::CONST_4 | 4 << 12 | 0 << 8,
+ Instruction::ADD_INT_LIT8, 3 << 8 | 0,
+ Instruction::RETURN);
+
+ TestCode(data, true, 7);
+}
+
+TEST(CodegenTest, ReturnAdd4) {
+ const uint16_t data[] = ONE_REGISTER_CODE_ITEM(
+ Instruction::CONST_4 | 4 << 12 | 0 << 8,
+ Instruction::ADD_INT_LIT16, 3,
+ Instruction::RETURN);
+
+ TestCode(data, true, 7);
+}
+
} // namespace art
diff --git a/compiler/optimizing/nodes.h b/compiler/optimizing/nodes.h
index e74ed82..fc67486 100644
--- a/compiler/optimizing/nodes.h
+++ b/compiler/optimizing/nodes.h
@@ -180,11 +180,13 @@
};
#define FOR_EACH_INSTRUCTION(M) \
+ M(Add) \
M(Equal) \
M(Exit) \
M(Goto) \
M(If) \
M(IntConstant) \
+ M(InvokeStatic) \
M(LoadLocal) \
M(Local) \
M(Return) \
@@ -476,14 +478,36 @@
DISALLOW_COPY_AND_ASSIGN(HIf);
};
-// Instruction to check if two inputs are equal to each other.
-class HEqual : public HTemplateInstruction<2> {
+class HBinaryOperation : public HTemplateInstruction<2> {
public:
- HEqual(HInstruction* first, HInstruction* second) {
- SetRawInputAt(0, first);
- SetRawInputAt(1, second);
+ HBinaryOperation(Primitive::Type result_type,
+ HInstruction* left,
+ HInstruction* right) : result_type_(result_type) {
+ SetRawInputAt(0, left);
+ SetRawInputAt(1, right);
}
+ HInstruction* GetLeft() const { return InputAt(0); }
+ HInstruction* GetRight() const { return InputAt(1); }
+ Primitive::Type GetResultType() const { return result_type_; }
+
+ virtual bool IsCommutative() { return false; }
+
+ private:
+ const Primitive::Type result_type_;
+
+ DISALLOW_COPY_AND_ASSIGN(HBinaryOperation);
+};
+
+
+// Instruction to check if two inputs are equal to each other.
+class HEqual : public HBinaryOperation {
+ public:
+ HEqual(HInstruction* first, HInstruction* second)
+ : HBinaryOperation(Primitive::kPrimBoolean, first, second) {}
+
+ virtual bool IsCommutative() { return true; }
+
DECLARE_INSTRUCTION(Equal)
private:
@@ -554,6 +578,55 @@
DISALLOW_COPY_AND_ASSIGN(HIntConstant);
};
+class HInvoke : public HInstruction {
+ public:
+ HInvoke(ArenaAllocator* arena, uint32_t number_of_arguments, int32_t dex_pc)
+ : inputs_(arena, number_of_arguments),
+ dex_pc_(dex_pc) {
+ inputs_.SetSize(number_of_arguments);
+ }
+
+ virtual intptr_t InputCount() const { return inputs_.Size(); }
+ virtual HInstruction* InputAt(intptr_t i) const { return inputs_.Get(i); }
+
+ int32_t GetDexPc() const { return dex_pc_; }
+
+ protected:
+ GrowableArray<HInstruction*> inputs_;
+ const int32_t dex_pc_;
+
+ private:
+ DISALLOW_COPY_AND_ASSIGN(HInvoke);
+};
+
+class HInvokeStatic : public HInvoke {
+ public:
+ HInvokeStatic(ArenaAllocator* arena, uint32_t number_of_arguments, int32_t dex_pc, int32_t index_in_dex_cache)
+ : HInvoke(arena, number_of_arguments, dex_pc), index_in_dex_cache_(index_in_dex_cache) { }
+
+ uint32_t GetIndexInDexCache() const { return index_in_dex_cache_; }
+
+ DECLARE_INSTRUCTION(InvokeStatic)
+
+ private:
+ uint32_t index_in_dex_cache_;
+
+ DISALLOW_COPY_AND_ASSIGN(HInvokeStatic);
+};
+
+class HAdd : public HBinaryOperation {
+ public:
+ HAdd(Primitive::Type result_type, HInstruction* left, HInstruction* right)
+ : HBinaryOperation(result_type, left, right) {}
+
+ virtual bool IsCommutative() { return true; }
+
+ DECLARE_INSTRUCTION(Add);
+
+ private:
+ DISALLOW_COPY_AND_ASSIGN(HAdd);
+};
+
class HGraphVisitor : public ValueObject {
public:
explicit HGraphVisitor(HGraph* graph) : graph_(graph) { }
diff --git a/compiler/optimizing/optimizing_compiler.cc b/compiler/optimizing/optimizing_compiler.cc
index 334b185..d19c40c 100644
--- a/compiler/optimizing/optimizing_compiler.cc
+++ b/compiler/optimizing/optimizing_compiler.cc
@@ -62,17 +62,31 @@
nullptr, class_loader, art::Runtime::Current()->GetClassLinker(), dex_file, code_item,
class_def_idx, method_idx, access_flags, driver.GetVerifiedMethod(&dex_file, method_idx));
+ // For testing purposes, we put a special marker on method names that should be compiled
+ // with this compiler. This makes sure we're not regressing.
+ bool shouldCompile = dex_compilation_unit.GetSymbol().find("00024opt_00024") != std::string::npos;
+
ArenaPool pool;
ArenaAllocator arena(&pool);
- HGraphBuilder builder(&arena);
+ HGraphBuilder builder(&arena, &dex_compilation_unit, &dex_file);
HGraph* graph = builder.BuildGraph(*code_item);
if (graph == nullptr) {
+ if (shouldCompile) {
+ LOG(FATAL) << "Could not build graph in optimizing compiler";
+ }
return nullptr;
}
InstructionSet instruction_set = driver.GetInstructionSet();
+ // The optimizing compiler currently does not have a Thumb2 assembler.
+ if (instruction_set == kThumb2) {
+ instruction_set = kArm;
+ }
CodeGenerator* codegen = CodeGenerator::Create(&arena, graph, instruction_set);
if (codegen == nullptr) {
+ if (shouldCompile) {
+ LOG(FATAL) << "Could not find code generator for optimizing compiler";
+ }
return nullptr;
}
@@ -90,7 +104,7 @@
instruction_set,
allocator.GetMemory(),
codegen->GetFrameSize(),
- 0, /* GPR spill mask, unused */
+ codegen->GetCoreSpillMask(),
0, /* FPR spill mask, unused */
mapping_table,
vmap_table,
diff --git a/compiler/utils/arm/assembler_arm.cc b/compiler/utils/arm/assembler_arm.cc
index dbd078a..872a557 100644
--- a/compiler/utils/arm/assembler_arm.cc
+++ b/compiler/utils/arm/assembler_arm.cc
@@ -1438,7 +1438,7 @@
void ArmAssembler::BuildFrame(size_t frame_size, ManagedRegister method_reg,
const std::vector<ManagedRegister>& callee_save_regs,
- const std::vector<ManagedRegister>& entry_spills) {
+ const ManagedRegisterEntrySpills& entry_spills) {
CHECK_ALIGNED(frame_size, kStackAlignment);
CHECK_EQ(R0, method_reg.AsArm().AsCoreRegister());
diff --git a/compiler/utils/arm/assembler_arm.h b/compiler/utils/arm/assembler_arm.h
index 757a8a2..bb9207c 100644
--- a/compiler/utils/arm/assembler_arm.h
+++ b/compiler/utils/arm/assembler_arm.h
@@ -440,7 +440,7 @@
// Emit code that will create an activation on the stack
virtual void BuildFrame(size_t frame_size, ManagedRegister method_reg,
const std::vector<ManagedRegister>& callee_save_regs,
- const std::vector<ManagedRegister>& entry_spills);
+ const ManagedRegisterEntrySpills& entry_spills);
// Emit code that will remove an activation from the stack
virtual void RemoveFrame(size_t frame_size,
diff --git a/compiler/utils/arm64/assembler_arm64.cc b/compiler/utils/arm64/assembler_arm64.cc
index 00ce923..f8b91d7 100644
--- a/compiler/utils/arm64/assembler_arm64.cc
+++ b/compiler/utils/arm64/assembler_arm64.cc
@@ -577,7 +577,7 @@
void Arm64Assembler::BuildFrame(size_t frame_size, ManagedRegister method_reg,
const std::vector<ManagedRegister>& callee_save_regs,
- const std::vector<ManagedRegister>& entry_spills) {
+ const ManagedRegisterEntrySpills& entry_spills) {
CHECK_ALIGNED(frame_size, kStackAlignment);
CHECK(X0 == method_reg.AsArm64().AsCoreRegister());
diff --git a/compiler/utils/arm64/assembler_arm64.h b/compiler/utils/arm64/assembler_arm64.h
index 1c47e77..44eb6ff 100644
--- a/compiler/utils/arm64/assembler_arm64.h
+++ b/compiler/utils/arm64/assembler_arm64.h
@@ -111,7 +111,7 @@
// Emit code that will create an activation on the stack.
void BuildFrame(size_t frame_size, ManagedRegister method_reg,
const std::vector<ManagedRegister>& callee_save_regs,
- const std::vector<ManagedRegister>& entry_spills);
+ const ManagedRegisterEntrySpills& entry_spills);
// Emit code that will remove an activation from the stack.
void RemoveFrame(size_t frame_size,
diff --git a/compiler/utils/assembler.cc b/compiler/utils/assembler.cc
index a7cb278..1921b28 100644
--- a/compiler/utils/assembler.cc
+++ b/compiler/utils/assembler.cc
@@ -23,6 +23,7 @@
#include "arm64/assembler_arm64.h"
#include "mips/assembler_mips.h"
#include "x86/assembler_x86.h"
+#include "x86_64/assembler_x86_64.h"
#include "globals.h"
#include "memory_region.h"
@@ -111,9 +112,10 @@
return new arm64::Arm64Assembler();
case kMips:
return new mips::MipsAssembler();
- case kX86: // Fall-through.
- case kX86_64:
+ case kX86:
return new x86::X86Assembler();
+ case kX86_64:
+ return new x86_64::X86_64Assembler();
default:
LOG(FATAL) << "Unknown InstructionSet: " << instruction_set;
return NULL;
diff --git a/compiler/utils/assembler.h b/compiler/utils/assembler.h
index cd4fc12..c23fd44 100644
--- a/compiler/utils/assembler.h
+++ b/compiler/utils/assembler.h
@@ -24,6 +24,7 @@
#include "arm/constants_arm.h"
#include "mips/constants_mips.h"
#include "x86/constants_x86.h"
+#include "x86_64/constants_x86_64.h"
#include "instruction_set.h"
#include "managed_register.h"
#include "memory_region.h"
@@ -47,6 +48,26 @@
namespace x86 {
class X86Assembler;
}
+namespace x86_64 {
+ class X86_64Assembler;
+}
+
+class ExternalLabel {
+ public:
+ ExternalLabel(const char* name, uword address)
+ : name_(name), address_(address) {
+ DCHECK(name != nullptr);
+ }
+
+ const char* name() const { return name_; }
+ uword address() const {
+ return address_;
+ }
+
+ private:
+ const char* name_;
+ const uword address_;
+};
class Label {
public:
@@ -95,6 +116,7 @@
friend class arm::ArmAssembler;
friend class mips::MipsAssembler;
friend class x86::X86Assembler;
+ friend class x86_64::X86_64Assembler;
DISALLOW_COPY_AND_ASSIGN(Label);
};
@@ -335,7 +357,7 @@
// Emit code that will create an activation on the stack
virtual void BuildFrame(size_t frame_size, ManagedRegister method_reg,
const std::vector<ManagedRegister>& callee_save_regs,
- const std::vector<ManagedRegister>& entry_spills) = 0;
+ const ManagedRegisterEntrySpills& entry_spills) = 0;
// Emit code that will remove an activation from the stack
virtual void RemoveFrame(size_t frame_size,
diff --git a/compiler/utils/managed_register.h b/compiler/utils/managed_register.h
index 04c9723..f007d28 100644
--- a/compiler/utils/managed_register.h
+++ b/compiler/utils/managed_register.h
@@ -17,6 +17,8 @@
#ifndef ART_COMPILER_UTILS_MANAGED_REGISTER_H_
#define ART_COMPILER_UTILS_MANAGED_REGISTER_H_
+#include <vector>
+
namespace art {
namespace arm {
@@ -28,10 +30,15 @@
namespace mips {
class MipsManagedRegister;
}
+
namespace x86 {
class X86ManagedRegister;
}
+namespace x86_64 {
+class X86_64ManagedRegister;
+}
+
class ManagedRegister {
public:
// ManagedRegister is a value class. There exists no method to change the
@@ -48,6 +55,7 @@
arm64::Arm64ManagedRegister AsArm64() const;
mips::MipsManagedRegister AsMips() const;
x86::X86ManagedRegister AsX86() const;
+ x86_64::X86_64ManagedRegister AsX86_64() const;
// It is valid to invoke Equals on and with a NoRegister.
bool Equals(const ManagedRegister& other) const {
@@ -71,6 +79,44 @@
int id_;
};
+class ManagedRegisterSpill : public ManagedRegister {
+ public:
+ // ManagedRegisterSpill contains information about data type size and location in caller frame
+ // These additional attributes could be defined by calling convention (EntrySpills)
+ ManagedRegisterSpill(const ManagedRegister& other, uint32_t size, uint32_t spill_offset)
+ : ManagedRegister(other), size_(size), spill_offset_(spill_offset) { }
+
+ explicit ManagedRegisterSpill(const ManagedRegister& other)
+ : ManagedRegister(other), size_(-1), spill_offset_(-1) { }
+
+ int32_t getSpillOffset() {
+ return spill_offset_;
+ }
+
+ int32_t getSize() {
+ return size_;
+ }
+
+ private:
+ int32_t size_;
+ int32_t spill_offset_;
+};
+
+class ManagedRegisterEntrySpills : public std::vector<ManagedRegisterSpill> {
+ public:
+ // The ManagedRegister does not have information about size and offset.
+ // In this case it's size and offset determined by BuildFrame (assembler)
+ void push_back(ManagedRegister __x) {
+ ManagedRegisterSpill spill(__x);
+ std::vector<ManagedRegisterSpill>::push_back(spill);
+ }
+
+ void push_back(ManagedRegisterSpill __x) {
+ std::vector<ManagedRegisterSpill>::push_back(__x);
+ }
+ private:
+};
+
} // namespace art
#endif // ART_COMPILER_UTILS_MANAGED_REGISTER_H_
diff --git a/compiler/utils/mips/assembler_mips.cc b/compiler/utils/mips/assembler_mips.cc
index ce21b84..dfd3306 100644
--- a/compiler/utils/mips/assembler_mips.cc
+++ b/compiler/utils/mips/assembler_mips.cc
@@ -538,7 +538,7 @@
void MipsAssembler::BuildFrame(size_t frame_size, ManagedRegister method_reg,
const std::vector<ManagedRegister>& callee_save_regs,
- const std::vector<ManagedRegister>& entry_spills) {
+ const ManagedRegisterEntrySpills& entry_spills) {
CHECK_ALIGNED(frame_size, kStackAlignment);
// Increase frame to required size.
diff --git a/compiler/utils/mips/assembler_mips.h b/compiler/utils/mips/assembler_mips.h
index 0f5f2fe..0d1a94c 100644
--- a/compiler/utils/mips/assembler_mips.h
+++ b/compiler/utils/mips/assembler_mips.h
@@ -357,7 +357,7 @@
// Emit code that will create an activation on the stack
virtual void BuildFrame(size_t frame_size, ManagedRegister method_reg,
const std::vector<ManagedRegister>& callee_save_regs,
- const std::vector<ManagedRegister>& entry_spills);
+ const ManagedRegisterEntrySpills& entry_spills);
// Emit code that will remove an activation from the stack
virtual void RemoveFrame(size_t frame_size,
diff --git a/compiler/utils/scoped_arena_allocator.cc b/compiler/utils/scoped_arena_allocator.cc
index a78d287..bd78eae 100644
--- a/compiler/utils/scoped_arena_allocator.cc
+++ b/compiler/utils/scoped_arena_allocator.cc
@@ -99,6 +99,7 @@
}
CurrentStats()->RecordAlloc(bytes, kind);
top_ptr_ = ptr + rounded_bytes;
+ VALGRIND_MAKE_MEM_UNDEFINED(ptr, bytes);
VALGRIND_MAKE_MEM_NOACCESS(ptr + bytes, rounded_bytes - bytes);
return ptr;
}
diff --git a/compiler/utils/x86/assembler_x86.cc b/compiler/utils/x86/assembler_x86.cc
index db8956d..ebbb43a 100644
--- a/compiler/utils/x86/assembler_x86.cc
+++ b/compiler/utils/x86/assembler_x86.cc
@@ -54,6 +54,16 @@
}
+void X86Assembler::call(const ExternalLabel& label) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ intptr_t call_start = buffer_.GetPosition();
+ EmitUint8(0xE8);
+ EmitInt32(label.address());
+ static const intptr_t kCallExternalLabelSize = 5;
+ DCHECK_EQ((buffer_.GetPosition() - call_start), kCallExternalLabelSize);
+}
+
+
void X86Assembler::pushl(Register reg) {
AssemblerBuffer::EnsureCapacity ensured(&buffer_);
EmitUint8(0x50 + reg);
@@ -1388,7 +1398,7 @@
void X86Assembler::BuildFrame(size_t frame_size, ManagedRegister method_reg,
const std::vector<ManagedRegister>& spill_regs,
- const std::vector<ManagedRegister>& entry_spills) {
+ const ManagedRegisterEntrySpills& entry_spills) {
CHECK_ALIGNED(frame_size, kStackAlignment);
for (int i = spill_regs.size() - 1; i >= 0; --i) {
pushl(spill_regs.at(i).AsX86().AsCpuRegister());
diff --git a/compiler/utils/x86/assembler_x86.h b/compiler/utils/x86/assembler_x86.h
index e284d8c..f906a6f 100644
--- a/compiler/utils/x86/assembler_x86.h
+++ b/compiler/utils/x86/assembler_x86.h
@@ -227,6 +227,7 @@
void call(Register reg);
void call(const Address& address);
void call(Label* label);
+ void call(const ExternalLabel& label);
void pushl(Register reg);
void pushl(const Address& address);
@@ -466,7 +467,7 @@
// Emit code that will create an activation on the stack
virtual void BuildFrame(size_t frame_size, ManagedRegister method_reg,
const std::vector<ManagedRegister>& callee_save_regs,
- const std::vector<ManagedRegister>& entry_spills);
+ const ManagedRegisterEntrySpills& entry_spills);
// Emit code that will remove an activation from the stack
virtual void RemoveFrame(size_t frame_size,
diff --git a/compiler/utils/x86_64/assembler_x86_64.cc b/compiler/utils/x86_64/assembler_x86_64.cc
new file mode 100644
index 0000000..fa302c9
--- /dev/null
+++ b/compiler/utils/x86_64/assembler_x86_64.cc
@@ -0,0 +1,1951 @@
+/*
+ * Copyright (C) 2014 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.
+ */
+
+#include "assembler_x86_64.h"
+
+#include "base/casts.h"
+#include "entrypoints/quick/quick_entrypoints.h"
+#include "memory_region.h"
+#include "thread.h"
+
+namespace art {
+namespace x86_64 {
+
+std::ostream& operator<<(std::ostream& os, const XmmRegister& reg) {
+ return os << "XMM" << static_cast<int>(reg);
+}
+
+std::ostream& operator<<(std::ostream& os, const X87Register& reg) {
+ return os << "ST" << static_cast<int>(reg);
+}
+
+void X86_64Assembler::call(Register reg) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xFF);
+ EmitRegisterOperand(2, reg);
+}
+
+
+void X86_64Assembler::call(const Address& address) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xFF);
+ EmitOperand(2, address);
+}
+
+
+void X86_64Assembler::call(Label* label) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xE8);
+ static const int kSize = 5;
+ EmitLabel(label, kSize);
+}
+
+
+void X86_64Assembler::pushq(Register reg) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ rex_rm(reg);
+ EmitUint8(0x50 + reg);
+}
+
+
+void X86_64Assembler::pushq(const Address& address) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xFF);
+ EmitOperand(6, address);
+}
+
+
+void X86_64Assembler::pushq(const Immediate& imm) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ if (imm.is_int8()) {
+ EmitUint8(0x6A);
+ EmitUint8(imm.value() & 0xFF);
+ } else {
+ EmitUint8(0x68);
+ EmitImmediate(imm);
+ }
+}
+
+
+void X86_64Assembler::popq(Register reg) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ rex_rm(reg);
+ EmitUint8(0x58 + reg);
+}
+
+
+void X86_64Assembler::popq(const Address& address) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x8F);
+ EmitOperand(0, address);
+}
+
+
+void X86_64Assembler::movq(Register dst, const Immediate& imm) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x48); // REX.W
+ EmitUint8(0xB8 + dst);
+ EmitImmediate(imm);
+}
+
+
+void X86_64Assembler::movl(Register dst, const Immediate& imm) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xB8 + dst);
+ EmitImmediate(imm);
+}
+
+
+void X86_64Assembler::movq(Register dst, Register src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x48); // REX.W
+ EmitUint8(0x89);
+ EmitRegisterOperand(src, dst);
+}
+
+
+void X86_64Assembler::movl(Register dst, Register src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x89);
+ EmitRegisterOperand(src, dst);
+}
+
+
+void X86_64Assembler::movq(Register dst, const Address& src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ rex_reg(dst, 8);
+ EmitUint8(0x8B);
+ EmitOperand(dst, src);
+}
+
+
+void X86_64Assembler::movl(Register dst, const Address& src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ rex_reg(dst, 4);
+ EmitUint8(0x8B);
+ EmitOperand(dst, src);
+}
+
+
+void X86_64Assembler::movq(const Address& dst, Register src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ rex_reg(src, 8);
+ EmitUint8(0x89);
+ EmitOperand(src, dst);
+}
+
+
+void X86_64Assembler::movl(const Address& dst, Register src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ rex_reg(src, 4);
+ EmitUint8(0x89);
+ EmitOperand(src, dst);
+}
+
+
+void X86_64Assembler::movl(const Address& dst, const Immediate& imm) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xC7);
+ EmitOperand(0, dst);
+ EmitImmediate(imm);
+}
+
+void X86_64Assembler::movl(const Address& dst, Label* lbl) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xC7);
+ EmitOperand(0, dst);
+ EmitLabel(lbl, dst.length_ + 5);
+}
+
+void X86_64Assembler::movzxb(Register dst, ByteRegister src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x0F);
+ EmitUint8(0xB6);
+ EmitRegisterOperand(dst, src);
+}
+
+
+void X86_64Assembler::movzxb(Register dst, const Address& src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x0F);
+ EmitUint8(0xB6);
+ EmitOperand(dst, src);
+}
+
+
+void X86_64Assembler::movsxb(Register dst, ByteRegister src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x0F);
+ EmitUint8(0xBE);
+ EmitRegisterOperand(dst, src);
+}
+
+
+void X86_64Assembler::movsxb(Register dst, const Address& src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x0F);
+ EmitUint8(0xBE);
+ EmitOperand(dst, src);
+}
+
+
+void X86_64Assembler::movb(Register /*dst*/, const Address& /*src*/) {
+ LOG(FATAL) << "Use movzxb or movsxb instead.";
+}
+
+
+void X86_64Assembler::movb(const Address& dst, ByteRegister src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x88);
+ EmitOperand(src, dst);
+}
+
+
+void X86_64Assembler::movb(const Address& dst, const Immediate& imm) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xC6);
+ EmitOperand(RAX, dst);
+ CHECK(imm.is_int8());
+ EmitUint8(imm.value() & 0xFF);
+}
+
+
+void X86_64Assembler::movzxw(Register dst, Register src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x0F);
+ EmitUint8(0xB7);
+ EmitRegisterOperand(dst, src);
+}
+
+
+void X86_64Assembler::movzxw(Register dst, const Address& src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x0F);
+ EmitUint8(0xB7);
+ EmitOperand(dst, src);
+}
+
+
+void X86_64Assembler::movsxw(Register dst, Register src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x0F);
+ EmitUint8(0xBF);
+ EmitRegisterOperand(dst, src);
+}
+
+
+void X86_64Assembler::movsxw(Register dst, const Address& src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x0F);
+ EmitUint8(0xBF);
+ EmitOperand(dst, src);
+}
+
+
+void X86_64Assembler::movw(Register /*dst*/, const Address& /*src*/) {
+ LOG(FATAL) << "Use movzxw or movsxw instead.";
+}
+
+
+void X86_64Assembler::movw(const Address& dst, Register src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitOperandSizeOverride();
+ EmitUint8(0x89);
+ EmitOperand(src, dst);
+}
+
+
+void X86_64Assembler::leaq(Register dst, const Address& src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ rex_reg(dst, 8);
+ EmitUint8(0x8D);
+ EmitOperand(dst, src);
+}
+
+
+void X86_64Assembler::cmovl(Condition condition, Register dst, Register src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x0F);
+ EmitUint8(0x40 + condition);
+ EmitRegisterOperand(dst, src);
+}
+
+
+void X86_64Assembler::setb(Condition condition, Register dst) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x0F);
+ EmitUint8(0x90 + condition);
+ EmitOperand(0, Operand(dst));
+}
+
+
+void X86_64Assembler::movss(XmmRegister dst, const Address& src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF3);
+ EmitUint8(0x0F);
+ EmitUint8(0x10);
+ EmitOperand(dst, src);
+}
+
+
+void X86_64Assembler::movss(const Address& dst, XmmRegister src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF3);
+ EmitUint8(0x0F);
+ EmitUint8(0x11);
+ EmitOperand(src, dst);
+}
+
+
+void X86_64Assembler::movss(XmmRegister dst, XmmRegister src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF3);
+ EmitUint8(0x0F);
+ EmitUint8(0x11);
+ EmitXmmRegisterOperand(src, dst);
+}
+
+
+void X86_64Assembler::movd(XmmRegister dst, Register src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x66);
+ EmitUint8(0x0F);
+ EmitUint8(0x6E);
+ EmitOperand(dst, Operand(src));
+}
+
+
+void X86_64Assembler::movd(Register dst, XmmRegister src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x66);
+ EmitUint8(0x0F);
+ EmitUint8(0x7E);
+ EmitOperand(src, Operand(dst));
+}
+
+
+void X86_64Assembler::addss(XmmRegister dst, XmmRegister src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF3);
+ EmitUint8(0x0F);
+ EmitUint8(0x58);
+ EmitXmmRegisterOperand(dst, src);
+}
+
+
+void X86_64Assembler::addss(XmmRegister dst, const Address& src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF3);
+ EmitUint8(0x0F);
+ EmitUint8(0x58);
+ EmitOperand(dst, src);
+}
+
+
+void X86_64Assembler::subss(XmmRegister dst, XmmRegister src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF3);
+ EmitUint8(0x0F);
+ EmitUint8(0x5C);
+ EmitXmmRegisterOperand(dst, src);
+}
+
+
+void X86_64Assembler::subss(XmmRegister dst, const Address& src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF3);
+ EmitUint8(0x0F);
+ EmitUint8(0x5C);
+ EmitOperand(dst, src);
+}
+
+
+void X86_64Assembler::mulss(XmmRegister dst, XmmRegister src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF3);
+ EmitUint8(0x0F);
+ EmitUint8(0x59);
+ EmitXmmRegisterOperand(dst, src);
+}
+
+
+void X86_64Assembler::mulss(XmmRegister dst, const Address& src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF3);
+ EmitUint8(0x0F);
+ EmitUint8(0x59);
+ EmitOperand(dst, src);
+}
+
+
+void X86_64Assembler::divss(XmmRegister dst, XmmRegister src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF3);
+ EmitUint8(0x0F);
+ EmitUint8(0x5E);
+ EmitXmmRegisterOperand(dst, src);
+}
+
+
+void X86_64Assembler::divss(XmmRegister dst, const Address& src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF3);
+ EmitUint8(0x0F);
+ EmitUint8(0x5E);
+ EmitOperand(dst, src);
+}
+
+
+void X86_64Assembler::flds(const Address& src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xD9);
+ EmitOperand(0, src);
+}
+
+
+void X86_64Assembler::fstps(const Address& dst) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xD9);
+ EmitOperand(3, dst);
+}
+
+
+void X86_64Assembler::movsd(XmmRegister dst, const Address& src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF2);
+ EmitUint8(0x0F);
+ EmitUint8(0x10);
+ EmitOperand(dst, src);
+}
+
+
+void X86_64Assembler::movsd(const Address& dst, XmmRegister src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF2);
+ EmitUint8(0x0F);
+ EmitUint8(0x11);
+ EmitOperand(src, dst);
+}
+
+
+void X86_64Assembler::movsd(XmmRegister dst, XmmRegister src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF2);
+ EmitUint8(0x0F);
+ EmitUint8(0x11);
+ EmitXmmRegisterOperand(src, dst);
+}
+
+
+void X86_64Assembler::addsd(XmmRegister dst, XmmRegister src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF2);
+ EmitUint8(0x0F);
+ EmitUint8(0x58);
+ EmitXmmRegisterOperand(dst, src);
+}
+
+
+void X86_64Assembler::addsd(XmmRegister dst, const Address& src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF2);
+ EmitUint8(0x0F);
+ EmitUint8(0x58);
+ EmitOperand(dst, src);
+}
+
+
+void X86_64Assembler::subsd(XmmRegister dst, XmmRegister src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF2);
+ EmitUint8(0x0F);
+ EmitUint8(0x5C);
+ EmitXmmRegisterOperand(dst, src);
+}
+
+
+void X86_64Assembler::subsd(XmmRegister dst, const Address& src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF2);
+ EmitUint8(0x0F);
+ EmitUint8(0x5C);
+ EmitOperand(dst, src);
+}
+
+
+void X86_64Assembler::mulsd(XmmRegister dst, XmmRegister src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF2);
+ EmitUint8(0x0F);
+ EmitUint8(0x59);
+ EmitXmmRegisterOperand(dst, src);
+}
+
+
+void X86_64Assembler::mulsd(XmmRegister dst, const Address& src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF2);
+ EmitUint8(0x0F);
+ EmitUint8(0x59);
+ EmitOperand(dst, src);
+}
+
+
+void X86_64Assembler::divsd(XmmRegister dst, XmmRegister src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF2);
+ EmitUint8(0x0F);
+ EmitUint8(0x5E);
+ EmitXmmRegisterOperand(dst, src);
+}
+
+
+void X86_64Assembler::divsd(XmmRegister dst, const Address& src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF2);
+ EmitUint8(0x0F);
+ EmitUint8(0x5E);
+ EmitOperand(dst, src);
+}
+
+
+void X86_64Assembler::cvtsi2ss(XmmRegister dst, Register src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF3);
+ EmitUint8(0x0F);
+ EmitUint8(0x2A);
+ EmitOperand(dst, Operand(src));
+}
+
+
+void X86_64Assembler::cvtsi2sd(XmmRegister dst, Register src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF2);
+ EmitUint8(0x0F);
+ EmitUint8(0x2A);
+ EmitOperand(dst, Operand(src));
+}
+
+
+void X86_64Assembler::cvtss2si(Register dst, XmmRegister src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF3);
+ EmitUint8(0x0F);
+ EmitUint8(0x2D);
+ EmitXmmRegisterOperand(dst, src);
+}
+
+
+void X86_64Assembler::cvtss2sd(XmmRegister dst, XmmRegister src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF3);
+ EmitUint8(0x0F);
+ EmitUint8(0x5A);
+ EmitXmmRegisterOperand(dst, src);
+}
+
+
+void X86_64Assembler::cvtsd2si(Register dst, XmmRegister src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF2);
+ EmitUint8(0x0F);
+ EmitUint8(0x2D);
+ EmitXmmRegisterOperand(dst, src);
+}
+
+
+void X86_64Assembler::cvttss2si(Register dst, XmmRegister src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF3);
+ EmitUint8(0x0F);
+ EmitUint8(0x2C);
+ EmitXmmRegisterOperand(dst, src);
+}
+
+
+void X86_64Assembler::cvttsd2si(Register dst, XmmRegister src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF2);
+ EmitUint8(0x0F);
+ EmitUint8(0x2C);
+ EmitXmmRegisterOperand(dst, src);
+}
+
+
+void X86_64Assembler::cvtsd2ss(XmmRegister dst, XmmRegister src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF2);
+ EmitUint8(0x0F);
+ EmitUint8(0x5A);
+ EmitXmmRegisterOperand(dst, src);
+}
+
+
+void X86_64Assembler::cvtdq2pd(XmmRegister dst, XmmRegister src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF3);
+ EmitUint8(0x0F);
+ EmitUint8(0xE6);
+ EmitXmmRegisterOperand(dst, src);
+}
+
+
+void X86_64Assembler::comiss(XmmRegister a, XmmRegister b) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x0F);
+ EmitUint8(0x2F);
+ EmitXmmRegisterOperand(a, b);
+}
+
+
+void X86_64Assembler::comisd(XmmRegister a, XmmRegister b) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x66);
+ EmitUint8(0x0F);
+ EmitUint8(0x2F);
+ EmitXmmRegisterOperand(a, b);
+}
+
+
+void X86_64Assembler::sqrtsd(XmmRegister dst, XmmRegister src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF2);
+ EmitUint8(0x0F);
+ EmitUint8(0x51);
+ EmitXmmRegisterOperand(dst, src);
+}
+
+
+void X86_64Assembler::sqrtss(XmmRegister dst, XmmRegister src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF3);
+ EmitUint8(0x0F);
+ EmitUint8(0x51);
+ EmitXmmRegisterOperand(dst, src);
+}
+
+
+void X86_64Assembler::xorpd(XmmRegister dst, const Address& src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x66);
+ EmitUint8(0x0F);
+ EmitUint8(0x57);
+ EmitOperand(dst, src);
+}
+
+
+void X86_64Assembler::xorpd(XmmRegister dst, XmmRegister src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x66);
+ EmitUint8(0x0F);
+ EmitUint8(0x57);
+ EmitXmmRegisterOperand(dst, src);
+}
+
+
+void X86_64Assembler::xorps(XmmRegister dst, const Address& src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x0F);
+ EmitUint8(0x57);
+ EmitOperand(dst, src);
+}
+
+
+void X86_64Assembler::xorps(XmmRegister dst, XmmRegister src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x0F);
+ EmitUint8(0x57);
+ EmitXmmRegisterOperand(dst, src);
+}
+
+
+void X86_64Assembler::andpd(XmmRegister dst, const Address& src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x66);
+ EmitUint8(0x0F);
+ EmitUint8(0x54);
+ EmitOperand(dst, src);
+}
+
+
+void X86_64Assembler::fldl(const Address& src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xDD);
+ EmitOperand(0, src);
+}
+
+
+void X86_64Assembler::fstpl(const Address& dst) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xDD);
+ EmitOperand(3, dst);
+}
+
+
+void X86_64Assembler::fnstcw(const Address& dst) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xD9);
+ EmitOperand(7, dst);
+}
+
+
+void X86_64Assembler::fldcw(const Address& src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xD9);
+ EmitOperand(5, src);
+}
+
+
+void X86_64Assembler::fistpl(const Address& dst) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xDF);
+ EmitOperand(7, dst);
+}
+
+
+void X86_64Assembler::fistps(const Address& dst) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xDB);
+ EmitOperand(3, dst);
+}
+
+
+void X86_64Assembler::fildl(const Address& src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xDF);
+ EmitOperand(5, src);
+}
+
+
+void X86_64Assembler::fincstp() {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xD9);
+ EmitUint8(0xF7);
+}
+
+
+void X86_64Assembler::ffree(const Immediate& index) {
+ CHECK_LT(index.value(), 7);
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xDD);
+ EmitUint8(0xC0 + index.value());
+}
+
+
+void X86_64Assembler::fsin() {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xD9);
+ EmitUint8(0xFE);
+}
+
+
+void X86_64Assembler::fcos() {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xD9);
+ EmitUint8(0xFF);
+}
+
+
+void X86_64Assembler::fptan() {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xD9);
+ EmitUint8(0xF2);
+}
+
+
+void X86_64Assembler::xchgl(Register dst, Register src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x87);
+ EmitRegisterOperand(dst, src);
+}
+
+void X86_64Assembler::xchgl(Register reg, const Address& address) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x87);
+ EmitOperand(reg, address);
+}
+
+
+void X86_64Assembler::cmpl(Register reg, const Immediate& imm) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitComplex(7, Operand(reg), imm);
+}
+
+
+void X86_64Assembler::cmpl(Register reg0, Register reg1) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x3B);
+ EmitOperand(reg0, Operand(reg1));
+}
+
+
+void X86_64Assembler::cmpl(Register reg, const Address& address) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x3B);
+ EmitOperand(reg, address);
+}
+
+
+void X86_64Assembler::addl(Register dst, Register src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x03);
+ EmitRegisterOperand(dst, src);
+}
+
+
+void X86_64Assembler::addl(Register reg, const Address& address) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x03);
+ EmitOperand(reg, address);
+}
+
+
+void X86_64Assembler::cmpl(const Address& address, Register reg) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x39);
+ EmitOperand(reg, address);
+}
+
+
+void X86_64Assembler::cmpl(const Address& address, const Immediate& imm) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitComplex(7, address, imm);
+}
+
+
+void X86_64Assembler::testl(Register reg1, Register reg2) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ rex(reg1, reg2, 4);
+ EmitUint8(0x85);
+ EmitRegisterOperand(reg1, reg2);
+}
+
+
+void X86_64Assembler::testl(Register reg, const Immediate& immediate) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ // For registers that have a byte variant (RAX, RBX, RCX, and RDX)
+ // we only test the byte register to keep the encoding short.
+ if (immediate.is_uint8() && reg < 4) {
+ // Use zero-extended 8-bit immediate.
+ if (reg == RAX) {
+ EmitUint8(0xA8);
+ } else {
+ EmitUint8(0xF6);
+ EmitUint8(0xC0 + reg);
+ }
+ EmitUint8(immediate.value() & 0xFF);
+ } else if (reg == RAX) {
+ // Use short form if the destination is RAX.
+ EmitUint8(0xA9);
+ EmitImmediate(immediate);
+ } else {
+ EmitUint8(0xF7);
+ EmitOperand(0, Operand(reg));
+ EmitImmediate(immediate);
+ }
+}
+
+
+void X86_64Assembler::andl(Register dst, Register src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x23);
+ EmitOperand(dst, Operand(src));
+}
+
+
+void X86_64Assembler::andl(Register dst, const Immediate& imm) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitComplex(4, Operand(dst), imm);
+}
+
+
+void X86_64Assembler::orl(Register dst, Register src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x0B);
+ EmitOperand(dst, Operand(src));
+}
+
+
+void X86_64Assembler::orl(Register dst, const Immediate& imm) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitComplex(1, Operand(dst), imm);
+}
+
+
+void X86_64Assembler::xorl(Register dst, Register src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ rex(dst, src, 4);
+ EmitUint8(0x33);
+ EmitOperand(dst, Operand(src));
+}
+
+void X86_64Assembler::rex_reg(Register &dst, size_t size) {
+ Register src = kNoRegister;
+ rex(dst, src, size);
+}
+
+void X86_64Assembler::rex_rm(Register &src, size_t size) {
+ Register dst = kNoRegister;
+ rex(dst, src, size);
+}
+
+void X86_64Assembler::rex(Register &dst, Register &src, size_t size) {
+ uint8_t rex = 0;
+ // REX.WRXB
+ // W - 64-bit operand
+ // R - MODRM.reg
+ // X - SIB.index
+ // B - MODRM.rm/SIB.base
+ if (size == 8) {
+ rex |= 0x48; // REX.W000
+ }
+ if (dst >= Register::R8 && dst < Register::kNumberOfCpuRegisters) {
+ rex |= 0x44; // REX.0R00
+ dst = static_cast<Register>(dst - 8);
+ }
+ if (src >= Register::R8 && src < Register::kNumberOfCpuRegisters) {
+ rex |= 0x41; // REX.000B
+ src = static_cast<Register>(src - 8);
+ }
+ if (rex != 0) {
+ EmitUint8(rex);
+ }
+}
+
+void X86_64Assembler::addl(Register reg, const Immediate& imm) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitComplex(0, Operand(reg), imm);
+}
+
+
+void X86_64Assembler::addq(Register reg, const Immediate& imm) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x48); // REX.W
+ EmitComplex(0, Operand(reg), imm);
+}
+
+
+void X86_64Assembler::addl(const Address& address, Register reg) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x01);
+ EmitOperand(reg, address);
+}
+
+
+void X86_64Assembler::addl(const Address& address, const Immediate& imm) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitComplex(0, address, imm);
+}
+
+
+void X86_64Assembler::adcl(Register reg, const Immediate& imm) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitComplex(2, Operand(reg), imm);
+}
+
+
+void X86_64Assembler::adcl(Register dst, Register src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x13);
+ EmitOperand(dst, Operand(src));
+}
+
+
+void X86_64Assembler::adcl(Register dst, const Address& address) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x13);
+ EmitOperand(dst, address);
+}
+
+
+void X86_64Assembler::subl(Register dst, Register src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x2B);
+ EmitOperand(dst, Operand(src));
+}
+
+
+void X86_64Assembler::subl(Register reg, const Immediate& imm) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x48); // REX.W
+ EmitComplex(5, Operand(reg), imm);
+}
+
+
+void X86_64Assembler::subl(Register reg, const Address& address) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x2B);
+ EmitOperand(reg, address);
+}
+
+
+void X86_64Assembler::cdq() {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x99);
+}
+
+
+void X86_64Assembler::idivl(Register reg) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF7);
+ EmitUint8(0xF8 | reg);
+}
+
+
+void X86_64Assembler::imull(Register dst, Register src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x0F);
+ EmitUint8(0xAF);
+ EmitOperand(dst, Operand(src));
+}
+
+
+void X86_64Assembler::imull(Register reg, const Immediate& imm) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x69);
+ EmitOperand(reg, Operand(reg));
+ EmitImmediate(imm);
+}
+
+
+void X86_64Assembler::imull(Register reg, const Address& address) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x0F);
+ EmitUint8(0xAF);
+ EmitOperand(reg, address);
+}
+
+
+void X86_64Assembler::imull(Register reg) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF7);
+ EmitOperand(5, Operand(reg));
+}
+
+
+void X86_64Assembler::imull(const Address& address) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF7);
+ EmitOperand(5, address);
+}
+
+
+void X86_64Assembler::mull(Register reg) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF7);
+ EmitOperand(4, Operand(reg));
+}
+
+
+void X86_64Assembler::mull(const Address& address) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF7);
+ EmitOperand(4, address);
+}
+
+
+void X86_64Assembler::sbbl(Register dst, Register src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x1B);
+ EmitOperand(dst, Operand(src));
+}
+
+
+void X86_64Assembler::sbbl(Register reg, const Immediate& imm) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitComplex(3, Operand(reg), imm);
+}
+
+
+void X86_64Assembler::sbbl(Register dst, const Address& address) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x1B);
+ EmitOperand(dst, address);
+}
+
+
+void X86_64Assembler::incl(Register reg) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x40 + reg);
+}
+
+
+void X86_64Assembler::incl(const Address& address) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xFF);
+ EmitOperand(0, address);
+}
+
+
+void X86_64Assembler::decl(Register reg) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x48 + reg);
+}
+
+
+void X86_64Assembler::decl(const Address& address) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xFF);
+ EmitOperand(1, address);
+}
+
+
+void X86_64Assembler::shll(Register reg, const Immediate& imm) {
+ EmitGenericShift(4, reg, imm);
+}
+
+
+void X86_64Assembler::shll(Register operand, Register shifter) {
+ EmitGenericShift(4, operand, shifter);
+}
+
+
+void X86_64Assembler::shrl(Register reg, const Immediate& imm) {
+ EmitGenericShift(5, reg, imm);
+}
+
+
+void X86_64Assembler::shrl(Register operand, Register shifter) {
+ EmitGenericShift(5, operand, shifter);
+}
+
+
+void X86_64Assembler::sarl(Register reg, const Immediate& imm) {
+ EmitGenericShift(7, reg, imm);
+}
+
+
+void X86_64Assembler::sarl(Register operand, Register shifter) {
+ EmitGenericShift(7, operand, shifter);
+}
+
+
+void X86_64Assembler::shld(Register dst, Register src) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x0F);
+ EmitUint8(0xA5);
+ EmitRegisterOperand(src, dst);
+}
+
+
+void X86_64Assembler::negl(Register reg) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF7);
+ EmitOperand(3, Operand(reg));
+}
+
+
+void X86_64Assembler::notl(Register reg) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF7);
+ EmitUint8(0xD0 | reg);
+}
+
+
+void X86_64Assembler::enter(const Immediate& imm) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xC8);
+ CHECK(imm.is_uint16());
+ EmitUint8(imm.value() & 0xFF);
+ EmitUint8((imm.value() >> 8) & 0xFF);
+ EmitUint8(0x00);
+}
+
+
+void X86_64Assembler::leave() {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xC9);
+}
+
+
+void X86_64Assembler::ret() {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xC3);
+}
+
+
+void X86_64Assembler::ret(const Immediate& imm) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xC2);
+ CHECK(imm.is_uint16());
+ EmitUint8(imm.value() & 0xFF);
+ EmitUint8((imm.value() >> 8) & 0xFF);
+}
+
+
+
+void X86_64Assembler::nop() {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x90);
+}
+
+
+void X86_64Assembler::int3() {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xCC);
+}
+
+
+void X86_64Assembler::hlt() {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF4);
+}
+
+
+void X86_64Assembler::j(Condition condition, Label* label) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ if (label->IsBound()) {
+ static const int kShortSize = 2;
+ static const int kLongSize = 6;
+ int offset = label->Position() - buffer_.Size();
+ CHECK_LE(offset, 0);
+ if (IsInt(8, offset - kShortSize)) {
+ EmitUint8(0x70 + condition);
+ EmitUint8((offset - kShortSize) & 0xFF);
+ } else {
+ EmitUint8(0x0F);
+ EmitUint8(0x80 + condition);
+ EmitInt32(offset - kLongSize);
+ }
+ } else {
+ EmitUint8(0x0F);
+ EmitUint8(0x80 + condition);
+ EmitLabelLink(label);
+ }
+}
+
+
+void X86_64Assembler::jmp(Register reg) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xFF);
+ EmitRegisterOperand(4, reg);
+}
+
+void X86_64Assembler::jmp(const Address& address) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xFF);
+ EmitOperand(4, address);
+}
+
+void X86_64Assembler::jmp(Label* label) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ if (label->IsBound()) {
+ static const int kShortSize = 2;
+ static const int kLongSize = 5;
+ int offset = label->Position() - buffer_.Size();
+ CHECK_LE(offset, 0);
+ if (IsInt(8, offset - kShortSize)) {
+ EmitUint8(0xEB);
+ EmitUint8((offset - kShortSize) & 0xFF);
+ } else {
+ EmitUint8(0xE9);
+ EmitInt32(offset - kLongSize);
+ }
+ } else {
+ EmitUint8(0xE9);
+ EmitLabelLink(label);
+ }
+}
+
+
+X86_64Assembler* X86_64Assembler::lock() {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0xF0);
+ return this;
+}
+
+
+void X86_64Assembler::cmpxchgl(const Address& address, Register reg) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x0F);
+ EmitUint8(0xB1);
+ EmitOperand(reg, address);
+}
+
+void X86_64Assembler::mfence() {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x0F);
+ EmitUint8(0xAE);
+ EmitUint8(0xF0);
+}
+
+X86_64Assembler* X86_64Assembler::gs() {
+ // TODO: fs is a prefix and not an instruction
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ EmitUint8(0x65);
+ return this;
+}
+
+void X86_64Assembler::AddImmediate(Register reg, const Immediate& imm) {
+ int value = imm.value();
+ if (value > 0) {
+ if (value == 1) {
+ incl(reg);
+ } else if (value != 0) {
+ addl(reg, imm);
+ }
+ } else if (value < 0) {
+ value = -value;
+ if (value == 1) {
+ decl(reg);
+ } else if (value != 0) {
+ subl(reg, Immediate(value));
+ }
+ }
+}
+
+
+void X86_64Assembler::LoadDoubleConstant(XmmRegister dst, double value) {
+ // TODO: Need to have a code constants table.
+ int64_t constant = bit_cast<int64_t, double>(value);
+ pushq(Immediate(High32Bits(constant)));
+ pushq(Immediate(Low32Bits(constant)));
+ movsd(dst, Address(RSP, 0));
+ addq(RSP, Immediate(2 * kWordSize));
+}
+
+
+void X86_64Assembler::FloatNegate(XmmRegister f) {
+ static const struct {
+ uint32_t a;
+ uint32_t b;
+ uint32_t c;
+ uint32_t d;
+ } float_negate_constant __attribute__((aligned(16))) =
+ { 0x80000000, 0x00000000, 0x80000000, 0x00000000 };
+ xorps(f, Address::Absolute(reinterpret_cast<uword>(&float_negate_constant)));
+}
+
+
+void X86_64Assembler::DoubleNegate(XmmRegister d) {
+ static const struct {
+ uint64_t a;
+ uint64_t b;
+ } double_negate_constant __attribute__((aligned(16))) =
+ {0x8000000000000000LL, 0x8000000000000000LL};
+ xorpd(d, Address::Absolute(reinterpret_cast<uword>(&double_negate_constant)));
+}
+
+
+void X86_64Assembler::DoubleAbs(XmmRegister reg) {
+ static const struct {
+ uint64_t a;
+ uint64_t b;
+ } double_abs_constant __attribute__((aligned(16))) =
+ {0x7FFFFFFFFFFFFFFFLL, 0x7FFFFFFFFFFFFFFFLL};
+ andpd(reg, Address::Absolute(reinterpret_cast<uword>(&double_abs_constant)));
+}
+
+
+void X86_64Assembler::Align(int alignment, int offset) {
+ CHECK(IsPowerOfTwo(alignment));
+ // Emit nop instruction until the real position is aligned.
+ while (((offset + buffer_.GetPosition()) & (alignment-1)) != 0) {
+ nop();
+ }
+}
+
+
+void X86_64Assembler::Bind(Label* label) {
+ int bound = buffer_.Size();
+ CHECK(!label->IsBound()); // Labels can only be bound once.
+ while (label->IsLinked()) {
+ int position = label->LinkPosition();
+ int next = buffer_.Load<int32_t>(position);
+ buffer_.Store<int32_t>(position, bound - (position + 4));
+ label->position_ = next;
+ }
+ label->BindTo(bound);
+}
+
+
+void X86_64Assembler::EmitOperand(int reg_or_opcode, const Operand& operand) {
+ CHECK_GE(reg_or_opcode, 0);
+ CHECK_LT(reg_or_opcode, 8);
+ const int length = operand.length_;
+ CHECK_GT(length, 0);
+ // Emit the ModRM byte updated with the given reg value.
+ CHECK_EQ(operand.encoding_[0] & 0x38, 0);
+ EmitUint8(operand.encoding_[0] + (reg_or_opcode << 3));
+ // Emit the rest of the encoded operand.
+ for (int i = 1; i < length; i++) {
+ EmitUint8(operand.encoding_[i]);
+ }
+}
+
+
+void X86_64Assembler::EmitImmediate(const Immediate& imm) {
+ EmitInt32(imm.value());
+}
+
+
+void X86_64Assembler::EmitComplex(int reg_or_opcode,
+ const Operand& operand,
+ const Immediate& immediate) {
+ CHECK_GE(reg_or_opcode, 0);
+ CHECK_LT(reg_or_opcode, 8);
+ if (immediate.is_int8()) {
+ // Use sign-extended 8-bit immediate.
+ EmitUint8(0x83);
+ EmitOperand(reg_or_opcode, operand);
+ EmitUint8(immediate.value() & 0xFF);
+ } else if (operand.IsRegister(RAX)) {
+ // Use short form if the destination is eax.
+ EmitUint8(0x05 + (reg_or_opcode << 3));
+ EmitImmediate(immediate);
+ } else {
+ EmitUint8(0x81);
+ EmitOperand(reg_or_opcode, operand);
+ EmitImmediate(immediate);
+ }
+}
+
+
+void X86_64Assembler::EmitLabel(Label* label, int instruction_size) {
+ if (label->IsBound()) {
+ int offset = label->Position() - buffer_.Size();
+ CHECK_LE(offset, 0);
+ EmitInt32(offset - instruction_size);
+ } else {
+ EmitLabelLink(label);
+ }
+}
+
+
+void X86_64Assembler::EmitLabelLink(Label* label) {
+ CHECK(!label->IsBound());
+ int position = buffer_.Size();
+ EmitInt32(label->position_);
+ label->LinkTo(position);
+}
+
+
+void X86_64Assembler::EmitGenericShift(int reg_or_opcode,
+ Register reg,
+ const Immediate& imm) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ CHECK(imm.is_int8());
+ if (imm.value() == 1) {
+ EmitUint8(0xD1);
+ EmitOperand(reg_or_opcode, Operand(reg));
+ } else {
+ EmitUint8(0xC1);
+ EmitOperand(reg_or_opcode, Operand(reg));
+ EmitUint8(imm.value() & 0xFF);
+ }
+}
+
+
+void X86_64Assembler::EmitGenericShift(int reg_or_opcode,
+ Register operand,
+ Register shifter) {
+ AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+ CHECK_EQ(shifter, RCX);
+ EmitUint8(0xD3);
+ EmitOperand(reg_or_opcode, Operand(operand));
+}
+
+void X86_64Assembler::BuildFrame(size_t frame_size, ManagedRegister method_reg,
+ const std::vector<ManagedRegister>& spill_regs,
+ const ManagedRegisterEntrySpills& entry_spills) {
+ CHECK_ALIGNED(frame_size, kStackAlignment);
+ for (int i = spill_regs.size() - 1; i >= 0; --i) {
+ pushq(spill_regs.at(i).AsX86_64().AsCpuRegister());
+ }
+ // return address then method on stack
+ addq(RSP, Immediate(-frame_size + (spill_regs.size() * kPointerSize) +
+ kPointerSize /*method*/ + kPointerSize /*return address*/));
+ pushq(method_reg.AsX86_64().AsCpuRegister());
+
+ for (size_t i = 0; i < entry_spills.size(); ++i) {
+ ManagedRegisterSpill spill = entry_spills.at(i);
+ if (spill.AsX86_64().IsCpuRegister()) {
+ if (spill.getSize() == 8) {
+ movq(Address(RSP, frame_size + spill.getSpillOffset()), spill.AsX86_64().AsCpuRegister());
+ } else {
+ CHECK_EQ(spill.getSize(), 4);
+ movl(Address(RSP, frame_size + spill.getSpillOffset()), spill.AsX86_64().AsCpuRegister());
+ }
+ } else {
+ if (spill.getSize() == 8) {
+ movsd(Address(RSP, frame_size + spill.getSpillOffset()), spill.AsX86_64().AsXmmRegister());
+ } else {
+ CHECK_EQ(spill.getSize(), 4);
+ movss(Address(RSP, frame_size + spill.getSpillOffset()), spill.AsX86_64().AsXmmRegister());
+ }
+ }
+ }
+}
+
+void X86_64Assembler::RemoveFrame(size_t frame_size,
+ const std::vector<ManagedRegister>& spill_regs) {
+ CHECK_ALIGNED(frame_size, kStackAlignment);
+ addq(RSP, Immediate(frame_size - (spill_regs.size() * kPointerSize) - kPointerSize));
+ for (size_t i = 0; i < spill_regs.size(); ++i) {
+ popq(spill_regs.at(i).AsX86_64().AsCpuRegister());
+ }
+ ret();
+}
+
+void X86_64Assembler::IncreaseFrameSize(size_t adjust) {
+ CHECK_ALIGNED(adjust, kStackAlignment);
+ addq(RSP, Immediate(-adjust));
+}
+
+void X86_64Assembler::DecreaseFrameSize(size_t adjust) {
+ CHECK_ALIGNED(adjust, kStackAlignment);
+ addq(RSP, Immediate(adjust));
+}
+
+void X86_64Assembler::Store(FrameOffset offs, ManagedRegister msrc, size_t size) {
+ X86_64ManagedRegister src = msrc.AsX86_64();
+ if (src.IsNoRegister()) {
+ CHECK_EQ(0u, size);
+ } else if (src.IsCpuRegister()) {
+ if (size == 4) {
+ CHECK_EQ(4u, size);
+ movl(Address(RSP, offs), src.AsCpuRegister());
+ } else {
+ CHECK_EQ(8u, size);
+ movq(Address(RSP, offs), src.AsCpuRegister());
+ }
+ } else if (src.IsRegisterPair()) {
+ CHECK_EQ(0u, size);
+ movq(Address(RSP, offs), src.AsRegisterPairLow());
+ movq(Address(RSP, FrameOffset(offs.Int32Value()+4)),
+ src.AsRegisterPairHigh());
+ } else if (src.IsX87Register()) {
+ if (size == 4) {
+ fstps(Address(RSP, offs));
+ } else {
+ fstpl(Address(RSP, offs));
+ }
+ } else {
+ CHECK(src.IsXmmRegister());
+ if (size == 4) {
+ movss(Address(RSP, offs), src.AsXmmRegister());
+ } else {
+ movsd(Address(RSP, offs), src.AsXmmRegister());
+ }
+ }
+}
+
+void X86_64Assembler::StoreRef(FrameOffset dest, ManagedRegister msrc) {
+ X86_64ManagedRegister src = msrc.AsX86_64();
+ CHECK(src.IsCpuRegister());
+ movq(Address(RSP, dest), src.AsCpuRegister());
+}
+
+void X86_64Assembler::StoreRawPtr(FrameOffset dest, ManagedRegister msrc) {
+ X86_64ManagedRegister src = msrc.AsX86_64();
+ CHECK(src.IsCpuRegister());
+ movq(Address(RSP, dest), src.AsCpuRegister());
+}
+
+void X86_64Assembler::StoreImmediateToFrame(FrameOffset dest, uint32_t imm,
+ ManagedRegister) {
+ movl(Address(RSP, dest), Immediate(imm)); // TODO(64) movq?
+}
+
+void X86_64Assembler::StoreImmediateToThread(ThreadOffset dest, uint32_t imm,
+ ManagedRegister) {
+ gs()->movl(Address::Absolute(dest, true), Immediate(imm)); // TODO(64) movq?
+}
+
+void X86_64Assembler::StoreStackOffsetToThread(ThreadOffset thr_offs,
+ FrameOffset fr_offs,
+ ManagedRegister mscratch) {
+ X86_64ManagedRegister scratch = mscratch.AsX86_64();
+ CHECK(scratch.IsCpuRegister());
+ leaq(scratch.AsCpuRegister(), Address(RSP, fr_offs));
+ gs()->movq(Address::Absolute(thr_offs, true), scratch.AsCpuRegister());
+}
+
+void X86_64Assembler::StoreStackPointerToThread(ThreadOffset thr_offs) {
+ gs()->movq(Address::Absolute(thr_offs, true), RSP);
+}
+
+void X86_64Assembler::StoreLabelToThread(ThreadOffset thr_offs, Label* lbl) {
+ gs()->movl(Address::Absolute(thr_offs, true), lbl); // TODO(64) movq?
+}
+
+void X86_64Assembler::StoreSpanning(FrameOffset /*dst*/, ManagedRegister /*src*/,
+ FrameOffset /*in_off*/, ManagedRegister /*scratch*/) {
+ UNIMPLEMENTED(FATAL); // this case only currently exists for ARM
+}
+
+void X86_64Assembler::Load(ManagedRegister mdest, FrameOffset src, size_t size) {
+ X86_64ManagedRegister dest = mdest.AsX86_64();
+ if (dest.IsNoRegister()) {
+ CHECK_EQ(0u, size);
+ } else if (dest.IsCpuRegister()) {
+ if (size == 4) {
+ CHECK_EQ(4u, size);
+ movl(dest.AsCpuRegister(), Address(RSP, src));
+ } else {
+ CHECK_EQ(8u, size);
+ movq(dest.AsCpuRegister(), Address(RSP, src));
+ }
+ } else if (dest.IsRegisterPair()) {
+ CHECK_EQ(0u, size);
+ movq(dest.AsRegisterPairLow(), Address(RSP, src));
+ movq(dest.AsRegisterPairHigh(), Address(RSP, FrameOffset(src.Int32Value()+4)));
+ } else if (dest.IsX87Register()) {
+ if (size == 4) {
+ flds(Address(RSP, src));
+ } else {
+ fldl(Address(RSP, src));
+ }
+ } else {
+ CHECK(dest.IsXmmRegister());
+ if (size == 4) {
+ movss(dest.AsXmmRegister(), Address(RSP, src));
+ } else {
+ movsd(dest.AsXmmRegister(), Address(RSP, src));
+ }
+ }
+}
+
+void X86_64Assembler::Load(ManagedRegister mdest, ThreadOffset src, size_t size) {
+ X86_64ManagedRegister dest = mdest.AsX86_64();
+ if (dest.IsNoRegister()) {
+ CHECK_EQ(0u, size);
+ } else if (dest.IsCpuRegister()) {
+ CHECK_EQ(4u, size);
+ gs()->movq(dest.AsCpuRegister(), Address::Absolute(src, true));
+ } else if (dest.IsRegisterPair()) {
+ CHECK_EQ(8u, size);
+ gs()->movq(dest.AsRegisterPairLow(), Address::Absolute(src, true));
+ gs()->movq(dest.AsRegisterPairHigh(), Address::Absolute(ThreadOffset(src.Int32Value()+4), true));
+ } else if (dest.IsX87Register()) {
+ if (size == 4) {
+ gs()->flds(Address::Absolute(src, true));
+ } else {
+ gs()->fldl(Address::Absolute(src, true));
+ }
+ } else {
+ CHECK(dest.IsXmmRegister());
+ if (size == 4) {
+ gs()->movss(dest.AsXmmRegister(), Address::Absolute(src, true));
+ } else {
+ gs()->movsd(dest.AsXmmRegister(), Address::Absolute(src, true));
+ }
+ }
+}
+
+void X86_64Assembler::LoadRef(ManagedRegister mdest, FrameOffset src) {
+ X86_64ManagedRegister dest = mdest.AsX86_64();
+ CHECK(dest.IsCpuRegister());
+ movq(dest.AsCpuRegister(), Address(RSP, src));
+}
+
+void X86_64Assembler::LoadRef(ManagedRegister mdest, ManagedRegister base,
+ MemberOffset offs) {
+ X86_64ManagedRegister dest = mdest.AsX86_64();
+ CHECK(dest.IsCpuRegister() && dest.IsCpuRegister());
+ movq(dest.AsCpuRegister(), Address(base.AsX86_64().AsCpuRegister(), offs));
+}
+
+void X86_64Assembler::LoadRawPtr(ManagedRegister mdest, ManagedRegister base,
+ Offset offs) {
+ X86_64ManagedRegister dest = mdest.AsX86_64();
+ CHECK(dest.IsCpuRegister() && dest.IsCpuRegister());
+ movq(dest.AsCpuRegister(), Address(base.AsX86_64().AsCpuRegister(), offs));
+}
+
+void X86_64Assembler::LoadRawPtrFromThread(ManagedRegister mdest,
+ ThreadOffset offs) {
+ X86_64ManagedRegister dest = mdest.AsX86_64();
+ CHECK(dest.IsCpuRegister());
+ gs()->movq(dest.AsCpuRegister(), Address::Absolute(offs, true));
+}
+
+void X86_64Assembler::SignExtend(ManagedRegister mreg, size_t size) {
+ X86_64ManagedRegister reg = mreg.AsX86_64();
+ CHECK(size == 1 || size == 2) << size;
+ CHECK(reg.IsCpuRegister()) << reg;
+ if (size == 1) {
+ movsxb(reg.AsCpuRegister(), reg.AsByteRegister());
+ } else {
+ movsxw(reg.AsCpuRegister(), reg.AsCpuRegister());
+ }
+}
+
+void X86_64Assembler::ZeroExtend(ManagedRegister mreg, size_t size) {
+ X86_64ManagedRegister reg = mreg.AsX86_64();
+ CHECK(size == 1 || size == 2) << size;
+ CHECK(reg.IsCpuRegister()) << reg;
+ if (size == 1) {
+ movzxb(reg.AsCpuRegister(), reg.AsByteRegister());
+ } else {
+ movzxw(reg.AsCpuRegister(), reg.AsCpuRegister());
+ }
+}
+
+void X86_64Assembler::Move(ManagedRegister mdest, ManagedRegister msrc, size_t size) {
+ X86_64ManagedRegister dest = mdest.AsX86_64();
+ X86_64ManagedRegister src = msrc.AsX86_64();
+ if (!dest.Equals(src)) {
+ if (dest.IsCpuRegister() && src.IsCpuRegister()) {
+ movq(dest.AsCpuRegister(), src.AsCpuRegister());
+ } else if (src.IsX87Register() && dest.IsXmmRegister()) {
+ // Pass via stack and pop X87 register
+ subl(RSP, Immediate(16));
+ if (size == 4) {
+ CHECK_EQ(src.AsX87Register(), ST0);
+ fstps(Address(RSP, 0));
+ movss(dest.AsXmmRegister(), Address(RSP, 0));
+ } else {
+ CHECK_EQ(src.AsX87Register(), ST0);
+ fstpl(Address(RSP, 0));
+ movsd(dest.AsXmmRegister(), Address(RSP, 0));
+ }
+ addq(RSP, Immediate(16));
+ } else {
+ // TODO: x87, SSE
+ UNIMPLEMENTED(FATAL) << ": Move " << dest << ", " << src;
+ }
+ }
+}
+
+void X86_64Assembler::CopyRef(FrameOffset dest, FrameOffset src,
+ ManagedRegister mscratch) {
+ X86_64ManagedRegister scratch = mscratch.AsX86_64();
+ CHECK(scratch.IsCpuRegister());
+ movl(scratch.AsCpuRegister(), Address(RSP, src));
+ movl(Address(RSP, dest), scratch.AsCpuRegister());
+}
+
+void X86_64Assembler::CopyRawPtrFromThread(FrameOffset fr_offs,
+ ThreadOffset thr_offs,
+ ManagedRegister mscratch) {
+ X86_64ManagedRegister scratch = mscratch.AsX86_64();
+ CHECK(scratch.IsCpuRegister());
+ gs()->movq(scratch.AsCpuRegister(), Address::Absolute(thr_offs, true));
+ Store(fr_offs, scratch, 8);
+}
+
+void X86_64Assembler::CopyRawPtrToThread(ThreadOffset thr_offs,
+ FrameOffset fr_offs,
+ ManagedRegister mscratch) {
+ X86_64ManagedRegister scratch = mscratch.AsX86_64();
+ CHECK(scratch.IsCpuRegister());
+ Load(scratch, fr_offs, 8);
+ gs()->movq(Address::Absolute(thr_offs, true), scratch.AsCpuRegister());
+}
+
+void X86_64Assembler::Copy(FrameOffset dest, FrameOffset src,
+ ManagedRegister mscratch,
+ size_t size) {
+ X86_64ManagedRegister scratch = mscratch.AsX86_64();
+ if (scratch.IsCpuRegister() && size == 8) {
+ Load(scratch, src, 4);
+ Store(dest, scratch, 4);
+ Load(scratch, FrameOffset(src.Int32Value() + 4), 4);
+ Store(FrameOffset(dest.Int32Value() + 4), scratch, 4);
+ } else {
+ Load(scratch, src, size);
+ Store(dest, scratch, size);
+ }
+}
+
+void X86_64Assembler::Copy(FrameOffset /*dst*/, ManagedRegister /*src_base*/, Offset /*src_offset*/,
+ ManagedRegister /*scratch*/, size_t /*size*/) {
+ UNIMPLEMENTED(FATAL);
+}
+
+void X86_64Assembler::Copy(ManagedRegister dest_base, Offset dest_offset, FrameOffset src,
+ ManagedRegister scratch, size_t size) {
+ CHECK(scratch.IsNoRegister());
+ CHECK_EQ(size, 4u);
+ pushq(Address(RSP, src));
+ popq(Address(dest_base.AsX86_64().AsCpuRegister(), dest_offset));
+}
+
+void X86_64Assembler::Copy(FrameOffset dest, FrameOffset src_base, Offset src_offset,
+ ManagedRegister mscratch, size_t size) {
+ Register scratch = mscratch.AsX86_64().AsCpuRegister();
+ CHECK_EQ(size, 4u);
+ movq(scratch, Address(RSP, src_base));
+ movq(scratch, Address(scratch, src_offset));
+ movq(Address(RSP, dest), scratch);
+}
+
+void X86_64Assembler::Copy(ManagedRegister dest, Offset dest_offset,
+ ManagedRegister src, Offset src_offset,
+ ManagedRegister scratch, size_t size) {
+ CHECK_EQ(size, 4u);
+ CHECK(scratch.IsNoRegister());
+ pushq(Address(src.AsX86_64().AsCpuRegister(), src_offset));
+ popq(Address(dest.AsX86_64().AsCpuRegister(), dest_offset));
+}
+
+void X86_64Assembler::Copy(FrameOffset dest, Offset dest_offset, FrameOffset src, Offset src_offset,
+ ManagedRegister mscratch, size_t size) {
+ Register scratch = mscratch.AsX86_64().AsCpuRegister();
+ CHECK_EQ(size, 4u);
+ CHECK_EQ(dest.Int32Value(), src.Int32Value());
+ movq(scratch, Address(RSP, src));
+ pushq(Address(scratch, src_offset));
+ popq(Address(scratch, dest_offset));
+}
+
+void X86_64Assembler::MemoryBarrier(ManagedRegister) {
+#if ANDROID_SMP != 0
+ mfence();
+#endif
+}
+
+void X86_64Assembler::CreateSirtEntry(ManagedRegister mout_reg,
+ FrameOffset sirt_offset,
+ ManagedRegister min_reg, bool null_allowed) {
+ X86_64ManagedRegister out_reg = mout_reg.AsX86_64();
+ X86_64ManagedRegister in_reg = min_reg.AsX86_64();
+ if (in_reg.IsNoRegister()) { // TODO(64): && null_allowed
+ // Use out_reg as indicator of NULL
+ in_reg = out_reg;
+ // TODO: movzwl
+ movl(in_reg.AsCpuRegister(), Address(RSP, sirt_offset));
+ }
+ CHECK(in_reg.IsCpuRegister());
+ CHECK(out_reg.IsCpuRegister());
+ VerifyObject(in_reg, null_allowed);
+ if (null_allowed) {
+ Label null_arg;
+ if (!out_reg.Equals(in_reg)) {
+ xorl(out_reg.AsCpuRegister(), out_reg.AsCpuRegister());
+ }
+ testl(in_reg.AsCpuRegister(), in_reg.AsCpuRegister());
+ j(kZero, &null_arg);
+ leaq(out_reg.AsCpuRegister(), Address(RSP, sirt_offset));
+ Bind(&null_arg);
+ } else {
+ leaq(out_reg.AsCpuRegister(), Address(RSP, sirt_offset));
+ }
+}
+
+void X86_64Assembler::CreateSirtEntry(FrameOffset out_off,
+ FrameOffset sirt_offset,
+ ManagedRegister mscratch,
+ bool null_allowed) {
+ X86_64ManagedRegister scratch = mscratch.AsX86_64();
+ CHECK(scratch.IsCpuRegister());
+ if (null_allowed) {
+ Label null_arg;
+ movl(scratch.AsCpuRegister(), Address(RSP, sirt_offset));
+ testl(scratch.AsCpuRegister(), scratch.AsCpuRegister());
+ j(kZero, &null_arg);
+ leaq(scratch.AsCpuRegister(), Address(RSP, sirt_offset));
+ Bind(&null_arg);
+ } else {
+ leaq(scratch.AsCpuRegister(), Address(RSP, sirt_offset));
+ }
+ Store(out_off, scratch, 8);
+}
+
+// Given a SIRT entry, load the associated reference.
+void X86_64Assembler::LoadReferenceFromSirt(ManagedRegister mout_reg,
+ ManagedRegister min_reg) {
+ X86_64ManagedRegister out_reg = mout_reg.AsX86_64();
+ X86_64ManagedRegister in_reg = min_reg.AsX86_64();
+ CHECK(out_reg.IsCpuRegister());
+ CHECK(in_reg.IsCpuRegister());
+ Label null_arg;
+ if (!out_reg.Equals(in_reg)) {
+ xorl(out_reg.AsCpuRegister(), out_reg.AsCpuRegister());
+ }
+ testl(in_reg.AsCpuRegister(), in_reg.AsCpuRegister());
+ j(kZero, &null_arg);
+ movq(out_reg.AsCpuRegister(), Address(in_reg.AsCpuRegister(), 0));
+ Bind(&null_arg);
+}
+
+void X86_64Assembler::VerifyObject(ManagedRegister /*src*/, bool /*could_be_null*/) {
+ // TODO: not validating references
+}
+
+void X86_64Assembler::VerifyObject(FrameOffset /*src*/, bool /*could_be_null*/) {
+ // TODO: not validating references
+}
+
+void X86_64Assembler::Call(ManagedRegister mbase, Offset offset, ManagedRegister) {
+ X86_64ManagedRegister base = mbase.AsX86_64();
+ CHECK(base.IsCpuRegister());
+ call(Address(base.AsCpuRegister(), offset.Int32Value()));
+ // TODO: place reference map on call
+}
+
+void X86_64Assembler::Call(FrameOffset base, Offset offset, ManagedRegister mscratch) {
+ Register scratch = mscratch.AsX86_64().AsCpuRegister();
+ movq(scratch, Address(RSP, base));
+ call(Address(scratch, offset));
+}
+
+void X86_64Assembler::Call(ThreadOffset offset, ManagedRegister /*mscratch*/) {
+ gs()->call(Address::Absolute(offset, true));
+}
+
+void X86_64Assembler::GetCurrentThread(ManagedRegister tr) {
+ gs()->movq(tr.AsX86_64().AsCpuRegister(),
+ Address::Absolute(Thread::SelfOffset(), true));
+}
+
+void X86_64Assembler::GetCurrentThread(FrameOffset offset,
+ ManagedRegister mscratch) {
+ X86_64ManagedRegister scratch = mscratch.AsX86_64();
+ gs()->movq(scratch.AsCpuRegister(), Address::Absolute(Thread::SelfOffset(), true));
+ movq(Address(RSP, offset), scratch.AsCpuRegister());
+}
+
+void X86_64Assembler::ExceptionPoll(ManagedRegister /*scratch*/, size_t stack_adjust) {
+ X86ExceptionSlowPath* slow = new X86ExceptionSlowPath(stack_adjust);
+ buffer_.EnqueueSlowPath(slow);
+ gs()->cmpl(Address::Absolute(Thread::ExceptionOffset(), true), Immediate(0));
+ j(kNotEqual, slow->Entry());
+}
+
+void X86ExceptionSlowPath::Emit(Assembler *sasm) {
+ X86_64Assembler* sp_asm = down_cast<X86_64Assembler*>(sasm);
+#define __ sp_asm->
+ __ Bind(&entry_);
+ // Note: the return value is dead
+ if (stack_adjust_ != 0) { // Fix up the frame.
+ __ DecreaseFrameSize(stack_adjust_);
+ }
+ // Pass exception as argument in RAX
+ __ gs()->movq(RAX, Address::Absolute(Thread::ExceptionOffset(), true)); // TODO(64): Pass argument via RDI
+ __ gs()->call(Address::Absolute(QUICK_ENTRYPOINT_OFFSET(pDeliverException), true));
+ // this call should never return
+ __ int3();
+#undef __
+}
+
+static const char* kRegisterNames[] = {
+ "rax", "rcx", "rdx", "rbx", "rsp", "rbp", "rsi", "rdi",
+ "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15",
+};
+
+std::ostream& operator<<(std::ostream& os, const Register& rhs) {
+ if (rhs >= RAX && rhs <= R15) {
+ os << kRegisterNames[rhs];
+ } else {
+ os << "Register[" << static_cast<int>(rhs) << "]";
+ }
+ return os;
+}
+} // namespace x86_64
+} // namespace art
+
diff --git a/compiler/utils/x86_64/assembler_x86_64.h b/compiler/utils/x86_64/assembler_x86_64.h
new file mode 100644
index 0000000..d48ba72
--- /dev/null
+++ b/compiler/utils/x86_64/assembler_x86_64.h
@@ -0,0 +1,657 @@
+/*
+ * Copyright (C) 2014 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.
+ */
+
+#ifndef ART_COMPILER_UTILS_X86_64_ASSEMBLER_X86_64_H_
+#define ART_COMPILER_UTILS_X86_64_ASSEMBLER_X86_64_H_
+
+#include <vector>
+#include "base/macros.h"
+#include "constants_x86_64.h"
+#include "globals.h"
+#include "managed_register_x86_64.h"
+#include "offsets.h"
+#include "utils/assembler.h"
+#include "utils.h"
+
+namespace art {
+namespace x86_64 {
+
+class Immediate {
+ public:
+ explicit Immediate(int32_t value) : value_(value) {}
+
+ int32_t value() const { return value_; }
+
+ bool is_int8() const { return IsInt(8, value_); }
+ bool is_uint8() const { return IsUint(8, value_); }
+ bool is_uint16() const { return IsUint(16, value_); }
+
+ private:
+ const int32_t value_;
+
+ DISALLOW_COPY_AND_ASSIGN(Immediate);
+};
+
+
+class Operand {
+ public:
+ uint8_t mod() const {
+ return (encoding_at(0) >> 6) & 3;
+ }
+
+ Register rm() const {
+ return static_cast<Register>(encoding_at(0) & 7);
+ }
+
+ ScaleFactor scale() const {
+ return static_cast<ScaleFactor>((encoding_at(1) >> 6) & 3);
+ }
+
+ Register index() const {
+ return static_cast<Register>((encoding_at(1) >> 3) & 7);
+ }
+
+ Register base() const {
+ return static_cast<Register>(encoding_at(1) & 7);
+ }
+
+ int8_t disp8() const {
+ CHECK_GE(length_, 2);
+ return static_cast<int8_t>(encoding_[length_ - 1]);
+ }
+
+ int32_t disp32() const {
+ CHECK_GE(length_, 5);
+ int32_t value;
+ memcpy(&value, &encoding_[length_ - 4], sizeof(value));
+ return value;
+ }
+
+ bool IsRegister(Register reg) const {
+ return ((encoding_[0] & 0xF8) == 0xC0) // Addressing mode is register only.
+ && ((encoding_[0] & 0x07) == reg); // Register codes match.
+ }
+
+ protected:
+ // Operand can be sub classed (e.g: Address).
+ Operand() : length_(0) { }
+
+ void SetModRM(int mod, Register rm) {
+ CHECK_EQ(mod & ~3, 0);
+ encoding_[0] = (mod << 6) | rm;
+ length_ = 1;
+ }
+
+ void SetSIB(ScaleFactor scale, Register index, Register base) {
+ CHECK_EQ(length_, 1);
+ CHECK_EQ(scale & ~3, 0);
+ encoding_[1] = (scale << 6) | (index << 3) | base;
+ length_ = 2;
+ }
+
+ void SetDisp8(int8_t disp) {
+ CHECK(length_ == 1 || length_ == 2);
+ encoding_[length_++] = static_cast<uint8_t>(disp);
+ }
+
+ void SetDisp32(int32_t disp) {
+ CHECK(length_ == 1 || length_ == 2);
+ int disp_size = sizeof(disp);
+ memmove(&encoding_[length_], &disp, disp_size);
+ length_ += disp_size;
+ }
+
+ private:
+ byte length_;
+ byte encoding_[6];
+ byte padding_;
+
+ explicit Operand(Register reg) { SetModRM(3, reg); }
+
+ // Get the operand encoding byte at the given index.
+ uint8_t encoding_at(int index) const {
+ CHECK_GE(index, 0);
+ CHECK_LT(index, length_);
+ return encoding_[index];
+ }
+
+ friend class X86_64Assembler;
+
+ DISALLOW_COPY_AND_ASSIGN(Operand);
+};
+
+
+class Address : public Operand {
+ public:
+ Address(Register base, int32_t disp) {
+ Init(base, disp);
+ }
+
+ Address(Register base, Offset disp) {
+ Init(base, disp.Int32Value());
+ }
+
+ Address(Register base, FrameOffset disp) {
+ CHECK_EQ(base, RSP);
+ Init(RSP, disp.Int32Value());
+ }
+
+ Address(Register base, MemberOffset disp) {
+ Init(base, disp.Int32Value());
+ }
+
+ void Init(Register base, int32_t disp) {
+ if (disp == 0 && base != RBP) {
+ SetModRM(0, base);
+ if (base == RSP) SetSIB(TIMES_1, RSP, base);
+ } else if (disp >= -128 && disp <= 127) {
+ SetModRM(1, base);
+ if (base == RSP) SetSIB(TIMES_1, RSP, base);
+ SetDisp8(disp);
+ } else {
+ SetModRM(2, base);
+ if (base == RSP) SetSIB(TIMES_1, RSP, base);
+ SetDisp32(disp);
+ }
+ }
+
+
+ Address(Register index, ScaleFactor scale, int32_t disp) {
+ CHECK_NE(index, RSP); // Illegal addressing mode.
+ SetModRM(0, RSP);
+ SetSIB(scale, index, RBP);
+ SetDisp32(disp);
+ }
+
+ Address(Register base, Register index, ScaleFactor scale, int32_t disp) {
+ CHECK_NE(index, RSP); // Illegal addressing mode.
+ if (disp == 0 && base != RBP) {
+ SetModRM(0, RSP);
+ SetSIB(scale, index, base);
+ } else if (disp >= -128 && disp <= 127) {
+ SetModRM(1, RSP);
+ SetSIB(scale, index, base);
+ SetDisp8(disp);
+ } else {
+ SetModRM(2, RSP);
+ SetSIB(scale, index, base);
+ SetDisp32(disp);
+ }
+ }
+
+ static Address Absolute(uword addr, bool has_rip = false) {
+ Address result;
+ if (has_rip) {
+ result.SetModRM(0, RSP);
+ result.SetSIB(TIMES_1, RSP, RBP);
+ result.SetDisp32(addr);
+ } else {
+ result.SetModRM(0, RBP);
+ result.SetDisp32(addr);
+ }
+ return result;
+ }
+
+ static Address Absolute(ThreadOffset addr, bool has_rip = false) {
+ return Absolute(addr.Int32Value(), has_rip);
+ }
+
+ private:
+ Address() {}
+
+ DISALLOW_COPY_AND_ASSIGN(Address);
+};
+
+
+class X86_64Assembler : public Assembler {
+ public:
+ X86_64Assembler() {}
+ virtual ~X86_64Assembler() {}
+
+ /*
+ * Emit Machine Instructions.
+ */
+ void call(Register reg);
+ void call(const Address& address);
+ void call(Label* label);
+
+ void pushq(Register reg);
+ void pushq(const Address& address);
+ void pushq(const Immediate& imm);
+
+ void popq(Register reg);
+ void popq(const Address& address);
+
+ void movq(Register dst, const Immediate& src);
+ void movl(Register dst, const Immediate& src);
+ void movq(Register dst, Register src);
+ void movl(Register dst, Register src);
+
+ void movq(Register dst, const Address& src);
+ void movl(Register dst, const Address& src);
+ void movq(const Address& dst, Register src);
+ void movl(const Address& dst, Register src);
+ void movl(const Address& dst, const Immediate& imm);
+ void movl(const Address& dst, Label* lbl);
+
+ void movzxb(Register dst, ByteRegister src);
+ void movzxb(Register dst, const Address& src);
+ void movsxb(Register dst, ByteRegister src);
+ void movsxb(Register dst, const Address& src);
+ void movb(Register dst, const Address& src);
+ void movb(const Address& dst, ByteRegister src);
+ void movb(const Address& dst, const Immediate& imm);
+
+ void movzxw(Register dst, Register src);
+ void movzxw(Register dst, const Address& src);
+ void movsxw(Register dst, Register src);
+ void movsxw(Register dst, const Address& src);
+ void movw(Register dst, const Address& src);
+ void movw(const Address& dst, Register src);
+
+ void leaq(Register dst, const Address& src);
+
+ void cmovl(Condition condition, Register dst, Register src);
+
+ void setb(Condition condition, Register dst);
+
+ void movss(XmmRegister dst, const Address& src);
+ void movss(const Address& dst, XmmRegister src);
+ void movss(XmmRegister dst, XmmRegister src);
+
+ void movd(XmmRegister dst, Register src);
+ void movd(Register dst, XmmRegister src);
+
+ void addss(XmmRegister dst, XmmRegister src);
+ void addss(XmmRegister dst, const Address& src);
+ void subss(XmmRegister dst, XmmRegister src);
+ void subss(XmmRegister dst, const Address& src);
+ void mulss(XmmRegister dst, XmmRegister src);
+ void mulss(XmmRegister dst, const Address& src);
+ void divss(XmmRegister dst, XmmRegister src);
+ void divss(XmmRegister dst, const Address& src);
+
+ void movsd(XmmRegister dst, const Address& src);
+ void movsd(const Address& dst, XmmRegister src);
+ void movsd(XmmRegister dst, XmmRegister src);
+
+ void addsd(XmmRegister dst, XmmRegister src);
+ void addsd(XmmRegister dst, const Address& src);
+ void subsd(XmmRegister dst, XmmRegister src);
+ void subsd(XmmRegister dst, const Address& src);
+ void mulsd(XmmRegister dst, XmmRegister src);
+ void mulsd(XmmRegister dst, const Address& src);
+ void divsd(XmmRegister dst, XmmRegister src);
+ void divsd(XmmRegister dst, const Address& src);
+
+ void cvtsi2ss(XmmRegister dst, Register src);
+ void cvtsi2sd(XmmRegister dst, Register src);
+
+ void cvtss2si(Register dst, XmmRegister src);
+ void cvtss2sd(XmmRegister dst, XmmRegister src);
+
+ void cvtsd2si(Register dst, XmmRegister src);
+ void cvtsd2ss(XmmRegister dst, XmmRegister src);
+
+ void cvttss2si(Register dst, XmmRegister src);
+ void cvttsd2si(Register dst, XmmRegister src);
+
+ void cvtdq2pd(XmmRegister dst, XmmRegister src);
+
+ void comiss(XmmRegister a, XmmRegister b);
+ void comisd(XmmRegister a, XmmRegister b);
+
+ void sqrtsd(XmmRegister dst, XmmRegister src);
+ void sqrtss(XmmRegister dst, XmmRegister src);
+
+ void xorpd(XmmRegister dst, const Address& src);
+ void xorpd(XmmRegister dst, XmmRegister src);
+ void xorps(XmmRegister dst, const Address& src);
+ void xorps(XmmRegister dst, XmmRegister src);
+
+ void andpd(XmmRegister dst, const Address& src);
+
+ void flds(const Address& src);
+ void fstps(const Address& dst);
+
+ void fldl(const Address& src);
+ void fstpl(const Address& dst);
+
+ void fnstcw(const Address& dst);
+ void fldcw(const Address& src);
+
+ void fistpl(const Address& dst);
+ void fistps(const Address& dst);
+ void fildl(const Address& src);
+
+ void fincstp();
+ void ffree(const Immediate& index);
+
+ void fsin();
+ void fcos();
+ void fptan();
+
+ void xchgl(Register dst, Register src);
+ void xchgl(Register reg, const Address& address);
+
+ void cmpl(Register reg, const Immediate& imm);
+ void cmpl(Register reg0, Register reg1);
+ void cmpl(Register reg, const Address& address);
+
+ void cmpl(const Address& address, Register reg);
+ void cmpl(const Address& address, const Immediate& imm);
+
+ void testl(Register reg1, Register reg2);
+ void testl(Register reg, const Immediate& imm);
+
+ void andl(Register dst, const Immediate& imm);
+ void andl(Register dst, Register src);
+
+ void orl(Register dst, const Immediate& imm);
+ void orl(Register dst, Register src);
+
+ void xorl(Register dst, Register src);
+
+ void addl(Register dst, Register src);
+ void addq(Register reg, const Immediate& imm);
+ void addl(Register reg, const Immediate& imm);
+ void addl(Register reg, const Address& address);
+
+ void addl(const Address& address, Register reg);
+ void addl(const Address& address, const Immediate& imm);
+
+ void adcl(Register dst, Register src);
+ void adcl(Register reg, const Immediate& imm);
+ void adcl(Register dst, const Address& address);
+
+ void subl(Register dst, Register src);
+ void subl(Register reg, const Immediate& imm);
+ void subl(Register reg, const Address& address);
+
+ void cdq();
+
+ void idivl(Register reg);
+
+ void imull(Register dst, Register src);
+ void imull(Register reg, const Immediate& imm);
+ void imull(Register reg, const Address& address);
+
+ void imull(Register reg);
+ void imull(const Address& address);
+
+ void mull(Register reg);
+ void mull(const Address& address);
+
+ void sbbl(Register dst, Register src);
+ void sbbl(Register reg, const Immediate& imm);
+ void sbbl(Register reg, const Address& address);
+
+ void incl(Register reg);
+ void incl(const Address& address);
+
+ void decl(Register reg);
+ void decl(const Address& address);
+
+ void shll(Register reg, const Immediate& imm);
+ void shll(Register operand, Register shifter);
+ void shrl(Register reg, const Immediate& imm);
+ void shrl(Register operand, Register shifter);
+ void sarl(Register reg, const Immediate& imm);
+ void sarl(Register operand, Register shifter);
+ void shld(Register dst, Register src);
+
+ void negl(Register reg);
+ void notl(Register reg);
+
+ void enter(const Immediate& imm);
+ void leave();
+
+ void ret();
+ void ret(const Immediate& imm);
+
+ void nop();
+ void int3();
+ void hlt();
+
+ void j(Condition condition, Label* label);
+
+ void jmp(Register reg);
+ void jmp(const Address& address);
+ void jmp(Label* label);
+
+ X86_64Assembler* lock();
+ void cmpxchgl(const Address& address, Register reg);
+
+ void mfence();
+
+ X86_64Assembler* gs();
+
+ //
+ // Macros for High-level operations.
+ //
+
+ void AddImmediate(Register reg, const Immediate& imm);
+
+ void LoadDoubleConstant(XmmRegister dst, double value);
+
+ void DoubleNegate(XmmRegister d);
+ void FloatNegate(XmmRegister f);
+
+ void DoubleAbs(XmmRegister reg);
+
+ void LockCmpxchgl(const Address& address, Register reg) {
+ lock()->cmpxchgl(address, reg);
+ }
+
+ //
+ // Misc. functionality
+ //
+ int PreferredLoopAlignment() { return 16; }
+ void Align(int alignment, int offset);
+ void Bind(Label* label);
+
+ //
+ // Overridden common assembler high-level functionality
+ //
+
+ // Emit code that will create an activation on the stack
+ virtual void BuildFrame(size_t frame_size, ManagedRegister method_reg,
+ const std::vector<ManagedRegister>& callee_save_regs,
+ const ManagedRegisterEntrySpills& entry_spills);
+
+ // Emit code that will remove an activation from the stack
+ virtual void RemoveFrame(size_t frame_size,
+ const std::vector<ManagedRegister>& callee_save_regs);
+
+ virtual void IncreaseFrameSize(size_t adjust);
+ virtual void DecreaseFrameSize(size_t adjust);
+
+ // Store routines
+ virtual void Store(FrameOffset offs, ManagedRegister src, size_t size);
+ virtual void StoreRef(FrameOffset dest, ManagedRegister src);
+ virtual void StoreRawPtr(FrameOffset dest, ManagedRegister src);
+
+ virtual void StoreImmediateToFrame(FrameOffset dest, uint32_t imm,
+ ManagedRegister scratch);
+
+ virtual void StoreImmediateToThread(ThreadOffset dest, uint32_t imm,
+ ManagedRegister scratch);
+
+ virtual void StoreStackOffsetToThread(ThreadOffset thr_offs,
+ FrameOffset fr_offs,
+ ManagedRegister scratch);
+
+ virtual void StoreStackPointerToThread(ThreadOffset thr_offs);
+
+ void StoreLabelToThread(ThreadOffset thr_offs, Label* lbl);
+
+ virtual void StoreSpanning(FrameOffset dest, ManagedRegister src,
+ FrameOffset in_off, ManagedRegister scratch);
+
+ // Load routines
+ virtual void Load(ManagedRegister dest, FrameOffset src, size_t size);
+
+ virtual void Load(ManagedRegister dest, ThreadOffset src, size_t size);
+
+ virtual void LoadRef(ManagedRegister dest, FrameOffset src);
+
+ virtual void LoadRef(ManagedRegister dest, ManagedRegister base,
+ MemberOffset offs);
+
+ virtual void LoadRawPtr(ManagedRegister dest, ManagedRegister base,
+ Offset offs);
+
+ virtual void LoadRawPtrFromThread(ManagedRegister dest,
+ ThreadOffset offs);
+
+ // Copying routines
+ virtual void Move(ManagedRegister dest, ManagedRegister src, size_t size);
+
+ virtual void CopyRawPtrFromThread(FrameOffset fr_offs, ThreadOffset thr_offs,
+ ManagedRegister scratch);
+
+ virtual void CopyRawPtrToThread(ThreadOffset thr_offs, FrameOffset fr_offs,
+ ManagedRegister scratch);
+
+ virtual void CopyRef(FrameOffset dest, FrameOffset src,
+ ManagedRegister scratch);
+
+ virtual void Copy(FrameOffset dest, FrameOffset src, ManagedRegister scratch, size_t size);
+
+ virtual void Copy(FrameOffset dest, ManagedRegister src_base, Offset src_offset,
+ ManagedRegister scratch, size_t size);
+
+ virtual void Copy(ManagedRegister dest_base, Offset dest_offset, FrameOffset src,
+ ManagedRegister scratch, size_t size);
+
+ virtual void Copy(FrameOffset dest, FrameOffset src_base, Offset src_offset,
+ ManagedRegister scratch, size_t size);
+
+ virtual void Copy(ManagedRegister dest, Offset dest_offset,
+ ManagedRegister src, Offset src_offset,
+ ManagedRegister scratch, size_t size);
+
+ virtual void Copy(FrameOffset dest, Offset dest_offset, FrameOffset src, Offset src_offset,
+ ManagedRegister scratch, size_t size);
+
+ virtual void MemoryBarrier(ManagedRegister);
+
+ // Sign extension
+ virtual void SignExtend(ManagedRegister mreg, size_t size);
+
+ // Zero extension
+ virtual void ZeroExtend(ManagedRegister mreg, size_t size);
+
+ // Exploit fast access in managed code to Thread::Current()
+ virtual void GetCurrentThread(ManagedRegister tr);
+ virtual void GetCurrentThread(FrameOffset dest_offset,
+ ManagedRegister scratch);
+
+ // Set up out_reg to hold a Object** into the SIRT, or to be NULL if the
+ // value is null and null_allowed. in_reg holds a possibly stale reference
+ // that can be used to avoid loading the SIRT entry to see if the value is
+ // NULL.
+ virtual void CreateSirtEntry(ManagedRegister out_reg, FrameOffset sirt_offset,
+ ManagedRegister in_reg, bool null_allowed);
+
+ // Set up out_off to hold a Object** into the SIRT, or to be NULL if the
+ // value is null and null_allowed.
+ virtual void CreateSirtEntry(FrameOffset out_off, FrameOffset sirt_offset,
+ ManagedRegister scratch, bool null_allowed);
+
+ // src holds a SIRT entry (Object**) load this into dst
+ virtual void LoadReferenceFromSirt(ManagedRegister dst,
+ ManagedRegister src);
+
+ // Heap::VerifyObject on src. In some cases (such as a reference to this) we
+ // know that src may not be null.
+ virtual void VerifyObject(ManagedRegister src, bool could_be_null);
+ virtual void VerifyObject(FrameOffset src, bool could_be_null);
+
+ // Call to address held at [base+offset]
+ virtual void Call(ManagedRegister base, Offset offset,
+ ManagedRegister scratch);
+ virtual void Call(FrameOffset base, Offset offset,
+ ManagedRegister scratch);
+ virtual void Call(ThreadOffset offset, ManagedRegister scratch);
+
+ // Generate code to check if Thread::Current()->exception_ is non-null
+ // and branch to a ExceptionSlowPath if it is.
+ virtual void ExceptionPoll(ManagedRegister scratch, size_t stack_adjust);
+
+ private:
+ inline void EmitUint8(uint8_t value);
+ inline void EmitInt32(int32_t value);
+ inline void EmitRegisterOperand(int rm, int reg);
+ inline void EmitXmmRegisterOperand(int rm, XmmRegister reg);
+ inline void EmitFixup(AssemblerFixup* fixup);
+ inline void EmitOperandSizeOverride();
+
+ void EmitOperand(int rm, const Operand& operand);
+ void EmitImmediate(const Immediate& imm);
+ void EmitComplex(int rm, const Operand& operand, const Immediate& immediate);
+ void EmitLabel(Label* label, int instruction_size);
+ void EmitLabelLink(Label* label);
+ void EmitNearLabelLink(Label* label);
+
+ void EmitGenericShift(int rm, Register reg, const Immediate& imm);
+ void EmitGenericShift(int rm, Register operand, Register shifter);
+ void rex(Register &dst, Register &src, size_t size = 4);
+ void rex_reg(Register &dst, size_t size = 4);
+ void rex_rm(Register &src, size_t size = 4);
+
+ DISALLOW_COPY_AND_ASSIGN(X86_64Assembler);
+};
+
+inline void X86_64Assembler::EmitUint8(uint8_t value) {
+ buffer_.Emit<uint8_t>(value);
+}
+
+inline void X86_64Assembler::EmitInt32(int32_t value) {
+ buffer_.Emit<int32_t>(value);
+}
+
+inline void X86_64Assembler::EmitRegisterOperand(int rm, int reg) {
+ CHECK_GE(rm, 0);
+ CHECK_LT(rm, 8);
+ buffer_.Emit<uint8_t>(0xC0 + (rm << 3) + reg);
+}
+
+inline void X86_64Assembler::EmitXmmRegisterOperand(int rm, XmmRegister reg) {
+ EmitRegisterOperand(rm, static_cast<Register>(reg));
+}
+
+inline void X86_64Assembler::EmitFixup(AssemblerFixup* fixup) {
+ buffer_.EmitFixup(fixup);
+}
+
+inline void X86_64Assembler::EmitOperandSizeOverride() {
+ EmitUint8(0x66);
+}
+
+// Slowpath entered when Thread::Current()->_exception is non-null
+class X86ExceptionSlowPath : public SlowPath {
+ public:
+ explicit X86ExceptionSlowPath(size_t stack_adjust) : stack_adjust_(stack_adjust) {}
+ virtual void Emit(Assembler *sp_asm);
+ private:
+ const size_t stack_adjust_;
+};
+
+} // namespace x86_64
+} // namespace art
+
+#endif // ART_COMPILER_UTILS_X86_64_ASSEMBLER_X86_64_H_
diff --git a/compiler/utils/x86_64/assembler_x86_64_test.cc b/compiler/utils/x86_64/assembler_x86_64_test.cc
new file mode 100644
index 0000000..df0d14e
--- /dev/null
+++ b/compiler/utils/x86_64/assembler_x86_64_test.cc
@@ -0,0 +1,32 @@
+/*
+ * Copyright (C) 2014 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.
+ */
+
+#include "assembler_x86_64.h"
+
+#include "gtest/gtest.h"
+
+namespace art {
+
+TEST(AssemblerX86_64, CreateBuffer) {
+ AssemblerBuffer buffer;
+ AssemblerBuffer::EnsureCapacity ensured(&buffer);
+ buffer.Emit<uint8_t>(0x42);
+ ASSERT_EQ(static_cast<size_t>(1), buffer.Size());
+ buffer.Emit<int32_t>(42);
+ ASSERT_EQ(static_cast<size_t>(5), buffer.Size());
+}
+
+} // namespace art
diff --git a/compiler/utils/x86_64/constants_x86_64.h b/compiler/utils/x86_64/constants_x86_64.h
new file mode 100644
index 0000000..3340802
--- /dev/null
+++ b/compiler/utils/x86_64/constants_x86_64.h
@@ -0,0 +1,126 @@
+/*
+ * Copyright (C) 2014 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.
+ */
+
+#ifndef ART_COMPILER_UTILS_X86_64_CONSTANTS_X86_64_H_
+#define ART_COMPILER_UTILS_X86_64_CONSTANTS_X86_64_H_
+
+#include <iosfwd>
+
+#include "arch/x86_64/registers_x86_64.h"
+#include "base/logging.h"
+#include "base/macros.h"
+#include "globals.h"
+
+namespace art {
+namespace x86_64 {
+
+enum ByteRegister {
+ AL = 0,
+ CL = 1,
+ DL = 2,
+ BL = 3,
+ AH = 4,
+ CH = 5,
+ DH = 6,
+ BH = 7,
+ kNoByteRegister = -1 // Signals an illegal register.
+};
+
+
+enum XmmRegister {
+ _XMM0 = 0,
+ _XMM1 = 1,
+ _XMM2 = 2,
+ _XMM3 = 3,
+ _XMM4 = 4,
+ _XMM5 = 5,
+ _XMM6 = 6,
+ _XMM7 = 7,
+ kNumberOfXmmRegisters = 8,
+ kNoXmmRegister = -1 // Signals an illegal register.
+};
+std::ostream& operator<<(std::ostream& os, const XmmRegister& reg);
+
+enum X87Register {
+ ST0 = 0,
+ ST1 = 1,
+ ST2 = 2,
+ ST3 = 3,
+ ST4 = 4,
+ ST5 = 5,
+ ST6 = 6,
+ ST7 = 7,
+ kNumberOfX87Registers = 8,
+ kNoX87Register = -1 // Signals an illegal register.
+};
+std::ostream& operator<<(std::ostream& os, const X87Register& reg);
+
+enum ScaleFactor {
+ TIMES_1 = 0,
+ TIMES_2 = 1,
+ TIMES_4 = 2,
+ TIMES_8 = 3
+};
+
+enum Condition {
+ kOverflow = 0,
+ kNoOverflow = 1,
+ kBelow = 2,
+ kAboveEqual = 3,
+ kEqual = 4,
+ kNotEqual = 5,
+ kBelowEqual = 6,
+ kAbove = 7,
+ kSign = 8,
+ kNotSign = 9,
+ kParityEven = 10,
+ kParityOdd = 11,
+ kLess = 12,
+ kGreaterEqual = 13,
+ kLessEqual = 14,
+ kGreater = 15,
+
+ kZero = kEqual,
+ kNotZero = kNotEqual,
+ kNegative = kSign,
+ kPositive = kNotSign
+};
+
+
+class Instr {
+ public:
+ static const uint8_t kHltInstruction = 0xF4;
+ // We prefer not to use the int3 instruction since it conflicts with gdb.
+ static const uint8_t kBreakPointInstruction = kHltInstruction;
+
+ bool IsBreakPoint() {
+ return (*reinterpret_cast<const uint8_t*>(this)) == kBreakPointInstruction;
+ }
+
+ // Instructions are read out of a code stream. The only way to get a
+ // reference to an instruction is to convert a pointer. There is no way
+ // to allocate or create instances of class Instr.
+ // Use the At(pc) function to create references to Instr.
+ static Instr* At(uintptr_t pc) { return reinterpret_cast<Instr*>(pc); }
+
+ private:
+ DISALLOW_IMPLICIT_CONSTRUCTORS(Instr);
+};
+
+} // namespace x86_64
+} // namespace art
+
+#endif // ART_COMPILER_UTILS_X86_64_CONSTANTS_X86_64_H_
diff --git a/compiler/utils/x86_64/managed_register_x86_64.cc b/compiler/utils/x86_64/managed_register_x86_64.cc
new file mode 100644
index 0000000..057a894
--- /dev/null
+++ b/compiler/utils/x86_64/managed_register_x86_64.cc
@@ -0,0 +1,115 @@
+/*
+ * Copyright (C) 2014 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.
+ */
+
+#include "managed_register_x86_64.h"
+
+#include "globals.h"
+
+namespace art {
+namespace x86_64 {
+
+// Define register pairs.
+// This list must be kept in sync with the RegisterPair enum.
+#define REGISTER_PAIR_LIST(P) \
+ P(RAX, RDX) \
+ P(RAX, RCX) \
+ P(RAX, RBX) \
+ P(RAX, RDI) \
+ P(RDX, RCX) \
+ P(RDX, RBX) \
+ P(RDX, RDI) \
+ P(RCX, RBX) \
+ P(RCX, RDI) \
+ P(RBX, RDI)
+
+
+struct RegisterPairDescriptor {
+ RegisterPair reg; // Used to verify that the enum is in sync.
+ Register low;
+ Register high;
+};
+
+
+static const RegisterPairDescriptor kRegisterPairs[] = {
+#define REGISTER_PAIR_ENUMERATION(low, high) { low##_##high, low, high },
+ REGISTER_PAIR_LIST(REGISTER_PAIR_ENUMERATION)
+#undef REGISTER_PAIR_ENUMERATION
+};
+
+std::ostream& operator<<(std::ostream& os, const RegisterPair& reg) {
+ os << X86_64ManagedRegister::FromRegisterPair(reg);
+ return os;
+}
+
+bool X86_64ManagedRegister::Overlaps(const X86_64ManagedRegister& other) const {
+ if (IsNoRegister() || other.IsNoRegister()) return false;
+ CHECK(IsValidManagedRegister());
+ CHECK(other.IsValidManagedRegister());
+ if (Equals(other)) return true;
+ if (IsRegisterPair()) {
+ Register low = AsRegisterPairLow();
+ Register high = AsRegisterPairHigh();
+ return X86_64ManagedRegister::FromCpuRegister(low).Overlaps(other) ||
+ X86_64ManagedRegister::FromCpuRegister(high).Overlaps(other);
+ }
+ if (other.IsRegisterPair()) {
+ return other.Overlaps(*this);
+ }
+ return false;
+}
+
+
+int X86_64ManagedRegister::AllocIdLow() const {
+ CHECK(IsRegisterPair());
+ const int r = RegId() - (kNumberOfCpuRegIds + kNumberOfXmmRegIds +
+ kNumberOfX87RegIds);
+ CHECK_EQ(r, kRegisterPairs[r].reg);
+ return kRegisterPairs[r].low;
+}
+
+
+int X86_64ManagedRegister::AllocIdHigh() const {
+ CHECK(IsRegisterPair());
+ const int r = RegId() - (kNumberOfCpuRegIds + kNumberOfXmmRegIds +
+ kNumberOfX87RegIds);
+ CHECK_EQ(r, kRegisterPairs[r].reg);
+ return kRegisterPairs[r].high;
+}
+
+
+void X86_64ManagedRegister::Print(std::ostream& os) const {
+ if (!IsValidManagedRegister()) {
+ os << "No Register";
+ } else if (IsXmmRegister()) {
+ os << "XMM: " << static_cast<int>(AsXmmRegister());
+ } else if (IsX87Register()) {
+ os << "X87: " << static_cast<int>(AsX87Register());
+ } else if (IsCpuRegister()) {
+ os << "CPU: " << static_cast<int>(AsCpuRegister());
+ } else if (IsRegisterPair()) {
+ os << "Pair: " << AsRegisterPairLow() << ", " << AsRegisterPairHigh();
+ } else {
+ os << "??: " << RegId();
+ }
+}
+
+std::ostream& operator<<(std::ostream& os, const X86_64ManagedRegister& reg) {
+ reg.Print(os);
+ return os;
+}
+
+} // namespace x86_64
+} // namespace art
diff --git a/compiler/utils/x86_64/managed_register_x86_64.h b/compiler/utils/x86_64/managed_register_x86_64.h
new file mode 100644
index 0000000..d68c59d
--- /dev/null
+++ b/compiler/utils/x86_64/managed_register_x86_64.h
@@ -0,0 +1,218 @@
+/*
+ * Copyright (C) 2014 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.
+ */
+
+#ifndef ART_COMPILER_UTILS_X86_64_MANAGED_REGISTER_X86_64_H_
+#define ART_COMPILER_UTILS_X86_64_MANAGED_REGISTER_X86_64_H_
+
+#include "constants_x86_64.h"
+#include "utils/managed_register.h"
+
+namespace art {
+namespace x86_64 {
+
+// Values for register pairs.
+// The registers in kReservedCpuRegistersArray in x86.cc are not used in pairs.
+// The table kRegisterPairs in x86.cc must be kept in sync with this enum.
+enum RegisterPair {
+ RAX_RDX = 0,
+ RAX_RCX = 1,
+ RAX_RBX = 2,
+ RAX_RDI = 3,
+ RDX_RCX = 4,
+ RDX_RBX = 5,
+ RDX_RDI = 6,
+ RCX_RBX = 7,
+ RCX_RDI = 8,
+ RBX_RDI = 9,
+ kNumberOfRegisterPairs = 10,
+ kNoRegisterPair = -1,
+};
+
+std::ostream& operator<<(std::ostream& os, const RegisterPair& reg);
+
+const int kNumberOfCpuRegIds = kNumberOfCpuRegisters;
+const int kNumberOfCpuAllocIds = kNumberOfCpuRegisters;
+
+const int kNumberOfXmmRegIds = kNumberOfXmmRegisters;
+const int kNumberOfXmmAllocIds = kNumberOfXmmRegisters;
+
+const int kNumberOfX87RegIds = kNumberOfX87Registers;
+const int kNumberOfX87AllocIds = kNumberOfX87Registers;
+
+const int kNumberOfPairRegIds = kNumberOfRegisterPairs;
+
+const int kNumberOfRegIds = kNumberOfCpuRegIds + kNumberOfXmmRegIds +
+ kNumberOfX87RegIds + kNumberOfPairRegIds;
+const int kNumberOfAllocIds = kNumberOfCpuAllocIds + kNumberOfXmmAllocIds +
+ kNumberOfX87RegIds;
+
+// Register ids map:
+// [0..R[ cpu registers (enum Register)
+// [R..X[ xmm registers (enum XmmRegister)
+// [X..S[ x87 registers (enum X87Register)
+// [S..P[ register pairs (enum RegisterPair)
+// where
+// R = kNumberOfCpuRegIds
+// X = R + kNumberOfXmmRegIds
+// S = X + kNumberOfX87RegIds
+// P = X + kNumberOfRegisterPairs
+
+// Allocation ids map:
+// [0..R[ cpu registers (enum Register)
+// [R..X[ xmm registers (enum XmmRegister)
+// [X..S[ x87 registers (enum X87Register)
+// where
+// R = kNumberOfCpuRegIds
+// X = R + kNumberOfXmmRegIds
+// S = X + kNumberOfX87RegIds
+
+
+// An instance of class 'ManagedRegister' represents a single cpu register (enum
+// Register), an xmm register (enum XmmRegister), or a pair of cpu registers
+// (enum RegisterPair).
+// 'ManagedRegister::NoRegister()' provides an invalid register.
+// There is a one-to-one mapping between ManagedRegister and register id.
+class X86_64ManagedRegister : public ManagedRegister {
+ public:
+ ByteRegister AsByteRegister() const {
+ CHECK(IsCpuRegister());
+ CHECK_LT(AsCpuRegister(), RSP); // RSP, RBP, ESI and RDI cannot be encoded as byte registers.
+ return static_cast<ByteRegister>(id_);
+ }
+
+ Register AsCpuRegister() const {
+ CHECK(IsCpuRegister());
+ return static_cast<Register>(id_);
+ }
+
+ XmmRegister AsXmmRegister() const {
+ CHECK(IsXmmRegister());
+ return static_cast<XmmRegister>(id_ - kNumberOfCpuRegIds);
+ }
+
+ X87Register AsX87Register() const {
+ CHECK(IsX87Register());
+ return static_cast<X87Register>(id_ -
+ (kNumberOfCpuRegIds + kNumberOfXmmRegIds));
+ }
+
+ Register AsRegisterPairLow() const {
+ CHECK(IsRegisterPair());
+ // Appropriate mapping of register ids allows to use AllocIdLow().
+ return FromRegId(AllocIdLow()).AsCpuRegister();
+ }
+
+ Register AsRegisterPairHigh() const {
+ CHECK(IsRegisterPair());
+ // Appropriate mapping of register ids allows to use AllocIdHigh().
+ return FromRegId(AllocIdHigh()).AsCpuRegister();
+ }
+
+ bool IsCpuRegister() const {
+ CHECK(IsValidManagedRegister());
+ return (0 <= id_) && (id_ < kNumberOfCpuRegIds);
+ }
+
+ bool IsXmmRegister() const {
+ CHECK(IsValidManagedRegister());
+ const int test = id_ - kNumberOfCpuRegIds;
+ return (0 <= test) && (test < kNumberOfXmmRegIds);
+ }
+
+ bool IsX87Register() const {
+ CHECK(IsValidManagedRegister());
+ const int test = id_ - (kNumberOfCpuRegIds + kNumberOfXmmRegIds);
+ return (0 <= test) && (test < kNumberOfX87RegIds);
+ }
+
+ bool IsRegisterPair() const {
+ CHECK(IsValidManagedRegister());
+ const int test = id_ -
+ (kNumberOfCpuRegIds + kNumberOfXmmRegIds + kNumberOfX87RegIds);
+ return (0 <= test) && (test < kNumberOfPairRegIds);
+ }
+
+ void Print(std::ostream& os) const;
+
+ // Returns true if the two managed-registers ('this' and 'other') overlap.
+ // Either managed-register may be the NoRegister. If both are the NoRegister
+ // then false is returned.
+ bool Overlaps(const X86_64ManagedRegister& other) const;
+
+ static X86_64ManagedRegister FromCpuRegister(Register r) {
+ CHECK_NE(r, kNoRegister);
+ return FromRegId(r);
+ }
+
+ static X86_64ManagedRegister FromXmmRegister(XmmRegister r) {
+ CHECK_NE(r, kNoXmmRegister);
+ return FromRegId(r + kNumberOfCpuRegIds);
+ }
+
+ static X86_64ManagedRegister FromX87Register(X87Register r) {
+ CHECK_NE(r, kNoX87Register);
+ return FromRegId(r + kNumberOfCpuRegIds + kNumberOfXmmRegIds);
+ }
+
+ static X86_64ManagedRegister FromRegisterPair(RegisterPair r) {
+ CHECK_NE(r, kNoRegisterPair);
+ return FromRegId(r + (kNumberOfCpuRegIds + kNumberOfXmmRegIds +
+ kNumberOfX87RegIds));
+ }
+
+ private:
+ bool IsValidManagedRegister() const {
+ return (0 <= id_) && (id_ < kNumberOfRegIds);
+ }
+
+ int RegId() const {
+ CHECK(!IsNoRegister());
+ return id_;
+ }
+
+ int AllocId() const {
+ CHECK(IsValidManagedRegister() && !IsRegisterPair());
+ CHECK_LT(id_, kNumberOfAllocIds);
+ return id_;
+ }
+
+ int AllocIdLow() const;
+ int AllocIdHigh() const;
+
+ friend class ManagedRegister;
+
+ explicit X86_64ManagedRegister(int reg_id) : ManagedRegister(reg_id) {}
+
+ static X86_64ManagedRegister FromRegId(int reg_id) {
+ X86_64ManagedRegister reg(reg_id);
+ CHECK(reg.IsValidManagedRegister());
+ return reg;
+ }
+};
+
+std::ostream& operator<<(std::ostream& os, const X86_64ManagedRegister& reg);
+
+} // namespace x86_64
+
+inline x86_64::X86_64ManagedRegister ManagedRegister::AsX86_64() const {
+ x86_64::X86_64ManagedRegister reg(id_);
+ CHECK(reg.IsNoRegister() || reg.IsValidManagedRegister());
+ return reg;
+}
+
+} // namespace art
+
+#endif // ART_COMPILER_UTILS_X86_64_MANAGED_REGISTER_X86_64_H_
diff --git a/compiler/utils/x86_64/managed_register_x86_64_test.cc b/compiler/utils/x86_64/managed_register_x86_64_test.cc
new file mode 100644
index 0000000..2dc7581
--- /dev/null
+++ b/compiler/utils/x86_64/managed_register_x86_64_test.cc
@@ -0,0 +1,359 @@
+/*
+ * Copyright (C) 2014 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.
+ */
+
+#include "globals.h"
+#include "managed_register_x86_64.h"
+#include "gtest/gtest.h"
+
+namespace art {
+namespace x86_64 {
+
+TEST(X86_64ManagedRegister, NoRegister) {
+ X86_64ManagedRegister reg = ManagedRegister::NoRegister().AsX86();
+ EXPECT_TRUE(reg.IsNoRegister());
+ EXPECT_TRUE(!reg.Overlaps(reg));
+}
+
+TEST(X86_64ManagedRegister, CpuRegister) {
+ X86_64ManagedRegister reg = X86_64ManagedRegister::FromCpuRegister(RAX);
+ EXPECT_TRUE(!reg.IsNoRegister());
+ EXPECT_TRUE(reg.IsCpuRegister());
+ EXPECT_TRUE(!reg.IsXmmRegister());
+ EXPECT_TRUE(!reg.IsX87Register());
+ EXPECT_TRUE(!reg.IsRegisterPair());
+ EXPECT_EQ(RAX, reg.AsCpuRegister());
+
+ reg = X86_64ManagedRegister::FromCpuRegister(RBX);
+ EXPECT_TRUE(!reg.IsNoRegister());
+ EXPECT_TRUE(reg.IsCpuRegister());
+ EXPECT_TRUE(!reg.IsXmmRegister());
+ EXPECT_TRUE(!reg.IsX87Register());
+ EXPECT_TRUE(!reg.IsRegisterPair());
+ EXPECT_EQ(RBX, reg.AsCpuRegister());
+
+ reg = X86_64ManagedRegister::FromCpuRegister(RCX);
+ EXPECT_TRUE(!reg.IsNoRegister());
+ EXPECT_TRUE(reg.IsCpuRegister());
+ EXPECT_TRUE(!reg.IsXmmRegister());
+ EXPECT_TRUE(!reg.IsX87Register());
+ EXPECT_TRUE(!reg.IsRegisterPair());
+ EXPECT_EQ(RCX, reg.AsCpuRegister());
+
+ reg = X86_64ManagedRegister::FromCpuRegister(RDI);
+ EXPECT_TRUE(!reg.IsNoRegister());
+ EXPECT_TRUE(reg.IsCpuRegister());
+ EXPECT_TRUE(!reg.IsXmmRegister());
+ EXPECT_TRUE(!reg.IsX87Register());
+ EXPECT_TRUE(!reg.IsRegisterPair());
+ EXPECT_EQ(RDI, reg.AsCpuRegister());
+}
+
+TEST(X86_64ManagedRegister, XmmRegister) {
+ X86_64ManagedRegister reg = X86_64ManagedRegister::FromXmmRegister(XMM0);
+ EXPECT_TRUE(!reg.IsNoRegister());
+ EXPECT_TRUE(!reg.IsCpuRegister());
+ EXPECT_TRUE(reg.IsXmmRegister());
+ EXPECT_TRUE(!reg.IsX87Register());
+ EXPECT_TRUE(!reg.IsRegisterPair());
+ EXPECT_EQ(XMM0, reg.AsXmmRegister());
+
+ reg = X86_64ManagedRegister::FromXmmRegister(XMM1);
+ EXPECT_TRUE(!reg.IsNoRegister());
+ EXPECT_TRUE(!reg.IsCpuRegister());
+ EXPECT_TRUE(reg.IsXmmRegister());
+ EXPECT_TRUE(!reg.IsX87Register());
+ EXPECT_TRUE(!reg.IsRegisterPair());
+ EXPECT_EQ(XMM1, reg.AsXmmRegister());
+
+ reg = X86_64ManagedRegister::FromXmmRegister(XMM7);
+ EXPECT_TRUE(!reg.IsNoRegister());
+ EXPECT_TRUE(!reg.IsCpuRegister());
+ EXPECT_TRUE(reg.IsXmmRegister());
+ EXPECT_TRUE(!reg.IsX87Register());
+ EXPECT_TRUE(!reg.IsRegisterPair());
+ EXPECT_EQ(XMM7, reg.AsXmmRegister());
+}
+
+TEST(X86_64ManagedRegister, X87Register) {
+ X86_64ManagedRegister reg = X86_64ManagedRegister::FromX87Register(ST0);
+ EXPECT_TRUE(!reg.IsNoRegister());
+ EXPECT_TRUE(!reg.IsCpuRegister());
+ EXPECT_TRUE(!reg.IsXmmRegister());
+ EXPECT_TRUE(reg.IsX87Register());
+ EXPECT_TRUE(!reg.IsRegisterPair());
+ EXPECT_EQ(ST0, reg.AsX87Register());
+
+ reg = X86_64ManagedRegister::FromX87Register(ST1);
+ EXPECT_TRUE(!reg.IsNoRegister());
+ EXPECT_TRUE(!reg.IsCpuRegister());
+ EXPECT_TRUE(!reg.IsXmmRegister());
+ EXPECT_TRUE(reg.IsX87Register());
+ EXPECT_TRUE(!reg.IsRegisterPair());
+ EXPECT_EQ(ST1, reg.AsX87Register());
+
+ reg = X86_64ManagedRegister::FromX87Register(ST7);
+ EXPECT_TRUE(!reg.IsNoRegister());
+ EXPECT_TRUE(!reg.IsCpuRegister());
+ EXPECT_TRUE(!reg.IsXmmRegister());
+ EXPECT_TRUE(reg.IsX87Register());
+ EXPECT_TRUE(!reg.IsRegisterPair());
+ EXPECT_EQ(ST7, reg.AsX87Register());
+}
+
+TEST(X86_64ManagedRegister, RegisterPair) {
+ X86_64ManagedRegister reg = X86_64ManagedRegister::FromRegisterPair(EAX_EDX);
+ EXPECT_TRUE(!reg.IsNoRegister());
+ EXPECT_TRUE(!reg.IsCpuRegister());
+ EXPECT_TRUE(!reg.IsXmmRegister());
+ EXPECT_TRUE(!reg.IsX87Register());
+ EXPECT_TRUE(reg.IsRegisterPair());
+ EXPECT_EQ(RAX, reg.AsRegisterPairLow());
+ EXPECT_EQ(RDX, reg.AsRegisterPairHigh());
+
+ reg = X86_64ManagedRegister::FromRegisterPair(EAX_ECX);
+ EXPECT_TRUE(!reg.IsNoRegister());
+ EXPECT_TRUE(!reg.IsCpuRegister());
+ EXPECT_TRUE(!reg.IsXmmRegister());
+ EXPECT_TRUE(!reg.IsX87Register());
+ EXPECT_TRUE(reg.IsRegisterPair());
+ EXPECT_EQ(RAX, reg.AsRegisterPairLow());
+ EXPECT_EQ(RCX, reg.AsRegisterPairHigh());
+
+ reg = X86_64ManagedRegister::FromRegisterPair(EAX_EBX);
+ EXPECT_TRUE(!reg.IsNoRegister());
+ EXPECT_TRUE(!reg.IsCpuRegister());
+ EXPECT_TRUE(!reg.IsXmmRegister());
+ EXPECT_TRUE(!reg.IsX87Register());
+ EXPECT_TRUE(reg.IsRegisterPair());
+ EXPECT_EQ(RAX, reg.AsRegisterPairLow());
+ EXPECT_EQ(RBX, reg.AsRegisterPairHigh());
+
+ reg = X86_64ManagedRegister::FromRegisterPair(EAX_EDI);
+ EXPECT_TRUE(!reg.IsNoRegister());
+ EXPECT_TRUE(!reg.IsCpuRegister());
+ EXPECT_TRUE(!reg.IsXmmRegister());
+ EXPECT_TRUE(!reg.IsX87Register());
+ EXPECT_TRUE(reg.IsRegisterPair());
+ EXPECT_EQ(RAX, reg.AsRegisterPairLow());
+ EXPECT_EQ(RDI, reg.AsRegisterPairHigh());
+
+ reg = X86_64ManagedRegister::FromRegisterPair(EDX_ECX);
+ EXPECT_TRUE(!reg.IsNoRegister());
+ EXPECT_TRUE(!reg.IsCpuRegister());
+ EXPECT_TRUE(!reg.IsXmmRegister());
+ EXPECT_TRUE(!reg.IsX87Register());
+ EXPECT_TRUE(reg.IsRegisterPair());
+ EXPECT_EQ(RDX, reg.AsRegisterPairLow());
+ EXPECT_EQ(RCX, reg.AsRegisterPairHigh());
+
+ reg = X86_64ManagedRegister::FromRegisterPair(EDX_EBX);
+ EXPECT_TRUE(!reg.IsNoRegister());
+ EXPECT_TRUE(!reg.IsCpuRegister());
+ EXPECT_TRUE(!reg.IsXmmRegister());
+ EXPECT_TRUE(!reg.IsX87Register());
+ EXPECT_TRUE(reg.IsRegisterPair());
+ EXPECT_EQ(RDX, reg.AsRegisterPairLow());
+ EXPECT_EQ(RBX, reg.AsRegisterPairHigh());
+
+ reg = X86_64ManagedRegister::FromRegisterPair(EDX_EDI);
+ EXPECT_TRUE(!reg.IsNoRegister());
+ EXPECT_TRUE(!reg.IsCpuRegister());
+ EXPECT_TRUE(!reg.IsXmmRegister());
+ EXPECT_TRUE(!reg.IsX87Register());
+ EXPECT_TRUE(reg.IsRegisterPair());
+ EXPECT_EQ(RDX, reg.AsRegisterPairLow());
+ EXPECT_EQ(RDI, reg.AsRegisterPairHigh());
+
+ reg = X86_64ManagedRegister::FromRegisterPair(ECX_EBX);
+ EXPECT_TRUE(!reg.IsNoRegister());
+ EXPECT_TRUE(!reg.IsCpuRegister());
+ EXPECT_TRUE(!reg.IsXmmRegister());
+ EXPECT_TRUE(!reg.IsX87Register());
+ EXPECT_TRUE(reg.IsRegisterPair());
+ EXPECT_EQ(RCX, reg.AsRegisterPairLow());
+ EXPECT_EQ(RBX, reg.AsRegisterPairHigh());
+
+ reg = X86_64ManagedRegister::FromRegisterPair(ECX_EDI);
+ EXPECT_TRUE(!reg.IsNoRegister());
+ EXPECT_TRUE(!reg.IsCpuRegister());
+ EXPECT_TRUE(!reg.IsXmmRegister());
+ EXPECT_TRUE(!reg.IsX87Register());
+ EXPECT_TRUE(reg.IsRegisterPair());
+ EXPECT_EQ(RCX, reg.AsRegisterPairLow());
+ EXPECT_EQ(RDI, reg.AsRegisterPairHigh());
+
+ reg = X86_64ManagedRegister::FromRegisterPair(EBX_EDI);
+ EXPECT_TRUE(!reg.IsNoRegister());
+ EXPECT_TRUE(!reg.IsCpuRegister());
+ EXPECT_TRUE(!reg.IsXmmRegister());
+ EXPECT_TRUE(!reg.IsX87Register());
+ EXPECT_TRUE(reg.IsRegisterPair());
+ EXPECT_EQ(RBX, reg.AsRegisterPairLow());
+ EXPECT_EQ(RDI, reg.AsRegisterPairHigh());
+}
+
+TEST(X86_64ManagedRegister, Equals) {
+ X86_64ManagedRegister reg_eax = X86_64ManagedRegister::FromCpuRegister(RAX);
+ EXPECT_TRUE(reg_eax.Equals(X86_64ManagedRegister::FromCpuRegister(RAX)));
+ EXPECT_TRUE(!reg_eax.Equals(X86_64ManagedRegister::FromCpuRegister(RBX)));
+ EXPECT_TRUE(!reg_eax.Equals(X86_64ManagedRegister::FromCpuRegister(RDI)));
+ EXPECT_TRUE(!reg_eax.Equals(X86_64ManagedRegister::FromXmmRegister(XMM0)));
+ EXPECT_TRUE(!reg_eax.Equals(X86_64ManagedRegister::FromXmmRegister(XMM7)));
+ EXPECT_TRUE(!reg_eax.Equals(X86_64ManagedRegister::FromX87Register(ST0)));
+ EXPECT_TRUE(!reg_eax.Equals(X86_64ManagedRegister::FromX87Register(ST7)));
+ EXPECT_TRUE(!reg_eax.Equals(X86_64ManagedRegister::FromRegisterPair(EAX_EDX)));
+ EXPECT_TRUE(!reg_eax.Equals(X86_64ManagedRegister::FromRegisterPair(EBX_EDI)));
+
+ X86_64ManagedRegister reg_xmm0 = X86_64ManagedRegister::FromXmmRegister(XMM0);
+ EXPECT_TRUE(!reg_xmm0.Equals(X86_64ManagedRegister::FromCpuRegister(RAX)));
+ EXPECT_TRUE(!reg_xmm0.Equals(X86_64ManagedRegister::FromCpuRegister(RBX)));
+ EXPECT_TRUE(!reg_xmm0.Equals(X86_64ManagedRegister::FromCpuRegister(RDI)));
+ EXPECT_TRUE(reg_xmm0.Equals(X86_64ManagedRegister::FromXmmRegister(XMM0)));
+ EXPECT_TRUE(!reg_xmm0.Equals(X86_64ManagedRegister::FromXmmRegister(XMM7)));
+ EXPECT_TRUE(!reg_xmm0.Equals(X86_64ManagedRegister::FromX87Register(ST0)));
+ EXPECT_TRUE(!reg_xmm0.Equals(X86_64ManagedRegister::FromX87Register(ST7)));
+ EXPECT_TRUE(!reg_xmm0.Equals(X86_64ManagedRegister::FromRegisterPair(EAX_EDX)));
+ EXPECT_TRUE(!reg_xmm0.Equals(X86_64ManagedRegister::FromRegisterPair(EBX_EDI)));
+
+ X86_64ManagedRegister reg_st0 = X86_64ManagedRegister::FromX87Register(ST0);
+ EXPECT_TRUE(!reg_st0.Equals(X86_64ManagedRegister::FromCpuRegister(RAX)));
+ EXPECT_TRUE(!reg_st0.Equals(X86_64ManagedRegister::FromCpuRegister(RBX)));
+ EXPECT_TRUE(!reg_st0.Equals(X86_64ManagedRegister::FromCpuRegister(RDI)));
+ EXPECT_TRUE(!reg_st0.Equals(X86_64ManagedRegister::FromXmmRegister(XMM0)));
+ EXPECT_TRUE(!reg_st0.Equals(X86_64ManagedRegister::FromXmmRegister(XMM7)));
+ EXPECT_TRUE(reg_st0.Equals(X86_64ManagedRegister::FromX87Register(ST0)));
+ EXPECT_TRUE(!reg_st0.Equals(X86_64ManagedRegister::FromX87Register(ST7)));
+ EXPECT_TRUE(!reg_st0.Equals(X86_64ManagedRegister::FromRegisterPair(EAX_EDX)));
+ EXPECT_TRUE(!reg_st0.Equals(X86_64ManagedRegister::FromRegisterPair(EBX_EDI)));
+
+ X86_64ManagedRegister reg_pair = X86_64ManagedRegister::FromRegisterPair(EAX_EDX);
+ EXPECT_TRUE(!reg_pair.Equals(X86_64ManagedRegister::FromCpuRegister(RAX)));
+ EXPECT_TRUE(!reg_pair.Equals(X86_64ManagedRegister::FromCpuRegister(RBX)));
+ EXPECT_TRUE(!reg_pair.Equals(X86_64ManagedRegister::FromCpuRegister(RDI)));
+ EXPECT_TRUE(!reg_pair.Equals(X86_64ManagedRegister::FromXmmRegister(XMM0)));
+ EXPECT_TRUE(!reg_pair.Equals(X86_64ManagedRegister::FromXmmRegister(XMM7)));
+ EXPECT_TRUE(!reg_pair.Equals(X86_64ManagedRegister::FromX87Register(ST0)));
+ EXPECT_TRUE(!reg_pair.Equals(X86_64ManagedRegister::FromX87Register(ST7)));
+ EXPECT_TRUE(reg_pair.Equals(X86_64ManagedRegister::FromRegisterPair(EAX_EDX)));
+ EXPECT_TRUE(!reg_pair.Equals(X86_64ManagedRegister::FromRegisterPair(EBX_EDI)));
+}
+
+TEST(X86_64ManagedRegister, Overlaps) {
+ X86_64ManagedRegister reg = X86_64ManagedRegister::FromCpuRegister(RAX);
+ EXPECT_TRUE(reg.Overlaps(X86_64ManagedRegister::FromCpuRegister(RAX)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromCpuRegister(RBX)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromCpuRegister(RDI)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromXmmRegister(XMM0)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromXmmRegister(XMM7)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromX87Register(ST0)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromX87Register(ST7)));
+ EXPECT_TRUE(reg.Overlaps(X86_64ManagedRegister::FromRegisterPair(EAX_EDX)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromRegisterPair(EBX_EDI)));
+
+ reg = X86_64ManagedRegister::FromCpuRegister(RDX);
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromCpuRegister(RAX)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromCpuRegister(RBX)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromCpuRegister(RDI)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromXmmRegister(XMM0)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromXmmRegister(XMM7)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromX87Register(ST0)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromX87Register(ST7)));
+ EXPECT_TRUE(reg.Overlaps(X86_64ManagedRegister::FromRegisterPair(EAX_EDX)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromRegisterPair(EBX_EDI)));
+
+ reg = X86_64ManagedRegister::FromCpuRegister(RDI);
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromCpuRegister(RAX)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromCpuRegister(RBX)));
+ EXPECT_TRUE(reg.Overlaps(X86_64ManagedRegister::FromCpuRegister(RDI)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromXmmRegister(XMM0)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromXmmRegister(XMM7)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromX87Register(ST0)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromX87Register(ST7)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromRegisterPair(EAX_EDX)));
+ EXPECT_TRUE(reg.Overlaps(X86_64ManagedRegister::FromRegisterPair(EBX_EDI)));
+
+ reg = X86_64ManagedRegister::FromCpuRegister(RBX);
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromCpuRegister(RAX)));
+ EXPECT_TRUE(reg.Overlaps(X86_64ManagedRegister::FromCpuRegister(RBX)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromCpuRegister(RDI)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromXmmRegister(XMM0)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromXmmRegister(XMM7)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromX87Register(ST0)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromX87Register(ST7)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromRegisterPair(EAX_EDX)));
+ EXPECT_TRUE(reg.Overlaps(X86_64ManagedRegister::FromRegisterPair(EBX_EDI)));
+
+ reg = X86_64ManagedRegister::FromXmmRegister(XMM0);
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromCpuRegister(RAX)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromCpuRegister(RBX)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromCpuRegister(RDI)));
+ EXPECT_TRUE(reg.Overlaps(X86_64ManagedRegister::FromXmmRegister(XMM0)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromXmmRegister(XMM7)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromX87Register(ST0)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromX87Register(ST7)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromRegisterPair(EAX_EDX)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromRegisterPair(EBX_EDI)));
+
+ reg = X86_64ManagedRegister::FromX87Register(ST0);
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromCpuRegister(RAX)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromCpuRegister(RBX)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromCpuRegister(RDI)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromXmmRegister(XMM0)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromXmmRegister(XMM7)));
+ EXPECT_TRUE(reg.Overlaps(X86_64ManagedRegister::FromX87Register(ST0)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromX87Register(ST7)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromRegisterPair(EAX_EDX)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromRegisterPair(EBX_EDI)));
+
+ reg = X86_64ManagedRegister::FromRegisterPair(EAX_EDX);
+ EXPECT_TRUE(reg.Overlaps(X86_64ManagedRegister::FromCpuRegister(RAX)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromCpuRegister(RBX)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromCpuRegister(RDI)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromXmmRegister(XMM0)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromXmmRegister(XMM7)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromX87Register(ST0)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromX87Register(ST7)));
+ EXPECT_TRUE(reg.Overlaps(X86_64ManagedRegister::FromRegisterPair(EAX_EDX)));
+ EXPECT_TRUE(reg.Overlaps(X86_64ManagedRegister::FromRegisterPair(EDX_ECX)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromRegisterPair(EBX_EDI)));
+
+ reg = X86_64ManagedRegister::FromRegisterPair(EBX_EDI);
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromCpuRegister(RAX)));
+ EXPECT_TRUE(reg.Overlaps(X86_64ManagedRegister::FromCpuRegister(RBX)));
+ EXPECT_TRUE(reg.Overlaps(X86_64ManagedRegister::FromCpuRegister(RDI)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromXmmRegister(XMM0)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromXmmRegister(XMM7)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromX87Register(ST0)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromX87Register(ST7)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromRegisterPair(EAX_EDX)));
+ EXPECT_TRUE(reg.Overlaps(X86_64ManagedRegister::FromRegisterPair(EBX_EDI)));
+ EXPECT_TRUE(reg.Overlaps(X86_64ManagedRegister::FromRegisterPair(EDX_EBX)));
+
+ reg = X86_64ManagedRegister::FromRegisterPair(EDX_ECX);
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromCpuRegister(RAX)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromCpuRegister(RBX)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromCpuRegister(RDI)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromXmmRegister(XMM0)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromXmmRegister(XMM7)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromX87Register(ST0)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromX87Register(ST7)));
+ EXPECT_TRUE(reg.Overlaps(X86_64ManagedRegister::FromRegisterPair(EAX_EDX)));
+ EXPECT_TRUE(!reg.Overlaps(X86_64ManagedRegister::FromRegisterPair(EBX_EDI)));
+ EXPECT_TRUE(reg.Overlaps(X86_64ManagedRegister::FromRegisterPair(EDX_EBX)));
+}
+
+} // namespace x86_64
+} // namespace art
diff --git a/runtime/Android.mk b/runtime/Android.mk
index 1ca8e07..1576905 100644
--- a/runtime/Android.mk
+++ b/runtime/Android.mk
@@ -52,6 +52,7 @@
gc/accounting/mod_union_table.cc \
gc/accounting/remembered_set.cc \
gc/accounting/space_bitmap.cc \
+ gc/collector/concurrent_copying.cc \
gc/collector/garbage_collector.cc \
gc/collector/immune_region.cc \
gc/collector/mark_sweep.cc \
diff --git a/runtime/arch/arm/fault_handler_arm.cc b/runtime/arch/arm/fault_handler_arm.cc
index c748ce9..abce838 100644
--- a/runtime/arch/arm/fault_handler_arm.cc
+++ b/runtime/arch/arm/fault_handler_arm.cc
@@ -18,9 +18,12 @@
#include "fault_handler.h"
#include <sys/ucontext.h>
#include "base/macros.h"
+#include "base/hex_dump.h"
#include "globals.h"
#include "base/logging.h"
#include "base/hex_dump.h"
+#include "mirror/art_method.h"
+#include "mirror/art_method-inl.h"
#include "thread.h"
#include "thread-inl.h"
@@ -31,16 +34,38 @@
namespace art {
extern "C" void art_quick_throw_null_pointer_exception();
+extern "C" void art_quick_throw_stack_overflow(void*);
extern "C" void art_quick_test_suspend();
+// Get the size of a thumb2 instruction in bytes.
+static uint32_t GetInstructionSize(uint8_t* pc) {
+ uint16_t instr = pc[0] | pc[1] << 8;
+ bool is_32bit = ((instr & 0xF000) == 0xF000) || ((instr & 0xF800) == 0xE800);
+ uint32_t instr_size = is_32bit ? 4 : 2;
+ return instr_size;
+}
+
void FaultManager::GetMethodAndReturnPC(void* context, uintptr_t& method, uintptr_t& return_pc) {
struct ucontext *uc = (struct ucontext *)context;
struct sigcontext *sc = reinterpret_cast<struct sigcontext*>(&uc->uc_mcontext);
uintptr_t* sp = reinterpret_cast<uint32_t*>(sc->arm_sp);
+ LOG(DEBUG) << "sp: " << sp;
if (sp == nullptr) {
return;
}
+ // In the case of a stack overflow, the stack is not valid and we can't
+ // get the method from the top of the stack. However it's in r0.
+ uintptr_t* fault_addr = reinterpret_cast<uintptr_t*>(sc->fault_address);
+ uintptr_t* overflow_addr = reinterpret_cast<uintptr_t*>(
+ reinterpret_cast<uint8_t*>(sp) - Thread::kStackOverflowReservedBytes);
+ if (overflow_addr == fault_addr) {
+ method = sc->arm_r0;
+ } else {
+ // The method is at the top of the stack.
+ method = sp[0];
+ }
+
// Work out the return PC. This will be the address of the instruction
// following the faulting ldr/str instruction. This is in thumb mode so
// the instruction might be a 16 or 32 bit one. Also, the GC map always
@@ -48,13 +73,8 @@
// Need to work out the size of the instruction that caused the exception.
uint8_t* ptr = reinterpret_cast<uint8_t*>(sc->arm_pc);
-
- uint16_t instr = ptr[0] | ptr[1] << 8;
- bool is_32bit = ((instr & 0xF000) == 0xF000) || ((instr & 0xF800) == 0xE800);
- uint32_t instr_size = is_32bit ? 4 : 2;
-
- // The method is at the top of the stack.
- method = sp[0];
+ LOG(DEBUG) << "pc: " << std::hex << static_cast<void*>(ptr);
+ uint32_t instr_size = GetInstructionSize(ptr);
return_pc = (sc->arm_pc + instr_size) | 1;
}
@@ -71,9 +91,7 @@
struct sigcontext *sc = reinterpret_cast<struct sigcontext*>(&uc->uc_mcontext);
uint8_t* ptr = reinterpret_cast<uint8_t*>(sc->arm_pc);
- uint16_t instr = ptr[0] | ptr[1] << 8;
- bool is_32bit = ((instr & 0xF000) == 0xF000) || ((instr & 0xF800) == 0xE800);
- uint32_t instr_size = is_32bit ? 4 : 2;
+ uint32_t instr_size = GetInstructionSize(ptr);
sc->arm_lr = (sc->arm_pc + instr_size) | 1; // LR needs to point to gc map location
sc->arm_pc = reinterpret_cast<uintptr_t>(art_quick_throw_null_pointer_exception);
LOG(DEBUG) << "Generating null pointer exception";
@@ -142,7 +160,116 @@
return false;
}
+// Stack overflow fault handler.
+//
+// This checks that the fault address is equal to the current stack pointer
+// minus the overflow region size (16K typically). The instruction sequence
+// that generates this signal is:
+//
+// sub r12,sp,#16384
+// ldr.w r12,[r12,#0]
+//
+// The second instruction will fault if r12 is inside the protected region
+// on the stack.
+//
+// If we determine this is a stack overflow we need to move the stack pointer
+// to the overflow region below the protected region. Because we now have
+// a gap in the stack (skips over protected region), we need to arrange
+// for the rest of the system to be unaware of the new stack arrangement
+// and behave as if there is a fully valid stack. We do this by placing
+// a unique address onto the stack followed by
+// the size of the gap. The stack walker will detect this and skip over the
+// gap.
+
+// NB. We also need to be careful of stack alignment as the ARM EABI specifies that
+// stack must be 8 byte aligned when making any calls.
+
+// NB. The size of the gap is the difference between the previous frame's SP and
+// the SP at which the size word is pushed.
+
bool StackOverflowHandler::Action(int sig, siginfo_t* info, void* context) {
- return false;
+ struct ucontext *uc = (struct ucontext *)context;
+ struct sigcontext *sc = reinterpret_cast<struct sigcontext*>(&uc->uc_mcontext);
+ LOG(DEBUG) << "stack overflow handler with sp at " << std::hex << &uc;
+ LOG(DEBUG) << "sigcontext: " << std::hex << sc;
+
+ uint8_t* sp = reinterpret_cast<uint8_t*>(sc->arm_sp);
+ LOG(DEBUG) << "sp: " << static_cast<void*>(sp);
+
+ uintptr_t* fault_addr = reinterpret_cast<uintptr_t*>(sc->fault_address);
+ LOG(DEBUG) << "fault_addr: " << std::hex << fault_addr;
+ LOG(DEBUG) << "checking for stack overflow, sp: " << std::hex << static_cast<void*>(sp) <<
+ ", fault_addr: " << fault_addr;
+ uintptr_t* overflow_addr = reinterpret_cast<uintptr_t*>(sp - Thread::kStackOverflowReservedBytes);
+
+ // Check that the fault address is the value expected for a stack overflow.
+ if (fault_addr != overflow_addr) {
+ LOG(DEBUG) << "Not a stack overflow";
+ return false;
+ }
+
+ // We know this is a stack overflow. We need to move the sp to the overflow region
+ // the exists below the protected region. R9 contains the current Thread* so
+ // we can read the stack_end from that and subtract the size of the
+ // protected region. This creates a gap in the stack that needs to be marked.
+ Thread* self = reinterpret_cast<Thread*>(sc->arm_r9);
+
+ uint8_t* prevsp = sp;
+ sp = self->GetStackEnd() - Thread::kStackOverflowProtectedSize;
+ LOG(DEBUG) << "setting sp to overflow region at " << std::hex << static_cast<void*>(sp);
+
+ // We need to find the previous frame. Remember that
+ // this has not yet been fully constructed because the SP has not been
+ // decremented. So we need to work out the size of the spill portion of the
+ // frame. This consists of something like:
+ //
+ // 0xb6a1d49c: e92d40e0 push {r5, r6, r7, lr}
+ // 0xb6a1d4a0: ed2d8a06 vpush.f32 {s16-s21}
+ //
+ // The first is encoded in the ArtMethod as the spill_mask, the second as the
+ // fp_spill_mask. A population count on each will give the number of registers
+ // in each mask. Each register is 4 bytes on ARM32.
+
+ mirror::ArtMethod* method = reinterpret_cast<mirror::ArtMethod*>(sc->arm_r0);
+ uint32_t spill_mask = method->GetCoreSpillMask();
+ uint32_t numcores = __builtin_popcount(spill_mask);
+ uint32_t fp_spill_mask = method->GetFpSpillMask();
+ uint32_t numfps = __builtin_popcount(fp_spill_mask);
+ uint32_t spill_size = (numcores + numfps) * 4;
+ LOG(DEBUG) << "spill size: " << spill_size;
+ uint8_t* prevframe = prevsp + spill_size;
+ LOG(DEBUG) << "previous frame: " << static_cast<void*>(prevframe);
+
+ // NOTE: the ARM EABI needs an 8 byte alignment. In the case of ARM32 a pointer
+ // is 4 bytes so that, together with the offset to the previous frame is 8
+ // bytes. On other architectures we will need to align the stack.
+
+ // Push a marker onto the stack to tell the stack walker that there is a stack
+ // overflow and the stack is not contiguous.
+
+ // First the offset from SP to the previous frame.
+ sp -= sizeof(uint32_t);
+ LOG(DEBUG) << "push gap of " << static_cast<uint32_t>(prevframe - sp);
+ *reinterpret_cast<uint32_t*>(sp) = static_cast<uint32_t>(prevframe - sp);
+
+ // Now the gap marker (pointer sized).
+ sp -= sizeof(mirror::ArtMethod*);
+ *reinterpret_cast<void**>(sp) = stack_overflow_gap_marker;
+
+ // Now establish the stack pointer for the signal return.
+ sc->arm_sp = reinterpret_cast<uintptr_t>(sp);
+
+ // Now arrange for the signal handler to return to art_quick_throw_stack_overflow.
+ // We need the LR to point to the GC map just after the fault instruction.
+ uint8_t* ptr = reinterpret_cast<uint8_t*>(sc->arm_pc);
+ uint32_t instr_size = GetInstructionSize(ptr);
+ sc->arm_lr = (sc->arm_pc + instr_size) | 1; // LR needs to point to gc map location
+ sc->arm_pc = reinterpret_cast<uintptr_t>(art_quick_throw_stack_overflow);
+
+ // The kernel will now return to the address in sc->arm_pc. We have arranged the
+ // stack pointer to be in the overflow region. Throwing the exception will perform
+ // a longjmp which will restore the stack pointer to the correct location for the
+ // exception catch.
+ return true;
}
} // namespace art
diff --git a/runtime/arch/x86_64/asm_support_x86_64.S b/runtime/arch/x86_64/asm_support_x86_64.S
index ca2489c..d03a474 100644
--- a/runtime/arch/x86_64/asm_support_x86_64.S
+++ b/runtime/arch/x86_64/asm_support_x86_64.S
@@ -29,7 +29,7 @@
// Clang's as(1) uses $0, $1, and so on for macro arguments.
#define VAR(name,index) SYMBOL($index)
- #define PLT_VAR(name, index) SYMBOL($index)
+ #define PLT_VAR(name, index) SYMBOL($index)@PLT
#define REG_VAR(name,index) %$index
#define CALL_MACRO(name,index) $index
#define FUNCTION_TYPE(name,index) .type $index, @function
diff --git a/runtime/arch/x86_64/quick_entrypoints_x86_64.S b/runtime/arch/x86_64/quick_entrypoints_x86_64.S
index 5fbf8cb..0d75a89 100644
--- a/runtime/arch/x86_64/quick_entrypoints_x86_64.S
+++ b/runtime/arch/x86_64/quick_entrypoints_x86_64.S
@@ -142,8 +142,13 @@
MACRO2(ONE_ARG_RUNTIME_EXCEPTION, c_name, cxx_name)
DEFINE_FUNCTION VAR(c_name, 0)
- int3
- int3
+ SETUP_SAVE_ALL_CALLEE_SAVE_FRAME // save all registers as basis for long jump context
+ // Outgoing argument set up
+ mov %rsp, %rdx // pass SP
+ mov %gs:THREAD_SELF_OFFSET, %rsi // pass Thread::Current()
+ mov %rax, %rdi // pass arg1
+ call PLT_VAR(cxx_name, 1) // cxx_name(arg1, Thread*, SP)
+ int3 // unreached
END_FUNCTION VAR(c_name, 0)
END_MACRO
diff --git a/runtime/arch/x86_64/registers_x86_64.h b/runtime/arch/x86_64/registers_x86_64.h
index 8b0dc07..b9d06b5 100644
--- a/runtime/arch/x86_64/registers_x86_64.h
+++ b/runtime/arch/x86_64/registers_x86_64.h
@@ -67,7 +67,7 @@
XMM15 = 15,
kNumberOfFloatRegisters = 16
};
-std::ostream& operator<<(std::ostream& os, const FloatRegister& rhs);
+std::ostream& operator<<(std::ostream& os, const Register& rhs);
} // namespace x86_64
} // namespace art
diff --git a/runtime/base/mutex.cc b/runtime/base/mutex.cc
index 52a1672..fdd0249 100644
--- a/runtime/base/mutex.cc
+++ b/runtime/base/mutex.cc
@@ -649,7 +649,13 @@
sequence_ = 0;
num_waiters_ = 0;
#else
- CHECK_MUTEX_CALL(pthread_cond_init, (&cond_, NULL));
+ pthread_condattr_t cond_attrs;
+ CHECK_MUTEX_CALL(pthread_condattr_init(&cond_attrs));
+#if !defined(__APPLE__)
+ // Apple doesn't have CLOCK_MONOTONIC or pthread_condattr_setclock.
+ CHECK_MUTEX_CALL(pthread_condattr_setclock(&cond_attrs, CLOCK_MONOTONIC));
+#endif
+ CHECK_MUTEX_CALL(pthread_cond_init, (&cond_, &cond_attrs));
#endif
}
@@ -787,17 +793,15 @@
CHECK_GE(guard_.num_contenders_, 0);
guard_.num_contenders_--;
#else
-#ifdef HAVE_TIMEDWAIT_MONOTONIC
-#define TIMEDWAIT pthread_cond_timedwait_monotonic
+#if !defined(__APPLE__)
int clock = CLOCK_MONOTONIC;
#else
-#define TIMEDWAIT pthread_cond_timedwait
int clock = CLOCK_REALTIME;
#endif
guard_.recursion_count_ = 0;
timespec ts;
InitTimeSpec(true, clock, ms, ns, &ts);
- int rc = TEMP_FAILURE_RETRY(TIMEDWAIT(&cond_, &guard_.mutex_, &ts));
+ int rc = TEMP_FAILURE_RETRY(pthread_cond_timedwait(&cond_, &guard_.mutex_, &ts));
if (rc != 0 && rc != ETIMEDOUT) {
errno = rc;
PLOG(FATAL) << "TimedWait failed for " << name_;
diff --git a/runtime/base/timing_logger.cc b/runtime/base/timing_logger.cc
index c769f64..a155002 100644
--- a/runtime/base/timing_logger.cc
+++ b/runtime/base/timing_logger.cc
@@ -54,7 +54,7 @@
void CumulativeLogger::End() {
MutexLock mu(Thread::Current(), lock_);
- iterations_++;
+ ++iterations_;
}
void CumulativeLogger::Reset() {
@@ -66,13 +66,12 @@
void CumulativeLogger::AddLogger(const TimingLogger &logger) {
MutexLock mu(Thread::Current(), lock_);
- const TimingLogger::SplitTimings& splits = logger.GetSplits();
- for (auto it = splits.begin(), end = splits.end(); it != end; ++it) {
- TimingLogger::SplitTiming split = *it;
+ for (const TimingLogger::SplitTiming& split : logger.GetSplits()) {
uint64_t split_time = split.first;
const char* split_name = split.second;
AddPair(split_name, split_time);
}
+ ++iterations_;
}
size_t CumulativeLogger::GetIterations() const {
@@ -161,8 +160,7 @@
uint64_t TimingLogger::GetTotalNs() const {
uint64_t total_ns = 0;
- for (auto it = splits_.begin(), end = splits_.end(); it != end; ++it) {
- TimingLogger::SplitTiming split = *it;
+ for (const TimingLogger::SplitTiming& split : splits_) {
total_ns += split.first;
}
return total_ns;
@@ -171,8 +169,7 @@
void TimingLogger::Dump(std::ostream &os) const {
uint64_t longest_split = 0;
uint64_t total_ns = 0;
- for (auto it = splits_.begin(), end = splits_.end(); it != end; ++it) {
- TimingLogger::SplitTiming split = *it;
+ for (const SplitTiming& split : splits_) {
uint64_t split_time = split.first;
longest_split = std::max(longest_split, split_time);
total_ns += split_time;
@@ -181,8 +178,7 @@
TimeUnit tu = GetAppropriateTimeUnit(longest_split);
uint64_t divisor = GetNsToTimeUnitDivisor(tu);
// Print formatted splits.
- for (auto it = splits_.begin(), end = splits_.end(); it != end; ++it) {
- const TimingLogger::SplitTiming& split = *it;
+ for (const SplitTiming& split : splits_) {
uint64_t split_time = split.first;
if (!precise_ && divisor >= 1000) {
// Make the fractional part 0.
@@ -194,7 +190,6 @@
os << name_ << ": end, " << NsToMs(total_ns) << " ms\n";
}
-
TimingLogger::ScopedSplit::ScopedSplit(const char* label, TimingLogger* timing_logger) {
DCHECK(label != NULL) << "New scoped split (" << label << ") with null label.";
CHECK(timing_logger != NULL) << "New scoped split (" << label << ") without TimingLogger.";
diff --git a/runtime/common_throws.cc b/runtime/common_throws.cc
index 24d16c4..4b6d82b 100644
--- a/runtime/common_throws.cc
+++ b/runtime/common_throws.cc
@@ -175,6 +175,12 @@
va_end(args);
}
+// IllegalAccessException
+
+void ThrowIllegalAccessException(const ThrowLocation* throw_location, const char* msg) {
+ ThrowException(throw_location, "Ljava/lang/IllegalAccessException;", NULL, msg);
+}
+
// IllegalArgumentException
void ThrowIllegalArgumentException(const ThrowLocation* throw_location, const char* msg) {
diff --git a/runtime/common_throws.h b/runtime/common_throws.h
index 792cdef..c06763e 100644
--- a/runtime/common_throws.h
+++ b/runtime/common_throws.h
@@ -92,6 +92,11 @@
__attribute__((__format__(__printf__, 2, 3)))
SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) COLD_ATTR;
+// IllegalAccessException
+
+void ThrowIllegalAccessException(const ThrowLocation* throw_location, const char* msg)
+ SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) COLD_ATTR;
+
// IllegalArgumentException
void ThrowIllegalArgumentException(const ThrowLocation* throw_location, const char* msg)
diff --git a/runtime/debugger.cc b/runtime/debugger.cc
index 26170de..024f830 100644
--- a/runtime/debugger.cc
+++ b/runtime/debugger.cc
@@ -3843,7 +3843,6 @@
} else {
gc::Heap* heap = Runtime::Current()->GetHeap();
const std::vector<gc::space::ContinuousSpace*>& spaces = heap->GetContinuousSpaces();
- ReaderMutexLock mu(self, *Locks::heap_bitmap_lock_);
typedef std::vector<gc::space::ContinuousSpace*>::const_iterator It;
for (It cur = spaces.begin(), end = spaces.end(); cur != end; ++cur) {
if ((*cur)->IsMallocSpace()) {
diff --git a/runtime/entrypoints/entrypoint_utils.cc b/runtime/entrypoints/entrypoint_utils.cc
index 9e5f54c..c81706f 100644
--- a/runtime/entrypoints/entrypoint_utils.cc
+++ b/runtime/entrypoints/entrypoint_utils.cc
@@ -134,7 +134,9 @@
LOG(ERROR) << "Couldn't throw new StackOverflowError because JNI ThrowNew failed.";
CHECK(self->IsExceptionPending());
}
- self->ResetDefaultStackEnd(); // Return to default stack size.
+
+ bool explicit_overflow_check = Runtime::Current()->ExplicitStackOverflowChecks();
+ self->ResetDefaultStackEnd(!explicit_overflow_check); // Return to default stack size.
}
JValue InvokeProxyInvocationHandler(ScopedObjectAccessUnchecked& soa, const char* shorty,
diff --git a/runtime/fault_handler.cc b/runtime/fault_handler.cc
index f9f3e25..fcb567e 100644
--- a/runtime/fault_handler.cc
+++ b/runtime/fault_handler.cc
@@ -61,8 +61,11 @@
void FaultManager::HandleFault(int sig, siginfo_t* info, void* context) {
bool handled = false;
+ LOG(DEBUG) << "Handling fault";
if (IsInGeneratedCode(context)) {
+ LOG(DEBUG) << "in generated code, looking for handler";
for (auto& handler : handlers_) {
+ LOG(DEBUG) << "invoking Action on handler " << handler;
handled = handler->Action(sig, info, context);
if (handled) {
return;
@@ -71,7 +74,7 @@
}
if (!handled) {
- LOG(INFO)<< "Caught unknown SIGSEGV in ART fault handler";
+ LOG(ERROR)<< "Caught unknown SIGSEGV in ART fault handler";
oldaction_.sa_sigaction(sig, info, context);
}
}
@@ -96,19 +99,23 @@
bool FaultManager::IsInGeneratedCode(void *context) {
// We can only be running Java code in the current thread if it
// is in Runnable state.
+ LOG(DEBUG) << "Checking for generated code";
Thread* thread = Thread::Current();
if (thread == nullptr) {
+ LOG(DEBUG) << "no current thread";
return false;
}
ThreadState state = thread->GetState();
if (state != kRunnable) {
+ LOG(DEBUG) << "not runnable";
return false;
}
// Current thread is runnable.
// Make sure it has the mutator lock.
if (!Locks::mutator_lock_->IsSharedHeld(thread)) {
+ LOG(DEBUG) << "no lock";
return false;
}
@@ -120,7 +127,9 @@
GetMethodAndReturnPC(context, /*out*/potential_method, /*out*/return_pc);
// If we don't have a potential method, we're outta here.
+ LOG(DEBUG) << "potential method: " << potential_method;
if (potential_method == 0) {
+ LOG(DEBUG) << "no method";
return false;
}
@@ -133,19 +142,23 @@
// Check that the class pointer inside the object is not null and is aligned.
mirror::Class* cls = method_obj->GetClass<kVerifyNone>();
if (cls == nullptr) {
+ LOG(DEBUG) << "not a class";
return false;
}
if (!IsAligned<kObjectAlignment>(cls)) {
+ LOG(DEBUG) << "not aligned";
return false;
}
if (!VerifyClassClass(cls)) {
+ LOG(DEBUG) << "not a class class";
return false;
}
// Now make sure the class is a mirror::ArtMethod.
if (!cls->IsArtMethodClass()) {
+ LOG(DEBUG) << "not a method";
return false;
}
@@ -153,7 +166,15 @@
// at the return PC address.
mirror::ArtMethod* method =
reinterpret_cast<mirror::ArtMethod*>(potential_method);
- return method->ToDexPc(return_pc, false) != DexFile::kDexNoIndex;
+ if (true || kIsDebugBuild) {
+ LOG(DEBUG) << "looking for dex pc for return pc " << std::hex << return_pc;
+ const void* code = Runtime::Current()->GetInstrumentation()->GetQuickCodeFor(method);
+ uint32_t sought_offset = return_pc - reinterpret_cast<uintptr_t>(code);
+ LOG(DEBUG) << "pc offset: " << std::hex << sought_offset;
+ }
+ uint32_t dexpc = method->ToDexPc(return_pc, false);
+ LOG(DEBUG) << "dexpc: " << dexpc;
+ return dexpc != DexFile::kDexNoIndex;
}
//
diff --git a/runtime/gc/accounting/heap_bitmap-inl.h b/runtime/gc/accounting/heap_bitmap-inl.h
index 18b93d4..04e85d2 100644
--- a/runtime/gc/accounting/heap_bitmap-inl.h
+++ b/runtime/gc/accounting/heap_bitmap-inl.h
@@ -19,6 +19,8 @@
#include "heap_bitmap.h"
+#include "space_bitmap-inl.h"
+
namespace art {
namespace gc {
namespace accounting {
@@ -34,6 +36,55 @@
}
}
+inline bool HeapBitmap::Test(const mirror::Object* obj) {
+ SpaceBitmap* bitmap = GetContinuousSpaceBitmap(obj);
+ if (LIKELY(bitmap != nullptr)) {
+ return bitmap->Test(obj);
+ } else {
+ return GetDiscontinuousSpaceObjectSet(obj) != NULL;
+ }
+}
+
+inline void HeapBitmap::Clear(const mirror::Object* obj) {
+ SpaceBitmap* bitmap = GetContinuousSpaceBitmap(obj);
+ if (LIKELY(bitmap != nullptr)) {
+ bitmap->Clear(obj);
+ } else {
+ ObjectSet* set = GetDiscontinuousSpaceObjectSet(obj);
+ DCHECK(set != NULL);
+ set->Clear(obj);
+ }
+}
+
+inline void HeapBitmap::Set(const mirror::Object* obj) {
+ SpaceBitmap* bitmap = GetContinuousSpaceBitmap(obj);
+ if (LIKELY(bitmap != NULL)) {
+ bitmap->Set(obj);
+ } else {
+ ObjectSet* set = GetDiscontinuousSpaceObjectSet(obj);
+ DCHECK(set != NULL);
+ set->Set(obj);
+ }
+}
+
+inline SpaceBitmap* HeapBitmap::GetContinuousSpaceBitmap(const mirror::Object* obj) const {
+ for (const auto& bitmap : continuous_space_bitmaps_) {
+ if (bitmap->HasAddress(obj)) {
+ return bitmap;
+ }
+ }
+ return nullptr;
+}
+
+inline ObjectSet* HeapBitmap::GetDiscontinuousSpaceObjectSet(const mirror::Object* obj) const {
+ for (const auto& space_set : discontinuous_space_sets_) {
+ if (space_set->Test(obj)) {
+ return space_set;
+ }
+ }
+ return nullptr;
+}
+
} // namespace accounting
} // namespace gc
} // namespace art
diff --git a/runtime/gc/accounting/heap_bitmap.h b/runtime/gc/accounting/heap_bitmap.h
index b23b12e..f729c0e 100644
--- a/runtime/gc/accounting/heap_bitmap.h
+++ b/runtime/gc/accounting/heap_bitmap.h
@@ -31,54 +31,11 @@
class HeapBitmap {
public:
- bool Test(const mirror::Object* obj) SHARED_LOCKS_REQUIRED(Locks::heap_bitmap_lock_) {
- SpaceBitmap* bitmap = GetContinuousSpaceBitmap(obj);
- if (LIKELY(bitmap != nullptr)) {
- return bitmap->Test(obj);
- } else {
- return GetDiscontinuousSpaceObjectSet(obj) != NULL;
- }
- }
-
- void Clear(const mirror::Object* obj) EXCLUSIVE_LOCKS_REQUIRED(Locks::heap_bitmap_lock_) {
- SpaceBitmap* bitmap = GetContinuousSpaceBitmap(obj);
- if (LIKELY(bitmap != NULL)) {
- bitmap->Clear(obj);
- } else {
- ObjectSet* set = GetDiscontinuousSpaceObjectSet(obj);
- DCHECK(set != NULL);
- set->Clear(obj);
- }
- }
-
- void Set(const mirror::Object* obj) EXCLUSIVE_LOCKS_REQUIRED(Locks::heap_bitmap_lock_) {
- SpaceBitmap* bitmap = GetContinuousSpaceBitmap(obj);
- if (LIKELY(bitmap != NULL)) {
- bitmap->Set(obj);
- } else {
- ObjectSet* set = GetDiscontinuousSpaceObjectSet(obj);
- DCHECK(set != NULL);
- set->Set(obj);
- }
- }
-
- SpaceBitmap* GetContinuousSpaceBitmap(const mirror::Object* obj) {
- for (const auto& bitmap : continuous_space_bitmaps_) {
- if (bitmap->HasAddress(obj)) {
- return bitmap;
- }
- }
- return nullptr;
- }
-
- ObjectSet* GetDiscontinuousSpaceObjectSet(const mirror::Object* obj) {
- for (const auto& space_set : discontinuous_space_sets_) {
- if (space_set->Test(obj)) {
- return space_set;
- }
- }
- return nullptr;
- }
+ bool Test(const mirror::Object* obj) SHARED_LOCKS_REQUIRED(Locks::heap_bitmap_lock_);
+ void Clear(const mirror::Object* obj) EXCLUSIVE_LOCKS_REQUIRED(Locks::heap_bitmap_lock_);
+ void Set(const mirror::Object* obj) EXCLUSIVE_LOCKS_REQUIRED(Locks::heap_bitmap_lock_);
+ SpaceBitmap* GetContinuousSpaceBitmap(const mirror::Object* obj) const;
+ ObjectSet* GetDiscontinuousSpaceObjectSet(const mirror::Object* obj) const;
void Walk(ObjectCallback* callback, void* arg)
SHARED_LOCKS_REQUIRED(Locks::heap_bitmap_lock_);
diff --git a/runtime/gc/accounting/mod_union_table-inl.h b/runtime/gc/accounting/mod_union_table-inl.h
index 76719b6..c756127 100644
--- a/runtime/gc/accounting/mod_union_table-inl.h
+++ b/runtime/gc/accounting/mod_union_table-inl.h
@@ -32,39 +32,8 @@
space::ContinuousSpace* space)
: ModUnionTableReferenceCache(name, heap, space) {}
- bool AddReference(const mirror::Object* /* obj */, const mirror::Object* ref) ALWAYS_INLINE {
- for (space::ContinuousSpace* space : GetHeap()->GetContinuousSpaces()) {
- if (space->HasAddress(ref)) {
- return !space->IsImageSpace();
- }
- }
- // Assume it points to a large object.
- // TODO: Check.
- return true;
- }
-};
-
-// A mod-union table to record Zygote references to the alloc space.
-class ModUnionTableToAllocspace : public ModUnionTableReferenceCache {
- public:
- explicit ModUnionTableToAllocspace(const std::string& name, Heap* heap,
- space::ContinuousSpace* space)
- : ModUnionTableReferenceCache(name, heap, space) {}
-
- bool AddReference(const mirror::Object* /* obj */, const mirror::Object* ref) ALWAYS_INLINE {
- const std::vector<space::ContinuousSpace*>& spaces = GetHeap()->GetContinuousSpaces();
- typedef std::vector<space::ContinuousSpace*>::const_iterator It;
- for (It it = spaces.begin(); it != spaces.end(); ++it) {
- space::ContinuousSpace* space = *it;
- if (space->Contains(ref)) {
- // The allocation space is always considered for collection whereas the Zygote space is
- // only considered for full GC.
- return space->GetGcRetentionPolicy() == space::kGcRetentionPolicyAlwaysCollect;
- }
- }
- // Assume it points to a large object.
- // TODO: Check.
- return true;
+ bool ShouldAddReference(const mirror::Object* ref) const OVERRIDE ALWAYS_INLINE {
+ return !space_->HasAddress(ref);
}
};
diff --git a/runtime/gc/accounting/mod_union_table.cc b/runtime/gc/accounting/mod_union_table.cc
index 314f3c5..34ca654 100644
--- a/runtime/gc/accounting/mod_union_table.cc
+++ b/runtime/gc/accounting/mod_union_table.cc
@@ -76,7 +76,7 @@
}
// Extra parameters are required since we use this same visitor signature for checking objects.
- void operator()(Object* obj, MemberOffset offset, bool /* static */) const
+ void operator()(Object* obj, MemberOffset offset, bool /*is_static*/) const
SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
// Only add the reference if it is non null and fits our criteria.
mirror::HeapReference<Object>* obj_ptr = obj->GetFieldObjectReferenceAddr(offset);
@@ -123,12 +123,12 @@
}
// Extra parameters are required since we use this same visitor signature for checking objects.
- void operator()(Object* obj, MemberOffset offset, bool /* static */) const
+ void operator()(Object* obj, MemberOffset offset, bool /*is_static*/) const
SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
mirror::HeapReference<Object>* ref_ptr = obj->GetFieldObjectReferenceAddr(offset);
mirror::Object* ref = ref_ptr->AsMirrorPtr();
// Only add the reference if it is non null and fits our criteria.
- if (ref != nullptr && mod_union_table_->AddReference(obj, ref)) {
+ if (ref != nullptr && mod_union_table_->ShouldAddReference(ref)) {
// Push the adddress of the reference.
references_->push_back(ref_ptr);
}
@@ -168,10 +168,10 @@
}
// Extra parameters are required since we use this same visitor signature for checking objects.
- void operator()(Object* obj, MemberOffset offset, bool /* is_static */) const
+ void operator()(Object* obj, MemberOffset offset, bool /*is_static*/) const
SHARED_LOCKS_REQUIRED(Locks::heap_bitmap_lock_, Locks::mutator_lock_) {
mirror::Object* ref = obj->GetFieldObject<mirror::Object>(offset, false);
- if (ref != nullptr && mod_union_table_->AddReference(obj, ref) &&
+ if (ref != nullptr && mod_union_table_->ShouldAddReference(ref) &&
references_.find(ref) == references_.end()) {
Heap* heap = mod_union_table_->GetHeap();
space::ContinuousSpace* from_space = heap->FindContinuousSpaceFromObject(obj, false);
@@ -260,8 +260,7 @@
void ModUnionTableReferenceCache::UpdateAndMarkReferences(MarkHeapReferenceCallback* callback,
void* arg) {
- Heap* heap = GetHeap();
- CardTable* card_table = heap->GetCardTable();
+ CardTable* card_table = heap_->GetCardTable();
std::vector<mirror::HeapReference<Object>*> cards_references;
ModUnionReferenceVisitor add_visitor(this, &cards_references);
@@ -271,7 +270,7 @@
cards_references.clear();
uintptr_t start = reinterpret_cast<uintptr_t>(card_table->AddrFromCard(card));
uintptr_t end = start + CardTable::kCardSize;
- auto* space = heap->FindContinuousSpaceFromObject(reinterpret_cast<Object*>(start), false);
+ auto* space = heap_->FindContinuousSpaceFromObject(reinterpret_cast<Object*>(start), false);
DCHECK(space != nullptr);
SpaceBitmap* live_bitmap = space->GetLiveBitmap();
live_bitmap->VisitMarkedRange(start, end, add_visitor);
diff --git a/runtime/gc/accounting/mod_union_table.h b/runtime/gc/accounting/mod_union_table.h
index c4b020b..c3a90e2 100644
--- a/runtime/gc/accounting/mod_union_table.h
+++ b/runtime/gc/accounting/mod_union_table.h
@@ -117,7 +117,7 @@
EXCLUSIVE_LOCKS_REQUIRED(Locks::heap_bitmap_lock_);
// Function that tells whether or not to add a reference to the table.
- virtual bool AddReference(const mirror::Object* obj, const mirror::Object* ref) = 0;
+ virtual bool ShouldAddReference(const mirror::Object* ref) const = 0;
void Dump(std::ostream& os) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
diff --git a/runtime/gc/collector/concurrent_copying.cc b/runtime/gc/collector/concurrent_copying.cc
new file mode 100644
index 0000000..079eeba
--- /dev/null
+++ b/runtime/gc/collector/concurrent_copying.cc
@@ -0,0 +1,25 @@
+/*
+ * Copyright (C) 2014 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.
+ */
+
+#include "concurrent_copying.h"
+
+namespace art {
+namespace gc {
+namespace collector {
+
+} // namespace collector
+} // namespace gc
+} // namespace art
diff --git a/runtime/gc/collector/concurrent_copying.h b/runtime/gc/collector/concurrent_copying.h
new file mode 100644
index 0000000..ab26a9c
--- /dev/null
+++ b/runtime/gc/collector/concurrent_copying.h
@@ -0,0 +1,56 @@
+/*
+ * Copyright (C) 2014 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.
+ */
+
+#ifndef ART_RUNTIME_GC_COLLECTOR_CONCURRENT_COPYING_H_
+#define ART_RUNTIME_GC_COLLECTOR_CONCURRENT_COPYING_H_
+
+#include "garbage_collector.h"
+
+namespace art {
+namespace gc {
+namespace collector {
+
+class ConcurrentCopying : public GarbageCollector {
+ public:
+ explicit ConcurrentCopying(Heap* heap, bool generational = false,
+ const std::string& name_prefix = "")
+ : GarbageCollector(heap,
+ name_prefix + (name_prefix.empty() ? "" : " ") +
+ "concurrent copying + mark sweep") {}
+
+ ~ConcurrentCopying() {}
+
+ virtual void InitializePhase() OVERRIDE {}
+ virtual void MarkingPhase() OVERRIDE {}
+ virtual void ReclaimPhase() OVERRIDE {}
+ virtual void FinishPhase() OVERRIDE {}
+ virtual GcType GetGcType() const OVERRIDE {
+ return kGcTypePartial;
+ }
+ virtual CollectorType GetCollectorType() const OVERRIDE {
+ return kCollectorTypeCC;
+ }
+ virtual void RevokeAllThreadLocalBuffers() OVERRIDE {}
+
+ private:
+ DISALLOW_COPY_AND_ASSIGN(ConcurrentCopying);
+};
+
+} // namespace collector
+} // namespace gc
+} // namespace art
+
+#endif // ART_RUNTIME_GC_COLLECTOR_CONCURRENT_COPYING_H_
diff --git a/runtime/gc/collector/garbage_collector.cc b/runtime/gc/collector/garbage_collector.cc
index 0c7565c..07951e0 100644
--- a/runtime/gc/collector/garbage_collector.cc
+++ b/runtime/gc/collector/garbage_collector.cc
@@ -46,8 +46,7 @@
ResetCumulativeStatistics();
}
-void GarbageCollector::HandleDirtyObjectsPhase() {
- LOG(FATAL) << "Unreachable";
+void GarbageCollector::PausePhase() {
}
void GarbageCollector::RegisterPause(uint64_t nano_length) {
@@ -86,12 +85,13 @@
// Pause is the entire length of the GC.
uint64_t pause_start = NanoTime();
ATRACE_BEGIN("Application threads suspended");
- // Mutator lock may be already exclusively held when we do garbage collections for changing the
- // current collector / allocator during process state updates.
+ // Mutator lock may be already exclusively held when we do garbage collections for changing
+ // the current collector / allocator during process state updates.
if (Locks::mutator_lock_->IsExclusiveHeld(self)) {
// PreGcRosAllocVerification() is called in Heap::TransitionCollector().
RevokeAllThreadLocalBuffers();
MarkingPhase();
+ PausePhase();
ReclaimPhase();
// PostGcRosAllocVerification() is called in Heap::TransitionCollector().
} else {
@@ -101,6 +101,7 @@
GetHeap()->PreGcRosAllocVerification(&timings_);
RevokeAllThreadLocalBuffers();
MarkingPhase();
+ PausePhase();
ReclaimPhase();
GetHeap()->PostGcRosAllocVerification(&timings_);
ATRACE_BEGIN("Resuming mutator threads");
@@ -125,7 +126,7 @@
ATRACE_END();
ATRACE_BEGIN("All mutator threads suspended");
GetHeap()->PreGcRosAllocVerification(&timings_);
- HandleDirtyObjectsPhase();
+ PausePhase();
RevokeAllThreadLocalBuffers();
GetHeap()->PostGcRosAllocVerification(&timings_);
ATRACE_END();
@@ -141,12 +142,20 @@
FinishPhase();
break;
}
+ case kCollectorTypeCC: {
+ // To be implemented.
+ break;
+ }
default: {
LOG(FATAL) << "Unreachable collector type=" << static_cast<size_t>(collector_type);
break;
}
}
-
+ // Add the current timings to the cumulative timings.
+ cumulative_timings_.AddLogger(timings_);
+ // Update cumulative statistics with how many bytes the GC iteration freed.
+ total_freed_objects_ += GetFreedObjects() + GetFreedLargeObjects();
+ total_freed_bytes_ += GetFreedBytes() + GetFreedLargeObjectBytes();
uint64_t end_time = NanoTime();
duration_ns_ = end_time - start_time;
total_time_ns_ += GetDurationNs();
diff --git a/runtime/gc/collector/garbage_collector.h b/runtime/gc/collector/garbage_collector.h
index f8c4579..ccfa9cf 100644
--- a/runtime/gc/collector/garbage_collector.h
+++ b/runtime/gc/collector/garbage_collector.h
@@ -129,7 +129,7 @@
virtual void MarkingPhase() = 0;
// Only called for concurrent GCs.
- virtual void HandleDirtyObjectsPhase() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+ virtual void PausePhase();
// Called with mutators running.
virtual void ReclaimPhase() = 0;
diff --git a/runtime/gc/collector/mark_sweep-inl.h b/runtime/gc/collector/mark_sweep-inl.h
index 1cb2adb..974952d 100644
--- a/runtime/gc/collector/mark_sweep-inl.h
+++ b/runtime/gc/collector/mark_sweep-inl.h
@@ -21,7 +21,7 @@
#include "gc/heap.h"
#include "mirror/art_field.h"
-#include "mirror/class.h"
+#include "mirror/class-inl.h"
#include "mirror/object_array-inl.h"
#include "mirror/reference.h"
diff --git a/runtime/gc/collector/mark_sweep.cc b/runtime/gc/collector/mark_sweep.cc
index 8abf5e2..91ccd64 100644
--- a/runtime/gc/collector/mark_sweep.cc
+++ b/runtime/gc/collector/mark_sweep.cc
@@ -27,31 +27,20 @@
#include "base/mutex-inl.h"
#include "base/timing_logger.h"
#include "gc/accounting/card_table-inl.h"
-#include "gc/accounting/heap_bitmap.h"
+#include "gc/accounting/heap_bitmap-inl.h"
#include "gc/accounting/mod_union_table.h"
#include "gc/accounting/space_bitmap-inl.h"
#include "gc/heap.h"
#include "gc/space/image_space.h"
#include "gc/space/large_object_space.h"
#include "gc/space/space-inl.h"
-#include "indirect_reference_table.h"
-#include "intern_table.h"
-#include "jni_internal.h"
-#include "monitor.h"
#include "mark_sweep-inl.h"
-#include "mirror/art_field.h"
#include "mirror/art_field-inl.h"
-#include "mirror/class-inl.h"
-#include "mirror/class_loader.h"
-#include "mirror/dex_cache.h"
-#include "mirror/reference-inl.h"
#include "mirror/object-inl.h"
-#include "mirror/object_array.h"
-#include "mirror/object_array-inl.h"
#include "runtime.h"
+#include "scoped_thread_state_change.h"
#include "thread-inl.h"
#include "thread_list.h"
-#include "verifier/method_verifier.h"
using ::art::mirror::ArtField;
using ::art::mirror::Class;
@@ -163,49 +152,38 @@
}
}
-void MarkSweep::HandleDirtyObjectsPhase() {
- TimingLogger::ScopedSplit split("(Paused)HandleDirtyObjectsPhase", &timings_);
+void MarkSweep::PausePhase() {
+ TimingLogger::ScopedSplit split("(Paused)PausePhase", &timings_);
Thread* self = Thread::Current();
Locks::mutator_lock_->AssertExclusiveHeld(self);
-
- {
+ if (IsConcurrent()) {
+ // Handle the dirty objects if we are a concurrent GC.
WriterMutexLock mu(self, *Locks::heap_bitmap_lock_);
-
// Re-mark root set.
ReMarkRoots();
-
// Scan dirty objects, this is only required if we are not doing concurrent GC.
RecursiveMarkDirtyObjects(true, accounting::CardTable::kCardDirty);
}
-
ProcessReferences(self);
-
- // Only need to do this if we have the card mark verification on, and only during concurrent GC.
- if (GetHeap()->verify_missing_card_marks_ || GetHeap()->verify_pre_gc_heap_||
- GetHeap()->verify_post_gc_heap_) {
+ {
+ timings_.NewSplit("SwapStacks");
WriterMutexLock mu(self, *Locks::heap_bitmap_lock_);
- // This second sweep makes sure that we don't have any objects in the live stack which point to
- // freed objects. These cause problems since their references may be previously freed objects.
- SweepArray(GetHeap()->allocation_stack_.get(), false);
- // Since SweepArray() above resets the (active) allocation
- // stack. Need to revoke the thread-local allocation stacks that
- // point into it.
+ heap_->SwapStacks(self);
+ live_stack_freeze_size_ = heap_->GetLiveStack()->Size();
+ // Need to revoke all the thread local allocation stacks since we just swapped the allocation
+ // stacks and don't want anybody to allocate into the live stack.
RevokeAllThreadLocalAllocationStacks(self);
}
-
timings_.StartSplit("PreSweepingGcVerification");
heap_->PreSweepingGcVerification(this);
timings_.EndSplit();
-
- // Ensure that nobody inserted items in the live stack after we swapped the stacks.
- ReaderMutexLock mu(self, *Locks::heap_bitmap_lock_);
- CHECK_GE(live_stack_freeze_size_, GetHeap()->GetLiveStack()->Size());
-
- // Disallow new system weaks to prevent a race which occurs when someone adds a new system
- // weak before we sweep them. Since this new system weak may not be marked, the GC may
- // incorrectly sweep it. This also fixes a race where interning may attempt to return a strong
- // reference to a string that is about to be swept.
- Runtime::Current()->DisallowNewSystemWeaks();
+ if (IsConcurrent()) {
+ // Disallow new system weaks to prevent a race which occurs when someone adds a new system
+ // weak before we sweep them. Since this new system weak may not be marked, the GC may
+ // incorrectly sweep it. This also fixes a race where interning may attempt to return a strong
+ // reference to a string that is about to be swept.
+ Runtime::Current()->DisallowNewSystemWeaks();
+ }
}
void MarkSweep::PreCleanCards() {
@@ -227,8 +205,7 @@
// reference write are visible to the GC before the card is scanned (this is due to locks being
// acquired / released in the checkpoint code).
// The other roots are also marked to help reduce the pause.
- MarkThreadRoots(self);
- // TODO: Only mark the dirty roots.
+ MarkRootsCheckpoint(self, false);
MarkNonThreadRoots();
MarkConcurrentRoots(
static_cast<VisitRootFlags>(kVisitRootFlagClearRootLog | kVisitRootFlagNewRoots));
@@ -241,6 +218,7 @@
void MarkSweep::RevokeAllThreadLocalAllocationStacks(Thread* self) {
if (kUseThreadLocalAllocationStack) {
+ timings_.NewSplit("RevokeAllThreadLocalAllocationStacks");
Locks::mutator_lock_->AssertExclusiveHeld(self);
heap_->RevokeAllThreadLocalAllocationStacks(self);
}
@@ -256,15 +234,8 @@
// Process dirty cards and add dirty cards to mod union tables.
heap_->ProcessCards(timings_, false);
- // Need to do this before the checkpoint since we don't want any threads to add references to
- // the live stack during the recursive mark.
- timings_.NewSplit("SwapStacks");
- heap_->SwapStacks(self);
-
WriterMutexLock mu(self, *Locks::heap_bitmap_lock_);
MarkRoots(self);
- live_stack_freeze_size_ = heap_->GetLiveStack()->Size();
- UpdateAndMarkModUnion();
MarkReachableObjects();
// Pre-clean dirtied cards to reduce pauses.
PreCleanCards();
@@ -284,18 +255,8 @@
}
}
-void MarkSweep::MarkThreadRoots(Thread* self) {
- MarkRootsCheckpoint(self);
-}
-
void MarkSweep::MarkReachableObjects() {
- // Mark everything allocated since the last as GC live so that we can sweep concurrently,
- // knowing that new allocations won't be marked as live.
- timings_.StartSplit("MarkStackAsLive");
- accounting::ObjectStack* live_stack = heap_->GetLiveStack();
- heap_->MarkAllocStackAsLive(live_stack);
- live_stack->Reset();
- timings_.EndSplit();
+ UpdateAndMarkModUnion();
// Recursively mark all the non-image bits set in the mark bitmap.
RecursiveMark();
}
@@ -303,44 +264,10 @@
void MarkSweep::ReclaimPhase() {
TimingLogger::ScopedSplit split("ReclaimPhase", &timings_);
Thread* self = Thread::Current();
-
- if (!IsConcurrent()) {
- ProcessReferences(self);
- }
-
- {
- WriterMutexLock mu(self, *Locks::heap_bitmap_lock_);
- SweepSystemWeaks();
- }
-
+ SweepSystemWeaks(self);
if (IsConcurrent()) {
Runtime::Current()->AllowNewSystemWeaks();
-
- TimingLogger::ScopedSplit split("UnMarkAllocStack", &timings_);
- WriterMutexLock mu(self, *Locks::heap_bitmap_lock_);
- accounting::ObjectStack* allocation_stack = GetHeap()->allocation_stack_.get();
- if (!kPreCleanCards) {
- // The allocation stack contains things allocated since the start of the GC. These may have
- // been marked during this GC meaning they won't be eligible for reclaiming in the next
- // sticky GC. Unmark these objects so that they are eligible for reclaiming in the next
- // sticky GC.
- // There is a race here which is safely handled. Another thread such as the hprof could
- // have flushed the alloc stack after we resumed the threads. This is safe however, since
- // reseting the allocation stack zeros it out with madvise. This means that we will either
- // read NULLs or attempt to unmark a newly allocated object which will not be marked in the
- // first place.
- // We can't do this if we pre-clean cards since we will unmark objects which are no longer on
- // a dirty card since we aged cards during the pre-cleaning process.
- mirror::Object** end = allocation_stack->End();
- for (mirror::Object** it = allocation_stack->Begin(); it != end; ++it) {
- const Object* obj = *it;
- if (obj != nullptr) {
- UnMarkObjectNonNull(obj);
- }
- }
- }
}
-
{
WriterMutexLock mu(self, *Locks::heap_bitmap_lock_);
@@ -414,28 +341,6 @@
reinterpret_cast<MarkSweep*>(arg)->MarkObject(ref->AsMirrorPtr());
}
-inline void MarkSweep::UnMarkObjectNonNull(const Object* obj) {
- DCHECK(!immune_region_.ContainsObject(obj));
- if (kUseBrooksPointer) {
- // Verify all the objects have the correct Brooks pointer installed.
- obj->AssertSelfBrooksPointer();
- }
- // Try to take advantage of locality of references within a space, failing this find the space
- // the hard way.
- accounting::SpaceBitmap* object_bitmap = current_space_bitmap_;
- if (UNLIKELY(!object_bitmap->HasAddress(obj))) {
- accounting::SpaceBitmap* new_bitmap = mark_bitmap_->GetContinuousSpaceBitmap(obj);
- if (LIKELY(new_bitmap != NULL)) {
- object_bitmap = new_bitmap;
- } else {
- MarkLargeObject(obj, false);
- return;
- }
- }
- DCHECK(object_bitmap->HasAddress(obj));
- object_bitmap->Clear(obj);
-}
-
inline void MarkSweep::MarkObjectNonNull(Object* obj) {
DCHECK(obj != nullptr);
if (kUseBrooksPointer) {
@@ -590,7 +495,7 @@
timings_.EndSplit();
RevokeAllThreadLocalAllocationStacks(self);
} else {
- MarkThreadRoots(self);
+ MarkRootsCheckpoint(self, kRevokeRosAllocThreadLocalBuffersAtCheckpoint);
// At this point the live stack should no longer have any mutators which push into it.
MarkNonThreadRoots();
MarkConcurrentRoots(
@@ -1020,10 +925,10 @@
}
}
-void MarkSweep::SweepSystemWeaks() {
- Runtime* runtime = Runtime::Current();
+void MarkSweep::SweepSystemWeaks(Thread* self) {
+ WriterMutexLock mu(self, *Locks::heap_bitmap_lock_);
timings_.StartSplit("SweepSystemWeaks");
- runtime->SweepSystemWeaks(IsMarkedCallback, this);
+ Runtime::Current()->SweepSystemWeaks(IsMarkedCallback, this);
timings_.EndSplit();
}
@@ -1034,14 +939,13 @@
}
void MarkSweep::VerifyIsLive(const Object* obj) {
- Heap* heap = GetHeap();
- if (!heap->GetLiveBitmap()->Test(obj)) {
- space::LargeObjectSpace* large_object_space = GetHeap()->GetLargeObjectsSpace();
+ if (!heap_->GetLiveBitmap()->Test(obj)) {
+ space::LargeObjectSpace* large_object_space = heap_->GetLargeObjectsSpace();
if (!large_object_space->GetLiveObjects()->Test(obj)) {
- if (std::find(heap->allocation_stack_->Begin(), heap->allocation_stack_->End(), obj) ==
- heap->allocation_stack_->End()) {
+ if (std::find(heap_->allocation_stack_->Begin(), heap_->allocation_stack_->End(), obj) ==
+ heap_->allocation_stack_->End()) {
// Object not found!
- heap->DumpSpaces();
+ heap_->DumpSpaces();
LOG(FATAL) << "Found dead object " << obj;
}
}
@@ -1055,9 +959,14 @@
class CheckpointMarkThreadRoots : public Closure {
public:
- explicit CheckpointMarkThreadRoots(MarkSweep* mark_sweep) : mark_sweep_(mark_sweep) {}
+ explicit CheckpointMarkThreadRoots(MarkSweep* mark_sweep,
+ bool revoke_ros_alloc_thread_local_buffers_at_checkpoint)
+ : mark_sweep_(mark_sweep),
+ revoke_ros_alloc_thread_local_buffers_at_checkpoint_(
+ revoke_ros_alloc_thread_local_buffers_at_checkpoint) {
+ }
- virtual void Run(Thread* thread) NO_THREAD_SAFETY_ANALYSIS {
+ virtual void Run(Thread* thread) OVERRIDE NO_THREAD_SAFETY_ANALYSIS {
ATRACE_BEGIN("Marking thread roots");
// Note: self is not necessarily equal to thread since thread may be suspended.
Thread* self = Thread::Current();
@@ -1065,21 +974,22 @@
<< thread->GetState() << " thread " << thread << " self " << self;
thread->VisitRoots(MarkSweep::MarkRootParallelCallback, mark_sweep_);
ATRACE_END();
- if (kUseThreadLocalAllocationStack) {
- thread->RevokeThreadLocalAllocationStack();
- }
- if (kRevokeRosAllocThreadLocalBuffersAtCheckpoint) {
+ if (revoke_ros_alloc_thread_local_buffers_at_checkpoint_) {
+ ATRACE_BEGIN("RevokeRosAllocThreadLocalBuffers");
mark_sweep_->GetHeap()->RevokeRosAllocThreadLocalBuffers(thread);
+ ATRACE_END();
}
mark_sweep_->GetBarrier().Pass(self);
}
private:
- MarkSweep* mark_sweep_;
+ MarkSweep* const mark_sweep_;
+ const bool revoke_ros_alloc_thread_local_buffers_at_checkpoint_;
};
-void MarkSweep::MarkRootsCheckpoint(Thread* self) {
- CheckpointMarkThreadRoots check_point(this);
+void MarkSweep::MarkRootsCheckpoint(Thread* self,
+ bool revoke_ros_alloc_thread_local_buffers_at_checkpoint) {
+ CheckpointMarkThreadRoots check_point(this, revoke_ros_alloc_thread_local_buffers_at_checkpoint);
timings_.StartSplit("MarkRootsCheckpoint");
ThreadList* thread_list = Runtime::Current()->GetThreadList();
// Request the check point is run on all threads returning a count of the threads that must
@@ -1089,10 +999,10 @@
// TODO: optimize to not release locks when there are no threads to wait for.
Locks::heap_bitmap_lock_->ExclusiveUnlock(self);
Locks::mutator_lock_->SharedUnlock(self);
- ThreadState old_state = self->SetState(kWaitingForCheckPointsToRun);
- CHECK_EQ(old_state, kWaitingPerformingGc);
- gc_barrier_->Increment(self, barrier_count);
- self->SetState(kWaitingPerformingGc);
+ {
+ ScopedThreadStateChange tsc(self, kWaitingForCheckPointsToRun);
+ gc_barrier_->Increment(self, barrier_count);
+ }
Locks::mutator_lock_->SharedLock(self);
Locks::heap_bitmap_lock_->ExclusiveLock(self);
timings_.EndSplit();
@@ -1108,7 +1018,7 @@
size_t freed_objects = 0;
size_t freed_large_objects = 0;
// How many objects are left in the array, modified after each space is swept.
- Object** objects = const_cast<Object**>(allocations->Begin());
+ Object** objects = allocations->Begin();
size_t count = allocations->Size();
// Change the order to ensure that the non-moving space last swept as an optimization.
std::vector<space::ContinuousSpace*> sweep_spaces;
@@ -1206,6 +1116,16 @@
}
void MarkSweep::Sweep(bool swap_bitmaps) {
+ // Ensure that nobody inserted items in the live stack after we swapped the stacks.
+ CHECK_GE(live_stack_freeze_size_, GetHeap()->GetLiveStack()->Size());
+ // Mark everything allocated since the last as GC live so that we can sweep concurrently,
+ // knowing that new allocations won't be marked as live.
+ timings_.StartSplit("MarkStackAsLive");
+ accounting::ObjectStack* live_stack = heap_->GetLiveStack();
+ heap_->MarkAllocStackAsLive(live_stack);
+ live_stack->Reset();
+ timings_.EndSplit();
+
DCHECK(mark_stack_->IsEmpty());
TimingLogger::ScopedSplit("Sweep", &timings_);
for (const auto& space : GetHeap()->GetContinuousSpaces()) {
@@ -1326,7 +1246,7 @@
}
obj = mark_stack_->PopBack();
}
- DCHECK(obj != NULL);
+ DCHECK(obj != nullptr);
ScanObject(obj);
}
}
@@ -1347,14 +1267,8 @@
void MarkSweep::FinishPhase() {
TimingLogger::ScopedSplit split("FinishPhase", &timings_);
// Can't enqueue references if we hold the mutator lock.
- Heap* heap = GetHeap();
timings_.NewSplit("PostGcVerification");
- heap->PostGcVerification(this);
- // Update the cumulative statistics.
- total_freed_objects_ += GetFreedObjects() + GetFreedLargeObjects();
- total_freed_bytes_ += GetFreedBytes() + GetFreedLargeObjectBytes();
- // Ensure that the mark stack is empty.
- CHECK(mark_stack_->IsEmpty());
+ heap_->PostGcVerification(this);
if (kCountScannedTypes) {
VLOG(gc) << "MarkSweep scanned classes=" << class_count_ << " arrays=" << array_count_
<< " other=" << other_count_;
@@ -1375,22 +1289,10 @@
VLOG(gc) << "Marked: null=" << mark_null_count_ << " immune=" << mark_immune_count_
<< " fastpath=" << mark_fastpath_count_ << " slowpath=" << mark_slowpath_count_;
}
- // Update the cumulative loggers.
- cumulative_timings_.Start();
- cumulative_timings_.AddLogger(timings_);
- cumulative_timings_.End();
- // Clear all of the spaces' mark bitmaps.
- for (const auto& space : GetHeap()->GetContinuousSpaces()) {
- accounting::SpaceBitmap* bitmap = space->GetMarkBitmap();
- if (bitmap != nullptr &&
- space->GetGcRetentionPolicy() != space::kGcRetentionPolicyNeverCollect) {
- bitmap->Clear();
- }
- }
+ CHECK(mark_stack_->IsEmpty()); // Ensure that the mark stack is empty.
mark_stack_->Reset();
- // Reset the marked large objects.
- space::LargeObjectSpace* large_objects = GetHeap()->GetLargeObjectsSpace();
- large_objects->GetMarkObjects()->Clear();
+ WriterMutexLock mu(Thread::Current(), *Locks::heap_bitmap_lock_);
+ heap_->ClearMarkedObjects();
}
void MarkSweep::RevokeAllThreadLocalBuffers() {
diff --git a/runtime/gc/collector/mark_sweep.h b/runtime/gc/collector/mark_sweep.h
index 84b775a..f1fd546 100644
--- a/runtime/gc/collector/mark_sweep.h
+++ b/runtime/gc/collector/mark_sweep.h
@@ -32,33 +32,22 @@
namespace mirror {
class Class;
class Object;
- template<class T> class ObjectArray;
class Reference;
} // namespace mirror
-class StackVisitor;
class Thread;
enum VisitRootFlags : uint8_t;
namespace gc {
+class Heap;
+
namespace accounting {
- template <typename T> class AtomicStack;
- class MarkIfReachesAllocspaceVisitor;
- class ModUnionClearCardVisitor;
- class ModUnionVisitor;
- class ModUnionTableBitmap;
- class MarkStackChunk;
+ template<typename T> class AtomicStack;
typedef AtomicStack<mirror::Object*> ObjectStack;
class SpaceBitmap;
} // namespace accounting
-namespace space {
- class ContinuousSpace;
-} // namespace space
-
-class Heap;
-
namespace collector {
class MarkSweep : public GarbageCollector {
@@ -69,7 +58,7 @@
virtual void InitializePhase() OVERRIDE;
virtual void MarkingPhase() OVERRIDE SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
- virtual void HandleDirtyObjectsPhase() OVERRIDE EXCLUSIVE_LOCKS_REQUIRED(Locks::mutator_lock_);
+ virtual void PausePhase() OVERRIDE EXCLUSIVE_LOCKS_REQUIRED(Locks::mutator_lock_);
virtual void ReclaimPhase() OVERRIDE SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
virtual void FinishPhase() OVERRIDE SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
virtual void MarkReachableObjects()
@@ -107,7 +96,7 @@
EXCLUSIVE_LOCKS_REQUIRED(Locks::heap_bitmap_lock_)
SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
- void MarkRootsCheckpoint(Thread* self)
+ void MarkRootsCheckpoint(Thread* self, bool revoke_ros_alloc_thread_local_buffers_at_checkpoint)
EXCLUSIVE_LOCKS_REQUIRED(Locks::heap_bitmap_lock_)
SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
@@ -137,8 +126,8 @@
EXCLUSIVE_LOCKS_REQUIRED(Locks::heap_bitmap_lock_)
SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
- // Update and mark references from immune spaces. Virtual as overridden by StickyMarkSweep.
- virtual void UpdateAndMarkModUnion()
+ // Update and mark references from immune spaces.
+ void UpdateAndMarkModUnion()
SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
// Pre clean cards to reduce how much work is needed in the pause.
@@ -170,8 +159,8 @@
SHARED_LOCKS_REQUIRED(Locks::mutator_lock_)
EXCLUSIVE_LOCKS_REQUIRED(Locks::heap_bitmap_lock_);
- void SweepSystemWeaks()
- SHARED_LOCKS_REQUIRED(Locks::mutator_lock_, Locks::heap_bitmap_lock_);
+ void SweepSystemWeaks(Thread* self)
+ SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) LOCKS_EXCLUDED(Locks::heap_bitmap_lock_);
static mirror::Object* VerifySystemWeakIsLiveCallback(mirror::Object* obj, void* arg)
SHARED_LOCKS_REQUIRED(Locks::heap_bitmap_lock_);
@@ -235,17 +224,6 @@
SHARED_LOCKS_REQUIRED(Locks::mutator_lock_)
EXCLUSIVE_LOCKS_REQUIRED(Locks::heap_bitmap_lock_);
- // Unmarks an object by clearing the bit inside of the corresponding bitmap, or if it is in a
- // space set, removing the object from the set.
- void UnMarkObjectNonNull(const mirror::Object* obj)
- SHARED_LOCKS_REQUIRED(Locks::mutator_lock_)
- EXCLUSIVE_LOCKS_REQUIRED(Locks::heap_bitmap_lock_);
-
- // Mark the vm thread roots.
- void MarkThreadRoots(Thread* self)
- EXCLUSIVE_LOCKS_REQUIRED(Locks::heap_bitmap_lock_)
- SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
-
// Marks an object atomically, safe to use from multiple threads.
void MarkObjectNonNullParallel(mirror::Object* obj);
@@ -311,7 +289,7 @@
accounting::ObjectStack* mark_stack_;
- // Immune range, every object inside the immune range is assumed to be marked.
+ // Immune region, every object inside the immune range is assumed to be marked.
ImmuneRegion immune_region_;
// Parallel finger.
diff --git a/runtime/gc/collector/semi_space.cc b/runtime/gc/collector/semi_space.cc
index d4f47ef..222bd63 100644
--- a/runtime/gc/collector/semi_space.cc
+++ b/runtime/gc/collector/semi_space.cc
@@ -25,7 +25,7 @@
#include "base/macros.h"
#include "base/mutex-inl.h"
#include "base/timing_logger.h"
-#include "gc/accounting/heap_bitmap.h"
+#include "gc/accounting/heap_bitmap-inl.h"
#include "gc/accounting/mod_union_table.h"
#include "gc/accounting/remembered_set.h"
#include "gc/accounting/space_bitmap-inl.h"
@@ -726,8 +726,8 @@
return obj;
}
if (from_space_->HasAddress(obj)) {
- mirror::Object* forwarding_address = GetForwardingAddressInFromSpace(const_cast<Object*>(obj));
- return forwarding_address; // Returns either the forwarding address or nullptr.
+ // Returns either the forwarding address or nullptr.
+ return GetForwardingAddressInFromSpace(obj);
} else if (to_space_->HasAddress(obj)) {
// Should be unlikely.
// Already forwarded, must be marked.
@@ -751,38 +751,12 @@
Heap* heap = GetHeap();
timings_.NewSplit("PostGcVerification");
heap->PostGcVerification(this);
-
// Null the "to" and "from" spaces since compacting from one to the other isn't valid until
// further action is done by the heap.
to_space_ = nullptr;
from_space_ = nullptr;
-
- // Update the cumulative statistics
- total_freed_objects_ += GetFreedObjects() + GetFreedLargeObjects();
- total_freed_bytes_ += GetFreedBytes() + GetFreedLargeObjectBytes();
-
- // Ensure that the mark stack is empty.
CHECK(mark_stack_->IsEmpty());
-
- // Update the cumulative loggers.
- cumulative_timings_.Start();
- cumulative_timings_.AddLogger(timings_);
- cumulative_timings_.End();
-
- // Clear all of the spaces' mark bitmaps.
- for (const auto& space : GetHeap()->GetContinuousSpaces()) {
- accounting::SpaceBitmap* bitmap = space->GetMarkBitmap();
- if (bitmap != nullptr &&
- space->GetGcRetentionPolicy() != space::kGcRetentionPolicyNeverCollect) {
- bitmap->Clear();
- }
- }
mark_stack_->Reset();
-
- // Reset the marked large objects.
- space::LargeObjectSpace* large_objects = GetHeap()->GetLargeObjectsSpace();
- large_objects->GetMarkObjects()->Clear();
-
if (generational_) {
// Decide whether to do a whole heap collection or a bump pointer
// only space collection at the next collection by updating
@@ -800,6 +774,9 @@
whole_heap_collection_ = false;
}
}
+ // Clear all of the spaces' mark bitmaps.
+ WriterMutexLock mu(Thread::Current(), *Locks::heap_bitmap_lock_);
+ heap_->ClearMarkedObjects();
}
void SemiSpace::RevokeAllThreadLocalBuffers() {
diff --git a/runtime/gc/collector/semi_space.h b/runtime/gc/collector/semi_space.h
index 52b53aa..f067cb2 100644
--- a/runtime/gc/collector/semi_space.h
+++ b/runtime/gc/collector/semi_space.h
@@ -28,37 +28,28 @@
namespace art {
+class Thread;
+
namespace mirror {
class Class;
class Object;
- template<class T> class ObjectArray;
} // namespace mirror
-class StackVisitor;
-class Thread;
-
namespace gc {
+class Heap;
+
namespace accounting {
template <typename T> class AtomicStack;
- class MarkIfReachesAllocspaceVisitor;
- class ModUnionClearCardVisitor;
- class ModUnionVisitor;
- class ModUnionTableBitmap;
- class MarkStackChunk;
typedef AtomicStack<mirror::Object*> ObjectStack;
class SpaceBitmap;
} // namespace accounting
namespace space {
- class BumpPointerSpace;
class ContinuousMemMapAllocSpace;
class ContinuousSpace;
- class MallocSpace;
} // namespace space
-class Heap;
-
namespace collector {
class SemiSpace : public GarbageCollector {
@@ -189,12 +180,6 @@
void ProcessMarkStack()
EXCLUSIVE_LOCKS_REQUIRED(Locks::mutator_lock_, Locks::heap_bitmap_lock_);
- void ProcessReferences(mirror::Object** soft_references, bool clear_soft_references,
- mirror::Object** weak_references,
- mirror::Object** finalizer_references,
- mirror::Object** phantom_references)
- EXCLUSIVE_LOCKS_REQUIRED(Locks::mutator_lock_, Locks::heap_bitmap_lock_);
-
inline mirror::Object* GetForwardingAddressInFromSpace(mirror::Object* obj) const;
// Revoke all the thread-local buffers.
diff --git a/runtime/gc/collector/sticky_mark_sweep.h b/runtime/gc/collector/sticky_mark_sweep.h
index 934b1bd..4f9dabf 100644
--- a/runtime/gc/collector/sticky_mark_sweep.h
+++ b/runtime/gc/collector/sticky_mark_sweep.h
@@ -46,10 +46,6 @@
SHARED_LOCKS_REQUIRED(Locks::mutator_lock_)
EXCLUSIVE_LOCKS_REQUIRED(Locks::heap_bitmap_lock_);
- // Don't need to do anything special here since we scan all the cards which may have references
- // to the newly allocated objects.
- void UpdateAndMarkModUnion() OVERRIDE { }
-
private:
DISALLOW_COPY_AND_ASSIGN(StickyMarkSweep);
};
diff --git a/runtime/gc/collector_type.h b/runtime/gc/collector_type.h
index 98c27fb..c0a6b6a 100644
--- a/runtime/gc/collector_type.h
+++ b/runtime/gc/collector_type.h
@@ -36,6 +36,8 @@
kCollectorTypeGSS,
// Heap trimming collector, doesn't do any actual collecting.
kCollectorTypeHeapTrim,
+ // A (mostly) concurrent copying collector.
+ kCollectorTypeCC,
};
std::ostream& operator<<(std::ostream& os, const CollectorType& collector_type);
diff --git a/runtime/gc/heap.cc b/runtime/gc/heap.cc
index ff4b4ce..1a32a9a 100644
--- a/runtime/gc/heap.cc
+++ b/runtime/gc/heap.cc
@@ -35,6 +35,7 @@
#include "gc/accounting/mod_union_table-inl.h"
#include "gc/accounting/remembered_set.h"
#include "gc/accounting/space_bitmap-inl.h"
+#include "gc/collector/concurrent_copying.h"
#include "gc/collector/mark_sweep-inl.h"
#include "gc/collector/partial_mark_sweep.h"
#include "gc/collector/semi_space.h"
@@ -328,6 +329,9 @@
bool generational = post_zygote_collector_type_ == kCollectorTypeGSS;
semi_space_collector_ = new collector::SemiSpace(this, generational);
garbage_collectors_.push_back(semi_space_collector_);
+
+ concurrent_copying_collector_ = new collector::ConcurrentCopying(this);
+ garbage_collectors_.push_back(concurrent_copying_collector_);
}
if (running_on_valgrind_) {
@@ -1153,13 +1157,29 @@
ptr = TryToAllocate<true, false>(self, allocator, alloc_size, bytes_allocated, usable_size);
}
+ collector::GcType tried_type = next_gc_type_;
+ if (ptr == nullptr) {
+ const bool gc_ran =
+ CollectGarbageInternal(tried_type, kGcCauseForAlloc, false) != collector::kGcTypeNone;
+ if (was_default_allocator && allocator != GetCurrentAllocator()) {
+ *klass = sirt_klass.get();
+ return nullptr;
+ }
+ if (gc_ran) {
+ ptr = TryToAllocate<true, false>(self, allocator, alloc_size, bytes_allocated, usable_size);
+ }
+ }
+
// Loop through our different Gc types and try to Gc until we get enough free memory.
for (collector::GcType gc_type : gc_plan_) {
if (ptr != nullptr) {
break;
}
+ if (gc_type == tried_type) {
+ continue;
+ }
// Attempt to run the collector, if we succeed, re-try the allocation.
- bool gc_ran =
+ const bool gc_ran =
CollectGarbageInternal(gc_type, kGcCauseForAlloc, false) != collector::kGcTypeNone;
if (was_default_allocator && allocator != GetCurrentAllocator()) {
*klass = sirt_klass.get();
@@ -1430,7 +1450,8 @@
break;
}
default: {
- LOG(FATAL) << "Attempted to transition to invalid collector type";
+ LOG(FATAL) << "Attempted to transition to invalid collector type "
+ << static_cast<size_t>(collector_type);
break;
}
}
@@ -1460,6 +1481,7 @@
collector_type_ = collector_type;
gc_plan_.clear();
switch (collector_type_) {
+ case kCollectorTypeCC: // Fall-through.
case kCollectorTypeSS: // Fall-through.
case kCollectorTypeGSS: {
gc_plan_.push_back(collector::kGcTypeFull);
@@ -1812,12 +1834,19 @@
if (compacting_gc) {
DCHECK(current_allocator_ == kAllocatorTypeBumpPointer ||
current_allocator_ == kAllocatorTypeTLAB);
- gc_type = semi_space_collector_->GetGcType();
- CHECK(temp_space_->IsEmpty());
- semi_space_collector_->SetFromSpace(bump_pointer_space_);
- semi_space_collector_->SetToSpace(temp_space_);
+ if (collector_type_ == kCollectorTypeSS || collector_type_ == kCollectorTypeGSS) {
+ gc_type = semi_space_collector_->GetGcType();
+ semi_space_collector_->SetFromSpace(bump_pointer_space_);
+ semi_space_collector_->SetToSpace(temp_space_);
+ collector = semi_space_collector_;
+ } else if (collector_type_ == kCollectorTypeCC) {
+ gc_type = concurrent_copying_collector_->GetGcType();
+ collector = concurrent_copying_collector_;
+ } else {
+ LOG(FATAL) << "Unreachable - invalid collector type " << static_cast<size_t>(collector_type_);
+ }
temp_space_->GetMemMap()->Protect(PROT_READ | PROT_WRITE);
- collector = semi_space_collector_;
+ CHECK(temp_space_->IsEmpty());
gc_type = collector::kGcTypeFull;
} else if (current_allocator_ == kAllocatorTypeRosAlloc ||
current_allocator_ == kAllocatorTypeDlMalloc) {
@@ -2821,5 +2850,19 @@
CHECK(remembered_sets_.find(space) == remembered_sets_.end());
}
+void Heap::ClearMarkedObjects() {
+ // Clear all of the spaces' mark bitmaps.
+ for (const auto& space : GetContinuousSpaces()) {
+ accounting::SpaceBitmap* mark_bitmap = space->GetMarkBitmap();
+ if (space->GetLiveBitmap() != mark_bitmap) {
+ mark_bitmap->Clear();
+ }
+ }
+ // Clear the marked objects in the discontinous space object sets.
+ for (const auto& space : GetDiscontinuousSpaces()) {
+ space->GetMarkObjects()->Clear();
+ }
+}
+
} // namespace gc
} // namespace art
diff --git a/runtime/gc/heap.h b/runtime/gc/heap.h
index a522750..5879757 100644
--- a/runtime/gc/heap.h
+++ b/runtime/gc/heap.h
@@ -60,6 +60,7 @@
} // namespace accounting
namespace collector {
+ class ConcurrentCopying;
class GarbageCollector;
class MarkSweep;
class SemiSpace;
@@ -254,6 +255,9 @@
void IncrementDisableMovingGC(Thread* self);
void DecrementDisableMovingGC(Thread* self);
+ // Clear all of the mark bits, doesn't clear bitmaps which have the same live bits as mark bits.
+ void ClearMarkedObjects() EXCLUSIVE_LOCKS_REQUIRED(Locks::heap_bitmap_lock_);
+
// Initiates an explicit garbage collection.
void CollectGarbage(bool clear_soft_references);
@@ -575,7 +579,8 @@
return AllocatorHasAllocationStack(allocator_type);
}
static bool IsCompactingGC(CollectorType collector_type) {
- return collector_type == kCollectorTypeSS || collector_type == kCollectorTypeGSS;
+ return collector_type == kCollectorTypeSS || collector_type == kCollectorTypeGSS ||
+ collector_type == kCollectorTypeCC;
}
bool ShouldAllocLargeObject(mirror::Class* c, size_t byte_count) const
SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
@@ -691,7 +696,7 @@
// What kind of concurrency behavior is the runtime after? Currently true for concurrent mark
// sweep GC, false for other GC types.
bool IsGcConcurrent() const ALWAYS_INLINE {
- return collector_type_ == kCollectorTypeCMS;
+ return collector_type_ == kCollectorTypeCMS || collector_type_ == kCollectorTypeCC;
}
// All-known continuous spaces, where objects lie within fixed bounds.
@@ -935,6 +940,7 @@
std::vector<collector::GarbageCollector*> garbage_collectors_;
collector::SemiSpace* semi_space_collector_;
+ collector::ConcurrentCopying* concurrent_copying_collector_;
const bool running_on_valgrind_;
const bool use_tlab_;
diff --git a/runtime/gc/space/rosalloc_space.cc b/runtime/gc/space/rosalloc_space.cc
index 3c65205..012267b 100644
--- a/runtime/gc/space/rosalloc_space.cc
+++ b/runtime/gc/space/rosalloc_space.cc
@@ -231,7 +231,7 @@
if (!rosalloc_->DoesReleaseAllPages()) {
VLOG(heap) << "RosAllocSpace::Trim() ";
size_t reclaimed = 0;
- InspectAllRosAlloc(DlmallocMadviseCallback, &reclaimed);
+ InspectAllRosAlloc(DlmallocMadviseCallback, &reclaimed, false);
return reclaimed;
}
return 0;
@@ -239,8 +239,7 @@
void RosAllocSpace::Walk(void(*callback)(void *start, void *end, size_t num_bytes, void* callback_arg),
void* arg) {
- InspectAllRosAlloc(callback, arg);
- callback(NULL, NULL, 0, arg); // Indicate end of a space.
+ InspectAllRosAlloc(callback, arg, true);
}
size_t RosAllocSpace::GetFootprint() {
@@ -268,35 +267,56 @@
uint64_t RosAllocSpace::GetBytesAllocated() {
size_t bytes_allocated = 0;
- InspectAllRosAlloc(art::gc::allocator::RosAlloc::BytesAllocatedCallback, &bytes_allocated);
+ InspectAllRosAlloc(art::gc::allocator::RosAlloc::BytesAllocatedCallback, &bytes_allocated, false);
return bytes_allocated;
}
uint64_t RosAllocSpace::GetObjectsAllocated() {
size_t objects_allocated = 0;
- InspectAllRosAlloc(art::gc::allocator::RosAlloc::ObjectsAllocatedCallback, &objects_allocated);
+ InspectAllRosAlloc(art::gc::allocator::RosAlloc::ObjectsAllocatedCallback, &objects_allocated, false);
return objects_allocated;
}
+void RosAllocSpace::InspectAllRosAllocWithSuspendAll(
+ void (*callback)(void *start, void *end, size_t num_bytes, void* callback_arg),
+ void* arg, bool do_null_callback_at_end) NO_THREAD_SAFETY_ANALYSIS {
+ // TODO: NO_THREAD_SAFETY_ANALYSIS.
+ Thread* self = Thread::Current();
+ ThreadList* tl = Runtime::Current()->GetThreadList();
+ tl->SuspendAll();
+ {
+ MutexLock mu(self, *Locks::runtime_shutdown_lock_);
+ MutexLock mu2(self, *Locks::thread_list_lock_);
+ rosalloc_->InspectAll(callback, arg);
+ if (do_null_callback_at_end) {
+ callback(NULL, NULL, 0, arg); // Indicate end of a space.
+ }
+ }
+ tl->ResumeAll();
+}
+
void RosAllocSpace::InspectAllRosAlloc(void (*callback)(void *start, void *end, size_t num_bytes, void* callback_arg),
- void* arg) NO_THREAD_SAFETY_ANALYSIS {
+ void* arg, bool do_null_callback_at_end) NO_THREAD_SAFETY_ANALYSIS {
// TODO: NO_THREAD_SAFETY_ANALYSIS.
Thread* self = Thread::Current();
if (Locks::mutator_lock_->IsExclusiveHeld(self)) {
// The mutators are already suspended. For example, a call path
// from SignalCatcher::HandleSigQuit().
rosalloc_->InspectAll(callback, arg);
- } else {
- // The mutators are not suspended yet.
- DCHECK(!Locks::mutator_lock_->IsSharedHeld(self));
- ThreadList* tl = Runtime::Current()->GetThreadList();
- tl->SuspendAll();
- {
- MutexLock mu(self, *Locks::runtime_shutdown_lock_);
- MutexLock mu2(self, *Locks::thread_list_lock_);
- rosalloc_->InspectAll(callback, arg);
+ if (do_null_callback_at_end) {
+ callback(NULL, NULL, 0, arg); // Indicate end of a space.
}
- tl->ResumeAll();
+ } else if (Locks::mutator_lock_->IsSharedHeld(self)) {
+ // The mutators are not suspended yet and we have a shared access
+ // to the mutator lock. Temporarily release the shared access by
+ // transitioning to the suspend state, and suspend the mutators.
+ self->TransitionFromRunnableToSuspended(kSuspended);
+ InspectAllRosAllocWithSuspendAll(callback, arg, do_null_callback_at_end);
+ self->TransitionFromSuspendedToRunnable();
+ Locks::mutator_lock_->AssertSharedHeld(self);
+ } else {
+ // The mutators are not suspended yet. Suspend the mutators.
+ InspectAllRosAllocWithSuspendAll(callback, arg, do_null_callback_at_end);
}
}
diff --git a/runtime/gc/space/rosalloc_space.h b/runtime/gc/space/rosalloc_space.h
index 949ec08..900e7a9 100644
--- a/runtime/gc/space/rosalloc_space.h
+++ b/runtime/gc/space/rosalloc_space.h
@@ -124,7 +124,11 @@
size_t maximum_size, bool low_memory_mode);
void InspectAllRosAlloc(void (*callback)(void *start, void *end, size_t num_bytes, void* callback_arg),
- void* arg)
+ void* arg, bool do_null_callback_at_end)
+ LOCKS_EXCLUDED(Locks::runtime_shutdown_lock_, Locks::thread_list_lock_);
+ void InspectAllRosAllocWithSuspendAll(
+ void (*callback)(void *start, void *end, size_t num_bytes, void* callback_arg),
+ void* arg, bool do_null_callback_at_end)
LOCKS_EXCLUDED(Locks::runtime_shutdown_lock_, Locks::thread_list_lock_);
// Underlying rosalloc.
diff --git a/runtime/native/java_lang_reflect_Constructor.cc b/runtime/native/java_lang_reflect_Constructor.cc
index b7e8ac2..2445b53 100644
--- a/runtime/native/java_lang_reflect_Constructor.cc
+++ b/runtime/native/java_lang_reflect_Constructor.cc
@@ -34,7 +34,8 @@
* check. We can also safely assume the constructor isn't associated
* with an interface, array, or primitive class.
*/
-static jobject Constructor_newInstance(JNIEnv* env, jobject javaMethod, jobjectArray javaArgs) {
+static jobject Constructor_newInstance(JNIEnv* env, jobject javaMethod, jobjectArray javaArgs,
+ jboolean accessible) {
ScopedFastNativeObjectAccess soa(env);
mirror::ArtMethod* m = mirror::ArtMethod::FromReflectedMethod(soa, javaMethod);
SirtRef<mirror::Class> c(soa.Self(), m->GetDeclaringClass());
@@ -67,14 +68,14 @@
}
jobject javaReceiver = soa.AddLocalReference<jobject>(receiver);
- InvokeMethod(soa, javaMethod, javaReceiver, javaArgs);
+ InvokeMethod(soa, javaMethod, javaReceiver, javaArgs, (accessible == JNI_TRUE));
// Constructors are ()V methods, so we shouldn't touch the result of InvokeMethod.
return javaReceiver;
}
static JNINativeMethod gMethods[] = {
- NATIVE_METHOD(Constructor, newInstance, "!([Ljava/lang/Object;)Ljava/lang/Object;"),
+ NATIVE_METHOD(Constructor, newInstance, "!([Ljava/lang/Object;Z)Ljava/lang/Object;"),
};
void register_java_lang_reflect_Constructor(JNIEnv* env) {
diff --git a/runtime/native/java_lang_reflect_Field.cc b/runtime/native/java_lang_reflect_Field.cc
index 48b58bf..ce622d9 100644
--- a/runtime/native/java_lang_reflect_Field.cc
+++ b/runtime/native/java_lang_reflect_Field.cc
@@ -20,6 +20,7 @@
#include "dex_file-inl.h"
#include "jni_internal.h"
#include "mirror/art_field-inl.h"
+#include "mirror/art_method-inl.h"
#include "mirror/class-inl.h"
#include "object_utils.h"
#include "reflection.h"
@@ -27,6 +28,21 @@
namespace art {
+static bool VerifyFieldAccess(mirror::ArtField* field, mirror::Object* obj, bool is_set)
+ SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
+ if (field->IsFinal() && is_set) {
+ ThrowIllegalAccessException(nullptr, StringPrintf("Cannot set final field: %s",
+ PrettyField(field).c_str()).c_str());
+ return false;
+ }
+ if (!VerifyAccess(obj, field->GetDeclaringClass(), field->GetAccessFlags())) {
+ ThrowIllegalAccessException(nullptr, StringPrintf("Cannot access field: %s",
+ PrettyField(field).c_str()).c_str());
+ return false;
+ }
+ return true;
+}
+
static bool GetFieldValue(const ScopedFastNativeObjectAccess& soa, mirror::Object* o,
mirror::ArtField* f, Primitive::Type field_type, bool allow_references,
JValue* value)
@@ -98,7 +114,7 @@
return true;
}
-static jobject Field_get(JNIEnv* env, jobject javaField, jobject javaObj) {
+static jobject Field_get(JNIEnv* env, jobject javaField, jobject javaObj, jboolean accessible) {
ScopedFastNativeObjectAccess soa(env);
CHECK(!kMovingFields) << "CheckReceiver may trigger thread suspension for initialization";
mirror::ArtField* f = mirror::ArtField::FromReflectedField(soa, javaField);
@@ -107,6 +123,11 @@
DCHECK(soa.Self()->IsExceptionPending());
return nullptr;
}
+ // If field is not set to be accessible, verify it can be accessed by the caller.
+ if ((accessible == JNI_FALSE) && !VerifyFieldAccess(f, o, false)) {
+ DCHECK(soa.Self()->IsExceptionPending());
+ return nullptr;
+ }
// We now don't expect suspension unless an exception is thrown.
// Get the field's value, boxing if necessary.
Primitive::Type field_type = FieldHelper(f).GetTypeAsPrimitiveType();
@@ -119,7 +140,7 @@
}
static JValue GetPrimitiveField(JNIEnv* env, jobject javaField, jobject javaObj,
- char dst_descriptor) {
+ char dst_descriptor, jboolean accessible) {
ScopedFastNativeObjectAccess soa(env);
CHECK(!kMovingFields) << "CheckReceiver may trigger thread suspension for initialization";
mirror::ArtField* f = mirror::ArtField::FromReflectedField(soa, javaField);
@@ -128,6 +149,13 @@
DCHECK(soa.Self()->IsExceptionPending());
return JValue();
}
+
+ // If field is not set to be accessible, verify it can be accessed by the caller.
+ if ((accessible == JNI_FALSE) && !VerifyFieldAccess(f, o, false)) {
+ DCHECK(soa.Self()->IsExceptionPending());
+ return JValue();
+ }
+
// We now don't expect suspension unless an exception is thrown.
// Read the value.
Primitive::Type field_type = FieldHelper(f).GetTypeAsPrimitiveType();
@@ -147,36 +175,38 @@
return wide_value;
}
-static jboolean Field_getBoolean(JNIEnv* env, jobject javaField, jobject javaObj) {
- return GetPrimitiveField(env, javaField, javaObj, 'Z').GetZ();
+static jboolean Field_getBoolean(JNIEnv* env, jobject javaField, jobject javaObj,
+ jboolean accessible) {
+ return GetPrimitiveField(env, javaField, javaObj, 'Z', accessible).GetZ();
}
-static jbyte Field_getByte(JNIEnv* env, jobject javaField, jobject javaObj) {
- return GetPrimitiveField(env, javaField, javaObj, 'B').GetB();
+static jbyte Field_getByte(JNIEnv* env, jobject javaField, jobject javaObj, jboolean accessible) {
+ return GetPrimitiveField(env, javaField, javaObj, 'B', accessible).GetB();
}
-static jchar Field_getChar(JNIEnv* env, jobject javaField, jobject javaObj) {
- return GetPrimitiveField(env, javaField, javaObj, 'C').GetC();
+static jchar Field_getChar(JNIEnv* env, jobject javaField, jobject javaObj, jboolean accessible) {
+ return GetPrimitiveField(env, javaField, javaObj, 'C', accessible).GetC();
}
-static jdouble Field_getDouble(JNIEnv* env, jobject javaField, jobject javaObj) {
- return GetPrimitiveField(env, javaField, javaObj, 'D').GetD();
+static jdouble Field_getDouble(JNIEnv* env, jobject javaField, jobject javaObj,
+ jboolean accessible) {
+ return GetPrimitiveField(env, javaField, javaObj, 'D', accessible).GetD();
}
-static jfloat Field_getFloat(JNIEnv* env, jobject javaField, jobject javaObj) {
- return GetPrimitiveField(env, javaField, javaObj, 'F').GetF();
+static jfloat Field_getFloat(JNIEnv* env, jobject javaField, jobject javaObj, jboolean accessible) {
+ return GetPrimitiveField(env, javaField, javaObj, 'F', accessible).GetF();
}
-static jint Field_getInt(JNIEnv* env, jobject javaField, jobject javaObj) {
- return GetPrimitiveField(env, javaField, javaObj, 'I').GetI();
+static jint Field_getInt(JNIEnv* env, jobject javaField, jobject javaObj, jboolean accessible) {
+ return GetPrimitiveField(env, javaField, javaObj, 'I', accessible).GetI();
}
-static jlong Field_getLong(JNIEnv* env, jobject javaField, jobject javaObj) {
- return GetPrimitiveField(env, javaField, javaObj, 'J').GetJ();
+static jlong Field_getLong(JNIEnv* env, jobject javaField, jobject javaObj, jboolean accessible) {
+ return GetPrimitiveField(env, javaField, javaObj, 'J', accessible).GetJ();
}
-static jshort Field_getShort(JNIEnv* env, jobject javaField, jobject javaObj) {
- return GetPrimitiveField(env, javaField, javaObj, 'S').GetS();
+static jshort Field_getShort(JNIEnv* env, jobject javaField, jobject javaObj, jboolean accessible) {
+ return GetPrimitiveField(env, javaField, javaObj, 'S', accessible).GetS();
}
static void SetFieldValue(ScopedFastNativeObjectAccess& soa, mirror::Object* o,
@@ -223,7 +253,8 @@
}
}
-static void Field_set(JNIEnv* env, jobject javaField, jobject javaObj, jobject javaValue) {
+static void Field_set(JNIEnv* env, jobject javaField, jobject javaObj, jobject javaValue,
+ jboolean accessible) {
ScopedFastNativeObjectAccess soa(env);
CHECK(!kMovingFields) << "CheckReceiver may trigger thread suspension for initialization";
mirror::ArtField* f = mirror::ArtField::FromReflectedField(soa, javaField);
@@ -260,11 +291,16 @@
DCHECK(soa.Self()->IsExceptionPending());
return;
}
+ // If field is not set to be accessible, verify it can be accessed by the caller.
+ if ((accessible == JNI_FALSE) && !VerifyFieldAccess(f, o, true)) {
+ DCHECK(soa.Self()->IsExceptionPending());
+ return;
+ }
SetFieldValue(soa, o, f, field_prim_type, true, unboxed_value);
}
static void SetPrimitiveField(JNIEnv* env, jobject javaField, jobject javaObj, char src_descriptor,
- const JValue& new_value) {
+ const JValue& new_value, jboolean accessible) {
ScopedFastNativeObjectAccess soa(env);
mirror::ArtField* f = mirror::ArtField::FromReflectedField(soa, javaField);
mirror::Object* o = nullptr;
@@ -286,77 +322,91 @@
return;
}
+ // If field is not set to be accessible, verify it can be accessed by the caller.
+ if ((accessible == JNI_FALSE) && !VerifyFieldAccess(f, o, true)) {
+ DCHECK(soa.Self()->IsExceptionPending());
+ return;
+ }
+
// Write the value.
SetFieldValue(soa, o, f, field_type, false, wide_value);
}
-static void Field_setBoolean(JNIEnv* env, jobject javaField, jobject javaObj, jboolean z) {
+static void Field_setBoolean(JNIEnv* env, jobject javaField, jobject javaObj, jboolean z,
+ jboolean accessible) {
JValue value;
value.SetZ(z);
- SetPrimitiveField(env, javaField, javaObj, 'Z', value);
+ SetPrimitiveField(env, javaField, javaObj, 'Z', value, accessible);
}
-static void Field_setByte(JNIEnv* env, jobject javaField, jobject javaObj, jbyte b) {
+static void Field_setByte(JNIEnv* env, jobject javaField, jobject javaObj, jbyte b,
+ jboolean accessible) {
JValue value;
value.SetB(b);
- SetPrimitiveField(env, javaField, javaObj, 'B', value);
+ SetPrimitiveField(env, javaField, javaObj, 'B', value, accessible);
}
-static void Field_setChar(JNIEnv* env, jobject javaField, jobject javaObj, jchar c) {
+static void Field_setChar(JNIEnv* env, jobject javaField, jobject javaObj, jchar c,
+ jboolean accessible) {
JValue value;
value.SetC(c);
- SetPrimitiveField(env, javaField, javaObj, 'C', value);
+ SetPrimitiveField(env, javaField, javaObj, 'C', value, accessible);
}
-static void Field_setDouble(JNIEnv* env, jobject javaField, jobject javaObj, jdouble d) {
+static void Field_setDouble(JNIEnv* env, jobject javaField, jobject javaObj, jdouble d,
+ jboolean accessible) {
JValue value;
value.SetD(d);
- SetPrimitiveField(env, javaField, javaObj, 'D', value);
+ SetPrimitiveField(env, javaField, javaObj, 'D', value, accessible);
}
-static void Field_setFloat(JNIEnv* env, jobject javaField, jobject javaObj, jfloat f) {
+static void Field_setFloat(JNIEnv* env, jobject javaField, jobject javaObj, jfloat f,
+ jboolean accessible) {
JValue value;
value.SetF(f);
- SetPrimitiveField(env, javaField, javaObj, 'F', value);
+ SetPrimitiveField(env, javaField, javaObj, 'F', value, accessible);
}
-static void Field_setInt(JNIEnv* env, jobject javaField, jobject javaObj, jint i) {
+static void Field_setInt(JNIEnv* env, jobject javaField, jobject javaObj, jint i,
+ jboolean accessible) {
JValue value;
value.SetI(i);
- SetPrimitiveField(env, javaField, javaObj, 'I', value);
+ SetPrimitiveField(env, javaField, javaObj, 'I', value, accessible);
}
-static void Field_setLong(JNIEnv* env, jobject javaField, jobject javaObj, jlong j) {
+static void Field_setLong(JNIEnv* env, jobject javaField, jobject javaObj, jlong j,
+ jboolean accessible) {
JValue value;
value.SetJ(j);
- SetPrimitiveField(env, javaField, javaObj, 'J', value);
+ SetPrimitiveField(env, javaField, javaObj, 'J', value, accessible);
}
-static void Field_setShort(JNIEnv* env, jobject javaField, jobject javaObj, jshort s) {
+static void Field_setShort(JNIEnv* env, jobject javaField, jobject javaObj, jshort s,
+ jboolean accessible) {
JValue value;
value.SetS(s);
- SetPrimitiveField(env, javaField, javaObj, 'S', value);
+ SetPrimitiveField(env, javaField, javaObj, 'S', value, accessible);
}
static JNINativeMethod gMethods[] = {
- NATIVE_METHOD(Field, get, "!(Ljava/lang/Object;)Ljava/lang/Object;"),
- NATIVE_METHOD(Field, getBoolean, "!(Ljava/lang/Object;)Z"),
- NATIVE_METHOD(Field, getByte, "!(Ljava/lang/Object;)B"),
- NATIVE_METHOD(Field, getChar, "!(Ljava/lang/Object;)C"),
- NATIVE_METHOD(Field, getDouble, "!(Ljava/lang/Object;)D"),
- NATIVE_METHOD(Field, getFloat, "!(Ljava/lang/Object;)F"),
- NATIVE_METHOD(Field, getInt, "!(Ljava/lang/Object;)I"),
- NATIVE_METHOD(Field, getLong, "!(Ljava/lang/Object;)J"),
- NATIVE_METHOD(Field, getShort, "!(Ljava/lang/Object;)S"),
- NATIVE_METHOD(Field, set, "!(Ljava/lang/Object;Ljava/lang/Object;)V"),
- NATIVE_METHOD(Field, setBoolean, "!(Ljava/lang/Object;Z)V"),
- NATIVE_METHOD(Field, setByte, "!(Ljava/lang/Object;B)V"),
- NATIVE_METHOD(Field, setChar, "!(Ljava/lang/Object;C)V"),
- NATIVE_METHOD(Field, setDouble, "!(Ljava/lang/Object;D)V"),
- NATIVE_METHOD(Field, setFloat, "!(Ljava/lang/Object;F)V"),
- NATIVE_METHOD(Field, setInt, "!(Ljava/lang/Object;I)V"),
- NATIVE_METHOD(Field, setLong, "!(Ljava/lang/Object;J)V"),
- NATIVE_METHOD(Field, setShort, "!(Ljava/lang/Object;S)V"),
+ NATIVE_METHOD(Field, get, "!(Ljava/lang/Object;Z)Ljava/lang/Object;"),
+ NATIVE_METHOD(Field, getBoolean, "!(Ljava/lang/Object;Z)Z"),
+ NATIVE_METHOD(Field, getByte, "!(Ljava/lang/Object;Z)B"),
+ NATIVE_METHOD(Field, getChar, "!(Ljava/lang/Object;Z)C"),
+ NATIVE_METHOD(Field, getDouble, "!(Ljava/lang/Object;Z)D"),
+ NATIVE_METHOD(Field, getFloat, "!(Ljava/lang/Object;Z)F"),
+ NATIVE_METHOD(Field, getInt, "!(Ljava/lang/Object;Z)I"),
+ NATIVE_METHOD(Field, getLong, "!(Ljava/lang/Object;Z)J"),
+ NATIVE_METHOD(Field, getShort, "!(Ljava/lang/Object;Z)S"),
+ NATIVE_METHOD(Field, set, "!(Ljava/lang/Object;Ljava/lang/Object;Z)V"),
+ NATIVE_METHOD(Field, setBoolean, "!(Ljava/lang/Object;ZZ)V"),
+ NATIVE_METHOD(Field, setByte, "!(Ljava/lang/Object;BZ)V"),
+ NATIVE_METHOD(Field, setChar, "!(Ljava/lang/Object;CZ)V"),
+ NATIVE_METHOD(Field, setDouble, "!(Ljava/lang/Object;DZ)V"),
+ NATIVE_METHOD(Field, setFloat, "!(Ljava/lang/Object;FZ)V"),
+ NATIVE_METHOD(Field, setInt, "!(Ljava/lang/Object;IZ)V"),
+ NATIVE_METHOD(Field, setLong, "!(Ljava/lang/Object;JZ)V"),
+ NATIVE_METHOD(Field, setShort, "!(Ljava/lang/Object;SZ)V"),
};
void register_java_lang_reflect_Field(JNIEnv* env) {
diff --git a/runtime/native/java_lang_reflect_Method.cc b/runtime/native/java_lang_reflect_Method.cc
index abb73b6..22e81e4 100644
--- a/runtime/native/java_lang_reflect_Method.cc
+++ b/runtime/native/java_lang_reflect_Method.cc
@@ -29,10 +29,10 @@
namespace art {
-static jobject Method_invoke(JNIEnv* env,
- jobject javaMethod, jobject javaReceiver, jobject javaArgs) {
+static jobject Method_invoke(JNIEnv* env, jobject javaMethod, jobject javaReceiver,
+ jobject javaArgs, jboolean accessible) {
ScopedFastNativeObjectAccess soa(env);
- return InvokeMethod(soa, javaMethod, javaReceiver, javaArgs);
+ return InvokeMethod(soa, javaMethod, javaReceiver, javaArgs, (accessible == JNI_TRUE));
}
static jobject Method_getExceptionTypesNative(JNIEnv* env, jobject javaMethod) {
@@ -56,7 +56,7 @@
}
static JNINativeMethod gMethods[] = {
- NATIVE_METHOD(Method, invoke, "!(Ljava/lang/Object;[Ljava/lang/Object;)Ljava/lang/Object;"),
+ NATIVE_METHOD(Method, invoke, "!(Ljava/lang/Object;[Ljava/lang/Object;Z)Ljava/lang/Object;"),
NATIVE_METHOD(Method, getExceptionTypesNative, "!()[Ljava/lang/Class;"),
};
diff --git a/runtime/parsed_options.cc b/runtime/parsed_options.cc
index 9b1c013..e2086f1 100644
--- a/runtime/parsed_options.cc
+++ b/runtime/parsed_options.cc
@@ -107,6 +107,8 @@
return gc::kCollectorTypeSS;
} else if (option == "GSS") {
return gc::kCollectorTypeGSS;
+ } else if (option == "CC") {
+ return gc::kCollectorTypeCC;
} else {
return gc::kCollectorTypeNone;
}
diff --git a/runtime/reflection.cc b/runtime/reflection.cc
index 7f39e70..6ed61f6 100644
--- a/runtime/reflection.cc
+++ b/runtime/reflection.cc
@@ -26,6 +26,7 @@
#include "mirror/class-inl.h"
#include "mirror/object_array.h"
#include "mirror/object_array-inl.h"
+#include "nth_caller_visitor.h"
#include "object_utils.h"
#include "scoped_thread_state_change.h"
#include "stack.h"
@@ -461,7 +462,7 @@
}
jobject InvokeMethod(const ScopedObjectAccess& soa, jobject javaMethod,
- jobject javaReceiver, jobject javaArgs) {
+ jobject javaReceiver, jobject javaArgs, bool accessible) {
mirror::ArtMethod* m = mirror::ArtMethod::FromReflectedMethod(soa, javaMethod);
mirror::Class* declaring_class = m->GetDeclaringClass();
@@ -499,6 +500,13 @@
return NULL;
}
+ // If method is not set to be accessible, verify it can be accessed by the caller.
+ if (!accessible && !VerifyAccess(receiver, declaring_class, m->GetAccessFlags())) {
+ ThrowIllegalAccessException(nullptr, StringPrintf("Cannot access method: %s",
+ PrettyMethod(m).c_str()).c_str());
+ return nullptr;
+ }
+
// Invoke the method.
JValue result;
ArgArray arg_array(mh.GetShorty(), mh.GetShortyLength());
@@ -786,4 +794,30 @@
return UnboxPrimitive(&throw_location, o, dst_class, nullptr, unboxed_value);
}
+bool VerifyAccess(mirror::Object* obj, mirror::Class* declaring_class, uint32_t access_flags) {
+ NthCallerVisitor visitor(Thread::Current(), 2);
+ visitor.WalkStack();
+ mirror::Class* caller_class = visitor.caller->GetDeclaringClass();
+
+ if ((((access_flags & kAccPublic) != 0) && declaring_class->IsPublic()) ||
+ caller_class == declaring_class) {
+ return true;
+ }
+ if ((access_flags & kAccPrivate) != 0) {
+ return false;
+ }
+ if ((access_flags & kAccProtected) != 0) {
+ if (obj != nullptr && !obj->InstanceOf(caller_class) &&
+ !declaring_class->IsInSamePackage(caller_class)) {
+ return false;
+ } else if (declaring_class->IsAssignableFrom(caller_class)) {
+ return true;
+ }
+ }
+ if (!declaring_class->IsInSamePackage(caller_class)) {
+ return false;
+ }
+ return true;
+}
+
} // namespace art
diff --git a/runtime/reflection.h b/runtime/reflection.h
index 325998f..d9a7228 100644
--- a/runtime/reflection.h
+++ b/runtime/reflection.h
@@ -68,12 +68,15 @@
SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
jobject InvokeMethod(const ScopedObjectAccess& soa, jobject method, jobject receiver,
- jobject args)
+ jobject args, bool accessible)
SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
bool VerifyObjectIsClass(mirror::Object* o, mirror::Class* c)
SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+bool VerifyAccess(mirror::Object* obj, mirror::Class* declaring_class, uint32_t access_flags)
+ SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+
} // namespace art
#endif // ART_RUNTIME_REFLECTION_H_
diff --git a/runtime/stack.cc b/runtime/stack.cc
index c33d1ab..ab3bd85 100644
--- a/runtime/stack.cc
+++ b/runtime/stack.cc
@@ -16,6 +16,7 @@
#include "stack.h"
+#include "base/hex_dump.h"
#include "mirror/art_method-inl.h"
#include "mirror/class-inl.h"
#include "mirror/object.h"
@@ -23,6 +24,7 @@
#include "mirror/object_array-inl.h"
#include "object_utils.h"
#include "runtime.h"
+#include "thread.h"
#include "thread_list.h"
#include "throw_location.h"
#include "verify_object-inl.h"
@@ -30,6 +32,14 @@
namespace art {
+// Define a piece of memory, the address of which can be used as a marker
+// for the gap in the stack added during stack overflow handling.
+static uint32_t stack_overflow_object;
+
+// The stack overflow gap marker is simply a valid unique address.
+void* stack_overflow_gap_marker = &stack_overflow_object;
+
+
mirror::Object* ShadowFrame::GetThisObject() const {
mirror::ArtMethod* m = GetMethod();
if (m->IsStatic()) {
@@ -294,20 +304,56 @@
CHECK_EQ(cur_depth_, 0U);
bool exit_stubs_installed = Runtime::Current()->GetInstrumentation()->AreExitStubsInstalled();
uint32_t instrumentation_stack_depth = 0;
+
+ bool kDebugStackWalk = false;
+ bool kDebugStackWalkVeryVerbose = false; // The name says it all.
+
+ if (kDebugStackWalk) {
+ LOG(INFO) << "walking stack";
+ }
for (const ManagedStack* current_fragment = thread_->GetManagedStack(); current_fragment != NULL;
current_fragment = current_fragment->GetLink()) {
cur_shadow_frame_ = current_fragment->GetTopShadowFrame();
cur_quick_frame_ = current_fragment->GetTopQuickFrame();
cur_quick_frame_pc_ = current_fragment->GetTopQuickFramePc();
+ if (kDebugStackWalkVeryVerbose) {
+ LOG(INFO) << "cur_quick_frame: " << cur_quick_frame_;
+ LOG(INFO) << "cur_quick_frame_pc: " << std::hex << cur_quick_frame_pc_;
+ }
+
if (cur_quick_frame_ != NULL) { // Handle quick stack frames.
// Can't be both a shadow and a quick fragment.
DCHECK(current_fragment->GetTopShadowFrame() == NULL);
mirror::ArtMethod* method = *cur_quick_frame_;
while (method != NULL) {
- SanityCheckFrame();
- bool should_continue = VisitFrame();
- if (UNLIKELY(!should_continue)) {
- return;
+ // Check for a stack overflow gap marker.
+ if (method == reinterpret_cast<mirror::ArtMethod*>(stack_overflow_gap_marker)) {
+ // Marker for a stack overflow. This is followed by the offset from the
+ // current SP to the next frame. There is a gap in the stack here. Jump
+ // the gap silently.
+ // Caveat coder: the layout of the overflow marker depends on the architecture.
+ // The first element is address sized (8 bytes on a 64 bit machine). The second
+ // element is 32 bits. So be careful with those address calculations.
+
+ // Get the address of the offset, just beyond the marker pointer.
+ byte* gapsizeaddr = reinterpret_cast<byte*>(cur_quick_frame_) + sizeof(uintptr_t);
+ uint32_t gap = *reinterpret_cast<uint32_t*>(gapsizeaddr);
+ CHECK_GT(gap, Thread::kStackOverflowProtectedSize);
+ mirror::ArtMethod** next_frame = reinterpret_cast<mirror::ArtMethod**>(
+ reinterpret_cast<byte*>(gapsizeaddr) + gap);
+ if (kDebugStackWalk) {
+ LOG(INFO) << "stack overflow marker hit, gap: " << gap << ", next_frame: " <<
+ next_frame;
+ }
+ cur_quick_frame_ = next_frame;
+ method = *next_frame;
+ CHECK(method != nullptr);
+ } else {
+ SanityCheckFrame();
+ bool should_continue = VisitFrame();
+ if (UNLIKELY(!should_continue)) {
+ return;
+ }
}
if (context_ != NULL) {
context_->FillCalleeSaves(*this);
@@ -317,6 +363,9 @@
size_t return_pc_offset = method->GetReturnPcOffsetInBytes();
byte* return_pc_addr = reinterpret_cast<byte*>(cur_quick_frame_) + return_pc_offset;
uintptr_t return_pc = *reinterpret_cast<uintptr_t*>(return_pc_addr);
+ if (kDebugStackWalkVeryVerbose) {
+ LOG(INFO) << "frame size: " << frame_size << ", return_pc: " << std::hex << return_pc;
+ }
if (UNLIKELY(exit_stubs_installed)) {
// While profiling, the return pc is restored from the side stack, except when walking
// the stack for an exception where the side stack will be unwound in VisitFrame.
@@ -349,6 +398,10 @@
cur_quick_frame_ = reinterpret_cast<mirror::ArtMethod**>(next_frame);
cur_depth_++;
method = *cur_quick_frame_;
+ if (kDebugStackWalkVeryVerbose) {
+ LOG(INFO) << "new cur_quick_frame_: " << cur_quick_frame_;
+ LOG(INFO) << "new cur_quick_frame_pc_: " << std::hex << cur_quick_frame_pc_;
+ }
}
} else if (cur_shadow_frame_ != NULL) {
do {
diff --git a/runtime/stack.h b/runtime/stack.h
index 4ee5de1..ab903d6 100644
--- a/runtime/stack.h
+++ b/runtime/stack.h
@@ -100,6 +100,14 @@
kVRegNonSpecialTempBaseReg = -3,
};
+// Special object used to mark the gap in the stack placed when a stack
+// overflow fault occurs during implicit stack checking. This is not
+// a real object - it is used simply as a valid address to which a
+// mirror::ArtMethod* can be compared during a stack walk. It is inserted
+// into the stack during the stack overflow signal handling to mark the gap
+// in which the memory is protected against read and write.
+extern void* stack_overflow_gap_marker;
+
// A reference from the shadow stack to a MirrorType object within the Java heap.
template<class MirrorType>
class MANAGED StackReference : public mirror::ObjectReference<false, MirrorType> {
diff --git a/runtime/thread.cc b/runtime/thread.cc
index 38e4204..3692b9f 100644
--- a/runtime/thread.cc
+++ b/runtime/thread.cc
@@ -215,10 +215,16 @@
stack_size = PTHREAD_STACK_MIN;
}
- // It's likely that callers are trying to ensure they have at least a certain amount of
- // stack space, so we should add our reserved space on top of what they requested, rather
- // than implicitly take it away from them.
- stack_size += Thread::kStackOverflowReservedBytes;
+ if (Runtime::Current()->ExplicitStackOverflowChecks()) {
+ // It's likely that callers are trying to ensure they have at least a certain amount of
+ // stack space, so we should add our reserved space on top of what they requested, rather
+ // than implicitly take it away from them.
+ stack_size += Thread::kStackOverflowReservedBytes;
+ } else {
+ // If we are going to use implicit stack checks, allocate space for the protected
+ // region at the bottom of the stack.
+ stack_size += Thread::kStackOverflowImplicitCheckSize;
+ }
// Some systems require the stack size to be a multiple of the system page size, so round up.
stack_size = RoundUp(stack_size, kPageSize);
@@ -226,6 +232,39 @@
return stack_size;
}
+// Install a protected region in the stack. This is used to trigger a SIGSEGV if a stack
+// overflow is detected. It is located right below the stack_end_. Just below that
+// is the StackOverflow reserved region used when creating the StackOverflow
+// exception.
+void Thread::InstallImplicitProtection(bool is_main_stack) {
+ byte* pregion = stack_end_;
+
+ constexpr uint32_t kMarker = 0xdadadada;
+ uintptr_t *marker = reinterpret_cast<uintptr_t*>(pregion);
+ if (*marker == kMarker) {
+ // The region has already been set up.
+ return;
+ }
+ // Add marker so that we can detect a second attempt to do this.
+ *marker = kMarker;
+
+ pregion -= kStackOverflowProtectedSize;
+
+ // Touch the pages in the region to map them in. Otherwise mprotect fails. Only
+ // need to do this on the main stack.
+ if (is_main_stack) {
+ memset(pregion, 0x55, kStackOverflowProtectedSize);
+ }
+ VLOG(threads) << "installing stack protected region at " << std::hex <<
+ static_cast<void*>(pregion) << " to " <<
+ static_cast<void*>(pregion + kStackOverflowProtectedSize - 1);
+
+ if (mprotect(pregion, kStackOverflowProtectedSize, PROT_NONE) == -1) {
+ LOG(FATAL) << "Unable to create protected region in stack for implicit overflow check. Reason:"
+ << strerror(errno);
+ }
+}
+
void Thread::CreateNativeThread(JNIEnv* env, jobject java_peer, size_t stack_size, bool is_daemon) {
CHECK(java_peer != nullptr);
Thread* self = static_cast<JNIEnvExt*>(env)->self;
@@ -472,7 +511,22 @@
#endif
// Set stack_end_ to the bottom of the stack saving space of stack overflows
- ResetDefaultStackEnd();
+ bool implicit_stack_check = !Runtime::Current()->ExplicitStackOverflowChecks();
+ ResetDefaultStackEnd(implicit_stack_check);
+
+ // Install the protected region if we are doing implicit overflow checks.
+ if (implicit_stack_check) {
+ if (is_main_thread) {
+ // The main thread has a 16K protected region at the bottom. We need
+ // to install our own region so we need to move the limits
+ // of the stack to make room for it.
+ constexpr uint32_t kDelta = 16 * KB;
+ stack_begin_ += kDelta;
+ stack_end_ += kDelta;
+ stack_size_ -= kDelta;
+ }
+ InstallImplicitProtection(is_main_thread);
+ }
// Sanity check.
int stack_variable;
@@ -967,6 +1021,7 @@
pthread_self_(0),
no_thread_suspension_(0),
last_no_thread_suspension_cause_(nullptr),
+ suspend_trigger_(reinterpret_cast<uintptr_t*>(&suspend_trigger_)),
thread_exit_check_count_(0),
thread_local_start_(nullptr),
thread_local_pos_(nullptr),
diff --git a/runtime/thread.h b/runtime/thread.h
index 32875e6..63d22c5 100644
--- a/runtime/thread.h
+++ b/runtime/thread.h
@@ -112,6 +112,14 @@
static constexpr size_t kStackOverflowReservedUsableBytes =
kStackOverflowReservedBytes - kStackOverflowSignalReservedBytes;
+ // For implicit overflow checks we reserve an extra piece of memory at the bottom
+ // of the stack (lowest memory). The higher portion of the memory
+ // is protected against reads and the lower is available for use while
+ // throwing the StackOverflow exception.
+ static constexpr size_t kStackOverflowProtectedSize = 32 * KB;
+ static constexpr size_t kStackOverflowImplicitCheckSize = kStackOverflowProtectedSize +
+ kStackOverflowReservedBytes;
+
// Creates a new native thread corresponding to the given managed peer.
// Used to implement Thread.start.
static void CreateNativeThread(JNIEnv* env, jobject peer, size_t stack_size, bool daemon);
@@ -461,12 +469,21 @@
void SetStackEndForStackOverflow() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
// Set the stack end to that to be used during regular execution
- void ResetDefaultStackEnd() {
+ void ResetDefaultStackEnd(bool implicit_overflow_check) {
// Our stacks grow down, so we want stack_end_ to be near there, but reserving enough room
// to throw a StackOverflowError.
- stack_end_ = stack_begin_ + kStackOverflowReservedBytes;
+ if (implicit_overflow_check) {
+ // For implicit checks we also need to add in the protected region above the
+ // overflow region.
+ stack_end_ = stack_begin_ + kStackOverflowImplicitCheckSize;
+ } else {
+ stack_end_ = stack_begin_ + kStackOverflowReservedBytes;
+ }
}
+ // Install the protected region for implicit stack checks.
+ void InstallImplicitProtection(bool is_main_stack);
+
bool IsHandlingStackOverflow() const {
return stack_end_ == stack_begin_;
}
diff --git a/test/046-reflect/expected.txt b/test/046-reflect/expected.txt
index fcfaf2f..55b0dbe 100644
--- a/test/046-reflect/expected.txt
+++ b/test/046-reflect/expected.txt
@@ -81,7 +81,8 @@
Field type is int
Access flags are 0x11
cantTouchThis is 77
- cantTouchThis is now 99
+ as expected: set-final throws exception
+ cantTouchThis is still 77
public final int Target.cantTouchThis accessible=false
public final int Target.cantTouchThis accessible=true
cantTouchThis is now 87
diff --git a/test/046-reflect/src/Main.java b/test/046-reflect/src/Main.java
index dfb0d8f..d60fcb4 100644
--- a/test/046-reflect/src/Main.java
+++ b/test/046-reflect/src/Main.java
@@ -335,11 +335,12 @@
System.out.println(" cantTouchThis is " + intVal);
try {
field.setInt(instance, 99);
+ System.out.println("ERROR: set-final did not throw exception");
} catch (IllegalAccessException iae) {
- System.out.println("ERROR: set-final failed");
+ System.out.println(" as expected: set-final throws exception");
}
intVal = field.getInt(instance);
- System.out.println(" cantTouchThis is now " + intVal);
+ System.out.println(" cantTouchThis is still " + intVal);
System.out.println(" " + field + " accessible=" + field.isAccessible());
field.setAccessible(true);
diff --git a/test/064-field-access/src/Main.java b/test/064-field-access/src/Main.java
index 3b9a475..c9b93ba 100644
--- a/test/064-field-access/src/Main.java
+++ b/test/064-field-access/src/Main.java
@@ -16,6 +16,7 @@
import other.PublicClass;
import java.lang.reflect.Field;
+import java.lang.reflect.Method;
/*
* Test field access through reflection.
@@ -192,6 +193,11 @@
/* package */ static float samePackagePackageFloatStaticField = 63.0f;
/* package */ static double samePackagePackageDoubleStaticField = 64.0;
/* package */ static Object samePackagePackageObjectStaticField = "65";
+
+ public void samePublicMethod() { }
+ protected void sameProtectedMethod() { }
+ private void samePrivateMethod() { }
+ /* package */ void samePackageMethod() { }
}
/*
@@ -510,23 +516,32 @@
for (int round = 0; round < 3; round++) {
Object validInst;
Field[] fields;
+ Method[] methods;
boolean same_package = false;
+ boolean protected_class = false;
switch (round) {
case 0:
validInst = new SamePackage();
fields = SamePackage.class.getDeclaredFields();
check(fields.length == 72);
+ methods = SamePackage.class.getDeclaredMethods();
+ check(methods.length == 4);
same_package = true;
break;
case 1:
validInst = new PublicClass();
fields = PublicClass.class.getDeclaredFields();
check(fields.length == 72);
+ methods = PublicClass.class.getDeclaredMethods();
+ check(methods.length == 4);
break;
default:
validInst = new PublicClass();
fields = PublicClass.class.getSuperclass().getDeclaredFields();
check(fields.length == 72);
+ methods = PublicClass.class.getSuperclass().getDeclaredMethods();
+ check(methods.length == 4);
+ protected_class = true;
break;
}
for (Field f : fields) {
@@ -540,16 +555,15 @@
// Check access or lack of to field.
Class<?> subClassAccessExceptionClass = null;
if (f.getName().contains("Private") ||
- (!same_package && f.getName().contains("Package"))) {
- // ART deliberately doesn't throw IllegalAccessException.
- // subClassAccessExceptionClass = IllegalAccessException.class;
+ (!same_package && f.getName().contains("Package")) ||
+ (!same_package && f.getName().contains("Protected"))) {
+ subClassAccessExceptionClass = IllegalAccessException.class;
}
Class<?> mainClassAccessExceptionClass = null;
if (f.getName().contains("Private") ||
(!same_package && f.getName().contains("Package")) ||
(!same_package && f.getName().contains("Protected"))) {
- // ART deliberately doesn't throw IllegalAccessException.
- // mainClassAccessExceptionClass = IllegalAccessException.class;
+ mainClassAccessExceptionClass = IllegalAccessException.class;
}
this.getValue(f, validInst, typeChar, subClassAccessExceptionClass);
@@ -588,6 +602,16 @@
}
}
}
+
+ for (Method m : methods) {
+ Class<?> subClassAccessExceptionClass = null;
+ if (protected_class || m.getName().contains("Private") ||
+ (!same_package && m.getName().contains("Package")) ||
+ (!same_package && m.getName().contains("Protected"))) {
+ subClassAccessExceptionClass = IllegalAccessException.class;
+ }
+ this.invoke(m, validInst, subClassAccessExceptionClass);
+ }
}
System.out.println("good");
}
@@ -598,7 +622,6 @@
*/
public Object getValue(Field field, Object obj, char type,
Class expectedException) {
-
Object result = null;
try {
switch (type) {
@@ -657,4 +680,29 @@
return result;
}
+
+ public Object invoke(Method method, Object obj, Class expectedException) {
+ Object result = null;
+ try {
+ result = method.invoke(obj);
+ /* success; expected? */
+ if (expectedException != null) {
+ System.err.println("ERROR: call succeeded for method " + method + "', was expecting " +
+ expectedException);
+ Thread.dumpStack();
+ }
+ } catch (Exception ex) {
+ if (expectedException == null) {
+ System.err.println("ERROR: call failed unexpectedly: " + ex.getClass());
+ ex.printStackTrace();
+ } else {
+ if (!expectedException.equals(ex.getClass())) {
+ System.err.println("ERROR: incorrect exception: wanted " + expectedException.getName() +
+ ", got " + ex.getClass());
+ ex.printStackTrace();
+ }
+ }
+ }
+ return result;
+ }
}
diff --git a/test/064-field-access/src/other/ProtectedClass.java b/test/064-field-access/src/other/ProtectedClass.java
index 779aa1d..756c97f 100644
--- a/test/064-field-access/src/other/ProtectedClass.java
+++ b/test/064-field-access/src/other/ProtectedClass.java
@@ -97,4 +97,10 @@
/* package */ static float otherProtectedClassPackageFloatStaticField = 63.0f;
/* package */ static double otherProtectedClassPackageDoubleStaticField = 64.0;
/* package */ static Object otherProtectedClassPackageObjectStaticField = "65";
+
+ public void otherPublicMethod() { }
+ protected void otherProtectedMethod() { }
+ private void otherPrivateMethod() { }
+ /* package */ void otherPackageMethod() { }
+
}
diff --git a/test/064-field-access/src/other/PublicClass.java b/test/064-field-access/src/other/PublicClass.java
index 1653973..dbcb4fd 100644
--- a/test/064-field-access/src/other/PublicClass.java
+++ b/test/064-field-access/src/other/PublicClass.java
@@ -97,4 +97,9 @@
/* package */ static float otherPublicClassPackageFloatStaticField = -63.0f;
/* package */ static double otherPublicClassPackageDoubleStaticField = -64.0;
/* package */ static Object otherPublicClassPackageObjectStaticField = "-65";
+
+ public void otherPublicMethod() { }
+ protected void otherProtectedMethod() { }
+ private void otherPrivateMethod() { }
+ /* package */ void otherPackageMethod() { }
}
diff --git a/test/100-reflect2/expected.txt b/test/100-reflect2/expected.txt
index 3d87ebc..bed0689 100644
--- a/test/100-reflect2/expected.txt
+++ b/test/100-reflect2/expected.txt
@@ -22,10 +22,7 @@
30
62
14
-java.lang.IllegalArgumentException: Invalid primitive conversion from int to short
- at java.lang.reflect.Field.set(Native Method)
- at Main.testFieldReflection(Main.java:121)
- at Main.main(Main.java:269)
+got expected IllegalArgumentException
true (class java.lang.Boolean)
6 (class java.lang.Byte)
z (class java.lang.Character)
@@ -66,12 +63,6 @@
true 0 1 2.0 hello world 3.0 4 5 6
null (null)
[]
-java.lang.reflect.InvocationTargetException
- at java.lang.reflect.Method.invoke(Native Method)
- at Main.testMethodReflection(Main.java:210)
- at Main.main(Main.java:270)
-Caused by: java.lang.ArithmeticException: surprise!
- at Main.thrower(Main.java:218)
- ... 3 more
+got expected InvocationTargetException
(class java.lang.String)
yz (class java.lang.String)
diff --git a/test/100-reflect2/src/Main.java b/test/100-reflect2/src/Main.java
index 0404591..0cc1488 100644
--- a/test/100-reflect2/src/Main.java
+++ b/test/100-reflect2/src/Main.java
@@ -119,8 +119,9 @@
try {
f = Main.class.getDeclaredField("s");
f.set(null, Integer.valueOf(14));
+ System.out.println("************* should have thrown!");
} catch (IllegalArgumentException expected) {
- expected.printStackTrace();
+ System.out.println("got expected IllegalArgumentException");
}
f = Main.class.getDeclaredField("z");
@@ -209,8 +210,8 @@
System.out.println(Arrays.toString(m.getParameterTypes()));
show(m.invoke(null));
System.out.println("************* should have thrown!");
- } catch (Exception expected) {
- expected.printStackTrace();
+ } catch (InvocationTargetException expected) {
+ System.out.println("got expected InvocationTargetException");
}
}
diff --git a/test/401-optimizing-compiler/expected.txt b/test/401-optimizing-compiler/expected.txt
new file mode 100644
index 0000000..7b3a018
--- /dev/null
+++ b/test/401-optimizing-compiler/expected.txt
@@ -0,0 +1,3 @@
+In static method
+Forced GC
+java.lang.Error: Error
diff --git a/test/401-optimizing-compiler/info.txt b/test/401-optimizing-compiler/info.txt
new file mode 100644
index 0000000..2a89eb5
--- /dev/null
+++ b/test/401-optimizing-compiler/info.txt
@@ -0,0 +1,2 @@
+A set of tests for the optimizing compiler. They will incrementally cover what the
+optimizing compiler covers.
diff --git a/test/401-optimizing-compiler/src/Main.java b/test/401-optimizing-compiler/src/Main.java
new file mode 100644
index 0000000..2609e0f
--- /dev/null
+++ b/test/401-optimizing-compiler/src/Main.java
@@ -0,0 +1,54 @@
+/*
+ * Copyright (C) 2014 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.
+ */
+
+// Note that $opt$ is a marker for the optimizing compiler to ensure
+// it does compile the method.
+
+public class Main {
+ public static void main(String[] args) {
+ Error error = null;
+ try {
+ $opt$TestInvokeStatic();
+ } catch (Error e) {
+ error = e;
+ }
+ System.out.println(error);
+ }
+
+ public static void $opt$TestInvokeStatic() {
+ printStaticMethod();
+ forceGCStaticMethod();
+ throwStaticMethod();
+ }
+
+ public static void printStaticMethod() {
+ System.out.println("In static method");
+ }
+
+ public static void forceGCStaticMethod() {
+ Runtime.getRuntime().gc();
+ Runtime.getRuntime().gc();
+ Runtime.getRuntime().gc();
+ Runtime.getRuntime().gc();
+ Runtime.getRuntime().gc();
+ Runtime.getRuntime().gc();
+ System.out.println("Forced GC");
+ }
+
+ public static void throwStaticMethod() {
+ throw new Error("Error");
+ }
+}
diff --git a/test/etc/push-and-run-test-jar b/test/etc/push-and-run-test-jar
index 2a2aa70..9e30f65 100755
--- a/test/etc/push-and-run-test-jar
+++ b/test/etc/push-and-run-test-jar
@@ -18,6 +18,7 @@
QUIET="n"
DEV_MODE="n"
INVOKE_WITH=""
+FLAGS=""
while true; do
if [ "x$1" = "x--quiet" ]; then
@@ -31,6 +32,11 @@
fi
LIB="$1"
shift
+ elif [ "x$1" = "x-Xcompiler-option" ]; then
+ shift
+ option="$1"
+ FLAGS="${FLAGS} -Xcompiler-option $option"
+ shift
elif [ "x$1" = "x--boot" ]; then
shift
BOOT_OPT="$1"
@@ -141,7 +147,7 @@
JNI_OPTS="-Xjnigreflimit:512 -Xcheck:jni"
cmdline="cd $DEX_LOCATION && mkdir dalvik-cache && export ANDROID_DATA=$DEX_LOCATION && export DEX_LOCATION=$DEX_LOCATION && \
- $INVOKE_WITH $gdb /system/bin/dalvikvm $gdbargs -XXlib:$LIB $ZYGOTE $JNI_OPTS $INT_OPTS $DEBUGGER_OPTS $BOOT_OPT -cp $DEX_LOCATION/$TEST_NAME.jar Main"
+ $INVOKE_WITH $gdb /system/bin/dalvikvm $FLAGS $gdbargs -XXlib:$LIB $ZYGOTE $JNI_OPTS $INT_OPTS $DEBUGGER_OPTS $BOOT_OPT -cp $DEX_LOCATION/$TEST_NAME.jar Main"
if [ "$DEV_MODE" = "y" ]; then
echo $cmdline "$@"
fi
diff --git a/test/run-test b/test/run-test
index cc15e58..8ff0915 100755
--- a/test/run-test
+++ b/test/run-test
@@ -36,7 +36,7 @@
tmp_dir="/tmp/${test_dir}"
export JAVA="java"
-export JAVAC="javac -g -source 1.5 -target 1.5"
+export JAVAC="javac -g"
export RUN="${progdir}/etc/push-and-run-test-jar"
export DEX_LOCATION=/data/run-test/${test_dir}
export NEED_DEX="true"
diff --git a/tools/art b/tools/art
index aa53a39..c9c0d4f 100755
--- a/tools/art
+++ b/tools/art
@@ -46,12 +46,15 @@
PROG_DIR="$(cd "${PROG_NAME%/*}" ; pwd -P)"
ANDROID_BUILD_TOP="$(cd "${PROG_DIR}/../../../../" ; pwd -P)/"
ANDROID_HOST_OUT=$PROG_DIR/..
+ANDROID_DATA=$PWD/android-data$$
-mkdir -p /tmp/android-data/dalvik-cache
-ANDROID_DATA=/tmp/android-data \
+mkdir -p $ANDROID_DATA/dalvik-cache
+ANDROID_DATA=$ANDROID_DATA \
ANDROID_ROOT=$ANDROID_HOST_OUT \
LD_LIBRARY_PATH=$ANDROID_HOST_OUT/lib \
- exec $invoke_with $ANDROID_HOST_OUT/bin/dalvikvm $lib \
- -Xbootclasspath:$ANDROID_HOST_OUT/core-hostdex.jar \
+ $invoke_with $ANDROID_HOST_OUT/bin/dalvikvm $lib \
-Ximage:$ANDROID_HOST_OUT/framework/core.art \
"$@"
+EXIT_STATUS=$?
+rm -rf $ANDROID_DATA
+exit $EXIT_STATUS