Add method info to oat files
The method info data is stored separately from the code info to
reduce oat size by improving deduplication of stack maps.
To reduce code size, this moves the invoke info and inline info
method indices to this table.
Oat size for a large app (arm64): 77746816 -> 74023552 (-4.8%)
Average oat size reduction for golem (arm64): 2%
Repurposed unused SrcMapElem deduping to be for MethodInfo.
TODO: Delete SrcMapElem in a follow up CL.
Bug: 36124906
Test: clean-oat-host && test-art-host-run-test
Change-Id: I2241362e728389030b959f42161ce817cf6e2009
diff --git a/compiler/optimizing/code_generator.cc b/compiler/optimizing/code_generator.cc
index 424b850..b7c8075 100644
--- a/compiler/optimizing/code_generator.cc
+++ b/compiler/optimizing/code_generator.cc
@@ -654,8 +654,12 @@
}
}
-size_t CodeGenerator::ComputeStackMapsSize() {
- return stack_map_stream_.PrepareForFillIn();
+void CodeGenerator::ComputeStackMapAndMethodInfoSize(size_t* stack_map_size,
+ size_t* method_info_size) {
+ DCHECK(stack_map_size != nullptr);
+ DCHECK(method_info_size != nullptr);
+ *stack_map_size = stack_map_stream_.PrepareForFillIn();
+ *method_info_size = stack_map_stream_.ComputeMethodInfoSize();
}
static void CheckCovers(uint32_t dex_pc,
@@ -723,10 +727,13 @@
}
}
-void CodeGenerator::BuildStackMaps(MemoryRegion region, const DexFile::CodeItem& code_item) {
- stack_map_stream_.FillIn(region);
+void CodeGenerator::BuildStackMaps(MemoryRegion stack_map_region,
+ MemoryRegion method_info_region,
+ const DexFile::CodeItem& code_item) {
+ stack_map_stream_.FillInCodeInfo(stack_map_region);
+ stack_map_stream_.FillInMethodInfo(method_info_region);
if (kIsDebugBuild) {
- CheckLoopEntriesCanBeUsedForOsr(*graph_, CodeInfo(region), code_item);
+ CheckLoopEntriesCanBeUsedForOsr(*graph_, CodeInfo(stack_map_region), code_item);
}
}
diff --git a/compiler/optimizing/code_generator.h b/compiler/optimizing/code_generator.h
index b912672..ea463ee 100644
--- a/compiler/optimizing/code_generator.h
+++ b/compiler/optimizing/code_generator.h
@@ -341,8 +341,10 @@
slow_paths_.push_back(std::unique_ptr<SlowPathCode>(slow_path));
}
- void BuildStackMaps(MemoryRegion region, const DexFile::CodeItem& code_item);
- size_t ComputeStackMapsSize();
+ void BuildStackMaps(MemoryRegion stack_map_region,
+ MemoryRegion method_info_region,
+ const DexFile::CodeItem& code_item);
+ void ComputeStackMapAndMethodInfoSize(size_t* stack_map_size, size_t* method_info_size);
size_t GetNumberOfJitRoots() const {
return jit_string_roots_.size() + jit_class_roots_.size();
}
diff --git a/compiler/optimizing/optimizing_compiler.cc b/compiler/optimizing/optimizing_compiler.cc
index d6153b0..8b1b04b 100644
--- a/compiler/optimizing/optimizing_compiler.cc
+++ b/compiler/optimizing/optimizing_compiler.cc
@@ -1,3 +1,4 @@
+
/*
* Copyright (C) 2014 The Android Open Source Project
*
@@ -856,8 +857,15 @@
const DexFile::CodeItem* code_item) const {
ArenaVector<LinkerPatch> linker_patches = EmitAndSortLinkerPatches(codegen);
ArenaVector<uint8_t> stack_map(arena->Adapter(kArenaAllocStackMaps));
- stack_map.resize(codegen->ComputeStackMapsSize());
- codegen->BuildStackMaps(MemoryRegion(stack_map.data(), stack_map.size()), *code_item);
+ ArenaVector<uint8_t> method_info(arena->Adapter(kArenaAllocStackMaps));
+ size_t stack_map_size = 0;
+ size_t method_info_size = 0;
+ codegen->ComputeStackMapAndMethodInfoSize(&stack_map_size, &method_info_size);
+ stack_map.resize(stack_map_size);
+ method_info.resize(method_info_size);
+ codegen->BuildStackMaps(MemoryRegion(stack_map.data(), stack_map.size()),
+ MemoryRegion(method_info.data(), method_info.size()),
+ *code_item);
CompiledMethod* compiled_method = CompiledMethod::SwapAllocCompiledMethod(
compiler_driver,
@@ -869,7 +877,7 @@
codegen->HasEmptyFrame() ? 0 : codegen->GetFrameSize(),
codegen->GetCoreSpillMask(),
codegen->GetFpuSpillMask(),
- ArrayRef<const SrcMapElem>(),
+ ArrayRef<const uint8_t>(method_info),
ArrayRef<const uint8_t>(stack_map),
ArrayRef<const uint8_t>(*codegen->GetAssembler()->cfi().data()),
ArrayRef<const LinkerPatch>(linker_patches));
@@ -1200,7 +1208,9 @@
}
}
- size_t stack_map_size = codegen->ComputeStackMapsSize();
+ size_t stack_map_size = 0;
+ size_t method_info_size = 0;
+ codegen->ComputeStackMapAndMethodInfoSize(&stack_map_size, &method_info_size);
size_t number_of_roots = codegen->GetNumberOfJitRoots();
ClassLinker* class_linker = Runtime::Current()->GetClassLinker();
// We allocate an object array to ensure the JIT roots that we will collect in EmitJitRoots
@@ -1216,20 +1226,30 @@
return false;
}
uint8_t* stack_map_data = nullptr;
+ uint8_t* method_info_data = nullptr;
uint8_t* roots_data = nullptr;
- uint32_t data_size = code_cache->ReserveData(
- self, stack_map_size, number_of_roots, method, &stack_map_data, &roots_data);
+ uint32_t data_size = code_cache->ReserveData(self,
+ stack_map_size,
+ method_info_size,
+ number_of_roots,
+ method,
+ &stack_map_data,
+ &method_info_data,
+ &roots_data);
if (stack_map_data == nullptr || roots_data == nullptr) {
return false;
}
MaybeRecordStat(MethodCompilationStat::kCompiled);
- codegen->BuildStackMaps(MemoryRegion(stack_map_data, stack_map_size), *code_item);
+ codegen->BuildStackMaps(MemoryRegion(stack_map_data, stack_map_size),
+ MemoryRegion(method_info_data, method_info_size),
+ *code_item);
codegen->EmitJitRoots(code_allocator.GetData(), roots, roots_data);
const void* code = code_cache->CommitCode(
self,
method,
stack_map_data,
+ method_info_data,
roots_data,
codegen->HasEmptyFrame() ? 0 : codegen->GetFrameSize(),
codegen->GetCoreSpillMask(),
diff --git a/compiler/optimizing/stack_map_stream.cc b/compiler/optimizing/stack_map_stream.cc
index 4d12ad6..b7840d7 100644
--- a/compiler/optimizing/stack_map_stream.cc
+++ b/compiler/optimizing/stack_map_stream.cc
@@ -152,6 +152,9 @@
encoding.location_catalog.num_entries = location_catalog_entries_.size();
encoding.location_catalog.num_bytes = ComputeDexRegisterLocationCatalogSize();
encoding.inline_info.num_entries = inline_infos_.size();
+ // Must be done before calling ComputeInlineInfoEncoding since ComputeInlineInfoEncoding requires
+ // dex_method_index_idx to be filled in.
+ PrepareMethodIndices();
ComputeInlineInfoEncoding(&encoding.inline_info.encoding,
encoding.dex_register_map.num_bytes);
CodeOffset max_native_pc_offset = ComputeMaxNativePcCodeOffset();
@@ -245,7 +248,7 @@
for (size_t j = 0; j < entry.inlining_depth; ++j) {
InlineInfoEntry inline_entry = inline_infos_[inline_info_index++];
if (inline_entry.method == nullptr) {
- method_index_max = std::max(method_index_max, inline_entry.method_index);
+ method_index_max = std::max(method_index_max, inline_entry.dex_method_index_idx);
extra_data_max = std::max(extra_data_max, 1u);
} else {
method_index_max = std::max(
@@ -288,7 +291,25 @@
return entry.offset;
}
-void StackMapStream::FillIn(MemoryRegion region) {
+void StackMapStream::FillInMethodInfo(MemoryRegion region) {
+ {
+ MethodInfo info(region.begin(), method_indices_.size());
+ for (size_t i = 0; i < method_indices_.size(); ++i) {
+ info.SetMethodIndex(i, method_indices_[i]);
+ }
+ }
+ if (kIsDebugBuild) {
+ // Check the data matches.
+ MethodInfo info(region.begin());
+ const size_t count = info.NumMethodIndices();
+ DCHECK_EQ(count, method_indices_.size());
+ for (size_t i = 0; i < count; ++i) {
+ DCHECK_EQ(info.GetMethodIndex(i), method_indices_[i]);
+ }
+ }
+}
+
+void StackMapStream::FillInCodeInfo(MemoryRegion region) {
DCHECK_EQ(0u, current_entry_.dex_pc) << "EndStackMapEntry not called after BeginStackMapEntry";
DCHECK_NE(0u, needed_size_) << "PrepareForFillIn not called before FillIn";
@@ -345,7 +366,7 @@
InvokeInfo invoke_info(code_info.GetInvokeInfo(encoding, invoke_info_idx));
invoke_info.SetNativePcCodeOffset(encoding.invoke_info.encoding, entry.native_pc_code_offset);
invoke_info.SetInvokeType(encoding.invoke_info.encoding, entry.invoke_type);
- invoke_info.SetMethodIndex(encoding.invoke_info.encoding, entry.dex_method_index);
+ invoke_info.SetMethodIndexIdx(encoding.invoke_info.encoding, entry.dex_method_index_idx);
++invoke_info_idx;
}
@@ -364,7 +385,7 @@
for (size_t depth = 0; depth < entry.inlining_depth; ++depth) {
InlineInfoEntry inline_entry = inline_infos_[depth + entry.inline_infos_start_index];
if (inline_entry.method != nullptr) {
- inline_info.SetMethodIndexAtDepth(
+ inline_info.SetMethodIndexIdxAtDepth(
encoding.inline_info.encoding,
depth,
High32Bits(reinterpret_cast<uintptr_t>(inline_entry.method)));
@@ -373,9 +394,9 @@
depth,
Low32Bits(reinterpret_cast<uintptr_t>(inline_entry.method)));
} else {
- inline_info.SetMethodIndexAtDepth(encoding.inline_info.encoding,
- depth,
- inline_entry.method_index);
+ inline_info.SetMethodIndexIdxAtDepth(encoding.inline_info.encoding,
+ depth,
+ inline_entry.dex_method_index_idx);
inline_info.SetExtraDataAtDepth(encoding.inline_info.encoding, depth, 1);
}
inline_info.SetDexPcAtDepth(encoding.inline_info.encoding, depth, inline_entry.dex_pc);
@@ -533,6 +554,29 @@
return dedupe.size();
}
+void StackMapStream::PrepareMethodIndices() {
+ CHECK(method_indices_.empty());
+ method_indices_.resize(stack_maps_.size() + inline_infos_.size());
+ ArenaUnorderedMap<uint32_t, size_t> dedupe(allocator_->Adapter(kArenaAllocStackMapStream));
+ for (StackMapEntry& stack_map : stack_maps_) {
+ const size_t index = dedupe.size();
+ const uint32_t method_index = stack_map.dex_method_index;
+ if (method_index != DexFile::kDexNoIndex) {
+ stack_map.dex_method_index_idx = dedupe.emplace(method_index, index).first->second;
+ method_indices_[index] = method_index;
+ }
+ }
+ for (InlineInfoEntry& inline_info : inline_infos_) {
+ const size_t index = dedupe.size();
+ const uint32_t method_index = inline_info.method_index;
+ CHECK_NE(method_index, DexFile::kDexNoIndex);
+ inline_info.dex_method_index_idx = dedupe.emplace(method_index, index).first->second;
+ method_indices_[index] = method_index;
+ }
+ method_indices_.resize(dedupe.size());
+}
+
+
size_t StackMapStream::PrepareStackMasks(size_t entry_size_in_bits) {
// Preallocate memory since we do not want it to move (the dedup map will point into it).
const size_t byte_entry_size = RoundUp(entry_size_in_bits, kBitsPerByte) / kBitsPerByte;
@@ -590,7 +634,8 @@
DCHECK_EQ(invoke_info.GetNativePcOffset(encoding.invoke_info.encoding, instruction_set_),
entry.native_pc_code_offset.Uint32Value(instruction_set_));
DCHECK_EQ(invoke_info.GetInvokeType(encoding.invoke_info.encoding), entry.invoke_type);
- DCHECK_EQ(invoke_info.GetMethodIndex(encoding.invoke_info.encoding), entry.dex_method_index);
+ DCHECK_EQ(invoke_info.GetMethodIndexIdx(encoding.invoke_info.encoding),
+ entry.dex_method_index_idx);
invoke_info_index++;
}
CheckDexRegisterMap(code_info,
@@ -615,8 +660,10 @@
DCHECK_EQ(inline_info.GetArtMethodAtDepth(encoding.inline_info.encoding, d),
inline_entry.method);
} else {
- DCHECK_EQ(inline_info.GetMethodIndexAtDepth(encoding.inline_info.encoding, d),
- inline_entry.method_index);
+ const size_t method_index_idx =
+ inline_info.GetMethodIndexIdxAtDepth(encoding.inline_info.encoding, d);
+ DCHECK_EQ(method_index_idx, inline_entry.dex_method_index_idx);
+ DCHECK_EQ(method_indices_[method_index_idx], inline_entry.method_index);
}
CheckDexRegisterMap(code_info,
@@ -633,4 +680,9 @@
}
}
+size_t StackMapStream::ComputeMethodInfoSize() const {
+ DCHECK_NE(0u, needed_size_) << "PrepareForFillIn not called before " << __FUNCTION__;
+ return MethodInfo::ComputeSize(method_indices_.size());
+}
+
} // namespace art
diff --git a/compiler/optimizing/stack_map_stream.h b/compiler/optimizing/stack_map_stream.h
index 4225a87..e6471e1 100644
--- a/compiler/optimizing/stack_map_stream.h
+++ b/compiler/optimizing/stack_map_stream.h
@@ -22,6 +22,7 @@
#include "base/hash_map.h"
#include "base/value_object.h"
#include "memory_region.h"
+#include "method_info.h"
#include "nodes.h"
#include "stack_map.h"
@@ -70,6 +71,7 @@
inline_infos_(allocator->Adapter(kArenaAllocStackMapStream)),
stack_masks_(allocator->Adapter(kArenaAllocStackMapStream)),
register_masks_(allocator->Adapter(kArenaAllocStackMapStream)),
+ method_indices_(allocator->Adapter(kArenaAllocStackMapStream)),
dex_register_entries_(allocator->Adapter(kArenaAllocStackMapStream)),
stack_mask_max_(-1),
dex_pc_max_(0),
@@ -120,6 +122,7 @@
size_t dex_register_map_index;
InvokeType invoke_type;
uint32_t dex_method_index;
+ uint32_t dex_method_index_idx; // Index into dex method index table.
};
struct InlineInfoEntry {
@@ -128,6 +131,7 @@
uint32_t method_index;
DexRegisterMapEntry dex_register_entry;
size_t dex_register_map_index;
+ uint32_t dex_method_index_idx; // Index into the dex method index table.
};
void BeginStackMapEntry(uint32_t dex_pc,
@@ -164,7 +168,10 @@
// Prepares the stream to fill in a memory region. Must be called before FillIn.
// Returns the size (in bytes) needed to store this stream.
size_t PrepareForFillIn();
- void FillIn(MemoryRegion region);
+ void FillInCodeInfo(MemoryRegion region);
+ void FillInMethodInfo(MemoryRegion region);
+
+ size_t ComputeMethodInfoSize() const;
private:
size_t ComputeDexRegisterLocationCatalogSize() const;
@@ -180,6 +187,9 @@
// Returns the number of unique register masks.
size_t PrepareRegisterMasks();
+ // Prepare and deduplicate method indices.
+ void PrepareMethodIndices();
+
// Deduplicate entry if possible and return the corresponding index into dex_register_entries_
// array. If entry is not a duplicate, a new entry is added to dex_register_entries_.
size_t AddDexRegisterMapEntry(const DexRegisterMapEntry& entry);
@@ -232,6 +242,7 @@
ArenaVector<InlineInfoEntry> inline_infos_;
ArenaVector<uint8_t> stack_masks_;
ArenaVector<uint32_t> register_masks_;
+ ArenaVector<uint32_t> method_indices_;
ArenaVector<DexRegisterMapEntry> dex_register_entries_;
int stack_mask_max_;
uint32_t dex_pc_max_;
diff --git a/compiler/optimizing/stack_map_test.cc b/compiler/optimizing/stack_map_test.cc
index 330f7f2..a842c6e 100644
--- a/compiler/optimizing/stack_map_test.cc
+++ b/compiler/optimizing/stack_map_test.cc
@@ -60,7 +60,7 @@
size_t size = stream.PrepareForFillIn();
void* memory = arena.Alloc(size, kArenaAllocMisc);
MemoryRegion region(memory, size);
- stream.FillIn(region);
+ stream.FillInCodeInfo(region);
CodeInfo code_info(region);
CodeInfoEncoding encoding = code_info.ExtractEncoding();
@@ -173,7 +173,7 @@
size_t size = stream.PrepareForFillIn();
void* memory = arena.Alloc(size, kArenaAllocMisc);
MemoryRegion region(memory, size);
- stream.FillIn(region);
+ stream.FillInCodeInfo(region);
CodeInfo code_info(region);
CodeInfoEncoding encoding = code_info.ExtractEncoding();
@@ -433,7 +433,7 @@
size_t size = stream.PrepareForFillIn();
void* memory = arena.Alloc(size, kArenaAllocMisc);
MemoryRegion region(memory, size);
- stream.FillIn(region);
+ stream.FillInCodeInfo(region);
CodeInfo code_info(region);
CodeInfoEncoding encoding = code_info.ExtractEncoding();
@@ -519,7 +519,7 @@
size_t size = stream.PrepareForFillIn();
void* memory = arena.Alloc(size, kArenaAllocMisc);
MemoryRegion region(memory, size);
- stream.FillIn(region);
+ stream.FillInCodeInfo(region);
CodeInfo code_info(region);
CodeInfoEncoding encoding = code_info.ExtractEncoding();
@@ -611,7 +611,7 @@
size_t size = stream.PrepareForFillIn();
void* memory = arena.Alloc(size, kArenaAllocMisc);
MemoryRegion region(memory, size);
- stream.FillIn(region);
+ stream.FillInCodeInfo(region);
CodeInfo code_info(region);
CodeInfoEncoding encoding = code_info.ExtractEncoding();
@@ -672,7 +672,7 @@
size_t size = stream.PrepareForFillIn();
void* memory = arena.Alloc(size, kArenaAllocMisc);
MemoryRegion region(memory, size);
- stream.FillIn(region);
+ stream.FillInCodeInfo(region);
CodeInfo ci(region);
CodeInfoEncoding encoding = ci.ExtractEncoding();
@@ -721,7 +721,7 @@
size_t size = stream.PrepareForFillIn();
void* memory = arena.Alloc(size, kArenaAllocMisc);
MemoryRegion region(memory, size);
- stream.FillIn(region);
+ stream.FillInCodeInfo(region);
CodeInfo code_info(region);
CodeInfoEncoding encoding = code_info.ExtractEncoding();
@@ -823,7 +823,7 @@
size_t size = stream.PrepareForFillIn();
void* memory = arena.Alloc(size, kArenaAllocMisc);
MemoryRegion region(memory, size);
- stream.FillIn(region);
+ stream.FillInCodeInfo(region);
CodeInfo ci(region);
CodeInfoEncoding encoding = ci.ExtractEncoding();
@@ -950,7 +950,7 @@
size_t size = stream.PrepareForFillIn();
void* memory = arena.Alloc(size, kArenaAllocMisc);
MemoryRegion region(memory, size);
- stream.FillIn(region);
+ stream.FillInCodeInfo(region);
CodeInfo code_info(region);
CodeInfoEncoding encoding = code_info.ExtractEncoding();
@@ -979,11 +979,16 @@
stream.AddInvoke(kDirect, 65535);
stream.EndStackMapEntry();
- const size_t size = stream.PrepareForFillIn();
- MemoryRegion region(arena.Alloc(size, kArenaAllocMisc), size);
- stream.FillIn(region);
+ const size_t code_info_size = stream.PrepareForFillIn();
+ MemoryRegion code_info_region(arena.Alloc(code_info_size, kArenaAllocMisc), code_info_size);
+ stream.FillInCodeInfo(code_info_region);
- CodeInfo code_info(region);
+ const size_t method_info_size = stream.ComputeMethodInfoSize();
+ MemoryRegion method_info_region(arena.Alloc(method_info_size, kArenaAllocMisc), method_info_size);
+ stream.FillInMethodInfo(method_info_region);
+
+ CodeInfo code_info(code_info_region);
+ MethodInfo method_info(method_info_region.begin());
CodeInfoEncoding encoding = code_info.ExtractEncoding();
ASSERT_EQ(3u, code_info.GetNumberOfStackMaps(encoding));
@@ -996,13 +1001,13 @@
EXPECT_TRUE(invoke2.IsValid());
EXPECT_TRUE(invoke3.IsValid());
EXPECT_EQ(invoke1.GetInvokeType(encoding.invoke_info.encoding), kSuper);
- EXPECT_EQ(invoke1.GetMethodIndex(encoding.invoke_info.encoding), 1u);
+ EXPECT_EQ(invoke1.GetMethodIndex(encoding.invoke_info.encoding, method_info), 1u);
EXPECT_EQ(invoke1.GetNativePcOffset(encoding.invoke_info.encoding, kRuntimeISA), 4u);
EXPECT_EQ(invoke2.GetInvokeType(encoding.invoke_info.encoding), kStatic);
- EXPECT_EQ(invoke2.GetMethodIndex(encoding.invoke_info.encoding), 3u);
+ EXPECT_EQ(invoke2.GetMethodIndex(encoding.invoke_info.encoding, method_info), 3u);
EXPECT_EQ(invoke2.GetNativePcOffset(encoding.invoke_info.encoding, kRuntimeISA), 8u);
EXPECT_EQ(invoke3.GetInvokeType(encoding.invoke_info.encoding), kDirect);
- EXPECT_EQ(invoke3.GetMethodIndex(encoding.invoke_info.encoding), 65535u);
+ EXPECT_EQ(invoke3.GetMethodIndex(encoding.invoke_info.encoding, method_info), 65535u);
EXPECT_EQ(invoke3.GetNativePcOffset(encoding.invoke_info.encoding, kRuntimeISA), 16u);
}