ART: Implement a fixed size string dex cache
Previously, the string dex cache was dex_file->NumStringIds() size, and
@ruhler found that only ~1% of that cache was ever getting filled. Since
many of these string dex caches were previously 100,000+ indices in
length, we're wasting a few hundred KB per app by storing null pointers.
The intent of this project was to reduce the space the string dex cache
is using, while not regressing on time that much. This is the first of a
few CLs, which implements the new fixed size array and disables the
compiled code so it always goes slow path. In four other CLs, I
implemented a "medium path" that regresses from the previous "fast path"
only a bit in assembly in the entrypoints. @vmarko will introduce new
compiled code in the future so that we ultimately won't be regressing on
time at all. Overall, space savings have been confirmed as on the order
of 100 KB per application.
A 4-5% slow down in art-opt on Golem, and no noticeable slow down in the
interpreter. The opt slow down should be diminished once the new
compiled code is introduced.
Test: m test-art-host
Bug: 20323084
Change-Id: Ic654a1fb9c1ae127dde59290bf36a23edb55ca8e
diff --git a/compiler/optimizing/code_generator_arm64.cc b/compiler/optimizing/code_generator_arm64.cc
index d95e7df..004d427 100644
--- a/compiler/optimizing/code_generator_arm64.cc
+++ b/compiler/optimizing/code_generator_arm64.cc
@@ -4195,7 +4195,6 @@
}
void InstructionCodeGeneratorARM64::VisitLoadString(HLoadString* load) {
- Location out_loc = load->GetLocations()->Out();
Register out = OutputRegister(load);
switch (load->GetLoadKind()) {
@@ -4231,63 +4230,15 @@
__ Ldr(out.W(), codegen_->DeduplicateBootImageAddressLiteral(load->GetAddress()));
return; // No dex cache slow path.
}
- case HLoadString::LoadKind::kDexCacheAddress: {
- DCHECK_NE(load->GetAddress(), 0u);
- // LDR immediate has a 12-bit offset multiplied by the size and for 32-bit loads
- // that gives a 16KiB range. To try and reduce the number of literals if we load
- // multiple strings, simply split the dex cache address to a 16KiB aligned base
- // loaded from a literal and the remaining offset embedded in the load.
- static_assert(sizeof(GcRoot<mirror::String>) == 4u, "Expected GC root to be 4 bytes.");
- DCHECK_ALIGNED(load->GetAddress(), 4u);
- constexpr size_t offset_bits = /* encoded bits */ 12 + /* scale */ 2;
- uint64_t base_address = load->GetAddress() & ~MaxInt<uint64_t>(offset_bits);
- uint32_t offset = load->GetAddress() & MaxInt<uint64_t>(offset_bits);
- __ Ldr(out.X(), codegen_->DeduplicateDexCacheAddressLiteral(base_address));
- // /* GcRoot<mirror::String> */ out = *(base_address + offset)
- GenerateGcRootFieldLoad(load, out_loc, out.X(), offset);
- break;
- }
- case HLoadString::LoadKind::kDexCachePcRelative: {
- // Add ADRP with its PC-relative DexCache access patch.
- const DexFile& dex_file = load->GetDexFile();
- uint32_t element_offset = load->GetDexCacheElementOffset();
- vixl::aarch64::Label* adrp_label =
- codegen_->NewPcRelativeDexCacheArrayPatch(dex_file, element_offset);
- {
- SingleEmissionCheckScope guard(GetVIXLAssembler());
- __ Bind(adrp_label);
- __ adrp(out.X(), /* offset placeholder */ 0);
- }
- // Add LDR with its PC-relative DexCache access patch.
- vixl::aarch64::Label* ldr_label =
- codegen_->NewPcRelativeDexCacheArrayPatch(dex_file, element_offset, adrp_label);
- // /* GcRoot<mirror::String> */ out = *(base_address + offset) /* PC-relative */
- GenerateGcRootFieldLoad(load, out_loc, out.X(), /* offset placeholder */ 0, ldr_label);
- break;
- }
- case HLoadString::LoadKind::kDexCacheViaMethod: {
- Register current_method = InputRegisterAt(load, 0);
- // /* GcRoot<mirror::Class> */ out = current_method->declaring_class_
- GenerateGcRootFieldLoad(
- load, out_loc, current_method, ArtMethod::DeclaringClassOffset().Int32Value());
- // /* GcRoot<mirror::String>[] */ out = out->dex_cache_strings_
- __ Ldr(out.X(), HeapOperand(out, mirror::Class::DexCacheStringsOffset().Uint32Value()));
- // /* GcRoot<mirror::String> */ out = out[string_index]
- GenerateGcRootFieldLoad(
- load, out_loc, out.X(), CodeGenerator::GetCacheOffset(load->GetStringIndex()));
- break;
- }
default:
- LOG(FATAL) << "Unexpected load kind: " << load->GetLoadKind();
- UNREACHABLE();
+ break;
}
- if (!load->IsInDexCache()) {
- SlowPathCodeARM64* slow_path = new (GetGraph()->GetArena()) LoadStringSlowPathARM64(load);
- codegen_->AddSlowPath(slow_path);
- __ Cbz(out, slow_path->GetEntryLabel());
- __ Bind(slow_path->GetExitLabel());
- }
+ // TODO: Re-add the compiler code to do string dex cache lookup again.
+ SlowPathCodeARM64* slow_path = new (GetGraph()->GetArena()) LoadStringSlowPathARM64(load);
+ codegen_->AddSlowPath(slow_path);
+ __ B(slow_path->GetEntryLabel());
+ __ Bind(slow_path->GetExitLabel());
}
void LocationsBuilderARM64::VisitLongConstant(HLongConstant* constant) {