Use trampolines for calls to helpers

This is an ARM specific optimization to the compiler
that uses trampoline islands to make calls to runtime
helper functions.  The intention is to reduce the size
of the generated code (by 2 bytes per call) without
affecting performance.

By default this is on when generating an OAT file.  It is
off when compiling to memory.

To switch this off in dex2oat, use the command line option:
--no-helper-trampolines

Enhances disassembler to print the trampoline entry on the
BL instruction like this:

0xb6a850c0: f7ffff9e  bl      -196 (0xb6a85000)  ; pTestSuspend

Bug: 12607709
Change-Id: I9202bdb7cf21252ad807bd48701f1f6ce8e3d0fe

1 files changed, 39 insertions, 0 deletions

diff --git a/compiler/driver/compiler_driver.cc b/compiler/driver/compiler_driver.cc
index a241d51..8bf3b04 100644
--- a/compiler/driver/compiler_driver.cc
+++ b/compiler/driver/compiler_driver.cc
@@ -24,6 +24,8 @@
 #include <unistd.h>
 #include <utility>
 
+#include "arch/arm/final_relocations_arm.h"
+#include "base/hex_dump.h"
 #include "base/stl_util.h"
 #include "base/timing_logger.h"
 #include "class_linker.h"
@@ -506,6 +508,7 @@ void CompilerDriver::CompileAll(jobject class_loader,
   UniquePtr<ThreadPool> thread_pool(new ThreadPool("Compiler driver thread pool", thread_count_ - 1));
   PreCompile(class_loader, dex_files, thread_pool.get(), timings);
   Compile(class_loader, dex_files, thread_pool.get(), timings);
+  PostCompile();
   if (dump_stats_) {
     stats_->Dump();
   }
@@ -617,6 +620,10 @@ void CompilerDriver::PreCompile(jobject class_loader, const std::vector<const De
   UpdateImageClasses(timings);
 }
 
+void CompilerDriver::PostCompile() {
+  BuildEntrypointTrampolineCode();
+}
+
 bool CompilerDriver::IsImageClass(const char* descriptor) const {
   if (!IsImage()) {
     return true;
@@ -1240,6 +1247,25 @@ bool CompilerDriver::IsSafeCast(const DexCompilationUnit* mUnit, uint32_t dex_pc
   return result;
 }
 
+uint32_t CompilerDriver::AddEntrypointTrampoline(uint32_t entrypoint) {
+  return entrypoint_trampolines_.AddEntrypoint(Thread::Current(), entrypoint);
+}
+
+
+void CompilerDriver::BuildEntrypointTrampolineCode() {
+  const auto& table = entrypoint_trampolines_.GetTrampolineTable();
+  for (uint32_t offset : table) {
+    switch (instruction_set_) {
+      case kThumb2:
+        BuildArmEntrypointTrampolineCall(ThreadOffset<4>(offset));
+        break;
+      default:
+        UNIMPLEMENTED(FATAL) << "No entrypoint trampolines for this architecture";
+    }
+  }
+}
+
+
 void CompilerDriver::AddCodePatch(const DexFile* dex_file,
                                   uint16_t referrer_class_def_idx,
                                   uint32_t referrer_method_idx,
@@ -2150,4 +2176,17 @@ bool CompilerDriver::SkipCompilation(const std::string& method_name) {
   }
   return !compile;
 }
+
+FinalEntrypointRelocationSet* CompilerDriver::AllocateFinalEntrypointRelocationSet(
+    CompilationUnit* cu) const {
+  switch (instruction_set_) {
+    case kArm:
+    case kThumb2:
+      return new FinalEntrypointRelocationSetArm(this);
+    default:
+      UNIMPLEMENTED(FATAL) << "Cannot allocate FinalEntrypointRelocationSet for non-ARM";
+      return nullptr;
+  }
+}
+
 }  // namespace art