PnaclCoordinator: Use llc's module-splitting for pexe compilation

When there is more than 1 core in the system, request extra temp files
and use LLC's StreamInitWithSplit rpc to enable module splitting.
If the RPC fails, fall back to StreamInitWithOverrides.

Add a private Pepper interface to get the cpu count rather than
duplicating base::SysInfo in the plugin.

TEST=browser_tests --gtest_filter=NaClBrowserTstPnacl.*
BUG= https://code.google.com/p/nativeclient/issues/detail?id=3777
R=jschuh@chromium.org, jvoung@chromium.org, teravest@chromium.org

Review URL: https://codereview.chromium.org/165433003

git-svn-id: svn://svn.chromium.org/chrome/trunk/src@251458 0039d316-1c4b-4281-b951-d872f2087c98

7 files changed, 169 insertions, 41 deletions

diff --git a/ppapi/api/private/ppb_nacl_private.idl b/ppapi/api/private/ppb_nacl_private.idl
index 6dfbff7..a674328 100644
--- a/ppapi/api/private/ppb_nacl_private.idl
+++ b/ppapi/api/private/ppb_nacl_private.idl
@@ -110,6 +110,9 @@ interface PPB_NaCl_Private {
    */
   PP_FileHandle CreateTemporaryFile([in] PP_Instance instance);
 
+  /* Return the number of processors in the system as reported by the OS */
+  int32_t GetNumberOfProcessors();
+
   /* Create a temporary file, which will be deleted by the time the
    * last handle is closed (or earlier on POSIX systems), to use for
    * the nexe with the cache information given by |pexe_url|,
diff --git a/ppapi/c/private/ppb_nacl_private.h b/ppapi/c/private/ppb_nacl_private.h
index 0001148..dde686b 100644
--- a/ppapi/c/private/ppb_nacl_private.h
+++ b/ppapi/c/private/ppb_nacl_private.h
@@ -3,7 +3,7 @@
  * found in the LICENSE file.
  */
 
-/* From private/ppb_nacl_private.idl modified Mon Feb 10 11:05:29 2014. */
+/* From private/ppb_nacl_private.idl modified Thu Feb 13 15:50:26 2014. */
 
 #ifndef PPAPI_C_PRIVATE_PPB_NACL_PRIVATE_H_
 #define PPAPI_C_PRIVATE_PPB_NACL_PRIVATE_H_
@@ -127,6 +127,8 @@ struct PPB_NaCl_Private_1_0 {
    * returns a posix handle to that temporary file.
    */
   PP_FileHandle (*CreateTemporaryFile)(PP_Instance instance);
+  /* Return the number of processors in the system as reported by the OS */
+  int32_t (*GetNumberOfProcessors)(void);
   /* Create a temporary file, which will be deleted by the time the
    * last handle is closed (or earlier on POSIX systems), to use for
    * the nexe with the cache information given by |pexe_url|,
diff --git a/ppapi/native_client/src/trusted/plugin/pnacl_coordinator.cc b/ppapi/native_client/src/trusted/plugin/pnacl_coordinator.cc
index 9e96f0d..91112a0 100644
--- a/ppapi/native_client/src/trusted/plugin/pnacl_coordinator.cc
+++ b/ppapi/native_client/src/trusted/plugin/pnacl_coordinator.cc
@@ -200,6 +200,9 @@ PnaclCoordinator* PnaclCoordinator::BitcodeToNative(
   PLUGIN_PRINTF(("PnaclCoordinator::BitcodeToNative (manifest=%p, ",
                  reinterpret_cast<const void*>(coordinator->manifest_.get())));
 
+  int cpus = plugin->nacl_interface()->GetNumberOfProcessors();
+  coordinator->split_module_count_ = std::min(4, std::max(1, cpus));
+
   // First start a network request for the pexe, to tickle the component
   // updater's On-Demand resource throttler, and to get Last-Modified/ETag
   // cache information. We can cancel the request later if there's
@@ -220,6 +223,8 @@ PnaclCoordinator::PnaclCoordinator(
     manifest_(new PnaclManifest()),
     pexe_url_(pexe_url),
     pnacl_options_(pnacl_options),
+    split_module_count_(1),
+    num_object_files_opened_(0),
     is_cache_hit_(PP_FALSE),
     error_already_reported_(false),
     pnacl_init_time_(0),
@@ -248,6 +253,9 @@ PnaclCoordinator::~PnaclCoordinator() {
         plugin_->pp_instance(),
         PP_FALSE);
   }
+  for (int i = 0; i < num_object_files_opened_; i++) {
+    delete obj_files_[i];
+  }
 }
 
 nacl::DescWrapper* PnaclCoordinator::ReleaseTranslatedFD() {
@@ -531,10 +539,14 @@ void PnaclCoordinator::NexeFdDidOpen(int32_t pp_error) {
   } else {
     // Open an object file first so the translator can start writing to it
     // during streaming translation.
-    obj_file_.reset(new TempFile(plugin_));
-    pp::CompletionCallback obj_cb =
-        callback_factory_.NewCallback(&PnaclCoordinator::ObjectFileDidOpen);
-    obj_file_->Open(obj_cb, true);
+    for (int i = 0; i < split_module_count_; i++) {
+      obj_files_.push_back(new TempFile(plugin_));
+
+      pp::CompletionCallback obj_cb =
+          callback_factory_.NewCallback(&PnaclCoordinator::ObjectFileDidOpen);
+      obj_files_[i]->Open(obj_cb, true);
+    }
+    invalid_desc_wrapper_.reset(plugin_->wrapper_factory()->MakeInvalid());
 
     // Meanwhile, a miss means we know we need to stream the bitcode, so stream
     // the rest of it now. (Calling FinishStreaming means that the downloader
@@ -643,11 +655,14 @@ void PnaclCoordinator::ObjectFileDidOpen(int32_t pp_error) {
                      "Failed to open scratch object file.");
     return;
   }
-  // Open the nexe file for connecting ld and sel_ldr.
-  // Start translation when done with this last step of setup!
-  pp::CompletionCallback cb =
-      callback_factory_.NewCallback(&PnaclCoordinator::RunTranslate);
-  temp_nexe_file_->Open(cb, true);
+  num_object_files_opened_++;
+  if (num_object_files_opened_ == split_module_count_) {
+    // Open the nexe file for connecting ld and sel_ldr.
+    // Start translation when done with this last step of setup!
+    pp::CompletionCallback cb =
+        callback_factory_.NewCallback(&PnaclCoordinator::RunTranslate);
+    temp_nexe_file_->Open(cb, true);
+  }
 }
 
 void PnaclCoordinator::RunTranslate(int32_t pp_error) {
@@ -661,8 +676,9 @@ void PnaclCoordinator::RunTranslate(int32_t pp_error) {
   CHECK(translate_thread_ != NULL);
   translate_thread_->RunTranslate(report_translate_finished,
                                   manifest_.get(),
-                                  obj_file_.get(),
+                                  &obj_files_,
                                   temp_nexe_file_.get(),
+                                  invalid_desc_wrapper_.get(),
                                   &error_info_,
                                   resources_.get(),
                                   &pnacl_options_,
diff --git a/ppapi/native_client/src/trusted/plugin/pnacl_coordinator.h b/ppapi/native_client/src/trusted/plugin/pnacl_coordinator.h
index e38cbc0..9475b9e 100644
--- a/ppapi/native_client/src/trusted/plugin/pnacl_coordinator.h
+++ b/ppapi/native_client/src/trusted/plugin/pnacl_coordinator.h
@@ -77,6 +77,9 @@ class TempFile;
 //   Complete when NexeReadDidOpen is invoked.
 class PnaclCoordinator: public CallbackSource<FileStreamData> {
  public:
+  // Maximum number of object files passable to the translator. Cannot be
+  // changed without changing the RPC signatures.
+  const static size_t kMaxTranslatorObjectFiles = 16;
   virtual ~PnaclCoordinator();
 
   // The factory method for translations.
@@ -202,7 +205,12 @@ class PnaclCoordinator: public CallbackSource<FileStreamData> {
   PnaclOptions pnacl_options_;
 
   // Object file, produced by the translator and consumed by the linker.
-  nacl::scoped_ptr<TempFile> obj_file_;
+  std::vector<TempFile*> obj_files_;
+  nacl::scoped_ptr<nacl::DescWrapper> invalid_desc_wrapper_;
+  // Number of split modules (threads) for llc
+  int split_module_count_;
+  int num_object_files_opened_;
+
   // Translated nexe file, produced by the linker.
   nacl::scoped_ptr<TempFile> temp_nexe_file_;
   // Passed to the browser, which sets it to true if there is a translation
diff --git a/ppapi/native_client/src/trusted/plugin/pnacl_translate_thread.cc b/ppapi/native_client/src/trusted/plugin/pnacl_translate_thread.cc
index d51efe4..2093aaf 100644
--- a/ppapi/native_client/src/trusted/plugin/pnacl_translate_thread.cc
+++ b/ppapi/native_client/src/trusted/plugin/pnacl_translate_thread.cc
@@ -4,6 +4,8 @@
 
 #include "ppapi/native_client/src/trusted/plugin/pnacl_translate_thread.h"
 
+#include <iterator>
+
 #include "native_client/src/trusted/desc/nacl_desc_wrapper.h"
 #include "ppapi/native_client/src/trusted/plugin/plugin.h"
 #include "ppapi/native_client/src/trusted/plugin/plugin_error.h"
@@ -19,7 +21,7 @@ PnaclTranslateThread::PnaclTranslateThread() : llc_subprocess_active_(false),
                                                done_(false),
                                                time_stats_(),
                                                manifest_(NULL),
-                                               obj_file_(NULL),
+                                               obj_files_(NULL),
                                                nexe_file_(NULL),
                                                coordinator_error_info_(NULL),
                                                resources_(NULL),
@@ -33,8 +35,9 @@ PnaclTranslateThread::PnaclTranslateThread() : llc_subprocess_active_(false),
 void PnaclTranslateThread::RunTranslate(
     const pp::CompletionCallback& finish_callback,
     const Manifest* manifest,
-    TempFile* obj_file,
+    const std::vector<TempFile*>* obj_files,
     TempFile* nexe_file,
+    nacl::DescWrapper* invalid_desc_wrapper,
     ErrorInfo* error_info,
     PnaclResources* resources,
     PnaclOptions* pnacl_options,
@@ -42,8 +45,9 @@ void PnaclTranslateThread::RunTranslate(
     Plugin* plugin) {
   PLUGIN_PRINTF(("PnaclStreamingTranslateThread::RunTranslate)\n"));
   manifest_ = manifest;
-  obj_file_ = obj_file;
+  obj_files_ = obj_files;
   nexe_file_ = nexe_file;
+  invalid_desc_wrapper_ = invalid_desc_wrapper;
   coordinator_error_info_ = error_info;
   resources_ = resources;
   pnacl_options_ = pnacl_options;
@@ -130,7 +134,14 @@ void WINAPI PnaclTranslateThread::DoTranslateThread(void* arg) {
 void PnaclTranslateThread::DoTranslate() {
   ErrorInfo error_info;
   SrpcParams params;
-  nacl::DescWrapper* llc_out_file = obj_file_->write_wrapper();
+  std::vector<nacl::DescWrapper*> llc_out_files;
+  size_t i;
+  for (i = 0; i < obj_files_->size(); i++) {
+    llc_out_files.push_back((*obj_files_)[i]->write_wrapper());
+  }
+  for (; i < PnaclCoordinator::kMaxTranslatorObjectFiles; i++) {
+    llc_out_files.push_back(invalid_desc_wrapper_);
+  }
 
   {
     nacl::MutexLocker ml(&subprocess_mu_);
@@ -149,19 +160,60 @@ void PnaclTranslateThread::DoTranslate() {
     // Run LLC.
     PluginReverseInterface* llc_reverse =
         llc_subprocess_->service_runtime()->rev_interface();
-    llc_reverse->AddTempQuotaManagedFile(obj_file_->identifier());
+    for (size_t i = 0; i < obj_files_->size(); i++) {
+      llc_reverse->AddTempQuotaManagedFile((*obj_files_)[i]->identifier());
+    }
   }
 
   int64_t compile_start_time = NaClGetTimeOfDayMicroseconds();
   bool init_success;
   std::vector<char> options = pnacl_options_->GetOptCommandline();
+
+  // Try to init with splitting
+  // TODO(dschuff): This CL override is ugly. Change llc to default to using
+  // the number of modules specified in the first param, and ignore multiple
+  // uses of -split-module
+  std::vector<char> split_args;
+  nacl::stringstream ss;
+  ss << "-split-module=" << obj_files_->size();
+  nacl::string split_arg = ss.str();
+  std::copy(split_arg.begin(), split_arg.end(), std::back_inserter(split_args));
+  split_args.push_back('\x00');
+  std::copy(options.begin(), options.end(), std::back_inserter(split_args));
+  int modules_used = static_cast<int>(obj_files_->size());
   init_success = llc_subprocess_->InvokeSrpcMethod(
-      "StreamInitWithOverrides",
-      "hC",
+      "StreamInitWithSplit",
+      "ihhhhhhhhhhhhhhhhC",
       &params,
-      llc_out_file->desc(),
-      &options[0],
-      options.size());
+      modules_used,
+      llc_out_files[0]->desc(),
+      llc_out_files[1]->desc(),
+      llc_out_files[2]->desc(),
+      llc_out_files[3]->desc(),
+      llc_out_files[4]->desc(),
+      llc_out_files[5]->desc(),
+      llc_out_files[6]->desc(),
+      llc_out_files[7]->desc(),
+      llc_out_files[8]->desc(),
+      llc_out_files[9]->desc(),
+      llc_out_files[10]->desc(),
+      llc_out_files[11]->desc(),
+      llc_out_files[12]->desc(),
+      llc_out_files[13]->desc(),
+      llc_out_files[14]->desc(),
+      llc_out_files[15]->desc(),
+      &split_args[0],
+      split_args.size());
+  if (!init_success) {
+    init_success = llc_subprocess_->InvokeSrpcMethod(
+        "StreamInitWithOverrides",
+        "hC",
+        &params,
+        llc_out_files[0]->desc(),
+        &options[0],
+        options.size());
+    modules_used = 1;
+  }
 
   if (!init_success) {
     if (llc_subprocess_->srpc_client()->GetLastError() ==
@@ -256,25 +308,36 @@ void PnaclTranslateThread::DoTranslate() {
   llc_subprocess_.reset(NULL);
   NaClXMutexUnlock(&subprocess_mu_);
 
-  if(!RunLdSubprocess(is_shared_library, soname, lib_dependencies)) {
+  if(!RunLdSubprocess(
+         modules_used, is_shared_library, soname, lib_dependencies)) {
     return;
   }
   core->CallOnMainThread(0, report_translate_finished_, PP_OK);
 }
 
-bool PnaclTranslateThread::RunLdSubprocess(int is_shared_library,
+bool PnaclTranslateThread::RunLdSubprocess(int modules_used,
+                                           int is_shared_library,
                                            const nacl::string& soname,
                                            const nacl::string& lib_dependencies
                                            ) {
   ErrorInfo error_info;
   SrpcParams params;
-  // Reset object file for reading first.
-  if (!obj_file_->Reset()) {
-    TranslateFailed(ERROR_PNACL_LD_SETUP,
-                    "Link process could not reset object file");
-    return false;
+
+  std::vector<nacl::DescWrapper*> ld_in_files;
+  size_t i;
+  for (i = 0; i < obj_files_->size(); i++) {
+    // Reset object file for reading first.
+    if (!(*obj_files_)[i]->Reset()) {
+      TranslateFailed(ERROR_PNACL_LD_SETUP,
+                      "Link process could not reset object file");
+      return false;
+    }
+    ld_in_files.push_back((*obj_files_)[i]->read_wrapper());
   }
-  nacl::DescWrapper* ld_in_file = obj_file_->read_wrapper();
+  for (; i < PnaclCoordinator::kMaxTranslatorObjectFiles; i++) {
+    ld_in_files.push_back(invalid_desc_wrapper_);
+  }
+
   nacl::DescWrapper* ld_out_file = nexe_file_->write_wrapper();
 
   {
@@ -299,14 +362,41 @@ bool PnaclTranslateThread::RunLdSubprocess(int is_shared_library,
 
   int64_t link_start_time = NaClGetTimeOfDayMicroseconds();
   // Run LD.
-  if (!ld_subprocess_->InvokeSrpcMethod("RunWithDefaultCommandLine",
-                                       "hhiss",
-                                       &params,
-                                       ld_in_file->desc(),
-                                       ld_out_file->desc(),
-                                       is_shared_library,
-                                       soname.c_str(),
-                                       lib_dependencies.c_str())) {
+  bool success;
+  // If we ran LLC with module splitting, we can't fall back here.
+  if (modules_used > 1) {
+    success = ld_subprocess_->InvokeSrpcMethod("RunWithSplit",
+                                               "ihhhhhhhhhhhhhhhhh",
+                                               &params,
+                                               modules_used,
+                                               ld_in_files[0]->desc(),
+                                               ld_in_files[1]->desc(),
+                                               ld_in_files[2]->desc(),
+                                               ld_in_files[3]->desc(),
+                                               ld_in_files[4]->desc(),
+                                               ld_in_files[5]->desc(),
+                                               ld_in_files[6]->desc(),
+                                               ld_in_files[7]->desc(),
+                                               ld_in_files[8]->desc(),
+                                               ld_in_files[9]->desc(),
+                                               ld_in_files[10]->desc(),
+                                               ld_in_files[11]->desc(),
+                                               ld_in_files[12]->desc(),
+                                               ld_in_files[13]->desc(),
+                                               ld_in_files[14]->desc(),
+                                               ld_in_files[15]->desc(),
+                                               ld_out_file->desc());
+  } else {
+    success = ld_subprocess_->InvokeSrpcMethod("RunWithDefaultCommandLine",
+                                               "hhiss",
+                                               &params,
+                                               ld_in_files[0]->desc(),
+                                               ld_out_file->desc(),
+                                               is_shared_library,
+                                               soname.c_str(),
+                                               lib_dependencies.c_str());
+  }
+  if (!success) {
     TranslateFailed(ERROR_PNACL_LD_INTERNAL,
                     "link failed.");
     return false;
diff --git a/ppapi/native_client/src/trusted/plugin/pnacl_translate_thread.h b/ppapi/native_client/src/trusted/plugin/pnacl_translate_thread.h
index 31981eb..11473b4 100644
--- a/ppapi/native_client/src/trusted/plugin/pnacl_translate_thread.h
+++ b/ppapi/native_client/src/trusted/plugin/pnacl_translate_thread.h
@@ -50,8 +50,9 @@ class PnaclTranslateThread {
   // as it is passed in with PutBytes.
   void RunTranslate(const pp::CompletionCallback& finish_callback,
                     const Manifest* manifest,
-                    TempFile* obj_file,
+                    const std::vector<TempFile*>* obj_files,
                     TempFile* nexe_file,
+                    nacl::DescWrapper* invalid_desc_wrapper,
                     ErrorInfo* error_info,
                     PnaclResources* resources,
                     PnaclOptions* pnacl_options,
@@ -85,7 +86,8 @@ class PnaclTranslateThread {
   void TranslateFailed(enum PluginErrorCode err_code,
                        const nacl::string& error_string);
   // Run the LD subprocess, returning true on success
-  bool RunLdSubprocess(int is_shared_library,
+  bool RunLdSubprocess(int modules_used,
+                       int is_shared_library,
                        const nacl::string& soname,
                        const nacl::string& lib_dependencies);
 
@@ -121,8 +123,9 @@ class PnaclTranslateThread {
 
   // Data about the translation files, owned by the coordinator
   const Manifest* manifest_;
-  TempFile* obj_file_;
+  const std::vector<TempFile*>* obj_files_;
   TempFile* nexe_file_;
+  nacl::DescWrapper* invalid_desc_wrapper_;
   ErrorInfo* coordinator_error_info_;
   PnaclResources* resources_;
   PnaclOptions* pnacl_options_;
diff --git a/ppapi/native_client/src/untrusted/pnacl_irt_shim/pnacl_shim.c b/ppapi/native_client/src/untrusted/pnacl_irt_shim/pnacl_shim.c
index 8696279..330c966 100644
--- a/ppapi/native_client/src/untrusted/pnacl_irt_shim/pnacl_shim.c
+++ b/ppapi/native_client/src/untrusted/pnacl_irt_shim/pnacl_shim.c
@@ -3144,6 +3144,11 @@ static PP_FileHandle Pnacl_M25_PPB_NaCl_Private_CreateTemporaryFile(PP_Instance
   return iface->CreateTemporaryFile(instance);
 }
 
+static int32_t Pnacl_M25_PPB_NaCl_Private_GetNumberOfProcessors(void) {
+  const struct PPB_NaCl_Private_1_0 *iface = Pnacl_WrapperInfo_PPB_NaCl_Private_1_0.real_iface;
+  return iface->GetNumberOfProcessors();
+}
+
 static int32_t Pnacl_M25_PPB_NaCl_Private_GetNexeFd(PP_Instance instance, const char* pexe_url, uint32_t abi_version, uint32_t opt_level, const char* last_modified, const char* etag, PP_Bool has_no_store_header, PP_Bool* is_hit, PP_FileHandle* nexe_handle, struct PP_CompletionCallback* callback) {
   const struct PPB_NaCl_Private_1_0 *iface = Pnacl_WrapperInfo_PPB_NaCl_Private_1_0.real_iface;
   return iface->GetNexeFd(instance, pexe_url, abi_version, opt_level, last_modified, etag, has_no_store_header, is_hit, nexe_handle, *callback);
@@ -5045,6 +5050,7 @@ static struct PPB_NaCl_Private_1_0 Pnacl_Wrappers_PPB_NaCl_Private_1_0 = {
     .BrokerDuplicateHandle = (int32_t (*)(PP_FileHandle source_handle, uint32_t process_id, PP_FileHandle* target_handle, uint32_t desired_access, uint32_t options))&Pnacl_M25_PPB_NaCl_Private_BrokerDuplicateHandle,
     .GetReadonlyPnaclFd = (PP_FileHandle (*)(const char* filename))&Pnacl_M25_PPB_NaCl_Private_GetReadonlyPnaclFd,
     .CreateTemporaryFile = (PP_FileHandle (*)(PP_Instance instance))&Pnacl_M25_PPB_NaCl_Private_CreateTemporaryFile,
+    .GetNumberOfProcessors = (int32_t (*)(void))&Pnacl_M25_PPB_NaCl_Private_GetNumberOfProcessors,
     .GetNexeFd = (int32_t (*)(PP_Instance instance, const char* pexe_url, uint32_t abi_version, uint32_t opt_level, const char* last_modified, const char* etag, PP_Bool has_no_store_header, PP_Bool* is_hit, PP_FileHandle* nexe_handle, struct PP_CompletionCallback callback))&Pnacl_M25_PPB_NaCl_Private_GetNexeFd,
     .ReportTranslationFinished = (void (*)(PP_Instance instance, PP_Bool success))&Pnacl_M25_PPB_NaCl_Private_ReportTranslationFinished,
     .ReportNaClError = (PP_ExternalPluginResult (*)(PP_Instance instance, PP_NaClError message_id))&Pnacl_M25_PPB_NaCl_Private_ReportNaClError,