[wasm] Add uma stat for wasm lazy throughput

This CL adds an uma stat to track the throughput
of lazy compilation for wasm functions in KB/s.

BUG=chromium:770618
R=kschimpf@chromium.org,bradnelson@chromium.org,isherman@chromium.org

Change-Id: Iad06cfb1f185f7e2ab6b0198282c03c3d8f29e2e
Reviewed-on: https://chromium-review.googlesource.com/706276
Commit-Queue: Aseem Garg <aseemgarg@chromium.org>
Reviewed-by: Clemens Hammacher <clemensh@chromium.org>
Reviewed-by: Brad Nelson <bradnelson@chromium.org>
Cr-Commit-Position: refs/heads/master@{#48475}

.gitignore

@@ -99,3 +99,4 @@ turbo*.cfg
 turbo*.dot
 turbo*.json
 v8.ignition_dispatches_table.json
+/Default/

src/asmjs/asm-js.cc

@@ -271,11 +271,13 @@ CompilationJob::Status AsmJsCompilationJob::ExecuteJobImpl() {
                     compilation_info()->literal()->start_position();
   compilation_info()->isolate()->counters()->asm_module_size_bytes()->AddSample(
       module_size);
-  int64_t translate_time_micro = translate_timer.Elapsed().InMicroseconds();
+  int64_t translation_time_micro = translate_timer.Elapsed().InMicroseconds();
+  // translation_throughput is not exact (assumes MB == 1000000). But that is ok
+  // since the metric is stored in buckets that lose some precision anyways.
   int translation_throughput =
-      translate_time_micro != 0
+      translation_time_micro != 0
           ? static_cast<int>(static_cast<int64_t>(module_size) /
-                             translate_time_micro)
+                             translation_time_micro)
           : 0;
   compilation_info()
       ->isolate()

src/counters.h

@@ -1032,7 +1032,9 @@ class RuntimeCallTimerScope {
      V8.WasmCompileFunctionPeakMemoryBytes, 1, GB, 51)                         \
   HR(asm_module_size_bytes, V8.AsmModuleSizeBytes, 1, GB, 51)                  \
   HR(asm_wasm_translation_throughput, V8.AsmWasmTranslationThroughput, 1, 100, \
-     20)
+     20)                                                                       \
+  HR(wasm_lazy_compilation_throughput, V8.WasmLazyCompilationThroughput, 1,    \
+     10000, 50)
 
 #define HISTOGRAM_TIMER_LIST(HT)                                               \
   /* Garbage collection timers. */                                             \

src/wasm/module-compiler.cc

@@ -729,6 +729,8 @@ compiler::ModuleEnv CreateModuleEnvFromCompiledModule(
 
 void LazyCompilationOrchestrator::CompileFunction(
     Isolate* isolate, Handle<WasmInstanceObject> instance, int func_index) {
+  base::ElapsedTimer compilation_timer;
+  compilation_timer.Start();
   Handle<WasmCompiledModule> compiled_module(instance->compiled_module(),
                                              isolate);
   if (Code::cast(compiled_module->code_table()->get(func_index))->kind() ==
@@ -789,10 +791,16 @@ void LazyCompilationOrchestrator::CompileFunction(
   code_specialization.ApplyToWasmCode(*code, SKIP_ICACHE_FLUSH);
   Assembler::FlushICache(isolate, code->instruction_start(),
                          code->instruction_size());
+  int64_t func_size =
+      static_cast<int64_t>(func->code.end_offset() - func->code.offset());
+  int64_t compilation_time = compilation_timer.Elapsed().InMicroseconds();
   auto counters = isolate->counters();
   counters->wasm_lazily_compiled_functions()->Increment();
   counters->wasm_generated_code_size()->Increment(code->body_size());
   counters->wasm_reloc_size()->Increment(code->relocation_info()->length());
+  counters->wasm_lazy_compilation_throughput()->AddSample(
+      compilation_time != 0 ? static_cast<int>(func_size / compilation_time)
+                            : 0);
 }
 
 int AdvanceSourcePositionTableIterator(SourcePositionTableIterator& iterator,