[wasm][pgo] Introduce a separate header and cc file

This moves the existing PGO code to a separate cc file with a separate
header. As the implementation will be further extended in follow-up CLs,
it's better to have it separated.

R=jkummerow@chromium.org

Bug: v8:13209
Change-Id: I7b7b5bf9c8d3d542dae734f3874499dccee152a9
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3899321Reviewed-by: Jakob Kummerow <jkummerow@chromium.org>
Commit-Queue: Clemens Backes <clemensb@chromium.org>
Cr-Commit-Position: refs/heads/main@{#83250}

BUILD.bazel

@@ -2527,6 +2527,8 @@ filegroup(
             "src/wasm/names-provider.cc",
             "src/wasm/names-provider.h",
             "src/wasm/object-access.h",
+            "src/wasm/pgo.cc",
+            "src/wasm/pgo.h",
             "src/wasm/simd-shuffle.cc",
             "src/wasm/simd-shuffle.h",
             "src/wasm/stacks.cc",

BUILD.gn

@@ -3625,6 +3625,7 @@ v8_header_set("v8_internal_headers") {
       "src/wasm/module-instantiate.h",
       "src/wasm/names-provider.h",
       "src/wasm/object-access.h",
+      "src/wasm/pgo.h",
       "src/wasm/simd-shuffle.h",
       "src/wasm/stacks.h",
       "src/wasm/streaming-decoder.h",
@@ -4766,6 +4767,7 @@ v8_source_set("v8_base_without_compiler") {
       "src/wasm/module-decoder.cc",
       "src/wasm/module-instantiate.cc",
       "src/wasm/names-provider.cc",
+      "src/wasm/pgo.cc",
       "src/wasm/simd-shuffle.cc",
       "src/wasm/stacks.cc",
       "src/wasm/streaming-decoder.cc",

src/wasm/pgo.cc

+// Copyright 2022 the V8 project authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#include "src/wasm/pgo.h"
+
+#include "src/wasm/decoder.h"
+#include "src/wasm/wasm-module-builder.h"  // For {ZoneBuffer}.
+
+namespace v8::internal::wasm {
+
+base::OwnedVector<uint8_t> GetProfileData(const WasmModule* module) {
+  const TypeFeedbackStorage& type_feedback = module->type_feedback;
+  AccountingAllocator allocator;
+  Zone zone{&allocator, "wasm::GetProfileData"};
+  ZoneBuffer buffer{&zone};
+  base::MutexGuard mutex_guard{&type_feedback.mutex};
+
+  // Get an ordered list of function indexes, so we generate deterministic data.
+  std::vector<uint32_t> ordered_func_indexes;
+  ordered_func_indexes.reserve(type_feedback.feedback_for_function.size());
+  for (const auto& entry : type_feedback.feedback_for_function) {
+    ordered_func_indexes.push_back(entry.first);
+  }
+  std::sort(ordered_func_indexes.begin(), ordered_func_indexes.end());
+
+  buffer.write_u32v(static_cast<uint32_t>(ordered_func_indexes.size()));
+  for (const uint32_t func_index : ordered_func_indexes) {
+    buffer.write_u32v(func_index);
+    // Serialize {feedback_vector}.
+    const FunctionTypeFeedback& feedback =
+        type_feedback.feedback_for_function.at(func_index);
+    buffer.write_u32v(static_cast<uint32_t>(feedback.feedback_vector.size()));
+    for (const CallSiteFeedback& call_site_feedback :
+         feedback.feedback_vector) {
+      int cases = call_site_feedback.num_cases();
+      buffer.write_i32v(cases);
+      for (int i = 0; i < cases; ++i) {
+        buffer.write_i32v(call_site_feedback.function_index(i));
+        buffer.write_i32v(call_site_feedback.call_count(i));
+      }
+    }
+    // Serialize {call_targets}.
+    buffer.write_u32v(static_cast<uint32_t>(feedback.call_targets.size()));
+    for (uint32_t call_target : feedback.call_targets) {
+      buffer.write_u32v(call_target);
+    }
+  }
+  return base::OwnedVector<uint8_t>::Of(buffer);
+}
+
+void RestoreProfileData(WasmModule* module,
+                        base::Vector<uint8_t> profile_data) {
+  TypeFeedbackStorage& type_feedback = module->type_feedback;
+  Decoder decoder{profile_data.begin(), profile_data.end()};
+  uint32_t num_entries = decoder.consume_u32v("num function entries");
+  CHECK_LE(num_entries, module->num_declared_functions);
+  for (uint32_t missing_entries = num_entries; missing_entries > 0;
+       --missing_entries) {
+    uint32_t function_index = decoder.consume_u32v("function index");
+    CHECK(!type_feedback.feedback_for_function.count(function_index));
+    FunctionTypeFeedback& feedback =
+        type_feedback.feedback_for_function[function_index];
+    // Deserialize {feedback_vector}.
+    uint32_t feedback_vector_size =
+        decoder.consume_u32v("feedback vector size");
+    feedback.feedback_vector.resize(feedback_vector_size);
+    for (CallSiteFeedback& feedback : feedback.feedback_vector) {
+      int num_cases = decoder.consume_i32v("num cases");
+      if (num_cases == 0) continue;  // no feedback
+      if (num_cases == 1) {          // monomorphic
+        int called_function_index = decoder.consume_i32v("function index");
+        int call_count = decoder.consume_i32v("call count");
+        feedback = CallSiteFeedback{called_function_index, call_count};
+      } else {  // polymorphic
+        auto* polymorphic = new CallSiteFeedback::PolymorphicCase[num_cases];
+        for (int i = 0; i < num_cases; ++i) {
+          polymorphic[i].function_index =
+              decoder.consume_i32v("function index");
+          polymorphic[i].absolute_call_frequency =
+              decoder.consume_i32v("call count");
+        }
+        feedback = CallSiteFeedback{polymorphic, num_cases};
+      }
+    }
+    // Deserialize {call_targets}.
+    uint32_t num_call_targets = decoder.consume_u32v("num call targets");
+    feedback.call_targets =
+        base::OwnedVector<uint32_t>::NewForOverwrite(num_call_targets);
+    for (uint32_t& call_target : feedback.call_targets) {
+      call_target = decoder.consume_u32v("call target");
+    }
+  }
+  CHECK(decoder.ok());
+  CHECK_EQ(decoder.pc(), decoder.end());
+}
+
+void DumpProfileToFile(const WasmModule* module,
+                       base::Vector<const uint8_t> wire_bytes) {
+  CHECK(!wire_bytes.empty());
+  // File are named `profile-wasm-<hash>`.
+  // We use the same hash as for reported scripts, to make it easier to
+  // correlate files to wasm modules (see {CreateWasmScript}).
+  uint32_t hash = static_cast<uint32_t>(GetWireBytesHash(wire_bytes));
+  base::EmbeddedVector<char, 32> filename;
+  SNPrintF(filename, "profile-wasm-%08x", hash);
+  base::OwnedVector<uint8_t> profile_data = GetProfileData(module);
+  PrintF("Dumping Wasm PGO data to file '%s' (%zu bytes)\n", filename.begin(),
+         profile_data.size());
+  if (FILE* file = base::OS::FOpen(filename.begin(), "wb")) {
+    CHECK_EQ(profile_data.size(),
+             fwrite(profile_data.begin(), 1, profile_data.size(), file));
+    base::Fclose(file);
+  }
+}
+
+void LoadProfileFromFile(WasmModule* module,
+                         base::Vector<const uint8_t> wire_bytes) {
+  CHECK(!wire_bytes.empty());
+  // File are named `profile-wasm-<hash>`.
+  // We use the same hash as for reported scripts, to make it easier to
+  // correlate files to wasm modules (see {CreateWasmScript}).
+  uint32_t hash = static_cast<uint32_t>(GetWireBytesHash(wire_bytes));
+  base::EmbeddedVector<char, 32> filename;
+  SNPrintF(filename, "profile-wasm-%08x", hash);
+
+  FILE* file = base::OS::FOpen(filename.begin(), "rb");
+  if (!file) {
+    PrintF("No Wasm PGO data found: Cannot open file '%s'\n", filename.begin());
+    return;
+  }
+
+  fseek(file, 0, SEEK_END);
+  size_t size = ftell(file);
+  rewind(file);
+
+  PrintF("Loading Wasm PGO data from file '%s' (%zu bytes)\n", filename.begin(),
+         size);
+  base::OwnedVector<uint8_t> profile_data =
+      base::OwnedVector<uint8_t>::NewForOverwrite(size);
+  for (size_t read = 0; read < size;) {
+    read += fread(profile_data.begin() + read, 1, size - read, file);
+    CHECK(!ferror(file));
+  }
+
+  base::Fclose(file);
+
+  RestoreProfileData(module, profile_data.as_vector());
+
+  // Check that the generated profile is deterministic.
+  DCHECK_EQ(profile_data.as_vector(), GetProfileData(module).as_vector());
+}
+
+}  // namespace v8::internal::wasm

src/wasm/pgo.h

+// Copyright 2022 the V8 project authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#if !V8_ENABLE_WEBASSEMBLY
+#error This header should only be included if WebAssembly is enabled.
+#endif  // !V8_ENABLE_WEBASSEMBLY
+
+#ifndef V8_WASM_PGO_H_
+#define V8_WASM_PGO_H_
+
+#include "src/base/vector.h"
+
+namespace v8::internal::wasm {
+
+struct WasmModule;
+
+void DumpProfileToFile(const WasmModule* module,
+                       base::Vector<const uint8_t> wire_bytes);
+
+void LoadProfileFromFile(WasmModule* module,
+                         base::Vector<const uint8_t> wire_bytes);
+
+}  // namespace v8::internal::wasm
+
+#endif  // V8_WASM_PGO_H_

src/wasm/wasm-code-manager.cc

@@ -34,6 +34,7 @@
 #include "src/wasm/jump-table-assembler.h"
 #include "src/wasm/module-compiler.h"
 #include "src/wasm/names-provider.h"
+#include "src/wasm/pgo.h"
 #include "src/wasm/wasm-debug.h"
 #include "src/wasm/wasm-engine.h"
 #include "src/wasm/wasm-import-wrapper-cache.h"

src/wasm/wasm-engine.cc

@@ -23,6 +23,7 @@
 #include "src/wasm/module-compiler.h"
 #include "src/wasm/module-decoder.h"
 #include "src/wasm/module-instantiate.h"
+#include "src/wasm/pgo.h"
 #include "src/wasm/stacks.h"
 #include "src/wasm/streaming-decoder.h"
 #include "src/wasm/wasm-debug.h"

src/wasm/wasm-module.cc

@@ -678,146 +678,4 @@ size_t GetWireBytesHash(base::Vector<const uint8_t> wire_bytes) {
       kZeroHashSeed);
 }
 
-base::OwnedVector<uint8_t> GetProfileData(const WasmModule* module) {
-  const TypeFeedbackStorage& type_feedback = module->type_feedback;
-  AccountingAllocator allocator;
-  Zone zone{&allocator, "wasm::GetProfileData"};
-  ZoneBuffer buffer{&zone};
-  base::MutexGuard mutex_guard{&type_feedback.mutex};
-
-  // Get an ordered list of function indexes, so we generate deterministic data.
-  std::vector<uint32_t> ordered_func_indexes;
-  ordered_func_indexes.reserve(type_feedback.feedback_for_function.size());
-  for (const auto& entry : type_feedback.feedback_for_function) {
-    ordered_func_indexes.push_back(entry.first);
-  }
-  std::sort(ordered_func_indexes.begin(), ordered_func_indexes.end());
-
-  buffer.write_u32v(static_cast<uint32_t>(ordered_func_indexes.size()));
-  for (const uint32_t func_index : ordered_func_indexes) {
-    buffer.write_u32v(func_index);
-    // Serialize {feedback_vector}.
-    const FunctionTypeFeedback& feedback =
-        type_feedback.feedback_for_function.at(func_index);
-    buffer.write_u32v(static_cast<uint32_t>(feedback.feedback_vector.size()));
-    for (const CallSiteFeedback& call_site_feedback :
-         feedback.feedback_vector) {
-      int cases = call_site_feedback.num_cases();
-      buffer.write_i32v(cases);
-      for (int i = 0; i < cases; ++i) {
-        buffer.write_i32v(call_site_feedback.function_index(i));
-        buffer.write_i32v(call_site_feedback.call_count(i));
-      }
-    }
-    // Serialize {call_targets}.
-    buffer.write_u32v(static_cast<uint32_t>(feedback.call_targets.size()));
-    for (uint32_t call_target : feedback.call_targets) {
-      buffer.write_u32v(call_target);
-    }
-  }
-  return base::OwnedVector<uint8_t>::Of(buffer);
-}
-
-void RestoreProfileData(WasmModule* module,
-                        base::Vector<uint8_t> profile_data) {
-  TypeFeedbackStorage& type_feedback = module->type_feedback;
-  Decoder decoder{profile_data.begin(), profile_data.end()};
-  uint32_t num_entries = decoder.consume_u32v("num function entries");
-  CHECK_LE(num_entries, module->num_declared_functions);
-  for (uint32_t missing_entries = num_entries; missing_entries > 0;
-       --missing_entries) {
-    uint32_t function_index = decoder.consume_u32v("function index");
-    CHECK(!type_feedback.feedback_for_function.count(function_index));
-    FunctionTypeFeedback& feedback =
-        type_feedback.feedback_for_function[function_index];
-    // Deserialize {feedback_vector}.
-    uint32_t feedback_vector_size =
-        decoder.consume_u32v("feedback vector size");
-    feedback.feedback_vector.resize(feedback_vector_size);
-    for (CallSiteFeedback& feedback : feedback.feedback_vector) {
-      int num_cases = decoder.consume_i32v("num cases");
-      if (num_cases == 0) continue;  // no feedback
-      if (num_cases == 1) {          // monomorphic
-        int called_function_index = decoder.consume_i32v("function index");
-        int call_count = decoder.consume_i32v("call count");
-        feedback = CallSiteFeedback{called_function_index, call_count};
-      } else {  // polymorphic
-        auto* polymorphic = new CallSiteFeedback::PolymorphicCase[num_cases];
-        for (int i = 0; i < num_cases; ++i) {
-          polymorphic[i].function_index =
-              decoder.consume_i32v("function index");
-          polymorphic[i].absolute_call_frequency =
-              decoder.consume_i32v("call count");
-        }
-        feedback = CallSiteFeedback{polymorphic, num_cases};
-      }
-    }
-    // Deserialize {call_targets}.
-    uint32_t num_call_targets = decoder.consume_u32v("num call targets");
-    feedback.call_targets =
-        base::OwnedVector<uint32_t>::NewForOverwrite(num_call_targets);
-    for (uint32_t& call_target : feedback.call_targets) {
-      call_target = decoder.consume_u32v("call target");
-    }
-  }
-  CHECK(decoder.ok());
-  CHECK_EQ(decoder.pc(), decoder.end());
-}
-
-void DumpProfileToFile(const WasmModule* module,
-                       base::Vector<const uint8_t> wire_bytes) {
-  CHECK(!wire_bytes.empty());
-  // File are named `profile-wasm-<hash>`.
-  // We use the same hash as for reported scripts, to make it easier to
-  // correlate files to wasm modules (see {CreateWasmScript}).
-  uint32_t hash = static_cast<uint32_t>(GetWireBytesHash(wire_bytes));
-  base::EmbeddedVector<char, 32> filename;
-  SNPrintF(filename, "profile-wasm-%08x", hash);
-  base::OwnedVector<uint8_t> profile_data = GetProfileData(module);
-  PrintF("Dumping Wasm PGO data to file '%s' (%zu bytes)\n", filename.begin(),
-         profile_data.size());
-  if (FILE* file = base::OS::FOpen(filename.begin(), "wb")) {
-    CHECK_EQ(profile_data.size(),
-             fwrite(profile_data.begin(), 1, profile_data.size(), file));
-    base::Fclose(file);
-  }
-}
-
-void LoadProfileFromFile(WasmModule* module,
-                         base::Vector<const uint8_t> wire_bytes) {
-  CHECK(!wire_bytes.empty());
-  // File are named `profile-wasm-<hash>`.
-  // We use the same hash as for reported scripts, to make it easier to
-  // correlate files to wasm modules (see {CreateWasmScript}).
-  uint32_t hash = static_cast<uint32_t>(GetWireBytesHash(wire_bytes));
-  base::EmbeddedVector<char, 32> filename;
-  SNPrintF(filename, "profile-wasm-%08x", hash);
-
-  FILE* file = base::OS::FOpen(filename.begin(), "rb");
-  if (!file) {
-    PrintF("No Wasm PGO data found: Cannot open file '%s'\n", filename.begin());
-    return;
-  }
-
-  fseek(file, 0, SEEK_END);
-  size_t size = ftell(file);
-  rewind(file);
-
-  PrintF("Loading Wasm PGO data from file '%s' (%zu bytes)\n", filename.begin(),
-         size);
-  base::OwnedVector<uint8_t> profile_data =
-      base::OwnedVector<uint8_t>::NewForOverwrite(size);
-  for (size_t read = 0; read < size;) {
-    read += fread(profile_data.begin() + read, 1, size - read, file);
-    CHECK(!ferror(file));
-  }
-
-  base::Fclose(file);
-
-  RestoreProfileData(module, profile_data.as_vector());
-
-  // Check that the generated profile is deterministic.
-  DCHECK_EQ(profile_data.as_vector(), GetProfileData(module).as_vector());
-}
-
 }  // namespace v8::internal::wasm

src/wasm/wasm-module.h

@@ -783,12 +783,6 @@ size_t PrintSignature(base::Vector<char> buffer, const wasm::FunctionSig*,
 V8_EXPORT_PRIVATE size_t
 GetWireBytesHash(base::Vector<const uint8_t> wire_bytes);
 
-void DumpProfileToFile(const WasmModule* module,
-                       base::Vector<const uint8_t> wire_bytes);
-
-void LoadProfileFromFile(WasmModule* module,
-                         base::Vector<const uint8_t> wire_bytes);
-
 }  // namespace v8::internal::wasm
 
 #endif  // V8_WASM_WASM_MODULE_H_