[Impeller] GPUTracer for Metal.

ci/licenses_golden/licenses_flutter

@@ -2127,6 +2127,8 @@ ORIGIN: ../../../flutter/impeller/renderer/backend/metal/device_buffer_mtl.h + .
 ORIGIN: ../../../flutter/impeller/renderer/backend/metal/device_buffer_mtl.mm + ../../../flutter/LICENSE
 ORIGIN: ../../../flutter/impeller/renderer/backend/metal/formats_mtl.h + ../../../flutter/LICENSE
 ORIGIN: ../../../flutter/impeller/renderer/backend/metal/formats_mtl.mm + ../../../flutter/LICENSE
+ORIGIN: ../../../flutter/impeller/renderer/backend/metal/gpu_tracer_mtl.h + ../../../flutter/LICENSE
+ORIGIN: ../../../flutter/impeller/renderer/backend/metal/gpu_tracer_mtl.mm + ../../../flutter/LICENSE
 ORIGIN: ../../../flutter/impeller/renderer/backend/metal/pipeline_library_mtl.h + ../../../flutter/LICENSE
 ORIGIN: ../../../flutter/impeller/renderer/backend/metal/pipeline_library_mtl.mm + ../../../flutter/LICENSE
 ORIGIN: ../../../flutter/impeller/renderer/backend/metal/pipeline_mtl.h + ../../../flutter/LICENSE
@@ -4887,6 +4889,8 @@ FILE: ../../../flutter/impeller/renderer/backend/metal/device_buffer_mtl.h
 FILE: ../../../flutter/impeller/renderer/backend/metal/device_buffer_mtl.mm
 FILE: ../../../flutter/impeller/renderer/backend/metal/formats_mtl.h
 FILE: ../../../flutter/impeller/renderer/backend/metal/formats_mtl.mm
+FILE: ../../../flutter/impeller/renderer/backend/metal/gpu_tracer_mtl.h
+FILE: ../../../flutter/impeller/renderer/backend/metal/gpu_tracer_mtl.mm
 FILE: ../../../flutter/impeller/renderer/backend/metal/pipeline_library_mtl.h
 FILE: ../../../flutter/impeller/renderer/backend/metal/pipeline_library_mtl.mm
 FILE: ../../../flutter/impeller/renderer/backend/metal/pipeline_mtl.h

impeller/renderer/backend/metal/BUILD.gn

@@ -24,6 +24,8 @@ impeller_component("metal") {
     "device_buffer_mtl.mm",
     "formats_mtl.h",
     "formats_mtl.mm",
+    "gpu_tracer_mtl.h",
+    "gpu_tracer_mtl.mm",
     "pipeline_library_mtl.h",
     "pipeline_library_mtl.mm",
     "pipeline_mtl.h",

impeller/renderer/backend/metal/command_buffer_mtl.mm

@@ -159,6 +159,12 @@ static bool LogMTLCommandBufferErrorIfPresent(id<MTLCommandBuffer> buffer) {
 }
 
 bool CommandBufferMTL::OnSubmitCommands(CompletionCallback callback) {
+  auto context = context_.lock();
+  if (!context) {
+    return false;
+  }
+  ContextMTL::Cast(*context).GetGPUTracer()->RecordCmdBuffer(buffer_);
+
   if (callback) {
     [buffer_
         addCompletedHandler:^(id<MTLCommandBuffer> buffer) {
@@ -190,6 +196,8 @@ static bool LogMTLCommandBufferErrorIfPresent(id<MTLCommandBuffer> buffer) {
   auto buffer = buffer_;
   buffer_ = nil;
 
+  ContextMTL::Cast(*context).GetGPUTracer()->RecordCmdBuffer(buffer);
+
   auto worker_task_runner = ContextMTL::Cast(*context).GetWorkerTaskRunner();
   auto mtl_render_pass = static_cast<RenderPassMTL*>(render_pass.get());
 

impeller/renderer/backend/metal/context_mtl.h

@@ -16,6 +16,7 @@
 #include "impeller/core/sampler.h"
 #include "impeller/renderer/backend/metal/allocator_mtl.h"
 #include "impeller/renderer/backend/metal/command_buffer_mtl.h"
+#include "impeller/renderer/backend/metal/gpu_tracer_mtl.h"
 #include "impeller/renderer/backend/metal/pipeline_library_mtl.h"
 #include "impeller/renderer/backend/metal/shader_library_mtl.h"
 #include "impeller/renderer/capabilities.h"
@@ -93,6 +94,8 @@ class ContextMTL final : public Context,
 
   std::shared_ptr<const fml::SyncSwitch> GetIsGpuDisabledSyncSwitch() const;
 
+  std::shared_ptr<GPUTracerMTL> GetGPUTracer() const;
+
  private:
   id<MTLDevice> device_ = nullptr;
   id<MTLCommandQueue> command_queue_ = nullptr;
@@ -103,6 +106,7 @@ class ContextMTL final : public Context,
   std::shared_ptr<const Capabilities> device_capabilities_;
   std::shared_ptr<fml::ConcurrentMessageLoop> raster_message_loop_;
   std::shared_ptr<const fml::SyncSwitch> is_gpu_disabled_sync_switch_;
+  std::shared_ptr<GPUTracerMTL> gpu_tracer_;
   bool is_valid_ = false;
 
   ContextMTL(

impeller/renderer/backend/metal/context_mtl.mm

@@ -5,13 +5,15 @@
 #include "impeller/renderer/backend/metal/context_mtl.h"
 
 #include <Foundation/Foundation.h>
+#include <memory>
 
 #include "flutter/fml/concurrent_message_loop.h"
 #include "flutter/fml/file.h"
 #include "flutter/fml/logging.h"
 #include "flutter/fml/paths.h"
 #include "flutter/fml/synchronization/sync_switch.h"
 #include "impeller/core/sampler_descriptor.h"
+#include "impeller/renderer/backend/metal/gpu_tracer_mtl.h"
 #include "impeller/renderer/backend/metal/sampler_library_mtl.h"
 #include "impeller/renderer/capabilities.h"
 
@@ -142,7 +144,7 @@ static bool DeviceSupportsComputeSubgroups(id<MTLDevice> device) {
 
   device_capabilities_ =
       InferMetalCapabilities(device_, PixelFormat::kB8G8R8A8UNormInt);
-
+  gpu_tracer_ = std::make_shared<GPUTracerMTL>();
   is_valid_ = true;
 }
 
@@ -325,6 +327,10 @@ new ContextMTL(device, command_queue,
   raster_message_loop_.reset();
 }
 
+std::shared_ptr<GPUTracerMTL> ContextMTL::GetGPUTracer() const {
+  return gpu_tracer_;
+}
+
 const std::shared_ptr<fml::ConcurrentTaskRunner>
 ContextMTL::GetWorkerTaskRunner() const {
   return raster_message_loop_->GetTaskRunner();

impeller/renderer/backend/metal/gpu_tracer_mtl.h

@@ -0,0 +1,53 @@
+// Copyright 2013 The Flutter Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#pragma once
+
+#include <Metal/Metal.h>
+
+#include <memory>
+#include <optional>
+#include "impeller/base/thread.h"
+#include "impeller/base/thread_safety.h"
+#include "impeller/geometry/scalar.h"
+
+namespace impeller {
+
+class ContextMTL;
+
+/// @brief Approximate the GPU frame time by computing a difference between the
+/// smallest
+///        GPUStartTime and largest GPUEndTime for all cmd buffers submitted in
+///        a frame workload.
+class GPUTracerMTL : public std::enable_shared_from_this<GPUTracerMTL> {
+ public:
+  GPUTracerMTL() = default;
+
+  ~GPUTracerMTL() = default;
+
+  /// @brief Record that the current frame has ended. Any additional cmd buffers
+  /// will be
+  ///        attributed to the "next" frame.
+  void MarkFrameEnd();
+
+  /// @brief Record the current cmd buffer GPU execution timestamps into an
+  /// aggregate
+  ///        frame workload metric.
+  void RecordCmdBuffer(id<MTLCommandBuffer> buffer);
+
+ private:
+  struct GPUTraceState {
+    Scalar smallest_timestamp = std::numeric_limits<float>::max();
+    Scalar largest_timestamp = 0;
+    size_t pending_buffers = 0;
+  };
+
+  [[maybe_unused]] mutable Mutex trace_state_mutex_;
+  [[maybe_unused]] GPUTraceState trace_states_[16] IPLR_GUARDED_BY(
+      trace_state_mutex_);
+  [[maybe_unused]] size_t current_state_ IPLR_GUARDED_BY(trace_state_mutex_) =
+      0u;
+};
+
+}  // namespace impeller

impeller/renderer/backend/metal/gpu_tracer_mtl.mm

@@ -0,0 +1,60 @@
+// Copyright 2013 The Flutter 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 <Metal/Metal.h>
+#include "fml/trace_event.h"
+#include "impeller/renderer/backend/metal/context_mtl.h"
+#include "impeller/renderer/backend/metal/formats_mtl.h"
+
+#include <memory>
+
+#include "impeller/renderer/backend/metal/gpu_tracer_mtl.h"
+
+namespace impeller {
+
+void GPUTracerMTL::MarkFrameEnd() {
+#ifdef IMPELLER_DEBUG
+  if (@available(ios 10.3, tvos 10.2, macos 10.15, macCatalyst 13.0, *)) {
+    Lock lock(trace_state_mutex_);
+    current_state_ = (current_state_ + 1) % 16;
+  }
+#endif  // IMPELLER_DEBUG
+}
+
+void GPUTracerMTL::RecordCmdBuffer(id<MTLCommandBuffer> buffer) {
+#ifdef IMPELLER_DEBUG
+  if (@available(ios 10.3, tvos 10.2, macos 10.15, macCatalyst 13.0, *)) {
+    Lock lock(trace_state_mutex_);
+    auto current_state = current_state_;
+    trace_states_[current_state].pending_buffers += 1;
+
+    auto weak_self = weak_from_this();
+    [buffer addCompletedHandler:^(id<MTLCommandBuffer> buffer) {
+      auto self = weak_self.lock();
+      if (!self) {
+        return;
+      }
+      Lock lock(self->trace_state_mutex_);
+      auto& state = self->trace_states_[current_state];
+      state.pending_buffers--;
+      state.smallest_timestamp = std::min(
+          state.smallest_timestamp, static_cast<Scalar>(buffer.GPUStartTime));
+      state.largest_timestamp = std::max(
+          state.largest_timestamp, static_cast<Scalar>(buffer.GPUEndTime));
+
+      if (state.pending_buffers == 0) {
+        auto gpu_ms =
+            (state.largest_timestamp - state.smallest_timestamp) * 1000;
+        state.smallest_timestamp = std::numeric_limits<float>::max();
+        state.largest_timestamp = 0;
+        FML_TRACE_COUNTER("flutter", "GPUTracer",
+                          reinterpret_cast<int64_t>(this),  // Trace Counter ID
+                          "FrameTimeMS", gpu_ms);
+      }
+    }];
+  }
+#endif  // IMPELLER_DEBUG
+}
+
+}  // namespace impeller

impeller/renderer/backend/metal/surface_mtl.mm

@@ -246,6 +246,7 @@
       return false;
     }
   }
+  ContextMTL::Cast(context.get())->GetGPUTracer()->MarkFrameEnd();
 
   if (drawable_) {
     id<MTLCommandBuffer> command_buffer =