src/ui/lib/escher/profiling/timestamp_profiler.h

// Copyright 2017 The Fuchsia Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.

#ifndef SRC_UI_LIB_ESCHER_PROFILING_TIMESTAMP_PROFILER_H_
#define SRC_UI_LIB_ESCHER_PROFILING_TIMESTAMP_PROFILER_H_

#include <vector>

#include "src/lib/fxl/memory/ref_counted.h"
#include "src/ui/lib/escher/forward_declarations.h"
#include "src/ui/lib/escher/third_party/granite/vk/command_buffer.h"
#include "src/ui/lib/escher/vk/vulkan_context.h"

namespace escher {

class TimestampProfiler : public fxl::RefCountedThreadSafe<TimestampProfiler> {
 public:
  TimestampProfiler(vk::Device device, float timestamp_period);
  ~TimestampProfiler();

  // Add a Vulkan timestamp-query that will mark the time that all previous
  // commands in |cmd_buf| have finished the pipeline stage specified by
  // |flags|.  For example, use eVertexShader to mark the time that vertex
  // shaders have been applied to all vertices from previous draw-calls.
  //
  // |name| needs to be a global constant. The trace engine that ultimately
  // consumes Result structs caches it, so if the contents were overwritten the
  // records would be inaccurate.
  //
  // NOTE: you should understand the caveats in the Vulkan spec regarding the
  // accuracy of these timestamps.  For example, many implementations will treat
  // any set of flags as equivalent to eBottomOfPipe.
  void AddTimestamp(CommandBufferPtr cmd_buf, vk::PipelineStageFlagBits flags, const char* name);

  struct Result {
    uint64_t raw_nanoseconds;  // nanoseconds according to some timebase.
    uint64_t time;             // microseconds elapsed since the first timestamp.
    uint64_t elapsed;          // microseconds elapsed since the previous timestamp.
    const char* name;
  };

  struct TraceEvent {
    uint64_t start_elapsed_ticks;    // ticks elapsed from the first timestamp to start
    uint64_t end_elapsed_ticks;      // ticks elapsed from the first timestamp to end
    std::vector<const char*> names;  // all trace event names for the time range
  };

  // GetQueryResults() returns the raw events generated by a series of
  // AddTimeStamp() calls.
  std::vector<Result> GetQueryResults();

  // Transforms a vector of Results into a vector of TraceEvents more suitable
  // for the trace system to use.
  //
  // NOTE: Currently only supported on Fuchsia.
  static std::vector<TraceEvent> ProcessTraceEvents(const std::vector<Result>& timestamps);

  // Uses VTHREAD trace macros to register all GPU work represented in
  // |trace_events|. |trace_events| should be returned from a call to
  // ProcessTraceEvents() |trace_literal| is the name that the event will take
  // and should be a string literal.
  static void TraceGpuQueryResults(const std::vector<TraceEvent>& trace_events,
                                   uint64_t frame_number, uint64_t escher_frame_number,
                                   const char* trace_literal, const char* gpu_vthread_literal,
                                   uint64_t gpu_vthread_id);

  static void LogGpuQueryResults(uint64_t escher_frame_number,
                                 const std::vector<Result>& timestamps);

 private:
  // Each QueryRange keeps track of current usage of a separate vk::QueryPool.
  // A new pool (and hence a new QueryRange) is used whenever:
  //   - the capacity of the previous pool is reached.
  //   - a different CommandBuffer is passed to AddTimestamp().
  struct QueryRange {
    vk::QueryPool pool;
    vk::CommandBuffer command_buffer;
    uint32_t start_index;  // within the pool
    uint32_t count;
  };

  QueryRange* ObtainRange(CommandBufferPtr cmd_buf);
  QueryRange* CreateRange(CommandBufferPtr cmd_buf);
  QueryRange* CreateRangeAndPool(CommandBufferPtr cmd_buf);

  std::vector<QueryRange> ranges_;
  std::vector<vk::QueryPool> pools_;

  // Remembers timestamp names, and will eventually be filled in with timestamp
  // values before returning results.
  std::vector<Result> results_;

  uint32_t query_count_ = 0;
  uint32_t current_pool_index_ = 0;

  const vk::Device device_;
  const float timestamp_period_;
};

}  // namespace escher

#endif  // SRC_UI_LIB_ESCHER_PROFILING_TIMESTAMP_PROFILER_H_