perfetto/pps-producer: add optimized cpu/gpu timestamp correlation support

The Intel Xe driver added the ability to do cpu/gpu timestamp
correlation giving a much better alignment of timestamps (we use to
have ~20us delta between the 2 samples, just because of the ioctl
barrier potentially sneaking in some work).

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24591>

src/freedreno/ds/fd_pps_driver.cc

@@ -710,4 +710,11 @@ FreedrenoDriver::gpu_timestamp() const
    return perfetto::base::GetBootTimeNs().count();
 }
 
+bool
+FreedrenoDriver::cpu_gpu_timestamp(uint64_t &, uint64_t &) const
+{
+   /* Not supported */
+   return false;
+}
+
 } // namespace pps

src/freedreno/ds/fd_pps_driver.h

@@ -31,6 +31,8 @@ public:
    uint64_t next() override;
    uint32_t gpu_clock_id() const override;
    uint64_t gpu_timestamp() const override;
+   bool cpu_gpu_timestamp(uint64_t &cpu_timestamp,
+                          uint64_t &gpu_timestamp) const override;
 
 private:
    struct fd_device *dev;

src/intel/ds/intel_pps_driver.cc

@@ -342,4 +342,21 @@ uint64_t IntelDriver::gpu_timestamp() const
    return intel_device_info_timebase_scale(&perf->devinfo, timestamp);
 }
 
+bool IntelDriver::cpu_gpu_timestamp(uint64_t &cpu_timestamp,
+                                    uint64_t &gpu_timestamp) const
+{
+   if (!intel_gem_read_correlate_cpu_gpu_timestamp(drm_device.fd,
+                                                   perf->devinfo.kmd_type,
+                                                   INTEL_ENGINE_CLASS_RENDER, 0,
+                                                   CLOCK_BOOTTIME,
+                                                   &cpu_timestamp,
+                                                   &gpu_timestamp,
+                                                   NULL))
+      return false;
+
+   gpu_timestamp =
+      intel_device_info_timebase_scale(&perf->devinfo, gpu_timestamp);
+   return true;
+}
+
 } // namespace pps

src/intel/ds/intel_pps_driver.h

@@ -49,6 +49,8 @@ class IntelDriver : public Driver
    uint64_t next() override;
    uint32_t gpu_clock_id() const override;
    uint64_t gpu_timestamp() const override;
+   bool cpu_gpu_timestamp(uint64_t &cpu_timestamp,
+                          uint64_t &gpu_timestamp) const override;
 
    private:
    /// @brief Requests the next perf sample

src/panfrost/ds/pan_pps_driver.cpp

@@ -180,4 +180,11 @@ PanfrostDriver::gpu_timestamp() const
    return perfetto::base::GetBootTimeNs().count();
 }
 
+bool
+PanfrostDriver::cpu_gpu_timestamp(uint64_t &, uint64_t &) const
+{
+   /* Not supported */
+   return false;
+}
+
 } // namespace pps

src/panfrost/ds/pan_pps_driver.h

@@ -42,6 +42,8 @@ class PanfrostDriver : public Driver {
    uint64_t next() override;
    uint32_t gpu_clock_id() const override;
    uint64_t gpu_timestamp() const override;
+   bool cpu_gpu_timestamp(uint64_t &cpu_timestamp,
+                          uint64_t &gpu_timestamp) const override;
 
    uint64_t last_dump_ts = 0;
 

src/tool/pps/pps_datasource.cc

@@ -245,8 +245,14 @@ void add_timestamp(perfetto::protos::pbzero::ClockSnapshot *event, const Driver
       return;
 
    // Send a correlation event between GPU & CPU timestamps
-   uint64_t cpu_ts = perfetto::base::GetBootTimeNs().count();
+   uint64_t cpu_ts, gpu_ts;
-   uint64_t gpu_ts = driver->gpu_timestamp();
+
+   // Try to use the optimized driver correlation if available, otherwise do a
+   // separate CPU & GPU sample
+   if (!driver->cpu_gpu_timestamp(cpu_ts, gpu_ts)) {
+      cpu_ts = perfetto::base::GetBootTimeNs().count();
+      gpu_ts = driver->gpu_timestamp();
+   }
 
    {
       auto clock = event->add_clocks();

src/tool/pps/pps_driver.h

@@ -84,6 +84,12 @@ class Driver
    /// Sample a timestamp from the GPU
    virtual uint64_t gpu_timestamp() const = 0;
 
+   /// Sample a timestamp from both the CPU & the GPU
+   ///
+   /// This is useful when the driver can do a better timestamp correlation
+   /// than sampling separately CPU & GPU timestamps.
+   virtual bool cpu_gpu_timestamp(uint64_t &cpu_timestamp, uint64_t &gpu_timestamp) const = 0;
+
    DrmDevice drm_device;
 
    /// List of counter groups