Use GpuTimers in CUDA and OpenCL versions of NBNXM directly

[alexxy/gromacs.git] / src / gromacs / nbnxm / cuda / nbnxm_cuda.cu

diff --git a/src/gromacs/nbnxm/cuda/nbnxm_cuda.cu b/src/gromacs/nbnxm/cuda/nbnxm_cuda.cu
index ccb1fdf26d19f4e6354eeab609a3b756a41109ad..0ddaa17b146e746cd5230692b037e787cc35fd3e 100644 (file)
--- a/src/gromacs/nbnxm/cuda/nbnxm_cuda.cu
+++ b/src/gromacs/nbnxm/cuda/nbnxm_cuda.cu
@@ -457,7 +457,7 @@ void gpu_copy_xq_to_gpu(NbnxmGpu* nb, const nbnxn_atomdata_t* nbatom, const Atom
 
     NBAtomData*         adat         = nb->atdat;
     gpu_plist*          plist        = nb->plist[iloc];
-    cu_timers_t*        t            = nb->timers;
+    Nbnxm::GpuTimers*   timers       = nb->timers;
     const DeviceStream& deviceStream = *nb->deviceStreams[iloc];
 
     bool bDoTime = nb->bDoTime;
@@ -489,7 +489,7 @@ void gpu_copy_xq_to_gpu(NbnxmGpu* nb, const nbnxn_atomdata_t* nbatom, const Atom
     /* beginning of timed HtoD section */
     if (bDoTime)
     {
-        t->xf[atomLocality].nb_h2d.openTimingRegion(deviceStream);
+        timers->xf[atomLocality].nb_h2d.openTimingRegion(deviceStream);
     }
 
     /* HtoD x, q */
@@ -505,7 +505,7 @@ void gpu_copy_xq_to_gpu(NbnxmGpu* nb, const nbnxn_atomdata_t* nbatom, const Atom
 
     if (bDoTime)
     {
-        t->xf[atomLocality].nb_h2d.closeTimingRegion(deviceStream);
+        timers->xf[atomLocality].nb_h2d.closeTimingRegion(deviceStream);
     }
 
     /* When we get here all misc operations issued in the local stream as well as
@@ -539,7 +539,7 @@ void gpu_launch_kernel(NbnxmGpu* nb, const gmx::StepWorkload& stepWork, const In
     NBAtomData*         adat         = nb->atdat;
     NBParamGpu*         nbp          = nb->nbparam;
     gpu_plist*          plist        = nb->plist[iloc];
-    cu_timers_t*        t            = nb->timers;
+    Nbnxm::GpuTimers*   timers       = nb->timers;
     const DeviceStream& deviceStream = *nb->deviceStreams[iloc];
 
     bool bDoTime = nb->bDoTime;
@@ -578,7 +578,7 @@ void gpu_launch_kernel(NbnxmGpu* nb, const gmx::StepWorkload& stepWork, const In
     /* beginning of timed nonbonded calculation section */
     if (bDoTime)
     {
-        t->interaction[iloc].nb_k.openTimingRegion(deviceStream);
+        timers->interaction[iloc].nb_k.openTimingRegion(deviceStream);
     }
 
     /* Kernel launch config:
@@ -619,7 +619,7 @@ void gpu_launch_kernel(NbnxmGpu* nb, const gmx::StepWorkload& stepWork, const In
                 config.sharedMemorySize);
     }
 
-    auto*      timingEvent = bDoTime ? t->interaction[iloc].nb_k.fetchNextEvent() : nullptr;
+    auto*      timingEvent = bDoTime ? timers->interaction[iloc].nb_k.fetchNextEvent() : nullptr;
     const auto kernel =
             select_nbnxn_kernel(nbp->elecType,
                                 nbp->vdwType,
@@ -632,7 +632,7 @@ void gpu_launch_kernel(NbnxmGpu* nb, const gmx::StepWorkload& stepWork, const In
 
     if (bDoTime)
     {
-        t->interaction[iloc].nb_k.closeTimingRegion(deviceStream);
+        timers->interaction[iloc].nb_k.closeTimingRegion(deviceStream);
     }
 
     if (GMX_NATIVE_WINDOWS)
@@ -660,7 +660,7 @@ void gpu_launch_kernel_pruneonly(NbnxmGpu* nb, const InteractionLocality iloc, c
     NBAtomData*         adat         = nb->atdat;
     NBParamGpu*         nbp          = nb->nbparam;
     gpu_plist*          plist        = nb->plist[iloc];
-    cu_timers_t*        t            = nb->timers;
+    Nbnxm::GpuTimers*   timers       = nb->timers;
     const DeviceStream& deviceStream = *nb->deviceStreams[iloc];
 
     bool bDoTime = nb->bDoTime;
@@ -711,7 +711,8 @@ void gpu_launch_kernel_pruneonly(NbnxmGpu* nb, const InteractionLocality iloc, c
     GpuRegionTimer* timer = nullptr;
     if (bDoTime)
     {
-        timer = &(plist->haveFreshList ? t->interaction[iloc].prune_k : t->interaction[iloc].rollingPrune_k);
+        timer = &(plist->haveFreshList ? timers->interaction[iloc].prune_k
+                                       : timers->interaction[iloc].rollingPrune_k);
     }
 
     /* beginning of timed prune calculation section */
@@ -800,7 +801,7 @@ void gpu_launch_cpyback(NbnxmGpu*                nb,
 
     /* extract the data */
     NBAtomData*         adat         = nb->atdat;
-    cu_timers_t*        t            = nb->timers;
+    Nbnxm::GpuTimers*   timers       = nb->timers;
     bool                bDoTime      = nb->bDoTime;
     const DeviceStream& deviceStream = *nb->deviceStreams[iloc];
 
@@ -817,7 +818,7 @@ void gpu_launch_cpyback(NbnxmGpu*                nb,
     /* beginning of timed D2H section */
     if (bDoTime)
     {
-        t->xf[atomLocality].nb_d2h.openTimingRegion(deviceStream);
+        timers->xf[atomLocality].nb_d2h.openTimingRegion(deviceStream);
     }
 
     /* With DD the local D2H transfer can only start after the non-local
@@ -884,7 +885,7 @@ void gpu_launch_cpyback(NbnxmGpu*                nb,
 
     if (bDoTime)
     {
-        t->xf[atomLocality].nb_d2h.closeTimingRegion(deviceStream);
+        timers->xf[atomLocality].nb_d2h.closeTimingRegion(deviceStream);
     }
 }