[flang-commits] [flang] Improve omp offload profiler (PR #68016)

via flang-commits flang-commits at lists.llvm.org
Mon Oct 2 11:46:01 PDT 2023


github-actions[bot] wrote:


<!--LLVM CODE FORMAT COMMENT: {clang-format}-->

:warning: C/C++ code formatter, clang-format found issues in your code. :warning:

<details>
<summary>
You can test this locally with the following command:
</summary>

``````````bash
git-clang-format --diff 0b07b06effe5fdf779b75bb5ac6cf15e477cb0be 7c550d3190614d7572bc2d63051d7ce56c8f73b6 -- llvm/lib/Support/TimeProfiler.cpp openmp/libomptarget/src/api.cpp openmp/libomptarget/src/interface.cpp openmp/libomptarget/src/omptarget.cpp openmp/libomptarget/src/private.h
``````````

</details>

<details>
<summary>
View the diff from clang-format here.
</summary>

``````````diff
diff --git a/llvm/lib/Support/TimeProfiler.cpp b/llvm/lib/Support/TimeProfiler.cpp
index 4446583102a8..330a4d93378a 100644
--- a/llvm/lib/Support/TimeProfiler.cpp
+++ b/llvm/lib/Support/TimeProfiler.cpp
@@ -226,7 +226,7 @@ struct llvm::TimeTraceProfiler {
         J.attribute("tid", int64_t(TotalTid));
         J.attribute("ph", "X");
         J.attribute("ts", 0);
-        J.attribute("dur", DurNs / 1000 );
+        J.attribute("dur", DurNs / 1000);
         J.attribute("name", "Total: " + Total.first);
         J.attributeObject("args", [&] {
           J.attribute("count", int64_t(Count));
@@ -261,10 +261,10 @@ struct llvm::TimeTraceProfiler {
     // Emit the absolute time when this TimeProfiler started.
     // This can be used to combine the profiling data from
     // multiple processes and preserve actual time intervals.
-    J.attribute("beginningOfTime",
-                time_point_cast<nanoseconds>(BeginningOfTime)
-                    .time_since_epoch()
-                    .count()/1000);
+    J.attribute("beginningOfTime", time_point_cast<nanoseconds>(BeginningOfTime)
+                                           .time_since_epoch()
+                                           .count() /
+                                       1000);
 
     J.objectEnd();
   }
diff --git a/openmp/libomptarget/src/api.cpp b/openmp/libomptarget/src/api.cpp
index 5dd918808492..06de1f8f20b7 100644
--- a/openmp/libomptarget/src/api.cpp
+++ b/openmp/libomptarget/src/api.cpp
@@ -50,8 +50,8 @@ EXTERN int omp_get_initial_device(void) {
 }
 
 EXTERN void *omp_target_alloc(size_t Size, int DeviceNum) {
-  TIMESCOPE_WITH_DETAILS("dst_dev="+std::to_string(DeviceNum)
-                        +";size="+std::to_string(Size));
+  TIMESCOPE_WITH_DETAILS("dst_dev=" + std::to_string(DeviceNum) +
+                         ";size=" + std::to_string(Size));
   return targetAllocExplicit(Size, DeviceNum, TARGET_ALLOC_DEFAULT, __func__);
 }
 
@@ -136,9 +136,9 @@ EXTERN int omp_target_is_present(const void *Ptr, int DeviceNum) {
 EXTERN int omp_target_memcpy(void *Dst, const void *Src, size_t Length,
                              size_t DstOffset, size_t SrcOffset, int DstDevice,
                              int SrcDevice) {
-  TIMESCOPE_WITH_DETAILS("dst_dev="+std::to_string(DstDevice)
-                          +";src_dev="+std::to_string(SrcDevice)
-                          +";size="+std::to_string(Length));
+  TIMESCOPE_WITH_DETAILS("dst_dev=" + std::to_string(DstDevice) +
+                         ";src_dev=" + std::to_string(SrcDevice) +
+                         ";size=" + std::to_string(Length));
   DP("Call to omp_target_memcpy, dst device %d, src device %d, "
      "dst addr " DPxMOD ", src addr " DPxMOD ", dst offset %zu, "
      "src offset %zu, length %zu\n",
@@ -291,9 +291,9 @@ EXTERN int omp_target_memcpy_async(void *Dst, const void *Src, size_t Length,
                                    size_t DstOffset, size_t SrcOffset,
                                    int DstDevice, int SrcDevice,
                                    int DepObjCount, omp_depend_t *DepObjList) {
-  TIMESCOPE_WITH_DETAILS("dst_dev="+std::to_string(DstDevice)
-                          +";src_dev="+std::to_string(SrcDevice)
-                          +";size="+std::to_string(Length));
+  TIMESCOPE_WITH_DETAILS("dst_dev=" + std::to_string(DstDevice) +
+                         ";src_dev=" + std::to_string(SrcDevice) +
+                         ";size=" + std::to_string(Length));
   DP("Call to omp_target_memcpy_async, dst device %d, src device %d, "
      "dst addr " DPxMOD ", src addr " DPxMOD ", dst offset %zu, "
      "src offset %zu, length %zu\n",
@@ -379,10 +379,10 @@ EXTERN int omp_target_memcpy_rect_async(
     const size_t *Volume, const size_t *DstOffsets, const size_t *SrcOffsets,
     const size_t *DstDimensions, const size_t *SrcDimensions, int DstDevice,
     int SrcDevice, int DepObjCount, omp_depend_t *DepObjList) {
-  TIMESCOPE_WITH_DETAILS("dst_dev="+std::to_string(DstDevice)
-                          +";src_dev="+std::to_string(SrcDevice)
-                          +";size="+std::to_string(ElementSize)
-                          +";num_dims="+std::to_string(NumDims));
+  TIMESCOPE_WITH_DETAILS("dst_dev=" + std::to_string(DstDevice) +
+                         ";src_dev=" + std::to_string(SrcDevice) +
+                         ";size=" + std::to_string(ElementSize) +
+                         ";num_dims=" + std::to_string(NumDims));
   DP("Call to omp_target_memcpy_rect_async, dst device %d, src device %d, "
      "dst addr " DPxMOD ", src addr " DPxMOD ", dst offsets " DPxMOD ", "
      "src offsets " DPxMOD ", dst dims " DPxMOD ", src dims " DPxMOD ", "
diff --git a/openmp/libomptarget/src/interface.cpp b/openmp/libomptarget/src/interface.cpp
index 99a7abc7e0bc..2c7ab7a49d0b 100644
--- a/openmp/libomptarget/src/interface.cpp
+++ b/openmp/libomptarget/src/interface.cpp
@@ -80,8 +80,7 @@ targetData(ident_t *Loc, int64_t DeviceId, int32_t ArgNum, void **ArgsBase,
                 "TargetAsyncInfoTy must be convertible to AsyncInfoTy.");
 
   TIMESCOPE_WITH_DETAILS_AND_IDENT("Runtime: Data Copy",
-                                  "NumArgs="+
-                                  std::to_string(ArgNum), Loc);
+                                   "NumArgs=" + std::to_string(ArgNum), Loc);
 
   DP("Entering data %s region for device %" PRId64 " with %d mappings\n",
      RegionName, DeviceId, ArgNum);
@@ -274,10 +273,11 @@ static inline int targetKernel(ident_t *Loc, int64_t DeviceId, int32_t NumTeams,
   assert(KernelArgs->ThreadLimit[0] == static_cast<uint32_t>(ThreadLimit) &&
          !KernelArgs->ThreadLimit[1] && !KernelArgs->ThreadLimit[2] &&
          "OpenMP interface should not use multiple dimensions");
-  TIMESCOPE_WITH_DETAILS_AND_IDENT("Runtime target exe",
-                                   "NumTeams="+std::to_string(NumTeams)+
-                                   ";NumArgs="+
-                                   std::to_string(KernelArgs->NumArgs), Loc);
+  TIMESCOPE_WITH_DETAILS_AND_IDENT(
+      "Runtime target exe",
+      "NumTeams=" + std::to_string(NumTeams) +
+          ";NumArgs=" + std::to_string(KernelArgs->NumArgs),
+      Loc);
 
   if (getInfoLevel() & OMP_INFOTYPE_KERNEL_ARGS)
     printKernelArguments(Loc, DeviceId, KernelArgs->NumArgs,
@@ -305,7 +305,7 @@ static inline int targetKernel(ident_t *Loc, int64_t DeviceId, int32_t NumTeams,
 
   int Rc = OFFLOAD_SUCCESS;
   Rc = target(Loc, Device, HostPtr, *KernelArgs, AsyncInfo);
-  { //required to show syncronization
+  { // required to show syncronization
     TIMESCOPE_WITH_RTM_AND_IDENT("syncronize", Loc);
     if (Rc == OFFLOAD_SUCCESS)
       Rc = AsyncInfo.synchronize();
diff --git a/openmp/libomptarget/src/omptarget.cpp b/openmp/libomptarget/src/omptarget.cpp
index b5a2dfc68569..630ded281e1a 100644
--- a/openmp/libomptarget/src/omptarget.cpp
+++ b/openmp/libomptarget/src/omptarget.cpp
@@ -574,10 +574,8 @@ int targetDataBegin(ident_t *Loc, DeviceTy &Device, int32_t ArgNum,
     if ((ArgTypes[I] & OMP_TGT_MAPTYPE_LITERAL) ||
         (ArgTypes[I] & OMP_TGT_MAPTYPE_PRIVATE))
       continue;
-    TIMESCOPE_WITH_DETAILS_AND_IDENT("HostToDev",
-                                     "Size="+
-                                     std::to_string(ArgSizes[I])+
-                                     "B", Loc);
+    TIMESCOPE_WITH_DETAILS_AND_IDENT(
+        "HostToDev", "Size=" + std::to_string(ArgSizes[I]) + "B", Loc);
     if (ArgMappers && ArgMappers[I]) {
       // Instead of executing the regular path of targetDataBegin, call the
       // targetDataMapper variant which will call targetDataBegin again
@@ -953,8 +951,8 @@ int targetDataEnd(ident_t *Loc, DeviceTy &Device, int32_t ArgNum,
         !TPR.Flags.IsHostPointer && DataSize != 0) {
       DP("Moving %" PRId64 " bytes (tgt:" DPxMOD ") -> (hst:" DPxMOD ")\n",
          DataSize, DPxPTR(TgtPtrBegin), DPxPTR(HstPtrBegin));
-        TIMESCOPE_WITH_DETAILS_AND_IDENT("DevToHost","Size="+
-                                         std::to_string(DataSize)+"B", Loc);
+      TIMESCOPE_WITH_DETAILS_AND_IDENT(
+          "DevToHost", "Size=" + std::to_string(DataSize) + "B", Loc);
       // Wait for any previous transfer if an event is present.
       if (void *Event = TPR.getEntry()->getEvent()) {
         if (Device.waitEvent(Event, AsyncInfo) != OFFLOAD_SUCCESS) {
@@ -1669,14 +1667,12 @@ int target(ident_t *Loc, DeviceTy &Device, void *HostPtr,
 
   {
     assert(KernelArgs.NumArgs == TgtArgs.size() && "Argument count mismatch!");
-    TIMESCOPE_WITH_DETAILS_AND_IDENT("Kernel Target",
-                                     "NumArguments="+
-                                     std::to_string(KernelArgs.NumArgs)+
-                                     ";NumTeams="+
-                                     std::to_string(KernelArgs.NumTeams[0])+
-                                     ";TripCount="+
-                                     std::to_string(KernelArgs.Tripcount)
-                                     , Loc);
+    TIMESCOPE_WITH_DETAILS_AND_IDENT(
+        "Kernel Target",
+        "NumArguments=" + std::to_string(KernelArgs.NumArgs) +
+            ";NumTeams=" + std::to_string(KernelArgs.NumTeams[0]) +
+            ";TripCount=" + std::to_string(KernelArgs.Tripcount),
+        Loc);
 
 #ifdef OMPT_SUPPORT
     assert(KernelArgs.NumTeams[1] == 0 && KernelArgs.NumTeams[2] == 0 &&
diff --git a/openmp/libomptarget/src/private.h b/openmp/libomptarget/src/private.h
index c8d07138b180..8657390dde17 100644
--- a/openmp/libomptarget/src/private.h
+++ b/openmp/libomptarget/src/private.h
@@ -435,7 +435,7 @@ public:
   std::string ProfileName = SI.getName();                                      \
   std::string RTM = RegionTypeMsg;                                             \
   llvm::TimeTraceScope TimeScope(__FUNCTION__, ProfileLocation + RTM)
-  //llvm::TimeTraceScope TimeScope(RegionTypeMsg, ProfileLocation + RTM)
+// llvm::TimeTraceScope TimeScope(RegionTypeMsg, ProfileLocation + RTM)
 #define TIMESCOPE_WITH_DETAILS_AND_IDENT(RegionTypeMsg, Details, IDENT)        \
   SourceInfo SI(IDENT);                                                        \
   std::string ProfileLocation = SI.getProfileLocation();                       \

``````````

</details>


https://github.com/llvm/llvm-project/pull/68016


More information about the flang-commits mailing list