[Openmp-commits] [PATCH] D66857: [libomptarget] Refactor syncwarp macro to inline function

Jon Chesterfield via Phabricator via Openmp-commits openmp-commits at lists.llvm.org
Tue Aug 27 19:02:40 PDT 2019


This revision was automatically updated to reflect the committed changes.
Closed by commit rL370149: [libomptarget] Refactor syncwarp macro to inline function (authored by JonChesterfield, committed by ).
Herald added a project: LLVM.
Herald added a subscriber: llvm-commits.

Changed prior to commit:
  https://reviews.llvm.org/D66857?vs=217555&id=217565#toc

Repository:
  rL LLVM

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D66857/new/

https://reviews.llvm.org/D66857

Files:
  openmp/trunk/libomptarget/deviceRTLs/nvptx/src/omptarget-nvptx.h
  openmp/trunk/libomptarget/deviceRTLs/nvptx/src/supporti.h
  openmp/trunk/libomptarget/deviceRTLs/nvptx/src/target_impl.h


Index: openmp/trunk/libomptarget/deviceRTLs/nvptx/src/omptarget-nvptx.h
===================================================================
--- openmp/trunk/libomptarget/deviceRTLs/nvptx/src/omptarget-nvptx.h
+++ openmp/trunk/libomptarget/deviceRTLs/nvptx/src/omptarget-nvptx.h
@@ -52,11 +52,8 @@
 #error CUDA_VERSION macro is undefined, something wrong with cuda.
 #elif CUDA_VERSION >= 9000
 #define __ACTIVEMASK() __activemask()
-#define __SYNCWARP(Mask) __syncwarp(Mask)
 #else
 #define __ACTIVEMASK() __ballot(1)
-// In Cuda < 9.0 no need to sync threads in warps.
-#define __SYNCWARP(Mask)
 #endif // CUDA_VERSION
 
 #define __SYNCTHREADS_N(n) asm volatile("bar.sync %0;" : : "r"(n) : "memory");
Index: openmp/trunk/libomptarget/deviceRTLs/nvptx/src/target_impl.h
===================================================================
--- openmp/trunk/libomptarget/deviceRTLs/nvptx/src/target_impl.h
+++ openmp/trunk/libomptarget/deviceRTLs/nvptx/src/target_impl.h
@@ -63,6 +63,12 @@
 #endif // CUDA_VERSION
 }
 
-INLINE void __kmpc_impl_syncwarp(int32_t Mask) { __SYNCWARP(Mask); }
+INLINE void __kmpc_impl_syncwarp(__kmpc_impl_lanemask_t Mask) {
+#if CUDA_VERSION >= 9000
+  __syncwarp(Mask);
+#else
+  // In Cuda < 9.0 no need to sync threads in warps.
+#endif // CUDA_VERSION
+}
 
 #endif
Index: openmp/trunk/libomptarget/deviceRTLs/nvptx/src/supporti.h
===================================================================
--- openmp/trunk/libomptarget/deviceRTLs/nvptx/src/supporti.h
+++ openmp/trunk/libomptarget/deviceRTLs/nvptx/src/supporti.h
@@ -14,6 +14,8 @@
 // Execution Parameters
 ////////////////////////////////////////////////////////////////////////////////
 
+#include "target_impl.h"
+
 INLINE void setExecutionParameters(ExecutionMode EMode, RuntimeMode RMode) {
   execution_param = EMode;
   execution_param |= RMode;
@@ -203,7 +205,7 @@
 
 INLINE void IncParallelLevel(bool ActiveParallel) {
   unsigned Active = __ACTIVEMASK();
-  __SYNCWARP(Active);
+  __kmpc_impl_syncwarp(Active);
   unsigned LaneMaskLt;
   asm("mov.u32 %0, %%lanemask_lt;" : "=r"(LaneMaskLt));
   unsigned Rank = __popc(Active & LaneMaskLt);
@@ -212,12 +214,12 @@
         (1 + (ActiveParallel ? OMP_ACTIVE_PARALLEL_LEVEL : 0));
     __threadfence();
   }
-  __SYNCWARP(Active);
+  __kmpc_impl_syncwarp(Active);
 }
 
 INLINE void DecParallelLevel(bool ActiveParallel) {
   unsigned Active = __ACTIVEMASK();
-  __SYNCWARP(Active);
+  __kmpc_impl_syncwarp(Active);
   unsigned LaneMaskLt;
   asm("mov.u32 %0, %%lanemask_lt;" : "=r"(LaneMaskLt));
   unsigned Rank = __popc(Active & LaneMaskLt);
@@ -226,7 +228,7 @@
         (1 + (ActiveParallel ? OMP_ACTIVE_PARALLEL_LEVEL : 0));
     __threadfence();
   }
-  __SYNCWARP(Active);
+  __kmpc_impl_syncwarp(Active);
 }
 
 ////////////////////////////////////////////////////////////////////////////////


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D66857.217565.patch
Type: text/x-patch
Size: 2845 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/openmp-commits/attachments/20190828/c3a6ae78/attachment.bin>


More information about the Openmp-commits mailing list