[libomptarget] Refactor shfl_down_sync macro to inline function

[JonChesterfield]

Summary

[libomptarget] Refactor shfl_down_sync macro to inline function
See also abandoned D66846, split into this diff and others.

[ABataev]

LG with a nit

openmp/libomptarget/deviceRTLs/nvptx/src/target_impl.h

54 ↗

(On Diff #217552)

#endif // CUDA_VERSION

openmp/trunk/libomptarget/deviceRTLs/nvptx/src/omptarget-nvptx.h

 #define ORDERED_COUNTER 1
 
 // Macros for Cuda intrinsics
 // In Cuda 9.0, the *_sync() version takes an extra argument 'mask'.
 // Also, __ballot(1) in Cuda 8.0 is replaced with __activemask().
 #ifndef CUDA_VERSION
 #error CUDA_VERSION macro is undefined, something wrong with cuda.
 #elif CUDA_VERSION >= 9000
-#define __SHFL_DOWN_SYNC(mask, var, delta, width)                              \
-  __shfl_down_sync((mask), (var), (delta), (width))
 #define __ACTIVEMASK() __activemask()
 #define __SYNCWARP(Mask) __syncwarp(Mask)
 #else
-#define __SHFL_DOWN_SYNC(mask, var, delta, width)                              \
-  __shfl_down((var), (delta), (width))
 #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");
 // Use original __syncthreads if compiled by nvcc or clang >= 9.0.
 #if !defined(__clang__) || __clang_major__ >= 9

openmp/trunk/libomptarget/deviceRTLs/nvptx/src/reduction.cu

 // This file contains the implementation of reduction with KMPC interface.
 //
 //===----------------------------------------------------------------------===//
 
 #include <complex.h>
 #include <stdio.h>
 
 #include "omptarget-nvptx.h"
+#include "target_impl.h"
 
 EXTERN
 void __kmpc_nvptx_end_reduce(int32_t global_tid) {}
 
 EXTERN
 void __kmpc_nvptx_end_reduce_nowait(int32_t global_tid) {}
 
 EXTERN int32_t __kmpc_shuffle_int32(int32_t val, int16_t delta, int16_t size) {
-  return __SHFL_DOWN_SYNC(0xFFFFFFFF, val, delta, size);
+  return __kmpc_impl_shfl_down_sync(0xFFFFFFFF, val, delta, size);
 }
 
 EXTERN int64_t __kmpc_shuffle_int64(int64_t val, int16_t delta, int16_t size) {
    int lo, hi;
    asm volatile("mov.b64 {%0,%1}, %2;" : "=r"(lo), "=r"(hi) : "l"(val));
-   hi = __SHFL_DOWN_SYNC(0xFFFFFFFF, hi, delta, size);
-   lo = __SHFL_DOWN_SYNC(0xFFFFFFFF, lo, delta, size);
+   hi = __kmpc_impl_shfl_down_sync(0xFFFFFFFF, hi, delta, size);
+   lo = __kmpc_impl_shfl_down_sync(0xFFFFFFFF, lo, delta, size);
    asm volatile("mov.b64 %0, {%1,%2};" : "=l"(val) : "r"(lo), "r"(hi));
    return val;
 }
 
 INLINE static void gpu_regular_warp_reduce(void *reduce_data,
                                            kmp_ShuffleReductFctPtr shflFct) {
   for (uint32_t mask = WARPSIZE / 2; mask > 0; mask /= 2) {
     shflFct(reduce_data, /*LaneId - not used= */ 0,

openmp/trunk/libomptarget/deviceRTLs/nvptx/src/target_impl.h

 
 INLINE int __kmpc_impl_popc(uint32_t x) { return __popc(x); }
 
 #ifndef CUDA_VERSION
 #error CUDA_VERSION macro is undefined, something wrong with cuda.
 #endif
 
 // In Cuda 9.0, the *_sync() version takes an extra argument 'mask'.
+
 INLINE int32_t __kmpc_impl_shfl_sync(__kmpc_impl_lanemask_t Mask, int32_t Var,
                                      int32_t SrcLane) {
 #if CUDA_VERSION >= 9000
   return __shfl_sync(Mask, Var, SrcLane);
 #else
   return __shfl(Var, SrcLane);
 #endif // CUDA_VERSION
+
+INLINE int32_t __kmpc_impl_shfl_down_sync(__kmpc_impl_lanemask_t Mask,
+                                          int32_t Var, uint32_t Delta,
+                                          int32_t Width) {
+#if CUDA_VERSION >= 9000
+  return __shfl_down_sync(Mask, Var, Delta, Width);
+#else
+  return __shfl_down(Var, Delta, Width);
+#endif // CUDA_VERSION
 }
 
 INLINE void __kmpc_impl_syncwarp(int32_t Mask) { __SYNCWARP(Mask); }
 
 #endif