[OPENMP][NVPTX]Enable fast shuffles on 64bit values only if CUDA >= 9.

[ABataev]

Summary

Shuffle on 64bit data is allowed only for CUDA >= 9.0. Also, fixed the
constant for the mask, need one extra L in the end.

[gtbercea]

LG

libomptarget/deviceRTLs/nvptx/src/reduction.cu

 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);
 }
 
 EXTERN int64_t __kmpc_shuffle_int64(int64_t val, int16_t delta, int16_t size) {
-  return __SHFL_DOWN_SYNC(0xFFFFFFFFFFFFFFFFL, val, delta, size);
+#if defined(CUDART_VERSION) && CUDART_VERSION >= 9000
+  return __SHFL_DOWN_SYNC(0xFFFFFFFFFFFFFFFFLL, (long long)val, (unsigned)delta,
+                          (int)size);
+#else
+   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);
+   asm volatile("mov.b64 %0, {%1,%2};" : "=l"(val) : "r"(lo), "r"(hi));
+   return val;
+#endif
 }
 
 static INLINE 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,
             /*Offset = */ mask, /*AlgoVersion=*/0);
   }