AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Initial support for clique-based kernels (#276)

Browse Source 
* Initial support for clique-based kernels
This commit is contained in:
gilbertlee-amd
2020-10-13 11:22:04 -06:00
committed by GitHub
parent ae008fd2db
commit 2b8184808d
27 changed files with 1969 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/collectives/device/common_kernel.h
+5
View File
@@ -350,9 +350,14 @@ __device__ int ptrAlign128(T* ptr) { return (uint64_t)ptr % alignof(int32_t); }
__device__ int ptrAlign128(T* ptr) { return (uint64_t)ptr % alignof(Pack128); }
#endif
#if defined(__HIP_PLATFORM_HCC__) || defined(__HCC__) || defined(__HIPCC__)
// Multiply UNROLL by 2 if single source/single destination
#define AUTOUNROLL (UNROLL*((MINSRCS==1 && MINDSTS==1) ? 2 : 1))
#else
// Try to limit consecutive load/stores to 8.
// Use UNROLL 8 when we have a single source and a single destination, 4 otherwise
#define AUTOUNROLL (UNROLL*(4/(MINDSTS+MINSRCS)))
#endif
template<int UNROLL, class FUNC, typename T, int MINSRCS, int MAXSRCS, int MINDSTS, int MAXDSTS>
__device__ void ReduceOrCopyMulti(const int tid, const int nthreads,