AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
2,047 Commits 3 Branches 0 Tags
b738633bcb057465ef22c0aa4b24cc795b341f33
Commit Graph

7 Commits

Author SHA1 Message Date
Alex Voicu b738633bcb This switches HIP from its currently convoluted macro + pfe based dispatch mechanism to a more natural one partially based on the existing module API. The basic idea is that HCC will always correctly emit __global__ functions: as empty-bodied stubs, on host, and as kernels, on device. It then becomes trivial to obtain the mangled name on host, at dispatch, from the function's address, and then to use the mangled name to retrieve the kernel. This should address all problems stemming from serialisation, dubious mismatches due to the manufactured functor, macro-isms et al. It also immediately enables support for generalised globals as a consequence of that being available in the module API. Finally, it will make debug much easier, since the actual names of the __global__ functions will automatically be used in traces etc. One detail is that due to how dispatch works now (hipLaunchKernel and hipLaunchKernelGGL are themselves variadic function templates which deduce the function type of the callee), in certain cases it may be necesssary to insert explicit casts to ensure that the variadic argument list selects a viable overload - this can be observed in some unit tests. Eventually we may be able to remove this limitation, but for now it does not appear terribly onerous. The code is not extremely HIPpie, nor is it fully optimised, but rather is intended as a starting point for the HIP team to make its own. 
[ROCm/clr commit: 28f87f7d2e]
 2017-11-01 15:09:59 +00:00
Ben Sander 6079b4b4f6 Some fixes to tracing. 
[ROCm/clr commit: 9e9f384899]
 2017-07-28 22:13:43 -05:00
Ben Sander c098221a46 Fix HIP_TRACE_API so kernel launch only printed when requested. 
[ROCm/clr commit: fad9104b5f]
 2017-05-16 19:04:25 -05:00
Ben Sander 4ac6ac9d1d Add initial HIP_SYNC_NULL_STREAM=0 mode. 
This eliminates host-synchronization for null stream.  Instead, the
null-stream uses GPU-side events to wait for other streams.
Default is OFF pending additional testing.

Add enhanced null-stream test.

Also refine HIP_TRACE_API.


[ROCm/clr commit: 8bc6ee5932]
 2017-05-16 19:04:25 -05:00
Sun, Peng 9fd6c7179d update GGL to log launched kernel information 
Change-Id: Ied0aa6055673c687071b4a579aecd17f0f3f09ce


[ROCm/clr commit: f699e02715]
 2017-04-07 14:06:31 -05:00
Sun, Peng 72d4fe0213 Add copy right in grid_launch.cpp 
Change-Id: I7de3fc32f13182b5c41a4e44147b642ba15e8636


[ROCm/clr commit: 9762667431]
 2017-04-01 15:12:00 -05:00
Sun, Peng a8708e7bdd Add grid_launch.cpp for GGL 
Change-Id: I87ff9b3f1203d0909f998c96c839f7b321fc3f09


[ROCm/clr commit: ae8448a9af]
 2017-04-01 14:57:47 -05:00