diff options
author | Cedric Nugteren <web@cedricnugteren.nl> | 2016-06-17 12:32:06 +0200 |
---|---|---|
committer | Cedric Nugteren <web@cedricnugteren.nl> | 2016-06-17 12:32:06 +0200 |
commit | 98a95c89fc0633efdc8439c942762bef9a1e5e1d (patch) | |
tree | d37775c4bf07229f7eae38c335da85eaf0c468a9 /src/routines/level3/xherk.cc | |
parent | 520e28e7a72f288f04d04d86d4e7560d78159820 (diff) |
Moved the RunKernel and PadCopyTransposeMatrix functions out of the Routine class
Diffstat (limited to 'src/routines/level3/xherk.cc')
-rw-r--r-- | src/routines/level3/xherk.cc | 10 |
1 files changed, 5 insertions, 5 deletions
diff --git a/src/routines/level3/xherk.cc b/src/routines/level3/xherk.cc index 9d64af95..8ebcbfa8 100644 --- a/src/routines/level3/xherk.cc +++ b/src/routines/level3/xherk.cc @@ -121,7 +121,7 @@ StatusCode Xherk<T,U>::DoHerk(const Layout layout, const Triangle triangle, cons // case nothing has to be done, these kernels can be skipped. Two copies are created. if (!a_no_temp) { auto eventProcessA = Event(); - status = PadCopyTransposeMatrix(eventProcessA.pointer(), emptyEventList, + status = PadCopyTransposeMatrix(queue_, device_, context_, db_, eventProcessA.pointer(), emptyEventList, a_one, a_two, a_ld, a_offset, a_buffer, n_ceiled, k_ceiled, n_ceiled, 0, a_temp, ConstantOne<T>(), program, @@ -131,7 +131,7 @@ StatusCode Xherk<T,U>::DoHerk(const Layout layout, const Triangle triangle, cons } if (!b_no_temp) { auto eventProcessB = Event(); - status = PadCopyTransposeMatrix(eventProcessB.pointer(), emptyEventList, + status = PadCopyTransposeMatrix(queue_, device_, context_, db_, eventProcessB.pointer(), emptyEventList, a_one, a_two, a_ld, a_offset, a_buffer, n_ceiled, k_ceiled, n_ceiled, 0, b_temp, ConstantOne<T>(), program, @@ -143,7 +143,7 @@ StatusCode Xherk<T,U>::DoHerk(const Layout layout, const Triangle triangle, cons // Furthermore, also creates a (possibly padded) copy of matrix C, since it is not allowed to // modify the other triangle. auto eventProcessC = Event(); - status = PadCopyTransposeMatrix(eventProcessC.pointer(), emptyEventList, + status = PadCopyTransposeMatrix(queue_, device_, context_, db_, eventProcessC.pointer(), emptyEventList, n, n, c_ld, c_offset, c_buffer, n_ceiled, n_ceiled, n_ceiled, 0, c_temp, ConstantOne<T>(), program, @@ -173,14 +173,14 @@ StatusCode Xherk<T,U>::DoHerk(const Layout layout, const Triangle triangle, cons // Launches the kernel auto eventKernel = Event(); - status = RunKernel(kernel, global, local, eventKernel.pointer(), eventWaitList); + status = RunKernel(kernel, queue_, device_, global, local, eventKernel.pointer(), eventWaitList); if (ErrorIn(status)) { return status; } eventWaitList.push_back(eventKernel); // Runs the post-processing kernel auto upper = (triangle == Triangle::kUpper); auto lower = (triangle == Triangle::kLower); - status = PadCopyTransposeMatrix(event_, eventWaitList, + status = PadCopyTransposeMatrix(queue_, device_, context_, db_, event_, eventWaitList, n_ceiled, n_ceiled, n_ceiled, 0, c_temp, n, n, c_ld, c_offset, c_buffer, ConstantOne<T>(), program, |