diff options
Diffstat (limited to 'src/routines/level3/xsyr2k.cpp')
-rw-r--r-- | src/routines/level3/xsyr2k.cpp | 22 |
1 files changed, 8 insertions, 14 deletions
diff --git a/src/routines/level3/xsyr2k.cpp b/src/routines/level3/xsyr2k.cpp index 424d4d2d..158cd9e5 100644 --- a/src/routines/level3/xsyr2k.cpp +++ b/src/routines/level3/xsyr2k.cpp @@ -31,6 +31,7 @@ Xsyr2k<T>::Xsyr2k(Queue &queue, EventPointer event, const std::string &name): #include "../../kernels/level3/transpose_pad.opencl" #include "../../kernels/level3/xgemm_part1.opencl" #include "../../kernels/level3/xgemm_part2.opencl" + #include "../../kernels/level3/xgemm_part3.opencl" ; } @@ -97,12 +98,6 @@ StatusCode Xsyr2k<T>::DoSyr2k(const Layout layout, const Triangle triangle, cons auto b_temp = (b_no_temp) ? b_buffer : Buffer<T>(context_, k_ceiled*n_ceiled); auto c_temp = Buffer<T>(context_, n_ceiled*n_ceiled); - // Upload the scalar arguments as constant buffers to the device (needed for half-precision) - auto alpha_buffer = Buffer<T>(context_, 1); - auto beta_buffer = Buffer<T>(context_, 1); - alpha_buffer.Write(queue_, 1, &alpha); - beta_buffer.Write(queue_, 1, &beta); - // Events of all kernels (including pre/post processing kernels) auto eventWaitList = std::vector<Event>(); auto emptyEventList = std::vector<Event>(); @@ -112,7 +107,7 @@ StatusCode Xsyr2k<T>::DoSyr2k(const Layout layout, const Triangle triangle, cons // case nothing has to be done, these kernels can be skipped. if (!a_no_temp) { auto eventProcessA = Event(); - status = PadCopyTransposeMatrix(queue_, device_, context_, db_, eventProcessA.pointer(), emptyEventList, + status = PadCopyTransposeMatrix(queue_, device_, db_, eventProcessA.pointer(), emptyEventList, ab_one, ab_two, a_ld, a_offset, a_buffer, n_ceiled, k_ceiled, n_ceiled, 0, a_temp, ConstantOne<T>(), program, @@ -122,7 +117,7 @@ StatusCode Xsyr2k<T>::DoSyr2k(const Layout layout, const Triangle triangle, cons } if (!b_no_temp) { auto eventProcessB = Event(); - status = PadCopyTransposeMatrix(queue_, device_, context_, db_, eventProcessB.pointer(), emptyEventList, + status = PadCopyTransposeMatrix(queue_, device_, db_, eventProcessB.pointer(), emptyEventList, ab_one, ab_two, b_ld, b_offset, b_buffer, n_ceiled, k_ceiled, n_ceiled, 0, b_temp, ConstantOne<T>(), program, @@ -134,7 +129,7 @@ StatusCode Xsyr2k<T>::DoSyr2k(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(queue_, device_, context_, db_, eventProcessC.pointer(), emptyEventList, + status = PadCopyTransposeMatrix(queue_, device_, db_, eventProcessC.pointer(), emptyEventList, n, n, c_ld, c_offset, c_buffer, n_ceiled, n_ceiled, n_ceiled, 0, c_temp, ConstantOne<T>(), program, @@ -149,8 +144,8 @@ StatusCode Xsyr2k<T>::DoSyr2k(const Layout layout, const Triangle triangle, cons // Sets the kernel arguments kernel.SetArgument(0, static_cast<int>(n_ceiled)); kernel.SetArgument(1, static_cast<int>(k_ceiled)); - kernel.SetArgument(2, alpha_buffer()); - kernel.SetArgument(3, beta_buffer()); + kernel.SetArgument(2, GetRealArg(alpha)); + kernel.SetArgument(3, GetRealArg(beta)); kernel.SetArgument(4, a_temp()); kernel.SetArgument(5, b_temp()); kernel.SetArgument(6, c_temp()); @@ -170,8 +165,7 @@ StatusCode Xsyr2k<T>::DoSyr2k(const Layout layout, const Triangle triangle, cons // Swaps the arguments for matrices A and B, and sets 'beta' to 1 auto one = static_cast<T>(1); - beta_buffer.Write(queue_, 1, &one); - kernel.SetArgument(3, beta_buffer()); + kernel.SetArgument(3, GetRealArg(one)); kernel.SetArgument(4, b_temp()); kernel.SetArgument(5, a_temp()); @@ -184,7 +178,7 @@ StatusCode Xsyr2k<T>::DoSyr2k(const Layout layout, const Triangle triangle, cons // Runs the post-processing kernel auto upper = (triangle == Triangle::kUpper); auto lower = (triangle == Triangle::kLower); - status = PadCopyTransposeMatrix(queue_, device_, context_, db_, event_, eventWaitList, + status = PadCopyTransposeMatrix(queue_, device_, db_, event_, eventWaitList, n_ceiled, n_ceiled, n_ceiled, 0, c_temp, n, n, c_ld, c_offset, c_buffer, ConstantOne<T>(), program, |