From b84d2296b87ac212474af855d916b12adf96bdb7 Mon Sep 17 00:00:00 2001 From: Cedric Nugteren Date: Sat, 1 Apr 2017 13:36:24 +0200 Subject: Separated host-device and device-host memory copies from execution of the CBLAS reference code; for fair timing and code de-duplication --- test/routines/level2/xtpmv.hpp | 13 +++++-------- 1 file changed, 5 insertions(+), 8 deletions(-) (limited to 'test/routines/level2/xtpmv.hpp') diff --git a/test/routines/level2/xtpmv.hpp b/test/routines/level2/xtpmv.hpp index 3821e1a4..7f4842f0 100644 --- a/test/routines/level2/xtpmv.hpp +++ b/test/routines/level2/xtpmv.hpp @@ -44,6 +44,8 @@ class TestXtpmv { kArgXInc, kArgAPOffset, kArgXOffset}; } + static std::vector BuffersIn() { return {kBufMatAP, kBufVecX}; } + static std::vector BuffersOut() { return {kBufVecX}; } // Describes how to obtain the sizes of the buffers static size_t GetSizeX(const Arguments &args) { @@ -107,19 +109,14 @@ class TestXtpmv { // Describes how to run the CPU BLAS routine (for correctness/performance comparison) #ifdef CLBLAST_REF_CBLAS - static StatusCode RunReference2(const Arguments &args, Buffers &buffers, Queue &queue) { - std::vector ap_mat_cpu(args.ap_size, static_cast(0)); - std::vector x_vec_cpu(args.x_size, static_cast(0)); - buffers.ap_mat.Read(queue, args.ap_size, ap_mat_cpu); - buffers.x_vec.Read(queue, args.x_size, x_vec_cpu); + static StatusCode RunReference2(const Arguments &args, BuffersHost &buffers_host, Queue &) { cblasXtpmv(convertToCBLAS(args.layout), convertToCBLAS(args.triangle), convertToCBLAS(args.a_transpose), convertToCBLAS(args.diagonal), args.n, - ap_mat_cpu, args.ap_offset, - x_vec_cpu, args.x_offset, args.x_inc); - buffers.x_vec.Write(queue, args.x_size, x_vec_cpu); + buffers_host.ap_mat, args.ap_offset, + buffers_host.x_vec, args.x_offset, args.x_inc); return StatusCode::kSuccess; } #endif -- cgit v1.2.3