diff options
author | Cedric Nugteren <web@cedricnugteren.nl> | 2017-07-12 21:53:39 +0200 |
---|---|---|
committer | Cedric Nugteren <web@cedricnugteren.nl> | 2017-07-12 21:53:39 +0200 |
commit | f77b48692b5dbc9f13f5f93a8242ea546a39236e (patch) | |
tree | 6642aa51003134c550abbd37dad57c7c29c091f6 /src | |
parent | f2477f663672fd37301d6e2ce4646519f71d5cce (diff) |
Relaxed requirement on a_ld and b_ld for batched GEMM
Diffstat (limited to 'src')
-rw-r--r-- | src/routines/levelx/xgemmbatched.cpp | 4 | ||||
-rw-r--r-- | src/utilities/buffer_test.hpp | 8 |
2 files changed, 6 insertions, 6 deletions
diff --git a/src/routines/levelx/xgemmbatched.cpp b/src/routines/levelx/xgemmbatched.cpp index 0fea1922..ee8448d2 100644 --- a/src/routines/levelx/xgemmbatched.cpp +++ b/src/routines/levelx/xgemmbatched.cpp @@ -94,8 +94,8 @@ void XgemmBatched<T>::DoGemmBatched(const Layout layout, const Transpose a_trans // Tests the matrices for validity for (auto batch = size_t{0}; batch < batch_count; ++batch) { - TestMatrixA(a_one, a_two, a_buffer, a_offsets[batch], a_ld); - TestMatrixB(b_one, b_two, b_buffer, b_offsets[batch], b_ld); + TestMatrixA(a_one, a_two, a_buffer, a_offsets[batch], a_ld, false); // don't test for invalid LD + TestMatrixB(b_one, b_two, b_buffer, b_offsets[batch], b_ld, false); // don't test for invalid LD TestMatrixC(c_one, c_two, c_buffer, c_offsets[batch], c_ld); } diff --git a/src/utilities/buffer_test.hpp b/src/utilities/buffer_test.hpp index 652ab8c6..b5693181 100644 --- a/src/utilities/buffer_test.hpp +++ b/src/utilities/buffer_test.hpp @@ -23,8 +23,8 @@ namespace clblast { // Tests matrix 'A' for validity template <typename T> void TestMatrixA(const size_t one, const size_t two, const Buffer<T> &buffer, - const size_t offset, const size_t ld) { - if (ld < one) { throw BLASError(StatusCode::kInvalidLeadDimA); } + const size_t offset, const size_t ld, const bool test_lead_dim = true) { + if (test_lead_dim && ld < one) { throw BLASError(StatusCode::kInvalidLeadDimA); } try { const auto required_size = (ld * (two - 1) + one + offset) * sizeof(T); if (buffer.GetSize() < required_size) { throw BLASError(StatusCode::kInsufficientMemoryA); } @@ -34,8 +34,8 @@ void TestMatrixA(const size_t one, const size_t two, const Buffer<T> &buffer, // Tests matrix 'B' for validity template <typename T> void TestMatrixB(const size_t one, const size_t two, const Buffer<T> &buffer, - const size_t offset, const size_t ld) { - if (ld < one) { throw BLASError(StatusCode::kInvalidLeadDimB); } + const size_t offset, const size_t ld, const bool test_lead_dim = true) { + if (test_lead_dim && ld < one) { throw BLASError(StatusCode::kInvalidLeadDimB); } try { const auto required_size = (ld * (two - 1) + one + offset) * sizeof(T); if (buffer.GetSize() < required_size) { throw BLASError(StatusCode::kInsufficientMemoryB); } |