summaryrefslogtreecommitdiff
path: root/src/clblast.cpp
diff options
context:
space:
mode:
authorCedric Nugteren <web@cedricnugteren.nl>2018-04-13 22:09:16 +0200
committerCedric Nugteren <web@cedricnugteren.nl>2018-04-13 22:27:11 +0200
commit0dff7f1ac43bb7d074db36ce2ce44c52e6760e7e (patch)
treee9c33e6067bd6dbc9fdb8306df5d14d45fab3ad4 /src/clblast.cpp
parent0f49dd24e5307e52d748654aca303f15fa629b36 (diff)
Made GEMM rotation expectations kernel-specific
Diffstat (limited to 'src/clblast.cpp')
-rw-r--r--src/clblast.cpp3
1 files changed, 2 insertions, 1 deletions
diff --git a/src/clblast.cpp b/src/clblast.cpp
index 331a39ef..10bb8cba 100644
--- a/src/clblast.cpp
+++ b/src/clblast.cpp
@@ -2490,7 +2490,8 @@ StatusCode GemmTempBufferSize(const Layout layout, const Transpose a_transpose,
else {
temp_buffer_size = Xgemm<T>::GetTempSize(layout, a_transpose, b_transpose, m, n, k,
a_offset, a_ld, b_offset, b_ld, c_offset, c_ld,
- db["MWG"], db["NWG"], db["KWG"]);
+ db["MWG"], db["NWG"], db["KWG"] * db["KREG"],
+ db["GEMMK"]);
}
temp_buffer_size *= sizeof(T); // translate from num-elements to bytes
return StatusCode::kSuccess;