diff options
author | Cedric Nugteren <web@cedricnugteren.nl> | 2023-05-07 20:03:16 +0200 |
---|---|---|
committer | GitHub <noreply@github.com> | 2023-05-07 20:03:16 +0200 |
commit | 4f24d927302078a416dea1bc29d714a95732b8e9 (patch) | |
tree | ae8b34c8810c080a6d71a91b2fe16939f4d3d49d | |
parent | 3d0c227fa5004067d857c74f7963876b34ed4170 (diff) |
TRMV: Use the minimum x buffer size for copying to a temp buffer (#458)
-rw-r--r-- | CHANGELOG | 1 | ||||
-rw-r--r-- | src/routines/level2/xtrmv.cpp | 5 |
2 files changed, 4 insertions, 2 deletions
@@ -1,4 +1,5 @@ Development version (next version) +- Fixes a minor issue with the expected input buffer size in the TRMV routine - Fixes two small issues in the plotting script - Modifications to improve performance on Qualcomm Adreno GPUs: * Unique database entries for specific Adreno devices diff --git a/src/routines/level2/xtrmv.cpp b/src/routines/level2/xtrmv.cpp index 80e29009..3be176b0 100644 --- a/src/routines/level2/xtrmv.cpp +++ b/src/routines/level2/xtrmv.cpp @@ -36,8 +36,9 @@ void Xtrmv<T>::DoTrmv(const Layout layout, const Triangle triangle, const Buffer<T> &x_buffer, const size_t x_offset, const size_t x_inc) { // Creates a copy of X: a temporary scratch buffer - auto scratch_buffer = Buffer<T>(context_, n*x_inc + x_offset); - x_buffer.CopyTo(queue_, n*x_inc + x_offset, scratch_buffer); + const auto x_size = (1 + (n - 1) * x_inc) + x_offset; + auto scratch_buffer = Buffer<T>(context_, x_size); + x_buffer.CopyTo(queue_, x_size, scratch_buffer); // The data is either in the upper or lower triangle size_t is_upper = ((triangle == Triangle::kUpper && layout != Layout::kRowMajor) || |