From 4f24d927302078a416dea1bc29d714a95732b8e9 Mon Sep 17 00:00:00 2001 From: Cedric Nugteren Date: Sun, 7 May 2023 20:03:16 +0200 Subject: TRMV: Use the minimum x buffer size for copying to a temp buffer (#458) --- CHANGELOG | 1 + src/routines/level2/xtrmv.cpp | 5 +++-- 2 files changed, 4 insertions(+), 2 deletions(-) diff --git a/CHANGELOG b/CHANGELOG index 3b84ffc7..39813f94 100644 --- a/CHANGELOG +++ b/CHANGELOG @@ -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::DoTrmv(const Layout layout, const Triangle triangle, const Buffer &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(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(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) || -- cgit v1.2.3