Properly copy the entire vector in the int16_t case (#845)

Fixes #842 #843 #844
marian-nmt · Mar 23, 2021 · ffd997e · ffd997e
1 parent b36d0bb
commit ffd997e
Show file tree

Hide file tree

Showing 3 changed files with 3 additions and 2 deletions.
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -20,6 +20,7 @@ and this project adheres to [Semantic Versioning](http://semver.org/spec/v2.0.0.
 - Fix compilation with OMP
 
 ### Fixed
+- Fixed an issue when loading intgemm16 models from unaligned memory.
 - Fix building marian with gcc 9.3+ and FBGEMM
 - Find MKL installed under Ubuntu 20.04 via apt-get
 - Support for CUDA 11.

diff --git a/VERSION b/VERSION
@@ -1 +1 @@
-v1.10.12
+v1.10.13
diff --git a/src/tensors/cpu/integer_common.h b/src/tensors/cpu/integer_common.h
@@ -198,7 +198,7 @@ void prepareAndTransposeB(io::Item& item, const char * input) {
                                                    cols(item.shape)); //rows here returns the columns of the transposed input matrix, and cols -> the rows
     } else {
         Integer * aligned_input = reinterpret_cast<Integer *>(genericMalloc(512, rows(item.shape)*cols(item.shape)*sizeof(Integer)));
-        std::copy(input, input + rows(item.shape)*cols(item.shape), aligned_input);
+        std::copy(reinterpret_cast<const Integer *>(input), reinterpret_cast<const Integer *>(input) + rows(item.shape)*cols(item.shape), aligned_input);
         Integer * aligned_output = reinterpret_cast<Integer *>(genericMalloc(512, rows(item.shape)*cols(item.shape)*sizeof(Integer)));
         intgemm_<vtype>::width::PrepareBQuantizedTransposed(reinterpret_cast<const Integer *>(aligned_input),
                                                    reinterpret_cast<Integer *>(aligned_output),