Fix coomv bug (#446) (#273)

* fix coomv where under certain nnz sizes we were calculating the wrong number of loops * updated changelog and version bump Co-authored-by: jsandham <[email protected]> Co-authored-by: Nico Trost <[email protected]> Co-authored-by: jsandham <[email protected]> Co-authored-by: Nico Trost <[email protected]>
ROCm · Jan 20, 2023 · 57ff53b · 57ff53b
1 parent 72e3bd2
commit 57ff53b
Show file tree

Hide file tree

Showing 3 changed files with 3 additions and 2 deletions.
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -15,6 +15,7 @@ Full documentation for rocSPARSE is available at [rocsparse.readthedocs.io](http
 - Fixed bug in ellmv
 - Optimized bsr2csr routine
 - Fixed integer overflow bugs
+- Fixes a bug in COO SpMV gridsize
 
 ## rocSPARSE 2.3.2 for ROCm 5.3.0
 ### Added

diff --git a/library/src/level2/rocsparse_coomv.cpp b/library/src/level2/rocsparse_coomv.cpp
@@ -424,7 +424,7 @@ rocsparse_status rocsparse_coomv_segmented_dispatch(rocsparse_handle          ha
 
         I minblocks = (nnz - 1) / COOMVN_DIM + 1;
         I nblocks   = maxblocks < minblocks ? maxblocks : minblocks;
-        I nloops    = (nnz / COOMVN_DIM + 1) / nblocks + 1;
+        I nloops    = (nnz - 1) / (COOMVN_DIM * nblocks) + 1;
 
         // Buffer
         char* ptr = reinterpret_cast<char*>(handle->buffer);

diff --git a/library/src/level2/rocsparse_coomv_aos.cpp b/library/src/level2/rocsparse_coomv_aos.cpp
@@ -272,7 +272,7 @@ rocsparse_status rocsparse_coomv_aos_segmented_dispatch(rocsparse_handle
 
         I minblocks = (nnz - 1) / COOMVN_DIM + 1;
         I nblocks   = maxblocks < minblocks ? maxblocks : minblocks;
-        I nloops    = (nnz / COOMVN_DIM + 1) / nblocks + 1;
+        I nloops    = (nnz - 1) / (COOMVN_DIM * nblocks) + 1;
 
         // Buffer
         char* ptr = reinterpret_cast<char*>(handle->buffer);