Fix segmentation fault. (#39)

* Fix segmentation fault.
* Update pyproject.toml

Author: duburcqa

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "gs-madrona"
-version = "0.0.4"
+version = "0.0.5"
 description = "A fork from the official Madrona. The fork will be used for Genesis's own Madrona integration."
 readme = {file = "README.md", content-type = "text/markdown"}
 license = { text = "MIT" }

src/mw/device/sort_archetype.cpp

@@ -168,7 +168,7 @@ struct BlockRadixRankMatchEarlyCountsCustom
                 for (int u = 0; u < WARP_BINS_PER_THREAD; ++u)
                 {
                     int bin = lane + u * WARP_THREADS;
-                    bins[u] = ThreadReduce(warp_histograms[bin], Sum());
+                    bins[u] = internal::ThreadReduce(warp_histograms[bin], Sum());
                 }
                 CTA_SYNC();