diff --git a/CUDA/GB_cuda_common_jitFactory.hpp b/CUDA/GB_cuda_common_jitFactory.hpp
index 896dc1609f..a4dbf7f421 100644
--- a/CUDA/GB_cuda_common_jitFactory.hpp
+++ b/CUDA/GB_cuda_common_jitFactory.hpp
@@ -43,6 +43,7 @@ static const std::vector<std::string> compiler_flags{
    "-I.",
    "-I..",
    "-I../templates",
+   "-I../../Source/Shared",
 
    // Add includes relative to GRAPHBLAS_SOURCE_PATH variable
    "-I" + jit::get_user_graphblas_source_path() + "/CUDA",
diff --git a/CUDA/GB_hash.h b/CUDA/GB_hash.h
deleted file mode 120000
index 390a314548..0000000000
--- a/CUDA/GB_hash.h
+++ /dev/null
@@ -1 +0,0 @@
-../Source/Shared/GB_hash.h
\ No newline at end of file
diff --git a/CUDA/GB_hash.h b/CUDA/GB_hash.h
new file mode 100644
index 0000000000..5e46e721af
--- /dev/null
+++ b/CUDA/GB_hash.h
@@ -0,0 +1,48 @@
+//------------------------------------------------------------------------------
+// GB_hash.h: definitions for hashing
+//------------------------------------------------------------------------------
+
+// SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2022, All Rights Reserved.
+// SPDX-License-Identifier: Apache-2.0
+
+//------------------------------------------------------------------------------
+
+#ifndef GB_HASH_H
+#define GB_HASH_H
+
+//------------------------------------------------------------------------------
+// functions for the Hash method for C=A*B, and for the A->Y hyperhash
+//------------------------------------------------------------------------------
+
+// initial hash function, for where to place the integer i in the hash table.
+// hash_bits is a bit mask to compute the result modulo the hash table size,
+// which is always a power of 2.  The function is (i*257) & (hash_bits).
+#define GB_HASHF(i,hash_bits) ((((i) << 8) + (i)) & (hash_bits))
+
+// #define GB_HASHF2(i,hash_bits) ((i) & (hash_bits))
+// lots of intentional collisions:
+// #define GB_HASHF2(i,hash_bits) ((i >> 2) & (hash_bits))
+
+// lots of intentional collisions: but blocks are scattered
+   #define GB_HASHF2(i,hash_bits) ((((i) >> 2) + 17L*((i) >> 8)) & (hash_bits))
+
+// rehash function, for subsequent hash lookups if the initial hash function
+// refers to a hash entry that is already occupied.  Linear probing is used,
+// so the function does not currently depend on i.  On input, hash is equal
+// to the current value of the hash function, and on output, hash is set to
+// the new hash value.
+#define GB_REHASH(hash,i,hash_bits) hash = ((hash + 1) & (hash_bits))
+
+// The hash functions and their parameters are modified from this paper:
+
+// [2] Yusuke Nagasaka, Satoshi Matsuoka, Ariful Azad, and Aydin Buluc. 2018.
+// High-Performance Sparse Matrix-Matrix Products on Intel KNL and Multicore
+// Architectures. In Proc. 47th Intl. Conf. on Parallel Processing (ICPP '18).
+// Association for Computing Machinery, New York, NY, USA, Article 34, 1–10.
+// DOI:https://doi.org/10.1145/3229710.3229720
+
+// The hash function in that paper is (i*107)&(hash_bits).  Here, the term
+// 107 is replaced with 257 to allow for a faster hash function computation.
+
+#endif
+
diff --git a/CUDA/TODO.txt b/CUDA/TODO.txt
index b4bb666ca1..75babdcf12 100644
--- a/CUDA/TODO.txt
+++ b/CUDA/TODO.txt
@@ -8,3 +8,4 @@ test complex and user-defined ops (wildtype) on the GPU
 reduce: do any monoid
     terminal condition?
 
+GxB_VOID
diff --git a/Makefile b/Makefile
index 1b3177e2e6..7b1272db8b 100644
--- a/Makefile
+++ b/Makefile
@@ -38,6 +38,10 @@ cuda:
 local:
 	( cd build && cmake $(CMAKE_OPTIONS) -DSUITESPARSE_LOCAL=1 .. && $(MAKE) --jobs=${JOBS} )
 
+# install in SuiteSparse/lib and SuiteSparse/include and /usr/local
+global:
+	( cd build && cmake $(CMAKE_OPTIONS) -DSUITESPARSE_LOCAL=0 .. && $(MAKE) --jobs=${JOBS} )
+
 # compile with -g 
 debug:
 	( cd build && cmake -DCMAKE_BUILD_TYPE=Debug $(CMAKE_OPTIONS) .. && $(MAKE) --jobs=$(JOBS) )