Version 2024.10.29

CHANGELOG.md

@@ -1,7 +1,17 @@
+2024.10.29
+  - Fixed norm to correctly propagate NaN and Inf values.
+  - Fixed matrix generators to 1-based i,j indices, to match Matlab.
+  - Added new matrix generators (minij, hilb, ..., gcdmat).
+  - Require MPI in CMake/Makefile. The non-MPI build was always broken.
+  - Improved GitHub continuous testing.
+  - Refactored norm test code.
+  - Refactored ScaLAPACK wrappers with enums, namespace.
+  - Replaced Fortran steqr2 with C++ steqr.
+
 2024.05.31
-  - Add shared library version (ABI version 1.0.0)
-  - Update enum parameters to have `to_string`, `from_string`;
-    deprecate `<enum>2str`, `str2<enum>`
+  - Added shared library version (ABI version 1.0.0)
+  - Updated enum parameters to have `to_string`, `from_string`;
+    deprecated `<enum>2str`, `str2<enum>`
   - Changed methods to enums; renamed some values and deprecated old values
   - Added "all vectors" case to SVD
   - Fixed SVD for slightly tall case (m > n but not m >> n)
@@ -18,29 +28,29 @@
   - Added internal timers to testers; use `tester --timer-level 2`
 
 2023.11.05
-  - Fix variable block sizes
-  - Fix tau in LQ tester
-  - Update examples for Users Guide
-  - Fix CUDA sync in Frobenius norm
-  - Add random butterfly transform (RBT) solver
-  - Use `blas_int` in scalapack wrappers, towards supporting int64
-  - Fix Cholesky QR test with well-conditioned matrix
-  - Add info check in LU for singular matrix
-  - Fix SVD tester for all vectors
-  - Use multi-threaded Intel MKL to improve eig and svd
-  - Add arbitrary batch regions in `set`
-  - Add timers in `gesv`, `posv`, `gels`, `heev`, `svd`
-  - Improve support for 2D GPU grids and lambda constructors
-  - Fix ROCm complex for ROCm 5.6
-  - Merge Cholesky potrf Host and Device implementations
-  - Remove tile life from QR, LQ, add routines
-  - Fix test matrix generation
-  - Cleanup MOSI, move to Tile class
-  - Add zerocol test matrix variant
-  - Fix receive count
-  - Use GPU-to-GPU copies
-  - Fix `tileMB`, `tileNb`
-  - Improve LU left pivoting for target device
+  - Fixed variable block sizes
+  - Fixed tau in LQ tester
+  - Updated examples for Users Guide
+  - Fixed CUDA sync in Frobenius norm
+  - Added random butterfly transform (RBT) solver
+  - Used `blas_int` in scalapack wrappers, towards supporting int64
+  - Fixed Cholesky QR test with well-conditioned matrix
+  - Added info check in LU for singular matrix
+  - Fixed SVD tester for all vectors
+  - Used multi-threaded Intel MKL to improve eig and svd
+  - Added arbitrary batch regions in `set`
+  - Added timers in `gesv`, `posv`, `gels`, `heev`, `svd`
+  - Improved support for 2D GPU grids and lambda constructors
+  - Fixed ROCm complex for ROCm 5.6
+  - Merged Cholesky potrf Host and Device implementations
+  - Removed tile life from QR, LQ, add routines
+  - Fixed test matrix generation
+  - Cleaned up MOSI, moved to Tile class
+  - Added zerocol test matrix variant
+  - Fixed receive count
+  - Used GPU-to-GPU copies
+  - Fixed `tileMB`, `tileNb`
+  - Improved LU left pivoting for target device
 
 2023.08.25
   - Added oneMKL/SYCL support

CMakeLists.txt

@@ -16,7 +16,7 @@ cmake_minimum_required( VERSION 3.18 )
 
 project(
     slate
-    VERSION 2024.05.31
+    VERSION 2024.10.29
     LANGUAGES CXX Fortran
 )
 

GNUmakefile

@@ -1335,7 +1335,7 @@ LDFLAGS_clean  = ${filter-out -fPIC -L./%, ${LDFLAGS}}
 
 .PHONY: ${pkg}
 ${pkg}:
-	perl -pe "s'#VERSION'2024.05.31'; \
+	perl -pe "s'#VERSION'2024.10.29'; \
 	          s'#PREFIX'${abs_prefix}'; \
 	          s'#CXX\b'${CXX}'; \
 	          s'#CXXFLAGS'${CXXFLAGS_clean}'; \

docs/doxygen/doxyfile.conf

@@ -40,7 +40,7 @@ PROJECT_NAME           = "SLATE"
 # could be handy for archiving the generated documentation or if some version
 # control system is used.
 
-PROJECT_NUMBER         = "2024.05.31"
+PROJECT_NUMBER         = "2024.10.29"
 
 # Using the PROJECT_BRIEF tag one can provide an optional one line description
 # for a project that appears at the top of each page and should give viewer a

include/slate/slate.hh

@@ -26,8 +26,8 @@
 namespace slate {
 
 // Version is updated by make_release.py; DO NOT EDIT.
-// Version 2024.05.31
-#define SLATE_VERSION 20240531
+// Version 2024.10.29
+#define SLATE_VERSION 20241029
 
 int version();
 const char* id();

test/idle_gpus.py

@@ -48,15 +48,16 @@
             if (s):
                 gpu = s.group( 1 )
 
-            # If using >= 10 MiB or 5% utilization, assume it is not idle.
+            # If using >= 50% memory or utilization, mark it as not idle.
+            # This allows some sharing of GPUs with other users.
             # Typically idle is 1 MiB and 0% utilization.
             # Docker can't see processes in section 2.
             s = re.search( '^\| +N/A +\d+C +\w+ +\d+W +/ +\d+W *\| +(\d+)MiB +/ +(\d+)MiB *\| +(\d+)%', line )
             if (s):
                 used_mem  = int( s.group( 1 ) )
                 total_mem = int( s.group( 2 ) )
                 percent   = int( s.group( 3 ) )
-                if (used_mem >= 10 or percent >= 5):
+                if (used_mem >= 0.5*total_mem or percent >= 50):
                     gpus[ gpu ] = 0
         else:
             # Match process lines:

tools/release.py

@@ -219,7 +219,7 @@ def make( project, version_h, version_c ):
 
     # Build Doxygen docs. Create dummy 'make.inc' to avoid 'make config'.
     open( 'make.inc', mode='a' ).close()
-    myrun( 'make docs blas=openblas' )
+    myrun( 'make docs blas=openblas mpi=1' )
     os.unlink( 'make.inc' )
 
     os.chdir( '..' )