Initial build script changes (untested on PyPi).

4 files changed, 216 insertions, 251 deletions

diff --git a/Makefile b/Makefile
index 195009f..04a6fa7 100644
--- a/Makefile
+++ b/Makefile
@@ -1,10 +1,16 @@
 MKFILE_PATH := $(abspath $(lastword $(MAKEFILE_LIST)))
 ROOT_DIR := $(patsubst %/,%,$(dir $(MKFILE_PATH)))
 
+ifndef CUDA_VERSION
+$(error ERROR: CUDA_VERSION not set. Call make with CUDA string, for example: make cuda11x CUDA_VERSION=115 or make cpuonly CUDA_VERSION=CPU)
+endif
+
 GPP:= /usr/bin/g++
 ifeq ($(CUDA_HOME),)
 	CUDA_HOME:= $(shell which nvcc | rev | cut -d'/' -f3- | rev)
 endif
+
+
 NVCC := $(CUDA_HOME)/bin/nvcc
 
 ###########################################
@@ -53,44 +59,46 @@ CC_cublasLt111 += -gencode arch=compute_86,code=sm_86
 all: $(ROOT_DIR)/dependencies/cub $(BUILD_DIR) env
 	$(NVCC) $(COMPUTE_CAPABILITY) -Xcompiler '-fPIC' --use_fast_math -Xptxas=-v -dc $(FILES_CUDA) $(INCLUDE) $(LIB) --output-directory $(BUILD_DIR)
 	$(NVCC) $(COMPUTE_CAPABILITY) -Xcompiler '-fPIC' -dlink $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o -o $(BUILD_DIR)/link.o 
-	$(GPP) -std=c++14 -DBUILD_CUDA -shared -fPIC $(INCLUDE) $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o $(BUILD_DIR)/link.o $(FILES_CPP) -o ./bitsandbytes/libbitsandbytes.so $(LIB)
+	$(GPP) -std=c++14 -DBUILD_CUDA -shared -fPIC $(INCLUDE) $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o $(BUILD_DIR)/link.o $(FILES_CPP) -o ./bitsandbytes/libbitsandbytes_cuda$(CUDA_VERSION).so $(LIB)
 
 cuda92: $(ROOT_DIR)/dependencies/cub $(BUILD_DIR) env
 	$(NVCC) $(COMPUTE_CAPABILITY) $(CC_CUDA92) -Xcompiler '-fPIC' --use_fast_math -Xptxas=-v -dc $(FILES_CUDA) $(INCLUDE) $(LIB) --output-directory $(BUILD_DIR) -D NO_CUBLASLT
 	$(NVCC) $(COMPUTE_CAPABILITY) $(CC_CUDA92) -Xcompiler '-fPIC' -dlink $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o -o $(BUILD_DIR)/link.o 
-	$(GPP) -std=c++14 -DBUILD_CUDA -shared -fPIC $(INCLUDE) $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o $(BUILD_DIR)/link.o $(FILES_CPP) -o ./bitsandbytes/libbitsandbytes.so $(LIB)
+	$(GPP) -std=c++14 -DBUILD_CUDA -shared -fPIC $(INCLUDE) $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o $(BUILD_DIR)/link.o $(FILES_CPP) -o ./bitsandbytes/libbitsandbytes_cuda$(CUDA_VERSION)_nocublaslt.so $(LIB)
 
 cuda10x_nomatmul: $(ROOT_DIR)/dependencies/cub $(BUILD_DIR) env
 	$(NVCC) $(COMPUTE_CAPABILITY) $(CC_CUDA10x) -Xcompiler '-fPIC' --use_fast_math -Xptxas=-v -dc $(FILES_CUDA) $(INCLUDE) $(LIB) --output-directory $(BUILD_DIR) -D NO_CUBLASLT
 	$(NVCC) $(COMPUTE_CAPABILITY) $(CC_CUDA10x) -Xcompiler '-fPIC' -dlink $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o -o $(BUILD_DIR)/link.o 
-	$(GPP) -std=c++14 -DBUILD_CUDA -shared -fPIC $(INCLUDE) $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o $(BUILD_DIR)/link.o $(FILES_CPP) -o ./bitsandbytes/libbitsandbytes.so $(LIB)
+	$(GPP) -std=c++14 -DBUILD_CUDA -shared -fPIC $(INCLUDE) $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o $(BUILD_DIR)/link.o $(FILES_CPP) -o ./bitsandbytes/libbitsandbytes_cuda$(CUDA_VERSION)_nocublaslt.so $(LIB)
 
 cuda110_nomatmul: $(BUILD_DIR) env
 	$(NVCC) $(COMPUTE_CAPABILITY) $(CC_CUDA110) -Xcompiler '-fPIC' --use_fast_math -Xptxas=-v -dc $(FILES_CUDA) $(INCLUDE) $(LIB) --output-directory $(BUILD_DIR) -D NO_CUBLASLT
 	$(NVCC) $(COMPUTE_CAPABILITY) $(CC_CUDA110) -Xcompiler '-fPIC' -dlink $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o -o $(BUILD_DIR)/link.o 
-	$(GPP) -std=c++14 -DBUILD_CUDA -shared -fPIC $(INCLUDE) $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o $(BUILD_DIR)/link.o $(FILES_CPP) -o ./bitsandbytes/libbitsandbytes.so $(LIB)
+	$(GPP) -std=c++14 -DBUILD_CUDA -shared -fPIC $(INCLUDE) $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o $(BUILD_DIR)/link.o $(FILES_CPP) -o ./bitsandbytes/libbitsandbytes_cuda$(CUDA_VERSION)_nocublaslt.so $(LIB)
 
 cuda11x_nomatmul: $(BUILD_DIR) env
 	$(NVCC) $(COMPUTE_CAPABILITY) $(CC_CUDA11x) -Xcompiler '-fPIC' --use_fast_math -Xptxas=-v -dc $(FILES_CUDA) $(INCLUDE) $(LIB) --output-directory $(BUILD_DIR) -D NO_CUBLASLT
 	$(NVCC) $(COMPUTE_CAPABILITY) $(CC_CUDA11x) -Xcompiler '-fPIC' -dlink $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o -o $(BUILD_DIR)/link.o 
-	$(GPP) -std=c++14 -DBUILD_CUDA -shared -fPIC $(INCLUDE) $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o $(BUILD_DIR)/link.o $(FILES_CPP) -o ./bitsandbytes/libbitsandbytes.so $(LIB)
+	$(GPP) -std=c++14 -DBUILD_CUDA -shared -fPIC $(INCLUDE) $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o $(BUILD_DIR)/link.o $(FILES_CPP) -o ./bitsandbytes/libbitsandbytes_cuda$(CUDA_VERSION)_nocublaslt.so $(LIB)
 
 cuda110: $(BUILD_DIR) env
 	$(NVCC) $(CC_cublasLt110) -Xcompiler '-fPIC' --use_fast_math -Xptxas=-v -dc $(FILES_CUDA) $(INCLUDE) $(LIB) --output-directory $(BUILD_DIR)
 	$(NVCC) $(CC_cublasLt110) -Xcompiler '-fPIC' -dlink $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o -o $(BUILD_DIR)/link.o 
-	$(GPP) -std=c++14 -DBUILD_CUDA -shared -fPIC $(INCLUDE) $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o $(BUILD_DIR)/link.o $(FILES_CPP) -o ./bitsandbytes/libbitsandbytes.so $(LIB)
+	$(GPP) -std=c++14 -DBUILD_CUDA -shared -fPIC $(INCLUDE) $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o $(BUILD_DIR)/link.o $(FILES_CPP) -o ./bitsandbytes/libbitsandbytes_cuda$(CUDA_VERSION).so $(LIB)
 
 cuda11x: $(BUILD_DIR) env
 	$(NVCC) $(CC_cublasLt111) -Xcompiler '-fPIC' --use_fast_math -Xptxas=-v -dc $(FILES_CUDA) $(INCLUDE) $(LIB) --output-directory $(BUILD_DIR)
 	$(NVCC) $(CC_cublasLt111) -Xcompiler '-fPIC' -dlink $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o -o $(BUILD_DIR)/link.o 
-	$(GPP) -std=c++14 -DBUILD_CUDA -shared -fPIC $(INCLUDE) $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o $(BUILD_DIR)/link.o $(FILES_CPP) -o ./bitsandbytes/libbitsandbytes.so $(LIB)
+	$(GPP) -std=c++14 -DBUILD_CUDA -shared -fPIC $(INCLUDE) $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o $(BUILD_DIR)/link.o $(FILES_CPP) -o ./bitsandbytes/libbitsandbytes_cuda$(CUDA_VERSION).so $(LIB)
 
 cpuonly: $(BUILD_DIR) env
-	$(GPP) -std=c++14 -shared -fPIC -I $(ROOT_DIR)/csrc -I $(ROOT_DIR)/include $(FILES_CPP) -o ./bitsandbytes/libbitsandbytes.so
+	$(GPP) -std=c++14 -shared -fPIC -I $(ROOT_DIR)/csrc -I $(ROOT_DIR)/include $(FILES_CPP) -o ./bitsandbytes/libbitsandbytes_$(CUDA_VERSION).so
 
 env:
 	@echo "ENVIRONMENT"
 	@echo "============================"
+	@echo "CUDA_VERSION: $(CUDA_VERSION)"
+	@echo "============================"
 	@echo "NVCC path: $(NVCC)"
 	@echo "GPP path: $(GPP) VERSION: `$(GPP) --version | head -n 1`"
 	@echo "CUDA_HOME: $(CUDA_HOME)"
@@ -108,7 +116,7 @@ $(ROOT_DIR)/dependencies/cub:
 	cd dependencies/cub; git checkout 1.11.0
 
 clean:
-	rm cuda_build/* ./bitsandbytes/libbitsandbytes.so
+	rm build/* ./bitsandbytes/libbitsandbytes.so
 
 cleaneggs:
 	rm -rf *.egg*
diff --git a/bitsandbytes/cuda_setup.py b/bitsandbytes/cuda_setup.py
index 6f67275..08f3ed2 100644
--- a/bitsandbytes/cuda_setup.py
+++ b/bitsandbytes/cuda_setup.py
@@ -133,7 +133,7 @@ def evaluate_cuda_setup():
 
     has_gpu = cc != ''
     if not has_gpu:
-        print('WARNING: No GPU detected! Check our CUDA paths. Processding to load CPU-only library...')
+        print('WARNING: No GPU detected! Check our CUDA paths. Processing to load CPU-only library...')
         return binary_name
 
     has_cublaslt = cc in ['7.5', '8.0', '8.6']
diff --git a/deploy_from_slurm.sh b/deploy_from_slurm.sh
index 37311bc..4aab04f 100644
--- a/deploy_from_slurm.sh
+++ b/deploy_from_slurm.sh
@@ -14,256 +14,204 @@ module unload cuda
 module unload gcc
 
 rm -rf dist build
-make clean
-make cleaneggs
-export CUDA_HOME=
-make cpuonly
-
-if [ ! -f "./bitsandbytes/libbitsandbytes.so" ]; then
-  # Control will enter here if $DIRECTORY doesn't exist.
-  echo "Compilation unsuccessul!" 1>&2
-  exit 64
-fi
-CUDA_VERSION=cpu python -m build
-python -m twine upload dist/* --verbose --repository testpypi
-
-rm -rf dist build
-make clean
-make cleaneggs
-export CUDA_HOME=$BASE_PATH/cuda-11.0
-make cuda110 
-
-if [ ! -f "./bitsandbytes/libbitsandbytes.so" ]; then
-  # Control will enter here if $DIRECTORY doesn't exist.
-  echo "Compilation unsuccessul!" 1>&2
-  exit 64
-fi
-CUDA_VERSION=110 python -m build
-python -m twine upload dist/* --verbose --repository testpypi
-
-rm -rf dist build
-make clean
-make cleaneggs
-export CUDA_HOME=$BASE_PATH/cuda-11.1
-make cuda11x
-
-if [ ! -f "./bitsandbytes/libbitsandbytes.so" ]; then
-  # Control will enter here if $DIRECTORY doesn't exist.
-  echo "Compilation unsuccessul!" 1>&2
-  exit 64
-fi
-CUDA_VERSION=111 python -m build
-python -m twine upload dist/* --verbose --repository testpypi
-
-rm -rf dist build
-make clean
 make cleaneggs
-export CUDA_HOME=$BASE_PATH/cuda-11.2
-make cuda11x
 
-if [ ! -f "./bitsandbytes/libbitsandbytes.so" ]; then
-  # Control will enter here if $DIRECTORY doesn't exist.
-  echo "Compilation unsuccessul!" 1>&2
-  exit 64
-fi
-CUDA_VERSION=112 python -m build
-python -m twine upload dist/* --verbose --repository testpypi
-
-rm -rf dist build
 make clean
-make cleaneggs
-export CUDA_HOME=$BASE_PATH/cuda-11.3
-make cuda11x
-
-if [ ! -f "./bitsandbytes/libbitsandbytes.so" ]; then
-  # Control will enter here if $DIRECTORY doesn't exist.
-  echo "Compilation unsuccessul!" 1>&2
-  exit 64
-fi
-CUDA_VERSION=113 python -m build
-python -m twine upload dist/* --verbose --repository testpypi
-
-rm -rf dist build
-make clean
-make cleaneggs
-export CUDA_HOME=$BASE_PATH/cuda-11.4
-make cuda11x
-
-if [ ! -f "./bitsandbytes/libbitsandbytes.so" ]; then
-  # Control will enter here if $DIRECTORY doesn't exist.
-  echo "Compilation unsuccessul!" 1>&2
-  exit 64
-fi
-CUDA_VERSION=114 python -m build
-python -m twine upload dist/* --verbose --repository testpypi
-
-rm -rf dist build
-make clean
-make cleaneggs
-export CUDA_HOME=$BASE_PATH/cuda-11.5
-make cuda11x
-
-if [ ! -f "./bitsandbytes/libbitsandbytes.so" ]; then
-  # Control will enter here if $DIRECTORY doesn't exist.
-  echo "Compilation unsuccessul!" 1>&2
-  exit 64
-fi
-CUDA_VERSION=115 python -m build
-python -m twine upload dist/* --verbose --repository testpypi
-
-rm -rf dist build
-make clean
-make cleaneggs
-export CUDA_HOME=$BASE_PATH/cuda-11.6
-
-make cuda11x
-if [ ! -f "./bitsandbytes/libbitsandbytes.so" ]; then
-  # Control will enter here if $DIRECTORY doesn't exist.
-  echo "Compilation unsuccessul!" 1>&2
-  exit 64
-fi
-CUDA_VERSION=116 python -m build
-python -m twine upload dist/* --verbose --repository testpypi
-
-rm -rf dist build
-make clean
-make cleaneggs
-export CUDA_HOME=$BASE_PATH/cuda-11.7
-make cuda11x
-
-if [ ! -f "./bitsandbytes/libbitsandbytes.so" ]; then
-  # Control will enter here if $DIRECTORY doesn't exist.
-  echo "Compilation unsuccessul!" 1>&2
-  exit 64
-fi
-CUDA_VERSION=117 python -m build
-python -m twine upload dist/* --verbose --repository testpypi
-
-
-rm -rf dist build
-make clean
-make cleaneggs
-export CUDA_HOME=$BASE_PATH/cuda-10.2
-make cuda10x_nomatmul
+export CUDA_HOME=
+make cpuonly CUDA_VERSION=CPU
 
 if [ ! -f "./bitsandbytes/libbitsandbytes.so" ]; then
   # Control will enter here if $DIRECTORY doesn't exist.
   echo "Compilation unsuccessul!" 1>&2
   exit 64
 fi
-CUDA_VERSION=102-nomatmul python -m build
-python -m twine upload dist/* --verbose --repository testpypi
-
 
-rm -rf dist build
 make clean
-make cleaneggs
 export CUDA_HOME=$BASE_PATH/cuda-11.0
-make cuda110_nomatmul
-
-if [ ! -f "./bitsandbytes/libbitsandbytes.so" ]; then
-  # Control will enter here if $DIRECTORY doesn't exist.
-  echo "Compilation unsuccessul!" 1>&2
-  exit 64
-fi
-CUDA_VERSION=110-nomatmul python -m build
-python -m twine upload dist/* --verbose --repository testpypi
-
-
-rm -rf dist build
-make clean
-make cleaneggs
-export CUDA_HOME=$BASE_PATH/cuda-11.1
-make cuda11x_nomatmul
-
-if [ ! -f "./bitsandbytes/libbitsandbytes.so" ]; then
-  # Control will enter here if $DIRECTORY doesn't exist.
-  echo "Compilation unsuccessul!" 1>&2
-  exit 64
-fi
-CUDA_VERSION=111-nomatmul python -m build
-python -m twine upload dist/* --verbose --repository testpypi
-
-rm -rf dist build
-make clean
-make cleaneggs
-export CUDA_HOME=$BASE_PATH/cuda-11.2
-make cuda11x_nomatmul
-
-if [ ! -f "./bitsandbytes/libbitsandbytes.so" ]; then
-  # Control will enter here if $DIRECTORY doesn't exist.
-  echo "Compilation unsuccessul!" 1>&2
-  exit 64
-fi
-CUDA_VERSION=112-nomatmul python -m build
-python -m twine upload dist/* --verbose --repository testpypi
-
-rm -rf dist build
-make clean
-make cleaneggs
-export CUDA_HOME=$BASE_PATH/cuda-11.3
-make cuda11x_nomatmul
-
-if [ ! -f "./bitsandbytes/libbitsandbytes.so" ]; then
-  # Control will enter here if $DIRECTORY doesn't exist.
-  echo "Compilation unsuccessul!" 1>&2
-  exit 64
-fi
-CUDA_VERSION=113-nomatmul python -m build
-python -m twine upload dist/* --verbose --repository testpypi
-
-rm -rf dist build
-make clean
-make cleaneggs
-export CUDA_HOME=$BASE_PATH/cuda-11.4
-make cuda11x_nomatmul
-
-if [ ! -f "./bitsandbytes/libbitsandbytes.so" ]; then
-  # Control will enter here if $DIRECTORY doesn't exist.
-  echo "Compilation unsuccessul!" 1>&2
-  exit 64
-fi
-CUDA_VERSION=114-nomatmul python -m build
-python -m twine upload dist/* --verbose --repository testpypi
-
-rm -rf dist build
-make clean
-make cleaneggs
-export CUDA_HOME=$BASE_PATH/cuda-11.5
-make cuda11x_nomatmul
+make cuda110 CUDA_VERSION=110
+
+if [ ! -f "./bitsandbytes/libbitsandbytes.so" ]; then
+  # Control will enter here if $DIRECTORY doesn't exist.
+  echo "Compilation unsuccessul!" 1>&2
+  exit 64
+fi
+
+#make clean
+#export CUDA_HOME=$BASE_PATH/cuda-11.1
+#make cuda11x CUDA_VERSION=111
+#
+#if [ ! -f "./bitsandbytes/libbitsandbytes.so" ]; then
+#  # Control will enter here if $DIRECTORY doesn't exist.
+#  echo "Compilation unsuccessul!" 1>&2
+#  exit 64
+#fi
+#
+#make clean
+#export CUDA_HOME=$BASE_PATH/cuda-11.2
+#make cuda11x CUDA_VERSION=112
+#
+#if [ ! -f "./bitsandbytes/libbitsandbytes.so" ]; then
+#  # Control will enter here if $DIRECTORY doesn't exist.
+#  echo "Compilation unsuccessul!" 1>&2
+#  exit 64
+#fi
+#CUDA_VERSION=112 python -m build
+#
+#make clean
+#export CUDA_HOME=$BASE_PATH/cuda-11.3
+#make cuda11x CUDA_VERSION=113
+#
+#if [ ! -f "./bitsandbytes/libbitsandbytes.so" ]; then
+#  # Control will enter here if $DIRECTORY doesn't exist.
+#  echo "Compilation unsuccessul!" 1>&2
+#  exit 64
+#fi
+#CUDA_VERSION=113 python -m build
+#
+#make clean
+#export CUDA_HOME=$BASE_PATH/cuda-11.4
+#make cuda11x CUDA_VERSION=114
+#
+#if [ ! -f "./bitsandbytes/libbitsandbytes.so" ]; then
+#  # Control will enter here if $DIRECTORY doesn't exist.
+#  echo "Compilation unsuccessul!" 1>&2
+#  exit 64
+#fi
+#CUDA_VERSION=114 python -m build
+#
+#make clean
+#export CUDA_HOME=$BASE_PATH/cuda-11.5
+#make cuda11x CUDA_VERSION=115
+#
+#if [ ! -f "./bitsandbytes/libbitsandbytes.so" ]; then
+#  # Control will enter here if $DIRECTORY doesn't exist.
+#  echo "Compilation unsuccessul!" 1>&2
+#  exit 64
+#fi
+#CUDA_VERSION=115 python -m build
+#
+#make clean
+#export CUDA_HOME=$BASE_PATH/cuda-11.6
+#
+#make cuda11x CUDA_VERSION=116
+#if [ ! -f "./bitsandbytes/libbitsandbytes.so" ]; then
+#  # Control will enter here if $DIRECTORY doesn't exist.
+#  echo "Compilation unsuccessul!" 1>&2
+#  exit 64
+#fi
+#CUDA_VERSION=116 python -m build
+#
+#make clean
+#export CUDA_HOME=$BASE_PATH/cuda-11.7
+#make cuda11x CUDA_VERSION=117
+#
+#if [ ! -f "./bitsandbytes/libbitsandbytes.so" ]; then
+#  # Control will enter here if $DIRECTORY doesn't exist.
+#  echo "Compilation unsuccessul!" 1>&2
+#  exit 64
+#fi
+#CUDA_VERSION=117 python -m build
+#
+#
+#make clean
+#export CUDA_HOME=$BASE_PATH/cuda-10.2
+#make cuda10x_nomatmul CUDA_VERSION=102
+#
+#if [ ! -f "./bitsandbytes/libbitsandbytes.so" ]; then
+#  # Control will enter here if $DIRECTORY doesn't exist.
+#  echo "Compilation unsuccessul!" 1>&2
+#  exit 64
+#fi
+#CUDA_VERSION=102-nomatmul python -m build
+#
+#
+#make clean
+#export CUDA_HOME=$BASE_PATH/cuda-11.0
+#make cuda110_nomatmul CUDA_VERSION=110
+#
+#if [ ! -f "./bitsandbytes/libbitsandbytes.so" ]; then
+#  # Control will enter here if $DIRECTORY doesn't exist.
+#  echo "Compilation unsuccessul!" 1>&2
+#  exit 64
+#fi
+#CUDA_VERSION=110-nomatmul python -m build
+#
+#
+#make clean
+#export CUDA_HOME=$BASE_PATH/cuda-11.1
+#make cuda11x_nomatmul CUDA_VERSION=111
+#
+#if [ ! -f "./bitsandbytes/libbitsandbytes.so" ]; then
+#  # Control will enter here if $DIRECTORY doesn't exist.
+#  echo "Compilation unsuccessul!" 1>&2
+#  exit 64
+#fi
+#CUDA_VERSION=111-nomatmul python -m build
+#
+#make clean
+#export CUDA_HOME=$BASE_PATH/cuda-11.2
+#make cuda11x_nomatmul CUDA_VERSION=112
+#
+#if [ ! -f "./bitsandbytes/libbitsandbytes.so" ]; then
+#  # Control will enter here if $DIRECTORY doesn't exist.
+#  echo "Compilation unsuccessul!" 1>&2
+#  exit 64
+#fi
+#CUDA_VERSION=112-nomatmul python -m build
+#
+#make clean
+#export CUDA_HOME=$BASE_PATH/cuda-11.3
+#make cuda11x_nomatmul CUDA_VERSION=113
+#
+#if [ ! -f "./bitsandbytes/libbitsandbytes.so" ]; then
+#  # Control will enter here if $DIRECTORY doesn't exist.
+#  echo "Compilation unsuccessul!" 1>&2
+#  exit 64
+#fi
+#CUDA_VERSION=113-nomatmul python -m build
+#
+#make clean
+#export CUDA_HOME=$BASE_PATH/cuda-11.4
+#make cuda11x_nomatmul CUDA_VERSION=114
+#
+#if [ ! -f "./bitsandbytes/libbitsandbytes.so" ]; then
+#  # Control will enter here if $DIRECTORY doesn't exist.
+#  echo "Compilation unsuccessul!" 1>&2
+#  exit 64
+#fi
+#CUDA_VERSION=114-nomatmul python -m build
+#
+#make clean
+#export CUDA_HOME=$BASE_PATH/cuda-11.5
+#make cuda11x_nomatmul CUDA_VERSION=115
+#
+#if [ ! -f "./bitsandbytes/libbitsandbytes.so" ]; then
+#  # Control will enter here if $DIRECTORY doesn't exist.
+#  echo "Compilation unsuccessul!" 1>&2
+#  exit 64
+#fi
+#CUDA_VERSION=115-nomatmul python -m build
+#
+#make clean
+#export CUDA_HOME=$BASE_PATH/cuda-11.6
+#
+#make cuda11x_nomatmul CUDA_VERSION=116
+#if [ ! -f "./bitsandbytes/libbitsandbytes.so" ]; then
+#  # Control will enter here if $DIRECTORY doesn't exist.
+#  echo "Compilation unsuccessul!" 1>&2
+#  exit 64
+#fi
+#CUDA_VERSION=116-nomatmul python -m build
+#
+#make clean
+#export CUDA_HOME=$BASE_PATH/cuda-11.7
+#make cuda11x_nomatmul CUDA_VERSION=117
+#
+#if [ ! -f "./bitsandbytes/libbitsandbytes.so" ]; then
+#  # Control will enter here if $DIRECTORY doesn't exist.
+#  echo "Compilation unsuccessul!" 1>&2
+#  exit 64
+#fi
+#CUDA_VERSION=117-nomatmul python -m build
 
-if [ ! -f "./bitsandbytes/libbitsandbytes.so" ]; then
-  # Control will enter here if $DIRECTORY doesn't exist.
-  echo "Compilation unsuccessul!" 1>&2
-  exit 64
-fi
-CUDA_VERSION=115-nomatmul python -m build
-python -m twine upload dist/* --verbose --repository testpypi
-
-rm -rf dist build
-make clean
-make cleaneggs
-export CUDA_HOME=$BASE_PATH/cuda-11.6
-
-make cuda11x_nomatmul
-if [ ! -f "./bitsandbytes/libbitsandbytes.so" ]; then
-  # Control will enter here if $DIRECTORY doesn't exist.
-  echo "Compilation unsuccessul!" 1>&2
-  exit 64
-fi
-CUDA_VERSION=116-nomatmul python -m build
-python -m twine upload dist/* --verbose --repository testpypi
-
-rm -rf dist build
-make clean
-make cleaneggs
-export CUDA_HOME=$BASE_PATH/cuda-11.7
-make cuda11x_nomatmul
-
-if [ ! -f "./bitsandbytes/libbitsandbytes.so" ]; then
-  # Control will enter here if $DIRECTORY doesn't exist.
-  echo "Compilation unsuccessul!" 1>&2
-  exit 64
-fi
-CUDA_VERSION=117-nomatmul python -m build
 python -m twine upload dist/* --verbose --repository testpypi
diff --git a/setup.py b/setup.py
index 3292e30..eefc6df 100644
--- a/setup.py
+++ b/setup.py
@@ -6,6 +6,15 @@ import os
 from setuptools import setup, find_packages
 
 
+import os
+
+def all_libs(directory):
+    paths = []
+    for (path, directories, filenames) in os.walk(directory):
+        for filename in filenames:
+            paths.append(os.path.join('..', path, filename))
+    return paths
+
 def read(fname):
     return open(os.path.join(os.path.dirname(__file__), fname)).read()
 
@@ -15,7 +24,7 @@ prefix = '' if version == 'cpu' else 'cuda'
 
 setup(
     name=f"bitsandbytes-{prefix}{version}",
-    version=f"0.30.0",
+    version=f"0.30.2",
     author="Tim Dettmers",
     author_email="dettmers@cs.washington.edu",
     description="8-bit optimizers and matrix multiplication routines.",
@@ -26,7 +35,7 @@ setup(
     entry_points={
         "console_scripts": ["debug_cuda = bitsandbytes.debug_cli:cli"],
     },
-    package_data={'': ['libbitsandbytes.so']},
+    package_data={'': ['libbitsandbytes*.so']},
     long_description=read('README.md'),
     long_description_content_type='text/markdown',
     classifiers=[