Bump CUDA 12.2.0 to 12.2.1, fix setup support for Cuda 12.1 (#703), Sort compute capabilities sets to select max

* Add support for CUDA 12.1

* Update README to include CUDA 12.1 version

* Add support for >= 12x

Co-authored-by: Jeongseok Kang <[email protected]>

* Temporary version of bitsandbytes PR 527: Sort compute capabilities sets to select max

* Modify PR 506 to support C++20

* Add Cuda 12.2

---------

Co-authored-by: PriNova <[email protected]>
Co-authored-by: PriNova <[email protected]>
Co-authored-by: Jeongseok Kang <[email protected]>

Author: 3 people

Makefile

@@ -110,7 +110,7 @@ cuda118: $(BUILD_DIR) env
 cuda12x: $(BUILD_DIR) env
 	$(NVCC) $(CC_cublasLt111) $(CC_ADA_HOPPER) -Xcompiler '-fPIC' --use_fast_math -Xptxas=-v -dc $(FILES_CUDA) $(INCLUDE) $(LIB) --output-directory $(BUILD_DIR)
 	$(NVCC) $(CC_cublasLt111) $(CC_ADA_HOPPER) -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 $(FILES_CPP) -o ./bitsandbytes/libbitsandbytes_cuda$(CUDA_VERSION).so $(LIB)
+	$(GPP) -std=c++20 -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_cpu.so

README.md

@@ -26,7 +26,7 @@ Compilation quickstart:
 git clone https://github.com/timdettmers/bitsandbytes.git
 cd bitsandbytes
 
-# CUDA_VERSIONS in {110, 111, 112, 113, 114, 115, 116, 117, 118, 119, 120, 120}
+# CUDA_VERSIONS in {110, 111, 112, 113, 114, 115, 116, 117, 118, 119, 120, 121, 122}
 # make argument in {cuda110, cuda11x, cuda12x}
 # if you do not know what CUDA you have, try looking at the output of: python -m bitsandbytes
 CUDA_VERSION=117 make cuda11x
@@ -83,7 +83,7 @@ Hardware requirements:
  - LLM.int8(): NVIDIA Turing (RTX 20xx; T4) or Ampere GPU (RTX 30xx; A4-A100); (a GPU from 2018 or newer).
  - 8-bit optimizers and quantization: NVIDIA Kepler GPU or newer (>=GTX 78X).
 
-Supported CUDA versions: 10.2 - 12.0
+Supported CUDA versions: 10.2 - 12.2
 
 The bitsandbytes library is currently only supported on Linux distributions. Windows is not supported at the moment.
 

bitsandbytes/cuda_setup/main.py

@@ -31,7 +31,7 @@
 # libcudart.so is missing by default for a conda install with PyTorch 2.0 and instead
 # we have libcudart.so.11.0 which causes a lot of errors before
 # not sure if libcudart.so.12.0 exists in pytorch installs, but it does not hurt
-CUDA_RUNTIME_LIBS: list = ["libcudart.so", 'libcudart.so.11.0', 'libcudart.so.12.0']
+CUDA_RUNTIME_LIBS: list = ["libcudart.so", 'libcudart.so.11.0', 'libcudart.so.12.0', 'libcudart.so.12.1', 'libcudart.so.12.2']
 
 # this is a order list of backup paths to search CUDA in, if it cannot be found in the main environmental paths
 backup_paths = []
@@ -77,6 +77,8 @@ def generate_instructions(self):
             make_cmd += ' make cuda110'
         elif self.cuda_version_string[:2] == '11' and int(self.cuda_version_string[2]) > 0:
             make_cmd += ' make cuda11x'
+        elif self.cuda_version_string[:2] == '12' and 1 >= int(self.cuda_version_string[2]) >= 0:
+            make_cmd += ' make cuda12x'
         elif self.cuda_version_string == '100':
             self.add_log_entry('CUDA SETUP: CUDA 10.0 not supported. Please use a different CUDA version.')
             self.add_log_entry('CUDA SETUP: Before you try again running bitsandbytes, make sure old CUDA 10.0 versions are uninstalled and removed from $LD_LIBRARY_PATH variables.')
@@ -327,6 +329,8 @@ def get_compute_capabilities():
         cc_major, cc_minor = torch.cuda.get_device_capability(torch.cuda.device(i))
         ccs.append(f"{cc_major}.{cc_minor}")
 
+    ccs.sort(key=lambda v: tuple(map(int, str(v).split("."))))
+
     return ccs
 
 

compile_from_source.md

@@ -11,7 +11,7 @@ You can install CUDA locally without sudo by following the following steps:
 ```bash
 wget https://raw.githubusercontent.com/TimDettmers/bitsandbytes/main/install_cuda.sh
 # Syntax cuda_install CUDA_VERSION INSTALL_PREFIX EXPORT_TO_BASH
-#   CUDA_VERSION in {110, 111, 112, 113, 114, 115, 116, 117, 118, 120, 121}
+#   CUDA_VERSION in {110, 111, 112, 113, 114, 115, 116, 117, 118, 120, 121, 122}
 #   EXPORT_TO_BASH in {0, 1} with 0=False and 1=True 
 
 # For example, the following installs CUDA 11.7 to ~/local/cuda-11.7 and exports the path to your .bashrc

install_cuda.sh

@@ -12,8 +12,8 @@ URL116=https://developer.download.nvidia.com/compute/cuda/11.6.2/local_installer
 URL117=https://developer.download.nvidia.com/compute/cuda/11.7.0/local_installers/cuda_11.7.0_515.43.04_linux.run
 URL118=https://developer.download.nvidia.com/compute/cuda/11.8.0/local_installers/cuda_11.8.0_520.61.05_linux.run
 URL120=https://developer.download.nvidia.com/compute/cuda/12.0.0/local_installers/cuda_12.0.0_525.60.13_linux.run
-URL121=https://developer.download.nvidia.com/compute/cuda/12.1.0/local_installers/cuda_12.1.0_530.30.02_linux.run
-URL122=https://developer.download.nvidia.com/compute/cuda/12.2.0/local_installers/cuda_12.2.0_535.54.03_linux.run
+URL121=https://developer.download.nvidia.com/compute/cuda/12.1.1/local_installers/cuda_12.1.1_530.30.02_linux.run
+URL122=https://developer.download.nvidia.com/compute/cuda/12.2.1/local_installers/cuda_12.2.1_535.86.10_linux.run
 URL123=https://developer.download.nvidia.com/compute/cuda/12.3.1/local_installers/cuda_12.3.1_545.23.08_linux.run