Make CUDA and GPU checks optional, to enable compilation on non-GPU systems

Signed-off-by: Santi Villalba <sdvillal@gmail.com>

Author: sdvillal

docs/_tutorials/ds4sci_evoformerattention.md

@@ -27,6 +27,7 @@ export CUTLASS_PATH=/path/to/cutlass
 The kernels will be compiled when `DS4Sci_EvoformerAttention` is called for the first time.
 
 `DS4Sci_EvoformerAttention` requires GPUs with compute capability 7.0 or higher (NVIDIA V100 or later GPUs) and the minimal CUDA version is 11.3. It is recommended to use CUDA 11.7 or later for better performance. Besides, the performance of backward kernel on V100 kernel is not as good as that on A100 for now.
+The extension checks both requirements and fails if any is not met. To disable the check, for example for cross-compiling in a system without GPUs, you can set the environment variable ```DS_IGNORE_CUDA_DETECTION=TRUE```.
 
 ### 3.2 Unit test and benchmark
 

op_builder/evoformer_attn.py

@@ -57,19 +57,21 @@ def is_compatible(self, verbose=False):
                             self.warning("Please use CUTLASS version >= 3.1.0")
                         return False
 
+        # Check CUDA and GPU capabilities
         cuda_okay = True
-        if not self.is_rocm_pytorch() and torch.cuda.is_available():  #ignore-cuda
-            sys_cuda_major, _ = installed_cuda_version()
-            torch_cuda_major = int(torch.version.cuda.split('.')[0])
-            cuda_capability = torch.cuda.get_device_properties(0).major  #ignore-cuda
-            if cuda_capability < 7:
-                if verbose:
-                    self.warning("Please use a GPU with compute capability >= 7.0")
-                cuda_okay = False
-            if torch_cuda_major < 11 or sys_cuda_major < 11:
-                if verbose:
-                    self.warning("Please use CUDA 11+")
-                cuda_okay = False
+        if not os.environ.get("DS_IGNORE_CUDA_DETECTION"):
+            if not self.is_rocm_pytorch() and torch.cuda.is_available():  #ignore-cuda
+                sys_cuda_major, _ = installed_cuda_version()
+                torch_cuda_major = int(torch.version.cuda.split('.')[0])
+                cuda_capability = torch.cuda.get_device_properties(0).major  #ignore-cuda
+                if cuda_capability < 7:
+                    if verbose:
+                        self.warning("Please use a GPU with compute capability >= 7.0")
+                    cuda_okay = False
+                if torch_cuda_major < 11 or sys_cuda_major < 11:
+                    if verbose:
+                        self.warning("Please use CUDA 11+")
+                    cuda_okay = False
         return super().is_compatible(verbose) and cuda_okay
 
     def include_paths(self):