connor (he/him) (UTC-5) tiny-cuda-nn fails on aarch64-linux with:

-- Build files have been written to: /build/source/build
cmake: enabled parallel building
cmake: enabled parallel installing
Running phase: buildPhase
build flags: -j80
[1/20] Building CUDA object CMakeFiles/tiny-cuda-nn.dir/src/encoding.cu.o
FAILED: CMakeFiles/tiny-cuda-nn.dir/src/encoding.cu.o 
/nix/store/z8g6ma876kbi5mxwq388aadn1h35yqy9-cuda-redist/bin/nvcc -forward-unknown-to-host-compiler -ccbin=/nix/store/k0kxskxvmkw97h3z3b5y4hwd56fh9x33-gcc-wrapper-13.3.0/bin/c++ -DTCNN_MIN_GPU_ARCH=60 -DTCNN_SHAMPOO -I/build/source/include -I/build/source/dependencies -I/build/source/dependencies/cutlass/include -I/build/source/dependencies/cutlass/tools/util/include -I/build/source/dependencies/fmt/include -O3 -DNDEBUG -std=c++14 "--generate-code=arch=compute_60,code=[compute_60,sm_60]" "--generate-code=arch=compute_61,code=[compute_61,sm_61]" "--generate-code=arch=compute_70,code=[compute_70,sm_70]" "--generate-code=arch=compute_75,code=[compute_75,sm_75]" "--generate-code=arch=compute_80,code=[compute_80,sm_80]" "--generate-code=arch=compute_86,code=[compute_86,sm_86]" "--generate-code=arch=compute_89,code=[compute_89,sm_89]" "--generate-code=arch=compute_90,code=[compute_90,sm_90]" "--generate-code=arch=compute_90a,code=[compute_90a,sm_90a]" -Xcompiler=-mf16c -Xcompiler=-Wno-float-conversion -Xcompiler=-fno-strict-aliasing -Xcudafe=--diag_suppress=unrecognized_gcc_pragma --extended-lambda --expt-relaxed-constexpr -MD -MT CMakeFiles/tiny-cuda-nn.dir/src/encoding.cu.o -MF CMakeFiles/tiny-cuda-nn.dir/src/encoding.cu.o.d -x cu -c /build/source/src/encoding.cu -o CMakeFiles/tiny-cuda-nn.dir/src/encoding.cu.o
nvcc warning : incompatible redefinition for option 'compiler-bindir', the last value of this option was used
g++: error: unrecognized command-line option '-mf16c'
...

_______ TestKernelLinearOperatorLinOpReturn.test_solve_matrix_broadcast ________

self = <test.operators.test_kernel_linear_operator.TestKernelLinearOperatorLinOpReturn testMethod=test_solve_matrix_broadcast>

    def test_solve_matrix_broadcast(self):
        linear_op = self.create_linear_op()
    
        # Right hand size has one more batch dimension
        batch_shape = torch.Size((3, *linear_op.batch_shape))
        rhs = torch.randn(*batch_shape, linear_op.size(-1), 5)
        self._test_solve(rhs)
    
        if linear_op.ndimension() > 2:
            # Right hand size has one fewer batch dimension
            batch_shape = torch.Size(linear_op.batch_shape[1:])
            rhs = torch.randn(*batch_shape, linear_op.size(-1), 5)
>           self._test_solve(rhs)

linear_operator/test/linear_operator_test_case.py:1115: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
linear_operator/test/linear_operator_test_case.py:615: in _test_solve
    self.assertAllClose(arg.grad, arg_copy.grad, **self.tolerances["grad"])
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

self = <test.operators.test_kernel_linear_operator.TestKernelLinearOperatorLinOpReturn testMethod=test_solve_matrix_broadcast>
tensor1 = tensor([[[[ 1.8514e+04,  7.1797e+03, -1.1073e+04, -6.6690e+03,  1.2985e+04,
            6.8468e+03],
          [ 1.685...  -3.0153e+04],
          [-9.0042e+04, -1.3429e+04, -3.1822e+04,  1.3839e+04,  5.9735e+04,
           -5.4315e+04]]]])
tensor2 = tensor([[[[ 1.8514e+04,  7.1797e+03, -1.1073e+04, -6.6690e+03,  1.2985e+04,
            6.8468e+03],
          [ 1.685...  -3.0153e+04],
          [-9.0042e+04, -1.3429e+04, -3.1822e+04,  1.3839e+04,  5.9735e+04,
           -5.4315e+04]]]])
rtol = 0.03, atol = 1e-05, equal_nan = False

    def assertAllClose(self, tensor1, tensor2, rtol=1e-4, atol=1e-5, equal_nan=False):
        if not tensor1.shape == tensor2.shape:
            raise ValueError(f"tensor1 ({tensor1.shape}) and tensor2 ({tensor2.shape}) do not have the same shape.")
    
        if torch.allclose(tensor1, tensor2, rtol=rtol, atol=atol, equal_nan=equal_nan):
            return True
    
        if not equal_nan:
            if not torch.equal(tensor1, tensor1):
                raise AssertionError(f"tensor1 ({tensor1.shape}) contains NaNs")
            if not torch.equal(tensor2, tensor2):
                raise AssertionError(f"tensor2 ({tensor2.shape}) contains NaNs")
    
        rtol_diff = (torch.abs(tensor1 - tensor2) / torch.abs(tensor2)).view(-1)
        rtol_diff = rtol_diff[torch.isfinite(rtol_diff)]
        rtol_max = rtol_diff.max().item()
    
        atol_diff = (torch.abs(tensor1 - tensor2) - torch.abs(tensor2).mul(rtol)).view(-1)
        atol_diff = atol_diff[torch.isfinite(atol_diff)]
        atol_max = atol_diff.max().item()
    
>       raise AssertionError(
            f"tensor1 ({tensor1.shape}) and tensor2 ({tensor2.shape}) are not close enough. \n"
            f"max rtol: {rtol_max:0.8f}\t\tmax atol: {atol_max:0.8f}"
        )
E       AssertionError: tensor1 (torch.Size([2, 3, 4, 6])) and tensor2 (torch.Size([2, 3, 4, 6])) are not close enough. 
E       max rtol: 0.03577567            max atol: 0.00741313

linear_operator/test/base_test_case.py:46: AssertionError