Error when installing selective_scan on Windows using conda environment

[WqxLoveCoding]

Error (same with #84)：

platform： Windows 10

pytorch install command (successful)： conda install pytorch==2.2.0 torchvision==0.17.0 torchaudio==2.2.0 pytorch-cuda=12.1 -c pytorch -c nvidia

python version： 3.8.18

nvcc -V

nvidia-smi

[WqxLoveCoding]

Note： I've found that following the given installation steps on Linux, all steps work fine, so I'm not sure if it's platform related, but I'm currently desperate to experiment on Windows...

[MzeroMiko]

M_LOG2E is actually math_log_2{e}, which is approximately 1.4426950408889634. You can #define M_LOG2E 1.4426950408889634; to use it.

[WqxLoveCoding]

M_LOG2E is actually math_log_2{e}, which is approximately 1.4426950408889634. You can #define M_LOG2E 1.4426950408889634; to use it.

I have added #define M_LOG2E 1.4426950408889634; to selective_scan_bwd_kernel.cuh, another error came up...

[MzeroMiko]

Can you change constexpr bool in static_switch.h to static constexpr bool?

By the way, you can try changing all constexpr to static constexpr and install again.

[WqxLoveCoding]

Can you change constexpr bool in static_switch.h to static constexpr bool?

By the way, you can try changing all constexpr to static constexpr and install again.

God, it finally worked! Also, the modifications that made the compilation process work are documented here in the hope that they may help others. I'll run training using any of the models to verify that the compilation is okay.

1. Change constexpr bool in kernels/selective_scan/csrc/selective_scan/static_switch.h to static const bool;
2. Add the following code

#ifndef M_LOG2E
#define M_LOG2E 1.4426950408889634074
#endif

to：

• kernels/selective_scan/csrc/selective_scan/cus/selective_scan_bwd_kernel.cuh
• kernels/selective_scan/csrc/selective_scan/cus/selective_scan_fwd_kernel.cuh
• kernels/selective_scan/csrc/selective_scan/cusndstate/selective_scan_bwd_kernel_ndstate.cuh
• kernels/selective_scan/csrc/selective_scan/cusndstate/selective_scan_fwd_kernel_ndstate.cuh
• kernels/selective_scan/csrc/selective_scan/cusoflex/selective_scan_bwd_kernel_oflex.cuh
• kernels/selective_scan/csrc/selective_scan/cusoflex/selective_scan_fwd_kernel_oflex.cuh

Useful links:

1. This issue
2. link 1
3. link 2
4. link 3

[CosmoWood]

Can you change constexpr bool in static_switch.h to static constexpr bool?
By the way, you can try changing all constexpr to static constexpr and install again.

God, it finally worked! Also, the modifications that made the compilation process work are documented here in the hope that they may help others. I'll run training using any of the models to verify that the compilation is okay.

1. Change constexpr bool in kernels/selective_scan/csrc/selective_scan/static_switch.h to static const bool;
2. Add the following code

#ifndef M_LOG2E
#define M_LOG2E 1.4426950408889634074
#endif

to：

• kernels/selective_scan/csrc/selective_scan/cus/selective_scan_bwd_kernel.cuh
• kernels/selective_scan/csrc/selective_scan/cus/selective_scan_fwd_kernel.cuh
• kernels/selective_scan/csrc/selective_scan/cusndstate/selective_scan_bwd_kernel_ndstate.cuh
• kernels/selective_scan/csrc/selective_scan/cusndstate/selective_scan_fwd_kernel_ndstate.cuh
• kernels/selective_scan/csrc/selective_scan/cusoflex/selective_scan_bwd_kernel_oflex.cuh
• kernels/selective_scan/csrc/selective_scan/cusoflex/selective_scan_fwd_kernel_oflex.cuh

Useful links:

1. This issue
2. link 1
3. link 2
4. link 3

It worked Thanks

[wenqiuL]

oh my god! Finally solved! Very useful, thank you, successfully installed on Windows. my successful version:

[ddcx00]

错误（同#84）：

平台： Windows 10

pytorch install命令（成功）： conda install pytorch==2.2.0 torchvision==0.17.0 torchaudio==2.2.0 pytorch-cuda=12.1 -c pytorch -c nvidia

python版本： 3.8.18

nvcc -V

nvidia-smi

Excuse me? I noticed that the environment you are using is Python 3.8 version，how do I install the Triton library on the Windows platform？

[Si1ive]

我C！！！！牛逼