changing workflow

Author: jafioti

.github/workflows/test.yml

@@ -46,9 +46,7 @@ jobs:
 
         steps:
             - uses: actions/checkout@v4
-            - name: GPU sanity check
-              run: sudo apt-get update; sudo apt-get install -y cuda-nvrtc-12-8; sudo apt install tree; ls /usr/local/cuda-12.8/lib64
-            - name: GPU sanity check 1
-              run: sudo apt install tree; ls /usr/local; ls /usr/local/lib; tree /usr/lib -L 2
+            - name: Install nvrtc
+              run: sudo apt-get update; sudo apt-get install -y cuda-nvrtc-12-8
             - name: Run CUDA crate tests
               run: export LD_LIBRARY_PATH=/usr/local/cuda-12.8/lib64:$LD_LIBRARY_PATH; curl https://sh.rustup.rs -sSf | sh -s -- -y && source "$HOME/.cargo/env"; sudo apt-get update && sudo apt-get install -y protobuf-compiler; rustup update; cargo test -p luminal_cuda --verbose

crates/luminal_cuda/src/kernel/ops.rs

@@ -2,7 +2,7 @@ use std::sync::Arc;
 
 use cudarc::{
     driver::{CudaContext, CudaFunction, CudaSlice, CudaStream},
-    nvrtc::{compile_ptx_with_opts, CompileOptions},
+    nvrtc::{compile_ptx, compile_ptx_with_opts, CompileOptions},
 };
 use itertools::Itertools;
 use luminal::{
@@ -115,19 +115,7 @@ extern \"C\" {{
             flatten_strides(&self.out_shape, &self.a_stride).to_kernel(),
             flatten_strides(&self.out_shape, &self.b_stride).to_kernel()
         );
-        let ptx = compile_ptx_with_opts(
-            &kernel,
-            CompileOptions {
-                // arch: Some("sm_90a"),
-                // options: vec!["--std=c++17".to_string(), "-default-device".to_string()],
-                // include_paths: vec![
-                //     "/usr/local/cuda/include".to_string(),
-                //     "/usr/include".to_string(),
-                // ],
-                ..Default::default()
-            },
-        )
-        .unwrap();
+        let ptx = compile_ptx(&kernel).unwrap();
         let module = ctx.load_module(ptx).unwrap();
         let func = module.load_function("add_k").unwrap();
         let constants = vars