Fix: use core::arch::asm when building cuda target

Author: jorge-ortega

crates/cuda_std/src/warp.rs

@@ -5,6 +5,8 @@
 
 use crate::gpu_only;
 use half::{bf16, f16};
+#[cfg(target_os = "cuda")]
+use core::arch::asm;
 
 /// Synchronizes all of the threads inside of this warp according to `mask`.
 ///

crates/optix_device/src/hit.rs

@@ -1,6 +1,7 @@
 use cuda_std::gpu_only;
 use glam::Vec3;
-
+#[cfg(target_os = "cuda")]
+use core::arch::asm;
 /// The type of primitive that a ray hit.
 #[repr(u32)]
 #[derive(Debug, Copy, Clone, PartialEq, Eq, Hash)]

crates/optix_device/src/intersect.rs

@@ -1,6 +1,8 @@
 use cuda_std::gpu_only;
 use paste::paste;
 use seq_macro::seq;
+#[cfg(target_os = "cuda")]
+use core::arch::asm;
 
 #[gpu_only]
 pub fn primitive_index() -> u32 {

crates/optix_device/src/lib.rs

@@ -1,9 +1,5 @@
-#![cfg_attr(
-    target_arch = "nvptx64",
-    no_std,
-    feature(asm_experimental_arch),
-    register_attr(nvvm_internal)
-)]
+#[cfg(target_os = "cuda")]
+use core::arch::asm;
 
 extern crate alloc;
 

crates/optix_device/src/misc.rs

@@ -1,4 +1,6 @@
 use cuda_std::gpu_only;
+#[cfg(target_os = "cuda")]
+use core::arch::asm;
 
 /// Retrieves the data past the SBT header for this particular program
 ///

crates/optix_device/src/payload.rs

@@ -1,5 +1,7 @@
 use crate::sys;
 use cuda_std::gpu_only;
+#[cfg(target_os = "cuda")]
+use core::arch::asm;
 
 /// Overrides the payload for the given register to a value.
 ///

crates/optix_device/src/ray.rs

@@ -1,6 +1,8 @@
 use crate::trace::*;
 use cuda_std::gpu_only;
 use glam::Vec3;
+#[cfg(target_os = "cuda")]
+use core::arch::asm;
 
 /// Returns the ray origin that was passed into [`trace`] in world-space.
 ///

crates/optix_device/src/sys.rs

@@ -4,6 +4,8 @@ use crate::trace::{RayFlags, TraversableHandle};
 use cuda_std::gpu_only;
 use glam::Vec3;
 use paste::paste;
+#[cfg(target_os = "cuda")]
+use core::arch::asm;
 
 macro_rules! set_payload {
     ($($number:literal),*) => {

crates/optix_device/src/transform.rs

@@ -1,6 +1,8 @@
 // use std::hint::unreachable_unchecked;
 use cuda_std::gpu_only;
 use glam::{Vec3, Vec4};
+#[cfg(target_os = "cuda")]
+use core::arch::asm;
 
 use crate::{intersection::ray_time, trace::TraversableHandle};