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

Author: jorge-ortega

crates/cuda_std/src/warp.rs

@@ -4,6 +4,8 @@
 //! thread blocks and execute in SIMT fashion.
 
 use crate::gpu_only;
+#[cfg(target_os = "cuda")]
+use core::arch::asm;
 use half::{bf16, f16};
 
 /// Synchronizes all of the threads inside of this warp according to `mask`.

crates/optix_device/src/hit.rs

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

crates/optix_device/src/intersect.rs

@@ -1,3 +1,5 @@
+#[cfg(target_os = "cuda")]
+use core::arch::asm;
 use cuda_std::gpu_only;
 use paste::paste;
 use seq_macro::seq;

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

@@ -1,3 +1,5 @@
+#[cfg(target_os = "cuda")]
+use core::arch::asm;
 use cuda_std::gpu_only;
 
 /// Retrieves the data past the SBT header for this particular program

crates/optix_device/src/payload.rs

@@ -1,4 +1,6 @@
 use crate::sys;
+#[cfg(target_os = "cuda")]
+use core::arch::asm;
 use cuda_std::gpu_only;
 
 /// Overrides the payload for the given register to a value.

crates/optix_device/src/ray.rs

@@ -1,4 +1,6 @@
 use crate::trace::*;
+#[cfg(target_os = "cuda")]
+use core::arch::asm;
 use cuda_std::gpu_only;
 use glam::Vec3;
 

crates/optix_device/src/sys.rs

@@ -1,6 +1,8 @@
 #![allow(clippy::missing_safety_doc)]
 
 use crate::trace::{RayFlags, TraversableHandle};
+#[cfg(target_os = "cuda")]
+use core::arch::asm;
 use cuda_std::gpu_only;
 use glam::Vec3;
 use paste::paste;

crates/optix_device/src/transform.rs

@@ -1,4 +1,6 @@
 // use std::hint::unreachable_unchecked;
+#[cfg(target_os = "cuda")]
+use core::arch::asm;
 use cuda_std::gpu_only;
 use glam::{Vec3, Vec4};