Updates rust-toolchain to nightly-02-20.

This includes several updates to the backend based on changes in
rustc_codegen_llvm, which brings some bug fixes and a refactor to debug
info. This fixes failing dependencies that resolved to newer versions
that failed to build on the previous nightly.

Removes depending on rustc_codegen_llvm as it can't be imported with the
new toolchain for some reason.

The example cuda kernals fail to build. Their dependencies
cause nvvm to abort with ICE.

Author: jorge-ortega

Cargo.toml

@@ -1,7 +1,5 @@
 [workspace]
-
 resolver = "2"
-
 members = [
   "crates/*",
   "crates/optix/examples/ex*",

crates/cuda_std/src/atomic/intrinsics.rs

@@ -2,6 +2,7 @@
 
 use crate::gpu_only;
 use core::concat;
+use core::arch::asm;
 use paste::paste;
 
 #[gpu_only]

crates/cuda_std/src/lib.rs

@@ -21,13 +21,12 @@
 //! In order to simplify imports, we provide a prelude module which contains GPU analogues to standard library
 //! structures as well as common imports such as [`thread`].
 
+#![allow(internal_features)]
 #![cfg_attr(
     target_os = "cuda",
     no_std,
     feature(
-        register_attr,
         alloc_error_handler,
-        asm,
         asm_experimental_arch,
         link_llvm_intrinsics
     ),

crates/cuda_std/src/mem.rs

@@ -37,7 +37,7 @@ pub static GLOBAL_ALLOCATOR: CUDAAllocator = CUDAAllocator;
 pub fn dynamic_smem_size() -> u32 {
     let mut out;
     unsafe {
-        asm!(
+        core::arch::asm!(
             "mov.u32 {}, %dynamic_smem_size",
             out(reg32) out
         )
@@ -55,7 +55,7 @@ pub fn dynamic_smem_size() -> u32 {
 pub fn total_smem_size() -> u32 {
     let mut out;
     unsafe {
-        asm!(
+        core::arch::asm!(
             "mov.u32 {}, %total_smem_size",
             out(reg32) out
         )

crates/cuda_std/src/misc.rs

@@ -1,6 +1,7 @@
 //! Misc functions that do not exactly fit into other categories.
 
 use crate::gpu_only;
+use core::arch::asm;
 
 /// Suspends execution of the kernel, usually to pause at a specific point when debugging in a debugger.
 #[gpu_only]

crates/cuda_std/src/ptr.rs

@@ -1,6 +1,7 @@
 //! CUDA-specific pointer handling logic.
 
 use crate::gpu_only;
+use core::arch::asm;
 
 /// Special areas of GPU memory where a pointer could reside.
 #[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]

crates/cuda_std/src/thread.rs

@@ -344,7 +344,7 @@ pub fn system_fence() {
 #[inline(always)]
 pub fn nanosleep(nanos: u32) {
     unsafe {
-        asm!(
+        core::arch::asm!(
             "nanosleep {}",
             in(reg32) nanos
         )

crates/cuda_std/src/warp.rs

@@ -4,6 +4,7 @@
 //! thread blocks and execute in SIMT fashion.
 
 use crate::gpu_only;
+use core::arch::asm;
 use half::{bf16, f16};
 
 /// Synchronizes all of the threads inside of this warp according to `mask`.

crates/cust/Cargo.toml

@@ -35,7 +35,7 @@ impl_num_complex = ["cust_core/num-complex", "num-complex"]
 find_cuda_helper = { path = "../find_cuda_helper", version = "0.2" }
 
 [dev-dependencies]
-image = "0.23.14"
+image = "0.25.5"
 
 [package.metadata.docs.rs]
 rustdoc-args = ["--cfg", "docsrs"]

crates/find_cuda_helper/src/lib.rs

@@ -1,6 +1,5 @@
 //! Tiny crate for common logic for finding and including CUDA.
 
-use std::process::Command;
 use std::{
     env,
     path::{Path, PathBuf},
@@ -154,6 +153,7 @@ pub fn find_cuda_lib_dirs() -> Vec<PathBuf> {
 
 #[cfg(not(target_os = "windows"))]
 fn detect_cuda_root_via_which_nvcc() -> PathBuf {
+    use std::process::Command;
     let output = Command::new("which")
         .arg("nvcc")
         .output()