Port to more current rust-nightly

crates/cuda_builder/src/lib.rs

@@ -376,10 +376,12 @@ fn invoke_rustc(builder: &CudaBuilder) -> Result<PathBuf, CudaBuilderError> {
 
     let new_path = get_new_path_var();
 
-    let mut rustflags = vec![format!(
-        "-Zcodegen-backend={}",
-        rustc_codegen_nvvm.display(),
-    )];
+    let mut rustflags = vec![
+        format!("-Zcodegen-backend={}", rustc_codegen_nvvm.display(),),
+        "-Zcrate-attr=feature(register_tool)".to_string(),
+        "-Zcrate-attr=register_tool(rust_cuda)".to_string(),
+        "-Zcrate-attr=no_std".to_string(),
+    ];
 
     if let Some(emit) = &builder.emit {
         let string = match emit {

crates/cuda_std/src/atomic/intrinsics.rs

@@ -1,6 +1,7 @@
 //! Raw CUDA-specific atomic functions that map to PTX instructions.
 
 use crate::gpu_only;
+use core::arch::asm;
 use core::concat;
 use paste::paste;
 

crates/cuda_std/src/lib.rs

@@ -23,15 +23,11 @@
 
 #![cfg_attr(
     target_os = "cuda",
-    no_std,
     feature(
-        register_attr,
         alloc_error_handler,
-        asm,
         asm_experimental_arch,
         link_llvm_intrinsics
     ),
-    register_attr(nvvm_internal)
 )]
 
 extern crate alloc;
@@ -49,7 +45,7 @@ pub mod cfg;
 pub mod ptr;
 pub mod shared;
 pub mod thread;
-pub mod warp;
+// pub mod warp;
 
 mod float_ext;
 

crates/cuda_std/src/mem.rs

@@ -1,8 +1,8 @@
 //! Support for allocating memory and using `alloc` using CUDA memory allocation system-calls.
-
 use crate::gpu_only;
 #[cfg(any(target_arch = "nvptx", target_arch = "nvptx64"))]
 use alloc::alloc::*;
+use core::arch::asm;
 #[cfg(any(target_arch = "nvptx", target_arch = "nvptx64"))]
 use core::ffi::c_void;
 

crates/cuda_std/src/misc.rs

@@ -1,6 +1,6 @@
 //! 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 +1,6 @@
 //! 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/shared.rs

@@ -77,7 +77,7 @@ pub fn dynamic_shared_mem<T>() -> *mut T {
     extern "C" {
         // need to use nvvm_internal and not address_space because address_space only parses
         // static definitions, not extern static definitions.
-        #[nvvm_internal(addrspace(3))]
+        #[rust_cuda::nvvm_internal(addrspace(3))]
         #[allow(improper_ctypes)]
         // mangle it a bit to make sure nobody makes the same thing
         #[link_name = "_Zcuda_std_dyn_shared"]

crates/cuda_std/src/thread.rs

@@ -17,7 +17,7 @@
 //! The most important structure after threads, thread blocks arrange
 
 // TODO: write some docs about the terms used in this module.
-
+use core::arch::asm;
 use cuda_std_macros::gpu_only;
 use vek::{Vec2, Vec3};
 

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/cuda_std_macros/src/lib.rs

@@ -1,6 +1,6 @@
 use proc_macro::TokenStream;
-use proc_macro2::Span;
-use quote::{quote_spanned, ToTokens};
+use proc_macro2::{Span, TokenTree, Delimiter, Group};
+use quote::{quote_spanned, ToTokens, quote};
 use syn::{
     parse::Parse, parse_macro_input, parse_quote, punctuated::Punctuated, spanned::Spanned, Error,
     FnArg, Ident, ItemFn, ReturnType, Stmt, Token,
@@ -27,7 +27,7 @@ pub fn kernel(input: proc_macro::TokenStream, item: proc_macro::TokenStream) ->
     let mut item = parse_macro_input!(item as ItemFn);
     let no_mangle = parse_quote!(#[no_mangle]);
     item.attrs.push(no_mangle);
-    let internal = parse_quote!(#[cfg_attr(any(target_arch="nvptx", target_arch="nvptx64"), nvvm_internal(kernel(#input)))]);
+    let internal = parse_quote!(#[cfg_attr(any(target_arch="nvptx", target_arch="nvptx64"), rust_cuda::nvvm_internal(kernel(#input)))]);
     item.attrs.push(internal);
 
     // used to guarantee some things about how params are passed in the codegen.
@@ -231,7 +231,7 @@ pub fn address_space(attr: proc_macro::TokenStream, item: proc_macro::TokenStrea
     };
 
     let new_attr =
-        parse_quote!(#[cfg_attr(target_os = "cuda", nvvm_internal(addrspace(#addrspace_num)))]);
+        parse_quote!(#[cfg_attr(target_os = "cuda", rust_cuda::nvvm_internal(addrspace(#addrspace_num)))]);
     global.attrs.push(new_attr);
 
     global.into_token_stream().into()

crates/cust/Cargo.toml

@@ -17,7 +17,7 @@ cust_core = { path = "../cust_core", version = "0.1.0"}
 cust_raw = { path = "../cust_raw", version = "0.11.2"}
 bitflags = "1.2"
 cust_derive = { path = "../cust_derive", version = "0.2" }
-glam = { version = "0.20", features=["cuda"], optional = true }
+glam = { version = "0.21", features=["cuda"], optional = true }
 mint = { version = "^0.5", optional = true }
 num-complex = { version = "0.4", optional = true }
 vek = { version = "0.15.1", optional = true, default-features = false }

crates/cust_core/Cargo.toml

@@ -9,7 +9,7 @@ readme = "../../README.md"
 
 [dependencies]
 vek = { version = "0.15.1", default-features=false, features=["libm"], optional = true }
-glam = { version = "0.20", features=["cuda", "libm"], default-features=false, optional=true }
+glam = { version = "0.21", features=["cuda", "libm"], default-features=false, optional=true }
 mint = { version = "^0.5", optional = true }
 half = { version = "1.8", optional = true }
 num-complex = { version = "0.4", optional = true }

crates/cust_core/src/lib.rs

@@ -172,7 +172,7 @@ pub mod _hidden {
 
     #[cfg(feature = "glam")]
     impl_device_copy! {
-        glam::Vec2, glam::Vec3, glam::Vec4, glam::IVec2, glam::IVec3, glam::IVec4,
+        glam::Vec2, glam::Vec3, glam::Vec4, glam::IVec2, glam::IVec3, glam::IVec4, glam::Mat3, glam::Mat4
     }
 
     #[cfg(feature = "mint")]

crates/optix/Cargo.toml

@@ -20,7 +20,7 @@ cust = { version = "0.3", path = "../cust", features=["impl_mint"] }
 cust_raw = { version = "0.11.2", path = "../cust_raw" }
 cfg-if = "1.0.0"
 bitflags = "1.3.2"
-glam = { version = "0.20", features=["cuda", "libm"], default-features=false, optional=true }
+glam = { version = "0.21", features=["cuda", "libm"], default-features=false, optional=true }
 half = { version = "^1.8", optional = true }
 memoffset = "0.6.4"
 mint = "0.5.8"

crates/optix/examples/ex02_pipeline/device/src/lib.rs

@@ -1,9 +1,8 @@
-#![feature(asm)]
 #![cfg_attr(
     target_os = "cuda",
     no_std,
-    feature(register_attr),
-    register_attr(nvvm_internal)
+    feature(register_tool),
+    register_tool(nvvm_internal)
 )]
 // #![deny(warnings)]
 #![allow(clippy::missing_safety_doc)]

crates/optix/examples/rust/ex04_mesh_gpu/src/lib.rs

@@ -1,8 +1,8 @@
 #![cfg_attr(
     target_os = "cuda",
     no_std,
-    feature(register_attr),
-    register_attr(nvvm_internal)
+    feature(register_tool),
+    register_tool(nvvm_internal)
 )]
 #![allow(non_snake_case, clippy::missing_safety_doc)]
 

crates/optix_device/Cargo.toml

@@ -7,7 +7,7 @@ authors = ["Anders Langlands <[email protected]>", "Riccardo D'Ambrosio
 [dependencies]
 bitflags = "1.3.2"
 cuda_std = { version = "0.2", path = "../cuda_std" }
-glam = { version = "0.20", features=["cuda", "libm"], default-features=false }
+glam = { version = "0.21", features=["cuda", "libm"], default-features=false }
 paste = "1.0.6"
 seq-macro = "0.3.0"
 cust_core = { version = "0.1", path = "../cust_core" }

crates/optix_device/src/lib.rs

@@ -1,8 +1,8 @@
 #![cfg_attr(
     target_arch = "nvptx64",
     no_std,
-    feature(register_attr, asm, asm_experimental_arch),
-    register_attr(nvvm_internal)
+    feature(register_tool, asm, asm_experimental_arch),
+    register_tool(nvvm_internal)
 )]
 
 extern crate alloc;

crates/rustc_codegen_nvvm/Cargo.toml

@@ -17,7 +17,9 @@ crate-type = ["dylib"]
 [dependencies]
 nvvm = { version = "0.1", path = "../nvvm" }
 rustc-demangle = "0.1.20"
+cstr = "0.2"
 libc = "0.2.97"
+libloading = "0.7"
 tar = "0.4.35"
 once_cell = "1.8.0"
 bitflags = "1.3.2"

crates/rustc_codegen_nvvm/src/abi.rs

@@ -18,6 +18,24 @@ use rustc_target::abi::{self, HasDataLayout, Int};
 pub use rustc_target::spec::abi::Abi;
 use tracing::trace;
 
+fn clone_pass_mode(original: &PassMode) -> PassMode {
+    match original {
+        PassMode::Ignore => PassMode::Ignore,
+        PassMode::Direct(attrs) => PassMode::Direct(*attrs),
+        PassMode::Pair(attrs1, attrs2) => PassMode::Pair(*attrs1, *attrs2),
+        PassMode::Cast(target, bool) => PassMode::Cast(target.clone(), *bool),
+        PassMode::Indirect {
+            attrs,
+            extra_attrs,
+            on_stack,
+        } => PassMode::Indirect {
+            attrs: *attrs,
+            extra_attrs: *extra_attrs,
+            on_stack: *on_stack,
+        },
+    }
+}
+
 pub(crate) fn readjust_fn_abi<'tcx>(
     tcx: TyCtxt<'tcx>,
     fn_abi: &'tcx FnAbi<'tcx, Ty<'tcx>>,
@@ -29,8 +47,7 @@ pub(crate) fn readjust_fn_abi<'tcx>(
     let readjust_arg_abi = |arg: &ArgAbi<'tcx, Ty<'tcx>>| {
         let mut arg = ArgAbi {
             layout: arg.layout,
-            mode: arg.mode,
-            pad: arg.pad,
+            mode: clone_pass_mode(&arg.mode),
         };
 
         // ignore zsts
@@ -265,7 +282,7 @@ pub(crate) trait FnAbiLlvmExt<'ll, 'tcx> {
 impl<'ll, 'tcx> FnAbiLlvmExt<'ll, 'tcx> for FnAbi<'tcx, Ty<'tcx>> {
     fn llvm_type(&self, cx: &CodegenCx<'ll, 'tcx>) -> &'ll Type {
         let args_capacity: usize = self.args.iter().map(|arg|
-            if arg.pad.is_some() { 1 } else { 0 } +
+            // if arg.pad.is_some() { 1 } else { 0 } +
             if let PassMode::Pair(_, _) = arg.mode { 2 } else { 1 }
         ).sum();
 
@@ -281,10 +298,10 @@ impl<'ll, 'tcx> FnAbiLlvmExt<'ll, 'tcx> for FnAbi<'tcx, Ty<'tcx>> {
             } + args_capacity,
         );
 
-        let mut llreturn_ty = match self.ret.mode {
+        let mut llreturn_ty = match &self.ret.mode {
             PassMode::Ignore => cx.type_void(),
             PassMode::Direct(_) | PassMode::Pair(..) => self.ret.layout.immediate_llvm_type(cx),
-            PassMode::Cast(cast) => cast.llvm_type(cx),
+            PassMode::Cast(cast, _) => cast.llvm_type(cx),
             PassMode::Indirect { .. } => {
                 idx += 1;
                 llargument_tys.push(cx.type_ptr_to(self.ret.memory_ty(cx)));
@@ -302,13 +319,7 @@ impl<'ll, 'tcx> FnAbiLlvmExt<'ll, 'tcx> for FnAbi<'tcx, Ty<'tcx>> {
         }
 
         for arg in self.args.iter() {
-            // add padding
-            if let Some(ty) = arg.pad {
-                idx += 1;
-                llargument_tys.push(ty.llvm_type(cx));
-            }
-
-            let llarg_ty = match arg.mode {
+            let llarg_ty = match &arg.mode {
                 PassMode::Ignore => continue,
                 PassMode::Direct(_) => arg.layout.immediate_llvm_type(cx),
                 PassMode::Pair(..) => {
@@ -329,7 +340,7 @@ impl<'ll, 'tcx> FnAbiLlvmExt<'ll, 'tcx> for FnAbi<'tcx, Ty<'tcx>> {
                     idx += 2;
                     continue;
                 }
-                PassMode::Cast(cast) => cast.llvm_type(cx),
+                PassMode::Cast(cast, _) => cast.llvm_type(cx),
                 PassMode::Indirect {
                     attrs: _,
                     extra_attrs: None,
@@ -398,10 +409,7 @@ impl<'ll, 'tcx> FnAbiLlvmExt<'ll, 'tcx> for FnAbi<'tcx, Ty<'tcx>> {
             }
             _ => {}
         }
-        for arg in &self.args {
-            if arg.pad.is_some() {
-                apply(&ArgAttributes::new());
-            }
+        for arg in self.args.iter() {
             match arg.mode {
                 PassMode::Ignore => {}
                 PassMode::Indirect {
@@ -436,7 +444,7 @@ impl<'ll, 'tcx> FnAbiLlvmExt<'ll, 'tcx> for FnAbi<'tcx, Ty<'tcx>> {
                     apply(a);
                     apply(b);
                 }
-                PassMode::Cast(_) => {
+                PassMode::Cast(..) => {
                     apply(&ArgAttributes::new());
                 }
             }
@@ -476,16 +484,13 @@ impl<'ll, 'tcx> FnAbiLlvmExt<'ll, 'tcx> for FnAbi<'tcx, Ty<'tcx>> {
             // If the value is a boolean, the range is 0..2 and that ultimately
             // become 0..0 when the type becomes i1, which would be rejected
             // by the LLVM verifier.
-            if let Int(..) = scalar.value {
+            if scalar.primitive().is_int() {
                 if !scalar.is_bool() && !scalar.is_always_valid(bx) {
-                    bx.range_metadata(callsite, scalar.valid_range);
+                    bx.range_metadata(callsite, scalar.valid_range(bx));
                 }
             }
         }
-        for arg in &self.args {
-            if arg.pad.is_some() {
-                apply(bx.cx, &ArgAttributes::new());
-            }
+        for arg in self.args.iter() {
             match arg.mode {
                 PassMode::Ignore => {}
                 PassMode::Indirect {
@@ -515,7 +520,7 @@ impl<'ll, 'tcx> FnAbiLlvmExt<'ll, 'tcx> for FnAbi<'tcx, Ty<'tcx>> {
                     apply(bx.cx, a);
                     apply(bx.cx, b);
                 }
-                PassMode::Cast(_) => {
+                PassMode::Cast(..) => {
                     apply(bx.cx, &ArgAttributes::new());
                 }
             }
@@ -524,10 +529,6 @@ impl<'ll, 'tcx> FnAbiLlvmExt<'ll, 'tcx> for FnAbi<'tcx, Ty<'tcx>> {
 }
 
 impl<'a, 'll, 'tcx> AbiBuilderMethods<'tcx> for Builder<'a, 'll, 'tcx> {
-    fn apply_attrs_callsite(&mut self, fn_abi: &FnAbi<'tcx, Ty<'tcx>>, callsite: Self::Value) {
-        fn_abi.apply_attrs_callsite(self, callsite)
-    }
-
     fn get_param(&mut self, index: usize) -> Self::Value {
         let val = llvm::get_param(self.llfn(), index as c_uint);
         trace!("Get param `{:?}`", val);
@@ -588,7 +589,7 @@ impl<'ll, 'tcx> ArgAbiExt<'ll, 'tcx> for ArgAbi<'tcx, Ty<'tcx>> {
             OperandValue::Ref(val, None, self.layout.align.abi).store(bx, dst)
         } else if self.is_unsized_indirect() {
             bug!("unsized `ArgAbi` must be handled through `store_fn_arg`");
-        } else if let PassMode::Cast(cast) = self.mode {
+        } else if let PassMode::Cast(cast, _) = &self.mode {
             let can_store_through_cast_ptr = false;
             if can_store_through_cast_ptr {
                 let cast_ptr_llty = bx.type_ptr_to(cast.llvm_type(bx));
@@ -647,7 +648,7 @@ impl<'ll, 'tcx> ArgAbiExt<'ll, 'tcx> for ArgAbi<'tcx, Ty<'tcx>> {
                 extra_attrs: None,
                 on_stack: _,
             }
-            | PassMode::Cast(_) => {
+            | PassMode::Cast(..) => {
                 let next_arg = next();
                 self.store(bx, next_arg, dst);
             }

crates/rustc_codegen_nvvm/src/allocator.rs

@@ -11,7 +11,7 @@ pub(crate) unsafe fn codegen(
     _tcx: TyCtxt<'_>,
     mods: &mut LlvmMod,
     kind: AllocatorKind,
-    has_alloc_error_handler: bool,
+    alloc_error_handler_kind: AllocatorKind,
 ) {
     let llcx = &*mods.llcx;
     let llmod = mods.llmod.as_ref().unwrap();
@@ -94,11 +94,7 @@ pub(crate) unsafe fn codegen(
     // -> ! DIFlagNoReturn
     llvm::Attribute::NoReturn.apply_llfn(llvm::AttributePlace::Function, llfn);
 
-    let kind = if has_alloc_error_handler {
-        AllocatorKind::Global
-    } else {
-        AllocatorKind::Default
-    };
+    let kind = alloc_error_handler_kind;
     let callee = kind.fn_name(sym::oom);
     let callee = llvm::LLVMRustGetOrInsertFunction(llmod, callee.as_ptr().cast(), callee.len(), ty);