Auto merge of #144269 - jieyouxu:rollup-137ysl2, r=jieyouxu

Rollup of 14 pull requests

Successful merges:

 - #142097 (gpu offload host code generation)
 - #143430 (Lower extra lifetimes before normal generic params.)
 - #143768 (Constify Try, From, TryFrom and relevant traits)
 - #143816 (Implement `check` for compiletest and RA using tool macro)
 - #143985 (rustc_public: de-StableMIR-ize)
 - #144027 (clippy: make tests work in stage 1)
 - #144080 (Mitigate `#[align]` name resolution ambiguity regression with a rename)
 - #144176 (Add approval blocking labels for new bors)
 - #144187 (fix handling of base address for TypeId allocations)
 - #144212 (Remove the ptr_unique lang item)
 - #144243 (Subtree update of `rust-analyzer`)
 - #144246 (Don't use another main test file as auxiliary)
 - #144251 (rustc-dev-guide subtree update)
 - #144254 (opt-dist: make `artifact-dir` an absolute path for `opt-dist local`)

r? `@ghost`
`@rustbot` modify labels: rollup

Author: bors

compiler/rustc_ast_lowering/src/lib.rs

@@ -874,25 +874,32 @@ impl<'a, 'hir> LoweringContext<'a, 'hir> {
     /// name resolver owing to lifetime elision; this also populates the resolver's node-id->def-id
     /// map, so that later calls to `opt_node_id_to_def_id` that refer to these extra lifetime
     /// parameters will be successful.
-    #[instrument(level = "debug", skip(self))]
+    #[instrument(level = "debug", skip(self), ret)]
     #[inline]
     fn lower_lifetime_binder(
         &mut self,
         binder: NodeId,
         generic_params: &[GenericParam],
     ) -> &'hir [hir::GenericParam<'hir>] {
-        let mut generic_params: Vec<_> = self
-            .lower_generic_params_mut(generic_params, hir::GenericParamSource::Binder)
-            .collect();
+        // Start by creating params for extra lifetimes params, as this creates the definitions
+        // that may be referred to by the AST inside `generic_params`.
         let extra_lifetimes = self.resolver.extra_lifetime_params(binder);
         debug!(?extra_lifetimes);
-        generic_params.extend(extra_lifetimes.into_iter().filter_map(|(ident, node_id, res)| {
-            self.lifetime_res_to_generic_param(ident, node_id, res, hir::GenericParamSource::Binder)
-        }));
-        let generic_params = self.arena.alloc_from_iter(generic_params);
-        debug!(?generic_params);
-
-        generic_params
+        let extra_lifetimes: Vec<_> = extra_lifetimes
+            .into_iter()
+            .filter_map(|(ident, node_id, res)| {
+                self.lifetime_res_to_generic_param(
+                    ident,
+                    node_id,
+                    res,
+                    hir::GenericParamSource::Binder,
+                )
+            })
+            .collect();
+        let arena = self.arena;
+        let explicit_generic_params =
+            self.lower_generic_params_mut(generic_params, hir::GenericParamSource::Binder);
+        arena.alloc_from_iter(explicit_generic_params.chain(extra_lifetimes.into_iter()))
     }
 
     fn with_dyn_type_scope<T>(&mut self, in_scope: bool, f: impl FnOnce(&mut Self) -> T) -> T {

compiler/rustc_attr_data_structures/src/attributes.rs

@@ -234,6 +234,7 @@ pub enum CfgEntry {
 pub enum AttributeKind {
     // tidy-alphabetical-start
     /// Represents `#[align(N)]`.
+    // FIXME(#82232, #143834): temporarily renamed to mitigate `#[align]` nameres ambiguity
     Align { align: Align, span: Span },
 
     /// Represents `#[rustc_allow_const_fn_unstable]`.

compiler/rustc_attr_parsing/src/attributes/codegen_attrs.rs

@@ -177,7 +177,8 @@ impl<S: Stage> AttributeParser<S> for NakedParser {
             sym::instruction_set,
             sym::repr,
             sym::rustc_std_internal_symbol,
-            sym::align,
+            // FIXME(#82232, #143834): temporarily renamed to mitigate `#[align]` nameres ambiguity
+            sym::rustc_align,
             // obviously compatible with self
             sym::naked,
             // documentation

compiler/rustc_attr_parsing/src/attributes/repr.rs

@@ -274,7 +274,7 @@ fn parse_alignment(node: &LitKind) -> Result<Align, &'static str> {
 pub(crate) struct AlignParser(Option<(Align, Span)>);
 
 impl AlignParser {
-    const PATH: &'static [Symbol] = &[sym::align];
+    const PATH: &'static [Symbol] = &[sym::rustc_align];
     const TEMPLATE: AttributeTemplate = template!(List: "<alignment in bytes>");
 
     fn parse<'c, S: Stage>(

compiler/rustc_codegen_llvm/src/back/lto.rs

@@ -654,6 +654,7 @@ pub(crate) fn run_pass_manager(
     // We then run the llvm_optimize function a second time, to optimize the code which we generated
     // in the enzyme differentiation pass.
     let enable_ad = config.autodiff.contains(&config::AutoDiff::Enable);
+    let enable_gpu = config.offload.contains(&config::Offload::Enable);
     let stage = if thin {
         write::AutodiffStage::PreAD
     } else {
@@ -668,6 +669,12 @@ pub(crate) fn run_pass_manager(
         write::llvm_optimize(cgcx, dcx, module, None, config, opt_level, opt_stage, stage)?;
     }
 
+    if enable_gpu && !thin {
+        let cx =
+            SimpleCx::new(module.module_llvm.llmod(), &module.module_llvm.llcx, cgcx.pointer_size);
+        crate::builder::gpu_offload::handle_gpu_code(cgcx, &cx);
+    }
+
     if cfg!(llvm_enzyme) && enable_ad && !thin {
         let cx =
             SimpleCx::new(module.module_llvm.llmod(), &module.module_llvm.llcx, cgcx.pointer_size);

compiler/rustc_codegen_llvm/src/builder.rs

@@ -3,6 +3,7 @@ use std::ops::Deref;
 use std::{iter, ptr};
 
 pub(crate) mod autodiff;
+pub(crate) mod gpu_offload;
 
 use libc::{c_char, c_uint, size_t};
 use rustc_abi as abi;
@@ -117,6 +118,74 @@ impl<'a, 'll, CX: Borrow<SCx<'ll>>> GenericBuilder<'a, 'll, CX> {
         }
         bx
     }
+
+    // The generic builder has less functionality and thus (unlike the other alloca) we can not
+    // easily jump to the beginning of the function to place our allocas there. We trust the user
+    // to manually do that. FIXME(offload): improve the genericCx and add more llvm wrappers to
+    // handle this.
+    pub(crate) fn direct_alloca(&mut self, ty: &'ll Type, align: Align, name: &str) -> &'ll Value {
+        let val = unsafe {
+            let alloca = llvm::LLVMBuildAlloca(self.llbuilder, ty, UNNAMED);
+            llvm::LLVMSetAlignment(alloca, align.bytes() as c_uint);
+            // Cast to default addrspace if necessary
+            llvm::LLVMBuildPointerCast(self.llbuilder, alloca, self.cx.type_ptr(), UNNAMED)
+        };
+        if name != "" {
+            let name = std::ffi::CString::new(name).unwrap();
+            llvm::set_value_name(val, &name.as_bytes());
+        }
+        val
+    }
+
+    pub(crate) fn inbounds_gep(
+        &mut self,
+        ty: &'ll Type,
+        ptr: &'ll Value,
+        indices: &[&'ll Value],
+    ) -> &'ll Value {
+        unsafe {
+            llvm::LLVMBuildGEPWithNoWrapFlags(
+                self.llbuilder,
+                ty,
+                ptr,
+                indices.as_ptr(),
+                indices.len() as c_uint,
+                UNNAMED,
+                GEPNoWrapFlags::InBounds,
+            )
+        }
+    }
+
+    pub(crate) fn store(&mut self, val: &'ll Value, ptr: &'ll Value, align: Align) -> &'ll Value {
+        debug!("Store {:?} -> {:?}", val, ptr);
+        assert_eq!(self.cx.type_kind(self.cx.val_ty(ptr)), TypeKind::Pointer);
+        unsafe {
+            let store = llvm::LLVMBuildStore(self.llbuilder, val, ptr);
+            llvm::LLVMSetAlignment(store, align.bytes() as c_uint);
+            store
+        }
+    }
+
+    pub(crate) fn load(&mut self, ty: &'ll Type, ptr: &'ll Value, align: Align) -> &'ll Value {
+        unsafe {
+            let load = llvm::LLVMBuildLoad2(self.llbuilder, ty, ptr, UNNAMED);
+            llvm::LLVMSetAlignment(load, align.bytes() as c_uint);
+            load
+        }
+    }
+
+    fn memset(&mut self, ptr: &'ll Value, fill_byte: &'ll Value, size: &'ll Value, align: Align) {
+        unsafe {
+            llvm::LLVMRustBuildMemSet(
+                self.llbuilder,
+                ptr,
+                align.bytes() as c_uint,
+                fill_byte,
+                size,
+                false,
+            );
+        }
+    }
 }
 
 /// Empty string, to be used where LLVM expects an instruction name, indicating