Auto merge of #137111 - jhpratt:rollup-izd98sx, r=jhpratt

Rollup of 5 pull requests

Successful merges:

 - #135797 (Import initial generated 1.85 relnotes)
 - #135909 (Export kernel descriptor for amdgpu kernels)
 - #136545 (nvptx64: update default alignment to match LLVM 21)
 - #137092 (abi_unsupported_vector_types: say which type is the problem)
 - #137097 (Ignore Self in bounds check for associated types with Self:Sized)

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

8 files changed, 91 insertions, 48 deletions

diff --git a/compiler/rustc_codegen_llvm/src/context.rs b/compiler/rustc_codegen_llvm/src/context.rs
index 7fe527a4c07..e7952bc95e7 100644
--- a/compiler/rustc_codegen_llvm/src/context.rs
+++ b/compiler/rustc_codegen_llvm/src/context.rs
@@ -194,6 +194,12 @@ pub(crate) unsafe fn create_module<'ll>(
             target_data_layout = target_data_layout.replace("-i128:128", "");
         }
     }
+    if llvm_version < (21, 0, 0) {
+        if sess.target.arch == "nvptx64" {
+            // LLVM 21 updated the default layout on nvptx: https://github.com/llvm/llvm-project/pull/124961
+            target_data_layout = target_data_layout.replace("e-p6:32:32-i64", "e-i64");
+        }
+    }
 
     // Ensure the data-layout values hardcoded remain the defaults.
     {
diff --git a/compiler/rustc_codegen_ssa/src/back/linker.rs b/compiler/rustc_codegen_ssa/src/back/linker.rs
index 05d6ff35751..e3ace01c1eb 100644
--- a/compiler/rustc_codegen_ssa/src/back/linker.rs
+++ b/compiler/rustc_codegen_ssa/src/back/linker.rs
@@ -1776,6 +1776,7 @@ fn exported_symbols_for_non_proc_macro(tcx: TyCtxt<'_>, crate_type: CrateType) -
             symbols.push(symbol_export::exporting_symbol_name_for_instance_in_crate(
                 tcx, symbol, cnum,
             ));
+            symbol_export::extend_exported_symbols(&mut symbols, tcx, symbol, cnum);
         }
     });
 
diff --git a/compiler/rustc_codegen_ssa/src/back/symbol_export.rs b/compiler/rustc_codegen_ssa/src/back/symbol_export.rs
index 1dbaffaa577..12ee872d531 100644
--- a/compiler/rustc_codegen_ssa/src/back/symbol_export.rs
+++ b/compiler/rustc_codegen_ssa/src/back/symbol_export.rs
@@ -10,9 +10,10 @@ use rustc_middle::middle::exported_symbols::{
     ExportedSymbol, SymbolExportInfo, SymbolExportKind, SymbolExportLevel, metadata_symbol_name,
 };
 use rustc_middle::query::LocalCrate;
-use rustc_middle::ty::{self, GenericArgKind, GenericArgsRef, Instance, SymbolName, TyCtxt};
+use rustc_middle::ty::{self, GenericArgKind, GenericArgsRef, Instance, SymbolName, Ty, TyCtxt};
 use rustc_middle::util::Providers;
 use rustc_session::config::{CrateType, OomStrategy};
+use rustc_target::callconv::Conv;
 use rustc_target::spec::{SanitizerSet, TlsModel};
 use tracing::debug;
 
@@ -584,6 +585,42 @@ pub(crate) fn symbol_name_for_instance_in_crate<'tcx>(
     }
 }
 
+fn calling_convention_for_symbol<'tcx>(
+    tcx: TyCtxt<'tcx>,
+    symbol: ExportedSymbol<'tcx>,
+) -> (Conv, &'tcx [rustc_target::callconv::ArgAbi<'tcx, Ty<'tcx>>]) {
+    let instance = match symbol {
+        ExportedSymbol::NonGeneric(def_id) | ExportedSymbol::Generic(def_id, _)
+            if tcx.is_static(def_id) =>
+        {
+            None
+        }
+        ExportedSymbol::NonGeneric(def_id) => Some(Instance::mono(tcx, def_id)),
+        ExportedSymbol::Generic(def_id, args) => Some(Instance::new(def_id, args)),
+        // DropGlue always use the Rust calling convention and thus follow the target's default
+        // symbol decoration scheme.
+        ExportedSymbol::DropGlue(..) => None,
+        // AsyncDropGlueCtorShim always use the Rust calling convention and thus follow the
+        // target's default symbol decoration scheme.
+        ExportedSymbol::AsyncDropGlueCtorShim(..) => None,
+        // NoDefId always follow the target's default symbol decoration scheme.
+        ExportedSymbol::NoDefId(..) => None,
+        // ThreadLocalShim always follow the target's default symbol decoration scheme.
+        ExportedSymbol::ThreadLocalShim(..) => None,
+    };
+
+    instance
+        .map(|i| {
+            tcx.fn_abi_of_instance(
+                ty::TypingEnv::fully_monomorphized().as_query_input((i, ty::List::empty())),
+            )
+            .unwrap_or_else(|_| bug!("fn_abi_of_instance({i:?}) failed"))
+        })
+        .map(|fnabi| (fnabi.conv, &fnabi.args[..]))
+        // FIXME(workingjubilee): why don't we know the convention here?
+        .unwrap_or((Conv::Rust, &[]))
+}
+
 /// This is the symbol name of the given instance as seen by the linker.
 ///
 /// On 32-bit Windows symbols are decorated according to their calling conventions.
@@ -592,8 +629,6 @@ pub(crate) fn linking_symbol_name_for_instance_in_crate<'tcx>(
     symbol: ExportedSymbol<'tcx>,
     instantiating_crate: CrateNum,
 ) -> String {
-    use rustc_target::callconv::Conv;
-
     let mut undecorated = symbol_name_for_instance_in_crate(tcx, symbol, instantiating_crate);
 
     // thread local will not be a function call,
@@ -617,35 +652,7 @@ pub(crate) fn linking_symbol_name_for_instance_in_crate<'tcx>(
         _ => return undecorated,
     };
 
-    let instance = match symbol {
-        ExportedSymbol::NonGeneric(def_id) | ExportedSymbol::Generic(def_id, _)
-            if tcx.is_static(def_id) =>
-        {
-            None
-        }
-        ExportedSymbol::NonGeneric(def_id) => Some(Instance::mono(tcx, def_id)),
-        ExportedSymbol::Generic(def_id, args) => Some(Instance::new(def_id, args)),
-        // DropGlue always use the Rust calling convention and thus follow the target's default
-        // symbol decoration scheme.
-        ExportedSymbol::DropGlue(..) => None,
-        // AsyncDropGlueCtorShim always use the Rust calling convention and thus follow the
-        // target's default symbol decoration scheme.
-        ExportedSymbol::AsyncDropGlueCtorShim(..) => None,
-        // NoDefId always follow the target's default symbol decoration scheme.
-        ExportedSymbol::NoDefId(..) => None,
-        // ThreadLocalShim always follow the target's default symbol decoration scheme.
-        ExportedSymbol::ThreadLocalShim(..) => None,
-    };
-
-    let (conv, args) = instance
-        .map(|i| {
-            tcx.fn_abi_of_instance(
-                ty::TypingEnv::fully_monomorphized().as_query_input((i, ty::List::empty())),
-            )
-            .unwrap_or_else(|_| bug!("fn_abi_of_instance({i:?}) failed"))
-        })
-        .map(|fnabi| (fnabi.conv, &fnabi.args[..]))
-        .unwrap_or((Conv::Rust, &[]));
+    let (conv, args) = calling_convention_for_symbol(tcx, symbol);
 
     // Decorate symbols with prefixes, suffixes and total number of bytes of arguments.
     // Reference: https://docs.microsoft.com/en-us/cpp/build/reference/decorated-names?view=msvc-170
@@ -677,6 +684,27 @@ pub(crate) fn exporting_symbol_name_for_instance_in_crate<'tcx>(
     maybe_emutls_symbol_name(tcx, symbol, &undecorated).unwrap_or(undecorated)
 }
 
+/// On amdhsa, `gpu-kernel` functions have an associated metadata object with a `.kd` suffix.
+/// Add it to the symbols list for all kernel functions, so that it is exported in the linked
+/// object.
+pub(crate) fn extend_exported_symbols<'tcx>(
+    symbols: &mut Vec<String>,
+    tcx: TyCtxt<'tcx>,
+    symbol: ExportedSymbol<'tcx>,
+    instantiating_crate: CrateNum,
+) {
+    let (conv, _) = calling_convention_for_symbol(tcx, symbol);
+
+    if conv != Conv::GpuKernel || tcx.sess.target.os != "amdhsa" {
+        return;
+    }
+
+    let undecorated = symbol_name_for_instance_in_crate(tcx, symbol, instantiating_crate);
+
+    // Add the symbol for the kernel descriptor (with .kd suffix)
+    symbols.push(format!("{undecorated}.kd"));
+}
+
 fn maybe_emutls_symbol_name<'tcx>(
     tcx: TyCtxt<'tcx>,
     symbol: ExportedSymbol<'tcx>,
diff --git a/compiler/rustc_monomorphize/messages.ftl b/compiler/rustc_monomorphize/messages.ftl
index 540ce7cc4ce..463c42b69b1 100644
--- a/compiler/rustc_monomorphize/messages.ftl
+++ b/compiler/rustc_monomorphize/messages.ftl
@@ -1,17 +1,17 @@
 monomorphize_abi_error_disabled_vector_type_call =
-  this function call uses a SIMD vector type that (with the chosen ABI) requires the `{$required_feature}` target feature, which is not enabled in the caller
+  this function call uses SIMD vector type `{$ty}` which (with the chosen ABI) requires the `{$required_feature}` target feature, which is not enabled in the caller
   .label = function called here
   .help = consider enabling it globally (`-C target-feature=+{$required_feature}`) or locally (`#[target_feature(enable="{$required_feature}")]`)
 monomorphize_abi_error_disabled_vector_type_def =
-  this function definition uses a SIMD vector type that (with the chosen ABI) requires the `{$required_feature}` target feature, which is not enabled
+  this function definition uses SIMD vector type `{$ty}` which (with the chosen ABI) requires the `{$required_feature}` target feature, which is not enabled
   .label = function defined here
   .help = consider enabling it globally (`-C target-feature=+{$required_feature}`) or locally (`#[target_feature(enable="{$required_feature}")]`)
 
 monomorphize_abi_error_unsupported_vector_type_call =
-  this function call uses a SIMD vector type that is not currently supported with the chosen ABI
+  this function call uses SIMD vector type `{$ty}` which is not currently supported with the chosen ABI
   .label = function called here
 monomorphize_abi_error_unsupported_vector_type_def =
-  this function definition uses a SIMD vector type that is not currently supported with the chosen ABI
+  this function definition uses SIMD vector type `{$ty}` which is not currently supported with the chosen ABI
   .label = function defined here
 
 monomorphize_couldnt_dump_mono_stats =
diff --git a/compiler/rustc_monomorphize/src/errors.rs b/compiler/rustc_monomorphize/src/errors.rs
index fc8d63b5888..75687a80b82 100644
--- a/compiler/rustc_monomorphize/src/errors.rs
+++ b/compiler/rustc_monomorphize/src/errors.rs
@@ -1,6 +1,7 @@
 use std::path::PathBuf;
 
 use rustc_macros::{Diagnostic, LintDiagnostic};
+use rustc_middle::ty::Ty;
 use rustc_span::{Span, Symbol};
 
 #[derive(Diagnostic)]
@@ -75,6 +76,7 @@ pub(crate) struct AbiErrorDisabledVectorTypeDef<'a> {
     #[label]
     pub span: Span,
     pub required_feature: &'a str,
+    pub ty: Ty<'a>,
 }
 
 #[derive(LintDiagnostic)]
@@ -84,18 +86,21 @@ pub(crate) struct AbiErrorDisabledVectorTypeCall<'a> {
     #[label]
     pub span: Span,
     pub required_feature: &'a str,
+    pub ty: Ty<'a>,
 }
 
 #[derive(LintDiagnostic)]
 #[diag(monomorphize_abi_error_unsupported_vector_type_def)]
-pub(crate) struct AbiErrorUnsupportedVectorTypeDef {
+pub(crate) struct AbiErrorUnsupportedVectorTypeDef<'a> {
     #[label]
     pub span: Span,
+    pub ty: Ty<'a>,
 }
 
 #[derive(LintDiagnostic)]
 #[diag(monomorphize_abi_error_unsupported_vector_type_call)]
-pub(crate) struct AbiErrorUnsupportedVectorTypeCall {
+pub(crate) struct AbiErrorUnsupportedVectorTypeCall<'a> {
     #[label]
     pub span: Span,
+    pub ty: Ty<'a>,
 }
diff --git a/compiler/rustc_monomorphize/src/mono_checks/abi_check.rs b/compiler/rustc_monomorphize/src/mono_checks/abi_check.rs
index a0be7f11d70..36cd3e00b76 100644
--- a/compiler/rustc_monomorphize/src/mono_checks/abi_check.rs
+++ b/compiler/rustc_monomorphize/src/mono_checks/abi_check.rs
@@ -34,7 +34,7 @@ fn do_check_abi<'tcx>(
     tcx: TyCtxt<'tcx>,
     abi: &FnAbi<'tcx, Ty<'tcx>>,
     target_feature_def: DefId,
-    mut emit_err: impl FnMut(Option<&'static str>),
+    mut emit_err: impl FnMut(Ty<'tcx>, Option<&'static str>),
 ) {
     let feature_def = tcx.sess.target.features_for_correct_vector_abi();
     let codegen_attrs = tcx.codegen_fn_attrs(target_feature_def);
@@ -45,7 +45,7 @@ fn do_check_abi<'tcx>(
             let feature = match feature_def.iter().find(|(bits, _)| size.bits() <= *bits) {
                 Some((_, feature)) => feature,
                 None => {
-                    emit_err(None);
+                    emit_err(arg_abi.layout.ty, None);
                     continue;
                 }
             };
@@ -53,7 +53,7 @@ fn do_check_abi<'tcx>(
             if !tcx.sess.unstable_target_features.contains(&feature_sym)
                 && !codegen_attrs.target_features.iter().any(|x| x.name == feature_sym)
             {
-                emit_err(Some(&feature));
+                emit_err(arg_abi.layout.ty, Some(&feature));
             }
         }
     }
@@ -69,21 +69,21 @@ fn check_instance_abi<'tcx>(tcx: TyCtxt<'tcx>, instance: Instance<'tcx>) {
         // function.
         return;
     };
-    do_check_abi(tcx, abi, instance.def_id(), |required_feature| {
+    do_check_abi(tcx, abi, instance.def_id(), |ty, required_feature| {
         let span = tcx.def_span(instance.def_id());
         if let Some(required_feature) = required_feature {
             tcx.emit_node_span_lint(
                 ABI_UNSUPPORTED_VECTOR_TYPES,
                 CRATE_HIR_ID,
                 span,
-                AbiErrorDisabledVectorTypeDef { span, required_feature },
+                AbiErrorDisabledVectorTypeDef { span, required_feature, ty },
             );
         } else {
             tcx.emit_node_span_lint(
                 ABI_UNSUPPORTED_VECTOR_TYPES,
                 CRATE_HIR_ID,
                 span,
-                AbiErrorUnsupportedVectorTypeDef { span },
+                AbiErrorUnsupportedVectorTypeDef { span, ty },
             );
         }
     })
@@ -123,20 +123,20 @@ fn check_call_site_abi<'tcx>(
         // ABI failed to compute; this will not get through codegen.
         return;
     };
-    do_check_abi(tcx, callee_abi, caller.def_id(), |required_feature| {
+    do_check_abi(tcx, callee_abi, caller.def_id(), |ty, required_feature| {
         if let Some(required_feature) = required_feature {
             tcx.emit_node_span_lint(
                 ABI_UNSUPPORTED_VECTOR_TYPES,
                 CRATE_HIR_ID,
                 span,
-                AbiErrorDisabledVectorTypeCall { span, required_feature },
+                AbiErrorDisabledVectorTypeCall { span, required_feature, ty },
             );
         } else {
             tcx.emit_node_span_lint(
                 ABI_UNSUPPORTED_VECTOR_TYPES,
                 CRATE_HIR_ID,
                 span,
-                AbiErrorUnsupportedVectorTypeCall { span },
+                AbiErrorUnsupportedVectorTypeCall { span, ty },
             );
         }
     });
diff --git a/compiler/rustc_target/src/spec/targets/nvptx64_nvidia_cuda.rs b/compiler/rustc_target/src/spec/targets/nvptx64_nvidia_cuda.rs
index 289251e906a..80bfa358243 100644
--- a/compiler/rustc_target/src/spec/targets/nvptx64_nvidia_cuda.rs
+++ b/compiler/rustc_target/src/spec/targets/nvptx64_nvidia_cuda.rs
@@ -5,7 +5,7 @@ use crate::spec::{
 pub(crate) fn target() -> Target {
     Target {
         arch: "nvptx64".into(),
-        data_layout: "e-i64:64-i128:128-v16:16-v32:32-n16:32:64".into(),
+        data_layout: "e-p6:32:32-i64:64-i128:128-v16:16-v32:32-n16:32:64".into(),
         llvm_target: "nvptx64-nvidia-cuda".into(),
         metadata: crate::spec::TargetMetadata {
             description: Some("--emit=asm generates PTX code that runs on NVIDIA GPUs".into()),
diff --git a/compiler/rustc_trait_selection/src/traits/dyn_compatibility.rs b/compiler/rustc_trait_selection/src/traits/dyn_compatibility.rs
index 617bc87a9d2..a0df74835bc 100644
--- a/compiler/rustc_trait_selection/src/traits/dyn_compatibility.rs
+++ b/compiler/rustc_trait_selection/src/traits/dyn_compatibility.rs
@@ -187,7 +187,10 @@ fn predicates_reference_self(
 fn bounds_reference_self(tcx: TyCtxt<'_>, trait_def_id: DefId) -> SmallVec<[Span; 1]> {
     tcx.associated_items(trait_def_id)
         .in_definition_order()
+        // We're only looking at associated type bounds
         .filter(|item| item.kind == ty::AssocKind::Type)
+        // Ignore GATs with `Self: Sized`
+        .filter(|item| !tcx.generics_require_sized_self(item.def_id))
         .flat_map(|item| tcx.explicit_item_bounds(item.def_id).iter_identity_copied())
         .filter_map(|(clause, sp)| {
             // Item bounds *can* have self projections, since they never get