Auto merge of #112400 - WaffleLapkin:vtable_stats, r=compiler-errors

Collect VTable stats & add `-Zprint-vtable-sizes`

This is a bit hacky/buggy, but I'm not entirely sure how to fix it, so I want to ask reviewers for help...

To try this, use either of those:
- `cargo clean && RUSTFLAGS="-Zprint-vtable-sizes" cargo +toolchain b`
- `cargo clean && cargo rustc +toolchain -Zprint-vtable-sizes`
- `rustc +toolchain -Zprint-vtable-sizes ./file.rs`

Author: bors

compiler/rustc_driver_impl/src/lib.rs

@@ -430,6 +430,13 @@ fn run_compiler(
                 sess.code_stats.print_type_sizes();
             }
 
+            if sess.opts.unstable_opts.print_vtable_sizes {
+                let crate_name =
+                    compiler.session().opts.crate_name.as_deref().unwrap_or("<UNKNOWN_CRATE>");
+
+                sess.code_stats.print_vtable_sizes(crate_name);
+            }
+
             let linker = queries.linker()?;
             Ok(Some(linker))
         })?;

compiler/rustc_interface/src/interface.rs

@@ -333,6 +333,7 @@ pub fn run_compiler<R: Send>(config: Config, f: impl FnOnce(&Compiler) -> R + Se
                 };
 
                 let prof = compiler.sess.prof.clone();
+
                 prof.generic_activity("drop_compiler").run(move || drop(compiler));
                 r
             })

compiler/rustc_interface/src/passes.rs

@@ -24,6 +24,7 @@ use rustc_parse::{parse_crate_from_file, parse_crate_from_source_str, validate_a
 use rustc_passes::{self, hir_stats, layout_test};
 use rustc_plugin_impl as plugin;
 use rustc_resolve::Resolver;
+use rustc_session::code_stats::VTableSizeInfo;
 use rustc_session::config::{CrateType, Input, OutFileName, OutputFilenames, OutputType};
 use rustc_session::cstore::{MetadataLoader, Untracked};
 use rustc_session::output::filename_for_input;
@@ -866,6 +867,92 @@ fn analysis(tcx: TyCtxt<'_>, (): ()) -> Result<()> {
         sess.time("check_lint_expectations", || tcx.check_expectations(None));
     });
 
+    if sess.opts.unstable_opts.print_vtable_sizes {
+        let traits = tcx.traits(LOCAL_CRATE);
+
+        for &tr in traits {
+            if !tcx.check_is_object_safe(tr) {
+                continue;
+            }
+
+            let name = ty::print::with_no_trimmed_paths!(tcx.def_path_str(tr));
+
+            let mut first_dsa = true;
+
+            // Number of vtable entries, if we didn't have upcasting
+            let mut entries_ignoring_upcasting = 0;
+            // Number of vtable entries needed solely for upcasting
+            let mut entries_for_upcasting = 0;
+
+            let trait_ref = ty::Binder::dummy(ty::TraitRef::identity(tcx, tr));
+
+            // A slightly edited version of the code in `rustc_trait_selection::traits::vtable::vtable_entries`,
+            // that works without self type and just counts number of entries.
+            //
+            // Note that this is technically wrong, for traits which have associated types in supertraits:
+            //
+            //   trait A: AsRef<Self::T> + AsRef<()> { type T; }
+            //
+            // Without self type we can't normalize `Self::T`, so we can't know if `AsRef<Self::T>` and
+            // `AsRef<()>` are the same trait, thus we assume that those are different, and potentially
+            // over-estimate how many vtable entries there are.
+            //
+            // Similarly this is wrong for traits that have methods with possibly-impossible bounds.
+            // For example:
+            //
+            //   trait B<T> { fn f(&self) where T: Copy; }
+            //
+            // Here `dyn B<u8>` will have 4 entries, while `dyn B<String>` will only have 3.
+            // However, since we don't know `T`, we can't know if `T: Copy` holds or not,
+            // thus we lean on the bigger side and say it has 4 entries.
+            traits::vtable::prepare_vtable_segments(tcx, trait_ref, |segment| {
+                match segment {
+                    traits::vtable::VtblSegment::MetadataDSA => {
+                        // If this is the first dsa, it would be included either way,
+                        // otherwise it's needed for upcasting
+                        if std::mem::take(&mut first_dsa) {
+                            entries_ignoring_upcasting += 3;
+                        } else {
+                            entries_for_upcasting += 3;
+                        }
+                    }
+
+                    traits::vtable::VtblSegment::TraitOwnEntries { trait_ref, emit_vptr } => {
+                        // Lookup the shape of vtable for the trait.
+                        let own_existential_entries =
+                            tcx.own_existential_vtable_entries(trait_ref.def_id());
+
+                        // The original code here ignores the method if its predicates are impossible.
+                        // We can't really do that as, for example, all not trivial bounds on generic
+                        // parameters are impossible (since we don't know the parameters...),
+                        // see the comment above.
+                        entries_ignoring_upcasting += own_existential_entries.len();
+
+                        if emit_vptr {
+                            entries_for_upcasting += 1;
+                        }
+                    }
+                }
+
+                std::ops::ControlFlow::Continue::<std::convert::Infallible>(())
+            });
+
+            sess.code_stats.record_vtable_size(
+                tr,
+                &name,
+                VTableSizeInfo {
+                    trait_name: name.clone(),
+                    entries: entries_ignoring_upcasting + entries_for_upcasting,
+                    entries_ignoring_upcasting,
+                    entries_for_upcasting,
+                    upcasting_cost_percent: entries_for_upcasting as f64
+                        / entries_ignoring_upcasting as f64
+                        * 100.,
+                },
+            )
+        }
+    }
+
     Ok(())
 }
 

compiler/rustc_session/src/code_stats.rs

@@ -1,5 +1,6 @@
-use rustc_data_structures::fx::FxHashSet;
+use rustc_data_structures::fx::{FxHashMap, FxHashSet};
 use rustc_data_structures::sync::Lock;
+use rustc_span::def_id::DefId;
 use rustc_span::Symbol;
 use rustc_target::abi::{Align, Size};
 use std::cmp;
@@ -65,9 +66,29 @@ pub struct TypeSizeInfo {
     pub variants: Vec<VariantInfo>,
 }
 
+pub struct VTableSizeInfo {
+    pub trait_name: String,
+
+    /// Number of entries in a vtable with the current algorithm
+    /// (i.e. with upcasting).
+    pub entries: usize,
+
+    /// Number of entries in a vtable, as-if we did not have trait upcasting.
+    pub entries_ignoring_upcasting: usize,
+
+    /// Number of entries in a vtable needed solely for upcasting
+    /// (i.e. `entries - entries_ignoring_upcasting`).
+    pub entries_for_upcasting: usize,
+
+    /// Cost of having upcasting in % relative to the number of entries without
+    /// upcasting (i.e. `entries_for_upcasting / entries_ignoring_upcasting * 100%`).
+    pub upcasting_cost_percent: f64,
+}
+
 #[derive(Default)]
 pub struct CodeStats {
     type_sizes: Lock<FxHashSet<TypeSizeInfo>>,
+    vtable_sizes: Lock<FxHashMap<DefId, VTableSizeInfo>>,
 }
 
 impl CodeStats {
@@ -101,6 +122,14 @@ impl CodeStats {
         self.type_sizes.borrow_mut().insert(info);
     }
 
+    pub fn record_vtable_size(&self, trait_did: DefId, trait_name: &str, info: VTableSizeInfo) {
+        let prev = self.vtable_sizes.lock().insert(trait_did, info);
+        assert!(
+            prev.is_none(),
+            "size of vtable for `{trait_name}` ({trait_did:?}) is already recorded"
+        );
+    }
+
     pub fn print_type_sizes(&self) {
         let type_sizes = self.type_sizes.borrow();
         let mut sorted: Vec<_> = type_sizes.iter().collect();
@@ -196,4 +225,33 @@ impl CodeStats {
             }
         }
     }
+
+    pub fn print_vtable_sizes(&self, crate_name: &str) {
+        let mut infos = std::mem::take(&mut *self.vtable_sizes.lock())
+            .into_iter()
+            .map(|(_did, stats)| stats)
+            .collect::<Vec<_>>();
+
+        // Primary sort: cost % in reverse order (from largest to smallest)
+        // Secondary sort: trait_name
+        infos.sort_by(|a, b| {
+            a.upcasting_cost_percent
+                .total_cmp(&b.upcasting_cost_percent)
+                .reverse()
+                .then_with(|| a.trait_name.cmp(&b.trait_name))
+        });
+
+        for VTableSizeInfo {
+            trait_name,
+            entries,
+            entries_ignoring_upcasting,
+            entries_for_upcasting,
+            upcasting_cost_percent,
+        } in infos
+        {
+            println!(
+                r#"print-vtable-sizes {{ "crate_name": "{crate_name}", "trait_name": "{trait_name}", "entries": "{entries}", "entries_ignoring_upcasting": "{entries_ignoring_upcasting}", "entries_for_upcasting": "{entries_for_upcasting}", "upcasting_cost_percent": "{upcasting_cost_percent}" }}"#
+            );
+        }
+    }
 }

compiler/rustc_session/src/lib.rs

@@ -27,7 +27,7 @@ pub use lint::{declare_lint, declare_lint_pass, declare_tool_lint, impl_lint_pas
 pub use rustc_lint_defs as lint;
 pub mod parse;
 
-mod code_stats;
+pub mod code_stats;
 #[macro_use]
 pub mod config;
 pub mod cstore;

compiler/rustc_session/src/options.rs

@@ -1632,6 +1632,8 @@ options! {
         "print the result of the monomorphization collection pass"),
     print_type_sizes: bool = (false, parse_bool, [UNTRACKED],
         "print layout information for each type encountered (default: no)"),
+    print_vtable_sizes: bool = (false, parse_bool, [UNTRACKED],
+        "print size comparison between old and new vtable layouts (default: no)"),
     proc_macro_backtrace: bool = (false, parse_bool, [UNTRACKED],
          "show backtraces for panics during proc-macro execution (default: no)"),
     proc_macro_execution_strategy: ProcMacroExecutionStrategy = (ProcMacroExecutionStrategy::SameThread,

compiler/rustc_trait_selection/src/traits/mod.rs

@@ -21,7 +21,7 @@ mod structural_match;
 mod structural_normalize;
 #[cfg_attr(not(bootstrap), allow(hidden_glob_reexports))]
 mod util;
-mod vtable;
+pub mod vtable;
 pub mod wf;
 
 use crate::infer::outlives::env::OutlivesEnvironment;

compiler/rustc_trait_selection/src/traits/vtable.rs

@@ -15,13 +15,13 @@ use std::fmt::Debug;
 use std::ops::ControlFlow;
 
 #[derive(Clone, Debug)]
-pub(super) enum VtblSegment<'tcx> {
+pub enum VtblSegment<'tcx> {
     MetadataDSA,
     TraitOwnEntries { trait_ref: ty::PolyTraitRef<'tcx>, emit_vptr: bool },
 }
 
 /// Prepare the segments for a vtable
-pub(super) fn prepare_vtable_segments<'tcx, T>(
+pub fn prepare_vtable_segments<'tcx, T>(
     tcx: TyCtxt<'tcx>,
     trait_ref: ty::PolyTraitRef<'tcx>,
     mut segment_visitor: impl FnMut(VtblSegment<'tcx>) -> ControlFlow<T>,

tests/ui/traits/object/print_vtable_sizes.rs

@@ -0,0 +1,61 @@
+// check-pass
+// compile-flags: -Z print-vtable-sizes
+#![crate_type = "lib"]
+
+trait A<T: help::V>: AsRef<[T::V]> + AsMut<[T::V]> {}
+
+trait B<T>: AsRef<T> + AsRef<T> + AsRef<T> + AsRef<T> {}
+
+trait C {
+    fn x() {} // not object safe, shouldn't be reported
+}
+
+// This ideally should not have any upcasting cost,
+// but currently does due to a bug
+trait D: Send + Sync + help::MarkerWithSuper {}
+
+// This can't have no cost without reordering,
+// because `Super::f`.
+trait E: help::MarkerWithSuper + Send + Sync {}
+
+trait F {
+    fn a(&self);
+    fn b(&self);
+    fn c(&self);
+
+    fn d() -> Self
+    where
+        Self: Sized;
+}
+
+trait G: AsRef<u8> + AsRef<u16> + help::MarkerWithSuper {
+    fn a(&self);
+    fn b(&self);
+    fn c(&self);
+    fn d(&self);
+    fn e(&self);
+
+    fn f() -> Self
+    where
+        Self: Sized;
+}
+
+// Traits with the same name
+const _: () = {
+    trait S {}
+};
+const _: () = {
+    trait S {}
+};
+
+mod help {
+    pub trait V {
+        type V;
+    }
+
+    pub trait MarkerWithSuper: Super {}
+
+    pub trait Super {
+        fn f(&self);
+    }
+}

tests/ui/traits/object/print_vtable_sizes.stdout

@@ -0,0 +1,11 @@
+print-vtable-sizes { "crate_name": "<UNKNOWN_CRATE>", "trait_name": "D", "entries": "7", "entries_ignoring_upcasting": "4", "entries_for_upcasting": "3", "upcasting_cost_percent": "75" }
+print-vtable-sizes { "crate_name": "<UNKNOWN_CRATE>", "trait_name": "E", "entries": "6", "entries_ignoring_upcasting": "4", "entries_for_upcasting": "2", "upcasting_cost_percent": "50" }
+print-vtable-sizes { "crate_name": "<UNKNOWN_CRATE>", "trait_name": "G", "entries": "14", "entries_ignoring_upcasting": "11", "entries_for_upcasting": "3", "upcasting_cost_percent": "27.27272727272727" }
+print-vtable-sizes { "crate_name": "<UNKNOWN_CRATE>", "trait_name": "A", "entries": "6", "entries_ignoring_upcasting": "5", "entries_for_upcasting": "1", "upcasting_cost_percent": "20" }
+print-vtable-sizes { "crate_name": "<UNKNOWN_CRATE>", "trait_name": "B", "entries": "4", "entries_ignoring_upcasting": "4", "entries_for_upcasting": "0", "upcasting_cost_percent": "0" }
+print-vtable-sizes { "crate_name": "<UNKNOWN_CRATE>", "trait_name": "F", "entries": "6", "entries_ignoring_upcasting": "6", "entries_for_upcasting": "0", "upcasting_cost_percent": "0" }
+print-vtable-sizes { "crate_name": "<UNKNOWN_CRATE>", "trait_name": "_::S", "entries": "3", "entries_ignoring_upcasting": "3", "entries_for_upcasting": "0", "upcasting_cost_percent": "0" }
+print-vtable-sizes { "crate_name": "<UNKNOWN_CRATE>", "trait_name": "_::S", "entries": "3", "entries_ignoring_upcasting": "3", "entries_for_upcasting": "0", "upcasting_cost_percent": "0" }
+print-vtable-sizes { "crate_name": "<UNKNOWN_CRATE>", "trait_name": "help::MarkerWithSuper", "entries": "4", "entries_ignoring_upcasting": "4", "entries_for_upcasting": "0", "upcasting_cost_percent": "0" }
+print-vtable-sizes { "crate_name": "<UNKNOWN_CRATE>", "trait_name": "help::Super", "entries": "4", "entries_ignoring_upcasting": "4", "entries_for_upcasting": "0", "upcasting_cost_percent": "0" }
+print-vtable-sizes { "crate_name": "<UNKNOWN_CRATE>", "trait_name": "help::V", "entries": "3", "entries_ignoring_upcasting": "3", "entries_for_upcasting": "0", "upcasting_cost_percent": "0" }