Auto merge of #37770 - pnkfelix:print-type-sizes, r=arielb1

Add debug flag `-Z print-type-sizes` for instrumention type/variant sizes

Add debug flag `-Z print-type-sizes` for instrumention type/variant sizes

This is meant to help with things like #36799 in a very local way; namely, once you have a hypothesis as to which types have a large population or are "too large", you can use `-Z print-type-sizes` to learn how large each type is, and how much each variant in an enum contributes to the size of that overall enum.

src/librustc/session/code_stats.rs

@@ -0,0 +1,173 @@
+// Copyright 2016 The Rust Project Developers. See the COPYRIGHT
+// file at the top-level directory of this distribution and at
+// http://rust-lang.org/COPYRIGHT.
+//
+// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
+// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
+// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
+// option. This file may not be copied, modified, or distributed
+// except according to those terms.
+
+use ty::AdtKind;
+use ty::layout::{Align, Size};
+
+use rustc_data_structures::fx::{FxHashSet};
+
+use std::cmp::{self, Ordering};
+
+#[derive(Clone, PartialEq, Eq, Hash, Debug)]
+pub struct VariantInfo {
+    pub name: Option<String>,
+    pub kind: SizeKind,
+    pub size: u64,
+    pub align: u64,
+    pub fields: Vec<FieldInfo>,
+}
+
+#[derive(Copy, Clone, PartialEq, Eq, Hash, Debug)]
+pub enum SizeKind { Exact, Min }
+
+#[derive(Clone, PartialEq, Eq, Hash, Debug)]
+pub struct FieldInfo {
+    pub name: String,
+    pub offset: u64,
+    pub size: u64,
+    pub align: u64,
+}
+
+impl From<AdtKind> for DataTypeKind {
+    fn from(kind: AdtKind) -> Self {
+        match kind {
+            AdtKind::Struct => DataTypeKind::Struct,
+            AdtKind::Enum => DataTypeKind::Enum,
+            AdtKind::Union => DataTypeKind::Union,
+        }
+    }
+}
+
+#[derive(Copy, Clone, PartialEq, Eq, Hash, Debug)]
+pub enum DataTypeKind {
+    Struct,
+    Union,
+    Enum,
+    Closure,
+}
+
+#[derive(PartialEq, Eq, Hash, Debug)]
+pub struct TypeSizeInfo {
+    pub kind: DataTypeKind,
+    pub type_description: String,
+    pub align: u64,
+    pub overall_size: u64,
+    pub opt_discr_size: Option<u64>,
+    pub variants: Vec<VariantInfo>,
+}
+
+#[derive(PartialEq, Eq, Debug)]
+pub struct CodeStats {
+    type_sizes: FxHashSet<TypeSizeInfo>,
+}
+
+impl CodeStats {
+    pub fn new() -> Self { CodeStats { type_sizes: FxHashSet() } }
+
+    pub fn record_type_size<S: ToString>(&mut self,
+                                         kind: DataTypeKind,
+                                         type_desc: S,
+                                         align: Align,
+                                         overall_size: Size,
+                                         opt_discr_size: Option<Size>,
+                                         variants: Vec<VariantInfo>) {
+        let info = TypeSizeInfo {
+            kind: kind,
+            type_description: type_desc.to_string(),
+            align: align.abi(),
+            overall_size: overall_size.bytes(),
+            opt_discr_size: opt_discr_size.map(|s| s.bytes()),
+            variants: variants,
+        };
+        self.type_sizes.insert(info);
+    }
+
+    pub fn print_type_sizes(&self) {
+        let mut sorted: Vec<_> = self.type_sizes.iter().collect();
+
+        // Primary sort: large-to-small.
+        // Secondary sort: description (dictionary order)
+        sorted.sort_by(|info1, info2| {
+            // (reversing cmp order to get large-to-small ordering)
+            match info2.overall_size.cmp(&info1.overall_size) {
+                Ordering::Equal => info1.type_description.cmp(&info2.type_description),
+                other => other,
+            }
+        });
+
+        for info in &sorted {
+            println!("print-type-size type: `{}`: {} bytes, alignment: {} bytes",
+                     info.type_description, info.overall_size, info.align);
+            let indent = "    ";
+
+            let discr_size = if let Some(discr_size) = info.opt_discr_size {
+                println!("print-type-size {}discriminant: {} bytes",
+                         indent, discr_size);
+                discr_size
+            } else {
+                0
+            };
+
+            // We start this at discr_size (rather than 0) because
+            // things like C-enums do not have variants but we still
+            // want the max_variant_size at the end of the loop below
+            // to reflect the presence of the discriminant.
+            let mut max_variant_size = discr_size;
+
+            let struct_like = match info.kind {
+                DataTypeKind::Struct | DataTypeKind::Closure => true,
+                DataTypeKind::Enum | DataTypeKind::Union => false,
+            };
+            for (i, variant_info) in info.variants.iter().enumerate() {
+                let VariantInfo { ref name, kind: _, align: _, size, ref fields } = *variant_info;
+                let indent = if !struct_like {
+                    let name = match name.as_ref() {
+                        Some(name) => format!("{}", name),
+                        None => format!("{}", i),
+                    };
+                    println!("print-type-size {}variant `{}`: {} bytes",
+                             indent, name, size - discr_size);
+                    "        "
+                } else {
+                    assert!(i < 1);
+                    "    "
+                };
+                max_variant_size = cmp::max(max_variant_size, size);
+
+                let mut min_offset = discr_size;
+                for field in fields {
+                    let FieldInfo { ref name, offset, size, align } = *field;
+
+                    // Include field alignment in output only if it caused padding injection
+                    if min_offset != offset {
+                        let pad = offset - min_offset;
+                        println!("print-type-size {}padding bytes: {}",
+                                 indent, pad);
+                        println!("print-type-size {}field `.{}`: {} bytes, alignment: {} bytes",
+                                 indent, name, size, align);
+                    } else {
+                        println!("print-type-size {}field `.{}`: {} bytes",
+                                 indent, name, size);
+                    }
+
+                    min_offset = offset + size;
+                }
+            }
+
+            assert!(max_variant_size <= info.overall_size,
+                    "max_variant_size {} !<= {} overall_size",
+                    max_variant_size, info.overall_size);
+            if max_variant_size < info.overall_size {
+                println!("print-type-size {}end padding bytes: {}",
+                         indent, info.overall_size - max_variant_size);
+            }
+        }
+    }
+}

src/librustc/session/config.rs

@@ -908,6 +908,8 @@ options! {DebuggingOptions, DebuggingSetter, basic_debugging_options,
           "keep the AST after lowering it to HIR"),
     show_span: Option<String> = (None, parse_opt_string, [TRACKED],
           "show spans for compiler debugging (expr|pat|ty)"),
+    print_type_sizes: bool = (false, parse_bool, [UNTRACKED],
+          "print layout information for each type encountered"),
     print_trans_items: Option<String> = (None, parse_opt_string, [UNTRACKED],
           "print the result of the translation item collection pass"),
     mir_opt_level: Option<usize> = (None, parse_opt_uint, [TRACKED],

src/librustc/session/mod.rs

@@ -8,6 +8,9 @@
 // option. This file may not be copied, modified, or distributed
 // except according to those terms.
 
+pub use self::code_stats::{CodeStats, DataTypeKind, FieldInfo};
+pub use self::code_stats::{SizeKind, TypeSizeInfo, VariantInfo};
+
 use dep_graph::DepGraph;
 use hir::def_id::{CrateNum, DefIndex};
 use hir::svh::Svh;
@@ -49,6 +52,7 @@ use std::fmt;
 use std::time::Duration;
 use libc::c_int;
 
+mod code_stats;
 pub mod config;
 pub mod filesearch;
 pub mod search_paths;
@@ -112,6 +116,9 @@ pub struct Session {
     /// Some measurements that are being gathered during compilation.
     pub perf_stats: PerfStats,
 
+    /// Data about code being compiled, gathered during compilation.
+    pub code_stats: RefCell<CodeStats>,
+
     next_node_id: Cell<ast::NodeId>,
 }
 
@@ -624,7 +631,8 @@ pub fn build_session_(sopts: config::Options,
             incr_comp_hashes_count: Cell::new(0),
             incr_comp_bytes_hashed: Cell::new(0),
             symbol_hash_time: Cell::new(Duration::from_secs(0)),
-        }
+        },
+        code_stats: RefCell::new(CodeStats::new()),
     };
 
     init_llvm(&sess);

src/librustc/ty/layout.rs

@@ -559,11 +559,14 @@ impl<'a, 'gcx, 'tcx> Struct {
 
             self.offsets.push(offset);
 
+            debug!("Struct::extend offset: {:?} field: {:?} {:?}", offset, field, field.size(dl));
 
             offset = offset.checked_add(field.size(dl), dl)
                            .map_or(Err(LayoutError::SizeOverflow(scapegoat)), Ok)?;
         }
 
+        debug!("Struct::extend min_size: {:?}", offset);
+
         self.min_size = offset;
 
         Ok(())
@@ -707,12 +710,16 @@ impl<'a, 'gcx, 'tcx> Union {
                      index, scapegoat);
             }
 
+            debug!("Union::extend field: {:?} {:?}", field, field.size(dl));
+
             if !self.packed {
                 self.align = self.align.max(field.align(dl));
             }
             self.min_size = cmp::max(self.min_size, field.size(dl));
         }
 
+        debug!("Union::extend min-size: {:?}", self.min_size);
+
         Ok(())
     }
 

src/librustc_driver/driver.rs

@@ -215,6 +215,10 @@ pub fn compile_input(sess: &Session,
         })??
     };
 
+    if sess.opts.debugging_opts.print_type_sizes {
+        sess.code_stats.borrow().print_type_sizes();
+    }
+
     let phase5_result = phase_5_run_llvm_passes(sess, &trans, &outputs);
 
     controller_entry_point!(after_llvm,

src/librustc_trans/adt.rs

@@ -247,6 +247,7 @@ fn generic_type_of<'a, 'tcx>(cx: &CrateContext<'a, 'tcx>,
             // of the size.
             let size = size.bytes();
             let align = align.abi();
+            assert!(align <= std::u32::MAX as u64);
             let discr_ty = Type::from_integer(cx, discr);
             let discr_size = discr.size().bytes();
             let padded_discr_size = roundup(discr_size, align as u32);