Auto merge of #113562 - saethlin:larger-incr-comp-offset, r=nnethercote

Use u64 for incr comp allocation offsets

Fixes https://github.com/rust-lang/rust/issues/76037
Fixes https://github.com/rust-lang/rust/issues/95780
Fixes https://github.com/rust-lang/rust/issues/111613

These issues are all reporting ICEs caused by using `u32` to store offsets to allocations in the incremental compilation cache. This PR aims to lift that limitation by changing the offset type in question to `u64`.

There are two perf runs in this PR. The first reports a regression, and the second does not. The changes are the same in both. I rebased the PR then did the second perf run because I noticed that the primary regression in it was very commonly seen in spurious regression reports.

I do not know what the perf run will report when this is merged. I would not be surprised to see regression or neutral, but the cachegrind diffs for the regression point at `try_mark_previous_green` which is a common source of inexplicable regressions and I don't think should be perturbed by this PR.

I'm not opposed to adding a regression test such as
```rust
fn main() {
    println!("{}", [37; 1 << 30].len());
}
```
But that program takes 1 minute to compile and consumes 4.6 GB of memory then writes that much to disk. Is that a concerning amount of resource use for a test?

r? `@nnethercote`

5 files changed, 9 insertions, 6 deletions

diff --git a/compiler/rustc_metadata/src/rmeta/encoder.rs b/compiler/rustc_metadata/src/rmeta/encoder.rs
index b5f955d14fb..2785732727f 100644
--- a/compiler/rustc_metadata/src/rmeta/encoder.rs
+++ b/compiler/rustc_metadata/src/rmeta/encoder.rs
@@ -608,7 +608,7 @@ impl<'a, 'tcx> EncodeContext<'a, 'tcx> {
                 trace!("encoding {} further alloc ids", new_n - n);
                 for idx in n..new_n {
                     let id = self.interpret_allocs[idx];
-                    let pos = self.position() as u32;
+                    let pos = self.position() as u64;
                     interpret_alloc_index.push(pos);
                     interpret::specialized_encode_alloc_id(self, tcx, id);
                 }
diff --git a/compiler/rustc_metadata/src/rmeta/mod.rs b/compiler/rustc_metadata/src/rmeta/mod.rs
index 9cffd96f4a3..f6087fbe8f6 100644
--- a/compiler/rustc_metadata/src/rmeta/mod.rs
+++ b/compiler/rustc_metadata/src/rmeta/mod.rs
@@ -264,7 +264,7 @@ pub(crate) struct CrateRoot {
     traits: LazyArray<DefIndex>,
     impls: LazyArray<TraitImpls>,
     incoherent_impls: LazyArray<IncoherentImpls>,
-    interpret_alloc_index: LazyArray<u32>,
+    interpret_alloc_index: LazyArray<u64>,
     proc_macro_data: Option<ProcMacroData>,
 
     tables: LazyTables,
diff --git a/compiler/rustc_middle/src/mir/interpret/mod.rs b/compiler/rustc_middle/src/mir/interpret/mod.rs
index fbd667a87fe..69c15e9cc06 100644
--- a/compiler/rustc_middle/src/mir/interpret/mod.rs
+++ b/compiler/rustc_middle/src/mir/interpret/mod.rs
@@ -274,7 +274,7 @@ pub struct AllocDecodingState {
     // For each `AllocId`, we keep track of which decoding state it's currently in.
     decoding_state: Vec<Lock<State>>,
     // The offsets of each allocation in the data stream.
-    data_offsets: Vec<u32>,
+    data_offsets: Vec<u64>,
 }
 
 impl AllocDecodingState {
@@ -289,7 +289,7 @@ impl AllocDecodingState {
         AllocDecodingSession { state: self, session_id }
     }
 
-    pub fn new(data_offsets: Vec<u32>) -> Self {
+    pub fn new(data_offsets: Vec<u64>) -> Self {
         let decoding_state =
             std::iter::repeat_with(|| Lock::new(State::Empty)).take(data_offsets.len()).collect();
 
diff --git a/compiler/rustc_middle/src/query/on_disk_cache.rs b/compiler/rustc_middle/src/query/on_disk_cache.rs
index 8751d3b7890..995b2140f61 100644
--- a/compiler/rustc_middle/src/query/on_disk_cache.rs
+++ b/compiler/rustc_middle/src/query/on_disk_cache.rs
@@ -104,7 +104,9 @@ struct Footer {
     query_result_index: EncodedDepNodeIndex,
     side_effects_index: EncodedDepNodeIndex,
     // The location of all allocations.
-    interpret_alloc_index: Vec<u32>,
+    // Most uses only need values up to u32::MAX, but benchmarking indicates that we can use a u64
+    // without measurable overhead. This permits larger const allocations without ICEing.
+    interpret_alloc_index: Vec<u64>,
     // See `OnDiskCache.syntax_contexts`
     syntax_contexts: FxHashMap<u32, AbsoluteBytePos>,
     // See `OnDiskCache.expn_data`
@@ -301,7 +303,7 @@ impl<'sess> OnDiskCache<'sess> {
                     interpret_alloc_index.reserve(new_n - n);
                     for idx in n..new_n {
                         let id = encoder.interpret_allocs[idx];
-                        let pos: u32 = encoder.position().try_into().unwrap();
+                        let pos: u64 = encoder.position().try_into().unwrap();
                         interpret_alloc_index.push(pos);
                         interpret::specialized_encode_alloc_id(&mut encoder, tcx, id);
                     }
diff --git a/compiler/rustc_middle/src/ty/parameterized.rs b/compiler/rustc_middle/src/ty/parameterized.rs
index cc2b26a5e14..f1c38984296 100644
--- a/compiler/rustc_middle/src/ty/parameterized.rs
+++ b/compiler/rustc_middle/src/ty/parameterized.rs
@@ -52,6 +52,7 @@ trivially_parameterized_over_tcx! {
     usize,
     (),
     u32,
+    u64,
     bool,
     std::string::String,
     crate::metadata::ModChild,