Carry the `b_offset` inside `BackendRepr::ScalarPair` by scottmcm · Pull Request #158666 · rust-lang/rust

scottmcm · 2026-07-01T16:28:12Z

Inspired by rust-lang/compiler-team#1007 but doesn't actually change any of the layout rules just yet.

This turned out to be a nice change even if we didn't use the extra flexibility, IMHO, because it allowed so many things like

@@ -222,12 +224,12 @@ fn from_const_alloc<Bx: BuilderMethods<'a, 'tcx, Value = V>>(
                 let val = read_scalar(offset, size, s, bx.immediate_backend_type(layout));
                 OperandRef { val: OperandValue::Immediate(val), layout, move_annotation: None }
             }
-            BackendRepr::ScalarPair(
-                a @ abi::Scalar::Initialized { .. },
-                b @ abi::Scalar::Initialized { .. },
-            ) => {
+            BackendRepr::ScalarPair {
+                a: a @ abi::Scalar::Initialized { .. },
+                b: b @ abi::Scalar::Initialized { .. },
+                b_offset,
+            } => {
                 let (a_size, b_size) = (a.size(bx), b.size(bx));
-                let b_offset = (offset + a_size).align_to(b.default_align(bx).abi);
                 assert!(b_offset.bytes() > 0);
                 let a_val = read_scalar(
                     offset,

as oh my was that little magic incantation copy-pasted all over the place.

Apologies for the pretty-giant PR. I tried to make it as direct a change as I could: if it was (..) before it's { .. } now, if it was (_, _) before it's { a: _, b: _, b_offset: _ } now. I kept the names the same so the code lines were unchanged even if normally I might have just renamed things, etc. I'll add some inline notes for places of particular interest.

r? @workingjubilee

rustbot · 2026-07-01T16:28:15Z

rustc_codegen_cranelift is developed in its own repository. If possible, consider making this change to rust-lang/rustc_codegen_cranelift instead.

cc @bjorn3

Some changes occurred to MIR optimizations

cc @rust-lang/wg-mir-opt

Some changes occurred to the CTFE machinery

cc @RalfJung, @oli-obk, @lcnr

Some changes occurred to the CTFE / Miri interpreter

cc @rust-lang/miri

rustc_codegen_gcc is developed in its own repository. If possible, consider making this change to rust-lang/rustc_codegen_gcc instead.

cc @antoyo, @GuillaumeGomez

This PR changes rustc_public

cc @oli-obk, @celinval, @ouz-a, @makai410

Some changes occurred in compiler/rustc_codegen_llvm/src/builder/autodiff.rs

cc @ZuseZ4

rustbot · 2026-07-01T16:28:18Z

workingjubilee is currently at their maximum review capacity.
They may take a while to respond.

scottmcm · 2026-07-01T16:29:48Z

+            ) => {
+                l1.primitive() == r1.primitive()
+                    && l2.primitive() == r2.primitive()
+                    && l_offset == r_offset


note: since this is an eq-like method, I also check that the offsets are the same here. (They always will be right now, but it makes sense to check it regardless IMHO.)

View changes since the review

scottmcm · 2026-07-01T16:45:06Z

+                abi::BackendRepr::ScalarPair { a: out_a, b: out_b, b_offset: out_offset },
+            ) if in_a.size(cx) == out_a.size(cx)
+                && in_b.size(cx) == out_b.size(cx)
+                && in_offset == out_offset =>


note: when transmuting we only want to transmute the immediates individually when the second one is at the same offset in the source and destination.

(It'll probably be UB if they're not, but not necessarily since the out_b might allow uninit, so it's not worth trying to do something smart for those cases. They can fall through to the via-alloca implementation.)

View changes since the review

oh cool, so the code is now more correct also! neat.

I think it was correct before as well because same scalars implied same offset, but it's clearer that it's correct now, and definitely more resiliently correct to subtle changes.

If there is any target where we have scalars of equal size but different alignment (u64 vs f64?), this was wrong before. But I agree the new code is right. :)

scottmcm · 2026-07-01T16:45:54Z

-            }
+            (
+                Immediate::ScalarPair(a_val, b_val),
+                BackendRepr::ScalarPair { a: _, b: _, b_offset },


Note: by not needing to recompute the offset the two scalars actually ended up unused here.

View changes since the review

scottmcm · 2026-07-01T16:48:20Z

                                a1.size(&self.ecx) == a2.size(&self.ecx)
                                    && b1.size(&self.ecx) == b2.size(&self.ecx)
-                                    // The alignment of the second component determines its offset, so that also needs to match.
-                                    && b1.default_align(&self.ecx) == b2.default_align(&self.ecx)


Note: like the other spot this is thinking about transmutes, so now we can check the offset directly instead of indirectly via the default alignment.

View changes since the review

neat side effect!

scottmcm · 2026-07-01T16:53:53Z

                    "`ScalarPair` second field at bad offset in {inner:#?}",
                );
+                assert_eq!(
+                    b_offset, field2_offset,


Note: for this PR I kept all the existing checks here, since it's not actually changing any layouts. (That's wanting on the MCP FCP.) Just added this new one that the offset stored in the variant matches the expected field offset.

View changes since the review

rustbot · 2026-07-01T17:14:54Z

clippy is developed in its own repository. If possible, consider making this change to rust-lang/rust-clippy instead.

cc @rust-lang/clippy

rustbot · 2026-07-01T17:48:48Z

miri is developed in its own repository. If possible, consider making this change to rust-lang/miri instead.

cc @rust-lang/miri

workingjubilee

added some incantations to the PR since the code is now a bit short on them. :^)

a couple minor wording nitpicks, because you wrote r? workingjubilee and I'd hate to disappoint. I will give it another scan once you're done fighting CI and r+ then, but it seems good! my eyes have indeed confirmed "this sure is a lot of whitespace changes thanks to rustfmt".

oh, the callconv code does need a further scan, I'll get to that about then but I don't expect it to need any changes.

View changes since this review

workingjubilee · 2026-07-01T17:56:26Z

-                Immediate::from(if offset.bytes() == 0 {
-                    a_val
-                } else {
-                    assert_eq!(offset, a.size(cx).align_to(b.default_align(cx).abi));


by the shades of the Seraphim!

Oh you found some of the more cursed code in the interpreter. ;)

But, this is not on my "needs a rewrite" list. It's all inherent complexity that I think exists for a good reason. But I may of course have missed a way to do this nicer.

workingjubilee · 2026-07-01T17:56:46Z

+                let BackendRepr::ScalarPair { a: _, b: _, b_offset } = dest.layout.backend_repr
+                else {
                    bug!("store_with_flags: invalid ScalarPair layout: {:#?}", dest.layout);
                };
-                let b_offset = a_scalar.size(bx).align_to(b_scalar.default_align(bx).abi);


by the hoary hosts of Hoggoth!

workingjubilee · 2026-07-01T17:57:15Z

                        assert_eq!(field.size, a.size(bx.cx()));
                        (Some(a), a_llval)
                    } else {
-                        assert_eq!(offset, a.size(bx.cx()).align_to(b.default_align(bx.cx()).abi));


by the crimson bands of Cyttorak!

workingjubilee · 2026-07-01T18:00:50Z

                                a1.size(&self.ecx) == a2.size(&self.ecx)
                                    && b1.size(&self.ecx) == b2.size(&self.ecx)
-                                    // The alignment of the second component determines its offset, so that also needs to match.
-                                    && b1.default_align(&self.ecx) == b2.default_align(&self.ecx)


neat side effect!

workingjubilee · 2026-07-01T18:03:23Z

            BackendRepr::Scalar(scalar) => PassMode::Direct(scalar_attrs(scalar, Size::ZERO)),
-            BackendRepr::ScalarPair(a, b) => PassMode::Pair(
-                scalar_attrs(a, Size::ZERO),
-                scalar_attrs(b, a.size(cx).align_to(b.default_align(cx).abi)),
-            ),
+            BackendRepr::ScalarPair { a, b, b_offset } => {
+                PassMode::Pair(scalar_attrs(a, Size::ZERO), scalar_attrs(b, b_offset))
+            }


chintap ...I need to look more closely at the surrounding code here because this is actually something that makes me wonder how correct it is.

Or was, for that matter.

Rechecked the maze of code this is entangled with. Yes, this version is correct, and will remain correct.

workingjubilee · 2026-07-01T18:10:33Z

+                abi::BackendRepr::ScalarPair { a: out_a, b: out_b, b_offset: out_offset },
+            ) if in_a.size(cx) == out_a.size(cx)
+                && in_b.size(cx) == out_b.size(cx)
+                && in_offset == out_offset =>


oh cool, so the code is now more correct also! neat.

oli-obk · 2026-07-01T18:26:38Z

@bors try @rust-timer queue

Carry the `b_offset` inside `BackendRepr::ScalarPair`

rust-bors · 2026-07-01T20:37:03Z

☀️ Try build successful (CI)
Build commit: e1ccfbf (e1ccfbf5eacfbdc172c8ed08bfba6e211f662bd7)
Base parent: e2b71ad (e2b71ade2db4ea263ab0d561d889f3e3795a500d)

rust-timer · 2026-07-01T21:17:37Z

Finished benchmarking commit (e1ccfbf): comparison URL.

Overall result: no relevant changes - no action needed

Benchmarking means the PR may be perf-sensitive. Consider adding rollup=never if this change is not fit for rolling up.

@rustbot label: -S-waiting-on-perf -perf-regression

Instruction count

This perf run didn't have relevant results for this metric.

Max RSS (memory usage)

Results (primary -6.9%, secondary -4.3%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-6.9%	[-6.9%, -6.9%]	1
Improvements ✅ (secondary)	-4.3%	[-8.8%, -2.1%]	4
All ❌✅ (primary)	-6.9%	[-6.9%, -6.9%]	1

Cycles

Results (primary -2.3%, secondary -3.6%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-2.3%	[-2.8%, -2.0%]	3
Improvements ✅ (secondary)	-3.6%	[-4.0%, -3.2%]	2
All ❌✅ (primary)	-2.3%	[-2.8%, -2.0%]	3

Binary size

This perf run didn't have relevant results for this metric.

Bootstrap: 490.049s -> 485.092s (-1.01%)
Artifact size: 393.28 MiB -> 393.26 MiB (-0.01%)

RalfJung · 2026-07-08T06:44:42Z

@bors r=workingjubilee

rust-bors · 2026-07-08T06:44:45Z

📌 Commit 47a510b has been approved by workingjubilee

It is now in the queue for this repository.

@workingjubilee

…ingjubilee Carry the `b_offset` inside `BackendRepr::ScalarPair` Inspired by rust-lang/compiler-team#1007 but doesn't actually change any of the layout rules just yet. This turned out to be a nice change even if we didn't use the extra flexibility, IMHO, because it allowed so many things like ```diff @@ -222,12 +224,12 @@ fn from_const_alloc<Bx: BuilderMethods<'a, 'tcx, Value = V>>( let val = read_scalar(offset, size, s, bx.immediate_backend_type(layout)); OperandRef { val: OperandValue::Immediate(val), layout, move_annotation: None } } - BackendRepr::ScalarPair( - a @ abi::Scalar::Initialized { .. }, - b @ abi::Scalar::Initialized { .. }, - ) => { + BackendRepr::ScalarPair { + a: a @ abi::Scalar::Initialized { .. }, + b: b @ abi::Scalar::Initialized { .. }, + b_offset, + } => { let (a_size, b_size) = (a.size(bx), b.size(bx)); - let b_offset = (offset + a_size).align_to(b.default_align(bx).abi); assert!(b_offset.bytes() > 0); let a_val = read_scalar( offset, ``` as *oh my* was that little magic incantation copy-pasted all over the place. Apologies for the pretty-giant PR. I tried to make it as direct a change as I could: if it was `(..)` before it's `{ .. }` now, if it was `(_, _)` before it's `{ a: _, b: _, b_offset: _ }` now. I kept the names the same so the code lines were unchanged even if normally I might have just renamed things, etc. I'll add some inline notes for places of particular interest. r? @workingjubilee

@workingjubilee

…ingjubilee Carry the `b_offset` inside `BackendRepr::ScalarPair` Inspired by rust-lang/compiler-team#1007 but doesn't actually change any of the layout rules just yet. This turned out to be a nice change even if we didn't use the extra flexibility, IMHO, because it allowed so many things like ```diff @@ -222,12 +224,12 @@ fn from_const_alloc<Bx: BuilderMethods<'a, 'tcx, Value = V>>( let val = read_scalar(offset, size, s, bx.immediate_backend_type(layout)); OperandRef { val: OperandValue::Immediate(val), layout, move_annotation: None } } - BackendRepr::ScalarPair( - a @ abi::Scalar::Initialized { .. }, - b @ abi::Scalar::Initialized { .. }, - ) => { + BackendRepr::ScalarPair { + a: a @ abi::Scalar::Initialized { .. }, + b: b @ abi::Scalar::Initialized { .. }, + b_offset, + } => { let (a_size, b_size) = (a.size(bx), b.size(bx)); - let b_offset = (offset + a_size).align_to(b.default_align(bx).abi); assert!(b_offset.bytes() > 0); let a_val = read_scalar( offset, ``` as *oh my* was that little magic incantation copy-pasted all over the place. Apologies for the pretty-giant PR. I tried to make it as direct a change as I could: if it was `(..)` before it's `{ .. }` now, if it was `(_, _)` before it's `{ a: _, b: _, b_offset: _ }` now. I kept the names the same so the code lines were unchanged even if normally I might have just renamed things, etc. I'll add some inline notes for places of particular interest. r? @workingjubilee

@workingjubilee

…ingjubilee Carry the `b_offset` inside `BackendRepr::ScalarPair` Inspired by rust-lang/compiler-team#1007 but doesn't actually change any of the layout rules just yet. This turned out to be a nice change even if we didn't use the extra flexibility, IMHO, because it allowed so many things like ```diff @@ -222,12 +224,12 @@ fn from_const_alloc<Bx: BuilderMethods<'a, 'tcx, Value = V>>( let val = read_scalar(offset, size, s, bx.immediate_backend_type(layout)); OperandRef { val: OperandValue::Immediate(val), layout, move_annotation: None } } - BackendRepr::ScalarPair( - a @ abi::Scalar::Initialized { .. }, - b @ abi::Scalar::Initialized { .. }, - ) => { + BackendRepr::ScalarPair { + a: a @ abi::Scalar::Initialized { .. }, + b: b @ abi::Scalar::Initialized { .. }, + b_offset, + } => { let (a_size, b_size) = (a.size(bx), b.size(bx)); - let b_offset = (offset + a_size).align_to(b.default_align(bx).abi); assert!(b_offset.bytes() > 0); let a_val = read_scalar( offset, ``` as *oh my* was that little magic incantation copy-pasted all over the place. Apologies for the pretty-giant PR. I tried to make it as direct a change as I could: if it was `(..)` before it's `{ .. }` now, if it was `(_, _)` before it's `{ a: _, b: _, b_offset: _ }` now. I kept the names the same so the code lines were unchanged even if normally I might have just renamed things, etc. I'll add some inline notes for places of particular interest. r? @workingjubilee

…uwer Rollup of 18 pull requests Successful merges: - #158871 (add relnotes for 1.97.0) - #150946 (intrinsics: Add a fallback for non-const libm float functions) - #158617 (allow mGCA const arguments to fall back to anon consts) - #158645 (Fix splat ICEs and ban it in closures) - #158655 (Fix coroutine MIR saved local remapping) - #158666 (Carry the `b_offset` inside `BackendRepr::ScalarPair`) - #158920 (Update wasm-component-ld to 0.5.26) - #158926 (wrapping_sh* methods: clarify underspecified reference) - #158927 (add core test run with `-Zforce-intrinsic-fallback`) - #151379 (Stabilize `VecDeque::retain_back` from `truncate_front`) - #158807 (Add regression test for CString::clone_into unwind safety) - #158862 (Fix the span for parameter suggestion ) - #158883 (tests: fix enum-match.rs to handle LLVM 23) - #158894 (Make the ordering of non-terminal binds in ambiguity error messages deterministic) - #158902 (add codegen test for range length bound propagation) - #158913 (Update `browser-ui-test` version to `0.24.1`) - #158935 (std: support real fd methods on Emscripten) - #158951 (Merge three `MaxUniverse`s into one)

@workingjubilee

…ingjubilee Carry the `b_offset` inside `BackendRepr::ScalarPair` Inspired by rust-lang/compiler-team#1007 but doesn't actually change any of the layout rules just yet. This turned out to be a nice change even if we didn't use the extra flexibility, IMHO, because it allowed so many things like ```diff @@ -222,12 +224,12 @@ fn from_const_alloc<Bx: BuilderMethods<'a, 'tcx, Value = V>>( let val = read_scalar(offset, size, s, bx.immediate_backend_type(layout)); OperandRef { val: OperandValue::Immediate(val), layout, move_annotation: None } } - BackendRepr::ScalarPair( - a @ abi::Scalar::Initialized { .. }, - b @ abi::Scalar::Initialized { .. }, - ) => { + BackendRepr::ScalarPair { + a: a @ abi::Scalar::Initialized { .. }, + b: b @ abi::Scalar::Initialized { .. }, + b_offset, + } => { let (a_size, b_size) = (a.size(bx), b.size(bx)); - let b_offset = (offset + a_size).align_to(b.default_align(bx).abi); assert!(b_offset.bytes() > 0); let a_val = read_scalar( offset, ``` as *oh my* was that little magic incantation copy-pasted all over the place. Apologies for the pretty-giant PR. I tried to make it as direct a change as I could: if it was `(..)` before it's `{ .. }` now, if it was `(_, _)` before it's `{ a: _, b: _, b_offset: _ }` now. I kept the names the same so the code lines were unchanged even if normally I might have just renamed things, etc. I'll add some inline notes for places of particular interest. r? @workingjubilee

@workingjubilee

…ingjubilee Carry the `b_offset` inside `BackendRepr::ScalarPair` Inspired by rust-lang/compiler-team#1007 but doesn't actually change any of the layout rules just yet. This turned out to be a nice change even if we didn't use the extra flexibility, IMHO, because it allowed so many things like ```diff @@ -222,12 +224,12 @@ fn from_const_alloc<Bx: BuilderMethods<'a, 'tcx, Value = V>>( let val = read_scalar(offset, size, s, bx.immediate_backend_type(layout)); OperandRef { val: OperandValue::Immediate(val), layout, move_annotation: None } } - BackendRepr::ScalarPair( - a @ abi::Scalar::Initialized { .. }, - b @ abi::Scalar::Initialized { .. }, - ) => { + BackendRepr::ScalarPair { + a: a @ abi::Scalar::Initialized { .. }, + b: b @ abi::Scalar::Initialized { .. }, + b_offset, + } => { let (a_size, b_size) = (a.size(bx), b.size(bx)); - let b_offset = (offset + a_size).align_to(b.default_align(bx).abi); assert!(b_offset.bytes() > 0); let a_val = read_scalar( offset, ``` as *oh my* was that little magic incantation copy-pasted all over the place. Apologies for the pretty-giant PR. I tried to make it as direct a change as I could: if it was `(..)` before it's `{ .. }` now, if it was `(_, _)` before it's `{ a: _, b: _, b_offset: _ }` now. I kept the names the same so the code lines were unchanged even if normally I might have just renamed things, etc. I'll add some inline notes for places of particular interest. r? @workingjubilee

…uwer Rollup of 25 pull requests Successful merges: - #158871 (add relnotes for 1.97.0) - #158968 (stdarch subtree update) - #154445 (rustdoc: Represent `--output-format=json` coverage and ir differently) - #156370 (Reject linked dylib EII default overrides) - #157153 (allow `Allocator`s to be used as `#[global_allocator]`s) - #158495 (Rename HAS_CT_PROJECTION to HAS_CONST_ALIAS) - #158617 (allow mGCA const arguments to fall back to anon consts) - #158645 (Fix splat ICEs and ban it in closures) - #158655 (Fix coroutine MIR saved local remapping) - #158666 (Carry the `b_offset` inside `BackendRepr::ScalarPair`) - #158912 (Introduce new bootstrap config section for PGO configuration) - #158920 (Update wasm-component-ld to 0.5.26) - #158926 (wrapping_sh* methods: clarify underspecified reference) - #158927 (add core test run with `-Zforce-intrinsic-fallback`) - #158932 (Do not build the compiler when invoking `x perf compare`) - #158937 (Emit the emscripten entry point as `__main_argc_argv`) - #151379 (Stabilize `VecDeque::retain_back` from `truncate_front`) - #156548 ( Library support for aarch64-unknown-linux-pauthtest target) - #158307 (CI job for parallel frontend ui tests) - #158347 (Improve generic parameters handling for #[diagnostic::on_const]) - #158722 (delegation: do not always inherit `ConstArgHasType` predicates) - #158741 (Simplify `Option::into_flat_iter` signature) - #158807 (Add regression test for CString::clone_into unwind safety) - #158862 (Fix the span for parameter suggestion ) - #158883 (tests: fix enum-match.rs to handle LLVM 23)

@workingjubilee

…ingjubilee Carry the `b_offset` inside `BackendRepr::ScalarPair` Inspired by rust-lang/compiler-team#1007 but doesn't actually change any of the layout rules just yet. This turned out to be a nice change even if we didn't use the extra flexibility, IMHO, because it allowed so many things like ```diff @@ -222,12 +224,12 @@ fn from_const_alloc<Bx: BuilderMethods<'a, 'tcx, Value = V>>( let val = read_scalar(offset, size, s, bx.immediate_backend_type(layout)); OperandRef { val: OperandValue::Immediate(val), layout, move_annotation: None } } - BackendRepr::ScalarPair( - a @ abi::Scalar::Initialized { .. }, - b @ abi::Scalar::Initialized { .. }, - ) => { + BackendRepr::ScalarPair { + a: a @ abi::Scalar::Initialized { .. }, + b: b @ abi::Scalar::Initialized { .. }, + b_offset, + } => { let (a_size, b_size) = (a.size(bx), b.size(bx)); - let b_offset = (offset + a_size).align_to(b.default_align(bx).abi); assert!(b_offset.bytes() > 0); let a_val = read_scalar( offset, ``` as *oh my* was that little magic incantation copy-pasted all over the place. Apologies for the pretty-giant PR. I tried to make it as direct a change as I could: if it was `(..)` before it's `{ .. }` now, if it was `(_, _)` before it's `{ a: _, b: _, b_offset: _ }` now. I kept the names the same so the code lines were unchanged even if normally I might have just renamed things, etc. I'll add some inline notes for places of particular interest. r? @workingjubilee

Rollup of 25 pull requests Successful merges: - #158871 (add relnotes for 1.97.0) - #158968 (stdarch subtree update) - #157690 (codegen_ssa: pack small const aggregates into immediate stores) - #158541 (Move `std::io::Write` to `core::io`) - #154445 (rustdoc: Represent `--output-format=json` coverage and ir differently) - #156370 (Reject linked dylib EII default overrides) - #157153 (allow `Allocator`s to be used as `#[global_allocator]`s) - #158495 (Rename HAS_CT_PROJECTION to HAS_CONST_ALIAS) - #158617 (allow mGCA const arguments to fall back to anon consts) - #158645 (Fix splat ICEs and ban it in closures) - #158655 (Fix coroutine MIR saved local remapping) - #158666 (Carry the `b_offset` inside `BackendRepr::ScalarPair`) - #158870 (std: merge the unix-like io::error modules into one file) - #158920 (Update wasm-component-ld to 0.5.26) - #158926 (wrapping_sh* methods: clarify underspecified reference) - #158927 (add core test run with `-Zforce-intrinsic-fallback`) - #158932 (Do not build the compiler when invoking `x perf compare`) - #158937 (Emit the emscripten entry point as `__main_argc_argv`) - #151379 (Stabilize `VecDeque::retain_back` from `truncate_front`) - #156144 (Better docs for PartialEq (includes macro rename)) - #156548 ( Library support for aarch64-unknown-linux-pauthtest target) - #158307 (CI job for parallel frontend ui tests) - #158347 (Improve generic parameters handling for #[diagnostic::on_const]) - #158722 (delegation: do not always inherit `ConstArgHasType` predicates) - #158741 (Simplify `Option::into_flat_iter` signature)

@workingjubilee

…ingjubilee Carry the `b_offset` inside `BackendRepr::ScalarPair` Inspired by rust-lang/compiler-team#1007 but doesn't actually change any of the layout rules just yet. This turned out to be a nice change even if we didn't use the extra flexibility, IMHO, because it allowed so many things like ```diff @@ -222,12 +224,12 @@ fn from_const_alloc<Bx: BuilderMethods<'a, 'tcx, Value = V>>( let val = read_scalar(offset, size, s, bx.immediate_backend_type(layout)); OperandRef { val: OperandValue::Immediate(val), layout, move_annotation: None } } - BackendRepr::ScalarPair( - a @ abi::Scalar::Initialized { .. }, - b @ abi::Scalar::Initialized { .. }, - ) => { + BackendRepr::ScalarPair { + a: a @ abi::Scalar::Initialized { .. }, + b: b @ abi::Scalar::Initialized { .. }, + b_offset, + } => { let (a_size, b_size) = (a.size(bx), b.size(bx)); - let b_offset = (offset + a_size).align_to(b.default_align(bx).abi); assert!(b_offset.bytes() > 0); let a_val = read_scalar( offset, ``` as *oh my* was that little magic incantation copy-pasted all over the place. Apologies for the pretty-giant PR. I tried to make it as direct a change as I could: if it was `(..)` before it's `{ .. }` now, if it was `(_, _)` before it's `{ a: _, b: _, b_offset: _ }` now. I kept the names the same so the code lines were unchanged even if normally I might have just renamed things, etc. I'll add some inline notes for places of particular interest. r? @workingjubilee

Rollup of 23 pull requests Successful merges: - #158968 (stdarch subtree update) - #154445 (rustdoc: Represent `--output-format=json` coverage and ir differently) - #158495 (Rename HAS_CT_PROJECTION to HAS_CONST_ALIAS) - #158666 (Carry the `b_offset` inside `BackendRepr::ScalarPair`) - #158870 (std: merge the unix-like io::error modules into one file) - #158920 (Update wasm-component-ld to 0.5.26) - #158926 (wrapping_sh* methods: clarify underspecified reference) - #158927 (add core test run with `-Zforce-intrinsic-fallback`) - #158932 (Do not build the compiler when invoking `x perf compare`) - #158937 (Emit the emscripten entry point as `__main_argc_argv`) - #151379 (Stabilize `VecDeque::retain_back` from `truncate_front`) - #156144 (Better docs for PartialEq (includes macro rename)) - #156548 ( Library support for aarch64-unknown-linux-pauthtest target) - #157995 (`Vec::dedup_by` docs explicit function argument order) - #158307 (CI job for parallel frontend ui tests) - #158741 (Simplify `Option::into_flat_iter` signature) - #158807 (Add regression test for CString::clone_into unwind safety) - #158862 (Fix the span for parameter suggestion ) - #158894 (Make the ordering of non-terminal binds in ambiguity error messages deterministic) - #158902 (add codegen test for range length bound propagation) - #158913 (Update `browser-ui-test` version to `0.24.1`) - #158935 (std: support real fd methods on Emscripten) - #158978 (Add regression test for too-big by-value ABI args)

@workingjubilee

Rollup merge of #158666 - scottmcm:scalar-but-packed, r=workingjubilee Carry the `b_offset` inside `BackendRepr::ScalarPair` Inspired by rust-lang/compiler-team#1007 but doesn't actually change any of the layout rules just yet. This turned out to be a nice change even if we didn't use the extra flexibility, IMHO, because it allowed so many things like ```diff @@ -222,12 +224,12 @@ fn from_const_alloc<Bx: BuilderMethods<'a, 'tcx, Value = V>>( let val = read_scalar(offset, size, s, bx.immediate_backend_type(layout)); OperandRef { val: OperandValue::Immediate(val), layout, move_annotation: None } } - BackendRepr::ScalarPair( - a @ abi::Scalar::Initialized { .. }, - b @ abi::Scalar::Initialized { .. }, - ) => { + BackendRepr::ScalarPair { + a: a @ abi::Scalar::Initialized { .. }, + b: b @ abi::Scalar::Initialized { .. }, + b_offset, + } => { let (a_size, b_size) = (a.size(bx), b.size(bx)); - let b_offset = (offset + a_size).align_to(b.default_align(bx).abi); assert!(b_offset.bytes() > 0); let a_val = read_scalar( offset, ``` as *oh my* was that little magic incantation copy-pasted all over the place. Apologies for the pretty-giant PR. I tried to make it as direct a change as I could: if it was `(..)` before it's `{ .. }` now, if it was `(_, _)` before it's `{ a: _, b: _, b_offset: _ }` now. I kept the names the same so the code lines were unchanged even if normally I might have just renamed things, etc. I'll add some inline notes for places of particular interest. r? @workingjubilee

rustbot added A-LLVM Area: Code generation parts specific to LLVM. Both correctness bugs and optimization-related issues. F-autodiff `#![feature(autodiff)]` S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. labels Jul 1, 2026

rustbot assigned workingjubilee Jul 1, 2026

rustbot added the T-compiler Relevant to the compiler team, which will review and decide on the PR/issue. label Jul 1, 2026

scottmcm commented Jul 1, 2026

View reviewed changes

This comment has been minimized.

Sign in to view

scottmcm commented Jul 1, 2026

View reviewed changes

scottmcm force-pushed the scalar-but-packed branch from 455c80c to ede96e9 Compare July 1, 2026 17:14

rustbot added the T-clippy Relevant to the Clippy team. label Jul 1, 2026

This comment has been minimized.

Sign in to view

workingjubilee approved these changes Jul 1, 2026

View reviewed changes

fbstj reviewed Jul 1, 2026

View reviewed changes

Comment thread compiler/rustc_abi/src/lib.rs Outdated

This comment has been minimized.

Sign in to view

rustbot added the S-waiting-on-perf Status: Waiting on a perf run to be completed. label Jul 1, 2026

rust-bors Bot pushed a commit that referenced this pull request Jul 1, 2026

Auto merge of #158666 - scottmcm:scalar-but-packed, r=<try>

e1ccfbf

Carry the `b_offset` inside `BackendRepr::ScalarPair`

This comment has been minimized.

Sign in to view

rustbot removed the S-waiting-on-perf Status: Waiting on a perf run to be completed. label Jul 1, 2026

rust-bors Bot added S-waiting-on-bors Status: Waiting on bors to run and complete tests. Bors will change the label on completion. and removed S-waiting-on-author Status: This is awaiting some action (such as code changes or more information) from the author. labels Jul 8, 2026

jhpratt mentioned this pull request Jul 8, 2026

Rollup of 8 pull requests #158946

Closed

GuillaumeGomez mentioned this pull request Jul 8, 2026

Rollup of 16 pull requests #158953

Closed

JonathanBrouwer mentioned this pull request Jul 8, 2026

Rollup of 18 pull requests #158958

Closed

JonathanBrouwer mentioned this pull request Jul 8, 2026

Rollup of 25 pull requests #158973

Closed

JonathanBrouwer mentioned this pull request Jul 8, 2026

Rollup of 25 pull requests #158977

Closed

jhpratt mentioned this pull request Jul 8, 2026

Rollup of 25 pull requests #158986

Closed

This was referenced Jul 9, 2026

Rollup of 25 pull requests #158989

Closed

Rollup of 23 pull requests #158991

Merged

scottmcm mentioned this pull request Jul 9, 2026

Tracking issue for MCP 1007: Decouple BackendRepr from ABI alignment #158985

Open

6 tasks

rust-bors Bot merged commit c688130 into rust-lang:main Jul 9, 2026
13 checks passed

rustbot added this to the 1.99.0 milestone Jul 9, 2026

scottmcm deleted the scalar-but-packed branch July 9, 2026 16:44

Uh oh!

Uh oh!

Conversation

scottmcm commented Jul 1, 2026 • edited by rustbot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

rustbot commented Jul 1, 2026

Uh oh!

rustbot commented Jul 1, 2026

Uh oh!

scottmcm Jul 1, 2026 • edited by rustbot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

scottmcm Jul 1, 2026 • edited by rustbot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

scottmcm Jul 1, 2026 • edited by rustbot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

scottmcm Jul 1, 2026 • edited by rustbot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

This comment has been minimized.

scottmcm Jul 1, 2026 • edited by rustbot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

rustbot commented Jul 1, 2026

Uh oh!

This comment has been minimized.

rustbot commented Jul 1, 2026

Uh oh!

workingjubilee left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

This comment has been minimized.

oli-obk commented Jul 1, 2026

Uh oh!

This comment has been minimized.

This comment has been minimized.

rust-bors Bot commented Jul 1, 2026

Uh oh!

This comment has been minimized.

rust-timer commented Jul 1, 2026

Overall result: no relevant changes - no action needed

Uh oh!

scottmcm commented Jul 1, 2026 •

edited by rustbot

Loading

scottmcm Jul 1, 2026 •

edited by rustbot

Loading

scottmcm Jul 1, 2026 •

edited by rustbot

Loading

scottmcm Jul 1, 2026 •

edited by rustbot

Loading

scottmcm Jul 1, 2026 •

edited by rustbot

Loading

scottmcm Jul 1, 2026 •

edited by rustbot

Loading

workingjubilee left a comment •

edited

Loading