Stop needing materialized places for most intrinsics by scottmcm · Pull Request #156116 · rust-lang/rust

scottmcm · 2026-05-03T19:08:25Z

Today even to return a constant from size_of_val it takes an alloca. That's wasteful.

This updates the cg_ssa traits to allow returning an OperandValue instead, which is possible for the vast majority of intrinsics -- this PR moves all but 5 over to doing that in LLVM.

The first commit adds a test to help show the difference here.

cc rust-lang/compiler-team#970 #153250

scottmcm · 2026-05-04T03:33:24Z

@bors try @rust-timer queue

Stop needing materialized places for most intrinsics

rust-bors · 2026-05-04T05:43:28Z

☀️ Try build successful (CI)
Build commit: ce9f4cc (ce9f4cc0887ef1020c405689861eeb71f9095196, parent: 045b17737dab5fcc28e4cbee0cfe2ce4ed363b32)

rust-timer · 2026-05-04T06:26:30Z

Finished benchmarking commit (ce9f4cc): comparison URL.

Overall result: ✅ improvements - 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

Our most reliable metric. Used to determine the overall result above. However, even this metric can be noisy.

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-0.1%	[-0.1%, -0.1%]	1
Improvements ✅ (secondary)	-0.3%	[-0.3%, -0.2%]	2
All ❌✅ (primary)	-0.1%	[-0.1%, -0.1%]	1

Max RSS (memory usage)

Results (secondary -0.0%)

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)	2.7%	[0.5%, 7.5%]	5
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-3.5%	[-6.3%, -0.6%]	4
All ❌✅ (primary)	-	-	0

Cycles

Results (secondary -1.2%)

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)	0.7%	[0.6%, 0.7%]	2
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-1.9%	[-5.7%, -0.4%]	5
All ❌✅ (primary)	-	-	0

Binary size

Results (primary -0.1%, secondary -0.1%)

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)	-0.1%	[-0.2%, -0.1%]	18
Improvements ✅ (secondary)	-0.1%	[-0.2%, -0.0%]	25
All ❌✅ (primary)	-0.1%	[-0.2%, -0.1%]	18

Bootstrap: 496.439s -> 493.878s (-0.52%)
Artifact size: 394.45 MiB -> 394.37 MiB (-0.02%)

scottmcm · 2026-05-04T06:29:44Z

r? codegen

rustbot · 2026-05-04T06:29:54Z

The GCC codegen subtree was changed

cc @antoyo, @GuillaumeGomez

scottmcm · 2026-05-04T22:53:52Z

+                let func_ty = func.ty(&self.fx.mir.local_decls, tcx);
+                if let ty::FnDef(def_id, _args) = *func_ty.kind()
+                    && let Some(intrinsic) = tcx.intrinsic(def_id)
+                    && self.fx.cx.intrinsic_call_expects_place_always(intrinsic.name)


annot: needing this check is unfortunate here, since we have to do it for all Calls. But perf says it doesn't cause a regression, so I think it's fine for now to keep this PR manageable, and we can hopefully remove it eventually with more updates to the backends.

(Worst come to worst the backends can make their own allocas for some intrinsics if really necessary. And extra typed copy on some intrinsics occasionally would be well worth it to get not making all the unnecessary ones on most of the intrinsics.)

View changes since the review

rust-bors · 2026-05-05T11:08:09Z

☔ The latest upstream changes (presumably #156166) made this pull request unmergeable. Please resolve the merge conflicts.

Add some basic tests for {size,align}_of_val

5b9f56b

scottmcm force-pushed the tweak-layout-of-alternative branch from 0bc61de to a8a0062 Compare May 3, 2026 19:13