codegen: tell LLVM that the discriminant is a power of two - #162741

dianqk · 2026-09-14T00:51:20Z

We should not disable the pattern in MatchBranchSimplification. The two forms exhaustive match and cast are equivalent, but the cast is better represented in MIR: no branches, only one statement.

AFAIK, LLVM does not handle multiple disjoint intervals well. As an workaround, the patch emits assume(ctpop(val) == 1) to tell LLVM that the discriminant is a power of two.

LLM disclosure: search the codebase and fix grammer.

dianqk · 2026-09-14T13:11:06Z

@bors try @rust-timer queue

codegen: tell LLVM that the discriminant is a power of two

rust-bors · 2026-09-14T14:10:02Z

☀️ Try build successful (CI)
Build commit: fb82f6b (fb82f6bd3bc4448e1f07ef7b824296245782bc7c)
Base parent: ed77b7b (ed77b7b8699e342c2dc842c83cb5fe4025252ee8)

rust-timer · 2026-09-14T15:24:54Z

Finished benchmarking commit (fb82f6b): comparison URL.

Overall result: ❌ regressions - 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)	0.2%	[0.2%, 0.2%]	2
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-	-	0

Max RSS (memory usage)

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

Cycles

Results (secondary -5.3%)

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)	2.3%	[2.3%, 2.3%]	1
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-9.1%	[-9.1%, -9.0%]	2
All ❌✅ (primary)	-	-	0

Binary size

Results (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
Improvements ✅ (secondary)	-0.1%	[-0.1%, -0.1%]	1
All ❌✅ (primary)	-	-	0

Bootstrap: 495.332s -> 496.261s (0.19%)
Artifact size: 406.96 MiB -> 406.98 MiB (0.01%)

rustbot · 2026-09-15T10:17:33Z

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

rustbot · 2026-09-15T10:17:37Z

r? @adwinwhite

rustbot has assigned @adwinwhite.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

Why was this reviewer chosen?

The reviewer was selected based on:

Owners of files modified in this PR: compiler
compiler expanded to 76 candidates
Random selection from 19 candidates

adwinwhite · 2026-09-16T01:11:19Z

@rustbot reroll

saethlin · 2026-10-05T01:40:13Z

r? saethlin
r=me but you might want to take my suggestion. I think it compiles.

We should not disable the pattern in MatchBranchSimplification. The two forms exhaustive match and cast are equivalent, but the cast is better represented in MIR: no branches, only one statement. AFAIK, LLVM does not handle multiple disjoint intervals well. As an workaround, the patch emits `assume(ctpop(val) == 1)` to tell LLVM that the discriminant is a power of two.

dianqk · 2026-10-07T20:11:17Z

@bors r=saethlin

rust-bors · 2026-10-07T20:11:21Z

📌 Commit 23fa0f4 has been approved by saethlin

It is now in the queue for this repository.

… r=saethlin codegen: tell LLVM that the discriminant is a power of two Fixes rust-lang#162513. We should not disable the pattern in MatchBranchSimplification. The two forms exhaustive match and cast are equivalent, but the cast is better represented in MIR: no branches, only one statement. AFAIK, LLVM does not handle multiple disjoint intervals well. As an workaround, the patch emits `assume(ctpop(val) == 1)` to tell LLVM that the discriminant is a power of two. LLM disclosure: search the codebase and fix grammer.

…uwer Rollup of 10 pull requests Successful merges: - #162741 (codegen: tell LLVM that the discriminant is a power of two) - #162758 (Split PassMode::IndirectUnsized out of PassMode::Indirect) - #163931 (Move `loop_hint_attrs` from `Terminator` into `TerminatorKind::Goto`) - #163028 (autodiff: fix sysroot candidates in libEnzyme not found error) - #163538 (Add `rustc::missing_generic_type_visitable` lint) - #163841 ([rustdoc] Fix more `incoherent_impl` linking) - #163908 (Update Enzyme submodule) - #163935 (Use BasicBlock::start_location and Location::START) - #163949 (disable next-solver with higher-ranked-assumptions) - #163951 (relax_elf_relocations is x86-only)

…uwer Rollup of 11 pull requests Successful merges: - #162741 (codegen: tell LLVM that the discriminant is a power of two) - #162758 (Split PassMode::IndirectUnsized out of PassMode::Indirect) - #163931 (Move `loop_hint_attrs` from `Terminator` into `TerminatorKind::Goto`) - #163028 (autodiff: fix sysroot candidates in libEnzyme not found error) - #163257 (abby DSL: sanity checks on forall where clauses) - #163841 ([rustdoc] Fix more `incoherent_impl` linking) - #163908 (Update Enzyme submodule) - #163935 (Use BasicBlock::start_location and Location::START) - #163949 (disable next-solver with higher-ranked-assumptions) - #163951 (relax_elf_relocations is x86-only) - #163968 (Remove needs-triage label from relnotes issues)

Rollup merge of #162741 - dianqk:codegen-discriminants-cast, r=saethlin codegen: tell LLVM that the discriminant is a power of two Fixes #162513. We should not disable the pattern in MatchBranchSimplification. The two forms exhaustive match and cast are equivalent, but the cast is better represented in MIR: no branches, only one statement. AFAIK, LLVM does not handle multiple disjoint intervals well. As an workaround, the patch emits `assume(ctpop(val) == 1)` to tell LLVM that the discriminant is a power of two. LLM disclosure: search the codebase and fix grammer.

…uwer Rollup of 11 pull requests Successful merges: - rust-lang/rust#162741 (codegen: tell LLVM that the discriminant is a power of two) - rust-lang/rust#162758 (Split PassMode::IndirectUnsized out of PassMode::Indirect) - rust-lang/rust#163931 (Move `loop_hint_attrs` from `Terminator` into `TerminatorKind::Goto`) - rust-lang/rust#163028 (autodiff: fix sysroot candidates in libEnzyme not found error) - rust-lang/rust#163257 (abby DSL: sanity checks on forall where clauses) - rust-lang/rust#163841 ([rustdoc] Fix more `incoherent_impl` linking) - rust-lang/rust#163908 (Update Enzyme submodule) - rust-lang/rust#163935 (Use BasicBlock::start_location and Location::START) - rust-lang/rust#163949 (disable next-solver with higher-ranked-assumptions) - rust-lang/rust#163951 (relax_elf_relocations is x86-only) - rust-lang/rust#163968 (Remove needs-triage label from relnotes issues)

codegen: tell LLVM that the discriminant is a power of two Fixes #162513. We should not disable the pattern in MatchBranchSimplification. The two forms exhaustive match and cast are equivalent, but the cast is better represented in MIR: no branches, only one statement. AFAIK, LLVM does not handle multiple disjoint intervals well. As an workaround, the patch emits `assume(ctpop(val) == 1)` to tell LLVM that the discriminant is a power of two. LLM disclosure: search the codebase and fix grammer.