Update stdarch

[ehuss]

Updates stdarch

• core_arch: Add LoongArch basic intrinsics: core_arch: Add LoongArch basic intrinsics stdarch#1688
• New ARM intrinsic generator: New ARM intrinsic generator stdarch#1693
• Fix the bug in CMPINT intrinsics with IMM3=7: Fix the bug in CMPINT intrinsics with IMM3=7 stdarch#1694
• Expand feature detection on AArch64 Darwin: Expand feature detection on AArch64 Darwin stdarch#1695
• Tidying x86 as_* functions: Tidying x86 as_* functions stdarch#1696
• Fix typo and prettify comment: Fix typo and prettify comment stdarch#1697
• add is_s390x_feature_detected: add is_s390x_feature_detected stdarch#1699
• add vec_add for s390x: add vec_add for s390x stdarch#1703
• s390x: add vec_sub, vec_mul, vec_min, vec_max, vec_abs and vec_splats: s390x: add vec_sub, vec_mul, vec_min, vec_max, vec_abs and vec_splats stdarch#1704
• Fix build and CLI behaviour for stdarch-gen-arm. Fix build and CLI behaviour for stdarch-gen-arm. stdarch#1705
• Fix some test naming, and refactor stdarch-verify in general: Fix some test naming, and refactor stdarch-verify in general stdarch#1707
• Update all stdarch crates to Rust 2024: Update all stdarch crates to Rust 2024 stdarch#1710
• Add keylocker (kl and widekl) intrinsics and runtime feature detection: Add keylocker (kl and widekl) intrinsics and runtime feature detection stdarch#1706
• S390x vector bitwise operations: S390x vector bitwise operations stdarch#1709
• Update CI to FreeBSD 13.4: Update CI to FreeBSD 13.4 stdarch#1715
• Update wasm sub sat intrinsics for LLVM 20: Update wasm sub sat intrinsics for LLVM 20 stdarch#1719
• powerpc: use more target-independent llvm intrinsics (min, max, round, countlz): powerpc: use more target-independent llvm intrinsics (min, max, round, countlz) stdarch#1713
• S390x float rounding: S390x float rounding stdarch#1712
• mark riscv intrinsics as safe: mark riscv intrinsics as safe stdarch#1717
• change redundant transmutations of sign to cast_unsigned: change redundant transmutations of sign to cast_unsigned stdarch#1711
• Fix - AArch64 Big Endian Intrinsics: Fix - AArch64 Big Endian Intrinsics  stdarch#1708
• mark x86 intrinsics as safe: mark x86 intrinsics as safe stdarch#1714
• AArch64: Add NEON fp16 intrinsics: AArch64: Add NEON fp16 intrinsics stdarch#1726
• wasm: use simd_as for float to integer conversions: wasm: use simd_as for float to integer conversions stdarch#1724
• nvptx: use simd_fmin and simd_fmax for minnum and maxnum: nvptx: use simd_fmin and simd_fmax for minnum and maxnum stdarch#1725
• powerpc: use simd_ceil and simd_floor: powerpc: use simd_ceil and simd_floor stdarch#1723
• Changed altivec.rs to new intrinsic declaration: Changed altivec.rs to new intrinsic declaration stdarch#1722
• Remove some allow(unsafe_op_in_unsafe_fn)s and use target_feature 1.1 in examples: Remove some allow(unsafe_op_in_unsafe_fn)s and use target_feature 1.1 in examples stdarch#1727
• fix - neon type signed unsigned conversions: fix - neon type signed unsigned conversions stdarch#1729
• s390x_is_feature_detected!: detect more features: s390x_is_feature_detected!: detect more features stdarch#1720
• Fix doctests failing due to unused_unsafe: Fix doctests failing due to unused_unsafe stdarch#1731
• fix compilation on armebv7r-none-eabi: fix compilation on armebv7r-none-eabi stdarch#1733
• wasm: update for rintf intrinsic rename: wasm: update for rintf intrinsic rename stdarch#1721
• powerpc: use the simd_fma intrinsic for vec_madd: powerpc: use the simd_fma intrinsic for vec_madd stdarch#1734
• powerpc: use llvm.fshl for vec_rl: powerpc: use llvm.fshl for vec_rl stdarch#1735
• s390x: add more intrinsics: s390x: add more intrinsics stdarch#1728
• make _mm256_zero{upper,all} safe: make _mm256_zero{upper,all} safe stdarch#1736
• fix unnecessary unsafe error in doctest: fix unnecessary unsafe error in doctest stdarch#1739
• Feat - Aarch64 FEAT_FAMINMAX: Feat - Aarch64 FEAT_FAMINMAX stdarch#1732
• feat - FEAT_LUT neon instrinsics: feat - FEAT_LUT neon instrinsics stdarch#1741

try-job: dist-x86_64-msvc
try-job: aarch64-gnu
try-job: aarch64-gnu-debug
try-job: dist-aarch64-linux
try-job: dist-arm-linux

[rustbot]

r? @thomcc

rustbot has assigned @thomcc.
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

[rustbot]

⚠️ Warning ⚠️

• These commits modify submodules.

[ehuss]

cc @Amanieu

[Amanieu]

@bors r+

[bors]

📌 Commit f410520 has been approved by Amanieu

It is now in the queue for this repository.

[workingjubilee]

@bors r-
@bors try

[bors]

⌛ Trying commit f410520 with merge 90aa469...

[GuillaumeGomez]

Seems like it failed in #136930.

[workingjubilee]

whoops, wrong job name.

@bors try

[bors]

⌛ Trying commit f410520 with merge 45a457d...

[ehuss]

I would suspect it is related to rust-lang/stdarch#1693.
I wonder if the definition of vqsub_u8 changed?

[ehuss]

That particular test (test_vqsub_u8) passes with the LLVM backend, but not cranelift.

@bjorn3 Do you have any ideas?

[bors]

💔 Test failed - checks-actions

[workingjubilee]

bjorn3 thinks it was this miscompilation: rust-lang/rustc_codegen_cranelift#1560

[ehuss]

bjorn3 thinks it was this miscompilation: rust-lang/rustc_codegen_cranelift#1560

From what I can tell, with the latest cranelift (or #137078), the test still fails.

[bors]

📌 Commit c4cee08 has been approved by Amanieu

It is now in the queue for this repository.

[bors]

⌛ Testing commit c4cee08 with merge d9bebb6...

[bors]

💔 Test failed - checks-actions

[RalfJung]

"unnecessary unsafe block" in a stdarch file -- why did stdrach CI not catch this? (library/core/src/../../stdarch/crates/core_arch/src/x86/sse41.rs:192:28)

[usamoi]

"unnecessary unsafe block" in a stdarch file -- why did stdrach CI not catch this? (library/core/src/../../stdarch/crates/core_arch/src/x86/sse41.rs:192:28)

It seems that rustdoc doesn't build doctests for #[target_feature] functions, if target features are not set at compilation. This doctest only builds for x86_64-apple-darwin because the baseline of this target contains sse4.2.

[RalfJung]

I think the reason is that docstests in stdarch don't use the "local" stdarch, they use the one form libcore, so when stdarch itself changes, doctests only get that update with a delay. Due to stdarch being a submodule, that makes changes that affect doctests quite tricky, as they have to be done in stages.

[usamoi]

I think the reason is that docstests in stdarch don't use the "local" stdarch, they use the one form libcore, so when stdarch itself changes, doctests only get that update with a delay. Due to stdarch being a submodule, that makes changes that affect doctests quite tricky, as they have to be done in stages.

You are right. I just realized

• docs of _mm_extract_ps is not built for x86_64-linux-gnu
• stdarch CI did not catch this

are two different things.

[ehuss]

Dang, I didn't think to try a target that had a different base set of target features enabled by default.

Now blocked on rust-lang/stdarch#1739

[ehuss]

@bors try

[bors]

⌛ Trying commit e3fd813 with merge ac896e7aecf3a47a74110b13d105db9e2a6cbaeb...

[bors]

☀️ Try build successful - checks-actions
Build commit: ac896e7 (ac896e7aecf3a47a74110b13d105db9e2a6cbaeb)

[ehuss]

@bors try

[bors]

⌛ Trying commit e3fd813 with merge 25a50037c041fcc495cc02d50f5d1dc75670a725...

[bors]

☀️ Try build successful - checks-actions
Build commit: 25a5003 (25a50037c041fcc495cc02d50f5d1dc75670a725)