Auto merge of rust-lang#136831 - ehuss:update-stdarch, r=Amanieu

Update stdarch

Updates stdarch

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

Author: bors

compiler/rustc_span/src/analyze_source_file.rs

@@ -83,17 +83,17 @@ cfg_match! {
 
                 // For character in the chunk, see if its byte value is < 0, which
                 // indicates that it's part of a UTF-8 char.
-                let multibyte_test = unsafe { _mm_cmplt_epi8(chunk, _mm_set1_epi8(0)) };
+                let multibyte_test = _mm_cmplt_epi8(chunk, _mm_set1_epi8(0));
                 // Create a bit mask from the comparison results.
-                let multibyte_mask = unsafe { _mm_movemask_epi8(multibyte_test) };
+                let multibyte_mask = _mm_movemask_epi8(multibyte_test);
 
                 // If the bit mask is all zero, we only have ASCII chars here:
                 if multibyte_mask == 0 {
                     assert!(intra_chunk_offset == 0);
 
                     // Check for newlines in the chunk
-                    let newlines_test = unsafe { _mm_cmpeq_epi8(chunk, _mm_set1_epi8(b'\n' as i8)) };
-                    let mut newlines_mask = unsafe { _mm_movemask_epi8(newlines_test) };
+                    let newlines_test = _mm_cmpeq_epi8(chunk, _mm_set1_epi8(b'\n' as i8));
+                    let mut newlines_mask = _mm_movemask_epi8(newlines_test);
 
                     let output_offset = RelativeBytePos::from_usize(chunk_index * CHUNK_SIZE + 1);
 

library/core/Cargo.toml

@@ -32,6 +32,8 @@ check-cfg = [
     'cfg(bootstrap)',
     'cfg(no_fp_fmt_parse)',
     'cfg(stdarch_intel_sde)',
+    # #[cfg(bootstrap)]
+    'cfg(target_feature, values("vector-enhancements-1"))',
     # core use #[path] imports to portable-simd `core_simd` crate
     # and to stdarch `core_arch` crate which messes-up with Cargo list
     # of declared features, we therefor expect any feature cfg

library/core/src/lib.rs

@@ -202,14 +202,17 @@
 //
 // Target features:
 // tidy-alphabetical-start
+#![feature(aarch64_unstable_target_feature)]
 #![feature(arm_target_feature)]
 #![feature(avx512_target_feature)]
 #![feature(hexagon_target_feature)]
+#![feature(keylocker_x86)]
 #![feature(loongarch_target_feature)]
 #![feature(mips_target_feature)]
 #![feature(powerpc_target_feature)]
 #![feature(riscv_target_feature)]
 #![feature(rtm_target_feature)]
+#![feature(s390x_target_feature)]
 #![feature(sha512_sm_x86)]
 #![feature(sse4a_target_feature)]
 #![feature(tbm_target_feature)]

library/stdarch

@@ -257,6 +257,8 @@ fn analyze_source_file_dispatch(
 /// SSE2 intrinsics to quickly find all newlines.
 #[target_feature(enable = "sse2")]
 #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+// This can be removed once 1.87 is stable due to some intrinsics switching to safe.
+#[allow(unsafe_op_in_unsafe_fn)]
 unsafe fn analyze_source_file_sse2(
     src: &str,
     lines: &mut Vec<TextSize>,
@@ -287,17 +289,17 @@ unsafe fn analyze_source_file_sse2(
 
         // For character in the chunk, see if its byte value is < 0, which
         // indicates that it's part of a UTF-8 char.
-        let multibyte_test = unsafe { _mm_cmplt_epi8(chunk, _mm_set1_epi8(0)) };
+        let multibyte_test = _mm_cmplt_epi8(chunk, _mm_set1_epi8(0));
         // Create a bit mask from the comparison results.
-        let multibyte_mask = unsafe { _mm_movemask_epi8(multibyte_test) };
+        let multibyte_mask = _mm_movemask_epi8(multibyte_test);
 
         // If the bit mask is all zero, we only have ASCII chars here:
         if multibyte_mask == 0 {
             assert!(intra_chunk_offset == 0);
 
             // Check for newlines in the chunk
-            let newlines_test = unsafe { _mm_cmpeq_epi8(chunk, _mm_set1_epi8(b'\n' as i8)) };
-            let newlines_mask = unsafe { _mm_movemask_epi8(newlines_test) };
+            let newlines_test = _mm_cmpeq_epi8(chunk, _mm_set1_epi8(b'\n' as i8));
+            let newlines_mask = _mm_movemask_epi8(newlines_test);
 
             if newlines_mask != 0 {
                 // All control characters are newlines, record them
@@ -354,15 +356,19 @@ unsafe fn analyze_source_file_sse2(
 // The mask is a 64-bit integer, where each 4-bit corresponds to a u8 in the
 // input vector. The least significant 4 bits correspond to the first byte in
 // the vector.
+// This can be removed once 1.87 is stable due to some intrinsics switching to safe.
+#[allow(unsafe_op_in_unsafe_fn)]
 unsafe fn move_mask(v: std::arch::aarch64::uint8x16_t) -> u64 {
     use std::arch::aarch64::*;
 
-    let nibble_mask = unsafe { vshrn_n_u16(vreinterpretq_u16_u8(v), 4) };
-    unsafe { vget_lane_u64(vreinterpret_u64_u8(nibble_mask), 0) }
+    let nibble_mask = vshrn_n_u16(vreinterpretq_u16_u8(v), 4);
+    vget_lane_u64(vreinterpret_u64_u8(nibble_mask), 0)
 }
 
 #[target_feature(enable = "neon")]
 #[cfg(all(target_arch = "aarch64", target_endian = "little"))]
+// This can be removed once 1.87 is stable due to some intrinsics switching to safe.
+#[allow(unsafe_op_in_unsafe_fn)]
 unsafe fn analyze_source_file_neon(
     src: &str,
     lines: &mut Vec<TextSize>,
@@ -376,7 +382,7 @@ unsafe fn analyze_source_file_neon(
 
     let chunk_count = src.len() / CHUNK_SIZE;
 
-    let newline = unsafe { vdupq_n_s8(b'\n' as i8) };
+    let newline = vdupq_n_s8(b'\n' as i8);
 
     // This variable keeps track of where we should start decoding a
     // chunk. If a multi-byte character spans across chunk boundaries,
@@ -390,7 +396,7 @@ unsafe fn analyze_source_file_neon(
 
         // For character in the chunk, see if its byte value is < 0, which
         // indicates that it's part of a UTF-8 char.
-        let multibyte_test = unsafe { vcltzq_s8(chunk) };
+        let multibyte_test = vcltzq_s8(chunk);
         // Create a bit mask from the comparison results.
         let multibyte_mask = unsafe { move_mask(multibyte_test) };
 
@@ -399,7 +405,7 @@ unsafe fn analyze_source_file_neon(
             assert!(intra_chunk_offset == 0);
 
             // Check for newlines in the chunk
-            let newlines_test = unsafe { vceqq_s8(chunk, newline) };
+            let newlines_test = vceqq_s8(chunk, newline);
             let mut newlines_mask = unsafe { move_mask(newlines_test) };
 
             // If the bit mask is not all zero, there are newlines in this chunk.