chore: update rust toolchain (#272)

AVX512 is stablized in Rust 1.89. So we could switch to stable Rust
toolchain on 7 August.

job: +psql_macos
job: +psql_windows

Signed-off-by: usamoi <usamoi@outlook.com>

Author: usamoi

.github/workflows/check.yml

@@ -22,6 +22,7 @@ jobs:
       - name: Set up Environment
         run: |
           rustup set profile
+          rustup default nightly-2025-06-08
 
           curl -fsSL https://github.com/tamasfe/taplo/releases/latest/download/taplo-full-linux-$(uname -m).gz | gzip -d - | install -m 755 /dev/stdin /usr/local/bin/taplo
 
@@ -112,6 +113,8 @@ jobs:
       - name: Set up Environment
         run: |
           rustup set profile
+          rustup default nightly-2025-06-08
+
           sudo apt-get update
 
           if [ "$(uname -m)" == "x86_64" ]; then
@@ -152,6 +155,11 @@ jobs:
           fi
 
   psql:
+    if: |
+      (github.event_name == 'push' && !contains(github.event.head_commit.message, 'job: -psql')) ||
+      (github.event_name == 'pull_request' && !contains(github.event.pull_request.body, 'job: -psql')) ||
+      github.event_name == 'workflow_dispatch'
+
     strategy:
       matrix:
         version: ["13", "14", "15", "16", "17"]
@@ -166,6 +174,8 @@ jobs:
       - name: Set up Environment
         run: |
           rustup set profile
+          rustup default nightly-2025-06-08
+
           sudo apt-get update
 
           sudo apt-get remove -y '^postgres.*' '^libpq.*'
@@ -254,8 +264,8 @@ jobs:
 
   psql_macos:
     if: |
-      (github.event_name == 'push' && contains(github.event.head_commit.message, 'try-job: psql_macos')) ||
-      (github.event_name == 'pull_request' && contains(github.event.pull_request.body, 'try-job: psql_macos')) ||
+      (github.event_name == 'push' && contains(github.event.head_commit.message, 'job: +psql_macos')) ||
+      (github.event_name == 'pull_request' && contains(github.event.pull_request.body, 'job: +psql_macos')) ||
       github.event_name == 'workflow_dispatch'
 
     strategy:
@@ -273,6 +283,7 @@ jobs:
       - name: Set up Environment
         run: |
           rustup set profile
+          rustup default nightly-2025-06-08
 
           echo CC=$(brew --prefix llvm@18)/bin/clang >> $GITHUB_ENV
 
@@ -337,3 +348,52 @@ jobs:
             ./build/postgresql-${{ matrix.version }}-vchord_0.0.0_${{ matrix.arch }}-apple-darwin.zip
           compression-level: 9
           retention-days: 14
+
+  psql_windows:
+    if: |
+      (github.event_name == 'push' && contains(github.event.head_commit.message, 'job: +psql_windows')) ||
+      (github.event_name == 'pull_request' && contains(github.event.pull_request.body, 'job: +psql_windows')) ||
+      github.event_name == 'workflow_dispatch'
+
+    strategy:
+      matrix:
+        version: ["16"]
+        arch: ["x86_64"]
+
+    runs-on: "windows-2022"
+
+    env:
+      RUSTC_WRAPPER: sccache
+      SCCACHE_GHA_ENABLED: true
+
+    steps:
+      - name: Set up Environment
+        run: |
+          rustup set profile
+          rustup default nightly-2025-06-08
+
+          Invoke-WebRequest -Uri "https://get.enterprisedb.com/postgresql/postgresql-16.9-1-windows-x64-binaries.zip" -OutFile "$env:TEMP\postgresql-install.zip"
+          Expand-Archive -Path "$env:TEMP\postgresql-install.zip" -DestinationPath "D:\postgresql-install" -Force
+          Add-Content -Path $env:GITHUB_ENV -Value "PGRX_PG_CONFIG_PATH=D:\postgresql-install\pgsql\bin\pg_config.exe"
+          Add-Content -Path $env:GITHUB_ENV -Value "PGBIN=D:\postgresql-install\pgsql\bin"
+          Add-Content -Path $env:GITHUB_ENV -Value "PGDATA=D:\postgresql-install\pgsql\data"
+          Add-Content -Path $env:GITHUB_ENV -Value "PGROOT=D:\postgresql-install\pgsql"
+
+          Invoke-WebRequest -Uri https://github.com/risinglightdb/sqllogictest-rs/releases/download/v0.26.4/sqllogictest-bin-v0.26.4-${{ matrix.arch }}-pc-windows-msvc.zip -OutFile "$env:TEMP\sqllogictest-install.zip"
+          Expand-Archive -Path "$env:TEMP\sqllogictest-install.zip" -DestinationPath "D:\sqllogictest-install" -Force
+          Add-Content -Path $env:GITHUB_PATH -Value "D:\sqllogictest-install"
+
+      - name: Set up Sccache
+        uses: mozilla-actions/sccache-action@v0.0.9
+
+      - name: Checkout
+        uses: actions/checkout@v4
+
+      - name: Clippy
+        run: |
+          cargo clippy -p vchord --features pg${{ matrix.version }} -- --no-deps
+
+      - name: Install
+        run: |
+          cargo make build -o ./build/raw
+          make PG_CONFIG="$env:PGRX_PG_CONFIG_PATH" install

.github/workflows/release.yml

@@ -51,6 +51,9 @@ jobs:
     steps:
       - name: Set up Environment
         run: |
+          rustup set profile
+          rustup default nightly-2025-06-08
+
           sudo apt-get update
 
           sudo apt-get remove -y '^postgres.*' '^libpq.*'

crates/algorithm/src/lib.rs

@@ -12,7 +12,6 @@
 //
 // Copyright (c) 2025 TensorChord Inc.
 
-#![feature(select_unpredictable)]
 #![allow(clippy::type_complexity)]
 
 mod build;

crates/make/src/main.rs

@@ -83,6 +83,7 @@ fn main() -> Result<()> {
         eprintln!("Fork: {fork}");
         fork
     };
+    #[allow(clippy::collapsible_if)]
     let version = 'version: {
         for line in std::fs::read_to_string("./vchord.control")?.lines() {
             if let Some(prefix_stripped) = line.strip_prefix("default_version = '") {

crates/rabitq/src/lib.rs

@@ -12,9 +12,6 @@
 //
 // Copyright (c) 2025 TensorChord Inc.
 
-#![feature(slice_as_chunks)]
-#![feature(select_unpredictable)]
-
 pub mod binary;
 pub mod block;
 pub mod packing;

crates/simd/cshim/x86_64.c

@@ -16,19 +16,50 @@
 #if !(__clang_major__ >= 16)
 #error "Clang version must be at least 16."
 #endif
+#elif defined(__GNUC__)
+#if !(__GNUC__ >= 12)
+#error "GCC version must be at least 12."
+#endif
 #else
-#error "This file requires Clang."
+#error "This file requires Clang or GCC."
 #endif
 
 #include <immintrin.h>
 #include <stddef.h>
 #include <stdint.h>
 
+#if defined(__clang__) && defined(_MSC_VER) && (__clang_major__ <= 19)
+// https://github.com/llvm/llvm-project/issues/53520
+// clang-format off
+#include <pmmintrin.h>
+#include <tmmintrin.h>
+#include <smmintrin.h>
+#include <lzcntintrin.h>
+#include <avxintrin.h>
+#include <avx2intrin.h>
+#include <f16cintrin.h>
+#include <fmaintrin.h>
+#include <bmiintrin.h>
+#include <bmi2intrin.h>
+#include <lzcntintrin.h>
+#include <avx512fintrin.h>
+#include <avx512bwintrin.h>
+#include <avx512cdintrin.h>
+#include <avx512dqintrin.h>
+#include <avx512vlintrin.h>
+#include <avx512vlbwintrin.h>
+#include <avx512vlcdintrin.h>
+#include <avx512vldqintrin.h>
+#include <avx512fp16intrin.h>
+#include <avx512vlfp16intrin.h>
+// clang-format on
+#endif
+
 typedef _Float16 f16;
 typedef float f32;
 
 __attribute__((target("arch=x86-64-v4,avx512fp16"))) float
-fp16_reduce_sum_of_xy_v4fp16(f16 *restrict a, f16 *restrict b, size_t n) {
+fp16_reduce_sum_of_xy_v4_avx512fp16(f16 *restrict a, f16 *restrict b, size_t n) {
   __m512h xy = _mm512_setzero_ph();
   while (n >= 32) {
     __m512h x = _mm512_loadu_ph(a);
@@ -48,7 +79,7 @@ fp16_reduce_sum_of_xy_v4fp16(f16 *restrict a, f16 *restrict b, size_t n) {
 }
 
 __attribute__((target("arch=x86-64-v4,avx512fp16"))) float
-fp16_reduce_sum_of_d2_v4fp16(f16 *restrict a, f16 *restrict b, size_t n) {
+fp16_reduce_sum_of_d2_v4_avx512fp16(f16 *restrict a, f16 *restrict b, size_t n) {
   __m512h d2 = _mm512_setzero_ph();
   while (n >= 32) {
     __m512h x = _mm512_loadu_ph(a);

crates/simd/src/f16.rs

@@ -239,25 +239,29 @@ mod reduce_sum_of_xy {
 
     #[inline]
     #[cfg(target_arch = "x86_64")]
-    #[crate::target_cpu(enable = "v4fp16")]
-    pub fn reduce_sum_of_xy_v4fp16(lhs: &[f16], rhs: &[f16]) -> f32 {
+    #[crate::target_cpu(enable = "v4")]
+    #[target_feature(enable = "avx512fp16")]
+    pub fn reduce_sum_of_xy_v4_avx512fp16(lhs: &[f16], rhs: &[f16]) -> f32 {
         assert!(lhs.len() == rhs.len());
         unsafe {
             unsafe extern "C" {
-                unsafe fn fp16_reduce_sum_of_xy_v4fp16(a: *const (), b: *const (), n: usize)
-                -> f32;
+                unsafe fn fp16_reduce_sum_of_xy_v4_avx512fp16(
+                    a: *const (),
+                    b: *const (),
+                    n: usize,
+                ) -> f32;
             }
-            fp16_reduce_sum_of_xy_v4fp16(lhs.as_ptr().cast(), rhs.as_ptr().cast(), lhs.len())
+            fp16_reduce_sum_of_xy_v4_avx512fp16(lhs.as_ptr().cast(), rhs.as_ptr().cast(), lhs.len())
         }
     }
 
     #[cfg(all(target_arch = "x86_64", test, not(miri)))]
     #[test]
-    fn reduce_sum_of_xy_v4fp16_test() {
+    fn reduce_sum_of_xy_v4_avx512fp16_test() {
         use rand::Rng;
         const EPSILON: f32 = 2.0;
-        if !crate::is_cpu_detected!("v4fp16") {
-            println!("test {} ... skipped (v4fp16)", module_path!());
+        if !crate::is_cpu_detected!("v4") || !crate::is_feature_detected!("avx512fp16") {
+            println!("test {} ... skipped (v4:avx512fp16)", module_path!());
             return;
         }
         let mut rng = rand::rng();
@@ -272,7 +276,7 @@ mod reduce_sum_of_xy {
             for z in 3984..4016 {
                 let lhs = &lhs[..z];
                 let rhs = &rhs[..z];
-                let specialized = unsafe { reduce_sum_of_xy_v4fp16(lhs, rhs) };
+                let specialized = unsafe { reduce_sum_of_xy_v4_avx512fp16(lhs, rhs) };
                 let fallback = fallback(lhs, rhs);
                 assert!(
                     (specialized - fallback).abs() < EPSILON,
@@ -494,7 +498,7 @@ mod reduce_sum_of_xy {
         }
     }
 
-    #[crate::multiversion(@"v4fp16", @"v4", @"v3", @"a3.512", @"a2:fp16")]
+    #[crate::multiversion(@"v4:avx512fp16", @"v4", @"v3", @"a3.512", @"a2:fp16")]
     pub fn reduce_sum_of_xy(lhs: &[f16], rhs: &[f16]) -> f32 {
         assert!(lhs.len() == rhs.len());
         let n = lhs.len();
@@ -511,25 +515,29 @@ mod reduce_sum_of_d2 {
 
     #[inline]
     #[cfg(target_arch = "x86_64")]
-    #[crate::target_cpu(enable = "v4fp16")]
-    pub fn reduce_sum_of_d2_v4fp16(lhs: &[f16], rhs: &[f16]) -> f32 {
+    #[crate::target_cpu(enable = "v4")]
+    #[target_feature(enable = "avx512fp16")]
+    pub fn reduce_sum_of_d2_v4_avx512fp16(lhs: &[f16], rhs: &[f16]) -> f32 {
         assert!(lhs.len() == rhs.len());
         unsafe {
             unsafe extern "C" {
-                unsafe fn fp16_reduce_sum_of_d2_v4fp16(a: *const (), b: *const (), n: usize)
-                -> f32;
+                unsafe fn fp16_reduce_sum_of_d2_v4_avx512fp16(
+                    a: *const (),
+                    b: *const (),
+                    n: usize,
+                ) -> f32;
             }
-            fp16_reduce_sum_of_d2_v4fp16(lhs.as_ptr().cast(), rhs.as_ptr().cast(), lhs.len())
+            fp16_reduce_sum_of_d2_v4_avx512fp16(lhs.as_ptr().cast(), rhs.as_ptr().cast(), lhs.len())
         }
     }
 
     #[cfg(all(target_arch = "x86_64", test, not(miri)))]
     #[test]
-    fn reduce_sum_of_d2_v4fp16_test() {
+    fn reduce_sum_of_d2_v4_avx512fp16_test() {
         use rand::Rng;
         const EPSILON: f32 = 6.4;
-        if !crate::is_cpu_detected!("v4fp16") {
-            println!("test {} ... skipped (v4fp16)", module_path!());
+        if !crate::is_cpu_detected!("v4") || !crate::is_feature_detected!("avx512fp16") {
+            println!("test {} ... skipped (v4:avx512fp16)", module_path!());
             return;
         }
         let mut rng = rand::rng();
@@ -544,7 +552,7 @@ mod reduce_sum_of_d2 {
             for z in 3984..4016 {
                 let lhs = &lhs[..z];
                 let rhs = &rhs[..z];
-                let specialized = unsafe { reduce_sum_of_d2_v4fp16(lhs, rhs) };
+                let specialized = unsafe { reduce_sum_of_d2_v4_avx512fp16(lhs, rhs) };
                 let fallback = fallback(lhs, rhs);
                 assert!(
                     (specialized - fallback).abs() < EPSILON,
@@ -774,7 +782,7 @@ mod reduce_sum_of_d2 {
         }
     }
 
-    #[crate::multiversion(@"v4fp16", @"v4", @"v3", @"a3.512", @"a2:fp16")]
+    #[crate::multiversion(@"v4:avx512fp16", @"v4", @"v3", @"a3.512", @"a2:fp16")]
     pub fn reduce_sum_of_d2(lhs: &[f16], rhs: &[f16]) -> f32 {
         assert!(lhs.len() == rhs.len());
         let n = lhs.len();

crates/simd/src/lib.rs

@@ -12,10 +12,6 @@
 //
 // Copyright (c) 2025 TensorChord Inc.
 
-#![cfg_attr(target_arch = "x86_64", feature(avx512_target_feature))]
-#![cfg_attr(target_arch = "x86_64", feature(stdarch_x86_avx512))]
-#![feature(slice_as_chunks)]
-#![feature(select_unpredictable)]
 #![allow(unsafe_code)]
 
 mod aligned;
@@ -85,20 +81,6 @@ mod internal {
     #[allow(unused_imports)]
     pub use is_riscv64_cpu_detected;
 
-    #[cfg(target_arch = "x86_64")]
-    pub fn is_v4fp16_detected() -> bool {
-        std::arch::is_x86_feature_detected!("avx512fp16")
-            && std::arch::is_x86_feature_detected!("avx512bw")
-            && std::arch::is_x86_feature_detected!("avx512cd")
-            && std::arch::is_x86_feature_detected!("avx512dq")
-            && std::arch::is_x86_feature_detected!("avx512vl")
-            && std::arch::is_x86_feature_detected!("bmi1")
-            && std::arch::is_x86_feature_detected!("bmi2")
-            && std::arch::is_x86_feature_detected!("lzcnt")
-            && std::arch::is_x86_feature_detected!("movbe")
-            && std::arch::is_x86_feature_detected!("popcnt")
-    }
-
     #[cfg(target_arch = "x86_64")]
     pub fn is_v4_detected() -> bool {
         std::arch::is_x86_feature_detected!("avx512bw")

crates/simd_macros/src/target.rs

@@ -19,15 +19,6 @@ pub struct TargetCpu {
 }
 
 pub const TARGET_CPUS: &[TargetCpu] = &[
-    // This is a temporary hack, for using AVX512FP16 without unstable features.
-    TargetCpu {
-        target_cpu: "v4fp16",
-        target_arch: "x86_64",
-        target_features: &[
-            "avx512bw", "avx512cd", "avx512dq", "avx512vl", // simd
-            "bmi1", "bmi2", "lzcnt", "movbe", "popcnt", // bit-operations
-        ],
-    },
     TargetCpu {
         target_cpu: "v4",
         target_arch: "x86_64",

rust-toolchain.toml

@@ -1,2 +1,2 @@
 [toolchain]
-channel = "nightly-2025-04-26"
+channel = "nightly-2025-06-08"