fixes

sayantn · sayantn · commit a764f33d98fd · 2026-04-25T02:09:46.000+05:30
diff --git a/.github/workflows/main.yml b/.github/workflows/main.yml
@@ -274,6 +274,7 @@ jobs:
     name: Intrinsic Test
     runs-on: ubuntu-latest
     strategy:
+      fail-fast: false
       matrix:
         target:
           - aarch64-unknown-linux-gnu
diff --git a/ci/docker/aarch64-unknown-linux-gnu/Dockerfile b/ci/docker/aarch64-unknown-linux-gnu/Dockerfile
@@ -8,8 +8,12 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
   qemu-user \
   make \
   file \
-  clang \
-  lld
+  xz-utils \
+  wget
+
+RUN wget https://mirrors.edge.kernel.org/pub/tools/llvm/files/llvm-22.1.1-x86_64.tar.gz -O llvm.tar.xz
+RUN mkdir llvm
+RUN tar -xvf llvm.tar.xz --strip-components=1 -C llvm
 
 ENV CARGO_TARGET_AARCH64_UNKNOWN_LINUX_GNU_LINKER=aarch64-linux-gnu-gcc \
     CARGO_TARGET_AARCH64_UNKNOWN_LINUX_GNU_RUNNER="qemu-aarch64 -cpu max -L /usr/aarch64-linux-gnu" \
diff --git a/ci/docker/aarch64_be-unknown-linux-gnu/Dockerfile b/ci/docker/aarch64_be-unknown-linux-gnu/Dockerfile
@@ -8,10 +8,9 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
   qemu-user \
   make \
   file \
-  clang \
   curl \
   xz-utils \
-  lld
+  wget
 
 ENV TOOLCHAIN="arm-gnu-toolchain-14.3.rel1-x86_64-aarch64_be-none-linux-gnu"
 
@@ -20,6 +19,10 @@ RUN curl -L "https://developer.arm.com/-/media/Files/downloads/gnu/14.3.rel1/bin
 RUN tar -xvf "${TOOLCHAIN}.tar.xz"
 RUN mkdir /toolchains && mv "./${TOOLCHAIN}" /toolchains
 
+RUN wget https://mirrors.edge.kernel.org/pub/tools/llvm/files/llvm-22.1.1-x86_64.tar.gz -O llvm.tar.xz
+RUN mkdir llvm
+RUN tar -xvf llvm.tar.xz --strip-components=1 -C llvm
+
 ENV AARCH64_BE_TOOLCHAIN="/toolchains/${TOOLCHAIN}"
 ENV AARCH64_BE_LIBC="${AARCH64_BE_TOOLCHAIN}/aarch64_be-none-linux-gnu/libc"
 
diff --git a/ci/docker/armv7-unknown-linux-gnueabihf/Dockerfile b/ci/docker/armv7-unknown-linux-gnueabihf/Dockerfile
@@ -8,8 +8,12 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
   qemu-user \
   make \
   file \
-  clang \
-  lld
+  wget
+
+RUN wget https://mirrors.edge.kernel.org/pub/tools/llvm/files/llvm-22.1.1-x86_64.tar.gz -O llvm.tar.xz
+RUN mkdir llvm
+RUN tar -xvf llvm.tar.xz --strip-components=1 -C llvm
+
 ENV CARGO_TARGET_ARMV7_UNKNOWN_LINUX_GNUEABIHF_LINKER=arm-linux-gnueabihf-gcc \
     CARGO_TARGET_ARMV7_UNKNOWN_LINUX_GNUEABIHF_RUNNER="qemu-arm -cpu max -L /usr/arm-linux-gnueabihf" \
     OBJDUMP=arm-linux-gnueabihf-objdump
diff --git a/ci/docker/x86_64-unknown-linux-gnu/Dockerfile b/ci/docker/x86_64-unknown-linux-gnu/Dockerfile
@@ -6,15 +6,16 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
   make \
   ca-certificates \
   wget \
-  xz-utils \
-  clang \
-  libstdc++-14-dev \
-  build-essential \
-  lld
+  xz-utils
 
 RUN wget http://ci-mirrors.rust-lang.org/sde-external-10.8.0-2026-03-15-lin.tar.xz -O sde.tar.xz
 RUN mkdir intel-sde
 RUN tar -xJf sde.tar.xz --strip-components=1 -C intel-sde
+
+RUN wget https://mirrors.edge.kernel.org/pub/tools/llvm/files/llvm-22.1.1-x86_64.tar.gz -O llvm.tar.xz
+RUN mkdir llvm
+RUN tar -xvf llvm.tar.xz --strip-components=1 -C llvm
+
 ENV CARGO_TARGET_X86_64_UNKNOWN_LINUX_GNU_RUNNER="/intel-sde/sde64 \
             -cpuid-in /checkout/ci/docker/x86_64-unknown-linux-gnu/cpuid.def \
             -rtm-mode full -tsx --"
diff --git a/ci/intrinsic-test.sh b/ci/intrinsic-test.sh
@@ -12,7 +12,7 @@ echo "PROFILE=${PROFILE}"
 
 INTRINSIC_TEST="--manifest-path=crates/intrinsic-test/Cargo.toml"
 
-export CC="clang"
+export CC="/llvm/bin/clang"
 
 case ${TARGET} in
     aarch64_be*)
@@ -67,4 +67,4 @@ case ${TARGET} in
         ;;
 esac
 
-cargo test --manifest-path=rust_programs/Cargo.toml --target "${TARGET}" --profile "${PROFILE}"
+cargo test --manifest-path=rust_programs/Cargo.toml --target "${TARGET}" --profile "${PROFILE}" --no-fail-fast
diff --git a/crates/intrinsic-test/missing_x86.txt b/crates/intrinsic-test/missing_x86.txt
@@ -52,8 +52,19 @@ _mm_sm3rnds2_epi32
 _xgetbv
 
 # top bits are undefined, unclear how to test these
+_mm256_castph128_ph256
+_mm256_castps128_ps256
+_mm256_castpd128_pd256
 _mm256_castsi128_si256
+
+_mm512_castph128_ph512
+_mm512_castps128_ps512
+_mm512_castpd128_pd512
 _mm512_castsi128_si512
+
+_mm512_castph256_ph512
+_mm512_castps256_ps512
+_mm512_castpd256_pd512
 _mm512_castsi256_si512
 
 # Clang bug
diff --git a/crates/intrinsic-test/src/arm/types.rs b/crates/intrinsic-test/src/arm/types.rs
@@ -1,5 +1,4 @@
 use super::intrinsic::ArmIntrinsicType;
-use crate::common::cli::Language;
 use crate::common::intrinsic_helpers::{IntrinsicType, IntrinsicTypeDefinition, Sign, TypeKind};
 
 impl IntrinsicTypeDefinition for ArmIntrinsicType {
@@ -36,7 +35,7 @@ impl IntrinsicTypeDefinition for ArmIntrinsicType {
     }
 
     /// Determines the load function for this type.
-    fn get_load_function(&self, language: Language) -> String {
+    fn get_load_function(&self) -> String {
         if let IntrinsicType {
             kind: k,
             bit_len: Some(bl),
@@ -51,16 +50,13 @@ impl IntrinsicTypeDefinition for ArmIntrinsicType {
                 ""
             };
 
-            let choose_workaround = language == Language::C && self.target.contains("v7");
             format!(
                 "vld{len}{quad}_{type}{size}",
                 type = match k {
                     TypeKind::Int(Sign::Unsigned) => "u",
                     TypeKind::Int(Sign::Signed) => "s",
                     TypeKind::Float => "f",
-                    // The ACLE doesn't support 64-bit polynomial loads on Armv7
-                    // if armv7 and bl == 64, use "s", else "p"
-                    TypeKind::Poly => if choose_workaround && *bl == 64 {"s"} else {"p"},
+                    TypeKind::Poly => "p",
                     x => todo!("get_load_function TypeKind: {x:#?}"),
                 },
                 size = bl,
diff --git a/crates/intrinsic-test/src/common/argument.rs b/crates/intrinsic-test/src/common/argument.rs
@@ -1,6 +1,5 @@
 use itertools::Itertools;
 
-use super::cli::Language;
 use super::constraint::Constraint;
 use super::indentation::Indentation;
 use super::intrinsic_helpers::IntrinsicTypeDefinition;
@@ -171,7 +170,7 @@ where
             .enumerate()
             .map(|(idx, arg)| {
                 let load = if arg.is_simd() {
-                    arg.ty.get_load_function(Language::Rust)
+                    arg.ty.get_load_function()
                 } else {
                     "*".to_string()
                 };
diff --git a/crates/intrinsic-test/src/common/cli.rs b/crates/intrinsic-test/src/common/cli.rs
@@ -1,12 +1,6 @@
 use itertools::Itertools;
 use std::path::PathBuf;
 
-#[derive(Debug, PartialEq)]
-pub enum Language {
-    Rust,
-    C,
-}
-
 /// Intrinsic test tool
 #[derive(clap::Parser)]
 #[command(
diff --git a/crates/intrinsic-test/src/common/gen_rust.rs b/crates/intrinsic-test/src/common/gen_rust.rs
@@ -275,7 +275,7 @@ pub fn write_build_rs(
     i: usize,
     arch_flags: &[&str],
 ) -> std::io::Result<()> {
-    const COMMON_FLAGS: &[&str] = &["-ffp-contract=off", "-Wno-narrowing", "-fuse-ld=lld"];
+    const COMMON_FLAGS: &[&str] = &["-ffp-contract=off", "-Wno-narrowing"];
 
     write!(
         w,
diff --git a/crates/intrinsic-test/src/common/intrinsic_helpers.rs b/crates/intrinsic-test/src/common/intrinsic_helpers.rs
@@ -5,7 +5,6 @@ use std::str::FromStr;
 
 use itertools::Itertools as _;
 
-use super::cli::Language;
 use super::indentation::Indentation;
 use super::values::value_for_array;
 
@@ -252,7 +251,7 @@ impl IntrinsicType {
 pub trait IntrinsicTypeDefinition: Deref<Target = IntrinsicType> {
     /// Determines the load function for this type.
     /// can be implemented in an `impl` block
-    fn get_load_function(&self, _language: Language) -> String;
+    fn get_load_function(&self) -> String;
 
     /// Gets a string containing the typename for this type in C format.
     /// can be directly defined in `impl` blocks
diff --git a/crates/intrinsic-test/src/common/mod.rs b/crates/intrinsic-test/src/common/mod.rs
@@ -105,6 +105,6 @@ pub trait SupportedArchitectureTest {
 }
 
 pub fn manual_chunk(intrinsic_count: usize) -> (usize, usize) {
-    let ncores = std::thread::available_parallelism().unwrap();
-    (intrinsic_count / ncores, ncores.into())
+    let ncores = std::thread::available_parallelism().unwrap().into();
+    (intrinsic_count.div_ceil(ncores), ncores)
 }
diff --git a/crates/intrinsic-test/src/x86/config.rs b/crates/intrinsic-test/src/x86/config.rs
@@ -131,14 +131,8 @@ unsafe fn _mm512_loadu_epi64_to___m512(mem_addr: *const i64) -> __m512 {
 "#;
 
 pub const PLATFORM_RUST_CFGS: &str = r#"
-#![cfg_attr(target_arch = "x86", feature(avx))]
-#![cfg_attr(target_arch = "x86", feature(sse))]
-#![cfg_attr(target_arch = "x86", feature(sse2))]
-#![cfg_attr(target_arch = "x86", feature(stdarch_x86_avx512_bf16))]
-#![cfg_attr(target_arch = "x86", feature(stdarch_x86_avx512_f16))]
-#![cfg_attr(target_arch = "x86", feature(stdarch_x86_rtm))]
-#![cfg_attr(target_arch = "x86", feature(stdarch_x86_rtm))]
-#![cfg_attr(target_arch = "x86_64", feature(x86_amx_intrinsics))]
-#![cfg_attr(target_arch = "x86_64", feature(stdarch_x86_avx512_f16))]
-#![feature(fmt_helpers_for_derive)]
+#![feature(stdarch_x86_avx512_bf16)]
+#![feature(stdarch_x86_avx512_f16)]
+#![feature(stdarch_x86_rtm)]
+#![feature(x86_amx_intrinsics)]
 "#;
diff --git a/crates/intrinsic-test/src/x86/mod.rs b/crates/intrinsic-test/src/x86/mod.rs
@@ -62,6 +62,7 @@ impl SupportedArchitectureTest for X86ArchitectureTest {
             "-mgfni",
             "-mvaes",
             "-mvpclmulqdq",
+            "-mlzcnt",
         ]
     }
 
diff --git a/crates/intrinsic-test/src/x86/types.rs b/crates/intrinsic-test/src/x86/types.rs
@@ -3,7 +3,6 @@ use std::str::FromStr;
 use itertools::Itertools;
 
 use super::intrinsic::X86IntrinsicType;
-use crate::common::cli::Language;
 use crate::common::intrinsic_helpers::{IntrinsicType, IntrinsicTypeDefinition, Sign, TypeKind};
 use crate::x86::xml_parser::Parameter;
 
@@ -33,7 +32,7 @@ impl IntrinsicTypeDefinition for X86IntrinsicType {
     }
 
     /// Determines the load function for this type.
-    fn get_load_function(&self, _language: Language) -> String {
+    fn get_load_function(&self) -> String {
         let type_value = self.param.type_data.clone();
         if type_value.len() == 0 {
             unimplemented!("the value for key 'type' is not present!");
