Merge pull request #2106 from sayantn/cvtf16s16

Use LLVM intrinsics for `f16` to `{i,u}16` intrinsics

Author: folkertdev

ci/docker/aarch64-unknown-linux-gnu/Dockerfile

@@ -10,8 +10,14 @@ 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-22.1.4-x86_64.tar.gz -O llvm.tar.xz
+RUN mkdir llvm
+RUN tar -xvf llvm.tar.xz --strip-components=1 -C llvm
+
+ENV PATH="/llvm/bin:$PATH"
 
 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" \

ci/docker/aarch64_be-unknown-linux-gnu/Dockerfile

@@ -9,10 +9,9 @@ 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"
 
@@ -21,6 +20,12 @@ 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-22.1.4-x86_64.tar.gz -O llvm.tar.xz
+RUN mkdir llvm
+RUN tar -xvf llvm.tar.xz --strip-components=1 -C llvm
+
+ENV PATH="/llvm/bin:$PATH"
+
 ENV AARCH64_BE_TOOLCHAIN="/toolchains/${TOOLCHAIN}"
 ENV AARCH64_BE_LIBC="${AARCH64_BE_TOOLCHAIN}/aarch64_be-none-linux-gnu/libc"
 

ci/docker/armv7-unknown-linux-gnueabihf/Dockerfile

@@ -10,8 +10,14 @@ 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-22.1.4-x86_64.tar.gz -O llvm.tar.xz
+RUN mkdir llvm
+RUN tar -xvf llvm.tar.xz --strip-components=1 -C llvm
+
+ENV PATH="/llvm/bin:$PATH"
+
 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

ci/docker/x86_64-unknown-linux-gnu/Dockerfile

@@ -15,6 +15,13 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
 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-22.1.4-x86_64.tar.gz -O llvm.tar.xz
+RUN mkdir llvm
+RUN tar -xvf llvm.tar.xz --strip-components=1 -C llvm
+
+ENV PATH="/llvm/bin:$PATH"
+
 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 --"

crates/core_arch/src/aarch64/neon/generated.rs

@@ -5516,7 +5516,14 @@ pub fn vcvtaq_u64_f64(a: float64x2_t) -> uint64x2_t {
 #[unstable(feature = "stdarch_neon_f16", issue = "136306")]
 #[cfg(not(target_arch = "arm64ec"))]
 pub fn vcvtah_s16_f16(a: f16) -> i16 {
-    vcvtah_s32_f16(a) as i16
+    unsafe extern "unadjusted" {
+        #[cfg_attr(
+            any(target_arch = "aarch64", target_arch = "arm64ec"),
+            link_name = "llvm.aarch64.neon.fcvtas.i16.f16"
+        )]
+        fn _vcvtah_s16_f16(a: f16) -> i16;
+    }
+    unsafe { _vcvtah_s16_f16(a) }
 }
 #[doc = "Floating-point convert to integer, rounding to nearest with ties to away"]
 #[doc = "[Arm's documentation](https://developer.arm.com/architectures/instruction-sets/intrinsics/vcvtah_s32_f16)"]
@@ -5560,7 +5567,14 @@ pub fn vcvtah_s64_f16(a: f16) -> i64 {
 #[unstable(feature = "stdarch_neon_f16", issue = "136306")]
 #[cfg(not(target_arch = "arm64ec"))]
 pub fn vcvtah_u16_f16(a: f16) -> u16 {
-    vcvtah_u32_f16(a) as u16
+    unsafe extern "unadjusted" {
+        #[cfg_attr(
+            any(target_arch = "aarch64", target_arch = "arm64ec"),
+            link_name = "llvm.aarch64.neon.fcvtau.i16.f16"
+        )]
+        fn _vcvtah_u16_f16(a: f16) -> u16;
+    }
+    unsafe { _vcvtah_u16_f16(a) }
 }
 #[doc = "Floating-point convert to integer, rounding to nearest with ties to away"]
 #[doc = "[Arm's documentation](https://developer.arm.com/architectures/instruction-sets/intrinsics/vcvtah_u32_f16)"]
@@ -6202,7 +6216,14 @@ pub fn vcvtmq_u64_f64(a: float64x2_t) -> uint64x2_t {
 #[unstable(feature = "stdarch_neon_f16", issue = "136306")]
 #[cfg(not(target_arch = "arm64ec"))]
 pub fn vcvtmh_s16_f16(a: f16) -> i16 {
-    vcvtmh_s32_f16(a) as i16
+    unsafe extern "unadjusted" {
+        #[cfg_attr(
+            any(target_arch = "aarch64", target_arch = "arm64ec"),
+            link_name = "llvm.aarch64.neon.fcvtms.i16.f16"
+        )]
+        fn _vcvtmh_s16_f16(a: f16) -> i16;
+    }
+    unsafe { _vcvtmh_s16_f16(a) }
 }
 #[doc = "Floating-point convert to integer, rounding towards minus infinity"]
 #[doc = "[Arm's documentation](https://developer.arm.com/architectures/instruction-sets/intrinsics/vcvtmh_s32_f16)"]
@@ -6238,15 +6259,22 @@ pub fn vcvtmh_s64_f16(a: f16) -> i64 {
     }
     unsafe { _vcvtmh_s64_f16(a) }
 }
-#[doc = "Floating-point convert to integer, rounding towards minus infinity"]
+#[doc = "Floating-point convert to unsigned integer, rounding towards minus infinity"]
 #[doc = "[Arm's documentation](https://developer.arm.com/architectures/instruction-sets/intrinsics/vcvtmh_u16_f16)"]
 #[inline]
 #[cfg_attr(test, assert_instr(fcvtmu))]
 #[target_feature(enable = "neon,fp16")]
 #[unstable(feature = "stdarch_neon_f16", issue = "136306")]
 #[cfg(not(target_arch = "arm64ec"))]
 pub fn vcvtmh_u16_f16(a: f16) -> u16 {
-    vcvtmh_u32_f16(a) as u16
+    unsafe extern "unadjusted" {
+        #[cfg_attr(
+            any(target_arch = "aarch64", target_arch = "arm64ec"),
+            link_name = "llvm.aarch64.neon.fcvtmu.i16.f16"
+        )]
+        fn _vcvtmh_u16_f16(a: f16) -> u16;
+    }
+    unsafe { _vcvtmh_u16_f16(a) }
 }
 #[doc = "Floating-point convert to unsigned integer, rounding towards minus infinity"]
 #[doc = "[Arm's documentation](https://developer.arm.com/architectures/instruction-sets/intrinsics/vcvtmh_u32_f16)"]
@@ -6550,7 +6578,14 @@ pub fn vcvtnq_u64_f64(a: float64x2_t) -> uint64x2_t {
 #[unstable(feature = "stdarch_neon_f16", issue = "136306")]
 #[cfg(not(target_arch = "arm64ec"))]
 pub fn vcvtnh_s16_f16(a: f16) -> i16 {
-    vcvtnh_s32_f16(a) as i16
+    unsafe extern "unadjusted" {
+        #[cfg_attr(
+            any(target_arch = "aarch64", target_arch = "arm64ec"),
+            link_name = "llvm.aarch64.neon.fcvtns.i16.f16"
+        )]
+        fn _vcvtnh_s16_f16(a: f16) -> i16;
+    }
+    unsafe { _vcvtnh_s16_f16(a) }
 }
 #[doc = "Floating-point convert to integer, rounding to nearest with ties to even"]
 #[doc = "[Arm's documentation](https://developer.arm.com/architectures/instruction-sets/intrinsics/vcvtnh_s32_f16)"]
@@ -6594,7 +6629,14 @@ pub fn vcvtnh_s64_f16(a: f16) -> i64 {
 #[unstable(feature = "stdarch_neon_f16", issue = "136306")]
 #[cfg(not(target_arch = "arm64ec"))]
 pub fn vcvtnh_u16_f16(a: f16) -> u16 {
-    vcvtnh_u32_f16(a) as u16
+    unsafe extern "unadjusted" {
+        #[cfg_attr(
+            any(target_arch = "aarch64", target_arch = "arm64ec"),
+            link_name = "llvm.aarch64.neon.fcvtnu.i16.f16"
+        )]
+        fn _vcvtnh_u16_f16(a: f16) -> u16;
+    }
+    unsafe { _vcvtnh_u16_f16(a) }
 }
 #[doc = "Floating-point convert to unsigned integer, rounding to nearest with ties to even"]
 #[doc = "[Arm's documentation](https://developer.arm.com/architectures/instruction-sets/intrinsics/vcvtnh_u32_f16)"]
@@ -6898,7 +6940,14 @@ pub fn vcvtpq_u64_f64(a: float64x2_t) -> uint64x2_t {
 #[unstable(feature = "stdarch_neon_f16", issue = "136306")]
 #[cfg(not(target_arch = "arm64ec"))]
 pub fn vcvtph_s16_f16(a: f16) -> i16 {
-    vcvtph_s32_f16(a) as i16
+    unsafe extern "unadjusted" {
+        #[cfg_attr(
+            any(target_arch = "aarch64", target_arch = "arm64ec"),
+            link_name = "llvm.aarch64.neon.fcvtps.i16.f16"
+        )]
+        fn _vcvtph_s16_f16(a: f16) -> i16;
+    }
+    unsafe { _vcvtph_s16_f16(a) }
 }
 #[doc = "Floating-point convert to integer, rounding to plus infinity"]
 #[doc = "[Arm's documentation](https://developer.arm.com/architectures/instruction-sets/intrinsics/vcvtph_s32_f16)"]
@@ -6942,7 +6991,14 @@ pub fn vcvtph_s64_f16(a: f16) -> i64 {
 #[unstable(feature = "stdarch_neon_f16", issue = "136306")]
 #[cfg(not(target_arch = "arm64ec"))]
 pub fn vcvtph_u16_f16(a: f16) -> u16 {
-    vcvtph_u32_f16(a) as u16
+    unsafe extern "unadjusted" {
+        #[cfg_attr(
+            any(target_arch = "aarch64", target_arch = "arm64ec"),
+            link_name = "llvm.aarch64.neon.fcvtpu.i16.f16"
+        )]
+        fn _vcvtph_u16_f16(a: f16) -> u16;
+    }
+    unsafe { _vcvtph_u16_f16(a) }
 }
 #[doc = "Floating-point convert to unsigned integer, rounding to plus infinity"]
 #[doc = "[Arm's documentation](https://developer.arm.com/architectures/instruction-sets/intrinsics/vcvtph_u32_f16)"]