Skip to content

loongarch: Use intrinsics::simd for selected LSX/LASX intrinsics #2078

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
heiher wants to merge 2 commits into
base: main
Choose a base branch
from
+1,936 −4,132
Open

loongarch: Use intrinsics::simd for selected LSX/LASX intrinsics #2078

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
a690147
stdarch-gen-loongarch: Support marking intrinsics as portable
heiher Mar 20, 2026
4dd9159
loongarch: Use `intrinsics::simd` for selected LSX/LASX intrinsics
heiher Apr 9, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2,204 changes: 203 additions & 2,001 deletions crates/core_arch/src/loongarch64/lasx/generated.rs
View file
Open in desktop

Large diffs are not rendered by default.

7 changes: 7 additions & 0 deletions crates/core_arch/src/loongarch64/lasx/mod.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,6 +16,13 @@ mod generated;
#[unstable(feature = "stdarch_loongarch", issue = "117427")]
pub use self::generated::*;

#[rustfmt::skip]
mod portable;

#[rustfmt::skip]
#[unstable(feature = "stdarch_loongarch", issue = "117427")]
pub use self::portable::*;

#[rustfmt::skip]
#[cfg(test)]
mod tests;
202 changes: 202 additions & 0 deletions crates/core_arch/src/loongarch64/lasx/portable.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,202 @@
//! LoongArch64 LASX intrinsics - intrinsics::simd implementation

use super::super::{simd::*, *};
use crate::core_arch::simd::*;
use crate::intrinsics::simd::*;
use crate::mem::transmute;

impl_vv!("lasx", lasx_xvpcnt_b, simd_ctpop, m256i, i8x32);
impl_vv!("lasx", lasx_xvpcnt_h, simd_ctpop, m256i, i16x16);
impl_vv!("lasx", lasx_xvpcnt_w, simd_ctpop, m256i, i32x8);
impl_vv!("lasx", lasx_xvpcnt_d, simd_ctpop, m256i, i64x4);
impl_vv!("lasx", lasx_xvclz_b, simd_ctlz, m256i, i8x32);
impl_vv!("lasx", lasx_xvclz_h, simd_ctlz, m256i, i16x16);
impl_vv!("lasx", lasx_xvclz_w, simd_ctlz, m256i, i32x8);
impl_vv!("lasx", lasx_xvclz_d, simd_ctlz, m256i, i64x4);
impl_vv!("lasx", lasx_xvneg_b, simd_neg, m256i, i8x32);
impl_vv!("lasx", lasx_xvneg_h, simd_neg, m256i, i16x16);
impl_vv!("lasx", lasx_xvneg_w, simd_neg, m256i, i32x8);
impl_vv!("lasx", lasx_xvneg_d, simd_neg, m256i, i64x4);
impl_vv!("lasx", lasx_xvfsqrt_s, simd_fsqrt, m256, f32x8);
impl_vv!("lasx", lasx_xvfsqrt_d, simd_fsqrt, m256d, f64x4);

impl_gv!("lasx", lasx_xvreplgr2vr_b, simd_splat, m256i, i8x32, i8, i32);
impl_gv!("lasx", lasx_xvreplgr2vr_h, simd_splat, m256i, i16x16, i16, i32);
impl_gv!("lasx", lasx_xvreplgr2vr_w, simd_splat, m256i, i32x8, i32, i32);
impl_gv!("lasx", lasx_xvreplgr2vr_d, simd_splat, m256i, i64x4, i64, i64);

impl_sv!("lasx", lasx_xvrepli_b, simd_splat, m256i, i8x32, i8, 10);
impl_sv!("lasx", lasx_xvrepli_h, simd_splat, m256i, i16x16, i16, 10);
impl_sv!("lasx", lasx_xvrepli_w, simd_splat, m256i, i32x8, i32, 10);
impl_sv!("lasx", lasx_xvrepli_d, simd_splat, m256i, i64x4, i64, 10);

impl_vvv!("lasx", lasx_xvadd_b, simd_add, m256i, i8x32);
impl_vvv!("lasx", lasx_xvadd_h, simd_add, m256i, i16x16);
impl_vvv!("lasx", lasx_xvadd_w, simd_add, m256i, i32x8);
impl_vvv!("lasx", lasx_xvadd_d, simd_add, m256i, i64x4);
impl_vvv!("lasx", lasx_xvsub_b, simd_sub, m256i, i8x32);
impl_vvv!("lasx", lasx_xvsub_h, simd_sub, m256i, i16x16);
impl_vvv!("lasx", lasx_xvsub_w, simd_sub, m256i, i32x8);
impl_vvv!("lasx", lasx_xvsub_d, simd_sub, m256i, i64x4);
impl_vvv!("lasx", lasx_xvmax_b, simd_imax, m256i, i8x32);
impl_vvv!("lasx", lasx_xvmax_h, simd_imax, m256i, i16x16);
impl_vvv!("lasx", lasx_xvmax_w, simd_imax, m256i, i32x8);
impl_vvv!("lasx", lasx_xvmax_d, simd_imax, m256i, i64x4);
impl_vvv!("lasx", lasx_xvmax_bu, simd_imax, m256i, u8x32);
impl_vvv!("lasx", lasx_xvmax_hu, simd_imax, m256i, u16x16);
impl_vvv!("lasx", lasx_xvmax_wu, simd_imax, m256i, u32x8);
impl_vvv!("lasx", lasx_xvmax_du, simd_imax, m256i, u64x4);
impl_vvv!("lasx", lasx_xvmin_b, simd_imin, m256i, i8x32);
impl_vvv!("lasx", lasx_xvmin_h, simd_imin, m256i, i16x16);
impl_vvv!("lasx", lasx_xvmin_w, simd_imin, m256i, i32x8);
impl_vvv!("lasx", lasx_xvmin_d, simd_imin, m256i, i64x4);
impl_vvv!("lasx", lasx_xvmin_bu, simd_imin, m256i, u8x32);
impl_vvv!("lasx", lasx_xvmin_hu, simd_imin, m256i, u16x16);
impl_vvv!("lasx", lasx_xvmin_wu, simd_imin, m256i, u32x8);
impl_vvv!("lasx", lasx_xvmin_du, simd_imin, m256i, u64x4);
impl_vvv!("lasx", lasx_xvseq_b, simd_eq, m256i, i8x32);
impl_vvv!("lasx", lasx_xvseq_h, simd_eq, m256i, i16x16);
impl_vvv!("lasx", lasx_xvseq_w, simd_eq, m256i, i32x8);
impl_vvv!("lasx", lasx_xvseq_d, simd_eq, m256i, i64x4);
impl_vvv!("lasx", lasx_xvslt_b, simd_lt, m256i, i8x32);
impl_vvv!("lasx", lasx_xvslt_h, simd_lt, m256i, i16x16);
impl_vvv!("lasx", lasx_xvslt_w, simd_lt, m256i, i32x8);
impl_vvv!("lasx", lasx_xvslt_d, simd_lt, m256i, i64x4);
impl_vvv!("lasx", lasx_xvslt_bu, simd_lt, m256i, u8x32);
impl_vvv!("lasx", lasx_xvslt_hu, simd_lt, m256i, u16x16);
impl_vvv!("lasx", lasx_xvslt_wu, simd_lt, m256i, u32x8);
impl_vvv!("lasx", lasx_xvslt_du, simd_lt, m256i, u64x4);
impl_vvv!("lasx", lasx_xvsle_b, simd_le, m256i, i8x32);
impl_vvv!("lasx", lasx_xvsle_h, simd_le, m256i, i16x16);
impl_vvv!("lasx", lasx_xvsle_w, simd_le, m256i, i32x8);
impl_vvv!("lasx", lasx_xvsle_d, simd_le, m256i, i64x4);
impl_vvv!("lasx", lasx_xvsle_bu, simd_le, m256i, u8x32);
impl_vvv!("lasx", lasx_xvsle_hu, simd_le, m256i, u16x16);
impl_vvv!("lasx", lasx_xvsle_wu, simd_le, m256i, u32x8);
impl_vvv!("lasx", lasx_xvsle_du, simd_le, m256i, u64x4);
impl_vvv!("lasx", lasx_xvmul_b, simd_mul, m256i, i8x32);
impl_vvv!("lasx", lasx_xvmul_h, simd_mul, m256i, i16x16);
impl_vvv!("lasx", lasx_xvmul_w, simd_mul, m256i, i32x8);
impl_vvv!("lasx", lasx_xvmul_d, simd_mul, m256i, i64x4);
impl_vvv!("lasx", lasx_xvdiv_b, simd_div, m256i, i8x32);
impl_vvv!("lasx", lasx_xvdiv_h, simd_div, m256i, i16x16);
impl_vvv!("lasx", lasx_xvdiv_w, simd_div, m256i, i32x8);
impl_vvv!("lasx", lasx_xvdiv_d, simd_div, m256i, i64x4);
impl_vvv!("lasx", lasx_xvdiv_bu, simd_div, m256i, u8x32);
impl_vvv!("lasx", lasx_xvdiv_hu, simd_div, m256i, u16x16);
impl_vvv!("lasx", lasx_xvdiv_wu, simd_div, m256i, u32x8);
impl_vvv!("lasx", lasx_xvdiv_du, simd_div, m256i, u64x4);
impl_vvv!("lasx", lasx_xvmod_b, simd_rem, m256i, i8x32);
impl_vvv!("lasx", lasx_xvmod_h, simd_rem, m256i, i16x16);
impl_vvv!("lasx", lasx_xvmod_w, simd_rem, m256i, i32x8);
impl_vvv!("lasx", lasx_xvmod_d, simd_rem, m256i, i64x4);
impl_vvv!("lasx", lasx_xvmod_bu, simd_rem, m256i, u8x32);
impl_vvv!("lasx", lasx_xvmod_hu, simd_rem, m256i, u16x16);
impl_vvv!("lasx", lasx_xvmod_wu, simd_rem, m256i, u32x8);
impl_vvv!("lasx", lasx_xvmod_du, simd_rem, m256i, u64x4);
impl_vvv!("lasx", lasx_xvand_v, simd_and, m256i, u8x32);
impl_vvv!("lasx", lasx_xvandn_v, simd_andn, m256i, u8x32);
impl_vvv!("lasx", lasx_xvor_v, simd_or, m256i, u8x32);
impl_vvv!("lasx", lasx_xvorn_v, simd_orn, m256i, u8x32);
impl_vvv!("lasx", lasx_xvnor_v, simd_nor, m256i, u8x32);
impl_vvv!("lasx", lasx_xvxor_v, simd_xor, m256i, u8x32);
impl_vvv!("lasx", lasx_xvfadd_s, simd_add, m256, f32x8);
impl_vvv!("lasx", lasx_xvfadd_d, simd_add, m256d, f64x4);
impl_vvv!("lasx", lasx_xvfsub_s, simd_sub, m256, f32x8);
impl_vvv!("lasx", lasx_xvfsub_d, simd_sub, m256d, f64x4);
impl_vvv!("lasx", lasx_xvfmul_s, simd_mul, m256, f32x8);
impl_vvv!("lasx", lasx_xvfmul_d, simd_mul, m256d, f64x4);
impl_vvv!("lasx", lasx_xvfdiv_s, simd_div, m256, f32x8);
impl_vvv!("lasx", lasx_xvfdiv_d, simd_div, m256d, f64x4);

impl_vvv_s!("lasx", lasx_xvsll_b, simd_shl, m256i, i8x32, i8);
impl_vvv_s!("lasx", lasx_xvsll_h, simd_shl, m256i, i16x16, i16);
impl_vvv_s!("lasx", lasx_xvsll_w, simd_shl, m256i, i32x8, i32);
impl_vvv_s!("lasx", lasx_xvsll_d, simd_shl, m256i, i64x4, i64);
impl_vvv_s!("lasx", lasx_xvsra_b, simd_shr, m256i, i8x32, i8);
impl_vvv_s!("lasx", lasx_xvsra_h, simd_shr, m256i, i16x16, i16);
impl_vvv_s!("lasx", lasx_xvsra_w, simd_shr, m256i, i32x8, i32);
impl_vvv_s!("lasx", lasx_xvsra_d, simd_shr, m256i, i64x4, i64);
impl_vvv_s!("lasx", lasx_xvsrl_b, simd_shr, m256i, u8x32, u8);
impl_vvv_s!("lasx", lasx_xvsrl_h, simd_shr, m256i, u16x16, u16);
impl_vvv_s!("lasx", lasx_xvsrl_w, simd_shr, m256i, u32x8, u32);
impl_vvv_s!("lasx", lasx_xvsrl_d, simd_shr, m256i, u64x4, u64);

impl_vuv!("lasx", lasx_xvslli_b, simd_shl, m256i, i8x32, i8);
impl_vuv!("lasx", lasx_xvslli_h, simd_shl, m256i, i16x16, i16);
impl_vuv!("lasx", lasx_xvslli_w, simd_shl, m256i, i32x8, i32);
impl_vuv!("lasx", lasx_xvslli_d, simd_shl, m256i, i64x4, i64);
impl_vuv!("lasx", lasx_xvsrai_b, simd_shr, m256i, i8x32, i8);
impl_vuv!("lasx", lasx_xvsrai_h, simd_shr, m256i, i16x16, i16);
impl_vuv!("lasx", lasx_xvsrai_w, simd_shr, m256i, i32x8, i32);
impl_vuv!("lasx", lasx_xvsrai_d, simd_shr, m256i, i64x4, i64);
impl_vuv!("lasx", lasx_xvsrli_b, simd_shr, m256i, u8x32, u8);
impl_vuv!("lasx", lasx_xvsrli_h, simd_shr, m256i, u16x16, u16);
impl_vuv!("lasx", lasx_xvsrli_w, simd_shr, m256i, u32x8, u32);
impl_vuv!("lasx", lasx_xvsrli_d, simd_shr, m256i, u64x4, u64);
impl_vuv!("lasx", lasx_xvaddi_bu, simd_add, m256i, u8x32, u8, 5);
impl_vuv!("lasx", lasx_xvaddi_hu, simd_add, m256i, u16x16, u16, 5);
impl_vuv!("lasx", lasx_xvaddi_wu, simd_add, m256i, u32x8, u32, 5);
impl_vuv!("lasx", lasx_xvaddi_du, simd_add, m256i, u64x4, u64, 5);
impl_vuv!("lasx", lasx_xvslti_bu, simd_lt, m256i, u8x32, u8, 5);
impl_vuv!("lasx", lasx_xvslti_hu, simd_lt, m256i, u16x16, u16, 5);
impl_vuv!("lasx", lasx_xvslti_wu, simd_lt, m256i, u32x8, u32, 5);
impl_vuv!("lasx", lasx_xvslti_du, simd_lt, m256i, u64x4, u64, 5);
impl_vuv!("lasx", lasx_xvslei_bu, simd_le, m256i, u8x32, u8, 5);
impl_vuv!("lasx", lasx_xvslei_hu, simd_le, m256i, u16x16, u16, 5);
impl_vuv!("lasx", lasx_xvslei_wu, simd_le, m256i, u32x8, u32, 5);
impl_vuv!("lasx", lasx_xvslei_du, simd_le, m256i, u64x4, u64, 5);
impl_vuv!("lasx", lasx_xvmaxi_bu, simd_imax, m256i, u8x32, u8, 5);
impl_vuv!("lasx", lasx_xvmaxi_hu, simd_imax, m256i, u16x16, u16, 5);
impl_vuv!("lasx", lasx_xvmaxi_wu, simd_imax, m256i, u32x8, u32, 5);
impl_vuv!("lasx", lasx_xvmaxi_du, simd_imax, m256i, u64x4, u64, 5);
impl_vuv!("lasx", lasx_xvmini_bu, simd_imin, m256i, u8x32, u8, 5);
impl_vuv!("lasx", lasx_xvmini_hu, simd_imin, m256i, u16x16, u16, 5);
impl_vuv!("lasx", lasx_xvmini_wu, simd_imin, m256i, u32x8, u32, 5);
impl_vuv!("lasx", lasx_xvmini_du, simd_imin, m256i, u64x4, u64, 5);

impl_vug!("lasx", lasx_xvpickve2gr_w, simd_extract, m256i, i32x8, i32, i32, 3);
impl_vug!("lasx", lasx_xvpickve2gr_d, simd_extract, m256i, i64x4, i64, i64, 2);
impl_vug!("lasx", lasx_xvpickve2gr_wu, simd_extract, m256i, u32x8, u32, u32, 3);
impl_vug!("lasx", lasx_xvpickve2gr_du, simd_extract, m256i, u64x4, u64, u64, 2);

impl_vsv!("lasx", lasx_xvseqi_b, simd_eq, m256i, i8x32, i8, 5);
impl_vsv!("lasx", lasx_xvseqi_h, simd_eq, m256i, i16x16, i16, 5);
impl_vsv!("lasx", lasx_xvseqi_w, simd_eq, m256i, i32x8, i32, 5);
impl_vsv!("lasx", lasx_xvseqi_d, simd_eq, m256i, i64x4, i64, 5);
impl_vsv!("lasx", lasx_xvslti_b, simd_lt, m256i, i8x32, i8, 5);
impl_vsv!("lasx", lasx_xvslti_h, simd_lt, m256i, i16x16, i16, 5);
impl_vsv!("lasx", lasx_xvslti_w, simd_lt, m256i, i32x8, i32, 5);
impl_vsv!("lasx", lasx_xvslti_d, simd_lt, m256i, i64x4, i64, 5);
impl_vsv!("lasx", lasx_xvslei_b, simd_le, m256i, i8x32, i8, 5);
impl_vsv!("lasx", lasx_xvslei_h, simd_le, m256i, i16x16, i16, 5);
impl_vsv!("lasx", lasx_xvslei_w, simd_le, m256i, i32x8, i32, 5);
impl_vsv!("lasx", lasx_xvslei_d, simd_le, m256i, i64x4, i64, 5);
impl_vsv!("lasx", lasx_xvmaxi_b, simd_imax, m256i, i8x32, i8, 5);
impl_vsv!("lasx", lasx_xvmaxi_h, simd_imax, m256i, i16x16, i16, 5);
impl_vsv!("lasx", lasx_xvmaxi_w, simd_imax, m256i, i32x8, i32, 5);
impl_vsv!("lasx", lasx_xvmaxi_d, simd_imax, m256i, i64x4, i64, 5);
impl_vsv!("lasx", lasx_xvmini_b, simd_imin, m256i, i8x32, i8, 5);
impl_vsv!("lasx", lasx_xvmini_h, simd_imin, m256i, i16x16, i16, 5);
impl_vsv!("lasx", lasx_xvmini_w, simd_imin, m256i, i32x8, i32, 5);
impl_vsv!("lasx", lasx_xvmini_d, simd_imin, m256i, i64x4, i64, 5);

impl_vvvv!("lasx", lasx_xvmadd_b, simd_madd, m256i, i8x32);
impl_vvvv!("lasx", lasx_xvmadd_h, simd_madd, m256i, i16x16);
impl_vvvv!("lasx", lasx_xvmadd_w, simd_madd, m256i, i32x8);
impl_vvvv!("lasx", lasx_xvmadd_d, simd_madd, m256i, i64x4);
impl_vvvv!("lasx", lasx_xvmsub_b, simd_msub, m256i, i8x32);
impl_vvvv!("lasx", lasx_xvmsub_h, simd_msub, m256i, i16x16);
impl_vvvv!("lasx", lasx_xvmsub_w, simd_msub, m256i, i32x8);
impl_vvvv!("lasx", lasx_xvmsub_d, simd_msub, m256i, i64x4);
impl_vvvv!("lasx", lasx_xvfmadd_s, simd_fma, m256, f32x8);
impl_vvvv!("lasx", lasx_xvfmadd_d, simd_fma, m256d, f64x4);
impl_vvvv!("lasx", lasx_xvfmsub_s, simd_fms, m256, f32x8);
impl_vvvv!("lasx", lasx_xvfmsub_d, simd_fms, m256d, f64x4);
impl_vvvv!("lasx", lasx_xvfnmadd_s, simd_nfma, m256, f32x8);
impl_vvvv!("lasx", lasx_xvfnmadd_d, simd_nfma, m256d, f64x4);
impl_vvvv!("lasx", lasx_xvfnmsub_s, simd_nfms, m256, f32x8);
impl_vvvv!("lasx", lasx_xvfnmsub_d, simd_nfms, m256d, f64x4);

impl_vugv!("lasx", lasx_xvinsgr2vr_w, simd_insert, m256i, i32x8, i32, i32, 3);
impl_vugv!("lasx", lasx_xvinsgr2vr_d, simd_insert, m256i, i64x4, i64, i64, 2);
Loading