From e91f93777997ee647768afa60d565a0a62aefa49 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?J=C3=B6rn=20Horstmann?= Date: Sat, 2 Mar 2024 10:59:11 +0100 Subject: [PATCH] Add tests for the generated assembly of mask related simd instructions. The tests show that the code generation currently uses the least significant bits of vector masks when converting to . This leads to an additional left shift operation in the assembly for x86, since mask operations on x86 operate based on the most significant bit. On aarch64 the left shift is followed by a comparison against zero, which repeats the sign bit across the whole lane. The exception, which does not introduce an unneeded shift, is simd_bitmask, because the code generation already shifts before truncating. By using the "C" calling convention the tests should be stable regarding changes in register allocation, but it is possible that future llvm updates will require updating some of the checks. This additional instruction would be removed by the fix in #104693, which uses the most significant bit for all mask operations. --- tests/assembly/simd-bitmask.rs | 149 +++++++++++++++++++ tests/assembly/simd-intrinsic-gather.rs | 44 ++++++ tests/assembly/simd-intrinsic-mask-load.rs | 88 +++++++++++ tests/assembly/simd-intrinsic-mask-reduce.rs | 60 ++++++++ tests/assembly/simd-intrinsic-mask-store.rs | 86 +++++++++++ tests/assembly/simd-intrinsic-scatter.rs | 40 +++++ tests/assembly/simd-intrinsic-select.rs | 130 ++++++++++++++++ 7 files changed, 597 insertions(+) create mode 100644 tests/assembly/simd-bitmask.rs create mode 100644 tests/assembly/simd-intrinsic-gather.rs create mode 100644 tests/assembly/simd-intrinsic-mask-load.rs create mode 100644 tests/assembly/simd-intrinsic-mask-reduce.rs create mode 100644 tests/assembly/simd-intrinsic-mask-store.rs create mode 100644 tests/assembly/simd-intrinsic-scatter.rs create mode 100644 tests/assembly/simd-intrinsic-select.rs diff --git a/tests/assembly/simd-bitmask.rs b/tests/assembly/simd-bitmask.rs new file mode 100644 index 00000000000..8264a706852 --- /dev/null +++ b/tests/assembly/simd-bitmask.rs @@ -0,0 +1,149 @@ +//@ revisions: x86 x86-avx2 x86-avx512 aarch64 +//@ [x86] compile-flags: --target=x86_64-unknown-linux-gnu -C llvm-args=-x86-asm-syntax=intel +//@ [x86] needs-llvm-components: x86 +//@ [x86-avx2] compile-flags: --target=x86_64-unknown-linux-gnu -C llvm-args=-x86-asm-syntax=intel +//@ [x86-avx2] compile-flags: -C target-feature=+avx2 +//@ [x86-avx2] needs-llvm-components: x86 +//@ [x86-avx512] compile-flags: --target=x86_64-unknown-linux-gnu -C llvm-args=-x86-asm-syntax=intel +//@ [x86-avx512] compile-flags: -C target-feature=+avx512f,+avx512vl,+avx512bw,+avx512dq +//@ [x86-avx512] needs-llvm-components: x86 +//@ [aarch64] compile-flags: --target=aarch64-unknown-linux-gnu +//@ [aarch64] needs-llvm-components: aarch64 +//@ [aarch64] min-llvm-version: 18.0 +//@ assembly-output: emit-asm +//@ compile-flags: --crate-type=lib -O + +#![feature(no_core, lang_items, repr_simd, intrinsics)] +#![no_core] +#![allow(non_camel_case_types)] + +// Because we don't have core yet. +#[lang = "sized"] +pub trait Sized {} + +#[lang = "copy"] +trait Copy {} + +#[repr(simd)] +pub struct m8x16([i8; 16]); + +#[repr(simd)] +pub struct m8x64([i8; 64]); + +#[repr(simd)] +pub struct m32x4([i32; 4]); + +#[repr(simd)] +pub struct m64x2([i64; 2]); + +#[repr(simd)] +pub struct m64x4([i64; 4]); + +extern "rust-intrinsic" { + fn simd_bitmask(mask: V) -> B; +} + +// CHECK-LABEL: bitmask_m8x16 +#[no_mangle] +pub unsafe extern "C" fn bitmask_m8x16(mask: m8x16) -> u16 { + // The simd_bitmask intrinsic already uses the most significant bit, so no shift is necessary. + // Note that x86 has no byte shift, llvm uses a word shift to move the least significant bit + // of each byte into the right position. + // + // x86-NOT: psllw + // x86: movmskb eax, xmm0 + // + // x86-avx2-NOT: vpsllw + // x86-avx2: vpmovmskb eax, xmm0 + // + // x86-avx512-NOT: vpsllw xmm0 + // x86-avx512: vpmovmskb eax, xmm0 + // + // aarch64: adrp + // aarch64-NEXT: cmlt + // aarch64-NEXT: ldr + // aarch64-NEXT: and + // aarch64-NEXT: ext + // aarch64-NEXT: zip1 + // aarch64-NEXT: addv + // aarch64-NEXT: fmov + simd_bitmask(mask) +} + +// CHECK-LABEL: bitmask_m8x64 +#[no_mangle] +pub unsafe extern "C" fn bitmask_m8x64(mask: m8x64) -> u64 { + // The simd_bitmask intrinsic already uses the most significant bit, so no shift is necessary. + // Note that x86 has no byte shift, llvm uses a word shift to move the least significant bit + // of each byte into the right position. + // + // The parameter is a 512 bit vector which in the C abi is only valid for avx512 targets. + // + // x86-avx512-NOT: vpsllw + // x86-avx512: vpmovb2m k0, zmm0 + // x86-avx512: kmovq rax, k0 + simd_bitmask(mask) +} + +// CHECK-LABEL: bitmask_m32x4 +#[no_mangle] +pub unsafe extern "C" fn bitmask_m32x4(mask: m32x4) -> u8 { + // The simd_bitmask intrinsic already uses the most significant bit, so no shift is necessary. + // + // x86-NOT: psllq + // x86: movmskps eax, xmm0 + // + // x86-avx2-NOT: vpsllq + // x86-avx2: vmovmskps eax, xmm0 + // + // x86-avx512-NOT: vpsllq + // x86-avx512: vmovmskps eax, xmm0 + // + // aarch64: adrp + // aarch64-NEXT: cmlt + // aarch64-NEXT: ldr + // aarch64-NEXT: and + // aarch64-NEXT: addv + // aarch64-NEXT: fmov + // aarch64-NEXT: and + simd_bitmask(mask) +} + +// CHECK-LABEL: bitmask_m64x2 +#[no_mangle] +pub unsafe extern "C" fn bitmask_m64x2(mask: m64x2) -> u8 { + // The simd_bitmask intrinsic already uses the most significant bit, so no shift is necessary. + // + // x86-NOT: psllq + // x86: movmskpd eax, xmm0 + // + // x86-avx2-NOT: vpsllq + // x86-avx2: vmovmskpd eax, xmm0 + // + // x86-avx512-NOT: vpsllq + // x86-avx512: vmovmskpd eax, xmm0 + // + // aarch64: adrp + // aarch64-NEXT: cmlt + // aarch64-NEXT: ldr + // aarch64-NEXT: and + // aarch64-NEXT: addp + // aarch64-NEXT: fmov + // aarch64-NEXT: and + simd_bitmask(mask) +} + +// CHECK-LABEL: bitmask_m64x4 +#[no_mangle] +pub unsafe extern "C" fn bitmask_m64x4(mask: m64x4) -> u8 { + // The simd_bitmask intrinsic already uses the most significant bit, so no shift is necessary. + // + // The parameter is a 256 bit vector which in the C abi is only valid for avx/avx512 targets. + // + // x86-avx2-NOT: vpsllq + // x86-avx2: vmovmskpd eax, ymm0 + // + // x86-avx512-NOT: vpsllq + // x86-avx512: vmovmskpd eax, ymm0 + simd_bitmask(mask) +} diff --git a/tests/assembly/simd-intrinsic-gather.rs b/tests/assembly/simd-intrinsic-gather.rs new file mode 100644 index 00000000000..ef6b597c25f --- /dev/null +++ b/tests/assembly/simd-intrinsic-gather.rs @@ -0,0 +1,44 @@ +//@ revisions: x86-avx512 +//@ [x86-avx512] compile-flags: --target=x86_64-unknown-linux-gnu -C llvm-args=-x86-asm-syntax=intel +//@ [x86-avx512] compile-flags: -C target-feature=+avx512f,+avx512vl,+avx512bw,+avx512dq +//@ [x86-avx512] needs-llvm-components: x86 +//@ [x86-avx512] min-llvm-version: 18.0 +//@ assembly-output: emit-asm +//@ compile-flags: --crate-type=lib -O + +#![feature(no_core, lang_items, repr_simd, intrinsics)] +#![no_core] +#![allow(non_camel_case_types)] + +// Because we don't have core yet. +#[lang = "sized"] +pub trait Sized {} + +#[lang = "copy"] +trait Copy {} + +#[repr(simd)] +pub struct f64x4([f64; 4]); + +#[repr(simd)] +pub struct m64x4([i64; 4]); + +#[repr(simd)] +pub struct pf64x4([*const f64; 4]); + +extern "rust-intrinsic" { + fn simd_gather(values: V, mask: M, pointer: P) -> V; +} + +// CHECK-LABEL: gather_f64x4 +#[no_mangle] +pub unsafe extern "C" fn gather_f64x4(mask: m64x4, ptrs: pf64x4) -> f64x4 { + // FIXME: This should also get checked to generate a gather instruction for avx2. + // Currently llvm scalarizes this code, see https://github.com/llvm/llvm-project/issues/59789 + // + // x86-avx512: vpsllq ymm0, ymm0, 63 + // x86-avx512-NEXT: vpmovq2m k1, ymm0 + // x86-avx512-NEXT: vpxor xmm0, xmm0, xmm0 + // x86-avx512-NEXT: vgatherqpd ymm0 {k1}, ymmword ptr [1*ymm1] + simd_gather(f64x4([0_f64, 0_f64, 0_f64, 0_f64]), ptrs, mask) +} diff --git a/tests/assembly/simd-intrinsic-mask-load.rs b/tests/assembly/simd-intrinsic-mask-load.rs new file mode 100644 index 00000000000..49d231c45f8 --- /dev/null +++ b/tests/assembly/simd-intrinsic-mask-load.rs @@ -0,0 +1,88 @@ +//@ revisions: x86-avx2 x86-avx512 +//@ [x86-avx2] compile-flags: --target=x86_64-unknown-linux-gnu -C llvm-args=-x86-asm-syntax=intel +//@ [x86-avx2] compile-flags: -C target-feature=+avx2 +//@ [x86-avx2] needs-llvm-components: x86 +//@ [x86-avx512] compile-flags: --target=x86_64-unknown-linux-gnu -C llvm-args=-x86-asm-syntax=intel +//@ [x86-avx512] compile-flags: -C target-feature=+avx512f,+avx512vl,+avx512bw,+avx512dq +//@ [x86-avx512] needs-llvm-components: x86 +//@ assembly-output: emit-asm +//@ compile-flags: --crate-type=lib -O + +#![feature(no_core, lang_items, repr_simd, intrinsics)] +#![no_core] +#![allow(non_camel_case_types)] + +// Because we don't have core yet. +#[lang = "sized"] +pub trait Sized {} + +#[lang = "copy"] +trait Copy {} + +#[repr(simd)] +pub struct i8x16([i8; 16]); + +#[repr(simd)] +pub struct m8x16([i8; 16]); + +#[repr(simd)] +pub struct f32x8([f32; 8]); + +#[repr(simd)] +pub struct m32x8([i32; 8]); + +#[repr(simd)] +pub struct f64x4([f64; 4]); + +#[repr(simd)] +pub struct m64x4([i64; 4]); + +extern "rust-intrinsic" { + fn simd_masked_load(mask: M, pointer: P, values: T) -> T; +} + +// CHECK-LABEL: load_i8x16 +#[no_mangle] +pub unsafe extern "C" fn load_i8x16(mask: m8x16, pointer: *const i8) -> i8x16 { + // Since avx2 supports no masked loads for bytes, the code tests each individual bit + // and jumps to code that inserts individual bytes. + // x86-avx2: vpsllw xmm0, xmm0, 7 + // x86-avx2-NEXT: vpmovmskb eax, xmm0 + // x86-avx2-NEXT: vpxor xmm0, xmm0 + // x86-avx2-NEXT: test al, 1 + // x86-avx2-NEXT: jne + // x86-avx2-NEXT: test al, 2 + // x86-avx2-NEXT: jne + // x86-avx2-DAG: movzx [[REG:[a-z]+]], byte ptr [rdi] + // x86-avx2-NEXT: vmovd xmm0, [[REG]] + // x86-avx2-DAG: vpinsrb xmm0, xmm0, byte ptr [rdi + 1], 1 + // + // x86-avx512: vpsllw xmm0, xmm0, 7 + // x86-avx512-NEXT: vpmovb2m k1, xmm0 + // x86-avx512-NEXT: vmovdqu8 xmm0 {k1} {z}, xmmword ptr [rdi] + simd_masked_load(mask, pointer, i8x16([0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0])) +} + +// CHECK-LABEL: load_f32x8 +#[no_mangle] +pub unsafe extern "C" fn load_f32x8(mask: m32x8, pointer: *const f32) -> f32x8 { + // x86-avx2: vpslld ymm0, ymm0, 31 + // x86-avx2-NEXT: vmaskmovps ymm0, ymm0, ymmword ptr [rdi] + // + // x86-avx512: vpslld ymm0, ymm0, 31 + // x86-avx512-NEXT: vpmovd2m k1, ymm0 + // x86-avx512-NEXT: vmovups ymm0 {k1} {z}, ymmword ptr [rdi] + simd_masked_load(mask, pointer, f32x8([0_f32, 0_f32, 0_f32, 0_f32, 0_f32, 0_f32, 0_f32, 0_f32])) +} + +// CHECK-LABEL: load_f64x4 +#[no_mangle] +pub unsafe extern "C" fn load_f64x4(mask: m64x4, pointer: *const f64) -> f64x4 { + // x86-avx2: vpsllq ymm0, ymm0, 63 + // x86-avx2-NEXT: vmaskmovpd ymm0, ymm0, ymmword ptr [rdi] + // + // x86-avx512: vpsllq ymm0, ymm0, 63 + // x86-avx512-NEXT: vpmovq2m k1, ymm0 + // x86-avx512-NEXT: vmovupd ymm0 {k1} {z}, ymmword ptr [rdi] + simd_masked_load(mask, pointer, f64x4([0_f64, 0_f64, 0_f64, 0_f64])) +} diff --git a/tests/assembly/simd-intrinsic-mask-reduce.rs b/tests/assembly/simd-intrinsic-mask-reduce.rs new file mode 100644 index 00000000000..763401755fa --- /dev/null +++ b/tests/assembly/simd-intrinsic-mask-reduce.rs @@ -0,0 +1,60 @@ +// verify that simd mask reductions do not introduce additional bit shift operations +//@ revisions: x86 aarch64 +//@ [x86] compile-flags: --target=x86_64-unknown-linux-gnu -C llvm-args=-x86-asm-syntax=intel +//@ [x86] needs-llvm-components: x86 +//@ [aarch64] compile-flags: --target=aarch64-unknown-linux-gnu +//@ [aarch64] needs-llvm-components: aarch64 +//@ [aarch64] min-llvm-version: 18.0 +//@ assembly-output: emit-asm +//@ compile-flags: --crate-type=lib -O + +#![feature(no_core, lang_items, repr_simd, intrinsics)] +#![no_core] +#![allow(non_camel_case_types)] + +// Because we don't have core yet. +#[lang = "sized"] +pub trait Sized {} + +#[lang = "copy"] +trait Copy {} + +#[repr(simd)] +pub struct mask8x16([i8; 16]); + +extern "rust-intrinsic" { + fn simd_reduce_all(x: T) -> bool; + fn simd_reduce_any(x: T) -> bool; +} + +// CHECK-LABEL: mask_reduce_all: +#[no_mangle] +pub unsafe extern "C" fn mask_reduce_all(m: mask8x16) -> bool { + // x86: psllw xmm0, 7 + // x86-NEXT: pmovmskb eax, xmm0 + // x86-NEXT: {{cmp ax, -1|xor eax, 65535}} + // x86-NEXT: sete al + // + // aarch64: shl v0.16b, v0.16b, #7 + // aarch64-NEXT: cmlt v0.16b, v0.16b, #0 + // aarch64-NEXT: uminv b0, v0.16b + // aarch64-NEXT: fmov [[REG:[a-z0-9]+]], s0 + // aarch64-NEXT: and w0, [[REG]], #0x1 + simd_reduce_all(m) +} + +// CHECK-LABEL: mask_reduce_any: +#[no_mangle] +pub unsafe extern "C" fn mask_reduce_any(m: mask8x16) -> bool { + // x86: psllw xmm0, 7 + // x86-NEXT: pmovmskb + // x86-NEXT: test eax, eax + // x86-NEXT: setne al + // + // aarch64: shl v0.16b, v0.16b, #7 + // aarch64-NEXT: cmlt v0.16b, v0.16b, #0 + // aarch64-NEXT: umaxv b0, v0.16b + // aarch64-NEXT: fmov [[REG:[a-z0-9]+]], s0 + // aarch64-NEXT: and w0, [[REG]], #0x1 + simd_reduce_any(m) +} diff --git a/tests/assembly/simd-intrinsic-mask-store.rs b/tests/assembly/simd-intrinsic-mask-store.rs new file mode 100644 index 00000000000..a6611e1c23d --- /dev/null +++ b/tests/assembly/simd-intrinsic-mask-store.rs @@ -0,0 +1,86 @@ +//@ revisions: x86-avx2 x86-avx512 +//@ [x86-avx2] compile-flags: --target=x86_64-unknown-linux-gnu -C llvm-args=-x86-asm-syntax=intel +//@ [x86-avx2] compile-flags: -C target-feature=+avx2 +//@ [x86-avx2] needs-llvm-components: x86 +//@ [x86-avx512] compile-flags: --target=x86_64-unknown-linux-gnu -C llvm-args=-x86-asm-syntax=intel +//@ [x86-avx512] compile-flags: -C target-feature=+avx512f,+avx512vl,+avx512bw,+avx512dq +//@ [x86-avx512] needs-llvm-components: x86 +//@ assembly-output: emit-asm +//@ compile-flags: --crate-type=lib -O + +#![feature(no_core, lang_items, repr_simd, intrinsics)] +#![no_core] +#![allow(non_camel_case_types)] + +// Because we don't have core yet. +#[lang = "sized"] +pub trait Sized {} + +#[lang = "copy"] +trait Copy {} + +#[repr(simd)] +pub struct i8x16([i8; 16]); + +#[repr(simd)] +pub struct m8x16([i8; 16]); + +#[repr(simd)] +pub struct f32x8([f32; 8]); + +#[repr(simd)] +pub struct m32x8([i32; 8]); + +#[repr(simd)] +pub struct f64x4([f64; 4]); + +#[repr(simd)] +pub struct m64x4([i64; 4]); + +extern "rust-intrinsic" { + fn simd_masked_store(mask: M, pointer: P, values: T); +} + +// CHECK-LABEL: store_i8x16 +#[no_mangle] +pub unsafe extern "C" fn store_i8x16(mask: m8x16, pointer: *mut i8, value: i8x16) { + // Since avx2 supports no masked stores for bytes, the code tests each individual bit + // and jumps to code that extracts individual bytes to memory. + // x86-avx2: vpsllw xmm0, xmm0, 7 + // x86-avx2-NEXT: vpmovmskb eax, xmm0 + // x86-avx2-NEXT: test al, 1 + // x86-avx2-NEXT: jne + // x86-avx2-NEXT: test al, 2 + // x86-avx2-NEXT: jne + // x86-avx2-DAG: vpextrb byte ptr [rdi + 1], xmm1, 1 + // x86-avx2-DAG: vpextrb byte ptr [rdi], xmm1, 0 + // + // x86-avx512: vpsllw xmm0, xmm0, 7 + // x86-avx512-NEXT: vpmovb2m k1, xmm0 + // x86-avx512-NEXT: vmovdqu8 xmmword ptr [rdi] {k1}, xmm1 + simd_masked_store(mask, pointer, value) +} + +// CHECK-LABEL: store_f32x8 +#[no_mangle] +pub unsafe extern "C" fn store_f32x8(mask: m32x8, pointer: *mut f32, value: f32x8) { + // x86-avx2: vpslld ymm0, ymm0, 31 + // x86-avx2-NEXT: vmaskmovps ymmword ptr [rdi], ymm0, ymm1 + // + // x86-avx512: vpslld ymm0, ymm0, 31 + // x86-avx512-NEXT: vpmovd2m k1, ymm0 + // x86-avx512-NEXT: vmovups ymmword ptr [rdi] {k1}, ymm1 + simd_masked_store(mask, pointer, value) +} + +// CHECK-LABEL: store_f64x4 +#[no_mangle] +pub unsafe extern "C" fn store_f64x4(mask: m64x4, pointer: *mut f64, value: f64x4) { + // x86-avx2: vpsllq ymm0, ymm0, 63 + // x86-avx2-NEXT: vmaskmovpd ymmword ptr [rdi], ymm0, ymm1 + // + // x86-avx512: vpsllq ymm0, ymm0, 63 + // x86-avx512-NEXT: vpmovq2m k1, ymm0 + // x86-avx512-NEXT: vmovupd ymmword ptr [rdi] {k1}, ymm1 + simd_masked_store(mask, pointer, value) +} diff --git a/tests/assembly/simd-intrinsic-scatter.rs b/tests/assembly/simd-intrinsic-scatter.rs new file mode 100644 index 00000000000..6ffefb0801a --- /dev/null +++ b/tests/assembly/simd-intrinsic-scatter.rs @@ -0,0 +1,40 @@ +//@ revisions: x86-avx512 +//@ [x86-avx512] compile-flags: --target=x86_64-unknown-linux-gnu -C llvm-args=-x86-asm-syntax=intel +//@ [x86-avx512] compile-flags: -C target-feature=+avx512f,+avx512vl,+avx512bw,+avx512dq +//@ [x86-avx512] needs-llvm-components: x86 +//@ [x86-avx512] min-llvm-version: 18.0 +//@ assembly-output: emit-asm +//@ compile-flags: --crate-type=lib -O + +#![feature(no_core, lang_items, repr_simd, intrinsics)] +#![no_core] +#![allow(non_camel_case_types)] + +// Because we don't have core yet. +#[lang = "sized"] +pub trait Sized {} + +#[lang = "copy"] +trait Copy {} + +#[repr(simd)] +pub struct f64x4([f64; 4]); + +#[repr(simd)] +pub struct m64x4([i64; 4]); + +#[repr(simd)] +pub struct pf64x4([*mut f64; 4]); + +extern "rust-intrinsic" { + fn simd_scatter(values: V, pointer: P, mask: M); +} + +// CHECK-LABEL: scatter_f64x4 +#[no_mangle] +pub unsafe extern "C" fn scatter_f64x4(values: f64x4, ptrs: pf64x4, mask: m64x4) { + // x86-avx512: vpsllq ymm2, ymm2, 63 + // x86-avx512-NEXT: vpmovq2m k1, ymm2 + // x86-avx512-NEXT: vscatterqpd ymmword ptr [1*ymm1] {k1}, ymm0 + simd_scatter(values, ptrs, mask) +} diff --git a/tests/assembly/simd-intrinsic-select.rs b/tests/assembly/simd-intrinsic-select.rs new file mode 100644 index 00000000000..3f36402e3d0 --- /dev/null +++ b/tests/assembly/simd-intrinsic-select.rs @@ -0,0 +1,130 @@ +//@ revisions: x86-avx2 x86-avx512 aarch64 +//@ [x86-avx2] compile-flags: --target=x86_64-unknown-linux-gnu -C llvm-args=-x86-asm-syntax=intel +//@ [x86-avx2] compile-flags: -C target-feature=+avx2 +//@ [x86-avx2] needs-llvm-components: x86 +//@ [x86-avx512] compile-flags: --target=x86_64-unknown-linux-gnu -C llvm-args=-x86-asm-syntax=intel +//@ [x86-avx512] compile-flags: -C target-feature=+avx512f,+avx512vl,+avx512bw,+avx512dq +//@ [x86-avx512] needs-llvm-components: x86 +//@ [aarch64] compile-flags: --target=aarch64-unknown-linux-gnu +//@ [aarch64] needs-llvm-components: aarch64 +//@ [aarch64] min-llvm-version: 18.0 +//@ assembly-output: emit-asm +//@ compile-flags: --crate-type=lib -O + +#![feature(no_core, lang_items, repr_simd, intrinsics)] +#![no_core] +#![allow(non_camel_case_types)] + +// Because we don't have core yet. +#[lang = "sized"] +pub trait Sized {} + +#[lang = "copy"] +trait Copy {} + +#[repr(simd)] +pub struct i8x16([i8; 16]); + +#[repr(simd)] +pub struct m8x16([i8; 16]); + +#[repr(simd)] +pub struct f32x4([f32; 4]); + +#[repr(simd)] +pub struct m32x4([i32; 4]); + +#[repr(simd)] +pub struct f64x2([f64; 2]); + +#[repr(simd)] +pub struct m64x2([i64; 2]); + +#[repr(simd)] +pub struct f64x4([f64; 4]); + +#[repr(simd)] +pub struct m64x4([i64; 4]); + +#[repr(simd)] +pub struct f64x8([f64; 8]); + +#[repr(simd)] +pub struct m64x8([i64; 8]); + +extern "rust-intrinsic" { + fn simd_select(mask: M, a: V, b: V) -> V; +} + +// CHECK-LABEL: select_i8x16 +#[no_mangle] +pub unsafe extern "C" fn select_i8x16(mask: m8x16, a: i8x16, b: i8x16) -> i8x16 { + // x86-avx2: vpsllw xmm0, xmm0, 7 + // x86-avx2-NEXT: vpblendvb xmm0, xmm2, xmm1, xmm0 + // + // x86-avx512: vpsllw xmm0, xmm0, 7 + // x86-avx512-NEXT: vpmovb2m k1, xmm0 + // x86-avx512-NEXT: vpblendmb xmm0 {k1}, xmm2, xmm1 + // + // aarch64: shl v0.16b, v0.16b, #7 + // aarch64-NEXT: cmlt v0.16b, v0.16b, #0 + // aarch64-NEXT: bsl v0.16b, v1.16b, v2.16b + simd_select(mask, a, b) +} + +// CHECK-LABEL: select_f32x4 +#[no_mangle] +pub unsafe extern "C" fn select_f32x4(mask: m32x4, a: f32x4, b: f32x4) -> f32x4 { + // x86-avx2: vpslld xmm0, xmm0, 31 + // x86-avx2-NEXT: vblendvps xmm0, xmm2, xmm1, xmm0 + // + // x86-avx512: vpslld xmm0, xmm0, 31 + // x86-avx512-NEXT: vpmovd2m k1, xmm0 + // x86-avx512-NEXT: vblendmps xmm0 {k1}, xmm2, xmm1 + // + // aarch64: shl v0.4s, v0.4s, #31 + // aarch64-NEXT: cmlt v0.4s, v0.4s, #0 + // aarch64-NEXT: bsl v0.16b, v1.16b, v2.16b + simd_select(mask, a, b) +} + +// CHECK-LABEL: select_f64x2 +#[no_mangle] +pub unsafe extern "C" fn select_f64x2(mask: m64x2, a: f64x2, b: f64x2) -> f64x2 { + // x86-avx2: vpsllq xmm0, xmm0, 63 + // x86-avx2-NEXT: vblendvpd xmm0, xmm2, xmm1, xmm0 + // + // x86-avx512: vpsllq xmm0, xmm0, 63 + // x86-avx512-NEXT: vpmovq2m k1, xmm0 + // x86-avx512-NEXT: vblendmpd xmm0 {k1}, xmm2, xmm1 + // + // aarch64: shl v0.2d, v0.2d, #63 + // aarch64-NEXT: cmlt v0.2d, v0.2d, #0 + // aarch64-NEXT: bsl v0.16b, v1.16b, v2.16b + simd_select(mask, a, b) +} + +// CHECK-LABEL: select_f64x4 +#[no_mangle] +pub unsafe extern "C" fn select_f64x4(mask: m64x4, a: f64x4, b: f64x4) -> f64x4 { + // The parameter is a 256 bit vector which in the C abi is only valid for avx targets. + // + // x86-avx2: vpsllq ymm0, ymm0, 63 + // x86-avx2-NEXT: vblendvpd ymm0, ymm2, ymm1, ymm0 + // + // x86-avx512: vpsllq ymm0, ymm0, 63 + // x86-avx512-NEXT: vpmovq2m k1, ymm0 + // x86-avx512-NEXT: vblendmpd ymm0 {k1}, ymm2, ymm1 + simd_select(mask, a, b) +} + +// CHECK-LABEL: select_f64x8 +#[no_mangle] +pub unsafe extern "C" fn select_f64x8(mask: m64x8, a: f64x8, b: f64x8) -> f64x8 { + // The parameter is a 256 bit vector which in the C abi is only valid for avx512 targets. + // + // x86-avx512: vpsllq zmm0, zmm0, 63 + // x86-avx512-NEXT: vpmovq2m k1, zmm0 + // x86-avx512-NEXT: vblendmpd zmm0 {k1}, zmm2, zmm1 + simd_select(mask, a, b) +}