Skip to content

Commit

Permalink
Unrolled build for rust-lang#121953
Browse files Browse the repository at this point in the history 
Rollup merge of rust-lang#121953 - jhorstmann:assembly-tests-for-masked-simd-instructions, r=workingjubilee

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 <iX x N> vector masks when converting to <i1 xN>. 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.

The exception is simd_bitmask, which already uses the most-significant bit.

This additional instruction would be removed by the changes in rust-lang#104693, which makes all mask operations consistently use the most significant bits.

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.
  • Loading branch information
@rust-timer
rust-timer authored Mar 12, 2024
2 parents 7de1a1f + e91f937 commit 3ec1402
Show file tree
Hide file tree
Showing 7 changed files with 597 additions and 0 deletions.
149 changes: 149 additions & 0 deletions tests/assembly/simd-bitmask.rs
View file
Original file line number Diff line number Diff line change
@@ -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<V, B>(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)
}
44 changes: 44 additions & 0 deletions tests/assembly/simd-intrinsic-gather.rs
View file
Original file line number Diff line number Diff line change
@@ -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<V, M, P>(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)
}
88 changes: 88 additions & 0 deletions tests/assembly/simd-intrinsic-mask-load.rs
View file
Original file line number Diff line number Diff line change
@@ -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<M, P, T>(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]))
}
60 changes: 60 additions & 0 deletions tests/assembly/simd-intrinsic-mask-reduce.rs
View file
Original file line number Diff line number Diff line change
@@ -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<T>(x: T) -> bool;
fn simd_reduce_any<T>(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)
}
Loading

0 comments on commit 3ec1402

Please sign in to comment.