File: packed-simd-alignment.rs

package info (click to toggle)
rustc 1.85.0%2Bdfsg3-1
  • links: PTS, VCS
  • area: main
  • in suites: experimental, sid, trixie
  • size: 893,396 kB
  • sloc: xml: 158,127; python: 35,830; javascript: 19,497; cpp: 19,002; sh: 17,245; ansic: 13,127; asm: 4,376; makefile: 1,051; perl: 29; lisp: 29; ruby: 19; sql: 11
file content (44 lines) | stat: -rw-r--r-- 1,263 bytes parent folder | download | duplicates (3)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
//@ compile-flags: -Cno-prepopulate-passes

#![crate_type = "lib"]
#![feature(repr_simd, core_intrinsics)]
// make sure that codegen emits correctly-aligned loads and stores for repr(packed, simd) types
// the alignment of a load should be no less than T, and no more than the size of the vector type
use std::intrinsics::simd as intrinsics;

#[derive(Copy, Clone)]
#[repr(packed, simd)]
struct f32x3([f32; 3]);

#[derive(Copy, Clone)]
#[repr(packed, simd)]
struct f32x4([f32; 4]);

// CHECK-LABEL: load_f32x3
#[no_mangle]
pub fn load_f32x3(floats: &f32x3) -> f32x3 {
    // FIXME: Is a memcpy really the best we can do?
    // CHECK: @llvm.memcpy.{{.*}}ptr align 4 {{.*}}ptr align 4
    *floats
}

// CHECK-LABEL: load_f32x4
#[no_mangle]
pub fn load_f32x4(floats: &f32x4) -> f32x4 {
    // CHECK: load <4 x float>, ptr %{{[a-z0-9_]*}}, align {{4|8|16}}
    *floats
}

// CHECK-LABEL: add_f32x3
#[no_mangle]
pub fn add_f32x3(x: f32x3, y: f32x3) -> f32x3 {
    // CHECK: load <3 x float>, ptr %{{[a-z0-9_]*}}, align 4
    unsafe { intrinsics::simd_add(x, y) }
}

// CHECK-LABEL: add_f32x4
#[no_mangle]
pub fn add_f32x4(x: f32x4, y: f32x4) -> f32x4 {
    // CHECK: load <4 x float>, ptr %{{[a-z0-9_]*}}, align {{4|8|16}}
    unsafe { intrinsics::simd_add(x, y) }
}