|
| 1 | +//@ compile-flags: -C opt-level=0 |
| 2 | +//@ only-x86_64 |
| 3 | + |
| 4 | +#![feature(link_llvm_intrinsics, abi_unadjusted, repr_simd, simd_ffi, portable_simd, f16)] |
| 5 | +#![crate_type = "lib"] |
| 6 | + |
| 7 | +use std::simd::i64x2; |
| 8 | + |
| 9 | +#[repr(simd)] |
| 10 | +pub struct Tile([i8; 1024]); |
| 11 | + |
| 12 | +#[repr(C, packed)] |
| 13 | +pub struct Bar(u32, i64x2, i64x2, i64x2, i64x2, i64x2, i64x2); |
| 14 | +// CHECK: %Bar = type <{ i32, <2 x i64>, <2 x i64>, <2 x i64>, <2 x i64>, <2 x i64>, <2 x i64> }> |
| 15 | + |
| 16 | +#[repr(simd)] |
| 17 | +pub struct f16x8([f16; 8]); |
| 18 | + |
| 19 | +// CHECK-LABEL: @struct_with_i1_vector_autocast |
| 20 | +#[no_mangle] |
| 21 | +pub unsafe fn struct_with_i1_vector_autocast(a: i64x2, b: i64x2) -> (u8, u8) { |
| 22 | + extern "unadjusted" { |
| 23 | + #[link_name = "llvm.x86.avx512.vp2intersect.q.128"] |
| 24 | + fn foo(a: i64x2, b: i64x2) -> (u8, u8); |
| 25 | + } |
| 26 | + |
| 27 | + // CHECK: [[A:%[0-9]+]] = call { <2 x i1>, <2 x i1> } @llvm.x86.avx512.vp2intersect.q.128(<2 x i64> {{.*}}, <2 x i64> {{.*}}) |
| 28 | + // CHECK: [[B:%[0-9]+]] = extractvalue { <2 x i1>, <2 x i1> } [[A]], 0 |
| 29 | + // CHECK: [[C:%[0-9]+]] = shufflevector <2 x i1> [[B]], <2 x i1> zeroinitializer, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3> |
| 30 | + // CHECK: [[D:%[0-9]+]] = bitcast <8 x i1> [[C]] to i8 |
| 31 | + // CHECK: [[E:%[0-9]+]] = insertvalue { i8, i8 } poison, i8 [[D]], 0 |
| 32 | + // CHECK: [[F:%[0-9]+]] = extractvalue { <2 x i1>, <2 x i1> } [[A]], 1 |
| 33 | + // CHECK: [[G:%[0-9]+]] = shufflevector <2 x i1> [[F]], <2 x i1> zeroinitializer, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3> |
| 34 | + // CHECK: [[H:%[0-9]+]] = bitcast <8 x i1> [[G]] to i8 |
| 35 | + // CHECK: insertvalue { i8, i8 } [[E]], i8 [[H]], 1 |
| 36 | + foo(a, b) |
| 37 | +} |
| 38 | + |
| 39 | +// CHECK-LABEL: @struct_autocast |
| 40 | +#[no_mangle] |
| 41 | +pub unsafe fn struct_autocast(key_metadata: u32, key: i64x2) -> Bar { |
| 42 | + extern "unadjusted" { |
| 43 | + #[link_name = "llvm.x86.encodekey128"] |
| 44 | + fn foo(key_metadata: u32, key: i64x2) -> Bar; |
| 45 | + } |
| 46 | + |
| 47 | + // CHECK: [[A:%[0-9]+]] = call { i32, <2 x i64>, <2 x i64>, <2 x i64>, <2 x i64>, <2 x i64>, <2 x i64> } @llvm.x86.encodekey128(i32 {{.*}}, <2 x i64> {{.*}}) |
| 48 | + // CHECK: [[B:%[0-9]+]] = extractvalue { i32, <2 x i64>, <2 x i64>, <2 x i64>, <2 x i64>, <2 x i64>, <2 x i64> } [[A]], 0 |
| 49 | + // CHECK: [[C:%[0-9]+]] = insertvalue %Bar poison, i32 [[B]], 0 |
| 50 | + // CHECK: [[D:%[0-9]+]] = extractvalue { i32, <2 x i64>, <2 x i64>, <2 x i64>, <2 x i64>, <2 x i64>, <2 x i64> } [[A]], 1 |
| 51 | + // CHECK: [[E:%[0-9]+]] = insertvalue %Bar [[C]], <2 x i64> [[D]], 1 |
| 52 | + // CHECK: [[F:%[0-9]+]] = extractvalue { i32, <2 x i64>, <2 x i64>, <2 x i64>, <2 x i64>, <2 x i64>, <2 x i64> } [[A]], 2 |
| 53 | + // CHECK: [[G:%[0-9]+]] = insertvalue %Bar [[E]], <2 x i64> [[F]], 2 |
| 54 | + // CHECK: [[H:%[0-9]+]] = extractvalue { i32, <2 x i64>, <2 x i64>, <2 x i64>, <2 x i64>, <2 x i64>, <2 x i64> } [[A]], 3 |
| 55 | + // CHECK: [[I:%[0-9]+]] = insertvalue %Bar [[G]], <2 x i64> [[H]], 3 |
| 56 | + // CHECK: [[J:%[0-9]+]] = extractvalue { i32, <2 x i64>, <2 x i64>, <2 x i64>, <2 x i64>, <2 x i64>, <2 x i64> } [[A]], 4 |
| 57 | + // CHECK: [[K:%[0-9]+]] = insertvalue %Bar [[I]], <2 x i64> [[J]], 4 |
| 58 | + // CHECK: [[L:%[0-9]+]] = extractvalue { i32, <2 x i64>, <2 x i64>, <2 x i64>, <2 x i64>, <2 x i64>, <2 x i64> } [[A]], 5 |
| 59 | + // CHECK: [[M:%[0-9]+]] = insertvalue %Bar [[K]], <2 x i64> [[L]], 5 |
| 60 | + // CHECK: [[N:%[0-9]+]] = extractvalue { i32, <2 x i64>, <2 x i64>, <2 x i64>, <2 x i64>, <2 x i64>, <2 x i64> } [[A]], 6 |
| 61 | + // CHECK: insertvalue %Bar [[M]], <2 x i64> [[N]], 6 |
| 62 | + foo(key_metadata, key) |
| 63 | +} |
| 64 | + |
| 65 | +// CHECK-LABEL: @i1_vector_autocast |
| 66 | +#[no_mangle] |
| 67 | +pub unsafe fn i1_vector_autocast(a: f16x8) -> u8 { |
| 68 | + extern "unadjusted" { |
| 69 | + #[link_name = "llvm.x86.avx512fp16.fpclass.ph.128"] |
| 70 | + fn foo(a: f16x8, b: i32) -> u8; |
| 71 | + } |
| 72 | + |
| 73 | + // CHECK: [[A:%[0-9]+]] = call <8 x i1> @llvm.x86.avx512fp16.fpclass.ph.128(<8 x half> {{.*}}, i32 1) |
| 74 | + // CHECK: bitcast <8 x i1> [[A]] to i8 |
| 75 | + foo(a, 1) |
| 76 | +} |
| 77 | + |
| 78 | +// CHECK: declare { <2 x i1>, <2 x i1> } @llvm.x86.avx512.vp2intersect.q.128(<2 x i64>, <2 x i64>) |
| 79 | + |
| 80 | +// CHECK: declare { i32, <2 x i64>, <2 x i64>, <2 x i64>, <2 x i64>, <2 x i64>, <2 x i64> } @llvm.x86.encodekey128(i32, <2 x i64>) |
| 81 | + |
| 82 | +// CHECK: declare <8 x i1> @llvm.x86.avx512fp16.fpclass.ph.128(<8 x half>, i32 immarg) |
0 commit comments