1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc < %s -mtriple=x86_64-unknown | FileCheck %s --check-prefixes=CHECK,SSE
3 ; RUN: llc < %s -mtriple=x86_64-unknown -mattr=avx512f | FileCheck %s --check-prefixes=CHECK,AVX512
5 ; This test makes sure that a vector that needs to be promoted that is bitcasted to fp16 is legalized correctly without causing a width mismatch.
6 define void @constant_fold_vector_to_half() {
7 ; CHECK-LABEL: constant_fold_vector_to_half:
9 ; CHECK-NEXT: movw $16384, (%rax) # imm = 0x4000
11 store volatile half bitcast (<4 x i4> <i4 0, i4 0, i4 0, i4 4> to half), half* undef
15 ; Similarly this makes sure that the opposite bitcast of the above is also legalized without crashing.
16 define void @pr38533_2(half %x) {
17 ; SSE-LABEL: pr38533_2:
19 ; SSE-NEXT: pushq %rax
20 ; SSE-NEXT: .cfi_def_cfa_offset 16
21 ; SSE-NEXT: callq __gnu_f2h_ieee
22 ; SSE-NEXT: movw %ax, (%rax)
24 ; SSE-NEXT: .cfi_def_cfa_offset 8
27 ; AVX512-LABEL: pr38533_2:
29 ; AVX512-NEXT: vcvtps2ph $4, %xmm0, %xmm0
30 ; AVX512-NEXT: vmovd %xmm0, %eax
31 ; AVX512-NEXT: movw %ax, (%rax)
33 %a = bitcast half %x to <4 x i4>
34 store volatile <4 x i4> %a, <4 x i4>* undef
38 ; This case is a bitcast from fp16 to a 16-bit wide legal vector type. In this case the result type is legal when the bitcast gets type legalized.
39 define void @pr38533_3(half %x) {
40 ; SSE-LABEL: pr38533_3:
42 ; SSE-NEXT: pushq %rax
43 ; SSE-NEXT: .cfi_def_cfa_offset 16
44 ; SSE-NEXT: callq __gnu_f2h_ieee
45 ; SSE-NEXT: movw %ax, (%rax)
47 ; SSE-NEXT: .cfi_def_cfa_offset 8
50 ; AVX512-LABEL: pr38533_3:
52 ; AVX512-NEXT: vcvtps2ph $4, %xmm0, %xmm0
53 ; AVX512-NEXT: vmovd %xmm0, %eax
54 ; AVX512-NEXT: movw %ax, (%rax)
56 %a = bitcast half %x to <16 x i1>
57 store volatile <16 x i1> %a, <16 x i1>* undef