1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4 2; RUN: llc < %s -mtriple=x86_64-linux-gnu -mattr=avx512bf16 | FileCheck %s 3 4define void @add(ptr %pa, ptr %pb, ptr %pc) nounwind { 5; CHECK-LABEL: add: 6; CHECK: # %bb.0: 7; CHECK-NEXT: pushq %rbx 8; CHECK-NEXT: movq %rdx, %rbx 9; CHECK-NEXT: movzwl (%rsi), %eax 10; CHECK-NEXT: shll $16, %eax 11; CHECK-NEXT: vmovd %eax, %xmm0 12; CHECK-NEXT: movzwl (%rdi), %eax 13; CHECK-NEXT: shll $16, %eax 14; CHECK-NEXT: vmovd %eax, %xmm1 15; CHECK-NEXT: vaddss %xmm0, %xmm1, %xmm0 16; CHECK-NEXT: callq __truncsfbf2@PLT 17; CHECK-NEXT: vpextrw $0, %xmm0, (%rbx) 18; CHECK-NEXT: popq %rbx 19; CHECK-NEXT: retq 20 %a = load bfloat, ptr %pa 21 %b = load bfloat, ptr %pb 22 %add = fadd bfloat %a, %b 23 store bfloat %add, ptr %pc 24 ret void 25} 26 27define <4 x bfloat> @fptrunc_v4f32(<4 x float> %a) nounwind { 28; CHECK-LABEL: fptrunc_v4f32: 29; CHECK: # %bb.0: 30; CHECK-NEXT: pushq %rbp 31; CHECK-NEXT: pushq %r14 32; CHECK-NEXT: pushq %rbx 33; CHECK-NEXT: subq $64, %rsp 34; CHECK-NEXT: vmovaps %xmm0, (%rsp) # 16-byte Spill 35; CHECK-NEXT: callq __truncsfbf2@PLT 36; CHECK-NEXT: vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill 37; CHECK-NEXT: vmovshdup (%rsp), %xmm0 # 16-byte Folded Reload 38; CHECK-NEXT: # xmm0 = mem[1,1,3,3] 39; CHECK-NEXT: callq __truncsfbf2@PLT 40; CHECK-NEXT: vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill 41; CHECK-NEXT: vpshufd $255, (%rsp), %xmm0 # 16-byte Folded Reload 42; CHECK-NEXT: # xmm0 = mem[3,3,3,3] 43; CHECK-NEXT: callq __truncsfbf2@PLT 44; CHECK-NEXT: vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill 45; CHECK-NEXT: callq __truncsfbf2@PLT 46; CHECK-NEXT: vpextrw $0, %xmm0, %ebx 47; CHECK-NEXT: vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload 48; CHECK-NEXT: vpextrw $0, %xmm0, %ebp 49; CHECK-NEXT: vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload 50; CHECK-NEXT: vpextrw $0, %xmm0, %r14d 51; CHECK-NEXT: vpermilpd $1, (%rsp), %xmm0 # 16-byte Folded Reload 52; CHECK-NEXT: # xmm0 = mem[1,0] 53; CHECK-NEXT: callq __truncsfbf2@PLT 54; CHECK-NEXT: vpextrw $0, %xmm0, %eax 55; CHECK-NEXT: vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload 56; CHECK-NEXT: vpinsrw $1, %r14d, %xmm0, %xmm0 57; CHECK-NEXT: vpinsrw $2, %eax, %xmm0, %xmm0 58; CHECK-NEXT: vpinsrw $3, %ebp, %xmm0, %xmm0 59; CHECK-NEXT: vpinsrw $4, %ebx, %xmm0, %xmm0 60; CHECK-NEXT: vpinsrw $5, %ebx, %xmm0, %xmm0 61; CHECK-NEXT: vpinsrw $6, %ebx, %xmm0, %xmm0 62; CHECK-NEXT: vpinsrw $7, %ebx, %xmm0, %xmm0 63; CHECK-NEXT: addq $64, %rsp 64; CHECK-NEXT: popq %rbx 65; CHECK-NEXT: popq %r14 66; CHECK-NEXT: popq %rbp 67; CHECK-NEXT: retq 68 %b = fptrunc <4 x float> %a to <4 x bfloat> 69 ret <4 x bfloat> %b 70} 71