71 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 42; RUN: llc < %s -mtriple=x86_64-linux-gnu -mattr=avx512bf16 | FileCheck %s3 4define void @add(ptr %pa, ptr %pb, ptr %pc) nounwind {5; CHECK-LABEL: add:6; CHECK: # %bb.0:7; CHECK-NEXT: pushq %rbx8; CHECK-NEXT: movq %rdx, %rbx9; CHECK-NEXT: movzwl (%rsi), %eax10; CHECK-NEXT: shll $16, %eax11; CHECK-NEXT: vmovd %eax, %xmm012; CHECK-NEXT: movzwl (%rdi), %eax13; CHECK-NEXT: shll $16, %eax14; CHECK-NEXT: vmovd %eax, %xmm115; CHECK-NEXT: vaddss %xmm0, %xmm1, %xmm016; CHECK-NEXT: callq __truncsfbf2@PLT17; CHECK-NEXT: vpextrw $0, %xmm0, (%rbx)18; CHECK-NEXT: popq %rbx19; CHECK-NEXT: retq20 %a = load bfloat, ptr %pa21 %b = load bfloat, ptr %pb22 %add = fadd bfloat %a, %b23 store bfloat %add, ptr %pc24 ret void25}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 %rbp31; CHECK-NEXT: pushq %r1432; CHECK-NEXT: pushq %rbx33; CHECK-NEXT: subq $64, %rsp34; CHECK-NEXT: vmovaps %xmm0, (%rsp) # 16-byte Spill35; CHECK-NEXT: callq __truncsfbf2@PLT36; CHECK-NEXT: vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill37; CHECK-NEXT: vmovshdup (%rsp), %xmm0 # 16-byte Folded Reload38; CHECK-NEXT: # xmm0 = mem[1,1,3,3]39; CHECK-NEXT: callq __truncsfbf2@PLT40; CHECK-NEXT: vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill41; CHECK-NEXT: vpshufd $255, (%rsp), %xmm0 # 16-byte Folded Reload42; CHECK-NEXT: # xmm0 = mem[3,3,3,3]43; CHECK-NEXT: callq __truncsfbf2@PLT44; CHECK-NEXT: vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill45; CHECK-NEXT: callq __truncsfbf2@PLT46; CHECK-NEXT: vpextrw $0, %xmm0, %ebx47; CHECK-NEXT: vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload48; CHECK-NEXT: vpextrw $0, %xmm0, %ebp49; CHECK-NEXT: vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload50; CHECK-NEXT: vpextrw $0, %xmm0, %r14d51; CHECK-NEXT: vpermilpd $1, (%rsp), %xmm0 # 16-byte Folded Reload52; CHECK-NEXT: # xmm0 = mem[1,0]53; CHECK-NEXT: callq __truncsfbf2@PLT54; CHECK-NEXT: vpextrw $0, %xmm0, %eax55; CHECK-NEXT: vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload56; CHECK-NEXT: vpinsrw $1, %r14d, %xmm0, %xmm057; CHECK-NEXT: vpinsrw $2, %eax, %xmm0, %xmm058; CHECK-NEXT: vpinsrw $3, %ebp, %xmm0, %xmm059; CHECK-NEXT: vpinsrw $4, %ebx, %xmm0, %xmm060; CHECK-NEXT: vpinsrw $5, %ebx, %xmm0, %xmm061; CHECK-NEXT: vpinsrw $6, %ebx, %xmm0, %xmm062; CHECK-NEXT: vpinsrw $7, %ebx, %xmm0, %xmm063; CHECK-NEXT: addq $64, %rsp64; CHECK-NEXT: popq %rbx65; CHECK-NEXT: popq %r1466; CHECK-NEXT: popq %rbp67; CHECK-NEXT: retq68 %b = fptrunc <4 x float> %a to <4 x bfloat>69 ret <4 x bfloat> %b70}71