1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4 2; RUN: llc -mtriple x86_64-pc-linux < %s | FileCheck %s 3 4define <2 x half> @test_atomicrmw_fadd_v2f16_align4(ptr addrspace(1) %ptr, <2 x half> %value) #0 { 5; CHECK-LABEL: test_atomicrmw_fadd_v2f16_align4: 6; CHECK: # %bb.0: 7; CHECK-NEXT: pushq %rbp 8; CHECK-NEXT: pushq %rbx 9; CHECK-NEXT: subq $88, %rsp 10; CHECK-NEXT: movq %rdi, %rbx 11; CHECK-NEXT: movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill 12; CHECK-NEXT: psrld $16, %xmm0 13; CHECK-NEXT: movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill 14; CHECK-NEXT: pinsrw $0, 2(%rdi), %xmm1 15; CHECK-NEXT: pinsrw $0, (%rdi), %xmm0 16; CHECK-NEXT: .p2align 4 17; CHECK-NEXT: .LBB0_1: # %atomicrmw.start 18; CHECK-NEXT: # =>This Inner Loop Header: Depth=1 19; CHECK-NEXT: movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill 20; CHECK-NEXT: movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill 21; CHECK-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload 22; CHECK-NEXT: callq __extendhfsf2@PLT 23; CHECK-NEXT: movss %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill 24; CHECK-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload 25; CHECK-NEXT: callq __extendhfsf2@PLT 26; CHECK-NEXT: addss {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload 27; CHECK-NEXT: callq __truncsfhf2@PLT 28; CHECK-NEXT: pextrw $0, %xmm0, %eax 29; CHECK-NEXT: movzwl %ax, %ebp 30; CHECK-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload 31; CHECK-NEXT: callq __extendhfsf2@PLT 32; CHECK-NEXT: movss %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill 33; CHECK-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload 34; CHECK-NEXT: callq __extendhfsf2@PLT 35; CHECK-NEXT: addss {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload 36; CHECK-NEXT: callq __truncsfhf2@PLT 37; CHECK-NEXT: pextrw $0, %xmm0, %ecx 38; CHECK-NEXT: shll $16, %ecx 39; CHECK-NEXT: orl %ebp, %ecx 40; CHECK-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload 41; CHECK-NEXT: pextrw $0, %xmm0, %edx 42; CHECK-NEXT: shll $16, %edx 43; CHECK-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload 44; CHECK-NEXT: pextrw $0, %xmm0, %eax 45; CHECK-NEXT: movzwl %ax, %eax 46; CHECK-NEXT: orl %edx, %eax 47; CHECK-NEXT: lock cmpxchgl %ecx, (%rbx) 48; CHECK-NEXT: setne %cl 49; CHECK-NEXT: pinsrw $0, %eax, %xmm0 50; CHECK-NEXT: shrl $16, %eax 51; CHECK-NEXT: pinsrw $0, %eax, %xmm1 52; CHECK-NEXT: testb %cl, %cl 53; CHECK-NEXT: jne .LBB0_1 54; CHECK-NEXT: # %bb.2: # %atomicrmw.end 55; CHECK-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3] 56; CHECK-NEXT: addq $88, %rsp 57; CHECK-NEXT: popq %rbx 58; CHECK-NEXT: popq %rbp 59; CHECK-NEXT: retq 60 %res = atomicrmw fadd ptr addrspace(1) %ptr, <2 x half> %value seq_cst, align 4 61 ret <2 x half> %res 62} 63 64define <2 x float> @test_atomicrmw_fadd_v2f32_align8(ptr addrspace(1) %ptr, <2 x float> %value) #0 { 65; CHECK-LABEL: test_atomicrmw_fadd_v2f32_align8: 66; CHECK: # %bb.0: 67; CHECK-NEXT: movq {{.*#+}} xmm1 = mem[0],zero 68; CHECK-NEXT: .p2align 4 69; CHECK-NEXT: .LBB1_1: # %atomicrmw.start 70; CHECK-NEXT: # =>This Inner Loop Header: Depth=1 71; CHECK-NEXT: movq %xmm1, %rax 72; CHECK-NEXT: addps %xmm0, %xmm1 73; CHECK-NEXT: movq %xmm1, %rcx 74; CHECK-NEXT: lock cmpxchgq %rcx, (%rdi) 75; CHECK-NEXT: movq %rax, %xmm1 76; CHECK-NEXT: jne .LBB1_1 77; CHECK-NEXT: # %bb.2: # %atomicrmw.end 78; CHECK-NEXT: movdqa %xmm1, %xmm0 79; CHECK-NEXT: retq 80 %res = atomicrmw fadd ptr addrspace(1) %ptr, <2 x float> %value seq_cst, align 8 81 ret <2 x float> %res 82} 83 84attributes #0 = { nounwind } 85