86 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 42; RUN: llc -mtriple x86_64-pc-linux < %s | FileCheck %s3 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 %rbp8; CHECK-NEXT: pushq %rbx9; CHECK-NEXT: subq $88, %rsp10; CHECK-NEXT: movq %rdi, %rbx11; CHECK-NEXT: movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill12; CHECK-NEXT: psrld $16, %xmm013; CHECK-NEXT: movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill14; CHECK-NEXT: pinsrw $0, 2(%rdi), %xmm115; CHECK-NEXT: pinsrw $0, (%rdi), %xmm016; CHECK-NEXT: .p2align 417; CHECK-NEXT: .LBB0_1: # %atomicrmw.start18; CHECK-NEXT: # =>This Inner Loop Header: Depth=119; CHECK-NEXT: movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill20; CHECK-NEXT: movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill21; CHECK-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload22; CHECK-NEXT: callq __extendhfsf2@PLT23; CHECK-NEXT: movss %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill24; CHECK-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload25; CHECK-NEXT: callq __extendhfsf2@PLT26; CHECK-NEXT: addss {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload27; CHECK-NEXT: callq __truncsfhf2@PLT28; CHECK-NEXT: pextrw $0, %xmm0, %eax29; CHECK-NEXT: movzwl %ax, %ebp30; CHECK-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload31; CHECK-NEXT: callq __extendhfsf2@PLT32; CHECK-NEXT: movss %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill33; CHECK-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload34; CHECK-NEXT: callq __extendhfsf2@PLT35; CHECK-NEXT: addss {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Folded Reload36; CHECK-NEXT: callq __truncsfhf2@PLT37; CHECK-NEXT: pextrw $0, %xmm0, %ecx38; CHECK-NEXT: shll $16, %ecx39; CHECK-NEXT: orl %ebp, %ecx40; CHECK-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload41; CHECK-NEXT: pextrw $0, %xmm0, %edx42; CHECK-NEXT: shll $16, %edx43; CHECK-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload44; CHECK-NEXT: pextrw $0, %xmm0, %eax45; CHECK-NEXT: movzwl %ax, %eax46; CHECK-NEXT: orl %edx, %eax47; CHECK-NEXT: lock cmpxchgl %ecx, (%rbx)48; CHECK-NEXT: setne %cl49; CHECK-NEXT: movl %eax, %edx50; CHECK-NEXT: shrl $16, %eax51; CHECK-NEXT: pinsrw $0, %edx, %xmm052; CHECK-NEXT: pinsrw $0, %eax, %xmm153; CHECK-NEXT: testb %cl, %cl54; CHECK-NEXT: jne .LBB0_155; CHECK-NEXT: # %bb.2: # %atomicrmw.end56; CHECK-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]57; CHECK-NEXT: addq $88, %rsp58; CHECK-NEXT: popq %rbx59; CHECK-NEXT: popq %rbp60; CHECK-NEXT: retq61 %res = atomicrmw fadd ptr addrspace(1) %ptr, <2 x half> %value seq_cst, align 462 ret <2 x half> %res63}64 65define <2 x float> @test_atomicrmw_fadd_v2f32_align8(ptr addrspace(1) %ptr, <2 x float> %value) #0 {66; CHECK-LABEL: test_atomicrmw_fadd_v2f32_align8:67; CHECK: # %bb.0:68; CHECK-NEXT: movq {{.*#+}} xmm1 = mem[0],zero69; CHECK-NEXT: .p2align 470; CHECK-NEXT: .LBB1_1: # %atomicrmw.start71; CHECK-NEXT: # =>This Inner Loop Header: Depth=172; CHECK-NEXT: movq %xmm1, %rax73; CHECK-NEXT: addps %xmm0, %xmm174; CHECK-NEXT: movq %xmm1, %rcx75; CHECK-NEXT: lock cmpxchgq %rcx, (%rdi)76; CHECK-NEXT: movq %rax, %xmm177; CHECK-NEXT: jne .LBB1_178; CHECK-NEXT: # %bb.2: # %atomicrmw.end79; CHECK-NEXT: movdqa %xmm1, %xmm080; CHECK-NEXT: retq81 %res = atomicrmw fadd ptr addrspace(1) %ptr, <2 x float> %value seq_cst, align 882 ret <2 x float> %res83}84 85attributes #0 = { nounwind }86