469 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc < %s -mtriple=i386-pc-win32 -mattr=+sse | FileCheck --check-prefix=WIN32 %s3; RUN: llc < %s -mtriple=x86_64-win32 -mattr=+sse | FileCheck --check-prefix=WIN64 %s4; RUN: llc < %s -mtriple=x86_64-linux-gnu -mattr=+sse | FileCheck --check-prefix=LINUXOSX %s5 6; Test regcall when receiving/returning i17define x86_regcallcc i1 @test_argReti1(i1 %a) {8; WIN32-LABEL: test_argReti1:9; WIN32: # %bb.0:10; WIN32-NEXT: xorb $1, %al11; WIN32-NEXT: # kill: def $al killed $al killed $eax12; WIN32-NEXT: retl13;14; WIN64-LABEL: test_argReti1:15; WIN64: # %bb.0:16; WIN64-NEXT: xorb $1, %al17; WIN64-NEXT: # kill: def $al killed $al killed $eax18; WIN64-NEXT: retq19;20; LINUXOSX-LABEL: test_argReti1:21; LINUXOSX: # %bb.0:22; LINUXOSX-NEXT: xorb $1, %al23; LINUXOSX-NEXT: # kill: def $al killed $al killed $eax24; LINUXOSX-NEXT: retq25 %add = add i1 %a, 126 ret i1 %add27}28 29; Test regcall when passing/retrieving i130define x86_regcallcc i1 @test_CallargReti1(i1 %a) {31; WIN32-LABEL: test_CallargReti1:32; WIN32: # %bb.0:33; WIN32-NEXT: xorb $1, %al34; WIN32-NEXT: movzbl %al, %eax35; WIN32-NEXT: calll _test_argReti136; WIN32-NEXT: xorb $1, %al37; WIN32-NEXT: retl38;39; WIN64-LABEL: test_CallargReti1:40; WIN64: # %bb.0:41; WIN64-NEXT: pushq %rax42; WIN64-NEXT: .seh_stackalloc 843; WIN64-NEXT: .seh_endprologue44; WIN64-NEXT: xorb $1, %al45; WIN64-NEXT: movzbl %al, %eax46; WIN64-NEXT: callq test_argReti147; WIN64-NEXT: xorb $1, %al48; WIN64-NEXT: .seh_startepilogue49; WIN64-NEXT: popq %rcx50; WIN64-NEXT: .seh_endepilogue51; WIN64-NEXT: retq52; WIN64-NEXT: .seh_endproc53;54; LINUXOSX-LABEL: test_CallargReti1:55; LINUXOSX: # %bb.0:56; LINUXOSX-NEXT: pushq %rax57; LINUXOSX-NEXT: .cfi_def_cfa_offset 1658; LINUXOSX-NEXT: xorb $1, %al59; LINUXOSX-NEXT: movzbl %al, %eax60; LINUXOSX-NEXT: callq *test_argReti1@GOTPCREL(%rip)61; LINUXOSX-NEXT: xorb $1, %al62; LINUXOSX-NEXT: popq %rcx63; LINUXOSX-NEXT: .cfi_def_cfa_offset 864; LINUXOSX-NEXT: retq65 %b = add i1 %a, 166 %c = call x86_regcallcc i1 @test_argReti1(i1 %b)67 %d = add i1 %c, 168 ret i1 %d69}70 71;test calling conventions - input parameters, callee saved xmms72define x86_regcallcc <16 x float> @testf32_inp(<16 x float> %a, <16 x float> %b, <16 x float> %c) nounwind {73; WIN32-LABEL: testf32_inp:74; WIN32: # %bb.0:75; WIN32-NEXT: pushl %ebp76; WIN32-NEXT: movl %esp, %ebp77; WIN32-NEXT: andl $-16, %esp78; WIN32-NEXT: subl $32, %esp79; WIN32-NEXT: movaps %xmm7, (%esp) # 16-byte Spill80; WIN32-NEXT: movaps %xmm6, %xmm781; WIN32-NEXT: movaps %xmm5, %xmm682; WIN32-NEXT: movaps %xmm4, %xmm583; WIN32-NEXT: movaps %xmm1, %xmm484; WIN32-NEXT: movaps %xmm0, %xmm185; WIN32-NEXT: addps %xmm5, %xmm086; WIN32-NEXT: mulps %xmm5, %xmm187; WIN32-NEXT: subps %xmm1, %xmm088; WIN32-NEXT: movups 8(%ebp), %xmm189; WIN32-NEXT: addps %xmm1, %xmm090; WIN32-NEXT: movaps %xmm4, %xmm191; WIN32-NEXT: addps %xmm6, %xmm192; WIN32-NEXT: mulps %xmm6, %xmm493; WIN32-NEXT: subps %xmm4, %xmm194; WIN32-NEXT: movups 24(%ebp), %xmm495; WIN32-NEXT: addps %xmm4, %xmm196; WIN32-NEXT: movaps %xmm2, %xmm497; WIN32-NEXT: addps %xmm7, %xmm498; WIN32-NEXT: mulps %xmm7, %xmm299; WIN32-NEXT: subps %xmm2, %xmm4100; WIN32-NEXT: movups 40(%ebp), %xmm2101; WIN32-NEXT: addps %xmm2, %xmm4102; WIN32-NEXT: movaps %xmm3, %xmm5103; WIN32-NEXT: movaps (%esp), %xmm2 # 16-byte Reload104; WIN32-NEXT: addps %xmm2, %xmm5105; WIN32-NEXT: mulps %xmm2, %xmm3106; WIN32-NEXT: subps %xmm3, %xmm5107; WIN32-NEXT: movups 56(%ebp), %xmm2108; WIN32-NEXT: addps %xmm2, %xmm5109; WIN32-NEXT: movaps %xmm4, %xmm2110; WIN32-NEXT: movaps %xmm5, %xmm3111; WIN32-NEXT: movl %ebp, %esp112; WIN32-NEXT: popl %ebp113; WIN32-NEXT: retl114;115; WIN64-LABEL: testf32_inp:116; WIN64: # %bb.0:117; WIN64-NEXT: subq $72, %rsp118; WIN64-NEXT: movaps %xmm15, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill119; WIN64-NEXT: movaps %xmm14, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill120; WIN64-NEXT: movaps %xmm13, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill121; WIN64-NEXT: movaps %xmm12, (%rsp) # 16-byte Spill122; WIN64-NEXT: movaps %xmm0, %xmm12123; WIN64-NEXT: addps %xmm4, %xmm12124; WIN64-NEXT: movaps %xmm1, %xmm13125; WIN64-NEXT: addps %xmm5, %xmm13126; WIN64-NEXT: movaps %xmm2, %xmm14127; WIN64-NEXT: addps %xmm6, %xmm14128; WIN64-NEXT: movaps %xmm3, %xmm15129; WIN64-NEXT: addps %xmm7, %xmm15130; WIN64-NEXT: mulps %xmm4, %xmm0131; WIN64-NEXT: subps %xmm0, %xmm12132; WIN64-NEXT: mulps %xmm5, %xmm1133; WIN64-NEXT: subps %xmm1, %xmm13134; WIN64-NEXT: mulps %xmm6, %xmm2135; WIN64-NEXT: subps %xmm2, %xmm14136; WIN64-NEXT: mulps %xmm7, %xmm3137; WIN64-NEXT: subps %xmm3, %xmm15138; WIN64-NEXT: addps %xmm8, %xmm12139; WIN64-NEXT: addps %xmm9, %xmm13140; WIN64-NEXT: addps %xmm10, %xmm14141; WIN64-NEXT: addps %xmm11, %xmm15142; WIN64-NEXT: movaps %xmm12, %xmm0143; WIN64-NEXT: movaps %xmm13, %xmm1144; WIN64-NEXT: movaps %xmm14, %xmm2145; WIN64-NEXT: movaps %xmm15, %xmm3146; WIN64-NEXT: movaps (%rsp), %xmm12 # 16-byte Reload147; WIN64-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm13 # 16-byte Reload148; WIN64-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm14 # 16-byte Reload149; WIN64-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm15 # 16-byte Reload150; WIN64-NEXT: addq $72, %rsp151; WIN64-NEXT: retq152;153; LINUXOSX-LABEL: testf32_inp:154; LINUXOSX: # %bb.0:155; LINUXOSX-NEXT: movaps %xmm15, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill156; LINUXOSX-NEXT: movaps %xmm14, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill157; LINUXOSX-NEXT: movaps %xmm13, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill158; LINUXOSX-NEXT: movaps %xmm12, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill159; LINUXOSX-NEXT: movaps %xmm0, %xmm12160; LINUXOSX-NEXT: addps %xmm4, %xmm12161; LINUXOSX-NEXT: movaps %xmm1, %xmm13162; LINUXOSX-NEXT: addps %xmm5, %xmm13163; LINUXOSX-NEXT: movaps %xmm2, %xmm14164; LINUXOSX-NEXT: addps %xmm6, %xmm14165; LINUXOSX-NEXT: movaps %xmm3, %xmm15166; LINUXOSX-NEXT: addps %xmm7, %xmm15167; LINUXOSX-NEXT: mulps %xmm4, %xmm0168; LINUXOSX-NEXT: subps %xmm0, %xmm12169; LINUXOSX-NEXT: mulps %xmm5, %xmm1170; LINUXOSX-NEXT: subps %xmm1, %xmm13171; LINUXOSX-NEXT: mulps %xmm6, %xmm2172; LINUXOSX-NEXT: subps %xmm2, %xmm14173; LINUXOSX-NEXT: mulps %xmm7, %xmm3174; LINUXOSX-NEXT: subps %xmm3, %xmm15175; LINUXOSX-NEXT: addps %xmm8, %xmm12176; LINUXOSX-NEXT: addps %xmm9, %xmm13177; LINUXOSX-NEXT: addps %xmm10, %xmm14178; LINUXOSX-NEXT: addps %xmm11, %xmm15179; LINUXOSX-NEXT: movaps %xmm12, %xmm0180; LINUXOSX-NEXT: movaps %xmm13, %xmm1181; LINUXOSX-NEXT: movaps %xmm14, %xmm2182; LINUXOSX-NEXT: movaps %xmm15, %xmm3183; LINUXOSX-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm12 # 16-byte Reload184; LINUXOSX-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm13 # 16-byte Reload185; LINUXOSX-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm14 # 16-byte Reload186; LINUXOSX-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm15 # 16-byte Reload187; LINUXOSX-NEXT: retq188 %x1 = fadd <16 x float> %a, %b189 %x2 = fmul <16 x float> %a, %b190 %x3 = fsub <16 x float> %x1, %x2191 %x4 = fadd <16 x float> %x3, %c192 ret <16 x float> %x4193}194 195;test calling conventions - input parameters, callee saved GPRs196define x86_regcallcc i32 @testi32_inp(i32 %a1, i32 %a2, i32 %a3, i32 %a4, i32 %a5, i32 %a6,197; WIN32-LABEL: testi32_inp:198; WIN32: # %bb.0:199; WIN32-NEXT: pushl %ebp200; WIN32-NEXT: pushl %ebx201; WIN32-NEXT: subl $12, %esp202; WIN32-NEXT: movl %esi, (%esp) # 4-byte Spill203; WIN32-NEXT: movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill204; WIN32-NEXT: movl %eax, %ebp205; WIN32-NEXT: leal (%edx,%edi), %eax206; WIN32-NEXT: movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill207; WIN32-NEXT: movl %edx, %eax208; WIN32-NEXT: subl %edi, %eax209; WIN32-NEXT: movl %ebp, %edx210; WIN32-NEXT: subl %ecx, %edx211; WIN32-NEXT: movl {{[0-9]+}}(%esp), %ebx212; WIN32-NEXT: subl {{[0-9]+}}(%esp), %ebx213; WIN32-NEXT: imull %edx, %ebx214; WIN32-NEXT: movl {{[0-9]+}}(%esp), %esi215; WIN32-NEXT: movl %esi, %edx216; WIN32-NEXT: subl {{[0-9]+}}(%esp), %edx217; WIN32-NEXT: imull %eax, %edx218; WIN32-NEXT: addl %ebx, %edx219; WIN32-NEXT: movl {{[0-9]+}}(%esp), %ebx220; WIN32-NEXT: movl (%esp), %edi # 4-byte Reload221; WIN32-NEXT: subl %ebx, %edi222; WIN32-NEXT: movl {{[0-9]+}}(%esp), %ecx223; WIN32-NEXT: movl %ecx, %eax224; WIN32-NEXT: subl {{[0-9]+}}(%esp), %eax225; WIN32-NEXT: imull %edi, %eax226; WIN32-NEXT: addl %edx, %eax227; WIN32-NEXT: addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload228; WIN32-NEXT: addl (%esp), %ebx # 4-byte Folded Reload229; WIN32-NEXT: movl {{[0-9]+}}(%esp), %edx230; WIN32-NEXT: addl {{[0-9]+}}(%esp), %edx231; WIN32-NEXT: imull %edx, %ebp232; WIN32-NEXT: addl {{[0-9]+}}(%esp), %esi233; WIN32-NEXT: imull {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload234; WIN32-NEXT: addl %esi, %ebp235; WIN32-NEXT: addl {{[0-9]+}}(%esp), %ecx236; WIN32-NEXT: imull %ebx, %ecx237; WIN32-NEXT: addl %ecx, %ebp238; WIN32-NEXT: addl %eax, %ebp239; WIN32-NEXT: movl %ebp, %eax240; WIN32-NEXT: addl $12, %esp241; WIN32-NEXT: popl %ebx242; WIN32-NEXT: popl %ebp243; WIN32-NEXT: retl244;245; WIN64-LABEL: testi32_inp:246; WIN64: # %bb.0:247; WIN64-NEXT: pushq %rbp248; WIN64-NEXT: pushq %rbx249; WIN64-NEXT: # kill: def $edx killed $edx def $rdx250; WIN64-NEXT: # kill: def $esi killed $esi def $rsi251; WIN64-NEXT: # kill: def $r15d killed $r15d def $r15252; WIN64-NEXT: # kill: def $r14d killed $r14d def $r14253; WIN64-NEXT: # kill: def $r12d killed $r12d def $r12254; WIN64-NEXT: # kill: def $r11d killed $r11d def $r11255; WIN64-NEXT: # kill: def $r10d killed $r10d def $r10256; WIN64-NEXT: # kill: def $r9d killed $r9d def $r9257; WIN64-NEXT: # kill: def $r8d killed $r8d def $r8258; WIN64-NEXT: # kill: def $edi killed $edi def $rdi259; WIN64-NEXT: leal (%rdx,%rdi), %ebx260; WIN64-NEXT: movl %edx, %ebp261; WIN64-NEXT: subl %edi, %ebp262; WIN64-NEXT: leal (%rsi,%r8), %edx263; WIN64-NEXT: # kill: def $esi killed $esi killed $rsi264; WIN64-NEXT: subl %r8d, %esi265; WIN64-NEXT: leal (%r9,%r10), %edi266; WIN64-NEXT: movl %r9d, %r8d267; WIN64-NEXT: subl %r10d, %r8d268; WIN64-NEXT: movl %eax, %r9d269; WIN64-NEXT: subl %ecx, %r9d270; WIN64-NEXT: imull %r9d, %r8d271; WIN64-NEXT: leal (%r11,%r12), %r9d272; WIN64-NEXT: movl %r11d, %r10d273; WIN64-NEXT: subl %r12d, %r10d274; WIN64-NEXT: imull %ebp, %r10d275; WIN64-NEXT: addl %r8d, %r10d276; WIN64-NEXT: leal (%r14,%r15), %r8d277; WIN64-NEXT: movl %r14d, %r11d278; WIN64-NEXT: subl %r15d, %r11d279; WIN64-NEXT: imull %esi, %r11d280; WIN64-NEXT: addl %r10d, %r11d281; WIN64-NEXT: addl %ecx, %eax282; WIN64-NEXT: imull %edi, %eax283; WIN64-NEXT: imull %ebx, %r9d284; WIN64-NEXT: addl %r9d, %eax285; WIN64-NEXT: imull %edx, %r8d286; WIN64-NEXT: addl %r8d, %eax287; WIN64-NEXT: addl %r11d, %eax288; WIN64-NEXT: popq %rbx289; WIN64-NEXT: popq %rbp290; WIN64-NEXT: retq291;292; LINUXOSX-LABEL: testi32_inp:293; LINUXOSX: # %bb.0:294; LINUXOSX-NEXT: # kill: def $edx killed $edx def $rdx295; LINUXOSX-NEXT: # kill: def $esi killed $esi def $rsi296; LINUXOSX-NEXT: # kill: def $r14d killed $r14d def $r14297; LINUXOSX-NEXT: # kill: def $r13d killed $r13d def $r13298; LINUXOSX-NEXT: # kill: def $r12d killed $r12d def $r12299; LINUXOSX-NEXT: # kill: def $r9d killed $r9d def $r9300; LINUXOSX-NEXT: # kill: def $r8d killed $r8d def $r8301; LINUXOSX-NEXT: # kill: def $edi killed $edi def $rdi302; LINUXOSX-NEXT: leal (%rdx,%rdi), %r10d303; LINUXOSX-NEXT: movl %edx, %r11d304; LINUXOSX-NEXT: subl %edi, %r11d305; LINUXOSX-NEXT: leal (%rsi,%r8), %edx306; LINUXOSX-NEXT: # kill: def $esi killed $esi killed $rsi307; LINUXOSX-NEXT: subl %r8d, %esi308; LINUXOSX-NEXT: leal (%r9,%r12), %edi309; LINUXOSX-NEXT: movl %r9d, %r8d310; LINUXOSX-NEXT: subl %r12d, %r8d311; LINUXOSX-NEXT: movl %eax, %r9d312; LINUXOSX-NEXT: subl %ecx, %r9d313; LINUXOSX-NEXT: imull %r9d, %r8d314; LINUXOSX-NEXT: leal (%r13,%r14), %r9d315; LINUXOSX-NEXT: movl %r13d, %r12d316; LINUXOSX-NEXT: subl %r14d, %r12d317; LINUXOSX-NEXT: imull %r11d, %r12d318; LINUXOSX-NEXT: movl {{[0-9]+}}(%rsp), %r11d319; LINUXOSX-NEXT: addl %r8d, %r12d320; LINUXOSX-NEXT: movl %r15d, %r8d321; LINUXOSX-NEXT: subl %r11d, %r8d322; LINUXOSX-NEXT: imull %esi, %r8d323; LINUXOSX-NEXT: addl %r12d, %r8d324; LINUXOSX-NEXT: addl %ecx, %eax325; LINUXOSX-NEXT: imull %edi, %eax326; LINUXOSX-NEXT: imull %r10d, %r9d327; LINUXOSX-NEXT: addl %r9d, %eax328; LINUXOSX-NEXT: addl %r15d, %r11d329; LINUXOSX-NEXT: imull %edx, %r11d330; LINUXOSX-NEXT: addl %r11d, %eax331; LINUXOSX-NEXT: addl %r8d, %eax332; LINUXOSX-NEXT: retq333 i32 %b1, i32 %b2, i32 %b3, i32 %b4, i32 %b5, i32 %b6) nounwind {334 %x1 = sub i32 %a1, %a2335 %x2 = sub i32 %a3, %a4336 %x3 = sub i32 %a5, %a6337 %y1 = sub i32 %b1, %b2338 %y2 = sub i32 %b3, %b4339 %y3 = sub i32 %b5, %b6340 %v1 = add i32 %a1, %a2341 %v2 = add i32 %a3, %a4342 %v3 = add i32 %a5, %a6343 %w1 = add i32 %b1, %b2344 %w2 = add i32 %b3, %b4345 %w3 = add i32 %b5, %b6346 %s1 = mul i32 %x1, %y1347 %s2 = mul i32 %x2, %y2348 %s3 = mul i32 %x3, %y3349 %t1 = mul i32 %v1, %w1350 %t2 = mul i32 %v2, %w2351 %t3 = mul i32 %v3, %w3352 %m1 = add i32 %s1, %s2353 %m2 = add i32 %m1, %s3354 %n1 = add i32 %t1, %t2355 %n2 = add i32 %n1, %t3356 %r1 = add i32 %m2, %n2357 ret i32 %r1358}359 360; Test that parameters, overflowing register capacity, are passed through the stack361define x86_regcallcc <32 x float> @testf32_stack(<32 x float> %a, <32 x float> %b, <32 x float> %c) nounwind {362; WIN32-LABEL: testf32_stack:363; WIN32: # %bb.0:364; WIN32-NEXT: pushl %ebp365; WIN32-NEXT: movl %esp, %ebp366; WIN32-NEXT: andl $-16, %esp367; WIN32-NEXT: subl $48, %esp368; WIN32-NEXT: movaps %xmm7, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill369; WIN32-NEXT: movaps %xmm6, (%esp) # 16-byte Spill370; WIN32-NEXT: movaps %xmm5, %xmm6371; WIN32-NEXT: movaps %xmm4, %xmm5372; WIN32-NEXT: movaps %xmm3, %xmm4373; WIN32-NEXT: movaps %xmm2, %xmm3374; WIN32-NEXT: movaps %xmm1, %xmm2375; WIN32-NEXT: movaps %xmm0, %xmm1376; WIN32-NEXT: movups 120(%ebp), %xmm7377; WIN32-NEXT: movaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm0 # 16-byte Reload378; WIN32-NEXT: addps %xmm7, %xmm0379; WIN32-NEXT: movups 248(%ebp), %xmm7380; WIN32-NEXT: addps %xmm7, %xmm0381; WIN32-NEXT: movaps %xmm0, {{[-0-9]+}}(%e{{[sb]}}p) # 16-byte Spill382; WIN32-NEXT: movups 104(%ebp), %xmm7383; WIN32-NEXT: movaps (%esp), %xmm0 # 16-byte Reload384; WIN32-NEXT: addps %xmm7, %xmm0385; WIN32-NEXT: movups 232(%ebp), %xmm7386; WIN32-NEXT: addps %xmm7, %xmm0387; WIN32-NEXT: movaps %xmm0, (%esp) # 16-byte Spill388; WIN32-NEXT: movups 88(%ebp), %xmm7389; WIN32-NEXT: addps %xmm7, %xmm6390; WIN32-NEXT: movups 216(%ebp), %xmm7391; WIN32-NEXT: addps %xmm7, %xmm6392; WIN32-NEXT: movups 72(%ebp), %xmm7393; WIN32-NEXT: addps %xmm7, %xmm5394; WIN32-NEXT: movups 200(%ebp), %xmm7395; WIN32-NEXT: addps %xmm7, %xmm5396; WIN32-NEXT: movups 56(%ebp), %xmm7397; WIN32-NEXT: addps %xmm7, %xmm4398; WIN32-NEXT: movups 184(%ebp), %xmm7399; WIN32-NEXT: addps %xmm7, %xmm4400; WIN32-NEXT: movups 40(%ebp), %xmm7401; WIN32-NEXT: addps %xmm7, %xmm3402; WIN32-NEXT: movups 168(%ebp), %xmm7403; WIN32-NEXT: addps %xmm7, %xmm3404; WIN32-NEXT: movups 24(%ebp), %xmm7405; WIN32-NEXT: addps %xmm7, %xmm2406; WIN32-NEXT: movups 152(%ebp), %xmm7407; WIN32-NEXT: addps %xmm7, %xmm2408; WIN32-NEXT: movups 8(%ebp), %xmm7409; WIN32-NEXT: addps %xmm7, %xmm1410; WIN32-NEXT: movups 136(%ebp), %xmm7411; WIN32-NEXT: addps %xmm7, %xmm1412; WIN32-NEXT: movaps %xmm1, %xmm0413; WIN32-NEXT: movaps %xmm2, %xmm1414; WIN32-NEXT: movaps %xmm3, %xmm2415; WIN32-NEXT: movaps %xmm4, %xmm3416; WIN32-NEXT: movaps %xmm5, %xmm4417; WIN32-NEXT: movaps %xmm6, %xmm5418; WIN32-NEXT: movaps (%esp), %xmm6 # 16-byte Reload419; WIN32-NEXT: movaps {{[-0-9]+}}(%e{{[sb]}}p), %xmm7 # 16-byte Reload420; WIN32-NEXT: movl %ebp, %esp421; WIN32-NEXT: popl %ebp422; WIN32-NEXT: retl423;424; WIN64-LABEL: testf32_stack:425; WIN64: # %bb.0:426; WIN64-NEXT: pushq %rax427; WIN64-NEXT: addps %xmm15, %xmm7428; WIN64-NEXT: addps %xmm14, %xmm6429; WIN64-NEXT: addps %xmm13, %xmm5430; WIN64-NEXT: addps %xmm12, %xmm4431; WIN64-NEXT: addps %xmm11, %xmm3432; WIN64-NEXT: addps %xmm10, %xmm2433; WIN64-NEXT: addps %xmm9, %xmm1434; WIN64-NEXT: addps %xmm8, %xmm0435; WIN64-NEXT: addps {{[0-9]+}}(%rsp), %xmm0436; WIN64-NEXT: addps {{[0-9]+}}(%rsp), %xmm1437; WIN64-NEXT: addps {{[0-9]+}}(%rsp), %xmm2438; WIN64-NEXT: addps {{[0-9]+}}(%rsp), %xmm3439; WIN64-NEXT: addps {{[0-9]+}}(%rsp), %xmm4440; WIN64-NEXT: addps {{[0-9]+}}(%rsp), %xmm5441; WIN64-NEXT: addps {{[0-9]+}}(%rsp), %xmm6442; WIN64-NEXT: addps {{[0-9]+}}(%rsp), %xmm7443; WIN64-NEXT: popq %rax444; WIN64-NEXT: retq445;446; LINUXOSX-LABEL: testf32_stack:447; LINUXOSX: # %bb.0:448; LINUXOSX-NEXT: addps %xmm15, %xmm7449; LINUXOSX-NEXT: addps %xmm14, %xmm6450; LINUXOSX-NEXT: addps %xmm13, %xmm5451; LINUXOSX-NEXT: addps %xmm12, %xmm4452; LINUXOSX-NEXT: addps %xmm11, %xmm3453; LINUXOSX-NEXT: addps %xmm10, %xmm2454; LINUXOSX-NEXT: addps %xmm9, %xmm1455; LINUXOSX-NEXT: addps %xmm8, %xmm0456; LINUXOSX-NEXT: addps {{[0-9]+}}(%rsp), %xmm0457; LINUXOSX-NEXT: addps {{[0-9]+}}(%rsp), %xmm1458; LINUXOSX-NEXT: addps {{[0-9]+}}(%rsp), %xmm2459; LINUXOSX-NEXT: addps {{[0-9]+}}(%rsp), %xmm3460; LINUXOSX-NEXT: addps {{[0-9]+}}(%rsp), %xmm4461; LINUXOSX-NEXT: addps {{[0-9]+}}(%rsp), %xmm5462; LINUXOSX-NEXT: addps {{[0-9]+}}(%rsp), %xmm6463; LINUXOSX-NEXT: addps {{[0-9]+}}(%rsp), %xmm7464; LINUXOSX-NEXT: retq465 %x1 = fadd <32 x float> %a, %b466 %x2 = fadd <32 x float> %x1, %c467 ret <32 x float> %x2468}469