1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2; RUN: llc -mtriple=i686-linux-gnu -mattr=-ermsb < %s -o - | FileCheck %s --check-prefix=NOFAST32 3; RUN: llc -mtriple=i686-linux-gnu -mattr=+ermsb < %s -o - | FileCheck %s --check-prefix=FAST32 4; RUN: llc -mtriple=x86_64-linux-gnu -mattr=-ermsb < %s -o - | FileCheck %s --check-prefix=NOFAST 5; RUN: llc -mtriple=x86_64-linux-gnu -mattr=+ermsb < %s -o - | FileCheck %s --check-prefix=FAST 6; RUN: llc -mtriple=x86_64-linux-gnu -mcpu=generic < %s -o - | FileCheck %s --check-prefix=NOFAST 7; RUN: llc -mtriple=x86_64-linux-gnu -mcpu=haswell < %s -o - | FileCheck %s --check-prefix=FAST 8; RUN: llc -mtriple=x86_64-linux-gnu -mcpu=skylake < %s -o - | FileCheck %s --check-prefix=FAST 9; FIXME: The documentation states that ivybridge has ermsb, but this is not 10; enabled right now since I could not confirm by testing. 11; RUN: llc -mtriple=x86_64-linux-gnu -mcpu=ivybridge < %s -o - | FileCheck %s --check-prefix=NOFAST 12 13%struct.large = type { [4096 x i8] } 14 15declare void @foo(%struct.large* align 8 byval(%struct.large)) nounwind 16 17define void @test1(%struct.large* nocapture %x) nounwind { 18; NOFAST32-LABEL: test1: 19; NOFAST32: # %bb.0: 20; NOFAST32-NEXT: pushl %edi 21; NOFAST32-NEXT: pushl %esi 22; NOFAST32-NEXT: subl $4100, %esp # imm = 0x1004 23; NOFAST32-NEXT: movl {{[0-9]+}}(%esp), %esi 24; NOFAST32-NEXT: movl $1024, %ecx # imm = 0x400 25; NOFAST32-NEXT: movl %esp, %edi 26; NOFAST32-NEXT: rep;movsl (%esi), %es:(%edi) 27; NOFAST32-NEXT: calll foo 28; NOFAST32-NEXT: addl $4100, %esp # imm = 0x1004 29; NOFAST32-NEXT: popl %esi 30; NOFAST32-NEXT: popl %edi 31; NOFAST32-NEXT: retl 32; 33; FAST32-LABEL: test1: 34; FAST32: # %bb.0: 35; FAST32-NEXT: pushl %edi 36; FAST32-NEXT: pushl %esi 37; FAST32-NEXT: subl $4100, %esp # imm = 0x1004 38; FAST32-NEXT: movl {{[0-9]+}}(%esp), %esi 39; FAST32-NEXT: movl $4096, %ecx # imm = 0x1000 40; FAST32-NEXT: movl %esp, %edi 41; FAST32-NEXT: rep;movsb (%esi), %es:(%edi) 42; FAST32-NEXT: calll foo 43; FAST32-NEXT: addl $4100, %esp # imm = 0x1004 44; FAST32-NEXT: popl %esi 45; FAST32-NEXT: popl %edi 46; FAST32-NEXT: retl 47; 48; NOFAST-LABEL: test1: 49; NOFAST: # %bb.0: 50; NOFAST-NEXT: subq $4104, %rsp # imm = 0x1008 51; NOFAST-NEXT: movq %rdi, %rsi 52; NOFAST-NEXT: movl $512, %ecx # imm = 0x200 53; NOFAST-NEXT: movq %rsp, %rdi 54; NOFAST-NEXT: rep;movsq (%rsi), %es:(%rdi) 55; NOFAST-NEXT: callq foo 56; NOFAST-NEXT: addq $4104, %rsp # imm = 0x1008 57; NOFAST-NEXT: retq 58; 59; FAST-LABEL: test1: 60; FAST: # %bb.0: 61; FAST-NEXT: subq $4104, %rsp # imm = 0x1008 62; FAST-NEXT: movq %rdi, %rsi 63; FAST-NEXT: movl $4096, %ecx # imm = 0x1000 64; FAST-NEXT: movq %rsp, %rdi 65; FAST-NEXT: rep;movsb (%rsi), %es:(%rdi) 66; FAST-NEXT: callq foo 67; FAST-NEXT: addq $4104, %rsp # imm = 0x1008 68; FAST-NEXT: retq 69 call void @foo(%struct.large* align 8 byval(%struct.large) %x) 70 ret void 71 72} 73 74define void @test2(%struct.large* nocapture %x) nounwind minsize { 75; NOFAST32-LABEL: test2: 76; NOFAST32: # %bb.0: 77; NOFAST32-NEXT: pushl %edi 78; NOFAST32-NEXT: pushl %esi 79; NOFAST32-NEXT: subl $4100, %esp # imm = 0x1004 80; NOFAST32-NEXT: movl {{[0-9]+}}(%esp), %esi 81; NOFAST32-NEXT: movl $1024, %ecx # imm = 0x400 82; NOFAST32-NEXT: movl %esp, %edi 83; NOFAST32-NEXT: rep;movsl (%esi), %es:(%edi) 84; NOFAST32-NEXT: calll foo 85; NOFAST32-NEXT: addl $4100, %esp # imm = 0x1004 86; NOFAST32-NEXT: popl %esi 87; NOFAST32-NEXT: popl %edi 88; NOFAST32-NEXT: retl 89; 90; FAST32-LABEL: test2: 91; FAST32: # %bb.0: 92; FAST32-NEXT: pushl %edi 93; FAST32-NEXT: pushl %esi 94; FAST32-NEXT: subl $4100, %esp # imm = 0x1004 95; FAST32-NEXT: movl {{[0-9]+}}(%esp), %esi 96; FAST32-NEXT: movl $4096, %ecx # imm = 0x1000 97; FAST32-NEXT: movl %esp, %edi 98; FAST32-NEXT: rep;movsb (%esi), %es:(%edi) 99; FAST32-NEXT: calll foo 100; FAST32-NEXT: addl $4100, %esp # imm = 0x1004 101; FAST32-NEXT: popl %esi 102; FAST32-NEXT: popl %edi 103; FAST32-NEXT: retl 104; 105; NOFAST-LABEL: test2: 106; NOFAST: # %bb.0: 107; NOFAST-NEXT: subq $4104, %rsp # imm = 0x1008 108; NOFAST-NEXT: movq %rdi, %rsi 109; NOFAST-NEXT: movl $512, %ecx # imm = 0x200 110; NOFAST-NEXT: movq %rsp, %rdi 111; NOFAST-NEXT: rep;movsq (%rsi), %es:(%rdi) 112; NOFAST-NEXT: callq foo 113; NOFAST-NEXT: addq $4104, %rsp # imm = 0x1008 114; NOFAST-NEXT: retq 115; 116; FAST-LABEL: test2: 117; FAST: # %bb.0: 118; FAST-NEXT: subq $4104, %rsp # imm = 0x1008 119; FAST-NEXT: movq %rdi, %rsi 120; FAST-NEXT: movl $4096, %ecx # imm = 0x1000 121; FAST-NEXT: movq %rsp, %rdi 122; FAST-NEXT: rep;movsb (%rsi), %es:(%rdi) 123; FAST-NEXT: callq foo 124; FAST-NEXT: addq $4104, %rsp # imm = 0x1008 125; FAST-NEXT: retq 126 call void @foo(%struct.large* align 8 byval(%struct.large) %x) 127 ret void 128 129} 130 131%struct.large_oddsize = type { [4095 x i8] } 132 133declare void @foo_oddsize(%struct.large_oddsize* align 8 byval(%struct.large_oddsize)) nounwind 134 135define void @test3(%struct.large_oddsize* nocapture %x) nounwind minsize { 136; NOFAST32-LABEL: test3: 137; NOFAST32: # %bb.0: 138; NOFAST32-NEXT: pushl %edi 139; NOFAST32-NEXT: pushl %esi 140; NOFAST32-NEXT: subl $4100, %esp # imm = 0x1004 141; NOFAST32-NEXT: movl {{[0-9]+}}(%esp), %esi 142; NOFAST32-NEXT: movl $4095, %ecx # imm = 0xFFF 143; NOFAST32-NEXT: movl %esp, %edi 144; NOFAST32-NEXT: rep;movsb (%esi), %es:(%edi) 145; NOFAST32-NEXT: calll foo_oddsize 146; NOFAST32-NEXT: addl $4100, %esp # imm = 0x1004 147; NOFAST32-NEXT: popl %esi 148; NOFAST32-NEXT: popl %edi 149; NOFAST32-NEXT: retl 150; 151; FAST32-LABEL: test3: 152; FAST32: # %bb.0: 153; FAST32-NEXT: pushl %edi 154; FAST32-NEXT: pushl %esi 155; FAST32-NEXT: subl $4100, %esp # imm = 0x1004 156; FAST32-NEXT: movl {{[0-9]+}}(%esp), %esi 157; FAST32-NEXT: movl $4095, %ecx # imm = 0xFFF 158; FAST32-NEXT: movl %esp, %edi 159; FAST32-NEXT: rep;movsb (%esi), %es:(%edi) 160; FAST32-NEXT: calll foo_oddsize 161; FAST32-NEXT: addl $4100, %esp # imm = 0x1004 162; FAST32-NEXT: popl %esi 163; FAST32-NEXT: popl %edi 164; FAST32-NEXT: retl 165; 166; NOFAST-LABEL: test3: 167; NOFAST: # %bb.0: 168; NOFAST-NEXT: subq $4104, %rsp # imm = 0x1008 169; NOFAST-NEXT: movq %rdi, %rsi 170; NOFAST-NEXT: movl $4095, %ecx # imm = 0xFFF 171; NOFAST-NEXT: movq %rsp, %rdi 172; NOFAST-NEXT: rep;movsb (%rsi), %es:(%rdi) 173; NOFAST-NEXT: callq foo_oddsize 174; NOFAST-NEXT: addq $4104, %rsp # imm = 0x1008 175; NOFAST-NEXT: retq 176; 177; FAST-LABEL: test3: 178; FAST: # %bb.0: 179; FAST-NEXT: subq $4104, %rsp # imm = 0x1008 180; FAST-NEXT: movq %rdi, %rsi 181; FAST-NEXT: movl $4095, %ecx # imm = 0xFFF 182; FAST-NEXT: movq %rsp, %rdi 183; FAST-NEXT: rep;movsb (%rsi), %es:(%rdi) 184; FAST-NEXT: callq foo_oddsize 185; FAST-NEXT: addq $4104, %rsp # imm = 0x1008 186; FAST-NEXT: retq 187 call void @foo_oddsize(%struct.large_oddsize* align 8 byval(%struct.large_oddsize) %x) 188 ret void 189 190} 191