1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2; RUN: llc < %s -mtriple=i686-- | FileCheck %s --check-prefixes=X86,X86-BASE 3; RUN: llc < %s -mtriple=i686-- -mattr=movbe | FileCheck %s --check-prefixes=X86,X86-MOVBE 4; RUN: llc < %s -mtriple=x86_64-- | FileCheck %s --check-prefixes=X64,X64-BASE 5; RUN: llc < %s -mtriple=x86_64-- -mattr=movbe | FileCheck %s --check-prefixes=X64,X64-MOVBE 6 7define i16 @foo(i16 %x, i16 %y, i16 %z) nounwind { 8; X86-LABEL: foo: 9; X86: # %bb.0: 10; X86-NEXT: movb {{[0-9]+}}(%esp), %cl 11; X86-NEXT: movzwl {{[0-9]+}}(%esp), %eax 12; X86-NEXT: rolw %cl, %ax 13; X86-NEXT: retl 14; 15; X64-LABEL: foo: 16; X64: # %bb.0: 17; X64-NEXT: movl %edx, %ecx 18; X64-NEXT: movl %edi, %eax 19; X64-NEXT: # kill: def $cl killed $cl killed $ecx 20; X64-NEXT: rolw %cl, %ax 21; X64-NEXT: # kill: def $ax killed $ax killed $eax 22; X64-NEXT: retq 23 %t0 = shl i16 %x, %z 24 %t1 = sub i16 16, %z 25 %t2 = lshr i16 %x, %t1 26 %t3 = or i16 %t2, %t0 27 ret i16 %t3 28} 29 30define i16 @bar(i16 %x, i16 %y, i16 %z) nounwind { 31; X86-LABEL: bar: 32; X86: # %bb.0: 33; X86-NEXT: movzwl {{[0-9]+}}(%esp), %edx 34; X86-NEXT: movzwl {{[0-9]+}}(%esp), %eax 35; X86-NEXT: movb {{[0-9]+}}(%esp), %cl 36; X86-NEXT: andb $15, %cl 37; X86-NEXT: shldw %cl, %dx, %ax 38; X86-NEXT: retl 39; 40; X64-LABEL: bar: 41; X64: # %bb.0: 42; X64-NEXT: movl %edx, %ecx 43; X64-NEXT: movl %esi, %eax 44; X64-NEXT: andb $15, %cl 45; X64-NEXT: # kill: def $cl killed $cl killed $ecx 46; X64-NEXT: shldw %cl, %di, %ax 47; X64-NEXT: # kill: def $ax killed $ax killed $eax 48; X64-NEXT: retq 49 %t0 = shl i16 %y, %z 50 %t1 = sub i16 16, %z 51 %t2 = lshr i16 %x, %t1 52 %t3 = or i16 %t2, %t0 53 ret i16 %t3 54} 55 56define i16 @un(i16 %x, i16 %y, i16 %z) nounwind { 57; X86-LABEL: un: 58; X86: # %bb.0: 59; X86-NEXT: movb {{[0-9]+}}(%esp), %cl 60; X86-NEXT: movzwl {{[0-9]+}}(%esp), %eax 61; X86-NEXT: rorw %cl, %ax 62; X86-NEXT: retl 63; 64; X64-LABEL: un: 65; X64: # %bb.0: 66; X64-NEXT: movl %edx, %ecx 67; X64-NEXT: movl %edi, %eax 68; X64-NEXT: # kill: def $cl killed $cl killed $ecx 69; X64-NEXT: rorw %cl, %ax 70; X64-NEXT: # kill: def $ax killed $ax killed $eax 71; X64-NEXT: retq 72 %t0 = lshr i16 %x, %z 73 %t1 = sub i16 16, %z 74 %t2 = shl i16 %x, %t1 75 %t3 = or i16 %t2, %t0 76 ret i16 %t3 77} 78 79define i16 @bu(i16 %x, i16 %y, i16 %z) nounwind { 80; X86-LABEL: bu: 81; X86: # %bb.0: 82; X86-NEXT: movzwl {{[0-9]+}}(%esp), %edx 83; X86-NEXT: movzwl {{[0-9]+}}(%esp), %eax 84; X86-NEXT: movb {{[0-9]+}}(%esp), %cl 85; X86-NEXT: andb $15, %cl 86; X86-NEXT: shrdw %cl, %dx, %ax 87; X86-NEXT: retl 88; 89; X64-LABEL: bu: 90; X64: # %bb.0: 91; X64-NEXT: movl %edx, %ecx 92; X64-NEXT: movl %esi, %eax 93; X64-NEXT: andb $15, %cl 94; X64-NEXT: # kill: def $cl killed $cl killed $ecx 95; X64-NEXT: shrdw %cl, %di, %ax 96; X64-NEXT: # kill: def $ax killed $ax killed $eax 97; X64-NEXT: retq 98 %t0 = lshr i16 %y, %z 99 %t1 = sub i16 16, %z 100 %t2 = shl i16 %x, %t1 101 %t3 = or i16 %t2, %t0 102 ret i16 %t3 103} 104 105define i16 @xfoo(i16 %x, i16 %y, i16 %z) nounwind { 106; X86-LABEL: xfoo: 107; X86: # %bb.0: 108; X86-NEXT: movzwl {{[0-9]+}}(%esp), %eax 109; X86-NEXT: rolw $5, %ax 110; X86-NEXT: retl 111; 112; X64-LABEL: xfoo: 113; X64: # %bb.0: 114; X64-NEXT: movl %edi, %eax 115; X64-NEXT: rolw $5, %ax 116; X64-NEXT: # kill: def $ax killed $ax killed $eax 117; X64-NEXT: retq 118 %t0 = lshr i16 %x, 11 119 %t1 = shl i16 %x, 5 120 %t2 = or i16 %t0, %t1 121 ret i16 %t2 122} 123 124define i16 @xbar(i16 %x, i16 %y, i16 %z) nounwind { 125; X86-LABEL: xbar: 126; X86: # %bb.0: 127; X86-NEXT: movzwl {{[0-9]+}}(%esp), %ecx 128; X86-NEXT: movzwl {{[0-9]+}}(%esp), %eax 129; X86-NEXT: shldw $5, %cx, %ax 130; X86-NEXT: retl 131; 132; X64-LABEL: xbar: 133; X64: # %bb.0: 134; X64-NEXT: movl %esi, %eax 135; X64-NEXT: shldw $5, %di, %ax 136; X64-NEXT: # kill: def $ax killed $ax killed $eax 137; X64-NEXT: retq 138 %t0 = shl i16 %y, 5 139 %t1 = lshr i16 %x, 11 140 %t2 = or i16 %t0, %t1 141 ret i16 %t2 142} 143 144define i16 @xun(i16 %x, i16 %y, i16 %z) nounwind { 145; X86-LABEL: xun: 146; X86: # %bb.0: 147; X86-NEXT: movzwl {{[0-9]+}}(%esp), %eax 148; X86-NEXT: rolw $11, %ax 149; X86-NEXT: retl 150; 151; X64-LABEL: xun: 152; X64: # %bb.0: 153; X64-NEXT: movl %edi, %eax 154; X64-NEXT: rolw $11, %ax 155; X64-NEXT: # kill: def $ax killed $ax killed $eax 156; X64-NEXT: retq 157 %t0 = lshr i16 %x, 5 158 %t1 = shl i16 %x, 11 159 %t2 = or i16 %t0, %t1 160 ret i16 %t2 161} 162 163define i16 @xbu(i16 %x, i16 %y, i16 %z) nounwind { 164; X86-LABEL: xbu: 165; X86: # %bb.0: 166; X86-NEXT: movzwl {{[0-9]+}}(%esp), %ecx 167; X86-NEXT: movzwl {{[0-9]+}}(%esp), %eax 168; X86-NEXT: shldw $11, %cx, %ax 169; X86-NEXT: retl 170; 171; X64-LABEL: xbu: 172; X64: # %bb.0: 173; X64-NEXT: movl %edi, %eax 174; X64-NEXT: shldw $11, %si, %ax 175; X64-NEXT: # kill: def $ax killed $ax killed $eax 176; X64-NEXT: retq 177 %t0 = lshr i16 %y, 5 178 %t1 = shl i16 %x, 11 179 %t2 = or i16 %t0, %t1 180 ret i16 %t2 181} 182 183define i32 @rot16_demandedbits(i32 %x, i32 %y) nounwind { 184; X86-LABEL: rot16_demandedbits: 185; X86: # %bb.0: 186; X86-NEXT: movl {{[0-9]+}}(%esp), %eax 187; X86-NEXT: movl %eax, %ecx 188; X86-NEXT: shrl $11, %ecx 189; X86-NEXT: shll $5, %eax 190; X86-NEXT: orl %ecx, %eax 191; X86-NEXT: movzwl %ax, %eax 192; X86-NEXT: retl 193; 194; X64-LABEL: rot16_demandedbits: 195; X64: # %bb.0: 196; X64-NEXT: movl %edi, %eax 197; X64-NEXT: shrl $11, %eax 198; X64-NEXT: shll $5, %edi 199; X64-NEXT: orl %eax, %edi 200; X64-NEXT: movzwl %di, %eax 201; X64-NEXT: retq 202 %t0 = lshr i32 %x, 11 203 %t1 = shl i32 %x, 5 204 %t2 = or i32 %t0, %t1 205 %t3 = and i32 %t2, 65535 206 ret i32 %t3 207} 208 209define i16 @rot16_trunc(i32 %x, i32 %y) nounwind { 210; X86-LABEL: rot16_trunc: 211; X86: # %bb.0: 212; X86-NEXT: movl {{[0-9]+}}(%esp), %eax 213; X86-NEXT: movl %eax, %ecx 214; X86-NEXT: shrl $11, %ecx 215; X86-NEXT: shll $5, %eax 216; X86-NEXT: orl %ecx, %eax 217; X86-NEXT: # kill: def $ax killed $ax killed $eax 218; X86-NEXT: retl 219; 220; X64-LABEL: rot16_trunc: 221; X64: # %bb.0: 222; X64-NEXT: movl %edi, %eax 223; X64-NEXT: movl %edi, %ecx 224; X64-NEXT: shrl $11, %ecx 225; X64-NEXT: shll $5, %eax 226; X64-NEXT: orl %ecx, %eax 227; X64-NEXT: # kill: def $ax killed $ax killed $eax 228; X64-NEXT: retq 229 %t0 = lshr i32 %x, 11 230 %t1 = shl i32 %x, 5 231 %t2 = or i32 %t0, %t1 232 %t3 = trunc i32 %t2 to i16 233 ret i16 %t3 234} 235 236define i16 @rotate16(i16 %x) { 237; X86-BASE-LABEL: rotate16: 238; X86-BASE: # %bb.0: 239; X86-BASE-NEXT: movzwl {{[0-9]+}}(%esp), %eax 240; X86-BASE-NEXT: rolw $8, %ax 241; X86-BASE-NEXT: retl 242; 243; X86-MOVBE-LABEL: rotate16: 244; X86-MOVBE: # %bb.0: 245; X86-MOVBE-NEXT: movbew {{[0-9]+}}(%esp), %ax 246; X86-MOVBE-NEXT: retl 247; 248; X64-LABEL: rotate16: 249; X64: # %bb.0: 250; X64-NEXT: movl %edi, %eax 251; X64-NEXT: rolw $8, %ax 252; X64-NEXT: # kill: def $ax killed $ax killed $eax 253; X64-NEXT: retq 254 %r = call i16 @llvm.fshl.i16(i16 %x, i16 %x, i16 8) 255 ret i16 %r 256} 257 258; TODO: Should this always be rolw with memory operand? 259 260define void @rotate16_in_place_memory(i8* %p) { 261; X86-BASE-LABEL: rotate16_in_place_memory: 262; X86-BASE: # %bb.0: 263; X86-BASE-NEXT: movl {{[0-9]+}}(%esp), %eax 264; X86-BASE-NEXT: rolw $8, (%eax) 265; X86-BASE-NEXT: retl 266; 267; X86-MOVBE-LABEL: rotate16_in_place_memory: 268; X86-MOVBE: # %bb.0: 269; X86-MOVBE-NEXT: movl {{[0-9]+}}(%esp), %eax 270; X86-MOVBE-NEXT: movzwl (%eax), %ecx 271; X86-MOVBE-NEXT: movbew %cx, (%eax) 272; X86-MOVBE-NEXT: retl 273; 274; X64-BASE-LABEL: rotate16_in_place_memory: 275; X64-BASE: # %bb.0: 276; X64-BASE-NEXT: rolw $8, (%rdi) 277; X64-BASE-NEXT: retq 278; 279; X64-MOVBE-LABEL: rotate16_in_place_memory: 280; X64-MOVBE: # %bb.0: 281; X64-MOVBE-NEXT: movzwl (%rdi), %eax 282; X64-MOVBE-NEXT: movbew %ax, (%rdi) 283; X64-MOVBE-NEXT: retq 284 %p0 = getelementptr i8, i8* %p, i64 0 285 %p1 = getelementptr i8, i8* %p, i64 1 286 %i0 = load i8, i8* %p0, align 1 287 %i1 = load i8, i8* %p1, align 1 288 store i8 %i1, i8* %p0, align 1 289 store i8 %i0, i8* %p1, align 1 290 ret void 291} 292 293define void @rotate16_memory(i8* %p, i8* %q) { 294; X86-BASE-LABEL: rotate16_memory: 295; X86-BASE: # %bb.0: 296; X86-BASE-NEXT: movl {{[0-9]+}}(%esp), %eax 297; X86-BASE-NEXT: movl {{[0-9]+}}(%esp), %ecx 298; X86-BASE-NEXT: movzwl (%ecx), %ecx 299; X86-BASE-NEXT: rolw $8, %cx 300; X86-BASE-NEXT: movw %cx, (%eax) 301; X86-BASE-NEXT: retl 302; 303; X86-MOVBE-LABEL: rotate16_memory: 304; X86-MOVBE: # %bb.0: 305; X86-MOVBE-NEXT: movl {{[0-9]+}}(%esp), %eax 306; X86-MOVBE-NEXT: movl {{[0-9]+}}(%esp), %ecx 307; X86-MOVBE-NEXT: movzwl (%ecx), %ecx 308; X86-MOVBE-NEXT: movbew %cx, (%eax) 309; X86-MOVBE-NEXT: retl 310; 311; X64-BASE-LABEL: rotate16_memory: 312; X64-BASE: # %bb.0: 313; X64-BASE-NEXT: movzwl (%rdi), %eax 314; X64-BASE-NEXT: rolw $8, %ax 315; X64-BASE-NEXT: movw %ax, (%rsi) 316; X64-BASE-NEXT: retq 317; 318; X64-MOVBE-LABEL: rotate16_memory: 319; X64-MOVBE: # %bb.0: 320; X64-MOVBE-NEXT: movzwl (%rdi), %eax 321; X64-MOVBE-NEXT: movbew %ax, (%rsi) 322; X64-MOVBE-NEXT: retq 323 %p0 = getelementptr i8, i8* %p, i64 0 324 %p1 = getelementptr i8, i8* %p, i64 1 325 %q0 = getelementptr i8, i8* %q, i64 0 326 %q1 = getelementptr i8, i8* %q, i64 1 327 %i0 = load i8, i8* %p0, align 1 328 %i1 = load i8, i8* %p1, align 1 329 store i8 %i1, i8* %q0, align 1 330 store i8 %i0, i8* %q1, align 1 331 ret void 332} 333 334declare i16 @llvm.fshl.i16(i16, i16, i16) 335