371 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+amx-int8 -mattr=+avx512f -verify-machineinstrs | FileCheck %s3@buf = dso_local global [3072 x i8] zeroinitializer, align 164 5define dso_local void @test1(i16 signext %0, i16 signext %1) nounwind {6; CHECK-LABEL: test1:7; CHECK: # %bb.0:8; CHECK-NEXT: vxorps %xmm0, %xmm0, %xmm09; CHECK-NEXT: vmovups %zmm0, -{{[0-9]+}}(%rsp)10; CHECK-NEXT: movb $1, -{{[0-9]+}}(%rsp)11; CHECK-NEXT: movw $8, -{{[0-9]+}}(%rsp)12; CHECK-NEXT: movb $8, -{{[0-9]+}}(%rsp)13; CHECK-NEXT: movw %si, -{{[0-9]+}}(%rsp)14; CHECK-NEXT: movb %dil, -{{[0-9]+}}(%rsp)15; CHECK-NEXT: movw %si, -{{[0-9]+}}(%rsp)16; CHECK-NEXT: movb %dil, -{{[0-9]+}}(%rsp)17; CHECK-NEXT: ldtilecfg -{{[0-9]+}}(%rsp)18; CHECK-NEXT: movl $buf, %eax19; CHECK-NEXT: movl $32, %ecx20; CHECK-NEXT: movw $8, %dx21; CHECK-NEXT: tileloadd (%rax,%rcx), %tmm022; CHECK-NEXT: movl $buf+1024, %eax23; CHECK-NEXT: tileloadd (%rax,%rcx), %tmm124; CHECK-NEXT: movl $buf+2048, %eax25; CHECK-NEXT: tileloadd (%rax,%rcx), %tmm226; CHECK-NEXT: tdpbssd %tmm1, %tmm0, %tmm227; CHECK-NEXT: tilestored %tmm2, (%rax,%rcx)28; CHECK-NEXT: tilerelease29; CHECK-NEXT: vzeroupper30; CHECK-NEXT: jmp foo # TAILCALL31 %3 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 %0, i16 8, ptr @buf, i64 32)32 %4 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 8, i16 %1, ptr getelementptr inbounds ([3072 x i8], ptr @buf, i64 0, i64 1024), i64 32)33 %5 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 %0, i16 %1, ptr getelementptr inbounds ([3072 x i8], ptr @buf, i64 0, i64 2048), i64 32)34 %6 = tail call x86_amx @llvm.x86.tdpbssd.internal(i16 %0, i16 %1, i16 8, x86_amx %5, x86_amx %3, x86_amx %4)35 call void @llvm.dbg.value(metadata x86_amx %6, metadata !DILocalVariable(name: "1", scope: !2), metadata !DIExpression()), !dbg !336 tail call void @llvm.x86.tilestored64.internal(i16 %0, i16 %1, ptr getelementptr inbounds ([3072 x i8], ptr @buf, i64 0, i64 2048), i64 32, x86_amx %6)37 tail call void @foo()38 ret void39}40 41define dso_local void @test2(i16 signext %0, i16 signext %1) nounwind {42; CHECK-LABEL: test2:43; CHECK: # %bb.0:44; CHECK-NEXT: pushq %rbp45; CHECK-NEXT: pushq %rbx46; CHECK-NEXT: subq $72, %rsp47; CHECK-NEXT: movl %esi, %ebx48; CHECK-NEXT: movl %edi, %ebp49; CHECK-NEXT: vxorps %xmm0, %xmm0, %xmm050; CHECK-NEXT: vmovups %zmm0, {{[0-9]+}}(%rsp)51; CHECK-NEXT: movb $1, {{[0-9]+}}(%rsp)52; CHECK-NEXT: movw $8, {{[0-9]+}}(%rsp)53; CHECK-NEXT: movb $8, {{[0-9]+}}(%rsp)54; CHECK-NEXT: movw $8, {{[0-9]+}}(%rsp)55; CHECK-NEXT: movb $8, {{[0-9]+}}(%rsp)56; CHECK-NEXT: movw %bx, {{[0-9]+}}(%rsp)57; CHECK-NEXT: movb %bpl, {{[0-9]+}}(%rsp)58; CHECK-NEXT: movw %bx, {{[0-9]+}}(%rsp)59; CHECK-NEXT: movb %bpl, {{[0-9]+}}(%rsp)60; CHECK-NEXT: movw %bx, {{[0-9]+}}(%rsp)61; CHECK-NEXT: movb %bpl, {{[0-9]+}}(%rsp)62; CHECK-NEXT: vzeroupper63; CHECK-NEXT: callq foo64; CHECK-NEXT: ldtilecfg {{[0-9]+}}(%rsp)65; CHECK-NEXT: xorl %eax, %eax66; CHECK-NEXT: testb %al, %al67; CHECK-NEXT: jne .LBB1_368; CHECK-NEXT: # %bb.1: # %if.true69; CHECK-NEXT: movw $8, %ax70; CHECK-NEXT: tilezero %tmm071; CHECK-NEXT: movl $32, %ecx72; CHECK-NEXT: movl $buf+1024, %edx73; CHECK-NEXT: tileloadd (%rdx,%rcx), %tmm174; CHECK-NEXT: movl $buf+2048, %edx75; CHECK-NEXT: tileloadd (%rdx,%rcx), %tmm276; CHECK-NEXT: tdpbssd %tmm2, %tmm1, %tmm077; CHECK-NEXT: tilestored %tmm0, (%rdx,%rcx)78; CHECK-NEXT: jmp .LBB1_279; CHECK-NEXT: .LBB1_3: # %if.false80; CHECK-NEXT: movl $buf, %eax81; CHECK-NEXT: movl $32, %ecx82; CHECK-NEXT: movw $8, %dx83; CHECK-NEXT: tileloadd (%rax,%rcx), %tmm384; CHECK-NEXT: movl $buf+1024, %eax85; CHECK-NEXT: tileloadd (%rax,%rcx), %tmm486; CHECK-NEXT: movl $buf+2048, %eax87; CHECK-NEXT: tileloadd (%rax,%rcx), %tmm288; CHECK-NEXT: tdpbssd %tmm2, %tmm4, %tmm389; CHECK-NEXT: tilestored %tmm3, (%rax,%rcx)90; CHECK-NEXT: .LBB1_2: # %if.true91; CHECK-NEXT: addq $72, %rsp92; CHECK-NEXT: popq %rbx93; CHECK-NEXT: popq %rbp94; CHECK-NEXT: tilerelease95; CHECK-NEXT: retq96 call void @foo()97 br i1 undef, label %if.true, label %if.false98 99if.true:100 %t1 = tail call x86_amx @llvm.x86.tilezero.internal(i16 %0, i16 8)101 %t2 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 8, i16 %1, ptr getelementptr inbounds ([3072 x i8], ptr @buf, i64 0, i64 1024), i64 32)102 %t3 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 %0, i16 %1, ptr getelementptr inbounds ([3072 x i8], ptr @buf, i64 0, i64 2048), i64 32)103 %t4 = tail call x86_amx @llvm.x86.tdpbssd.internal(i16 %0, i16 %1, i16 8, x86_amx %t1, x86_amx %t2, x86_amx %t3)104 tail call void @llvm.x86.tilestored64.internal(i16 %0, i16 %1, ptr getelementptr inbounds ([3072 x i8], ptr @buf, i64 0, i64 2048), i64 32, x86_amx %t4)105 br label %exit106 107if.false:108 %t5 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 %0, i16 8, ptr @buf, i64 32)109 %t6 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 8, i16 %1, ptr getelementptr inbounds ([3072 x i8], ptr @buf, i64 0, i64 1024), i64 32)110 %t7 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 %0, i16 %1, ptr getelementptr inbounds ([3072 x i8], ptr @buf, i64 0, i64 2048), i64 32)111 %t8 = tail call x86_amx @llvm.x86.tdpbssd.internal(i16 %0, i16 %1, i16 8, x86_amx %t5, x86_amx %t6, x86_amx %t7)112 tail call void @llvm.x86.tilestored64.internal(i16 %0, i16 %1, ptr getelementptr inbounds ([3072 x i8], ptr @buf, i64 0, i64 2048), i64 32, x86_amx %t8)113 br label %exit114 115exit:116 ret void117}118 119define dso_local void @test3(i16 signext %0, i16 signext %1) nounwind {120; CHECK-LABEL: test3:121; CHECK: # %bb.0:122; CHECK-NEXT: vxorps %xmm0, %xmm0, %xmm0123; CHECK-NEXT: vmovups %zmm0, -{{[0-9]+}}(%rsp)124; CHECK-NEXT: movb $1, -{{[0-9]+}}(%rsp)125; CHECK-NEXT: movw %si, -{{[0-9]+}}(%rsp)126; CHECK-NEXT: movb %dil, -{{[0-9]+}}(%rsp)127; CHECK-NEXT: xorl %eax, %eax128; CHECK-NEXT: testb %al, %al129; CHECK-NEXT: jne .LBB2_2130; CHECK-NEXT: # %bb.1: # %if.true131; CHECK-NEXT: incl %edi132; CHECK-NEXT: jmp .LBB2_3133; CHECK-NEXT: .LBB2_2: # %if.false134; CHECK-NEXT: decl %edi135; CHECK-NEXT: .LBB2_3: # %exit136; CHECK-NEXT: movb %dil, -{{[0-9]+}}(%rsp)137; CHECK-NEXT: ldtilecfg -{{[0-9]+}}(%rsp)138; CHECK-NEXT: tilezero %tmm0139; CHECK-NEXT: movl $buf, %eax140; CHECK-NEXT: movl $32, %ecx141; CHECK-NEXT: tilestored %tmm0, (%rax,%rcx)142; CHECK-NEXT: tilerelease143; CHECK-NEXT: vzeroupper144; CHECK-NEXT: retq145 br i1 undef, label %if.true, label %if.false146 147if.true:148 %3 = add i16 %0, 1149 br label %exit150 151if.false:152 %4 = sub i16 %0, 1153 br label %exit154 155exit:156 %5 = phi i16 [ %3, %if.true ], [ %4, %if.false ]157 %6 = tail call x86_amx @llvm.x86.tilezero.internal(i16 %5, i16 %1)158 tail call void @llvm.x86.tilestored64.internal(i16 %5, i16 %1, ptr @buf, i64 32, x86_amx %6)159 ret void160}161 162define dso_local void @test4(i16 signext %0, i16 signext %1) nounwind {163; CHECK-LABEL: test4:164; CHECK: # %bb.0:165; CHECK-NEXT: vxorps %xmm0, %xmm0, %xmm0166; CHECK-NEXT: vmovups %zmm0, -{{[0-9]+}}(%rsp)167; CHECK-NEXT: movb $1, -{{[0-9]+}}(%rsp)168; CHECK-NEXT: movw %si, -{{[0-9]+}}(%rsp)169; CHECK-NEXT: movb %dil, -{{[0-9]+}}(%rsp)170; CHECK-NEXT: xorl %eax, %eax171; CHECK-NEXT: testb %al, %al172; CHECK-NEXT: jne .LBB3_3173; CHECK-NEXT: # %bb.1: # %if.true174; CHECK-NEXT: incl %edi175; CHECK-NEXT: movb %dil, -{{[0-9]+}}(%rsp)176; CHECK-NEXT: ldtilecfg -{{[0-9]+}}(%rsp)177; CHECK-NEXT: xorl %eax, %eax178; CHECK-NEXT: testb %al, %al179; CHECK-NEXT: jne .LBB3_4180; CHECK-NEXT: .LBB3_2: # %amx2181; CHECK-NEXT: movl $32, %eax182; CHECK-NEXT: movl $buf+1024, %ecx183; CHECK-NEXT: tileloadd (%rcx,%rax), %tmm0184; CHECK-NEXT: movl $buf, %ecx185; CHECK-NEXT: tilestored %tmm0, (%rcx,%rax)186; CHECK-NEXT: tilerelease187; CHECK-NEXT: vzeroupper188; CHECK-NEXT: retq189; CHECK-NEXT: .LBB3_3: # %if.false190; CHECK-NEXT: decl %edi191; CHECK-NEXT: movb %dil, -{{[0-9]+}}(%rsp)192; CHECK-NEXT: ldtilecfg -{{[0-9]+}}(%rsp)193; CHECK-NEXT: xorl %eax, %eax194; CHECK-NEXT: testb %al, %al195; CHECK-NEXT: jne .LBB3_2196; CHECK-NEXT: .LBB3_4: # %amx1197; CHECK-NEXT: tilezero %tmm0198; CHECK-NEXT: movl $buf, %eax199; CHECK-NEXT: movl $32, %ecx200; CHECK-NEXT: tilestored %tmm0, (%rax,%rcx)201; CHECK-NEXT: tilerelease202; CHECK-NEXT: vzeroupper203; CHECK-NEXT: retq204 br i1 undef, label %if.true, label %if.false205 206if.true:207 %3 = add i16 %0, 1208 br i1 undef, label %amx1, label %amx2209 210if.false:211 %4 = sub i16 %0, 1212 br i1 undef, label %amx2, label %amx1213 214amx1:215 %5 = phi i16 [ %3, %if.true ], [ %4, %if.false ]216 %6 = tail call x86_amx @llvm.x86.tilezero.internal(i16 %5, i16 %1)217 tail call void @llvm.x86.tilestored64.internal(i16 %5, i16 %1, ptr @buf, i64 32, x86_amx %6)218 br label %exit219 220amx2:221 %7 = phi i16 [ %3, %if.true ], [ %4, %if.false ]222 %8 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 %7, i16 %1, ptr getelementptr inbounds ([3072 x i8], ptr @buf, i64 0, i64 1024), i64 32)223 tail call void @llvm.x86.tilestored64.internal(i16 %7, i16 %1, ptr @buf, i64 32, x86_amx %8)224 br label %exit225 226exit:227 ret void228}229 230define dso_local void @test5(i16 signext %0, i16 signext %1) nounwind {231; CHECK-LABEL: test5:232; CHECK: # %bb.0: # %entry233; CHECK-NEXT: vxorps %xmm0, %xmm0, %xmm0234; CHECK-NEXT: vmovups %zmm0, -{{[0-9]+}}(%rsp)235; CHECK-NEXT: movb $1, -{{[0-9]+}}(%rsp)236; CHECK-NEXT: movw %si, -{{[0-9]+}}(%rsp)237; CHECK-NEXT: movb %dil, -{{[0-9]+}}(%rsp)238; CHECK-NEXT: xorl %eax, %eax239; CHECK-NEXT: movl $buf, %ecx240; CHECK-NEXT: movl $32, %edx241; CHECK-NEXT: movl %esi, %r8d242; CHECK-NEXT: decl %r8d243; CHECK-NEXT: jmp .LBB4_1244; CHECK-NEXT: .p2align 4245; CHECK-NEXT: .LBB4_3: # %if.false246; CHECK-NEXT: # in Loop: Header=BB4_1 Depth=1247; CHECK-NEXT: movl %r8d, %esi248; CHECK-NEXT: movw %r8w, -{{[0-9]+}}(%rsp)249; CHECK-NEXT: cmpw $7, %si250; CHECK-NEXT: jne .LBB4_5251; CHECK-NEXT: .LBB4_1: # %loop.bb1252; CHECK-NEXT: # =>This Inner Loop Header: Depth=1253; CHECK-NEXT: ldtilecfg -{{[0-9]+}}(%rsp)254; CHECK-NEXT: testb %al, %al255; CHECK-NEXT: jne .LBB4_3256; CHECK-NEXT: # %bb.2: # %if.true257; CHECK-NEXT: # in Loop: Header=BB4_1 Depth=1258; CHECK-NEXT: tilezero %tmm0259; CHECK-NEXT: tilestored %tmm0, (%rcx,%rdx)260; CHECK-NEXT: cmpw $7, %si261; CHECK-NEXT: je .LBB4_1262; CHECK-NEXT: .LBB4_5: # %exit263; CHECK-NEXT: tilerelease264; CHECK-NEXT: vzeroupper265; CHECK-NEXT: retq266entry:267 br label %loop.bb1268 269loop.bb1:270 %2 = phi i16 [ %1, %entry ], [ %5, %loop.bb2 ]271 br i1 undef, label %if.true, label %if.false272 273if.true:274 %3 = tail call x86_amx @llvm.x86.tilezero.internal(i16 %0, i16 %2)275 tail call void @llvm.x86.tilestored64.internal(i16 %0, i16 %2, ptr @buf, i64 32, x86_amx %3)276 br label %loop.bb2277 278if.false:279 %4 = sub i16 %1, 1280 br label %loop.bb2281 282loop.bb2:283 %5 = phi i16 [ %2, %if.true ], [ %4, %if.false ]284 %6 = icmp eq i16 %5, 7285 br i1 %6, label %loop.bb1, label %exit286 287exit:288 ret void289}290 291define dso_local void @test6(i16 signext %0) nounwind {292; CHECK-LABEL: test6:293; CHECK: # %bb.0: # %entry294; CHECK-NEXT: # kill: def $edi killed $edi def $rdi295; CHECK-NEXT: vxorps %xmm0, %xmm0, %xmm0296; CHECK-NEXT: vmovups %zmm0, -{{[0-9]+}}(%rsp)297; CHECK-NEXT: movb $1, -{{[0-9]+}}(%rsp)298; CHECK-NEXT: movb %dil, -{{[0-9]+}}(%rsp)299; CHECK-NEXT: xorl %eax, %eax300; CHECK-NEXT: xorl %ecx, %ecx301; CHECK-NEXT: movl $buf, %edx302; CHECK-NEXT: movl $32, %esi303; CHECK-NEXT: jmp .LBB5_1304; CHECK-NEXT: .p2align 4305; CHECK-NEXT: .LBB5_3: # %if.false306; CHECK-NEXT: # in Loop: Header=BB5_1 Depth=1307; CHECK-NEXT: decl %eax308; CHECK-NEXT: .LBB5_4: # %loop.bb2309; CHECK-NEXT: # in Loop: Header=BB5_1 Depth=1310; CHECK-NEXT: leal (%rdi,%rax), %r8d311; CHECK-NEXT: movw %r8w, -{{[0-9]+}}(%rsp)312; CHECK-NEXT: cmpw $7, %ax313; CHECK-NEXT: ldtilecfg -{{[0-9]+}}(%rsp)314; CHECK-NEXT: tilezero %tmm0315; CHECK-NEXT: tilestored %tmm0, (%rdx,%rsi)316; CHECK-NEXT: jne .LBB5_5317; CHECK-NEXT: .LBB5_1: # %loop.bb1318; CHECK-NEXT: # =>This Inner Loop Header: Depth=1319; CHECK-NEXT: testb %cl, %cl320; CHECK-NEXT: jne .LBB5_3321; CHECK-NEXT: # %bb.2: # %if.true322; CHECK-NEXT: # in Loop: Header=BB5_1 Depth=1323; CHECK-NEXT: incl %eax324; CHECK-NEXT: jmp .LBB5_4325; CHECK-NEXT: .LBB5_5: # %exit326; CHECK-NEXT: tilerelease327; CHECK-NEXT: vzeroupper328; CHECK-NEXT: retq329entry:330 br label %loop.bb1331 332loop.bb1:333 %1 = phi i16 [ 0, %entry ], [ %4, %loop.bb2 ]334 br i1 undef, label %if.true, label %if.false335 336if.true:337 %2 = add i16 %1, 1338 br label %loop.bb2339 340if.false:341 %3 = sub i16 %1, 1342 br label %loop.bb2343 344loop.bb2:345 %4 = phi i16 [ %2, %if.true ], [ %3, %if.false ]346 %5 = icmp eq i16 %4, 7347 %6 = add i16 %0, %4348 %7 = tail call x86_amx @llvm.x86.tilezero.internal(i16 %0, i16 %6)349 tail call void @llvm.x86.tilestored64.internal(i16 %0, i16 %6, ptr @buf, i64 32, x86_amx %7)350 br i1 %5, label %loop.bb1, label %exit351 352exit:353 ret void354}355 356 357declare dso_local void @foo() nounwind358declare void @llvm.dbg.value(metadata, metadata, metadata)359declare x86_amx @llvm.x86.tilezero.internal(i16, i16)360declare x86_amx @llvm.x86.tileloadd64.internal(i16, i16, ptr, i64)361declare x86_amx @llvm.x86.tdpbssd.internal(i16, i16, i16, x86_amx, x86_amx, x86_amx)362declare void @llvm.x86.tilestored64.internal(i16, i16, ptr, i64, x86_amx)363 364!llvm.dbg.cu = !{!0}365!llvm.module.flags = !{!1}366 367!0 = distinct !DICompileUnit(language: DW_LANG_C, file: !DIFile(filename: "1", directory: "1"))368!1 = !{i32 2, !"Debug Info Version", i32 3}369!2 = distinct !DISubprogram(unit: !0)370!3 = !DILocation(line: 1, column: 1, scope: !2)371