100 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 4@buf = dso_local global [1024 x i8] zeroinitializer, align 645@buf2 = dso_local global [1024 x i8] zeroinitializer, align 646 7define dso_local void @test_api(i32 %0, i16 signext %1, i16 signext %2) nounwind {8; CHECK-LABEL: test_api:9; CHECK: # %bb.0:10; CHECK-NEXT: subq $968, %rsp # imm = 0x3C811; CHECK-NEXT: vxorps %xmm0, %xmm0, %xmm012; CHECK-NEXT: vmovups %zmm0, -{{[0-9]+}}(%rsp)13; CHECK-NEXT: movb $1, -{{[0-9]+}}(%rsp)14; CHECK-NEXT: movw %dx, -{{[0-9]+}}(%rsp)15; CHECK-NEXT: movb %sil, -{{[0-9]+}}(%rsp)16; CHECK-NEXT: movw %si, -{{[0-9]+}}(%rsp)17; CHECK-NEXT: movb %sil, -{{[0-9]+}}(%rsp)18; CHECK-NEXT: movw %dx, -{{[0-9]+}}(%rsp)19; CHECK-NEXT: movb %dl, -{{[0-9]+}}(%rsp)20; CHECK-NEXT: movw %dx, -{{[0-9]+}}(%rsp)21; CHECK-NEXT: movb %dl, -{{[0-9]+}}(%rsp)22; CHECK-NEXT: movw %dx, -{{[0-9]+}}(%rsp)23; CHECK-NEXT: movb %sil, -{{[0-9]+}}(%rsp)24; CHECK-NEXT: movw %dx, -{{[0-9]+}}(%rsp)25; CHECK-NEXT: movb %dl, -{{[0-9]+}}(%rsp)26; CHECK-NEXT: movw %dx, -{{[0-9]+}}(%rsp)27; CHECK-NEXT: movb %sil, -{{[0-9]+}}(%rsp)28; CHECK-NEXT: movw %dx, -{{[0-9]+}}(%rsp)29; CHECK-NEXT: movb %dl, -{{[0-9]+}}(%rsp)30; CHECK-NEXT: ldtilecfg -{{[0-9]+}}(%rsp)31; CHECK-NEXT: movl $buf, %ecx32; CHECK-NEXT: movl $32, %eax33; CHECK-NEXT: tileloadd (%rcx,%rax), %tmm134; CHECK-NEXT: tileloadd (%rcx,%rax), %tmm135; CHECK-NEXT: movabsq $64, %r836; CHECK-NEXT: tilestored %tmm1, -64(%rsp,%r8) # 1024-byte Folded Spill37; CHECK-NEXT: tileloadd (%rcx,%rax), %tmm338; CHECK-NEXT: tileloadd (%rcx,%rax), %tmm439; CHECK-NEXT: tileloadd (%rcx,%rax), %tmm240; CHECK-NEXT: tileloadd (%rcx,%rax), %tmm541; CHECK-NEXT: tileloadd (%rcx,%rax), %tmm042; CHECK-NEXT: testl %edi, %edi43; CHECK-NEXT: jne .LBB0_244; CHECK-NEXT: # %bb.1:45; CHECK-NEXT: movl $buf2, %ecx46; CHECK-NEXT: .LBB0_2:47; CHECK-NEXT: tileloadd (%rcx,%rax), %tmm648; CHECK-NEXT: tileloadd (%rcx,%rax), %tmm749; CHECK-NEXT: tileloadd (%rcx,%rax), %tmm150; CHECK-NEXT: tdpbssd %tmm7, %tmm6, %tmm151; CHECK-NEXT: movabsq $64, %rax52; CHECK-NEXT: tileloadd -64(%rsp,%rax), %tmm7 # 1024-byte Folded Reload53; CHECK-NEXT: tdpbssd %tmm7, %tmm1, %tmm354; CHECK-NEXT: tdpbssd %tmm4, %tmm3, %tmm255; CHECK-NEXT: tdpbssd %tmm5, %tmm2, %tmm056; CHECK-NEXT: movl $buf, %eax57; CHECK-NEXT: movl $32, %ecx58; CHECK-NEXT: tilestored %tmm0, (%rax,%rcx)59; CHECK-NEXT: addq $968, %rsp # imm = 0x3C860; CHECK-NEXT: tilerelease61; CHECK-NEXT: vzeroupper62; CHECK-NEXT: retq63 %4 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 %1, i16 %2, ptr @buf, i64 32)64 %5 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 %1, i16 %2, ptr @buf, i64 32)65 %6 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 %1, i16 %2, ptr @buf, i64 32)66 %7 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 %2, i16 %2, ptr @buf, i64 32)67 %8 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 %2, i16 %2, ptr @buf, i64 32)68 %9 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 %2, i16 %2, ptr @buf, i64 32)69 %10 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 %2, i16 %2, ptr @buf, i64 32)70 %11 = icmp eq i32 %0, 071 br i1 %11, label %16, label %1272 7312: ; preds = %374 %13 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 %1, i16 %1, ptr @buf, i64 32)75 %14 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 %1, i16 %2, ptr @buf, i64 32)76 %15 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 %1, i16 %2, ptr @buf, i64 32)77 br label %2078 7916: ; preds = %380 %17 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 %1, i16 %1, ptr @buf2, i64 32)81 %18 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 %1, i16 %2, ptr @buf2, i64 32)82 %19 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 %1, i16 %2, ptr @buf2, i64 32)83 br label %2084 8520: ; preds = %16, %1286 %21 = phi x86_amx [ %17, %16 ], [ %13, %12 ]87 %22 = phi x86_amx [ %18, %16 ], [ %14, %12 ]88 %23 = phi x86_amx [ %19, %16 ], [ %15, %12 ]89 %24 = tail call x86_amx @llvm.x86.tdpbssd.internal(i16 %1, i16 %2, i16 %1, x86_amx %23, x86_amx %21, x86_amx %22)90 %25 = tail call x86_amx @llvm.x86.tdpbssd.internal(i16 %1, i16 %2, i16 %2, x86_amx %6, x86_amx %24, x86_amx %5)91 %26 = tail call x86_amx @llvm.x86.tdpbssd.internal(i16 %1, i16 %2, i16 %2, x86_amx %8, x86_amx %25, x86_amx %7)92 %27 = tail call x86_amx @llvm.x86.tdpbssd.internal(i16 %2, i16 %2, i16 %2, x86_amx %10, x86_amx %26, x86_amx %9)93 tail call void @llvm.x86.tilestored64.internal(i16 %2, i16 %2, ptr @buf, i64 32, x86_amx %27)94 ret void95}96 97declare x86_amx @llvm.x86.tileloadd64.internal(i16, i16, ptr, i64)98declare x86_amx @llvm.x86.tdpbssd.internal(i16, i16, i16, x86_amx, x86_amx, x86_amx)99declare void @llvm.x86.tilestored64.internal(i16, i16, ptr, i64, x86_amx)100