1 //==- AArch64SchedKryo.td - Qualcomm Kryo Scheduling Defs ---*- tablegen -*-==//
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
7 //===----------------------------------------------------------------------===//
9 // This file defines the machine model for Qualcomm Kryo to support
10 // instruction scheduling and other instruction cost heuristics.
12 //===----------------------------------------------------------------------===//
14 //===----------------------------------------------------------------------===//
15 // The issue width is set to five, matching the five issue queues for expanded
16 // uops. Now, the latency spreadsheet has information based on fragmented uops,
17 // but these do not actually take up an issue queue.
19 def KryoModel : SchedMachineModel {
20 let IssueWidth = 5; // 5-wide issue for expanded uops
21 let MicroOpBufferSize = 128; // Out-of-order with temporary unified issue buffer
22 let LoadLatency = 4; // Optimistic load latency
23 let MispredictPenalty = 14; // Fetch + Decode/Rename/Dispatch + Branch
25 // Enable partial & runtime unrolling. The magic number is chosen based on
26 // experiments and benchmarking data.
27 let LoopMicroOpBufferSize = 16;
28 let CompleteModel = 1;
30 list<Predicate> UnsupportedFeatures = SVEUnsupported.F;
32 // FIXME: Remove when all errors have been fixed.
33 let FullInstRWOverlapCheck = 0;
36 //===----------------------------------------------------------------------===//
37 // Define each kind of processor resource and number available on Kryo.
39 let SchedModel = KryoModel in {
40 def KryoUnitXA : ProcResource<1>; // Type X(A) micro-ops
41 def KryoUnitXB : ProcResource<1>; // Type X(B) micro-ops
42 def KryoUnitYA : ProcResource<1>; // Type Y(A) micro-ops
43 def KryoUnitYB : ProcResource<1>; // Type Y(B) micro-ops
44 def KryoUnitX : ProcResGroup<[KryoUnitXA, // Type X micro-ops
46 def KryoUnitY : ProcResGroup<[KryoUnitYA, // Type Y micro-ops
48 def KryoUnitXY : ProcResGroup<[KryoUnitXA, // Type XY micro-ops
52 def KryoUnitLSA : ProcResource<1>; // Type LS(A) micro-ops
53 def KryoUnitLSB : ProcResource<1>; // Type LS(B) micro-ops
54 def KryoUnitLS : ProcResGroup<[KryoUnitLSA, // Type LS micro-ops
58 let SchedModel = KryoModel in {
60 //===----------------------------------------------------------------------===//
61 // Map the target-defined scheduler read/write resources and latency for
64 def : WriteRes<WriteImm, [KryoUnitXY]> { let Latency = 1; }
65 def : WriteRes<WriteI, [KryoUnitXY]> { let Latency = 1; }
66 def : WriteRes<WriteISReg, [KryoUnitXY, KryoUnitXY]>
67 { let Latency = 2; let NumMicroOps = 2; }
68 def : WriteRes<WriteIEReg, [KryoUnitXY, KryoUnitXY]>
69 { let Latency = 2; let NumMicroOps = 2; }
70 def : WriteRes<WriteExtr, [KryoUnitXY, KryoUnitX]>
71 { let Latency = 2; let NumMicroOps = 2; }
72 def : WriteRes<WriteIS, [KryoUnitXY]> { let Latency = 2; }
73 def : WriteRes<WriteID32, [KryoUnitXA, KryoUnitY]>
74 { let Latency = 8; let NumMicroOps = 1; } // Fragent -1
75 def : WriteRes<WriteID64, [KryoUnitXA, KryoUnitY]>
76 { let Latency = 8; let NumMicroOps = 1; } // Fragent -1
77 def : WriteRes<WriteIM32, [KryoUnitX]> { let Latency = 5; }
78 def : WriteRes<WriteIM64, [KryoUnitX]> { let Latency = 5; }
79 def : WriteRes<WriteBr, [KryoUnitXY]> { let Latency = 1; }
80 def : WriteRes<WriteBrReg, [KryoUnitXY]> { let Latency = 1; }
81 def : WriteRes<WriteLD, [KryoUnitLS]> { let Latency = 4; }
82 def : WriteRes<WriteST, [KryoUnitLS]> { let Latency = 4; }
83 def : WriteRes<WriteSTP, [KryoUnitLS]> { let Latency = 4; }
84 def : WriteRes<WriteAdr, [KryoUnitXY]> { let Latency = 6; }
85 def : WriteRes<WriteLDIdx, [KryoUnitLS]> { let Latency = 4; }
86 def : WriteRes<WriteSTIdx, [KryoUnitLS]> { let Latency = 4; }
87 def : WriteRes<WriteF, [KryoUnitXY, KryoUnitXY]>
88 { let Latency = 3; let NumMicroOps = 2; }
89 def : WriteRes<WriteFCmp, [KryoUnitXY]> { let Latency = 2; }
90 def : WriteRes<WriteFCvt, [KryoUnitX]> { let Latency = 4; }
91 def : WriteRes<WriteFCopy, [KryoUnitXY]> { let Latency = 6; }
92 def : WriteRes<WriteFImm, [KryoUnitXY]> { let Latency = 6; }
93 def : WriteRes<WriteFMul, [KryoUnitX, KryoUnitX]>
94 { let Latency = 6; let NumMicroOps = 2; }
95 def : WriteRes<WriteFDiv, [KryoUnitXA, KryoUnitY]>
96 { let Latency = 12; let NumMicroOps = 2; } // Fragent -1 / NoRSV +1
97 def : WriteRes<WriteV, [KryoUnitXY]> { let Latency = 6; }
98 def : WriteRes<WriteVLD, [KryoUnitLS]> { let Latency = 4; }
99 def : WriteRes<WriteVST, [KryoUnitLS]> { let Latency = 4; }
101 def : WriteRes<WriteSys, []> { let Latency = 1; }
102 def : WriteRes<WriteBarrier, []> { let Latency = 1; }
103 def : WriteRes<WriteHint, []> { let Latency = 1; }
105 def : WriteRes<WriteLDHi, []> { let Latency = 4; }
107 def : WriteRes<WriteAtomic, []> { let Unsupported = 1; }
109 // No forwarding logic is modelled yet.
110 def : ReadAdvance<ReadI, 0>;
111 def : ReadAdvance<ReadISReg, 0>;
112 def : ReadAdvance<ReadIEReg, 0>;
113 def : ReadAdvance<ReadIM, 0>;
114 def : ReadAdvance<ReadIMA, 0>;
115 def : ReadAdvance<ReadID, 0>;
116 def : ReadAdvance<ReadExtrHi, 0>;
117 def : ReadAdvance<ReadAdrBase, 0>;
118 def : ReadAdvance<ReadVLD, 0>;
121 //===----------------------------------------------------------------------===//
122 // Specialize the coarse model by associating instruction groups with the
123 // subtarget-defined types. As the modeled is refined, this will override most
124 // of the above SchedWriteRes and SchedAlias mappings.
127 // -----------------------------------------------------------------------------
129 def : InstRW<[WriteI], (instrs COPY)>;
132 // Detailed Refinedments
133 // -----------------------------------------------------------------------------
134 include "AArch64SchedKryoDetails.td"
137 } // SchedModel = KryoModel