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 = !listconcat(SVEUnsupported.F,
34 // FIXME: Remove when all errors have been fixed.
35 let FullInstRWOverlapCheck = 0;
38 //===----------------------------------------------------------------------===//
39 // Define each kind of processor resource and number available on Kryo.
41 let SchedModel = KryoModel in {
42 def KryoUnitXA : ProcResource<1>; // Type X(A) micro-ops
43 def KryoUnitXB : ProcResource<1>; // Type X(B) micro-ops
44 def KryoUnitYA : ProcResource<1>; // Type Y(A) micro-ops
45 def KryoUnitYB : ProcResource<1>; // Type Y(B) micro-ops
46 def KryoUnitX : ProcResGroup<[KryoUnitXA, // Type X micro-ops
48 def KryoUnitY : ProcResGroup<[KryoUnitYA, // Type Y micro-ops
50 def KryoUnitXY : ProcResGroup<[KryoUnitXA, // Type XY micro-ops
54 def KryoUnitLSA : ProcResource<1>; // Type LS(A) micro-ops
55 def KryoUnitLSB : ProcResource<1>; // Type LS(B) micro-ops
56 def KryoUnitLS : ProcResGroup<[KryoUnitLSA, // Type LS micro-ops
60 let SchedModel = KryoModel in {
62 //===----------------------------------------------------------------------===//
63 // Map the target-defined scheduler read/write resources and latency for
66 def : WriteRes<WriteImm, [KryoUnitXY]> { let Latency = 1; }
67 def : WriteRes<WriteI, [KryoUnitXY]> { let Latency = 1; }
68 def : WriteRes<WriteISReg, [KryoUnitXY, KryoUnitXY]>
69 { let Latency = 2; let NumMicroOps = 2; }
70 def : WriteRes<WriteIEReg, [KryoUnitXY, KryoUnitXY]>
71 { let Latency = 2; let NumMicroOps = 2; }
72 def : WriteRes<WriteExtr, [KryoUnitXY, KryoUnitX]>
73 { let Latency = 2; let NumMicroOps = 2; }
74 def : WriteRes<WriteIS, [KryoUnitXY]> { let Latency = 2; }
75 def : WriteRes<WriteID32, [KryoUnitXA, KryoUnitY]>
76 { let Latency = 8; let NumMicroOps = 1; } // Fragent -1
77 def : WriteRes<WriteID64, [KryoUnitXA, KryoUnitY]>
78 { let Latency = 8; let NumMicroOps = 1; } // Fragent -1
79 def : WriteRes<WriteIM32, [KryoUnitX]> { let Latency = 5; }
80 def : WriteRes<WriteIM64, [KryoUnitX]> { let Latency = 5; }
81 def : WriteRes<WriteBr, [KryoUnitXY]> { let Latency = 1; }
82 def : WriteRes<WriteBrReg, [KryoUnitXY]> { let Latency = 1; }
83 def : WriteRes<WriteLD, [KryoUnitLS]> { let Latency = 4; }
84 def : WriteRes<WriteST, [KryoUnitLS]> { let Latency = 4; }
85 def : WriteRes<WriteSTP, [KryoUnitLS]> { let Latency = 4; }
86 def : WriteRes<WriteAdr, [KryoUnitXY]> { let Latency = 6; }
87 def : WriteRes<WriteLDIdx, [KryoUnitLS]> { let Latency = 4; }
88 def : WriteRes<WriteSTIdx, [KryoUnitLS]> { let Latency = 4; }
89 def : WriteRes<WriteF, [KryoUnitXY, KryoUnitXY]>
90 { let Latency = 3; let NumMicroOps = 2; }
91 def : WriteRes<WriteFCmp, [KryoUnitXY]> { let Latency = 2; }
92 def : WriteRes<WriteFCvt, [KryoUnitX]> { let Latency = 4; }
93 def : WriteRes<WriteFCopy, [KryoUnitXY]> { let Latency = 6; }
94 def : WriteRes<WriteFImm, [KryoUnitXY]> { let Latency = 6; }
95 def : WriteRes<WriteFMul, [KryoUnitX, KryoUnitX]>
96 { let Latency = 6; let NumMicroOps = 2; }
97 def : WriteRes<WriteFDiv, [KryoUnitXA, KryoUnitY]>
98 { let Latency = 12; let NumMicroOps = 2; } // Fragent -1 / NoRSV +1
99 def : WriteRes<WriteVd, [KryoUnitXY]> { let Latency = 6; }
100 def : WriteRes<WriteVq, [KryoUnitXY]> { let Latency = 6; }
101 def : WriteRes<WriteVLD, [KryoUnitLS]> { let Latency = 4; }
102 def : WriteRes<WriteVST, [KryoUnitLS]> { let Latency = 4; }
104 def : WriteRes<WriteSys, []> { let Latency = 1; }
105 def : WriteRes<WriteBarrier, []> { let Latency = 1; }
106 def : WriteRes<WriteHint, []> { let Latency = 1; }
108 def : WriteRes<WriteLDHi, []> { let Latency = 4; }
110 def : WriteRes<WriteAtomic, []> { let Unsupported = 1; }
112 // No forwarding logic is modelled yet.
113 def : ReadAdvance<ReadI, 0>;
114 def : ReadAdvance<ReadISReg, 0>;
115 def : ReadAdvance<ReadIEReg, 0>;
116 def : ReadAdvance<ReadIM, 0>;
117 def : ReadAdvance<ReadIMA, 0>;
118 def : ReadAdvance<ReadID, 0>;
119 def : ReadAdvance<ReadExtrHi, 0>;
120 def : ReadAdvance<ReadAdrBase, 0>;
121 def : ReadAdvance<ReadVLD, 0>;
122 def : ReadAdvance<ReadST, 0>;
125 //===----------------------------------------------------------------------===//
126 // Specialize the coarse model by associating instruction groups with the
127 // subtarget-defined types. As the modeled is refined, this will override most
128 // of the above SchedWriteRes and SchedAlias mappings.
131 // -----------------------------------------------------------------------------
133 def : InstRW<[WriteI], (instrs COPY)>;
136 // Detailed Refinedments
137 // -----------------------------------------------------------------------------
138 include "AArch64SchedKryoDetails.td"
141 } // SchedModel = KryoModel