1 //==- AArch64SchedFalkor.td - Falkor Scheduling Definitions -*- tablegen -*-==//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file defines the machine model for Qualcomm Falkor to support
11 // instruction scheduling and other instruction cost heuristics.
13 //===----------------------------------------------------------------------===//
15 //===----------------------------------------------------------------------===//
16 // Define the SchedMachineModel and provide basic properties for coarse grained
17 // instruction cost model.
19 def FalkorModel : SchedMachineModel {
20 let IssueWidth = 8; // 8 uops are dispatched per cycle.
21 let MicroOpBufferSize = 128; // Out-of-order with temporary unified issue buffer.
22 let LoopMicroOpBufferSize = 16;
23 let LoadLatency = 3; // Optimistic load latency.
24 let MispredictPenalty = 11; // Minimum branch misprediction penalty.
25 let CompleteModel = 1;
28 //===----------------------------------------------------------------------===//
29 // Define each kind of processor resource and number available on Falkor.
31 let SchedModel = FalkorModel in {
33 def FalkorUnitB : ProcResource<1>; // Branch
34 def FalkorUnitLD : ProcResource<1>; // Load pipe
35 def FalkorUnitSD : ProcResource<1>; // Store data
36 def FalkorUnitST : ProcResource<1>; // Store pipe
37 def FalkorUnitX : ProcResource<1>; // Complex arithmetic
38 def FalkorUnitY : ProcResource<1>; // Simple arithmetic
39 def FalkorUnitZ : ProcResource<1>; // Simple arithmetic
41 def FalkorUnitVSD : ProcResource<1>; // Vector store data
42 def FalkorUnitVX : ProcResource<1>; // Vector X-pipe
43 def FalkorUnitVY : ProcResource<1>; // Vector Y-pipe
45 def FalkorUnitGTOV : ProcResource<1>; // Scalar to Vector
46 def FalkorUnitVTOG : ProcResource<1>; // Vector to Scalar
48 // Define the resource groups.
49 def FalkorUnitXY : ProcResGroup<[FalkorUnitX, FalkorUnitY]>;
50 def FalkorUnitXYZ : ProcResGroup<[FalkorUnitX, FalkorUnitY, FalkorUnitZ]>;
51 def FalkorUnitXYZB : ProcResGroup<[FalkorUnitX, FalkorUnitY, FalkorUnitZ,
53 def FalkorUnitZB : ProcResGroup<[FalkorUnitZ, FalkorUnitB]>;
54 def FalkorUnitVXVY : ProcResGroup<[FalkorUnitVX, FalkorUnitVY]>;
58 //===----------------------------------------------------------------------===//
59 // Map the target-defined scheduler read/write resources and latency for
62 let SchedModel = FalkorModel in {
64 def : WriteRes<WriteImm, [FalkorUnitXYZ]> { let Latency = 1; }
65 def : WriteRes<WriteI, [FalkorUnitXYZ]> { let Latency = 1; }
66 def : WriteRes<WriteISReg, [FalkorUnitVXVY, FalkorUnitVXVY]>
67 { let Latency = 1; let NumMicroOps = 2; }
68 def : WriteRes<WriteIEReg, [FalkorUnitXYZ, FalkorUnitXYZ]>
69 { let Latency = 2; let NumMicroOps = 2; }
70 def : WriteRes<WriteExtr, [FalkorUnitXYZ, FalkorUnitXYZ]>
71 { let Latency = 2; let NumMicroOps = 2; }
72 def : WriteRes<WriteIS, [FalkorUnitXYZ]> { let Latency = 1; }
73 def : WriteRes<WriteID32, [FalkorUnitX, FalkorUnitZ]>
74 { let Latency = 8; let NumMicroOps = 2; }
75 def : WriteRes<WriteID64, [FalkorUnitX, FalkorUnitZ]>
76 { let Latency = 16; let NumMicroOps = 2; }
77 def : WriteRes<WriteIM32, [FalkorUnitX]> { let Latency = 4; }
78 def : WriteRes<WriteIM64, [FalkorUnitX]> { let Latency = 5; }
79 def : WriteRes<WriteBr, [FalkorUnitB]> { let Latency = 1; }
80 def : WriteRes<WriteBrReg, [FalkorUnitB]> { let Latency = 1; }
81 def : WriteRes<WriteLD, [FalkorUnitLD]> { let Latency = 3; }
82 def : WriteRes<WriteST, [FalkorUnitST, FalkorUnitSD]>
83 { let Latency = 0; let NumMicroOps = 2; }
84 def : WriteRes<WriteSTP, [FalkorUnitST, FalkorUnitSD]>
85 { let Latency = 0; let NumMicroOps = 2; }
86 def : WriteRes<WriteAdr, [FalkorUnitXYZ]> { let Latency = 1; }
87 def : WriteRes<WriteLDIdx, [FalkorUnitLD]> { let Latency = 5; }
88 def : WriteRes<WriteSTIdx, [FalkorUnitST, FalkorUnitSD]>
89 { let Latency = 0; let NumMicroOps = 2; }
90 def : WriteRes<WriteF, [FalkorUnitVXVY, FalkorUnitVXVY]>
91 { let Latency = 3; let NumMicroOps = 2; }
92 def : WriteRes<WriteFCmp, [FalkorUnitVXVY]> { let Latency = 2; }
93 def : WriteRes<WriteFCvt, [FalkorUnitVXVY]> { let Latency = 4; }
94 def : WriteRes<WriteFCopy, [FalkorUnitVXVY]> { let Latency = 4; }
95 def : WriteRes<WriteFImm, [FalkorUnitVXVY]> { let Latency = 4; }
96 def : WriteRes<WriteFMul, [FalkorUnitVXVY, FalkorUnitVXVY]>
97 { let Latency = 6; let NumMicroOps = 2; }
98 def : WriteRes<WriteFDiv, [FalkorUnitVXVY, FalkorUnitVXVY]>
99 { let Latency = 12; let NumMicroOps = 2; } // Fragent -1 / NoRSV +1
100 def : WriteRes<WriteV, [FalkorUnitVXVY]> { let Latency = 6; }
101 def : WriteRes<WriteVLD, [FalkorUnitLD]> { let Latency = 3; }
102 def : WriteRes<WriteVST, [FalkorUnitST, FalkorUnitVSD]>
103 { let Latency = 0; let NumMicroOps = 2; }
105 def : WriteRes<WriteSys, []> { let Latency = 1; }
106 def : WriteRes<WriteBarrier, []> { let Latency = 1; }
107 def : WriteRes<WriteHint, []> { let Latency = 1; }
109 def : WriteRes<WriteLDHi, []> { let Latency = 3; }
111 def : WriteRes<WriteAtomic, []> { let Unsupported = 1; }
113 // No forwarding logic is modelled yet.
114 def : ReadAdvance<ReadI, 0>;
115 def : ReadAdvance<ReadISReg, 0>;
116 def : ReadAdvance<ReadIEReg, 0>;
117 def : ReadAdvance<ReadIM, 0>;
118 def : ReadAdvance<ReadIMA, 0>;
119 def : ReadAdvance<ReadID, 0>;
120 def : ReadAdvance<ReadExtrHi, 0>;
121 def : ReadAdvance<ReadAdrBase, 0>;
122 def : ReadAdvance<ReadVLD, 0>;
124 // Detailed Refinements
125 // -----------------------------------------------------------------------------
126 include "AArch64SchedFalkorDetails.td"