1 //==- ARMScheduleM4.td - Cortex-M4 Scheduling Definitions -*- 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 SchedRead/Write data for the ARM Cortex-M4 processor.
11 //===----------------------------------------------------------------------===//
13 def CortexM4Model : SchedMachineModel {
14 let IssueWidth = 1; // Only IT can be dual-issued, so assume single-issue
15 let MicroOpBufferSize = 0; // In-order
16 let LoadLatency = 2; // Latency when not pipelined, not pc-relative
17 let MispredictPenalty = 2; // Best case branch taken cost
18 let PostRAScheduler = 1;
20 let CompleteModel = 0;
24 // We model the entire cpu as a single pipeline with a BufferSize = 0 since
25 // Cortex-M4 is in-order.
27 def M4Unit : ProcResource<1> { let BufferSize = 0; }
30 let SchedModel = CortexM4Model in {
32 // Some definitions of latencies we apply to different instructions
34 class M4UnitL1<SchedWrite write> : WriteRes<write, [M4Unit]> { let Latency = 1; }
35 class M4UnitL2<SchedWrite write> : WriteRes<write, [M4Unit]> { let Latency = 2; }
36 class M4UnitL3<SchedWrite write> : WriteRes<write, [M4Unit]> { let Latency = 3; }
37 class M4UnitL14<SchedWrite write> : WriteRes<write, [M4Unit]> { let Latency = 14; }
38 def M4UnitL1_wr : SchedWriteRes<[M4Unit]> { let Latency = 1; }
39 def M4UnitL2_wr : SchedWriteRes<[M4Unit]> { let Latency = 2; }
40 class M4UnitL1I<dag instr> : InstRW<[M4UnitL1_wr], instr>;
41 class M4UnitL2I<dag instr> : InstRW<[M4UnitL2_wr], instr>;
44 // Loads, MAC's and DIV all get a higher latency of 2
45 def : M4UnitL2<WriteLd>;
46 def : M4UnitL2<WriteMAC32>;
47 def : M4UnitL2<WriteMAC64Hi>;
48 def : M4UnitL2<WriteMAC64Lo>;
49 def : M4UnitL2<WriteMAC16>;
50 def : M4UnitL2<WriteDIV>;
52 def : M4UnitL2I<(instregex "(t|t2)LDM")>;
55 // Stores we use a latency of 1 as they have no outputs
57 def : M4UnitL1<WriteST>;
58 def : M4UnitL1I<(instregex "(t|t2)STM")>;
61 // Everything else has a Latency of 1
63 def : M4UnitL1<WriteALU>;
64 def : M4UnitL1<WriteALUsi>;
65 def : M4UnitL1<WriteALUsr>;
66 def : M4UnitL1<WriteALUSsr>;
67 def : M4UnitL1<WriteBr>;
68 def : M4UnitL1<WriteBrL>;
69 def : M4UnitL1<WriteBrTbl>;
70 def : M4UnitL1<WriteCMPsi>;
71 def : M4UnitL1<WriteCMPsr>;
72 def : M4UnitL1<WriteCMP>;
73 def : M4UnitL1<WriteMUL32>;
74 def : M4UnitL1<WriteMUL64Hi>;
75 def : M4UnitL1<WriteMUL64Lo>;
76 def : M4UnitL1<WriteMUL16>;
77 def : M4UnitL1<WriteNoop>;
78 def : M4UnitL1<WritePreLd>;
79 def : M4UnitL1I<(instregex "(t|t2)MOV")>;
80 def : M4UnitL1I<(instrs COPY)>;
81 def : M4UnitL1I<(instregex "t2IT")>;
82 def : M4UnitL1I<(instregex "t2SEL", "t2USAD8",
83 "t2(S|Q|SH|U|UQ|UH)(ADD16|ASX|SAX|SUB16|ADD8|SUB8)", "t2USADA8", "(t|t2)REV")>;
85 def : ReadAdvance<ReadALU, 0>;
86 def : ReadAdvance<ReadALUsr, 0>;
87 def : ReadAdvance<ReadMUL, 0>;
88 def : ReadAdvance<ReadMAC, 0>;
90 // Most FP instructions are single-cycle latency, except MAC's, Div's and Sqrt's.
91 // Loads still take 2 cycles.
93 def : M4UnitL1<WriteFPCVT>;
94 def : M4UnitL1<WriteFPMOV>;
95 def : M4UnitL1<WriteFPALU32>;
96 def : M4UnitL1<WriteFPALU64>;
97 def : M4UnitL1<WriteFPMUL32>;
98 def : M4UnitL1<WriteFPMUL64>;
99 def : M4UnitL2I<(instregex "VLD")>;
100 def : M4UnitL1I<(instregex "VST")>;
101 def : M4UnitL3<WriteFPMAC32>;
102 def : M4UnitL3<WriteFPMAC64>;
103 def : M4UnitL14<WriteFPDIV32>;
104 def : M4UnitL14<WriteFPDIV64>;
105 def : M4UnitL14<WriteFPSQRT32>;
106 def : M4UnitL14<WriteFPSQRT64>;
107 def : M4UnitL1<WriteVLD1>;
108 def : M4UnitL1<WriteVLD2>;
109 def : M4UnitL1<WriteVLD3>;
110 def : M4UnitL1<WriteVLD4>;
111 def : M4UnitL1<WriteVST1>;
112 def : M4UnitL1<WriteVST2>;
113 def : M4UnitL1<WriteVST3>;
114 def : M4UnitL1<WriteVST4>;
116 def : ReadAdvance<ReadFPMUL, 0>;
117 def : ReadAdvance<ReadFPMAC, 0>;