InstrItinData<IIC_iMOVsr, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [1, 1, 1]>,
InstrItinData<IIC_iMOVix2,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2]>,
+ InstrItinData<IIC_iMOVix2addpc,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
+ InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
+ InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [3]>,
+ InstrItinData<IIC_iMOVix2ld,[InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
+ InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
+ InstrStage<1, [A8_LSPipe]>], [5]>,
//
// Move instructions, conditional
InstrItinData<IIC_iCMOVi , [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2]>,
// Load multiple, def is the 5th operand. Pipeline 0 only.
// FIXME: A8_LSPipe cycle time is dynamic, this assumes 3 to 4 registers.
InstrItinData<IIC_iLoad_m , [InstrStage<2, [A8_Pipe0], 0>,
- InstrStage<2, [A8_LSPipe]>], [1, 1, 1, 1, 3]>,
+ InstrStage<2, [A8_LSPipe]>],
+ [1, 1, 1, 1, 3], [], -1>, // dynamic uops
//
// Load multiple + update, defs are the 1st and 5th operands.
InstrItinData<IIC_iLoad_mu , [InstrStage<3, [A8_Pipe0], 0>,
- InstrStage<3, [A8_LSPipe]>], [2, 1, 1, 1, 3]>,
+ InstrStage<3, [A8_LSPipe]>],
+ [2, 1, 1, 1, 3], [], -1>, // dynamic uops
//
// Load multiple plus branch
InstrItinData<IIC_iLoad_mBr, [InstrStage<3, [A8_Pipe0], 0>,
InstrStage<3, [A8_LSPipe]>,
InstrStage<1, [A8_Pipe0, A8_Pipe1]>],
- [1, 2, 1, 1, 3]>,
+ [1, 2, 1, 1, 3], [], -1>, // dynamic uops
//
// Pop, def is the 3rd operand.
InstrItinData<IIC_iPop , [InstrStage<3, [A8_Pipe0], 0>,
- InstrStage<3, [A8_LSPipe]>], [1, 1, 3]>,
+ InstrStage<3, [A8_LSPipe]>],
+ [1, 1, 3], [], -1>, // dynamic uops
//
// Push, def is the 3th operand.
InstrItinData<IIC_iPop_Br, [InstrStage<3, [A8_Pipe0], 0>,
InstrStage<3, [A8_LSPipe]>,
InstrStage<1, [A8_Pipe0, A8_Pipe1]>],
- [1, 1, 3]>,
-
+ [1, 1, 3], [], -1>, // dynamic uops
//
// iLoadi + iALUr for t2LDRpci_pic.
InstrItinData<IIC_iLoadiALU, [InstrStage<1, [A8_Pipe0, A8_Pipe1], 0>,
// Store multiple. Pipeline 0 only.
// FIXME: A8_LSPipe cycle time is dynamic, this assumes 3 to 4 registers.
InstrItinData<IIC_iStore_m , [InstrStage<2, [A8_Pipe0], 0>,
- InstrStage<2, [A8_LSPipe]>]>,
+ InstrStage<2, [A8_LSPipe]>],
+ [], [], -1>, // dynamic uops
//
// Store multiple + update
InstrItinData<IIC_iStore_mu, [InstrStage<2, [A8_Pipe0], 0>,
- InstrStage<2, [A8_LSPipe]>], [2]>,
-
+ InstrStage<2, [A8_LSPipe]>],
+ [2], [], -1>, // dynamic uops
//
// Preload
InstrItinData<IIC_Preload, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>], [2, 2]>,
InstrStage<19, [A8_NPipe], 0>,
InstrStage<19, [A8_NLSPipe]>], [19, 2, 1, 1]>,
//
+ // Single-precision Fused FP MAC
+ InstrItinData<IIC_fpFMAC32, [InstrStage<1, [A8_Pipe0, A8_Pipe1], 0>,
+ InstrStage<1, [A8_NPipe]>], [7, 2, 1, 1]>,
+ //
+ // Double-precision Fused FP MAC
+ InstrItinData<IIC_fpFMAC64, [InstrStage<1, [A8_Pipe0, A8_Pipe1], 0>,
+ InstrStage<19, [A8_NPipe], 0>,
+ InstrStage<19, [A8_NLSPipe]>], [19, 2, 1, 1]>,
+ //
// Single-precision FP DIV
InstrItinData<IIC_fpDIV32 , [InstrStage<1, [A8_Pipe0, A8_Pipe1], 0>,
InstrStage<20, [A8_NPipe], 0>,
InstrStage<1, [A8_NLSPipe], 0>,
InstrStage<1, [A8_LSPipe]>,
InstrStage<1, [A8_NLSPipe], 0>,
- InstrStage<1, [A8_LSPipe]>], [1, 1, 1, 2]>,
+ InstrStage<1, [A8_LSPipe]>],
+ [1, 1, 1, 2], [], -1>, // dynamic uops
//
// FP Load Multiple + update
InstrItinData<IIC_fpLoad_mu,[InstrStage<1, [A8_Pipe0, A8_Pipe1], 0>,
InstrStage<1, [A8_NLSPipe], 0>,
InstrStage<1, [A8_LSPipe]>,
InstrStage<1, [A8_NLSPipe], 0>,
- InstrStage<1, [A8_LSPipe]>], [2, 1, 1, 1, 2]>,
+ InstrStage<1, [A8_LSPipe]>],
+ [2, 1, 1, 1, 2], [], -1>, // dynamic uops
//
// Single-precision FP Store
InstrItinData<IIC_fpStore32,[InstrStage<1, [A8_Pipe0, A8_Pipe1], 0>,
InstrStage<1, [A8_NLSPipe], 0>,
InstrStage<1, [A8_LSPipe]>,
InstrStage<1, [A8_NLSPipe], 0>,
- InstrStage<1, [A8_LSPipe]>], [1, 1, 1, 1]>,
+ InstrStage<1, [A8_LSPipe]>],
+ [1, 1, 1, 1], [], -1>, // dynamic uops
//
// FP Store Multiple + update
InstrItinData<IIC_fpStore_mu,[InstrStage<1, [A8_Pipe0, A8_Pipe1], 0>,
InstrStage<1, [A8_NLSPipe], 0>,
InstrStage<1, [A8_LSPipe]>,
InstrStage<1, [A8_NLSPipe], 0>,
- InstrStage<1, [A8_LSPipe]>], [2, 1, 1, 1, 1]>,
-
+ InstrStage<1, [A8_LSPipe]>],
+ [2, 1, 1, 1, 1], [], -1>, // dynamic uops
// NEON
// Issue through integer pipeline, and execute in NEON unit.
//
//
// VLD1ln
InstrItinData<IIC_VLD1ln, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
- InstrStage<3, [A8_NLSPipe], 1>,
+ InstrStage<3, [A8_NLSPipe], 0>,
InstrStage<3, [A8_LSPipe]>],
[3, 1, 1, 1]>,
//
// VLD1lnu
InstrItinData<IIC_VLD1lnu, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
- InstrStage<3, [A8_NLSPipe], 1>,
+ InstrStage<3, [A8_NLSPipe], 0>,
InstrStage<3, [A8_LSPipe]>],
[3, 2, 1, 1, 1, 1]>,
//
+ // VLD1dup
+ InstrItinData<IIC_VLD1dup, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
+ InstrStage<2, [A8_NLSPipe], 0>,
+ InstrStage<2, [A8_LSPipe]>],
+ [2, 1]>,
+ //
+ // VLD1dupu
+ InstrItinData<IIC_VLD1dupu, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
+ InstrStage<2, [A8_NLSPipe], 0>,
+ InstrStage<2, [A8_LSPipe]>],
+ [2, 2, 1, 1]>,
+ //
// VLD2
InstrItinData<IIC_VLD2, [InstrStage<1, [A8_Pipe0, A8_Pipe1], 0>,
InstrStage<2, [A8_NLSPipe], 0>,
InstrStage<3, [A8_LSPipe]>],
[3, 3, 2, 1, 1, 1, 1, 1]>,
//
+ // VLD2dup
+ InstrItinData<IIC_VLD2dup, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
+ InstrStage<2, [A8_NLSPipe], 0>,
+ InstrStage<2, [A8_LSPipe]>],
+ [2, 2, 1]>,
+ //
+ // VLD2dupu
+ InstrItinData<IIC_VLD2dupu, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
+ InstrStage<2, [A8_NLSPipe], 0>,
+ InstrStage<2, [A8_LSPipe]>],
+ [2, 2, 2, 1, 1]>,
+ //
// VLD3
InstrItinData<IIC_VLD3, [InstrStage<1, [A8_Pipe0, A8_Pipe1], 0>,
InstrStage<4, [A8_NLSPipe], 0>,
InstrStage<5, [A8_LSPipe]>],
[4, 4, 5, 2, 1, 1, 1, 1, 1, 2]>,
//
+ // VLD3dup
+ InstrItinData<IIC_VLD3dup, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
+ InstrStage<3, [A8_NLSPipe], 0>,
+ InstrStage<3, [A8_LSPipe]>],
+ [2, 2, 3, 1]>,
+ //
+ // VLD3dupu
+ InstrItinData<IIC_VLD3dupu, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
+ InstrStage<3, [A8_NLSPipe], 0>,
+ InstrStage<3, [A8_LSPipe]>],
+ [2, 2, 3, 2, 1, 1]>,
+ //
// VLD4
InstrItinData<IIC_VLD4, [InstrStage<1, [A8_Pipe0, A8_Pipe1], 0>,
InstrStage<4, [A8_NLSPipe], 0>,
InstrStage<5, [A8_LSPipe]>],
[4, 4, 5, 5, 2, 1, 1, 1, 1, 1, 2, 2]>,
//
+ // VLD4dup
+ InstrItinData<IIC_VLD4dup, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
+ InstrStage<3, [A8_NLSPipe], 0>,
+ InstrStage<3, [A8_LSPipe]>],
+ [2, 2, 3, 3, 1]>,
+ //
+ // VLD4dupu
+ InstrItinData<IIC_VLD4dupu, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
+ InstrStage<3, [A8_NLSPipe], 0>,
+ InstrStage<3, [A8_LSPipe]>],
+ [2, 2, 3, 3, 2, 1, 1]>,
+ //
// VST1
InstrItinData<IIC_VST1, [InstrStage<1, [A8_Pipe0, A8_Pipe1], 0>,
InstrStage<2, [A8_NLSPipe], 0>,
//
// VST1ln
InstrItinData<IIC_VST1ln, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
- InstrStage<2, [A8_NLSPipe], 1>,
+ InstrStage<2, [A8_NLSPipe], 0>,
InstrStage<2, [A8_LSPipe]>],
[1, 1, 1]>,
//
// VST1lnu
InstrItinData<IIC_VST1lnu, [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
- InstrStage<2, [A8_NLSPipe], 1>,
+ InstrStage<2, [A8_NLSPipe], 0>,
InstrStage<2, [A8_LSPipe]>],
[2, 1, 1, 1, 1]>,
//
InstrItinData<IIC_VMACQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1], 0>,
InstrStage<2, [A8_NPipe]>], [10, 3, 2, 2]>,
//
+ // Double-register Fused FP Multiple-Accumulate
+ InstrItinData<IIC_VFMACD, [InstrStage<1, [A8_Pipe0, A8_Pipe1], 0>,
+ InstrStage<1, [A8_NPipe]>], [9, 3, 2, 2]>,
+ //
+ // Quad-register Fused FP Multiple-Accumulate
+ // Result written in N9, but that is relative to the last cycle of multicycle,
+ // so we use 10 for those cases
+ InstrItinData<IIC_VFMACQ, [InstrStage<1, [A8_Pipe0, A8_Pipe1], 0>,
+ InstrStage<2, [A8_NPipe]>], [10, 3, 2, 2]>,
+ //
// Double-register Reciprical Step
InstrItinData<IIC_VRECSD, [InstrStage<1, [A8_Pipe0, A8_Pipe1], 0>,
InstrStage<1, [A8_NPipe]>], [9, 2, 2]>,
InstrStage<1, [A8_NPipe], 0>,
InstrStage<2, [A8_NLSPipe]>], [4, 1, 2, 2, 3, 3, 1]>
]>;
+
+// ===---------------------------------------------------------------------===//
+// This following definitions describe the simple machine model which
+// will replace itineraries.
+
+// Cortex-A8 machine model for scheduling and other instruction cost heuristics.
+def CortexA8Model : SchedMachineModel {
+ let IssueWidth = 2; // 2 micro-ops are dispatched per cycle.
+ let MinLatency = -1; // OperandCycles are interpreted as MinLatency.
+ let LoadLatency = 2; // Optimistic load latency assuming bypass.
+ // This is overriden by OperandCycles if the
+ // Itineraries are queried instead.
+ let MispredictPenalty = 13; // Based on estimate of pipeline depth.
+
+ let Itineraries = CortexA8Itineraries;
+}
projects / oota-llvm.git / blobdiff