llvm-6502/lib/Target/ARM/ARMScheduleA9.td
Andrew Trick 573931394f MI-Sched: handle latency of in-order operations with the new machine model.
The per-operand machine model allows the target to define "unbuffered"
processor resources. This change is a quick, cheap way to model stalls
caused by the latency of operations that use such resources. This only
applies when the processor's micro-op buffer size is non-zero
(Out-of-Order). We can't precisely model in-order stalls during
out-of-order execution, but this is an easy and effective
heuristic. It benefits cortex-a9 scheduling when using the new
machine model, which is not yet on by default.

MI-Sched for armv7 was evaluated on Swift (and only not enabled because
of a performance bug related to predication). However, we never
evaluated Cortex-A9 performance on MI-Sched in its current form. This
change adds MI-Sched functionality to reach performance goals on
A9. The only remaining change is to allow MI-Sched to run as a PostRA
pass.

I evaluated performance using a set of options to estimate the performance impact once MI sched is default on armv7:
-mcpu=cortex-a9 -disable-post-ra -misched-bench -scheditins=false

For a simple saxpy loop I see a 1.7x speedup. Here are the llvm-testsuite results:
(min run time over 2 runs, filtering tiny changes)

Speedups:
| Benchmarks/BenchmarkGame/recursive         |  52.39% |
| Benchmarks/VersaBench/beamformer           |  20.80% |
| Benchmarks/Misc/pi                         |  19.97% |
| Benchmarks/Misc/mandel-2                   |  19.95% |
| SPEC/CFP2000/188.ammp                      |  18.72% |
| Benchmarks/McCat/08-main/main              |  18.58% |
| Benchmarks/Misc-C++/Large/sphereflake      |  18.46% |
| Benchmarks/Olden/power                     |  17.11% |
| Benchmarks/Misc-C++/mandel-text            |  16.47% |
| Benchmarks/Misc/oourafft                   |  15.94% |
| Benchmarks/Misc/flops-7                    |  14.99% |
| Benchmarks/FreeBench/distray               |  14.26% |
| SPEC/CFP2006/470.lbm                       |  14.00% |
| mediabench/mpeg2/mpeg2dec/mpeg2decode      |  12.28% |
| Benchmarks/SmallPT/smallpt                 |  10.36% |
| Benchmarks/Misc-C++/Large/ray              |   8.97% |
| Benchmarks/Misc/fp-convert                 |   8.75% |
| Benchmarks/Olden/perimeter                 |   7.10% |
| Benchmarks/Bullet/bullet                   |   7.03% |
| Benchmarks/Misc/mandel                     |   6.75% |
| Benchmarks/Olden/voronoi                   |   6.26% |
| Benchmarks/Misc/flops-8                    |   5.77% |
| Benchmarks/Misc/matmul_f64_4x4             |   5.19% |
| Benchmarks/MiBench/security-rijndael       |   5.15% |
| Benchmarks/Misc/flops-6                    |   5.10% |
| Benchmarks/Olden/tsp                       |   4.46% |
| Benchmarks/MiBench/consumer-lame           |   4.28% |
| Benchmarks/Misc/flops-5                    |   4.27% |
| Benchmarks/mafft/pairlocalalign            |   4.19% |
| Benchmarks/Misc/himenobmtxpa               |   4.07% |
| Benchmarks/Misc/lowercase                  |   4.06% |
| SPEC/CFP2006/433.milc                      |   3.99% |
| Benchmarks/tramp3d-v4                      |   3.79% |
| Benchmarks/FreeBench/pifft                 |   3.66% |
| Benchmarks/Ptrdist/ks                      |   3.21% |
| Benchmarks/Adobe-C++/loop_unroll           |   3.12% |
| SPEC/CINT2000/175.vpr                      |   3.12% |
| Benchmarks/nbench                          |   2.98% |
| SPEC/CFP2000/183.equake                    |   2.91% |
| Benchmarks/Misc/perlin                     |   2.85% |
| Benchmarks/Misc/flops-1                    |   2.82% |
| Benchmarks/Misc-C++-EH/spirit              |   2.80% |
| Benchmarks/Misc/flops-2                    |   2.77% |
| Benchmarks/NPB-serial/is                   |   2.42% |
| Benchmarks/ASC_Sequoia/CrystalMk           |   2.33% |
| Benchmarks/BenchmarkGame/n-body            |   2.28% |
| Benchmarks/SciMark2-C/scimark2             |   2.27% |
| Benchmarks/Olden/bh                        |   2.03% |
| skidmarks10/skidmarks                      |   1.81% |
| Benchmarks/Misc/flops                      |   1.72% |

Slowdowns:
| Benchmarks/llubenchmark/llu                | -14.14% |
| Benchmarks/Polybench/stencils/seidel-2d    |  -5.67% |
| Benchmarks/Adobe-C++/functionobjects       |  -5.25% |
| Benchmarks/Misc-C++/oopack_v1p8            |  -5.00% |
| Benchmarks/Shootout/hash                   |  -2.35% |
| Benchmarks/Prolangs-C++/ocean              |  -2.01% |
| Benchmarks/Polybench/medley/floyd-warshall |  -1.98% |
| Polybench/linear-algebra/kernels/3mm       |  -1.95% |
| Benchmarks/McCat/09-vor/vor                |  -1.68% |

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@196516 91177308-0d34-0410-b5e6-96231b3b80d8
2013-12-05 17:55:58 +00:00

2519 lines
128 KiB
TableGen

//=- ARMScheduleA9.td - ARM Cortex-A9 Scheduling Definitions -*- tablegen -*-=//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file defines the itinerary class data for the ARM Cortex A9 processors.
//
//===----------------------------------------------------------------------===//
// ===---------------------------------------------------------------------===//
// This section contains legacy support for itineraries. This is
// required until SD and PostRA schedulers are replaced by MachineScheduler.
//
// Ad-hoc scheduling information derived from pretty vague "Cortex-A9 Technical
// Reference Manual".
//
// Functional units
def A9_Issue0 : FuncUnit; // Issue 0
def A9_Issue1 : FuncUnit; // Issue 1
def A9_Branch : FuncUnit; // Branch
def A9_ALU0 : FuncUnit; // ALU / MUL pipeline 0
def A9_ALU1 : FuncUnit; // ALU pipeline 1
def A9_AGU : FuncUnit; // Address generation unit for ld / st
def A9_NPipe : FuncUnit; // NEON pipeline
def A9_MUX0 : FuncUnit; // AGU + NEON/FPU multiplexer
def A9_LSUnit : FuncUnit; // L/S Unit
def A9_DRegsVFP: FuncUnit; // FP register set, VFP side
def A9_DRegsN : FuncUnit; // FP register set, NEON side
// Bypasses
def A9_LdBypass : Bypass;
def CortexA9Itineraries : ProcessorItineraries<
[A9_Issue0, A9_Issue1, A9_Branch, A9_ALU0, A9_ALU1, A9_AGU, A9_NPipe, A9_MUX0,
A9_LSUnit, A9_DRegsVFP, A9_DRegsN],
[A9_LdBypass], [
// Two fully-pipelined integer ALU pipelines
//
// Move instructions, unconditional
InstrItinData<IIC_iMOVi , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_ALU0, A9_ALU1]>], [1]>,
InstrItinData<IIC_iMOVr , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_ALU0, A9_ALU1]>], [1, 1]>,
InstrItinData<IIC_iMOVsi , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_ALU0, A9_ALU1]>], [1, 1]>,
InstrItinData<IIC_iMOVsr , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<2, [A9_ALU0, A9_ALU1]>], [2, 1, 1]>,
InstrItinData<IIC_iMOVix2 , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_ALU0, A9_ALU1]>,
InstrStage<1, [A9_ALU0, A9_ALU1]>], [2]>,
InstrItinData<IIC_iMOVix2addpc,[InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_ALU0, A9_ALU1]>,
InstrStage<1, [A9_ALU0, A9_ALU1]>,
InstrStage<1, [A9_ALU0, A9_ALU1]>], [3]>,
InstrItinData<IIC_iMOVix2ld,[InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_ALU0, A9_ALU1]>,
InstrStage<1, [A9_ALU0, A9_ALU1]>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_AGU], 0>,
InstrStage<1, [A9_LSUnit]>], [5]>,
//
// MVN instructions
InstrItinData<IIC_iMVNi , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_ALU0, A9_ALU1]>],
[1]>,
InstrItinData<IIC_iMVNr , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_ALU0, A9_ALU1]>],
[1, 1], [NoBypass, A9_LdBypass]>,
InstrItinData<IIC_iMVNsi , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<2, [A9_ALU0, A9_ALU1]>],
[2, 1]>,
InstrItinData<IIC_iMVNsr , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<3, [A9_ALU0, A9_ALU1]>],
[3, 1, 1]>,
//
// No operand cycles
InstrItinData<IIC_iALUx , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_ALU0, A9_ALU1]>]>,
//
// Binary Instructions that produce a result
InstrItinData<IIC_iALUi , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_ALU0, A9_ALU1]>],
[1, 1], [NoBypass, A9_LdBypass]>,
InstrItinData<IIC_iALUr , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_ALU0, A9_ALU1]>],
[1, 1, 1], [NoBypass, A9_LdBypass, A9_LdBypass]>,
InstrItinData<IIC_iALUsi, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<2, [A9_ALU0, A9_ALU1]>],
[2, 1, 1], [NoBypass, A9_LdBypass, NoBypass]>,
InstrItinData<IIC_iALUsir,[InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<2, [A9_ALU0, A9_ALU1]>],
[2, 1, 1], [NoBypass, NoBypass, A9_LdBypass]>,
InstrItinData<IIC_iALUsr, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<3, [A9_ALU0, A9_ALU1]>],
[3, 1, 1, 1],
[NoBypass, A9_LdBypass, NoBypass, NoBypass]>,
//
// Bitwise Instructions that produce a result
InstrItinData<IIC_iBITi , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_ALU0, A9_ALU1]>], [1, 1]>,
InstrItinData<IIC_iBITr , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_ALU0, A9_ALU1]>], [1, 1, 1]>,
InstrItinData<IIC_iBITsi, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<2, [A9_ALU0, A9_ALU1]>], [2, 1, 1]>,
InstrItinData<IIC_iBITsr, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<3, [A9_ALU0, A9_ALU1]>], [3, 1, 1, 1]>,
//
// Unary Instructions that produce a result
// CLZ, RBIT, etc.
InstrItinData<IIC_iUNAr , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_ALU0, A9_ALU1]>], [1, 1]>,
// BFC, BFI, UBFX, SBFX
InstrItinData<IIC_iUNAsi, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<2, [A9_ALU0, A9_ALU1]>], [2, 1]>,
//
// Zero and sign extension instructions
InstrItinData<IIC_iEXTr , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_ALU0, A9_ALU1]>], [2, 1]>,
InstrItinData<IIC_iEXTAr, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<2, [A9_ALU0, A9_ALU1]>], [3, 1, 1]>,
InstrItinData<IIC_iEXTAsr,[InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<3, [A9_ALU0, A9_ALU1]>], [3, 1, 1, 1]>,
//
// Compare instructions
InstrItinData<IIC_iCMPi , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_ALU0, A9_ALU1]>],
[1], [A9_LdBypass]>,
InstrItinData<IIC_iCMPr , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_ALU0, A9_ALU1]>],
[1, 1], [A9_LdBypass, A9_LdBypass]>,
InstrItinData<IIC_iCMPsi , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<2, [A9_ALU0, A9_ALU1]>],
[1, 1], [A9_LdBypass, NoBypass]>,
InstrItinData<IIC_iCMPsr , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<3, [A9_ALU0, A9_ALU1]>],
[1, 1, 1], [A9_LdBypass, NoBypass, NoBypass]>,
//
// Test instructions
InstrItinData<IIC_iTSTi , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_ALU0, A9_ALU1]>], [1]>,
InstrItinData<IIC_iTSTr , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_ALU0, A9_ALU1]>], [1, 1]>,
InstrItinData<IIC_iTSTsi , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<2, [A9_ALU0, A9_ALU1]>], [1, 1]>,
InstrItinData<IIC_iTSTsr , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<3, [A9_ALU0, A9_ALU1]>], [1, 1, 1]>,
//
// Move instructions, conditional
// FIXME: Correctly model the extra input dep on the destination.
InstrItinData<IIC_iCMOVi , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_ALU0, A9_ALU1]>], [1]>,
InstrItinData<IIC_iCMOVr , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_ALU0, A9_ALU1]>], [1, 1]>,
InstrItinData<IIC_iCMOVsi , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_ALU0, A9_ALU1]>], [1, 1]>,
InstrItinData<IIC_iCMOVsr , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<2, [A9_ALU0, A9_ALU1]>], [2, 1, 1]>,
InstrItinData<IIC_iCMOVix2, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_ALU0, A9_ALU1]>,
InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_ALU0, A9_ALU1]>], [2]>,
// Integer multiply pipeline
//
InstrItinData<IIC_iMUL16 , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<2, [A9_ALU0]>], [3, 1, 1]>,
InstrItinData<IIC_iMAC16 , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<2, [A9_ALU0]>],
[3, 1, 1, 1]>,
InstrItinData<IIC_iMUL32 , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<2, [A9_ALU0]>], [4, 1, 1]>,
InstrItinData<IIC_iMAC32 , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<2, [A9_ALU0]>],
[4, 1, 1, 1]>,
InstrItinData<IIC_iMUL64 , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<3, [A9_ALU0]>], [4, 5, 1, 1]>,
InstrItinData<IIC_iMAC64 , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<3, [A9_ALU0]>],
[4, 5, 1, 1]>,
// Integer load pipeline
// FIXME: The timings are some rough approximations
//
// Immediate offset
InstrItinData<IIC_iLoad_i , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_AGU], 0>,
InstrStage<1, [A9_LSUnit]>],
[3, 1], [A9_LdBypass]>,
InstrItinData<IIC_iLoad_bh_i, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<2, [A9_AGU], 0>,
InstrStage<1, [A9_LSUnit]>],
[4, 1], [A9_LdBypass]>,
// FIXME: If address is 64-bit aligned, AGU cycles is 1.
InstrItinData<IIC_iLoad_d_i , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<2, [A9_AGU], 0>,
InstrStage<1, [A9_LSUnit]>],
[3, 3, 1], [A9_LdBypass]>,
//
// Register offset
InstrItinData<IIC_iLoad_r , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_AGU], 0>,
InstrStage<1, [A9_LSUnit]>],
[3, 1, 1], [A9_LdBypass]>,
InstrItinData<IIC_iLoad_bh_r, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<2, [A9_AGU], 0>,
InstrStage<1, [A9_LSUnit]>],
[4, 1, 1], [A9_LdBypass]>,
InstrItinData<IIC_iLoad_d_r , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<2, [A9_AGU], 0>,
InstrStage<1, [A9_LSUnit]>],
[3, 3, 1, 1], [A9_LdBypass]>,
//
// Scaled register offset
InstrItinData<IIC_iLoad_si , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_AGU], 0>,
InstrStage<1, [A9_LSUnit], 0>],
[4, 1, 1], [A9_LdBypass]>,
InstrItinData<IIC_iLoad_bh_si,[InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<2, [A9_AGU], 0>,
InstrStage<1, [A9_LSUnit]>],
[5, 1, 1], [A9_LdBypass]>,
//
// Immediate offset with update
InstrItinData<IIC_iLoad_iu , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_AGU], 0>,
InstrStage<1, [A9_LSUnit]>],
[3, 2, 1], [A9_LdBypass]>,
InstrItinData<IIC_iLoad_bh_iu,[InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<2, [A9_AGU], 0>,
InstrStage<1, [A9_LSUnit]>],
[4, 3, 1], [A9_LdBypass]>,
//
// Register offset with update
InstrItinData<IIC_iLoad_ru , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_AGU], 0>,
InstrStage<1, [A9_LSUnit]>],
[3, 2, 1, 1], [A9_LdBypass]>,
InstrItinData<IIC_iLoad_bh_ru,[InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<2, [A9_AGU], 0>,
InstrStage<1, [A9_LSUnit]>],
[4, 3, 1, 1], [A9_LdBypass]>,
InstrItinData<IIC_iLoad_d_ru, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<2, [A9_AGU], 0>,
InstrStage<1, [A9_LSUnit]>],
[3, 3, 1, 1], [A9_LdBypass]>,
//
// Scaled register offset with update
InstrItinData<IIC_iLoad_siu , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_AGU], 0>,
InstrStage<1, [A9_LSUnit]>],
[4, 3, 1, 1], [A9_LdBypass]>,
InstrItinData<IIC_iLoad_bh_siu,[InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<2, [A9_AGU], 0>,
InstrStage<1, [A9_LSUnit]>],
[5, 4, 1, 1], [A9_LdBypass]>,
//
// Load multiple, def is the 5th operand.
// FIXME: This assumes 3 to 4 registers.
InstrItinData<IIC_iLoad_m , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<2, [A9_AGU], 1>,
InstrStage<2, [A9_LSUnit]>],
[1, 1, 1, 1, 3],
[NoBypass, NoBypass, NoBypass, NoBypass, A9_LdBypass],
-1>, // dynamic uops
//
// Load multiple + update, defs are the 1st and 5th operands.
InstrItinData<IIC_iLoad_mu , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<2, [A9_AGU], 1>,
InstrStage<2, [A9_LSUnit]>],
[2, 1, 1, 1, 3],
[NoBypass, NoBypass, NoBypass, NoBypass, A9_LdBypass],
-1>, // dynamic uops
//
// Load multiple plus branch
InstrItinData<IIC_iLoad_mBr, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_AGU], 1>,
InstrStage<2, [A9_LSUnit]>,
InstrStage<1, [A9_Branch]>],
[1, 2, 1, 1, 3],
[NoBypass, NoBypass, NoBypass, NoBypass, A9_LdBypass],
-1>, // dynamic uops
//
// Pop, def is the 3rd operand.
InstrItinData<IIC_iPop , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<2, [A9_AGU], 1>,
InstrStage<2, [A9_LSUnit]>],
[1, 1, 3],
[NoBypass, NoBypass, A9_LdBypass],
-1>, // dynamic uops
//
// Pop + branch, def is the 3rd operand.
InstrItinData<IIC_iPop_Br, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<2, [A9_AGU], 1>,
InstrStage<2, [A9_LSUnit]>,
InstrStage<1, [A9_Branch]>],
[1, 1, 3],
[NoBypass, NoBypass, A9_LdBypass],
-1>, // dynamic uops
//
// iLoadi + iALUr for t2LDRpci_pic.
InstrItinData<IIC_iLoadiALU, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_AGU], 0>,
InstrStage<1, [A9_LSUnit]>,
InstrStage<1, [A9_ALU0, A9_ALU1]>],
[2, 1]>,
// Integer store pipeline
///
// Immediate offset
InstrItinData<IIC_iStore_i , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_AGU], 0>,
InstrStage<1, [A9_LSUnit]>], [1, 1]>,
InstrItinData<IIC_iStore_bh_i,[InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<2, [A9_AGU], 1>,
InstrStage<1, [A9_LSUnit]>], [1, 1]>,
// FIXME: If address is 64-bit aligned, AGU cycles is 1.
InstrItinData<IIC_iStore_d_i, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<2, [A9_AGU], 1>,
InstrStage<1, [A9_LSUnit]>], [1, 1]>,
//
// Register offset
InstrItinData<IIC_iStore_r , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_AGU], 0>,
InstrStage<1, [A9_LSUnit]>], [1, 1, 1]>,
InstrItinData<IIC_iStore_bh_r,[InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<2, [A9_AGU], 1>,
InstrStage<1, [A9_LSUnit]>], [1, 1, 1]>,
InstrItinData<IIC_iStore_d_r, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<2, [A9_AGU], 1>,
InstrStage<1, [A9_LSUnit]>], [1, 1, 1]>,
//
// Scaled register offset
InstrItinData<IIC_iStore_si , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_AGU], 0>,
InstrStage<1, [A9_LSUnit]>], [1, 1, 1]>,
InstrItinData<IIC_iStore_bh_si,[InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<2, [A9_AGU], 1>,
InstrStage<1, [A9_LSUnit]>], [1, 1, 1]>,
//
// Immediate offset with update
InstrItinData<IIC_iStore_iu , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_AGU], 0>,
InstrStage<1, [A9_LSUnit]>], [2, 1, 1]>,
InstrItinData<IIC_iStore_bh_iu,[InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<2, [A9_AGU], 1>,
InstrStage<1, [A9_LSUnit]>], [3, 1, 1]>,
//
// Register offset with update
InstrItinData<IIC_iStore_ru , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_AGU], 0>,
InstrStage<1, [A9_LSUnit]>],
[2, 1, 1, 1]>,
InstrItinData<IIC_iStore_bh_ru,[InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<2, [A9_AGU], 1>,
InstrStage<1, [A9_LSUnit]>],
[3, 1, 1, 1]>,
InstrItinData<IIC_iStore_d_ru, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<2, [A9_AGU], 1>,
InstrStage<1, [A9_LSUnit]>],
[3, 1, 1, 1]>,
//
// Scaled register offset with update
InstrItinData<IIC_iStore_siu, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_AGU], 0>,
InstrStage<1, [A9_LSUnit]>],
[2, 1, 1, 1]>,
InstrItinData<IIC_iStore_bh_siu, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<2, [A9_AGU], 1>,
InstrStage<1, [A9_LSUnit]>],
[3, 1, 1, 1]>,
//
// Store multiple
InstrItinData<IIC_iStore_m , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_AGU], 0>,
InstrStage<2, [A9_LSUnit]>],
[], [], -1>, // dynamic uops
//
// Store multiple + update
InstrItinData<IIC_iStore_mu, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_AGU], 0>,
InstrStage<2, [A9_LSUnit]>],
[2], [], -1>, // dynamic uops
//
// Preload
InstrItinData<IIC_Preload, [InstrStage<1, [A9_Issue0, A9_Issue1]>], [1, 1]>,
// Branch
//
// no delay slots, so the latency of a branch is unimportant
InstrItinData<IIC_Br , [InstrStage<1, [A9_Issue0], 0>,
InstrStage<1, [A9_Issue1], 0>,
InstrStage<1, [A9_Branch]>]>,
// VFP and NEON shares the same register file. This means that every VFP
// instruction should wait for full completion of the consecutive NEON
// instruction and vice-versa. We model this behavior with two artificial FUs:
// DRegsVFP and DRegsVFP.
//
// Every VFP instruction:
// - Acquires DRegsVFP resource for 1 cycle
// - Reserves DRegsN resource for the whole duration (including time to
// register file writeback!).
// Every NEON instruction does the same but with FUs swapped.
//
// Since the reserved FU cannot be acquired, this models precisely
// "cross-domain" stalls.
// VFP
// Issue through integer pipeline, and execute in NEON unit.
// FP Special Register to Integer Register File Move
InstrItinData<IIC_fpSTAT , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsVFP], 0, Required>,
InstrStage<2, [A9_DRegsN], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[1]>,
//
// Single-precision FP Unary
InstrItinData<IIC_fpUNA32 , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsVFP], 0, Required>,
// Extra latency cycles since wbck is 2 cycles
InstrStage<3, [A9_DRegsN], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[1, 1]>,
//
// Double-precision FP Unary
InstrItinData<IIC_fpUNA64 , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsVFP], 0, Required>,
// Extra latency cycles since wbck is 2 cycles
InstrStage<3, [A9_DRegsN], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[1, 1]>,
//
// Single-precision FP Compare
InstrItinData<IIC_fpCMP32 , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsVFP], 0, Required>,
// Extra latency cycles since wbck is 4 cycles
InstrStage<5, [A9_DRegsN], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[1, 1]>,
//
// Double-precision FP Compare
InstrItinData<IIC_fpCMP64 , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsVFP], 0, Required>,
// Extra latency cycles since wbck is 4 cycles
InstrStage<5, [A9_DRegsN], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[1, 1]>,
//
// Single to Double FP Convert
InstrItinData<IIC_fpCVTSD , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsVFP], 0, Required>,
InstrStage<5, [A9_DRegsN], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[4, 1]>,
//
// Double to Single FP Convert
InstrItinData<IIC_fpCVTDS , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsVFP], 0, Required>,
InstrStage<5, [A9_DRegsN], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[4, 1]>,
//
// Single to Half FP Convert
InstrItinData<IIC_fpCVTSH , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsVFP], 0, Required>,
InstrStage<5, [A9_DRegsN], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[4, 1]>,
//
// Half to Single FP Convert
InstrItinData<IIC_fpCVTHS , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsVFP], 0, Required>,
InstrStage<3, [A9_DRegsN], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[2, 1]>,
//
// Single-Precision FP to Integer Convert
InstrItinData<IIC_fpCVTSI , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsVFP], 0, Required>,
InstrStage<5, [A9_DRegsN], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[4, 1]>,
//
// Double-Precision FP to Integer Convert
InstrItinData<IIC_fpCVTDI , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsVFP], 0, Required>,
InstrStage<5, [A9_DRegsN], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[4, 1]>,
//
// Integer to Single-Precision FP Convert
InstrItinData<IIC_fpCVTIS , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsVFP], 0, Required>,
InstrStage<5, [A9_DRegsN], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[4, 1]>,
//
// Integer to Double-Precision FP Convert
InstrItinData<IIC_fpCVTID , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsVFP], 0, Required>,
InstrStage<5, [A9_DRegsN], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[4, 1]>,
//
// Single-precision FP ALU
InstrItinData<IIC_fpALU32 , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsVFP], 0, Required>,
InstrStage<5, [A9_DRegsN], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[4, 1, 1]>,
//
// Double-precision FP ALU
InstrItinData<IIC_fpALU64 , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsVFP], 0, Required>,
InstrStage<5, [A9_DRegsN], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[4, 1, 1]>,
//
// Single-precision FP Multiply
InstrItinData<IIC_fpMUL32 , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsVFP], 0, Required>,
InstrStage<6, [A9_DRegsN], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[5, 1, 1]>,
//
// Double-precision FP Multiply
InstrItinData<IIC_fpMUL64 , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsVFP], 0, Required>,
InstrStage<7, [A9_DRegsN], 0, Reserved>,
InstrStage<2, [A9_NPipe]>],
[6, 1, 1]>,
//
// Single-precision FP MAC
InstrItinData<IIC_fpMAC32 , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsVFP], 0, Required>,
InstrStage<9, [A9_DRegsN], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[8, 1, 1, 1]>,
//
// Double-precision FP MAC
InstrItinData<IIC_fpMAC64 , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsVFP], 0, Required>,
InstrStage<10, [A9_DRegsN], 0, Reserved>,
InstrStage<2, [A9_NPipe]>],
[9, 1, 1, 1]>,
//
// Single-precision Fused FP MAC
InstrItinData<IIC_fpFMAC32, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsVFP], 0, Required>,
InstrStage<9, [A9_DRegsN], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[8, 1, 1, 1]>,
//
// Double-precision Fused FP MAC
InstrItinData<IIC_fpFMAC64, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsVFP], 0, Required>,
InstrStage<10, [A9_DRegsN], 0, Reserved>,
InstrStage<2, [A9_NPipe]>],
[9, 1, 1, 1]>,
//
// Single-precision FP DIV
InstrItinData<IIC_fpDIV32 , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsVFP], 0, Required>,
InstrStage<16, [A9_DRegsN], 0, Reserved>,
InstrStage<10, [A9_NPipe]>],
[15, 1, 1]>,
//
// Double-precision FP DIV
InstrItinData<IIC_fpDIV64 , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsVFP], 0, Required>,
InstrStage<26, [A9_DRegsN], 0, Reserved>,
InstrStage<20, [A9_NPipe]>],
[25, 1, 1]>,
//
// Single-precision FP SQRT
InstrItinData<IIC_fpSQRT32, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsVFP], 0, Required>,
InstrStage<18, [A9_DRegsN], 0, Reserved>,
InstrStage<13, [A9_NPipe]>],
[17, 1]>,
//
// Double-precision FP SQRT
InstrItinData<IIC_fpSQRT64, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsVFP], 0, Required>,
InstrStage<33, [A9_DRegsN], 0, Reserved>,
InstrStage<28, [A9_NPipe]>],
[32, 1]>,
//
// Integer to Single-precision Move
InstrItinData<IIC_fpMOVIS, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsVFP], 0, Required>,
// Extra 1 latency cycle since wbck is 2 cycles
InstrStage<3, [A9_DRegsN], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[1, 1]>,
//
// Integer to Double-precision Move
InstrItinData<IIC_fpMOVID, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsVFP], 0, Required>,
// Extra 1 latency cycle since wbck is 2 cycles
InstrStage<3, [A9_DRegsN], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[1, 1, 1]>,
//
// Single-precision to Integer Move
//
// On A9 move-from-VFP is free to issue with no stall if other VFP
// operations are in flight. I assume it still can't dual-issue though.
InstrItinData<IIC_fpMOVSI, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>],
[2, 1]>,
//
// Double-precision to Integer Move
//
// On A9 move-from-VFP is free to issue with no stall if other VFP
// operations are in flight. I assume it still can't dual-issue though.
InstrItinData<IIC_fpMOVDI, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>],
[2, 1, 1]>,
//
// Single-precision FP Load
InstrItinData<IIC_fpLoad32, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsVFP], 0, Required>,
InstrStage<2, [A9_DRegsN], 0, Reserved>,
InstrStage<1, [A9_NPipe], 0>,
InstrStage<1, [A9_LSUnit]>],
[1, 1]>,
//
// Double-precision FP Load
// FIXME: Result latency is 1 if address is 64-bit aligned.
InstrItinData<IIC_fpLoad64, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsVFP], 0, Required>,
InstrStage<2, [A9_DRegsN], 0, Reserved>,
InstrStage<1, [A9_NPipe], 0>,
InstrStage<1, [A9_LSUnit]>],
[2, 1]>,
//
// FP Load Multiple
// FIXME: assumes 2 doubles which requires 2 LS cycles.
InstrItinData<IIC_fpLoad_m, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsVFP], 0, Required>,
InstrStage<2, [A9_DRegsN], 0, Reserved>,
InstrStage<1, [A9_NPipe], 0>,
InstrStage<2, [A9_LSUnit]>],
[1, 1, 1, 1], [], -1>, // dynamic uops
//
// FP Load Multiple + update
// FIXME: assumes 2 doubles which requires 2 LS cycles.
InstrItinData<IIC_fpLoad_mu,[InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsVFP], 0, Required>,
InstrStage<2, [A9_DRegsN], 0, Reserved>,
InstrStage<1, [A9_NPipe], 0>,
InstrStage<2, [A9_LSUnit]>],
[2, 1, 1, 1], [], -1>, // dynamic uops
//
// Single-precision FP Store
InstrItinData<IIC_fpStore32,[InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsVFP], 0, Required>,
InstrStage<2, [A9_DRegsN], 0, Reserved>,
InstrStage<1, [A9_NPipe], 0>,
InstrStage<1, [A9_LSUnit]>],
[1, 1]>,
//
// Double-precision FP Store
InstrItinData<IIC_fpStore64,[InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsVFP], 0, Required>,
InstrStage<2, [A9_DRegsN], 0, Reserved>,
InstrStage<1, [A9_NPipe], 0>,
InstrStage<1, [A9_LSUnit]>],
[1, 1]>,
//
// FP Store Multiple
// FIXME: assumes 2 doubles which requires 2 LS cycles.
InstrItinData<IIC_fpStore_m,[InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsVFP], 0, Required>,
InstrStage<2, [A9_DRegsN], 0, Reserved>,
InstrStage<1, [A9_NPipe], 0>,
InstrStage<2, [A9_LSUnit]>],
[1, 1, 1, 1], [], -1>, // dynamic uops
//
// FP Store Multiple + update
// FIXME: assumes 2 doubles which requires 2 LS cycles.
InstrItinData<IIC_fpStore_mu,[InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsVFP], 0, Required>,
InstrStage<2, [A9_DRegsN], 0, Reserved>,
InstrStage<1, [A9_NPipe], 0>,
InstrStage<2, [A9_LSUnit]>],
[2, 1, 1, 1], [], -1>, // dynamic uops
// NEON
// VLD1
InstrItinData<IIC_VLD1, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe], 0>,
InstrStage<1, [A9_LSUnit]>],
[1, 1]>,
// VLD1x2
InstrItinData<IIC_VLD1x2, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe], 0>,
InstrStage<1, [A9_LSUnit]>],
[1, 1, 1]>,
// VLD1x3
InstrItinData<IIC_VLD1x3, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<8, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe], 0>,
InstrStage<2, [A9_LSUnit]>],
[1, 1, 2, 1]>,
// VLD1x4
InstrItinData<IIC_VLD1x4, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<8, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe], 0>,
InstrStage<2, [A9_LSUnit]>],
[1, 1, 2, 2, 1]>,
// VLD1u
InstrItinData<IIC_VLD1u, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe], 0>,
InstrStage<1, [A9_LSUnit]>],
[1, 2, 1]>,
// VLD1x2u
InstrItinData<IIC_VLD1x2u, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe], 0>,
InstrStage<1, [A9_LSUnit]>],
[1, 1, 2, 1]>,
// VLD1x3u
InstrItinData<IIC_VLD1x3u, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<8, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe], 0>,
InstrStage<2, [A9_LSUnit]>],
[1, 1, 2, 2, 1]>,
// VLD1x4u
InstrItinData<IIC_VLD1x4u, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<8, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe], 0>,
InstrStage<2, [A9_LSUnit]>],
[1, 1, 2, 2, 2, 1]>,
//
// VLD1ln
InstrItinData<IIC_VLD1ln, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<8, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe], 0>,
InstrStage<2, [A9_LSUnit]>],
[3, 1, 1, 1]>,
//
// VLD1lnu
InstrItinData<IIC_VLD1lnu, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<8, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe], 0>,
InstrStage<2, [A9_LSUnit]>],
[3, 2, 1, 1, 1, 1]>,
//
// VLD1dup
InstrItinData<IIC_VLD1dup, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe], 0>,
InstrStage<1, [A9_LSUnit]>],
[2, 1]>,
//
// VLD1dupu
InstrItinData<IIC_VLD1dupu, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe], 0>,
InstrStage<1, [A9_LSUnit]>],
[2, 2, 1, 1]>,
//
// VLD2
InstrItinData<IIC_VLD2, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 7 cycles
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe], 0>,
InstrStage<1, [A9_LSUnit]>],
[2, 2, 1]>,
//
// VLD2x2
InstrItinData<IIC_VLD2x2, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<8, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe], 0>,
InstrStage<2, [A9_LSUnit]>],
[2, 3, 2, 3, 1]>,
//
// VLD2ln
InstrItinData<IIC_VLD2ln, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<8, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe], 0>,
InstrStage<2, [A9_LSUnit]>],
[3, 3, 1, 1, 1, 1]>,
//
// VLD2u
InstrItinData<IIC_VLD2u, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 7 cycles
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe], 0>,
InstrStage<1, [A9_LSUnit]>],
[2, 2, 2, 1, 1, 1]>,
//
// VLD2x2u
InstrItinData<IIC_VLD2x2u, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<8, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe], 0>,
InstrStage<2, [A9_LSUnit]>],
[2, 3, 2, 3, 2, 1]>,
//
// VLD2lnu
InstrItinData<IIC_VLD2lnu, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<8, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe], 0>,
InstrStage<2, [A9_LSUnit]>],
[3, 3, 2, 1, 1, 1, 1, 1]>,
//
// VLD2dup
InstrItinData<IIC_VLD2dup, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe], 0>,
InstrStage<1, [A9_LSUnit]>],
[2, 2, 1]>,
//
// VLD2dupu
InstrItinData<IIC_VLD2dupu, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe], 0>,
InstrStage<1, [A9_LSUnit]>],
[2, 2, 2, 1, 1]>,
//
// VLD3
InstrItinData<IIC_VLD3, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<9,[A9_DRegsVFP], 0, Reserved>,
InstrStage<3, [A9_NPipe], 0>,
InstrStage<3, [A9_LSUnit]>],
[3, 3, 4, 1]>,
//
// VLD3ln
InstrItinData<IIC_VLD3ln, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<11,[A9_DRegsVFP], 0, Reserved>,
InstrStage<5, [A9_NPipe], 0>,
InstrStage<5, [A9_LSUnit]>],
[5, 5, 6, 1, 1, 1, 1, 2]>,
//
// VLD3u
InstrItinData<IIC_VLD3u, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<9,[A9_DRegsVFP], 0, Reserved>,
InstrStage<3, [A9_NPipe], 0>,
InstrStage<3, [A9_LSUnit]>],
[3, 3, 4, 2, 1]>,
//
// VLD3lnu
InstrItinData<IIC_VLD3lnu, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<11,[A9_DRegsVFP], 0, Reserved>,
InstrStage<5, [A9_NPipe], 0>,
InstrStage<5, [A9_LSUnit]>],
[5, 5, 6, 2, 1, 1, 1, 1, 1, 2]>,
//
// VLD3dup
InstrItinData<IIC_VLD3dup, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<9, [A9_DRegsVFP], 0, Reserved>,
InstrStage<3, [A9_NPipe], 0>,
InstrStage<3, [A9_LSUnit]>],
[3, 3, 4, 1]>,
//
// VLD3dupu
InstrItinData<IIC_VLD3dupu, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<9, [A9_DRegsVFP], 0, Reserved>,
InstrStage<3, [A9_NPipe], 0>,
InstrStage<3, [A9_LSUnit]>],
[3, 3, 4, 2, 1, 1]>,
//
// VLD4
InstrItinData<IIC_VLD4, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<9,[A9_DRegsVFP], 0, Reserved>,
InstrStage<3, [A9_NPipe], 0>,
InstrStage<3, [A9_LSUnit]>],
[3, 3, 4, 4, 1]>,
//
// VLD4ln
InstrItinData<IIC_VLD4ln, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<10,[A9_DRegsVFP], 0, Reserved>,
InstrStage<4, [A9_NPipe], 0>,
InstrStage<4, [A9_LSUnit]>],
[4, 4, 5, 5, 1, 1, 1, 1, 2, 2]>,
//
// VLD4u
InstrItinData<IIC_VLD4u, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<9,[A9_DRegsVFP], 0, Reserved>,
InstrStage<3, [A9_NPipe], 0>,
InstrStage<3, [A9_LSUnit]>],
[3, 3, 4, 4, 2, 1]>,
//
// VLD4lnu
InstrItinData<IIC_VLD4lnu, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<10,[A9_DRegsVFP], 0, Reserved>,
InstrStage<4, [A9_NPipe], 0>,
InstrStage<4, [A9_LSUnit]>],
[4, 4, 5, 5, 2, 1, 1, 1, 1, 1, 2, 2]>,
//
// VLD4dup
InstrItinData<IIC_VLD4dup, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<8, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe], 0>,
InstrStage<2, [A9_LSUnit]>],
[2, 2, 3, 3, 1]>,
//
// VLD4dupu
InstrItinData<IIC_VLD4dupu, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<8, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe], 0>,
InstrStage<2, [A9_LSUnit]>],
[2, 2, 3, 3, 2, 1, 1]>,
//
// VST1
InstrItinData<IIC_VST1, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<1, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe], 0>,
InstrStage<1, [A9_LSUnit]>],
[1, 1, 1]>,
//
// VST1x2
InstrItinData<IIC_VST1x2, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<1, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe], 0>,
InstrStage<1, [A9_LSUnit]>],
[1, 1, 1, 1]>,
//
// VST1x3
InstrItinData<IIC_VST1x3, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<2, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe], 0>,
InstrStage<2, [A9_LSUnit]>],
[1, 1, 1, 1, 2]>,
//
// VST1x4
InstrItinData<IIC_VST1x4, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<2, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe], 0>,
InstrStage<2, [A9_LSUnit]>],
[1, 1, 1, 1, 2, 2]>,
//
// VST1u
InstrItinData<IIC_VST1u, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<1, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe], 0>,
InstrStage<1, [A9_LSUnit]>],
[2, 1, 1, 1, 1]>,
//
// VST1x2u
InstrItinData<IIC_VST1x2u, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<1, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe], 0>,
InstrStage<1, [A9_LSUnit]>],
[2, 1, 1, 1, 1, 1]>,
//
// VST1x3u
InstrItinData<IIC_VST1x3u, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<2, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe], 0>,
InstrStage<2, [A9_LSUnit]>],
[2, 1, 1, 1, 1, 1, 2]>,
//
// VST1x4u
InstrItinData<IIC_VST1x4u, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<2, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe], 0>,
InstrStage<2, [A9_LSUnit]>],
[2, 1, 1, 1, 1, 1, 2, 2]>,
//
// VST1ln
InstrItinData<IIC_VST1ln, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<1, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe], 0>,
InstrStage<1, [A9_LSUnit]>],
[1, 1, 1]>,
//
// VST1lnu
InstrItinData<IIC_VST1lnu, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<1, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe], 0>,
InstrStage<1, [A9_LSUnit]>],
[2, 1, 1, 1, 1]>,
//
// VST2
InstrItinData<IIC_VST2, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<1, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe], 0>,
InstrStage<1, [A9_LSUnit]>],
[1, 1, 1, 1]>,
//
// VST2x2
InstrItinData<IIC_VST2x2, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<3, [A9_DRegsVFP], 0, Reserved>,
InstrStage<3, [A9_NPipe], 0>,
InstrStage<3, [A9_LSUnit]>],
[1, 1, 1, 1, 2, 2]>,
//
// VST2u
InstrItinData<IIC_VST2u, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<1, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe], 0>,
InstrStage<1, [A9_LSUnit]>],
[2, 1, 1, 1, 1, 1]>,
//
// VST2x2u
InstrItinData<IIC_VST2x2u, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<3, [A9_DRegsVFP], 0, Reserved>,
InstrStage<3, [A9_NPipe], 0>,
InstrStage<3, [A9_LSUnit]>],
[2, 1, 1, 1, 1, 1, 2, 2]>,
//
// VST2ln
InstrItinData<IIC_VST2ln, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<1, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe], 0>,
InstrStage<1, [A9_LSUnit]>],
[1, 1, 1, 1]>,
//
// VST2lnu
InstrItinData<IIC_VST2lnu, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<1, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe], 0>,
InstrStage<1, [A9_LSUnit]>],
[2, 1, 1, 1, 1, 1]>,
//
// VST3
InstrItinData<IIC_VST3, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<2, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe], 0>,
InstrStage<2, [A9_LSUnit]>],
[1, 1, 1, 1, 2]>,
//
// VST3u
InstrItinData<IIC_VST3u, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<2, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe], 0>,
InstrStage<2, [A9_LSUnit]>],
[2, 1, 1, 1, 1, 1, 2]>,
//
// VST3ln
InstrItinData<IIC_VST3ln, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<3, [A9_DRegsVFP], 0, Reserved>,
InstrStage<3, [A9_NPipe], 0>,
InstrStage<3, [A9_LSUnit]>],
[1, 1, 1, 1, 2]>,
//
// VST3lnu
InstrItinData<IIC_VST3lnu, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<3, [A9_DRegsVFP], 0, Reserved>,
InstrStage<3, [A9_NPipe], 0>,
InstrStage<3, [A9_LSUnit]>],
[2, 1, 1, 1, 1, 1, 2]>,
//
// VST4
InstrItinData<IIC_VST4, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<2, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe], 0>,
InstrStage<2, [A9_LSUnit]>],
[1, 1, 1, 1, 2, 2]>,
//
// VST4u
InstrItinData<IIC_VST4u, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<2, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe], 0>,
InstrStage<2, [A9_LSUnit]>],
[2, 1, 1, 1, 1, 1, 2, 2]>,
//
// VST4ln
InstrItinData<IIC_VST4ln, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<2, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe], 0>,
InstrStage<2, [A9_LSUnit]>],
[1, 1, 1, 1, 2, 2]>,
//
// VST4lnu
InstrItinData<IIC_VST4lnu, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<2, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe], 0>,
InstrStage<2, [A9_LSUnit]>],
[2, 1, 1, 1, 1, 1, 2, 2]>,
//
// Double-register Integer Unary
InstrItinData<IIC_VUNAiD, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 6 cycles
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[4, 2]>,
//
// Quad-register Integer Unary
InstrItinData<IIC_VUNAiQ, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 6 cycles
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[4, 2]>,
//
// Double-register Integer Q-Unary
InstrItinData<IIC_VQUNAiD, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 6 cycles
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[4, 1]>,
//
// Quad-register Integer CountQ-Unary
InstrItinData<IIC_VQUNAiQ, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 6 cycles
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[4, 1]>,
//
// Double-register Integer Binary
InstrItinData<IIC_VBINiD, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 6 cycles
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[3, 2, 2]>,
//
// Quad-register Integer Binary
InstrItinData<IIC_VBINiQ, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 6 cycles
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[3, 2, 2]>,
//
// Double-register Integer Subtract
InstrItinData<IIC_VSUBiD, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 6 cycles
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[3, 2, 1]>,
//
// Quad-register Integer Subtract
InstrItinData<IIC_VSUBiQ, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 6 cycles
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[3, 2, 1]>,
//
// Double-register Integer Shift
InstrItinData<IIC_VSHLiD, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 6 cycles
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[3, 1, 1]>,
//
// Quad-register Integer Shift
InstrItinData<IIC_VSHLiQ, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 6 cycles
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[3, 1, 1]>,
//
// Double-register Integer Shift (4 cycle)
InstrItinData<IIC_VSHLi4D, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 6 cycles
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[4, 1, 1]>,
//
// Quad-register Integer Shift (4 cycle)
InstrItinData<IIC_VSHLi4Q, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 6 cycles
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[4, 1, 1]>,
//
// Double-register Integer Binary (4 cycle)
InstrItinData<IIC_VBINi4D, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 6 cycles
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[4, 2, 2]>,
//
// Quad-register Integer Binary (4 cycle)
InstrItinData<IIC_VBINi4Q, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 6 cycles
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[4, 2, 2]>,
//
// Double-register Integer Subtract (4 cycle)
InstrItinData<IIC_VSUBi4D, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 6 cycles
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[4, 2, 1]>,
//
// Quad-register Integer Subtract (4 cycle)
InstrItinData<IIC_VSUBi4Q, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 6 cycles
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[4, 2, 1]>,
//
// Double-register Integer Count
InstrItinData<IIC_VCNTiD, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 6 cycles
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[3, 2, 2]>,
//
// Quad-register Integer Count
// Result written in N3, but that is relative to the last cycle of multicycle,
// so we use 4 for those cases
InstrItinData<IIC_VCNTiQ, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 7 cycles
InstrStage<8, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe]>],
[4, 2, 2]>,
//
// Double-register Absolute Difference and Accumulate
InstrItinData<IIC_VABAD, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 6 cycles
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[6, 3, 2, 1]>,
//
// Quad-register Absolute Difference and Accumulate
InstrItinData<IIC_VABAQ, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 6 cycles
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe]>],
[6, 3, 2, 1]>,
//
// Double-register Integer Pair Add Long
InstrItinData<IIC_VPALiD, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 6 cycles
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[6, 3, 1]>,
//
// Quad-register Integer Pair Add Long
InstrItinData<IIC_VPALiQ, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 6 cycles
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe]>],
[6, 3, 1]>,
//
// Double-register Integer Multiply (.8, .16)
InstrItinData<IIC_VMULi16D, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 6 cycles
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[6, 2, 2]>,
//
// Quad-register Integer Multiply (.8, .16)
InstrItinData<IIC_VMULi16Q, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 7 cycles
InstrStage<8, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe]>],
[7, 2, 2]>,
//
// Double-register Integer Multiply (.32)
InstrItinData<IIC_VMULi32D, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 7 cycles
InstrStage<8, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe]>],
[7, 2, 1]>,
//
// Quad-register Integer Multiply (.32)
InstrItinData<IIC_VMULi32Q, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 9 cycles
InstrStage<10, [A9_DRegsVFP], 0, Reserved>,
InstrStage<4, [A9_NPipe]>],
[9, 2, 1]>,
//
// Double-register Integer Multiply-Accumulate (.8, .16)
InstrItinData<IIC_VMACi16D, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 6 cycles
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[6, 3, 2, 2]>,
//
// Double-register Integer Multiply-Accumulate (.32)
InstrItinData<IIC_VMACi32D, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 7 cycles
InstrStage<8, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe]>],
[7, 3, 2, 1]>,
//
// Quad-register Integer Multiply-Accumulate (.8, .16)
InstrItinData<IIC_VMACi16Q, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 7 cycles
InstrStage<8, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe]>],
[7, 3, 2, 2]>,
//
// Quad-register Integer Multiply-Accumulate (.32)
InstrItinData<IIC_VMACi32Q, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 9 cycles
InstrStage<10, [A9_DRegsVFP], 0, Reserved>,
InstrStage<4, [A9_NPipe]>],
[9, 3, 2, 1]>,
//
// Move
InstrItinData<IIC_VMOV, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<1, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[1,1]>,
//
// Move Immediate
InstrItinData<IIC_VMOVImm, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 6 cycles
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[3]>,
//
// Double-register Permute Move
InstrItinData<IIC_VMOVD, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 6 cycles
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[2, 1]>,
//
// Quad-register Permute Move
InstrItinData<IIC_VMOVQ, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 6 cycles
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[2, 1]>,
//
// Integer to Single-precision Move
InstrItinData<IIC_VMOVIS , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<3, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[1, 1]>,
//
// Integer to Double-precision Move
InstrItinData<IIC_VMOVID , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<3, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[1, 1, 1]>,
//
// Single-precision to Integer Move
InstrItinData<IIC_VMOVSI , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<3, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[2, 1]>,
//
// Double-precision to Integer Move
InstrItinData<IIC_VMOVDI , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<3, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[2, 2, 1]>,
//
// Integer to Lane Move
InstrItinData<IIC_VMOVISL , [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
InstrStage<4, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe]>],
[3, 1, 1]>,
//
// Vector narrow move
InstrItinData<IIC_VMOVN, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 6 cycles
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[3, 1]>,
//
// Double-register FP Unary
InstrItinData<IIC_VUNAD, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 6 cycles
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[5, 2]>,
//
// Quad-register FP Unary
// Result written in N5, but that is relative to the last cycle of multicycle,
// so we use 6 for those cases
InstrItinData<IIC_VUNAQ, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 7 cycles
InstrStage<8, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe]>],
[6, 2]>,
//
// Double-register FP Binary
// FIXME: We're using this itin for many instructions and [2, 2] here is too
// optimistic.
InstrItinData<IIC_VBIND, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 6 cycles
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[5, 2, 2]>,
//
// VPADD, etc.
InstrItinData<IIC_VPBIND, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 6 cycles
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[5, 1, 1]>,
//
// Double-register FP VMUL
InstrItinData<IIC_VFMULD, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 6 cycles
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[5, 2, 1]>,
//
// Quad-register FP Binary
// Result written in N5, but that is relative to the last cycle of multicycle,
// so we use 6 for those cases
// FIXME: We're using this itin for many instructions and [2, 2] here is too
// optimistic.
InstrItinData<IIC_VBINQ, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 7 cycles
InstrStage<8, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe]>],
[6, 2, 2]>,
//
// Quad-register FP VMUL
InstrItinData<IIC_VFMULQ, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 7 cycles
InstrStage<8, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[6, 2, 1]>,
//
// Double-register FP Multiple-Accumulate
InstrItinData<IIC_VMACD, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 7 cycles
InstrStage<8, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe]>],
[6, 3, 2, 1]>,
//
// Quad-register 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_VMACQ, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 9 cycles
InstrStage<10, [A9_DRegsVFP], 0, Reserved>,
InstrStage<4, [A9_NPipe]>],
[8, 4, 2, 1]>,
//
// Double-register Fused FP Multiple-Accumulate
InstrItinData<IIC_VFMACD, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 7 cycles
InstrStage<8, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe]>],
[6, 3, 2, 1]>,
//
// 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, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 9 cycles
InstrStage<10, [A9_DRegsVFP], 0, Reserved>,
InstrStage<4, [A9_NPipe]>],
[8, 4, 2, 1]>,
//
// Double-register Reciprical Step
InstrItinData<IIC_VRECSD, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 10 cycles
InstrStage<11, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[9, 2, 2]>,
//
// Quad-register Reciprical Step
InstrItinData<IIC_VRECSQ, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 11 cycles
InstrStage<12, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe]>],
[10, 2, 2]>,
//
// Double-register Permute
InstrItinData<IIC_VPERMD, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 6 cycles
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[2, 2, 1, 1]>,
//
// Quad-register Permute
// Result written in N2, but that is relative to the last cycle of multicycle,
// so we use 3 for those cases
InstrItinData<IIC_VPERMQ, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 7 cycles
InstrStage<8, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe]>],
[3, 3, 1, 1]>,
//
// Quad-register Permute (3 cycle issue)
// Result written in N2, but that is relative to the last cycle of multicycle,
// so we use 4 for those cases
InstrItinData<IIC_VPERMQ3, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 8 cycles
InstrStage<9, [A9_DRegsVFP], 0, Reserved>,
InstrStage<3, [A9_NPipe]>],
[4, 4, 1, 1]>,
//
// Double-register VEXT
InstrItinData<IIC_VEXTD, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 6 cycles
InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
InstrStage<1, [A9_NPipe]>],
[2, 1, 1]>,
//
// Quad-register VEXT
InstrItinData<IIC_VEXTQ, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 7 cycles
InstrStage<8, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe]>],
[3, 1, 2]>,
//
// VTB
InstrItinData<IIC_VTB1, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 7 cycles
InstrStage<8, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe]>],
[3, 2, 1]>,
InstrItinData<IIC_VTB2, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<2, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 7 cycles
InstrStage<8, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe]>],
[3, 2, 2, 1]>,
InstrItinData<IIC_VTB3, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<2, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 8 cycles
InstrStage<9, [A9_DRegsVFP], 0, Reserved>,
InstrStage<3, [A9_NPipe]>],
[4, 2, 2, 3, 1]>,
InstrItinData<IIC_VTB4, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 8 cycles
InstrStage<9, [A9_DRegsVFP], 0, Reserved>,
InstrStage<3, [A9_NPipe]>],
[4, 2, 2, 3, 3, 1]>,
//
// VTBX
InstrItinData<IIC_VTBX1, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 7 cycles
InstrStage<8, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe]>],
[3, 1, 2, 1]>,
InstrItinData<IIC_VTBX2, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 7 cycles
InstrStage<8, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe]>],
[3, 1, 2, 2, 1]>,
InstrItinData<IIC_VTBX3, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 8 cycles
InstrStage<9, [A9_DRegsVFP], 0, Reserved>,
InstrStage<3, [A9_NPipe]>],
[4, 1, 2, 2, 3, 1]>,
InstrItinData<IIC_VTBX4, [InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
InstrStage<1, [A9_MUX0], 0>,
InstrStage<1, [A9_DRegsN], 0, Required>,
// Extra latency cycles since wbck is 8 cycles
InstrStage<9, [A9_DRegsVFP], 0, Reserved>,
InstrStage<2, [A9_NPipe]>],
[4, 1, 2, 2, 3, 3, 1]>
]>;
// ===---------------------------------------------------------------------===//
// The following definitions describe the simpler per-operand machine model.
// This works with MachineScheduler and will eventually replace itineraries.
class A9WriteLMOpsListType<list<WriteSequence> writes> {
list <WriteSequence> Writes = writes;
SchedMachineModel SchedModel = ?;
}
// Cortex-A9 machine model for scheduling and other instruction cost heuristics.
def CortexA9Model : SchedMachineModel {
let IssueWidth = 2; // 2 micro-ops are dispatched per cycle.
let MicroOpBufferSize = 56; // Based on available renamed registers.
let LoadLatency = 2; // Optimistic load latency assuming bypass.
// This is overriden by OperandCycles if the
// Itineraries are queried instead.
let MispredictPenalty = 8; // Based on estimate of pipeline depth.
let Itineraries = CortexA9Itineraries;
}
//===----------------------------------------------------------------------===//
// Define each kind of processor resource and number available.
let SchedModel = CortexA9Model in {
def A9UnitALU : ProcResource<2>;
def A9UnitMul : ProcResource<1> { let Super = A9UnitALU; }
def A9UnitAGU : ProcResource<1>;
def A9UnitLS : ProcResource<1>;
def A9UnitFP : ProcResource<1> { let BufferSize = 1; }
def A9UnitB : ProcResource<1>;
//===----------------------------------------------------------------------===//
// Define scheduler read/write types with their resources and latency on A9.
// Consume an issue slot, but no processor resources. This is useful when all
// other writes associated with the operand have NumMicroOps = 0.
def A9WriteIssue : SchedWriteRes<[]> { let Latency = 0; }
// Write an integer register.
def A9WriteI : SchedWriteRes<[A9UnitALU]>;
// Write an integer shifted-by register
def A9WriteIsr : SchedWriteRes<[A9UnitALU]> { let Latency = 2; }
// Basic ALU.
def A9WriteALU : SchedWriteRes<[A9UnitALU]>;
// ALU with operand shifted by immediate.
def : WriteRes<WriteALUsi, [A9UnitALU]> { let Latency = 2; }
// ALU with operand shifted by register.
def A9WriteALUsr : SchedWriteRes<[A9UnitALU]> { let Latency = 3; }
// Multiplication
def A9WriteM : SchedWriteRes<[A9UnitMul, A9UnitMul]> { let Latency = 4; }
def A9WriteMHi : SchedWriteRes<[A9UnitMul]> { let Latency = 5;
let NumMicroOps = 0; }
def A9WriteM16 : SchedWriteRes<[A9UnitMul]> { let Latency = 3; }
def A9WriteM16Hi : SchedWriteRes<[A9UnitMul]> { let Latency = 4;
let NumMicroOps = 0; }
// Floating-point
// Only one FP or AGU instruction may issue per cycle. We model this
// by having FP instructions consume the AGU resource.
def A9WriteF : SchedWriteRes<[A9UnitFP, A9UnitAGU]> { let Latency = 4; }
def A9WriteFMov : SchedWriteRes<[A9UnitFP, A9UnitAGU]> { let Latency = 1; }
def A9WriteFMulS : SchedWriteRes<[A9UnitFP, A9UnitAGU]> { let Latency = 5; }
def A9WriteFMulD : SchedWriteRes<[A9UnitFP, A9UnitAGU]> { let Latency = 6; }
def A9WriteFMAS : SchedWriteRes<[A9UnitFP, A9UnitAGU]> { let Latency = 8; }
def A9WriteFMAD : SchedWriteRes<[A9UnitFP, A9UnitAGU]> { let Latency = 9; }
def A9WriteFDivS : SchedWriteRes<[A9UnitFP, A9UnitAGU]> { let Latency = 15; }
def A9WriteFDivD : SchedWriteRes<[A9UnitFP, A9UnitAGU]> { let Latency = 25; }
def A9WriteFSqrtS : SchedWriteRes<[A9UnitFP, A9UnitAGU]> { let Latency = 17; }
def A9WriteFSqrtD : SchedWriteRes<[A9UnitFP, A9UnitAGU]> { let Latency = 32; }
// NEON has an odd mix of latencies. Simply name the write types by latency.
def A9WriteV1 : SchedWriteRes<[A9UnitFP, A9UnitAGU]> { let Latency = 1; }
def A9WriteV2 : SchedWriteRes<[A9UnitFP, A9UnitAGU]> { let Latency = 2; }
def A9WriteV3 : SchedWriteRes<[A9UnitFP, A9UnitAGU]> { let Latency = 3; }
def A9WriteV4 : SchedWriteRes<[A9UnitFP, A9UnitAGU]> { let Latency = 4; }
def A9WriteV5 : SchedWriteRes<[A9UnitFP, A9UnitAGU]> { let Latency = 5; }
def A9WriteV6 : SchedWriteRes<[A9UnitFP, A9UnitAGU]> { let Latency = 6; }
def A9WriteV7 : SchedWriteRes<[A9UnitFP, A9UnitAGU]> { let Latency = 7; }
def A9WriteV9 : SchedWriteRes<[A9UnitFP, A9UnitAGU]> { let Latency = 9; }
def A9WriteV10 : SchedWriteRes<[A9UnitFP, A9UnitAGU]> { let Latency = 10; }
// Reserve A9UnitFP for 2 consecutive cycles.
def A9Write2V4 : SchedWriteRes<[A9UnitFP, A9UnitAGU]> {
let Latency = 4;
let ResourceCycles = [2];
}
def A9Write2V7 : SchedWriteRes<[A9UnitFP, A9UnitAGU]> {
let Latency = 7;
let ResourceCycles = [2];
}
def A9Write2V9 : SchedWriteRes<[A9UnitFP, A9UnitAGU]> {
let Latency = 9;
let ResourceCycles = [2];
}
// Branches don't have a def operand but still consume resources.
def A9WriteB : SchedWriteRes<[A9UnitB]>;
// Address generation.
def A9WriteAdr : SchedWriteRes<[A9UnitAGU]> { let NumMicroOps = 0; }
// Load Integer.
def A9WriteL : SchedWriteRes<[A9UnitLS]> { let Latency = 3; }
// Load the upper 32-bits using the same micro-op.
def A9WriteLHi : SchedWriteRes<[]> { let Latency = 3;
let NumMicroOps = 0; }
// Offset shifted by register.
def A9WriteLsi : SchedWriteRes<[A9UnitLS]> { let Latency = 4; }
// Load (and zero extend) a byte.
def A9WriteLb : SchedWriteRes<[A9UnitLS]> { let Latency = 4; }
def A9WriteLbsi : SchedWriteRes<[A9UnitLS]> { let Latency = 5; }
// Load or Store Float, aligned.
def A9WriteLSfp : SchedWriteRes<[A9UnitLS, A9UnitFP]> { let Latency = 1; }
// Store Integer.
def A9WriteS : SchedWriteRes<[A9UnitLS]>;
//===----------------------------------------------------------------------===//
// Define resources dynamically for load multiple variants.
// Define helpers for extra latency without consuming resources.
def A9WriteCycle1 : SchedWriteRes<[]> { let Latency = 1; let NumMicroOps = 0; }
foreach NumCycles = 2-8 in {
def A9WriteCycle#NumCycles : WriteSequence<[A9WriteCycle1], NumCycles>;
} // foreach NumCycles
// Define address generation sequences and predicates for 8 flavors of LDMs.
foreach NumAddr = 1-8 in {
// Define A9WriteAdr1-8 as a sequence of A9WriteAdr with additive
// latency for instructions that generate multiple loads or stores.
def A9WriteAdr#NumAddr : WriteSequence<[A9WriteAdr], NumAddr>;
// Define a predicate to select the LDM based on number of memory addresses.
def A9LMAdr#NumAddr#Pred :
SchedPredicate<"(TII->getNumLDMAddresses(MI)+1)/2 == "#NumAddr>;
} // foreach NumAddr
// Fall-back for unknown LDMs.
def A9LMUnknownPred : SchedPredicate<"TII->getNumLDMAddresses(MI) == 0">;
// LDM/VLDM/VLDn address generation latency & resources.
// Dynamically select the A9WriteAdrN sequence using a predicate.
def A9WriteLMAdr : SchedWriteVariant<[
SchedVar<A9LMAdr1Pred, [A9WriteAdr1]>,
SchedVar<A9LMAdr2Pred, [A9WriteAdr2]>,
SchedVar<A9LMAdr3Pred, [A9WriteAdr3]>,
SchedVar<A9LMAdr4Pred, [A9WriteAdr4]>,
SchedVar<A9LMAdr5Pred, [A9WriteAdr5]>,
SchedVar<A9LMAdr6Pred, [A9WriteAdr6]>,
SchedVar<A9LMAdr7Pred, [A9WriteAdr7]>,
SchedVar<A9LMAdr8Pred, [A9WriteAdr8]>,
// For unknown LDM/VLDM/VSTM, assume 2 32-bit registers.
SchedVar<A9LMUnknownPred, [A9WriteAdr2]>]>;
// Define LDM Resources.
// These take no issue resource, so they can be combined with other
// writes like WriteB.
// A9WriteLMLo takes a single LS resource and 2 cycles.
def A9WriteLMLo : SchedWriteRes<[A9UnitLS]> { let Latency = 2;
let NumMicroOps = 0; }
// Assuming aligned access, the upper half of each pair is free with
// the same latency.
def A9WriteLMHi : SchedWriteRes<[]> { let Latency = 2;
let NumMicroOps = 0; }
// Each A9WriteL#N variant adds N cycles of latency without consuming
// additional resources.
foreach NumAddr = 1-8 in {
def A9WriteL#NumAddr : WriteSequence<
[A9WriteLMLo, !cast<SchedWrite>("A9WriteCycle"#NumAddr)]>;
def A9WriteL#NumAddr#Hi : WriteSequence<
[A9WriteLMHi, !cast<SchedWrite>("A9WriteCycle"#NumAddr)]>;
}
//===----------------------------------------------------------------------===//
// LDM: Load multiple into 32-bit integer registers.
def A9WriteLMOpsList : A9WriteLMOpsListType<
[A9WriteL1, A9WriteL1Hi,
A9WriteL2, A9WriteL2Hi,
A9WriteL3, A9WriteL3Hi,
A9WriteL4, A9WriteL4Hi,
A9WriteL5, A9WriteL5Hi,
A9WriteL6, A9WriteL6Hi,
A9WriteL7, A9WriteL7Hi,
A9WriteL8, A9WriteL8Hi]>;
// A9WriteLM variants expand into a pair of writes for each 64-bit
// value loaded. When the number of registers is odd, the last
// A9WriteLnHi is naturally ignored because the instruction has no
// following def operands. These variants take no issue resource, so
// they may need to be part of a WriteSequence that includes A9WriteIssue.
def A9WriteLM : SchedWriteVariant<[
SchedVar<A9LMAdr1Pred, A9WriteLMOpsList.Writes[0-1]>,
SchedVar<A9LMAdr2Pred, A9WriteLMOpsList.Writes[0-3]>,
SchedVar<A9LMAdr3Pred, A9WriteLMOpsList.Writes[0-5]>,
SchedVar<A9LMAdr4Pred, A9WriteLMOpsList.Writes[0-7]>,
SchedVar<A9LMAdr5Pred, A9WriteLMOpsList.Writes[0-9]>,
SchedVar<A9LMAdr6Pred, A9WriteLMOpsList.Writes[0-11]>,
SchedVar<A9LMAdr7Pred, A9WriteLMOpsList.Writes[0-13]>,
SchedVar<A9LMAdr8Pred, A9WriteLMOpsList.Writes[0-15]>,
// For unknown LDMs, define the maximum number of writes, but only
// make the first two consume resources.
SchedVar<A9LMUnknownPred, [A9WriteL1, A9WriteL1Hi,
A9WriteL2, A9WriteL2Hi,
A9WriteL3Hi, A9WriteL3Hi,
A9WriteL4Hi, A9WriteL4Hi,
A9WriteL5Hi, A9WriteL5Hi,
A9WriteL6Hi, A9WriteL6Hi,
A9WriteL7Hi, A9WriteL7Hi,
A9WriteL8Hi, A9WriteL8Hi]>]> {
let Variadic = 1;
}
//===----------------------------------------------------------------------===//
// VFP Load/Store Multiple Variants, and NEON VLDn/VSTn support.
// A9WriteLfpOp is the same as A9WriteLSfp but takes no issue resources
// so can be used in WriteSequences for in single-issue instructions that
// encapsulate multiple loads.
def A9WriteLfpOp : SchedWriteRes<[A9UnitLS, A9UnitFP]> {
let Latency = 1;
let NumMicroOps = 0;
}
foreach NumAddr = 1-8 in {
// Helper for A9WriteLfp1-8: A sequence of fp loads with no micro-ops.
def A9WriteLfp#NumAddr#Seq : WriteSequence<[A9WriteLfpOp], NumAddr>;
// A9WriteLfp1-8 definitions are statically expanded into a sequence of
// A9WriteLfpOps with additive latency that takes a single issue slot.
// Used directly to describe NEON VLDn.
def A9WriteLfp#NumAddr : WriteSequence<
[A9WriteIssue, !cast<SchedWrite>("A9WriteLfp"#NumAddr#Seq)]>;
// A9WriteLfp1-8Mov adds a cycle of latency and FP resource for
// permuting loaded values.
def A9WriteLfp#NumAddr#Mov : WriteSequence<
[A9WriteF, !cast<SchedWrite>("A9WriteLfp"#NumAddr#Seq)]>;
} // foreach NumAddr
// Define VLDM/VSTM PreRA resources.
// A9WriteLMfpPreRA are dynamically expanded into the correct
// A9WriteLfp1-8 sequence based on a predicate. This supports the
// preRA VLDM variants in which all 64-bit loads are written to the
// same tuple of either single or double precision registers.
def A9WriteLMfpPreRA : SchedWriteVariant<[
SchedVar<A9LMAdr1Pred, [A9WriteLfp1]>,
SchedVar<A9LMAdr2Pred, [A9WriteLfp2]>,
SchedVar<A9LMAdr3Pred, [A9WriteLfp3]>,
SchedVar<A9LMAdr4Pred, [A9WriteLfp4]>,
SchedVar<A9LMAdr5Pred, [A9WriteLfp5]>,
SchedVar<A9LMAdr6Pred, [A9WriteLfp6]>,
SchedVar<A9LMAdr7Pred, [A9WriteLfp7]>,
SchedVar<A9LMAdr8Pred, [A9WriteLfp8]>,
// For unknown VLDM/VSTM PreRA, assume 2xS registers.
SchedVar<A9LMUnknownPred, [A9WriteLfp2]>]>;
// Define VLDM/VSTM PostRA Resources.
// A9WriteLMfpLo takes a LS and FP resource and one issue slot but no latency.
def A9WriteLMfpLo : SchedWriteRes<[A9UnitLS, A9UnitFP]> { let Latency = 0; }
foreach NumAddr = 1-8 in {
// Each A9WriteL#N variant adds N cycles of latency without consuming
// additional resources.
def A9WriteLMfp#NumAddr : WriteSequence<
[A9WriteLMfpLo, !cast<SchedWrite>("A9WriteCycle"#NumAddr)]>;
// Assuming aligned access, the upper half of each pair is free with
// the same latency.
def A9WriteLMfp#NumAddr#Hi : WriteSequence<
[A9WriteLMHi, !cast<SchedWrite>("A9WriteCycle"#NumAddr)]>;
} // foreach NumAddr
// VLDM PostRA Variants. These variants expand A9WriteLMfpPostRA into a
// pair of writes for each 64-bit data loaded. When the number of
// registers is odd, the last WriteLMfpnHi is naturally ignored because
// the instruction has no following def operands.
def A9WriteLMfpPostRAOpsList : A9WriteLMOpsListType<
[A9WriteLMfp1, A9WriteLMfp2, // 0-1
A9WriteLMfp3, A9WriteLMfp4, // 2-3
A9WriteLMfp5, A9WriteLMfp6, // 4-5
A9WriteLMfp7, A9WriteLMfp8, // 6-7
A9WriteLMfp1Hi, // 8-8
A9WriteLMfp2Hi, A9WriteLMfp2Hi, // 9-10
A9WriteLMfp3Hi, A9WriteLMfp3Hi, // 11-12
A9WriteLMfp4Hi, A9WriteLMfp4Hi, // 13-14
A9WriteLMfp5Hi, A9WriteLMfp5Hi, // 15-16
A9WriteLMfp6Hi, A9WriteLMfp6Hi, // 17-18
A9WriteLMfp7Hi, A9WriteLMfp7Hi, // 19-20
A9WriteLMfp8Hi, A9WriteLMfp8Hi]>; // 21-22
def A9WriteLMfpPostRA : SchedWriteVariant<[
SchedVar<A9LMAdr1Pred, A9WriteLMfpPostRAOpsList.Writes[0-0, 8-8]>,
SchedVar<A9LMAdr2Pred, A9WriteLMfpPostRAOpsList.Writes[0-1, 9-10]>,
SchedVar<A9LMAdr3Pred, A9WriteLMfpPostRAOpsList.Writes[0-2, 10-12]>,
SchedVar<A9LMAdr4Pred, A9WriteLMfpPostRAOpsList.Writes[0-3, 11-14]>,
SchedVar<A9LMAdr5Pred, A9WriteLMfpPostRAOpsList.Writes[0-4, 12-16]>,
SchedVar<A9LMAdr6Pred, A9WriteLMfpPostRAOpsList.Writes[0-5, 13-18]>,
SchedVar<A9LMAdr7Pred, A9WriteLMfpPostRAOpsList.Writes[0-6, 14-20]>,
SchedVar<A9LMAdr8Pred, A9WriteLMfpPostRAOpsList.Writes[0-7, 15-22]>,
// For unknown LDMs, define the maximum number of writes, but only
// make the first two consume resources. We are optimizing for the case
// where the operands are DPRs, and this determines the first eight
// types. The remaining eight types are filled to cover the case
// where the operands are SPRs.
SchedVar<A9LMUnknownPred, [A9WriteLMfp1, A9WriteLMfp2,
A9WriteLMfp3Hi, A9WriteLMfp4Hi,
A9WriteLMfp5Hi, A9WriteLMfp6Hi,
A9WriteLMfp7Hi, A9WriteLMfp8Hi,
A9WriteLMfp5Hi, A9WriteLMfp5Hi,
A9WriteLMfp6Hi, A9WriteLMfp6Hi,
A9WriteLMfp7Hi, A9WriteLMfp7Hi,
A9WriteLMfp8Hi, A9WriteLMfp8Hi]>]> {
let Variadic = 1;
}
// Distinguish between our multiple MI-level forms of the same
// VLDM/VSTM instructions.
def A9PreRA : SchedPredicate<
"TargetRegisterInfo::isVirtualRegister(MI->getOperand(0).getReg())">;
def A9PostRA : SchedPredicate<
"TargetRegisterInfo::isPhysicalRegister(MI->getOperand(0).getReg())">;
// VLDM represents all destination registers as a single register
// tuple, unlike LDM. So the number of write operands is not variadic.
def A9WriteLMfp : SchedWriteVariant<[
SchedVar<A9PreRA, [A9WriteLMfpPreRA]>,
SchedVar<A9PostRA, [A9WriteLMfpPostRA]>]>;
//===----------------------------------------------------------------------===//
// Resources for other (non-LDM/VLDM) Variants.
// These mov immediate writers are unconditionally expanded with
// additive latency.
def A9WriteI2 : WriteSequence<[A9WriteI, A9WriteI]>;
def A9WriteI2pc : WriteSequence<[A9WriteI, A9WriteI, WriteALU]>;
def A9WriteI2ld : WriteSequence<[A9WriteI, A9WriteI, A9WriteL]>;
// Some ALU operations can read loaded integer values one cycle early.
def A9ReadALU : SchedReadAdvance<1,
[A9WriteL, A9WriteLHi, A9WriteLsi, A9WriteLb, A9WriteLbsi,
A9WriteL1, A9WriteL2, A9WriteL3, A9WriteL4,
A9WriteL5, A9WriteL6, A9WriteL7, A9WriteL8,
A9WriteL1Hi, A9WriteL2Hi, A9WriteL3Hi, A9WriteL4Hi,
A9WriteL5Hi, A9WriteL6Hi, A9WriteL7Hi, A9WriteL8Hi]>;
// Read types for operands that are unconditionally read in cycle N
// after the instruction issues, decreases producer latency by N-1.
def A9Read2 : SchedReadAdvance<1>;
def A9Read3 : SchedReadAdvance<2>;
def A9Read4 : SchedReadAdvance<3>;
//===----------------------------------------------------------------------===//
// Map itinerary classes to scheduler read/write resources per operand.
//
// For ARM, we piggyback scheduler resources on the Itinerary classes
// to avoid perturbing the existing instruction definitions.
// This table follows the ARM Cortex-A9 Technical Reference Manuals,
// mostly in order.
def :ItinRW<[WriteALU], [IIC_iMOVi,IIC_iMOVr,IIC_iMOVsi,
IIC_iMVNi,IIC_iMVNsi,
IIC_iCMOVi,IIC_iCMOVr,IIC_iCMOVsi]>;
def :ItinRW<[WriteALU, A9ReadALU],[IIC_iMVNr]>;
def :ItinRW<[A9WriteIsr], [IIC_iMOVsr,IIC_iMVNsr,IIC_iCMOVsr]>;
def :ItinRW<[A9WriteI2], [IIC_iMOVix2,IIC_iCMOVix2]>;
def :ItinRW<[A9WriteI2pc], [IIC_iMOVix2addpc]>;
def :ItinRW<[A9WriteI2ld], [IIC_iMOVix2ld]>;
def :ItinRW<[WriteALU], [IIC_iBITi,IIC_iBITr,IIC_iUNAr,IIC_iTSTi,IIC_iTSTr]>;
def :ItinRW<[WriteALU, A9ReadALU], [IIC_iALUi, IIC_iCMPi, IIC_iCMPsi]>;
def :ItinRW<[WriteALU, A9ReadALU, A9ReadALU],[IIC_iALUr,IIC_iCMPr]>;
def :ItinRW<[WriteALUsi], [IIC_iBITsi,IIC_iUNAsi,IIC_iEXTr,IIC_iTSTsi]>;
def :ItinRW<[WriteALUsi, A9ReadALU], [IIC_iALUsi]>;
def :ItinRW<[WriteALUsi, ReadDefault, A9ReadALU], [IIC_iALUsir]>; // RSB
def :ItinRW<[A9WriteALUsr], [IIC_iBITsr,IIC_iTSTsr,IIC_iEXTAr,IIC_iEXTAsr]>;
def :ItinRW<[A9WriteALUsr, A9ReadALU], [IIC_iALUsr,IIC_iCMPsr]>;
// A9WriteHi ignored for MUL32.
def :ItinRW<[A9WriteM, A9WriteMHi], [IIC_iMUL32,IIC_iMAC32,
IIC_iMUL64,IIC_iMAC64]>;
// FIXME: SMLALxx needs itin classes
def :ItinRW<[A9WriteM16, A9WriteM16Hi], [IIC_iMUL16,IIC_iMAC16]>;
// TODO: For floating-point ops, we model the pipeline forwarding
// latencies here. WAW latencies are sometimes longer.
def :ItinRW<[A9WriteFMov], [IIC_fpSTAT, IIC_fpMOVIS, IIC_fpMOVID, IIC_fpMOVSI,
IIC_fpUNA32, IIC_fpUNA64,
IIC_fpCMP32, IIC_fpCMP64]>;
def :ItinRW<[A9WriteFMov, A9WriteFMov], [IIC_fpMOVDI]>;
def :ItinRW<[A9WriteF], [IIC_fpCVTSD, IIC_fpCVTDS, IIC_fpCVTSH, IIC_fpCVTHS,
IIC_fpCVTIS, IIC_fpCVTID, IIC_fpCVTSI, IIC_fpCVTDI,
IIC_fpALU32, IIC_fpALU64]>;
def :ItinRW<[A9WriteFMulS], [IIC_fpMUL32]>;
def :ItinRW<[A9WriteFMulD], [IIC_fpMUL64]>;
def :ItinRW<[A9WriteFMAS], [IIC_fpMAC32]>;
def :ItinRW<[A9WriteFMAD], [IIC_fpMAC64]>;
def :ItinRW<[A9WriteFDivS], [IIC_fpDIV32]>;
def :ItinRW<[A9WriteFDivD], [IIC_fpDIV64]>;
def :ItinRW<[A9WriteFSqrtS], [IIC_fpSQRT32]>;
def :ItinRW<[A9WriteFSqrtD], [IIC_fpSQRT64]>;
def :ItinRW<[A9WriteB], [IIC_Br]>;
// A9 PLD is processed in a dedicated unit.
def :ItinRW<[], [IIC_Preload]>;
// Note: We must assume that loads are aligned, since the machine
// model cannot know this statically and A9 ignores alignment hints.
// A9WriteAdr consumes AGU regardless address writeback. But it's
// latency is only relevant for users of an updated address.
def :ItinRW<[A9WriteL, A9WriteAdr], [IIC_iLoad_i,IIC_iLoad_r,
IIC_iLoad_iu,IIC_iLoad_ru]>;
def :ItinRW<[A9WriteLsi, A9WriteAdr], [IIC_iLoad_si,IIC_iLoad_siu]>;
def :ItinRW<[A9WriteLb, A9WriteAdr2], [IIC_iLoad_bh_i,IIC_iLoad_bh_r,
IIC_iLoad_bh_iu,IIC_iLoad_bh_ru]>;
def :ItinRW<[A9WriteLbsi, A9WriteAdr2], [IIC_iLoad_bh_si,IIC_iLoad_bh_siu]>;
def :ItinRW<[A9WriteL, A9WriteLHi, A9WriteAdr], [IIC_iLoad_d_i,IIC_iLoad_d_r,
IIC_iLoad_d_ru]>;
// Store either has no def operands, or the one def for address writeback.
def :ItinRW<[A9WriteAdr, A9WriteS], [IIC_iStore_i, IIC_iStore_r,
IIC_iStore_iu, IIC_iStore_ru,
IIC_iStore_d_i, IIC_iStore_d_r,
IIC_iStore_d_ru]>;
def :ItinRW<[A9WriteAdr2, A9WriteS], [IIC_iStore_si, IIC_iStore_siu,
IIC_iStore_bh_i, IIC_iStore_bh_r,
IIC_iStore_bh_iu, IIC_iStore_bh_ru]>;
def :ItinRW<[A9WriteAdr3, A9WriteS], [IIC_iStore_bh_si, IIC_iStore_bh_siu]>;
// A9WriteML will be expanded into a separate write for each def
// operand. Address generation consumes resources, but A9WriteLMAdr
// is listed after all def operands, so has no effective latency.
//
// Note: A9WriteLM expands into an even number of def operands. The
// actual number of def operands may be less by one.
def :ItinRW<[A9WriteLM, A9WriteLMAdr, A9WriteIssue], [IIC_iLoad_m, IIC_iPop]>;
// Load multiple with address writeback has an extra def operand in
// front of the loaded registers.
//
// Reuse the load-multiple variants for store-multiple because the
// resources are identical, For stores only the address writeback
// has a def operand so the WriteL latencies are unused.
def :ItinRW<[A9WriteLMAdr, A9WriteLM, A9WriteIssue], [IIC_iLoad_mu,
IIC_iStore_m,
IIC_iStore_mu]>;
def :ItinRW<[A9WriteLM, A9WriteLMAdr, A9WriteB], [IIC_iLoad_mBr, IIC_iPop_Br]>;
def :ItinRW<[A9WriteL, A9WriteAdr, WriteALU], [IIC_iLoadiALU]>;
def :ItinRW<[A9WriteLSfp, A9WriteAdr], [IIC_fpLoad32, IIC_fpLoad64]>;
def :ItinRW<[A9WriteLMfp, A9WriteLMAdr], [IIC_fpLoad_m]>;
def :ItinRW<[A9WriteLMAdr, A9WriteLMfp], [IIC_fpLoad_mu]>;
def :ItinRW<[A9WriteAdr, A9WriteLSfp], [IIC_fpStore32, IIC_fpStore64,
IIC_fpStore_m, IIC_fpStore_mu]>;
// Note: Unlike VLDM, VLD1 expects the writeback operand after the
// normal writes.
def :ItinRW<[A9WriteLfp1, A9WriteAdr1], [IIC_VLD1, IIC_VLD1u,
IIC_VLD1x2, IIC_VLD1x2u]>;
def :ItinRW<[A9WriteLfp2, A9WriteAdr2], [IIC_VLD1x3, IIC_VLD1x3u,
IIC_VLD1x4, IIC_VLD1x4u,
IIC_VLD4dup, IIC_VLD4dupu]>;
def :ItinRW<[A9WriteLfp1Mov, A9WriteAdr1], [IIC_VLD1dup, IIC_VLD1dupu,
IIC_VLD2, IIC_VLD2u,
IIC_VLD2dup, IIC_VLD2dupu]>;
def :ItinRW<[A9WriteLfp2Mov, A9WriteAdr1], [IIC_VLD1ln, IIC_VLD1lnu,
IIC_VLD2x2, IIC_VLD2x2u,
IIC_VLD2ln, IIC_VLD2lnu]>;
def :ItinRW<[A9WriteLfp3Mov, A9WriteAdr3], [IIC_VLD3, IIC_VLD3u,
IIC_VLD3dup, IIC_VLD3dupu]>;
def :ItinRW<[A9WriteLfp4Mov, A9WriteAdr4], [IIC_VLD4, IIC_VLD4u,
IIC_VLD4ln, IIC_VLD4lnu]>;
def :ItinRW<[A9WriteLfp5Mov, A9WriteAdr5], [IIC_VLD3ln, IIC_VLD3lnu]>;
// Vector stores use similar resources to vector loads, so use the
// same write types. The address write must be first for stores with
// address writeback.
def :ItinRW<[A9WriteAdr1, A9WriteLfp1], [IIC_VST1, IIC_VST1u,
IIC_VST1x2, IIC_VST1x2u,
IIC_VST1ln, IIC_VST1lnu,
IIC_VST2, IIC_VST2u,
IIC_VST2x2, IIC_VST2x2u,
IIC_VST2ln, IIC_VST2lnu]>;
def :ItinRW<[A9WriteAdr2, A9WriteLfp2], [IIC_VST1x3, IIC_VST1x3u,
IIC_VST1x4, IIC_VST1x4u,
IIC_VST3, IIC_VST3u,
IIC_VST3ln, IIC_VST3lnu,
IIC_VST4, IIC_VST4u,
IIC_VST4ln, IIC_VST4lnu]>;
// NEON moves.
def :ItinRW<[A9WriteV2], [IIC_VMOVSI, IIC_VMOVDI, IIC_VMOVD, IIC_VMOVQ]>;
def :ItinRW<[A9WriteV1], [IIC_VMOV, IIC_VMOVIS, IIC_VMOVID]>;
def :ItinRW<[A9WriteV3], [IIC_VMOVISL, IIC_VMOVN]>;
// NEON integer arithmetic
//
// VADD/VAND/VORR/VEOR/VBIC/VORN/VBIT/VBIF/VBSL
def :ItinRW<[A9WriteV3, A9Read2, A9Read2], [IIC_VBINiD, IIC_VBINiQ]>;
// VSUB/VMVN/VCLSD/VCLZD/VCNTD
def :ItinRW<[A9WriteV3, A9Read2], [IIC_VSUBiD, IIC_VSUBiQ, IIC_VCNTiD]>;
// VADDL/VSUBL/VNEG are mapped later under IIC_SHLi.
// ...
// VHADD/VRHADD/VQADD/VTST/VADH/VRADH
def :ItinRW<[A9WriteV4, A9Read2, A9Read2], [IIC_VBINi4D, IIC_VBINi4Q]>;
// VSBH/VRSBH/VHSUB/VQSUB/VABD/VCEQ/VCGE/VCGT/VMAX/VMIN/VPMAX/VPMIN/VABDL
def :ItinRW<[A9WriteV4, A9Read2], [IIC_VSUBi4D, IIC_VSUBi4Q]>;
// VQNEG/VQABS
def :ItinRW<[A9WriteV4], [IIC_VQUNAiD, IIC_VQUNAiQ]>;
// VABS
def :ItinRW<[A9WriteV4, A9Read2], [IIC_VUNAiD, IIC_VUNAiQ]>;
// VPADD/VPADDL are mapped later under IIC_SHLi.
// ...
// VCLSQ/VCLZQ/VCNTQ, takes two cycles.
def :ItinRW<[A9Write2V4, A9Read3], [IIC_VCNTiQ]>;
// VMOVimm/VMVNimm/VORRimm/VBICimm
def :ItinRW<[A9WriteV3], [IIC_VMOVImm]>;
def :ItinRW<[A9WriteV6, A9Read3, A9Read2], [IIC_VABAD, IIC_VABAQ]>;
def :ItinRW<[A9WriteV6, A9Read3], [IIC_VPALiD, IIC_VPALiQ]>;
// NEON integer multiply
//
// Note: these don't quite match the timing docs, but they do match
// the original A9 itinerary.
def :ItinRW<[A9WriteV6, A9Read2, A9Read2], [IIC_VMULi16D]>;
def :ItinRW<[A9WriteV7, A9Read2, A9Read2], [IIC_VMULi16Q]>;
def :ItinRW<[A9Write2V7, A9Read2], [IIC_VMULi32D]>;
def :ItinRW<[A9Write2V9, A9Read2], [IIC_VMULi32Q]>;
def :ItinRW<[A9WriteV6, A9Read3, A9Read2, A9Read2], [IIC_VMACi16D]>;
def :ItinRW<[A9WriteV7, A9Read3, A9Read2, A9Read2], [IIC_VMACi16Q]>;
def :ItinRW<[A9Write2V7, A9Read3, A9Read2], [IIC_VMACi32D]>;
def :ItinRW<[A9Write2V9, A9Read3, A9Read2], [IIC_VMACi32Q]>;
// NEON integer shift
// TODO: Q,Q,Q shifts should actually reserve FP for 2 cycles.
def :ItinRW<[A9WriteV3], [IIC_VSHLiD, IIC_VSHLiQ]>;
def :ItinRW<[A9WriteV4], [IIC_VSHLi4D, IIC_VSHLi4Q]>;
// NEON permute
def :ItinRW<[A9WriteV2, A9WriteV2], [IIC_VPERMD, IIC_VPERMQ, IIC_VEXTD]>;
def :ItinRW<[A9WriteV3, A9WriteV4, ReadDefault, A9Read2],
[IIC_VPERMQ3, IIC_VEXTQ]>;
def :ItinRW<[A9WriteV3, A9Read2], [IIC_VTB1]>;
def :ItinRW<[A9WriteV3, A9Read2, A9Read2], [IIC_VTB2]>;
def :ItinRW<[A9WriteV4, A9Read2, A9Read2, A9Read3], [IIC_VTB3]>;
def :ItinRW<[A9WriteV4, A9Read2, A9Read2, A9Read3, A9Read3], [IIC_VTB4]>;
def :ItinRW<[A9WriteV3, ReadDefault, A9Read2], [IIC_VTBX1]>;
def :ItinRW<[A9WriteV3, ReadDefault, A9Read2, A9Read2], [IIC_VTBX2]>;
def :ItinRW<[A9WriteV4, ReadDefault, A9Read2, A9Read2, A9Read3], [IIC_VTBX3]>;
def :ItinRW<[A9WriteV4, ReadDefault, A9Read2, A9Read2, A9Read3, A9Read3],
[IIC_VTBX4]>;
// NEON floating-point
def :ItinRW<[A9WriteV5, A9Read2, A9Read2], [IIC_VBIND]>;
def :ItinRW<[A9WriteV6, A9Read2, A9Read2], [IIC_VBINQ]>;
def :ItinRW<[A9WriteV5, A9Read2], [IIC_VUNAD, IIC_VFMULD]>;
def :ItinRW<[A9WriteV6, A9Read2], [IIC_VUNAQ, IIC_VFMULQ]>;
def :ItinRW<[A9WriteV9, A9Read3, A9Read2], [IIC_VMACD, IIC_VFMACD]>;
def :ItinRW<[A9WriteV10, A9Read3, A9Read2], [IIC_VMACQ, IIC_VFMACQ]>;
def :ItinRW<[A9WriteV9, A9Read2, A9Read2], [IIC_VRECSD]>;
def :ItinRW<[A9WriteV10, A9Read2, A9Read2], [IIC_VRECSQ]>;
// Map SchedRWs that are identical for cortexa9 to existing resources.
def : SchedAlias<WriteALU, A9WriteALU>;
def : SchedAlias<WriteALUsr, A9WriteALUsr>;
def : SchedAlias<WriteALUSsr, A9WriteALUsr>;
def : SchedAlias<ReadALU, A9ReadALU>;
def : SchedAlias<ReadALUsr, A9ReadALU>;
def : InstRW< [WriteALU],
(instregex "ANDri", "ORRri", "EORri", "BICri", "ANDrr", "ORRrr", "EORrr",
"BICrr")>;
def : InstRW< [WriteALUsi], (instregex "ANDrsi", "ORRrsi", "EORrsi", "BICrsi")>;
def : InstRW< [WriteALUsr], (instregex "ANDrsr", "ORRrsr", "EORrsr", "BICrsr")>;
def : SchedAlias<WriteCMP, A9WriteALU>;
def : SchedAlias<WriteCMPsi, A9WriteALU>;
def : SchedAlias<WriteCMPsr, A9WriteALU>;
def : InstRW< [A9WriteIsr], (instregex "MOVsr", "MOVsi", "MVNsr", "MOVCCsi",
"MOVCCsr")>;
def : InstRW< [WriteALU, A9ReadALU], (instregex "MVNr")>;
def : InstRW< [A9WriteI2], (instregex "MOVCCi32imm", "MOVi32imm",
"MOV_ga_dyn")>;
def : InstRW< [A9WriteI2pc], (instregex "MOV_ga_pcrel")>;
def : InstRW< [A9WriteI2ld], (instregex "MOV_ga_pcrel_ldr")>;
def : InstRW< [WriteALU], (instregex "SEL")>;
def : InstRW< [WriteALUsi], (instregex "BFC", "BFI", "UBFX", "SBFX")>;
def : InstRW< [A9WriteM],
(instregex "MUL", "MULv5", "SMMUL", "SMMULR", "MLA", "MLAv5", "MLS",
"SMMLA", "SMMLAR", "SMMLS", "SMMLSR")>;
def : InstRW< [A9WriteM, A9WriteMHi],
(instregex "SMULL", "SMULLv5", "UMULL", "UMULLv5", "SMLAL$", "UMLAL",
"UMAAL", "SMLALv5", "UMLALv5", "UMAALv5", "SMLALBB", "SMLALBT", "SMLALTB",
"SMLALTT")>;
// FIXME: These instructions used to have NoItinerary. Just copied the one from above.
def : InstRW< [A9WriteM, A9WriteMHi],
(instregex "SMLAD", "SMLADX", "SMLALD", "SMLALDX", "SMLSD", "SMLSDX",
"SMLSLD", "SMLLDX", "SMUAD", "SMUADX", "SMUSD", "SMUSDX")>;
def : InstRW<[A9WriteM16, A9WriteM16Hi],
(instregex "SMULBB", "SMULBT", "SMULTB", "SMULTT", "SMULWB", "SMULWT")>;
def : InstRW<[A9WriteM16, A9WriteM16Hi],
(instregex "SMLABB", "SMLABT", "SMLATB", "SMLATT", "SMLAWB", "SMLAWT")>;
def : InstRW<[A9WriteL], (instregex "LDRi12", "PICLDR$")>;
def : InstRW<[A9WriteLsi], (instregex "LDRrs")>;
def : InstRW<[A9WriteLb],
(instregex "LDRBi12", "PICLDRH", "PICLDRB", "PICLDRSH", "PICLDRSB",
"LDRH", "LDRSH", "LDRSB")>;
def : InstRW<[A9WriteLbsi], (instregex "LDRrs")>;
def : WriteRes<WriteDiv, []> { let Latency = 0; }
def : WriteRes<WriteBr, [A9UnitB]>;
def : WriteRes<WriteBrL, [A9UnitB]>;
def : WriteRes<WriteBrTbl, [A9UnitB]>;
def : WriteRes<WritePreLd, []>;
def : SchedAlias<WriteCvtFP, A9WriteF>;
def : WriteRes<WriteNoop, []> { let Latency = 0; let NumMicroOps = 0; }
} // SchedModel = CortexA9Model