llvm.org GIT mirror llvm / stable lib / Target / AMDGPU / SISchedule.td
stable

Tree @stable (Download .tar.gz)

SISchedule.td @stable

f98f2ce
 
6b54768
 
 
f98f2ce
 
 
aafca11
f98f2ce
 
 
0791b66
 
 
 
 
 
aafca11
 
 
 
 
 
55ec9f2
f98f2ce
aafca11
 
 
4fc498f
aafca11
 
 
4fc498f
 
 
 
aafca11
 
f43d543
 
 
4fc498f
 
 
 
 
 
 
7725fd8
00a96df
4a5c408
 
 
 
7725fd8
6ab99c7
4dc4396
 
 
 
 
eab2869
aafca11
7725fd8
 
f43d543
aafca11
 
7725fd8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
aafca11
f43d543
 
 
aafca11
 
 
 
 
 
 
 
 
 
 
4fc498f
aafca11
 
 
 
7725fd8
 
 
 
 
 
55ec9f2
aafca11
 
4fc498f
aafca11
 
 
0791b66
 
 
 
 
 
aafca11
 
 
 
 
 
 
 
f43d543
aafca11
0791b66
 
aafca11
 
 
 
 
 
 
 
 
f43d543
aafca11
0791b66
 
aafca11
f43d543
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
//===-- SISchedule.td - SI Scheduling definitons -------------------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// MachineModel definitions for Southern Islands (SI)
//
//===----------------------------------------------------------------------===//

def : PredicateProlog<[{
  const SIInstrInfo *TII =
    static_cast<const SIInstrInfo*>(SchedModel->getInstrInfo());
  (void)TII;
}]>;

def WriteBranch : SchedWrite;
def WriteExport : SchedWrite;
def WriteLDS    : SchedWrite;
def WriteSALU   : SchedWrite;
def WriteSMEM   : SchedWrite;
def WriteVMEM   : SchedWrite;
def WriteBarrier : SchedWrite;

// Vector ALU instructions
def Write32Bit         : SchedWrite;
def WriteQuarterRate32 : SchedWrite;
def WriteFullOrQuarterRate32 : SchedWrite;

def WriteFloatFMA   : SchedWrite;

// Slow quarter rate f64 instruction.
def WriteDouble : SchedWrite;

// half rate f64 instruction (same as v_add_f64)
def WriteDoubleAdd  : SchedWrite;

// Conversion to or from f64 instruction
def WriteDoubleCvt  : SchedWrite;

// Half rate 64-bit instructions.
def Write64Bit : SchedWrite;

// FIXME: Should there be a class for instructions which are VALU
// instructions and have VALU rates, but write to the SALU (i.e. VOPC
// instructions)

class SISchedMachineModel : SchedMachineModel {
  let CompleteModel = 0;
  // MicroOpBufferSize = 1 means that instructions will always be added
  // the ready queue when they become available.  This exposes them
  // to the register pressure analysis.
  let MicroOpBufferSize = 1;
  let IssueWidth = 1;
  let PostRAScheduler = 1;

  // FIXME:Approximate 2 * branch cost.  Try to hack around bad
  // early-ifcvt heuristics. These need improvement to avoid the OOE
  // heuristics.
  int MispredictPenalty = 20;
}

def SIFullSpeedModel : SISchedMachineModel;
def SIQuarterSpeedModel : SISchedMachineModel;
def GFX10SpeedModel : SISchedMachineModel;

// XXX: Are the resource counts correct?
def HWBranch : ProcResource<1> {
  let BufferSize = 1;
}
def HWExport : ProcResource<1> {
  let BufferSize = 7; // Taken from S_WAITCNT
}
def HWLGKM   : ProcResource<1> {
  let BufferSize = 31;  // Taken from S_WAITCNT
}
def HWSALU   : ProcResource<1> {
  let BufferSize = 1;
}
def HWVMEM   : ProcResource<1> {
  let BufferSize = 15;  // Taken from S_WAITCNT
}
def HWVALU   : ProcResource<1> {
  let BufferSize = 1;
}
def HWRC   : ProcResource<1> { // Register destination cache
  let BufferSize = 1;
}

class HWWriteRes<SchedWrite write, list<ProcResourceKind> resources,
                 int latency> : WriteRes<write, resources> {
  let Latency = latency;
}

class HWVALUWriteRes<SchedWrite write, int latency> :
  HWWriteRes<write, [HWVALU], latency>;


// The latency numbers are taken from AMD Accelerated Parallel Processing
// guide. They may not be accurate.

// The latency values are 1 / (operations / cycle) / 4.
multiclass SICommonWriteRes {

  def : HWWriteRes<WriteBranch,  [HWBranch], 8>;
  def : HWWriteRes<WriteExport,  [HWExport], 4>;
  def : HWWriteRes<WriteLDS,     [HWLGKM],   5>; // Can be between 2 and 64
  def : HWWriteRes<WriteSALU,    [HWSALU],   1>;
  def : HWWriteRes<WriteSMEM,    [HWLGKM],   5>;
  def : HWWriteRes<WriteVMEM,    [HWVMEM],   80>;
  def : HWWriteRes<WriteBarrier, [HWBranch], 500>; // XXX: Guessed ???

  def : HWVALUWriteRes<Write32Bit,         1>;
  def : HWVALUWriteRes<Write64Bit,         2>;
  def : HWVALUWriteRes<WriteQuarterRate32, 4>;
}

def PredIsVGPR32Copy : SchedPredicate<[{TII->isVGPRCopy(*MI) && TII->getOpSize(*MI, 0) <= 32}]>;
def PredIsVGPR64Copy : SchedPredicate<[{TII->isVGPRCopy(*MI) && TII->getOpSize(*MI, 0) > 32}]>;
def WriteCopy : SchedWriteVariant<[
    SchedVar<PredIsVGPR32Copy, [Write32Bit]>,
    SchedVar<PredIsVGPR64Copy, [Write64Bit]>,
    SchedVar<NoSchedPred, [WriteSALU]>]>;

let SchedModel = SIFullSpeedModel in {

defm : SICommonWriteRes;

def : HWVALUWriteRes<WriteFloatFMA,   1>;
def : HWVALUWriteRes<WriteDouble,     4>;
def : HWVALUWriteRes<WriteDoubleAdd,  2>;
def : HWVALUWriteRes<WriteDoubleCvt,  4>;

def : InstRW<[WriteCopy], (instrs COPY)>;

} // End SchedModel = SIFullSpeedModel

let SchedModel = SIQuarterSpeedModel in {

defm : SICommonWriteRes;

def : HWVALUWriteRes<WriteFloatFMA, 16>;
def : HWVALUWriteRes<WriteDouble,   16>;
def : HWVALUWriteRes<WriteDoubleAdd, 8>;
def : HWVALUWriteRes<WriteDoubleCvt, 4>;

def : InstRW<[WriteCopy], (instrs COPY)>;

}  // End SchedModel = SIQuarterSpeedModel

let SchedModel = GFX10SpeedModel in {

// The latency values are 1 / (operations / cycle).
// Add 1 stall cycle for VGPR read.
def : HWWriteRes<Write32Bit,         [HWVALU, HWRC],   5>;
def : HWWriteRes<Write64Bit,         [HWVALU, HWRC],   9>;
def : HWWriteRes<WriteQuarterRate32, [HWVALU, HWRC],   17>;
def : HWWriteRes<WriteFloatFMA,      [HWVALU, HWRC],   5>;
def : HWWriteRes<WriteDouble,        [HWVALU, HWRC],   17>;
def : HWWriteRes<WriteDoubleAdd,     [HWVALU, HWRC],   17>;
def : HWWriteRes<WriteDoubleCvt,     [HWVALU, HWRC],   17>;

def : HWWriteRes<WriteBranch,        [HWBranch],       32>;
def : HWWriteRes<WriteExport,        [HWExport, HWRC], 16>;
def : HWWriteRes<WriteLDS,           [HWLGKM,   HWRC], 20>;
def : HWWriteRes<WriteSALU,          [HWSALU,   HWRC], 5>;
def : HWWriteRes<WriteSMEM,          [HWLGKM,   HWRC], 20>;
def : HWWriteRes<WriteVMEM,          [HWVMEM,   HWRC], 320>;
def : HWWriteRes<WriteBarrier,       [HWBranch],       2000>;

def : InstRW<[WriteCopy], (instrs COPY)>;

}  // End SchedModel = GFX10SpeedModel