X-Git-Url: http://plrg.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FTarget%2FARM%2FARMSchedule.td;h=528c4ec7378194972aef5e8bec58033e00f72be9;hb=c3f2ad087923d6570f0e3a6ea1e004a892a5ef15;hp=e385ec7f6a26277b94523be02a6823eb161d7e36;hpb=7e1bf305cfecbaee859405468b769650efe68f1a;p=oota-llvm.git

diff --git a/lib/Target/ARM/ARMSchedule.td b/lib/Target/ARM/ARMSchedule.td
index e385ec7f6a2..528c4ec7378 100644
--- a/lib/Target/ARM/ARMSchedule.td
+++ b/lib/Target/ARM/ARMSchedule.td
@@ -1,11 +1,100 @@
-//===- ARMSchedule.td - ARM Scheduling Definitions ---------*- tablegen -*-===//
-// 
+//===-- ARMSchedule.td - ARM Scheduling Definitions --------*- tablegen -*-===//
+//
 //                     The LLVM Compiler Infrastructure
 //
 // This file is distributed under the University of Illinois Open Source
 // License. See LICENSE.TXT for details.
-// 
+//
+//===----------------------------------------------------------------------===//
 //===----------------------------------------------------------------------===//
+// Instruction scheduling annotations for out-of-order CPUs.
+// These annotations are independent of the itinerary class defined below.
+// Here we define the subtarget independent read/write per-operand resources.
+// The subtarget schedule definitions will then map these to the subtarget's
+// resource usages.
+// For example:
+// The instruction cycle timings table might contain an entry for an operation
+// like the following:
+// Rd <- ADD Rn, Rm, <shift> Rs
+//  Uops | Latency from register | Uops - resource requirements - latency
+//  2    | Rn: 1 Rm: 4 Rs: 4     | uop T0, Rm, Rs - P01 - 3
+//       |                       | uopc Rd, Rn, T0 -  P01 - 1
+// This is telling us that the result will be available in destination register
+// Rd after a minimum of three cycles after the result in Rm and Rs is available
+// and one cycle after the result in Rn is available. The micro-ops can execute
+// on resource P01.
+// To model this, we need to express that we need to dispatch two micro-ops,
+// that the resource P01 is needed and that the latency to Rn is different than
+// the latency to Rm and Rs. The scheduler can decrease Rn's producer latency by
+// two.
+// We will do this by assigning (abstract) resources to register defs/uses.
+// ARMSchedule.td:
+//   def WriteALUsr : SchedWrite;
+//   def ReadAdvanceALUsr : ScheRead;
+//
+// ARMInstrInfo.td:
+//   def ADDrs : I<>, Sched<[WriteALUsr, ReadAdvanceALUsr, ReadDefault,
+//                           ReadDefault]> { ...}
+// ReadAdvance read resources allow us to define "pipeline by-passes" or
+// shorter latencies to certain registers as needed in the example above.
+// The "ReadDefault" can be omitted.
+// Next, the subtarget td file assigns resources to the abstract resources
+// defined here.
+// ARMScheduleSubtarget.td:
+//  // Resources.
+//  def P01 : ProcResource<3>; // ALU unit (3 of it).
+//  ...
+//  // Resource usages.
+//  def : WriteRes<WriteALUsr, [P01, P01]> {
+//    Latency = 4; // Latency of 4.
+//    NumMicroOps = 2; // Dispatch 2 micro-ops.
+//    // The two instances of resource P01 are occupied for one cycle. It is one
+//    // cycle because these resources happen to be pipelined.
+//    ResourceCycles = [1, 1];
+//  }
+//  def : ReadAdvance<ReadAdvanceALUsr, 3>;
+
+// Basic ALU operation.
+def WriteALU : SchedWrite;
+def ReadALU : SchedRead;
+
+// Basic ALU with shifts.
+def WriteALUsi : SchedWrite; // Shift by immediate.
+def WriteALUsr : SchedWrite; // Shift by register.
+def WriteALUSsr : SchedWrite; // Shift by register (flag setting).
+def ReadALUsr : SchedRead; // Some operands are read later.
+
+// Compares.
+def WriteCMP : SchedWrite;
+def WriteCMPsi : SchedWrite;
+def WriteCMPsr : SchedWrite;
+
+// Division.
+def WriteDiv : SchedWrite;
+
+// Loads.
+def WriteLd : SchedWrite;
+def WritePreLd : SchedWrite;
+
+// Branches.
+def WriteBr : SchedWrite;
+def WriteBrL : SchedWrite;
+def WriteBrTbl : SchedWrite;
+
+// Fixpoint conversions.
+def WriteCvtFP : SchedWrite;
+
+// Noop.
+def WriteNoop : SchedWrite;
+
+// Define TII for use in SchedVariant Predicates.
+def : PredicateProlog<[{
+  const ARMBaseInstrInfo *TII =
+    static_cast<const ARMBaseInstrInfo*>(SchedModel->getInstrInfo());
+  (void)TII;
+}]>;
+
+def IsPredicatedPred : SchedPredicate<[{TII->isPredicated(MI)}]>;
 
 //===----------------------------------------------------------------------===//
 // Instruction Itinerary classes used for ARM
@@ -14,6 +103,7 @@ def IIC_iALUx      : InstrItinClass;
 def IIC_iALUi      : InstrItinClass;
 def IIC_iALUr      : InstrItinClass;
 def IIC_iALUsi     : InstrItinClass;
+def IIC_iALUsir    : InstrItinClass;
 def IIC_iALUsr     : InstrItinClass;
 def IIC_iBITi      : InstrItinClass;
 def IIC_iBITr      : InstrItinClass;
@@ -28,37 +118,72 @@ def IIC_iCMPi      : InstrItinClass;
 def IIC_iCMPr      : InstrItinClass;
 def IIC_iCMPsi     : InstrItinClass;
 def IIC_iCMPsr     : InstrItinClass;
+def IIC_iTSTi      : InstrItinClass;
+def IIC_iTSTr      : InstrItinClass;
+def IIC_iTSTsi     : InstrItinClass;
+def IIC_iTSTsr     : InstrItinClass;
 def IIC_iMOVi      : InstrItinClass;
-def IIC_iMOVix2    : InstrItinClass;
 def IIC_iMOVr      : InstrItinClass;
 def IIC_iMOVsi     : InstrItinClass;
 def IIC_iMOVsr     : InstrItinClass;
+def IIC_iMOVix2    : InstrItinClass;
+def IIC_iMOVix2addpc : InstrItinClass;
+def IIC_iMOVix2ld  : InstrItinClass;
+def IIC_iMVNi      : InstrItinClass;
+def IIC_iMVNr      : InstrItinClass;
+def IIC_iMVNsi     : InstrItinClass;
+def IIC_iMVNsr     : InstrItinClass;
 def IIC_iCMOVi     : InstrItinClass;
 def IIC_iCMOVr     : InstrItinClass;
 def IIC_iCMOVsi    : InstrItinClass;
 def IIC_iCMOVsr    : InstrItinClass;
+def IIC_iCMOVix2   : InstrItinClass;
 def IIC_iMUL16     : InstrItinClass;
 def IIC_iMAC16     : InstrItinClass;
 def IIC_iMUL32     : InstrItinClass;
 def IIC_iMAC32     : InstrItinClass;
 def IIC_iMUL64     : InstrItinClass;
 def IIC_iMAC64     : InstrItinClass;
-def IIC_iLoadi     : InstrItinClass;
-def IIC_iLoadr     : InstrItinClass;
-def IIC_iLoadsi    : InstrItinClass;
-def IIC_iLoadiu    : InstrItinClass;
-def IIC_iLoadru    : InstrItinClass;
-def IIC_iLoadsiu   : InstrItinClass;
-def IIC_iLoadm     : InstrItinClass<0>;  // micro-coded
-def IIC_iLoadmBr   : InstrItinClass<0>;  // micro-coded
+def IIC_iDIV     : InstrItinClass;
+def IIC_iLoad_i    : InstrItinClass;
+def IIC_iLoad_r    : InstrItinClass;
+def IIC_iLoad_si   : InstrItinClass;
+def IIC_iLoad_iu   : InstrItinClass;
+def IIC_iLoad_ru   : InstrItinClass;
+def IIC_iLoad_siu  : InstrItinClass;
+def IIC_iLoad_bh_i   : InstrItinClass;
+def IIC_iLoad_bh_r   : InstrItinClass;
+def IIC_iLoad_bh_si  : InstrItinClass;
+def IIC_iLoad_bh_iu  : InstrItinClass;
+def IIC_iLoad_bh_ru  : InstrItinClass;
+def IIC_iLoad_bh_siu : InstrItinClass;
+def IIC_iLoad_d_i  : InstrItinClass;
+def IIC_iLoad_d_r  : InstrItinClass;
+def IIC_iLoad_d_ru : InstrItinClass;
+def IIC_iLoad_m    : InstrItinClass;
+def IIC_iLoad_mu   : InstrItinClass;
+def IIC_iLoad_mBr  : InstrItinClass;
+def IIC_iPop       : InstrItinClass;
+def IIC_iPop_Br    : InstrItinClass;
 def IIC_iLoadiALU  : InstrItinClass;
-def IIC_iStorei    : InstrItinClass;
-def IIC_iStorer    : InstrItinClass;
-def IIC_iStoresi   : InstrItinClass;
-def IIC_iStoreiu   : InstrItinClass;
-def IIC_iStoreru   : InstrItinClass;
-def IIC_iStoresiu  : InstrItinClass;
-def IIC_iStorem    : InstrItinClass<0>;  // micro-coded
+def IIC_iStore_i   : InstrItinClass;
+def IIC_iStore_r   : InstrItinClass;
+def IIC_iStore_si  : InstrItinClass;
+def IIC_iStore_iu  : InstrItinClass;
+def IIC_iStore_ru  : InstrItinClass;
+def IIC_iStore_siu : InstrItinClass;
+def IIC_iStore_bh_i   : InstrItinClass;
+def IIC_iStore_bh_r   : InstrItinClass;
+def IIC_iStore_bh_si  : InstrItinClass;
+def IIC_iStore_bh_iu  : InstrItinClass;
+def IIC_iStore_bh_ru  : InstrItinClass;
+def IIC_iStore_bh_siu : InstrItinClass;
+def IIC_iStore_d_i   : InstrItinClass;
+def IIC_iStore_d_r   : InstrItinClass;
+def IIC_iStore_d_ru  : InstrItinClass;
+def IIC_iStore_m   : InstrItinClass;
+def IIC_iStore_mu  : InstrItinClass;
+def IIC_Preload    : InstrItinClass;
 def IIC_Br         : InstrItinClass;
 def IIC_fpSTAT     : InstrItinClass;
 def IIC_fpUNA32    : InstrItinClass;
@@ -83,25 +208,84 @@ def IIC_fpMUL32    : InstrItinClass;
 def IIC_fpMUL64    : InstrItinClass;
 def IIC_fpMAC32    : InstrItinClass;
 def IIC_fpMAC64    : InstrItinClass;
+def IIC_fpFMAC32   : InstrItinClass;
+def IIC_fpFMAC64   : InstrItinClass;
 def IIC_fpDIV32    : InstrItinClass;
 def IIC_fpDIV64    : InstrItinClass;
 def IIC_fpSQRT32   : InstrItinClass;
 def IIC_fpSQRT64   : InstrItinClass;
 def IIC_fpLoad32   : InstrItinClass;
 def IIC_fpLoad64   : InstrItinClass;
-def IIC_fpLoadm    : InstrItinClass<0>;  // micro-coded
+def IIC_fpLoad_m   : InstrItinClass;
+def IIC_fpLoad_mu  : InstrItinClass;
 def IIC_fpStore32  : InstrItinClass;
 def IIC_fpStore64  : InstrItinClass;
-def IIC_fpStorem   : InstrItinClass<0>;  // micro-coded
+def IIC_fpStore_m  : InstrItinClass;
+def IIC_fpStore_mu : InstrItinClass;
 def IIC_VLD1       : InstrItinClass;
+def IIC_VLD1x2     : InstrItinClass;
+def IIC_VLD1x3     : InstrItinClass;
+def IIC_VLD1x4     : InstrItinClass;
+def IIC_VLD1u      : InstrItinClass;
+def IIC_VLD1x2u    : InstrItinClass;
+def IIC_VLD1x3u    : InstrItinClass;
+def IIC_VLD1x4u    : InstrItinClass;
+def IIC_VLD1ln     : InstrItinClass;
+def IIC_VLD1lnu    : InstrItinClass;
+def IIC_VLD1dup    : InstrItinClass;
+def IIC_VLD1dupu   : InstrItinClass;
 def IIC_VLD2       : InstrItinClass;
+def IIC_VLD2x2     : InstrItinClass;
+def IIC_VLD2u      : InstrItinClass;
+def IIC_VLD2x2u    : InstrItinClass;
+def IIC_VLD2ln     : InstrItinClass;
+def IIC_VLD2lnu    : InstrItinClass;
+def IIC_VLD2dup    : InstrItinClass;
+def IIC_VLD2dupu   : InstrItinClass;
 def IIC_VLD3       : InstrItinClass;
+def IIC_VLD3ln     : InstrItinClass;
+def IIC_VLD3u      : InstrItinClass;
+def IIC_VLD3lnu    : InstrItinClass;
+def IIC_VLD3dup    : InstrItinClass;
+def IIC_VLD3dupu   : InstrItinClass;
 def IIC_VLD4       : InstrItinClass;
-def IIC_VST        : InstrItinClass;
+def IIC_VLD4ln     : InstrItinClass;
+def IIC_VLD4u      : InstrItinClass;
+def IIC_VLD4lnu    : InstrItinClass;
+def IIC_VLD4dup    : InstrItinClass;
+def IIC_VLD4dupu   : InstrItinClass;
+def IIC_VST1       : InstrItinClass;
+def IIC_VST1x2     : InstrItinClass;
+def IIC_VST1x3     : InstrItinClass;
+def IIC_VST1x4     : InstrItinClass;
+def IIC_VST1u      : InstrItinClass;
+def IIC_VST1x2u    : InstrItinClass;
+def IIC_VST1x3u    : InstrItinClass;
+def IIC_VST1x4u    : InstrItinClass;
+def IIC_VST1ln     : InstrItinClass;
+def IIC_VST1lnu    : InstrItinClass;
+def IIC_VST2       : InstrItinClass;
+def IIC_VST2x2     : InstrItinClass;
+def IIC_VST2u      : InstrItinClass;
+def IIC_VST2x2u    : InstrItinClass;
+def IIC_VST2ln     : InstrItinClass;
+def IIC_VST2lnu    : InstrItinClass;
+def IIC_VST3       : InstrItinClass;
+def IIC_VST3u      : InstrItinClass;
+def IIC_VST3ln     : InstrItinClass;
+def IIC_VST3lnu    : InstrItinClass;
+def IIC_VST4       : InstrItinClass;
+def IIC_VST4u      : InstrItinClass;
+def IIC_VST4ln     : InstrItinClass;
+def IIC_VST4lnu    : InstrItinClass;
 def IIC_VUNAD      : InstrItinClass;
 def IIC_VUNAQ      : InstrItinClass;
 def IIC_VBIND      : InstrItinClass;
 def IIC_VBINQ      : InstrItinClass;
+def IIC_VPBIND     : InstrItinClass;
+def IIC_VFMULD     : InstrItinClass;
+def IIC_VFMULQ     : InstrItinClass;
+def IIC_VMOV       : InstrItinClass;
 def IIC_VMOVImm    : InstrItinClass;
 def IIC_VMOVD      : InstrItinClass;
 def IIC_VMOVQ      : InstrItinClass;
@@ -110,11 +294,14 @@ def IIC_VMOVID     : InstrItinClass;
 def IIC_VMOVISL    : InstrItinClass;
 def IIC_VMOVSI     : InstrItinClass;
 def IIC_VMOVDI     : InstrItinClass;
+def IIC_VMOVN      : InstrItinClass;
 def IIC_VPERMD     : InstrItinClass;
 def IIC_VPERMQ     : InstrItinClass;
 def IIC_VPERMQ3    : InstrItinClass;
 def IIC_VMACD      : InstrItinClass;
 def IIC_VMACQ      : InstrItinClass;
+def IIC_VFMACD     : InstrItinClass;
+def IIC_VFMACQ     : InstrItinClass;
 def IIC_VRECSD     : InstrItinClass;
 def IIC_VRECSQ     : InstrItinClass;
 def IIC_VCNTiD     : InstrItinClass;
@@ -161,8 +348,7 @@ def IIC_VTBX4      : InstrItinClass;
 //===----------------------------------------------------------------------===//
 // Processor instruction itineraries.
 
-def GenericItineraries : ProcessorItineraries<[], [], []>;
-
 include "ARMScheduleV6.td"
 include "ARMScheduleA8.td"
 include "ARMScheduleA9.td"
+include "ARMScheduleSwift.td"