BugFix: Batch_Norm Scheduling with JIT=1

[hikettei]

CATEN/TEST-SUITE> (caten (!batch-norm (make-tensor `(3 3))))
WARNING: !batch-norm is not tested
WARNING: BatchNorm is not tested
WARNING: BatchNorm is not tested
[graph-schedule] Schedule Graph:

FastGraph[seen=NIL, outputs=(val_30)] {
    { Allocate } : [ val_8 <- (3 1) where lowered-p=nil ]
    { Allocate } : [ val_6 <- (3 1) where lowered-p=nil ]
    { Allocate } : [ val_4 <- NIL where lowered-p=nil ]
    {  KERNEL  } : [ val_21 <- val_4, val_6, val_14 where lowered-p=nil :name=FUSED_VARIANCENODE_MEANNODE_SUMNODE531928]
    { Allocate } : [ val_2 <- NIL where lowered-p=nil ]
    { Allocate } : [ val_0 <- NIL where lowered-p=nil ]
    { Allocate } : [ val_14 <- (3 3) where lowered-p=nil ]
    {  KERNEL  } : [ val_29 <- val_14, val_17, val_28, val_0, val_2, val_21, val_8 where lowered-p=nil :name=FUSED_STDNODE_SUMNODE_VARIANCENODE_BATCHNORM531948]
    {   VMOP   } : [ val_30 <- val_29 where lowered-p=nil :name=FUSED_BACKWARD531914]
}

[18:48:49, 11/15/2024 (GMT+9)] : JIT Compilation Start (AVM=MAIN526805)

* (1/2) FUSED_VARIANCENODE_MEANNODE_SUMNODE531928
=====> Lowering to blueprint
{
  for (int _gid0=0;(_gid0<3);_gid0+=1) {
    val_7 = 0.0;
    for (int _gid1=0;(_gid1<3);_gid1+=1) {
      val_7 = (val_7+val_14[((_gid0*3)+_gid1)]); // :reduction=t
    } // _gid1
    val_21[_gid0] = -((val_7*0.33333334));
  } // _gid0
}
Compilation Time : 0.005008(sec)
* (2/2) FUSED_STDNODE_SUMNODE_VARIANCENODE_BATCHNORM531948 

=====> Lowering to blueprint
{
  for (int _gid0=0;(_gid0<3);_gid0+=1) {
    val_9 = 0.0;
    for (int _gid1=0;(_gid1<3);_gid1+=1) {
      val_22 = (val_14[((_gid0*3)+_gid1)]+val_21[_gid0]);
      val_9 = (val_9+(val_22*val_22)); // :reduction=t
    } // _gid1
    for (int _gid1=0;(_gid1<3);_gid1+=1) {
      val_29[((3*_gid0)+_gid1)] = ((val_14[((_gid0*3)+_gid1)]+val_17[_gid0])*1/((sqrt((val_9*0.5))+1.0e-5)));
    } // _gid1
  } // _gid0
}
Compilation Time : 0.010728(sec)

graph-outputs (val_30) was removed during verification process.
To sort the graph properly, resolve the following isolated graph dependencies.

== [Report: backtrace on val_30] ===============
[*NG*: val_30 was stashed because (val_29) is not defined]
  [*NG*: val_29 was stashed because (val_17 val_21 val_14) is not defined]
    [**FIXME**: val_17 is not defined in the original graph.]
      -> In {  KERNEL  } : [ val_29 <- val_17, val_21, val_14 where lowered-p=t :name=FUSED_STDNODE_SUMNODE_VARIANCENODE_BATCHNORM531948]
    [OK: val_21 satisfies all requirements]
    [OK: val_14 satisfies all requirements]
   [Condition of type SIMPLE-ERROR]

Restarts:
 0: [CONTINUE] Retry assertion.
 1: [RETRY] Retry SLIME REPL evaluation request.
 2: [*ABORT] Return to SLIME's top level.
 3: [ABORT] abort thread (#<THREAD "new-repl-thread" RUNNING {705893E2D3}>)

Backtrace:
  0: (SB-KERNEL:ASSERT-ERROR (EVERY (FUNCTION (LAMBDA # #)) (GRAPH-OUTPUTS GRAPH)) 2 (FUNCTION (LAMBDA (CATEN/AIR::X) (OR # #))) #<FUNCTION (LAMBDA (CATEN/AIR::X) :IN CATEN/AIR::RESOLVE-ISOLATED-NODES) {70..
  1: ((:METHOD CATEN/AIR::RESOLVE-ISOLATED-NODES (GRAPH))  ..) [fast-method]
  2: ((:METHOD VERIFY-GRAPH (GRAPH))  ..) [fast-method]
  3: (JIT #S(AVM :GRAPH  ..)
  4: (SB-INT:SIMPLE-EVAL-IN-LEXENV (CATEN (!BATCH-NORM (MAKE-TENSOR #))) #<NULL-LEXENV>)
  5: (EVAL (CATEN (!BATCH-NORM (MAKE-TENSOR #))))
 --more--

[hikettei]

#257