Fixes to direct linearize

  * Fix a bug in pjit linearization rule
  * Handle multiple results and zeros in fallback rule
  * Handle `has_aux`
  * Implement process_custom_vjp_call

Author: dougalm

jax/_src/interpreters/ad.py

@@ -143,21 +143,31 @@ def new_arg(primal_aval, nz):
 
 def direct_linearize(traceable, *primals, **kwargs):
   has_aux = kwargs.pop('has_aux', False)
-  assert not has_aux
   with core.take_current_trace() as parent_trace:
     tangent_trace = pe.DynamicJaxprTrace()
     tangents = [tangent_trace.new_arg(get_aval(p).to_tangent_aval()) for p in primals]
     linearize_trace = LinearizeTrace(parent_trace, tangent_trace)
     tracers = [LinearizeTracer(linearize_trace, p, t) for p, t in zip(primals, tangents)]
     with core.set_current_trace(linearize_trace):
-      ans = traceable.call_wrapped(*tracers)
-
+      if has_aux:
+        ans, aux = traceable.call_wrapped(*tracers)
+        aux_primals = [x.primal
+                       if isinstance(x, LinearizeTracer)
+                       and x._trace.tag is linearize_trace.tag
+                       else x for x in aux]
+      else:
+        ans = traceable.call_wrapped(*tracers)
+        aux = None
   out_primals, out_tangents = unzip2(map(linearize_trace.to_primal_tangent_pair, ans))
+  out_tangents = map(instantiate_zeros, out_tangents)
   out_tangents = map(tangent_trace.to_jaxpr_tracer, out_tangents)
   jaxpr, consts, attrs_tracked = tangent_trace.to_jaxpr(out_tangents)
   out_tangents_pvals = [pe.PartialVal.unknown(core.get_aval(t)) for t in out_tangents]
   del attrs_tracked # TODO: attrs
-  return out_primals, out_tangents_pvals, jaxpr, consts
+  if has_aux:
+    return out_primals, out_tangents_pvals, jaxpr, consts, aux_primals
+  else:
+    return out_primals, out_tangents_pvals, jaxpr, consts
 
 def linearize(traceable, *primals, **kwargs):
   if config.use_direct_linearize.value:
@@ -532,22 +542,45 @@ def to_primal_tangent_pair(self, val):
 
   def process_primitive(self, primitive, args, params):
     primals_in, tangents_in = unzip2(map(self.to_primal_tangent_pair, args))
-    tangent_nonzeros = [type(t) is not Zero for t in tangents_in]
+    tangent_nzs = [type(t) is not Zero for t in tangents_in]
     if all(type(t) is Zero for t in tangents_in):
       return primitive.bind_with_trace(self.parent_trace, primals_in, params)
-    lin = primitive_linearizations.get(primitive)
-    if lin is None:
-      lin = partial(fallback_linearize_rule, primitive)
+    fallback = partial(fallback_linearize_rule, primitive)
+    lin = primitive_linearizations.get(primitive, fallback)
     with core.set_current_trace(self.parent_trace):
-      primal_out, tangent_nonzeros_out, residuals, linearized = lin(
-          tangent_nonzeros, *primals_in, **params)
+      primal_out, tangent_nzs_out, residuals, linearized = lin(
+          tangent_nzs, *primals_in, **params)
     with core.set_current_trace(self.tangent_trace):
       tangent_out = linearized(residuals, *tangents_in)
     if primitive.multiple_results:
       return [maybe_linearize_tracer(self, x, nz, t)
-              for x, nz, t in zip(primal_out, tangent_nonzeros, tangent_out)]
+              for x, nz, t in zip(primal_out, tangent_nzs_out, tangent_out)]
     else:
-      return maybe_linearize_tracer(self, primal_out, tangent_nonzeros, tangent_out)
+      return maybe_linearize_tracer(self, primal_out, tangent_nzs_out, tangent_out)
+
+  def process_custom_vjp_call(self, prim, fun, fwd, bwd, tracers, out_trees,
+                              symbolic_zeros):
+    primals_in, tangents_in = unzip2(map(self.to_primal_tangent_pair, tracers))
+    if all(type(t) is Zero for t in tangents_in):
+      return prim.bind_with_trace(self.parent_trace,
+                                  (fun, fwd, bwd, *primals_in),
+                                  dict(out_trees=out_trees, symbolic_zeros=symbolic_zeros))
+    fwd_in = [(p, type(t) is not Zero) for p, t in zip(primals_in, tangents_in)]
+    fwd_in = [x for pair in fwd_in for x in pair]   # flatten
+    with core.set_current_trace(self.parent_trace):
+      res_and_primals_out = fwd.call_wrapped(*fwd_in)
+
+    _, res_tree = out_trees()
+    res, primals_out = split_list(res_and_primals_out, [res_tree.num_leaves])
+    avals_out = [core.get_aval(x).to_tangent_aval() for x in primals_out]
+
+    with core.set_current_trace(self.tangent_trace):
+      tangents_in = map(instantiate_zeros, tangents_in)
+      tangents_out = custom_lin_p.bind(
+        *res, *tangents_in, num_res=res_tree.num_leaves, bwd=bwd,
+        out_avals=avals_out, symbolic_zeros=symbolic_zeros)
+    tangent_nzs_out = [type(t) is not Zero for t in tangents_out]
+    return map(partial(maybe_linearize_tracer, self), primals_out, tangent_nzs_out, tangents_out)
 
 def maybe_linearize_tracer(trace, primal, is_nonzero, tangent):
   if is_nonzero:
@@ -557,21 +590,50 @@ def maybe_linearize_tracer(trace, primal, is_nonzero, tangent):
     assert type(tangent) is Zero
     return primal
 
-def fallback_linearize_rule(prim, _, *args, **kwargs):
-  assert not prim.multiple_results
-
-  def call_prim(*args_):
-    return [prim.bind(*args_, **kwargs)]
-
-  with config.use_direct_linearize(False):
-    (out_primal,), (out_tangent_pval,), jaxpr, consts, *_maybe_aux = linearize(
-        lu.wrap_init(call_prim), *args, **kwargs)
+def fallback_linearize_rule(prim, nonzeros, *primals, **params):
+  jvp = primitive_jvps.get(prim)
+  if not jvp:
+    msg = f"Differentiation rule for '{prim}' not implemented"
+    raise NotImplementedError(msg)
+  current_name_stack = source_info_util.current_name_stack()
+  with core.take_current_trace() as parent_trace:
+    trace = pe.JaxprTrace(parent_trace, current_name_stack, core.TraceTag())
+    tangent_avals = [get_aval(p).to_tangent_aval() for p in primals]
+    tangent_args = [trace.new_arg(pe.PartialVal.unknown(aval)) if nz else Zero(aval)
+                    for aval, nz in zip(tangent_avals, nonzeros)]
+    with core.set_current_trace(trace):
+      out_primals, out_tangents = jvp(primals, tangent_args, **params)
+
+    if not prim.multiple_results:
+      out_primals = [out_primals]
+      out_tangents = [out_tangents]
+
+    out_primals = [trace.to_jaxpr_tracer(p).pval.get_known() for p in out_primals]
+    out_nzs = [type(r) is not Zero for r in out_tangents]
+    out_tangent_avals = [get_aval(p).to_tangent_aval() for p in out_primals]
+    out_nz_tracers = [trace.to_jaxpr_tracer(r) for (r, nz) in zip(out_tangents, out_nzs) if nz]
+    in_tracers = [t for t in tangent_args if type(t) is not Zero]
+    jaxpr, out_consts, _ = pe.tracers_to_jaxpr(in_tracers, out_nz_tracers)
+
+    def linearized(residuals, *tangents):
+      nz_tangents_in = [t for (t, nz) in zip(tangents, nonzeros) if nz]
+      nz_tangents_out = core.eval_jaxpr(jaxpr, residuals, *nz_tangents_in)
+      nz_tangents_out_iter = iter(nz_tangents_out)
+      all_out_tangents = [next(nz_tangents_out_iter) if nz else Zero(aval)
+                          for (aval, nz) in zip(out_tangent_avals, out_nzs)]
+      if prim.multiple_results:
+        return all_out_tangents
+      else:
+        out_tangent, = all_out_tangents
+        return out_tangent
 
-  def linearized(residuals, *tangents):
-    out_tangent, = core.eval_jaxpr(jaxpr, residuals, *tangents)
-    return out_tangent
+  if prim.multiple_results:
+    return out_primals, out_nzs, out_consts, linearized
+  else:
+    out_primal, = out_primals
+    out_nz, = out_nzs
+    return out_primal, out_nz, out_consts, linearized
 
-  return out_primal, True, consts, linearized
 
 class LinearizeTracer(Tracer):
   __slots__ = ['primal', 'tangent']