few more symbolic rules [pr] (#17161)

bitcast twice can skip middle one; bool != False is self; two more where combine

test/null/test_uop_symbolic.py

@@ -967,6 +967,28 @@ class TestSymbolic(unittest.TestCase):
     # not combining  # TODO: can combine if one is identity element const
     self.helper_test_variable(aa+ab, 0, 6, "((x<2).where(a, b)+(x<2).where(a, 0))")
 
+  def test_where_combine_cross_zero(self):
+    cond = Variable("x", 0, 3) < 2
+    a = Variable("a", 0, 3)
+    b = Variable("b", 0, 3)
+    self.helper_test_variable(cond.where(a, a.ufix(0)) + cond.where(b.ufix(0), b), 0, 3, "(x<2).where(a, b)")
+    self.helper_test_variable(cond.where(a, a.ufix(0)) + cond.where(a.ufix(0), a), 0, 3, "a")
+
+  def test_where_or_dual(self):
+    m1 = Variable("x", 0, 3) < 2
+    m2 = Variable("y", 0, 3) < 2
+    a = Variable("a", 0, 3)
+    b = Variable("b", 0, 3)
+    self.helper_test_variable(m1.where(a, m2.where(a, b)), 0, 3, "((x<2)|(y<2)).where(a, b)")
+
+  def test_bool_ne_false(self):
+    cond = Variable("x", 0, 3) < 2
+    self.helper_test_variable(cond.ne(False), 0, 1, "(x<2)")
+
+  def test_bitcast_chain(self):
+    a = Variable("a", 0, 3)
+    self.assertIs(graph_rewrite(a.bitcast(dtypes.float32).bitcast(a.dtype), sym), a)
+
   def test_negation_in_where(self):
     cond = Variable("x", 0, 3) < 2
     a = Variable("a", 0, 3)

tinygrad/uop/symbolic.py

@@ -109,6 +109,7 @@ symbolic_simple = pm_data_invalid + PatternMatcher([
   (UPat(Ops.ADD, dtype=dtypes.weakint, name="x"), fold_add_divmod_recombine),
   (UPat.var("x", dtype=dtypes.bool) & UPat.cvar("c"), lambda x,c: x if c.arg else c),
   (UPat.var("x", dtype=dtypes.bool) | UPat.cvar("c"), lambda x,c: c if c.arg else x),
+  (UPat.var("x", dtype=dtypes.bool) != UPat.const(dtypes.bool, False), lambda x: x),  # x != False -> x
   (UPat(GroupOp.Idempotent, src=(UPat.var("x"), UPat.var("x"))), lambda x: x),
   (UPat.var("x", dtype=dtypes.bool).logical_not().logical_not(), lambda x: x),
   (UPat.var("x", dtype=dtypes.bool).where(UPat.const(dtypes.bool, True), UPat.const(dtypes.bool, False)), lambda x: x),
@@ -156,6 +157,8 @@ symbolic_simple = pm_data_invalid + PatternMatcher([
   (UPat(Ops.BITCAST, name="root", src=(UPat.cvar("c"),)), fold_bitcast),
   # b.cast(a).cast(b) -> b if a preserves all values in b
   (UPat.var('x').cast(name="a").cast(name="b"), lambda x,a,b: x if x.dtype == b.dtype and can_lossless_cast(b.dtype, a.dtype) else None),
+  # bitcast twice
+  (UPat(Ops.BITCAST, name="b", src=(UPat.var('x').bitcast(),)), lambda x,b: x.bitcast(b.dtype)),
   (UPat.var("x").cast(dtypes.bool), lambda x: x != 0),
   # ** pow **
   (UPat.var("x").alu(Ops.POW, UPat.cvar("c")), simplify_pow),
@@ -174,6 +177,8 @@ symbolic_simple = pm_data_invalid + PatternMatcher([
   (UPat.cvar("gate").where(UPat.var("c0"), UPat.var("c1")), lambda gate, c0, c1: c0 if gate.arg else c1),
   # a.where(b.where(c, d), d) -> (a & b).where(c, d)
   (UPat.var("a").where(UPat.var("b").where(UPat.var("c"), UPat.var("d")), UPat.var("d")), lambda a,b,c,d: (a&b).where(c,d)),
+  # a.where(c, b.where(c, d)) -> (a | b).where(c, d)
+  (UPat.var("a").where(UPat.var("c"), UPat.var("b").where(UPat.var("c"), UPat.var("d"))), lambda a,b,c,d: (a|b).where(c,d)),
 ])+mop_cleanup
 
 # ******** phase 2 builds on phase 1, it includes the old "symbolic", rules that match deeper ********
@@ -237,6 +242,8 @@ symbolic = symbolic_simple+commutative+PatternMatcher([
   # if its a plus we add the associative variation too
   ((UPat.var("y")+UPat.var("c").where(UPat.var("t"), UPat.var("f"))) + UPat.var("c").where(UPat.var("tt"), UPat.var("ff")), \
    lambda y,c,t,tt,f,ff: y+c.where(t+tt, f+ff) if t.op == tt.op == Ops.CONST or f.op == ff.op == Ops.CONST else None),
+  # complementary zero branches under the same condition select directly
+  (UPat.var("c").where(UPat.var("t"), 0) + UPat.var("c").where(0, UPat.var("f")), lambda c,t,f: c.where(t, f)),
   # ALU/variable min==max -> CONST
   (UPat({Ops.CMPLT, Ops.CMPNE, Ops.FLOORDIV, Ops.FLOORMOD, Ops.PARAM, Ops.BIND, Ops.SPECIAL}, name="x"),
    lambda x: x.const_like(x.vmin) if x.vmin == x.vmax else None),