fix webgpu is_nan [pr] (#17551)

test/backend/test_const_folding.py

@@ -188,7 +188,6 @@ class TestTautologicalCompare(unittest.TestCase):
     np.testing.assert_equal((Tensor(True) < Tensor(False)).numpy(), False)
     np.testing.assert_equal((Tensor(True) < Tensor(True)).numpy(), False)
 
-  @unittest.skipIf(Device.DEFAULT == "WEBGPU", "WEBGPU doesn't support NaN comparison correctly")
   def test_a_eq_a(self):
     # self eq is always true for int or bool
     a = Tensor([1, 2, 3])

tinygrad/renderer/wgsl.py

@@ -50,8 +50,9 @@ wgsl_matcher = PatternMatcher([
   (UPat.store(UPat.var("b"), UPat.var("var"), name="s"), lambda b,var,s: packed_store(b,var) if is_packed(s) else None),
   (UPat.var("a") << UPat.var("b"),lambda a,b:(a.bitcast(dtypes.uint32)<<b.cast(dtypes.uint32)).bitcast(a.dtype) if b.dtype!=dtypes.uint32 else None),
   (UPat.var("x") >> UPat.var("y"), lambda x,y: UOp(Ops.SHR, x.dtype, (x,y.cast(dtypes.uint))) if y.dtype != dtypes.uint else None),
-  # fix nan check: 'a != a -> is_nan()'
-  (UPat.var("a") != UPat.var("a"), is_nan),
+  # fix nan check: 'a != a -> is_nan()'. the decomp rewrites (a != a).logical_not() to CMPEQ, so match both forms
+  (UPat.var("a", dtypes.floats) != UPat.var("a"), is_nan),
+  (UPat.var("a", dtypes.floats).alu(Ops.CMPEQ, UPat.var("a")), lambda a: is_nan(a).ne(True)),
   ])
 
 class WGSLRenderer(CStyleLanguage):