fix float16 in CLANG on linux

test/test_dtype.py

@@ -14,6 +14,7 @@ class TestDtype(unittest.TestCase):
     na = a.numpy()
     print(na, na.dtype, a.lazydata.realized)
     assert na.dtype == np.float16
+    np.testing.assert_allclose(na, [1,2,3,4])
 
   def test_half_add(self):
     a = Tensor([1,2,3,4], dtype=dtypes.float16)
@@ -21,6 +22,7 @@ class TestDtype(unittest.TestCase):
     c = a+b
     print(c.numpy())
     assert c.dtype == dtypes.float16
+    np.testing.assert_allclose(c.numpy(), [2,4,6,8])
 
   def test_upcast_float(self):
     # NOTE: there's no downcasting support
@@ -29,6 +31,7 @@ class TestDtype(unittest.TestCase):
     na = a.numpy()
     print(na, na.dtype)
     assert na.dtype == np.float32
+    np.testing.assert_allclose(na, [1,2,3,4])
 
   def test_half_add_upcast(self):
     a = Tensor([1,2,3,4], dtype=dtypes.float16)
@@ -36,6 +39,7 @@ class TestDtype(unittest.TestCase):
     c = a+b
     print(c.numpy())
     assert c.dtype == dtypes.float32
+    np.testing.assert_allclose(c.numpy(), [2,4,6,8])
 
 if __name__ == '__main__':
   unittest.main()

tinygrad/runtime/ops_clang.py

@@ -14,8 +14,9 @@ class ClangProgram:
     prg = "#include <math.h>\n#define max(x,y) ((x>y)?x:y)\n#define half __fp16\n" + prg
     # TODO: is there a way to not write this to disk?
     fn = f"/tmp/clang_{hashlib.md5(prg.encode('utf-8')).hexdigest()}.{'dylib' if platform.system() == 'Darwin' else 'so'}"
+    # NOTE: --rtlib=compiler-rt fixes float16 on Linux, it defines __gnu_h2f_ieee and __gnu_f2h_ieee
     if not os.path.exists(fn):
-      subprocess.check_output(['clang', '-shared', '-O2', '-Wall','-Werror', '-lm', '-fPIC', '-x', 'c', '-', '-o', fn+".tmp"], input=prg.encode('utf-8'))
+      subprocess.check_output(['clang', '-shared', '-O2', '-Wall','-Werror', '-lm', '--rtlib=compiler-rt', '-fPIC', '-x', 'c', '-', '-o', fn+".tmp"], input=prg.encode('utf-8'))
       os.rename(fn+".tmp", fn)
     self.lib = ctypes.CDLL(fn)
     self.fxn = self.lib[name]