Bring back WebGPU (#7063)

* Start from andredaprato:webgpu-clean

* Fix infs

* inf wgsl function is not needed

* Emulated ulong for threefry, more tests passing

* Randomness tests passing

* Update model export to support new changes in webgpu, efficientnet export works again

* Simplify shift emulation in wgsl

* Delete test file

* Fix bigger than u32 u32 literal

* Why was skip copies added here?

* Python3.12 for webgpu tests

* Fix model export syntax error

* Get test ops passing with some skips

* Fix lint

* Much simpler shift

* Run more tests

* Timestamp queries are not supported in CI, so skip search tests

* All fancy indexing passing

* r is ctx

* Run more dtype tests by using is_dtype_supported

* Cleanup ulong shift rendering

* UPat -> Pat, UOps -> Ops

* Pat -> UPat

* Refactor render_ushift if-else

* Pattern to avoid ulong mul

* Remove vals_dtype

* is_nan trick + rewrite, test_isnan passing

* Rewrite a * select(1, nan, gate) -> select(a, nan, gate)

* No arg, just op

* Support char, uchar, short, ushort

* Run test_index_mnis now that we have uint8

* Fix pyling

* Save 3 lines by using base Compiler

* No more long emulation

* Remove fixup_binops

* No more external_local_bufx wgsl specific cstyle modif, use base extra_pm

* Simpler, faster copyin/out

* Skip some new tests that use long

* Fix typo

* copyout touchup

* Save lines by using render_cast

* WebGL is not supported in core, delete it from is_dtype_supported

* More narrow test skips for some unary tests

* TernaryOps, UnaryOps -> Ops

* TinyGrad supports WebGPU

* StableDiffusion demo: f16tof32 gpu is a lib, update UI

* Packed load/store, no more scale_size, no core tinygrad changes

* Rename copyin, copyout

* Device -> dev

* Fix lint

* Pattern matcher rule for packed load/store

* Refactor

* Shorter packed load/store

* this should fix lint

* Fix mypy

* SD compile script working

* New SD webgpu UI

* New default prompt

* New SD weights

* Fix title when webgpu not available

* Run symbolic tests, simplify is_nan, use round_up

* Show step time on UI

* Bump minimum wgpu version to v0.19

* Fix latent

---------

Co-authored-by: George Hotz <72895+geohot@users.noreply.github.com>

test/test_dtype.py

@@ -35,11 +35,11 @@ def _test_to_np(a:Tensor, np_dtype, target):
   except AssertionError as e:
     raise AssertionError(f"\ntensor {a.numpy()} does not match target {target} with np_dtype {np_dtype}") from e
 
-def _assert_eq(tensor:Tensor, target_dtype:DType, target):
+def _assert_eq(tensor:Tensor, target_dtype:DType, target, tol_target_dtype:float=1e-7):
   if DEBUG >= 2: print(tensor.numpy())
   try:
     assert tensor.dtype == target_dtype
-    np.testing.assert_allclose(tensor.numpy(), target, rtol={dtypes.float16:1e-3, dtypes.bfloat16:1e-2}.get(target_dtype, 1e-7))
+    np.testing.assert_allclose(tensor.numpy(), target, rtol={dtypes.float16:1e-3, dtypes.bfloat16:1e-2}.get(target_dtype, tol_target_dtype))
   except AssertionError as e:
     raise AssertionError(f"\ntensor {tensor.numpy()} dtype {tensor.dtype} does not match target {target} with dtype {target_dtype}") from e
 
@@ -541,7 +541,7 @@ class TestTypeSpec(unittest.TestCase):
       _assert_eq(Tensor.arange(5, dtype=dtypes.int64), dtypes.int64, np.arange(5))
     if is_dtype_supported(dtypes.float16):
       _assert_eq(Tensor.arange(5, dtype=dtypes.float16), dtypes.float16, np.arange(5))
-    _assert_eq(Tensor.arange(3, 9, 0.7), dtypes.default_float, np.arange(3, 9, 0.7))
+    _assert_eq(Tensor.arange(3, 9, 0.7), dtypes.default_float, np.arange(3, 9, 0.7), 1e-6 if Device.DEFAULT == "WEBGPU" else 1e-7)
     _assert_eq(Tensor.arange(3, 8.5, 3), dtypes.default_float, np.arange(3, 8.5, 3))
     # stop-start and step have different signs
     _assert_eq(Tensor.arange(3, 5, -2), dtypes.default_int, np.arange(3, 5, -2))