[Gluon][Tutorial] Use autolayout in a few places (#7646)

cc @manman-ren

Author: Mogball

python/triton/experimental/gluon/language/_core.py

@@ -389,7 +389,7 @@ def convert_layout(value, layout, assert_trivial=False, _semantic=None):
 
 
 @builtin
-def full(shape, value, dtype, layout, _semantic=None):
+def full(shape, value, dtype, layout=None, _semantic=None):
     """
     Create a tensor filled with a scalar value, with specified shape, dtype, and layout.
 

python/triton/experimental/gluon/language/_semantic.py

@@ -158,6 +158,8 @@ def splat(self, value, shape, layout):
 
     def full(self, shape, value, dtype, layout):
         scalar = self.make_scalar(value, dtype)
+        if layout is None:
+            layout = AutoLayout()
         return self.splat(scalar, shape, layout)
 
     def convert_layout(self, value, layout, assert_trivial=False):

python/tutorials/gluon/01-attention-forward.py

@@ -753,7 +753,7 @@ def _softmax_tile(tile_id: gl.constexpr, config, M, desc_o, STAGE: gl.constexpr,
         prog = scheduler.get_program(pid)
 
         offs_m = prog.start_m * config.BLOCK_M
-        offs_m += gl.arange(tile_id * config.SPLIT_M, (1 + tile_id) * config.SPLIT_M, qk_slice_dim1)
+        offs_m += gl.arange(tile_id * config.SPLIT_M, (1 + tile_id) * config.SPLIT_M)
 
         m_i = gl.full([config.SPLIT_M], -float("inf"), gl.float32, qk_slice_dim1)
         l_i0 = gl.full([config.SPLIT_M], 0.0, gl.float32, gl.SliceLayout(1, sum_layout))