Update docs

Author: github-actions[bot]

_sources/autoapi/tilelang/language/copy_op/index.rst.txt

@@ -53,14 +53,16 @@ Module Contents
    - Accepts `Buffer`/`BufferRegion`/`BufferLoad` on either side. Extents are
      derived as follows: `Buffer -> shape`, `BufferRegion -> [r.extent]`,
      `BufferLoad -> extents from its inferred/encoded region`.
-   - If both `src` and `dst` are scalar `BufferLoad` without region extents,
-     lowers to a direct store: `dst[...] = src`.
-   - If one side is missing extents, it is treated as all-ones with the other
-     side's rank to enable broadcasting.
-   - Extents are right-aligned and legalized via `legalize_pairwise_extents`:
-     per tail-dimension, equal keeps as-is, a `1` broadcasts to the other,
-     otherwise a conservative `tir.max` is used to remain safe for dynamic
-     shapes.
+   - Normally, we require the extents of both sides to be the same. If they
+     differ, the copy instruction follows an internal rule to select one side
+     as the base range and create iteration space. This may generate unexpected
+     code. And if some dimensions are 1, unexpected errors may happen.
+   - Small Optimization: If both `src` and `dst` are scalar `BufferLoad` without
+     region extents, lowers to a direct store: `dst[...] = src[...]`.
+   - Syntactic Sugar: TileLang supports passing the head address of a buffer to represent
+     the whole buffer if there are no ambiguity. For example, T.copy(A, A_shared[i, j]).
+     To support this, we need some special shape checking. But remember currently we don't
+     support something like "broadcast".
    - The finalized extents are encoded with `tl.region` via `to_buffer_region`
      and passed through to the backend; low-level loop construction and any
      scope-specific decisions happen during lowering.

autoapi/tilelang/language/copy_op/index.html

@@ -612,14 +612,16 @@ <h2>Module Contents<a class="headerlink" href="#module-contents" title="Link to
 <cite>BufferLoad -&gt; extents from its inferred/encoded region</cite>.</p>
 </div></blockquote>
 <ul class="simple">
-<li><p>If both <cite>src</cite> and <cite>dst</cite> are scalar <cite>BufferLoad</cite> without region extents,
-lowers to a direct store: <cite>dst[…] = src</cite>.</p></li>
-<li><p>If one side is missing extents, it is treated as all-ones with the other
-side’s rank to enable broadcasting.</p></li>
-<li><p>Extents are right-aligned and legalized via <cite>legalize_pairwise_extents</cite>:
-per tail-dimension, equal keeps as-is, a <cite>1</cite> broadcasts to the other,
-otherwise a conservative <cite>tir.max</cite> is used to remain safe for dynamic
-shapes.</p></li>
+<li><p>Normally, we require the extents of both sides to be the same. If they
+differ, the copy instruction follows an internal rule to select one side
+as the base range and create iteration space. This may generate unexpected
+code. And if some dimensions are 1, unexpected errors may happen.</p></li>
+<li><p>Small Optimization: If both <cite>src</cite> and <cite>dst</cite> are scalar <cite>BufferLoad</cite> without
+region extents, lowers to a direct store: <cite>dst[…] = src[…]</cite>.</p></li>
+<li><p>Syntactic Sugar: TileLang supports passing the head address of a buffer to represent
+the whole buffer if there are no ambiguity. For example, T.copy(A, A_shared[i, j]).
+To support this, we need some special shape checking. But remember currently we don’t
+support something like “broadcast”.</p></li>
 <li><p>The finalized extents are encoded with <cite>tl.region</cite> via <cite>to_buffer_region</cite>
 and passed through to the backend; low-level loop construction and any
 scope-specific decisions happen during lowering.</p></li>