[globalisel][docs] Add a section about debugging with the block extractor

[dsanders]

Summary

Depends on D69644

[volkan]

Thanks Daniel, LGTM.

llvm/docs/GlobalISel/Pipeline.rst

163

Would it be better if we use the pictures from the talk instead of the code block?

llvm/docs/GlobalISel/Pipeline.rst

 testable:
 
 .. image:: testing-unit-level.png
 
 It's possible to create an imaginary target such as in `LegalizerHelperTest.cpp <https://github.com/llvm/llvm-project/blob/93b29d3882baf7df42e4e9bc26b977b00373ef56/llvm/unittests/CodeGen/GlobalISel/LegalizerHelperTest.cpp#L28-L57>`_
 and perform a single step of the algorithm and check the result. The MIR and
 FileCheck directives can be embedded using strings so you still have access to
 the convenience available in llvm-lit.
+
+Debugging
+---------
+
+One debugging technique that's proven particularly valuable is to use the
+BlockExtractor to extract basic blocks into new functions. This can be used
+to track down correctness bugs and can also be used to track down performance
+regressions.
+
+The command to do this is:
+
+.. code-block:: shell
+
+  ./bin/llvm-extract!-o!-!-S -b!‘foo:bb9;bb20’!<input>!>!extracted.ll
+
+This particular example extracts two basic blocks from a function named ``foo``.
+
+This can prevent some optimizations as GlobalISel is generally able to work on a
+single function at a time. This technique can be repeated for different
+combinations of basic blocks until you have identified the critical blocks
+involved in a bug.
+
+Once the critical blocks have been identified, you can further increase the
+resolution to the critical instructions by splitting the blocks like from:
+
+.. code-block:: llvm
+
+  bb1:
+    ... instructions group 1 ...
+    ... instructions group 2 ...
+
+into:
+
+.. code-block:: llvm
+
+  bb1:
+    ... instructions group 1 ...
+    br %bb2
+
+  bb2:
+    ... instructions group 2 ...
+
+and then repeating the process for the new blocks.
+
+It's also possible to use this technique in a mode where the main function
+is compiled with GlobalISel and the extracted basic blocks are compiled with
+SelectionDAG (or the other way around) to leverage the existing quality of
+another code generator to track down bugs. This technique can also be used to
+improve the similarity between fast and slow code when tracking down performance
+regressions and help you zero in on a particular cause of the regression.