[GitHub] [tvm] hogepodge commented on a change in pull request #7642: [docs] Getting Started With TVM: Tensor Expressions

[GitBox <gi...@apache.org>]

hogepodge commented on a change in pull request #7642:
URL: https://github.com/apache/tvm/pull/7642#discussion_r600033842



##########
File path: tutorials/get_started/tensor_expr_get_started.py
##########
@@ -255,41 +340,39 @@
 fadd1(a, b, c)
 tvm.testing.assert_allclose(c.asnumpy(), a.asnumpy() + b.asnumpy())
 
-######################################################################
+################################################################################
 # Pack Everything into One Library
-# --------------------------------
-# In the above example, we store the device and host code separately.
-# TVM also supports export everything as one shared library.
-# Under the hood, we pack the device modules into binary blobs and link
-# them together with the host code.
-# Currently we support packing of Metal, OpenCL and CUDA modules.
-#
+# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+# In the above example, we store the device and host code separately. TVM also
+# supports export everything as one shared library. Under the hood, we pack
+# the device modules into binary blobs and link them together with the host
+# code. Currently we support packing of Metal, OpenCL and CUDA modules.
+
 fadd.export_library(temp.relpath("myadd_pack.so"))
 fadd2 = tvm.runtime.load_module(temp.relpath("myadd_pack.so"))
 fadd2(a, b, c)
 tvm.testing.assert_allclose(c.asnumpy(), a.asnumpy() + b.asnumpy())
 
-######################################################################
+################################################################################
 # .. note:: Runtime API and Thread-Safety
 #
-#   The compiled modules of TVM do not depend on the TVM compiler.
-#   Instead, they only depend on a minimum runtime library.
-#   The TVM runtime library wraps the device drivers and provides
-#   thread-safe and device agnostic calls into the compiled functions.
-#
-#   This means that you can call the compiled TVM functions from any thread,
-#   on any GPUs.
+#   The compiled modules of TVM do not depend on the TVM compiler. Instead,
+#   they only depend on a minimum runtime library. The TVM runtime library
+#   wraps the device drivers and provides thread-safe and device agnostic calls
+#   into the compiled functions.
 #
+#   This means that you can call the compiled TVM functions from any thread, on
+#   any GPUs, provided that you have compiled the code for that GPU.
 
-######################################################################
+################################################################################
 # Generate OpenCL Code

Review comment:
       This came from an existing document, and since this is an incremental refactor my plan is to revisit and break out the CUDA and OpenCL sections into their own documents. I'm in agreement, but think that change may be out of the scope of the this PR.




-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

For queries about this service, please contact Infrastructure at:
users@infra.apache.org