@@ -166,19 +166,19 @@ A full list of available SYCL alias targets is available in the
166
166
CUDA build
167
167
~~~~~~~~~~
168
168
169
- ``dpctl `` can be built for CUDA devices using the ``DPCTL_TARGET_CUDA `` CMake option,
170
- which accepts a specific compute architecture string:
169
+ ``dpctl `` can be built for CUDA devices using the ``--target-cuda `` argument.
170
+
171
+ To target a specific architecture (e.g., ``sm_80 ``):
171
172
172
173
.. code-block :: bash
173
174
174
- python scripts/build_locally.py --verbose --cmake-opts= " -DDPCTL_TARGET_CUDA= sm_80"
175
+ python scripts/build_locally.py --verbose --target-cuda= sm_80
175
176
176
- To use the default architecture (``sm_50 ``),
177
- set ``DPCTL_TARGET_CUDA `` to a value such as ``ON ``, ``TRUE ``, ``YES ``, ``Y ``, or ``1 ``:
177
+ To use the default architecture (``sm_50 ``), omit the value:
178
178
179
179
.. code-block :: bash
180
180
181
- python scripts/build_locally.py --verbose --cmake-opts= " -DDPCTL_TARGET_CUDA=ON "
181
+ python scripts/build_locally.py --verbose --target-cuda
182
182
183
183
Note that kernels are built for the default architecture (``sm_50 ``), allowing them to work on a
184
184
wider range of architectures, but limiting the usage of more recent CUDA features.
@@ -225,8 +225,7 @@ devices at the same time:
225
225
226
226
.. code-block :: bash
227
227
228
- python scripts/build_locally.py --verbose --cmake-opts=" -DDPCTL_TARGET_CUDA=ON \
229
- -DDPCTL_TARGET_HIP=gfx1030"
228
+ python scripts/build_locally.py --verbose --target-cuda --cmake-opts=" -DDPCTL_TARGET_HIP=gfx1030"
230
229
231
230
Running Examples and Tests
232
231
==========================
0 commit comments