Nouveau NV50 Gets Basic Compute Support, A Step Towards OpenCL
Just one week after Mesa Git received Nouveau NVC0 compute support, the NV50 Gallium3D driver for pre-Fermi GPUs has also received basic compute support.
The purpose of this compute support isn't to go start launching your own, complex OpenCL kernels on this open-source NVIDIA Linux driver right away. Unfortunately, the code isn't that far along, but it is moving in the right direction. This NV50 compute support -- like the NVC0 code -- is currently just for launching very simple compute kernels in order to read the MP performance counters of the hardware. This compute work was originally done by Francisco Jerez back during his EVoC project four years ago and then picked up by Samuel Pitoiset with all the work he's been doing on performance counters.
Samuel Pitoiset noted when committing the thousand lines of new code, "I did some improvements on the original code, like fixing using both 3D and COMPUTE simultaneously, improving global buffers binding, and making the code closer to what nvc0 already does. This compute support has been tested by Pierre Moreau and myself with some compute kernels. This is a step towards OpenCL."
Following that basic compute support were commits for compute-related MP perf counters on G84+ and support for performance metrics on G84+. NVIDIA's G84 is what was introduced under the GeForce 8600 series.
The purpose of this compute support isn't to go start launching your own, complex OpenCL kernels on this open-source NVIDIA Linux driver right away. Unfortunately, the code isn't that far along, but it is moving in the right direction. This NV50 compute support -- like the NVC0 code -- is currently just for launching very simple compute kernels in order to read the MP performance counters of the hardware. This compute work was originally done by Francisco Jerez back during his EVoC project four years ago and then picked up by Samuel Pitoiset with all the work he's been doing on performance counters.
Samuel Pitoiset noted when committing the thousand lines of new code, "I did some improvements on the original code, like fixing using both 3D and COMPUTE simultaneously, improving global buffers binding, and making the code closer to what nvc0 already does. This compute support has been tested by Pierre Moreau and myself with some compute kernels. This is a step towards OpenCL."
Following that basic compute support were commits for compute-related MP perf counters on G84+ and support for performance metrics on G84+. NVIDIA's G84 is what was introduced under the GeForce 8600 series.
1 Comment