On 11/16/2015 11:55 AM, Emil Velikov wrote:
On 9 October 2015 at 14:10, Samuel Pitoiset <samuel.pitoi...@gmail.com> wrote:
Compute class value for GF110+ is 0x91c0 and not 0x92c0. This fixes
compute support and MP performance counters on GF110.
Signed-off-by: Samuel Pitoiset <samuel.pitoi...@gmail.com>
---
src/gallium/drivers/nouveau/nv_object.xml.h | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/src/gallium/drivers/nouveau/nv_object.xml.h
b/src/gallium/drivers/nouveau/nv_object.xml.h
index 0a0e187..92c0633 100644
--- a/src/gallium/drivers/nouveau/nv_object.xml.h
+++ b/src/gallium/drivers/nouveau/nv_object.xml.h
@@ -197,7 +197,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.
#define NV50_COMPUTE_CLASS 0x000050c0
#define NVA3_COMPUTE_CLASS 0x000085c0
#define NVC0_COMPUTE_CLASS 0x000090c0
-#define NVC8_COMPUTE_CLASS 0x000092c0
+#define NVC8_COMPUTE_CLASS 0x000091c0
Worth updating the classic one (src/mesa/drivers/dri/nouveau/) as well ?
I don't think it's useful because the classic one is for < NV50.
But one day, we should re-generate those headers.
There is a nasty looking comment in nvc0_screen_compute_setup about
the above class. Afaics although updated the define isn't used
anywhere so I'm dubious how it fixes compute on GF110.
This actually only fixes the value of the NVC8_COMPUTE_CLASS. I did this
patch before writing that comment in nvc0_screen_compute_setup().
In practice, GF100/GF110 should support 0x91c0 but for some weird
reason, a ILLEGAL_CLASS dmesg appears when using it.
Anyway, 0x91c0 only introduces some minor changes regarding 0x90c0.
That's why we use 0x90c0 for all NV50 chipsets.
Cheers,
Emil
--
-Samuel
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/mesa-dev
