[Mesa-dev] [PATCH] clover: clEnqueue* should block when asked for

[Tom Stellard]

On Mon, Aug 03, 2015 at 12:53:32PM +0300, Francisco Jerez wrote:
> EdB <edb+mesa at sigluy.net> writes:
> 
> > As a side effect, this fix clRetain/ReleaseEvent Piglit test
> > ---
> >  src/gallium/state_trackers/clover/api/transfer.cpp | 29 ++++++++++++++++++++--
> >  1 file changed, 27 insertions(+), 2 deletions(-)
> >
> > diff --git a/src/gallium/state_trackers/clover/api/transfer.cpp b/src/gallium/state_trackers/clover/api/transfer.cpp
> > index fdb9405..c2f4f13 100644
> > --- a/src/gallium/state_trackers/clover/api/transfer.cpp
> > +++ b/src/gallium/state_trackers/clover/api/transfer.cpp
> > @@ -295,6 +295,9 @@ clEnqueueReadBuffer(cl_command_queue d_q, cl_mem d_mem, cl_bool blocking,
> >                     &mem, obj_origin, obj_pitch,
> >                     region));
> >  
> > +   if (blocking)
> > +      hev().wait();
> > +
> >     ret_object(rd_ev, hev);
> >     return CL_SUCCESS;
> >  
> > @@ -325,6 +328,9 @@ clEnqueueWriteBuffer(cl_command_queue d_q, cl_mem d_mem, cl_bool blocking,
> >                     ptr, {}, obj_pitch,
> >                     region));
> >  
> > +   if (blocking)
> > +      hev().wait();
> > +
> >     ret_object(rd_ev, hev);
> >     return CL_SUCCESS;
> >  
> > @@ -362,6 +368,9 @@ clEnqueueReadBufferRect(cl_command_queue d_q, cl_mem d_mem, cl_bool blocking,
> >                     &mem, obj_origin, obj_pitch,
> >                     region));
> >  
> > +   if (blocking)
> > +      hev().wait();
> > +
> >     ret_object(rd_ev, hev);
> >     return CL_SUCCESS;
> >  
> > @@ -398,6 +407,8 @@ clEnqueueWriteBufferRect(cl_command_queue d_q, cl_mem d_mem, cl_bool blocking,
> >        soft_copy_op(q, &mem, obj_origin, obj_pitch,
> >                     ptr, host_origin, host_pitch,
> >                     region));
> > +   if (blocking)
> > +      hev().wait();
> >  
> >     ret_object(rd_ev, hev);
> >     return CL_SUCCESS;
> > @@ -504,6 +515,9 @@ clEnqueueReadImage(cl_command_queue d_q, cl_mem d_mem, cl_bool blocking,
> >                     &img, src_origin, src_pitch,
> >                     region));
> >  
> > +   if (blocking)
> > +      hev().wait();
> > +
> >     ret_object(rd_ev, hev);
> >     return CL_SUCCESS;
> >  
> > @@ -537,6 +551,8 @@ clEnqueueWriteImage(cl_command_queue d_q, cl_mem d_mem, cl_bool blocking,
> >        soft_copy_op(q, &img, dst_origin, dst_pitch,
> >                     ptr, {}, src_pitch,
> >                     region));
> > +   if (blocking)
> > +      hev().wait();
> >  
> >     ret_object(rd_ev, hev);
> >     return CL_SUCCESS;
> > @@ -666,8 +682,12 @@ clEnqueueMapBuffer(cl_command_queue d_q, cl_mem d_mem, cl_bool blocking,
> >     validate_map_flags(mem, flags);
> >  
> >     void *map = mem.resource(q).add_map(q, flags, blocking, obj_origin, region);
> > +   auto hev = create<hard_event>(q, CL_COMMAND_MAP_BUFFER, deps);
> > +
> > +   if (blocking)
> > +      hev().wait();
> >  
> > -   ret_object(rd_ev, create<hard_event>(q, CL_COMMAND_MAP_BUFFER, deps));
> > +   ret_object(rd_ev, hev);
> >     ret_error(r_errcode, CL_SUCCESS);
> >     return map;
> >  
> > @@ -695,7 +715,12 @@ clEnqueueMapImage(cl_command_queue d_q, cl_mem d_mem, cl_bool blocking,
> >  
> >     void *map = img.resource(q).add_map(q, flags, blocking, origin, region);
> >  
> > -   ret_object(rd_ev, create<hard_event>(q, CL_COMMAND_MAP_IMAGE, deps));
> > +   auto hev = create<hard_event>(q, CL_COMMAND_MAP_IMAGE, deps);
> > +
> > +   if (blocking)
> > +      hev().wait();
> > +
> > +   ret_object(rd_ev, hev);
> >     ret_error(r_errcode, CL_SUCCESS);
> >     return map;
> >  
> > -- 
> > 2.1.0
> 
> This has come up several times already, and the naive fix has a number
> of problems -- Last time [1] I proposed an alternative solution to avoid
> them, not sure if Grigori is still planning to look into it.
> 

I think someone should add a comment to these functions, so we don't
keep getting this same patch.

-Tom

> [1] http://lists.freedesktop.org/archives/mesa-dev/2015-June/086110.html




> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/mesa-dev