[Mesa-dev] [PATCH 6/8] radeonsi: use WRITE_DATA for small glBufferSubData sizes

[Bas Nieuwenhuizen]

Ack, patches 1-6 are

Reviewed-by: Bas Nieuwenhuizen <bas at basnieuwenhuizen.nl>

On Sat, Jan 19, 2019 at 2:08 AM Marek Olšák <maraeo at gmail.com> wrote:
>
> On Fri, Jan 18, 2019 at 6:05 PM Bas Nieuwenhuizen <bas at basnieuwenhuizen.nl> wrote:
>>
>> On Fri, Jan 18, 2019 at 5:44 PM Marek Olšák <maraeo at gmail.com> wrote:
>> >
>> > From: Marek Olšák <marek.olsak at amd.com>
>> >
>> > ---
>> >  src/gallium/drivers/radeonsi/si_buffer.c | 27 ++++++++++++++++++++++++
>> >  src/gallium/drivers/radeonsi/si_pipe.h   |  1 +
>> >  2 files changed, 28 insertions(+)
>> >
>> > diff --git a/src/gallium/drivers/radeonsi/si_buffer.c b/src/gallium/drivers/radeonsi/si_buffer.c
>> > index 4766cf4bdfa..a1e421b8b0d 100644
>> > --- a/src/gallium/drivers/radeonsi/si_buffer.c
>> > +++ b/src/gallium/drivers/radeonsi/si_buffer.c
>> > @@ -16,20 +16,22 @@
>> >   * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
>> >   * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
>> >   * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
>> >   * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
>> >   * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
>> >   * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
>> >   * USE OR OTHER DEALINGS IN THE SOFTWARE.
>> >   */
>> >
>> >  #include "radeonsi/si_pipe.h"
>> > +#include "sid.h"
>> > +
>> >  #include "util/u_memory.h"
>> >  #include "util/u_upload_mgr.h"
>> >  #include "util/u_transfer.h"
>> >  #include <inttypes.h>
>> >  #include <stdio.h>
>> >
>> >  bool si_rings_is_buffer_referenced(struct si_context *sctx,
>> >                                    struct pb_buffer *buf,
>> >                                    enum radeon_bo_usage usage)
>> >  {
>> > @@ -506,20 +508,38 @@ static void *si_buffer_transfer_map(struct pipe_context *ctx,
>> >         data = si_buffer_map_sync_with_rings(sctx, rbuffer, usage);
>> >         if (!data) {
>> >                 return NULL;
>> >         }
>> >         data += box->x;
>> >
>> >         return si_buffer_get_transfer(ctx, resource, usage, box,
>> >                                         ptransfer, data, NULL, 0);
>> >  }
>> >
>> > +static void si_buffer_write_data(struct si_context *sctx, struct r600_resource *buf,
>> > +                                unsigned offset, unsigned size, const void *data)
>> > +{
>> > +       struct radeon_cmdbuf *cs = sctx->gfx_cs;
>> > +
>> > +       si_need_gfx_cs_space(sctx);
>> > +
>> > +       sctx->flags |= SI_CONTEXT_PS_PARTIAL_FLUSH |
>> > +                      SI_CONTEXT_CS_PARTIAL_FLUSH |
>> > +                      si_get_flush_flags(sctx, SI_COHERENCY_SHADER, L2_LRU);
>> > +       si_emit_cache_flush(sctx);
>>
>> Maybe only do the cache flush if the buffer is referenced by the
>> current cmd buffer?
>
>
> We can't do that, because 2 consecutive IBs can execute simultaneously for a moment.
>
> Marek