[Mesa-dev] [PATCH 8/8] radeonsi: ignore PIPE_RESOURCE_FLAG_MAP_COHERENT
Dieter Nützel
Dieter at nuetzel-hh.de
Mon Jun 11 01:27:22 UTC 2018
For this series:
Tested-by: Dieter Nützel <Dieter at nuetzel-hh.de>
with glmark2, UH, UV, Blender 2.79, FreeCAD, Krita 4.0.3 under KDE
Plasma5
on my Polaris 20 / RX580.
Dieter
Am 09.06.2018 05:16, schrieb Marek Olšák:
> From: Marek Olšák <marek.olsak at amd.com>
>
> We treat coherent and non-coherent buffers the same.
>
> And move external_usage for better packing.
> ---
> src/gallium/drivers/radeonsi/si_buffer.c | 3 +--
> src/gallium/drivers/radeonsi/si_pipe.h | 6 +++---
> 2 files changed, 4 insertions(+), 5 deletions(-)
>
> diff --git a/src/gallium/drivers/radeonsi/si_buffer.c
> b/src/gallium/drivers/radeonsi/si_buffer.c
> index 0546fa9d336..5bd95bffc17 100644
> --- a/src/gallium/drivers/radeonsi/si_buffer.c
> +++ b/src/gallium/drivers/radeonsi/si_buffer.c
> @@ -135,22 +135,21 @@ void si_init_resource_fields(struct si_screen
> *sscreen,
> case PIPE_USAGE_IMMUTABLE:
> default:
> /* Not listing GTT here improves performance in some
> * apps. */
> res->domains = RADEON_DOMAIN_VRAM;
> res->flags |= RADEON_FLAG_GTT_WC;
> break;
> }
>
> if (res->b.b.target == PIPE_BUFFER &&
> - res->b.b.flags & (PIPE_RESOURCE_FLAG_MAP_PERSISTENT |
> - PIPE_RESOURCE_FLAG_MAP_COHERENT)) {
> + res->b.b.flags & PIPE_RESOURCE_FLAG_MAP_PERSISTENT) {
> /* Use GTT for all persistent mappings with older
> * kernels, because they didn't always flush the HDP
> * cache before CS execution.
> *
> * Write-combined CPU mappings are fine, the kernel
> * ensures all CPU writes finish before the GPU
> * executes a command stream.
> *
> * radeon doesn't have good BO move throttling, so put all
> * persistent buffers into GTT to prevent VRAM CPU page faults.
> diff --git a/src/gallium/drivers/radeonsi/si_pipe.h
> b/src/gallium/drivers/radeonsi/si_pipe.h
> index eb0c226ed79..5ff762296fc 100644
> --- a/src/gallium/drivers/radeonsi/si_pipe.h
> +++ b/src/gallium/drivers/radeonsi/si_pipe.h
> @@ -209,26 +209,26 @@ struct r600_resource {
> * Any hw block which doesn't use or bypasses TC L2 should check this
> * flag and flush the cache before using the buffer.
> *
> * For example, TC L2 must be flushed if a buffer which has been
> * modified by a shader store instruction is about to be used as
> * an index buffer. The reason is that VGT DMA index fetching doesn't
> * use TC L2.
> */
> bool TC_L2_dirty;
>
> - /* Whether the resource has been exported via resource_get_handle. */
> - unsigned external_usage; /* PIPE_HANDLE_USAGE_* */
> -
> /* Whether this resource is referenced by bindless handles. */
> bool texture_handle_allocated;
> bool image_handle_allocated;
> +
> + /* Whether the resource has been exported via resource_get_handle. */
> + unsigned external_usage; /* PIPE_HANDLE_USAGE_* */
> };
>
> struct r600_transfer {
> struct threaded_transfer b;
> struct r600_resource *staging;
> unsigned offset;
> };
>
> struct r600_cmask_info {
> uint64_t offset;
More information about the mesa-dev
mailing list