Signed-off-by: Andreas Baierl <ichgeh@imkreisrum.de>

Signed-off-by: Qiang Yu <yuq825@gmail.com>

staging size is smaller than actual texture, so we need to take it
into account.
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>

Unlike the pp disassembler, the syntax here is significantly different
from the old open-gpu-tools disassembler. Instead of printing
loads and stores as a separate part of the instruction, we combine them
with the ALU operations, to make it easier to see how the data flows.

This is based on the open-gpu-tools disassembler, although I've taken a
few liberties with the syntax.

Signed-off-by: Andreas Baierl <ichgeh@imkreisrum.de>

In preparation for future lower functions we make clear, that the existing
ones are executed after gpir_reduce_reg_pressure_prog.
Signed-off-by: Andreas Baierl <ichgeh@imkreisrum.de>

regression:
6afdaef3 lima: seperate lima_texture_desc_set_res
Signed-off-by: Qiang Yu <yuq825@gmail.com>

Signed-off-by: Qiang Yu <yuq825@gmail.com>

Signed-off-by: Qiang Yu <yuq825@gmail.com>

To fix the flush not called problem when user read a buffer
before swap buffer or flush explicitly, because lima uses
the bo recorded in submit to identify if it's dirty.
Signed-off-by: Qiang Yu <yuq825@gmail.com>

Signed-off-by: Qiang Yu <yuq825@gmail.com>

Signed-off-by: Qiang Yu <yuq825@gmail.com>

Signed-off-by: Qiang Yu <yuq825@gmail.com>

Signed-off-by: Qiang Yu <yuq825@gmail.com>

In order to deal with scissor clear (partial clear).
Signed-off-by: Qiang Yu <yuq825@gmail.com>

lima ppir and gpir often hit unsupported alu nir_op instructions at this
stage of development. Printing the raw nir_op is not very helpful, since
the list of nir_op is an unnumbered enum that is generated during the
build, so it's annoying to figure out what is the missing
implementation.
Just print the name as pointed out by cwabbott some time ago on IRC.
Signed-off-by: Erico Nunes <nunes.erico@gmail.com>

Method is only pick pp stream for damaged region,
so that fb outside damage region don't need reload.

In fact only fb between damaged region and tile
unit boundary (when damage region not aligned to
tile buffer) need to be reloaded.
Signed-off-by: Qiang Yu <yuq825@gmail.com>

Signed-off-by: Qiang Yu <yuq825@gmail.com>

This is the reload implementation of this use case.
Use this method instead of the previous gp stream
reuse method for:
1. simplicity
2. gp stream reuse method need more investigate on
   max draw num for buffer overflow
3. easier integration with the EGL_KHR_partial_update
   implementation

Shortcome is the performance maybe worse, but I don't
know how much.
Signed-off-by: Qiang Yu <yuq825@gmail.com>

For reused by reload FB as texture.
Signed-off-by: Qiang Yu <yuq825@gmail.com>

Signed-off-by: Qiang Yu <yuq825@gmail.com>

This reverts commit 659d7a53.

prepare for EGL_KHR_partial_update.
Signed-off-by: Qiang Yu <yuq825@gmail.com>

Signed-off-by: Qiang Yu <yuq825@gmail.com>

Enable dri2DamageExtension according to
PIPE_CAP_SET_DAMAGE.
Signed-off-by: Qiang Yu <yuq825@gmail.com>

Signed-off-by: Qiang Yu <yuq825@gmail.com>

When optional extension increases, use static extension
is not good. This is for the following adding another
optional extension.
Signed-off-by: Qiang Yu <yuq825@gmail.com>

Use the DRI2 interface callback to pass the damage rects to
the driver.

v2: check for dri2_dpy->damage_extension after checking for
    optional_core_extensions in dr2_setup_extensions.
Signed-off-by: Harish Krupo <harish.krupo.kps@intel.com>

Implementors of this interface provide ability to render only
to damaged regions on the framebuffer.
Signed-off-by: Harish Krupo <harish.krupo.kps@intel.com>

The intension of the KHR_partial_update was not to send the damage back
to the platform but to send the damage to the driver to ensure that the
following rendering could be restricted to those regions.
This patch removes the set_damage_region from the egl_dri vtbl and all
the platfrom_*.c files.
Then upcomming patches add a new dri2 interface for the drivers to
implement
Signed-off-by: Harish Krupo <harish.krupo.kps@intel.com>

When no EGL_EXT_buffer_age but has EGL_KHR_partial_update,
query buffer age should not fail.
Signed-off-by: Qiang Yu <yuq825@gmail.com>

For gallium drivers to expose EGL/GLX_EXT_buffer_age.
Signed-off-by: Qiang Yu <yuq825@gmail.com>

Add __DRI2_RENDERER_EXPOSE_BUFFER_AGE dri integer query for
driver to determine if want to expose EGL/GLX_EXT_buffer_age.

Without knowing damage region first, some tile buffer based
GPU like ARM Mali Utgard has to load whole render target to
tile buffer before partitial update, then write them back.

Expose this extension will mis-lead application to use it
which will make performance worse for this kind of GPU.

EGL_KHR_partial_update is better choice for this kind of GPU.
Signed-off-by: Qiang Yu <yuq825@gmail.com>

Signed-off-by: Qiang Yu <yuq825@gmail.com>

Signed-off-by: Qiang Yu <yuq825@gmail.com>

A mesa driver for rockchip is defined which will call into lima when 3D
rendering is requested from it.
This uses the renderonly library from mesa, which hides the buffer
sharing details between the two drivers from the application.
This allows applications that expect a single device node like
/dev/dri/card0 (such as kmscube) to run unmodified on Rockchip devices
that use the rockchip-drm driver for display and have a Mali400 gpu using
lima.

This follows the model of the lima+sun4i drivers existing in mesa.
The code implemented here is mostly borrowed from that particular
implementation.

Rockchip socs use the same drm/kms driver throughout all socs but
multiple gpus are in use like Mali-Utgard, Mali-Midgard and also
PowerVR and Vivante.
To help with this the winsys part is set up to iterate through multiple
gpus and try each until a valid one is found. Right now of course only
the Lima-supported GPUs are implemented but it can be easily adapted
for the other ones, once Mesa drivers are available for them.
Signed-off-by: Heiko Stuebner <heiko@sntech.de>

Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>