mesa.git
7 years agost/nine: Upload on secondary context for Draw*Up
Axel Davy [Sun, 18 Dec 2016 18:21:08 +0000 (19:21 +0100)]
st/nine: Upload on secondary context for Draw*Up

Avoid synchronization by using the secondary context
for uploading the vertex data for Draw*Up.

v2: Rely on u_upload_mgr to use persistent coherent
buffers. Do not flush.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Dirty MANAGED buffers at Lock time
Axel Davy [Tue, 13 Dec 2016 00:16:21 +0000 (01:16 +0100)]
st/nine: Dirty MANAGED buffers at Lock time

Tests suggest MANAGED buffers are made dirty
at Lock time, not at Unlock time.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Implement new buffer upload path
Axel Davy [Sun, 4 Dec 2016 19:34:59 +0000 (20:34 +0100)]
st/nine: Implement new buffer upload path

This new buffer upload path enables to lock
faster than the normal path when using
DISCARD/NOOVERWRITE.

v2: Diverse cleanups and fixes.
v3: Fix allocation size for 'lone' buffers and
add more debug info.
v4: Rewrite of the path to handle when DISCARD/NOOVERWRITE
is not used anymore. The resource content is copied to the
new resource used.
v5: flush for safety after unmap (not sure it is really required
here, but safer to flush).
v6: Do not use the path if persistent coherent mapping is unavailable.
Fix buffer creation flags.
v7: Do not flush since it is not needed.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Allow non-zero resource offset for vertex buffers
Axel Davy [Sun, 4 Dec 2016 18:23:11 +0000 (19:23 +0100)]
st/nine: Allow non-zero resource offset for vertex buffers

Next patches will introduce an offset.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Do not wait for DEFAULT lock for volumes when we can
Axel Davy [Sun, 4 Dec 2016 09:34:34 +0000 (10:34 +0100)]
st/nine: Do not wait for DEFAULT lock for volumes when we can

If the volumes (and the texture container) are not referenced,
then they are no pending operations on them. We can lock directly.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Do not wait for DEFAULT lock for surfaces when we can
Axel Davy [Sun, 4 Dec 2016 09:33:19 +0000 (10:33 +0100)]
st/nine: Do not wait for DEFAULT lock for surfaces when we can

If the surfaces (and the texture container) are not referenced,
then they are no pending operations on them. We can lock directly.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Add arguments to context's blit and copy_region
Axel Davy [Sun, 4 Dec 2016 09:32:01 +0000 (10:32 +0100)]
st/nine: Add arguments to context's blit and copy_region

The new arguments enable to reference the objects while
the function hasn't run.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Idem for nine_context_gen_mipmap
Axel Davy [Sun, 4 Dec 2016 00:13:25 +0000 (01:13 +0100)]
st/nine: Idem for nine_context_gen_mipmap

Will enable to use the bind count as an information for
whether the surface/volume is used in the worker thread.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Bind destination for surface/volume uploads
Axel Davy [Sun, 4 Dec 2016 00:10:34 +0000 (01:10 +0100)]
st/nine: Bind destination for surface/volume uploads

Will enable to use the bind count as an information for
whether the surface/volume is used in the worker thread.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Use nine_context_box_upload for volumes
Axel Davy [Sat, 3 Dec 2016 23:29:16 +0000 (00:29 +0100)]
st/nine: Use nine_context_box_upload for volumes

Use nine_context_box_upload for uploads:
. systemmem volume to default volume
. managed volume internal content to its resource.

Check the uploads are executed before any action
that can alter the data, that is LockBox and
volume destruction.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Fix leak with volume dtor
Axel Davy [Sun, 4 Dec 2016 23:12:52 +0000 (00:12 +0100)]
st/nine: Fix leak with volume dtor

The last level was not released.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Fix leak with cubetexture dtor
Axel Davy [Sun, 4 Dec 2016 23:12:07 +0000 (00:12 +0100)]
st/nine: Fix leak with cubetexture dtor

The last level was not released.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Use nine_context_box_upload for surfaces
Axel Davy [Sat, 3 Dec 2016 23:19:45 +0000 (00:19 +0100)]
st/nine: Use nine_context_box_upload for surfaces

Use nine_context_box_upload for uploads:
. systemmem surface to default surface
. managed surface internal content to its resource.

Check the uploads are executed before any action
that can alter the data, that is LockRect,
NineSurface9_CopyDefaultToMem and surface destruction.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Implement nine_context_box_upload
Axel Davy [Sat, 3 Dec 2016 22:18:30 +0000 (23:18 +0100)]
st/nine: Implement nine_context_box_upload

This function will be used for surface and volume uploads

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Use nine_context_gen_mipmap in BaseTexture9
Axel Davy [Sat, 3 Dec 2016 22:04:07 +0000 (23:04 +0100)]
st/nine: Use nine_context_gen_mipmap in BaseTexture9

Generate mipmaps in the worker thread.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Implement nine_context_gen_mipmap
Axel Davy [Sat, 3 Dec 2016 22:01:56 +0000 (23:01 +0100)]
st/nine: Implement nine_context_gen_mipmap

To offload mipmap generation as well.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Optimize managed buffer upload
Axel Davy [Thu, 1 Dec 2016 21:50:22 +0000 (22:50 +0100)]
st/nine: Optimize managed buffer upload

Do the upload in the other thread.

Usually managed buffers are used once per frame.
It is then very likely pending_upload is 0 at Lock
time.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Implement nine_context_range_upload
Axel Davy [Thu, 1 Dec 2016 21:44:06 +0000 (22:44 +0100)]
st/nine: Implement nine_context_range_upload

Will be used to upload buffers.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Do not bind the container if forward is false
Axel Davy [Sat, 3 Dec 2016 18:41:40 +0000 (19:41 +0100)]
st/nine: Do not bind the container if forward is false

This doesn't make sense to bind the container in that specific case.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Comment and simplify iunknown
Axel Davy [Sat, 3 Dec 2016 18:37:06 +0000 (19:37 +0100)]
st/nine: Comment and simplify iunknown

The behaviour is a bit less obscure now.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Detach buffers in swapchain dtor.
Axel Davy [Sat, 3 Dec 2016 17:58:00 +0000 (18:58 +0100)]
st/nine: Detach buffers in swapchain dtor.

BackBuffers can survive swapchain dtor if
the user has a reference on them.

The swapchain itself has no reference on the buffer.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Fix NineUnknown_Detach
Axel Davy [Sat, 3 Dec 2016 17:46:26 +0000 (18:46 +0100)]
st/nine: Fix NineUnknown_Detach

We don't bind the container in AddRef.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Simplify ARG_BIND_REF
Axel Davy [Sat, 3 Dec 2016 16:52:12 +0000 (17:52 +0100)]
st/nine: Simplify ARG_BIND_REF

Remove some noop operations.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Avoid flushing the queue for queries GetData
Axel Davy [Sun, 13 Nov 2016 15:28:33 +0000 (16:28 +0100)]
st/nine: Avoid flushing the queue for queries GetData

Use the newly introduced counter to know when we don't
need synchronization.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Add CSMT_NO_WAIT_WITH_COUNTER
Patrick Rudolph [Sun, 13 Nov 2016 14:41:58 +0000 (15:41 +0100)]
st/nine: Add CSMT_NO_WAIT_WITH_COUNTER

Similar to the other macros, but introduces a counter,
which enables to know when the instructions has been
executed.

Signed-off-by: Patrick Rudolph <siro@das-labor.org>
7 years agost/nine: Use nine_context_clear_render_target
Axel Davy [Sun, 13 Nov 2016 11:47:58 +0000 (12:47 +0100)]
st/nine: Use nine_context_clear_render_target

Enables to not wait for the worker thread for ColorFill.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Optimize ColorFill
Axel Davy [Sun, 13 Nov 2016 11:41:53 +0000 (12:41 +0100)]
st/nine: Optimize ColorFill

When we lock the whole surface to overwrite it, we can use DISCARD.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Simplify ColorFill
Axel Davy [Sun, 13 Nov 2016 11:40:10 +0000 (12:40 +0100)]
st/nine: Simplify ColorFill

For render targets, NineSurface9_GetSurface is not
expected to fail.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: use get_pipe_acquire/release when possible
Axel Davy [Sat, 12 Nov 2016 22:36:35 +0000 (23:36 +0100)]
st/nine: use get_pipe_acquire/release when possible

Use the acquire/release semantic when we don't need
to wait for any pending command.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Implement Fast path for dynamic buffers and csmt
Axel Davy [Thu, 3 Nov 2016 22:37:46 +0000 (23:37 +0100)]
st/nine: Implement Fast path for dynamic buffers and csmt

Use the secondary pipe for DISCARD/NOOVERWRITE, which
avoids stalling to get the pipe from the worker thread.

v2: flush at unmap. This is required for example if
the driver does hidden draw calls or copies. In the case
of unsynchronized it is probably not required, but
it is more safe.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Add secondary pipe for device
Axel Davy [Thu, 3 Nov 2016 21:12:01 +0000 (22:12 +0100)]
st/nine: Add secondary pipe for device

The secondary pipe will be used for operations
that don't need synchronization.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Add nine_context_get_pipe_acquire/release
Axel Davy [Sat, 12 Nov 2016 21:24:04 +0000 (22:24 +0100)]
st/nine: Add nine_context_get_pipe_acquire/release

See commit for description.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: SYSTEMMEM ignores DISCARD.
Axel Davy [Tue, 13 Dec 2016 00:10:15 +0000 (01:10 +0100)]
st/nine: SYSTEMMEM ignores DISCARD.

Tests show SYSTEMMEM should ignore DISCARD.

Prevents game bugs with following patches reimplementing
DISCARD. Halo is affected.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Upload Managed buffers just before draw call using them
Axel Davy [Sun, 6 Nov 2016 11:38:38 +0000 (12:38 +0100)]
st/nine: Upload Managed buffers just before draw call using them

Previously we were uploading Managed buffers at the next draw call
after they were set dirty.
This is not the expected behaviour. Instead upload just before
draw call needing the content.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Track bindings for buffers
Axel Davy [Thu, 3 Nov 2016 20:53:53 +0000 (21:53 +0100)]
st/nine: Track bindings for buffers

Similar code than for textures.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Fix BASETEX_REGISTER_UPDATE
Axel Davy [Sun, 6 Nov 2016 11:06:22 +0000 (12:06 +0100)]
st/nine: Fix BASETEX_REGISTER_UPDATE

BASETEX_REGISTER_UPDATE was adding the texture to the list
of textures to upload in too many cases. tex->base.base.bind
will be set to true if the texture is in a stateblock, whereas
we want to upload only if bound to the device, which is
what bind_count is for.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Simplify the logic to bind textures
Axel Davy [Sun, 6 Nov 2016 11:05:50 +0000 (12:05 +0100)]
st/nine: Simplify the logic to bind textures

This makes the code more readable.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Use nine_context for resource_copy_region
Patrick Rudolph [Wed, 2 Nov 2016 16:17:21 +0000 (17:17 +0100)]
st/nine: Use nine_context for resource_copy_region

Use nine_context wrapper for resource_copy_region.
Enables to offload it with CSMT.

Signed-off-by: Patrick Rudolph <siro@das-labor.org>
7 years agost/nine: Use nine_context for blit
Patrick Rudolph [Thu, 27 Oct 2016 05:39:03 +0000 (07:39 +0200)]
st/nine: Use nine_context for blit

Enables to offload it with CSMT.

Signed-off-by: Patrick Rudolph <siro@das-labor.org>
7 years agost/nine: Add NINE_DEBUG=tid to turn threadid on or off
Patrick Rudolph [Fri, 28 Oct 2016 16:04:10 +0000 (18:04 +0200)]
st/nine: Add NINE_DEBUG=tid to turn threadid on or off

To ease debugging.

Signed-off-by: Patrick Rudolph <siro@das-labor.org>
7 years agost/nine: Print threadid in debug log
Patrick Rudolph [Fri, 28 Oct 2016 15:54:18 +0000 (17:54 +0200)]
st/nine: Print threadid in debug log

To ease debugging.

Signed-off-by: Patrick Rudolph <siro@das-labor.org>
7 years agost/nine: Implement gallium nine CSMT
Patrick Rudolph [Wed, 26 Oct 2016 15:11:00 +0000 (17:11 +0200)]
st/nine: Implement gallium nine CSMT

Use an offloading thread for all nine_context functions.
Macros are used to ease the reading of the code.

Signed-off-by: Patrick Rudolph <siro@das-labor.org>
Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Call GetPipe for implicit pipe usages
Axel Davy [Sun, 18 Dec 2016 09:59:53 +0000 (10:59 +0100)]
st/nine: Call GetPipe for implicit pipe usages

With csmt, every usage of the pipe in the main thread
has to be protected by calling GetPipe.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Add struct nine_clipplane
Patrick Rudolph [Tue, 25 Oct 2016 18:06:31 +0000 (20:06 +0200)]
st/nine: Add struct nine_clipplane

Required to know the size exact size of the plane.

Signed-off-by: Patrick Rudolph <siro@das-labor.org>
7 years agost/nine: Add nine_queue
Patrick Rudolph [Tue, 25 Oct 2016 16:03:44 +0000 (18:03 +0200)]
st/nine: Add nine_queue

This queue mechanism will be used for CSMT.

Signed-off-by: Patrick Rudolph <siro@das-labor.org>
7 years agost/nine: Create pipe_surfaces on resource creation.
Axel Davy [Fri, 4 Nov 2016 21:39:14 +0000 (22:39 +0100)]
st/nine: Create pipe_surfaces on resource creation.

Create the pipe_surfaces on renderable resources creation.

This enables to avoid creating them on the fly.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Back swvp in nine_context
Axel Davy [Tue, 1 Nov 2016 17:30:33 +0000 (18:30 +0100)]
st/nine: Back swvp in nine_context

Part of the refactor to move all gallium calls to
nine_state.c, and have all internal states required
for those calls in nine_context.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Change the way nine_shader gets the pipe
Axel Davy [Tue, 1 Nov 2016 17:23:05 +0000 (18:23 +0100)]
st/nine: Change the way nine_shader gets the pipe

The change is required with csmt, where depending on the thread
you don't access the pipe the same way.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Reimplement nine_context_apply_stateblock
Axel Davy [Tue, 1 Nov 2016 17:00:02 +0000 (18:00 +0100)]
st/nine: Reimplement nine_context_apply_stateblock

The new version uses nine_context functions instead of
applying the changes directly to nine_context.
This will enable it to work with CSMT.

v2: Fix nine_context_light_enable_stateblock
The memcpy arguments were wrong, and the state
wasn't set dirty.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Decompose nine_context_set_texture
Axel Davy [Tue, 1 Nov 2016 15:57:37 +0000 (16:57 +0100)]
st/nine: Decompose nine_context_set_texture

Part of the refactor to move all gallium calls to
nine_state.c, and have all internal states required
for those calls in nine_context.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Decompose nine_context_set_indices
Axel Davy [Tue, 1 Nov 2016 15:38:57 +0000 (16:38 +0100)]
st/nine: Decompose nine_context_set_indices

Part of the refactor to move all gallium calls to
nine_state.c, and have all internal states required
for those calls in nine_context.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Decompose nine_context_set_stream_source
Axel Davy [Wed, 26 Oct 2016 21:25:05 +0000 (23:25 +0200)]
st/nine: Decompose nine_context_set_stream_source

Part of the refactor to move all gallium calls to
nine_state.c, and have all internal states required
for those calls in nine_context.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Do not use NineBaseTexture9 in nine_context
Axel Davy [Mon, 31 Oct 2016 16:04:29 +0000 (17:04 +0100)]
st/nine: Do not use NineBaseTexture9 in nine_context

Some fields are subject to modification outside of nine_context
(SetLod, etc).

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Move Managed Pool handling out of nine_context
Axel Davy [Sun, 30 Oct 2016 14:00:48 +0000 (15:00 +0100)]
st/nine: Move Managed Pool handling out of nine_context

Part of the refactor to move all gallium calls to
nine_state.c, and have all internal states required
for those calls in nine_context.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Integrate nine_pipe_context_clear to nine_context_clear
Axel Davy [Fri, 28 Oct 2016 17:19:39 +0000 (19:19 +0200)]
st/nine: Integrate nine_pipe_context_clear to nine_context_clear

Part of the refactor to move all gallium calls to
nine_state.c, and have all internal states required
for those calls in nine_context.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Move pipe and cso to nine_context
Axel Davy [Fri, 28 Oct 2016 17:03:59 +0000 (19:03 +0200)]
st/nine: Move pipe and cso to nine_context

Part of the refactor to move all gallium calls to
nine_state.c, and have all internal states required
for those calls in nine_context.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Rename pipe to pipe_data in nine_context
Axel Davy [Fri, 28 Oct 2016 17:05:31 +0000 (19:05 +0200)]
st/nine: Rename pipe to pipe_data in nine_context

This patch it to avoid name conflict when device->pipe
will be moved to nine_context.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Rename cso in nine_context to cso_shader
Axel Davy [Fri, 28 Oct 2016 16:54:52 +0000 (18:54 +0200)]
st/nine: Rename cso in nine_context to cso_shader

This patch it to avoid name conflict when device->cso
is moved to nine_context.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Access pipe_context via NineDevice9_GetPipe
Axel Davy [Thu, 27 Oct 2016 22:43:56 +0000 (00:43 +0200)]
st/nine: Access pipe_context via NineDevice9_GetPipe

Except for nine_ff and nine_state.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Remove NineDevice9_GetCSO
Axel Davy [Thu, 27 Oct 2016 21:36:39 +0000 (23:36 +0200)]
st/nine: Remove NineDevice9_GetCSO

Was useless. Remove useless usage in
swapchain9.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Move query9 pipe calls to nine_context
Axel Davy [Thu, 27 Oct 2016 20:52:53 +0000 (22:52 +0200)]
st/nine: Move query9 pipe calls to nine_context

This will enable to use threading for them.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Use atomics for nine_bind
Axel Davy [Wed, 26 Oct 2016 21:15:34 +0000 (23:15 +0200)]
st/nine: Use atomics for nine_bind

nine_bind didn't need atomics up to now,
because it's use what always within a protected
mutex. We need to use atomics because with the
next patches several threads may use nine_bind.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Track dirty state groups in nine_context
Axel Davy [Thu, 20 Oct 2016 20:10:34 +0000 (22:10 +0200)]
st/nine: Track dirty state groups in nine_context

Part of the refactor to move all gallium calls to
nine_state.c, and have all internal states required
for those calls in nine_context.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Back User Clip Planes to nine_context
Axel Davy [Thu, 20 Oct 2016 19:59:01 +0000 (21:59 +0200)]
st/nine: Back User Clip Planes to nine_context

Part of the refactor to move all gallium calls to
nine_state.c, and have all internal states required
for those calls in nine_context.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Back ps to nine_context
Axel Davy [Thu, 20 Oct 2016 19:41:19 +0000 (21:41 +0200)]
st/nine: Back ps to nine_context

Part of the refactor to move all gallium calls to
nine_state.c, and have all internal states required
for those calls in nine_context.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Back ds to nine_context
Axel Davy [Thu, 20 Oct 2016 19:31:43 +0000 (21:31 +0200)]
st/nine: Back ds to nine_context

Part of the refactor to move all gallium calls to
nine_state.c, and have all internal states required
for those calls in nine_context.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Back all ff states in nine_context
Axel Davy [Wed, 19 Oct 2016 21:40:27 +0000 (23:40 +0200)]
st/nine: Back all ff states in nine_context

Part of the refactor to move all gallium calls to
nine_state.c, and have all internal states required
for those calls in nine_context.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Refactor LightEnable
Axel Davy [Wed, 19 Oct 2016 21:36:50 +0000 (23:36 +0200)]
st/nine: Refactor LightEnable

Call a helper function.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Refactor SetLight
Axel Davy [Wed, 19 Oct 2016 21:29:58 +0000 (23:29 +0200)]
st/nine: Refactor SetLight

Call a helper function to set the light.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Put ff data in a separate structure
Axel Davy [Wed, 19 Oct 2016 21:06:05 +0000 (23:06 +0200)]
st/nine: Put ff data in a separate structure

And make nine_state_access_transform take this
new structure as input.
Part of the refactor to move all gallium calls to
nine_state.c, and have all internal states required
for those calls in nine_context.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Back viewport to nine_context
Axel Davy [Wed, 19 Oct 2016 19:57:52 +0000 (21:57 +0200)]
st/nine: Back viewport to nine_context

Part of the refactor to move all gallium calls to
nine_state.c, and have all internal states required
for those calls in nine_context.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Back scissor to nine_context
Axel Davy [Wed, 19 Oct 2016 19:47:02 +0000 (21:47 +0200)]
st/nine: Back scissor to nine_context

Part of the refactor to move all gallium calls to
nine_state.c, and have all internal states required
for those calls in nine_context.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Back RT to nine_context
Axel Davy [Wed, 19 Oct 2016 19:38:47 +0000 (21:38 +0200)]
st/nine: Back RT to nine_context

Part of the refactor to move all gallium calls to
nine_state.c, and have all internal states required
for those calls in nine_context.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Back current index buffer to nine_context
Axel Davy [Wed, 19 Oct 2016 18:41:40 +0000 (20:41 +0200)]
st/nine: Back current index buffer to nine_context

Part of the refactor to move all gallium calls to
nine_state.c, and have all internal states required
for those calls in nine_context.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Back all shader constants to nine_context
Axel Davy [Tue, 18 Oct 2016 20:43:53 +0000 (22:43 +0200)]
st/nine: Back all shader constants to nine_context

For device vs shader float constants and may_swvp,
the same tips than for the other constant types is
used.

Also memset the constants properly.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Back sampler states to nine_context
Axel Davy [Tue, 18 Oct 2016 18:48:54 +0000 (20:48 +0200)]
st/nine: Back sampler states to nine_context

Part of the refactor to move all gallium calls to
nine_state.c, and have all internal states required
for those calls in nine_context.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Back vs to nine_context
Axel Davy [Mon, 17 Oct 2016 19:43:11 +0000 (21:43 +0200)]
st/nine: Back vs to nine_context

And move programmable_vs storage and computation.
Part of the refactor to move all gallium calls to
nine_state.c, and have all internal states required
for those calls in nine_context.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Back vdecl to nine_context
Axel Davy [Mon, 17 Oct 2016 19:18:42 +0000 (21:18 +0200)]
st/nine: Back vdecl to nine_context

Part of the refactor to move all gallium calls to
nine_state.c, and have all internal states required
for those calls in nine_context.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Move stream freq data to nine_context
Axel Davy [Sun, 16 Oct 2016 17:16:31 +0000 (19:16 +0200)]
st/nine: Move stream freq data to nine_context

Part of the refactor to move all gallium calls to
nine_state.c, and have all internal states required
for those calls in nine_context.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Move vtxbuf to nine_context
Axel Davy [Sun, 16 Oct 2016 16:17:27 +0000 (18:17 +0200)]
st/nine: Move vtxbuf to nine_context

Part of the refactor to move all gallium calls to
nine_state.c, and have all internal states required
for those calls in nine_context.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Move stream_usage_mask to nine_context
Axel Davy [Sun, 16 Oct 2016 15:49:09 +0000 (17:49 +0200)]
st/nine: Move stream_usage_mask to nine_context

Part of the refactor to move all gallium calls to
nine_state.c, and have all internal states required
for those calls in nine_context.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Back textures into nine_context
Axel Davy [Sun, 16 Oct 2016 14:09:51 +0000 (16:09 +0200)]
st/nine: Back textures into nine_context

Part of the refactor to move all gallium calls to
nine_state.c, and have all internal states required
for those calls in nine_context.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Move texture setting to nine_context_*
Axel Davy [Sun, 16 Oct 2016 13:56:14 +0000 (15:56 +0200)]
st/nine: Move texture setting to nine_context_*

And move samplers_shadow to nine_context.
Part of the refactor to move all gallium calls to
nine_state.c, and have all internal states required
for those calls in nine_context.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Track changed.texture only for stateblocks
Axel Davy [Sun, 16 Oct 2016 13:29:46 +0000 (15:29 +0200)]
st/nine: Track changed.texture only for stateblocks

Part of the refactor to move all gallium calls to
nine_state.c, and have all internal states required
for those calls in nine_context.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Move draw calls to nine_state
Axel Davy [Sun, 16 Oct 2016 09:34:35 +0000 (11:34 +0200)]
st/nine: Move draw calls to nine_state

Part of the refactor to move all gallium calls to
nine_state.c, and have all internal states required
for those calls in nine_context.

v2: Release buffers for Draw*Up functions in device9.c,
instead of nine_context. This prevents a leak with csmt
where the wrong pointers were released.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Move core of device clear to nine_state
Axel Davy [Sat, 15 Oct 2016 11:52:04 +0000 (13:52 +0200)]
st/nine: Move core of device clear to nine_state

Part of the refactor to move all gallium calls to
nine_state.c, and have all internal states required
for those calls in nine_context.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Introduce nine_context
Axel Davy [Sat, 15 Oct 2016 11:17:52 +0000 (13:17 +0200)]
st/nine: Introduce nine_context

nine_context is a new structure which goal will be
to contain all internal states. It will be the states
of the second thread in the to-be-introduced CSMT mode.

This patch moves several internal states to nine_context,
while the next patches add the other fields.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Implement WFOG properly
Axel Davy [Tue, 29 Nov 2016 23:00:25 +0000 (00:00 +0100)]
st/nine: Implement WFOG properly

We were advertising support for WFOG (like all win drivers),
but we weren't implementing it.
This patch implements the behaviour. See comments.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Fix ff texture coordinate selection
Axel Davy [Sun, 27 Nov 2016 15:23:37 +0000 (16:23 +0100)]
st/nine: Fix ff texture coordinate selection

The code was wrongly detecting which texture coordinates
to generate when the coordinate index was different to
the stage index.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Convert redundant check to assert in ff ps
Axel Davy [Sun, 27 Nov 2016 12:14:21 +0000 (13:14 +0100)]
st/nine: Convert redundant check to assert in ff ps

We disable the alpha stage if the color stage is disabled.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Fix two special cases in ff ps
Axel Davy [Sun, 27 Nov 2016 12:13:13 +0000 (13:13 +0100)]
st/nine: Fix two special cases in ff ps

if first alpha stage is disabled and writes to temp,
diffuse alpha is written to temp.
Last stage always writes to current.

Behaviour was deduced by tests with a test app.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Remove useless code in ff ps
Axel Davy [Sun, 27 Nov 2016 12:10:52 +0000 (13:10 +0100)]
st/nine: Remove useless code in ff ps

Current is already initialized to Diffuse.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Fix ff cases when stages should be disabled
Axel Davy [Sat, 26 Nov 2016 20:58:29 +0000 (21:58 +0100)]
st/nine: Fix ff cases when stages should be disabled

When a texture is read by a stage for colorop, it should
be disabled, and disable following stages.

When a texture is read for alphaop, 1.0f is read for the input,
which is the behaviour for a dummy texture.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Always initialize current in ff ps
Axel Davy [Sun, 27 Nov 2016 11:55:46 +0000 (12:55 +0100)]
st/nine: Always initialize current in ff ps

The check was not catching all possible cases.
NVE4 should be fine.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Fix check for ff specular
Axel Davy [Sat, 26 Nov 2016 20:47:44 +0000 (21:47 +0100)]
st/nine: Fix check for ff specular

Fix the check for computing ff specular.

This seems to match the opengl behavior,
and give the correct output on windows.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Do not saturate illumination coefficients in ff
Axel Davy [Sat, 26 Nov 2016 13:28:38 +0000 (14:28 +0100)]
st/nine: Do not saturate illumination coefficients in ff

Fixes bad rendering of a test app.
Wine has the same behaviour.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Fix ff COLOR0 w component computation
Axel Davy [Sat, 26 Nov 2016 13:25:44 +0000 (14:25 +0100)]
st/nine: Fix ff COLOR0 w component computation

The computation was wrong. COLOR0's last component
should be equal to the material diffuse w component.

The behaviour was checked with a test app on Windows.
Wine has the same behaviour.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Fix specular enable for alpha
Axel Davy [Tue, 22 Nov 2016 20:53:03 +0000 (21:53 +0100)]
st/nine: Fix specular enable for alpha

Apparently specular enable doesn't affect
the alpha channel.

Fixes https://github.com/iXit/Mesa-3D/issues/253

Behaviour comfirmed looking in wine sources.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agost/nine: Ignore MULTISAMPLEMASK when RT is not multisampled
Axel Davy [Sun, 20 Nov 2016 09:55:54 +0000 (10:55 +0100)]
st/nine: Ignore MULTISAMPLEMASK when RT is not multisampled

We were ignoring MULTISAMPLEMASK for non-maskable multisample
modes, but we were missing the non-multisampled case.

Fixes a crash in Halo.

Signed-off-by: Axel Davy <axel.davy@ens.fr>
7 years agodriconf: Fix missing gettext
Axel Davy [Tue, 6 Dec 2016 20:58:25 +0000 (21:58 +0100)]
driconf: Fix missing gettext

DRI_CONF_NINE_OVERRIDEVENDOR was missing gettext for the
description.

Signed-off-by: Axel Davy <axel.davy@ens.fr>