mesa.git
10 years agovc4: Fix viewport handling in the uniforms upload.
Eric Anholt [Mon, 18 Aug 2014 19:46:24 +0000 (12:46 -0700)]
vc4: Fix viewport handling in the uniforms upload.

I had the right viewports in vc4_emit.c, but grabbed the wrong values in
the uniform setup, so primitives would claim to be in the wrong parts of
the screen.  (The vc4_emit.c state looks like it just decides how big the
clipping guardband is).

This gets fbo-viewport closer to working (which still has the problem that
the HW is always guard-band clipping), and fixes inverted FBO rendering in
general.

10 years agodocs/relnotes: document GLX_MESA_query_renderer
Marek Olšák [Mon, 18 Aug 2014 22:26:01 +0000 (00:26 +0200)]
docs/relnotes: document GLX_MESA_query_renderer

10 years agoclover: Refuse to build a program if there are kernel objects attached to it.
Francisco Jerez [Sat, 16 Aug 2014 13:25:34 +0000 (16:25 +0300)]
clover: Refuse to build a program if there are kernel objects attached to it.

Fixes piglit cl-api-build-program.

Tested-by: EdB <edb+mesa@sigluy.net>
10 years agoclover/util: Pass initial count value to ref_counter constructor.
Francisco Jerez [Sun, 17 Aug 2014 20:26:49 +0000 (23:26 +0300)]
clover/util: Pass initial count value to ref_counter constructor.

And mark the ref_count() method as const.

Tested-by: EdB <edb+mesa@sigluy.net>
10 years agoclover/util: Implement minimalist reference to clover::ref_counter object.
Francisco Jerez [Sun, 17 Aug 2014 20:18:45 +0000 (23:18 +0300)]
clover/util: Implement minimalist reference to clover::ref_counter object.

Tested-by: EdB <edb+mesa@sigluy.net>
10 years agoclover: clGetProgramInfo support for OpenCL 1.2.
EdB [Tue, 5 Aug 2014 17:09:38 +0000 (19:09 +0200)]
clover: clGetProgramInfo support for OpenCL 1.2.

[ Francisco Jerez: Rework using fold() for conciseness. ]

Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Tested-by: EdB <edb+mesa@sigluy.net>
10 years agonouveau: don't keep stale pointer to free'd data
Ilia Mirkin [Sat, 16 Aug 2014 06:46:01 +0000 (02:46 -0400)]
nouveau: don't keep stale pointer to free'd data

If ->sys is non-null, we might decide that it's where the data is
stored.

Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: <mesa-stable@lists.freedesktop.org>
10 years agoegl: don't exit process on initialization failure
Ilia Mirkin [Mon, 11 Aug 2014 00:10:24 +0000 (20:10 -0400)]
egl: don't exit process on initialization failure

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
10 years agomesa: fix compressed_subtexture_error_check() return value
Brian Paul [Fri, 15 Aug 2014 22:55:40 +0000 (16:55 -0600)]
mesa: fix compressed_subtexture_error_check() return value

The function should return GLboolean, not GLenum.
If we detect invalid compressed pixel storage parameters, we should
return GL_TRUE, not GL_FALSE so that the function is no-op'd.
An update to the piglit s3tc-errors test will check this.

Reviewed-by: Matt Turner <mattst88@gmail.com>
10 years agomesa: move _mesa_compressed_texture_pixel_storage_error_check()
Brian Paul [Fri, 15 Aug 2014 22:28:59 +0000 (16:28 -0600)]
mesa: move _mesa_compressed_texture_pixel_storage_error_check()

to pixelstore.c, add const qualifier to the 'packing' parameter.
Add comments.

Reviewed-by: Matt Turner <mattst88@gmail.com>
10 years agomesa: minor improvements to _mesa_compute_compressed_pixelstore()
Brian Paul [Fri, 15 Aug 2014 21:21:01 +0000 (15:21 -0600)]
mesa: minor improvements to _mesa_compute_compressed_pixelstore()

Replace the gl_texture_image parameter with mesa_format since we only
used the image's format.

Add some comments.

Reviewed-by: Matt Turner <mattst88@gmail.com>
10 years agoutil: whitespace and formatting fixes in u_math.h
Brian Paul [Fri, 15 Aug 2014 22:21:15 +0000 (16:21 -0600)]
util: whitespace and formatting fixes in u_math.h

Trivial.

10 years agonouveau: make sure to invalidate any vbo state as well
Ilia Mirkin [Sat, 16 Aug 2014 05:00:39 +0000 (01:00 -0400)]
nouveau: make sure to invalidate any vbo state as well

Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: <mesa-stable@lists.freedesktop.org>
10 years agoi965/gen6: Force ALL_SLICES_AT_EACH_LOD for separate stencil/hiz
Jordan Justen [Wed, 28 May 2014 17:44:13 +0000 (10:44 -0700)]
i965/gen6: Force ALL_SLICES_AT_EACH_LOD for separate stencil/hiz

For gen6 we will use the ALL_SLICES_AT_EACH_LOD miptree layout for
separate stencil/hiz. This is needed because gen6 hiz and separate
stencil only support a single miplevel. When accessing the other LODs,
we will program a tile aligned offset for the bo.

PRM Volume 1, Part 1, 7.18.3.7.2 For separate stencil buffer [DevILK]
to [DevSNB]:
 "The separate stencil buffer does not support mip mapping, thus the
  storage for LODs other than LOD 0 is not needed."

We still allocate storage for the other stencil mip-levels within a
single texture, but each mip-level will use non-mip-array spacing.

PRM Volume 2, Part 1, 7.5.3 Hierarchical Depth Buffer
 "[DevSNB]: The hierarchical depth buffer does not support the LOD
  field, it is assumed by hardware to be zero. A separate
  hierarachical depth buffer is required for each LOD used, and the
  corresponding buffer’s state delivered to hardware each time a new
  depth buffer state with modified LOD is delivered."

We allocate storage for the other hiz mip-levels within a single
texture, but each mip-level will use non-mip-array spacing.

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoi965/gen6: Stencil/hiz needs an offset for LOD > 0
Jordan Justen [Wed, 28 May 2014 17:19:37 +0000 (10:19 -0700)]
i965/gen6: Stencil/hiz needs an offset for LOD > 0

Since gen6 separate stencil & hiz only supports LOD0, we need to
program an offset to the LOD when emitting the separate stencil/hiz.

v3:
 * Use new array_layout enum

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoi965/gen6: Force tile alignment for each stencil/hiz LOD
Jordan Justen [Wed, 28 May 2014 17:36:44 +0000 (10:36 -0700)]
i965/gen6: Force tile alignment for each stencil/hiz LOD

Gen6 doesn't support multiple miplevels for hiz and stencil.

Therefore, we must point to the LOD directly during rendering.

But, we also have removed the tile offsets from normal depth surfaces,
so we need to align each LOD to a tile boundary for hiz and stencil.

v3:
 * Use new array_layout enum

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoi965: Support array_layout == ALL_SLICES_AT_EACH_LOD for multiple LODs
Jordan Justen [Wed, 28 May 2014 16:30:39 +0000 (09:30 -0700)]
i965: Support array_layout == ALL_SLICES_AT_EACH_LOD for multiple LODs

Previously array_layout ALL_SLICES_AT_EACH_LOD was only used for array
spacing lod0 on gen7+ and therefore was only used with a single mip
level.

gen6 separate stencil & hiz only support LOD0, so we need to allocate
the miptree similar to gen7+ array spacing lod0, except we also need
space for multiple mip levels. (Since OpenGL stencil and depth support
multiple LODs.)

The miptree is allocated with tightly packed array slice spacing, but
we still also pack the miplevels into the region similar to a normal
multi mip level packing.

A 2D Array texture with 2 slices and multiple LODs would look somewhat
like this:

+----------+
|          |
|          |
+----------+
|          |
|          |
+----------+
+---+ +-+
|   | +-+
+---+ +-+
|   | :
+---+

v3:
 * Use new array_layout enum
 * ASCII art!

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoi965: Allow forcing miptree->array_layout = ALL_SLICES_AT_EACH_LOD
Jordan Justen [Wed, 28 May 2014 16:05:37 +0000 (09:05 -0700)]
i965: Allow forcing miptree->array_layout = ALL_SLICES_AT_EACH_LOD

gen6 does not support multiple miplevels with separate
stencil/hiz. Therefore we need to layout its miptree with no mipmap
spacing between the slices of each miplevel.

v3:
 * Use new array_layout enum

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoi965: Change mipmap array_spacing_lod0 to array_layout (enum)
Jordan Justen [Fri, 13 Jun 2014 22:46:32 +0000 (15:46 -0700)]
i965: Change mipmap array_spacing_lod0 to array_layout (enum)

We will want to setup gen6 separate stencil and hiz miptrees in a
layout that is similar to array_spacing_lod0. This is needed because
gen6 hiz and stencil only support a single mip-level.

In both use cases (gen7+ LOD0 spacing & gen6 separate stencil/hiz),
the array slices will be packed at each LOD without reserving extra
space for LODs within each array slice.

So, we generalize the name of this field and add comments to indicate
the old and new uses.

Motivation for the gen6 change comes from the PRM:

PRM Volume 1, Part 1, 7.18.3.7.2 For separate stencil buffer [DevILK]
to [DevSNB]:
 "The separate stencil buffer does not support mip mapping, thus the
  storage for LODs other than LOD 0 is not needed."

PRM Volume 2, Part 1, 7.5.3 Hierarchical Depth Buffer
 "[DevSNB]: The hierarchical depth buffer does not support the LOD
  field, it is assumed by hardware to be zero. A separate
  hierarachical depth buffer is required for each LOD used, and the
  corresponding buffer’s state delivered to hardware each time a new
  depth buffer state with modified LOD is delivered."

v2:
 * Rename array_spacing_lod0 to non_mip_arrays
v3:
 * Instead, replace array_spacing_lod0 with array_layout enum

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoi965/gen6 depth surface: program 3DSTATE_DEPTH_BUFFER to top of surface
Jordan Justen [Tue, 9 Jul 2013 22:36:32 +0000 (15:36 -0700)]
i965/gen6 depth surface: program 3DSTATE_DEPTH_BUFFER to top of surface

(bf25ee2 for gen6)

Previously we would always find the 2D sub-surface of interest,
and then program the surface to this location. Now we always
program the 3DSTATE_DEPTH_BUFFER at the start of the surface.
To select the lod/slice, we utilize the lod & minimum array
element fields.

We also must disable brw_workaround_depthstencil_alignment for
gen >= 6. Now the hardware will handle alignment when rendering
to additional slices/LODs.

v3:
 * Set depth_mt bo RELOC offset to 0, as was done in bf25ee2

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=56127
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoi965/gen6 fbo: make unmatched depth/stencil configs return unsupported
Jordan Justen [Fri, 19 Jul 2013 22:44:56 +0000 (15:44 -0700)]
i965/gen6 fbo: make unmatched depth/stencil configs return unsupported

(f3c886b for gen6)

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoi965/gen6 blorp depth: calculate base surface width/height
Jordan Justen [Tue, 9 Jul 2013 22:32:42 +0000 (15:32 -0700)]
i965/gen6 blorp depth: calculate base surface width/height

(e3a49e1 for gen6)

This will be used in 3DSTATE_DEPTH_BUFFER in a later patch.

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoi965/gen6 depth surface: calculate minimum array element being rendered
Jordan Justen [Tue, 9 Jul 2013 22:24:56 +0000 (15:24 -0700)]
i965/gen6 depth surface: calculate minimum array element being rendered

(a23cfb8 for gen6)

In layered rendering this will be 0. Otherwise it will be the
selected slice.

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoi965/gen6 depth surface: calculate LOD being rendered to
Jordan Justen [Tue, 9 Jul 2013 22:19:55 +0000 (15:19 -0700)]
i965/gen6 depth surface: calculate LOD being rendered to

(08ef1dd for gen6)

This will be used in 3DSTATE_DEPTH_BUFFER in a later patch.

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoi965/gen6 depth surface: calculate depth (array size) for depth surface
Jordan Justen [Tue, 9 Jul 2013 22:16:35 +0000 (15:16 -0700)]
i965/gen6 depth surface: calculate depth (array size) for depth surface

(bc1acaa for gen6)

This will be used in 3DSTATE_DEPTH_BUFFER in a later patch.

Note: Cube maps are treated as 2D arrays with 6 times as
many array elements as the cube map array would have.

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoi965/gen6 depth surface: calculate more specific surface type
Jordan Justen [Tue, 9 Jul 2013 21:56:38 +0000 (14:56 -0700)]
i965/gen6 depth surface: calculate more specific surface type

(171e633 for gen6)

This will be used in 3DSTATE_DEPTH_BUFFER in a later patch.

Note: Cube maps are treated as 2D arrays with 6 times as
many array elements as the cube map array would have.

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoi965/gen6_depth_state.c: Remove (gen != 6) code paths
Jordan Justen [Wed, 30 Jul 2014 18:20:48 +0000 (11:20 -0700)]
i965/gen6_depth_state.c: Remove (gen != 6) code paths

Since this code was branched from brw_misc_state.c, it had support for
gen != 6. We can now remove this.

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoi965: Split gen6 depth hiz state out from brw
Jordan Justen [Wed, 28 May 2014 23:02:12 +0000 (16:02 -0700)]
i965: Split gen6 depth hiz state out from brw

We will program the gen6 hiz depth state differently to enable layered
rendering on gen6.

v2:
 * Remove unneeded gen6_emit_depthbuffer as suggested by Topi

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoi965/gen6: Adjust render height in errata case for MSAA
Jordan Justen [Tue, 25 Feb 2014 19:18:25 +0000 (11:18 -0800)]
i965/gen6: Adjust render height in errata case for MSAA

In the gen6 PRM Volume 1 Part 1: Graphics Core, Section
7.18.3.7.1 (Surface Arrays For all surfaces other than separate
stencil buffer):

 "[DevSNB] Errata: Sampler MSAA Qpitch will be 4 greater than the
  value calculated in the equation above , for every other odd Surface
  Height starting from 1 i.e. 1,5,9,13"

Since this Qpitch errata only impacts the sampler, we have to adjust
the input for the rendering surface to achieve the same qpitch. For
the affected heights, we increment the height by 1 for the rendering
surface.

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Acked-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoi965/gen6: Add support for layered renderbuffers
Jordan Justen [Fri, 19 Apr 2013 08:39:51 +0000 (01:39 -0700)]
i965/gen6: Add support for layered renderbuffers

Rather than pointing the surface_state directly at a single
sub-image of the texture for rendering, we now point the
surface_state at the top level of the texture, and configure
the surface_state as needed based on this.

v2:
 * Use SET_FIELD as suggested by Topi
 * Simplify min_array_element assignment as suggested by Topi
v3:
 * Use irb->layer_count for depth instead of rb->Depth
 * Make gl_target const
 * depth - 1, not depth
v4:
 * Merge in dd43900b & b875f39e fixes to prevent 3D texture piglit
   regressions

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoi965/gen6_surface_state.c: Remove (gen < 6) code path
Jordan Justen [Wed, 30 Jul 2014 18:20:48 +0000 (11:20 -0700)]
i965/gen6_surface_state.c: Remove (gen < 6) code path

Since this code was branched from brw_wm_surface_state.c, it had
support for gen < 6. We can now remove this.

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoi965: Split gen6 renderbuffer surface state from gen5 and older
Jordan Justen [Wed, 28 May 2014 21:18:05 +0000 (14:18 -0700)]
i965: Split gen6 renderbuffer surface state from gen5 and older

We will program the gen6 renderbuffer surface state differently to
enable layered rendering on gen6.

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agometa: Use instanced rendering for layered clears.
Kenneth Graunke [Wed, 13 Aug 2014 23:03:01 +0000 (16:03 -0700)]
meta: Use instanced rendering for layered clears.

Layered rendering is part of OpenGL 3.2; GL_ARB_draw_instanced is part
of OpenGL 3.1.  As such, all drivers supporting layered rendering
already support gl_InstanceID.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
10 years agomesa: Expose vbo_exec_DrawArraysInstanced as _mesa_DrawArraysInstanced.
Kenneth Graunke [Wed, 13 Aug 2014 23:02:17 +0000 (16:02 -0700)]
mesa: Expose vbo_exec_DrawArraysInstanced as _mesa_DrawArraysInstanced.

So we can use it in meta.c.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kristian Høgsberg <krh@bitplanet.net>
10 years agoRevert "hud: don't overrun malloced arrays"
Dave Airlie [Fri, 15 Aug 2014 23:14:17 +0000 (09:14 +1000)]
Revert "hud: don't overrun malloced arrays"

This reverts commit 1cfcd0164e1be7d7b05b693f60a262ad735b7565.

This seems to cause r600g lockups,
https://bugs.freedesktop.org/show_bug.cgi?id=82628

Signed-off-by: Dave Airlie <airlied@redhat.com>
10 years agoi965: Guard access to gl_Layer by extension #ifdef
Kristian Høgsberg [Fri, 15 Aug 2014 22:59:59 +0000 (15:59 -0700)]
i965: Guard access to gl_Layer by extension #ifdef

Only assign gl_Layer if we have GL_AMD_vertex_shader_layer.  Gen6 doesn't
(currently) have that extension, but it also doesn't support layered
rendering.

Signed-off-by: Kristian Høgsberg <krh@bitplanet.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Tested-by: Jordan Justen <jordan.l.justen@intel.com>
10 years agogallium/vc4: PIPE_CAP_VIDEO_MEMORY return the amount of system ram
Emil Velikov [Fri, 15 Aug 2014 19:07:09 +0000 (20:07 +0100)]
gallium/vc4: PIPE_CAP_VIDEO_MEMORY return the amount of system ram

Suggested-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
10 years agovc4: Add support for blending.
Eric Anholt [Thu, 14 Aug 2014 20:27:11 +0000 (13:27 -0700)]
vc4: Add support for blending.

Passes blendminmax and blendsquare.  glean's more serious blendFunc fails
in simulation due to binner memory overflow (I really need to work around
that), and fbo-blending-formats fails due to Mesa refusing one of the
getter requests, even before it could fail due to the driver not actually
supporting different formats yet.

10 years agovc4: Drop incorrect attempt to incorrectly invert the primconvert hw_mask.
Eric Anholt [Wed, 13 Aug 2014 17:43:26 +0000 (10:43 -0700)]
vc4: Drop incorrect attempt to incorrectly invert the primconvert hw_mask.

The hw_mask is the set of primitives you actually support, so this attempt
to provide the set of formats that's unsupported was wrong in two ways (it
was intended to be '~' not '!').  However, we only call this code when
prim isn't one of the actually supported hw_mask bits, so missing out on
the memcpy didn't matter anyway.

10 years agovc4: Use cl_f() instead of cl_u32(fui())
Eric Anholt [Wed, 13 Aug 2014 21:46:06 +0000 (14:46 -0700)]
vc4: Use cl_f() instead of cl_u32(fui())

10 years agovc4: Consistently use qir_uniform_f().
Eric Anholt [Wed, 13 Aug 2014 21:44:57 +0000 (14:44 -0700)]
vc4: Consistently use qir_uniform_f().

10 years agovc4: Consume the implicit varyings for points and lines.
Eric Anholt [Tue, 12 Aug 2014 22:50:53 +0000 (15:50 -0700)]
vc4: Consume the implicit varyings for points and lines.

We were triggering simulator assertion failures for not consuming these,
and presumably we want to actually make use of them some day (for things
like point/line antialiasing)

Note that this has the qreg index as 0, which is the same index as the
first GL varyings read.  This doesn't matter currently, since that number
isn't used for anything except dumping.

10 years agovc4: Move the deref of the color buffer for simulator into the simulator.
Eric Anholt [Tue, 12 Aug 2014 23:35:33 +0000 (16:35 -0700)]
vc4: Move the deref of the color buffer for simulator into the simulator.

At some point I'm going to want to move the information necessary for the
host buffer upload/download into the BO so that it's independent of the
current vc4->framebuffer, but for now this fixes pointless derefs on
non-simulator in vc4_context.c since the dump_fbo() removal

10 years agoi965: Implement fast color clears using meta operations
Kristian Høgsberg [Mon, 7 Jul 2014 23:44:58 +0000 (16:44 -0700)]
i965: Implement fast color clears using meta operations

This patch uses the infrastructure put in place by previous patches
to implement fast color clears and replicated color clears in terms of
meta operations.

This works all the way back to gen7 where fast clear was introduced and
adds support for fast clear on gen8.  It replaces the blorp path
completely and improves on a few cases.  Layered clears are now done
using instanced rendering and multiple render-target clears use a
MRT shader with rep16 writes.

Signed-off-by: Kristian Høgsberg <krh@bitplanet.net>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoi965: Add optimization pass to let us use the replicate data message
Kristian Høgsberg [Mon, 7 Jul 2014 22:27:17 +0000 (15:27 -0700)]
i965: Add optimization pass to let us use the replicate data message

The data port has a SIMD16 'replicate data' message, which lets us write
the same color for all 16 pixels by sending the four floats in the
lower half of a register instead of sending 4 times 16 identical
component values in 8 registers.

The message comes with a lot of restrictions and could be made generally
useful by recognizing when those restriction are satisfied.  For now,
this lets us enable the optimization when we know it's safe, but we don't
enable it by default.  The optimization works for simple color clear shaders
only, but does recognized and support multiple render targets.

Signed-off-by: Kristian Høgsberg <krh@bitplanet.net>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agometa: Export _mesa_meta_drawbuffers_from_bitfield()
Kristian Høgsberg [Sun, 10 Aug 2014 23:00:31 +0000 (16:00 -0700)]
meta: Export _mesa_meta_drawbuffers_from_bitfield()

We'll use this in the i965 fast clear implementation.

Signed-off-by: Kristian Høgsberg <krh@bitplanet.net>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
10 years agomesa: Use _mesa_lock_context_textures in _mesa_GetTexParameterfv()
Kristian Høgsberg [Fri, 8 Aug 2014 23:53:53 +0000 (16:53 -0700)]
mesa: Use _mesa_lock_context_textures in _mesa_GetTexParameterfv()

GetTexParamterfv() doesnt change texture state, so instead of
_mesa_lock_texture() we can use _mesa_lock_context_textures(),
which doesn't increase the texture stamp.  With this change,
_mesa_update_state_locked() is now only called from under
_mesa_lock_context_textures(), which is right thing to do.  Right now
it's the same mutex, but if we made texture locking more fine grained
locking one day, just locking one texture here would be wrong.

This all ignores the fact that texture locking seem a bit
flaky and broken, but we're trying to not blatantly make it worse.

This change allows us to reliably unlock the context textures in the
dd::UpdateState callback as is necessary for meta color resolves.

Signed-off-by: Kristian Høgsberg <krh@bitplanet.net>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoi965: Move pre-draw resolve buffers to dd::UpdateState
Kristian Høgsberg [Fri, 8 Aug 2014 23:24:00 +0000 (16:24 -0700)]
i965: Move pre-draw resolve buffers to dd::UpdateState

No functional change except for glBegin/glEnd style rendering, where we now
do the resolves at glBegin time instead of FLUSH_VERTICES time.  This is also
the reason for this change, so that when we later switch fast clear resolve to
use meta, we won't be doing meta operations in the middle of a begin/end
sequence.

Signed-off-by: Kristian Høgsberg <krh@bitplanet.net>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoi965: Provide a context flag to let us enable fast clear
Kristian Høgsberg [Mon, 7 Jul 2014 23:27:31 +0000 (16:27 -0700)]
i965: Provide a context flag to let us enable fast clear

GEN7+ has the fast clear functionality, which lets us clear the color
buffers using the MCS and a scaled down rectangle.  To enable this
we have to set the appropriate bits in the 3DSTATE_PS package.

Signed-off-by: Kristian Høgsberg <krh@bitplanet.net>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoi965: Disable clipping when rendering 3DPRIM_RECTLIST primitives
Kristian Høgsberg [Mon, 7 Jul 2014 22:15:31 +0000 (15:15 -0700)]
i965: Disable clipping when rendering 3DPRIM_RECTLIST primitives

The clipper doesn't support clipping 3DPRIM_RECTLIST primitives and must
be turned off when we use them.

Signed-off-by: Kristian Høgsberg <krh@bitplanet.net>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoi965: Add a mechanism for sending native primitives into the driver
Kristian Høgsberg [Mon, 7 Jul 2014 22:11:45 +0000 (15:11 -0700)]
i965: Add a mechanism for sending native primitives into the driver

The brw_draw_prims() function is the draw entry point into the driver,
and takes struct _mesa_prim for input.  We want to be able to feed
native primitives into the driver, and to that end we introduce
BRW_PRIM_OFFSET, which lets use describe geometry using the native
GEN primitive types.

Signed-off-by: Kristian Høgsberg <krh@bitplanet.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoi965: Add context flag to disable the viewport transform
Kristian Høgsberg [Mon, 7 Jul 2014 22:00:46 +0000 (15:00 -0700)]
i965: Add context flag to disable the viewport transform

This lets us disable the viewport transform, which will be useful for
emitting 3DPRIM_RECTLIST.

Signed-off-by: Kristian Høgsberg <krh@bitplanet.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoi965: Add an option to not generate the SIMD8 fragment shader
Kristian Høgsberg [Mon, 7 Jul 2014 21:49:12 +0000 (14:49 -0700)]
i965: Add an option to not generate the SIMD8 fragment shader

For now, this can only be triggered with a new 'no8' INTEL_DEBUG option
and a new context flag.  We'll use the context flag later, but introducing
it now lets us bisect to this commit if it breaks something.

Signed-off-by: Kristian Høgsberg <krh@bitplanet.net>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agodocs/autoconf: explicitly mention PKG_CONFIG_PATH for cross/multilib builds
Emil Velikov [Fri, 15 Aug 2014 16:58:14 +0000 (17:58 +0100)]
docs/autoconf: explicitly mention PKG_CONFIG_PATH for cross/multilib builds

... and squash a couple of typos.

Suggested-by: Eero Tamminen <eero.t.tamminen@intel.com>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
10 years agost/dri: Add __DRI2rendererQueryExtension support
Emil Velikov [Thu, 14 Aug 2014 20:20:37 +0000 (21:20 +0100)]
st/dri: Add __DRI2rendererQueryExtension support

The final step to get GLX_MESA_query_renderer working with gallium
drivers.

v2: Remove __DRI2_RENDERER_PREFERRED_PROFILE handling. It's already
handled in dri/common. Spotted by Marek.

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
10 years agogallium/softpipe/llvmpipe: handle query_renderer caps
Emil Velikov [Thu, 14 Aug 2014 20:15:06 +0000 (21:15 +0100)]
gallium/softpipe/llvmpipe: handle query_renderer caps

Both report 0xffffffff as both vendor and device id, and the maximum
amount of system memory as video memory.

v2: Use aux helper os_get_total_physical_memory().

Cc: Brian Paul <brianp@vmware.com>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
10 years agogallium/svga: handle query_rendered caps
Emil Velikov [Thu, 14 Aug 2014 20:09:43 +0000 (21:09 +0100)]
gallium/svga: handle query_rendered caps

All the values are are currently hardcoded. One could use
some heuristics to determine the amount of video memory if
a callback to the host is not available.

Do we what to advertise the driver as hardwar accelerated ?

Cc: Brian Paul <brianp@vmware.com>
Cc: José Fonseca <jose.r.fonseca@gmail.com>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
10 years agogallium/nouveau: handle query_renderer caps
Emil Velikov [Thu, 14 Aug 2014 20:05:35 +0000 (21:05 +0100)]
gallium/nouveau: handle query_renderer caps

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
10 years agogallium/vc4: handle query_renderer caps
Emil Velikov [Thu, 14 Aug 2014 20:36:07 +0000 (21:36 +0100)]
gallium/vc4: handle query_renderer caps

Cc: Eric Anholt <eric@anholt.net>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
10 years agogallium/r300/r600/radeonsi: handle query_renderer caps
Emil Velikov [Thu, 14 Aug 2014 19:57:29 +0000 (20:57 +0100)]
gallium/r300/r600/radeonsi: handle query_renderer caps

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
10 years agogallium/ilo: handle query_renderer caps
Emil Velikov [Thu, 14 Aug 2014 19:46:29 +0000 (20:46 +0100)]
gallium/ilo: handle query_renderer caps

Implementation based on the classic driver with the following
changes:
 - Use auxiliarry function os_get_total_physical_memory to get the
total amount of memory.
 - Move the libdrm_intel specific get_aperture_size to the winsys.

Cc: Chia-I Wu <olvaffe@gmail.com>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
10 years agogallium/i915: handle query_renderer caps
Emil Velikov [Thu, 14 Aug 2014 19:11:32 +0000 (20:11 +0100)]
gallium/i915: handle query_renderer caps

Implementation based on the classic driver with the following
changes:
 - Use auxiliarry function os_get_total_physical_memory to get the
total amount of memory.
 - Move the libdrm_intel specific get_aperture_size to the winsys.

Cc: Stephane Marchesin <stephane.marchesin@gmail.com>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
10 years agogallium/freedreno: handle query_renderer caps
Emil Velikov [Thu, 14 Aug 2014 18:42:39 +0000 (19:42 +0100)]
gallium/freedreno: handle query_renderer caps

Provide the real vendor and and hardcode the device id as
0xffffffff as the devices currently using freedreno are non-pci.
The device features UMA.

Cc: Rob Clark <robclark@freedesktop.org>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Rob Clark <robdclark@gmail.com>
10 years agoauxiliary/os: introduce os_get_total_physical_memory helper function
Emil Velikov [Fri, 28 Feb 2014 03:34:51 +0000 (03:34 +0000)]
auxiliary/os: introduce os_get_total_physical_memory helper function

Cc: Alexander von Gluck IV <kallisti5@unixzen.com>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
10 years agogallium: add GLX_MESA_query_renderer caps
Emil Velikov [Fri, 15 Aug 2014 16:36:34 +0000 (17:36 +0100)]
gallium: add GLX_MESA_query_renderer caps

Namely vendor/device id, accelerated and UMA, which will be used to describe
the underlying renderer.

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
10 years agodri/swrast: add GLX_MESA_query_renderer support
Emil Velikov [Fri, 14 Feb 2014 23:28:25 +0000 (23:28 +0000)]
dri/swrast: add GLX_MESA_query_renderer support

v2:
 - Drop __DRI2_RENDERER_PREFERRED_PROFILE case.
 - Cleanup return statements.

Cc: Brian Paul <brianp@vmware.com>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
10 years agodri/radeon: add GLX_MESA_query_renderer support
Emil Velikov [Thu, 20 Feb 2014 01:07:35 +0000 (01:07 +0000)]
dri/radeon: add GLX_MESA_query_renderer support

 - Create radeon{Vendor,GetRenderer}String helpers.
 - Drop __DRI2_RENDERER_PREFERRED_PROFILE case.
 - Cleanup return statements.

To be used by the upcomming GLX_MESA_query_renderer implementation.

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
10 years agodri/radeon: don't print TCL status on glGetString(GL_RENDERER)
Emil Velikov [Wed, 19 Feb 2014 21:53:38 +0000 (21:53 +0000)]
dri/radeon: don't print TCL status on glGetString(GL_RENDERER)

Printing the TCL involves that context is available at the time of
query. The GLX_MESA_query_renderer states that glGetString(GL_RENDERER)
and glXQueryRendererStringMESA(GLX_RENDERER_DEVICE_ID_MESA) will have
the same format, thus removing the context dependenicy will help us
achieve that.

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
10 years agodri/nouveau: add GLX_MESA_query_renderer support
Emil Velikov [Fri, 14 Feb 2014 23:34:47 +0000 (23:34 +0000)]
dri/nouveau: add GLX_MESA_query_renderer support

 - Create nouveau_{vendor,get_renderer}_string helpers.
 - Set correct max_gl*version.
 - Query the device PCIID via libdrm_nouveau/nouveau_getparam.

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
10 years agodri/common: Move __DRI2_RENDERER_PREFFERED_PROFILE handling to driQueryRendererIntege...
Emil Velikov [Sun, 23 Feb 2014 22:03:48 +0000 (22:03 +0000)]
dri/common: Move __DRI2_RENDERER_PREFFERED_PROFILE handling to driQueryRendererIntegerCommon

Essentially all drivers would like to use to opengl core profile if
available, so avoid duplication by moving the code to a common fallback
within driQueryRendererIntegerCommon.

If a driver uses different approach they can handle it separately.

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
10 years agoglx/drisw: add support for DRI2rendererQueryExtension
Emil Velikov [Thu, 14 Aug 2014 17:26:55 +0000 (18:26 +0100)]
glx/drisw: add support for DRI2rendererQueryExtension

The extension is used by GLX_MESA_query_renderer, which
can be provided for by hardware and software drivers.

v2: Use designated initializers.
v3: Move drisw_query_renderer_*() to dri2_query_renderer.c

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
10 years agoglx/dri2: use mapping table for dri2_convert_glx_query_renderer_attribs()
Emil Velikov [Thu, 20 Feb 2014 00:23:41 +0000 (00:23 +0000)]
glx/dri2: use mapping table for dri2_convert_glx_query_renderer_attribs()

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
10 years agoglx/drisw: Move private structure declarations to a header file
Emil Velikov [Fri, 21 Feb 2014 23:09:16 +0000 (23:09 +0000)]
glx/drisw: Move private structure declarations to a header file

v2: Reff the correct file wrt copyright, spotted by Chia-I

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
10 years agomesa: check if GL_ARB_copy_image is enabled in _mesa_CopyImageSubData()
Brian Paul [Fri, 15 Aug 2014 13:40:47 +0000 (07:40 -0600)]
mesa: check if GL_ARB_copy_image is enabled in _mesa_CopyImageSubData()

Generate a GL error and return rather than crashing on a null
ctx->Driver.CopyImageSubData pointer (gallium).  This allows apitraces
with glCopyImageSubData() calls to continue rather than crash.

Plus, fix a comment typo.

Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
10 years agoi965/blorp_clear: Use memcpy instead of assignment to copy clear value
Neil Roberts [Thu, 14 Aug 2014 14:01:39 +0000 (15:01 +0100)]
i965/blorp_clear: Use memcpy instead of assignment to copy clear value

Similar to the problem described in 2c50212b14da27de4e3, if we copy the clear
value through a regular assignment via a floating point value, then if an
integer clear value is being used that happens to contain a signalling NaN
value then it would get converted to a quiet NaN when stored via the x87
floating-point registers. This would corrupt the integer value. Instead we
should use a memcpy to ensure the exact bit representation is preserved.

This bug can be triggered on 32-bit builds with optimisations by using an
integer clear color with a value like 0x7f817f81.

Reviewed-by: Matt Turner <mattst88@gmail.com>
10 years agor600g: Implement ARB_derivative_control
Glenn Kennard [Fri, 15 Aug 2014 08:30:58 +0000 (10:30 +0200)]
r600g: Implement ARB_derivative_control

Requires Evergreen/Cayman

marek: update release notes

Signed-off-by: Glenn Kennard <glenn.kennard@gmail.com>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
10 years agodocs: Update relnotes for ARB_gpu_shader5
Chris Forbes [Sun, 10 Aug 2014 04:22:35 +0000 (16:22 +1200)]
docs: Update relnotes for ARB_gpu_shader5

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Matt Turner <mattst88@gmail.com>
10 years agodocs: Mark off ARB_gpu_shader5 for i965
Chris Forbes [Sun, 10 Aug 2014 04:08:36 +0000 (16:08 +1200)]
docs: Mark off ARB_gpu_shader5 for i965

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Matt Turner <mattst88@gmail.com>
10 years agoi965: Enable ARB_gpu_shader5 on Gen7
Chris Forbes [Sun, 10 Aug 2014 02:36:25 +0000 (14:36 +1200)]
i965: Enable ARB_gpu_shader5 on Gen7

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Matt Turner <mattst88@gmail.com>
10 years agoi965/fs: Add support for nonconst sampler indexing in FS visitor
Chris Forbes [Sun, 10 Aug 2014 00:12:06 +0000 (12:12 +1200)]
i965/fs: Add support for nonconst sampler indexing in FS visitor

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoi965/fs: Add support for non-const sampler indices in generator
Chris Forbes [Sun, 10 Aug 2014 00:02:22 +0000 (12:02 +1200)]
i965/fs: Add support for non-const sampler indices in generator

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoi965/fs: Refactor generate_tex in prep for nonconst sampler indexing
Chris Forbes [Sat, 9 Aug 2014 23:58:06 +0000 (11:58 +1200)]
i965/fs: Refactor generate_tex in prep for nonconst sampler indexing

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoi965/fs: Use brw_adjust_sampler_state_pointer in fs generator too
Chris Forbes [Sat, 9 Aug 2014 23:54:47 +0000 (11:54 +1200)]
i965/fs: Use brw_adjust_sampler_state_pointer in fs generator too

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoi965/vec4: Add support for nonconst sampler indexing in VS visitor
Chris Forbes [Sat, 9 Aug 2014 10:08:11 +0000 (22:08 +1200)]
i965/vec4: Add support for nonconst sampler indexing in VS visitor

V2: Set force_writemask_all on ADD; this *is* necessary in the VS case
too.

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoi965/vec4: Add support for non-const sampler indices in generator
Chris Forbes [Sat, 9 Aug 2014 23:42:08 +0000 (11:42 +1200)]
i965/vec4: Add support for non-const sampler indices in generator

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoi965: Generalize sampler state pointer mangling for non-const
Chris Forbes [Sat, 9 Aug 2014 23:39:44 +0000 (11:39 +1200)]
i965: Generalize sampler state pointer mangling for non-const

For now, assume that the addressed sampler can be in any of the
16-sampler banks. If we preserved range information this far, we
could avoid emitting these instructions if the sampler were known
to be contained within one bank.

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoi965/vec4: Refactor generate_tex in prep for non-const samplers
Chris Forbes [Sat, 9 Aug 2014 23:50:16 +0000 (11:50 +1200)]
i965/vec4: Refactor generate_tex in prep for non-const samplers

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoi965: Extract helper function for surface state pointer adjustment
Chris Forbes [Sat, 9 Aug 2014 23:49:31 +0000 (11:49 +1200)]
i965: Extract helper function for surface state pointer adjustment

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agodocs: Mark off ARB_gpu_shader5 UBO array indexing for i965
Chris Forbes [Sun, 3 Aug 2014 00:32:10 +0000 (12:32 +1200)]
docs: Mark off ARB_gpu_shader5 UBO array indexing for i965

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
10 years agoi965/vec4: Add visitor support for nonconst ubo block indexing
Chris Forbes [Sun, 29 Jun 2014 00:30:04 +0000 (12:30 +1200)]
i965/vec4: Add visitor support for nonconst ubo block indexing

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
10 years agoi965/vec4: Generate indirect sends for nonconstant UBO array access
Chris Forbes [Sat, 12 Jul 2014 23:43:11 +0000 (11:43 +1200)]
i965/vec4: Generate indirect sends for nonconstant UBO array access

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
10 years agoi965/fs: Add visitor support for nonconstant UBO indices
Chris Forbes [Sun, 18 May 2014 00:38:57 +0000 (12:38 +1200)]
i965/fs: Add visitor support for nonconstant UBO indices

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
10 years agoi965/fs: Generate indirect sends for nonconstant UBO array accesses
Chris Forbes [Sat, 2 Aug 2014 02:27:21 +0000 (14:27 +1200)]
i965/fs: Generate indirect sends for nonconstant UBO array accesses

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
10 years agoi965: Adjust set_message_descriptor to handle non-sends
Chris Forbes [Sun, 3 Aug 2014 00:02:25 +0000 (12:02 +1200)]
i965: Adjust set_message_descriptor to handle non-sends

We're about to be using this infrastructure to build descriptors in
src1 of non-send instructions, when preparing to do an indirect send.

Don't accidentally clobber the conditionalmod field of those
instructions with SFID bits, which aren't part of the descriptor.

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
10 years agoi965: Add low-level support for indirect sends
Chris Forbes [Sat, 2 Aug 2014 23:44:25 +0000 (11:44 +1200)]
i965: Add low-level support for indirect sends

This provides a reasonable place to enforce the hardware restriction
that indirect descriptors must be in a0.0

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
10 years agoi965/fs: Add pass to rename registers to break live ranges.
Kenneth Graunke [Mon, 14 Apr 2014 22:01:37 +0000 (15:01 -0700)]
i965/fs: Add pass to rename registers to break live ranges.

The pass breaks live ranges of virtual registers by allocating new
registers when it sees an assignment to a virtual GRF it's already seen
written.

total instructions in shared programs: 4337879 -> 4335014 (-0.07%)
instructions in affected programs:     343865 -> 341000 (-0.83%)
GAINED:                                46
LOST:                                  1

[mattst88]: Make pass not break in presence of control flow.
            invalidate_live_intervals() only if progress.
            Fix up delta_x/delta_y.

Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
10 years agoi965: Fix INTDIV math assertions on Broadwell.
Kenneth Graunke [Fri, 15 Aug 2014 03:14:34 +0000 (20:14 -0700)]
i965: Fix INTDIV math assertions on Broadwell.

Commit c66d928f2c9fa59e162c391fbdd37df969959718 ("i965: Enable INTDIV
in SIMD16 mode.") began using generate_math_gen6 to break SIMD16 INTDIV
into two SIMD8 operations.

generate_math_gen6 takes two registers - for unary operations, we pass
ARF null for the second operand.  Prior to Broadwell, real operands were
always GRF.  But now they can be IMM as well.

So, check for != ARF instead of == GRF.

+12 piglits.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
10 years agoRevert "i965/vec4: Use MOV, not OR, to set URB write channel mask bits."
Kenneth Graunke [Thu, 14 Aug 2014 19:22:16 +0000 (12:22 -0700)]
Revert "i965/vec4: Use MOV, not OR, to set URB write channel mask bits."

This reverts commit af13cf609f4257768ad8b80be8cec7f2e6ca8c81, which
appears to cause huge performance problems on Ivybridge.  I'd missed
that the FFTID bits are in the low byte.  The documentation doesn't
indicate that the URB write message header actually wants FFTID - it
just labels those bits as "Reserved."  But it appears necessary.

This does slightly more than revert the original change: originally,
Broadwell had separate code generation, which used MOV, and this patch
only changed it for Gen4-7.  Now that both are unified, reverting this
also makes Broadwell use OR.  Which should be fine.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agodocs: Mark off ARB_derivative_control for i965.
Chris Forbes [Thu, 14 Aug 2014 09:14:47 +0000 (21:14 +1200)]
docs: Mark off ARB_derivative_control for i965.

Also update 10.3 relnotes to match.

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Matt Turner <mattst88@gmail.com>
10 years agoi965: Enable ARB_derivative_control on Gen7+.
Chris Forbes [Thu, 14 Aug 2014 09:11:28 +0000 (21:11 +1200)]
i965: Enable ARB_derivative_control on Gen7+.

The extension says GL 4.0 is required. We'll meet the spirit
of that restriction by enabling on just those generations which will
soon support GL 4.0 (Gen7+), although it's technically supportable on
all generations.

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Matt Turner <mattst88@gmail.com>