mesa.git
9 years agoilo: align vertex buffer size in buf_create()
Chia-I Wu [Mon, 22 Jun 2015 06:15:52 +0000 (14:15 +0800)]
ilo: align vertex buffer size in buf_create()

With ilo_format.[ch] moved out of core, the aligning of vertex buffers does
not belong to core anymore.

9 years agoilo: move ilo_format.[ch] out of core
Chia-I Wu [Mon, 22 Jun 2015 06:06:13 +0000 (14:06 +0800)]
ilo: move ilo_format.[ch] out of core

They provide PIPE_FORMAT_x to GEN6_FORMAT_x translation as well as some
convenient helpers.  Move them out of core.

9 years agoilo: add ilo_state_surface_valid_format()
Chia-I Wu [Mon, 22 Jun 2015 05:37:05 +0000 (13:37 +0800)]
ilo: add ilo_state_surface_valid_format()

Check if a surface format can be used for the specified access type.

9 years agoilo: add ilo_state_vf_valid_element_format()
Chia-I Wu [Mon, 22 Jun 2015 05:15:24 +0000 (13:15 +0800)]
ilo: add ilo_state_vf_valid_element_format()

Check if a surface format can be used as a VE format.

9 years agonvc0: use NV_VRAM_DOMAIN() macro
Alexandre Courbot [Fri, 17 Oct 2014 06:05:32 +0000 (15:05 +0900)]
nvc0: use NV_VRAM_DOMAIN() macro

Use the newly-introduced NV_VRAM_DOMAIN() macro to support alternative
VRAM domains for chips that do not have dedicated video memory.

Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Martin Peres <martin.peres@free.fr>
9 years agonouveau: support for custom VRAM domains
Alexandre Courbot [Fri, 17 Oct 2014 05:58:11 +0000 (14:58 +0900)]
nouveau: support for custom VRAM domains

Some GPUs (e.g. GK20A, GM20B) do not embed VRAM of their own and use
the system memory as a backend instead. For such systems, allocating
objects in VRAM results in errors since the kernel will not allow
VRAM objects allocations.

This patch adds a vram_domain member to struct nouveau_screen that can
optionally be initialized to an alternative domain to use for VRAM
allocations. If left untouched, NOUVEAU_BO_VRAM will be used for
systems that embed VRAM, and NOUVEAU_BO_GART will be used for VRAM-less
systems.

Code that uses GPU objects is then expected to use the NV_VRAM_DOMAIN()
macro in place of NOUVEAU_BO_VRAM to ensure correct behavior on
VRAM-less chips.

Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Martin Peres <martin.peres@free.fr>
9 years agoilo: add ilo_state_compute
Chia-I Wu [Sat, 20 Jun 2015 15:27:08 +0000 (23:27 +0800)]
ilo: add ilo_state_compute

Replace gen6_idrt_data with ilo_state_compute, which has a bunch of
validations and is now preferred.

9 years agor600g: ignore sampler views for now.
Dave Airlie [Mon, 22 Jun 2015 03:36:41 +0000 (13:36 +1000)]
r600g: ignore sampler views for now.

This fixes a regression in that r600 stopped working when
sampler views were pushed.

Signed-off-by: Dave Airlie <airlied@redhat.com>
9 years agofreedreno/ir3: pass sz to split_dest()
Rob Clark [Sat, 13 Jun 2015 13:14:31 +0000 (09:14 -0400)]
freedreno/ir3: pass sz to split_dest()

For query_levels, we generate a getinfo with writemask of (z), which RA
will consider as size==3.  But we were still generating four fanouts.
Which meant that RA would see it as two different register classes,
depending on the path to definer.  Ie. on the getinfo instruction itself
it would see size==3, but when chasing back through the fanouts it would
see size==4.

Easiest way to solve that is to just generate the chain of neighboring
fanouts to have the correct size in the first place.

Note: we may eventually want split_dest() to take start/end or wrmask
instead, since really we only need size==1.  But RA is not clever enough
for that, query_levels is not that common, and the other two registers
that get allocated are never used so those register slots can be
immediately re-used.  So bunch of work for probably no real gain.

Signed-off-by: Rob Clark <robclark@freedesktop.org>
9 years agofreedreno/ir3/nir: add more opcodes
Rob Clark [Fri, 12 Jun 2015 18:27:44 +0000 (14:27 -0400)]
freedreno/ir3/nir: add more opcodes

Signed-off-by: Rob Clark <robclark@freedesktop.org>
9 years agofreedreno/ir3: only unminify txf coords on a3xx
Rob Clark [Mon, 8 Jun 2015 18:45:47 +0000 (14:45 -0400)]
freedreno/ir3: only unminify txf coords on a3xx

Seems like a4xx gets this right.

Signed-off-by: Rob Clark <robclark@freedesktop.org>
9 years agofreedreno: remove int sampler shader variants
Rob Clark [Mon, 8 Jun 2015 18:23:49 +0000 (14:23 -0400)]
freedreno: remove int sampler shader variants

We get this information from NIR (which gets it from sview decl in tgsi
when translating from tgsi), so no need to maintain shader variants for
this.

Signed-off-by: Rob Clark <robclark@freedesktop.org>
9 years agofreedreno/ir3: block reshuffling and loops!
Rob Clark [Tue, 9 Jun 2015 21:17:06 +0000 (17:17 -0400)]
freedreno/ir3: block reshuffling and loops!

This shuffles things around to allow the shader to have multiple basic
blocks.  We drop the entire CFG structure from nir and just preserve the
blocks.  At scheduling we know whether to schedule conditional branches
or unconditional jumps at the end of the block based on the # of block
successors.  (Dropping jumps to the following instruction, etc.)

One slight complication is that variables (load_var/store_var, ie.
arrays) are not in SSA form, so we have to figure out where to put the
phi's ourself.  For this, we use the predecessor set information from
nir_block.  (We could perhaps use NIR's dominance frontier information
to help with this?)

Signed-off-by: Rob Clark <robclark@freedesktop.org>
9 years agofreedreno/ir3: a4xx encodes larger immed offset
Rob Clark [Mon, 1 Jun 2015 16:35:19 +0000 (12:35 -0400)]
freedreno/ir3: a4xx encodes larger immed offset

Without this, negative branch/jump offsets look like very large positive
offsets.

Signed-off-by: Rob Clark <robclark@freedesktop.org>
9 years agofreedreno/ir3: simplify find_neighbors stop condition
Rob Clark [Mon, 25 May 2015 14:59:21 +0000 (10:59 -0400)]
freedreno/ir3: simplify find_neighbors stop condition

Signed-off-by: Rob Clark <robclark@freedesktop.org>
9 years agofreedreno/ir3: move inputs/outputs to shader
Rob Clark [Mon, 25 May 2015 14:30:54 +0000 (10:30 -0400)]
freedreno/ir3: move inputs/outputs to shader

These belong in the shader, rather than the block.  Mostly a lot of
churn and nothing too interesting.  But splitting this out from the
rest of ir3_block reshuffling to cut down the noise in the later
patch.

Signed-off-by: Rob Clark <robclark@freedesktop.org>
9 years agofreedreno/ir3/ra: use register_allocate
Rob Clark [Fri, 1 May 2015 16:21:12 +0000 (12:21 -0400)]
freedreno/ir3/ra: use register_allocate

Signed-off-by: Rob Clark <robclark@freedesktop.org>
9 years agofreedreno/ir3: introduce ir3_compiler object
Rob Clark [Sat, 23 May 2015 17:37:41 +0000 (13:37 -0400)]
freedreno/ir3: introduce ir3_compiler object

Right now, just provides a cleaner way to get at the gpu-id, given the
separation between compiler and context.  But we will need this also to
hold the reg-set for new register allocation.

Signed-off-by: Rob Clark <robclark@freedesktop.org>
9 years agofreedreno/ir3: dump nocp option
Rob Clark [Sat, 25 Apr 2015 20:30:55 +0000 (16:30 -0400)]
freedreno/ir3: dump nocp option

No longer used, or even possible, with NIR frontend.

Signed-off-by: Rob Clark <robclark@freedesktop.org>
9 years agofreedreno/ir3: silence warnings
Rob Clark [Tue, 9 Jun 2015 21:42:16 +0000 (17:42 -0400)]
freedreno/ir3: silence warnings

Signed-off-by: Rob Clark <robclark@freedesktop.org>
9 years agofreedreno/ir3: remove tgsi f/e
Rob Clark [Sat, 25 Apr 2015 14:22:49 +0000 (10:22 -0400)]
freedreno/ir3: remove tgsi f/e

Also remove ir3_flatten which was only used by tgsi f/e.

Signed-off-by: Rob Clark <robclark@freedesktop.org>
9 years agofreedreno/ir3/sched: convert to priority queue
Rob Clark [Thu, 30 Apr 2015 17:57:15 +0000 (13:57 -0400)]
freedreno/ir3/sched: convert to priority queue

Use a more standard priority-queue based scheduling algo.  It is simpler
and will make things easier once we have multiple basic blocks and flow
control.

Signed-off-by: Rob Clark <robclark@freedesktop.org>
9 years agofreedreno/ir3: use standard list implementation
Rob Clark [Thu, 30 Apr 2015 15:38:43 +0000 (11:38 -0400)]
freedreno/ir3: use standard list implementation

Use standard list_head double-linked list and related iterators,
helpers, etc, rather than weird combo of instruction array and next
pointers depending on stage.  Now block has an instrs_list.  In
certain stages where we want to remove and re-add to the blocks list
we just use list_replace() to copy the list to a new list_head.

Signed-off-by: Rob Clark <robclark@freedesktop.org>
9 years agofreedreno/ir3: drop dot graph dumping
Rob Clark [Thu, 30 Apr 2015 14:10:14 +0000 (10:10 -0400)]
freedreno/ir3: drop dot graph dumping

At least for now.. right now the instruction and instruction list
printing should suffice, and the re-working of ir3_block would require
a lot of changes in that code.

Signed-off-by: Rob Clark <robclark@freedesktop.org>
9 years agofreedreno/ir3: more builder helpers
Rob Clark [Sat, 25 Apr 2015 15:05:27 +0000 (11:05 -0400)]
freedreno/ir3: more builder helpers

Use ir3_MOV() builder in a couple of spots, rather than open-coding the
instruction construction.  Also add ir3_NOP() builder and use that
instead of open coding.

Signed-off-by: Rob Clark <robclark@freedesktop.org>
9 years agogallium/ttn: add missing SNE
Rob Clark [Thu, 30 Apr 2015 19:20:03 +0000 (15:20 -0400)]
gallium/ttn: add missing SNE

Signed-off-by: Rob Clark <robclark@freedesktop.org>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
9 years agoutil/list: add list_first/last_entry
Rob Clark [Wed, 29 Apr 2015 12:38:45 +0000 (08:38 -0400)]
util/list: add list_first/last_entry

I need an easier way to get at head/tail in ir3.

Signed-off-by: Rob Clark <robclark@freedesktop.org>
9 years agogallium/ttn: add texture-type support
Rob Clark [Mon, 8 Jun 2015 18:09:09 +0000 (14:09 -0400)]
gallium/ttn: add texture-type support

v2: rebased on using SVIEW to hold type information

Signed-off-by: Rob Clark <robclark@freedesktop.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
9 years agoglsl_to_tgsi: add SVIEW decl support
Rob Clark [Mon, 8 Jun 2015 17:20:30 +0000 (13:20 -0400)]
glsl_to_tgsi: add SVIEW decl support

Freedreno needs sampler type information to deal with int/uint textures.
To accomplish this, start creating sampler-view declarations, as
suggested here:

 http://lists.freedesktop.org/archives/mesa-dev/2014-November/071583.html

create a sampler-view with index matching the sampler, to encode the
texture type (ie. SINT/UINT/FLOAT).  Ie:

   DCL SVIEW[n], 2D, UINT
   DCL SAMP[n]
   TEX OUT[1], IN[1], SAMP[n]

For tgsi texture instructions which do not take an explicit SVIEW
argument, the SVIEW index is implied by the SAMP index.

Signed-off-by: Rob Clark <robclark@freedesktop.org>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
9 years agoutil/blitter (and friends): generate appropriate SVIEW decls
Rob Clark [Thu, 11 Jun 2015 00:02:55 +0000 (20:02 -0400)]
util/blitter (and friends): generate appropriate SVIEW decls

Some hardware needs to know the sampler type.  Update the blit related
shaders to include SVIEW decl.

Signed-off-by: Rob Clark <robclark@freedesktop.org>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
9 years agoutil/pstipple: updates for SVIEW decls
Rob Clark [Thu, 11 Jun 2015 00:01:11 +0000 (20:01 -0400)]
util/pstipple: updates for SVIEW decls

To allow for shaders which use SVIEW decls for TEX* instructions, we
need to preserve the constraint that the shader either has no SVIEW's or
it has one matching SVIEW for each SAMP.

Signed-off-by: Rob Clark <robclark@freedesktop.org>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
9 years agodraw: updates to support SVIEW decls
Rob Clark [Wed, 10 Jun 2015 23:59:20 +0000 (19:59 -0400)]
draw: updates to support SVIEW decls

To allow for shaders which use SVIEW decls for TEX* instructions, we
need to preserve the constraint that the shader either has no SVIEW's or
it has one matching SVIEW for each SAMP.

Signed-off-by: Rob Clark <robclark@freedesktop.org>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
9 years agotgsi/transform: add support for SVIEW decls
Rob Clark [Wed, 10 Jun 2015 23:51:32 +0000 (19:51 -0400)]
tgsi/transform: add support for SVIEW decls

TODO single return_type (use enum)

v2: single return_type arg, and use enum

Signed-off-by: Rob Clark <robclark@freedesktop.org>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
9 years agotgsi: update docs for SVIEW usage with TEX* instructions
Rob Clark [Wed, 10 Jun 2015 23:49:55 +0000 (19:49 -0400)]
tgsi: update docs for SVIEW usage with TEX* instructions

Based on mailing list discussion here:

http://lists.freedesktop.org/archives/mesa-dev/2014-November/071583.html

Signed-off-by: Rob Clark <robclark@freedesktop.org>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
9 years agomesa: Back out an accidental change I had in a VC4 commit.
Eric Anholt [Sat, 20 Jun 2015 22:02:50 +0000 (15:02 -0700)]
mesa: Back out an accidental change I had in a VC4 commit.

This was a hack as part of debugging some glamor-on-GLES2 behavior that
ended up being an xserver bug.  I suspect we can just flip this extension
on for GLES2, but the spec says it requires 3.1.

9 years agodocs: add news item and link release notes for mesa 10.5.8
Emil Velikov [Sat, 20 Jun 2015 15:40:56 +0000 (16:40 +0100)]
docs: add news item and link release notes for mesa 10.5.8

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
9 years agodocs: Add sha256sums for the 10.5.8 release
Emil Velikov [Sat, 20 Jun 2015 15:37:16 +0000 (16:37 +0100)]
docs: Add sha256sums for the 10.5.8 release

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
(cherry picked from commit a81b1d5512f64ffca1c13a5937e7eb0de24713ae)

9 years agoAdd release notes for the 10.5.8 release
Emil Velikov [Sat, 20 Jun 2015 14:14:45 +0000 (15:14 +0100)]
Add release notes for the 10.5.8 release

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
(cherry picked from commit 24b043aab73ce066ded6e4bc93f589008dfc8484)

9 years agovc4: Use a defined t value for 1D textures.
Eric Anholt [Sat, 20 Jun 2015 02:47:44 +0000 (19:47 -0700)]
vc4: Use a defined t value for 1D textures.

This doesn't fix the broken 1D cases of texsubimage, but it does prevent
segfaulting when dumping the QIR code generated in fbo-1d.

9 years agovc4: Fix write-only texsubimage when we had to align.
Eric Anholt [Sat, 20 Jun 2015 02:41:25 +0000 (19:41 -0700)]
vc4: Fix write-only texsubimage when we had to align.

We need to make sure that when we store the aligned box, we've got
initialized contents in the border.  We could potentially just load the
border area, but for now let's get text rendering working in X (and fix
the GL_TEXTURE_2D errors in piglit's texsubimage test and
gl-2.1-pbo/test_tex_image)

9 years agoilo: clean up header includes
Chia-I Wu [Thu, 18 Jun 2015 14:48:14 +0000 (22:48 +0800)]
ilo: clean up header includes

Core is more self-contained now.

9 years agoilo: avoid ilo_ib_state in genX_3DPRIMITIVE()
Chia-I Wu [Fri, 19 Jun 2015 16:34:29 +0000 (00:34 +0800)]
ilo: avoid ilo_ib_state in genX_3DPRIMITIVE()

ilo_ib_state is not in core.

9 years agoilo: move gen6_so_SURFACE_STATE() out of core
Chia-I Wu [Thu, 18 Jun 2015 14:47:20 +0000 (22:47 +0800)]
ilo: move gen6_so_SURFACE_STATE() out of core

It does not belong to core.

9 years agoilo: add ilo_state_sol_buffer
Chia-I Wu [Mon, 15 Jun 2015 07:17:45 +0000 (15:17 +0800)]
ilo: add ilo_state_sol_buffer

It serves the same purpose as ilo_state_vertex_buffer does.

9 years agoilo: add ilo_state_index_buffer
Chia-I Wu [Fri, 19 Jun 2015 07:10:02 +0000 (15:10 +0800)]
ilo: add ilo_state_index_buffer

It serves the same purpose as ilo_state_vertex_buffer does.

9 years agoilo: add ilo_state_vertex_buffer
Chia-I Wu [Fri, 19 Jun 2015 07:06:50 +0000 (15:06 +0800)]
ilo: add ilo_state_vertex_buffer

Being a parameter-like state, we may want to get rid of
ilo_state_vertex_buffer_info or ilo_state_vertex_buffer eventually.  But we
want them now as they are how we do cross-validation right now.

9 years agoilo: add 3DSTATE_VF_INSTANCING to ilo_state_vf
Chia-I Wu [Thu, 18 Jun 2015 06:26:29 +0000 (14:26 +0800)]
ilo: add 3DSTATE_VF_INSTANCING to ilo_state_vf

3DSTATE_VF_INSTANCING specifies instancing enable and step rate.  They are
specified along with 3DSTATE_VERTEX_BUFFERS instead prior to Gen8.  Both
commands are added.

9 years agoilo: add 3DSTATE_VF to ilo_state_vf
Chia-I Wu [Tue, 16 Jun 2015 15:11:06 +0000 (23:11 +0800)]
ilo: add 3DSTATE_VF to ilo_state_vf

3DSTATE_VF specifies cut index enable and cut index.  Cut index enable is
specified in 3DSTATE_INDEX_BUFFER instead prior to Gen7.5.  Both commands are
added.

9 years agoilo: embed pipe_index_buffer in ilo_ib_state
Chia-I Wu [Thu, 18 Jun 2015 05:55:32 +0000 (13:55 +0800)]
ilo: embed pipe_index_buffer in ilo_ib_state

Make it obvious that we save a copy of pipe_index_buffer.

9 years agoilo: fix a buffer overrun
Chia-I Wu [Fri, 19 Jun 2015 15:29:32 +0000 (23:29 +0800)]
ilo: fix a buffer overrun

Add missing parentheses in SURFTYPE_NULL initialization.

9 years agoilo: fix a -Wmaybe-uninitialized warning
Chia-I Wu [Fri, 19 Jun 2015 15:24:17 +0000 (23:24 +0800)]
ilo: fix a -Wmaybe-uninitialized warning

ilo_shader.c: In function ‘ilo_shader_select_kernel_sbe’:
ilo_shader.c:1140:27: warning: ‘src_skip’ may be used uninitialized in this
function [-Wmaybe-uninitialized]

9 years agoglsl: fix formatting glitch in _mesa_print_ir()
Brian Paul [Fri, 19 Jun 2015 22:45:44 +0000 (16:45 -0600)]
glsl: fix formatting glitch in _mesa_print_ir()

Print the closing ) before the newline.  Trivial.

9 years agoi965/gen8: Use HALIGN_16 for single sample mcs buffers
Ben Widawsky [Fri, 19 Jun 2015 01:45:47 +0000 (18:45 -0700)]
i965/gen8: Use HALIGN_16 for single sample mcs buffers

The original code meant to do this, but was only checking num_samples == 1 to
figure out if a surface was fast clear capable. However, we can allocate single
sample miptrees with num_samples == 0 (when it's an internally created buffer).

This fixes a bunch of the piglit tests on gen8. Other gens should have been
fine.

Here is the order of events that allowed this to slip through:
t0: I wrote halign patches and tested them. These alignment assertions are for
   gen8 fast clear surfaces, basically.
t1: I pushed bogus perf patch which made fast clears never happen
t2: Reworked halign patches based on Chad's feedback and introduced the bug this
   patch fixes.
t2.5: I tested reworked patches, but assertion wasn't hit because of t1.
t3. Matt fixed issue in t1 which made fast clears happen here:
commit 22af95af8316f2888a3935cdf774ff0997b3dd42
Author: Matt Turner <mattst88@gmail.com>
Date:   Thu Jun 18 16:14:50 2015 -0700

    i965: Add missing braces around if-statement.

This logic should match that of the v1 of my halign patch series.

Cc: Kenneth Graunke <kenneth@whitecape.org>
Cc: Matt Turner <mattst88@gmail.com>
Reported-by: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Tested-by: Mark Janes <mark.a.janes@intel.com>
9 years agomesa: move ARB_gs5 enums to core, EXT_polygon_offset_clamp to desktop
Ilia Mirkin [Fri, 19 Jun 2015 16:08:24 +0000 (12:08 -0400)]
mesa: move ARB_gs5 enums to core, EXT_polygon_offset_clamp to desktop

When adding EXT_polygon_offset_clamp, I first made it core-only, and
never moved the enum getter back to the GL/GL_CORE section. Similarly,
ARB_gs5 is a core-only extension, so move its getters to the GL_CORE
section.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Matt Turner <mattst88@gmail.com>
9 years agou_vbuf: fix src_offset alignment in u_vbuf_create_vertex_elements()
Brian Paul [Fri, 19 Jun 2015 00:03:29 +0000 (18:03 -0600)]
u_vbuf: fix src_offset alignment in u_vbuf_create_vertex_elements()

If the driver says PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY=1,
the driver should never receive a pipe_vertex_element::src_offset value
that's not a multiple of four.  But the vbuf code wasn't actually adjusting
the src_offset value when creating the vertex element state object.

We just need to align the src_offset values put in the driver_attribs[]
array.

See the piglit gl-1.5-vertex-buffer-offsets test.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
9 years agogallium: whitespace, formatting clean-up in p_state.h
Brian Paul [Thu, 18 Jun 2015 23:53:42 +0000 (17:53 -0600)]
gallium: whitespace, formatting clean-up in p_state.h

Remove trailing whitespace, move some braces, 78-column wrapping.
Trivial.

9 years agost/wgl: fix WGL_SWAP_METHOD_ARB query
Brian Paul [Tue, 16 Jun 2015 21:32:46 +0000 (15:32 -0600)]
st/wgl: fix WGL_SWAP_METHOD_ARB query

There are three possible return values (not two): WGL_SWAP_COPY_ARB,
WGL_SWAP_EXCHANGE_EXT and WGL_SWAP_UNDEFINED_ARB.

VMware bug 1431184

Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
9 years agostw: use new stw_get_nop_function() function to avoid Viewperf 12 crashes
Brian Paul [Tue, 16 Jun 2015 21:32:46 +0000 (15:32 -0600)]
stw: use new stw_get_nop_function() function to avoid Viewperf 12 crashes

Also, print a warning if we do return NULL from wglGetProcAddress() to
help spot this sort of problem in the future.

Reviewed-by: José Fonseca <jfonseca@vmware.com>
9 years agostw: add some no-op functions for GL_EXT_dsa, GL_NV_half_float
Brian Paul [Tue, 16 Jun 2015 21:32:46 +0000 (15:32 -0600)]
stw: add some no-op functions for GL_EXT_dsa, GL_NV_half_float

Viewperf 12 calls wglGetProcAddress() to get pointers to some unsupported
DSA and half-float functions.  We return NULL but Viewperf doesn't check
for null before trying to jump through the pointer.  That causes a crash.

This patch adds no-op functions to call instead (used by the next patch).
This avoids the crash but the rendering is incorrect.

Some DSA functions are being added to Mesa at this time so we may be
able to remove some of these no-ops in the future.

More no-op functions may be added as needed.

VMware PR1383421

Reviewed-by: José Fonseca <jfonseca@vmware.com>
9 years agost/wgl: Don't return core profile for 3.1 contexts.
Jose Fonseca [Tue, 16 Jun 2015 21:32:46 +0000 (15:32 -0600)]
st/wgl: Don't return core profile for 3.1 contexts.

WGL_CONTEXT_PROFILE_MASK_ARB doesn't apply to desktop OpenGL versions
less than 3.2 -- applications can't specify whether they want a core or
a compat 3.1 context -- instead they are supposed the check whether the
returned context advertises GL_ARB_compatibility extension.

Mesa doesn't support compatability contexts for version higher than 3.1,
so we used to return core profile context, but this makes several Windows
applications unhappy, because they just assume they got a compatability
context without checking.

So it seems safer to on Windows to never return core profile for 3.1,
ie, just fail the context creation.

VMware PR1365920.

Reviewed-by: Brian Paul <brianp@vmware.com>
9 years agost/wgl: set PIPE_BIND_SAMPLER_VIEW for window color buffers
Brian Paul [Tue, 16 Jun 2015 21:32:46 +0000 (15:32 -0600)]
st/wgl: set PIPE_BIND_SAMPLER_VIEW for window color buffers

To allow sampling from the surface for things like glCopyPixels
or glCopyTexSubImage.

Reviewed-by: Charmaine Lee <charmainel@vmware.com>
9 years agost/wgl: add support for multisample pixel formats
Brian Paul [Tue, 16 Jun 2015 21:32:45 +0000 (15:32 -0600)]
st/wgl: add support for multisample pixel formats

Create pixel formats with 0, 4, 8 and 16 samples per pixel.
Add a SVGA_FORCE_MSAA env var to force creating all pixel formats
with a particular sample count.  This is useful for testing Mesa/GLUT/
etc. programs which don't ordinarily use multisample.

Reviewed-by: Matthew McClure <mcclurem@vmware.com>
9 years agost/wgl: respect sample count when creating framebuffer surfaces
Brian Paul [Tue, 16 Jun 2015 21:32:45 +0000 (15:32 -0600)]
st/wgl: respect sample count when creating framebuffer surfaces

Use the visual/pixel format's sample count instead of zero.

Reviewed-by: Matthew McClure <mcclurem@vmware.com>
9 years agost/wgl: fix WGL_SAMPLE_BUFFERS_ARB query
Brian Paul [Tue, 16 Jun 2015 21:32:45 +0000 (15:32 -0600)]
st/wgl: fix WGL_SAMPLE_BUFFERS_ARB query

Only report 1 for WGL_SAMPLE_BUFFERS_ARB if the number of samples
per pixel > 1.

Reviewed-by: Matthew McClure <mcclurem@vmware.com>
9 years agotgsi: add comments for ureg_emit_label()
Brian Paul [Sat, 13 Jun 2015 14:07:08 +0000 (08:07 -0600)]
tgsi: add comments for ureg_emit_label()

9 years agotgsi: new comments, assertion for executing TGSI_OPCODE_CAL
Brian Paul [Sat, 13 Jun 2015 13:58:53 +0000 (07:58 -0600)]
tgsi: new comments, assertion for executing TGSI_OPCODE_CAL

9 years agodocs: update developer info
Timothy Arceri [Fri, 19 Jun 2015 03:03:36 +0000 (13:03 +1000)]
docs: update developer info

Update piglit link to the current Piglit website.

Add note about updating patchwork when sending patch revisions.

Acked-by: Matt Turner <mattst88@gmail.com>
9 years agollvmpipe: Truncate the binned constants to max const buffer size.
Jose Fonseca [Thu, 18 Jun 2015 14:47:00 +0000 (15:47 +0100)]
llvmpipe: Truncate the binned constants to max const buffer size.

Tested with Ilia Mirkin's gzdoom.trace and
"arb_uniform_buffer_object-maxuniformblocksize fsexceed" piglit test
without my earlier fix to fail linkage when UBO exceeds
GL_MAX_UNIFORM_BLOCK_SIZE.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
9 years agoglsl: Fail linkage when UBO exceeds GL_MAX_UNIFORM_BLOCK_SIZE.
Jose Fonseca [Mon, 15 Jun 2015 17:29:02 +0000 (18:29 +0100)]
glsl: Fail linkage when UBO exceeds GL_MAX_UNIFORM_BLOCK_SIZE.

It's not totally clear whether other Mesa drivers can safely cope with
over-sized UBOs, but at least for llvmpipe receiving a UBO larger than
its limit causes problems, as it won't fit into its internal display
lists.

This fixes piglit "arb_uniform_buffer_object-maxuniformblocksize
fsexceed" without regressions for llvmpipe.

NVIDIA driver also fails to link the shader from
"arb_uniform_buffer_object-maxuniformblocksize fsexceed".

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=65525

PS: I don't recommend cherry-picking this for Mesa stable, as some app
might inadvertently been relying on UBOs larger than
GL_MAX_UNIFORM_BLOCK_SIZE to work on other drivers, so even if this
commit is universally accepted it's probably best to let it mature in
master for a while.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
9 years agoglsl: guard gl_NumSamples enablement on ARB_sample_shading
Ilia Mirkin [Thu, 18 Jun 2015 23:08:24 +0000 (19:08 -0400)]
glsl: guard gl_NumSamples enablement on ARB_sample_shading

gl_NumSamples should only be enabled when ARB_sample_shading is enabled.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
9 years agoi965: Add missing braces around if-statement.
Matt Turner [Thu, 18 Jun 2015 23:14:50 +0000 (16:14 -0700)]
i965: Add missing braces around if-statement.

Fixes a performance problem caused by commit b639ed2f.

Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=90895

9 years agoi965/compute: Fix undefined code with right_mask for SIMD32
Jordan Justen [Tue, 16 Jun 2015 21:27:15 +0000 (14:27 -0700)]
i965/compute: Fix undefined code with right_mask for SIMD32

Although we don't support SIMD32, krh pointed out that the left shift
by 32 is undefined by C/C++ for 32-bit integers.

Suggested-by: Kristian Høgsberg <krh@bitplanet.net>
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
9 years agomesa: add GL_PROGRAM_PIPELINE support in KHR_debug calls
Ilia Mirkin [Thu, 18 Jun 2015 03:00:44 +0000 (23:00 -0400)]
mesa: add GL_PROGRAM_PIPELINE support in KHR_debug calls

This was apparently missed when ARB_sso support was added.
Add label support to pipeline objects just like all the other
debug-related objects.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Timothy Arceri <t_arceri@yahoo.com.au>
Cc: "10.5 10.6" <mesa-stable@lists.freedesktop.org>
9 years agoglsl: add version checks to conditionals for builtin variable enablement
Ilia Mirkin [Wed, 17 Jun 2015 19:09:26 +0000 (15:09 -0400)]
glsl: add version checks to conditionals for builtin variable enablement

A number of builtin variables have checks based on the extension being
enabled, but were missing enablement via a higher GLSL version.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Timothy Arceri <t_arceri@yahoo.com.au>
Cc: "10.5 10.6" <mesa-stable@lists.freedesktop.org>
9 years agoglsl: handle conversions to double when comparing param matches
Ilia Mirkin [Wed, 17 Jun 2015 19:07:14 +0000 (15:07 -0400)]
glsl: handle conversions to double when comparing param matches

This allows mod(int, int) to become selected as float mod when doubles
are supported.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
Cc: "10.6" <mesa-stable@lists.freedesktop.org>
9 years agoilo: remove missing ilo_fence.h from the sources list
Emil Velikov [Thu, 18 Jun 2015 11:59:28 +0000 (12:59 +0100)]
ilo: remove missing ilo_fence.h from the sources list

Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
9 years agoegl/x11: Set version of swrastLoader to 2
Boyan Ding [Tue, 16 Jun 2015 03:08:33 +0000 (11:08 +0800)]
egl/x11: Set version of swrastLoader to 2

which it actually implements instead of the newest version defined in
dri_interface.h

Cc: "10.5 10.6" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Boyan Ding <boyan.j.ding@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
9 years agovc4: Move tile state/alloc allocation into the kernel.
Eric Anholt [Wed, 17 Jun 2015 20:24:06 +0000 (13:24 -0700)]
vc4: Move tile state/alloc allocation into the kernel.

This avoids a security issue where userspace could have written the tile
state/tile alloc behind the GPU's back, and will apparently be necessary
for fixing stability bugs (tile state buffers are missing some top bits
for the tile alloc's address).

9 years agovc4: Move RCL generation into the kernel.
Eric Anholt [Wed, 10 Jun 2015 19:36:47 +0000 (12:36 -0700)]
vc4: Move RCL generation into the kernel.

There weren't that many variations of RCL generation, and this lets us
skip all the in-kernel validation for what we generated.

9 years agovc4: Add dumping of VC4_PACKET_TILE_BINNING_MODE_CONFIG.
Eric Anholt [Wed, 17 Jun 2015 20:51:55 +0000 (13:51 -0700)]
vc4: Add dumping of VC4_PACKET_TILE_BINNING_MODE_CONFIG.

9 years agovc4: Fix memory leak from simple_list conversion.
Eric Anholt [Thu, 18 Jun 2015 06:49:19 +0000 (23:49 -0700)]
vc4: Fix memory leak from simple_list conversion.

I accidentally shadowed the outside declaration, so we always returned
NULL even when we'd found something in the cache.

9 years agovc4: Track the number of BOs allocated and their size.
Eric Anholt [Thu, 18 Jun 2015 05:56:15 +0000 (22:56 -0700)]
vc4: Track the number of BOs allocated and their size.

This is useful for BO leak debugging.

9 years agoi965: Fix textureGrad with cube samplers
Iago Toral Quiroga [Tue, 24 Feb 2015 18:02:50 +0000 (19:02 +0100)]
i965: Fix textureGrad with cube samplers

We can't use sampler messages with gradient information (like
sample_g or sample_d) to deal with this scenario because according
to the PRM:

"The r coordinate and its gradients are required only for surface
types that use the third coordinate. Usage of this message type on
cube surfaces assumes that the u, v, and gradients have already been
transformed onto the appropriate face, but still in [-1,+1] range.
The r coordinate contains the faceid, and the r gradients are ignored
by hardware."

Instead, we should lower this to compute the LOD manually based on the
gradients and use a different sample message that takes the computed
LOD instead of the gradients. This is already being done in
brw_lower_texture_gradients.cpp, but it is restricted to shadow
samplers only, although there is a comment stating that we should
probably do this also for samplerCube and samplerCubeArray.

Because of this, both dEQP and Piglit test cases for textureGrad with
cube maps currently fail.

This patch does two things:
1) Activates the texturegrad lowering pass for all cube samplers.
2) Corrects the computation of the LOD value for cube samplers.

I had to do 2) because for cube maps the calculations implemented
in the lowering pass always compute a value of rho that is twice
the value we want (so we get a LOD value one unit larger than we
want). This only happens for cube map samplers (all kinds). I am
not sure about why we need to do this, but I suspect that it is
related to the fact that cube map coordinates, when transported
to a specific face in the cube, are in the range [-1, 1] instead of
[0, 1] so we probably need to divide the derivatives by 2 when
we compute the LOD. Doing that would produce the same result as
dividing the final rho computation by 2 (or removing a unit
from the computed LOD, which is what we are doing here).

Fixes the following piglit tests:
bin/tex-miplevel-selection textureGrad Cube -auto -fbo
bin/tex-miplevel-selection textureGrad CubeArray -auto -fbo
bin/tex-miplevel-selection textureGrad CubeShadow -auto -fbo

Fixes 10 dEQP tests in the following category:
dEQP-GLES3.functional.shaders.texture_functions.texturegrad.*cube*

Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
9 years agonvc0/ir: can't have a join on a load with an indirect source
Ilia Mirkin [Thu, 18 Jun 2015 02:18:09 +0000 (22:18 -0400)]
nvc0/ir: can't have a join on a load with an indirect source

Triggers an INVALID_OPCODE warning on GK208. Seems rare enough to not
warrant verification on other chips. Fixes the new piglits:

  ubo_array_indexing/fs-nonuniform-control-flow.shader_test
  ubo_array_indexing/vs-nonuniform-control-flow.shader_test

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "10.5 10.6" <mesa-stable@lists.freedesktop.org>
9 years agodocs: mark GL_ARB_framebuffer_no_attachments done for i965
Kevin Rogovin [Wed, 17 Jun 2015 10:29:59 +0000 (13:29 +0300)]
docs: mark GL_ARB_framebuffer_no_attachments done for i965

Mark GL_ARB_framebuffer_no_attachments as done for i965.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Signed-off-by: Kevin Rogovin <kevin.rogovin@intel.com>
9 years agoi965: enable ARB_framebuffer_no_attachments for Gen7+
Kevin Rogovin [Wed, 17 Jun 2015 10:29:58 +0000 (13:29 +0300)]
i965: enable ARB_framebuffer_no_attachments for Gen7+

Enable GL_ARB_framebuffer_no_attachments in i965 for Gen7 and higher.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Signed-off-by: Kevin Rogovin <kevin.rogovin@intel.com>
9 years agoi965: execution of frag-shader when it has atomic buffer
Kevin Rogovin [Wed, 17 Jun 2015 10:29:57 +0000 (13:29 +0300)]
i965: execution of frag-shader when it has atomic buffer

Ensure that the GPU spawns the fragment shader thread for those
fragment shaders with atomic buffer access.

Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Signed-off-by: Kevin Rogovin <kevin.rogovin@intel.com>
9 years agomesa: function for testing if current frag-shader has atomics
Kevin Rogovin [Wed, 17 Jun 2015 10:29:56 +0000 (13:29 +0300)]
mesa: function for testing if current frag-shader has atomics

Add helper function that checks if current fragment shader active
of gl_context has atomic buffer access.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Signed-off-by: Kevin Rogovin <kevin.rogovin@intel.com>
9 years agoi965: Use _mesa_geometric_ functions appropriately
Kevin Rogovin [Wed, 17 Jun 2015 10:29:55 +0000 (13:29 +0300)]
i965: Use _mesa_geometric_ functions appropriately

Change references to gl_framebuffer::Width, Height, MaxNumLayers
and Visual::samples to use the _mesa_geometry_ convenience functions
for those places where the geometry of the gl_framebuffer is needed
(in contrast to the geometry of the intersection of the attachments
of the gl_framebuffer).

This patch is to pave the way to enable GL_ARB_framebuffer_no_attachments
on Gen7 and higher in i965.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Signed-off-by: Kevin Rogovin <kevin.rogovin@intel.com>
9 years agomesa: helper function for scissor box of gl_framebuffer
Kevin Rogovin [Wed, 17 Jun 2015 10:29:54 +0000 (13:29 +0300)]
mesa: helper function for scissor box of gl_framebuffer

Add helper convenience function that intersects the scissor values
against a passed bounding box. In addition, to avoid replicated code,
make the function _mesa_scissor_bounding_box() use this new function.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Signed-off-by: Kevin Rogovin <kevin.rogovin@intel.com>
9 years agomesa: add helper functions for geometry of gl_framebuffer
Kevin Rogovin [Wed, 17 Jun 2015 10:29:53 +0000 (13:29 +0300)]
mesa: add helper functions for geometry of gl_framebuffer

Add convenience helper functions for fetching geometry of gl_framebuffer
that return the geometry of the gl_framebuffer instead of the geometry of
the buffers of the gl_framebuffer when then the gl_framebuffer has no
attachments.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Signed-off-by: Kevin Rogovin <kevin.rogovin@intel.com>
9 years agoPATCH 03/10] mesa: Complete ARB_framebuffer_no_attachments in Mesa core
Kevin Rogovin [Wed, 17 Jun 2015 10:29:52 +0000 (13:29 +0300)]
PATCH 03/10] mesa: Complete ARB_framebuffer_no_attachments in Mesa core

Implement GL_ARB_framebuffer_no_attachments in Mesa core
 - changes to conditions for framebuffer completenss
 - implement set/get functions for framebuffers for
   new functions in GL_ARB_framebuffer_no_attachments

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Signed-off-by: Kevin Rogovin <kevin.rogovin@intel.com>
9 years agomesa: Constants and functions for ARB_framebuffer_no_attachments
Kevin Rogovin [Wed, 17 Jun 2015 10:29:51 +0000 (13:29 +0300)]
mesa: Constants and functions for ARB_framebuffer_no_attachments

Define the enumeration constants, function entry points and
glGet for the GL_ARB_framebuffer_no_attachments.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Signed-off-by: Kevin Rogovin <kevin.rogovin@intel.com>
9 years agomesa: Define infrastructure for ARB_framebuffer_no_attachments
Kevin Rogovin [Wed, 17 Jun 2015 10:29:50 +0000 (13:29 +0300)]
mesa: Define infrastructure for ARB_framebuffer_no_attachments

Define the infrastructure for the extension GL_ARB_framebuffer_no_attachments:
 - extension table
 - additions to gl_framebuffer

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Signed-off-by: Kevin Rogovin <kevin.rogovin@intel.com>
9 years agovc4: Make sure that direct texture clamps have a minimum value of 0.
Eric Anholt [Mon, 15 Jun 2015 22:05:36 +0000 (15:05 -0700)]
vc4: Make sure that direct texture clamps have a minimum value of 0.

I was thinking of the MIN opcode in terms of unsigned math, but it's
signed, so if you used a negative array index, you could read before the
UBO.  Fixes segfaults under simulation in piglit array indexing tests with
mprotect-based guard pages.

9 years agovc4: Swap around which src we spill to ra31/rb31.
Eric Anholt [Tue, 16 Jun 2015 00:47:12 +0000 (17:47 -0700)]
vc4: Swap around which src we spill to ra31/rb31.

I wanted to assert that src1 came from a non-unspilled register in shader
validation, and this easily gets us that.  And, as a bonus:

total instructions in shared programs: 93347 -> 92723 (-0.67%)
instructions in affected programs:     60524 -> 59900 (-1.03%)

9 years agovc4: R4 is not a valid register for clamped direct texturing.
Eric Anholt [Tue, 16 Jun 2015 19:03:10 +0000 (12:03 -0700)]
vc4: R4 is not a valid register for clamped direct texturing.

Our array only goes to R3, and R4 is a special case that shouldn't be
used.

9 years agovc4: Factor out the live clamp register getter.
Eric Anholt [Mon, 15 Jun 2015 21:54:26 +0000 (14:54 -0700)]
vc4: Factor out the live clamp register getter.

9 years agovc4: Drop the unused "stride" field of surfaces.
Eric Anholt [Mon, 15 Jun 2015 18:41:06 +0000 (11:41 -0700)]
vc4: Drop the unused "stride" field of surfaces.

We're always looking at the slice anyway, when we would have needed it.

9 years agovc4: Handle refcounting the exec BO like we do in the kernel.
Eric Anholt [Fri, 12 Jun 2015 19:47:47 +0000 (12:47 -0700)]
vc4: Handle refcounting the exec BO like we do in the kernel.

This reduces the diff to the kernel, and will be useful when I make the
kernel allocate more BOs as part of validation.