mesa.git
7 years agost/mesa: fix reference counting bug in st_vdpau
Christian König [Wed, 13 Jul 2016 12:54:31 +0000 (14:54 +0200)]
st/mesa: fix reference counting bug in st_vdpau

Otherwise we leak the resources created for the DMA-buf descriptors.

Signed-off-by: Christian König <christian.koenig@amd.com>
Cc: 12.0 <mesa-stable@lists.freedesktop.org>
Tested-and-Reviewed by: Leo Liu <leo.liu@amd.com>
Ack-by: Tom St Denis <tom.stdenis@amd.com>
7 years agovc4: Emit resets of the uniform stream at the starts of blocks.
Eric Anholt [Fri, 6 May 2016 01:11:04 +0000 (18:11 -0700)]
vc4: Emit resets of the uniform stream at the starts of blocks.

If a block might be entered from multiple locations, then the uniform
stream will (probably) be at different points, and we need to make sure
that it's pointing where we expect it to be.  The kernel also enforces
that any block reading a uniform resets uniforms, to prevent reading
outside of the uniform stream by using looping.

7 years agovc4: Add support for scheduling of branch instructions.
Eric Anholt [Wed, 27 Apr 2016 19:14:07 +0000 (12:14 -0700)]
vc4: Add support for scheduling of branch instructions.

For now we don't fill the delay slots, and instead just drop in NOPs.

7 years agovc4: Move the QPU instructions to schedule into each block.
Eric Anholt [Wed, 16 Mar 2016 00:53:36 +0000 (17:53 -0700)]
vc4: Move the QPU instructions to schedule into each block.

We'll want to schedule them individually, to handle delay slots.

7 years agovc4: Disable vc4_opt_vpm in the presence of control flow.
Eric Anholt [Sat, 9 Jul 2016 00:06:18 +0000 (17:06 -0700)]
vc4: Disable vc4_opt_vpm in the presence of control flow.

It's a really valuable pass currently, but it will be a mess to rewrite
for control flow.  For now, just disable it if we have multiple blocks
present.

7 years agovc4: Convert vc4_opt_dead_code to work in the presence of control flow.
Eric Anholt [Fri, 8 Jul 2016 23:23:38 +0000 (16:23 -0700)]
vc4: Convert vc4_opt_dead_code to work in the presence of control flow.

With control flow, we can't be sure that we'll see the uses of a variable
before its def as we walk backwards.  Given that NIR is eliminating our
long chains of dead code, a simple solution for now seems fine.

This slightly changes the order of some optimizations, and so an opt_vpm
happens before opt_dce, causing 3 dead MOVs to be turned into dead FMAXes
in Minecraft:

instructions in affected programs:     52 -> 54 (3.85%)

7 years agovc4: Update copy propagation for control flow.
Eric Anholt [Wed, 13 Jul 2016 20:37:56 +0000 (13:37 -0700)]
vc4: Update copy propagation for control flow.

Previously, we could assume that a MOV from a temp was always an available
copy, because all temps were SSA in NIR, and their non-SSA state in QIR
was just due to the fact that they were from a bcsel or pack_unorm_4x8, so
we could use the current value of the temp after that series of QIR
instructions to define it.

However, this is no longer the case with control flow.  Instead, we track
a new array of MOVs defined within the block that haven't had their source
or dest killed yet, and use that primarily.  We fall back to looking
through the QIR defs array to handle across-block MOVs, but now require
that copies from the SSA defs have an SSA src as well.

7 years agoi965/fs: emit DIM instruction to load 64-bit immediates in HSW
Samuel Iglesias Gonsálvez [Thu, 7 Jul 2016 07:19:43 +0000 (09:19 +0200)]
i965/fs: emit DIM instruction to load 64-bit immediates in HSW

v2 (Matt):
- Use brw_imm_df() as source argument of DIM instruction.

Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
7 years agoi965/eu: set DF imm value to the source of DIM
Samuel Iglesias Gonsálvez [Thu, 7 Jul 2016 11:55:32 +0000 (13:55 +0200)]
i965/eu: set DF imm value to the source of DIM

According to HSW's PRM, vol02b, the DIM instruction has the following
restriction:

"Restriction : src0 must be immediate. src0 must specify the :f (F, Float)
type encoding but is an immediate 64-bit DF (Double Float) value. dst
must have type DF."

This commit allows to upload the immediate 64-bit DF value to the source
of a DIM instruction even when it is of float type encoding.

Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
7 years agoi965: enable the emission of the DIM instruction
Samuel Iglesias Gonsálvez [Thu, 7 Jul 2016 06:38:22 +0000 (08:38 +0200)]
i965: enable the emission of the DIM instruction

v2 (Matt):
- Take a DF source argument for the DIM instruction emission
in the visitors.
- Indentation.

Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
7 years agoanv: Add a stub for CmdCopyQueryPoolResults on Ivy Bridge
Jason Ekstrand [Thu, 14 Jul 2016 03:11:30 +0000 (20:11 -0700)]
anv: Add a stub for CmdCopyQueryPoolResults on Ivy Bridge

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Cc: "12.0" <mesa-stable@lists.freedesktop.org>
7 years agoi965: fix compiler warnings for 32bit build
Timothy Arceri [Thu, 14 Jul 2016 00:27:06 +0000 (10:27 +1000)]
i965: fix compiler warnings for 32bit build

Reviewed-by: Matt Turner <mattst88@gmail.com>
7 years agoRevert "gallium: Force blend color to 16-byte alignment"
Tim Rowley [Wed, 13 Jul 2016 15:30:46 +0000 (10:30 -0500)]
Revert "gallium: Force blend color to 16-byte alignment"

This reverts commit d8d6091a846ac2a40a011d512d6d57f6c8442e6a.

Heap allocations may be only 8-byte aligned on 32-bit system, and so having
members with 16-byte alignment (such as in the case where pipe_blend_color is
embedded in radeonsi's si_context) is undefined behavior which indeed causes
crashes when compiled with gcc -O3.

Cc: <mesa-stable@lists.freedesktop.org>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=96835
Signed-off-by: Tim Rowley <timothy.o.rowley@intel.com>
Acked-by: Chuck Atkins <chuck.atkins@kitware.com>
7 years agoisl/state: Add support for handling auxiliary surfaces
Jason Ekstrand [Fri, 3 Jun 2016 23:37:19 +0000 (16:37 -0700)]
isl/state: Add support for handling auxiliary surfaces

Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Chad Versace <chad.versace@intel.com>
7 years agoisl: Add an auxiliary surface usage enum
Jason Ekstrand [Tue, 21 Jun 2016 23:16:59 +0000 (16:16 -0700)]
isl: Add an auxiliary surface usage enum

Reviewed-by: Chad Versace <chad.versace@intel.com>
7 years agoisl: Add support for color control surfaces
Jason Ekstrand [Sat, 9 Jul 2016 05:12:10 +0000 (22:12 -0700)]
isl: Add support for color control surfaces

Reviewed-by: Chad Versace <chad.versace@intel.com>
7 years agoisl: Add support for multisample compression surfaces
Jason Ekstrand [Sat, 9 Jul 2016 02:37:27 +0000 (19:37 -0700)]
isl: Add support for multisample compression surfaces

Reviewed-by: Chad Versace <chad.versace@intel.com>
7 years agoisl: Add support for HiZ surfaces
Jason Ekstrand [Sat, 9 Jul 2016 02:36:33 +0000 (19:36 -0700)]
isl: Add support for HiZ surfaces

Reviewed-by: Chad Versace <chad.versace@intel.com>
7 years agoisl: Kill off isl_format_layout::bs
Jason Ekstrand [Sat, 9 Jul 2016 05:11:06 +0000 (22:11 -0700)]
isl: Kill off isl_format_layout::bs

Reviewed-by: Chad Versace <chad.versace@intel.com>
7 years agoisl: Take bpb rather than bs in tiling_get_info
Jason Ekstrand [Sat, 9 Jul 2016 05:04:18 +0000 (22:04 -0700)]
isl: Take bpb rather than bs in tiling_get_info

Reviewed-by: Chad Versace <chad.versace@intel.com>
7 years agoisl: Use bpb in a few places where it's more natural than bs
Jason Ekstrand [Sat, 9 Jul 2016 05:10:11 +0000 (22:10 -0700)]
isl: Use bpb in a few places where it's more natural than bs

Reviewed-by: Chad Versace <chad.versace@intel.com>
7 years agoisl: Use bpb for determining YUV image padding
Jason Ekstrand [Tue, 12 Jul 2016 21:17:18 +0000 (14:17 -0700)]
isl: Use bpb for determining YUV image padding

When we initially dropped bpb in favor of bs, we accidentally didn't change
this one line properly.  This brings it back to what it should be.

Reviewed-by: Chad Versace <chad.versace@intel.com>
7 years agoisl: Bring back isl_format_layout::bpb
Jason Ekstrand [Sat, 9 Jul 2016 04:55:34 +0000 (21:55 -0700)]
isl: Bring back isl_format_layout::bpb

A while ago we got rid of the bits-per-block because we thought we didn't
need it.  We're about to introduce some very useful 1 and 2-bit formats so
we really should be able to handle them again.

Reviewed-by: Chad Versace <chad.versace@intel.com>
7 years agoisl: Change the physical size of a W-tile to 128x32
Jason Ekstrand [Sat, 9 Jul 2016 00:24:19 +0000 (17:24 -0700)]
isl: Change the physical size of a W-tile to 128x32

Reviewed-by: Chad Versace <chad.versace@intel.com>
7 years agoisl: Rework the way we define tile sizes.
Jason Ekstrand [Sat, 9 Jul 2016 00:10:59 +0000 (17:10 -0700)]
isl: Rework the way we define tile sizes.

This is based on a very long set of discussions between Chad and myself
about how we should properly represent HiZ and CCS buffers.  The end result
of that discussion was that a tiling actually has two different sizes, a
logical size in elements, and a physical size in bytes and rows.  This
commit reworks ISL's pitch and size calculations to work in terms of these
two sizes.

Reviewed-by: Chad Versace <chad.versace@intel.com>
7 years agoisl: Rework the way we handle surface padding
Jason Ekstrand [Fri, 8 Jul 2016 22:08:20 +0000 (15:08 -0700)]
isl: Rework the way we handle surface padding

Reviewed-by: Chad Versace <chad.versace@intel.com>
7 years agoisl: Use ARRAY_PITCH_SPAN_FULL for depth/stencil surfaces on gen7
Jason Ekstrand [Tue, 12 Jul 2016 21:57:30 +0000 (14:57 -0700)]
isl: Use ARRAY_PITCH_SPAN_FULL for depth/stencil surfaces on gen7

We helpfully inserted a PRM quotation about how we need to use
ARRAY_PITCH_SPAN_FULL and then set it to COMPACT.  Oops...

Reviewed-by: Chad Versace <chad.versace@intel.com>
7 years agoisl: Stop multiplying height by block size
Jason Ekstrand [Fri, 8 Jul 2016 21:57:19 +0000 (14:57 -0700)]
isl: Stop multiplying height by block size

The row pitch already specifies the size of a row of elements.
Multiplying by the block height simply causes us to allocate as muc as 12
times more memory than needed for compressed textures.

Reviewed-by: Chad Versace <chad.versace@intel.com>
7 years agoisl: Get rid of tiling_get_extent
Jason Ekstrand [Fri, 8 Jul 2016 20:59:05 +0000 (13:59 -0700)]
isl: Get rid of tiling_get_extent

It was unused

Reviewed-by: Chad Versace <chad.versace@intel.com>
7 years agonir/spirv: Don't multiply the push constant block size by 4
Jason Ekstrand [Wed, 13 Jul 2016 18:35:29 +0000 (11:35 -0700)]
nir/spirv: Don't multiply the push constant block size by 4

I have no idea why we were multiplying by 4 before.  The offsets we get
from SPIR-V are in bytes and so is nir->num_uniforms so there's no need to
do any adjustment whatsoever.

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Cc: "12.0" <mesa-stable@lists.freedesktop.org>
7 years agoanv/pipeline: Assert that the number of uniforms from NIR fits
Jason Ekstrand [Wed, 13 Jul 2016 18:35:24 +0000 (11:35 -0700)]
anv/pipeline: Assert that the number of uniforms from NIR fits

7 years agoradeonsi: report accurate SGPR and VGPR spills
Marek Olšák [Tue, 5 Jul 2016 00:54:43 +0000 (02:54 +0200)]
radeonsi: report accurate SGPR and VGPR spills

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agoradeonsi: add a workaround for a compute VGPR-usage LLVM bug
Marek Olšák [Sun, 3 Jul 2016 18:00:17 +0000 (20:00 +0200)]
radeonsi: add a workaround for a compute VGPR-usage LLVM bug

v2: use abort(), describe which LLVM version is affected

Cc: 12.0 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agoradeonsi: use LLVMGetTypeKind to tell if an input is an array of descriptors
Marek Olšák [Tue, 12 Jul 2016 17:58:28 +0000 (19:58 +0200)]
radeonsi: use LLVMGetTypeKind to tell if an input is an array of descriptors

just a cleanup

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agoradeonsi: replace !tbaa with !invariant.load
Marek Olšák [Fri, 1 Jul 2016 14:25:01 +0000 (16:25 +0200)]
radeonsi: replace !tbaa with !invariant.load

no change in generated code thanks to dereferenceable(n)

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agoradeonsi: set dereferenceable attribute on descriptor arrays
Marek Olšák [Tue, 12 Jul 2016 16:52:55 +0000 (18:52 +0200)]
radeonsi: set dereferenceable attribute on descriptor arrays

This allows moving the loads arbitrarily in the Sinking pass.

26002 shaders in 14643 tests
Totals:
SGPRS: 2080160 -> 2080160 (0.00 %)
VGPRS: 798875 -> 797826 (-0.13 %)
Spilled SGPRs: 108485 -> 79165 (-27.03 %)
Spilled VGPRs: 327 -> 327 (0.00 %)
Scratch VGPRs: 1656 -> 1652 (-0.24 %) dwords per thread
Code Size: 36127192 -> 35559780 (-1.57 %) bytes
LDS: 767 -> 767 (0.00 %) blocks
Max Waves: 212464 -> 212672 (0.10 %)
Wait states: 0 -> 0 (0.00 %)

 PERCENTAGES / App    Shaders    SGPRs     VGPRs  SpillSGPR SpillVGPR  Scratch   CodeSize  MaxWaves    Waits
 (unknown)                  4     .         .         .         .         .         .         .         .
 0ad                        6     .         .         .         .         .         .         .         .
 alien_isolation         2938     .        0.04 %   -8.53 %     .         .       -0.71 %   -0.06 %     .
 anholt                    10     .         .         .         .         .         .         .         .
 batman_arkham_origins    589     .       -0.58 %  -79.54 %     .         .       -6.72 %    0.57 %     .
 bioshock-infinite       1769     .       -0.65 %  -89.32 %     .         .       -4.73 %    0.48 %     .
 borderlands2            3968     .       -0.31 %  -51.21 %     .         .       -4.09 %    0.22 %     .
 brutal-legend            338     .       -0.03 %   -2.95 %     .         .       -0.06 %     .         .
 civilization_beyond..    116     .         .      -14.17 %     .         .       -0.88 %     .         .
 counter_strike_glob..   1142     .         .         .         .         .         .         .         .
 dirt-showdown            541     .       -0.56 %  -40.14 %     .       -3.45 %   -1.82 %    0.35 %     .
 dolphin                   22     .         .         .         .         .        0.16 %     .         .
 dota2                   1747     .         .         .         .         .        0.01 %     .         .
 europa_universalis_4      76     .       -0.23 %  -42.11 %     .         .       -0.96 %     .         .
 f1-2015                  774     .       -0.09 %  -28.89 %     .         .       -2.60 %    0.09 %     .
 furmark-0.7.0              4     .         .         .         .         .         .         .         .
 gimark-0.7.0              10     .         .         .         .         .         .         .         .
 glamor                    16     .         .         .         .         .         .         .         .
 humus-celshading           4     .         .         .         .         .         .         .         .
 humus-domino               6     .         .         .         .         .         .         .         .
 humus-dynamicbranching    24     .        0.71 %     .         .         .        0.29 %   -0.45 %     .
 humus-hdr                 10     .         .         .         .         .         .         .         .
 humus-portals              2     .         .         .         .         .         .         .         .
 humus-volumetricfog..      6     .         .         .         .         .         .         .         .
 left_4_dead_2           1762     .         .         .         .         .         .         .         .
 metro_2033_redux        2670     .       -0.10 %   -7.15 %     .         .       -0.03 %     .         .
 nexuiz                    80     .         .         .         .         .         .         .         .
 pixmark-julia-fp32         2     .         .         .         .         .         .         .         .
 pixmark-julia-fp64         2     .         .         .         .         .         .         .         .
 pixmark-piano-0.7.0        2     .         .         .         .         .         .         .         .
 pixmark-volplosion-..      2     .         .         .         .         .         .         .         .
 plot3d-0.7.0               8     .         .         .         .         .         .         .         .
 portal                   474     .         .         .         .         .         .         .         .
 sauerbraten                7     .         .         .         .         .         .         .         .
 serious_sam_3_bfe        392     .         .      -13.20 %     .         .       -1.81 %     .         .
 supertuxkart               4     .         .         .         .         .         .         .         .
 talos_principle          324     .       -0.21 %  -18.39 %     .         .       -2.73 %    0.14 %     .
 team_fortress_2          808     .         .         .         .         .         .         .         .
 tesseract                430     .        0.08 %  -68.57 %     .         .       -0.45 %     .         .
 tessmark-0.7.0             6     .         .         .         .         .         .         .         .
 thea                     172     .         .         .         .         .        0.03 %     .         .
 ue4_effects_cave         299     .       -0.04 %  -10.15 %     .         .       -0.25 %    0.04 %     .
 ue4_elemental            586     .       -0.02 %  -13.93 %     .         .       -0.13 %    0.02 %     .
 ue4_lightroom_inter..     74     .       -0.17 %  -70.00 %     .         .       -1.27 %     .         .
 ue4_realistic_rende..     92     .         .      -32.58 %     .         .       -0.35 %     .         .
 unigine_heaven           322     .        0.12 %  -54.17 %     .         .       -1.42 %   -0.12 %     .
 unigine_sanctuary        264     .         .         .         .         .         .         .         .
 unigine_tropics          210     .         .         .         .         .         .         .         .
 unigine_valley           278     .       -0.15 %  -40.74 %     .         .       -2.00 %    0.09 %     .
 unity                     72     .         .         .         .         .        0.03 %     .         .
 warsow                   176     .         .         .         .         .         .         .         .
 warzone2100                4     .         .         .         .         .        0.13 %     .         .
 witcher2                1040     .       -0.03 %  -86.28 %     .         .       -0.28 %    0.01 %     .
 xcom_enemy_within       1236     .       -0.24 %  -63.54 %     .         .       -0.93 %    0.18 %     .
 yofrankie                 82     .       -0.61 % -100.00 %     .         .       -0.83 %    0.41 %     .
 -----------------------------------------------------------------------------------------------------------
 Total                  26002     .       -0.13 %  -27.03 %     .       -0.24 %   -1.57 %    0.10 %     .

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agogallivm: add helper lp_add_attr_dereferenceable
Marek Olšák [Tue, 12 Jul 2016 16:49:57 +0000 (18:49 +0200)]
gallivm: add helper lp_add_attr_dereferenceable

Not sure if this is the right way to do it, but it seems to work.

v2: make it a no-op on LLVM <= 3.5

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agoradeonsi: clean up shader value metadata code
Marek Olšák [Fri, 1 Jul 2016 14:22:05 +0000 (16:22 +0200)]
radeonsi: clean up shader value metadata code

No change in behavior.
BTW, tbaa_md_kind == 1, which was the magic number in the code.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agoradeonsi: remove LLVMNoUnwindAttribute uses
Marek Olšák [Fri, 1 Jul 2016 18:13:08 +0000 (20:13 +0200)]
radeonsi: remove LLVMNoUnwindAttribute uses

always set by gallivm

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agoradeonsi: fix a typo in SI_PARAM_LINEAR_* handling
Marek Olšák [Mon, 11 Jul 2016 17:14:33 +0000 (19:14 +0200)]
radeonsi: fix a typo in SI_PARAM_LINEAR_* handling

introduced in 476e9cee1d0cbe321c401277214e6c36ce5b18c9

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agogallium/radeon: normalize the code style
Marek Olšák [Fri, 1 Jul 2016 18:00:47 +0000 (20:00 +0200)]
gallium/radeon: normalize the code style

no change in behavior

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agoradeonsi: just save buffer sizes instead of buffers while recording IBs
Marek Olšák [Fri, 8 Jul 2016 16:03:56 +0000 (18:03 +0200)]
radeonsi: just save buffer sizes instead of buffers while recording IBs

whole buffer objects are not needed

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
7 years agoAdd c99_alloca.h include to fix compilation on Cygwin
Jon Turney [Tue, 5 Jul 2016 09:40:05 +0000 (10:40 +0100)]
Add c99_alloca.h include to fix compilation on Cygwin

Fix compilation on Cygwin, since 50b22354, by adding c99_alloca.h include,
which should know how to portably make the alloc() prototype available.

Signed-off-by: Jon Turney <jon.turney@dronecode.org.uk>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
7 years agoi965/blorp: Cleanup leftovers from push constant disabling
Topi Pohjolainen [Wed, 13 Jul 2016 08:03:54 +0000 (11:03 +0300)]
i965/blorp: Cleanup leftovers from push constant disabling

Setup for pixel shader push constants is the same as for other
stages. Note that on gen8+ the if-else branches were identical
and the generation check for packet size redundant.

Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
7 years agoi965/blorp/gen7+: Bring back push constant setup
Topi Pohjolainen [Tue, 12 Jul 2016 19:09:42 +0000 (22:09 +0300)]
i965/blorp/gen7+: Bring back push constant setup

This is partial revert of commit cc2d0e64.

It looks that even though blorp disables a stage the corresponding
3DSTATE_CONSTANT_XS packet is needed to be programmed. Hardware
seems to try to fetch the constants even for disabled stages.
Therefore care needs to be taken that the constant buffer is
set up properly. Blorp will continue to trash it into non-existing
such as before.
It is possible that this could be omitted on SKL where the
constant buffer is considered when the corresponding binding table
settings are changed. Bspec:

  "The 3DSTATE_CONSTANT_* command is not committed to the shader
   unit until the corresponding (same shader)
   3DSTATE_BINDING_TABLE_POINTER_* command is parsed."

However, as CONSTANT_XS packet itself does not seem to stall on its
own, it is safer to emit the packets for SKL also.

Possible alternative to blorp trashing could have been to setup
defaults in the beginning of each batch buffer. However, hardware
doesn't seem to tolerate these packets being programmed multiple
times per primitive. Bspec for IVB:

  "It is invalid to execute this command more than once between
   3D_PRIMITIVE commands."

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=96878
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
7 years agoradeonsi: silence Coverity warning
Nicolai Hähnle [Sun, 10 Jul 2016 14:13:28 +0000 (16:13 +0200)]
radeonsi: silence Coverity warning

Coverity's analysis is too weak to understand that
r600_init_flushed_depth(_, _, NULL) only returns true when
flushed_depth_texture was assigned a non-NULL value.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
7 years agoi965/fs: do d2x lowering before simd splitting
Samuel Iglesias Gonsálvez [Fri, 8 Jul 2016 07:33:45 +0000 (09:33 +0200)]
i965/fs: do d2x lowering before simd splitting

So that we can have gen7 split large writes produced by this lowering pass.

Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
7 years agoi965/fs: do pack lowering before simd splitting
Iago Toral Quiroga [Fri, 1 Apr 2016 09:54:47 +0000 (11:54 +0200)]
i965/fs: do pack lowering before simd splitting

So that we can have gen7 split large writes produced by the pack lowering.

Reviewed-by: Francisco Jerez <currojerez@riseup.net>
7 years agoi965/fs: do not require force_writemask_all with exec_size 4
Samuel Iglesias Gonsálvez [Mon, 20 Jun 2016 10:13:14 +0000 (12:13 +0200)]
i965/fs: do not require force_writemask_all with exec_size 4

So far we only used instructions with this size in situations where we
did not operate per-channel and we wanted to ignore the execution mask,
but gen7 fp64 will need to emit code with a width of 4 that needs
normal execution masking.

v2:
- Modify the assert instead of deleting it (Curro)

Reviewed-by: Francisco Jerez <currojerez@riseup.net>
7 years agoi965/fs/gen7: split instructions that run into exec masking bugs
Iago Toral Quiroga [Wed, 30 Mar 2016 12:00:31 +0000 (14:00 +0200)]
i965/fs/gen7: split instructions that run into exec masking bugs

In fp64 we can produce code like this:

mov(16) vgrf2<2>:UD, vgrf3<2>:UD

That our simd lowering pass would typically split in instructions with a
width of 8, writing to two consecutive registers each. Unfortunately, gen7
hardware has a bug affecting execution masking and as a result, the
second GRF register write won't work properly. Curro verified this:

"The problem is that pre-Gen8 EUs are hardwired to use the QtrCtrl+1
 (where QtrCtrl is the 8-bit quarter of the execution mask signals
 specified in the instruction control fields) for the second
 compressed half of any single-precision instruction (for
 double-precision instructions it's hardwired to use NibCtrl+1,
 at least on HSW), which means that the EU will apply the wrong
 execution controls for the second sequential GRF write if the number
 of channels per GRF is not exactly eight in single-precision mode (or
 four in double-float mode)."

In practice, this means that we cannot write more than one
consecutive GRF in a single instruction if the number of channels
per GRF is not exactly eight in single-precision mode (or four
in double-float mode).

This patch makes our SIMD lowering pass split this kind of instructions
so that the split versions only write to a single register. In the
example above this means that we split the write in 4 instructions, each
one writing 4 UD elements (width = 4) to a single register.

v2 (Curro):
 - Make explicit that the thing about hardwiring NibCtrl+1 for the second
   compressed half is known to happen in Haswell and the issue with IVB
   might not be exactly the same.
 - Assign max_width instead of returning early so that we can handle
   multiple restrictions affecting to the same instruction.
 - Avoid division by 0 if the instruction does not write any registers.
 - Ignore instructions what have WE_all set.
 - Use the instruction execution type size instead of the dst type size.

v3 (Curro):
 - Move the implementation down so it is not placed in the middle of another
   workaround.
 - Declare channels_per_grf as const.
 - Don't break the loop early if we find a BAD_FILE source.
 - Fix the number of channels that the hardware shifts for the second half
   of a compressed instruction to be 8 in single precision and 4 in double
   precision.

Reviewed-by: Francisco Jerez <currojerez@riseup.net>
7 years agoi965/fs: use the new helper function to create double immediates
Iago Toral Quiroga [Mon, 11 Jul 2016 11:25:52 +0000 (13:25 +0200)]
i965/fs: use the new helper function to create double immediates

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
7 years agoi965/fs: add a helper function to create double immediates
Iago Toral Quiroga [Wed, 9 Mar 2016 13:12:43 +0000 (14:12 +0100)]
i965/fs: add a helper function to create double immediates

Gen7 hardware does not support double immediates so these need
to be moved in 32-bit chunks to a regular vgrf instead. Instead
of doing this every time we need to create a DF immediate,
create a helper function that does the right thing depending
on the hardware generation.

v2:
- Define setup_imm_df() as an independent function (Curro)
- Create a specific builder to get rid of some instruction field
  assignments (Curro).

v3:
- Get devinfo from builder (Kenneth)

Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
7 years agovc4: Validate QPU uniform pointer updates.
Eric Anholt [Tue, 3 May 2016 18:30:32 +0000 (11:30 -0700)]
vc4: Validate QPU uniform pointer updates.

7 years agovc4: Add support for NIR loops and break/continue.
Eric Anholt [Mon, 2 May 2016 20:35:21 +0000 (13:35 -0700)]
vc4: Add support for NIR loops and break/continue.

7 years agovc4: Add support for emitting NIR IF nodes.
Eric Anholt [Fri, 11 Mar 2016 00:03:45 +0000 (16:03 -0800)]
vc4: Add support for emitting NIR IF nodes.

7 years agovc4: Add support for storing to NIR registers in a non-SSA fashion.
Eric Anholt [Wed, 27 Apr 2016 23:01:24 +0000 (16:01 -0700)]
vc4: Add support for storing to NIR registers in a non-SSA fashion.

Previously, there were occasionally NIR registers in our programs, but
they were always actually used SSA-only.  Now that we're trying to support
control flow, we need to actually conditionally move to registers based on
whether channels are active or not.

7 years agovc4: Add a flag in the screen to track control flow support.
Eric Anholt [Mon, 11 Jul 2016 17:55:18 +0000 (10:55 -0700)]
vc4: Add a flag in the screen to track control flow support.

For now it's still always false, but I need it in place for kernel
backwards compat support as I extend the backend for control flow.

7 years agovc4: Define a QIR branch instruction
Eric Anholt [Tue, 15 Mar 2016 20:55:28 +0000 (13:55 -0700)]
vc4: Define a QIR branch instruction

This uses the branch condition code in inst->cond to jump to either
successor[0] (condition matches) or successor[0] (condition doesn't
match).

7 years agovc4: Add kernel support for branching in shader validation.
Eric Anholt [Tue, 15 Mar 2016 20:53:02 +0000 (13:53 -0700)]
vc4: Add kernel support for branching in shader validation.

We're already checking that branch instructions are within the
contents of the shader and the proper PROG_END sequence is present.
The other thing we need in the presence of branching is to verify that
the shader doesn't overflow past the end of the uniforms stream.

To do that, we require that at the start of any basic block reading
uniforms have the following instructions:

load_imm temp, <offset within uniform stream>
add unif_addr, temp, unif

The instructions are generated by userspace, and the kernel verifies
that the load_imm is of the expected offset, and that the add adds it
to a uniform.  We track which uniform in the stream that is, and at
draw call time fix up the uniform stream to have the address of the
start of the shader's uniforms for that draw call.

Signed-off-by: Eric Anholt <eric@anholt.net>
7 years agovc4: Add a bitmap of branch targets in kernel validation.
Eric Anholt [Sat, 2 Jul 2016 17:03:05 +0000 (10:03 -0700)]
vc4: Add a bitmap of branch targets in kernel validation.

This isn't used yet, it's just a first step toward loop validation.
During the main parsing of instructions, we need to know when we hit a new
basic block so that we can reset validated state.

7 years agovc4: Track the current instruction into the validation_state.
Eric Anholt [Fri, 3 Jun 2016 19:48:59 +0000 (12:48 -0700)]
vc4: Track the current instruction into the validation_state.

This reduces how much we need to pass around as arguments, which was
becoming more of a problem with looping validation.

7 years agovc4: Add QPU support for generating BRANCH instructions.
Eric Anholt [Tue, 15 Mar 2016 20:43:54 +0000 (13:43 -0700)]
vc4: Add QPU support for generating BRANCH instructions.

7 years agovc4: Print live variable start/ends during QIR dumping.
Eric Anholt [Wed, 6 Jul 2016 19:49:52 +0000 (12:49 -0700)]
vc4: Print live variable start/ends during QIR dumping.

This only happens when live variables are set up, which is not in the
normal dump, but is set up when we've failed to register allocate.

7 years agovc4: Implement live intervals using a CFG.
Eric Anholt [Thu, 10 Mar 2016 20:53:57 +0000 (12:53 -0800)]
vc4: Implement live intervals using a CFG.

Right now our CFG is always a trivial single basic block, but that will
change when enable loops.

7 years agovc4: Make vc4_qir_schedule handle each block in the program.
Eric Anholt [Tue, 19 Jan 2016 22:15:54 +0000 (14:15 -0800)]
vc4: Make vc4_qir_schedule handle each block in the program.

Basically we just treat each block independently.  The only inter-block
scheduling I can think of that would be be interesting would be to move
texture result collection to after a short loop/if block that doesn't do
texturing.  However, the kernel disallows that as part of its security
validation.

7 years agovc4: Convert uniforms lowering to work with multiple blocks.
Eric Anholt [Sat, 9 Jul 2016 00:10:38 +0000 (17:10 -0700)]
vc4: Convert uniforms lowering to work with multiple blocks.

We still decide which uniform to lower based on how many
instructions-that-need-lowering use that uniform, but now we emit a new
temporary uniform load in each of the basic blocks containing an
instruction being lowered.

This commit is best reviewed with diff -b.

7 years agovc4: Convert vc4_opt_peephole_sf to work with control flow.
Eric Anholt [Fri, 8 Jul 2016 23:59:15 +0000 (16:59 -0700)]
vc4: Convert vc4_opt_peephole_sf to work with control flow.

We need to apply the peephole pass to each of the blocks in the program.
We don't do dataflow analysis for SF across blocks, but we also don't
generate code that would need us to do so.

7 years agovc4: Create a basic block structure and move the instructions into it.
Eric Anholt [Fri, 8 Jul 2016 22:42:15 +0000 (15:42 -0700)]
vc4: Create a basic block structure and move the instructions into it.

The optimization passes and scheduling aren't actually ready for multiple
blocks with control flow yet (as seen by the "cur_block" references in
them instead of iterating over blocks), but this creates the structures
necessary for converting them.

7 years agovc4: Add a "qir_for_each_inst_inorder" macro and use it in many places.
Eric Anholt [Fri, 8 Jul 2016 22:24:34 +0000 (15:24 -0700)]
vc4: Add a "qir_for_each_inst_inorder" macro and use it in many places.

We have the prior list_foreach() all over the code, but I need to move
where instructions live as part of adding support for control flow.  Start
by just converting to a helper iterator macro.  (The simpler
"qir_for_each_inst()" will be used for the for-each-inst-in-a-block
iterator macro later)

7 years agovc4: Also enable phi elimination.
Eric Anholt [Wed, 6 Jul 2016 22:21:53 +0000 (15:21 -0700)]
vc4: Also enable phi elimination.

This avoids a bunch of code gen regressions when enabling loops in vc4.

Prior to that, the GLSL that would have generated these optimizable phi
nodes was being lowered to csels between either (undef, a) or (a, a), and
those were being dealt with by nir_opt_undef and nir_opt_algebraic.

7 years agovc4: fix memory leak
Eric Engestrom [Tue, 12 Jul 2016 22:29:34 +0000 (23:29 +0100)]
vc4: fix memory leak

The allocation has succeeded by that point, so it needs to be freed.

CovID: 1358929
Signed-off-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Eric Anholt <eric@anholt.net>
7 years agovc4: Close our screen's fd on screen close.
Eric Anholt [Mon, 11 Jul 2016 23:46:20 +0000 (16:46 -0700)]
vc4: Close our screen's fd on screen close.

We're passed in a freshly dup()ed fd on screen create, so we should close
it on exit.  Debugged by Hugh Cole-Baker.

7 years agonir: Add optimization for (a || True == True)
Eric Anholt [Wed, 6 Jul 2016 22:00:45 +0000 (15:00 -0700)]
nir: Add optimization for (a || True == True)

This was appearing in vc4 VS/CS in mupen64, due to vertex attrib lowering
producing some constants that were getting compared.

total instructions in shared programs: 112276 -> 112198 (-0.07%)
instructions in affected programs:     2239 -> 2161 (-3.48%)
total estimated cycles in shared programs: 283102 -> 283038 (-0.02%)
estimated cycles in affected programs:     2365 -> 2301 (-2.71%)

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
7 years agoswr: [rasterizer core] correct MSAA behavior for conservative rasterization
Tim Rowley [Wed, 29 Jun 2016 15:47:31 +0000 (09:47 -0600)]
swr: [rasterizer core] correct MSAA behavior for conservative rasterization

Signed-off-by: Tim Rowley <timothy.o.rowley@intel.com>
7 years agoswr: [rasterizer core] conservative rast backend changes
Tim Rowley [Mon, 27 Jun 2016 21:50:58 +0000 (15:50 -0600)]
swr: [rasterizer core] conservative rast backend changes

Signed-off-by: Tim Rowley <timothy.o.rowley@intel.com>
7 years agoswr: [rasterizer] buckets cleanup
Tim Rowley [Mon, 20 Jun 2016 05:37:57 +0000 (23:37 -0600)]
swr: [rasterizer] buckets cleanup

Signed-off-by: Tim Rowley <timothy.o.rowley@intel.com>
7 years agoswr: [rasterizer core] make all api functions call GetContext
Tim Rowley [Thu, 16 Jun 2016 19:53:28 +0000 (13:53 -0600)]
swr: [rasterizer core] make all api functions call GetContext

Small api cleanup.  Make all api functions call GetContext instead
of locally casting handle.  Makes debugging easier by providing a
single point to track context changes.

Signed-off-by: Tim Rowley <timothy.o.rowley@intel.com>
7 years agoswr: [rasterizer] add support for llvm-3.9
Tim Rowley [Tue, 14 Jun 2016 23:54:34 +0000 (17:54 -0600)]
swr: [rasterizer] add support for llvm-3.9

v2: use signed compare, remove unneeded vmask

Signed-off-by: Tim Rowley <timothy.o.rowley@intel.com>
7 years agoswr: [rasterizer jitter] fix llvm-3.7 compile
Tim Rowley [Wed, 6 Jul 2016 21:26:16 +0000 (16:26 -0500)]
swr: [rasterizer jitter] fix llvm-3.7 compile

d3d97f8 broke llvm-3.7, which has a mismatched API for
setDataLayout/getDataLayout.

Signed-off-by: Tim Rowley <timothy.o.rowley@intel.com>
7 years agodocs: remove duplicated line in 12.0.1 release notes file
Brian Paul [Tue, 12 Jul 2016 15:42:16 +0000 (09:42 -0600)]
docs: remove duplicated line in 12.0.1 release notes file

Signed-off-by: Brian Paul <brianp@vmware.com>
7 years agost/omx/dec: convert decoder video buffer to progressive
Leo Liu [Mon, 4 Jul 2016 20:32:48 +0000 (16:32 -0400)]
st/omx/dec: convert decoder video buffer to progressive

with encode tunneling

The idea of encode tunneling is to use video buffer directly for encoder,
but currently the encoder doesn’t support interlaced surface, the OMX
decoder set progressive surface before on that purpose.

Since now we are polling the driver for interlacing information for
decoder, we got the interlaced as preferred as other APIs(VDPAU, VA-API),
thus breaking the transcode with tunneling.

The solution is when with tunnel detected, re-allocate progressive target
buffers, and then converting the interlaced decoder results to there.

This has been tested with transcode results bit to bit matching as before
with surface from progressive to progressive.

Signed-off-by: Leo Liu <leo.liu@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Tested-by: Julien Isorce <j.isorce@samsung.com>
7 years agovl/compositor: set layer of y or uv to render
Leo Liu [Wed, 6 Jul 2016 16:42:37 +0000 (12:42 -0400)]
vl/compositor: set layer of y or uv to render

Signed-off-by: Leo Liu <leo.liu@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Tested-by: Julien Isorce <j.isorce@samsung.com>
7 years agovl/compositor: add weave to yuv shader
Leo Liu [Wed, 6 Jul 2016 16:34:02 +0000 (12:34 -0400)]
vl/compositor: add weave to yuv shader

This shader will make interlaced yuv to progressive yuv.

Signed-off-by: Leo Liu <leo.liu@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Tested-by: Julien Isorce <j.isorce@samsung.com>
7 years agovl/compositor: move weave shader out from rgb weaving
Leo Liu [Mon, 4 Jul 2016 19:59:00 +0000 (15:59 -0400)]
vl/compositor: move weave shader out from rgb weaving

We'll use weave shader in the later patch.

Signed-off-by: Leo Liu <leo.liu@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Tested-by: Julien Isorce <j.isorce@samsung.com>
7 years agoglsl_to_tgsi: don't use the negate modifier in integer ops after bitcast
Marek Olšák [Sun, 3 Jul 2016 12:33:55 +0000 (14:33 +0200)]
glsl_to_tgsi: don't use the negate modifier in integer ops after bitcast

This bug is uncovered by glsl/lower_if_to_cond_assign.
I don't know if it can be reproduced in any other way.

Cc: <mesa-stable@lists.freedesktop.org>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
7 years agoclover/api: Implement clLinkProgram per-device binary presence validation rule.
Francisco Jerez [Mon, 6 Jun 2016 02:12:22 +0000 (19:12 -0700)]
clover/api: Implement clLinkProgram per-device binary presence validation rule.

Reviewed-by: Serge Martin <edb+mesa@sigluy.net>
Tested-by: Jan Vesely <jan.vesely@rutgers.edu>
7 years agoclover: Add clLinkProgram (CL 1.2).
Serge Martin [Mon, 9 May 2016 04:57:34 +0000 (21:57 -0700)]
clover: Add clLinkProgram (CL 1.2).

[ Francisco Jerez: Use validate_build_common for error checking,
  simplify control flow slightly and handle additional exception
  types. ]

Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Tested-by: Jan Vesely <jan.vesely@rutgers.edu>
7 years agoclover: Trivial cleanups for api/program.cpp.
Francisco Jerez [Tue, 17 May 2016 14:03:18 +0000 (16:03 +0200)]
clover: Trivial cleanups for api/program.cpp.

Reviewed-by: Serge Martin <edb+mesa@sigluy.net>
Tested-by: Jan Vesely <jan.vesely@rutgers.edu>
7 years agoclover/core: Remove compiler.hpp.
Francisco Jerez [Tue, 17 May 2016 14:03:17 +0000 (16:03 +0200)]
clover/core: Remove compiler.hpp.

header_map was the only definition left in compiler.hpp, move it into
program.hpp which is its only user in clover/core.

Reviewed-by: Serge Martin <edb+mesa@sigluy.net>
Tested-by: Jan Vesely <jan.vesely@rutgers.edu>
7 years agoclover/llvm: Get rid of compile_program_llvm().
Francisco Jerez [Tue, 17 May 2016 14:03:16 +0000 (16:03 +0200)]
clover/llvm: Get rid of compile_program_llvm().

Superseded by compile_program() and link_program().

Reviewed-by: Serge Martin <edb+mesa@sigluy.net>
Tested-by: Jan Vesely <jan.vesely@rutgers.edu>
7 years agoclover: Provide separate program methods for compilation and linking.
Francisco Jerez [Tue, 17 May 2016 14:03:14 +0000 (16:03 +0200)]
clover: Provide separate program methods for compilation and linking.

[ Serge Martin: Fix inverted opts and log build ctor args.
  Keep the log related to the build. Fix indentation ]

Reviewed-by: Serge Martin <edb+mesa@sigluy.net>
Tested-by: Jan Vesely <jan.vesely@rutgers.edu>
7 years agoclover: Unify program::build_* into a single method returning a struct.
Francisco Jerez [Tue, 17 May 2016 14:03:13 +0000 (16:03 +0200)]
clover: Unify program::build_* into a single method returning a struct.

This gets rid of the program::build_* query methods and replaces them
with the program::build() method that returns a single data structure
containing all parameters for the last build done on the given target
device (including build logs, options and the binary itself).

[ Serge Martin: Fix inverted opts and log build ctor args ]

Reviewed-by: Serge Martin <edb+mesa@sigluy.net>
Tested-by: Jan Vesely <jan.vesely@rutgers.edu>
7 years agoclover: Change program::build opts argument to std::string.
Serge Martin [Tue, 17 May 2016 14:03:12 +0000 (16:03 +0200)]
clover: Change program::build opts argument to std::string.

Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Tested-by: Jan Vesely <jan.vesely@rutgers.edu>
7 years agoclover: Define error subclass to signal build option parse failure.
Francisco Jerez [Tue, 17 May 2016 14:03:11 +0000 (16:03 +0200)]
clover: Define error subclass to signal build option parse failure.

Reviewed-by: Serge Martin <edb+mesa@sigluy.net>
Tested-by: Jan Vesely <jan.vesely@rutgers.edu>
7 years agoclover: Move back to using build_error to signal compilation failure.
Francisco Jerez [Sun, 19 Jun 2016 21:31:59 +0000 (14:31 -0700)]
clover: Move back to using build_error to signal compilation failure.

This partially reverts 7e0180d57d330bd8d3047e841086712376b2a1cc.
Having two different exception subclasses for compilation and linking
makes it more difficult to share or move code between the two
codepaths, because the exact same function under the same error
condition would need to throw one exception or the other depending on
what top-level API is being implemented with it.  There is little
benefit anyway because clCompileProgram() and clLinkProgram() can tell
whether they are linking or compiling a program.

Reviewed-by: Serge Martin <edb+mesa@sigluy.net>
Tested-by: Jan Vesely <jan.vesely@rutgers.edu>
7 years agoclover: Override ret_object.
Serge Martin [Tue, 17 May 2016 14:03:09 +0000 (16:03 +0200)]
clover: Override ret_object.

Return an API object from an intrusive reference to a Clover object,
incrementing the reference count of the object.

Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Tested-by: Jan Vesely <jan.vesely@rutgers.edu>
7 years agoclover/tgsi: Add stub link_program() function.
Francisco Jerez [Tue, 17 May 2016 14:03:08 +0000 (16:03 +0200)]
clover/tgsi: Add stub link_program() function.

Reviewed-by: Serge Martin <edb+mesa@sigluy.net>
Tested-by: Jan Vesely <jan.vesely@rutgers.edu>
7 years agoclover/tgsi: Move compiler entry point declaration into tgsi directory and namespace.
Francisco Jerez [Tue, 17 May 2016 14:03:07 +0000 (16:03 +0200)]
clover/tgsi: Move compiler entry point declaration into tgsi directory and namespace.

Reviewed-by: Serge Martin <edb+mesa@sigluy.net>
Tested-by: Jan Vesely <jan.vesely@rutgers.edu>
7 years agoclover/llvm: Implement the -create-library linker option.
Francisco Jerez [Fri, 3 Jun 2016 22:46:21 +0000 (15:46 -0700)]
clover/llvm: Implement the -create-library linker option.

[ Serge Martin: disable internalize pass when building a library.
  Otherwise some functions may be inlined and removed ]

Reviewed-by: Serge Martin <edb+mesa@sigluy.net>
Tested-by: Jan Vesely <jan.vesely@rutgers.edu>
7 years agoclover/llvm: Implement linkage of multiple clover modules.
Francisco Jerez [Mon, 13 Jun 2016 08:41:05 +0000 (01:41 -0700)]
clover/llvm: Implement linkage of multiple clover modules.

Reviewed-by: Serge Martin <edb+mesa@sigluy.net>
Tested-by: Jan Vesely <jan.vesely@rutgers.edu>