panfrost: Draw the wallpaper when only depth/stencil bufs are cleared

When only the depth/stencil bufs are cleared, we should make sure the
color content is reloaded into the tile buffers if we want to preserve
their content.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Make sure a clear does not re-use a pre-existing batch

glClear()s are expected to be the first thing GL apps do before drawing
new things. If there's already an existing batch targetting the same
FBO that has draws attached to it, we should make sure the new clear
gets a new batch assigned to guaranteed that the FB content is actually
cleared with the requested color/depth/stencil values.

We create a panfrost_get_fresh_batch_for_fbo() helper for that and
call it from panfrost_clear().

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

iris: Update comment about 3-component formats and buffer textures

You can't render to PIPE_BUFFER so there's no reason to prefer RGBX.
PBO upload would like to use proper RGB textures as source data.

iris: Allow packed RGB pbo uploads

Hitting any fallback path on Broxton as we require clflushing the whole
buffer even for an upload of a subtexture. However, since gallium
provides a pbo upload path, allow it to sample packed RGB if supported.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

anv/android: fix images created with external format support

This fixes a case where user first creates image and then later binds it
with memory created from AHW buffer.

Cc: <mesa-stable@lists.freedesktop.org>
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>

meson: Always add LLVM coroutines module.

It gets used by the gallium auxiliary draw module, which gets used
pretty much always when LLVM is used as JIT.

At the same time most builds don't hit the issue here because the
shared library of LLVM contains all modules.

Fixes: d32690b43c91 ("gallivm: add coroutine pass manager support")
Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/951
Reviewed-by: Gert Wollny <gert.wollny@collabora.com>

amd: Move all amd/common code that depends on LLVM to amd/llvm.

This commit is a step towards the goal of being able to build RADV
without LLVM. In the future we would like to offer the option to
use RADV solely with ACO. There is still a need for the common AMD
code located in amd/common but the LLVM specific parts need to be
separated.

Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Acked-by: Marek Olšák <marek.olsak@amd.com>
Acked-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>

nvc0: add support for GL_EXT_demote_to_helper_invocation

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>

gallium/tgsi: add support for DEMOTE and READ_HELPER opcodes

This mirrors the intrinsics in the GLSL IR. One could imagine an
alternate definition where reading the semantic would account for the
READ_HELPER functionality, but that feels potentially dodgy and could be
subject to CSE unpleasantness.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>

radeonsi: use simple_mtx_t instead of mtx_t

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

st/mesa: use simple_mtx_t instead of mtx_t

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

gallium: add PIPE_RESOURCE_FLAG_SINGLE_THREAD_USE to skip util_range lock

u_upload_mgr sets it, so that util_range_add can skip the lock.

The time spent in tc_transfer_flush_region decreases from 0.8% to 0.2%
in torcs on radeonsi.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

util: use simple_mtx_t for util_range

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

winsys/radeon: initialize SIMD properties in radeon_info

This was missed when I added them.

Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/1839
Fixes: 0692ae34e93 ("ac: move ac_get_num_physical_sgprs into radeon_info")
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>

iris: Drop vtbl usage for some load_register calls

We can just call the actual functions directly.

iris/state: Move reg/mem load/store functions earlier in file

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>

meson: drop unused inc_nir

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>

meson: drop duplicate inc_nir from spirv2nir

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>

meson: drop duplicate inc_nir from libglsl

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>

meson: drop duplicate inc_nir from libiris

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>

meson: rename libnir to _libnir to make it clear it's not meant to be used anywhere else

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>

meson: use idep_nir instead of libnir in pipe-loader

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>

meson: use idep_nir instead of libnir in haiku softpipe

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>

meson: use idep_nir instead of libnir in gallium nine

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>

meson: use idep_nir instead of libnir in libclnir

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>

meson: use idep_nir instead of libnir in libnouveau

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>

meson: add missing idep_nir_headers in iris_gen_libs

Fixes: 4929f020c37eac5b07f9 ("iris: better SBE")
Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>

script: drop get_reviewer.pl

This script doesn't make sense anymore in the age of GitLab.

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>

meson/loader: drop unneeded *.h file

Meson automatically tracks any file included by a file it already tracks,
and `pci_id_driver_map.h` & `loader.h` are included by `loader.c`, while
`loader_dri3_helper.h` is included by `loader_dri3_helper.c`.

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Acked-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>

loader: use ARRAY_SIZE instead of NULL sentinel

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Acked-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>

loader: s/int/bool/ for predicate result

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Acked-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>

loader: replace int/1/0 with bool/true/false

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Acked-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>

egl: replace MESA_EGL_NO_X11_HEADERS hack with upstream EGL_NO_X11

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>

iris: Drop bonus parameters from iris_init_*_context()

Nothing uses vtbl or dbg, and screen is available from the batch.

nir/constant_folding: fold load_constant intrinsics

These can appear after loop unrolling.

v2: stylistic changes
v2: replace state->mem_ctx with state->shader
v2: add bounds checking
v3: use nir_intrinsic_range() for bounds checking
v3: fix issue where partially out-of-bounds reads are replaced with undefs
v4: fix merge conflicts during rebase
v5: split into two commits
v6: set constant_data to NULL after freeing (fixes nir_sweep()/Iris)
v7: don't remove the constant data if there are no constant loads

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com> (v6)
Acked-by: Ian Romanick <ian.d.romanick@intel.com>

nir/constant_folding: add back and use constant_fold_state

Useful for load_constant folding.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>

anv: Implement VK_KHR_shader_clock

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>

spirv: Implement SPV_KHR_shader_clock

We only have the subgroup variant in NIR (equivalent to clockARB), so
only support that for now.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>

vulkan: Update the XML and headers to 1.1.124

Acked-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>

Revert "iris: Hack up a SKL/Gen9LP PS push constant fifo depth workaround"

This reverts commit 4f857423b3c095516e553b976b41969c2b9721fa.

It caused GPU hangs on all affected platforms, in e.g.
Piglit bin/stencil-twoside -auto -fbo.

gitlab-ci/lava: Fix image to use in test jobs

In the test stage, we can use any of the two container images as we
arent going to do anything architecture-dependent when submitting the
jobs to LAVA.

But if we are in a pipeline in which the images need to be rebuilt and
one finishes much earlier than the other, it could happen that the test
job that executes first fails to find the container image.

To avoid that, have each job in the test stage to use the image that has
been already implicitly built by depending on the build job for the
given arch.

Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>

Revert "Revert "st/dri2: Implement DRI2bufferDamageExtension""

This reverts commit 19546108d3dd5541a189e36df4ea83b3f519e48f.
This commit breaks the build because lima implements
->set_damage_region(). I guess we'll need more discussion before
removing the ->set_damage_region() hook.

Revert "st/dri2: Implement DRI2bufferDamageExtension"

This reverts commit 492ffbed63a2a62759224b1c7d45aa7923d8f542.

BACK_LEFT attachment can be outdated when the user calls
KHR_partial_update(), leading to a damage region update on the
wrong pipe_resource object.
Let's not expose the ->set_damage_region() method until the core is
fixed to handle that properly.

Cc: mesa-stable@lists.freedesktop.org
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Acked-by: Daniel Stone <daniels@collabora.com>

gitlab-ci: Move LAVA-related files into top-level ci dir

In preparation for testing drivers other than Panfrost in LAVA labs.

Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

gitlab-ci: Run dEQP on devices with Panfrost

Include Panfrost's gitlab.ci.yml file from Mesa's main .gitlab-ci.yml so
we test on devices with Panfrost.

This uses LAVA to schedule jobs in the devices and will be the base for
testing Etnaviv, Lima, etc.

Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

iris: Hack up a SKL/Gen9LP PS push constant fifo depth workaround

This is a port of Nanley's 904c2a617d86944fbdc2c955f327aacd0b3df318
from i965 to iris.

One concern is that iris uses larger batches, and also emits far fewer
commands, so we may come closer to the 500 limit within a batch, and
could need to supplement this with actual counting.  Manhattan 3.0 had
239 3DSTATE_CONSTANT_PS packets in a batch,  Unigine Valley had 155.
So it seems like we're still in the realm of safety.

iris: Refactor push constant allocation so we can reuse it

We'll need this for a workaround shortly.  While refactoring, also
improve the comment slightly.

intel/isl: set vertical surface alignment on null surfaces

Just following the spec. Somewhat unclear whether this applies to NULL
surfaces.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

intel/isl: set surface array appropriately

This doesn't seem to affect anything.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

intel/isl: Set null surface format to R32_UINT

It appears we never had a test in piglit or deqp sampling from a null
surface...

It turns out this triggers a hang on IVB only. Updating the null
surface format to R32_UINT fixes the hang on ivb and doesn't affect
other platforms, so set it by default for all platforms.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Gitlab: https://gitlab.freedesktop.org/mesa/mesa/issues/1872
Cc: <mesa-stable@lists.freedesktop.org>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

etnaviv: set texture INT_FILTER bit

This should improve texture sampling performance on GC3000.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>

etnaviv: implement texture comparator

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>

etnaviv: update headers from rnndb

Update to etna_viv commit 7ff8029.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>

intel: fix subslice computation from topology data

We're missing the offset of the slice in the subslice mask...

This worked for most platforms that don't have first slice fused off
because we would reread the same mask from slice0 again and again...

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: c1900f5b0f ("intel: devinfo: add helper functions to fill fusing masks values")
Gitlab: https://gitlab.freedesktop.org/mesa/mesa/issues/1869
Reviewed-by: Mark Janes <mark.a.janes@intel.com>

dri: Avoid swapbuffer throttling in glXCopySubBufferMESA

We were supplying __DRI2_THROTTLE_SWAPBUFFER, rather than the obvious
choice of __DRI2_THROTTLE_COPYSUBBUFFER.  This meant that we hit the
swap-based frame throttling.  glXCopySubBuffer doesn't seem like it's
intended to be a frame boundary, so we'd like to avoid this throttling.

Tested-by: Michel Dänzer <mdaenzer@redhat.com> # DRI3 only
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>

st/dri: Perform MSAA downsampling for __DRI2_THROTTLE_COPYSUBBUFFER

glXCopySubBufferMESA copies data from the back buffer to the front,
so it needs to perform a MSAA downsampling operation just like
glXSwapBuffers would.

Currently, the CopySubBuffer implementations supply a throttle reason
of __DRI2_THROTTLE_SWAPBUFFERS, so they hit this path and work today.
But we'd like to avoid swapbuffer throttling in this case, so the next
patch will change that reason.

Tested-by: Michel Dänzer <mdaenzer@redhat.com> # DRI3 only
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>

scons/MSYS2-MinGW-W64: Fix build options defaults
Signed-off-by: Prodea Alexandru-Liviu <liviuprodea@yahoo.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Cc: <mesa-stable@lists.freedesktop.org>
When building in a MSYS2 Mingw-w64 environment Mesa3D sets wrong default build options which inevitably lead to build failure.

intel/error2aub: add support for platforms without PPGTT

Not much to do to enable this, just make sure to always write to the
GGTT :)

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

aco: fix load_constant with multiple arrays

I thought I fixed this, but I guess I must have broken it again.

Fixes various dEQP-VK.draw.* tests

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>

nir: Fix some wonky whitespace in nir_search.h.

Reviewed-by: Ian Romanick <ian.d.romainck@intel.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>

nir: Factor out most of the algebraic passes C code to .c/.h.

Working on the algebraic implementation, I was being driven nuts by my
editor not highlighting and handling indentation for the C code.  It turns
out that it's basically not pass-specific code, and we can move it over to
the relevant .c file.  Replaces 30KB of code with 34KB of data on my i965
build.  No perf diff on shader-db (n=3)

Reviewed-by: Ian Romanick <ian.d.romainck@intel.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>

nir: Keep the range analysis HT around intra-pass until we make a change.

This lets us memoize range analysis work across instructions.  Reduces
runtime of shader-db on Intel by -30.0288% +/- 2.1693% (n=3).

Fixes: 405de7ccb6cb ("nir/range-analysis: Rudimentary value range analysis pass")
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>

nir: Skip emitting no-op movs from the builder.

Having passes generate these is just making more work for copy
propagation (and thus probably calling more optimization passes)
later.  Noticed while trying to debug nir_opt_algebraic()
top-to-bottom having O(n^2) behavior due to not finding new matches in
replacement code.

Reviewed-by: Ian Romanick <ian.d.romainck@intel.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>

nir: Make nir_search's dumping go to stderr.

Reviewed-by: Ian Romanick <ian.d.romainck@intel.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>

surfaceless: Support EGL_WL_bind_wayland_display

Feature parity with the drm, x11, and wayland platforms.

Gitlab: https://gitlab.freedesktop.org/mesa/mesa/issues/1870
Tested-by: Pekka Paalanen <pekka.paalanen@collabora.co.uk>

nir/print: always use the right FILE *

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>

nir: initialize needs_helper_invocations as well

Similar to the previous commit, we should also initialize
needs_helper_invocations here.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

nir: initialize uses_discard to false

This matches what we do for uses_sample_qualifier, and what we
do in ir_set_program_inouts.cpp as well.

Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

radv/aco,aco: set lower_fmod

This simplifies ACO and allows the lowered code to be optimized (in
particular, constant folded).

Totals from affected shaders:
SGPRS: 1776 -> 1776 (0.00 %)
VGPRS: 1436 -> 1436 (0.00 %)
Spilled SGPRs: 0 -> 0 (0.00 %)
Spilled VGPRs: 0 -> 0 (0.00 %)
Private memory VGPRs: 0 -> 0 (0.00 %)
Scratch size: 0 -> 0 (0.00 %) dwords per thread
Code Size: 203452 -> 203564 (0.06 %) bytes
LDS: 0 -> 0 (0.00 %) blocks
Max Waves: 103 -> 103 (0.00 %)

At least some of the code size increase seems to be from literals being
applied to instructions as a result of constant folding.

v2: remove fmod/frem handling in init_context()

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>

scons/windows: Fix build with LLVM>=8

Fixes eebe091d292609de8aaf5b5c537d867b23908947
("scons/windows: Enable compute shaders when possible.")
Signed-off-by: Prodea Alexandru-Liviu <liviuprodea@yahoo.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>

dri3: Pass __DRI2_THROTTLE_COPYSUBBUFFER from loader_dri3_copy_drawable

0 is __DRI2_THROTTLE_SWAPBUFFER, which doesn't really make sense here.

Avoids dri_flush() throttling twice for the same glFlush call with front
buffer rendering, as described in
https://gitlab.freedesktop.org/mesa/mesa/merge_requests/2057 .

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

r600: Fix interpolateAtCentroid

If the instruction interpolateAtCentroid is used the extra interpolator
must also be enabled in the state.

Fixes: fs-interpolateatcentroid-block
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>

meson: Only error building gallium video without libdrm when the platform is drm

Fixes: 3b265f61f5f61f08718fe5bb4b2726f9b8e016cc
       ("meson: gallium media state trackers require libdrm with x11")
Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/1878
Tested-by: Vinson Lee <vlee@freedesktop.org>

pan/midgard: Replace mir_is_live_after with new pass

Now that we have live_out calculated per block as metadata, calculating
liveness of an instruction at a given point in the program becomes O(n)
to the size of the block worst-case, rather than O(n) the program.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

pan/midgard: Calculate temp_count for liveness

This needs to be correct or the analysis fails.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

pan/midgard: Invalidate liveness for mir_is_live_after

Callers should have liveness info ready. Ideally we'd have a nice
metadata tracking framework like NIR to handle this automatically, but
for now this will allow us to make forward progress... when we're about
to do something with liveness, invalidate everything ahead to force a
clean calculation.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

pan/midgard: Begin tracking liveness metadata

This will allow us to explicitly invalidate liveness analysis results so
we can cache liveness results.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

pan/midgard: Don't try to OR live_in of successors

By definition, once liveness analysis has occurred:

   live_out = OR {succ} succ->live_in

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

pan/midgard: Move RA's liveness analysis into midgard_liveness.c

There are unfortunately two distinct liveness analysis passes in the
compiler right now -- one good (but complex) pass used by RA based on
solving data flow equations, and one awful (but simple) pass used for
dead code elimination and bundling based on an abstract walk of the AST.

Let's move RA's pass into shared code so we can work on unifying.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

pan/midgard: Add mir_calculate_temp_count helper

This allows us to fill in ctx->temp_count explicitly, even if we haven't
squished down the MIR.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

pan/midgard: Remove mir_has_multiple_writes

We already enforce this with the SSA/register distinction in the
backend. There is no need to duplicate this logic merely for an assert.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

.mailmap: add a couple of aliases for Jakob Bornecrantz

Reviewed-by: Jakob Bornecrantz <jakob@collabora.com>

.mailmap: add an alias for Tomeu Vizoso

Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>

.mailmap: add an alias for Gert Wollny

Reviewed-by: Gert Wollny <gert.wollny@collabora.com>

.mailmap: add an alias for Alexandros Frantzis

Reviewed-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

.mailmap: specify spelling for Elie Tournier

Reviewed-by: Elie Tournier <tournier.elie@gmail.com>

panfrost: Get rid of the flush in panfrost_set_framebuffer_state()

Now that we have track inter-batch dependencies, the flush done in
panfrost_set_framebuffer_state() is no longer needed. Let's get rid of
it.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Kill the explicit serialization in panfrost_batch_submit()

Now that we have all the pieces in place to support pipelining batches
we can get rid of the drmSyncobjWait() at the end of
panfrost_batch_submit().

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Do fine-grained flushing when preparing BO for CPU accesses

We don't have to flush all batches when we're only interested in
reading/writing a specific BO. Thanks to the
panfrost_flush_batches_accessing_bo() and panfrost_bo_wait() helpers
we can now flush only the batches touching the BO we want to access
from the CPU.

This fixes the dEQP-GLES2.functional.fbo.render.texsubimage.* tests.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Make sure the BO is 'ready' when picked from the cache

This is needed if we want to free the panfrost_batch object at submit
time in order to not have to GC the batch on the next job submission.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Add flags to reflect the BO imported/exported state

Will be useful to make the ioctl(WAIT_BO) call conditional on BOs that
are not exported/imported (meaning that all GPU accesses are known
by the context).

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Add a panfrost_flush_batches_accessing_bo() helper

This will allow us to only flush batches touching a specific resource,
which is particularly useful when the CPU needs to access a BO.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Add a panfrost_flush_all_batches() helper

And use it in panfrost_flush() to flush all batches, and not only the
one currently bound to the context.

We also replace all internal calls to panfrost_flush() by
panfrost_flush_all_batches() ones.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Prepare panfrost_fence for batch pipelining

The panfrost_fence logic currently waits on the last submitted batch,
but the batch serialization that was enforced in
panfrost_batch_submit() is about to go away, allowing for several
batches to be pipelined, and the last submitted one is not necessarily
the one that will finish last.

We need to make sure the fence logic waits on all flushed batches, not
only the last one.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Start tracking inter-batch dependencies

The idea is to track which BO are being accessed and the type of access
to determine when a dependency exists. Thanks to that we can build a
dependency graph that will allow us to flush batches in the correct
order.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Add a panfrost_freeze_batch() helper

We'll soon need to freeze a batch not only when it's flushed, but also
when another batch depends on us, so let's add a helper to avoid
duplicating the logic.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Use the per-batch fences to wait on the last submitted batch

We just replace the per-context out_sync object by a pointer to the
the fence of the last last submitted batch. Pipelining of batches will
come later.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Add a batch fence

So we can implement fine-grained dependency tracking between batches.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Make panfrost_batch->bos a hash table

So we can store the flags as data and keep the BO as a key. This way
we keep track of the type of access done on BOs.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>

panfrost: Extend the panfrost_batch_add_bo() API to pass access flags

The type of access being done on a BO has impacts on job scheduling
(shared resources being written enforce serialization while those
being read only allow for job parallelization) and BO lifetime (the
fragment job might last longer than the vertex/tiler ones, if we can,
it's good to release BOs earlier so that others can re-use them
through the BO re-use cache).

Let's pass extra access flags to panfrost_batch_add_bo() and
panfrost_batch_create_bo() so the batch submission logic can take the
appropriate when submitting batches. Note that this information is not
used yet, we're just patching callers to pass the correct flags here.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>