Kenneth Graunke [Mon, 19 Nov 2018 22:16:03 +0000 (14:16 -0800)]
iris: Drop vestiges of throttling code
Chris Wilson [Mon, 19 Nov 2018 22:12:52 +0000 (14:12 -0800)]
iris: Merge two walks of the exec_bos list
Kenneth Graunke [Thu, 8 Nov 2018 22:42:00 +0000 (14:42 -0800)]
iris: replace vestiges of fence fds with newer exec_fence API
patch by me and Chris Wilson
Kenneth Graunke [Tue, 20 Nov 2018 06:08:51 +0000 (22:08 -0800)]
iris: Avoid synchronizing due to the workaround BO
Kenneth Graunke [Tue, 20 Nov 2018 05:49:56 +0000 (21:49 -0800)]
iris: Avoid cross-batch synchronization on read/reads
This avoids flushing batches just because e.g. both are reading the same
dynamic state streaming buffer, or shader assembly buffer.
Kenneth Graunke [Tue, 20 Nov 2018 05:40:54 +0000 (21:40 -0800)]
iris: Combine iris_use_pinned_bo and add_exec_bo
Kenneth Graunke [Tue, 20 Nov 2018 05:32:52 +0000 (21:32 -0800)]
iris: Use iris_use_pinned_bo rather than add_exec_bo directly
less special this way
Chris Wilson [Tue, 13 Nov 2018 17:55:45 +0000 (17:55 +0000)]
iris: Fix assigning the output handle for exporting for KMS
Fixes gbm_bo_get_handle() used for KMS in glamor.
Chris Wilson [Tue, 13 Nov 2018 17:55:21 +0000 (17:55 +0000)]
iris: Tidy exporting the flink handle
Kenneth Graunke [Mon, 19 Nov 2018 19:22:56 +0000 (11:22 -0800)]
iris: Fix SLM
Now that Jason has set up the L3 we can do this. Also, my assert was
useless because we hadn't set up the field in the first place. Oops.
Jason Ekstrand [Wed, 14 Nov 2018 22:52:26 +0000 (16:52 -0600)]
iris: Don't set constant read lengths at upload time
They're set in derived_data as part of store_cs_state
Jason Ekstrand [Thu, 15 Nov 2018 05:09:10 +0000 (23:09 -0600)]
iris: Configure the L3$ on the compute context
Kenneth Graunke [Mon, 19 Nov 2018 08:23:54 +0000 (00:23 -0800)]
iris: properly pin stencil buffers
Kenneth Graunke [Sat, 17 Nov 2018 00:01:28 +0000 (16:01 -0800)]
iris: Fix TCS/TES slot unification
TCS outputs, TES inputs...not TCS inputs
Fixes some barrier tests
Kenneth Graunke [Fri, 16 Nov 2018 23:12:57 +0000 (15:12 -0800)]
iris: more todo notes
Kenneth Graunke [Fri, 9 Nov 2018 05:16:59 +0000 (21:16 -0800)]
iris: scissored and mirrored blits
Kenneth Graunke [Sat, 10 Nov 2018 19:41:22 +0000 (11:41 -0800)]
iris: more TODO
Kenneth Graunke [Mon, 12 Nov 2018 19:33:44 +0000 (11:33 -0800)]
iris: Fix independent alpha blending.
independent_blend_enable means per-RT blending, not RGB != A
Kenneth Graunke [Sat, 10 Nov 2018 10:25:24 +0000 (02:25 -0800)]
iris: "Fix" transfer maps of buffers
x should be in bytes, not cpp units
This generally worked out because PIPE_BUFFER is supposedly required
to be R8_UINT or R8_UNORM. I hear some state trackers pass
PIPE_FORMAT_NONE instead, however, which would make this break.
Just do the right thing directly, to be defensive and clear.
Kenneth Graunke [Sat, 10 Nov 2018 08:15:37 +0000 (00:15 -0800)]
iris: Fix SourceAlphaBlendFactor
Kenneth Graunke [Sat, 10 Nov 2018 08:08:03 +0000 (00:08 -0800)]
iris: leave another TODO
Kenneth Graunke [Sat, 10 Nov 2018 01:35:29 +0000 (17:35 -0800)]
iris: only clip lower if there's something to clip against
Kenneth Graunke [Sat, 10 Nov 2018 01:35:22 +0000 (17:35 -0800)]
iris: fix sysval only binding tables
Kenneth Graunke [Sat, 10 Nov 2018 00:56:29 +0000 (16:56 -0800)]
iris: don't forget to upload CS consts
Kenneth Graunke [Sat, 10 Nov 2018 00:50:19 +0000 (16:50 -0800)]
iris: drop param stuffs
Kenneth Graunke [Fri, 9 Nov 2018 20:49:49 +0000 (12:49 -0800)]
iris: don't trip on param asserts
I'd rather not rewrite i965's compute system value handling right now :(
Kenneth Graunke [Fri, 9 Nov 2018 20:09:50 +0000 (12:09 -0800)]
iris: don't support pull constants.
I don't think it matters, we won't have any params anyway, but let's
be sure it doesn't try
Kenneth Graunke [Fri, 9 Nov 2018 20:06:11 +0000 (12:06 -0800)]
iris: regather info so we get CLIP_DIST slots, not CLIP_VERTEX
Kenneth Graunke [Fri, 9 Nov 2018 19:57:03 +0000 (11:57 -0800)]
iris: enable push constants if we have sysvals but no uniforms
Kenneth Graunke [Fri, 9 Nov 2018 19:56:15 +0000 (11:56 -0800)]
iris: drop iris_setup_push_uniform_range
it doesn't do anything, we have no params. I guess I thought there
would be some, but they all get dead code eliminated even if we try
to make them exist in the first place.
Kenneth Graunke [Fri, 9 Nov 2018 10:40:29 +0000 (02:40 -0800)]
iris: fix more uniform setup
Kenneth Graunke [Fri, 9 Nov 2018 10:20:31 +0000 (02:20 -0800)]
iris: fix num clip plane consts
Kenneth Graunke [Fri, 9 Nov 2018 10:11:16 +0000 (02:11 -0800)]
iris: actually upload clip planes.
Kenneth Graunke [Fri, 9 Nov 2018 10:04:23 +0000 (02:04 -0800)]
iris: bypass params and do it ourselves
the backend keeps dead code eliminating them all, so we can't do that,
plus we don't want to because params[] is lame
Kenneth Graunke [Fri, 9 Nov 2018 09:35:14 +0000 (01:35 -0800)]
iris: dodge backend UCP lowering
Kenneth Graunke [Fri, 9 Nov 2018 09:27:39 +0000 (01:27 -0800)]
iris: fix system value remapping
Kenneth Graunke [Sat, 27 Oct 2018 05:18:56 +0000 (22:18 -0700)]
iris: hook up key stuff for clip plane lowering
Kenneth Graunke [Fri, 9 Nov 2018 09:08:36 +0000 (01:08 -0800)]
iris: lower user clip planes
Kenneth Graunke [Fri, 9 Nov 2018 08:51:58 +0000 (00:51 -0800)]
iris: only bother with params if there are any...
Kenneth Graunke [Fri, 9 Nov 2018 08:41:36 +0000 (00:41 -0800)]
iris: fill out params array with built-ins, like clip planes
Kenneth Graunke [Fri, 9 Nov 2018 07:19:53 +0000 (23:19 -0800)]
iris: add param domain defines
Kenneth Graunke [Fri, 9 Nov 2018 08:04:19 +0000 (00:04 -0800)]
iris: drop unnecessary param[] setup from iris_setup_uniforms
the backend just considers these dead anyway
Kenneth Graunke [Fri, 9 Nov 2018 07:10:46 +0000 (23:10 -0800)]
iris: Defer cbuf0 upload to draw time
Kenneth Graunke [Fri, 9 Nov 2018 05:48:37 +0000 (21:48 -0800)]
iris: Clone the NIR
The backend compiler used to do this for us, but after a rebase, it's
now the driver's responsibility. This lets us alter it for say, clip
vertex lowering, at the global level rather than the per-variant level.
Kenneth Graunke [Thu, 8 Nov 2018 23:32:59 +0000 (15:32 -0800)]
iris: Print the batch name when decoding
Kenneth Graunke [Thu, 8 Nov 2018 09:14:27 +0000 (01:14 -0800)]
iris: partial set_query_active_state
used to avoid OQ during clears for example
fixes occlusion_query_meta_no_fragments
Kenneth Graunke [Thu, 8 Nov 2018 08:15:50 +0000 (00:15 -0800)]
iris: Fix multiple RTs with non-independent blending
rt[i] isn't filled out in this case, so we have to use rt[0]
Kenneth Graunke [Thu, 8 Nov 2018 07:23:30 +0000 (23:23 -0800)]
iris: Fix TextureBarrier
I don't know how I came up with the old one, this is now what i965 does
Also we now do compute batches too
Kenneth Graunke [Thu, 8 Nov 2018 06:54:25 +0000 (22:54 -0800)]
iris: Fix MSAA smooth points
Fixes bin/ext_framebuffer_multisample-point-smooth 2 -auto -fbo
Kenneth Graunke [Thu, 8 Nov 2018 06:05:14 +0000 (22:05 -0800)]
iris: implement scratch space!
we borrow the approach from anv rather than i965, as it works better
with pre-baked state that needs to contain scratch BO addresses
fixes a bunch of varying packing tests
Kenneth Graunke [Thu, 8 Nov 2018 06:07:08 +0000 (22:07 -0800)]
iris: tidy more warnings
Kenneth Graunke [Thu, 8 Nov 2018 05:06:49 +0000 (21:06 -0800)]
iris: Enable msaa_map transfer helpers
This does the downsampling for us. It'll use BLORP anyway because
it uses blit(), and that uses BLORP.
Kenneth Graunke [Wed, 7 Nov 2018 17:35:19 +0000 (09:35 -0800)]
iris: Actually create/destroy HW contexts
The intention is that render and compute use their own contexts,
and each is PIPELINE_SELECT'd to the right pipeline. But we hadn't
actually made them, so we got the fd-default context.
Thanks to Chris Wilson for catching this!
Kenneth Graunke [Wed, 7 Nov 2018 17:56:37 +0000 (09:56 -0800)]
iris: Don't leak the compute batch
Kenneth Graunke [Wed, 7 Nov 2018 05:34:58 +0000 (21:34 -0800)]
iris: cross batch flushing
Kenneth Graunke [Wed, 7 Nov 2018 05:12:30 +0000 (21:12 -0800)]
iris: Cross-link iris_batches so they can potentially flush each other
This makes e.g. the render batch aware of the compute batch, so it can
ask questions like "is this BO referenced by some other batch?" and do
something about that.
Dave Airlie [Wed, 7 Nov 2018 01:49:04 +0000 (11:49 +1000)]
iris: fix crash in sparse vertex array
this fixes crash in array-stride piglit.
Kenneth Graunke [Tue, 6 Nov 2018 07:16:14 +0000 (23:16 -0800)]
iris: Use at least 1x1 size for null FB surface state.
Otherwise we get 0 - 1 = 0xffffffff and fail to pack SURFACE_STATE.
Fixes some object namespace pollution gltexsubimage2d tests
Kenneth Graunke [Sun, 4 Nov 2018 23:20:23 +0000 (15:20 -0800)]
iris: Drop B5G5R5X1 support
This is oddly renderable but not supported for sampling, which is the
opposite of other X formats. Just skip it and fall back to BGRA.
Kenneth Graunke [Sun, 4 Nov 2018 23:00:31 +0000 (15:00 -0800)]
iris: Enable A8/A16_UNORM in an inefficient manner
These are currently just use the 'A' hardware formats, rather than the
faster 'R' formats. glBitmap handling needs these, it seems. :(
Kenneth Graunke [Sun, 4 Nov 2018 22:43:10 +0000 (14:43 -0800)]
iris: Enable ARB_shader_stencil_export
Kenneth Graunke [Thu, 1 Nov 2018 23:11:31 +0000 (16:11 -0700)]
iris: Disable a PIPE_CONTROL workaround on Icelake
Kenneth Graunke [Sat, 3 Nov 2018 13:44:48 +0000 (06:44 -0700)]
iris: Flag constants dirty on program changes
3DSTATE_CONSTANT_* looks at prog_data->ubo_ranges. We were getting
saved by iris_set_constant_buffers() usually happening when changing
programs (as they usually change uniforms too), but with the clear
shader that doesn't use uniforms, we weren't getting one and were
leaving push constants enabled, screwing things up.
Also clean up a bit of a mess left by the hacks - we were missing
bindings in the VS/FS/CS case, among other issues...
Kenneth Graunke [Fri, 2 Nov 2018 21:54:35 +0000 (14:54 -0700)]
iris: allow binding a null vertex buffer
PBO upload apparently does this...
Kenneth Graunke [Sat, 3 Nov 2018 14:24:54 +0000 (07:24 -0700)]
iris: fix overhead regression from "don't stomp each other's dirty bits"
The change from dirty = 0ull to dirty &= ~NOT_MY_BITS broke the "nothing
to do? skip it!" optimization. thanks to Chris for noticing this!
Kenneth Graunke [Sun, 28 Oct 2018 11:45:07 +0000 (04:45 -0700)]
iris: delete dead code
Kenneth Graunke [Sat, 27 Oct 2018 05:43:34 +0000 (22:43 -0700)]
iris: Fix refcounting of grid surface
Jason Ekstrand [Sat, 27 Oct 2018 03:22:04 +0000 (22:22 -0500)]
iris/compute: Zero out the last grid size on indirect dispatches
Jason Ekstrand [Sat, 27 Oct 2018 03:18:41 +0000 (22:18 -0500)]
iris/compute: Don't increment the grid size offset
It may be in the dynamic state buffer but the fact that we have a
resource takes care of that. We don't need to add in the address of
the dynamic state buffer again.
Kenneth Graunke [Wed, 24 Oct 2018 21:45:00 +0000 (14:45 -0700)]
iris: SO_DECL_LIST fix
Kenneth Graunke [Wed, 24 Oct 2018 21:16:38 +0000 (14:16 -0700)]
iris: Fall back to 1x1x1 null surface if no framebuffer supplied
If the state tracker never gave us the framebuffer dimensions via
a set_framebuffer_state() call, just fall back to the unbound texture
null surface, which is 1x1x1. Otherwise we'd use a NULL resource
(no pun intended).
Kenneth Graunke [Tue, 23 Oct 2018 08:36:26 +0000 (01:36 -0700)]
iris: Fix off by one in scissoring, empty scissors, default scissors
Kenneth Graunke [Mon, 22 Oct 2018 22:55:54 +0000 (15:55 -0700)]
iris: Move snapshots_landed to the front.
Transform feedback overflow queries need to write additional data,
and it would be nice to have this field remain at a consistent offset.
Kenneth Graunke [Mon, 22 Oct 2018 21:35:33 +0000 (14:35 -0700)]
iris: Clamp UBO and SSBO access to the actual BO size, for safety
Kenneth Graunke [Mon, 22 Oct 2018 21:28:54 +0000 (14:28 -0700)]
iris: Fix texture buffer / image buffer sizes.
Also fix image buffers with offsets.
Kenneth Graunke [Mon, 22 Oct 2018 21:06:04 +0000 (14:06 -0700)]
iris: fix SF_CLIP_VIEWPORT array indexing with multiple VPs
fixes bunches of viewport stuffs
Kenneth Graunke [Mon, 22 Oct 2018 00:43:32 +0000 (17:43 -0700)]
iris: flag CC_VIEWPORT when changing num viewports
this also has a loop over num_viewports
Kenneth Graunke [Sun, 21 Oct 2018 22:14:02 +0000 (15:14 -0700)]
iris: fix UBOs with bindings that have an offset
Kenneth Graunke [Fri, 19 Oct 2018 09:12:29 +0000 (02:12 -0700)]
iris: try and avoid pointless compute submissions
if apps don't use compute shaders, we don't even want to kick off the
compute initialization batch
Kenneth Graunke [Fri, 19 Oct 2018 09:11:11 +0000 (02:11 -0700)]
iris: fix SBA flushing by refactoring code
Kenneth Graunke [Fri, 19 Oct 2018 09:00:13 +0000 (02:00 -0700)]
iris: do PIPELINE_SELECT for render engine, add flushes, GLK hacks
Kenneth Graunke [Fri, 19 Oct 2018 08:40:06 +0000 (01:40 -0700)]
iris: hack to avoid memorybarriers out the wazoo
we don't want to emit piles of pipe controls to a compute batch if
it isn't necessary...
prevents double-batch-wraps in cs-op-selection-bool-bvec4-bvec4
(but it's still kinda a big ol' hack...)
Kenneth Graunke [Fri, 19 Oct 2018 08:31:56 +0000 (01:31 -0700)]
iris: don't let render/compute contexts stomp each other's dirty bits
only clear what you process
Kenneth Graunke [Fri, 19 Oct 2018 08:29:05 +0000 (01:29 -0700)]
iris: better dirty checking
Kenneth Graunke [Fri, 19 Oct 2018 08:14:38 +0000 (01:14 -0700)]
iris: rewrite grid surface handling
now we only upload a new grid when it's actually changed, which saves us
from having to emit a new binding table every time it changes.
this also moves a bunch of non-gen-specific stuff out of iris_state.c
Kenneth Graunke [Fri, 19 Oct 2018 05:38:32 +0000 (22:38 -0700)]
iris: XXX for compute state tracking :/
Maybe we should just move dirty to batch, it would help with the
reset stuff too
Kenneth Graunke [Fri, 19 Oct 2018 05:32:50 +0000 (22:32 -0700)]
iris: fix whitespace
Kenneth Graunke [Fri, 19 Oct 2018 04:26:06 +0000 (21:26 -0700)]
iris: bail if SLM is needed
Kenneth Graunke [Fri, 19 Oct 2018 04:15:02 +0000 (21:15 -0700)]
iris: leave XXX about unnecessary binding table uploads
Kenneth Graunke [Fri, 19 Oct 2018 03:04:33 +0000 (20:04 -0700)]
iris: drop unnecessary #ifdefs
Kenneth Graunke [Mon, 15 Oct 2018 02:05:31 +0000 (19:05 -0700)]
iris: drop XXX that Jordan handled
Jordan Justen [Fri, 21 Sep 2018 09:28:44 +0000 (02:28 -0700)]
iris/compute: Support indirect compute dispatch
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Jordan Justen [Tue, 18 Sep 2018 23:24:13 +0000 (16:24 -0700)]
iris/compute: Push subgroup-id
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Jordan Justen [Tue, 18 Sep 2018 22:08:44 +0000 (15:08 -0700)]
iris/compute: Flush compute batch on memory-barriers
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Jordan Justen [Sun, 23 Sep 2018 22:25:10 +0000 (15:25 -0700)]
iris/compute: Provide binding table entry for gl_NumWorkGroups
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Jordan Justen [Tue, 18 Sep 2018 22:04:14 +0000 (15:04 -0700)]
iris/compute: Wait on compute batch when mapping
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Jordan Justen [Tue, 18 Sep 2018 21:23:58 +0000 (14:23 -0700)]
iris/program: Don't try to push ubo ranges for compute
We only can push constants for compute shaders from one range.
Gallium glsl-to-nir (src/mesa/state_tracker/st_glsl_to_nir.cpp) lowers
all uniform accesses to a ubo.
Unfortunately we also load the subgroup-id as a uniform in the
compiler. Since we use the 1 push range for this subgroup-id, we then
lose the ability to actually push the ubo with all the normal user
uniform values.
In other words, there is lots of room for performance improvement, but
at least retrieving the uniforms as pull-constants is functional for
now.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Jordan Justen [Tue, 18 Sep 2018 20:04:59 +0000 (13:04 -0700)]
iris/compute: Get group counts from grid->grid
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Jordan Justen [Thu, 13 Sep 2018 18:42:29 +0000 (11:42 -0700)]
iris/compute: Flush compute batches
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Jordan Justen [Sun, 23 Sep 2018 22:09:55 +0000 (15:09 -0700)]
iris/compute: Add MEDIA_STATE_FLUSH following WALKER
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>