Age | Commit message (Collapse) | Author |
|
Just emit the URB write at END time. Subroutine code that sits after
OPCODE_END won't be executed since we've ended the thread at the point
that the URB write is done.
|
|
MOV, MOV."
This reverts commit 8ef3b1834a896927bdd4f2aea552cdb732849da9. Fixes
piglit glsl-vs-if.
|
|
|
|
This is recommended by the B-Spec. I wasn't able to measure any
difference in ETQW.
|
|
We were patching up all the break and continues between the start of
our loop and the end of our loop, even if they were breaks/continues
for an inner loop. Avoiding patching already patched breaks/continues
fixes piglit glsl-vs-loop-nested.
|
|
|
|
|
|
This gets the VS to the point of accepting vertices. \o/
|
|
This is untested at this point.
|
|
|
|
|
|
The pull constants require sending out to an overworked shared unit
and waiting for a response, while push constants are nicely loaded in
for us at thread dispatch time. By putting things we access in every
VS invocation there, ETQW performance improved by 2.5% +/- 1.6% (n=6).
|
|
The codepaths in the function were almost entirely different.
|
|
|
|
Fixes piglit vp-arl-constant-array-huge-overwritten.
|
|
Saves another 600 bytes or so of code.
|
|
Saves ~480 bytes of code.
|
|
Bug #25628. Fixes piglit case glsl-vs-sqrt-zero.
|
|
Add a GLbitfield64 type and several macros to operate on 64-bit
fields. The OutputsWritten field of gl_program is changed to use that
type. This results in a fair amount of fallout in drivers that use
programs.
No changes are strictly necessary at this point as all bits used are
below the 32-bit boundary. Fairly soon several bits will be added for
clip distances written by a vertex shader. This will cause several
bits used for varyings to be pushed above the 32-bit boundary. This
will affect any drivers that support GLSL.
At this point, only the i965 driver has been modified to support this
eventuality.
I did this as a "squash" merge. There were several places through the
outputswritten64 branch where things were broken. I foresee this
causing difficulties later for bisecting. The history is still
available in the branch.
Conflicts:
src/mesa/drivers/dri/i965/brw_wm.h
|
|
This is a 2.9% (+/-.3%) performance win for my GL demo, which hits MAD
sequences for matrix transforms.
|
|
Fixes piglit vp-sge-alias test, and the googleearth ground shader. \o/
Bug #22228
|
|
Fixes piglit arl.vp.
|
|
Passes piglit glsl-vs-loop testcase.
Bug #20171
|
|
This should help with things like lightsmark, but I don't have a testcase
for this commit.
|
|
|
|
This showed a 1.9% (+/-.3%, n=3) improvement in OA performance with high
geometry settings.
|
|
|
|
Fixes piglit glsl-vs-if-bool and progs/glsl/twoside, and will likely be
useful for the looping code.
Bug #18992
|
|
Previously, we'd be branching based on whatever condition code happened to be
laying around.
|
|
I was getting tired of doing the dance of INTEL_DEBUG=batch, copying it out,
and running intel-gen4disasm on it.
|
|
See comment on Vertex URB Entry Read Length for VS_STATE.
This, combined with the previous three commits, fixes #22945.
|
|
This fix is just from code and docs inspection, but it may fix hangs on
some applications.
|
|
1. new PCI ids
2. fix some 3D commands on new chipset
3. fix send instruction on new chipset
4. new VUE vertex header
5. ff_sync message (added by Zou Nan Hai <nanhai.zou@intel.com>)
6. the offset in JMPI is in unit of 64bits on new chipset
7. new cube map layout
|
|
If we can't fit all the VS outputs into the MRF, we need to overflow into
temporary GRF registers, then use some MOVs and a second brw_urb_WRITE()
instruction to place the overflow vertex results into the URB.
This is hit when a vertex/fragment shader pair has a large number of varying
variables (12 or more).
There's still something broken here, but it seems close...
|
|
|
|
|
|
|
|
|
|
|
|
Conflicts:
src/mesa/drivers/dri/i965/brw_curbe.c
src/mesa/drivers/dri/i965/brw_vs_emit.c
src/mesa/drivers/dri/i965/brw_wm_glsl.c
|
|
Make the use_const_buffer field per-program and only call the code which
updates the constant buffer's data if the flag is set.
This should undo the perf regression from 20f3497e4b6756e330f7b3f54e8acaa1d6c92052
(cherry picked from master, commit dc9705d12d162ba6d087eb762e315de9f97bc456)
|
|
Make the use_const_buffer field per-program and only call the code which
updates the constant buffer's data if the flag is set.
This should undo the perf regression from 20f3497e4b6756e330f7b3f54e8acaa1d6c92052
|
|
|
|
In the VS constants can now be handled in two different ways:
1. If there's room in the GRF, put constants there. They're preloaded from
the CURBE prior to VS execution. This is the historical approach. The
problem is the GRF may not have room for all the shader's constants and
temps and misc registers. Hence...
2. Use a separate constant buffer which is read from using a READ message.
This allows a very large number of constants and frees up GRF regs for
shader temporaries. This is the new approach. May be a little slower
than 1.
1 vs. 2 is chosen according to how many constants and temps the shader needs.
|
|
The READ message's msg_control value can be 0 or 1 to indicate that the
Oword should be read into the lower or upper half of the target register.
It seems that the other half of the register gets clobbered though. So
we read into two dest registers then use a MOV to combine the upper/lower
halves.
|
|
Now that we have real constant buffers, the demands on the CURBE are lessened.
When we use real VS/WM constant buffers we only use the CURBE for clip planes.
|
|
|
|
A scatter-read should be possible, but we're just using two READs for
the time being.
|
|
|
|
Calls to release_tmps() were causing the temps holding constants to get
recycled.
|