| Commit message (Collapse) | Author | Age | Files | Lines |
... | |
|
|
|
|
|
|
|
|
|
|
| |
v2: - adjustments for exercising all important SDMA code paths
- decrease the probability of getting huge sizes (faster testing)
- increase the probability of getting power-of-two dimensions
- change the memory cap to 128MB (faster testing)
- better detect which engine has been used
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
|
|
|
|
|
| |
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
|
|
|
|
|
| |
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
|
|
|
|
| |
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
|
|
|
|
|
|
|
|
|
| |
Changes:
- don't flush DB for fast color clears
- don't flush any caches for initial clears
- remove the flag from si_copy_buffer, always assume shader coherency
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
|
|
|
|
| |
Acked-by: Jose Fonseca <jfonseca@vmware.com>
|
|
|
|
|
|
|
|
|
|
|
|
| |
v2: Use chip_class instead of family.
v3: Check kernel version for SI.
v4: Preemptively allow amdgpu winsys for SI.
Signed-off-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Based on work by Marek Olšák.
v2: Add preamble IB.
Leaves the load packet in the space calculation as the
radeon winsys might not be able to support a premable.
The added space calculation may look expensive, but
is converted to a constant with (at least) -O2 and -O3.
v3: - Fix code style.
- Remove needed space for vertex buffer descriptors.
- Fail when the preamble cannot be created.
Signed-off-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
|
|
|
|
| |
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
|
|
|
|
|
|
| |
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Reviewed-by: Grigori Goronzy <greg@chown.ath.cx>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
|
|
|
|
|
|
|
| |
Only CMASK and DCC clears can use this, because only textures can be so
large.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
|
|
|
|
|
|
|
| |
All of them are paused only between IBs.
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
For radeonsi, native and TGSI use different compilers and this results
in different limits for different IR's.
The set we strictly need for radeonsi is only the MAX_BLOCK_SIZE
and MAX_THREADS_PER_BLOCK params, but I added a few others as shader
related that seemed like they would also typically depend on the
compiler.
Signed-off-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Dave Airlie <airlied@redhat.com>
|
|
|
|
|
|
|
|
|
| |
v2: Polaris chips should be defined after Stoney
Signed-off-by: Sonny Jiang <sonny.jiang@amd.com> (v1)
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com> (v1)
Signed-off-by: Leo Liu <leo.liu@amd.com> (v2 diff)
Reviewed-by: Alex Deucher <alexander.deucher@amd.com> (v2 diff)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Cons:
- it was only integrated in r600g
- it doesn't work with GPUVM
- it records buffer contents at the end of IBs instead of at the beginning,
so the replay isn't exact
- it lacks an IB parser and user-friendliness
A better solution is apitrace in combination with gallium/ddebug, which
has a complete IB parser and can pinpoint hanging CP packets.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Shader stats from VERDE:
Default scheduler:
Totals:
SGPRS: 491272 -> 488672 (-0.53 %)
VGPRS: 289980 -> 311093 (7.28 %)
Code Size: 11091656 -> 11219948 (1.16 %) bytes
LDS: 97 -> 97 (0.00 %) blocks
Scratch: 1732608 -> 2246656 (29.67 %) bytes per wave
Max Waves: 78063 -> 77352 (-0.91 %)
Wait states: 0 -> 0 (0.00 %)
Looking at some of the worst regressions, I get:
- The VGPR increase seems to be caused by the fact that if PS has used less
than 16 VGPRs, now it will always use 16 VGPRs and sometimes even 20.
However, the wave count remains at 10 if VGPRs <= 24, so no harm there.
- The scratch increase seems to be caused by SGPR spilling.
The unnecessary SGPR spilling has been an ongoing issue with the compiler
and it's completely fixable by rematerializing s_loads or reordering
instructions.
SI scheduler:
Totals:
SGPRS: 374848 -> 374576 (-0.07 %)
VGPRS: 284456 -> 307515 (8.11 %)
Code Size: 11433068 -> 11535452 (0.90 %) bytes
LDS: 97 -> 97 (0.00 %) blocks
Scratch: 509952 -> 522240 (2.41 %) bytes per wave
Max Waves: 79456 -> 78217 (-1.56 %)
Wait states: 0 -> 0 (0.00 %)
VGPRs - same story as before. The SI scheduler doesn't spill SGPRs so much
and generally spills way less than the default scheduler.
(522240 spills vs 2246656 spills)
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
|
|
|
|
|
|
| |
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
v2: adjust the comment in the amdgpu winsys
|
|
|
|
|
|
|
| |
v2: don't use DIV_ROUND_UP (no so useful)
also return eviction stats
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
|
|
|
|
| |
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
|
|
|
|
| |
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
|
|
|
|
|
|
|
| |
amdgpu doesn't have to set this, because radeonsi gets it from tile mode
arrays by default.
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
|
|
|
|
| |
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
|
|
|
|
| |
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Set R600_DEBUG=preoptir to dump the LLVM IR before optimization passes,
to allow diagnosing problems caused by optimization passes.
Note that in order to compile the resulting IR with llc, you will first
have to run at least the mem2reg pass, e.g.
opt -mem2reg -S < shader.ll | llc -march=amdgcn -mcpu=bonaire
Signed-off-by: Michel Dänzer <michel.daenzer@amd.com> (original patch)
Signed-off-by: Nicolai Hähnle <nicolai.haehnle@amd.com> (w/ debug flag)
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
|
|
|
|
|
|
|
|
|
|
| |
Add a debug option to select the LLVM SI Machine Scheduler.
R600_DEBUG=sisched
Signed-off-by: Axel Davy <axel.davy@ens.fr>
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
|
|
|
|
|
|
|
|
|
|
|
| |
v2: drop inline keyword
drop radeon_llvm_dispose_kernel_module wrapper
v3: move definitions to .c file
use in radeonsi
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
|
|
|
|
|
|
| |
radeon sets this correctly, but not amdgpu
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
|
|
|
|
| |
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
|
|
|
|
|
| |
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
|
|
|
|
|
| |
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
|
|
|
|
| |
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
|
|
|
|
| |
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
|
|
|
|
|
|
|
|
|
|
| |
The fixed alignment of u_upload_mgr will go away.
This is the first step.
The motivation is that one u_upload_mgr can have multiple users,
each allocating from the same buffer, but requiring a different alignment.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
|
|
|
|
|
|
|
|
| |
"radeon_winsys_cs_handle *cs_buf" is now equivalent to "pb_buffer *buf".
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
|
|
|
|
|
|
| |
v2: fix dual source blending
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
|
|
|
|
|
|
|
|
|
|
| |
Use NULL tests of the form `if (ptr)' or `if (!ptr)'.
They do not depend on the definition of the symbol NULL.
Further, they provide the opportunity for the accidental
assignment, are clear and succinct.
Signed-off-by: Edward O'Callaghan <eocallaghan@alterapraxis.com>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
|
|
|
|
|
| |
Cc: 11.0 11.1 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
|
|
|
|
| |
Tested.
|
|
|
|
|
|
| |
we don't need 2 flags for dumping texture info
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Non-timer queries are suspended during blits. When the blits end, the queries
are resumed, but this resume operation itself might run out of CS space and
trigger a flush. When this happens, we must prevent a duplicate suspend during
preflush suspend, and we must also prevent a duplicate resume when the CS flush
returns back to the original resume operation.
This fixes a regression that was introduced by:
commit 8a125afa6e88a3eeddba8c7fdc1a75c9b99d5489
Author: Nicolai Hähnle <nhaehnle@gmail.com>
Date: Wed Nov 18 18:40:22 2015 +0100
radeon: ensure that timing/profiling queries are suspended on flush
The queries_suspended_for_flush flag is redundant because suspended queries
are not removed from their respective linked list.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reported-by: Axel Davy <axel.davy@ens.fr>
Cc: "11.1" <mesa-stable@lists.freedesktop.org>
Tested-by: Axel Davy <axel.davy@ens.fr>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Expose most of the performance counter groups that are exposed by Catalyst.
Ideally, the driver will work with GPUPerfStudio at some point, but we are not
quite there yet. In any case, this is the reason for grouping multiple
instances of hardware blocks in the way it is implemented.
The counters can also be shown using the Gallium HUD. If one is interested to
see how work is distributed across multiple shader engines, one can set the
environment variable RADEON_PC_SEPARATE_SE=1 to obtain finer-grained performance
counter groups.
Part of the implementation is in radeon because an implementation for
older hardware would largely follow along the same lines, but exposing
a different set of blocks which are programmed slightly differently.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
|
|
|
|
|
|
|
| |
The queries_suspended_for_flush flag is redundant because suspended queries
are not removed from their respective linked list.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
|
|
|
|
| |
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
|
|
|
|
| |
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
|
|
|
|
| |
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
|
|
|
|
| |
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
|
|
|
|
|
|
|
| |
Not needed anymore. A similar flag will be introduced in the next commit,
which will be private in radeonsi.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
|
|
|
|
|
|
|
|
| |
need_cs_space isn't invoked so often and is called before all commands too.
This is a lot cleaner. The code in radeon_add_to_buffer_list always seemed
dodgy to me.
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
|
|
|
|
|
|
|
| |
pipe->flush never returned SDMA fences. This fixes it.
This is only an issue on amdgpu where fences can signal out of order.
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
|
|
|
|
| |
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
|