| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
|
|
|
| |
Transfer commands can have associated GPU operations.
Enabled by passing GALLIUM_DDEBUG=transfers.
Reviewed-by: Marek Olšák <[email protected]>
|
| |
|
|
| |
Reviewed-by: Marek Olšák <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This patch has multiple goals:
1. Off-load the writing of records in 'always' mode to another thread
for performance.
2. Allow using ddebug with threaded contexts. This really forces us to
move some of the "after_draw" handling into another thread.
3. Simplify the different modes of ddebug, both in the code and in
the user interface, i.e. GALLIUM_DDEBUG. In particular, there's
no 'pipelined' anymore, since we're always pipelined; and 'noflush'
is replaced by 'flush', since we no longer flush by default.
4. Fix the fences in pipelining mode. They previously relied on writes
via pipe_context::clear_buffer. However, on radeonsi, those could
(quite reasonably) end up in the SDMA buffer. So we use the newly
added PIPE_FLUSH_{TOP,BOTTOM}_OF_PIPE fences instead.
5. Improve pipelined mode overall, using the finer grained information
provided by the new fences.
Overall, the result is that pipelined mode should be more useful, and
using ddebug in default mode is much less invasive, in the sense that
it changes the overall driver behavior less (which is kind of crucial
for a driver debugging tool).
An example of the new hang debug output:
Gallium debugger active.
Hang detection timeout is 1000ms.
GPU hang detected, collecting information...
Draw # driver prev BOP TOP BOP dump file
-------------------------------------------------------------
2 YES YES YES NO /home/nha/ddebug_dumps/shader_runner_19919_00000000
3 YES NO YES NO /home/nha/ddebug_dumps/shader_runner_19919_00000001
4 YES NO YES NO /home/nha/ddebug_dumps/shader_runner_19919_00000002
5 YES NO YES NO /home/nha/ddebug_dumps/shader_runner_19919_00000003
Done.
We can see that there were almost certainly 4 draws in flight when
the hang happened: the top-of-pipe fence was signaled for all 4 draws,
the bottom-of-pipe fence for none of them. In virtually all cases,
we'd expect the first draw in the list to be at fault, but due to the
GPU parallelism, it's possible (though highly unlikely) that one of
the later draws causes a component to get stuck in a way that prevents
the earlier draws from making progress as well.
(In the above example, there were actually only 3 draws truly in flight:
the last draw is a blit that waits for the earlier draws; however, its
top-of-pipe fence is emitted before the cache flush and wait, and so
the fact that the draw hasn't truly started yet can only be seen from a
closer inspection of GPU state.)
Acked-by: Marek Olšák <[email protected]>
|
| |
|
|
|
|
|
|
| |
If the last operation happens to be a non-draw, such as a
transfer_map that triggers a decompress blit, there may be
interesting messages left in the driver log.
Reviewed-by: Marek Olšák <[email protected]>
|
| |
|
|
|
|
| |
It is no longer used.
Reviewed-by: Marek Olšák <[email protected]>
|
| |
|
|
| |
Reviewed-by: Marek Olšák <[email protected]>
|
| |
|
|
| |
Reviewed-by: Marek Olšák <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
pipe_draw_info::indexed is replaced with index_size. index_size == 0 means
non-indexed.
Instead of pipe_index_buffer::offset, pipe_draw_info::start is used.
For indexed indirect draws, pipe_draw_info::start is added to the indirect
start. This is the only case when "start" affects indirect draws.
pipe_draw_info::index is a union. Use either index::resource or
index::user depending on the value of pipe_draw_info::has_user_indices.
v2: fixes for nine, svga
|
| |
|
|
| |
For faster initialization of non-indirect draws.
|
| |
|
|
| |
Reviewed-by: Nicolai Hähnle <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
This was made unnecessary with fd33a6bcd7f12.
This was mostly done with:
find ./src -type f -exec sed -i -- \
's:PIPE_THREAD_ROUTINE(\([^,]*\), \([^)]*\)):int\n\1(void \*\2):g' {} \;
With some small manual tidy ups.
Reviewed-by: Plamena Manolova <[email protected]>
Reviewed-by: Marek Olšák <[email protected]>
|
| |
|
|
|
|
|
|
| |
pipe_thread was made unnecessary with fd33a6bcd7f12.
V2: fix compile error in u_queue.c
Reviewed-by: Marek Olšák <[email protected]>
|
| |
|
|
|
|
| |
pipe_mutex was made unnecessary with fd33a6bcd7f12.
Reviewed-by: Marek Olšák <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
For good performance while being able to generate decent hang reports.
The report doesn't contain the parsed IB and the buffer list, but it
isolates the draw call and dumps shaders while not having to flush
the context.
This is for GPU hangs that are harder to reproduce and require interactive
playing for minutes or even hours.
dd_pipe.h explains some implementation details. Initializing, copying
(recording) and clearing states is most of the code.
The performance should be at least 50% of the normal performance depending
on the circumstances. (i.e. 50% is expected to be the worst case scenario,
not the best case) The majority of time is spent in
dump_debug_state(PIPE_DUMP_CURRENT_SHADERS) and that's after all
the optimizations in later patches. There is no obvious way to optimize
that further.
Reviewed-by: Nicolai Hähnle <[email protected]>
|
| |
|
|
| |
Reviewed-by: Nicolai Hähnle <[email protected]>
|
| |
|
|
| |
Reviewed-by: Nicolai Hähnle <[email protected]>
|
| |
|
|
| |
Reviewed-by: Nicolai Hähnle <[email protected]>
|
| |
|
|
|
| |
Reviewed-by: Bas Nieuwenhuizen <[email protected]>
Reviewed-by: Nicolai Hähnle <[email protected]>
|
| |
|
|
|
| |
Reviewed-by: Bas Nieuwenhuizen <[email protected]>
Reviewed-by: Nicolai Hähnle <[email protected]>
|
| |
|
|
|
|
|
|
|
| |
Make them akin to shader buffers, with no refcounting/etc. Just used to
pass data about the bound image in ->set_shader_images.
Signed-off-by: Ilia Mirkin <[email protected]>
Reviewed-by: Dave Airlie <[email protected]>
Reviewed-by: Marek Olšák <[email protected]>
|
| |
|
|
|
|
|
|
| |
This currently just writes out the name of dump files, which can be useful
to easily correlate those files with other log outputs (driver debug output,
apitrace calls, etc.)
Reviewed-by: Marek Olšák <[email protected]>
|
| |
|
|
|
|
|
|
| |
When we know that hangs occur only very late in a reproducible run (e.g.
apitrace), we can save a lot of debugging time by skipping the flush and hang
detection for earlier draw calls.
Reviewed-by: Marek Olšák <[email protected]>
|
| |
|
|
|
|
| |
This will be used by radeonsi for logging.
Reviewed-by: Michel Dänzer <[email protected]>
|
|
|
v2: lots of improvements
This is like identity or trace, but simpler. It doesn't wrap most states.
Run with:
GALLIUM_DDEBUG=1000 [executable]
where "executable" is the app and "1000" is in miliseconds, meaning that
the context will be considered hung if a fence fails to signal in 1000 ms.
If that happens, all shaders, context states, bound resources, draw
parameters, and driver debug information (if any) will be dumped into:
/home/$username/dd_dumps/$processname_$pid_$index.
Note that the context is flushed after every draw/clear/copy/blit operation
and then waited for to find the exact call that hangs.
You can also do:
GALLIUM_DDEBUG=always
to do the dumping after every draw/clear/copy/blit operation without
flushing and waiting.
Examples of driver states that can be dumped are:
- Hardware status registers saying which hw block is busy (hung).
- Disassembled shaders in a human-readable form.
- The last submitted command buffer in a human-readable form.
v2: drop pipe-loader changes, drop SConscript
rename dd.h -> dd_pipe.h
Acked-by: Christian König <[email protected]>
Acked-by: Alex Deucher <[email protected]>
|
