It now affects the GPU determinism mode as well as some miscellaneous
things that were calling IsNetPlayRunning. Probably incomplete.
Notably, this can change while paused, if the user starts recording a
movie. The movie code appears to have been missing locking between
setting g_playMode and doing other things, which probably had a small
chance of causing crashes or even desynced movies; fix that with
PauseAndLock.
The next commit will add a hidden config variable to override GPU
determinism mode.
It's a relatively big commit (less big with -w), but it's hard to test
any of this separately...
The basic problem is that in netplay or movies, the state of the CPU
must be deterministic, including when the game receives notification
that the GPU has processed FIFO data. Dual core mode notifies the game
whenever the GPU thread actually gets around to doing the work, so it
isn't deterministic. Single core mode is because it notifies the game
'instantly' (after processing the data synchronously), but it's too slow
for many systems and games.
My old dc-netplay branch worked as follows: everything worked as normal
except the state of the CP registers was a lie, and the CPU thread only
delivered results when idle detection triggered (waiting for the GPU if
they weren't ready at that point). Usually, a game is idle iff all the
work for the frame has been done, except for a small amount of work
depending on the GPU result, so neither the CPU or the GPU waiting on
the other affected performance much. However, it's possible that the
game could be waiting for some earlier interrupt, and any of several
games which, for whatever reason, never went into a detectable idle
(even when I tried to improve the detection) would never receive results
at all. (The current method should have better compatibility, but it
also has slightly higher overhead and breaks some other things, so I
want to reimplement this, hopefully with less impact on the code, in the
future.)
With this commit, the basic idea is that the CPU thread acts as if the
work has been done instantly, like single core mode, but actually hands
it off asynchronously to the GPU thread (after backing up some data that
the game might change in memory before it's actually done). Since the
work isn't done, any feedback from the GPU to the CPU, such as real
XFB/EFB copies (virtual are OK), EFB pokes, performance queries, etc. is
broken; but most games work with these options disabled, and there is no
need to try to detect what the CPU thread is doing.
Technically: when the flag g_use_deterministic_gpu_thread (currently
stuck on) is on, the CPU thread calls RunGpu like in single core mode.
This function synchronously copies the data from the FIFO to the
internal video buffer and updates the CP registers, interrupts, etc.
However, instead of the regular ReadDataFromFifo followed by running the
opcode decoder, it runs ReadDataFromFifoOnCPU ->
OpcodeDecoder_Preprocess, which relatively quickly scans through the
FIFO data, detects SetFinish calls etc., which are immediately fired,
and saves certain associated data from memory (e.g. display lists) in
AuxBuffers (a parallel stream to the main FIFO, which is a bit slow at
the moment), before handing the data off to the GPU thread to actually
render. That makes up the bulk of this commit.
In various circumstances, including the aforementioned EFB pokes and
performance queries as well as swap requests (i.e. the end of a frame -
we don't want the CPU potentially pumping out frames too quickly and the
GPU falling behind*), SyncGPU is called to wait for actual completion.
The overhead mainly comes from OpcodeDecoder_Preprocess (which is,
again, synchronous), as well as the actual copying.
Currently, display lists and such are escrowed from main memory even
though they usually won't change over the course of a frame, and
textures are not even though they might, resulting in a small chance of
graphical glitches. When the texture locking (i.e. fault on write) code
lands, I can make this all correct and maybe a little faster.
* This suggests an alternate determinism method of just delaying results
until a short time before the end of each frame. For all I know this
might mostly work - I haven't tried it - but if any significant work
hinges on the competion of render to texture etc., the frame will be
missed.
videoBuffer -> s_video_buffer
size -> s_video_buffer_write_ptr
g_pVideoData -> g_video_buffer_read_ptr (impl moved to Fifo.cpp)
This eradicates the wonderful use of 'size' as a global name, and makes
it clear that s_video_buffer_write_ptr and g_video_buffer_read_ptr are
the two ends of the FIFO buffer s_video_buffer.
Oh, and remove a useless namespace {}.
This state will be used to calculate sizes for skipping over commands on
a separate thread. An alternative to having these state variables would
be to have the preprocessor stash "state as we go" somewhere, but I
think that would be much uglier.
GetVertexSize now takes an extra argument to determine which state to
use, as does FifoCommandRunnable, which calls it. While I'm modifying
FifoCommandRunnable, I also change it to take a buffer and size as
parameters rather than using g_pVideoData, which will also be necessary
later. I also get rid of an unused overload.
VertexLoader::VertexLoader was setting loop_counter, a *static*
variable, to 0. This was nonsensical, but harmless until I started to
run it on a separate thread, where it had a chance of interfering with a
running vertex translator.
Switch to just using a register for the loop counter.
- Lazily create the native vertex format (which involves GL calls) from
RunVertices rather than RefreshLoader itself, freeing the latter to be
run from the CPU thread (hopefully).
- In order to avoid useless allocations while doing so, store the native
format inside the VertexLoader rather than using a cache entry.
- Wrap the s_vertex_loader_map in a lock, for similar reasons.
To avoid FPRs being pushed unnecessarily, I checked the uses: DSPEmitter
doesn't use FPRs, and VertexLoader doesn't use anything but RAX, so I
specified the register list accordingly. The regular JIT, however, does
use FPRs, and as far as I can tell, it was incorrect not to save them in
the outer routine. Since the dispatcher loop is only exited when
pausing or stopping, this should have no noticeable performance impact.
- Factor common work into a helper function.
- Replace confusingly named "noProlog" with "rsp_alignment". Now that
x86 is not supported, we can just specify it explicitly as 8 for
clarity.
- Add the option to include more frame size, which I'll need later.
- Revert a change by magumagu in March which replaced MOVAPD with MOVUPD
on account of 32-bit Windows, since it's no longer supported. True,
apparently recent processors don't execute the former any faster if the
pointer is, in fact, aligned, but there's no point using MOVUPD for
something that's guaranteed to be aligned...
(I discovered that GenFrsqrte and GenFres were incorrectly passing false
to noProlog - they were, in fact, functions without prologs, the
original meaning of the parameter - which caused the previous change to
break. This is now fixed.)
For a long time, we've had ugly and inconsistent function names here as
helpers, names like "decodebytesRGB5A3rgba" which are absolutely
incomprehensible to understand. Fix this by introducing a new consistent
naming scheme, where the above function now becomes "DecodeBytes_RGB5A3".
Instead of having three separate functions and checking the tlutfmt in a
variety of places, just do it once in a helper method. This is already
for the slow path either in our Generic decoder or in our Software
renderer, so it doesn't matter that this is slower.
x64 will continue using the separate functions for speed.
The D3D / OGL backends only ever used RGBA textures, and the Software
backend uses its own custom code for sampling. The ARGB path seems to
just be dead code.
Since ARGB and RGBA formats are similar, I don't think this will make
the code more difficult to read or unable to be used as
reference. Somebody who wants to use this code to output ARGB can simply
modify the MakeRGBA function to put the shift at the other end.
This pulls all the duplicate code from TextureDecoder_Generic /
TextureDecoder_x64 out and puts it in a common file. Out custom font
used for debugging the texture cache is also pulled out and put in a
common "sfont.inc" file. At some point we should also combine this font
with the other six binary fonts we ship.
GetPC_TexFormat was never used. It was added in commit d02426a, with the
only user being commented out code. The commented out code was later
removed in 9893122, but the implementation stayed.
We were decoding to BGRA32 textures in our RGBA32 texture decoder. Since
this is the same for the BGRA32 decoder implementation, this is most
likely a copy/paste typo, rather than the texture actually being
bit-swapped. Fix this.
I'm not sure of any games that use the C14X2 texture format, so I'm not
sure this fixes any games, but it does make the code cleaner for when we
clean it up in the future, and merge some of these similar loops.
Separated out from my gpu-determinism branch by request. It's not a big
commit; I just like to write long commit messages.
The main reason to kill it is hopefully a slight performance improvement
from avoiding the double switch (especially in single core mode);
however, this also improves cycle calculation, as described below.
- FifoCommandRunnable is removed; in its stead, Decode returns the
number of cycles (which only matters for "sync" GPU mode), or 0 if there
was not enough data, and is also responsible for unknown opcode alerts.
Decode and DecodeSemiNop are almost identical, so the latter is replaced
with a skipped_frame parameter to Decode. Doesn't mean we can't improve
skipped_frame mode to do less work; if, at such a point, branching on it
has too much overhead (it certainly won't now), it can always be changed
to a template parameter.
- FifoCommandRunnable used a fixed, large cycle count for display lists,
regardless of the contents. Presumably the actual hardware's processing
time is mostly the processing time of whatever commands are in the list,
and with this change InterpretDisplayList can just return the list's
cycle count to be added to the total. (Since the calculation for this
is part of Decode, it didn't seem easy to split this change up.)
To facilitate this, Decode also gains an explicit 'end' parameter in
lieu of FifoCommandRunnable's call to GetVideoBufferEndPtr, which can
point to there or to the end of a display list (or elsewhere in
gpu-determinism, but that's another story). Also, as a small
optimization, InterpretDisplayList now calls OpcodeDecoder_Run rather
than having its own Decode loop, to allow Decode to be inlined (haven't
checked whether this actually happens though).
skipped_frame mode still does not traverse display lists and uses the
old fake value of 45 cycles. degasus has suggested that this hack is
not essential for performance and can be removed, but I want to separate
any potential performance impact of that from this commit.
This is required to make packing consistent between compilers: with u32, MSVC
would not allocate a bitfield that spans two u32s (it would leave a "hole").
The only possible functionality change is that s_efbAccessRequested and
s_swapRequested are no longer reset at init and shutdown of the OGL
backend (only; this is the only interaction any files other than
MainBase.cpp have with them). I am fairly certain this was entirely
vestigial.
Possible performance implications: efbAccessReady now uses an Event
rather than spinning, which might be slightly slower, but considering
the slow loop the flags are being checked in from the GPU thread, I
doubt it's noticeable.
Also, this uses sequentially consistent rather than release/acquire
memory order, which might be slightly slower, especially on ARM...
something to improve in Event/Flag, really.
This shouldn't affect functionality. I'm not sure if the breakpoint
distinction is actually necessary (my commit messages from the old
dc-netplay last year claim that breakpoints are broken anyway, but I
don't remember why), but I don't actually need to change this part of
the code (yet), so I'll stick with the trimmings change for now.
This is effectively unused, as the window handles that we pass to the
GLInterface are window handles for the frame which isn't ever a real
toplevel window. Host_UpdateTitle is what actually sets the proper title
on the render window.
Now that MainNoGUI is properly architected and GLX doesn't need to
sometimes craft its own windows sometimes which we have to thread back
into MainNoGUI, we don't need to thread the window handle that GLX
creates at all.
This removes the reference to pass back here, and the g_pWindowHandle
always be the same as the window returned by Host_GetRenderHandle().
A future cleanup could remove g_pWindowHandle entirely.
The framebuffer is no longer rotated the wrong way around in Qualcomm's latest development drivers.
They did something right, only took them over a year.
This catches most instances of configuration failures that can happen in a post processing shader.
Gives a user a helpful error message that lets them know what they have failed to set up correctly
This class loads all the common PP shader configuration options and passes those options through to a inherited class that OpenGL or D3D will have.
Makes it so all the common code for PP shaders is in VideoCommon instead of duplicating the code across each backend.
Fixes a bug where "Use Fullscreen" would initialize into exclusive fullscreen regardless of the borderless fullscreen setting.
Also relieves the need for the video renderer to check the borderless fullscreen setting each time.
The hack was needed because the Nvidia 3D Vision heuristics are documented to only support surfaces that are the same size as the backbuffer. This would be the case if you enabled the hack and selected the "Auto (Window Size)" internal resolution.
However, on recent drivers the same effect is achieved by selecting the "Auto (Multiple)" internal resolution. Therefore the hack is no longer required.
Also have the renderer remember its own fullscreen state. This is done to prevent a case where we exit exclusive fullscreen through the configuration and a focus shift at the same time. In this case the renderer would fail to detect that the fullscreen state was changed.
In the cases where we support the binding layout keyword, use it for more than binding UBO location.
This changes it so it is supported for samplers as well.
Instances when this is enabled is if a device supports GL_ARB_shading_language_420pack, or if it supports GLES 3.10.
ffmpeg 2.0 changed requirements for the FFV1 encoder and made them more strict,
requiring more fields of the input frame to be initialized. Explicitly setting
pixfmt, width and height solve the EINVAL issues with FFV1 encoding.
Original fix from http://ffmpeg.org/pipermail/libav-user/2013-October/005759.html
We are used to have a 1:1 mapping of GX vertex formats and the native (OGL + D3D) ones, but there are by far more GX ones.
This new cache maps them directly so that we don't flush on GX vertex format changes as long as the native one doesn't change.
The idea is stolen from galop1n.
- Isolate it into it's own namespace
- Shorten function names, the namespace self-documents.
- Just use the std I/O, we can just write directly to the stream for
logging.
Some headers where using #ifndef to guard being including multiple times. But most were using pragma once. So for consistency I changed them all to use pragma once.
When I dropped ARM from a generic target, this caused the vertexloader to try using the JIT path.
Instead of !_M_GENERIC, check for _M_X86 instead. Since it is only for the x86 target
These were only compiled in on Windows and x86_32.
They provided "optimized" copies and compares based on blocksizes for the AMD Athlon and Duron CPU families.
The code was taken from something that AMD provides with a as-is license.
Just get rid of this crap.
For whatever reason, the hardware doesn't do a full divide by 255, but
instead uses an approximation with shifting, similar to the way it is done
in TEV.
Noticed this while messing with EFB to RAM.
We were having an implicit conversion from integer to float, GLSL ES doesn't allow implicit conversion.
Changes it to a explicit conversion to float.
This matches how ARM handles their naming in their drivers for different models.
Really it's that way because both Mali-T6xx and Mali-T7xx fall under Midgard.
While everything else (except Mali-55) fall under Utgard.
This was a mistake of mine when translating floating point values to integer values.
Also, the max() part of that line was just completely redundant because the sign of an absolute value is always greater than or equal to zero.
Fixes issue 7178.
They are similar enough that they will share bugs with their drivers, so make them fall under the same Mali-Txxx umbrella of bug issues.
If there is ever a need in the future for having separate bugs depending on family, we can support that then.
Fixes issue 6990.
This uses a bit of templating to remove the duplicate code that is the CodeBlocks in each emitter headers.
No actual functionality change in this.
GLSL ES 3.10 adds implicit support for the binding layout qualifier that we use.
Changes our GLSL version enums to bit values so we can check for both ES versions easily.
This is undefined behavior in C++, and a clang warning suggests it is
actually producing bad code as a result:
../Source/Core/VideoCommon/BPFunctions.cpp:164:45: warning: comparison of constant 4294967295 with expression of type 'PEControl::PixelFormat' is always false [-Wtautological-constant-out-of-range-compare]
if (new_format == old_format || old_format == (unsigned int)-1)
Fixes all warnings on Android build except for what is in externals.
Removes a function from TextureDecoder_Generic since it is unused and generates a warning.
Metroid: Other M was the only game which required this field, but the
issue in that game can be fixed properly by enabling format change
emulation. Hence, there's no point in having this around anymore.
Fixes issue 6644.
Our defines were never clear between what meant 64bit or x86_64
This makes a clear cut between bitness and architecture.
This commit also has the side effect of bringing up aarch64 compiling support.
- remove unused variables
- reduce the scope where it makes sense
- correct limits (did you know that strcat()'s last parameter does not
include the \0 that is always added?)
- set some free()'d pointers to NULL
inline halfxb
So we know which is the first pixel by masking.
inline xl
inline xb a bit
inline yl
inline uv1.x shift
remove likely wrong guessed ternary operator
add pixel layout comment
inline xel
optimize the shifts a bit
inline xb
optimize shifts in a second step
extract xb
rename all variables
calculate cache line by position.x
Revert 5115b459f40d53044cd7a858f52e6e876e1211b4 "optimize the shifts a bit"
It seems I was wrong, the other way is the more natural.
use x_virtual_position instead of uv1.x for x_offset_in_block
This looks more natural and the offset should be masked anyway.
substitude factor with cache_lines
move 32bit logic in a conditional block
This option was known to break every second game and only boost a bit.
It also seems to be broken because of streaming into pinned memory and buffer storage buffers.
v2: also remove dlc_desc
Older Qualcomm drivers rotated the framebuffer 90 degrees and this fix didn't work.
Now for some obscene reason it rotates a full 180 degrees.
This can at least be worked around by flipping around the image on our end.
On Windows, nvidia don't give us their driver version, so we can't workaround any issues.
As buffer_storage is broken on some drivers, we wanted to disble it for them.
So we can't.
Luckyly only "some" released driver versions are affected as this extension is only available since some months. Let's hope that nobody have to use one of this driver version, else they will get a black screen ...
OSX has their own driver, so performance issues aren't shared with the nvidia driver (unlike the closed source linux and windows nvidia driver). So now they'll also use the MapAndSync backend like all other osx drivers.
fixes issue 6596
I've also cleaned up the if/else block selecting the best backend a bit.
The new chapter title in Paper Mario TTYD had a small graphical bug due to the new code because it read one extra pixel, this fixes it.
I hope this gets everything, I though I had checked most bugs and yet here I am, commit-spamming...
Fixes some remaining bbox related bugs in Mickey's Magical Mirror and a slight graphical glitch in Paper Mario: TTYD when flipping and Vivian as your companion (I've been scratching my head for days to find this one).
Instead of being vertex-based, it is now primitive (point, line or dissected triangle) based, with proper clipping.
Also, screen position is now calculated based on viewport values, instead of "guesstimating".
This fixes many graphical glitches in Paper Mario: TTYD and Super Paper Mario.
Also, the new code allows Mickey's Magical Mirror and Disney's Hide & Sneak to work (mostly) bug-free. I changed their inis to use bbox.
These changes have a slight cost in performance when bbox is being used (rare), mostly due to the new clipping algorithm.
Please check for any regressions or crashes.
This branch is the final step of fully supporting both OpenGL and OpenGL ES in the same binary.
This of course only applies to EGL and won't work for GLX/AGL/WGL since they don't really support GL ES.
The changes here actually aren't too terrible, basically change every #ifdef USE_GLES to a runtime check.
This adds a DetectMode() function to the EGL context backend.
EGL will iterate through each of the configs and check for GL, GLES3_KHR, and GLES2 bits
After that it'll change the mode from _DETECT to whichever one is the best supported.
After that point we'll just create a context with the mode that was detected
As we do lots of writes to *Iptr, the compiler isn't allowed to cache any shared variable (neither index nor Iptr itself).
This commit inlines Iptr + index into the index generator functions, so the compiler know that they are const.
We are used to render them out of order as long as everything else matches, but rendering order does matter, so we have to flush on primitive switch. This commit implements this flush.
Also as we flush on primitive switch, we don't have to create three different index buffers. All indices are now stored in one buffer.
This will slow down games which switch often primitive types (eg ztp), but it should be more accurate.
This "u32 components" is a list of flags which attributes of the vertex loader are present.
We are used to append this variable to lots of vertex generation functions, but some of them don't need it at all.
The usual way to handle this kind of request is to rise a flag which the gpu thread polls.
The gpu thread itself either generates the result or just write zeros if disabled.
After this, it rise another flag which says that this work is done.
So if disabled, we still have the cpu-gpu round trip time. This commit just returns 0 on the cpu thread
instead of playing ping pong...
Some information on this bug since this isn't quite true.
Seemingly with the v53 driver, Qualcomm has actually fixed this bug. So we can dynamically access UBO array members.
The issue that is cropping up is actually converting our attribute 'fposmtx' to an integer.
int posmtx = int(fpostmtx);
This line causes some seemingly garbage values to enter in to the posmtx variable.
Not sure exactly why it is failing, probably them just not actually converting the float to an integer and just handling the float directly as a integer.
So the bug is going to stay active with Qualcomm devices until we convert this vertex attribute from a float to a integer.
Let's talk a bit about this bug. 12nd oldest bug not fixed in Dolphin, it was a
lot of fun to debug and it kept me busy for a while :)
Shoutout to Nintendo for framework.map, without which this could have taken a
lot longer.
Basic debugging using apitrace shows that the heat effect is rendered in an
interesting way:
* An EFB copy texture is created, using the hardware scaler to divide the
texture resolution by two and that way create the blur effect.
* This texture is then warped using indirect texturing: a deformation map is
used to "move" the texture coordinates used to sample the framebuffer copy.
Pixel shader: http://pastie.org/private/25oe1pqn6s0h5yieks1jfw
Interestingly, when looking at apitrace, the deformation texture was only 4x4
pixels... weird. It also does not have any feature that you would expect from a
deformation map. Seeing how the heat effect glitches, this deformation texture
being wrong looks like a good candidate for the problem. Let's see how it's
loaded!
By NOPing random calls to GXSetTevIndirect, we find a call that when removed
breaks the effect completely. The parameters used for this call come from the
results of methods of JPAExTexShapeArc objects. 3 different objects go through
this code path, by breaking each one we can notice that the one "controlling"
the heat effect is the one at 0x81575b98.
Following the path of this object a bit more, we can see that it has a method
called "getIndTexId". When this is called, the returned texture ID is used to
index a map and get a JPATextureArc object stored at 0x81577bec.
Nice feature of JPATextureArc: they have a getName method. For this object, it
returns "AK_kagerouInd01". We can probably use that to see how this texture
should look like, by loading it "manually" from the Wind Waker DVD.
Unfortunately I don't know how to do that. Fortunately @Abahbob got me the
texture I wanted in less than 10min after I asked him on Twitter.
AK_kagerouInd01 is a 32x32 texture that really looks like a deformation map:
http://i.imgur.com/0TfZEVj.png . Fun fact: "kagerou" means "heat haze" in JP.
So apparently we're not using the right texture object when rendering! The
GXTexObj that maps to the JPATextureArc is at offset 0x81577bf0 and points to
data at 0x80ed0460, but we're loading texture data from 0x0039d860 instead.
I started to suspect the BP write that loads the texture parameters "did not
work" somehow. Logged that and yes: nothing gets loaded to texture stage 1! ...
but it turns out this is normal, the deformation map is loaded to texture stage
5 (hardcoded in the DOL). Wait, why is the TextureCache trying to load from
texture stage 1 then?!
Because someone sucked at hex.
Fixes issue 2338.
At the moment, custom textures with:
- invalid mipmap size
- invalid aspect ratio
- non-fractional scaling factors
are allowed. But they can't be loaded fine by the backend, so generate a warning if someone trys to load them.
fixes issue 6898
OpenGL defaults are GL_REPEAT, which is even more unlikely than GL_CLAMP_TO_EDGE.
As I can't test the behavoir of the real hardware, I changed it to how it works before,
but I guess just clip the texture makes more sense.
This flag wasn't cleared at all, so we set our constants dirty every time...
This could fix some performance regressions because of revision 6798a4763e
This removes the redundant code and also implements this feature for OSX and Wayland.
But so it's dropped for non-wx builds...
imo DolphinWX still isn't the best place for this, but now it's in the same file as all other hotkeys. Maybe they'll be moved to InputCommon sometimes at once ...
comex
