The calculation of each address in lmw/stmw currently has a dependency
on the calculation of the previous address. By removing this dependency,
the host CPU should be able to pipeline the loads/stores better. The cost
we pay for this is up to one extra register and one extra MOV instruction
per guest instruction, but often nothing.
Making EmitBackpatchRoutine support using any register as the address
register would let us get rid of the MOV, but I consider that to be too
big of a task to do in one go at the same time as this.
Now that we've flipped the C++20 switch, let's start making use of
the nice new <bit> header.
I'm planning on handling this move away from BitUtils.h incrementally
in a series of PRs. There may be a few functions remaining in
BitUtils.h by the end that C++20 doesn't have any equivalents for.
This reverts commit 351d095fff.
In hindsight, my attempted optimization messes with the return
predictor, unlike real tail calls. So I think it does more bad than
good.
The "vector shift by immediate" category encodes the shift amount for
right shifts as `size - amount`, whereas left shifts use `amount`.
We're not actually using SHRN/SHRN2 anywhere, which is why this has gone
undetected.
This creates a new benchmark module that is responsible for generating baseline profiles and testing them. As part of this commit a baseline-prof.txt file has been included to speed up launch times with the app in its current state. Later, profile generation can be automated and keep up with the app as it changes.
The sliding panel layout was in the way of the top app bar and prevented the up button from being pressed. This anchors the panel to the bottom of the top app bar and removed unnecessary hardcoded padding.
Use: callstack(0x80000000).
!callstack(value) works as a 'does not contain'.
Add strings to expr.h conditionals.
Use quotations: callstack("anim") to check symbols/name.
For quite some time now, we've had a setting on x86-64 that makes Dolphin
handle NaNs in a more accurate but slower way. There's only one game that
cares about this, Dragon Ball: Revenge of King Piccolo, and what that game
cares about more specifically is that the default NaN (or "generated NaN"
as I believe it's called in PowerPC documentation) is the same as on
PowerPC. On ARM, the default NaN is the same as on PowerPC, so for the
longest time we didn't need to do anything special to get Dragon Ball:
Revenge of King Piccolo working. However, in 93e636a I changed how we
handle FMA instructions in a way that resulted in the sign of NaNs
becoming inverted for nmadd/nmsub instructions, breaking the game.
To fix this, let's implement the AccurateNaNs setting, like on x86-64.
This affected the memory and registers widgets (and possibly others). I'm pretty sure it regressed in 5f629abd8b.
The SetCodeVisible line is a new fix, but the equivalent already existed in the memory widget.
The call to analyzer.Analyze breaks when it attempts to read an instruction, as it eventually tries to read memory when Memory::m_pRAM is nullptr. Trying to read when execution is not paused in general seems like a bad idea (especially as analyzer.Analyze uses PowerPC::TryReadInstruction which can update icache - this is probably still a problem).
Operations that have two operands and can't generate a default NaN,
i.e. addition and subtraction, already have the desired NaN handling
on x86. We just need to make sure to not reverse the operands.
This fixes ps_sum0/ps_sum1 outputting NaNs in cases where they shouldn't.
(HandleNaNs assumes that a NaN in a ps0 input always results in a NaN in
the ps0 output, and correspondingly for ps1.)
1. In some cases, ps_merge01 can be implemented using one instruction.
2. When we need two instructions for ps_merge01, it's best to start with
a MOV to avoid false dependencies on the destination register.
3. ps_merge10 can be implemented using a single EXT instruction.
This regressed in 0a906f553f, I think (though I haven't confirmed it). Mario Tennis and Luigi's Mansion both use these for some reason (as far as I can tell, the data isn't actually used; it's just extra data included for no reason)
DataReader is generally jank - it has a start and end pointer, but the end pointer is generally not used, and all of the vertex loaders mostly bypassed it anyways.
Wrapper code (the vertex loaer test, as well as Fifo.cpp and OpcodeDecoding.cpp) still uses it, as does the software vertex loader (which is not a subclass of VertexLoader). These can probably be eliminated later.
Previously the app would crash because Material Dividers inherit from AppCompat and the leanback activity does not. This creates a new layout file with leanback-specific accommodations and code is duplicated in GameDetailsDialog to prevent inflation crashes.
This new function is like MOVP2R, except it masks out the lower 12 bits,
returning them instead of writing them to the register. These lower
12 bits can then be used as an offset for LDR/STR. This lets us turn
ADRP+ADD+LDR sequences with a zero offset into ADRP+LDR sequences with
a non-zero offset, saving one instruction.
When emulated GBAs were added to Dolphin, it was possible to control them
using the GC TAS input window. (Z was mapped to Select.) Unaware of this,
I broke the functionality in b296248.
To make it possible to control emulated GBAs using TAS input again,
I'm adding a proper TAS input window for GBAs, with a real Select button
and no analog controls.
0e02ddcf52 removed separate logic for tiled versus non-tiled EFB peek caches, and as part of that made it so that color peeks updated the frame access mask even when a non-tiled cache is in use. However, the same change was not made for depth peeks. I'm not sure if this affected anything in practice.
In order to avoid getting stuck making a new dimension file every time a new device is found we take a known value for how large the game card will be, take the screen size, and adjust the grid accordingly.
Sometimes a gecko code would have a title long enough to appear over the checkbox. This is now prevented by marking the text's boundary a 16dp before the start of the checkbox.
`ImGui::GetIO` performs an assertion that a context exists, and if one doesn't then things will likely crash. Unfortunately this crash is hard to consistently reproduce.
When inflating this layout, the layout inflater doesn't expect a View and rather a descendant of ViewGroup. This resulted in a crash which is resolved by using a FrameLayout instead.
I recently talked to a homebrew developer who was trying to add exception
handlers at link time but found out that Dolphin was overwriting their
exception handlers. I figure that's not the usual way to do exception
handlers, but... making us load the executable after setting up memory
rather than before is easy, and matches what we do when booting discs,
so I suppose there's no reason not to do it. It also matches the intent
of why Dolphin is writing default exception handlers – we're writing
them because some homebrew relies on exception handlers being left
around from whatever program was running before it (see 3dd777be70).
Let's take advantage of ARM64's input register shifting one last time,
shall we?
Before:
0x1280005b mov w27, #-0x3
0x1b1b7f18 mul w24, w24, w27
After:
0x4b180b18 sub w24, w24, w24, lsl #2
ARM64's flexible shifting of input registers also allows us to calculate
a negative power of two in one instruction; shift the input of a NEG
instruction.
Before:
0x128001f7 mov w23, #-0x10
0x1b1a7efa mul w26, w23, w26
0x93407f58 sxtw x24, w26
After:
0x4b1a13fa neg w26, w26, lsl #4
0x93407f58 sxtw x24, w26
If the destination register doesn't equal the input register, using it
to temporarily hold the immediate value is fair game as it'll be
overwritten with the result of the multiplication anyway. This can
slightly reduce register pressure.
Before:
0x52800659 mov w25, #0x32
0x1b197f5b mul w27, w26, w25
After:
0x5280065b mov w27, #0x32
0x1b1b7f5b mul w27, w26, w27
By taking advantage of ARM64's ability to shift an input register by any
amount, we can calculate multiplication by a number that is one more
than a power of two with a single instruction.
Before:
0x52800838 mov w24, #0x41
0x1b187f7b mul w27, w27, w24
After:
0x0b1b1b7b add w27, w27, w27, lsl #6
Turn multiplications by a power of two into bitshifts.
Before:
0x52800817 mov w23, #0x40
0x1b167ef6 mul w22, w23, w22
After:
0x531a66d6 lsl w22, w22, #6
Multiplication by one is also trivial. Depending on the registers
involved, either a single MOV or no instructions will be generated.
Before:
0x52800038 mov w24, #0x1
0x1b1a7f1b mul w27, w24, w26
After:
0x2a1a03fb mov w27, w26
Before:
0x52800039 mov w25, #0x1
0x1b1a7f3a mul w26, w25, w26
After:
Nothing!
Add a new function that will handle all the special cases regarding
multiplication. It does nothing for now, but will be expanded in
follow-up commits.
We can merge an SXTW with the SUB, eliminating one instruction. In
addition, it is no longer necessary to allocate a temporary register,
reducing register pressure.
Before:
0x93407f59 sxtw x25, w26
0x93407ebb sxtw x27, w21
0xcb1b033b sub x27, x25, x27
After:
0x93407f5b sxtw x27, w26
0xcb35c37b sub x27, x27, w21, sxtw
ARM64 can do perform various types of sign and zero extension on a
register value before using it. The Arm64Emitter already had support for
this, but it was kinda hidden away.
This commit exposes the functionality by making the ExtendSpecifier enum
available everywhere and adding a new ArithOption constructor.
[ VUID-VkDescriptorPoolCreateInfo-maxSets-00301 ] Object 0:
handle = 0x7f1,b8d,3cd,e70, type = VK_OBJECT_TYPE_DEVICE; |
MessageID = 0xa1,70e,236 | vkCreateDescriptorPool():
pCreateInfo->maxSets is not greater than 0.
The Vulkan spec states: maxSets must be greater than 0
BindFramebuffer depends on the pipeline which might not be set yet.
That's why the framebuffer dirty flag exists in the first place.
I assume BindFramebuffer was called directly here, in order to handle
the texture state transitions necessary for DiscardResource.
The state is tracked anyway, so we can just issue those transitions there
too and defer binding the actual framebuffer.
Fixes an issue in Zelda Twilight Princess with EFB depth peeks.
Dolphin would bind a frame buffer which doesn't have an integer format
descriptor for the color target before binding the new pipeline.
So it would accidentally use the 0 descriptor.
Debug layer error:
D3D12 ERROR: ID3D12CommandList::OMSetRenderTargets:
Specified CPU descriptor handle ptr=0x0000000000000000 does not refer to
a location in a descriptor heap. pRenderTargetDescriptors[0] is the issue.
[ EXECUTION ERROR #646: INVALID_DESCRIPTOR_HANDLE]
Fixes the following error in the D3D12 debug layer:
D3D12 WARNING: ID3D12Device::CreateCommittedResource:
Ignoring InitialState D3D12_RESOURCE_STATE_UNORDERED_ACCESS.
Buffers are effectively created in state D3D12_RESOURCE_STATE_COMMON.
[ STATE_CREATION WARNING #1328: CREATERESOURCE_STATE_IGNORED]
Fixes the following error in the D3D12 debug layer:
D3D12 ERROR: ID3D12DescriptorHeap::GetGPUDescriptorHandleForHeapStart:
GetGPUDescriptorHandleForHeapStart is invalid to call on a descriptor
heap that does not have DESCRIPTOR_HEAP_FLAG_SHADER_VISIBLE set.
If the heap is not supposed to be shader visible, then
GetCPUDescriptorHandleForHeapStart would be the appropriate method
to call. That call is valid both for shader visible and non shader
visible descriptor heaps.
[ STATE_GETTING ERROR #1315: DESCRIPTOR_HEAP_NOT_SHADER_VISIBLE]
When searching for a disc where the revision doesn't match any disc in
the datfile, the loop would never get to the part where serials_exist is
set to true, leading to a bogus error message.
Because of the previous commit, `regs_in_use` must not include `dest_reg`
when calling MMIOLoadToReg. There are also some other registers we can
skip including in regs_in_use just for efficiency's sake.
The `addr_reg_set = false` statements that I've added in this commit are
technically redundant – if `mmio_address` is non-zero then `addr_reg_set`
is already false – but it's just a coincidence that that's the case.
I originally added these in 2b1d1038a6, for both the TPipelineFunction and the size. The size was moved into the header in fdcd2b7d00 (making the size functions obsolete), but it seems that the functions themselves are no longer needed now.
I think I didn't use this approach before because it would have required ComponentFormatTable and ComponentCountRow to be templated, which would end up resulting in lines that were too long and thus wrapped in awkward places. (I *think* they didn't get inferred properly.) Now that we only need TPipelineFunction, the templating is not needed, and this ends up being a more readable version of the version with the wrapper functions.
This adds PR 10890's new setting to the Android GUI. I'm curious to see
if any Android users might get a performance improvement from it.
Due to how our settings work on Android, I haven't implemented disabling
the checkbox when the graphics backend doesn't support both GS and VS
for point/line expansion, but I don't think that's critical to have.
The old calculation was stride * (max_index + 1), which fails if stride is less than the size of a component (for instance, if float XYZ positions are used, and the stride was set to 4 (i.e. sizeof(float)) instead of 12 (i.e. 3 * sizeof(float)), it would be missing the last 8 bytes of the final element in the array. Or, if stride was set to 0, then no bytes would be recorded at all (though that's not a useful configuration so it's unlikely to actually exist).
I'm not aware of any games affected by this issue.
This should fix recording the wall in the staircase leading to the basement in Luigi's Mansion (though I haven't tested it, as I don't own a copy of Luigi's Mansion). This uses NormalIndex3, and the index for the normal vector (generally 0x02XX or 0x01XX) there is always lower than the tangent or binormal (generally 0x07XX). Other games seem to usually have a similar range of indices for the normal, tangent, and binormal, so this issue wouldn't affect them.
In most cases, games will use the same type for all vertex components (either Index8 or Index16 or Direct). However, RS2's deflection towers use Index16 for the texture coordinate and Index8 for everything else, meaning the texture coordinates were recorded incorrectly (the first byte was used, so only indices 0 and 1 were recorded instead of 0 through 0x0192). Worse still, some background elements in RS2 use direct positions but indexed normals or texture coordinates, and those would not be recorded at all.
This is a regression from b5fd35f951.
`count` is the number of stereo samples to write (where each stereo sample is two shorts), while `BUFFER_SIZE` is the size of the buffer in shorts. So `count` needs to be multiplied by `2`, not `BUFFER_SIZE`. Also, when this check was failed, the previous code just clobbered whatever was past the end of the buffer after logging the warning, which corrupted `basename`, eventually resulting in Dolphin crashing.
This affected Datel's Wii-compatible Action Replay, which uses a block size of 2298, or 18384 stereo samples, which is 36768 shorts, which is bigger than the buffer size of 32768. (However, the previous commit means that only one block is transfered at a time, eliminating this issue; fixing the bounds check is just a general safety thing instead of an actual bugfix now.)
The previous implementation of Force25BitPrecision was essentially a
translation of the x86-64 implementation. It worked, but we can make a
more efficient implementation by using an AArch64 instruction I don't
believe x86-64 has an equivalent of: URSHR. The latency is the same as
before, but the instruction count and register count are both reduced.
The new `dispatcher_no_timing_check` is the same as `dispatcher_no_check`
except it includes the "stepping check" in debug mode. This lets us avoid
the `m_enable_debugging ? dispatcher : dispatcher_no_check` dance.
Maybe "tail call" isn't quite the right term for what this code
is doing, since it's jumping to the dispatcher rather than
returning, but it's the same optimization as for a tail call.
fregsIn will include FD for double-precision instructions, since for
dependency tracking purposes the instruction does read the upper
half of FD. This is not what we want in HandleNaNs.
The consequence of this bug is that if an instruction was supposed to
output a NaN and FD happens to contain a NaN and FD happens to be the
same register as an unused register in the instruction encoding, the
NaN in FD could get used as the output instead of the correct NaN.
This isn't known to affect any games, which isn't especially surprising
considering that there's only one game that needs AccurateNaNs anyway.
Jumping to `dispatcher` requires first subtracting the downcount,
otherwise `dispatcher` may unpredictably jump to CoreTiming::Advance,
which could break determinism compatibility with JitArm64. We should
jump to `dispatcher_no_check` instead.
The breakpoint check in Jit.cpp makes it redundant.
Normally this redundant check doesn't cause any issues, but if you
create a breakpoint and enable logging without breaking, you get two
log messages if the breakpoint is at the beginning of a block. See
https://bugs.dolphin-emu.org/issues/13044.
This is also a tiny performance improvement for when debugging is
active, since we no longer check for breakpoints for blocks that never
had any breakpoints to begin with.
Nothing currently uses it. It could theoretically be replaced with fmt support, but I don't think the LOG_VULKAN_ERROR macro is that useful and it'd be better to replace it with regular logging instead.
base is an unsigned variable, so we can make things little more
consistent by making the loop index unsigned so we aren't doing bit
arithmetic with signed types.
MemoryInterface already does this, so we can leave it alone.
No behavioral changes, just a consistency thing.
Rather than makring some parts of VertexLoaderManager dirty in some places and some in others, do it all in VideoState. Also, since CPState no longer contains pointers/non-CP data after d039b1bc0d, we can just use p.Do on it instead of manually saving each field.
Micro-optimization. Some CPUs can fuse CMP+B, TST+B, arith+CBZ, etc.
I also moved things around for CMP+CSET and TST+CSET - which I'm not sure
if any CPUs support - but it doesn't hurt anything, so I might as well.
Improves accuracy but isn't known to affect any games.
This turned out to be fairly convenient to implement; ORing with the
PPC default NaN will quieten SNaNs and do nothing to QNaNs.
Shawn Hoffman