Commit Graph
922 Commits
Author SHA1 Message Date
JosJuice fd511a689f Jit: Skip redundant flushes
This makes codegen faster (by perhaps 10-20% in the case of Jit64,
I didn't measure too closely), which helps speed up NBA Live 2005
a little. But the game still has serious performance issues.
2022-02-14 19:17:59 +01:00
Pokechu22 ce52ea35ec Jit: Replace magic 32 with GPFifo::GATHER_PIPE_SIZE 2022-02-12 16:14:05 -08:00
Pokechu22 558de04cfc Common/Assert: Actually use the ASSERT_MSG's log type parameter
Since it was unused, nonexistent values were used in a few places.  I've replaced them.
2022-01-09 12:44:14 -08:00
Pokechu22 161c627466 Treewide: Remove unused inclusions of <cinttypes>
Most of these became unneeded when fmt was introduced.
2022-01-09 12:43:11 -08:00
Pokechu22 44e93e91d7 Common/Assert: Switch to fmt 2022-01-09 12:43:11 -08:00
Admiral H. Curtiss dc7e7d08ad Config: Port Fastmem setting to new config system. 2022-01-06 16:13:56 +01:00
Admiral H. Curtiss e08171fa24 Config: Port remaining Core settings to new config system (partial). 2022-01-05 00:54:15 +01:00
Admiral H. Curtiss d6331c1e71 Config: Port remaining Interface settings to new config system. 2021-12-31 17:40:04 +01:00
Admiral H. Curtiss 810dcfa0f6 Config: Port Debug settings to new config system. 2021-12-30 22:28:05 +01:00
JMC47 1714dc64bb Merge pull request #10120 from Sintendo/jit64arithcxrefactor
Jit64: Merge arithcx into addx/subfx
2021-12-20 14:20:49 -05:00
JMC47 9ed368ead7 Merge pull request #9376 from merryhime/rlwimix2
Jit_Integer: Optimize rlwimix
2021-12-20 14:20:26 -05:00
Admiral H. Curtiss 48560135b3 Jit64: Put the exception exit in twX into farcode.
As a byproduct, this fixes the dont_trap jump not having enough available jump distance to its target in some instances (eg. in Not64).
2021-12-18 06:05:08 +01:00
Pokechu22 2025763420 Treewide: Adjust order of includes 2021-12-10 14:49:57 -08:00
Merry 850d281cb8 Jit_Integer: Fix pure rotation rlwimix case
We are writing to a.
2021-11-07 14:10:19 +00:00
Sintendo dfb32040bf Jit64: divwx - Micro-optimize division by 2
Prefer using eax to isolate the sign bit. This saves a byte when the
destination ends up as r8-15, because those require a REX prefix.

Before:
41 8B C5             mov         eax,r13d
41 C1 ED 1F          shr         r13d,1Fh
44 03 E8             add         r13d,eax
41 D1 FD             sar         r13d,1

After:
41 8B C5             mov         eax,r13d
C1 E8 1F             shr         eax,1Fh
44 03 E8             add         r13d,eax
41 D1 FD             sar         r13d,1
2021-11-03 21:01:41 +01:00
Sintendo f18f6cd0a2 Jit64: divwx - Improve comments 2021-11-02 21:50:28 +01:00
Pokechu22 78bfd25964 Fix all uninitialized variable warnings (C26495) 2021-10-13 12:32:16 -07:00
JosJuice 9f525d69c8 Jit: Raise program exception on floating point exceptions
This is done entirely through interpreter fallbacks. It would
probably be possible to implement this using host exception
handlers instead, but I think it would be a lot of complexity
for a rarely used feature, so let's not do it for now.

For performance reasons, there are two settings for this feature:
One setting which does enables just what True Crime: New York City
needs and one setting which enables it all. The latter makes
almost all float instructions fall back to the interpreter.
2021-10-13 17:42:56 +02:00
JosJuice c3bcc67653 PowerPC: Update FEX on FPSCR store instead of FPSCR load
This is needed not only for the next commit, but also for
correctly emulating float instructions that write to CR1.
2021-10-13 17:42:56 +02:00
JosJuice 83c6df1965 PowerPC: Set SRR1 correctly for program exceptions 2021-10-13 17:42:56 +02:00
Sintendo da9546cb2f Jit64: Merge subfx and subfcx
Again, logic and optimizations are mostly the same so it makes sense.
2021-09-22 22:29:20 +02:00
Sintendo a725eb80ff Jit64: Merge addx and addcx
They are extremely similar and can share nearly all optimizations.
2021-09-22 22:28:54 +02:00
Sintendo a9e0f21373 Jit64: Split arithcx into addcx and subfcx
JitArm64 also opts to separate the two. The shared logic makes less
sense once we start adding more optimizations.
2021-09-21 22:11:11 +02:00
Léo Lam 6659f8019e Merge pull request #10035 from JosJuice/jitarm64-dcbx-in-loop
JitArm64: dcbx loop detection for improved performance when invalidating large memory regions
2021-08-30 22:39:20 +02:00
JosJuice 90fcaf7e96 Jit: Use one less register in dcbx
We were using a "value" register to avoid clobbering physical_addr,
but this isn't actually needed anymore. The only bits we need from
physical_addr after we start clobbering it are bits 5-9, and
those bits are identical in effective_addr and physical_addr,
so we can read them from effective_addr instead.
2021-08-19 23:26:11 +02:00
JMC47 a36855c983 Merge pull request #9818 from JosJuice/jits-cdts-double
Jits: Don't use fast double-to-single when input is double precision
2021-08-17 05:21:08 -04:00
Admiral H. Curtiss 8b2f5d5006 Jit64: Optimize dcbx being called in a loop over a large memory region. 2021-08-17 02:38:00 +02:00
JosJuice 125af42e4b Jit: Re-add dcbx masking
When making 92d1d60, I checked whether the ~0x1f masking in dcbx
actually was necessary. I came to the conclusion that it wasn't,
so I removed it. However, I hadn't checked the second half of
InvalidateICache closely enough - the masking is actually needed.

This commit re-adds the masking, but this time in C++ code instead
of in jitted code in order to save icache. Though I suppose the
difference doesn't matter all that much, since this is in farcode
and all...

Hopefully fixes https://bugs.dolphin-emu.org/issues/12612.
2021-08-06 14:55:07 +02:00
Admiral H. Curtiss 0e76dabbbb Jit64: Always pass effective address to InvalidateICache() in dcbx. 2021-08-04 22:17:42 +02:00
Mai M 627832355e Merge pull request #9973 from JosJuice/jit-fma-negation-order
Jit: Use accurate negation order for FMA instructions
2021-07-31 20:18:49 -04:00
JosJuice 5c5de35568 Jit: Use ibat_table for dcbf/dcbi/dcbst address check
Minor mistake in 92d1d60. We should be using ibat_table instead
of dbat_table, since we're dealing with invalidating icache.
2021-07-31 14:30:03 +02:00
Léo Lam a208ff5aab Merge pull request #9957 from JosJuice/dcbx-faster
Jit: Perform BAT lookup in dcbf/dcbi/dcbst
2021-07-31 03:27:24 +02:00
JosJuice 08b358a829 Jit64: Fix minor fmaddXX inefficiencies 2021-07-29 23:34:20 +02:00
JosJuice 93e636abc3 Jit: Use accurate negation order for FMA instructions
It was believed that this only mattered when the rounding mode was
set to round to infinity, which games generally don't do, but it
can also affect the sign of the output when the inputs are all zero.
2021-07-29 23:33:35 +02:00
JosJuice 3bb4a4e344 Jit64: Fix fmaddXX with accurate NaNs
So it turns out you have to pass XMM0 as the clobber register
to HandleNaNs, because HandleNaNs uses BLENDVPD and BLENDVPD
implicitly uses XMM0, and nobody noticed when I broke this in
2c38d64 because nobody plays the one game that needs accurate NaNs.
2021-07-28 23:03:03 +02:00
JosJuice 92d1d60ff1 Jit: Perform BAT lookup in dcbf/dcbi/dcbst
When 66b992c fixed https://bugs.dolphin-emu.org/issues/12133,
it did so by removing the broken address calculation entirely and
always using the slow path. This caused a performance regression,
https://bugs.dolphin-emu.org/issues/12477.

This commit instead replaces the broken address calculation with
a BAT lookup. If the BAT lookup succeeds, we can use the old fast
path. Otherwise we use the slow path.

Intends to improve https://bugs.dolphin-emu.org/issues/12477.
2021-07-25 15:15:15 +02:00
JosJuice b84a0704cd Revert "Jit: Fix correctness issue in dcbf/dcbi/dcbst"
This reverts commit 66b992cfe4.

A new (additional) correctness issue was revealed in the old
AArch64 code when applying it on top of modern JitArm64:
LSR was being used when LSRV was intended. This commit uses LSRV.
2021-07-25 15:13:57 +02:00
JosJuice 6fe96d12c6 Jits: Don't use fast double-to-single when input is double precision
The fast path of using CVTSD2SS/FCVTN rounds the significand if it
can't be exactly represented as a single, whereas the accurate path
instead truncates the significand. So we should only use the fast
path if we know that the lower bits of the significand are not set.

This is not known to affect any games.
2021-07-11 13:04:02 +02:00
Pierre Bourdon e149ad4f0a treewide: convert GPLv2+ license info to SPDX tags
SPDX standardizes how source code conveys its copyright and licensing
information. See https://spdx.github.io/spdx-spec/1-rationale/ . SPDX
tags are adopted in many large projects, including things like the Linux
kernel.
2021-07-05 04:35:56 +02:00
Tilka 901a4fb5f4 Merge pull request #9811 from JosJuice/fprf-denormal-singles
Fix FPRF handling of denormal singles
2021-06-29 03:18:18 +01:00
JosJuice 2ef2eee2e0 Merge pull request #9464 from Sintendo/jit64addxreloaded
Jit64: addx revisited
2021-06-28 13:53:44 +02:00
Tillmann Karras a68c1bf648 PowerPC: add enum values for emulator SO and LT shifts 2021-06-28 03:30:07 +01:00
Tillmann Karras 25a701aa1b PowerPC: move SO bit to a more convenient location
Spacing the LT and SO bits like in the native PowerPC format makes it easier to
convert to that format (see mfcr).
2021-06-28 03:06:45 +01:00
JosJuice ccd8233ea3 Jit64: Fix FPRF handling of denormal singles 2021-06-27 20:37:18 +02:00
Tilka aff39af5fb Merge pull request #9810 from JosJuice/fprf-cmp-mask
Jits: Fix fcmpX FPRF mask
2021-06-27 19:32:19 +01:00
Sintendo 7de4623d48 Jit64: mtfsfx - Optimized masking
The masking logic can be eliminated when the mask is known to be all
ones. This case is very common.

Before:
66 48 0F 7E F0       movq        rax,xmm6
8B 55 5C             mov         edx,dword ptr [rbp+5Ch]
83 E0 FF             and         eax,0FFFFFFFFh
83 E2 00             and         edx,0
0B C2                or          eax,edx
89 45 5C             mov         dword ptr [rbp+5Ch],eax
48 8D 15 C3 3C FE 01 lea         rdx,[19520020h]
83 E0 07             and         eax,7
0F AE 14 82          ldmxcsr     dword ptr [rdx+rax*4]

After:
66 48 0F 7E F0       movq        rax,xmm6
89 45 5C             mov         dword ptr [rbp+5Ch],eax
48 8D 15 2A D1 FD 01 lea         rdx,[19440000h]
83 E0 07             and         eax,7
0F AE 14 82          ldmxcsr     dword ptr [rdx+rax*4]
2021-06-26 17:31:31 +02:00
JosJuice b11b5955ae Jit64: Optimize mcrfs when mask == 0 2021-06-18 18:26:49 +02:00
Sintendo 18aaf488b0 Jit64: subfic - Optimize constants for d != a
These optimizations were already present, but only when d == a. They
also make sense when this condition does not hold.

- imm == 0
Before:
41 BB 00 00 00 00    mov         r11d,0
45 2B DF             sub         r11d,r15d

After:
45 8B DF             mov         r11d,r15d
41 F7 DB             neg         r11d

- imm == -1
Before:
41 BD FF FF FF FF    mov         r13d,0FFFFFFFFh
44 2B EE             sub         r13d,esi
0F 93 45 68          setae       byte ptr [rbp+68h]

After:
44 8B EE             mov         r13d,esi
41 F7 D5             not         r13d
C6 45 68 01          mov         byte ptr [rbp+68h],1
2021-06-16 20:33:02 +02:00
JosJuice 891a46596d Jits: Fix fcmpX FPRF mask
fcmpX only updates the FPCC bits, not the C bit.
This was already correctly implemented in the interpreter.
Not known to affect any games, but affects a hardware test.
2021-06-13 13:59:00 +02:00
JosJuice ac28b89fa5 NetPlay/Jit64: Avoid using software FMA
When I added the software FMA path in 2c38d64 and made us use
it when determinism is enabled, I was assuming that either the
performance impact of software FMA wouldn't be too large or CPUs
that were too old to have FMA instructions were too slow to run
Dolphin well anyway. This was wrong. To give an example, the
netplay performance went from 60 FPS to 30 FPS in one case.

This change makes netplay clients negotiate whether FMA should
be used. If all clients use an x64 CPU that supports FMA, or
AArch64, then FMA is enabled, and otherwise FMA is disabled.
In other words, we sacrifice accuracy if needed to avoid massive
slowdown, but not otherwise. When not using netplay, whether to
enable FMA is simply based on whether the host CPU supports it.

The only remaining case where the software FMA path gets used
under normal circumstances is when an input recording is created
on a CPU with FMA support and then played back on a CPU without.
This is not an especially common scenario (though it can happen),
and TASers are generally less picky about performance and more
picky about accuracy than other users anyway.

With this change, FMA desyncs are avoided between AArch64 and
modern x64 CPUs (unlike before 2c38d64), but we do get FMA
desyncs between AArch64 and old x64 CPUs (like before 2c38d64).
This desync can be avoided by adding a non-FMA path to JitArm64 as
an option, which I will wait with for another pull request so that
we can get the performance regression fixed as quickly as possible.

https://bugs.dolphin-emu.org/issues/12542
2021-06-09 22:56:26 +02:00