mtvc masks the written control register like the other JITs, vsgn tests the
exponent (so a denormal gives 0, as in the interpreter), and the hardware
half-to-float conversion quiets NaNs, so vh2f goes to the interpreter.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The ARM32 JIT only checked for unknown prefixes, so a swizzle naming a lane
past the op's size, or a T prefix on vscl, read the wrong register or
asserted in VMOV. It now uses the ARM64 JIT's gates, and sends what they
refuse to the interpreter.
Its vasin was a polynomial approximation, not the bit-exact function the
other cores use (cpu/vfpu/exact failed 364 lines), so it goes to the
interpreter too, like vsin and vcos already did.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ins with msb below pos encoded an invalid BFI instead of doing nothing,
as the interpreter and the other JITs do, and ext with pos + size past 32
an invalid UBFX. max and min of two constants compared them unsigned.
Found by CPUEquivalence.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
So vcmovt never moves and vcmovf always moves (cpu/vfpu/undefined). The
interpreter did nothing for either, and the JITs took it as imm3 = 6,
one CC bit per lane.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
On the PSP, zeros and denormals all compare equal, a tie returns the
second operand, and the result keeps its bits (cpu/vfpu/minmax_zero,
minmax_tie, fcr31). The interpreter got the ties and NaNs right but
compared denormals as numbers, and each JIT had its own fast path with
host min/max, which returned the first operand on a tie and flushed or
not depending on the host's mode.
All cores now compare by vfpu_minmax_key(), an integer key that is 0
for a zero exponent and orders the rest by sign and magnitude, and
select the operand by it, without branching on NaN.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
srl/srlv rotate when the low bit of their unused field (rs or sa) is
set and shift when it's clear, whatever the other bits hold
(cpu/cpu_alu/shift_encodings). The interpreter treated values above 1 as
invalid and the other cores shifted.
The Allegrex bit ops decode only some bits of sa, so all 32 values are
one of wsbh, wsbw, seb, seh and bitrev (cpu/cpu_alu/allegrex_encodings).
AllegrexBitOp() gives the canonical one for every core. Special3
function 0x18 does nothing on the PSP, where it was decoded as an alias
of 0x20.
The interpreter dispatch is regenerated, and its generator now lists
each instruction name once. CPUEquivalence no longer skips the srl/srlv
encodings.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
It was left out of the JIT harness tests in 2024 for a crash nobody tracked
down, and runs fine on both arm64 and x86-64 Macs now.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
They swap only the rounding mode bits in FPSCR now instead of resetting it,
which kept FZ off, so a denormal input is flushed like on the other hosts
(cpu/fpu/denormal).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
round.w.s gave 0 for a NaN instead of 0x7FFFFFFF like the other
conversions. round.w.s, ceil.w.s and floor.w.s reset FPSCR to
round-to-nearest and left it there, so cvt.w.s and the float ops after
them ignored the guest's mode (cpu/fpu/roundmode).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Runs FPU math after a ctc1, and after a mode set from outside on blocks
compiled before it, on every core with and without ops left to the
interpreter, and checks the result against the interpreter. A fake syscall
checks that HLE always runs in the host's own mode.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Merges the settings in an ini file over headless's own defaults, for runs that
need something like JitDisableFlags. Unlike in the app, the result is never
saved.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A native backend can run this op through its generic fallback after its own
asm applied the guest's mode, which RestoreHostRoundingMode() doesn't know
about, so it has to clear the bits even when it didn't save them.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A JIT only switched modes around syscalls, and only took the mode into account
in conversions, once it had compiled a ctc1 (hasSetRounding). A mode from an
interpreted ctc1, a savestate made on another core, a core switch or a thread
switch went unnoticed, and HLE then ran in the guest's mode. UpdateFCR31()
now sets the flag when fcr31 isn't the default, and clears the block cache at
the next safe point, since blocks compiled before don't switch either. A new
JIT gets an UpdateFCR31() too.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Interpreted ops run in the guest's mode. A syscall left to the interpreter
(with JitDisable flags, say) is host code and now gets the host's mode, like a
compiled one, and an interpreted ctc1 to fcr31 now takes effect right away:
nothing set hasSetRounding for it, so the mode only reached the host at the
next time slice. Every other interpreted op skips the two calls it used to
make after it. The IR frontend's Comp_Generic, which the IR interpreter and the
IR native backends share, gets the same treatment.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
RestoreHostRoundingMode read the control register back to clear our bits, even
when nothing had been applied. Now ApplyHostRoundingMode saves the host's mode
the first time it changes it, and restore writes that back only if it did. With
the guest in the default mode, both are now just a flag check.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ARM32's libm ceilf/floorf don't honor flush-to-zero, so the interpreters flush
the input themselves there.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ARMv7 NEON always flushes denormals and rounds to nearest, whatever FPSCR
says, so on ARM32 the scalar paths are used, which follow the guest's mode
like everything else.
The Vec4 ops only come from VFPU code, and the VFPU itself always rounds to
nearest and flushes denormals (cpu/vfpu/fcr31, see docs/cpu-equivalence.md),
so the NEON paths were arguably closer to the PSP there. But they disagreed
with every other host and core, and agreement between those comes first.
Running VFPU ops in the VFPU's own mode on every core would get both, and is
still to do.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ApplyHostRoundingMode fell back to fesetround, which can't set FPSCR.FZ,
so the interpreter and IR interpreter didn't flush with FS set as every
other host does. FPSCR has the mode and FZ at the same bits as ARM64's
FPCR. Found with CPUEquivalence on a Galaxy S8+ running the armeabi-v7a
build.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Comp_Generic restored the host's default mode around the interpreter call, so
an instruction the JIT doesn't compile rounded and flushed differently than in
the interpreter. Now the guest's mode stays in place, and is reapplied after the
call in case the instruction changed fcr31.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With FS set in fcr31, x86 flushed only denormal results (FTZ) while
ARM64's FZ also flushes denormal inputs, so the hosts disagreed whenever
a denormal reached an FPU or VFPU op. With DAZ too, x86 flushes the way
ARM64 does. That matches the PSP's VFPU, which always flushes inputs and
outputs (cpu/vfpu/fcr31); the PSP's FPU flushes only results
(cpu/fpu/denormal), which neither host does now.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
It only applied fcr31's rounding mode and flush-to-zero after a ctc1 in
the same block, so add.s, mul.s, cvt and friends rounded to nearest in
every other block. It's now applied once per time slice in RunLoopUntil,
where the native dispatchers do it, and the host's mode is put back
around CoreTiming, compiling, syscalls and replacement functions.
See docs/cpu-equivalence.md.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The same as GachiTora: without it, GPUs with cull distances draw parts the PSP
range-culls after clipping.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A strip through the near plane leaves a sliver at the screen edge on GPUs with
cull distances, where it isn't software clipped. The PSP range-culls it after
clipping.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A draw that blends with the source alpha while setting the stencil (the
alpha on the PSP) needs dual source blending to write both. Without it the
alpha came out wrong, and later draws that blend with the destination alpha
differed: WipEout Pure, Juiced 2 and 16131 UCES00001 on a Mali-G71 and a
PowerVR GE8320. With framebuffer fetch, blend those in the shader instead,
which matches the Mac's Vulkan render now. Without fetch it would take a
framebuffer copy per draw, too slow for something this common.
A draw that writes no color can't need the source alpha, so it writes the
stencil as alpha directly.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With fetch, the shader applies the colour masks and logic ops to the real
bits, and OutRun 2006's water mask comes out like the software renderer's.
BlueToAlpha put the first step in alpha while the masked logic-op steps after
it worked on blue, so on the Adreno 840's OpenGL the water was off by twice as
much as on Vulkan. Now it matches or beats Vulkan there.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ResizeFramebufFBO copied color and depth into the new FBO but left its
stencil cleared. Need for Speed Shift grows its framebuffer from 480 to 512
wide just before stencil testing, and the frame came out far too dark on
Android (112k px off the PSP on an Adreno 840, now 4k). MoltenVK happened
to keep it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
PerformWriteStencilFromMemory bound the framebuffer it wrote the stencil to
and, unless it blitted, left it bound. When that wasn't the current render
target, the following draws landed in it. Draws that copy the target for
shader blending rebind it themselves, which hid this; with framebuffer fetch
they don't, and OutRun 2006's masked water draws went into the old 5551
buffer at the address instead of the new 565 one.
Over the framedumps set on Vulkan, 9 dumps get closer to the PSP (OutRun
ULUS10064_0002 12295 -> 483 px, F1 2009, God of War) and one a little further
(16131 ULUS10193, a slight brightness shift).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Without clip distances, the fragment shader's min/max Z discard only came on
when both minz and maxz were set, while the clip planes take each side on
its own. Pursuit Force's sky has minz 0 and maxz 300 and lies past 300, so
the PSP drops it (its replays of 11399 UCES00019 and 11216 UCUS98703 show
that), but GPUs without clip distances drew it over the scene: 13.9k px off
the PSP on a Mali-G71, now 8 as with clip distances. The discard can now
come with the fragment depth clamp, for the side that's at the limit.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The fragment shader's min/max Z discard declared it with the default
precision, the vertex shader with highp, and PowerVR refuses to link such a
program, so every draw that needed the discard was dropped. On a PowerVR
GE8320 that was Street Riders, Valkyria Chronicles, Flatout and Frontier
Gate, and with the discard taking one-sided depth ranges too (next commit),
most of Pursuit Force.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Picking a byte out of a 16-bit framebuffer needs more than the defaults. The
texcoord reached the shader as a mediump varying, which Adreno keeps in fp16:
its byte coordinate (u times up to 2048) then picks the wrong byte for every
other one past 512, and SOCOM Tactical Strike's night vision came out striped
in the right half. On a PowerVR GE8320 the default lowp in the rest of the
shader made SOCOM's night vision and full screen effect far too bright (53k
and 17k px off the Mac's Vulkan render, now 0). Vulkan already uses highp.
Where the fragment shader has full precision, the vertex shader now writes the
texcoord in highp, and CLUT8 depal reads it that way and computes in highp.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
UpdateUniforms checked some locations against -1 before queueing the update,
but the GL thread fills them in when it links the program, which can be
later. On drivers that link slowly the check saw -1, the update was dropped
along with its dirty flag, and the uniform stayed unset. On a PowerVR GE8320
and a Mali-G71 the texture clamp values never arrived, so Harvest Moon's and
11655's composites sampled one point of the framebuffer (122k and 76k px off
the Mac's Vulkan render, now 0). The GL thread already skips -1.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
GL_TEXTURE_LOD_BIAS is desktop GL only, and the queue runner skipped it on
GLES, but GLES still reported sampler LOD bias as supported, so the bias was
dropped and textures used the wrong mip levels: Super Monkey Ball Adventure's
intro came out blurred on the Adreno 840. Report it unsupported, so the shader
applies it (GPU_USE_SHADER_LOD_BIAS), and give the GL backend the uniform that
Vulkan has in its uniform block.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With the default -1..1, window depth 0 comes from z = -w, where the
rasterizer's rounding is an ulp of 1.0. Puyo Puyo Fever 2's ground at z = 0
got depths just above 0 with depth clamp on, and the sprites on it that test
GEQUAL lost pixels (910 px off the PSP on an Adreno 840, now 0). Use
glClipControl (GL 4.5, ARB/EXT_clip_control) as in Vulkan. Not in libretro,
where the frontend shares the context.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With the default of 4, the driver reads past the end of any upload whose rows
aren't a multiple of 4 bytes, like a 16-bit texture of odd width. The Adreno
840 crashed in glTexImage2D on one in SOCOM Fireteam Bravo 3.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
GLES drivers with GL_EXT_depth_clamp reported depth clamp as supported, but
glEnable(GL_DEPTH_CLAMP) was compiled out with USING_GLES2. The shaders then
relied on clamping that never happened, and geometry outside the depth range
was clipped: on the Adreno 840, Hokuto no Ken's title screen came out black and
Hot Pixel's background was missing.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Mali-G71 and PowerVR GE8320 Vulkan lack it, and without it WipEout Pure,
Juiced 2 and 16131 UCES00001 render like they do there.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>