The SM86 lowering keeps the uniform registers in the general registers after the highest one the program uses
(HighestGpr), with one spare so pairs stay aligned. The scan read the register fields as single registers, so a
128-bit load or store whose upper registers the program names nowhere else (LDG, LDS, LDL and STG, STS, STL with size
6) left Rd+2 and Rd+3 to UR0 and UR1, and the load overwrote them. The scan now counts Rd..Rd+3 (Rd..Rd+1 for 64 bits)
of these forms. An overestimate only moves the uniform registers up.
Checked: no translation changes in the game's 520 task and mesh programs, its 4,238 NVN2 compute programs, or the
2,450 vertex and fragment programs whose stored input id starts with 0, all from an installed shader cache, and all
pass spirv-val (vulkan1.3): the case does not occur in what this game version has shown so far.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The loops the translator writes for narrow storage buffer stores and float atomic adds end only when their
compare-and-swap succeeds. With robust buffer access on, an atomic on a word outside its buffer may return anything,
so the compare can fail for ever and the GPU hangs until the device is lost, as reported right after the first
cutscene on an Intel Arc GPU (#610). New translator patch 0041 counts the attempts in a private word (a local cannot
count round a loop in that pass, which runs on SSA form) and leaves the loop after 4,096; contention on a word needs
about as many attempts as invocations race for it. Shared memory loops stay as they are: a shared word in range is
always settled, and the mesh programs run that loop for every primitive index byte.
Checked: all 4,238 NVN2 compute programs an installed shader cache stores translate and pass spirv-val (vulkan1.3); 45
contain a compare-and-swap, 11 of them in a storage loop, which grew by 256 bytes each and now compare the count with
4096; the other 34 are unchanged. tools/xlate_smoke.py passes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
An NVN2 compute program's control section gives its local memory in bytes a thread at +0x68, beside the shared memory
size at +0x64 that the job already carried. Without it the translator declared its default of 0x1000 bytes: one of the
game's compute programs needs 4,752 bytes a thread, and its translation declared a 1,024-word local array that it
indexed up to word 1,187, outside the array. The job now carries "local_mem" from +0x68, and the lowering starts a
compute program's spill area after its own local memory (at least 0x1000 bytes) instead of always at 0x1000, where a
larger program's own memory would have overlapped it.
Checked: that program's translation now declares 1,188 words (indices up to 1,187) and passes spirv-val; two programs
of 3,456 bytes a thread translate as before. All 266 compute programs whose stored input id starts with 0 in an
installed shader cache translate and pass spirv-val (vulkan1.3), and tools/xlate_smoke.py passes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A mesh program's header lists the components of each output vector it has; the fragment stage may read the others.
The native mesh entry function never wrote them, so they read as whatever the output held, and the fallback's vertex
stage wrote 0 for all of them, w included. Maxwell gives an unwritten component (0, 0, 0, 1), as the pipeline code
already pads a missing vertex output. Both paths now write that default: MeshSpirv.cs after copying each vertex's lines,
mesh_fb.c meshfb_vs with w as the bits of 1.0f whatever the type.
Checked: all 520 task and mesh programs of an installed shader cache translate and pass spirv-val (vulkan1.3); 500
mesh modules gained about 730 bytes each on average, so many of the game's mesh programs have partly listed vectors.
--probe-meshfb passes (fallback pictures equal the mesh pipeline's), and its module part converts all 520 new
translations with 0 failures, all 1,020 modules pass spirv-val and their compute pipelines build.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Vulkan leaves workgroup memory undefined, and a mesh program need not write every vertex it has room for or every
index byte. The mesh entry function copies all of them into the outputs, so what an earlier workgroup left in shared
memory became vertices and primitives: on AMD, long thin triangles across the foliage's shadow map. The entry function
now sets the vertex lines, the primitive count and the index bytes to zero in a loop over the workgroup's threads
before the payload copy and the barrier. Task programs keep clearing only their mesh count.
Checked by translating all 520 task and mesh programs an installed shader cache stores with the translator before and
after the change: every one translates and passes spirv-val (vulkan1.3) both times; each mesh module grows by about
88 bytes, task modules are unchanged. tools/xlate_smoke.py passes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A program that only asks a combined-handle texture for its size got that texture declared as plain 2D, the runtime bound a one-layer view, and the layer count read as 1 where the console answers the texture's real count. The game's particle vertex shaders read exactly that count, of a texture their pixel shaders sample as a 2D array, and use it as the frame count of a flipbook, so such particles stayed on their first frame. A size query that reads the component after a 1D or 2D type's dimensions now makes the declaration an array; a plain texture bound through an array view still reports one layer. 74 of the stored shader variants change (the contributor's sweep, all valid; a reader here found the same 74 in a shader cache on this PC), and the contributor's before-and-after pictures differ in the one scene tried, the Wallmaster's warning disc.
Goes in with the image read capability so that installs translate their shaders again once. Built here; tools/xlate_smoke.py's new case reports the texture as plain 2D with the translator before the patch and as a 2D array with it, and all three modules validate. Not played with the new translator.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
(cherry picked from commit e4fac75190b4baf3811feef45b3302a9216cb8b2)
The translator declares every float storage image without a format and turns a load of it into OpImageRead, which SPIR-V allows only in a module that declares StorageImageReadWithoutFormat; only the write-side capability was declared. Of the 44,206 shader variants a 0.7.1 install stores, 117 compute shaders were invalid SPIR-V for that reason (the contributor's sweep with spirv-val, for NVIDIA's, AMD's and Intel's host options: 117 before, 0 after). The drivers tried accept them, so nothing crashed, but an invalid module is whatever a driver makes of it. The read now declares the capability, once a module.
The translator's identity changes, so an updated install translates its shaders again (the carry pass serves the old translations meanwhile). Built here from the local package cache; tools/xlate_smoke.py's new compute case fails with the translator before the patch and passes with it, and both modules validate. Not played with the new translator.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
(cherry picked from commit 62e5c345a2ec3e2cc70be49214997ff76e691309)
The SPIR-V generator emitted OpCompositeExtract with the guest's component index even when it exceeded the image's size vector. A 2D texture's OpImageQuerySizeLod returns a v2int, so a guest request for the third component (depth) produced an extract of index 2, which is invalid SPIR-V. RADV crashed while compiling those variants during shader preparation.
Return the constant 1 for a component past the texture's dimensions, the depth of a 1D/2D/cube texture, and apply the same guard in the GLSL generator. build_xlate.py now expects the added patch.
The Ampere I2F lowering wrote (size | 4) << 10, so the signed flag landed in bit 12, which the Maxwell decoder
does not read (ISrcFmt = (op >> 11) & 4 | (op >> 10) & 3). I2F.F32.S16 and the other signed sources converted
as unsigned: in Jabu-Jabu's Belly the walls' deformation grid of signed 16-bit offsets turned -5 into 65531 and
the room shell into huge stretched triangles (issue 57). The size code stays in bits 10-11, the flag goes to 13,
as I2FP already does.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A census of every NVN2 program in the RomFS shader packages found 31 that failed in the SM86 lowering, none met in play yet. Each form was checked with nvdisasm. MUFU.F16 (the upscaler's colour decode: a half lane widened, the F32 function, rounded back to a half) in 11 upscaler programs. ATOMS.POPC.INC with the X16 index stride, the luminance histograms' three counters per 16-byte bin, in 8 programs. LDG and STG .STRONG.GPU, read as plain accesses, in 2. TLD.LZ.MS, a 2x MSAA resolve that reads samples 0 and 1, in 3. ATOMG.CAS for the foliage programs' slot claims in 4, through Maxwell ATOM.CAS, which the translator only logged as not implemented: patch 0037 emits a global compare-and-swap for it. SUATOM.D on a 1D image, the post-effect append counters, in 2. ISBERD.O, a task program reading back its own payload, in 1. HighestGpr also counts the registers these forms read beyond the base fields.
All 31 now pass spirv-val, and the other 42,146 programs give SPIR-V byte-identical to the previous translator. tensor_lowering_check passes 29 of 29, with four new MUFU.F16 cases checked against a model of the lowering (the warp emulator gains MUFU); a wrong half select fails all four.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Large compute shaders that no cache holds are built on one background thread after the boot prebuild, smallest first, so their first dispatch no longer compiles them on the replay thread. NX2NAVI_BIG_BUILD_MAX_MB stops that builder once the process has grown by that much: on a 31 GB machine one of these shaders took the process up by 12.8 GB and the game stuttered while Windows paged. The builder takes the queue under its lock and copies each entry out before compiling it, so a shader queued meanwhile cannot free the arrays under it. The same change reports what pipeline-cache.bin gives back, and starts over when the file grows past NX2NAVI_PIPELINE_CACHE_MAX_MB.
A tensor-core program is now tested before its shader is asked for, in both dispatch paths. It runs the stand-in while NX2NAVI_NATIVE_TENSOR is off, and translating it anyway recorded four AI-upscaler shaders of 0.87 to 4.4 million SPIR-V words that this game never dispatches natively. Process memory went from 29.3 GB to 24 GB and the multi-gigabyte builds during play are gone.
The frame-time summary says which part a slow frame went to, and with NX2NAVI_REPLAY_TIME how the replay splits between draws, binds and transfers; NX2NAVI_SVC_SLOW_MS logs the kernel services that block, with the thread they blocked on. prof_report.py reads the map this build writes and adds the image base to a runtime Rip, so a profile resolves to function names instead of question marks.
build_xlate.py applies the patches with GIT_CEILING_DIRECTORIES set to the build folder's parent. Inside a checkout git read 0021-prmt-byte-permute.patch's paths from the repository's top, skipped every file of it and still exited 0, so PRMT was missing from locally built translators and every byte permute was a no-op. A Skipped patch line now fails the build. The note in AmpereLowering.cs that said the translator emits nothing for PRMT is corrected.
The payload size in an NVN2 task or mesh program header counts only the program's own data, which starts after a
24-byte header (the mesh count and two words the mesh program reads). The foliage task writes three 1024-byte arrays
from there, up to byte 3096 of a payload sized 3072, so the last entries of the third array were lost and its mesh
programs read past the end of their buffer. Those reads came back different on each run, so a few leaves of each dense
patch landed in different places in the depth pre-pass and the colour pass, and the colour pass's equal depth test left
them unpainted.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The culling passes count each frame's visible foliage with atomics through the device address table into small records (an indirect draw's count, then 1, 1). Stores through the table already reached both the VRAM mirror and its guest copy, but atomics reached the mirror only, so guest memory held the CPU's template of zeros. The CPU writes other words of the same pages every frame, and the sync of such a page copied those zeros over the counts: in Hyrule Field the trees flickered, went missing or stretched into columns reaching the sky. NX2NAVI_MEMORY=verify found 5 to 10 such pages a minute, all in ten pools of 20 and 44 KiB.
Translator patch 0036 repeats every table atomic on the twin (a compare-exchange with the same operands), leaving its result unused so the driver need not wait on guest memory. A buffer fill into a mirror now fills its guest copy at once (vk_mem_guest_twin), so a counter the GPU resets that way counts on from the same value in both; that mismatch is why table atomics stayed out of the twin before. The memory check's mismatch lines now also give the pool's last written binding, its last use, the CPU's write stamp for the page and the first bytes of both copies.
With the change the trees stayed steady at the spot where they flickered, all 37,848 shaders translated without a failure, and the GPU time per frame stayed at about 9 ms (75 fps uncapped).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The SM86 lowering now takes MEMBAR (as Maxwell MEMBAR.GL), drops ERRBAR, takes F2I of an immediate and MATCH.ALL.U64,
and finds a switch's jump table bound past the BSSY the compiler puts between the IMNMX and the shift: three vertex
shaders, one of them drawn in Kokiri Forest, were skipped for it. A task program and the mesh programs it feeds whose
payload size field reads 0 (it counts 256-byte units) get one unit. Of the 42,177 programs, 31 still fail, all modes
whose meaning is not certain.
Translator patch 0035 inlines SHFL.IDX over the whole warp (mask 0x1f) on a 32-lane host as the subgroup shuffle,
instead of a call to the general shuffle helper. The tensor lowering shuffles thousands of times per program and the
driver inlined the helper at each call: a mid-size upscaler shader built in 185 s before and in 34 s now.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The float atomic add's compare-and-swap loop (patch 0033) carried its expected value round the loop in a local, but the
pass runs on SSA form and the optimizer's copy propagation takes every value as assigned once, so the loop compared
against a stale value and the froxel fog added up wrong: Kokiri Forest was washed out white. Each attempt now reads the
word with an atomic OR of 0, which also cannot return a stale cached copy.
Writes through the device address table go to the guest copy for stores only (patch 0034). An atomic on the guest copy
started from its stale value, so a counter the GPU reset with a buffer fill kept counting there, and the page's next
sync copied the wrong count into the mirror: culled foliage drew garbage for a frame.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The table's records gain a twin: a pool with a VRAM mirror names the mirror first and its imported buffer, guest
memory, second; any other record names one buffer twice. Shaders read through the first address and store and run
atomics through both (translator patch 0034); a page without a twin sends the second write to a 4 KiB scratch buffer.
The GPU's writes through the table were not recorded as GPU writes, so they stayed in the mirror: the CPU never saw
them, and a sync of a page the CPU wrote copied the stale guest page over them. Trees drawn from records lost that
way stretched into columns reaching the sky. With NX2NAVI_MEMORY=verify the Hyrule Field walk found 16 such mirror
pages before and none or one after; GPU time per frame 5.92 ms before, 6.17 ms after.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Shaders: a switch whose index is clamped against a register that a MOV of an immediate set before it now finds its
jump table (AmpereLowering.cs BrxTargets), so all 36,031 of the game's shader variants translate and no draw is skipped
for a failed one. The compare-and-swap loop behind storage-buffer atomics (patch 0033) takes the value a failed swap
returns as its next expected value instead of loading the word again, which on some drivers kept returning a stale
copy.
Memory: an unused physical pool outlives the submission being recorded, as virtual pages may still reach it through
device addresses; an image whose creation failed is not tried again for 2 s.
The installer and Linux texts name this game.
The README has the project icon, the current status, features, requirements and known issues.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The runtime's TXD gradient transpose (gfx_shader.c patch_txd_transpose) is off for Switch 2 titles: the SM86 lowering
already orders the interleaved derivatives for the translator, so transposing again swapped them back. The
translator's float min, max and clamp are the NaN-safe forms (patch 0031), as Maxwell FMNMX and saturate behave, so a
NaN no longer spreads through lighting and bloom on drivers that pass it on. Depth-compare gathers with per-tap offsets
are split into one gather per tap like the others (patch 0032), where all four taps took the first tap's offset.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Crashes: the render-target heap (HeapDef DynamicTextureAllocatorMem and its parent) grows with the widescreen
picture, so the pause menu no longer reads a null block at 5120x2160; texture uploads keep the game's memory mapped
while they copy it, so UnmapPhysicalMemory no longer pulls it away mid-copy; DLSS Frame Generation no longer crashes
in Streamline's set-up (physical devices and the window surface go through its proxies, its swapchain images are
counted first, resources are tagged per frame); a stale core-wait target is cleared before its thread is freed.
Freezes and races: a resize during a swapchain rebuild is kept, a failed frame generation swapchain is retried
instead of replaced by a plain one, three acquire timeouts in a row rebuild the swapchain, device waits take the
queue lock, a translator child that stops answering times out, an in-flight compute pipeline builds at normal
priority, a save Commit started before the window closes finishes and none starts after, a settings change just
before closing is saved, and a movie sample taken before a flush is dropped.
HDR: Enhanced HDR extends the game's own tonemap so highlights go above the picture's white (the hue slider,
which no longer applies, is gone from the menu); frame generation gets a HUD-less picture, the HDR flag and the HDR10
metadata under HDR too.
Memory: ASTC textures are encoded to BC3 unless the PC has 20 GiB of VRAM and 48 GiB of RAM, as Windows charges video
memory to the commit charge (a long session held 17 GiB of RGBA8 textures and 45.7 GiB of commit on a 32 GiB PC).
Shaders: the translator lowers spill-mode programs with more than eight uniform register groups in a wider spill
layout, and local loads and stores with a uniform-register address (36029 of the game's 36031 variants translate,
20 failed before).
The launcher release ships the shader tools and AMD's FidelityFX library (make_launcher.py --ffx); windres there gets
the toolchain on PATH.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ULDC's immediate form (0xab9) has no register index; the sub-word path required the field to hold URZ, as the 32-
and 64-bit paths do not, and compilers leave 0 there too. Of the game's 36,031 shaders, 40 more translate: 20 failures
are left, all the lowering's room for uniform register groups.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Kokiri Forest drew black, in tiles and in shadowed and indoor areas, where shaders read light, probe and instance data through pointers they load from other buffers at an index (a phi of a pushed pointer and a loaded one, a pointer spilled to local memory): those reads returned 0. Translator patch 0030 keeps such accesses global, with a new first source naming the table, and the SPIR-V back end looks the address's 4 KiB page up in it and loads through the buffer device address it gives (PhysicalStorageBuffer64). runtime/vk_gpa.c keeps the table: two levels over the 40-bit GPU address space, page entries naming immutable records, records reused only after the GPU finished with them. Imported pools, their VRAM mirrors and the pages of virtual pools go in it; pools in it are destroyed after the submissions that may still read them. The last failing compute program needed MATCH.ALL (subgroupAllEqual) and the aggregated shared increment ATOMS.POPC.INC, now lowered in AmpereLowering.cs; Kokiri Forest translates without a failure.
NX2NAVI_GPA=0 turns the table off and NX2NAVI_GPA_EMPTY=1 keeps it empty. A first try synced every VRAM mirror at each table reader's draw, which overwrote what the GPU had written into them and made the black worse; mirrors sync only when bound.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Ocarina's NVN2 compute programs reach more than 16 buffers through their pushed pointers and pointer tables; past the 16th the translator found no slot, dropped the store, and its loop then read the next node of a node it had just removed, so every later global access in the block stayed global and the whole program failed with "Invalid storage kind GlobalMemory". Ten compute programs of Kokiri Forest were skipped on every dispatch and retried on every one, which also cost 10-25 ms frames. Now one program still fails (an Ampere MATCH instruction the lowering does not handle). NX2NAVI_XLATE_KEEP=2 keeps every translated job with its code for running the translator by hand.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Comments say what the code does and why, docs/maxwell.md keeps the Maxwell facts, and fetch_refs.sh clones only the checkout the build reads. Code is unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Patch 0028: when GlobalToStorage finds no base for a global access whose address is a constant buffer load at a
non-constant offset, every table entry that offset can reach becomes a candidate of the multi-target helper ({u64
address, u32 size} entries 16 bytes apart: C to C + 240 for E + C, each constant for phi(constants) << 4), and the
helper uses the one whose range holds the address. Ocarina's NVN2 compute shaders pick c0 storage slots this way;
unresolved compute accesses fell from 445 in 21 shaders to 243 in 17 over a Kokiri boot.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Patch 0027: a float colour output turns NaN into 0 and clamps the rest to +-1e30 before it reaches its target, so one
bad texel cannot spread through bloom, blur and downsample passes as black or white blocks; integer targets keep their
bits. Task and mesh shaders start their mesh and primitive counts at 0 behind a barrier, and a task emits at most one
mesh group per thread, so a task that culls everything no longer launches what the shared memory last held. Without a
gamepad SDL can open, the first connected XInput controller (slots 0-3, empty slots asked once a second) is read
directly. Kokiri Forest: 60 fps, GPU 5.5 ms a frame, unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Patch 0023 turns a global RED or ATOM .F32 add, which the translator dropped, into a compare-and-swap loop with a float
add once GlobalToStorage knows its buffer; the light and fog accumulators of Ocarina use them, and their loss lit
Kokiri Forest's fog as if nothing shaded it. Patch 0024 reads RED's 20-bit offset as signed, as ATOM already did.
Patch 0025 keeps the block after a CAL as its successor, so code after a call is no longer dropped. Patch 0026 keeps
ballot, shuffle and vote operations where they are, so they see every lane. build_xlate.py now expects 25 patches.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The translator dropped 36113 global accesses per Kokiri boot (loads read 0, stores and atomics deleted); 837 remain.
AmpereLowering computes a predicated LEA's carry sum unconditionally, so a base like c[0][0x110] no longer reaches the
address as a phi; the edge AA's tile list is written and its blend weights follow the edges. Patch 0022 extends
GlobalToStorage: a constant-buffer base inside a sum of terms, and buffers reached through a pointer the program loads
from another storage buffer at a constant offset (the pointer tables Ocarina's vertex, pixel, mesh and compute shaders
read, also through register spills and phis). Such a buffer is listed with its outer buffer and offset, and the runtime
binds it at the address read from guest memory when the stage is bound (vk_exec_draw.c ptr_addr). The fog injection's
64x36x96 volumes now hold in-scattering.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The post pass stores the square root of each channel, so the stand-in for Ocarina's untranslatable upscaler now
filters those values and squares them in its compute pass (shaders/fallback_tonemap.spvasm, vk_tonemap.c) instead of
decoding them with the sRGB curve, which darkened mid-tones and lifted the darkest values; the pass is compiled in again,
its round-24 device loss having been the push descriptor call outside the recorder. NX2NAVI_FALLBACK_EXPOSURE is gone.
AmpereLowering emitted SURED as 0xeb48, which matches no Maxwell instruction, so the translator skipped every image
reduction; with 0xeb58 the edge AA's blend-weight pass writes its image. The translator's own warnings go to res.json
notes, and shader dumps are named by program serial so they match the frame dumps' program numbers.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The texture pool descriptors are now a ring of sets, one per submission, brought up to date from a change log, and the first submission of each frame waits for its GPU work before the game goes on: the GPU ran late and part of a frame read data of the next one, and distant characters drew NaN that bloom spread over the frame. The window acquire also waits for the image's previous frame. Function locals start at zero unless DLAA is on, which removes the per-warp light tiles on the Great Fairy Fountain floor and Link's dark, speckled look. Texture slots follow only live entries of their own target, so the volumetric fog gets its 3D volume. The pool refresh walks only the slots that changed.
The translator implements PRMT (patch 0021), which it only logged before; the button prompts render again. HFMA2.RELU no longer depends on it, HMMA.F16 rounds once and HMMA.F32 is lowered, checked against GPU vectors with a new Maxwell warp emulator.
The frame dump gains a re-arming white-frame trigger, a trigger on a depth target's first pass, NaN pixels in magenta, image handles and view targets, and depth copies of sampled depth textures.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The NVN2 front end and the SM86 to Maxwell shader lowering (tools/xlate/AmpereLowering.cs, mesh and task stages,
bindless texture pools, BRX jump tables, the NVN2 compute c0 layout) with four translator patches (0017-0020) bring the
opening of the game to 58-60 fps on the Vulkan renderer.
This round fixes the haze: SM86 IPA is lowered as a plain Maxwell IPA.PASS (texture coordinates were corrected for
perspective twice); LDSM, TXD, both F2IP byte forms, HMMA, MOVM and HFMA2.RELU are lowered and spill mode has eight
uniform register groups; conditional rendering keeps the newest counter report per address and treats a report as
pending until its value is in guest memory, so occlusion tests and the depth clears around the depth copy behave; and
the upscaler stand-in decodes the post pass's sRGB picture before writing the linear target. Tensor-core programs keep
the stand-in unless NX2NAVI_NATIVE_TENSOR=1, because their lowering does not match the hardware yet.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>