Commit Graph

29 Commits

Author SHA1 Message Date
chrg6
43a8096f37 Give DLSS Ocarina's camera, so it keeps its history along the camera's motion.
Ocarina's opaque scene draws bind a 0x3500-byte scene context at c3, and their vertex shaders take the world position
through its rows 0-2 (the view) and then rows 7-10 (the projection: an infinite-far projection whose z row is its w row
less twice the 0.1 near plane). The camera motion pass now uses that block's view-projection, from the block most of
the frame's light-target draws bind, and DLSS keeps its history instead of starting anew every frame. Measured with
NX2NAVI_MV_CHECK while turning: 176 to 227 px of camera motion a frame, a warp error of 0.006 to 0.015 with the vectors
against about 0.025 without.

There is no jitter on Ocarina's frame yet: with it the picture went white within seconds, with or without motion
vectors, while the same frames without it look right.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-07 10:16:34 -05:00
chrg6
56a46cadba Draw the picture above 1440p, and fold scaling, FSR 1 and DLSS into one Upscaling choice.
NX2NAVI_GAME_RES (1440, 1800, 2160, 2880 or window, read at start) sets the picture's height; the widescreen patch
sets now also run at 16:9 when the height is not the game's, with the width following the shape and the scene 3/4 of
both as the game draws it. Measured at 2160p with 21:9: a 5120x2160 picture and a 3840x1620 scene at 60 fps.

The Screen tab's Scaling-to-the-window row and the DLSS rows become one Upscaling row (Bilinear, AMD FSR 1, NVIDIA
DLSS) over NX2NAVI_UPSCALER and NX2NAVI_DLAA (option_rows: opt_choice_index and opt_apply for a row over several
settings), with the FSR strength greyed out unless FSR is chosen and a Resolution row beside the widescreen one. DLSS
always uses NVIDIA's model for its scale (no model row). There are no DLSS quality modes: the game sizes its scene
from the picture by its own 3/4 share, and writing a smaller scene into wide_scene's constant left the light target
as it was, so every mode would render the same.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-07 09:26:51 -05:00
chrg6
a4fde2090a Put DLSS back in the settings menu, and build the kit's runtime at -O2 without strict aliasing.
The Screen tab gets the DLSS switch and its model choice after the picture rows. There is no quality-mode row: this
build cannot change the game's scene size, so every mode would render the same. The description says that DLSS works
on each frame on its own for now and takes effect after a restart; the model row is greyed out while DLSS is off.

tools/make_game.py compiles with tools/build_game.py's aliasing rule: -fno-strict-aliasing everywhere, since the lifted
code reaches guest memory through casts of one address to several types, and the runtime at -O2 (the lifted code
stays at -O1). x86-64-v3 is left out so that the kit runs on CPUs without AVX2.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-07 09:26:51 -05:00
chrg6
3c0377abf8 Run DLSS on Ocarina's frame, which has no G-buffer and tonemaps straight into its 8-bit output.
Without a scene-size G-buffer the DLSS point arms on the first scene-size light-target pass (B10G11R11 with the D32F
scene depth) and runs at the pass that tonemaps it: into the output when that is larger (DLSS scales and the tonemap
samples its picture), or back into the light target for an sRGB output of the scene size (DLAA). The sizes come from
the frame, since Ocarina's builds have no scene or output patch sets for res_sizes. The deferred engine's shape is
kept as it was.

Ocarina's camera for the motion vectors is not found yet: the c3 view block its light-target draws bind gives a
reprojection that moves a still picture by tens of pixels (NX2NAVI_MV_CHECK), so there is no camera motion and no
jitter there, and DLSS works on each frame alone. dlaa stays off by default.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-07 09:26:51 -05:00
chrg6
375ecf27e2 Wait for the whole boot prewarm and show its progress on a boot screen.
The game used to start after at most 3 s of prewarm, so after a translator, GPU or driver change it met the rest of
its shaders in play and stalled on each one. It now waits for all of it: a warm start is ready in about half a second
and shows nothing, while a cold one shows a "Preparing shaders" screen in the menu's colours and the game's fonts, with
the shader and then the pipeline count and a bar. The shader and pipeline prewarm report their numbers every 100 ms.
The present path takes a frame without a game image for the screen. The RomFS opens on the first host-side read, so
the screen's fonts load before the guest's first file access. NX2NAVI_PREWARM_WAIT_MS now defaults to 0 (no cap). With
no progress for 60 s the game starts anyway and the prewarm finishes in the background.

Measured with an emptied cache: 2,988 shaders translated and 2,399 pipelines built in 9.4 s behind the screen, then
0 ms shader and pipeline waits in Kokiri Forest. Warm: the game starts 0.47 s after launch.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-07 09:26:51 -05:00
chrg6
29fe2b4396 Restyle the settings menu with the game's own interface fonts and textures, and give the window a gold pointer.
runtime/menu_assets.c reads them from the user's RomFS at runtime (nothing is shipped): .zs files with the dictionaries of Pack/ZsDic.pack.zs (streamed when a frame does not say its size), the fonts of Font2/Scalable behind their 8-byte header and per-word XOR key, and BNTX textures (Switch 2 archives say "Ounc" where Switch ones say "NX  ") through the runtime's block-linear copy and BC4 and ASTC decoders. ui/menu_art.cpp decodes the shared interface archive on a worker at the first present and uploads each texture for Dear ImGui at its first use. The menu now uses ClarimoUDPE for its text (with a sixth more line height) and CetraDisplayPro for its title and tabs, Roboto merged behind both; umber, brown and gold colours; the game's ornamented frame along the window's edge (the art holds two sides and is drawn again turned for the other two); its cursor glow and line around the selected row (edges only, so the label stays readable); and its control-guide bar under Continue and Quit. Without the files it keeps Roboto and plain shapes. The window's pointer is an original gold arrowhead drawn at start for the system DPI.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-07 09:26:51 -05:00
chrg6
020fe0eeb8 Give Ocarina's own frames-per-update the display interval, so its menus and interface keep their speed above 60 fps.
The base-step write fixed gameplay, but the interface and the frame counter step by the time manager's nominal frame, which the framework update (main+0x6e6be0) sets every frame as 60 over a target refresh rate it keeps at 60 (main+0x3b4e300 writes frames per update, its inverse, seconds and the rate). The new patch set ostep (recomp/patches-v111.txt, on by default) makes that update read frames per update from patch page +256, which the runtime writes at each present as the 5-frame average of the display interval in 60 Hz frames. The game then makes the whole nominal frame itself and the base step stays 1.0. At 144 fps the Plus menu opened in 0.17 s against 0.35 s at 60; now in 0.25 s, within the screenshot capture's noise. Builds without the set keep writing the base step. NX2NAVI_SHOT_AFTER delays the first NX2NAVI_SHOT_EVERY shot.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-07 09:26:51 -05:00
chrg6
897ad1d0ea Step Ocarina's game by the display interval, not one 60 Hz frame per present, so it keeps its speed above 60 fps.
The time manager **[main+0x4c7e890] holds the frame's base step in 60 Hz frames (+0x1c0), from which each time category's {frames, 1/frames, seconds, 1/seconds} is made (main+0x2c84c). The game leaves it at 1.0 whatever the frame rate, so at 144 fps it ran 2.4 times too fast. Each present now writes the 5-frame average of the display interval there (1/1000 to 1/15 s); the game never writes it itself. Scripted input follows that game clock. NX2NAVI_OCARINA_STEP=0 leaves the step alone; NX2NAVI_TRACE=perf reports game seconds against wall seconds. NX2NAVI_VI_HZ and NX2NAVI_VI_VRR set the refresh rate and VRR support vi reports (60 and off as before), for diagnosis.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-07 09:26:51 -05:00
chrg6
df2fbe4640 Play the game's own interface sounds in the settings menu.
Opening and closing the menu, moving the cursor, changing a value, pressing an item, cancelling a dialog and changing tabs play the game's system sounds (Plus menu open and close, cursor, decide, cancel, tab move), read from the user's RomFS at the first use and decoded from BWAV (DSP-ADPCM or PCM16, 48 kHz) in runtime/menu_sound.c. They are mixed over the game's output in the audio callback with up to eight voices started without locks. The Sound tab gets a Menu sounds volume (menu_sound_volume, 0 to 100, default 80, under the output volume). A RomFS without the files stays silent.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-07 05:04:40 -05:00
chrg6
fa6f9717fb Reach buffers through a device address table where the translator finds no storage buffer, and lower MATCH.ALL and ATOMS.POPC.INC.
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>
2026-11-07 05:04:40 -05:00
chrg6
f76d896ed5 Give the translator 64 storage buffer slots and keep it converting a block after a store it cannot place.
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>
2026-11-07 05:04:40 -05:00
chrg6
99de7b943f Give each file of copied code one license line.
Each file that contains copied code names its copyright holder and license and points at THIRD_PARTY_NOTICES.md; zlib code keeps its full notice and the mark that it was altered. Comments and probe text only.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-07 05:04:40 -05:00
chrg6
1b39aabd37 Describe what the code does in comments and docs.
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>
2026-11-07 05:04:40 -05:00
chrg6
91c147a0aa Draw Ocarina in the window's shape: a wider picture and camera, with the HUD, menus and movies kept 16:9 in the middle.
The patch sets of recomp/patches-v111.txt read the game's 2560x1440 picture and 1920x1080 scene sizes from the patch page, give every 16:9 camera the picture's ratio, squeeze the UI's full-screen projection into the middle 16:9 (fades and the menu backgrounds that redraw a capture stay full width) and pillarbox movies. runtime/widescreen.c sets them up from the new Screen tab row Widescreen (NX2NAVI_GAME_ASPECT: 16:9, window, 21:9 or 32:9, at the next start).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-07 05:04:39 -05:00
chrg6
29cab01da7 Guard float colour outputs against NaN, start mesh counts at zero, and read XInput pads SDL misses.
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>
2026-11-07 05:04:39 -05:00
chrg6
18b9ef3cd0 Log who sizes 16:9 textures with NX2NAVI_LOG_RT_SIZES.
A diagnosis for native ultrawide: each nvnTextureBuilderSetSize2D of a 16:9 size of at least 320 wide logs its
caller and the caller's caller (up to 200 lines).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-07 05:04:39 -05:00
chrg6
cbef6bf7da Remove the colour settings from the menu and show the game's own colours.
The Colour group (profile, brightness, contrast, saturation, gamma) is gone from the Screen tab; values left in
nx2navi.ini are no longer applied, since the menu could not undo them.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-07 05:04:39 -05:00
chrg6
4943b50ff2 Unlock the frame rate, shorten cold starts, and describe the Ocarina build in the docs.
Ocarina steps its game by the clock (the same frame-numbered walk covers half the distance at 120 fps), so in a build
without the fps60 and realtime patch sets the pacer holds frames to the frame limit alone: any rate from 30 to 1000,
the monitor's rate, or unlimited (164 fps average in Kokiri Forest). The menu always shows the row, with 165, 240 and
Unlimited added. NX2NAVI_PACE_HZ sets the pacer's rate for tests.

A compute pipeline is compiled outside the pipeline lock, and one compile serves the replay and the boot prebuild
alike; the prebuild uses half the CPUs. The guest waits at most NX2NAVI_PREWARM_WAIT_MS (3000) for the boot prewarm,
which then goes on in the background (a cold start waited 9.8 s). NX2NAVI_TEST_PAD_AT attaches the test gamepad late
and presses it, to test a controller switched on in play.

README, docs/architecture.md, docs/controls.md and AGENTS.md describe the Ocarina build.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-07 05:04:39 -05:00
chrg6
f5e3322bf5 Follow the gamepad the player presses and show the project's name.
Every connected gamepad is kept open and rescanned each second, and the game reads the one pressed last: taking only
the first one SDL listed left the game deaf when that was a virtual or idle device. SDL may deliver gamepad presses in
the background, since it cannot see the Win32 window's focus; the keyboard stays gated on it. The log notes each
gamepad's first press. The version line, the window title and the Discord status say the project's name.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-07 05:04:39 -05:00
chrg6
c7257fe652 Fix NVN2 compute constants, indirect mesh draws and empty descriptor sets, and cut the texture pool refresh.
NVN2 compute banks get the dispatch's group counts in bytes 0-0xb, for direct and indirect dispatches; the fog pass
after the integration divided by them. Indirect mesh draws read NVN's {count, first} record as {x, y, z}, so y = 0
drew nothing: each now draws from a ring entry holding {count, 1, 1}, copied on the GPU before the draw. An empty set 1
no longer leaves set 0 pushed with 0 bytes. Compute stages with subgroup operations whose width is a whole number of
subgroups require full subgroups.

The texture pool refresh fell from 3.1-3.7 ms to 1.0 ms a frame in Kokiri Forest: slots with the same descriptor,
address generation and entry generation share one lookup (about 14,500 of 15,000 refills a frame), and registering a
texture and view identical to what the slot holds keeps its resolution. Under 20 GiB of VRAM, ASTC textures are
encoded to BC3 by default (Kokiri: 15.9 to 8.2 GiB, same GPU time). The pool log counts refills by cause and what
invalidates them; frame dumps write BC textures as raw blocks and print a storage buffer's backing span; a profile with
NX2NAVI_PROFILE_FRAMES is written when its window ends; the alpha test logs once that it is not applied.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-07 05:04:39 -05:00
chrg6
12fbb31451 Stop texture slots registered before an entry was released from reviving it.
A CPU-written texture slot could resolve to a released entry and upload from storage the game was already freeing,
which crashed in tex_bl_copy during the pool refresh. Each released entry now records the registration serial at
release, and slot_resolve treats slots registered at or before it as stale.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-07 05:04:39 -05:00
chrg6
9b4cca39ca Map Ocarina's global memory accesses through pointer tables and summed bases to storage buffers.
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>
2026-11-07 05:04:39 -05:00
chrg6
eed3e6b786 Decode the upscaler stand-in's picture with a square, emit SURED with its real opcode and keep the translator's warnings.
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>
2026-11-07 05:04:39 -05:00
chrg6
5dfa690d96 Fix Ocarina's garbled UI caption, the mesh pipeline hitches, half the scene-load stall and Kokiri Forest's frame rate.
A texture slot whose content the CPU writes now follows another image at its address only when that image differs in
swizzle alone or holds GPU-written content; Ocarina reuses freed glyph atlas memory for captions of another width, and
following the older image drew the Basic Controls screen's "Target" with the wrong row length. Mesh-shader pipelines
are built on the workers and recorded in pipeline-keys.bin (version 4 with the task stage; version 3 files are
converted) instead of stopping the frame for 0.26-0.51 s. The ASTC decoder reuses one scratch buffer per upload command
buffer, and CPU-provided textures up to 32 MiB are sub-allocated unless the driver requires dedicated memory, which
takes the scene-load frame from about 1.25 s to 0.7 s. The feedback scan looks only at bound texture slots instead of
the bindless pool arrays, per-draw knob lookups are cached and the texture address buckets grow to 65536, so Kokiri
Forest's replay falls from about 16 to 7-9 ms a frame and the forest holds 60 fps. A float clear of zero on an integer
image now clears instead of being skipped.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-07 05:04:39 -05:00
chrg6
9a5b1163f3 Make every settings menu row work for Ocarina and give the controls Ocarina's actions and keys.
The Screen tab hides Game resolution when the build has no output and scene patch sets, and the resolution code no
longer grows the memory pool for it. FSR sharpness now also runs when the game image is shrunk to the window, which is
always the case for Ocarina's 2560x1440 picture at 1920x1080, and the overlay drops NX2navi's fixed 3D scene size. A colour
profile set in nx2navi.ini now applies without the menu. The Game tab hides the frame limit in builds without frame-rate
patch sets and lists the game languages the RomFS ships, adding Polish and Brazilian Portuguese. 5.1 sound is reported to
the game's virtual output device. The remap page describes each button as Ocarina's controls screen does, gives each
D-pad direction its own description and drops NX2navi's shake row; the keyboard defaults follow Ocarina (Shift dashes on A,
Ctrl puts away on B, O and N for the ocarina and Navi, middle mouse resets the camera). docs/controls.md is updated.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-07 05:04:39 -05:00
chrg6
618b9ba4e5 Fix Ocarina's white dialogue frames, the fountain floor speckle and the button prompt text
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>
2026-11-07 05:04:39 -05:00
chrg6
c8738a948c Name the settings, files and tools NX2navi
The names are NX2NAVI, NX2navi or nx2navi: the settings prefix (NX2NAVI_*), the ini file
(nx2navi.ini), the shader translator (nx2navi_xlate), the Windows resources, the installer and the docs. The 8-byte
file magics keep their size (NX2NSANS, NX2NSPRG, NX2NSPRP, NX2NPKEY, NX2NRFS1), so caches and the RomFS index from
older builds are not read. cfg.c and the settings pages took the prefix length as 5; it now follows the prefix.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-07 05:04:39 -05:00
chrg6
77d0c0b84e Port Ocarina of Time (Switch 2) to the recomp host
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>
2026-11-07 05:04:39 -05:00
Valiantsin Dzerakh
0478c13e45 Import the base tree (MIT, see LICENSE) 2026-11-07 05:04:38 -05:00