* refactor unit tests for dtypes
* add missing dtypes in llvmir.py and lib.py
* skip torch tests
* webgpu
* cleaner skips
* fix llvm bool casting issue using compare
* llvm 100% passing
* llvm segfault
* TEMP decrease timeout mins to 11
debug
* add bf16 to setup
* skip half tests in cuda cpu
* check for CUDACPU insetad
* add int16 to triton dtypes
* u16 for triton
* remove debug - diff is still hard to read
* derive from base class TestDType
* enhance test_upcast and downcast by running on every possible version
* dummy commit to rerun the flakey test
* skip the correct tests for CUDA
* bf16 should be skipped in the common TestDType cases
* re-enable bf16
* more consistent structure
* tiny changes to is_dtype_supported 1
* tiny changes 2
add reason
* fuzz
* fuzzer p2
* run fp32 twice
* remove duplicate fp32 run
* clang: use stdbool
* skip triton on bool casts
* merge and resolve conflicts
* Enable Multi-Output Export
* Add test
* Update examples and lint
* fix padding
* test ops
* dummy commit to rerun test
* revert cuda lint
* Enforce tuple/list of tensors
* subscripted generics
* put back webgpu test
* Re-enable WebGPU Efficientnet test
* optimizer: simplify GROUP and LOCAL to have one of each
Now that tensor cores only use LASTLOCAL, we can simplify to use
only that op everywhere.
The only use of GROUP is in matvec hand-coded opts and it doesn't
make a performance difference so switching to use only the top
behavior.
Also adds additional asserts to prevent tensor core dims from
being altered which causes bad kernels to be generated.
* search: remove duplicated actions
* stable diffusion < 324ms
* revert swap action
* fix tests due to more sum splitting
* REDUCEOP_SPLIT_THRESHOLD env var
* added from unaligned np test (#2134)
* align cpu buffer before copy into cl buffer (#2135)
* remove shelve from handcode_resnet50_opt.py (#2139)
* Add dictionary keys to reduce db size (#2131)
* work
* ignore beam cache
* dictionary keys are generic
* minor db cleanups
* fix baseline and extract dataset
* fix training
* log likelihood
* more lin to feats
* sts
* training policynet
* net sort of works
* dedup
* refactor, stupid new actions
* fix uops deduping
* BEAM_ESTIMATE
---------
Co-authored-by: chenyu <chenyu@fastmail.com>
Co-authored-by: imaolo <56898718+imaolo@users.noreply.github.com>
* wmma: refactor tensor cores using existing local dims
* optimizer: fix bad rebase and break after one late local
---------
Co-authored-by: George Hotz <72895+geohot@users.noreply.github.com>
* feat: move to hip
* feat: special path for RawBufferTransfer
* feat: initial rawbuffertransfer
* feat: hip ipc
* feat: working hip ipc
* feat: need to base device without args
* feat: close mem handle
* feat: modified test
* feat: more multihip stuff
* clean: cleanup
* feat: cleaner
* feat: don't crash
* feat: test more
* clean: way cleaner hip wrapper
* feat: barrier
* feat: barrier
* feat: this breaks stuff
* feat: we can use empty here
* feat: maybe fix tests
* feat: maybe fix tests again?
* fix: probably fix tests
* feat: no waiting here
* feat: wait here
* feat: much larger test
* feat: need to sync here
* feat: make this async
* feat: no waiting!
* feat: cut here
* feat: sync copy
* feat: random imports
* feat: much cleaner world
* feat: restore this
* feat: restore this
* clean: cleanup
* feat: set this
* optimizer: add test for correctness of opts
Also added OptOps.UPCASTMID to constrain valid axes for opts with
group_for_reduce.
* llvm: fix LinearizerOptions to correctly not has_shared
* optimizer: remove premature test scaffold for TC opts
* search: fix the action space
* uops graphing
* add_phi_node
* less phi nodes
* where graph uops should live
* naming
* move it to external
* fix triton yolo
* fix clang and preserve behavior
* add mops to graph, refactor IMAGE
* no reshape pushing
* add todo
* fix openpilot model alt
* push reshapes reduces kernels in new op
* IMAGE=2 is a first class citizen now
* Revert "disable flaky triton test"
This reverts commit 1e15fdaee7.
* Update test.yml
* check if has shared for matvec
* disable ocelot cache for triton
* disable ocelot cache
* disable ocelot cache
* pass shared to triton uops tests
* temporary debugs for CI crash
* Revert "temporary debugs for CI crash"
This reverts commit fee3ea96c8.
* Revert "triton isn't tested, and allows this refactor (#2007)"
This reverts commit dea8bb0938.
* add runtime_args to every renderer, move triton local size override to runtime args
* Add binary to args, correct type returned
* update to new loops
* Update test.yml
* some cleanup
* move continue back
* more more more
* added to CI
* try
* try intentionally break some tests
* wtf
* del True for test
* yay tests broke, now pls no break
* try AGAIN
* gahy
* lol
* try
* move over constant
* moved over MORE
* move shrink over
* trailing lines
* try CUDA CI
* try again
* boom
* oops
* improved comments
* try: disable some flags and disable CUDA
* try breaking tests
* traceback has too much info so add --tb=no
* revert forced CI failure
* add comments and del unused imports
* oooooooo using regular debug try enable tb
* intentionally break tests
* added tb back. Maybe not too verbose
* strip whitespcae
* missed something
* Shape op int32 -> int64
* oops missed something
* add some types
* get rid of crazy 1 liners in pad op
* actually test Split this time LOL
* strip that whitespace
* limit metal buffers
* look at the base, not the srcs
* Revert "Revert "openpilot kernel fix from 209 to 207 (#2006)" (#2065)"
This reverts commit 924ecc4d6a.
* add a test for that
* create cache for q learning
* make linter happy
* global beam
* where it belongs
* bugfix
* ditch the kopt, use the beam
* faster lin and DEBUG=2 okay
* remove kopt, move search to features
* Fix openpilot kernel from 209 to 206
1. Use push_movement_ops conditions in _movement_op. Don't push
PAD or check if the ops are safe to be pushed with PAD
2. Don't push if all the op.buffers are realized
* change ALLOWED_KERNEL_COUNT to 206 for openpilot
* don't push through sourceless buffers
* change the tests to adjust kernel counts for new behaviour
* restore pushing of movement ops through childless buffer
* don't push EXPAND, causes OOM
* allow push of intermediate movement ops
* adding new test behaviour
* modifying external_test_opt for new behaviour
* restore old tests
* Reenable push of EXPAND and introduce new tests
I was wrong intially thinking EXPAND can cause OOM and hence I had
disabled it. Since it is 0 stride and doesn't allocate memory its cool
* Don't push EXPAND above LoadOps LB. This is causing OOM
* Push should be decided on movement root of bufs
To check if ast.op.buffers is sourceless/ realized go the the movement
root and then decide if pushing should be done or not
* refactor for readability
* use .base instead
* don't push expand, bad memory/compute consumption
* restrict push of reshape, seeing improvement
* push reshape if unary without further check
* disable PAD solves convnext kernel count increase
* reenable test_cache_binaryop_transpose
* small nit
* init compiled cache
* clang not compile to stdout
* use kwrags in compile
* remove some useless lines
* slimmer
* fix
* tabs
* retry
* remove decorator
* no race in hip
* smaller hip
* unused import
* unused pathlib
* path to str
* add test
* fix linter
* less lines?
* decorator is back
* update tests
* no hip version
* better comments
* a bit better test
* linter
* work wo decorator
* linter happy
* simpler return type
* more tests
* better comment
* readable
* readable
* readable
* compile returns bytes
* no ununsed imports
* readable
* start work on auto opt
* lin failure
* not beating hcopt
* greedy
* timing is fast
* codegen.search
* greedy search in handcode_opt
* track running gflops
* clean up those files
* no failure
* testing with the test_ops pattern
* add assign test
* flake8 complaining about single line fn
* slice 2d and minor cleanup
* make assign_slice a one-liner
* we dont need to repeat the same lambda twice, default tinygrad_fxn to be np_fxn
* back assign fn for np array
* implement __setitem__ in tensor.py
* dont re-slice the ret tesnsor
* one liner assign
* drop the permute test
