* minor cleanup on test_tensor_core_opts tests
Tests now notify when skipped
Before, they silently skipped if backend didn't had half precision and
accumulation
Also cleaned up atol and rtol setup
* refactor test_tensor_core_opts_group
---------
Co-authored-by: George Hotz <72895+geohot@users.noreply.github.com>
* change clang -march flag to -mcpu with fp16 disassembly test
* fix
* add capstone to macos dependencies
* just check no cast in test
* rm import
* woops
* lets check
* move check
* llvm init before cpu chcek
* try this
* bump autogen llvm version
* also update libclang?
* revert
* add comment
* skip llvm test and add comment
* linter
* move index validation to load/stores
* add name
* add linearizer_failure
* add validate_store with implicit gates
* linearizer_failure_58 is fixed!
* add test_uop_graph test
* rename cond to gate
* test gated load/stores
* use or_casted()
* viz: non blocking UOp tracing
* u.arg
* no if Ops.KENREL
* drop replace
* switch to weakref.WeakKeyDictionary
* back
* remove ram usage skips, viz works here
* cache on reconstruct
* fix
* add early verbose demo test
* is this how to write tests :s
* is definition drift even a thing? gemini says it is
* clean up
* better
* even better
* try add to CI
* doesn't work quite yet
* much more work to be done
* whoops
* partition the test heh
* skipif
* some nits for better names
* add webgpu test for onnxrunner
* fix reference links
* flush for now
* work
* basic stuff
* work
* also reset
* moving through time
* cleanup
* proper zoom
* add livereload.js
pip install livereload
livereload tinygrad/viz
* minor
* fixed width, remove viewbox
* bit of flexbox magic
* show pid/tid
* merge loops
* min-height
* redo some layout stuff
* create cell groups
* text is hard
* javascript Math.min causes "Maximum call stack size"
bert repro: VIZ=1 PYTHONPATH=. DEFAULT_FLOAT=HALF BS=66 GPUS=6 BERT_LAYERS=2 FUSE_ARANGE=1 MODEL=bert python3 examples/mlperf/model_train.py
* fix recursion issue
* no viz/server changes
* fix test_viz
* everything is a g
* text is easy
* no it's still hard
* livereload+notes
* height: 100% fixes the device bug
* start canvas work
* base canvas
* take chrome's stuff
* serve chrome's thing
* fetch traces from get_profile
* remove junk
* remove some more
* bring everything back again
* dispatch resize events
* base ticks
* hook d3.zoom
* zoom on the x axis
* bring filter back, makes ctrl+drag possible
* remove junk
* Revert "remove junk"
This reverts commit 4987e7bec1.
* draws something, the zooms aren't right
* move to canvas
* fix zooming
* Revert "Revert "remove junk""
This reverts commit 5aac2034fb.
* space key resets zoom
* Divide timelines by device on y axis
* Show kernel names when the width allows
* Clicking on kernel opens it in the kernel graph
* remove livereload.js
* reset diff
* base diff:
- fetch traceEvents
- displayGraph
- flexbox layout
- rest of canvas
* rescale in-place is faster, d3's rescaleX creates a copy
* less
* aesthetics
* map names
* first viz is profiler
* this will work when i make canvas once
* initial cleanups
* factor out of loop
* refactor + only show devices with events
* properly align program rects
* cleaner tick lines
* padding
* listen for resize
* simple zoom
* space more
* i always end up making zoom globl
* how is this ever allowed
* clicking works again
* back button goes back to the same zoom level
* work
* more work
* bring that back
* coloring work and simplify
* black
* keep perfetto button for comparison
* better
* ph===X
* simplify history stuff
* temp: handcoded
* test: flamegraph style leveling
* Revert "temp: handcoded"
This reverts commit bdcd538e88.
* disable flamegraph
* group by pid
* factor y and height out of render
* now flamegraph is easy
* livereload stuff
* remove that
* less
