this fixes .split where self.shape[dim] is not perfectly divisible by
sizes - .chunk is always the wrong choice here:
- tensor((5,)).split(4) should result in (tensor((4,)), tensor((1,)))
was (tensor((3,)), tensor((2,)))
this also fixes issues in .split and .chunk where tensors with
shape[dim]==0 lead to empty tuples/lists when the tensor itself should
have been returned instead
because tinygrad is expected to fail in all cases where torch fails
tinygrad will now be strict regarding sizes having to sum up to passed
dimension in .split, num having to be non-null for .chunk and only
allowing valid dims in .unsqueeze
* test/external/fuzz_linearizer: add a FUZZ_MAX_SIZE option
this allows us to limit the size of the kernel and reduce running
times by avoiding ones that take a long time
* fix spacing and re-order to put parameters together
* support same uidx in multiple shape positions
* rename var
* update comment
* add contiguous index check to global_store too
* update comment
* small change
* is this better?
* smh
* smaller change?
* get rid of more changes
* get rid of more changes
* is this even making anything better
* comment
* fix test
---------
Co-authored-by: George Hotz <72895+geohot@users.noreply.github.com>
* fix LtNode simplification when lhs and rhs contain same variables
`(Variable("a", 1, 5) < Variable("a", 1, 5))` should eval to `NumNode(0)`
* fix with less perf impact
* Fix numpy uint/int overflow
* lol
* Works
* Update
* Move overflow test to float64/float32
* One line
* Update
* One more
---------
Co-authored-by: Patrick Tsai <patosai@users.noreply.github.com>
it's recommended to use __getnewargs__ to update the args of classes that use __new__ when unpickling.
It's preferred because it does not change the __new__ behavior.
* do not truncate float64 precision
* use l suffix to try avoid overload confusion
* long line, ruff bloats the function otherwise
* fmt
* remove long double suffix (l), it's sufficient to have the float32 (f) suffix to avoid function overload ambigouity; add test showcasing rtol=1e-12 precision increase, the test fails without the renderer changes
* use more reasonable test values, same as test_int_to_float_unary_func
* disable test for CUDACPU, does not support half and segfaults on some operations per dtypes_alu test
* disable test for HIP, renderer does not support f64 precision
* do not use noqa E501, break up condition
* remove cpu and torch backends
* don't copy to cpu
* use clang instead of cpu
* multitensor gathers on the first device
* clang is cpu + use default
* fixup
* bugfix
* remove float cast
* cast scalars to the correct value in creation time
* cast scalar in the correct place
* wrong, use y_dtype
* make consts have a unique cache key
* add cast_scalar back
* test_load_cache_const_bufs
* add bool dtype
* test_const_dtype
* fix linters
* generic rendering of half and bf16
hotfix
* fix uops + regression test
* fix the test for metal's half4
* uop.uop fixup
* mypy with --strict-equality, fix ops_gpu
* set metal fast math default to 0 (disabled)
It's a correctness fix because we use inf and nan. Let's see how slow it is
* skip failed onnx tests
* tmp DISABLE_COMPILER_CACHE=1 in metal benchmark
* Revert "tmp DISABLE_COMPILER_CACHE=1 in metal benchmark"
This reverts commit 22267df380.
* env var METAL_FAST_MATH to disable fastmath for metal
use this to test impact of fast math. might need to disable compiler cache with DISABLE_COMPILER_CACHE
* failed onnx test with fast math
METAL_FAST_MATH=0 DISABLE_COMPILER_CACHE=1 NOOPT=1 python -m pytest -n=auto test/external/external_test_onnx_backend.py -k test_MaxPool3d_stride_padding_cpu
Fully UNROLLing the first_reduce should not change the number of
local_dims.
Fully UNROLLing a GROUP dim should reduce the number of
group_for_reduces by one.
Also changed group_for_reduces to be a count as the axis number
isn't used anywhere (they are always the first reduce dims).
