George HotzandGitHub c32ea95d7d Python uop emulator (#3327) ...

* start uop emu

* tiny_add passes

* more ops

* emulate the whole warp

* test_gemm passes

* metal gemm test pass

* works on big gemm

* works on big gemm

* more tests pass

* touch ups

* fix mypy

* cleanups

* exp2 mypy

* arch is where it belongs

* actually emulate tensor cores

* fix test

* new style

2024-02-08 19:24:55 +01:00

..

codegen

Python uop emulator (#3327)

2024-02-08 19:24:55 +01:00

features

limit group_for_reduce bufs to 32kb (#3299)

2024-02-02 03:13:12 -05:00

nn

fix safetensor load uint32 and uint64 (#3315)

2024-02-04 10:46:27 -05:00

renderer

wmma: add HIP FP16 to FP16 tensor core (#3287)

2024-01-31 23:00:51 -05:00

runtime

Python uop emulator (#3327)

2024-02-08 19:24:55 +01:00

shape

fix import and long lines in view (#3337)

2024-02-07 06:50:21 -05:00

__init__.py

move dtypes to dtype.py (#2964)

2024-01-01 14:58:48 -08:00

device.py

type annotation for Compiler.cachekey and minor cleanup (#3298)

2024-02-01 21:31:21 -05:00

dtype.py

const cleanup with dtype.Scalar (#3257)

2024-01-26 21:16:22 -05:00

graph.py

remove unused expr node (#3170)

2024-01-18 14:18:43 -08:00

helpers.py

hip mutex signal (#3234)

2024-01-24 13:23:09 -08:00

jit.py

fix jit realize issue (#3258)

2024-01-26 18:27:35 -08:00

lazy.py

PoC UnaryOps before expand (#3319)

2024-02-06 19:05:13 -08:00

mlops.py

Fix backward fn for < and == (#3037)

2024-01-14 20:39:52 -08:00

ops.py

hip events work (#3229)

2024-01-24 11:49:53 -08:00

realize.py

minor hip cleanups (#3237)

2024-01-24 15:13:38 -08:00

tensor.py

test intermediate tensors created by function have same device as input (#3338)

2024-02-07 09:24:36 -05:00