From e1d2bec4a4889a0e91c5a0bb45b2634d8868d6b8 Mon Sep 17 00:00:00 2001 From: George Hotz Date: Sat, 22 Mar 2025 18:52:56 +0800 Subject: [PATCH] opt --- extra/replay_pkl.py | 14 ++++++++++++++ tinygrad/codegen/kernel.py | 2 +- 2 files changed, 15 insertions(+), 1 deletion(-) diff --git a/extra/replay_pkl.py b/extra/replay_pkl.py index a235558bb8..da192585e8 100644 --- a/extra/replay_pkl.py +++ b/extra/replay_pkl.py @@ -116,6 +116,11 @@ if __name__ == "__main__": k.apply_opt(Opt(OptOps.UNROLL, 0, 4)) k.apply_opt(Opt(OptOps.UPCAST, 1, 24)) k.apply_opt(Opt(OptOps.UPCAST, 0, 16)) + elif knum == 9: + # same as 6 + k.apply_opt(Opt(OptOps.UNROLL, 0, 4)) + k.apply_opt(Opt(OptOps.UPCAST, 1, 24)) + k.apply_opt(Opt(OptOps.UPCAST, 0, 16)) elif knum in [7,11]: k.apply_opt(Opt(OptOps.UNROLL, 0, 4)) k.apply_opt(Opt(OptOps.UPCAST, 1, 144)) @@ -129,6 +134,15 @@ if __name__ == "__main__": k.apply_opt(Opt(OptOps.UPCAST, 1, 64)) #k.apply_opt(Opt(OptOps.UPCAST, 0, 2)) pass + elif knum == 1: + k.apply_opt(Opt(OptOps.UPCAST, 2, 32)) + k.apply_opt(Opt(OptOps.UPCAST, 1, 4)) + k.apply_opt(Opt(OptOps.UPCAST, 0, 2)) + elif knum == 26: + #k.apply_opt(Opt(OptOps.UNROLL, 1, 0)) + #k.apply_opt(Opt(OptOps.UNROLL, 0, 0)) + #k.apply_opt(Opt(OptOps.UPCAST, 2, 128)) + pass #elif knum == 18: # k.apply_opt(Opt(OptOps.UNROLL, 0, 4)) # k.apply_opt(Opt(OptOps.UPCAST, 1, 192)) diff --git a/tinygrad/codegen/kernel.py b/tinygrad/codegen/kernel.py index df44027fd9..8fcc027599 100644 --- a/tinygrad/codegen/kernel.py +++ b/tinygrad/codegen/kernel.py @@ -666,7 +666,7 @@ class Kernel: print(self.name) if getenv("RAWAST"): print(self.ast) for i,(buf,st) in enumerate([(buf,st) for buf,st in zip(self.bufs, self.sts) if buf.op not in {Ops.CONST, Ops.VALID}]): - print(f"{i:2d}: {str(st.shape):25s} {str(buf.src[0].dtype).replace('dtypes.',''):20s} {str(st.real_strides()):20s}", + print(f"{i:2d}: {str(st.shape):25s} {str(buf.src[0].dtype).replace('dtypes.',''):20s} {str(st.real_strides()):30s}", str(st) if DEBUG >= 4 else "") print(self.applied_opts) if DEBUG >= 5: print(modified_ast)