github/tinygrad
1
1
Fork 0
mirror of https://github.com/tinygrad/tinygrad.git synced 2026-02-10 22:54:59 -05:00
Code Issues Packages Projects Releases Wiki Activity

Refactor GPU and Metal langauges in their own separate renderers (#2033)

Browse Source 
* Refactor GPU and Metal langauges in their own separate renderers

* remove CStyleLanguage imports

* move renderers too
This commit is contained in:
qazal
2023-10-10 17:46:41 +03:00
committed by GitHub
parent f139060103
commit 71d93ffd79
4 changed files with 41 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
tinygrad/runtime/ops_gpu.py
View File

@@ -1,13 +1,13 @@
from __future__ import annotations
import pathlib, functools
import pathlib
import numpy as np
import pyopencl as cl # type: ignore
from typing import Optional, List
from tinygrad.helpers import DEBUG, getenv, prod, ImageDType, OSX, fromimport
from tinygrad.ops import Compiled
from tinygrad.renderer.opencl import OpenCLRenderer
from tinygrad.runtime.lib import RawBufferCopyInOut, LRUAllocator, RawBufferTransfer
from tinygrad.codegen.kernel import LinearizerOptions
from tinygrad.renderer.cstyle import uops_to_cstyle, CStyleLanguage
OSX_TIMING_RATIO = (125/3) if OSX else 1.0 # see test/external/external_osx_profiling.py to determine this ratio. it's in like GPU clocks or something
@@ -103,9 +103,4 @@ class CLProgram:
return None
return None
renderer = functools.partial(uops_to_cstyle, CStyleLanguage(
kernel_prefix = "__kernel ", buffer_prefix = "__global ", smem_align = "__attribute__ ((aligned (16))) ", smem_prefix = "__local ", arg_int_prefix = "const int",
half_prekernel = "#pragma OPENCL EXTENSION cl_khr_fp16 : enable",
barrier = "barrier(CLK_LOCAL_MEM_FENCE);", float4 = "(float4)",
gid = [f'get_group_id({i})' for i in range(3)], lid = [f'get_local_id({i})' for i in range(3)], uses_vload=True))
GPUBuffer = Compiled(CLBuffer, LinearizerOptions(), renderer, CLProgram, CL.synchronize)
GPUBuffer = Compiled(CLBuffer, LinearizerOptions(), OpenCLRenderer, CLProgram, CL.synchronize)