github/tinygrad
1
1
Fork 0
mirror of https://github.com/tinygrad/tinygrad.git synced 2026-01-25 23:08:06 -05:00
Code Issues Packages Projects Releases Wiki Activity

consistent render of recip in cuda with CStyleLanguage (#6980)

Browse Source
This commit is contained in:
ignaciosica
2024-10-12 07:56:47 -03:00
committed by GitHub
parent a71bb09ec3
commit 334f499e6a
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
tinygrad/renderer/cstyle.py
View File

@@ -294,7 +294,7 @@ class MetalRenderer(CStyleLanguage):
return {arg[3].name}2(c.thread_elements()[0], c.thread_elements()[1]);\n}}""")
return super().render_kernel(function_name, kernel, bufs, uops, prefix)
code_for_op_half = {UnaryOps.RECIP: lambda x,dtype: f"hrcp({x})" if dtype in (dtypes.half, dtypes.bfloat16) else f"1/{x}",
code_for_op_half = {UnaryOps.RECIP: lambda x,dtype: f"hrcp({x})" if dtype in (dtypes.half, dtypes.bfloat16) else f"(1/{x})",
BinaryOps.MAX: lambda a,b,dtype: f"__hmax({a},{b})" if dtype in (dtypes.half, dtypes.bfloat16) else f"max({a},{b})",
UnaryOps.SQRT: lambda x,dtype: f"hsqrt({x})" if dtype in (dtypes.half, dtypes.bfloat16) else f"sqrt({x})",
UnaryOps.SIN: lambda x,dtype: f"hsin({x})" if dtype in (dtypes.half, dtypes.bfloat16) else f"sin({x})",