mawei/tinygrad
1
0
Fork 0
mirror of https://github.com/tinygrad/tinygrad.git synced 2026-06-13 00:15:35 +08:00
Code Issues Packages Projects Releases Wiki Activity

compile fixes (#10442)

Browse Source
This commit is contained in:
Sidharth N. Babu
2025-06-06 15:38:37 -07:00
committed by GitHub
parent eb7305e6a4
commit ef14dfb277
2 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
extra/gemm/metal_matmul.py
View File

@@ -26,7 +26,7 @@ metalalloc._copyin(c,nc.tobytes())
FLOPS = N*N*N*2
BW = N*N*3*4
prog = MetalProgram(device, "test", MetalCompiler(device).compile(f"""
prog = MetalProgram(device, "test", MetalCompiler().compile(f"""
#include <metal_stdlib>
#include <metal_simdgroup_matrix> // Available from Metal version 2.3 released with OS X 11.0+
using namespace metal;

2
extra/gemm/metal_matvec.py
View File

@@ -30,7 +30,7 @@ WORKSIZE_ROW = 16
WORKSIZE_COL = 1
LOCAL_SIZE = [32, WORKSIZE_COL, WORKSIZE_ROW]
GLOBAL_SIZE = [M//(LOCAL_SIZE[0]*LOCAL_SIZE[1]*4), 1, 1]
prog = MetalProgram(device, "test", MetalCompiler(device).compile(f"""
prog = MetalProgram(device, "test", MetalCompiler().compile(f"""
#include <metal_stdlib>
using namespace metal;
kernel void test(device float* data0, const device float* data1, const device float* data2, uint3 gid [[threadgroup_position_in_grid]], uint3 lid [[thread_position_in_threadgroup]]) {{