commaai
/
tinygrad
mirror of https://github.com/commaai/tinygrad.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

harmless to intertwine the math and the stores

This commit is contained in:
George Hotz 2023-01-21 09:31:56 -08:00
parent 708215d06b
commit f1196984e6
2 changed files with 2 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
extra/thneed.py
View File

@ -281,7 +281,7 @@ class Thneed:
for i, ((prg, args), e) in enumerate(zip(self.cl_cache, events)):
runtime = (e.profile.end - e.profile.start)
print(f"{i:3d} time {total_runtime/1e6:5.2f} ms running {prg.name:20s} with {str(args[0]):15s} {str(args[1]):15s} count {len(args)-2:2d} runtime {runtime/1e3:7.2f} us {(prg.op_estimate)/runtime:9.2f} GFLOPS {prg.options} -> {args[2].shape if hasattr(args[2], 'shape') else args[2].size}")
if DEBUGCL >= 2 and int(os.getenv("PRINT_KERNEL", "-1")) == i:
if (DEBUGCL >= 2 and int(os.getenv("PRINT_KERNEL", "-1")) == i) or DEBUGCL >= 3:
print(prg.prg)
total_runtime += runtime
print(f"total runtime: {total_runtime/1e6:.2f} ms wall time: {et*1000.0:.2f} ms")

8
tinygrad/llops/ops_gpu.py
View File

@ -354,13 +354,7 @@ class CLASTKernel(ASTKernel):
self.kernel.append(f"for (int mid = 0; mid < {self.group_for_reduce}; mid++) {{ {accumulators[0].tok} += vload4(0, &temp[mid*4]); }}\n")
# late ast
outs : List[Token] = []
out : List[Token] = self.ast_parse(self.ast, reduce=accumulators)
for i, out in enumerate(out):
self.kernel.append(f"{out.decltype()} outs{i} = {out.tok};\n")
outs.append(Token(f"outs{i}", out.typ))
self.store(0, outs)
self.store(0, self.ast_parse(self.ast, reduce=accumulators))
if self.group_for_reduce: self.kernel.append("}")
self.kernel.append("}")