gaspersic/triton
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[FRONTEND] Improve error message for CPU tensors (#654)

Browse Source 
Redo of #651 against master. Fixes #525 by catching CUDA error when we
check pytorch tensor size and rethrowing a more informative error that
says why we failed.
This commit is contained in:
Sophia Wisdom
2022-09-14 14:26:42 -07:00
committed by GitHub
parent cfbbc7b43a
commit 4580a04710
1 changed files with 7 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
python/src/triton.cc
View File

@@ -226,11 +226,16 @@ void parse_args(py::list& args, py::list do_not_specialize, const std::string& f
// copy param // copy param
std::memcpy(params_ptr, &value, 8); std::memcpy(params_ptr, &value, 8);
params_ptr += 8; params_ptr += 8;
// udpate cache key // update cache key
cache_key += dtype_cache_key_part(arg.attr("dtype")); cache_key += dtype_cache_key_part(arg.attr("dtype"));
cache_key += "*"; cache_key += "*";
cache_key += "[multipleof("; cache_key += "[multipleof(";
size_t range_size = get_pointer_range_size(value); size_t range_size;
try {
range_size = get_pointer_range_size(value);
} catch (...) {
throw std::runtime_error("argument tensor #" + std::to_string(i) + " is not on cuda! " + std::string(py::str(arg)));
}
cache_key += std::to_string(std::min(pow2_divisor(value), pow2_divisor(range_size))); cache_key += std::to_string(std::min(pow2_divisor(value), pow2_divisor(range_size)));
cache_key += ")]"; cache_key += ")]";
continue; continue;