Commit 5c22dd0f authored by Shucai Xiao's avatar Shucai Xiao
Browse files

Merge branch 'develop' of github.com:ROCmSoftwarePlatform/AMDMIGraphX into print_matmul_perf_flops

parents 4da0e5e6 b20e3d4d
......@@ -89,6 +89,13 @@ inline std::ptrdiff_t operator-(basic_iota_iterator<F, Iterator> x,
return x.index - y.index;
}
template <class F, class Iterator>
inline basic_iota_iterator<F, Iterator> operator-(basic_iota_iterator<F, Iterator> x,
std::ptrdiff_t y)
{
return x -= y;
}
template <class F, class Iterator>
inline bool operator==(basic_iota_iterator<F, Iterator> x, basic_iota_iterator<F, Iterator> y)
{
......
......@@ -183,6 +183,63 @@ void program::finalize()
mm->finalize(this->impl->ctx);
}
template <class T>
std::string classify(T x)
{
switch(std::fpclassify(x))
{
case FP_INFINITE: return "inf";
case FP_NAN: return "nan";
case FP_NORMAL: return "normal";
case FP_SUBNORMAL: return "subnormal";
case FP_ZERO: return "zero";
default: return "unknown";
}
}
std::unordered_set<std::string> classify_argument(const argument& a)
{
std::unordered_set<std::string> result;
a.visit(
[&](auto t) {
for(const auto& x : t)
result.insert(classify(x));
},
[&](const auto& xs) {
for(const auto& x : xs)
{
auto r = classify_argument(x);
result.insert(r.begin(), r.end());
}
});
return result;
}
void preview_argument(std::ostream& os, const argument& a)
{
a.visit(
[&](auto t) {
if(t.size() <= 10)
{
os << t;
}
else
{
os << to_string_range(t.begin(), t.begin() + 5);
os << ", ..., ";
os << to_string_range(t.end() - 5, t.end());
}
},
[&](const auto& xs) {
for(const auto& x : xs)
{
os << '{';
preview_argument(os, x);
os << '}';
}
});
}
template <class F>
std::vector<argument> generic_eval(const module* mod,
context& ctx,
......@@ -315,8 +372,21 @@ std::vector<argument> program::eval(parameter_map params) const
if(trace_level > 1 and ins->name().front() != '@' and
ins->name() != "load" and not result.empty())
{
target tgt = make_target(this->impl->target_name);
std::cout << "Output: " << tgt.copy_from(result) << std::endl;
target tgt = make_target(this->impl->target_name);
auto buffer = tgt.copy_from(result);
if(trace_level == 2)
{
std::cout << "Output has "
<< to_string_range(classify_argument(buffer))
<< std::endl;
std::cout << "Output: ";
preview_argument(std::cout, buffer);
std::cout << std::endl;
}
else
{
std::cout << "Output: " << buffer << std::endl;
}
}
return result;
}));
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment