https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116453
Bug ID: 116453
Summary: Large code when initializing array using user defined
literals
Product: gcc
Version: 14.2.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: JoernRath at gmx dot de
Target Milestone: ---
The following code returns a float array initialized from an array of uint32_t
values which are the binary representation of the target float values. The
conversion to float is done using a user defined literal. With "-O3" for every
four values two instructions are generated in addition to the stored constants.
This can be optimized by storing the constants below a single label and loading
the data using a loop.
float operator""_F(unsigned long long int float_data)
{
float tmp{};
std::memcpy(&tmp, &float_data, sizeof(float));
return tmp;
}
std::array<float, 168> foo()
{
std::array<float, num_entries> a = {
0x3B479C6D_F, ... ,0x3B47A0FE_F,
};
return a;
}
See https://godbolt.org/z/7avvhYnW8 for a full example and compiler
configuration.
