https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89809
Bug ID: 89809
Summary: movzwl is not utilized when uint16_t is loaded with
bit-shifts (while memcpy does)
Product: gcc
Version: 9.0
Status: UNCONFIRMED
Keywords: missed-optimization
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: nok.raven at gmail dot com
Target Milestone: ---
Host: x86_64
Target: x86_64
#include <cstdint>
#include <cstring>
// could be just a single movzwl
std::uint16_t foo(unsigned char const* p)
{
return static_cast<std::uint16_t>(p[0])
| (static_cast<std::uint16_t>(p[1]) << 8)
;
}
// the same, but for uint32_t is recognized
std::uint32_t bar(unsigned char const* p)
{
return static_cast<std::uint32_t>(p[0])
| (static_cast<std::uint32_t>(p[1]) << 8)
| (static_cast<std::uint32_t>(p[2]) << 16)
| (static_cast<std::uint32_t>(p[3]) << 24)
;
}
// memcpy for uint16_t is also good, and generates movzwl
std::uint16_t qaz(unsigned char const* p)
{
std::uint16_t tmp = 0;
std::memcpy(&tmp, p, sizeof(tmp));
return tmp;
}
https://godbolt.org/z/ZQl1tF
movzwl could be also utilized in other byte-size integers load. See the
attachment or https://godbolt.org/z/PQwc87
