https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95834
Bug ID: 95834
Summary: x86 immediates --- some redundant
Product: gcc
Version: 9.3.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: zero at smallinteger dot com
Target Milestone: ---
Created attachment 48774
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=48774&action=edit
sample code
In some cases, literals could be synthesized with simple instructions instead
of loading them from scratch. Consider the case of foo(), with 4 literals that
can be obtained by shifting the first one: gcc emit 5 eight-byte literals
instead.
In some other cases, the same literals are loaded. Consider the case of baz(),
where 6 is loaded twice in consecutive instructions when an instruction reorder
could eliminate one of the literals.
The Intel optimization manual suggests not to have many instructions that load
literals in a row due to decreased code density, more difficulty to cache the
decoded instructions due to lack of space, and because shorter instruction
encodings are better.
Verified with gcc 9.3.0 and gcc 10.x (at godbolt), with -O3.
