https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111500
Bug ID: 111500
Summary: [arm-none-eabi-gcc] / suboptimal optimization
Product: gcc
Version: unknown
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c
Assignee: unassigned at gcc dot gnu.org
Reporter: gnu.arne at wgboome dot org
Target Milestone: ---
Hi!
For my STM32G030K6T I need thumb2 machine code.
arm-none-eabi-gcc can convert my C code quite nicely.
When I inspected the resulting disassembled machine code, I found some funny
artifacts.
Is it worth making&sending minimal source code and/or that corresponding
"preprocessed file *.i*"?
I mean: Is somebody able/willing to do something about it?
Here are the 4 artifacts (when using "-O3 -g"):
1.
a "subs r2, #1" instruction is followed by a "cmp r2, #0 / bne.n" instruction,
although the "subs" already leaves the zero flag properly, doesn't it?
2.
a "b.n" instruction jumps to a "bx lr" instruction, although both have the same
length... i mean: instead of "b.n" it could have written "bx lr"...
3.
a "uxth r2, r2" is followed by a "strh r2, [r3]", although "strh" doesn't look
at the high half-word of r2, does it?
4.
when i use single bits in a struct (e. g.: "struct A { int a:1; int b:7; }"),
gcc reserves 16 bytes on the stack and sometimes writes into this area, without
ever reading from there... when i do it myself with "struct A { int ab; }" and
"if (S.ab&128)..." and "if (S.ab&127)..." then it doesnt touch the stack...
Thx.
Bye.
