[llvm-bugs] [Bug 46237] New: Size regression -Os constant propagation difference results in larger code
via llvm-bugs
llvm-bugs at lists.llvm.org
Sun Jun 7 17:33:41 PDT 2020
https://bugs.llvm.org/show_bug.cgi?id=46237
Bug ID: 46237
Summary: Size regression -Os constant propagation difference
results in larger code
Product: clang
Version: trunk
Hardware: PC
OS: Linux
Status: NEW
Severity: enhancement
Priority: P
Component: LLVM Codegen
Assignee: unassignedclangbugs at nondot.org
Reporter: michaeljclark at mac.com
CC: llvm-bugs at lists.llvm.org, neeilans at live.com,
richard-llvm at metafoo.co.uk
-Os is using more space than -O2 for this example.
--- BEGIN SCRIPT ---
cat >f.c <<EOF
typedef unsigned long long u64;
u64 f(u64 x, u64 y, u64 z)
{
return ((z & 3u) << 6) | ((y & 7u) << 3) | (z & 7u);
}
EOF
clang -O2 -c -o O2.o f.c
clang -Os -c -o Os.o f.c
objdump -d O2.o
objdump -d Os.o
--- END SCRIPT ---
It seems that lifting a small immediate constant into a register takes more
space due to the requirement of a 32-bit immediate for register immediate
constant synthesis on x86, whereas the inlined constant fits within an imm8
operand. It seems more profitable to not hoist constants <= 255 which are used
as operands in simple arith/logical ops 2 or 3 times.
O2.o: <f>:
0: 89 d0 mov %edx,%eax
2: c1 e0 06 shl $0x6,%eax
5: 0f b6 c8 movzbl %al,%ecx
8: 83 e6 07 and $0x7,%esi
b: 83 e2 07 and $0x7,%edx
e: 48 8d 04 f2 lea (%rdx,%rsi,8),%rax
12: 48 09 c8 or %rcx,%rax
15: c3 retq
Os.o: <f>:
0: 89 d0 mov %edx,%eax
2: c1 e0 06 shl $0x6,%eax
5: 0f b6 c8 movzbl %al,%ecx
8: b8 07 00 00 00 mov $0x7,%eax
d: 48 21 c6 and %rax,%rsi
10: 48 21 c2 and %rax,%rdx
13: 48 8d 04 f2 lea (%rdx,%rsi,8),%rax
17: 48 09 c8 or %rcx,%rax
1a: c3 retq
$ clang --version
Ubuntu clang version
11.0.0-++20200607081810+e664d0543f8-1~exp1~20200607063338.223
Target: x86_64-pc-linux-gnu
Thread model: posix
InstalledDir: /usr/bin
--
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20200608/13fa1619/attachment-0001.html>
More information about the llvm-bugs
mailing list