https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95435
Bug ID: 95435
Summary: bad builtin memcpy performance with znver1/znver2 and
32bit
Product: gcc
Version: 10.1.1
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: jan at jki dot io
Target Milestone: ---
Created attachment 48641
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=48641&action=edit
gcc -g -m32 -march=znver2 -O1 -s testmem_modified.c -o tm32
Hi,
I have found a regression with znver2 and 32bit compiles regarding builtin
memcpy.
Test program is this one:
https://github.com/level1wendell/memcpy_sse
results:
gcc -g -m32 -march=znver2 -O1 -s testmem_modified.c -o tm32
32 MB = 2.462717 ms
gcc -g -m32 -march=skylake -O1 -s testmem_modified.c -o tm32
32 MB = 1.135762 ms
gcc -fno-builtin-memcpy -g -m32 -march=znver2 -O1 -s testmem_modified.c -o
tm32
32 MB = 1.138656 ms
I have attached the generated assembler code for the first 2 cases.
