https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70140
Bug ID: 70140
Summary: Inefficient expansion of __builtin_mempcpy
Product: gcc
Version: 6.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: middle-end
Assignee: unassigned at gcc dot gnu.org
Reporter: wdijkstr at arm dot com
Target Milestone: ---
The expansion of __builtin_mempcpy is inefficient on many targets (eg. AArch64,
ARM, PPC). The issue is due to not using the same expansion options that memcpy
uses in builtins.c. As a result GCC6 produces for __builtin_mempcpy(x, y, 32):
PPC:
0: 38 a0 00 20 li r5,32
4: 48 00 00 00 b 4 <foo+0x4>
4: R_PPC_REL24 mempcpy
8: 60 00 00 00 nop
c: 60 42 00 00 ori r2,r2,0
AArch64:
mov x2, 32
b mempcpy
A second issue is that GCC always calls mempcpy. mempcpy is not supported or
implemented efficiently in many (if not most) library/target combinations.
GLIBC only has 3 targets which implement an optimized mempcpy, so GLIBC
currently inlines mempcpy into memcpy by default unless a target explicitly
disables this. It seems better to do this in GCC so it works for all libraries.
