| Message ID | 20200227171447.1793041-1-maurosr@linux.vnet.ibm.com |
|---|---|
| State | Accepted |
| Headers | show |
| Series | Honor DEAD_CODE_ELIMINATION flag | expand |
| Context | Check | Description |
|---|---|---|
| snowpatch_ozlabs/apply_patch | success | Successfully applied on branch master (82aed17a5468aff6b600ee1694a10a60f942c018) |
| snowpatch_ozlabs/snowpatch_job_snowpatch-skiboot | success | Test snowpatch/job/snowpatch-skiboot on branch master |
| snowpatch_ozlabs/snowpatch_job_snowpatch-skiboot-dco | success | Signed-off-by present |
Mauro S. M. Rodrigues's on February 28, 2020 3:14 am: > While trying to reduce the size of the final binary I found > DEAD_CODE_ELIMINATION=1 but it didn't change the binary size and > known ununsed functions were seen when inspecting the elf with nm. > > Even though the necessary parameters for compiler, -ffunction-sections > and -fdata-sections, are set, ld's --gc-sections wasn't, so add it in > order to honor the flag. > > Signed-off-by: Mauro S. M. Rodrigues <maurosr@linux.vnet.ibm.com> > --- > Makefile.main | 1 + > 1 file changed, 1 insertion(+) > > diff --git a/Makefile.main b/Makefile.main > index 7ff2d21be..52251aa7e 100644 > --- a/Makefile.main > +++ b/Makefile.main > @@ -171,6 +171,7 @@ endif > > ifeq ($(DEAD_CODE_ELIMINATION),1) > LDFLAGS += -Wl,--gc-sections > +LDFLAGS_FINAL += --gc-sections > endif > > AFLAGS := -D__ASSEMBLY__ -m64 Obviously required, I got the same thing lying around. Reviewed-by: Nicholas Piggin <npiggin@gmail.com> Did you get much savings? It wasn't a great deal when I last tried. Thanks, Nick
On Fri, Feb 28, 2020 at 12:00:44PM +1000, Nicholas Piggin wrote: > > Obviously required, I got the same thing lying around. > > Reviewed-by: Nicholas Piggin <npiggin@gmail.com> > > Did you get much savings? It wasn't a great deal when I last tried. > Thanks for your review Nick! In my last test I was able to save 55KiB out of 479KiB, so ~11%. But please bear in mind that this was achieved testing on top of some TSS changes we're developing - including mbedtls and ibmtpm20tss libraries as subtrees. Testing at v6.0.21 it is indeed irrelevant, I was able to save just 1KiB. Thanks, Mauro > Thanks, > Nick
Mauro Rodrigues's on February 29, 2020 12:54 am: > On Fri, Feb 28, 2020 at 12:00:44PM +1000, Nicholas Piggin wrote: >> >> Obviously required, I got the same thing lying around. >> >> Reviewed-by: Nicholas Piggin <npiggin@gmail.com> >> >> Did you get much savings? It wasn't a great deal when I last tried. >> > > Thanks for your review Nick! > > In my last test I was able to save 55KiB out of 479KiB, so ~11%. But > please bear in mind that this was achieved testing on top of some TSS > changes we're developing - including mbedtls and ibmtpm20tss libraries > as subtrees. > > Testing at v6.0.21 it is indeed irrelevant, I was able to save just > 1KiB. Re-tested with upstream plus your patch text data bss dec filename 1486029 259992 48552 1794573 skiboot.elf 1409501 270736 40080 1720317 skiboot.elf.dce ( -4.1%) 1302629 246784 48552 1597965 skiboot.elf.le ( -9.6%) 1243629 239928 40080 1523637 skiboot.elf.le.dce (-15.1%) Not too bad with LE :) Interesting how the data section gets a bit larger with DCE with BE, I didn't look why that is. Thanks, Nick
On Fri, Feb 28, 2020 at 4:15 AM Mauro S. M. Rodrigues <maurosr@linux.vnet.ibm.com> wrote: > > While trying to reduce the size of the final binary I found > DEAD_CODE_ELIMINATION=1 but it didn't change the binary size and > known ununsed functions were seen when inspecting the elf with nm. > > Even though the necessary parameters for compiler, -ffunction-sections > and -fdata-sections, are set, ld's --gc-sections wasn't, so add it in > order to honor the flag. > > Signed-off-by: Mauro S. M. Rodrigues <maurosr@linux.vnet.ibm.com> Thanks, merged as 22817f8d563a069fa9be5202095fd88ddadb21f6
diff --git a/Makefile.main b/Makefile.main index 7ff2d21be..52251aa7e 100644 --- a/Makefile.main +++ b/Makefile.main @@ -171,6 +171,7 @@ endif ifeq ($(DEAD_CODE_ELIMINATION),1) LDFLAGS += -Wl,--gc-sections +LDFLAGS_FINAL += --gc-sections endif AFLAGS := -D__ASSEMBLY__ -m64
While trying to reduce the size of the final binary I found DEAD_CODE_ELIMINATION=1 but it didn't change the binary size and known ununsed functions were seen when inspecting the elf with nm. Even though the necessary parameters for compiler, -ffunction-sections and -fdata-sections, are set, ld's --gc-sections wasn't, so add it in order to honor the flag. Signed-off-by: Mauro S. M. Rodrigues <maurosr@linux.vnet.ibm.com> --- Makefile.main | 1 + 1 file changed, 1 insertion(+)