[Bug target/86040] [avr]: RAMPZ is not always cleared after loading __flashN data
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86040 Georg-Johann Lay changed: What|Removed |Added Status|NEW |RESOLVED Resolution|--- |FIXED --- Comment #11 from Georg-Johann Lay --- Fixed in v7.5, v8.4 and v9.3+
[Bug target/86040] [avr]: RAMPZ is not always cleared after loading __flashN data
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86040 --- Comment #10 from Georg-Johann Lay --- Author: gjl Date: Fri Oct 18 09:16:16 2019 New Revision: 277149 URL: https://gcc.gnu.org/viewcvs?rev=277149=gcc=rev Log: Backport from 2019-10-18 trunk r277143. PR target/86040 * config/avr/avr.c (avr_out_lpm): Do not shortcut-return. Modified: branches/gcc-7-branch/gcc/ChangeLog branches/gcc-7-branch/gcc/config/avr/avr.c
[Bug target/86040] [avr]: RAMPZ is not always cleared after loading __flashN data
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86040 --- Comment #9 from Georg-Johann Lay --- Author: gjl Date: Fri Oct 18 09:12:34 2019 New Revision: 277148 URL: https://gcc.gnu.org/viewcvs?rev=277148=gcc=rev Log: Backport from 2019-10-18 trunk r277143. PR target/86040 * config/avr/avr.c (avr_out_lpm): Do not shortcut-return. Modified: branches/gcc-8-branch/gcc/ChangeLog branches/gcc-8-branch/gcc/config/avr/avr.c
[Bug target/86040] [avr]: RAMPZ is not always cleared after loading __flashN data
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86040 --- Comment #8 from Georg-Johann Lay --- Author: gjl Date: Fri Oct 18 09:10:20 2019 New Revision: 277147 URL: https://gcc.gnu.org/viewcvs?rev=277147=gcc=rev Log: Backport from 2019-10-18 trunk r277143. PR target/86040 * config/avr/avr.c (avr_out_lpm): Do not shortcut-return. Modified: branches/gcc-9-branch/gcc/ChangeLog branches/gcc-9-branch/gcc/config/avr/avr.c
[Bug target/86040] [avr]: RAMPZ is not always cleared after loading __flashN data
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86040 --- Comment #7 from Georg-Johann Lay --- Author: gjl Date: Fri Oct 18 06:53:34 2019 New Revision: 277143 URL: https://gcc.gnu.org/viewcvs?rev=277143=gcc=rev Log: PR target/86040 * config/avr/avr.c (avr_out_lpm): Do not shortcut-return. Modified: trunk/gcc/ChangeLog trunk/gcc/config/avr/avr.c
[Bug target/86040] [avr]: RAMPZ is not always cleared after loading __flashN data
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86040 --- Comment #6 from Georg-Johann Lay --- Created attachment 47009 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=47009=edit proposed patch
[Bug target/86040] [avr]: RAMPZ is not always cleared after loading __flashN data
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86040 --- Comment #5 from Georg-Johann Lay --- Created attachment 44416 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=44416=edit C test case for movmem The movmem from ASes __flash1 ... __flash5 is also affected. As the place to fix I'd propose the output function avr.c::avr_out_movmem().
[Bug target/86040] [avr]: RAMPZ is not always cleared after loading __flashN data
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86040 --- Comment #4 from Georg-Johann Lay --- Created attachment 44412 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=44412=edit C test case.
[Bug target/86040] [avr]: RAMPZ is not always cleared after loading __flashN data
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86040
--- Comment #3 from Georg-Johann Lay ---
...and here is code that triggers the wrong path of the 2-byte case:
typedef struct S
{
const __flash2 struct S *p;
struct S *q;
} S;
const __flash2 S* func2 (const S *s)
{
return s->p->q->p;
}
$ avr-gcc -std=gnu99 -mmcu=atxmega128a1 -S foo.c -Os -dp
func2:
movw r26,r24 ; 20 [c=4 l=1] *movhi/0
ld r30,X+; 6 [c=8 l=2] *movhi/2
ld r31,X
adiw r30,2 ; 8 [c=4 l=1] addhi3_clobber/0
ldi r18,2; 9 [c=8 l=5] *movhi/2
out __RAMPZ__,r18
elpm r0,Z+
elpm r31,Z
mov r30,r0
ld r24,Z ; 15 [c=8 l=2] *movhi/2
ldd r25,Z+1
/* epilogue start */
ret ; 23 [c=0 l=1] return
Insn 9 is missing the reset of RAMPZ so that insn 15 loads with a high-byte of
2.
[Bug target/86040] [avr]: RAMPZ is not always cleared after loading __flashN data
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86040
Georg-Johann Lay changed:
What|Removed |Added
Keywords||wrong-code
Status|UNCONFIRMED |NEW
Last reconfirmed||2018-07-16
Ever confirmed|0 |1
--- Comment #2 from Georg-Johann Lay ---
IMO the problem are the early returns in avr.c::avr_out_lpm() that bypass the
reset to 0 of RAMPZ at the end of that function if RAMPD is present:
http://gcc.gnu.org/viewcvs/gcc/trunk/gcc/config/avr/avr.c?view=markup=257301#l3703
Affected is code for the devices in -mmcu=avrxmega5 and avrxmega7. A test case
for 1-byte reads is
char func1 (const __flash2 char *p)
{
return *p;
}
func1:
movw r30,r24 ; 16 [c=4 l=1] *movhi/0
ldi r18,2; 11 [c=4 l=3] movqi_insn/3
out __RAMPZ__,r18
elpm r24,Z
/* epilogue start */
ret ; 19 [c=0 l=1] return
I found no code to cover the wrong parts of the 2-byte and 4-byte reads.
[Bug target/86040] [avr]: RAMPZ is not always cleared after loading __flashN data
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86040 --- Comment #1 from Luke Morrison --- Created attachment 44231 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=44231=edit Compiled output from avr-gcc 8.1.0 of the test case.
