[Bug target/101129] [11/12 Regression] wrong code at -O1 since r11-5839
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101129 Bill Schmidt changed: What|Removed |Added Status|NEW |RESOLVED Resolution|--- |FIXED Known to fail|11.1.1 | --- Comment #14 from Bill Schmidt --- Now fixed everywhere.
[Bug target/101129] [11/12 Regression] wrong code at -O1 since r11-5839
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101129 --- Comment #13 from CVS Commits --- The releases/gcc-9 branch has been updated by William Schmidt : https://gcc.gnu.org/g:3e4b43ec9a937866ead88e996bd200b116a1766c commit r9-9632-g3e4b43ec9a937866ead88e996bd200b116a1766c Author: Bill Schmidt Date: Mon Jul 19 16:14:36 2021 -0500 rs6000: Don't let swaps pass break multiply low-part (PR101129) Backport from mainline. 2021-07-15 Bill Schmidt gcc/ PR target/101129 * config/rs6000/rs6000-p8swap.c (has_part_mult): New. (rs6000_analyze_swaps): Insns containing a subreg of a mult are not swappable. gcc/testsuite/ PR target/101129 * gcc.target/powerpc/pr101129.c: New.
[Bug target/101129] [11/12 Regression] wrong code at -O1 since r11-5839
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101129 --- Comment #12 from CVS Commits --- The releases/gcc-10 branch has been updated by William Schmidt : https://gcc.gnu.org/g:ecad28561c47f5b2652a03e9b67cf37c13d88f53 commit r10-9990-gecad28561c47f5b2652a03e9b67cf37c13d88f53 Author: Bill Schmidt Date: Mon Jul 19 14:38:23 2021 -0500 rs6000: Don't let swaps pass break multiply low-part (PR101129) Backport from mainline. 2021-07-15 Bill Schmidt gcc/ PR target/101129 * config/rs6000/rs6000-p8swap.c (has_part_mult): New. (rs6000_analyze_swaps): Insns containing a subreg of a mult are not swappable. gcc/testsuite/ PR target/101129 * gcc.target/powerpc/pr101129.c: New.
[Bug target/101129] [11/12 Regression] wrong code at -O1 since r11-5839
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101129 --- Comment #11 from CVS Commits --- The releases/gcc-11 branch has been updated by William Schmidt : https://gcc.gnu.org/g:ac0efe3c6fc6231b20ffd684956a4a5c3c54a96b commit r11-8780-gac0efe3c6fc6231b20ffd684956a4a5c3c54a96b Author: Bill Schmidt Date: Mon Jul 19 12:49:17 2021 -0500 rs6000: Don't let swaps pass break multiply low-part (PR101129) Backport from mainline. 2021-07-15 Bill Schmidt gcc/ PR target/101129 * config/rs6000/rs6000-p8swap.c (has_part_mult): New. (rs6000_analyze_swaps): Insns containing a subreg of a mult are not swappable. gcc/testsuite/ PR target/101129 * gcc.target/powerpc/pr101129.c: New.
[Bug target/101129] [11/12 Regression] wrong code at -O1 since r11-5839
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101129 Bill Schmidt changed: What|Removed |Added Known to fail|12.0| Known to work||12.0 --- Comment #10 from Bill Schmidt --- Fixed on trunk, but will require backports.
[Bug target/101129] [11/12 Regression] wrong code at -O1 since r11-5839
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101129 --- Comment #9 from CVS Commits --- The master branch has been updated by William Schmidt : https://gcc.gnu.org/g:ad5f8ac1d2f2dc92d43663243b52f9e9eb3cf7c0 commit r12-2325-gad5f8ac1d2f2dc92d43663243b52f9e9eb3cf7c0 Author: Bill Schmidt Date: Thu Jul 15 10:16:17 2021 -0500 rs6000: Don't let swaps pass break multiply low-part (PR101129) 2021-07-15 Bill Schmidt gcc/ PR target/101129 * config/rs6000/rs6000-p8swap.c (has_part_mult): New. (rs6000_analyze_swaps): Insns containing a subreg of a mult are not swappable. gcc/testsuite/ PR target/101129 * gcc.target/powerpc/pr101129.c: New.
[Bug target/101129] [11/12 Regression] wrong code at -O1 since r11-5839
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101129 --- Comment #8 from Bill Schmidt --- Small change required to actually check that it's a SET insn. (oops) Otherwise looks like it passed regstrap. Testing the revised patch now. Thanks for the pre-review!
[Bug target/101129] [11/12 Regression] wrong code at -O1 since r11-5839
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101129 --- Comment #7 from Segher Boessenkool --- That patch is pre-approved (if it works ;-) ) Bill. Thanks!
[Bug target/101129] [11/12 Regression] wrong code at -O1 since r11-5839
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101129
--- Comment #6 from Bill Schmidt ---
Testing this patch.
diff --git a/gcc/config/rs6000/rs6000-p8swap.c
b/gcc/config/rs6000/rs6000-p8swap.c
index 21cbcb2e28a..00693e6dc60 100644
--- a/gcc/config/rs6000/rs6000-p8swap.c
+++ b/gcc/config/rs6000/rs6000-p8swap.c
@@ -1523,6 +1523,20 @@ replace_swap_with_copy (swap_web_entry *insn_entry,
unsigned i)
insn->set_deleted ();
}
+/* INSN is known to contain a SUBREG, which we can normally handle,
+ but if the SUBREG itself contains a MULT then we need to leave it alone
+ to avoid turning a mult_hipart into a mult_lopart, for example. */
+static bool
+has_part_mult (rtx_insn *insn)
+{
+ rtx body = PATTERN (insn);
+ rtx src = SET_SRC (body);
+ if (GET_CODE (src) != SUBREG)
+return false;
+ rtx inner = XEXP (src, 0);
+ return (GET_CODE (inner) == MULT);
+}
+
/* Make NEW_MEM_EXP's attributes and flags resemble those of
ORIGINAL_MEM_EXP. */
static void
@@ -2501,6 +2515,9 @@ rs6000_analyze_swaps (function *fun)
insn_entry[uid].is_swappable = 0;
else if (special != SH_NONE)
insn_entry[uid].special_handling = special;
+ else if (insn_entry[uid].contains_subreg
+ && has_part_mult (insn))
+ insn_entry[uid].is_swappable = 0;
else if (insn_entry[uid].contains_subreg)
insn_entry[uid].special_handling = SH_SUBREG;
}
[Bug target/101129] [11/12 Regression] wrong code at -O1 since r11-5839
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101129 --- Comment #5 from Bill Schmidt --- Uh, yeah, that is completely unexpected behavior for swaps. I'll try to look at this soon.
[Bug target/101129] [11/12 Regression] wrong code at -O1 since r11-5839
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101129
--- Comment #4 from Jakub Jelinek ---
I think this is a bug in the swaps rs6000 specific pass.
It in particular changes
(insn 20 19 21 2 (set (reg:DI 155)
(mult:DI (reg:DI 224)
-(subreg:DI (reg:V16QI 148) 0))) "pr101129.c":14:8 154 {muldi3}
+(subreg:DI (reg:V16QI 148) 8))) "pr101129.c":14:8 154 {muldi3}
(nil))
(insn 21 20 22 2 (set (reg:DI 156)
(subreg:DI (mult:TI (zero_extend:TI (reg:DI 224))
-(zero_extend:TI (subreg:DI (reg:V16QI 148) 0))) 8))
"pr101129.c":14:8 166 {umuldi3_highpart_le}
+(zero_extend:TI (subreg:DI (reg:V16QI 148) 8))) 0))
"pr101129.c":14:8 166 {umuldi3_highpart_le}
(nil))
The SUBREG_BYTE changes in subregs of pseudo 148 look reasonable as the pass
removed a xxswapd_v16qi that was swapping the value set to that pseudo.
But the 8 to 0 change of SUBREG_BYTE with SUBREG_REG of MULT looks wrong, that
wouldn't even match unless simplified and turns a highpart multiply into
lowpart multiply.
So either the swaps pass needs to be fixed not to do that, or the highpart
multiply instructions should be changed to e.g. use lshiftrt by 64 instead of
highpart subreg of the result.
[Bug target/101129] [11/12 Regression] wrong code at -O1 since r11-5839
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101129
Richard Biener changed:
What|Removed |Added
CC||rguenth at gcc dot gnu.org
--- Comment #3 from Richard Biener ---
The GIMPLE doesn't look wrong but eventually
+ _4 = _2 < { 5, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
+ _5 = VIEW_CONVERT_EXPR(_4);
goes wrong somewhere in RTL expansion?
[Bug target/101129] [11/12 Regression] wrong code at -O1 since r11-5839
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101129
--- Comment #2 from Jakub Jelinek ---
Optimized difference:
--- pr101129.c.244t.optimized.r11-5838 2021-06-18 22:40:20.915971652 +0200
+++ pr101129.c.244t.optimized.r11-5839 2021-06-18 22:40:34.739789553 +0200
@@ -8,6 +8,7 @@ void foo0 (u32 u32_0, U * ret)
__int128 _1;
vector(16) unsigned char _2;
unsigned char _3;
+ vector(16) _4;
vector(16) signed char _5;
__int128 _6;
__int128 _7;
@@ -31,7 +32,8 @@ void foo0 (u32 u32_0, U * ret)
_14 = u32_0_13(D) & 4;
_3 = (unsigned char) _14;
_2 = {_3, _3, _3, _3, _3, _3, _3, _3, _3, _3, _3, _3, _3, _3, _3, _3};
- _5 = .VCONDU (_2, { 5, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }, { -1,
-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1 }, { 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }, 108);
+ _4 = _2 < { 5, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
+ _5 = VIEW_CONVERT_EXPR(_4);
_6 = VIEW_CONVERT_EXPR<__int128>(_5);
u128_2_15 = _1 * _6;
_7 = u128_2_15 >> 64;
Assembly difference:
--- pr101129.s.r11-5838 2021-06-18 22:40:20.919971599 +0200
+++ pr101129.s.r11-5839 2021-06-18 22:40:34.743789500 +0200
@@ -14,19 +14,16 @@ foo0:
.localentry foo0,.-foo0
rlwinm 9,3,0,29,29
mtvsrd 32,9
- vspltb 0,0,7
+ vspltb 1,0,7
addis 9,2,.LC0@toc@ha
addi 9,9,.LC0@toc@l
- lxvd2x 33,0,9
- vcmpgtub 0,1,0
- addi 9,1,-16
- xxpermdi 32,32,32,2
- stxvd2x 32,0,9
- ld 9,-8(1)
- ld 10,-16(1)
+ lxvd2x 32,0,9
+ vcmpgtub 0,0,1
+ mfvsrd 11,32
+ xxpermdi 0,32,32,3
+ mfvsrd 10,0
+ add 10,11,10
mulld 10,10,3
- mulhdu 3,3,9
- add 10,10,3
addis 9,2,.LANCHOR0@toc@ha
lbz 9,.LANCHOR0@toc@l(9)
add 10,10,9
[Bug target/101129] [11/12 Regression] wrong code at -O1 since r11-5839
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101129 Jakub Jelinek changed: What|Removed |Added Summary|[11/12 Regression] wrong|[11/12 Regression] wrong |code at -O1 |code at -O1 since r11-5839 Last reconfirmed||2021-06-18 Priority|P3 |P2 Ever confirmed|0 |1 CC||jakub at gcc dot gnu.org Status|UNCONFIRMED |NEW --- Comment #1 from Jakub Jelinek --- Started with r11-5839-g3a6e3ad38a17a03ee0139b49a0946e7b9ded1eb1
