On Mon, 2012-10-08 at 09:45 +0900, Kaz Kojima wrote:
> Oleg Endo <oleg.e...@t-online.de> wrote:
> > The attached patch improves comparisons such as
> > 'unsigned int <= 0x7FFFFFFF' on SH.
> > As mentioned in the PR, for some reason, those comparisons do not go
> > through the cstore expander. As a consequence the comparison doesn't
> > get the chance to be canonicalized by the target code and ends up as
> > '(~x) >> 31'.
> > I've not investigated this further and just fixed the symptoms on SH. I
> > don't know whether it's also an issue on other targets.
> >
> > Tested on rev 192142 with
> > make -k check RUNTESTFLAGS="--target_board=sh-sim
> > \{-m2/-ml,-m2/-mb,-m2a/-mb,-m4/-ml,-m4/-mb,-m4a/-ml,-m4a/-mb}"
> >
> > and no new failures.
> > OK?
>
> I've run CSiBE with and without the patch for sh4-unknown-linux-gnu
> at -O2. Only one difference in the resulted sizes: jpeg-6b/jcphuff
> increases 5336 bytes to 5340 bytes with the patch. Could you look
> into it?
Yep, that's actually the only place in the CSiBE set where this case
hits. The function in question is encode_mcu_AC_refine. The increase
seems to be due to different register allocation and different spill
code :T
I've attached the asm diff.
Cheers,
Oleg
--- CSiBE/m4-single-ml-O2-trunk/jpeg-6b/jcphuff.s
+++ CSiBE/m4-single-ml-O2/jpeg-6b/jcphuff.s
@@ -2147,7 +2147,7 @@
bt/s .L611
mov.l r2,@(24,r15)
bra .L612
- mov.l @(44,r15),r0
+ mov.l @(44,r15),r3
.L611:
mov.l .L565,r4
mov r2,r5
@@ -2513,21 +2513,21 @@
mov r0,r1
mov r9,r0
and r2,r1
- mov.l @(24,r15),r3
+ mov.l @(28,r15),r3
mov.b r1,@(r0,r8)
mov r9,r11
- mov.l @(28,r15),r0
- add #1,r3
- mov.l @(36,r15),r1
+ mov.l @(24,r15),r2
+ add #4,r3
+ mov.l @(36,r15),r0
add #1,r11
- mov.l @(40,r15),r2
+ mov.l @(40,r15),r1
+ add #1,r2
add #4,r0
- add #4,r1
- mov.l r3,@(24,r15)
- mov.l r0,@(28,r15)
- cmp/ge r3,r2
+ mov.l r2,@(24,r15)
+ mov.l r3,@(28,r15)
+ cmp/ge r2,r1
bt/s .L599
- mov.l r1,@(36,r15)
+ mov.l r0,@(36,r15)
tst r11,r11
bt/s .L555
mov r12,r14
@@ -2545,21 +2545,23 @@
mov.w .L578,r3
cmp/hi r3,r2
bf/s .L612
- mov.l @(44,r15),r0
+ mov.l @(44,r15),r3
.L515:
mov.l .L582,r2
jsr @r2
mov r14,r4
.L459:
+ mov.l @(44,r15),r3
+.L612:
mov.l @(44,r15),r0
-.L612:
+ mov.l @(24,r3),r2
mov.l @(16,r14),r3
- mov.l @(24,r0),r2
mov.l r3,@r2
mov.l @(20,r14),r3
mov.l r3,@(4,r2)
mov.w .L580,r2
- mov.l @(r0,r2),r2
+ add r0,r2
+ mov.l @(8,r2),r2
tst r2,r2
bt .L544
add #64,r14
@@ -2594,18 +2596,18 @@
add #1,r2
mov.l r2,@(16,r15)
.L467:
- mov.l @(24,r15),r3
- mov.l @(28,r15),r0
- mov.l @(36,r15),r1
- add #1,r3
- mov.l @(40,r15),r2
+ mov.l @(24,r15),r2
+ mov.l @(28,r15),r3
+ mov.l @(36,r15),r0
+ add #1,r2
+ mov.l @(40,r15),r1
+ add #4,r3
add #4,r0
- add #4,r1
- mov.l r3,@(24,r15)
- mov.l r0,@(28,r15)
- cmp/ge r3,r2
+ mov.l r2,@(24,r15)
+ mov.l r3,@(28,r15)
+ cmp/ge r2,r1
bf/s .L603
- mov.l r1,@(36,r15)
+ mov.l r0,@(36,r15)
.L599:
bra .L617
mov.l @(28,r15),r1
@@ -2614,8 +2616,8 @@
bf/s .L523
mov r12,r14
.L555:
- mov.l @(16,r15),r3
- cmp/pl r3
+ mov.l @(16,r15),r2
+ cmp/pl r2
bf .L459
mov.l @(56,r14),r3
bra .L625
@@ -2642,13 +2644,13 @@
add #1,r2
mov.l r2,@r3
.L511:
- mov.l @(20,r15),r1
+ mov.l @(20,r15),r0
.L620:
- mov #0,r2
+ mov #0,r1
mov #0,r11
- mov.l r2,@(16,r15)
+ mov.l r1,@(16,r15)
bra .L467
- mov.l @(0,r1),r8
+ mov.l @(0,r0),r8
.align 1
.L522:
bra .L619
@@ -2659,7 +2661,7 @@
.L578:
.short 937
.L580:
- .short 196
+ .short 188
.L581:
.short 312
.L583:
@@ -2728,16 +2730,15 @@
tst r3,r3
mov.l r14,@(28,r12)
mov.l @r1,r0
- mov.l @(52,r15),r2
+ mov.l @(52,r15),r1
add r0,r0
mov.l r11,@(24,r12)
bf/s .L511
- mov.w @(r0,r2),r1
- not r1,r1
- mov r14,r10
- shll r1
+ mov.w @(r0,r1),r2
+ cmp/pz r2
neg r14,r3
movt r1
+ mov r14,r10
add #23,r3
shld r3,r1
add #1,r10
@@ -2784,7 +2785,7 @@
mov r9,r6
.L601:
bra .L620
- mov.l @(20,r15),r1
+ mov.l @(20,r15),r0
.align 1
.L556:
mov.l .L589,r1
@@ -2812,9 +2813,9 @@
add #-8,r10
.align 1
.L558:
- mov.l .L589,r3
+ mov.l .L589,r2
mov r12,r4
- jsr @r3
+ jsr @r2
mov.l r1,@(4,r15)
mov.l @(4,r15),r1
cmp/eq r13,r1
@@ -2830,8 +2831,8 @@
dt r2
bf/s .L507
mov.l r2,@(20,r12)
- mov.l .L589,r0
- jsr @r0
+ mov.l .L589,r3
+ jsr @r3
mov r12,r4
bra .L622
add #-8,r10
