Patchwork [alpha] : Improve sysdeps/alpha/divqu.S and sysdeps/alpha/remqu.S

login
register
mail settings
Submitter Uros Bizjak
Date Feb. 27, 2019, 7:54 p.m.
Message ID <CAFULd4acKuU=nVHNo_rgH4xbSQHN-DE0W=kfa5h9S3v83-rdrg@mail.gmail.com>
Download mbox | patch
Permalink /patch/31633/
State New
Headers show

Comments

Uros Bizjak - Feb. 27, 2019, 7:54 p.m.
Attached patch improves and unifies sysdeps/alpha/divqu.S and
sysdeps/alpha/remqu.S.

* sysdeps/alpha/divqu.S (__divqu): Move save of $f0 and excb after
conditional branch to DIVBYZERO.  Fix unwind info.
* sysdeps/alpha/remqu.S (__remqu): Move saves of $f0, $f1, $f2 and
excb after conditional branch to $powerof2.  Add missing unop
instructions and .align directives and reorder instructions to match
__divqu.

Signed-off-by: UroŇ° Bizjak <ubizjak@gmail.com>

Patch was tested on alphaev68-linux-gnu, also by running complete
libgo testsuite.

Uros.
Richard Henderson - April 1, 2019, 9:02 a.m.
On 2/28/19 2:54 AM, Uros Bizjak wrote:
> Attached patch improves and unifies sysdeps/alpha/divqu.S and
> sysdeps/alpha/remqu.S.
> 
> * sysdeps/alpha/divqu.S (__divqu): Move save of $f0 and excb after
> conditional branch to DIVBYZERO.  Fix unwind info.
> * sysdeps/alpha/remqu.S (__remqu): Move saves of $f0, $f1, $f2 and
> excb after conditional branch to $powerof2.  Add missing unop
> instructions and .align directives and reorder instructions to match
> __divqu.
> 
> Signed-off-by: UroŇ° Bizjak <ubizjak@gmail.com>

Thanks, committed.


r~

Patch

diff --git a/sysdeps/alpha/divqu.S b/sysdeps/alpha/divqu.S
index f5cedd0716..3165374b6d 100644
--- a/sysdeps/alpha/divqu.S
+++ b/sysdeps/alpha/divqu.S
@@ -56,10 +56,10 @@  __divqu:
 	   that's done, we have at least 22 cycles until its results are
 	   ready -- all the time in the world to figure out how we're
 	   going to use the results.  */
-	stt	$f0, 0(sp)
-	excb
 	beq	Y, DIVBYZERO
 
+	stt	$f0, 0(sp)
+	excb
 	stt	$f1, 8(sp)
 	stt	$f3, 48(sp)
 	cfi_rel_offset ($f0, 0)
@@ -70,6 +70,7 @@  __divqu:
 	_ITOFT2	X, $f0, 16, Y, $f1, 24
 	cvtqt	$f0, $f0
 	cvtqt	$f1, $f1
+
 	blt	X, $x_is_neg
 	divt/c	$f0, $f1, $f0
 
@@ -90,12 +91,12 @@  __divqu:
 
 	ldt	$f0, 0(sp)
 	ldt	$f3, 48(sp)
+	lda	sp, FRAME(sp)
 	cfi_remember_state
 	cfi_restore ($f0)
 	cfi_restore ($f1)
 	cfi_restore ($f3)
 	cfi_def_cfa_offset (0)
-	lda	sp, FRAME(sp)
 	ret	$31, (RA), 1
 
 	.align	4
diff --git a/sysdeps/alpha/remqu.S b/sysdeps/alpha/remqu.S
index a240ee9735..3b6a62dd88 100644
--- a/sysdeps/alpha/remqu.S
+++ b/sysdeps/alpha/remqu.S
@@ -57,19 +57,19 @@  __remqu:
 	   ready -- all the time in the world to figure out how we're
 	   going to use the results.  */
 	subq	Y, 1, AT
-	stt	$f0, 0(sp)
 	and	Y, AT, AT
+	beq	AT, $powerof2
 
-	stt	$f1, 8(sp)
+	stt	$f0, 0(sp)
 	excb
+	stt	$f1, 8(sp)
 	stt	$f3, 48(sp)
-	beq	AT, $powerof2
 	cfi_rel_offset ($f0, 0)
 	cfi_rel_offset ($f1, 8)
 	cfi_rel_offset ($f3, 48)
+	mf_fpcr	$f3
 
 	_ITOFT2	X, $f0, 16, Y, $f1, 24
-	mf_fpcr	$f3
 	cvtqt	$f0, $f0
 	cvtqt	$f1, $f1
 
@@ -116,11 +116,13 @@  $x_is_neg:
 	cfi_rel_offset ($f2, 24)
 	_ITOFS	AT, $f2, 16
 
+	.align	4
 	addt	$f0, $f2, $f0
+	unop
 	divt/c	$f0, $f1, $f0
+	unop
 
 	/* Ok, we've now the divide issued.  Continue with other checks.  */
-	.align	4
 	ldt	$f1, 8(sp)
 	unop
 	ldt	$f2, 24(sp)