* sysdeps/alpha/div_libc.h (_ITOFS, _ITOFT, _FTOIT, _ITOFT2): New. * sysdeps/alpha/divl.S, sysdeps/alpha/divq.S, sysdeps/alpha/divqu.S, sysdeps/alpha/reml.S, sysdeps/alpha/remq.S, sysdeps/alpha/remqu.S: Use them.

2004-05-06  Richard Henderson  <rth@redhat.com>

	* sysdeps/alpha/div_libc.h (_ITOFS, _ITOFT, _FTOIT, _ITOFT2): New.
	* sysdeps/alpha/divl.S, sysdeps/alpha/divq.S, sysdeps/alpha/divqu.S,
	sysdeps/alpha/reml.S, sysdeps/alpha/remq.S, sysdeps/alpha/remqu.S:
	Use them.

7 files changed, 81 insertions, 73 deletions

diff --git a/sysdeps/alpha/div_libc.h b/sysdeps/alpha/div_libc.h
index 98566435ce..27209bd9fa 100644
--- a/sysdeps/alpha/div_libc.h
+++ b/sysdeps/alpha/div_libc.h
@@ -111,3 +111,48 @@ DIVBYZERO:
 	cfi_endproc
 	.size	DIVBYZERO, .-DIVBYZERO
 .endm
+
+/* Like the ev6 instructions, but fall back to stack use on prior machines.  */
+
+	.arch	ev6
+
+.macro _ITOFS  gr, fr, slot
+#ifdef __alpha_fix__
+	itofs	\gr, \fr
+#else
+	stl	\gr, \slot($sp)
+	lds	\fr, \slot($sp)
+#endif
+.endm
+
+.macro _ITOFT  gr, fr, slot
+#ifdef __alpha_fix__
+	itoft	\gr, \fr
+#else
+	stq	\gr, \slot($sp)
+	ldt	\fr, \slot($sp)
+#endif
+.endm
+
+.macro _FTOIT  fr, gr, slot
+#ifdef __alpha_fix__
+	ftoit	\fr, \gr
+#else
+	stt	\fr, \slot($sp)
+	ldq	\gr, \slot($sp)
+#endif
+.endm
+
+/* Similarly, but move two registers.  Schedules better for pre-ev6.  */
+
+.macro _ITOFT2 gr1, fr1, slot1, gr2, fr2, slot2
+#ifdef __alpha_fix__
+	itoft	\gr1, \fr1
+	itoft	\gr2, \fr2
+#else
+	stq	\gr1, \slot1($sp)
+	stq	\gr2, \slot2($sp)
+	ldt	\fr1, \slot1($sp)
+	ldt	\fr2, \slot2($sp)
+#endif
+.endm
diff --git a/sysdeps/alpha/divl.S b/sysdeps/alpha/divl.S
index 33fa1187d9..90cd6862a7 100644
--- a/sysdeps/alpha/divl.S
+++ b/sysdeps/alpha/divl.S
@@ -48,25 +48,20 @@ __divl:
 
 	EXTEND	(X, RV)
 	EXTEND	(Y, AT)
-	stq	RV, 16(sp)
-	stq	AT, 24(sp)
-
-	ldt	$f0, 16(sp)
-	ldt	$f1, 24(sp)
+	_ITOFT2	RV, $f0, 16, AT, $f1, 24
 	cvtqt	$f0, $f0
 	cvtqt	$f1, $f1
-
 	divt/c	$f0, $f1, $f0
 	cvttq/c	$f0, $f0
-	stt	$f0, 16(sp)
-	ldt	$f0, 0(sp)
+	_FTOIT	$f0, RV, 16
 
+	ldt	$f0, 0(sp)
 	ldt	$f1, 8(sp)
-	ldl	RV, 16(sp)
 	lda	sp, FRAME(sp)
 	cfi_restore ($f0)
 	cfi_restore ($f1)
 	cfi_def_cfa_offset (0)
+	sextl	RV, RV
 	ret	$31, (RA), 1
 
 	cfi_endproc
diff --git a/sysdeps/alpha/divq.S b/sysdeps/alpha/divq.S
index 464536db3d..4df79829f9 100644
--- a/sysdeps/alpha/divq.S
+++ b/sysdeps/alpha/divq.S
@@ -52,17 +52,13 @@ __divq:
 	   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.  */
-	stq	X, 16(sp)
-	stq	Y, 24(sp)
-	beq	Y, DIVBYZERO
-
 	stt	$f0, 0(sp)
 	stt	$f1, 8(sp)
+	beq	Y, DIVBYZERO
 	cfi_rel_offset ($f0, 0)
 	cfi_rel_offset ($f1, 8)
-	ldt	$f0, 16(sp)
-	ldt	$f1, 24(sp)
 
+	_ITOFT2	X, $f0, 16, Y, $f1, 24
 	cvtqt	$f0, $f0
 	cvtqt	$f1, $f1
 	divt/c	$f0, $f1, $f0
@@ -77,9 +73,8 @@ __divq:
 	/* If we get here, we're expecting exact results from the division.
 	   Do nothing else besides convert and clean up.  */
 	cvttq/c	$f0, $f0
-	stt	$f0, 16(sp)
+	_FTOIT	$f0, RV, 16
 
-	ldq	RV, 16(sp)
 	ldt	$f0, 0(sp)
 	cfi_restore ($f1)
 	cfi_remember_state
@@ -117,8 +112,8 @@ $x_big:
 $fix_sign_in_ret1:
 	cvttq/c	$f0, $f0
 
-	stt	$f0, 8(sp)
-	ldq	Q, 8(sp)
+	_FTOIT	$f0, Q, 8
+	.align	3
 $fix_sign_in_ret2:
 	mulq	Q, Y, QY
 	stq	t4, 8(sp)
@@ -234,10 +229,8 @@ $fix_sign_in:
 	blbc	t5, $fix_sign_in_ret1
 
 	cvttq/c	$f0, $f0
-	stt	$f0, 8(sp)
-	ldq	Q, 8(sp)
-	unop
-
+	_FTOIT	$f0, Q, 8
+	.align	3
 	negq	Q, Q
 	br	$fix_sign_in_ret2
 
diff --git a/sysdeps/alpha/divqu.S b/sysdeps/alpha/divqu.S
index 6ff6c035e2..63b575faa7 100644
--- a/sysdeps/alpha/divqu.S
+++ b/sysdeps/alpha/divqu.S
@@ -52,23 +52,20 @@ __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.  */
-	stq	X, 16(sp)
-	stq	Y, 24(sp)
-	beq	Y, DIVBYZERO
-
 	stt	$f0, 0(sp)
 	stt	$f1, 8(sp)
+	beq	Y, DIVBYZERO
 	cfi_rel_offset ($f0, 0)
 	cfi_rel_offset ($f1, 8)
-	ldt	$f0, 16(sp)
-	ldt	$f1, 24(sp)
 
+	_ITOFT2	X, $f0, 16, Y, $f1, 24
 	cvtqt	$f0, $f0
 	cvtqt	$f1, $f1
 	blt	X, $x_is_neg
 	divt/c	$f0, $f1, $f0
 
 	/* Check to see if Y was mis-converted as signed value.  */
+	.align	4
 	ldt	$f1, 8(sp)
 	unop
 	nop
@@ -81,9 +78,7 @@ __divqu:
 	/* If we get here, we're expecting exact results from the division.
 	   Do nothing else besides convert and clean up.  */
 	cvttq/c	$f0, $f0
-	stt	$f0, 16(sp)
-
-	ldq	RV, 16(sp)
+	_FTOIT	$f0, RV, 16
 	ldt	$f0, 0(sp)
 	cfi_remember_state
 	cfi_restore ($f0)
@@ -101,9 +96,9 @@ $x_is_neg:
 	ldah	AT, 0x5f80		/* 2**64 as float.  */
 	stt	$f2, 24(sp)
 	cfi_rel_offset ($f2, 24)
-	stl	AT, 16(sp)
-	lds	$f2, 16(sp)
+	_ITOFS	AT, $f2, 16
 
+	.align	4
 	addt	$f0, $f2, $f0
 	unop
 	divt/c	$f0, $f1, $f0
@@ -140,10 +135,10 @@ $x_big:
 #define QY	t3		/* Q*Y */
 
 	cvttq/c	$f0, $f0
-	stt	$f0, 8(sp)
-	ldq	Q, 8(sp)
+	_FTOIT	$f0, Q, 8
 	mulq	Q, Y, QY
 
+	.align	4
 	stq	t4, 8(sp)
 	unop
 	ldt	$f0, 0(sp)
diff --git a/sysdeps/alpha/reml.S b/sysdeps/alpha/reml.S
index c4eb426c5a..1bbb978f66 100644
--- a/sysdeps/alpha/reml.S
+++ b/sysdeps/alpha/reml.S
@@ -50,18 +50,12 @@ __reml:
 
 	EXTEND	(X, RV)
 	EXTEND	(Y, AT)
-	stq	RV, 16(sp)
-	stq	AT, 24(sp)
-
-	ldt	$f0, 16(sp)
-	ldt	$f1, 24(sp)
+	_ITOFT2	RV, $f0, 16, AT, $f1, 24
 	cvtqt	$f0, $f0
 	cvtqt	$f1, $f1
-
 	divt/c	$f0, $f1, $f0
 	cvttq/c	$f0, $f0
-	stt	$f0, 16(sp)
-	ldq	RV, 16(sp)
+	_FTOIT	$f0, RV, 16
 
 	ldt	$f0, 0(sp)
 	mull	RV, Y, RV
@@ -70,7 +64,6 @@ __reml:
 	cfi_restore ($f0)
 	cfi_restore ($f1)
 	cfi_def_cfa_offset (0)
-
 	subl	X, RV, RV
 	ret	$31, (RA), 1
 
diff --git a/sysdeps/alpha/remq.S b/sysdeps/alpha/remq.S
index ce527d1055..a8795c8d24 100644
--- a/sysdeps/alpha/remq.S
+++ b/sysdeps/alpha/remq.S
@@ -52,17 +52,13 @@ __remq:
 	   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.  */
-	stq	X, 16(sp)
-	stq	Y, 24(sp)
-	beq	Y, DIVBYZERO
-
 	stt	$f0, 0(sp)
 	stt	$f1, 8(sp)
+	beq	Y, DIVBYZERO
 	cfi_rel_offset ($f0, 0)
 	cfi_rel_offset ($f1, 8)
-	ldt	$f0, 16(sp)
-	ldt	$f1, 24(sp)
 
+	_ITOFT2	X, $f0, 16, Y, $f1, 24
 	cvtqt	$f0, $f0
 	cvtqt	$f1, $f1
 	divt/c	$f0, $f1, $f0
@@ -77,9 +73,7 @@ __remq:
 	/* If we get here, we're expecting exact results from the division.
 	   Do nothing else besides convert, compute remainder, clean up.  */
 	cvttq/c	$f0, $f0
-	stt	$f0, 16(sp)
-
-	ldq	AT, 16(sp)
+	_FTOIT	$f0, AT, 16
 	mulq	AT, Y, AT
 	ldt	$f0, 0(sp)
 	cfi_restore ($f1)
@@ -87,7 +81,6 @@ __remq:
 	cfi_restore ($f0)
 	cfi_def_cfa_offset (0)
 	lda	sp, FRAME(sp)
-
 	subq	X, AT, RV
 	ret	$31, (RA), 1
 
@@ -120,8 +113,8 @@ $x_big:
 $fix_sign_in_ret1:
 	cvttq/c	$f0, $f0
 
-	stt	$f0, 8(sp)
-	ldq	Q, 8(sp)
+	_FTOIT	$f0, Q, 8
+	.align	3
 $fix_sign_in_ret2:
 	mulq	Q, Y, QY
 	stq	t4, 8(sp)
@@ -233,9 +226,8 @@ $fix_sign_in:
 
 	bge	t1, $fix_sign_in_ret1
 	cvttq/c	$f0, $f0
-	stt	$f0, 8(sp)
-	ldq	Q, 8(sp)
-
+	_FTOIT	$f0, Q, 8
+	.align	3
 	negq	Q, Q
 	br	$fix_sign_in_ret2
 
diff --git a/sysdeps/alpha/remqu.S b/sysdeps/alpha/remqu.S
index fa2eb626c7..f8deebbbc1 100644
--- a/sysdeps/alpha/remqu.S
+++ b/sysdeps/alpha/remqu.S
@@ -52,19 +52,16 @@ __remqu:
 	   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.  */
-	stq	X, 16(sp)
-	stq	Y, 24(sp)
 	subq	Y, 1, AT
-
 	stt	$f0, 0(sp)
 	and	Y, AT, AT
+
 	stt	$f1, 8(sp)
 	beq	AT, $powerof2
 	cfi_rel_offset ($f0, 0)
 	cfi_rel_offset ($f1, 8)
 
-	ldt	$f0, 16(sp)
-	ldt	$f1, 24(sp)
+	_ITOFT2	X, $f0, 16, Y, $f1, 24
 	cvtqt	$f0, $f0
 	cvtqt	$f1, $f1
 
@@ -82,9 +79,8 @@ __remqu:
 	/* If we get here, we're expecting exact results from the division.
 	   Do nothing else besides convert, compute remainder, clean up.  */
 	cvttq/c	$f0, $f0
-	stt	$f0, 16(sp)
+	_FTOIT	$f0, AT, 16
 
-	ldq	AT, 16(sp)
 	mulq	AT, Y, AT
 	ldt	$f0, 0(sp)
 	lda	sp, FRAME(sp)
@@ -93,6 +89,7 @@ __remqu:
 	cfi_restore ($f1)
 	cfi_def_cfa_offset (0)
 
+	.align	4
 	subq	X, AT, RV
 	ret	$31, (RA), 1
 
@@ -105,15 +102,13 @@ $x_is_neg:
 	ldah	AT, 0x5f80		/* 2**64 as float.  */
 	stt	$f2, 24(sp)
 	cfi_rel_offset ($f2, 24)
-	stl	AT, 16(sp)
-	lds	$f2, 16(sp)
+	_ITOFS	AT, $f2, 16
 
 	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)
@@ -144,10 +139,10 @@ $x_big:
 #define QY	t3		/* Q*Y */
 
 	cvttq/c	$f0, $f0
-	stt	$f0, 8(sp)
-	ldq	Q, 8(sp)
+	_FTOIT	$f0, Q, 8
 	mulq	Q, Y, QY
 
+	.align	4
 	stq	t4, 8(sp)
 	unop
 	ldt	$f0, 0(sp)