MIPS specific optimizations.

5 files changed, 333 insertions, 0 deletions

diff --git a/sysdeps/mips/fpu/e_sqrt.c b/sysdeps/mips/fpu/e_sqrt.c
new file mode 100644
index 0000000000..5449710aff
--- /dev/null
+++ b/sysdeps/mips/fpu/e_sqrt.c
@@ -0,0 +1,38 @@
+/* Copyright (C) 2002 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+   Contributed by Hartvig Ekner <hartvige@mips.com>, 2002.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, write to the Free
+   Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
+   02111-1307 USA.  */
+
+
+#include <sgidefs.h>
+
+
+#if (_MIPS_ISA >= _MIPS_ISA_MIPS2)
+
+double
+__ieee754_sqrt (double x)
+{
+  double z;
+  __asm__ ("sqrt.d %0,%1" : "=f" (z) : "f" (x));
+  return z;
+}
+
+#else
+
+#include <sysdeps/ieee754/dbl-64/e_sqrt.c>
+
+#endif
diff --git a/sysdeps/mips/fpu/e_sqrtf.c b/sysdeps/mips/fpu/e_sqrtf.c
new file mode 100644
index 0000000000..3590ad46d5
--- /dev/null
+++ b/sysdeps/mips/fpu/e_sqrtf.c
@@ -0,0 +1,39 @@
+/* Copyright (C) 2002 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+   Contributed by Hartvig Ekner <hartvige@mips.com>, 2002.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, write to the Free
+   Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
+   02111-1307 USA.  */
+
+
+#include <sgidefs.h>
+
+
+#if (_MIPS_ISA >= _MIPS_ISA_MIPS2)
+
+float
+__ieee754_sqrtf (float x)
+{
+  float z;
+  __asm__ ("sqrt.s %0,%1" : "=f" (z) : "f" (x));
+  return z;
+}
+
+#else
+
+#include <sysdeps/ieee754/flt-32/e_sqrtf.c>
+
+#endif
+
diff --git a/sysdeps/mips/fpu/fsetexcptflg.c b/sysdeps/mips/fpu/fsetexcptflg.c
new file mode 100644
index 0000000000..c65d793178
--- /dev/null
+++ b/sysdeps/mips/fpu/fsetexcptflg.c
@@ -0,0 +1,43 @@
+/* Set floating-point environment exception handling.
+   Copyright (C) 1998, 1999, 2000, 2002 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+   Contributed by Hartvig Ekner <hartvige@mips.com>, 2002.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, write to the Free
+   Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
+   02111-1307 USA.  */
+
+#include <fenv.h>
+#include <fpu_control.h>
+
+int
+fesetexceptflag (const fexcept_t *flagp, int excepts)
+{
+  fexcept_t temp;
+
+  /* Get the current exceptions.  */
+  _FPU_GETCW (temp);
+
+  /* Make sure the flags we want restored are legal.  */
+  excepts &= FE_ALL_EXCEPT;
+
+  /* Now clear the bits called for, and copy them in from flagp. Note that
+     we ignore all non-flag bits from *flagp, so they don't matter.  */
+  temp = (temp & ~excepts) | (*flagp & excepts);
+
+  _FPU_SETCW (temp);
+
+  /* Success.  */
+  return 0;
+}
diff --git a/sysdeps/mips/memcpy.S b/sysdeps/mips/memcpy.S
new file mode 100644
index 0000000000..394265eed7
--- /dev/null
+++ b/sysdeps/mips/memcpy.S
@@ -0,0 +1,130 @@
+/* Copyright (C) 2002 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+   Contributed by Hartvig Ekner <hartvige@mips.com>, 2002.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, write to the Free
+   Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
+   02111-1307 USA.  */
+
+#include <sysdep.h>
+#include <endian.h>
+
+
+/* void *memcpy(void *s1, const void *s2, size_t n);
+
+   This routine could be optimized further for MIPS64, but this is left
+   as an exercise for the future. When it is done, the file should be kept
+   as a sisterfile to this one, and placed in the sysdeps/mips/mips64 
+   directory.  */
+
+#if __BYTE_ORDER == __BIG_ENDIAN
+#  define LWHI	lwl		/* high part is left in big-endian	*/
+#  define SWHI	swl		/* high part is left in big-endian	*/
+#  define LWLO	lwr		/* low part is right in big-endian	*/
+#  define SWLO	swr		/* low part is right in big-endian	*/
+#else
+#  define LWHI	lwr		/* high part is right in little-endian	*/
+#  define SWHI	swr		/* high part is right in little-endian	*/
+#  define LWLO	lwl		/* low part is left in little-endian	*/
+#  define SWLO	swl		/* low part is left in little-endian	*/
+#endif
+
+ENTRY (memcpy)
+	.set	noreorder
+
+	slti	t0, a2, 8		# Less than 8?
+	bne	t0, zero, $last8
+	move	v0, a0			# Setup exit value before too late
+
+	xor	t0, a1, a0		# Find a0/a1 displacement
+	andi	t0, 0x3
+	bne	t0, zero, $shift	# Go handle the unaligned case
+	subu	t1, zero, a1
+	andi	t1, 0x3			# a0/a1 are aligned, but are we
+	beq	t1, zero, $chk8w	#  starting in the middle of a word?
+	subu	a2, t1
+	LWHI	t0, 0(a1)		# Yes we are... take care of that
+	addu	a1, t1
+	SWHI	t0, 0(a0)
+	addu	a0, t1
+
+$chk8w:	andi	t0, a2, 0x1f		# 32 or more bytes left?
+	beq	t0, a2, $chk1w
+	subu	a3, a2, t0		# Yes
+	addu	a3, a1			# a3 = end address of loop
+	move	a2, t0			# a2 = what will be left after loop
+$lop8w:	lw	t0,  0(a1)		# Loop taking 8 words at a time
+	lw	t1,  4(a1)
+	lw	t2,  8(a1)
+	lw	t3, 12(a1)
+	lw	t4, 16(a1)
+	lw	t5, 20(a1)
+	lw	t6, 24(a1)
+	lw	t7, 28(a1)
+	addiu	a0, 32
+	addiu	a1, 32
+	sw	t0, -32(a0)
+	sw	t1, -28(a0)
+	sw	t2, -24(a0)
+	sw	t3, -20(a0)
+	sw	t4, -16(a0)
+	sw	t5, -12(a0)
+	sw	t6,  -8(a0)
+	bne	a1, a3, $lop8w
+	sw	t7,  -4(a0)
+
+$chk1w:	andi	t0, a2, 0x3		# 4 or more bytes left?
+	beq	t0, a2, $last8
+	subu	a3, a2, t0		# Yes, handle them one word at a time
+	addu	a3, a1			# a3 again end address
+	move	a2, t0
+$lop1w:	lw	t0, 0(a1)
+	addiu	a0, 4
+	addiu	a1, 4
+	bne	a1, a3, $lop1w
+	sw	t0, -4(a0)
+
+$last8:	blez	a2, $lst8e		# Handle last 8 bytes, one at a time
+	addu	a3, a2, a1
+$lst8l:	lb	t0, 0(a1)
+	addiu	a0, 1
+	addiu	a1, 1
+	bne	a1, a3, $lst8l
+	sb	t0, -1(a0)
+$lst8e:	jr	ra			# Bye, bye
+	nop
+
+$shift:	subu	a3, zero, a0		# Src and Dest unaligned 
+	andi	a3, 0x3			#  (unoptimized case...)
+	beq	a3, zero, $shft1
+	subu	a2, a3			# a2 = bytes left
+	LWHI	t0, 0(a1)		# Take care of first odd part
+	LWLO	t0, 3(a1)
+	addu	a1, a3
+	SWHI	t0, 0(a0)
+	addu	a0, a3
+$shft1:	andi	t0, a2, 0x3
+	subu	a3, a2, t0
+	addu	a3, a1
+$shfth:	LWHI	t1, 0(a1)		# Limp through, word by word
+	LWLO	t1, 3(a1)
+	addiu	a0, 4
+	addiu	a1, 4
+	bne	a1, a3, $shfth
+	sw	t1, -4(a0)
+	b	$last8			# Handle anything which may be left
+	move	a2, t0
+
+	.set	reorder
+END (memcpy)
diff --git a/sysdeps/mips/memset.S b/sysdeps/mips/memset.S
new file mode 100644
index 0000000000..7e3f129af9
--- /dev/null
+++ b/sysdeps/mips/memset.S
@@ -0,0 +1,83 @@
+/* Copyright (C) 2002 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+   Contributed by Hartvig Ekner <hartvige@mips.com>, 2002.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, write to the Free
+   Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
+   02111-1307 USA.  */
+
+#include <sysdep.h>
+#include <endian.h>
+
+
+/* void *memset(void *s, int c, size_t n).
+
+   This routine could be optimized further for MIPS64, but this is left
+   as an exercise for the future. When it is done, the file should be kept
+   as a sisterfile to this one, and placed in the sysdeps/mips/mips64 
+   directory.  */
+
+#if __BYTE_ORDER == __BIG_ENDIAN
+# define SWHI	swl		/* high part is left in big-endian	*/
+#else
+# define SWHI	swr		/* high part is right in little-endian	*/
+#endif
+
+ENTRY (memset)
+	.set	noreorder
+
+	slti	t1, a2, 8		# Less than 8?
+	bne	t1, zero, $last8
+	move	v0, a0			# Setup exit value before too late
+
+	beq	a1, zero, $ueven	# If zero pattern, no need to extend
+	andi	a1, 0xff		# Avoid problems with bogus arguments
+	sll	t0, a1, 8
+	or	a1, t0
+	sll	t0, a1, 16
+	or	a1, t0			# a1 is now pattern in full word
+
+$ueven:	subu	t0, zero, a0		# Unaligned address?
+	andi	t0, 0x3
+	beq	t0, zero, $chkw
+	subu	a2, t0
+	SWHI	a1, 0(a0)		# Yes, handle first unaligned part
+	addu	a0, t0			# Now both a0 and a2 are updated
+
+$chkw:	andi	t0, a2, 0x7		# Enough left for one loop iteration?
+	beq	t0, a2, $chkl
+	subu	a3, a2, t0
+	addu	a3, a0			# a3 is last loop address +1
+	move	a2, t0			# a2 is now # of bytes left after loop
+$loopw:	addiu	a0, 8			# Handle 2 words pr. iteration
+	sw	a1, -8(a0)
+	bne	a0, a3, $loopw
+	sw	a1, -4(a0)
+
+$chkl:	andi	t0, a2, 0x4		# Check if there is at least a full
+	beq	t0, zero, $last8	#  word remaining after the loop
+	subu	a2, t0
+	sw	a1, 0(a0)		# Yes...
+	addiu	a0, 4
+
+$last8:	blez	a2, $exit		# Handle last 8 bytes (if cnt>0)
+	addu	a3, a2, a0		# a3 is last address +1
+$lst8l:	addiu	a0, 1
+	bne	a0, a3, $lst8l
+	sb	a1, -1(a0)
+$exit:	j	ra			# Bye, bye
+	nop
+
+	.set	reorder
+END (memset)