aboutsummaryrefslogtreecommitdiffstats
path: root/mpi/hppa
diff options
context:
space:
mode:
Diffstat (limited to 'mpi/hppa')
-rw-r--r--mpi/hppa/README84
-rw-r--r--mpi/hppa/distfiles6
-rw-r--r--mpi/hppa/mpih-shift.S124
-rw-r--r--mpi/hppa/mpih-sub1.S77
4 files changed, 289 insertions, 2 deletions
diff --git a/mpi/hppa/README b/mpi/hppa/README
new file mode 100644
index 000000000..5a2d5fd97
--- /dev/null
+++ b/mpi/hppa/README
@@ -0,0 +1,84 @@
+This directory contains mpn functions for various HP PA-RISC chips. Code
+that runs faster on the PA7100 and later implementations, is in the pa7100
+directory.
+
+RELEVANT OPTIMIZATION ISSUES
+
+ Load and Store timing
+
+On the PA7000 no memory instructions can issue the two cycles after a store.
+For the PA7100, this is reduced to one cycle.
+
+The PA7100 has a lookup-free cache, so it helps to schedule loads and the
+dependent instruction really far from each other.
+
+STATUS
+
+1. mpn_mul_1 could be improved to 6.5 cycles/limb on the PA7100, using the
+ instructions bwlow (but some sw pipelining is needed to avoid the
+ xmpyu-fstds delay):
+
+ fldds s1_ptr
+
+ xmpyu
+ fstds N(%r30)
+ xmpyu
+ fstds N(%r30)
+
+ ldws N(%r30)
+ ldws N(%r30)
+ ldws N(%r30)
+ ldws N(%r30)
+
+ addc
+ stws res_ptr
+ addc
+ stws res_ptr
+
+ addib Loop
+
+2. mpn_addmul_1 could be improved from the current 10 to 7.5 cycles/limb
+ (asymptotically) on the PA7100, using the instructions below. With proper
+ sw pipelining and the unrolling level below, the speed becomes 8
+ cycles/limb.
+
+ fldds s1_ptr
+ fldds s1_ptr
+
+ xmpyu
+ fstds N(%r30)
+ xmpyu
+ fstds N(%r30)
+ xmpyu
+ fstds N(%r30)
+ xmpyu
+ fstds N(%r30)
+
+ ldws N(%r30)
+ ldws N(%r30)
+ ldws N(%r30)
+ ldws N(%r30)
+ ldws N(%r30)
+ ldws N(%r30)
+ ldws N(%r30)
+ ldws N(%r30)
+ addc
+ addc
+ addc
+ addc
+ addc %r0,%r0,cy-limb
+
+ ldws res_ptr
+ ldws res_ptr
+ ldws res_ptr
+ ldws res_ptr
+ add
+ stws res_ptr
+ addc
+ stws res_ptr
+ addc
+ stws res_ptr
+ addc
+ stws res_ptr
+
+ addib
diff --git a/mpi/hppa/distfiles b/mpi/hppa/distfiles
index 7ca77f801..83c59ea62 100644
--- a/mpi/hppa/distfiles
+++ b/mpi/hppa/distfiles
@@ -1,4 +1,6 @@
-
-mpih-add1.S
+README
udiv-qrnnd.S
+mpih-add1.S
+mpih-sub1.S
+mpih-shift.S
diff --git a/mpi/hppa/mpih-shift.S b/mpi/hppa/mpih-shift.S
new file mode 100644
index 000000000..5039c7381
--- /dev/null
+++ b/mpi/hppa/mpih-shift.S
@@ -0,0 +1,124 @@
+/* hppa rshift, lshift
+ * Copyright (C) 1992, 1994 Free Software Foundation, Inc.
+ * Copyright (c) 1997 by Werner Koch (dd9jn)
+ *
+ * This file is part of G10.
+ *
+ * G10 is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * G10 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 General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA
+ *
+ * Note: This code is heavily based on the GNU MP Library.
+ * Actually it's the same code with only minor changes in the
+ * way the data is stored; this is to support the abstraction
+ * of an optional secure memory allocation which may be used
+ * to avoid revealing of sensitive data due to paging etc.
+ * The GNU MP Library itself is published under the LGPL;
+ * however I decided to publish this code under the plain GPL.
+ */
+
+
+
+/*******************
+ * mpi_limb_t
+ * mpihelp_lshift( mpi_ptr_t wp, (gr26)
+ * mpi_ptr_t up, (gr25)
+ * mpi_size_t usize, (gr24)
+ * unsigned cnt) (gr23)
+ */
+
+ .code
+ .export mpihelp_lshift
+mpihelp_lshift
+ .proc
+ .callinfo frame=64,no_calls
+ .entry
+
+ sh2add %r24,%r25,%r25
+ sh2add %r24,%r26,%r26
+ ldws,mb -4(0,%r25),%r22
+ subi 32,%r23,%r1
+ mtsar %r1
+ addib,= -1,%r24,L$0004
+ vshd %r0,%r22,%r28 ; compute carry out limb
+ ldws,mb -4(0,%r25),%r29
+ addib,= -1,%r24,L$0002
+ vshd %r22,%r29,%r20
+
+L$loop ldws,mb -4(0,%r25),%r22
+ stws,mb %r20,-4(0,%r26)
+ addib,= -1,%r24,L$0003
+ vshd %r29,%r22,%r20
+ ldws,mb -4(0,%r25),%r29
+ stws,mb %r20,-4(0,%r26)
+ addib,<> -1,%r24,L$loop
+ vshd %r22,%r29,%r20
+
+L$0002 stws,mb %r20,-4(0,%r26)
+ vshd %r29,%r0,%r20
+ bv 0(%r2)
+ stw %r20,-4(0,%r26)
+L$0003 stws,mb %r20,-4(0,%r26)
+L$0004 vshd %r22,%r0,%r20
+ bv 0(%r2)
+ stw %r20,-4(0,%r26)
+
+ .exit
+ .procend
+
+
+
+/*******************
+ * mpi_limb_t
+ * mpihelp_rshift( mpi_ptr_t wp, (gr26)
+ * mpi_ptr_t up, (gr25)
+ * mpi_size_t usize, (gr24)
+ * unsigned cnt) (gr23)
+ */
+
+ .code
+ .export mpihelp_rshift
+mpihelp_rshift
+ .proc
+ .callinfo frame=64,no_calls
+ .entry
+
+ ldws,ma 4(0,%r25),%r22
+ mtsar %r23
+ addib,= -1,%r24,L$r004
+ vshd %r22,%r0,%r28 ; compute carry out limb
+ ldws,ma 4(0,%r25),%r29
+ addib,= -1,%r24,L$r002
+ vshd %r29,%r22,%r20
+
+L$roop ldws,ma 4(0,%r25),%r22
+ stws,ma %r20,4(0,%r26)
+ addib,= -1,%r24,L$r003
+ vshd %r22,%r29,%r20
+ ldws,ma 4(0,%r25),%r29
+ stws,ma %r20,4(0,%r26)
+ addib,<> -1,%r24,L$roop
+ vshd %r29,%r22,%r20
+
+L$r002 stws,ma %r20,4(0,%r26)
+ vshd %r0,%r29,%r20
+ bv 0(%r2)
+ stw %r20,0(0,%r26)
+L$r003 stws,ma %r20,4(0,%r26)
+L$r004 vshd %r0,%r22,%r20
+ bv 0(%r2)
+ stw %r20,0(0,%r26)
+
+ .exit
+ .procend
+
diff --git a/mpi/hppa/mpih-sub1.S b/mpi/hppa/mpih-sub1.S
new file mode 100644
index 000000000..361cc6832
--- /dev/null
+++ b/mpi/hppa/mpih-sub1.S
@@ -0,0 +1,77 @@
+/* hppa sub_n -- Sub two limb vectors of the same length > 0 and store
+ * sum in a third limb vector.
+ * Copyright (C) 1992, 1994 Free Software Foundation, Inc.
+ * Copyright (c) 1998 by Werner Koch (dd9jn)
+ *
+ * This file is part of G10.
+ *
+ * G10 is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * G10 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 General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA
+ *
+ * Note: This code is heavily based on the GNU MP Library.
+ * Actually it's the same code with only minor changes in the
+ * way the data is stored; this is to support the abstraction
+ * of an optional secure memory allocation which may be used
+ * to avoid revealing of sensitive data due to paging etc.
+ * The GNU MP Library itself is published under the LGPL;
+ * however I decided to publish this code under the plain GPL.
+ */
+
+
+#include "sysdep.h"
+#include "asm-syntax.h"
+
+
+/*******************
+ * mpi_limb_t
+ * mpihelp_sub_n( mpi_ptr_t res_ptr, (gr26)
+ * mpi_ptr_t s1_ptr, (gr25)
+ * mpi_ptr_t s2_ptr, (gr24)
+ * mpi_size_t size) (gr23)
+ *
+ * One might want to unroll this as for other processors, but it turns
+ * out that the data cache contention after a store makes such
+ * unrolling useless. We can't come under 5 cycles/limb anyway.
+ */
+
+
+ .code
+ .export mpihelp_sub_n
+mpihelp_sub_n
+ .proc
+ .callinfo frame=0,no_calls
+ .entry
+
+ ldws,ma 4(0,%r25),%r20
+ ldws,ma 4(0,%r24),%r19
+
+ addib,= -1,%r23,L$end ; check for (SIZE == 1)
+ sub %r20,%r19,%r28 ; subtract first limbs ignoring cy
+
+L$loop ldws,ma 4(0,%r25),%r20
+ ldws,ma 4(0,%r24),%r19
+ stws,ma %r28,4(0,%r26)
+ addib,<> -1,%r23,L$loop
+ subb %r20,%r19,%r28
+
+L$end stws %r28,0(0,%r26)
+ addc %r0,%r0,%r28
+ bv 0(%r2)
+ subi 1,%r28,%r28
+
+ .exit
+ .procend
+
+
+
generated by cgit v1.2.3 (git 2.25.1) at 2025-06-19 00:13:37 +0000