Annotation of OpenXM/src/kan96xx/gmp-2.0.2-ssh-2/mpn/hppa/README, Revision 1.1.1.1
1.1 takayama 1: This directory contains mpn functions for various HP PA-RISC chips. Code
2: that runs faster on the PA7100 and later implementations, is in the pa7100
3: directory.
4:
5: RELEVANT OPTIMIZATION ISSUES
6:
7: Load and Store timing
8:
9: On the PA7000 no memory instructions can issue the two cycles after a store.
10: For the PA7100, this is reduced to one cycle.
11:
12: The PA7100 has a lookup-free cache, so it helps to schedule loads and the
13: dependent instruction really far from each other.
14:
15: STATUS
16:
17: 1. mpn_mul_1 could be improved to 6.5 cycles/limb on the PA7100, using the
18: instructions bwlow (but some sw pipelining is needed to avoid the
19: xmpyu-fstds delay):
20:
21: fldds s1_ptr
22:
23: xmpyu
24: fstds N(%r30)
25: xmpyu
26: fstds N(%r30)
27:
28: ldws N(%r30)
29: ldws N(%r30)
30: ldws N(%r30)
31: ldws N(%r30)
32:
33: addc
34: stws res_ptr
35: addc
36: stws res_ptr
37:
38: addib Loop
39:
40: 2. mpn_addmul_1 could be improved from the current 10 to 7.5 cycles/limb
41: (asymptotically) on the PA7100, using the instructions below. With proper
42: sw pipelining and the unrolling level below, the speed becomes 8
43: cycles/limb.
44:
45: fldds s1_ptr
46: fldds s1_ptr
47:
48: xmpyu
49: fstds N(%r30)
50: xmpyu
51: fstds N(%r30)
52: xmpyu
53: fstds N(%r30)
54: xmpyu
55: fstds N(%r30)
56:
57: ldws N(%r30)
58: ldws N(%r30)
59: ldws N(%r30)
60: ldws N(%r30)
61: ldws N(%r30)
62: ldws N(%r30)
63: ldws N(%r30)
64: ldws N(%r30)
65: addc
66: addc
67: addc
68: addc
69: addc %r0,%r0,cy-limb
70:
71: ldws res_ptr
72: ldws res_ptr
73: ldws res_ptr
74: ldws res_ptr
75: add
76: stws res_ptr
77: addc
78: stws res_ptr
79: addc
80: stws res_ptr
81: addc
82: stws res_ptr
83:
84: addib
FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>