1 //
2 // Test_3x3mulVM.cpp
3 // BulletTest
4 //
5 // Copyright (c) 2011 Apple Inc.
6 //
7
8 #include "LinearMath/btScalar.h"
9 #if defined(BT_USE_SSE_IN_API) || defined(BT_USE_NEON)
10
11 #include "Test_3x3mulVM.h"
12 #include "vector.h"
13 #include "Utils.h"
14 #include "main.h"
15 #include <math.h>
16 #include <string.h>
17
18 #include <LinearMath/btMatrix3x3.h>
19
20 #define LOOPCOUNT 1000
21 #define ARRAY_SIZE 128
22
rand_f4(void)23 static inline btSimdFloat4 rand_f4(void)
24 {
25 return btAssign128(RANDF_01, RANDF_01, RANDF_01, BT_NAN); // w channel NaN
26 }
27
M3x3mulVM_ref(const btVector3 & v,const btMatrix3x3 & m)28 static btVector3 M3x3mulVM_ref(const btVector3 &v, const btMatrix3x3 &m)
29 {
30 return btVector3(m.tdotx(v), m.tdoty(v), m.tdotz(v));
31 }
32
Test_3x3mulVM(void)33 int Test_3x3mulVM(void)
34 {
35 // Init an array flanked by guard pages
36 btVector3 in1[ARRAY_SIZE];
37 btMatrix3x3 in2[ARRAY_SIZE];
38 btVector3 out[ARRAY_SIZE];
39 btVector3 out2[ARRAY_SIZE];
40
41 // Init the data
42 size_t i, j;
43 for (i = 0; i < ARRAY_SIZE; i++)
44 {
45 in1[i] = btVector3(rand_f4());
46 in2[i] = btMatrix3x3(rand_f4(), rand_f4(), rand_f4());
47
48 out[i] = M3x3mulVM_ref(in1[i], in2[i]);
49 out2[i] = (in1[i] * in2[i]);
50
51 if (fabsf(out[i].m_floats[0] - out2[i].m_floats[0]) +
52 fabsf(out[i].m_floats[1] - out2[i].m_floats[1]) +
53 fabsf(out[i].m_floats[2] - out2[i].m_floats[2]) +
54 fabsf(out[i].m_floats[3] - out2[i].m_floats[3]) >
55 FLT_EPSILON * 4)
56 {
57 vlog("Error - M3x3mulVM result error! ");
58 vlog("failure @ %ld\n", i);
59 vlog(
60 "\ncorrect = (%10.4f, %10.4f, %10.4f, %10.4f) "
61 "\ntested = (%10.4f, %10.4f, %10.4f, %10.4f) \n",
62 out[i].m_floats[0], out[i].m_floats[1], out[i].m_floats[2], out[i].m_floats[3],
63 out2[i].m_floats[0], out2[i].m_floats[1], out2[i].m_floats[2], out2[i].m_floats[3]);
64
65 return 1;
66 }
67 }
68
69 uint64_t scalarTime, vectorTime;
70 uint64_t startTime, bestTime, currentTime;
71 bestTime = -1LL;
72 scalarTime = 0;
73 for (j = 0; j < LOOPCOUNT; j++)
74 {
75 startTime = ReadTicks();
76 for (i = 0; i < ARRAY_SIZE; i++)
77 out[i] = M3x3mulVM_ref(in1[i], in2[i]);
78 currentTime = ReadTicks() - startTime;
79 scalarTime += currentTime;
80 if (currentTime < bestTime)
81 bestTime = currentTime;
82 }
83 if (0 == gReportAverageTimes)
84 scalarTime = bestTime;
85 else
86 scalarTime /= LOOPCOUNT;
87
88 bestTime = -1LL;
89 vectorTime = 0;
90 for (j = 0; j < LOOPCOUNT; j++)
91 {
92 startTime = ReadTicks();
93 for (i = 0; i < ARRAY_SIZE; i++)
94 out2[i] = (in1[i] * in2[i]);
95 currentTime = ReadTicks() - startTime;
96 vectorTime += currentTime;
97 if (currentTime < bestTime)
98 bestTime = currentTime;
99 }
100 if (0 == gReportAverageTimes)
101 vectorTime = bestTime;
102 else
103 vectorTime /= LOOPCOUNT;
104
105 vlog("Timing:\n");
106 vlog("\t scalar\t vector\n");
107 vlog("\t%10.2f\t%10.2f\n", TicksToCycles(scalarTime) / ARRAY_SIZE, TicksToCycles(vectorTime) / ARRAY_SIZE);
108
109 return 0;
110 }
111 #endif //BT_USE_SSE
112