// // Test_3x3mulM1M2.cpp // BulletTest // // Copyright (c) 2011 Apple Inc. // #include "LinearMath/btScalar.h" #if defined(BT_USE_SSE_IN_API) || defined(BT_USE_NEON) #include "Test_3x3mulM1M2.h" #include "vector.h" #include "Utils.h" #include "main.h" #include #include #include #define LOOPCOUNT 1000 #define ARRAY_SIZE 128 static inline btSimdFloat4 rand_f4(void) { return btAssign128(RANDF_01, RANDF_01, RANDF_01, BT_NAN); // w channel NaN } static btMatrix3x3 M3x3mulM1M2_ref(const btMatrix3x3 &m1, const btMatrix3x3 &m2) { return btMatrix3x3( m2.tdotx(m1[0]), m2.tdoty(m1[0]), m2.tdotz(m1[0]), m2.tdotx(m1[1]), m2.tdoty(m1[1]), m2.tdotz(m1[1]), m2.tdotx(m1[2]), m2.tdoty(m1[2]), m2.tdotz(m1[2])); } static bool fuzzyEqualSlow(const btVector3 &ref, const btVector3 &other) { const btScalar epsilon = SIMD_EPSILON; return ((btFabs(ref.m_floats[3] - other.m_floats[3]) <= epsilon) && (btFabs(ref.m_floats[2] - other.m_floats[2]) <= epsilon) && (btFabs(ref.m_floats[1] - other.m_floats[1]) <= epsilon) && (btFabs(ref.m_floats[0] - other.m_floats[0]) <= epsilon)); } static int operator!=(const btMatrix3x3 &a, const btMatrix3x3 &b) { if (a.getRow(0) != b.getRow(0)) { if (!fuzzyEqualSlow(a.getRow(0), b.getRow(0))) { return 1; } } if (a.getRow(1) != b.getRow(1)) { if (!fuzzyEqualSlow(a.getRow(1), b.getRow(1))) return 1; } if (a.getRow(2) != b.getRow(2)) { if (!fuzzyEqualSlow(a.getRow(2), b.getRow(2))) { return 1; } } return 0; } int Test_3x3mulM1M2(void) { // Init an array flanked by guard pages btMatrix3x3 in1[ARRAY_SIZE]; btMatrix3x3 in2[ARRAY_SIZE]; btMatrix3x3 out[ARRAY_SIZE]; btMatrix3x3 out2[ARRAY_SIZE]; // Init the data size_t i, j; for (i = 0; i < ARRAY_SIZE; i++) { in1[i] = btMatrix3x3(rand_f4(), rand_f4(), rand_f4()); in2[i] = btMatrix3x3(rand_f4(), rand_f4(), rand_f4()); out[i] = M3x3mulM1M2_ref(in1[i], in2[i]); out2[i] = (in1[i] * in2[i]); if (out[i] != out2[i]) { vlog("Error - M3x3mulM1M2 result error! "); vlog("failure @ %ld\n", i); btVector3 m0, m1, m2; m0 = out[i].getRow(0); m1 = out[i].getRow(1); m2 = out[i].getRow(2); vlog( "\ncorrect = (%10.4f, %10.4f, %10.4f, %10.4f) " "\n (%10.4f, %10.4f, %10.4f, %10.4f) " "\n (%10.4f, %10.4f, %10.4f, %10.4f) \n", m0.m_floats[0], m0.m_floats[1], m0.m_floats[2], m0.m_floats[3], m1.m_floats[0], m1.m_floats[1], m1.m_floats[2], m1.m_floats[3], m2.m_floats[0], m2.m_floats[1], m2.m_floats[2], m2.m_floats[3]); m0 = out2[i].getRow(0); m1 = out2[i].getRow(1); m2 = out2[i].getRow(2); vlog( "\ntested = (%10.4f, %10.4f, %10.4f, %10.4f) " "\n (%10.4f, %10.4f, %10.4f, %10.4f) " "\n (%10.4f, %10.4f, %10.4f, %10.4f) \n", m0.m_floats[0], m0.m_floats[1], m0.m_floats[2], m0.m_floats[3], m1.m_floats[0], m1.m_floats[1], m1.m_floats[2], m1.m_floats[3], m2.m_floats[0], m2.m_floats[1], m2.m_floats[2], m2.m_floats[3]); return -1; } } uint64_t scalarTime, vectorTime; uint64_t startTime, bestTime, currentTime; bestTime = -1LL; scalarTime = 0; for (j = 0; j < LOOPCOUNT; j++) { startTime = ReadTicks(); for (i = 0; i < ARRAY_SIZE; i++) out[i] = M3x3mulM1M2_ref(in1[i], in2[i]); currentTime = ReadTicks() - startTime; scalarTime += currentTime; if (currentTime < bestTime) bestTime = currentTime; } if (0 == gReportAverageTimes) scalarTime = bestTime; else scalarTime /= LOOPCOUNT; bestTime = -1LL; vectorTime = 0; for (j = 0; j < LOOPCOUNT; j++) { startTime = ReadTicks(); for (i = 0; i < ARRAY_SIZE; i++) out2[i] = (in1[i] * in2[i]); currentTime = ReadTicks() - startTime; vectorTime += currentTime; if (currentTime < bestTime) bestTime = currentTime; } if (0 == gReportAverageTimes) vectorTime = bestTime; else vectorTime /= LOOPCOUNT; vlog("Timing:\n"); vlog("\t scalar\t vector\n"); vlog("\t%10.2f\t%10.2f\n", TicksToCycles(scalarTime) / ARRAY_SIZE, TicksToCycles(vectorTime) / ARRAY_SIZE); return 0; } #endif //BT_USE_SSE