diff options
Diffstat (limited to 'volk')
| -rw-r--r-- | volk/lib/qa_32fc_magnitude_16s_aligned16.cc | 6 | ||||
| -rw-r--r-- | volk/lib/qa_volk.cc | 1 | ||||
| -rw-r--r-- | volk/orc/volk_32fc_magnitude_16s_aligned16_orc_impl.orc | 6 |
3 files changed, 9 insertions, 4 deletions
diff --git a/volk/lib/qa_32fc_magnitude_16s_aligned16.cc b/volk/lib/qa_32fc_magnitude_16s_aligned16.cc index 53b3bf790..93d4ec150 100644 --- a/volk/lib/qa_32fc_magnitude_16s_aligned16.cc +++ b/volk/lib/qa_32fc_magnitude_16s_aligned16.cc @@ -63,9 +63,9 @@ void qa_32fc_magnitude_16s_aligned16::t1() { total = (double)(end-start)/(double)CLOCKS_PER_SEC; printf("sse3_time: %f\n", total); - for(int i = 0; i < 10; ++i) { - printf("inputs: %f, %f\n", input0[i].real(), input0[i].imag()); - printf("generic... %i, sse3... %i, orc... %i\n", output_generic[i], output_sse3[i], output_orc[i]); + for(int i = 0; i < 1; ++i) { + // printf("inputs: %f, %f\n", input0[i].real(), input0[i].imag()); + // printf("generic... %i, sse3... %i, orc... %i\n", output_generic[i], output_sse3[i], output_orc[i]); } for(int i = 0; i < vlen; ++i) { diff --git a/volk/lib/qa_volk.cc b/volk/lib/qa_volk.cc index f6a334da7..c3c27b69b 100644 --- a/volk/lib/qa_volk.cc +++ b/volk/lib/qa_volk.cc @@ -118,6 +118,7 @@ CppUnit::TestSuite * qa_volk::suite() { CppUnit::TestSuite *s = new CppUnit::TestSuite("volk"); + s->addTest(qa_16s_quad_max_star_aligned16::suite()); s->addTest(qa_32fc_dot_prod_aligned16::suite()); s->addTest(qa_32fc_square_dist_scalar_mult_aligned16::suite()); diff --git a/volk/orc/volk_32fc_magnitude_16s_aligned16_orc_impl.orc b/volk/orc/volk_32fc_magnitude_16s_aligned16_orc_impl.orc index 48b831021..15f8fdff0 100644 --- a/volk/orc/volk_32fc_magnitude_16s_aligned16_orc_impl.orc +++ b/volk/orc/volk_32fc_magnitude_16s_aligned16_orc_impl.orc @@ -9,11 +9,15 @@ .temp 4 sumf .temp 4 rootf .temp 4 rootl +.temp 4 maskl x2 mulf prodiqf, src, src splitql qf, if, prodiqf addf sumf, if, qf sqrtf rootf, sumf mulf rootf, rootf, scalar +cmpltf maskl, scalar, rootf +andl maskl, maskl, 0x80000000 +orl rootf, rootf, maskl convfl rootl, rootf -convlw dst, rootl +convssslw dst, rootl |