diff options
Diffstat (limited to 'volk/lib/qa_16sc_magnitude_32f_aligned16.cc')
| -rw-r--r-- | volk/lib/qa_16sc_magnitude_32f_aligned16.cc | 20 |
1 files changed, 20 insertions, 0 deletions
diff --git a/volk/lib/qa_16sc_magnitude_32f_aligned16.cc b/volk/lib/qa_16sc_magnitude_32f_aligned16.cc index 1ebe644c5..54cc2ba6e 100644 --- a/volk/lib/qa_16sc_magnitude_32f_aligned16.cc +++ b/volk/lib/qa_16sc_magnitude_32f_aligned16.cc @@ -16,6 +16,7 @@ void qa_16sc_magnitude_32f_aligned16::t1() { std::complex<int16_t> input0[vlen] __attribute__ ((aligned (16))); float output_generic[vlen] __attribute__ ((aligned (16))); + float output_orc[vlen] __attribute__ ((aligned (16))); float output_known[vlen] __attribute__ ((aligned (16))); int16_t* inputLoad = (int16_t*)input0; @@ -38,6 +39,14 @@ void qa_16sc_magnitude_32f_aligned16::t1() { end = clock(); total = (double)(end-start)/(double)CLOCKS_PER_SEC; printf("generic_time: %f\n", total); + + start = clock(); + for(int count = 0; count < ITERS; ++count) { + volk_16sc_magnitude_32f_aligned16_manual(output_orc, input0, scale, vlen, "orc"); + } + end = clock(); + total = (double)(end-start)/(double)CLOCKS_PER_SEC; + printf("orc_time: %f\n", total); /* for(int i = 0; i < 100; ++i) { @@ -49,6 +58,7 @@ void qa_16sc_magnitude_32f_aligned16::t1() { for(int i = 0; i < vlen; ++i) { //printf("%d...%d\n", output0[i], output01[i]); CPPUNIT_ASSERT_DOUBLES_EQUAL(output_generic[i], output_known[i], fabs(output_generic[i])*1e-4); + CPPUNIT_ASSERT_DOUBLES_EQUAL(output_orc[i], output_known[i], fabs(output_generic[i])*1e-4); } } @@ -64,6 +74,7 @@ void qa_16sc_magnitude_32f_aligned16::t1() { std::complex<int16_t> input0[vlen] __attribute__ ((aligned (16))); float output_generic[vlen] __attribute__ ((aligned (16))); + float output_orc[vlen] __attribute__ ((aligned (16))); float output_sse[vlen] __attribute__ ((aligned (16))); float output_sse3[vlen] __attribute__ ((aligned (16))); @@ -80,6 +91,14 @@ void qa_16sc_magnitude_32f_aligned16::t1() { end = clock(); total = (double)(end-start)/(double)CLOCKS_PER_SEC; printf("generic_time: %f\n", total); +/* start = clock(); + for(int count = 0; count < ITERS; ++count) { + volk_16sc_magnitude_32f_aligned16_manual(output_orc, input0, 32768.0, vlen, "orc"); + } + end = clock(); + total = (double)(end-start)/(double)CLOCKS_PER_SEC; + printf("orc_time: %f\n", total); +*/ start = clock(); for(int count = 0; count < ITERS; ++count) { volk_16sc_magnitude_32f_aligned16_manual(output_sse, input0, 32768.0, vlen, "sse"); @@ -105,6 +124,7 @@ void qa_16sc_magnitude_32f_aligned16::t1() { //printf("%d...%d\n", output0[i], output01[i]); CPPUNIT_ASSERT_DOUBLES_EQUAL(output_generic[i], output_sse[i], fabs(output_generic[i])*1e-4); CPPUNIT_ASSERT_DOUBLES_EQUAL(output_generic[i], output_sse3[i], fabs(output_generic[i])*1e-4); +// CPPUNIT_ASSERT_DOUBLES_EQUAL(output_generic[i], output_orc[i], fabs(output_generic[i])*1e-4); } } |