diff options
author | Tom Rondeau | 2010-12-07 18:50:28 -0500 |
---|---|---|
committer | Tom Rondeau | 2010-12-07 18:50:28 -0500 |
commit | 239144659b29c0a5ecd83a34e0e57387a1060ed7 (patch) | |
tree | 3476e1c123da4696c64cc1756ddec5d971bcf9f2 /volk/lib/qa_32f_interleave_16sc_aligned16.cc | |
parent | e13783aeb84a2c3656c3344a8d52fa2c9ee38a00 (diff) | |
download | gnuradio-239144659b29c0a5ecd83a34e0e57387a1060ed7.tar.gz gnuradio-239144659b29c0a5ecd83a34e0e57387a1060ed7.tar.bz2 gnuradio-239144659b29c0a5ecd83a34e0e57387a1060ed7.zip |
Initial checkin for VOLK - Vector-Optimized Library of Kernels. This is a new SIMD library.
It currently stands by itself under the GNU Radio tree and can be used separately. We will integrate the build process into GNU Raio and start building off of its functionality over time.
Diffstat (limited to 'volk/lib/qa_32f_interleave_16sc_aligned16.cc')
-rw-r--r-- | volk/lib/qa_32f_interleave_16sc_aligned16.cc | 75 |
1 files changed, 75 insertions, 0 deletions
diff --git a/volk/lib/qa_32f_interleave_16sc_aligned16.cc b/volk/lib/qa_32f_interleave_16sc_aligned16.cc new file mode 100644 index 000000000..2a937637f --- /dev/null +++ b/volk/lib/qa_32f_interleave_16sc_aligned16.cc @@ -0,0 +1,75 @@ +#include <volk/volk.h> +#include <qa_32f_interleave_16sc_aligned16.h> +#include <volk/volk_32f_interleave_16sc_aligned16.h> +#include <cstdlib> + +//test for sse + +#ifndef LV_HAVE_SSE2 + +void qa_32f_interleave_16sc_aligned16::t1() { + printf("sse2 not available... no test performed\n"); +} + +#else + +void qa_32f_interleave_16sc_aligned16::t1() { + + volk_environment_init(); + clock_t start, end; + double total; + const int vlen = 3201; + const int ITERS = 100000; + float input0[vlen] __attribute__ ((aligned (16))); + float input1[vlen] __attribute__ ((aligned (16))); + + std::complex<int16_t> output_generic[vlen] __attribute__ ((aligned (16))); + std::complex<int16_t> output_sse[vlen] __attribute__ ((aligned (16))); + std::complex<int16_t> output_sse2[vlen] __attribute__ ((aligned (16))); + + for(int i = 0; i < vlen; ++i) { + input0[i] = ((float) (rand() - (RAND_MAX/2))) / static_cast<float>((RAND_MAX/2)); + input1[i] = ((float) (rand() - (RAND_MAX/2))) / static_cast<float>((RAND_MAX/2)); + } + printf("32f_interleave_16sc_aligned\n"); + + start = clock(); + for(int count = 0; count < ITERS; ++count) { + volk_32f_interleave_16sc_aligned16_manual(output_generic, input0, input1, 32768.0, vlen, "generic"); + } + 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_32f_interleave_16sc_aligned16_manual(output_sse, input0, input1, 32768.0, vlen, "sse"); + } + end = clock(); + total = (double)(end-start)/(double)CLOCKS_PER_SEC; + printf("sse_time: %f\n", total); + + start = clock(); + for(int count = 0; count < ITERS; ++count) { + volk_32f_interleave_16sc_aligned16_manual(output_sse2, input0, input1, 32768.0, vlen, "sse2"); + } + end = clock(); + total = (double)(end-start)/(double)CLOCKS_PER_SEC; + printf("sse2_time: %f\n", total); + + for(int i = 0; i < 1; ++i) { + //printf("inputs: %d, %d\n", input0[i*2], input0[i*2 + 1]); + //printf("generic... %d, ssse3... %d\n", output0[i], output1[i]); + } + + for(int i = 0; i < vlen; ++i) { + //printf("%d...%d\n", output0[i], output01[i]); + CPPUNIT_ASSERT_DOUBLES_EQUAL(std::real(output_generic[i]), std::real(output_sse[i]), 1.01); + CPPUNIT_ASSERT_DOUBLES_EQUAL(std::imag(output_generic[i]), std::imag(output_sse[i]), 1.01); + + CPPUNIT_ASSERT_DOUBLES_EQUAL(std::real(output_generic[i]), std::real(output_sse2[i]), 1.01); + CPPUNIT_ASSERT_DOUBLES_EQUAL(std::imag(output_generic[i]), std::imag(output_sse2[i]), 1.01); + } +} + +#endif |