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.

1 files changed, 70 insertions, 0 deletions

diff --git a/volk/lib/qa_32f_convert_16s_aligned16.cc b/volk/lib/qa_32f_convert_16s_aligned16.cc
new file mode 100644
index 000000000..84a4c40c4
--- /dev/null
+++ b/volk/lib/qa_32f_convert_16s_aligned16.cc
@@ -0,0 +1,70 @@
+#include <volk/volk.h>
+#include <qa_32f_convert_16s_aligned16.h>
+#include <volk/volk_32f_convert_16s_aligned16.h>
+#include <cstdlib>
+
+//test for sse2
+
+#ifndef LV_HAVE_SSE2
+
+void qa_32f_convert_16s_aligned16::t1() {
+  printf("sse2 not available... no test performed\n");
+}
+
+#else
+
+void qa_32f_convert_16s_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)));
+  
+  int16_t output_generic[vlen] __attribute__ ((aligned (16)));
+  int16_t output_sse[vlen] __attribute__ ((aligned (16)));
+  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));
+  }
+  printf("32f_convert_16s_aligned\n");
+
+  start = clock();
+  for(int count = 0; count < ITERS; ++count) {
+    volk_32f_convert_16s_aligned16_manual(output_generic, input0, 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_convert_16s_aligned16_manual(output_sse, input0, 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_convert_16s_aligned16_manual(output_sse2, input0, 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 < vlen; ++i) {
+    //printf("inputs: %d, %d\n", input0[i*2], input0[i*2 + 1]);
+    //printf("%d generic... %d, sse... %d sse2... %d\n", i, output_generic[i], output_sse[i], output_sse2[i]);
+  }
+  
+  for(int i = 0; i < vlen; ++i) {
+    //printf("%d...%d\n", output0[i], output01[i]);
+    CPPUNIT_ASSERT(abs(output_generic[i] - output_sse[i]) <= 1);
+    CPPUNIT_ASSERT(abs(output_generic[i] - output_sse2[i]) <= 1);
+  }
+}
+
+#endif