Volk: Added QA tests for all the Orc stuff. Added a 16u_byteswap but it's broken right now.

8 files changed, 66 insertions, 2 deletions

diff --git a/volk/include/volk/volk_16u_byteswap_aligned16.h b/volk/include/volk/volk_16u_byteswap_aligned16.h
index 698e958e4..8e628ab17 100644
--- a/volk/include/volk/volk_16u_byteswap_aligned16.h
+++ b/volk/include/volk/volk_16u_byteswap_aligned16.h
@@ -61,5 +61,17 @@ static inline void volk_16u_byteswap_aligned16_generic(uint16_t* intsToSwap, uns
 }
 #endif /* LV_HAVE_GENERIC */
 
+#if LV_HAVE_ORC
+/*!
+  \brief Byteswaps (in-place) an aligned vector of int16_t's.
+  \param intsToSwap The vector of data to byte swap
+  \param numDataPoints The number of data points
+*/
+extern void volk_16u_byteswap_aligned16_orc_impl(uint16_t* intsToSwap, uint16_t* intsToSwapAgain, unsigned int num_points);
+static inline void volk_16u_byteswap_aligned16_orc(uint16_t* intsToSwap, unsigned int num_points){
+    volk_16u_byteswap_aligned16_orc_impl(intsToSwap, intsToSwap, num_points);
+}
+#endif /* LV_HAVE_ORC */
+
 
 #endif /* INCLUDED_VOLK_16u_BYTESWAP_ALIGNED16_H */
diff --git a/volk/include/volk/volk_32f_add_aligned16.h b/volk/include/volk/volk_32f_add_aligned16.h
index 721c60fd6..e7d8de265 100644
--- a/volk/include/volk/volk_32f_add_aligned16.h
+++ b/volk/include/volk/volk_32f_add_aligned16.h
@@ -63,7 +63,19 @@ static inline void volk_32f_add_aligned16_generic(float* cVector, const float* a
 }
 #endif /* LV_HAVE_GENERIC */
 
-
+#if LV_HAVE_ORC
+/*!
+  \brief Adds the two input vectors and store their results in the third vector
+  \param cVector The vector where the results will be stored
+  \param aVector One of the vectors to be added
+  \param bVector One of the vectors to be added
+  \param num_points The number of values in aVector and bVector to be added together and stored into cVector
+*/
+extern void volk_32f_add_aligned16_orc_impl(float* cVector, const float* aVector, const float* bVector, unsigned int num_points);
+static inline void volk_32f_add_aligned16_orc(float* cVector, const float* aVector, const float* bVector, unsigned int num_points){
+    volk_32f_add_aligned16_orc_impl(cVector, aVector, bVector, num_points);
+}
+#endif /* LV_HAVE_ORC */
 
 
 #endif /* INCLUDED_VOLK_32f_ADD_ALIGNED16_H */
diff --git a/volk/lib/qa_16u_byteswap_aligned16.cc b/volk/lib/qa_16u_byteswap_aligned16.cc
index 6b19828a4..c30b6ba41 100644
--- a/volk/lib/qa_16u_byteswap_aligned16.cc
+++ b/volk/lib/qa_16u_byteswap_aligned16.cc
@@ -24,6 +24,7 @@ void qa_16u_byteswap_aligned16::t1() {
   
   uint16_t output0[vlen] __attribute__ ((aligned (16)));
   uint16_t output01[vlen] __attribute__ ((aligned (16)));
+  uint16_t output02[vlen] __attribute__ ((aligned (16)));
 
   for(int i = 0; i < vlen; ++i) {   
     output0[i] = (uint16_t) ((rand() - (RAND_MAX/2)) / (RAND_MAX/2));
@@ -41,6 +42,13 @@ void qa_16u_byteswap_aligned16::t1() {
   printf("generic_time: %f\n", total);
   start = clock();
   for(int count = 0; count < ITERS; ++count) {
+    volk_16u_byteswap_aligned16_manual(output02, 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_16u_byteswap_aligned16_manual(output01, vlen, "sse2");
   }
   end = clock();
@@ -54,6 +62,7 @@ void qa_16u_byteswap_aligned16::t1() {
   for(int i = 0; i < vlen; ++i) {
     //printf("%d...%d\n", output0[i], output01[i]);
     CPPUNIT_ASSERT_EQUAL(output0[i], output01[i]);
+    CPPUNIT_ASSERT_EQUAL(output0[i], output02[i]);    
   }
 }
 
diff --git a/volk/lib/qa_32f_add_aligned16.cc b/volk/lib/qa_32f_add_aligned16.cc
index 002aebfc9..d9214e8a2 100644
--- a/volk/lib/qa_32f_add_aligned16.cc
+++ b/volk/lib/qa_32f_add_aligned16.cc
@@ -78,6 +78,7 @@ void qa_32f_add_aligned16::t1() {
   
   float output0[vlen] __attribute__ ((aligned (16)));
   float output01[vlen] __attribute__ ((aligned (16)));
+  float output02[vlen] __attribute__ ((aligned (16)));
 
   for(int i = 0; i < vlen; ++i) {   
     input0[i] = ((float) (rand() - (RAND_MAX/2))) / static_cast<float>((RAND_MAX/2));
@@ -94,6 +95,13 @@ void qa_32f_add_aligned16::t1() {
   printf("generic_time: %f\n", total);
   start = clock();
   for(int count = 0; count < ITERS; ++count) {
+    volk_32f_add_aligned16_manual(output02, input0, input1, 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_32f_add_aligned16_manual(output01, input0, input1, vlen, "sse");
   }
   end = clock();
@@ -107,6 +115,7 @@ void qa_32f_add_aligned16::t1() {
   for(int i = 0; i < vlen; ++i) {
     //printf("%d...%d\n", output0[i], output01[i]);
     CPPUNIT_ASSERT_EQUAL(output0[i], output01[i]);
+    CPPUNIT_ASSERT_EQUAL(output0[i], output02[i]);
   }
 }
 
diff --git a/volk/lib/qa_32s_and_aligned16.cc b/volk/lib/qa_32s_and_aligned16.cc
index 72d05cf6f..5720ee869 100644
--- a/volk/lib/qa_32s_and_aligned16.cc
+++ b/volk/lib/qa_32s_and_aligned16.cc
@@ -25,6 +25,7 @@ void qa_32s_and_aligned16::t1() {
   
   int32_t output0[vlen] __attribute__ ((aligned (16)));
   int32_t output01[vlen] __attribute__ ((aligned (16)));
+  int32_t output02[vlen] __attribute__ ((aligned (16)));
 
   for(int i = 0; i < vlen; ++i) {   
     input0[i] = ((int32_t) (rand() - (RAND_MAX/2)));
@@ -41,6 +42,13 @@ void qa_32s_and_aligned16::t1() {
   printf("generic_time: %f\n", total);
   start = clock();
   for(int count = 0; count < ITERS; ++count) {
+    volk_32s_and_aligned16_manual(output02, input0, input1, 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_32s_and_aligned16_manual(output01, input0, input1, vlen, "sse");
   }
   end = clock();
@@ -54,6 +62,7 @@ void qa_32s_and_aligned16::t1() {
   for(int i = 0; i < vlen; ++i) {
     //printf("%d...%d\n", output0[i], output01[i]);
     CPPUNIT_ASSERT_EQUAL(output0[i], output01[i]);
+    CPPUNIT_ASSERT_EQUAL(output0[i], output02[i]);
   }
 }
 
diff --git a/volk/lib/qa_8s_convert_32f_aligned16.cc b/volk/lib/qa_8s_convert_32f_aligned16.cc
index 522da0b9d..3b3aa6919 100644
--- a/volk/lib/qa_8s_convert_32f_aligned16.cc
+++ b/volk/lib/qa_8s_convert_32f_aligned16.cc
@@ -40,6 +40,14 @@ void qa_8s_convert_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_8s_convert_32f_aligned16_manual(output_generic, input0, 128.0, 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) {
diff --git a/volk/orc/Makefile.am b/volk/orc/Makefile.am
index 5d82f540f..3f155d0be 100644
--- a/volk/orc/Makefile.am
+++ b/volk/orc/Makefile.am
@@ -29,7 +29,8 @@ volk_32f_sqrt_aligned16_orc_impl.orc \
 volk_8s_convert_16s_aligned16_orc_impl.orc \
 volk_8s_convert_32f_aligned16_orc_impl.orc \
 volk_32s_and_aligned16_orc_impl.orc \
-volk_32f_add_aligned16_orc_impl.orc
+volk_32f_add_aligned16_orc_impl.orc \
+volk_16u_byteswap_aligned16_orc_impl.orc
 
 
 
diff --git a/volk/orc/volk_16u_byteswap_aligned16_orc_impl.orc b/volk/orc/volk_16u_byteswap_aligned16_orc_impl.orc
new file mode 100644
index 000000000..73c3e345e
--- /dev/null
+++ b/volk/orc/volk_16u_byteswap_aligned16_orc_impl.orc
@@ -0,0 +1,4 @@
+.function volk_16u_byteswap_aligned16_orc_impl
+.dest 2 dst
+.source 2 src
+swapw dst, src