Volk: Some new basic Orc implementations with QA code

4 files changed, 29 insertions, 0 deletions

diff --git a/volk/lib/qa_16u_byteswap_aligned16.cc b/volk/lib/qa_16u_byteswap_aligned16.cc
index c30b6ba41..b740f91df 100644
--- a/volk/lib/qa_16u_byteswap_aligned16.cc
+++ b/volk/lib/qa_16u_byteswap_aligned16.cc
@@ -30,6 +30,7 @@ void qa_16u_byteswap_aligned16::t1() {
     output0[i] = (uint16_t) ((rand() - (RAND_MAX/2)) / (RAND_MAX/2));
   }
   memcpy(output01, output0, vlen*sizeof(uint16_t));
+  memcpy(output02, output0, vlen*sizeof(uint16_t));
 
   printf("16u_byteswap_aligned\n");
 
diff --git a/volk/lib/qa_32f_divide_aligned16.cc b/volk/lib/qa_32f_divide_aligned16.cc
index 8826bf94f..f104e0443 100644
--- a/volk/lib/qa_32f_divide_aligned16.cc
+++ b/volk/lib/qa_32f_divide_aligned16.cc
@@ -35,6 +35,7 @@ void qa_32f_divide_aligned16::t1() {
   float input1[vlen] __attribute__ ((aligned (16)));
   
   float output0[vlen] __attribute__ ((aligned (16)));
+  float output1[vlen] __attribute__ ((aligned (16)));
   float output_known[vlen] __attribute__ ((aligned (16)));
 
   for(int i = 0; i < vlen; ++i) {   
@@ -51,6 +52,14 @@ void qa_32f_divide_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_32f_divide_aligned16_manual(output1, input0, input1, vlen, "orc");
+  }
+  end = clock();
+  total = (double)(end-start)/(double)CLOCKS_PER_SEC;
+  printf("orc_time: %f\n", total);
 
   /*
   for(int i = 0; i < 10; ++i) {
@@ -61,6 +70,7 @@ void qa_32f_divide_aligned16::t1() {
   
   for(int i = 0; i < vlen; ++i) {
     CPPUNIT_ASSERT_EQUAL(output0[i], output_known[i]);
+    CPPUNIT_ASSERT_EQUAL(output1[i], output_known[i]);
   }
 }
 
diff --git a/volk/lib/qa_32f_multiply_aligned16.cc b/volk/lib/qa_32f_multiply_aligned16.cc
index e52748466..f9c034d70 100644
--- a/volk/lib/qa_32f_multiply_aligned16.cc
+++ b/volk/lib/qa_32f_multiply_aligned16.cc
@@ -78,6 +78,7 @@ void qa_32f_multiply_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_multiply_aligned16::t1() {
   printf("generic_time: %f\n", total);
   start = clock();
   for(int count = 0; count < ITERS; ++count) {
+    volk_32f_multiply_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_multiply_aligned16_manual(output01, input0, input1, vlen, "sse");
   }
   end = clock();
@@ -107,6 +115,7 @@ void qa_32f_multiply_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_subtract_aligned16.cc b/volk/lib/qa_32f_subtract_aligned16.cc
index a7e1b5ae3..5a5a7c9b6 100644
--- a/volk/lib/qa_32f_subtract_aligned16.cc
+++ b/volk/lib/qa_32f_subtract_aligned16.cc
@@ -25,6 +25,7 @@ void qa_32f_subtract_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));
@@ -41,6 +42,13 @@ void qa_32f_subtract_aligned16::t1() {
   printf("generic_time: %f\n", total);
   start = clock();
   for(int count = 0; count < ITERS; ++count) {
+    volk_32f_subtract_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_subtract_aligned16_manual(output01, input0, input1, vlen, "sse");
   }
   end = clock();
@@ -54,6 +62,7 @@ void qa_32f_subtract_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]);
   }
 }