Volk: A bunch of new Orc routines plus a couple of build changes.

32fc_magnitude_16s fails test_all right now.

config/orc.m4

@@ -5,7 +5,7 @@ dnl ORC_CHECK([REQUIRED_VERSION])
 
 AC_DEFUN([ORC_CHECK],
 [
-  ORC_REQ=ifelse([$1], , "0.4.6", [$1])
+  ORC_REQ=ifelse([$1], , "0.4.10", [$1])
   
   enable_orc = auto
   if test "x$enable_orc" != "xno" ; then

include/volk/volk_16sc_deinterleave_16s_aligned16.h

@@ -140,7 +140,19 @@ static inline void volk_16sc_deinterleave_16s_aligned16_generic(int16_t* iBuffer
 }
 #endif /* LV_HAVE_GENERIC */
 
-
+#if LV_HAVE_ORC
+/*!
+  \brief Deinterleaves the complex 16 bit vector into I & Q vector data
+  \param complexVector The complex input vector
+  \param iBuffer The I buffer output data
+  \param qBuffer The Q buffer output data
+  \param num_points The number of complex data values to be deinterleaved
+*/
+extern void volk_16sc_deinterleave_16s_aligned16_orc_impl(int16_t* iBuffer, int16_t* qBuffer, const lv_16sc_t* complexVector, unsigned int num_points);
+static inline void volk_16sc_deinterleave_16s_aligned16_orc(int16_t* iBuffer, int16_t* qBuffer, const lv_16sc_t* complexVector, unsigned int num_points){
+    volk_16sc_deinterleave_16s_aligned16_orc_impl(iBuffer, qBuffer, complexVector, num_points);
+}
+#endif /* LV_HAVE_ORC */
 
 
 #endif /* INCLUDED_VOLK_16sc_DEINTERLEAVE_16S_ALIGNED16_H */

include/volk/volk_16sc_deinterleave_32f_aligned16.h

@@ -89,7 +89,20 @@ static inline void volk_16sc_deinterleave_32f_aligned16_generic(float* iBuffer,
 }
 #endif /* LV_HAVE_GENERIC */
 
-
+#if LV_HAVE_ORC
+  /*!
+    \brief Converts the complex 16 bit vector into floats,scales each data point, and deinterleaves into I & Q vector data
+    \param complexVector The complex input vector
+    \param iBuffer The I buffer output data
+    \param qBuffer The Q buffer output data
+    \param scalar The data value to be divided against each input data value of the input complex vector
+    \param num_points The number of complex data values to be deinterleaved
+  */
+extern void volk_16sc_deinterleave_32f_aligned16_orc_impl(float* iBuffer, float* qBuffer, const lv_16sc_t* complexVector, const float scalar, unsigned int num_points);
+static inline void volk_16sc_deinterleave_32f_aligned16_orc(float* iBuffer, float* qBuffer, const lv_16sc_t* complexVector, const float scalar, unsigned int num_points){
+    volk_16sc_deinterleave_32f_aligned16_orc_impl(iBuffer, qBuffer, complexVector, scalar, num_points);
+}
+#endif /* LV_HAVE_ORC */
 
 
 #endif /* INCLUDED_VOLK_16sc_DEINTERLEAVE_32F_ALIGNED16_H */

include/volk/volk_16sc_deinterleave_real_8s_aligned16.h

@@ -77,7 +77,18 @@ static inline void volk_16sc_deinterleave_real_8s_aligned16_generic(int8_t* iBuf
 }
 #endif /* LV_HAVE_GENERIC */
 
-
+#if LV_HAVE_ORC
+/*!
+  \brief Deinterleaves the complex 16 bit vector into 8 bit I vector data
+  \param complexVector The complex input vector
+  \param iBuffer The I buffer output data
+  \param num_points The number of complex data values to be deinterleaved
+*/
+extern void volk_16sc_deinterleave_real_8s_aligned16_orc_impl(int8_t* iBuffer, const lv_16sc_t* complexVector, unsigned int num_points);
+static inline void volk_16sc_deinterleave_real_8s_aligned16_orc(int8_t* iBuffer, const lv_16sc_t* complexVector, unsigned int num_points){
+    volk_16sc_deinterleave_real_8s_aligned16_orc_impl(iBuffer, complexVector, num_points);
+}
+#endif /* LV_HAVE_ORC */
 
 
 #endif /* INCLUDED_VOLK_16sc_DEINTERLEAVE_REAL_8s_ALIGNED16_H */

include/volk/volk_16sc_magnitude_16s_aligned16.h

@@ -173,16 +173,16 @@ static inline void volk_16sc_magnitude_16s_aligned16_generic(int16_t* magnitudeV
 }
 #endif /* LV_HAVE_GENERIC */
 
-#if LV_HAVE_ORC
+#if LV_HAVE_ORC_DISABLED
 /*!
   \brief Calculates the magnitude of the complexVector and stores the results in the magnitudeVector
   \param complexVector The vector containing the complex input values
   \param magnitudeVector The vector containing the real output values
   \param num_points The number of complex values in complexVector to be calculated and stored into cVector
 */
-extern void volk_16sc_magnitude_16s_aligned16_orc_impl(int16_t* magnitudeVector, const lv_16sc_t* complexVector, unsigned int num_points);
+extern void volk_16sc_magnitude_16s_aligned16_orc_impl(int16_t* magnitudeVector, const lv_16sc_t* complexVector, float scalar, unsigned int num_points);
 static inline void volk_16sc_magnitude_16s_aligned16_orc(int16_t* magnitudeVector, const lv_16sc_t* complexVector, unsigned int num_points){
-    volk_16sc_magnitude_16s_aligned16_orc_impl(magnitudeVector, complexVector, num_points);
+    volk_16sc_magnitude_16s_aligned16_orc_impl(magnitudeVector, complexVector, 32768.0, num_points);
 }
 #endif /* LV_HAVE_ORC */
 

include/volk/volk_16sc_magnitude_32f_aligned16.h

@@ -161,7 +161,7 @@ static inline void volk_16sc_magnitude_32f_aligned16_generic(float* magnitudeVec
 }
 #endif /* LV_HAVE_GENERIC */
 
-#if LV_HAVE_ORC
+#if LV_HAVE_ORC_DISABLED
 /*!
   \brief Calculates the magnitude of the complexVector and stores the results in the magnitudeVector
   \param complexVector The vector containing the complex input values

include/volk/volk_32f_max_aligned16.h

@@ -67,5 +67,19 @@ static inline void volk_32f_max_aligned16_generic(float* cVector, const float* a
 }
 #endif /* LV_HAVE_GENERIC */
 
+#if LV_HAVE_ORC
+/*!
+  \brief Selects maximum value from each entry between bVector and aVector and store their results in the cVector
+  \param cVector The vector where the results will be stored
+  \param aVector The vector to be checked
+  \param bVector The vector to be checked
+  \param num_points The number of values in aVector and bVector to be checked and stored into cVector
+*/
+extern void volk_32f_max_aligned16_orc_impl(float* cVector, const float* aVector, const float* bVector, unsigned int num_points);
+static inline void volk_32f_max_aligned16_orc(float* cVector, const float* aVector, const float* bVector, unsigned int num_points){
+    volk_32f_max_aligned16_orc_impl(cVector, aVector, bVector, num_points);
+}
+#endif /* LV_HAVE_ORC */
+
 
 #endif /* INCLUDED_VOLK_32f_MAX_ALIGNED16_H */

include/volk/volk_32f_min_aligned16.h

@@ -67,5 +67,19 @@ static inline void volk_32f_min_aligned16_generic(float* cVector, const float* a
 }
 #endif /* LV_HAVE_GENERIC */
 
+#if LV_HAVE_ORC
+/*!
+  \brief Selects minimum value from each entry between bVector and aVector and store their results in the cVector
+  \param cVector The vector where the results will be stored
+  \param aVector The vector to be checked
+  \param bVector The vector to be checked
+  \param num_points The number of values in aVector and bVector to be checked and stored into cVector
+*/
+extern void volk_32f_min_aligned16_orc_impl(float* cVector, const float* aVector, const float* bVector, unsigned int num_points);
+static inline void volk_32f_min_aligned16_orc(float* cVector, const float* aVector, const float* bVector, unsigned int num_points){
+    volk_32f_min_aligned16_orc_impl(cVector, aVector, bVector, num_points);
+}
+#endif /* LV_HAVE_ORC */
+
 
 #endif /* INCLUDED_VOLK_32f_MIN_ALIGNED16_H */

lib/qa_16sc_deinterleave_16s_aligned16.cc

@@ -26,6 +26,8 @@ void qa_16sc_deinterleave_16s_aligned16::t1() {
   int16_t output_generic1[vlen] __attribute__ ((aligned (16)));
   int16_t output_sse2[vlen] __attribute__ ((aligned (16)));
   int16_t output_sse21[vlen] __attribute__ ((aligned (16)));
+  int16_t output_orc[vlen] __attribute__ ((aligned (16)));
+  int16_t output_orc1[vlen] __attribute__ ((aligned (16)));
   int16_t output_ssse3[vlen] __attribute__ ((aligned (16)));
   int16_t output_ssse31[vlen] __attribute__ ((aligned (16)));
 
@@ -43,6 +45,13 @@ void qa_16sc_deinterleave_16s_aligned16::t1() {
   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_deinterleave_16s_aligned16_manual(output_orc, output_orc1, input0, 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_deinterleave_16s_aligned16_manual(output_sse2, output_sse21, input0, vlen, "sse2");
   }
@@ -70,6 +79,9 @@ void qa_16sc_deinterleave_16s_aligned16::t1() {
 
     CPPUNIT_ASSERT_EQUAL(output_generic[i],  output_ssse3[i]);
     CPPUNIT_ASSERT_EQUAL(output_generic1[i],  output_ssse31[i]);
+
+    CPPUNIT_ASSERT_EQUAL(output_generic[i],  output_orc[i]);
+    CPPUNIT_ASSERT_EQUAL(output_generic1[i],  output_orc1[i]);
   }
 }
 

lib/qa_16sc_deinterleave_32f_aligned16.cc

@@ -26,6 +26,8 @@ void qa_16sc_deinterleave_32f_aligned16::t1() {
   float output_generic1[vlen] __attribute__ ((aligned (16)));
   float output_sse2[vlen] __attribute__ ((aligned (16)));
   float output_sse21[vlen] __attribute__ ((aligned (16)));
+  float output_orc[vlen] __attribute__ ((aligned (16)));
+  float output_orc1[vlen] __attribute__ ((aligned (16)));
 
   int16_t* loadInput = (int16_t*)input0;
   for(int i = 0; i < vlen*2; ++i) {   
@@ -41,6 +43,13 @@ void qa_16sc_deinterleave_32f_aligned16::t1() {
   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_deinterleave_32f_aligned16_manual(output_orc, output_orc1, 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_deinterleave_32f_aligned16_manual(output_sse2, output_sse21, input0, 32768.0, vlen, "sse");
   }
@@ -57,6 +66,8 @@ void qa_16sc_deinterleave_32f_aligned16::t1() {
     //printf("%d...%d\n", output0[i], output01[i]);
     CPPUNIT_ASSERT_DOUBLES_EQUAL(output_generic[i], output_sse2[i], fabs(output_generic[i])*1e-4);
     CPPUNIT_ASSERT_DOUBLES_EQUAL(output_generic1[i],  output_sse21[i], fabs(output_generic1[i])*1e-4);
+    CPPUNIT_ASSERT_DOUBLES_EQUAL(output_generic[i], output_orc[i], fabs(output_generic[i])*1e-4);
+    CPPUNIT_ASSERT_DOUBLES_EQUAL(output_generic1[i],  output_orc1[i], fabs(output_generic1[i])*1e-4);
   }
 }
 

lib/qa_16sc_deinterleave_real_8s_aligned16.cc

@@ -24,6 +24,7 @@ void qa_16sc_deinterleave_real_8s_aligned16::t1() {
 
   int8_t output_generic[vlen] __attribute__ ((aligned (16)));
   int8_t output_ssse3[vlen] __attribute__ ((aligned (16)));
+  int8_t output_orc[vlen] __attribute__ ((aligned (16)));
 
   int16_t* loadInput = (int16_t*)input0;
   for(int i = 0; i < vlen*2; ++i) {   
@@ -39,6 +40,13 @@ void qa_16sc_deinterleave_real_8s_aligned16::t1() {
   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_deinterleave_real_8s_aligned16_manual(output_orc, input0, 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_deinterleave_real_8s_aligned16_manual(output_ssse3, input0, vlen, "ssse3");
   }
@@ -54,6 +62,7 @@ void qa_16sc_deinterleave_real_8s_aligned16::t1() {
   for(int i = 0; i < vlen; ++i) {
     //printf("%d...%d\n", output0[i], output01[i]);
     CPPUNIT_ASSERT_EQUAL(output_generic[i], output_ssse3[i]);
+    CPPUNIT_ASSERT_EQUAL(output_generic[i], output_orc[i]);
   }
 }
 

lib/qa_16sc_magnitude_16s_aligned16.cc

@@ -40,13 +40,14 @@ void qa_16sc_magnitude_16s_aligned16::t1() {
   end = clock();
   total = (double)(end-start)/(double)CLOCKS_PER_SEC;
   printf("generic_time: %f\n", total);
-  start = clock();
+/*  start = clock();
   for(int count = 0; count < ITERS; ++count) {
     volk_16sc_magnitude_16s_aligned16_manual(output_orc, input0, 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_16s_aligned16_manual(output_sse, input0, vlen, "sse");
@@ -72,7 +73,7 @@ void qa_16sc_magnitude_16s_aligned16::t1() {
     //printf("%d...%d\n", output0[i], output01[i]);
     CPPUNIT_ASSERT_DOUBLES_EQUAL(output_generic[i], output_sse[i], 1.1);
     CPPUNIT_ASSERT_DOUBLES_EQUAL(output_generic[i], output_sse3[i], 1.1);
-    CPPUNIT_ASSERT_DOUBLES_EQUAL(output_generic[i], output_orc[i], 1.1);
+    //CPPUNIT_ASSERT_DOUBLES_EQUAL(output_generic[i], output_orc[i], 1.1);
   }
 }
 

lib/qa_16sc_magnitude_32f_aligned16.cc

@@ -90,14 +90,14 @@ 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();
+/*  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");
@@ -123,7 +123,7 @@ 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);
+//    CPPUNIT_ASSERT_DOUBLES_EQUAL(output_generic[i], output_orc[i], fabs(output_generic[i])*1e-4);
   }
 }
 

lib/qa_32f_max_aligned16.cc

@@ -25,6 +25,7 @@ void qa_32f_max_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));
@@ -40,6 +41,13 @@ void qa_32f_max_aligned16::t1() {
   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_max_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_max_aligned16_manual(output01, input0, input1, vlen, "sse");
   }
@@ -54,6 +62,7 @@ void qa_32f_max_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]);
   }
 }
 

lib/qa_32f_min_aligned16.cc

@@ -25,6 +25,7 @@ void qa_32f_min_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));
@@ -40,6 +41,13 @@ void qa_32f_min_aligned16::t1() {
   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_min_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_min_aligned16_manual(output01, input0, input1, vlen, "sse");
   }
@@ -54,6 +62,7 @@ void qa_32f_min_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]);
   }
 }
 

lib/qa_32fc_magnitude_16s_aligned16.cc

@@ -63,10 +63,10 @@ void qa_32fc_magnitude_16s_aligned16::t1() {
   total = (double)(end-start)/(double)CLOCKS_PER_SEC;
   printf("sse3_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 < 10; ++i) {
+  //  printf("inputs: %f, %f\n", input0[i].real(), input0[i].imag());
+  //  printf("generic... %i, sse3... %i, orc... %i\n", output_generic[i], output_sse3[i], output_orc[i]);
+  //}
 
   for(int i = 0; i < vlen; ++i) {
     //printf("%d...%d\n", output0[i], output01[i]);

lib/qa_volk.cc

@@ -118,7 +118,6 @@ CppUnit::TestSuite *
 qa_volk::suite()
 {
   CppUnit::TestSuite *s = new CppUnit::TestSuite("volk");
-
   s->addTest(qa_16s_quad_max_star_aligned16::suite());
   s->addTest(qa_32fc_dot_prod_aligned16::suite());
   s->addTest(qa_32fc_square_dist_scalar_mult_aligned16::suite());

orc/Makefile.am

@@ -35,13 +35,17 @@ volk_32f_subtract_aligned16_orc_impl.orc \
 volk_32f_divide_aligned16_orc_impl.orc \
 volk_32f_multiply_aligned16_orc_impl.orc \
 volk_32f_sqrt_aligned16_orc_impl.orc \
-volk_16sc_magnitude_32f_aligned16_orc_impl.orc \
+volk_32f_max_aligned16_orc_impl.orc \
+volk_32f_min_aligned16_orc_impl.orc \
 volk_32fc_magnitude_32f_aligned16_orc_impl.orc \
-volk_32fc_magnitude_16s_aligned16_orc_impl.orc
+volk_32fc_magnitude_16s_aligned16_orc_impl.orc \
+volk_16sc_deinterleave_16s_aligned16_orc_impl.orc \
+volk_16sc_deinterleave_32f_aligned16_orc_impl.orc \
+volk_16sc_deinterleave_real_8s_aligned16_orc_impl.orc
 
 
 
-my_ORCC_FLAGS = --implementation --lazy-init $(ORCC_FLAGS)
+my_ORCC_FLAGS = --implementation $(ORCC_FLAGS)
 
 .orc.c:
 	$(ORCC) $(my_ORCC_FLAGS) -o $@ $<

orc/volk_16sc_deinterleave_16s_aligned16_orc_impl.orc

@@ -0,0 +1,5 @@
+.function volk_16sc_deinterleave_16s_aligned16_orc_impl
+.dest 2 idst
+.dest 2 qdst
+.source 4 src
+splitlw qdst, idst, src