summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorNick Foster <nick@nerdnetworks.org>2010-12-14 22:58:33 -0800
committerNick Foster <nick@nerdnetworks.org>2010-12-14 22:58:33 -0800
commitf9ee6a55cb397f9302769a25a8c959fa162354f0 (patch)
tree194211146c78f5cb56ea9ab8c72b231eaba31986
parent21426265324c883c91eeaaf75a81f2ccdc6e249d (diff)
Volk: Some new basic Orc implementations with QA code
Diffstat
-rw-r--r--volk/include/volk/volk_16u_byteswap_aligned16.h4
-rw-r--r--volk/include/volk/volk_32f_divide_aligned16.h13
-rw-r--r--volk/include/volk/volk_32f_multiply_aligned16.h14
-rw-r--r--volk/include/volk/volk_32f_subtract_aligned16.h14
-rw-r--r--volk/lib/qa_16u_byteswap_aligned16.cc1
-rw-r--r--volk/lib/qa_32f_divide_aligned16.cc10
-rw-r--r--volk/lib/qa_32f_multiply_aligned16.cc9
-rw-r--r--volk/lib/qa_32f_subtract_aligned16.cc9
-rw-r--r--volk/orc/Makefile.am7
-rw-r--r--volk/orc/volk_16u_byteswap_aligned16_orc_impl.orc3
-rw-r--r--volk/orc/volk_32f_divide_aligned16_orc_impl.orc5
-rw-r--r--volk/orc/volk_32f_multiply_aligned16_orc_impl.orc5
-rw-r--r--volk/orc/volk_32f_subtract_aligned16_orc_impl.orc5
13 files changed, 92 insertions, 7 deletions
diff --git a/volk/include/volk/volk_16u_byteswap_aligned16.h b/volk/include/volk/volk_16u_byteswap_aligned16.h
index 8e628ab17a..9d19d1a456 100644
--- a/volk/include/volk/volk_16u_byteswap_aligned16.h
+++ b/volk/include/volk/volk_16u_byteswap_aligned16.h
@@ -67,9 +67,9 @@ static inline void volk_16u_byteswap_aligned16_generic(uint16_t* intsToSwap, uns
\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);
+extern void volk_16u_byteswap_aligned16_orc_impl(uint16_t* intsToSwap, 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);
+ volk_16u_byteswap_aligned16_orc_impl(intsToSwap, num_points);
}
#endif /* LV_HAVE_ORC */
diff --git a/volk/include/volk/volk_32f_divide_aligned16.h b/volk/include/volk/volk_32f_divide_aligned16.h
index c00700cd8a..c595b5e92c 100644
--- a/volk/include/volk/volk_32f_divide_aligned16.h
+++ b/volk/include/volk/volk_32f_divide_aligned16.h
@@ -63,6 +63,19 @@ static inline void volk_32f_divide_aligned16_generic(float* cVector, const float
}
#endif /* LV_HAVE_GENERIC */
+#if LV_HAVE_ORC
+/*!
+ \brief Divides the two input vectors and store their results in the third vector
+ \param cVector The vector where the results will be stored
+ \param aVector The vector to be divideed
+ \param bVector The divisor vector
+ \param num_points The number of values in aVector and bVector to be divideed together and stored into cVector
+*/
+extern void volk_32f_divide_aligned16_orc_impl(float* cVector, const float* aVector, const float* bVector, unsigned int num_points);
+static inline void volk_32f_divide_aligned16_orc(float* cVector, const float* aVector, const float* bVector, unsigned int num_points){
+ volk_32f_divide_aligned16_orc_impl(cVector, aVector, bVector, num_points);
+}
+#endif /* LV_HAVE_ORC */
diff --git a/volk/include/volk/volk_32f_multiply_aligned16.h b/volk/include/volk/volk_32f_multiply_aligned16.h
index b557580aba..87ae7bcf8c 100644
--- a/volk/include/volk/volk_32f_multiply_aligned16.h
+++ b/volk/include/volk/volk_32f_multiply_aligned16.h
@@ -63,7 +63,19 @@ static inline void volk_32f_multiply_aligned16_generic(float* cVector, const flo
}
#endif /* LV_HAVE_GENERIC */
-
+#if LV_HAVE_ORC
+/*!
+ \brief Multiplys 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 multiplied
+ \param bVector One of the vectors to be multiplied
+ \param num_points The number of values in aVector and bVector to be multiplied together and stored into cVector
+*/
+extern void volk_32f_multiply_aligned16_orc_impl(float* cVector, const float* aVector, const float* bVector, unsigned int num_points);
+static inline void volk_32f_multiply_aligned16_orc(float* cVector, const float* aVector, const float* bVector, unsigned int num_points){
+ volk_32f_multiply_aligned16_orc_impl(cVector, aVector, bVector, num_points);
+}
+#endif /* LV_HAVE_ORC */
#endif /* INCLUDED_VOLK_32f_MULTIPLY_ALIGNED16_H */
diff --git a/volk/include/volk/volk_32f_subtract_aligned16.h b/volk/include/volk/volk_32f_subtract_aligned16.h
index ac3f5e5d17..e152429016 100644
--- a/volk/include/volk/volk_32f_subtract_aligned16.h
+++ b/volk/include/volk/volk_32f_subtract_aligned16.h
@@ -63,5 +63,19 @@ static inline void volk_32f_subtract_aligned16_generic(float* cVector, const flo
}
#endif /* LV_HAVE_GENERIC */
+#if LV_HAVE_ORC
+/*!
+ \brief Subtracts bVector form aVector and store their results in the cVector
+ \param cVector The vector where the results will be stored
+ \param aVector The initial vector
+ \param bVector The vector to be subtracted
+ \param num_points The number of values in aVector and bVector to be subtracted together and stored into cVector
+*/
+extern void volk_32f_subtract_aligned16_orc_impl(float* cVector, const float* aVector, const float* bVector, unsigned int num_points);
+static inline void volk_32f_subtract_aligned16_orc(float* cVector, const float* aVector, const float* bVector, unsigned int num_points){
+ volk_32f_subtract_aligned16_orc_impl(cVector, aVector, bVector, num_points);
+}
+#endif /* LV_HAVE_ORC */
+
#endif /* INCLUDED_VOLK_32f_SUBTRACT_ALIGNED16_H */
diff --git a/volk/lib/qa_16u_byteswap_aligned16.cc b/volk/lib/qa_16u_byteswap_aligned16.cc
index c30b6ba41f..b740f91df9 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 8826bf94f5..f104e04439 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 e527484663..f9c034d706 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 a7e1b5ae34..5a5a7c9b68 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]);
}
}
diff --git a/volk/orc/Makefile.am b/volk/orc/Makefile.am
index 3f155d0be6..c71625d87b 100644
--- a/volk/orc/Makefile.am
+++ b/volk/orc/Makefile.am
@@ -25,12 +25,15 @@ lib_LTLIBRARIES = libvolk_orc.la
libvolk_orc_la_LDFLAGS = $(ORC_LDFLAGS)
libvolk_orc_la_SOURCES = \
-volk_32f_sqrt_aligned16_orc_impl.orc \
volk_8s_convert_16s_aligned16_orc_impl.orc \
volk_8s_convert_32f_aligned16_orc_impl.orc \
+volk_16u_byteswap_aligned16_orc_impl.orc \
volk_32s_and_aligned16_orc_impl.orc \
volk_32f_add_aligned16_orc_impl.orc \
-volk_16u_byteswap_aligned16_orc_impl.orc
+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
diff --git a/volk/orc/volk_16u_byteswap_aligned16_orc_impl.orc b/volk/orc/volk_16u_byteswap_aligned16_orc_impl.orc
index 73c3e345ea..3ffd12ec0e 100644
--- a/volk/orc/volk_16u_byteswap_aligned16_orc_impl.orc
+++ b/volk/orc/volk_16u_byteswap_aligned16_orc_impl.orc
@@ -1,4 +1,3 @@
.function volk_16u_byteswap_aligned16_orc_impl
.dest 2 dst
-.source 2 src
-swapw dst, src
+swapw dst, dst
diff --git a/volk/orc/volk_32f_divide_aligned16_orc_impl.orc b/volk/orc/volk_32f_divide_aligned16_orc_impl.orc
new file mode 100644
index 0000000000..870843f2af
--- /dev/null
+++ b/volk/orc/volk_32f_divide_aligned16_orc_impl.orc
@@ -0,0 +1,5 @@
+.function volk_32f_divide_aligned16_orc_impl
+.dest 4 dst
+.source 4 src1
+.source 4 src2
+divf dst, src1, src2
diff --git a/volk/orc/volk_32f_multiply_aligned16_orc_impl.orc b/volk/orc/volk_32f_multiply_aligned16_orc_impl.orc
new file mode 100644
index 0000000000..23619af4ed
--- /dev/null
+++ b/volk/orc/volk_32f_multiply_aligned16_orc_impl.orc
@@ -0,0 +1,5 @@
+.function volk_32f_multiply_aligned16_orc_impl
+.dest 4 dst
+.source 4 src1
+.source 4 src2
+mulf dst, src1, src2
diff --git a/volk/orc/volk_32f_subtract_aligned16_orc_impl.orc b/volk/orc/volk_32f_subtract_aligned16_orc_impl.orc
new file mode 100644
index 0000000000..17dbcad465
--- /dev/null
+++ b/volk/orc/volk_32f_subtract_aligned16_orc_impl.orc
@@ -0,0 +1,5 @@
+.function volk_32f_subtract_aligned16_orc_impl
+.dest 4 dst
+.source 4 src1
+.source 4 src2
+subf dst, src1, src2
generated by cgit v1.2.3 (git 2.39.1) at 2025-03-16 21:43:45 +0000