Merge remote branch 'bistromath/volk_fix' into safe_align

Conflicts:
	volk/apps/volk_profile.cc

2 files changed, 32 insertions, 5 deletions

diff --git a/volk/lib/qa_utils.cc b/volk/lib/qa_utils.cc
index 9bb515e9f..bb37801c9 100644
--- a/volk/lib/qa_utils.cc
+++ b/volk/lib/qa_utils.cc
@@ -198,6 +198,18 @@ inline void run_cast_test3_s32f(volk_fn_3arg_s32f func, std::vector<void *> &buf
     while(iter--) func(buffs[0], buffs[1], buffs[2], scalar, vlen, arch.c_str());
 }
 
+inline void run_cast_test1_s32fc(volk_fn_1arg_s32fc func, std::vector<void *> &buffs, lv_32fc_t scalar, unsigned int vlen, unsigned int iter, std::string arch) {
+    while(iter--) func(buffs[0], scalar, vlen, arch.c_str());
+}
+
+inline void run_cast_test2_s32fc(volk_fn_2arg_s32fc func, std::vector<void *> &buffs, lv_32fc_t scalar, unsigned int vlen, unsigned int iter, std::string arch) {
+    while(iter--) func(buffs[0], buffs[1], scalar, vlen, arch.c_str());
+}
+
+inline void run_cast_test3_s32fc(volk_fn_3arg_s32fc func, std::vector<void *> &buffs, lv_32fc_t scalar, unsigned int vlen, unsigned int iter, std::string arch) {
+    while(iter--) func(buffs[0], buffs[1], buffs[2], scalar, vlen, arch.c_str());
+}
+
 template <class t>
 bool fcompare(t *in1, t *in2, unsigned int vlen, float tol) {
     bool fail = false;
@@ -246,7 +258,7 @@ bool run_volk_tests(struct volk_func_desc desc,
                     void (*manual_func)(),
                     std::string name,
                     float tol,
-                    float scalar,
+                    lv_32fc_t scalar,
                     int vlen,
                     int iter,
                     std::vector<std::string> *best_arch_vector = 0
@@ -316,21 +328,33 @@ bool run_volk_tests(struct volk_func_desc desc,
                 if(inputsc.size() == 0) {
                     run_cast_test1((volk_fn_1arg)(manual_func), test_data[i], vlen, iter, arch_list[i]); 
                 } else if(inputsc.size() == 1 && inputsc[0].is_float) {
-                    run_cast_test1_s32f((volk_fn_1arg_s32f)(manual_func), test_data[i], scalar, vlen, iter, arch_list[i]);
+                    if(inputsc[0].is_complex) {
+                        run_cast_test1_s32fc((volk_fn_1arg_s32fc)(manual_func), test_data[i], scalar, vlen, iter, arch_list[i]);
+                    } else {
+                        run_cast_test1_s32f((volk_fn_1arg_s32f)(manual_func), test_data[i], scalar.real(), vlen, iter, arch_list[i]);
+                    }
                 } else throw "unsupported 1 arg function >1 scalars";
                 break;
             case 2:
                 if(inputsc.size() == 0) {
                     run_cast_test2((volk_fn_2arg)(manual_func), test_data[i], vlen, iter, arch_list[i]);
                 } else if(inputsc.size() == 1 && inputsc[0].is_float) {
-                    run_cast_test2_s32f((volk_fn_2arg_s32f)(manual_func), test_data[i], scalar, vlen, iter, arch_list[i]);
+                    if(inputsc[0].is_complex) {
+                        run_cast_test2_s32fc((volk_fn_2arg_s32fc)(manual_func), test_data[i], scalar, vlen, iter, arch_list[i]);
+                    } else {
+                        run_cast_test2_s32f((volk_fn_2arg_s32f)(manual_func), test_data[i], scalar.real(), vlen, iter, arch_list[i]);
+                    }
                 } else throw "unsupported 2 arg function >1 scalars";
                 break;
             case 3:
                 if(inputsc.size() == 0) {
                     run_cast_test3((volk_fn_3arg)(manual_func), test_data[i], vlen, iter, arch_list[i]);
                 } else if(inputsc.size() == 1 && inputsc[0].is_float) {
-                    run_cast_test3_s32f((volk_fn_3arg_s32f)(manual_func), test_data[i], scalar, vlen, iter, arch_list[i]);
+                    if(inputsc[0].is_complex) {
+                        run_cast_test3_s32fc((volk_fn_3arg_s32fc)(manual_func), test_data[i], scalar, vlen, iter, arch_list[i]);
+                    } else {
+                        run_cast_test3_s32f((volk_fn_3arg_s32f)(manual_func), test_data[i], scalar.real(), vlen, iter, arch_list[i]);
+                    }
                 } else throw "unsupported 3 arg function >1 scalars";
                 break;
             case 4:
diff --git a/volk/lib/qa_utils.h b/volk/lib/qa_utils.h
index a1bc1f20c..b998df852 100644
--- a/volk/lib/qa_utils.h
+++ b/volk/lib/qa_utils.h
@@ -21,7 +21,7 @@ volk_type_t volk_type_from_string(std::string);
 float uniform(void);
 void random_floats(float *buf, unsigned n);
 
-bool run_volk_tests(struct volk_func_desc, void(*)(), std::string, float, float, int, int, std::vector<std::string> *);
+bool run_volk_tests(struct volk_func_desc, void(*)(), std::string, float, lv_32fc_t, int, int, std::vector<std::string> *);
 
 #define VOLK_RUN_TESTS(func, tol, scalar, len, iter) BOOST_AUTO_TEST_CASE(func##_test) { BOOST_CHECK_EQUAL(run_volk_tests(func##_get_func_desc(), (void (*)())func##_manual, std::string(#func), tol, scalar, len, iter, 0), 0); }
 #define VOLK_PROFILE(func, tol, scalar, len, iter, results) run_volk_tests(func##_get_func_desc(), (void (*)())func##_manual, std::string(#func), tol, scalar, len, iter, results)
@@ -32,5 +32,8 @@ typedef void (*volk_fn_4arg)(void *, void *, void *, void *, unsigned int, const
 typedef void (*volk_fn_1arg_s32f)(void *, float, unsigned int, const char*); //one input vector, one scalar float input
 typedef void (*volk_fn_2arg_s32f)(void *, void *, float, unsigned int, const char*);
 typedef void (*volk_fn_3arg_s32f)(void *, void *, void *, float, unsigned int, const char*);
+typedef void (*volk_fn_1arg_s32fc)(void *, lv_32fc_t, unsigned int, const char*); //one input vector, one scalar float input
+typedef void (*volk_fn_2arg_s32fc)(void *, void *, lv_32fc_t, unsigned int, const char*);
+typedef void (*volk_fn_3arg_s32fc)(void *, void *, void *, lv_32fc_t, unsigned int, const char*);
 
 #endif //VOLK_QA_UTILS_H