Merge pull request #642 from argilo/fix-polar

Add length checks to volk_8u_x2_encodeframepolar_8u
gnuradio · Oct 22, 2023 · da574b4 · da574b4
2 parents 1f554ce + 6e79380
commit da574b4
Show file tree

Hide file tree

Showing 2 changed files with 22 additions and 10 deletions.
diff --git a/kernels/volk/volk_32f_8u_polarbutterfly_32f.h b/kernels/volk/volk_32f_8u_polarbutterfly_32f.h
@@ -232,11 +232,7 @@ static inline void volk_32f_8u_polarbutterfly_32f_u_avx(float* llrs,
         unsigned char* u_temp = u + 2 * frame_size;
         memcpy(u_temp, u + u_num - stage_size, sizeof(unsigned char) * stage_size);
 
-        if (stage_size > 15) {
-            volk_8u_x2_encodeframepolar_8u_u_ssse3(u_target, u_temp, stage_size);
-        } else {
-            volk_8u_x2_encodeframepolar_8u_generic(u_target, u_temp, stage_size);
-        }
+        volk_8u_x2_encodeframepolar_8u_u_ssse3(u_target, u_temp, stage_size);
 
         src_llr_ptr = llrs + (max_stage_depth + 1) * frame_size + row - stage_size;
         dst_llr_ptr = llrs + max_stage_depth * frame_size + row;
@@ -331,11 +327,7 @@ static inline void volk_32f_8u_polarbutterfly_32f_u_avx2(float* llrs,
         unsigned char* u_temp = u + 2 * frame_size;
         memcpy(u_temp, u + u_num - stage_size, sizeof(unsigned char) * stage_size);
 
-        if (stage_size > 15) {
-            volk_8u_x2_encodeframepolar_8u_u_ssse3(u_target, u_temp, stage_size);
-        } else {
-            volk_8u_x2_encodeframepolar_8u_generic(u_target, u_temp, stage_size);
-        }
+        volk_8u_x2_encodeframepolar_8u_u_avx2(u_target, u_temp, stage_size);
 
         src_llr_ptr = llrs + (max_stage_depth + 1) * frame_size + row - stage_size;
         dst_llr_ptr = llrs + max_stage_depth * frame_size + row;

diff --git a/kernels/volk/volk_8u_x2_encodeframepolar_8u.h b/kernels/volk/volk_8u_x2_encodeframepolar_8u.h
@@ -77,6 +77,11 @@ static inline void volk_8u_x2_encodeframepolar_8u_u_ssse3(unsigned char* frame,
                                                           unsigned char* temp,
                                                           unsigned int frame_size)
 {
+    if (frame_size < 16) {
+        volk_8u_x2_encodeframepolar_8u_generic(frame, temp, frame_size);
+        return;
+    }
+
     const unsigned int po2 = log2_of_power_of_2(frame_size);
 
     unsigned int stage = po2;
@@ -256,6 +261,11 @@ static inline void volk_8u_x2_encodeframepolar_8u_u_avx2(unsigned char* frame,
                                                          unsigned char* temp,
                                                          unsigned int frame_size)
 {
+    if (frame_size < 32) {
+        volk_8u_x2_encodeframepolar_8u_generic(frame, temp, frame_size);
+        return;
+    }
+
     const unsigned int po2 = log2_of_power_of_2(frame_size);
 
     unsigned int stage = po2;
@@ -612,6 +622,11 @@ static inline void volk_8u_x2_encodeframepolar_8u_a_ssse3(unsigned char* frame,
                                                           unsigned char* temp,
                                                           unsigned int frame_size)
 {
+    if (frame_size < 16) {
+        volk_8u_x2_encodeframepolar_8u_generic(frame, temp, frame_size);
+        return;
+    }
+
     const unsigned int po2 = log2_of_power_of_2(frame_size);
 
     unsigned int stage = po2;
@@ -790,6 +805,11 @@ static inline void volk_8u_x2_encodeframepolar_8u_a_avx2(unsigned char* frame,
                                                          unsigned char* temp,
                                                          unsigned int frame_size)
 {
+    if (frame_size < 32) {
+        volk_8u_x2_encodeframepolar_8u_generic(frame, temp, frame_size);
+        return;
+    }
+
     const unsigned int po2 = log2_of_power_of_2(frame_size);
 
     unsigned int stage = po2;