fix memcpy_neon

3 files changed, 45 insertions, 16 deletions

diff --git a/src/camera/veyeimx287m.cpp b/src/camera/veyeimx287m.cpp
index d4c19fe..e89442c 100644
--- a/src/camera/veyeimx287m.cpp
+++ b/src/camera/veyeimx287m.cpp
@@ -81,8 +81,7 @@ std::vector<std::shared_ptr<ICamera> > VeyeIMX287m::search()
     if (!cam->init())
         return {};
 
-    // if (!cam->setExposureTimeUs(30))
-    if (!cam->setExposureTimeUs(250))
+    if (!cam->setExposureTimeUs(30))
         return {};
 
     if (!cam->setLaserLevel(1))
@@ -379,12 +378,25 @@ void VeyeIMX287m::calcFrameLoop(std::stop_token stopToken)
             t.start();
             // std::lock_guard buffer_lock{m_bufferMutexes[bufferIdx]};
             // get: 4100-4500
-            // memcpy(&image.data, m_videoBuffers[bufferIdx], img_size);
+            memcpy(&image.data, m_videoBuffers[bufferIdx], img_size);
             // get: 5000-5100
+            // memcpy_1by1<img_size / sizeof(ARRAY_TYPE)>((ARRAY_TYPE *) &image.data,
+            //                                            (ARRAY_TYPE *) m_videoBuffers[bufferIdx]);
             // memcpy_1by1<img_size>((std::byte *) &image.data,
             //                       (std::byte *) m_videoBuffers[bufferIdx]);
-            memcpy_neon<img_size / sizeof(ARRAY_TYPE)>((ARRAY_TYPE *) &image.data,
-                                                       (ARRAY_TYPE *) m_videoBuffers[bufferIdx]);
+
+            // memcpy_neon<img_size / sizeof(ARRAY_TYPE)>((ARRAY_TYPE *) &image.data,
+            //                                            (ARRAY_TYPE *) m_videoBuffers[bufferIdx]);
+            // std::cerr << "size: " << img_size / sizeof(ARRAY_TYPE) << std::endl;
+            // Image::data_t native;
+            // Image::data_t neon;
+            // memcpy(&native, m_videoBuffers[bufferIdx], img_size);
+            // memcpy_neon<img_size / sizeof(ARRAY_TYPE)>((ARRAY_TYPE *) &neon,
+            //                                            (ARRAY_TYPE *) m_videoBuffers[bufferIdx]);
+
+            // if (memcmp(&native, &neon, img_size) != 0) {
+            //     std::cerr << "different: " << img_size / sizeof(ARRAY_TYPE) << std::endl;
+            // }
             get_elapsed_ns += t.nsecsElapsed();
         }
 
diff --git a/src/image.h b/src/image.h
index a3c4476..e4cf9eb 100644
--- a/src/image.h
+++ b/src/image.h
@@ -17,8 +17,6 @@ struct Image
     Image(Image &other) = delete;
     Image(Image &&other) = delete;
     Image &operator=(Image &&other) = default;
-    int width{0};
-    int height{0};
     // uint8_t data[img_height][img_width] = {{0}};
 
     using row_t = std::array<uint8_t, img_width>;
@@ -27,7 +25,9 @@ struct Image
     using column_t = rotated_row_t;
     using rotated_data_t = std::array<column_t, img_width>;
     // data_t d;
-    data_t data;
+    alignas(128) data_t data;
+    int width{0};
+    int height{0};
     // data_t *data;
     // uint8_t *data = {nullptr};
     // uint8_t &dataAt(size_t row, size_t col);
diff --git a/src/mem_utils.h b/src/mem_utils.h
index 8601f78..cb5f179 100644
--- a/src/mem_utils.h
+++ b/src/mem_utils.h
@@ -1,25 +1,42 @@
 #pragma once
 
 #include <cstddef>
+#include <iostream>
 
 #include <arm_neon.h>
 
+// using ARRAY_TYPE = uint16_t;
+// using ARRAY_TYPE = uint32_t;
+using ARRAY_TYPE = uint64_t;
+// using ARRAY_TYPE = uint64_t;
+
 template<std::size_t S>
-void memcpy_1by1(std::byte *dst, const std::byte *src)
+void memcpy_1by1(ARRAY_TYPE *dst, const ARRAY_TYPE *src)
 {
     for (std::size_t i{0}; i < S; ++i) {
         dst[i] = src[i];
     }
 }
-
-using ARRAY_TYPE = uint32_t;
 template<std::size_t S>
-void memcpy_neon(ARRAY_TYPE *dst, const ARRAY_TYPE *src)
+inline void memcpy_neon(ARRAY_TYPE *dst, const ARRAY_TYPE *src)
 {
-    uint32x4_t tmp;
+    // for (std::size_t i{0}; i < S; i += 4) {
+    //     vst1q_u32(&dst[i], vld1q_u32(src + i));
+    // }
+    // if (uint64_t(src) % 16 != 0) {
+    //     std::cerr << "src misaligned" << std::endl;
+    // }
+    // if (uint64_t(dst) % 16 != 0) {
+    //     std::cerr << "dst misaligned" << std::endl;
+    // }
 
-    for (std::size_t i{0}; i < (S / 4); i += 4) {
-        tmp = vld1q_u32(src + i);
-        vst1q_u32(&dst[i], tmp);
+    for (std::size_t i{0}; i < S; i += 2) {
+        vst1q_u64(&dst[i], vld1q_u64(src + i));
     }
+    // for (std::size_t i{0}; i < S; i += 8) {
+    //     vst1q_u16(&dst[i], vld1q_u16(src + i));
+    // }
+    // for (std::size_t i{0}; i < S; i += 4) {
+    //     vst1q_u64(&dst[i], vld1q_u64(src + i));
+    // }
 }