Improve ANSCV with sotfware decoder:

Thread-local staging Mat (video_player.cpp:1400-1407) — single biggest win. Eliminates the 12 MB per-call malloc/free cycle.
Contiguous get_buffer2 allocator (video_decoder.cpp:35-102) — keeps the 3 bulk memcpys cache-friendly. Would also enable FAST/zero-copy for resolutions where visible_h % 64 == 0.
SW-decoder thread config (video_decoder.cpp:528-540) — thread_count=0, thread_type=FRAME|SLICE. FRAME is downgraded to SLICE-only by AV_CODEC_FLAG_LOW_DELAY, but decode throughput is sufficient for your input rate.
SetTargetFPS(100) delivery throttle (already there) — caps onVideoFrame post-decode work at 10 FPS. Keeps the caller path warm-cached.
Instrumentation — [MEDIA_DecInit] / [MEDIA_Convert] / [MEDIA_SWDec] / [MEDIA_Timing] / [MEDIA_JpegTiming] — always-on regression detector, zero cost when ANSCORE_DEBUGVIEW=OFF.

CMakeLists.txt

@@ -51,11 +51,19 @@ if(MSVC)
         "$<$<COMPILE_LANGUAGE:C,CXX>:/W3>"
         "$<$<COMPILE_LANGUAGE:C,CXX>:/utf-8>"
         "$<$<AND:$<COMPILE_LANGUAGE:C,CXX>,$<NOT:$<CONFIG:MINSIZEREL>>>:/Zi>"
+        # RelWithDebInfo: keep /O2 but disable inlining so debuggers can land
+        # breakpoints on small dispatch functions (e.g. avframeToCVMat).
+        "$<$<AND:$<COMPILE_LANGUAGE:C,CXX>,$<CONFIG:RelWithDebInfo>>:/Ob0>"
     )
     add_link_options(
         "$<$<NOT:$<CONFIG:DEBUG>>:/DEBUG:FULL>"
-        "$<$<NOT:$<CONFIG:DEBUG>>:/OPT:REF>"
-        "$<$<NOT:$<CONFIG:DEBUG>>:/OPT:ICF>"
+        # /OPT:REF and /OPT:ICF improve Release size/perf but confuse the
+        # debugger (folds identical functions, strips unused ones). Apply
+        # them to Release only, not RelWithDebInfo.
+        "$<$<CONFIG:Release>:/OPT:REF>"
+        "$<$<CONFIG:Release>:/OPT:ICF>"
+        "$<$<CONFIG:RelWithDebInfo>:/OPT:NOREF>"
+        "$<$<CONFIG:RelWithDebInfo>:/OPT:NOICF>"
     )
     add_compile_definitions(_CRT_SECURE_NO_WARNINGS _WINSOCK_DEPRECATED_NO_WARNINGS)
 elseif(CMAKE_CXX_COMPILER_ID MATCHES "Clang|GNU")
@@ -89,6 +97,35 @@ else()
     message(STATUS "ANSCORE_DEBUGVIEW = OFF — ANS_DBG verbose logging disabled (production)")
 endif()
 
+# ── Vendored libyuv (submodule: 3rdparty/libyuv) ────────────────
+# SIMD-optimized YUV conversion library. Only genuinely fast on toolchains
+# that understand GCC inline assembly (Clang/clang-cl/GCC) — 144 of its
+# SIMD row functions live in row_gcc.cc. MSVC (cl.exe) can only compile the
+# 8 routines in row_win.cc, so on MSVC libyuv silently falls back to the
+# scalar C code in row_common.cc and runs ~10× slower than OpenCV+IPP's
+# cv::cvtColor(COLOR_YUV2BGR_I420). Do NOT enable on MSVC builds.
+#
+# Default: OFF. Enable with `-DANSCORE_USE_LIBYUV=ON` only after switching
+# the project's compiler to clang-cl / clang / gcc.
+option(ANSCORE_USE_LIBYUV "Use libyuv for YUV→BGR conversion (only effective on Clang/GCC, NOT MSVC)" OFF)
+if(ANSCORE_USE_LIBYUV AND EXISTS "${CMAKE_SOURCE_DIR}/3rdparty/libyuv/CMakeLists.txt")
+    if(MSVC AND NOT CMAKE_CXX_COMPILER_ID STREQUAL "Clang")
+        message(WARNING "ANSCORE_USE_LIBYUV=ON but compiler is MSVC cl.exe — "
+                        "libyuv's SIMD paths (row_gcc.cc) won't compile. "
+                        "Expect ~10× slowdown vs cv::cvtColor. "
+                        "Use clang-cl (LLVM) instead, or keep ANSCORE_USE_LIBYUV=OFF.")
+    endif()
+    # Prevent libyuv from finding a system libjpeg and enabling HAVE_JPEG
+    # (we don't use libyuv's MJPEG codepaths; avoids an unnecessary runtime dep).
+    set(CMAKE_DISABLE_FIND_PACKAGE_JPEG TRUE)
+    add_subdirectory(3rdparty/libyuv EXCLUDE_FROM_ALL)
+    set(ANSCORE_HAS_LIBYUV ON)
+    message(STATUS "libyuv: ENABLED (vendored from 3rdparty/libyuv, static target 'yuv')")
+else()
+    set(ANSCORE_HAS_LIBYUV OFF)
+    message(STATUS "libyuv: DISABLED — using cv::cvtColor+IPP path (fast on MSVC)")
+endif()
+
 # ── External Dependencies ───────────────────────────────────────
 include(cmake/Dependencies.cmake)