合并kunpeng分支和主分支

uavs3 · Jun 29, 2022 · 38f4346 · 38f4346
1 parent cd29508
commit 38f4346
Show file tree

Hide file tree

Showing 26 changed files with 38,042 additions and 1,489 deletions.
diff --git a/README.md b/README.md
@@ -5,6 +5,8 @@
  2) supports to compile for Windows/Linux systems.
  3) optimized for SSE4/AVX2 chips.
  4) 10bit encoding on all supported platforms (without SIMD).
+ 5) The uavs3 codec has supported x86 and arm platforms, and has been tested and verified on the Kunpeng processor. 
+ 6) The ARM platform recommends the Kunpeng processor.
 
 # license
   Copyright reserved by “Peking University Shenzhen Graduate School”, “Peng Cheng Laboratory”, and “Guangdong Bohua UHD Innovation Corporation” <br><br>
@@ -13,7 +15,7 @@
 
 # compile
   The default configuration only support 8bit decoding. <br>
-  To support 10bit streams decoding, edit inc/com_api.h : #define BIT_DEPTH 10
+  To support 10bit streams decoding:cmake -DCOMPILE_10BIT=1 
 
 ## windows
 Prerequisites:
@@ -22,6 +24,8 @@ Prerequisites:
 build:
   1. ./version.bat (to generate version.h)
   2. solution file: build/x86_windows/uavs3e.sln 
+
+  To support 10bit streams decoding, edit inc/com_api.h : #define COMPILE_10BIT 1 
 
 ## linux
 Prerequisites:
@@ -30,9 +34,10 @@ Prerequisites:
 
 Build:
   1. mkdir build/linux
-  2. cd build/linux && cmake ../..
+  2. cd build/linux && cmake -DCOMPILE_10BIT=0 ../..
   3. make && make install
-
+
+  To support 10bit streams decoding:cmake -DCOMPILE_10BIT=1
   to build shared library, set BUILD_SHARED_LIBS=1 please.
 
 # Run tests
@@ -79,4 +84,4 @@ This program is originally developed by the team of Prof.Ronggang Wang (rgwang@p
 *This program also refers to the following fast algorithm:
   * 陈焕浜, 杨海涛. SMVD编码加速. AVS技术提案, M5218. 2020年3月.
   * 赵寅, 杨海涛. 块划分快速算法. AVS技术提案, M5137. 2020年3月.
-  * 张焕宸, 陈雅梅, 喻莉. 基于灰度共生矩阵与编码信息的帧内预测快速算法. AVS技术提案, M5327.
+  * 张焕宸, 陈雅梅, 喻莉. 基于灰度共生矩阵与编码信息的帧内预测快速算法. AVS技术提案, M5327.
diff --git a/inc/com_api.h b/inc/com_api.h
@@ -43,12 +43,25 @@ extern "C"
 {
 #endif
 
-#define BIT_DEPTH 8
+#if defined(__linux__)
+#if COMPILE_10BIT
+    typedef unsigned short pel;
+    #define BIT_DEPTH 10
+#else
+    typedef unsigned char pel;
+    #define BIT_DEPTH 8
+#endif
 
-#if (BIT_DEPTH == 8)
-    typedef unsigned char     pel; /* pixel type */
 #else
-    typedef unsigned short    pel; /* pixel type */
+#define COMPILE_10BIT 0
+#if COMPILE_10BIT
+    typedef unsigned short pel;
+    #define BIT_DEPTH 10
+#else
+    typedef unsigned char pel;
+    #define BIT_DEPTH 8
+#endif
+
 #endif
 
 /*****************************************************************************

diff --git a/inc/com_util.h b/inc/com_util.h
@@ -262,7 +262,7 @@ u8 com_tree_split(int w, int h, split_mode_t split, u8 slice_type);
 
 
 /* function selection define based on platforms */
-#if (defined(__ANDROID__) && defined(__aarch64__)) || (defined(__APPLE__) && defined(__arm64__))
+#if (defined(__ANDROID__) && defined(__aarch64__)) || (defined(__APPLE__) && defined(__arm64__)) || (defined(__linux__) && defined(__aarch64__))
 #define ENABLE_FUNCTION_C 1
 #define ENABLE_FUNCTION_ARM64 1
 #elif (defined(__ANDROID__) && defined(__arm__)) || (defined(__APPLE__) && defined(__ARM_NEON__))
@@ -373,6 +373,10 @@ void uavs3e_funs_init_sse();
 void uavs3e_funs_init_avx2();
 #endif
 
+#if ENABLE_FUNCTION_ARM64
+void uavs3e_funs_init_arm64();
+#endif
+
 void *uavs3e_align_malloc(int i_size);
 void uavs3e_align_free(void *p);
 
@@ -383,7 +387,7 @@ static avs3_always_inline int uavs3e_get_log2(int v)
     _BitScanReverse(&index, v);
     return index;
 #else
-    return 31 - __builtin_clz(v);
+    return 31 - __builtin_clz(v|1);
 #endif
 }
 
@@ -439,4 +443,4 @@ static int com_img_release(com_img_t *img)
     }
 }
 
-#endif /* _COM_UTIL_H_ */
+#endif /* _COM_UTIL_H_ */
diff --git a/inc/uavs3e.h b/inc/uavs3e.h
@@ -62,6 +62,29 @@ extern "C"
 
 #define PIC_ALIGN_SIZE 8
 
+#if defined(_WIN32) || defined(__linux__)
+#include <sys/timeb.h>
+#elif defined(__GNUC__)
+#include <sys/time.h>
+#endif
+
+#include "stddef.h"
+
+static __inline long long get_mdate(void)
+{
+#if defined(_WIN32) || defined(__linux__)
+    struct timeb tb;
+    ftime(&tb);
+    return ((long long)tb.time * 1000 + (long long)tb.millitm) * 1000;
+#elif defined(__GNUC__) 
+    struct timeval tv;
+    gettimeofday(&tv, NULL);
+    return (long long)tv.tv_sec * 1000000 + tv.tv_usec;
+#else
+    return -1;
+#endif
+}
+
 typedef struct uavs3e_enc_stat_t {
     void          *buf;                // [out] bitstream buffer                                       
     int            bytes;              // [out] size of bitstream

diff --git a/src/CMakeLists.txt b/src/CMakeLists.txt
@@ -3,13 +3,41 @@ set(LIBNAME uavs3e)
 
 # add source
 aux_source_directory(. DIR_UAVS3E_SRC)
-aux_source_directory(./sse DIR_X86_SRC)
-aux_source_directory(./avx2 DIR_X86_256_SRC)
-
-list(APPEND DIR_UAVS3E_SRC ${DIR_X86_CORE})
+if("${CMAKE_SYSTEM_PROCESSOR}" MATCHES "aarch64" OR "${CMAKE_SYSTEM_PROCESSOR}" MATCHES "arm64")
+  message("-- enable_armv8")
+  aux_source_directory(./armv8 DIR_ARM64_SRC)
+  list(APPEND DIR_ARM64_SRC "./armv8/arm64.c")
+  list(APPEND DIR_ARM64_SRC "./armv8/alf_arm64.S")
+  list(APPEND DIR_ARM64_SRC "./armv8/deblock_arm64.S")
+  list(APPEND DIR_ARM64_SRC "./armv8/def_arm64.S")
+  list(APPEND DIR_ARM64_SRC "./armv8/inter_pred_arm64.S")
+  list(APPEND DIR_ARM64_SRC "./armv8/intra_pred_arm64.S")
+  list(APPEND DIR_ARM64_SRC "./armv8/trans_arm64.c")
+  list(APPEND DIR_ARM64_SRC "./armv8/trans_dct_arm64.S")
+  list(APPEND DIR_ARM64_SRC "./armv8/cost_arm64.S")
+  list(APPEND DIR_ARM64_SRC "./armv8/pixel_arm64.S")
+  list(APPEND DIR_ARM64_SRC "./armv8/itrans_dct_arm64.S")
+  list(APPEND DIR_ARM64_SRC "./armv8/sao_arm64.c")
+  list(APPEND DIR_ARM64_SRC "./armv8/sao_kernel_arm64.S")
+  list(APPEND DIR_ARM64_SRC "./armv8/itrans_arm64.c")
+  include_directories("./armv8")
+  enable_language(ASM)
+else()
+  aux_source_directory(./sse DIR_X86_SRC)
+  aux_source_directory(./avx2 DIR_X86_256_SRC)
+  list(APPEND DIR_UAVS3E_SRC ${DIR_X86_CORE})
+endif()
 
 include_directories("../inc")
 
+if(${COMPILE_10BIT})
+    add_definitions(-DCOMPILE_10BIT=1)
+    message("-- compile 10bit")
+else()
+    add_definitions(-DCOMPILE_10BIT=0)
+    message("-- compile 8bit")
+endif()
+
 set_source_files_properties(${DIR_UAVS3E_SRC} PROPERTIES COMPILE_FLAGS "${CMAKE_C_FLAGS} -fPIC -std=c99 -O3")
 set_source_files_properties(${DIR_X86_SRC} PROPERTIES COMPILE_FLAGS "${CMAKE_C_FLAGS} -fPIC -std=c99 -O3 -msse4.2")
 set_source_files_properties(${DIR_X86_256_SRC} PROPERTIES COMPILE_FLAGS "${CMAKE_C_FLAGS} -fPIC -std=c99 -O3 -mavx2")
@@ -65,7 +93,7 @@ else()
 MESSAGE(STATUS "BUILD_SHARED_LIBS \t\t: false")
 endif()
 
-add_library(uavs3e ${DIR_UAVS3E_SRC} ${DIR_X86_256_SRC} ${DIR_X86_SRC})
+add_library(uavs3e ${DIR_UAVS3E_SRC} ${DIR_X86_256_SRC} ${DIR_X86_SRC} ${DIR_ARM64_SRC})
 
 target_link_libraries(uavs3e m)
 if(CMAKE_USE_PTHREADS_INIT)