diff --git a/cmake/cmake.define b/cmake/cmake.define
index d32200bb91..1b9a06d9e4 100644
--- a/cmake/cmake.define
+++ b/cmake/cmake.define
@@ -117,12 +117,18 @@ ELSE ()
     IF (${BUILD_SANITIZER})
         SET(CMAKE_C_FLAGS "${CMAKE_C_FLAGS}     -Werror -Werror=return-type -fPIC -gdwarf-2 -fsanitize=address -fsanitize=undefined -fsanitize-recover=all -fsanitize=float-divide-by-zero -fsanitize=float-cast-overflow -fno-sanitize=shift-base -fno-sanitize=alignment -g3 -Wformat=0")
         SET(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Werror -Wno-literal-suffix -Werror=return-type -fPIC -gdwarf-2 -fsanitize=address -fsanitize=undefined -fsanitize-recover=all -fsanitize=float-divide-by-zero -fsanitize=float-cast-overflow -fno-sanitize=shift-base -fno-sanitize=alignment -g3 -Wformat=0")
-	      MESSAGE(STATUS "Will compile with Address Sanitizer!")
+	      MESSAGE(STATUS "Compile with Address Sanitizer!")
     ELSE ()
         SET(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -Werror -Werror=return-type -fPIC -gdwarf-2 -g3 -Wformat=2 -Wno-format-nonliteral -Wno-format-truncation -Wno-format-y2k")
         SET(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Werror -Wno-literal-suffix -Werror=return-type -fPIC -gdwarf-2 -g3 -Wformat=2 -Wno-format-nonliteral -Wno-format-truncation -Wno-format-y2k")
     ENDIF ()
 
+    # disable all assert
+    IF ((${DISABLE_ASSERT} MATCHES "true") OR (${DISABLE_ASSERTS} MATCHES "true"))
+        ADD_DEFINITIONS(-DDISABLE_ASSERT)
+        MESSAGE(STATUS "Disable all asserts")
+    ENDIF()
+
     INCLUDE(CheckCCompilerFlag)
     IF (TD_ARM_64 OR TD_ARM_32)
         SET(COMPILER_SUPPORT_SSE42 false)
@@ -155,7 +161,7 @@ ELSE ()
             SET(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -mavx2")
             SET(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mavx2")
         ENDIF()
-        MESSAGE(STATUS "SIMD instructions (AVX/AVX2) is ACTIVATED")
+        MESSAGE(STATUS "SIMD instructions (FMA/AVX/AVX2) is ACTIVATED")
     ENDIF()
 
 ENDIF ()
diff --git a/include/util/tlog.h b/include/util/tlog.h
index c7158def29..6e9b304e1d 100644
--- a/include/util/tlog.h
+++ b/include/util/tlog.h
@@ -85,12 +85,19 @@ void taosPrintLongString(const char *flags, ELogLevel level, int32_t dflag, cons
 
 bool taosAssertDebug(bool condition, const char *file, int32_t line, const char *format, ...);
 bool taosAssertRelease(bool condition);
+
+// Disable all asserts that may compromise the performance.
+#if defined DISABLE_ASSERT
+#define ASSERT(condition)
+#define ASSERTS(condition, ...)
+#else
 #define ASSERTS(condition, ...) taosAssertDebug(condition, __FILE__, __LINE__, __VA_ARGS__)
 #ifdef NDEBUG
 #define ASSERT(condition) taosAssertRelease(condition)
 #else
 #define ASSERT(condition) taosAssertDebug(condition, __FILE__, __LINE__, "assert info not provided")
 #endif
+#endif
 
 // clang-format off
 #define uFatal(...) { if (uDebugFlag & DEBUG_FATAL) { taosPrintLog("UTL FATAL", DEBUG_FATAL, tsLogEmbedded ? 255 : uDebugFlag, __VA_ARGS__); }}
diff --git a/source/libs/executor/src/dataInserter.c b/source/libs/executor/src/dataInserter.c
index 346fcc9729..99d1968a5d 100644
--- a/source/libs/executor/src/dataInserter.c
+++ b/source/libs/executor/src/dataInserter.c
@@ -41,6 +41,7 @@ typedef struct SDataInserterHandle {
   SHashObj*           pCols;
   int32_t             status;
   bool                queryEnd;
+  bool                fullOrderColList;
   uint64_t            useconds;
   uint64_t            cachedSize;
   TdThreadMutex       mutex;
@@ -125,7 +126,6 @@ int32_t dataBlockToSubmit(SDataInserterHandle* pInserter, SSubmitReq** pReq) {
   int64_t         uid = pInserter->pNode->tableId;
   int64_t         suid = pInserter->pNode->stableId;
   int32_t         vgId = pInserter->pNode->vgId;
-  bool            fullCol = (pInserter->pNode->pCols->length == pTSchema->numOfCols);
 
   SSubmitReq* ret = NULL;
   int32_t     sz = taosArrayGetSize(pBlocks);
@@ -176,7 +176,7 @@ int32_t dataBlockToSubmit(SDataInserterHandle* pInserter, SSubmitReq** pReq) {
         const STColumn*  pColumn = &pTSchema->columns[k];
         SColumnInfoData* pColData = NULL;
         int16_t          colIdx = k;
-        if (!fullCol) {
+        if (!pInserter->fullOrderColList) {
           int16_t* slotId = taosHashGet(pInserter->pCols, &pColumn->colId, sizeof(pColumn->colId));
           if (NULL == slotId) {
             continue;
@@ -212,7 +212,7 @@ int32_t dataBlockToSubmit(SDataInserterHandle* pInserter, SSubmitReq** pReq) {
           tdAppendColValToRow(&rb, pColumn->colId, pColumn->type, TD_VTYPE_NORM, data, true, pColumn->offset, k);
         }
       }
-      if (!fullCol) {
+      if (!pInserter->fullOrderColList) {
         rb.hasNone = true;
       }
       tdSRowEnd(&rb);
@@ -346,12 +346,19 @@ int32_t createDataInserter(SDataSinkManager* pManager, const SDataSinkNode* pDat
     return TSDB_CODE_OUT_OF_MEMORY;
   }
 
+  inserter->fullOrderColList = pInserterNode->pCols->length == inserter->pSchema->numOfCols;
+
   inserter->pCols = taosHashInit(pInserterNode->pCols->length, taosGetDefaultHashFunction(TSDB_DATA_TYPE_SMALLINT),
                                  false, HASH_NO_LOCK);
   SNode* pNode = NULL;
+  int32_t i = 0;
   FOREACH(pNode, pInserterNode->pCols) {
     SColumnNode* pCol = (SColumnNode*)pNode;
     taosHashPut(inserter->pCols, &pCol->colId, sizeof(pCol->colId), &pCol->slotId, sizeof(pCol->slotId));
+    if (inserter->fullOrderColList && pCol->colId != inserter->pSchema->columns[i].colId) {
+      inserter->fullOrderColList = false;
+    }
+    ++i;
   }
 
   tsem_init(&inserter->ready, 0, 0);
diff --git a/source/util/src/tcompression.c b/source/util/src/tcompression.c
index edc99dbd93..64d550e874 100644
--- a/source/util/src/tcompression.c
+++ b/source/util/src/tcompression.c
@@ -273,42 +273,90 @@ int32_t tsDecompressINTImp(const char *const input, const int32_t nelements, cha
     char    bit = bit_per_integer[(int32_t)selector];  // bit = 3
     int32_t elems = selector_to_elems[(int32_t)selector];
 
-    for (int32_t i = 0; i < elems; i++) {
-      uint64_t zigzag_value;
+    // Optimize the performance, by remove the constantly switch operation.
+    int32_t v = 0;
+    uint64_t zigzag_value;
 
-      if (selector == 0 || selector == 1) {
-        zigzag_value = 0;
-      } else {
-        zigzag_value = ((w >> (4 + bit * i)) & INT64MASK(bit));
-      }
-      int64_t diff = ZIGZAG_DECODE(int64_t, zigzag_value);
-      int64_t curr_value = diff + prev_value;
-      prev_value = curr_value;
+    switch (type) {
+      case TSDB_DATA_TYPE_BIGINT: {
+        for (int32_t i = 0; i < elems; i++) {
+          if (selector == 0 || selector == 1) {
+            zigzag_value = 0;
+          } else {
+            zigzag_value = ((w >> (4 + v)) & INT64MASK(bit));
+          }
+
+          int64_t diff = ZIGZAG_DECODE(int64_t, zigzag_value);
+          int64_t curr_value = diff + prev_value;
+          prev_value = curr_value;
 
-      switch (type) {
-        case TSDB_DATA_TYPE_BIGINT:
           *((int64_t *)output + _pos) = (int64_t)curr_value;
           _pos++;
-          break;
-        case TSDB_DATA_TYPE_INT:
+
+          v += bit;
+          if ((++count) == nelements) break;
+        }
+      } break;
+      case TSDB_DATA_TYPE_INT: {
+        for (int32_t i = 0; i < elems; i++) {
+          if (selector == 0 || selector == 1) {
+            zigzag_value = 0;
+          } else {
+            zigzag_value = ((w >> (4 + v)) & INT64MASK(bit));
+          }
+
+          int64_t diff = ZIGZAG_DECODE(int64_t, zigzag_value);
+          int64_t curr_value = diff + prev_value;
+          prev_value = curr_value;
+
           *((int32_t *)output + _pos) = (int32_t)curr_value;
           _pos++;
-          break;
-        case TSDB_DATA_TYPE_SMALLINT:
+
+          v += bit;
+          if ((++count) == nelements) break;
+        }
+      } break;
+      case TSDB_DATA_TYPE_SMALLINT: {
+        for (int32_t i = 0; i < elems; i++) {
+          if (selector == 0 || selector == 1) {
+            zigzag_value = 0;
+          } else {
+            zigzag_value = ((w >> (4 + v)) & INT64MASK(bit));
+          }
+
+          int64_t diff = ZIGZAG_DECODE(int64_t, zigzag_value);
+          int64_t curr_value = diff + prev_value;
+          prev_value = curr_value;
+
           *((int16_t *)output + _pos) = (int16_t)curr_value;
           _pos++;
-          break;
-        case TSDB_DATA_TYPE_TINYINT:
+
+          v += bit;
+          if ((++count) == nelements) break;
+        }
+      } break;
+
+      case TSDB_DATA_TYPE_TINYINT: {
+        for (int32_t i = 0; i < elems; i++) {
+          if (selector == 0 || selector == 1) {
+            zigzag_value = 0;
+          } else {
+            zigzag_value = ((w >> (4 + v)) & INT64MASK(bit));
+          }
+
+          int64_t diff = ZIGZAG_DECODE(int64_t, zigzag_value);
+          int64_t curr_value = diff + prev_value;
+          prev_value = curr_value;
+
           *((int8_t *)output + _pos) = (int8_t)curr_value;
           _pos++;
-          break;
-        default:
-          perror("Wrong integer types.\n");
-          return -1;
-      }
-      count++;
-      if (count == nelements) break;
+
+          v += bit;
+          if ((++count) == nelements) break;
+        }
+      } break;
     }
+
     ip += LONG_BYTES;
   }
 
@@ -758,7 +806,7 @@ int32_t tsDecompressDoubleImp(const char *const input, const int32_t nelements,
   uint64_t prev_value = 0;
 
   for (int32_t i = 0; i < nelements; i++) {
-    if (i % 2 == 0) {
+    if ((i & 0x01) == 0) {
       flags = input[ipos++];
     }
 
diff --git a/tests/script/tsim/insert/insert_select.sim b/tests/script/tsim/insert/insert_select.sim
index e3374ee277..333964b1d6 100644
--- a/tests/script/tsim/insert/insert_select.sim
+++ b/tests/script/tsim/insert/insert_select.sim
@@ -42,4 +42,24 @@ if $rows != 2 then
   return -1
 endi
 
+print ======== step2
+sql drop database if exists db1;
+sql create database db1 vgroups 1;
+sql use db1;
+sql create table t1(ts timestamp, a int, b int );
+sql create table t2(ts timestamp, a int, b int );
+sql insert into t1 values(1648791211000,1,2);
+sql insert into t2 (ts, b, a) select ts, a, b from t1;
+sql select * from t2;
+if $rows != 1 then
+  return -1
+endi
+if $data01 != 2 then
+  return -1
+endi
+if $data02 != 1 then
+  return -1
+endi
+
+
 system sh/exec.sh -n dnode1 -s stop -x SIGINT