left and right use diff pack kernel

StgSharp.Native/sn-head/sn_intrinsic.context.x86.h

@@ -175,17 +175,39 @@ SN_MK_PROC_DECL_STD(float, sse, , quality);
 * 
 * CONVENSIONS:
 * 
-* buffer_1 is pointer to original colum, offset_1 is the offset to the original column, count_1 is the count can be packed
+* buffer_1 is pointer to original colum, 
+* offset_1 is the offset to the original column, 
+* count_1 is the count can be packed
+* 
 * C# shceduler is responsible for calculating the offset and count based on the actual buffer size and matrix dimension, 
 * and make sure that count_1 will not exceeds lenght of the column
 * 
 * scalar pack carries column length and offset data:
-* [column length, offset, buffer requires, empty]
+* [column length, empty, empty, empty]
 */
 SN_MK_PROC_DECL_STD(float, sse, , fma_pack_left);
 SN_MK_PROC_DECL_STD(float, avx, , fma_pack_left);
 SN_MK_PROC_DECL_STD(float, 512, , fma_pack_left);
 
+/*
+* pack right matrix data to L4 buffer
+* 
+* CONVENSIONS:
+* 
+* buffer_1 is pointer to original row, 
+* offset_1 is the offset to the original row, 
+* count_1 is the count can be packed
+* 
+* C# shceduler is responsible for calculating the offset and count based on the actual buffer size and matrix dimension, 
+* and make sure that count_1 will not exceeds lenght of the row
+* 
+* scalar pack carries column length and offset data:
+* [column length, empty, empty, empty]
+*/
+SN_MK_PROC_DECL_STD(float, sse, , fma_pack_right);
+SN_MK_PROC_DECL_STD(float, avx, , fma_pack_right);
+SN_MK_PROC_DECL_STD(float, 512, , fma_pack_right);
+
 SN_MK_PROC_DECL_STD(float, sse, , fma);
 SN_MK_PROC_DECL_STD(float, avx, , fma);
 SN_MK_PROC_DECL_STD(float, avx, _fma, fma);

StgSharp.Native/sn-head/sn_intrinsic.h

@@ -160,23 +160,11 @@ typedef int(SN_DECL *INDEX_PAIR)(short const *stram, uint32_t target, int length
 #pragma endregion
 
 typedef struct mat_mk_intrinsic {
-        UNI_MK_PROC add;
-        UNI_MK_PROC fill;
-        UNI_MK_PROC scalar_mul;
-        UNI_MK_PROC sub;
-        UNI_MK_PROC transpose;
-        UNI_MK_PROC fma;
-        UNI_MK_PROC pivot;
+        UNI_MK_PROC proc_array[10];
 } mat_mk_intrinsic;
 
 typedef struct mat_sk_intrinsic {
-        UNI_SK_PROC add;
-        UNI_SK_PROC fill;
-        UNI_SK_PROC scalar_mul;
-        UNI_SK_PROC sub;
-        UNI_SK_PROC transpose;
-        UNI_SK_PROC fma;
-        UNI_SK_PROC pivot;
+        UNI_SK_PROC proc_array[10];
 } mat_sk_intrinsic;
 
 // Mirrors C# IntrinsicContext field order so managed/unmanaged layouts stay in sync.
@@ -188,6 +176,19 @@ typedef struct sn_intrinsic {
         mat_sk_intrinsic mat_sk[0x0a];
 } sn_intrinsic;
 
+
+typedef enum mat_proc_index {
+        Add = 0,
+        Fill = 1,
+        ScalarMul = 2,
+        Sub = 3,
+        Transpose = 4,
+        ClearPanel = 5,
+        Fma = 6,
+        PackFmaLeft = 7,
+        PackFmaRight = 8,
+        Pivot = 9,
+} mat_proc_index;
 typedef enum elment_mask { F32 = 0, F64 = 1, I32 = 2, I64 = 3 } element_mask;
 
 #endif