fix function call of liboptimize

mindspore/lite/nnacl/int8/conv_int8.c

@@ -132,7 +132,8 @@ void IndirectGemmInt8Opt(int8_t *dst, int32_t *tmp_dst, const int8_t *src, const
     size_t asymmetric = conv_param->conv_quant_arg_.asymmetric_ & FILTER_ASYMMETRIC;
     size_t per_channel = conv_param->conv_quant_arg_.per_channel_ & FILTER_PER_CHANNEL;
     gemm_func(dst, src, weight, bias, kernel_plane, ic4, output_channel, output_channel * sizeof(int8_t), input_sum,
-              act_min, act_max, out_zp, out_multiplier, shift_before, shift_after, asymmetric, per_channel);
+              act_min, act_max, out_zp, out_multiplier, shift_before, shift_after, asymmetric, per_channel,
+              oc4 * C4NUM * sizeof(int32_t));
 #endif
   } else {
     int tile_num = conv_param->tile_num_;

mindspore/lite/nnacl/int8/conv_int8.h

@@ -31,7 +31,7 @@
 typedef void (*GEMM_FUNC)(int8_t *dst, const int8_t *src, const int8_t *weight, const int32_t *bias, size_t ksize,
                           size_t ic4, size_t output_channel, size_t offset, const int32_t *input_sum, size_t act_min,
                           size_t act_max, size_t out_zp, int32_t *out_multiplier, int32_t *shift_before,
-                          int32_t *shift_after, size_t asymmetric, size_t per_channel);
+                          int32_t *shift_after, size_t asymmetric, size_t per_channel, size_t per_channel_offset);
 
 #ifdef __cplusplus
 extern "C" {

mindspore/lite/nnacl/opt_op_handler.c

@@ -25,7 +25,7 @@ extern void IndirectGemmInt8_24x4_dp(int8_t *dst, const int8_t *src, const int8_
                                      size_t ksize, size_t ic4, size_t output_channel, size_t offset,
                                      const int32_t *input_sum, size_t act_min, size_t act_max, size_t out_zp,
                                      int32_t *out_multiplier, int32_t *shift_before, int32_t *shift_after,
-                                     size_t asymmetric, size_t per_channel);
+                                     size_t asymmetric, size_t per_channel, size_t per_channel_offset);
 
 extern void MatMulOptR4Int8Neon64(const int8_t *a, const int8_t *b, int *dst, int row4, int col4, int deep16,
                                   const int *input_sum, const int *bias);
@@ -43,9 +43,10 @@ void IndirectGemmInt8_optimize_handler(int8_t *dst, const int8_t *src, const int
                                        size_t ksize, size_t ic4, size_t output_channel, size_t offset,
                                        const int32_t *input_sum, size_t act_min, size_t act_max, size_t out_zp,
                                        int32_t *out_multiplier, int32_t *shift_before, int32_t *shift_after,
-                                       size_t asymmetric, size_t per_channel) {
+                                       size_t asymmetric, size_t per_channel, size_t per_channel_offset) {
   return IndirectGemmInt8_24x4_dp(dst, src, weight, bias, ksize, ic4, output_channel, offset, input_sum, act_min,
-                                  act_max, out_zp, out_multiplier, shift_before, shift_after, asymmetric, per_channel);
+                                  act_max, out_zp, out_multiplier, shift_before, shift_after, asymmetric, per_channel,
+                                  per_channel_offset);
 }
 
 void MatMulR4Int8_optimize_handler(const int8_t *a, const int8_t *b, int *dst, int row4, int col4, int deep16,