Build if supporting __ARM_FEATURE_FP16_VECTOR_ARITHMETIC (#1436)

* acl_executor.cc fixed bug: scratch_mem invalid memory released condition

* fc_hcl_arm.c fixed bug: failed build if supporting __ARM_FEATURE_FP16_VECTOR_ARITHMETIC

* Fixed build for TENGINE_ARCH_ARM_82

---------

Co-authored-by: Artem Kopytin <[email protected]>

Author: slipmetal

source/device/cpu/op/conv/cortex-a/armv8.2/conv_dw_kernel_fp16_arm82.c

@@ -25,7 +25,9 @@
 #include <stdint.h>
 #include <stdlib.h>
 #include <math.h>
-#include "compiler_fp16.h"
+#include "graph/tensor.h"
+#include "utility/log.h"
+#include "common/compiler_fp16.h"
 #include "conv_dw_kernel_fp16_arm82.h"
 
 void dw_k3s1p1_fp16_a76(__fp16* input, __fp16* kernel, __fp16* output, long channel_number, long input_w, long input_h, __fp16* bias);

source/device/cpu/op/conv/cortex-a/armv8.2/conv_kernel_fp16_arm82.c

@@ -27,8 +27,8 @@
 #include <arm_neon.h>
 #include <sys/time.h>
 
-#include "conv_kernel_arm.h"
-#include "compiler_fp16.h"
+#include "../conv_kernel_arm.h"
+#include "common/compiler_fp16.h"
 
 #define PER_OUT_CHAN 16
 

source/device/cpu/op/fc/cortex-a/fc_hcl_arm.c

@@ -42,7 +42,7 @@
 #include <string.h>
 
 #if __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
-#include "cortex_a/fc_kernel_fp16_arm82.h"
+#include "armv8.2/fc_kernel_fp16_arm82.h"
 #endif
 
 static int prerun(struct node_ops* node_ops, struct exec_node* exec_node, struct exec_graph* exec_graph)