Autotiler CNN Kernel Code Error

[cweekiat]

Hi,

I am trying to deploy an object detection model on the AI-deck.

I have a Yolov8n model which I exported as .tflite format. The tflite model is quantised using NNTool.

When I try to compile the files, I received the following error message:

BUILD_MODEL_SQ8BIT/detectionKernels.c: In function 'S249_Concat':
BUILD_MODEL_SQ8BIT/detectionKernels.c:15049:2: error: unknown type name 'CNN_Concat_Width_Arg_T'
  CNN_Concat_Width_Arg_T S_KerArg0, *KerArg0 = &S_KerArg0;
  ^~~~~~~~~~~~~~~~~~~~~~
BUILD_MODEL_SQ8BIT/detectionKernels.c:15095:9: error: request for member 'W1' in something not a structure or union
  KerArg0->W1 = (unsigned short int) (784);
         ^~
BUILD_MODEL_SQ8BIT/detectionKernels.c:15096:9: error: request for member 'W2' in something not a structure or union
  KerArg0->W2 = (unsigned short int) (196);
         ^~
BUILD_MODEL_SQ8BIT/detectionKernels.c:15097:9: error: request for member 'W3' in something not a structure or union
  KerArg0->W3 = (unsigned short int) (49);
         ^~
BUILD_MODEL_SQ8BIT/detectionKernels.c:15098:9: error: request for member 'W4' in something not a structure or union
  KerArg0->W4 = (unsigned short int) (0);
         ^~
BUILD_MODEL_SQ8BIT/detectionKernels.c:15099:9: error: request for member 'DataSize' in something not a structure or union
  KerArg0->DataSize = (unsigned char) (1);
         ^~
BUILD_MODEL_SQ8BIT/detectionKernels.c:15144:10: error: request for member 'In1' in something not a structure or union
   KerArg0->In1 = (char *__restrict__) (detection_L1_Memory+0+6272*((T0Ind_Total)%2));
          ^~
BUILD_MODEL_SQ8BIT/detectionKernels.c:15145:10: error: request for member 'In2' in something not a structure or union
   KerArg0->In2 = (char *__restrict__) (detection_L1_Memory+12544+1568*((T0Ind_Total)%2));
          ^~
BUILD_MODEL_SQ8BIT/detectionKernels.c:15146:10: error: request for member 'In3' in something not a structure or union
   KerArg0->In3 = (char *__restrict__) (detection_L1_Memory+15680+392*((T0Ind_Total)%2));
          ^~
BUILD_MODEL_SQ8BIT/detectionKernels.c:15147:10: error: request for member 'Out' in something not a structure or union
   KerArg0->Out = (char *__restrict__) (detection_L1_Memory+16464+8232*((T0Ind_Total)%2));
          ^~
BUILD_MODEL_SQ8BIT/detectionKernels.c:15148:10: error: request for member 'H' in something not a structure or union
   KerArg0->H = (unsigned short int) (T0Ind_Last?2:8);
          ^~
In file included from /gap_sdk/tools/autotiler_v3/Emulation/at_api.h:23:0,
                 from /gap_sdk/tools/autotiler_v3/Emulation/Gap.h:33,
                 from BUILD_MODEL_SQ8BIT/detectionKernels.h:5,
                 from BUILD_MODEL_SQ8BIT/detectionKernels.c:1:
BUILD_MODEL_SQ8BIT/detectionKernels.c:15149:33: error: 'CNN_Concat_Width' undeclared (first use in this function); did you mean 'CNN_Copy_void'?
   AT_FORK(gap_ncore(), (void *) CNN_Concat_Width, (void *) KerArg0);
                                 ^
/gap_sdk/tools/autotiler_v3/Emulation/at_api_pmsis.h:634:83: note: in definition of macro 'AT_FORK'
 #define AT_FORK(nb_cores,entry,arg) pi_cl_team_fork((nb_cores),(AT_FORK_FUN_TYPE)(entry),(AT_FORK_ARG_TYPE)(arg))
                                                                                   ^~~~~
BUILD_MODEL_SQ8BIT/detectionKernels.c:15149:33: note: each undeclared identifier is reported only once for each function it appears in
   AT_FORK(gap_ncore(), (void *) CNN_Concat_Width, (void *) KerArg0);
                                 ^
/gap_sdk/tools/autotiler_v3/Emulation/at_api_pmsis.h:634:83: note: in definition of macro 'AT_FORK'
 #define AT_FORK(nb_cores,entry,arg) pi_cl_team_fork((nb_cores),(AT_FORK_FUN_TYPE)(entry),(AT_FORK_ARG_TYPE)(arg))
                                                                                   ^~~~~
make: *** [/gap_sdk/rtos/freeRTOS/vendors/gwt/rules/freeRTOS_rules.mk:379: /module/examples/ai/detection/BUILD/GAP8_V2/GCC_RISCV_FREERTOS/BUILD_MODEL_SQ8BIT/detectionKernels.o] Error 1

Can you provide some assistance please?

nntool_script:

set log_level INFO
adjust
fusions -a expression_matcher 
fusions --scale8
aquant --force_width 8 --scheme SQ8 --transpose safmc_data/*
set input_norm_func 'x:(x-128)*2.00787401'

#set graph_produce_node_names true
#set graph_produce_operinfos true
#set graph_monitor_cycles true
#set l3_flash_device AT_MEM_L3_HFLASH
#set l3_ram_device AT_MEM_L3_HRAM
qshow 0
save_state 

[cweekiat]

FYI: I am using the GAP_SDK environment in the AI-deck docker container.