data2/text/range/15001-20000/679876.txt

Ticket Name: CCS/TDA2: DATA dma TO L2 fun time not change

Query Text:
Part Number: TDA2 Tool/software: Code Composer Studio Input input and output output data on DDR, I copy the data through EDMA to L2 l2input, L2output, and run the function fun (l2input, L2output), but find that the time is not changed, and the data is similar to DDR on the fun. I don't know what can cause this problem ，follow is the CMD ,i use JTAG run the program /* Linker for DSP1 instance, not DSP2 */ -stack 0x3000 -heap 0x7000000 MEMORY { L1P_SRAM : origin = 0x00E00000, len = 0x8000 L1D_SRAM : origin = 0x00F00000, len = 0x4000 /* 16 KB SRAM */ L1D_CACHE : origin = 0x00F04000, len = 0x4000 /* 16 KB cache */ L2_SRAM : origin = 0x00800000, len = 0x28000 /* SARAM in L2, = 256 + 32 - 128 = 160 KB*/ L2_CACHE : origin = 0x00828000, len = 0x20000 /* Cache for L2, which is configured as 128 KB*/ DSP2_L2_SRAM : origin = 0x40800000, len = 0x48000 SL2_SRAM : origin = 0x5B000000, len = 0x40000 EXT_MEM_CACHE : origin = 0x80000000, len = 0x08000000 /* DSP Used cachable area */ } SECTIONS { vectors :> EXT_MEM_CACHE .cinit :> EXT_MEM_CACHE .cio :> EXT_MEM_CACHE .bss :> EXT_MEM_CACHE .text :> EXT_MEM_CACHE .const :> EXT_MEM_CACHE .far :> EXT_MEM_CACHE .fardata :> EXT_MEM_CACHE .neardata :> EXT_MEM_CACHE .rodata :> EXT_MEM_CACHE .sysmem :> EXT_MEM_CACHE .switch :> EXT_MEM_CACHE .L2SramSect :> L2_SRAM .stack :> L2_SRAM }

Responses:
the fun is 12ms， i think it should be 1ms only，so i dont know whats wrong

Hi Shuai, Can you please give some more information as to how much data you are copying from L2 RAM to L2 RAM? I am guessing the l2input, L2output is part of .L2SramSect. Thanks and Regards, Piyali

Hi Piyali i made a project in ccs ,in the main i read a inputimage and call my funs, here is my cmd (i use just one C66 of TDA2X , L2 configure L2_SRAM no L2_CACHE ,i copy data to L2 by EDMA . L1 configure L1D_CACHE) -stack 0x3000 -heap 0x7000000 MEMORY { L1P_SRAM : origin = 0x00E00000, len = 0x8000 // L1D_SRAM : origin = 0x00F00000, len = 0x4000 /* 16 KB SRAM */ // L1D_CACHE : origin = 0x00F00000, len = 0x8000 /* 16 KB cache */ L1D_CACHE : origin = 0x00F04000, len = 0x4000 /* 16 KB cache */ L2_SRAM : origin = 0x00800000, len = 0x48000 /* SARAM in L2, = 256 + 32 - 128 = 160 KB*/ // L2_CACHE : origin = 0x00828000, len = 0x20000 /* Cache for L2, which is configured as 128 KB*/ DSP2_L2_SRAM : origin = 0x40800000, len = 0x48000 SL2_SRAM : origin = 0x5B000000, len = 0x40000 EXT_MEM_CACHE : origin = 0x80000000, len = 0x08000000 /* DSP Used cachable area */ } SECTIONS { vectors :> EXT_MEM_CACHE .cinit :> EXT_MEM_CACHE .cio :> EXT_MEM_CACHE .bss :> EXT_MEM_CACHE ////usually reserves space for uninitialized variables .text :> EXT_MEM_CACHE //////contains executable code .const :> EXT_MEM_CACHE .far :> EXT_MEM_CACHE .fardata :> EXT_MEM_CACHE /////usually contains initialized data .neardata :> EXT_MEM_CACHE ///////usually contains initialized data .rodata :> EXT_MEM_CACHE .sysmem :> EXT_MEM_CACHE .switch :> EXT_MEM_CACHE .L2SramSect :> L2_SRAM .stack :> L2_SRAM } #ifdef TI_DSP #define L2SRAM_SIZE (32*1024) #pragma DATA_SECTION (L2SRAM,".L2SramSect"); unsigned char L2SRAM[L2SRAM_SIZE]; #define L2TMP_SIZE (128*1024) #pragma DATA_SECTION (L2SRAM,".L2SramSect"); unsigned char L2TMP[L2TMP_SIZE]; #endif in the main i malloc DDR space unsigned char *DDR_Init = (unsigned char *)malloc(DDR_SIZE); load my net to DDR_Init , the net is big const float model_detection_global[] = { 0.0640991,-0.219671,0.00918405,0.164087,-0.00767695,0.142808,0.162083,0.145399,0.00453223,-1.25773, -0.0320548,0.161154,0.361267,1.714,0.197743,0.0784551,-0.0696055,-1.14334,-0.0957228,0.72948, -0.54699,0.508312,0.649834,-1.14706,0.405195,0.0574608,-0.567157,0.00642587,-0.0300398,-0.0680399, 0.0273895,-0.0263173,0.0327542,0.0387226,-0.00789206,-0.448488,-0.15092,0.0934956,1.32409,-0.244303, -1.82788,-0.4503,1.11225,0.178051,-0.0456231,0.0360474,-0.365231,-0.383318,0.00687606,0.00101387, -0.00782929,-0.00366878,0.00721993,-0.0535142,0.194668,0.524056,0.482211,-0.148617,0.136613,0.248017, -0.0469616,-0.86986,-0.494906,-0.0215785,0.0179301,-0.0402353,-0.157173,-0.284838,-0.247974,0.00699071, 0.048664,0.0686588,-0.403694,-0.0938685,0.0354045,0.42147,0.0852851,0.648701,0.03408,0.529695, 0.131282,0.284551,0.379638,0.142066,0.309521,0.347878,0.217491,0.219339,9.99999,0.465153, 0.334311,0.420712,0.74588,0.44612,1.1629,0.321116,0.423224,-0.422717,-0.0121956,0.0108994, -0.839461,0.0118817,-1.85014,0.187245,-0.0707816,0.474495,-0.945377,-0.809488,0.0574341,-1.04387, 0.114112,-0.110706,0.68189,-0.153875,-0.155772,-0.107221,-0.137917,-0.118754,-0.0194638,0.0568384, -0.0432765,0.0663145,0.09544 and so on } then call my funcs i Compile my project with CCS .out is 1.6MB here is the map TMS320C6x Linker PC v8.2.2 ****************************************************************************** >> Linked Fri Apr 13 16:57:12 2018 OUTPUT FILE NAME: <Drowsy_Detection_to_boader.out> ENTRY POINT SYMBOL: "_c_int00" address: 87156c60 MEMORY CONFIGURATION name origin length used unused attr fill ---------------------- -------- --------- -------- -------- ---- -------- L2_SRAM 00800000 00048000 0000b000 0003d000 RWIX L1P_SRAM 00e00000 00008000 00000000 00008000 RWIX L1D_CACHE 00f04000 00004000 00000000 00004000 RWIX DSP2_L2_SRAM 40800000 00048000 00000000 00048000 RWIX SL2_SRAM 5b000000 00040000 00000000 00040000 RWIX EXT_MEM_CACHE 80000000 08000000 071599d0 00ea6630 RWIX SEGMENT ALLOCATION MAP run origin load origin length init length attrs members ---------- ----------- ---------- ----------- ----- ------- 00800000 00800000 0000b000 00000000 rw- 00800000 00800000 00008000 00000000 rw- .L2SramSect 00808000 00808000 00003000 00000000 rw- .stack 80000000 80000000 07000000 00000000 rw- 80000000 80000000 07000000 00000000 rw- .sysmem 87000000 87000000 0012577c 0012577c r-- 87000000 87000000 0012577c 0012577c r-- .const 8712577c 8712577c 00000004 00000000 rw- 8712577c 8712577c 00000004 00000000 rw- .neardata 87125780 87125780 000202d0 00000000 rw- 87125780 87125780 000202d0 00000000 rw- .far 87145a60 87145a60 00011a00 00011a00 r-x 87145a60 87145a60 00011a00 00011a00 r-x .text 87157460 87157460 00001448 00000000 rw- 87157460 87157460 00001324 00000000 rw- .fardata 87158788 87158788 00000120 00000000 rw- .cio 871588a8 871588a8 00001140 00001140 r-- 871588a8 871588a8 00000084 00000084 r-- .switch 87158930 87158930 000010b8 000010b8 r-- .cinit SECTION ALLOCATION MAP output attributes/ section page origin length input sections -------- ---- ---------- ---------- ---------------- .L2SramSect * 0 00800000 00008000 UNINITIALIZED 00800000 00008000 main.obj (.L2SramSect) .stack 0 00808000 00003000 UNINITIALIZED 00808000 00000008 rts6600_elf.lib : boot.obj (.stack) 00808008 00002ff8 --HOLE-- .const 0 87000000 0012577c 87000000 00124854 main.obj (.const:model_detection_global) 87124854 00000004 dma_funcs.obj (.const) 87124858 000007d0 main.obj (.const:$P$T0$1) 87125028 00000408 main.obj (.const:proto_detection_global) 87125430 0000013c afq_caffe_cnn.obj (.const:.string) 8712556c 00000004 --HOLE-- [fill = 0] 87125570 00000101 rts6600_elf.lib : ctype.obj (.const:.string:_ctypes_) 87125671 00000003 --HOLE-- [fill = 0] 87125674 00000044 afq_caffe_cnn.obj (.const) 871256b8 00000034 main.obj (.const:.string) 871256ec 00000024 rts6600_elf.lib : _printfi.obj (.const:.string) 87125710 00000024 edma_utils_autoincrement_v2.obj (.const) 87125734 0000001c edma_utils_autoincrement.obj (.const) 87125750 0000001c main.obj (.const) 8712576c 0000000c edma_utils.obj (.const) 87125778 00000004 edma_utils.obj (.const:edmaBase) .neardata * 0 8712577c 00000004 UNINITIALIZED 8712577c 00000004 dma_resource_allocator.obj (.neardata) .far 0 87125780 000202d0 UNINITIALIZED 87125780 00020000 (.common:L2TMP) 87145780 00000140 (.common:__TI_tmpnams) 871458c0 000000c8 dma_resource_allocator.obj (.far:gDmaResourceState) 87145988 000000c0 (.common:gDmaUtilsStateStruct) 87145a48 00000008 (.common:parmbuf) .cio 0 87158788 00000120 UNINITIALIZED 87158788 00000120 rts6600_elf.lib : trgmsg.obj (.cio) .cinit 0 87158930 000010b8 87158930 00001068 (.cinit..fardata.load) [load image, compression = rle] 87159998 0000000d (.cinit..far.load) [load image, compression = rle] 871599a5 0000000b (.cinit..L2SramSect.load) [load image, compression = rle] 871599b0 00000009 (.cinit..neardata.load) [load image, compression = rle] 871599b9 00000003 --HOLE-- [fill = 0] 871599bc 00000008 (__TI_handler_table) 871599c4 00000004 --HOLE-- [fill = 0] 871599c8 00000020 (__TI_cinit_table) .text 0 87145a60 00011a00 87145a60 00006c00 afq_caffe_cnn.obj (.text) 8714c660 00001c40 rts6600_elf.lib : _printfi.obj (.text:__TI_printfi) 8714e2a0 000016c0 edma_utils_autoincrement.obj (.text) 8714f960 00000ac0 edma_utils_autoincrement_v2.obj (.text) 87150420 00000680 rts6600_elf.lib : _printfi.obj (.text:_pconv_a) 87150aa0 00000600 : divd.obj (.text:__c6xabi_divd) 871510a0 000005c0 : _printfi.obj (.text:_pconv_g) 87151660 00000460 : _printfi.obj (.text:_pconv_e) 87151ac0 00000440 : memory.obj (.text:free) 87151f00 000003e0 dma_resource_allocator.obj (.text) 871522e0 000003e0 rts6600_elf.lib : fputs.obj (.text:fputs) 871526c0 000002e0 dma_funcs.obj (.text) 871529a0 000002e0 rts6600_elf.lib : _printfi.obj (.text:fcvt) 87152c80 00000280 : divf.obj (.text:__c6xabi_divf) 87152f00 00000280 : _printfi.obj (.text:_pconv_f) 87153180 00000240 : imath64.obj (.text:__c6xabi_divull) 871533c0 00000240 : _printfi.obj (.text:_ltostr) 87153600 00000220 : fputc.obj (.text:fputc) 87153820 00000220 main.obj (.text) 87153a40 00000220 rts6600_elf.lib : open.obj (.text:open) 87153c60 00000220 : setvbuf.obj (.text:setvbuf) 87153e80 000001e0 : fopen.obj (.text:_openfile) 87154060 000001c0 : imath40.obj (.text:__c6xabi_divul) 87154220 000001a0 : hostrename.obj (.text:HOSTrename) 871543c0 000001a0 edma_utils_state.obj (.text) 87154560 00000180 rts6600_elf.lib : frexp.obj (.text:frexp) 871546e0 00000180 : memory.obj (.text:malloc) 87154860 00000160 : autoinit.obj (.text:_auto_init_elf) 871549c0 00000160 edma_utils_memcpy.obj (.text) 87154b20 00000160 rts6600_elf.lib : fopen.obj (.text:fopen) 87154c80 00000160 : sqrt.obj (.text:sqrt) 87154de0 00000140 : fclose.obj (.text:__TI_closefile) 87154f20 00000140 : getdevice.obj (.text:getdevice) 87155060 00000140 : ldexp.obj (.text:ldexp) 871551a0 00000120 : fseek.obj (.text:fseek) 871552c0 00000100 : copy_decompress_rle.obj (.text:__TI_decompress_rle_core) 871553c0 00000100 : _io_perm.obj (.text:__TI_wrt_ok) 871554c0 00000100 : atoi.obj (.text:atoi) 871555c0 00000100 : cpy_tbl.obj (.text:copy_in) 871556c0 00000100 : fprintf.obj (.text:fprintf) 871557c0 000000e0 : hostlseek.obj (.text:HOSTlseek) 871558a0 000000e0 : hostopen.obj (.text:HOSTopen) 87155980 000000e0 : hostwrite.obj (.text:HOSTwrite) 87155a60 000000e0 : divi.obj (.text:__divi) 87155b40 000000e0 : close.obj (.text:close) 87155c20 000000e0 : exit.obj (.text:exit) 87155d00 000000e0 : ltoa.obj (.text:ltoa) 87155de0 000000e0 : memset.obj (.text:memset) 87155ec0 000000e0 : printf.obj (.text:printf) 87155fa0 000000c0 : hostread.obj (.text:HOSTread) 87156060 000000c0 : hostunlink.obj (.text:HOSTunlink) 87156120 000000c0 : fflush.obj (.text:__TI_doflush) 871561e0 000000c0 : divu.obj (.text:__divu) 871562a0 000000c0 : remi.obj (.text:__remi) 87156360 000000c0 : _printfi.obj (.text:_ecpy) 87156420 000000c0 : memory.obj (.text:minsert) 871564e0 000000c0 : memory.obj (.text:mremove) 871565a0 000000c0 : tls.obj (.text:tls:init:__TI_tls_init) 87156660 000000a0 : hostclose.obj (.text:HOSTclose) 87156700 000000a0 : fopen.obj (.text:__TI_cleanup) 871567a0 000000a0 : fixfu.obj (.text:__c6xabi_fixfu) 87156840 000000a0 : fltulld.obj (.text:__c6xabi_fltulld) 871568e0 000000a0 : remu.obj (.text:__remu) 87156980 000000a0 : memory.obj (.text:_minit) 87156a20 000000a0 : memcpy64.obj (.text:memcpy) 87156ac0 000000a0 : modf.obj (.text:modf) 87156b60 00000080 : trgmsg.obj (.text:__TI_readmsg) 87156be0 00000080 : llshift.obj (.text:__c6xabi_llshl) 87156c60 00000080 : boot.obj (.text:_c_int00) 87156ce0 00000080 edma_utils.obj (.text) 87156d60 00000080 rts6600_elf.lib : fclose.obj (.text:fclose) 87156de0 00000080 : getdevice.obj (.text:finddevice) 87156e60 00000080 : unlink.obj (.text:unlink) 87156ee0 00000060 : frcmpyd_div.obj (.text:__TI_frcmpyd_div) 87156f40 00000060 : trgmsg.obj (.text:__TI_writemsg) 87156fa0 00000060 : llshift.obj (.text:__c6xabi_llshru) 87157000 00000060 : lseek.obj (.text:lseek) 87157060 00000060 : memccpy.obj (.text:memccpy) 871570c0 00000060 : write.obj (.text:write) 87157120 00000040 : isinf.obj (.text:__c6xabi_isinf) 87157160 00000040 : _printfi.obj (.text:__c6xabi_isnan) 871571a0 00000040 : call_stub.obj (.text:__call_stub) 871571e0 00000040 : strasg.obj (.text:__strasgi_64plus) 87157220 00000040 : args_main.obj (.text:_args_main) 87157260 00000040 : imath64.obj (.text:_subcull) 871572a0 00000020 : errno.obj (.text:__c6xabi_errno_addr) 871572c0 00000020 : push.obj (.text:__pop_rts) 871572e0 00000020 : push.obj (.text:__push_rts) 87157300 00000020 : _lock.obj (.text:_nop) 87157320 00000020 : fprintf.obj (.text:_outc) 87157340 00000020 : printf.obj (.text:_outc) 87157360 00000020 : fprintf.obj (.text:_outs) 87157380 00000020 : printf.obj (.text:_outs) 871573a0 00000020 : startup.obj (.text:_system_post_cinit) 871573c0 00000020 : pre_init.obj (.text:_system_pre_init) 871573e0 00000020 : exit.obj (.text:abort) 87157400 00000020 : copy_decompress_none.obj (.text:decompress:none:__TI_decompress_none) 87157420 00000020 : copy_decompress_rle.obj (.text:decompress:rle24:__TI_decompress_rle24) 87157440 00000020 edma3lld_support.obj (.text) .fardata 0 87157460 00001324 UNINITIALIZED 87157460 00001000 main.obj (.fardata:input_img) 87158460 000001e0 rts6600_elf.lib : defs.obj (.fardata:_ftable) 87158640 000000a0 : write.obj (.fardata:_stream) 871586e0 00000078 : write.obj (.fardata:_device) 87158758 0000000c : exit.obj (.fardata) 87158764 00000008 : memory.obj (.fardata) 8715876c 00000004 : _lock.obj (.fardata:_lock) 87158770 00000004 : _lock.obj (.fardata:_unlock) 87158774 00000004 : defs.obj (.fardata) 87158778 00000004 : errno.obj (.fardata) 8715877c 00000004 : fopen.obj (.fardata) 87158780 00000004 : open.obj (.fardata) .sysmem 0 80000000 07000000 UNINITIALIZED 80000000 00000008 rts6600_elf.lib : memory.obj (.sysmem) 80000008 06fffff8 --HOLE-- .switch 0 871588a8 00000084 871588a8 0000004c afq_caffe_cnn.obj (.switch:afq_caffecnn_load) 871588f4 00000038 rts6600_elf.lib : _printfi.obj (.switch:__TI_printfi) MODULE SUMMARY Module code ro data rw data ------ ---- ------- ------- .\ main.obj 544 1201276 167936 afq_caffe_cnn.obj 27648 460 0 edma_utils_autoincrement.obj 5824 28 0 edma_utils_autoincrement_v2.obj 2752 36 0 dma_resource_allocator.obj 992 0 204 dma_funcs.obj 736 4 0 edma_utils_state.obj 416 0 192 edma_utils_memcpy.obj 352 0 0 edma_utils.obj 128 16 0 edma3lld_support.obj 32 0 0 +--+---------------------------------+-------+---------+-----------+ Total: 39424 1201820 168332 D:\ti_ccsv7\ccsv7\tools\compiler\ti-cgt-c6000_8.2.2\lib\rts6600_elf.lib _printfi.obj 13696 92 0 memory.obj 2016 0 8 divd.obj 1536 0 0 fopen.obj 992 0 4 fputs.obj 992 0 0 defs.obj 0 0 804 divf.obj 640 0 0 imath64.obj 640 0 0 open.obj 544 0 4 fputc.obj 544 0 0 setvbuf.obj 544 0 0 trgmsg.obj 224 0 288 fclose.obj 448 0 0 getdevice.obj 448 0 0 imath40.obj 448 0 0 hostrename.obj 416 0 0 frexp.obj 384 0 0 write.obj 96 0 280 autoinit.obj 352 0 0 sqrt.obj 352 0 0 fprintf.obj 320 0 0 ldexp.obj 320 0 0 copy_decompress_rle.obj 288 0 0 fseek.obj 288 0 0 printf.obj 288 0 0 exit.obj 256 0 12 ctype.obj 0 257 0 _io_perm.obj 256 0 0 atoi.obj 256 0 0 cpy_tbl.obj 256 0 0 hostopen.obj 224 0 8 close.obj 224 0 0 divi.obj 224 0 0 hostlseek.obj 224 0 0 hostwrite.obj 224 0 0 llshift.obj 224 0 0 ltoa.obj 224 0 0 memset.obj 224 0 0 divu.obj 192 0 0 fflush.obj 192 0 0 hostread.obj 192 0 0 hostunlink.obj 192 0 0 remi.obj 192 0 0 tls.obj 192 0 0 fixfu.obj 160 0 0 fltulld.obj 160 0 0 hostclose.obj 160 0 0 memcpy64.obj 160 0 0 modf.obj 160 0 0 remu.obj 160 0 0 boot.obj 128 0 0 unlink.obj 128 0 0 frcmpyd_div.obj 96 0 0 lseek.obj 96 0 0 memccpy.obj 96 0 0 args_main.obj 64 0 0 call_stub.obj 64 0 0 isinf.obj 64 0 0 push.obj 64 0 0 strasg.obj 64 0 0 _lock.obj 32 0 8 errno.obj 32 0 4 copy_decompress_none.obj 32 0 0 pre_init.obj 32 0 0 startup.obj 32 0 0 +--+---------------------------------+-------+---------+-----------+ Total: 32768 349 1420 Heap: 0 0 117440512 Stack: 0 0 12288 Linker Generated: 0 4273 0 +--+---------------------------------+-------+---------+-----------+ Grand Total: 72192 1206442 117622552 LINKER GENERATED COPY TABLES __TI_cinit_table @ 871599c8 records: 4, size/record: 8, table size: 32 .fardata: load addr=87158930, load size=00001068 bytes, run addr=87157460, run size=00001324 bytes, compression=rle .far: load addr=87159998, load size=0000000d bytes, run addr=87125780, run size=000202d0 bytes, compression=rle .L2SramSect: load addr=871599a5, load size=0000000b bytes, run addr=00800000, run size=00008000 bytes, compression=rle .neardata: load addr=871599b0, load size=00000009 bytes, run addr=8712577c, run size=00000004 bytes, compression=rle LINKER GENERATED HANDLER TABLE __TI_handler_table @ 871599bc records: 2, size/record: 4, table size: 8 index: 0, handler: __TI_decompress_rle24 index: 1, handler: __TI_decompress_none i load .out to the evm whit JTAG and run follow is out time [C66xx_DSP1] Meanvalue_ti_times = 0.262565 ms im2col_ti_cov_times = 0.972233 ms mcvMatrixMulMatrixRowMajor_f32_afq_ti_cov_times = 12.518572 ms caffecnn_forward_layer_conv_ti_times = 13.755874 ms im2col_ti_cov_times = 2.175635 ms mcvMatrixMulMatrixRowMajor_f32_afq_ti_cov_times = 25.113652 ms caffecnn_forward_layer_conv_ti_times = 27.531509 ms im2col_ti_cov_times = 2.222183 ms ,,but i think the out is right but time is very big , I don't think it's normal. before i test some funcs alone in a simple project the time is samll （input image size is same） Meanvalue_ti_times +im2col_ti_cov_times +mcvMatrixMulMatrixRowMajor_f32_afq_ti_cov_times is 0.5ms so I what should I check?

Hi Shuai, Thanks for the details! Couple of things to check: 1. You seem to be profiling the code as well as the DMA operation. Do you have a split up of the Compute cycles of the operation and the DMA. This will help us identify whether the issue is with the DMA and the compute. 2. The other thing to check is the DSP Compiler options between the simple project and the other project. Do you have RTOS/tasks in the final project? 3. Currently what is your EDMA parameters? Are you comparing the DMA cycles between DDR to DDR vs L2RAM to L2RAM? Thanks and Regards, Piyali

Hi Piyali some mistake about l2RAM ,some data i want to put in L2RAM but infact it in DDR . so itis not problem of DMA. thanks shuai