Hexagon Arch : v73
TVM : Built from source
OS : QNX 7.1
Metaschedule : Performed via RPC
I’m trying to tune a resnet50 int8 onnx model for hexagon v73 dsp using metascheduling. I’m performing this tuning using vrmpy based intrinsics. Metaschedule is able to extract all the tasks, generate the .so file and dump in onto remote hexagon device and get inference to calculate cost. After some trials, Metascheduling is unable to load the compiled .so file and calculate the cost. More specifically, after tuning for some trials I’m getting following error from hexagon (logs are gathered from logcat) :
Jan 12 15:02:36.309 frpc_lib.4747299 frpc_lib 5 16135 SLOG2_ERROR htp_runner[apps_std_imp.c:232]: apps_std fopen_INFO : /mnt/etc/images/tvm_tmp_mod.so
Jan 12 15:02:36.437 frpc_lib.4747299 frpc_lib 5 16135 SLOG2_ERROR htp_runner[fastrpc_lib.c:874]: :error: -2147482611: 0 == (nErr = ctx->nRetVal)
Jan 12 15:02:36.437 frpc_lib.4747299 frpc_lib 5 16135 SLOG2_ERROR htp_runner[fastrpc_lib.c:900]: Returning nErr:-2147482611-0x8000040d for domain=3, client=0 sc=4020200
Jan 12 15:02:36.437 frpc_lib.4747299 frpc_lib 5 16135 SLOG2_ERROR htp_runner[listener_qnx.c:280]: listener protocol failure -2147482611
Jan 12 15:02:36.447 frpc_lib.4747299 frpc_lib 2 16135 SLOG2_ERROR htp_runner[fastrpc_lib.c:874]: :error: -2147482611: 0 == (nErr = ctx->nRetVal)
Jan 12 15:02:36.447 frpc_lib.4747299 frpc_lib 2 16135 SLOG2_ERROR htp_runner[fastrpc_lib.c:900]: Returning nErr:-2147482611-0x8000040d for domain=3, client=0 sc=2020000
Jan 12 15:02:36.447 frpc_lib.4747299 frpc_lib 5 16135 SLOG2_ERROR htp_runner[fastrpc_lib.c:874]: :error: -2147482611: 0 == (nErr = ctx->nRetVal)
What is the source of this error and how to resolve this error? @sanirudh @srkreddy1238 @Hzfengsy @tqchen