Consider the following example stats.json output, gathered by executing the 3SAT solver sample binary on the Pompia sample pipeline:
{
"message_size": {
"19": 1,
"18": 1,
"150": 1,
...
"972": 1,
"890": 1,
},
"cycle": 129550,
"0": {
"fetch": {
"l1ic_misses": 12035,
"l1ic_accesses": 29547
},
"l2": {
"l2_misses": 1229,
"l2_accesses": 18064
},
"decode": {
"decoded.insn": {
"ADD": 4834,
"JAL": 2456,
"LW": 2067,
...
"MUL": 6,
"SLLW": 2,
"SRAI": 1
},
"issued.insn": {
"ADD": 3546,
"JAL": 2022,
"LW": 1890,
...
"MUL": 4,
"SLLW": 2,
"SRAI": 1
}
},
"regfile": {
"get.register": {
"0": 5587,
"25": 230,
"8": 5074,
...
"27": 98,
"28": 28,
"30": 14
},
"set.register": {
"11": 688,
"1": 914,
"15": 6394,
...
"16": 713,
"30": 14,
"17": 77
}
},
"alu": {
"category": {
"do_rtype": 4486,
"do_jal": 2022,
"do_load": 9965,
"do_branch": 3366,
"do_itype": 6929,
"do_store": 4326,
"do_jalr": 971,
"do_nop": 336,
"do_lui": 700,
"do_shift": 1230,
"do_ecall": 6
}
},
"commit": {
"retires": 27828,
"flushes": 6507
},
"lsu": {
"l1dc_misses": 1945,
"l1dc_accesses": 11772
}
},
"-1": {
"mainmem": {
"peek.size": {
"16": 1229,
"153": 1
},
"poke.size": {
"1": 142,
"4": 400,
"8": 3535,
"2": 8
}
}
}
}
The JSON struct has several top-level keys including: message_size, cycle,
0, and -1. Themessage_size entry contains a histogram of the sizes of
the messages passed between the various components, and cycle contains the
count of simulated cycles when the simulator exited. 0 contains several
sub-keys corresponding to the pipeline components on simulated core 0:
fetch, l2, mainmem, decode, regfile, alu, commit, lsu. They
offer statistics about the number of cache misses, cache accesses, number of
times each register was read/written, number of instructions retired/flushed,
etc.
Finally, -1 contains sub-keys corresponding to components that are
accessible by all simulated cores, in this case, only mainmem.