From 0f27e06e0c374b2d87a929879750271a2237117f Mon Sep 17 00:00:00 2001
From: Joran Honig <joran.honig@gmail.com>
Date: Fri, 29 Mar 2019 13:46:22 +0100
Subject: [PATCH] implement basic version of the instruction profiler plugin

this is entirely based on the earlier implementation in the iprof
module made by @hzzhang
---
 .../instruction_profiler_plugin.py            | 89 ++++++++++++++++++-
 1 file changed, 88 insertions(+), 1 deletion(-)

diff --git a/mythril/laser/ethereum/plugins/implementations/instruction_profiler_plugin.py b/mythril/laser/ethereum/plugins/implementations/instruction_profiler_plugin.py
index 77c9a016..22790f01 100644
--- a/mythril/laser/ethereum/plugins/implementations/instruction_profiler_plugin.py
+++ b/mythril/laser/ethereum/plugins/implementations/instruction_profiler_plugin.py
@@ -2,6 +2,30 @@ from mythril.laser.ethereum.svm import LaserEVM
 from mythril.laser.ethereum.plugins.plugin import LaserPlugin
 from mythril.laser.ethereum.state.global_state import GlobalState
 
+from collections import namedtuple
+from datetime import datetime
+from typing import Dict, List, Tuple
+
+# Type annotations:
+#   start_time: datetime
+#   end_time: datetime
+_InstrExecRecord = namedtuple("_InstrExecRecord", ["start_time", "end_time"])
+
+# Type annotations:
+#   total_time: float
+#   total_nr: float
+#   min_time: float
+#   max_time: float
+_InstrExecStatistic = namedtuple(
+    "_InstrExecStatistic", ["total_time", "total_nr", "min_time", "max_time"]
+)
+
+# Map the instruction opcode to its records if all execution times
+_InstrExecRecords = Dict[str, List[_InstrExecRecord]]
+
+# Map the instruction opcode to the statistic of its execution times
+_InstrExecStatistics = Dict[str, _InstrExecStatistic]
+
 
 class InstructionProfilerPlugin(LaserPlugin):
     """ Instruction profiler plugin
@@ -10,10 +34,73 @@ class InstructionProfilerPlugin(LaserPlugin):
     It does so by measuring the frequency and duration of each operation execution
     """
 
+    def __init__(self):
+        self.records = dict()
+
+    def record(self, op: int, start_time: datetime, end_time: datetime):
+        try:
+            self.records[op].append(_InstrExecRecord(start_time, end_time))
+        except KeyError:
+            self.records[op] = [_InstrExecRecord(start_time, end_time)]
+
+    def _make_stats(self) -> Tuple[float, _InstrExecStatistics]:
+        periods = {
+            op: list(
+                map(lambda r: r.end_time.timestamp() - r.start_time.timestamp(), rs)
+            )
+            for op, rs in self.records.items()
+        }
+
+        stats = dict()
+        total_time = 0
+
+        for _, (op, times) in enumerate(periods.items()):
+            stat = _InstrExecStatistic(
+                total_time=sum(times),
+                total_nr=len(times),
+                min_time=min(times),
+                max_time=max(times),
+            )
+            total_time += stat.total_time
+            stats[op] = stat
+
+        return total_time, stats
+
+    def __str__(self):
+        total, stats = self._make_stats()
+
+        s = "Total: {} s\n".format(total)
+
+        for op in sorted(stats):
+            stat = stats[op]
+            s += "[{:12s}] {:>8.4f} %,  nr {:>6},  total {:>8.4f} s,  avg {:>8.4f} s,  min {:>8.4f} s,  max {:>8.4f} s\n".format(
+                op,
+                stat.total_time * 100 / total,
+                stat.total_nr,
+                stat.total_time,
+                stat.total_time / stat.total_nr,
+                stat.min_time,
+                stat.max_time,
+            )
+
+        return s
+
     def initialize(self, symbolic_vm: LaserEVM):
         """Initializes the instruction profiler
 
         Introduces hooks that measure the duration of each execution and the name of the instruction being executed
         :param symbolic_vm: The virtual machine to initialize the plugin for
         """
-        pass
+
+        @symbolic_vm.register_instruction_evaluate_wrapper
+        def introduce_wrap(evaluate):
+            def evaluate(*args, **kwargs):
+                start_time = datetime.now()
+                try:
+                    evaluate(*args, **kwargs)
+                finally:
+                    end_time = datetime.now()
+                    self.record(op=args[0], start_time=start_time, end_time=end_time)
+            return evaluate
+
+