2_testing_pytorch_output_softmax.py

import sys
import glob
import math
import re, random
from Ntk_Struct_PO_cmu import *
from Ntk_Parser_PO_cmu import *
from fflatch_only_graph_PO import *
import numpy as np
import networkx as nx
import collections
import h5py
from sklearn.model_selection import train_test_split
from numpy.random import seed
import h5py
from sklearn.preprocessing import MinMaxScaler
from numpy.random import seed
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from torchvision import datasets, transforms
from torch.optim.lr_scheduler import StepLR
from torch.utils.data.sampler import SubsetRandomSampler


class MLP(nn.Module):
    def __init__(self, input_dim, n_class):
        super(MLP, self).__init__()
        self.n_class=n_class

        self.fc1=nn.Linear(input_dim, 100)

        self.fc3 = nn.Linear(100, n_class)

    def forward(self, x):
        x=self.fc1(x)
        x = F.relu(x)
        x = self.fc3(x)

        return x

def customgraph2networkx(netlist_graph):

    G = nx.DiGraph()
    for node in netlist_graph.object_list:
        if node.name not in G.nodes:
            G.add_node(node.name, type=netlist_graph.gateType_reverse[node.gate_type])
        else: # already added by edge
            G.nodes[node.name]['type'] = netlist_graph.gateType_reverse[node.gate_type]

        for innode in node.fan_in_node:
            G.add_edge(innode.name,node.name)

        for outnode in node.fan_out_node:
            G.add_edge(node.name,outnode.name)

    return G

def feature_normalization(cur_data_X):


    # fit scaler on training data
    norm = MinMaxScaler().fit(cur_data_X)

    # transform training data
    cur_data_X = norm.transform(cur_data_X)

    return cur_data_X

def add_q2latchname(benchname, filepath, q2latchname):
    fi=open(filepath+'/'+benchname+'_latchname2Q_remove_LD','r')
    for line in fi:

        q, latchname=line.rstrip().split(':')
        #print (q, latchname)
        q2latchname[q]=latchname
    fi.close()

def add_delay(node, benchname, report_dir, q2latchname, seq_sig):
    latchname=q2latchname[node][:-7]
    PIflag=False # this latch connects to input

    f1=open(report_dir+latchname+'.from','r')
    f2 = open(report_dir + latchname + '.to', 'r')
    for line in f1:


        if 'Required Time' in line:
            #print (line.rstrip().split(' '))
            fromreqtime=int(line.rstrip().split(' ')[-1])

        if 'Launch Clock' in line:
            #print (line.rstrip().split(' '))
            fromlclk=int(line.rstrip().split(' ')[-1])

        if 'Data Path' in line:
            #print (line.rstrip().split(' '))
            fromdelay=int(line.rstrip().split(' ')[-1])


    for line in f2:
        if 'Startpoint' in line:
            if seq_sig not in line:
                #print (node, line)
                PIflag=True

        if 'Required Time' in line:
            #print (line.rstrip().split(' '))
            toreqtime=int(line.rstrip().split(' ')[-1])

        if 'Launch Clock' in line:
            #print (line.rstrip().split(' '))
            tolclk=int(line.rstrip().split(' ')[-1])

        if 'Data Path' in line:
            todelay=int(line.rstrip().split(' ')[-1])

    f1.close()
    f2.close()
    if PIflag:
        todelay=0

    return fromdelay, fromreqtime-fromlclk, todelay, toreqtime-tolclk


def construct_dataset(curfile, benchpath, data_X, data_Y):
    num_train=0
    for idx, file in enumerate(glob.glob(benchpath + '/*')):
        if file!=curfile:
            num_train+=1

            with h5py.File(file, 'r') as hf:
                X = hf['X_train'][:]
                Y = hf['Y_train'][:]

            data_X = np.vstack((data_X, X))
            data_Y = np.vstack((data_Y, Y))


    return data_X, data_Y

def get_accuracy(preds, Ys):
    max_preds = preds.argmax(dim=1, keepdim=True)
    numcorrect=max_preds.squeeze(1).eq(Ys)
    return numcorrect.sum()/torch.FloatTensor([Ys.shape[0]])

def train(model, train_loader, optimizer, criterion):
    model.train()
    epoch_loss = 0
    epoch_acc = 0

    for Xs, Ys in train_loader:
        #print (Xs.shape, Ys.shape)
        optimizer.zero_grad()
        preds=model(Xs.float())
        preds = preds.view(-1, preds.shape[-1])
        Ys=Ys.view(-1)
        loss = criterion(preds, Ys)
        acc=get_accuracy(preds, Ys)
        loss.backward()
        optimizer.step()
        epoch_loss += loss.item()
        epoch_acc += acc.item()

    return epoch_loss / len(train_loader), epoch_acc / len(train_loader)

def evaluate(model, validation_loader, criterion, all_results, benchname, num_classes):
    global misclassification
    epoch_loss = 0
    epoch_acc = 0

    allsoftmax_probs=[]
    probs = nn.Softmax(dim=1)

    confusion_matrix = [[0] * num_classes for i in range(num_classes)]

    #dicres = {"LATCH_L0": 0, "LATCH_L1": 0, "LATCH_DD": 1}
    model.eval()
    with torch.no_grad():
        for Xs, Ys in validation_loader:
            preds = model(Xs.float())
            preds = preds.view(-1, preds.shape[-1])
            predsoftmax = probs(preds)
            allsoftmax_probs.append(np.array(predsoftmax[0]))

            Ys = Ys.view(-1)

            preds_arg = np.argmax(preds, axis=1)
            confusion_matrix[Ys[0]][preds_arg[0]] += 1

            loss = criterion(preds, Ys)
            acc = get_accuracy(preds, Ys)
            epoch_loss += loss.item()
            epoch_acc += acc.item()
    print ("correct count:", epoch_acc, len(validation_loader))
    np_cm = np.array(confusion_matrix)
    print ('confusion matrix.')
    print (np_cm)
    all_results[benchname]=all_results.get(benchname, [])+[epoch_acc, len(validation_loader)]
    misclassification+=len(validation_loader)-epoch_acc
    return epoch_loss / len(validation_loader), epoch_acc / len(validation_loader), np.array(allsoftmax_probs)


def add_dataset(benchname, seq_sig, q2latchname,report_dir, DB, DF, G_fflatch_only, nxG, data_X, comb_data_X, data_Y, test_nodes):
    for node in G_fflatch_only:
        if not (G_fflatch_only.nodes[node]['type'] == 'LATCH_L0' or G_fflatch_only.nodes[node]['type'] == 'LATCH_L1' or G_fflatch_only.nodes[node]['type'] == 'LATCH_LD' or G_fflatch_only.nodes[node]['type'] == 'LATCH_DD'):
           continue

        tot_num_FI = len(G_fflatch_only.in_edges(node))
        tot_num_FO = len(G_fflatch_only.out_edges(node))

        visited = set() # add current node
        visited.add(node)
        # search backward
        bdepthidx = 0
        backl = []
        backq = collections.deque([node])

        while bdepthidx < DB:
            unseen_fanin_latch = 0
            seen_fanin_latch = 0
            unseen_fanin_ff=0
            seen_fanin_ff=0
            unseen_PIs = 0
            seen_PIs=0
            #cur_visited = set()
            for _ in range(len(backq)):
                cur=backq.popleft()

                #tot_fanin+=len(G_fflatch_only.in_edges(cur))


                for inedge in G_fflatch_only.in_edges(cur):
                    innode=inedge[0]
                    #print (innode, visited)


                    if innode not in visited:
                        #if innode not in cur_visited:
                        if G_fflatch_only.nodes[innode]['type'] == 'IPT':
                            unseen_PIs += 1
                        elif G_fflatch_only.nodes[innode]['type'] == 'DFF':  # FF
                            unseen_fanin_ff+=1
                        elif G_fflatch_only.nodes[innode]['type'] == 'LATCH_L0':
                            #unseen_fanin_latch0+=1
                            unseen_fanin_latch+=1
                        elif G_fflatch_only.nodes[innode]['type'] == 'LATCH_L1':
                            #unseen_fanin_latch1+=1
                            unseen_fanin_latch += 1
                        elif G_fflatch_only.nodes[innode]['type'] == 'LATCH_LD':
                            #unseen_fanin_latchld+=1
                            unseen_fanin_latch += 1
                        elif G_fflatch_only.nodes[innode]['type'] == 'LATCH_DD':
                            #unseen_fanin_latchdd+=1
                            unseen_fanin_latch += 1
                        backq.append(innode)

                        #cur_visited.add(innode)
                        visited.add(innode)
                    else: # seen this node before
                        #print ("seen", bdepthidx)
                        if G_fflatch_only.nodes[innode]['type'] == 'IPT':
                            seen_PIs += 1
                        elif G_fflatch_only.nodes[innode]['type'] == 'DFF':  # FF
                            seen_fanin_ff+=1
                        elif G_fflatch_only.nodes[innode]['type'] == 'LATCH_L0':
                            #seen_fanin_latch0+=1
                            seen_fanin_latch+=1
                        elif G_fflatch_only.nodes[innode]['type'] == 'LATCH_L1':
                            #seen_fanin_latch1+=1
                            seen_fanin_latch += 1
                        elif G_fflatch_only.nodes[innode]['type'] == 'LATCH_LD':
                            #seen_fanin_latchld+=1
                            seen_fanin_latch += 1
                        elif G_fflatch_only.nodes[innode]['type'] == 'LATCH_DD':
                            #seen_fanin_latchdd+=1
                            seen_fanin_latch += 1
                        backq.append(innode)


            backl.append(unseen_PIs)
            backl.append(unseen_fanin_ff)
            backl.append(unseen_fanin_latch)

            bdepthidx+=1

        # search forward
        visited = set()  # add current node
        visited.add(node)
        fdepthidx = 0
        forwl = []
        forwq = collections.deque([node])
        while fdepthidx<DF:
            unseen_fanout_latch = 0
            seen_fanout_latch = 0
            unseen_fanout_ff=0
            seen_fanout_ff=0
            unseen_POs = 0
            seen_POs=0
            for _ in range(len(forwq)):
                cur=forwq.popleft()


                for outedge in G_fflatch_only.out_edges(cur):
                    outnode=outedge[1]
                    if outnode not in visited:
                        #if '_PO' in outnode:
                        if G_fflatch_only.nodes[outnode]['type'] == 'PO':
                            unseen_POs += 1
                        elif G_fflatch_only.nodes[outnode]['type'] == 'DFF':  # FF
                            unseen_fanout_ff+=1
                        elif G_fflatch_only.nodes[outnode]['type'] == 'LATCH_L0':
                            #unseen_fanout_latch0+=1
                            unseen_fanout_latch+=1
                        elif G_fflatch_only.nodes[outnode]['type'] == 'LATCH_L1':
                            #unseen_fanout_latch1+=1
                            unseen_fanout_latch += 1
                        elif G_fflatch_only.nodes[outnode]['type'] == 'LATCH_LD':
                            #unseen_fanout_latchld+=1
                            unseen_fanout_latch += 1
                        elif G_fflatch_only.nodes[outnode]['type'] == 'LATCH_DD':
                            #unseen_fanout_latchdd+=1
                            unseen_fanout_latch += 1

                        forwq.append(outnode)
                            #if not count_visited_FF:
                        visited.add(outnode)
                    else:
                        if G_fflatch_only.nodes[outnode]['type'] == 'PO':
                            seen_POs += 1
                        elif G_fflatch_only.nodes[outnode]['type'] == 'DFF':  # FF
                            seen_fanout_ff+=1
                        elif G_fflatch_only.nodes[outnode]['type'] == 'LATCH_L0':
                            #seen_fanout_latch0+=1
                            seen_fanout_latch+=1
                        elif G_fflatch_only.nodes[outnode]['type'] == 'LATCH_L1':
                            #seen_fanout_latch1+=1
                            seen_fanout_latch += 1
                        elif G_fflatch_only.nodes[outnode]['type'] == 'LATCH_LD':
                            #seen_fanout_latchld+=1
                            seen_fanout_latch += 1
                        elif G_fflatch_only.nodes[outnode]['type'] == 'LATCH_DD':
                            #seen_fanout_latchdd+=1
                            seen_fanout_latch += 1
                        forwq.append(outnode)


            forwl.append(unseen_fanout_ff)
            forwl.append(unseen_fanout_latch)
            fdepthidx+=1
        curFO=set()
        childFO=set()



        for outedge in G_fflatch_only.out_edges(node):
            outnode=outedge[1]
            curFO.add(outnode)

            for choutedge in G_fflatch_only.out_edges(outnode):
                choutnode=choutedge[1]
                childFO.add(choutnode)


        # compute the fraction
        fraction=0
        Sfraction=0
        #Tfraction=0

        curFI = set()
        parentFI = set()

        for inedge in G_fflatch_only.in_edges(node):
            innode = inedge[0]

            curFI.add(innode)

            parentFO=set([outedge[1] for outedge in G_fflatch_only.out_edges(innode)])

            if len(curFO & parentFO)>0:
                fraction+=1

            if len(childFO & parentFO)>0:
                Sfraction+=1



            for pinedge in G_fflatch_only.in_edges(innode):
                pinnode=pinedge[0]
                parentFI.add(pinnode)


        Sfraction2=0 # detect 2nd DD in trapezoid shape
        #Tfraction3=0
        for outnode in curFO:
            childFI = set([inedge[0] for inedge in G_fflatch_only.in_edges(outnode)])

            if len(childFI & parentFI)>0:
                Sfraction2+=1




        latchloop=1 if len(childFO & curFI)>0 else 0

        backl.reverse()

        if tot_num_FI==0:
            #print (fraction, Sfraction)
            tot_num_FI=1
        vector = backl + forwl + [fraction / tot_num_FI, Sfraction/tot_num_FI, Sfraction2/tot_num_FO,latchloop]
        #print (node, vector)

        # single path feature
        if len(G_fflatch_only.out_edges(node))==1 or len(G_fflatch_only.in_edges(node))==1:
            vector.append(1)
        else:
            vector.append(0)

        #print (node)
        if '_PO' in node:
            nodename=node[:-3]
            #print (nodename)
        else:
            nodename=node
        fromdelay, fromdenominator, todelay, todenominator=add_delay(nodename, benchname, report_dir, q2latchname, seq_sig)
        vector.append(fromdelay)
        vector.append(todelay)

        # 13 self-loop feature
        all_fanout_self_loop=[]
        for outedge in G_fflatch_only.out_edges(node):
            outnode=outedge[1]
            child_selfloop=None
            child_num_fanin=len(G_fflatch_only.in_edges(outnode))
            for choutedge in G_fflatch_only.out_edges(outnode):
                choutnode = choutedge[1]
                if choutnode==outnode:
                    child_selfloop=True

            if child_selfloop:
                #print (G_fflatch_only.nodes[node]['type'], outnode, G_fflatch_only.out_edges(outnode))
                all_fanout_self_loop.append(round(1/child_num_fanin, 2))

        if all_fanout_self_loop:
            vector.append(max(all_fanout_self_loop))
            #print (G_fflatch_only.nodes[node]['type'], max(all_fanout_self_loop))
        else: # none fanout has self-loop
            vector.append(0)

        #print (vector)
        data_X.append(vector)
        test_nodes.append(node)

        # add labels according to the latch's type
        if G_fflatch_only.nodes[node]['type'] == 'LATCH_L0':
            data_Y.append(np.array([1, 0]))
        elif G_fflatch_only.nodes[node]['type'] == 'LATCH_L1':
            data_Y.append(np.array([1, 0]))
        elif G_fflatch_only.nodes[node]['type'] == 'LATCH_DD':
            data_Y.append(np.array([0, 1]))
        elif G_fflatch_only.nodes[node]['type'] == 'LATCH_LD':
            data_Y.append(np.array([0, 1]))

def generate_dataset(specific_bench,benchpath,DB, DF, seq_sig):
    all_results={}
    num_classes = 2
    global misclassification
    misclassification = 0


    for idx, filepath in enumerate(glob.glob(benchpath+'/*')):
        seed(12)
        torch.manual_seed(12)
        torch.cuda.manual_seed(12)
        torch.cuda.manual_seed_all(12)
        np.random.seed(12)
        random.seed(12)

        # torch.cuda.is_available() checks and returns a Boolean True if a GPU is available, else it'll return False
        is_cuda = torch.cuda.is_available()

        # If we have a GPU available, we'll set our device to GPU. We'll use this device variable later in our code.
        if is_cuda:
            device = torch.device("cuda")
        else:
            device = torch.device("cpu")
        print (is_cuda)

        #print (file)
        x = re.findall(r"^\.\/dataset_seed1\\([A-Za-z0-9\_]+)$", filepath)
        #print (x)
        benchname=x[0].split('_')[0]
        input_dim=DF*5+5+2+1
        n_class=2

        comb_data_X = []
        data_X = []
        data_Y = []

        model = MLP(input_dim, n_class)
        criterion = nn.CrossEntropyLoss()
        model.load_state_dict(torch.load(f"./best_models/8datasets_cweights_simple_best_MLP_model_pytorch_{n_class}.pt"))

        if specific_bench in filepath and '_2_' not in filepath and '_3_' not  in filepath:
            print (filepath)
            # testing

            q2latchname={}

            add_q2latchname(benchname, filepath, q2latchname)

            file=filepath+f'\{benchname}_clean_remove_LD.bench'
            report_dir=filepath+f'/{benchname}_time_reports_remove_LD/'
            print (file)

            netlist_graph = ntk_parser(file)

            nxG = customgraph2networkx(netlist_graph)

            G_fflatch_only = fflatch_graph(nxG)

            # remove clk and reset node
            all_nodes = list(G_fflatch_only.nodes())

            for node in all_nodes:
                if node == 'rst':
                    G_fflatch_only.remove_node(node)
                if node == 'reset':
                    G_fflatch_only.remove_node(node)
                if node == 'clk':
                    G_fflatch_only.remove_node(node)

            test_nodes = []
            add_dataset(benchname, seq_sig, q2latchname, report_dir, DB, DF, G_fflatch_only, nxG, data_X, comb_data_X, data_Y, test_nodes)

            data_X = np.array(data_X)
            data_Y = np.array(data_Y)

            # normalization
            data_X = feature_normalization(data_X)


            Y_test = np.argmax(data_Y, axis=1)

            test_batch_size=1
            test_data = torch.utils.data.TensorDataset(torch.from_numpy(data_X), torch.from_numpy(Y_test))
            test_loader = torch.utils.data.DataLoader(test_data, shuffle=False, batch_size=test_batch_size)
            test_loss, test_acc, allsoftmax_probs = evaluate(model, test_loader, criterion, all_results, x[0], n_class)

            #results = model.evaluate(X_test, Y_test, batch_size=32)
            print(f'test Loss:{test_loss :.4f} | test Acc: {test_acc :.4f}%')

            all_results[x[0]]=all_results.get(x[0], [])+[test_acc]

            fbo=open(f"./2_all_softmaxprobs/{benchname}_softmaxprobs_remove_LD", "w")
            for idx, node in enumerate(test_nodes):
                fbo.write(f'{node}: {allsoftmax_probs[idx][0]:.4f} {allsoftmax_probs[idx][1]:.4f} {G_fflatch_only.nodes[node]["type"]}\n')
            fbo.close()


    print (all_results)
    #return data_X, data_Y
    f = open(f"8datasets_cweights_simple_best_MLP_model_pytorch_{n_class}_softmax.txt", "w")
    for key, val in all_results.items():
        f.write(f'{key}:{val}\n')

    f.write(str(misclassification))

    f.close()





seed(1)
benchpath="./dataset_seed1"
all_bench='s'
seq_sig='reg' # signature of sequential logic, only re-elaborate can generate this signature

DB = 1  # depth for backward (towards inputs)
DF = 1  # depth for forward (towards outputs)


generate_dataset(all_bench, benchpath, DB, DF, seq_sig)