import pandas as pd
import numpy as np

np.random.seed(1)

# Setting up dummy data
df = pd.DataFrame(np.random.rand(9, 7), columns = [i for i in range(5)] + ['H1', 'H2'])
df['H1'] = ['M1'] * 5 + ['M2'] * 4
df['H2'] = ['All', 'c1', 'c2', 'c3', 'c4', 'All', 'c1', 'c2', 'c3']

# Mark columns to be aggregated.
numeric_cols = [0,1,2,3,4]

# Exclude columns with 'All' indicator and compute the summary.
summary = df.loc[df.H2 != 'All'].groupby('H1')[numeric_cols].sum()

# Assign the summary values; relies on H1 being sorted.
df.loc[df.H2 == 'All', numeric_cols] = summary.values

# Result

          0         1         2         3         4  H1   H2
0  1.511040  2.258802  1.746791  2.275381  2.678103  M1  All
1  0.345561  0.396767  0.538817  0.419195  0.685220  M1   c1
2  0.027388  0.670468  0.417305  0.558690  0.140387  M1   c2
3  0.968262  0.313424  0.692323  0.876389  0.894607  M1   c3
4  0.169830  0.878143  0.098347  0.421108  0.957890  M1   c4
5  0.984590  0.955375  1.371367  1.359280  1.276477  M2  All
6  0.280444  0.789279  0.103226  0.447894  0.908596  M2   c1
7  0.130029  0.019367  0.678836  0.211628  0.265547  M2   c2
8  0.574118  0.146729  0.589306  0.699758  0.102334  M2   c3

# Division
# Times two assumes that the current totals are correct

df2 = 2 * df.groupby('H1').apply(lambda gp: gp[numeric_cols].div(gp[numeric_cols].sum()))

          0         1         2         3         4
0  1.000000  1.000000  1.000000  1.000000  1.000000
1  0.228691  0.175654  0.308461  0.184230  0.255860
2  0.018125  0.296824  0.238898  0.245537  0.052420
3  0.640791  0.138757  0.396340  0.385161  0.334045
4  0.112393  0.388765  0.056301  0.185071  0.357675
5  1.000000  1.000000  1.000000  1.000000  1.000000
6  0.284833  0.826146  0.075272  0.329508  0.711800
7  0.132064  0.020272  0.495006  0.155691  0.208031
8  0.583103  0.153582  0.429721  0.514801  0.080169