Visualization

The generated CostModelEvaluation object(s) (from e.g. the API call) can be visualized in multiple ways. The visualizations described here are automatically generated when running the example.py script.

Temporal mapping

The temporal mapping can be visualized by a function that prints it to the terminal. The code block demonstrates how to use it (copied from example.py):

from zigzag.utils import pickle_load
from zigzag.visualization.results.print_mapping import print_mapping

pickle_filename = f"outputs/list_of_cmes.pickle"
with open(pickle_filename, "rb") as fp:
    cmes = pickle.load(fp)

print_mapping(cmes[0])

The function will show the loops of the temporal mapping and for each operand shows at which memory level it resides. For example:

Loop ordering for /conv1/Conv
==================================================================================
Temporal Loops                       I              W              O
==================================================================================
for OX in [0, 16):                   dram           dram           dram
----------------------------------------------------------------------------------
for OX in [0, 7):                  sram_256KB     dram           dram
----------------------------------------------------------------------------------
    for FY in [0, 7):                sram_256KB     dram           sram_256KB
----------------------------------------------------------------------------------
    for C in [0, 3):               sram_256KB     dram           sram_256KB
----------------------------------------------------------------------------------
        for OY in [0, 16):           sram_256KB     cells          sram_256KB
----------------------------------------------------------------------------------
        for OY in [0, 7):          sram_256KB     cells          sram_256KB
----------------------------------------------------------------------------------
==================================================================================
Spatial Loops
==================================================================================
            parfor K in [0, 64):
----------------------------------------------------------------------------------
            parfor FX in [0, 7):
----------------------------------------------------------------------------------

The top loop is the outer-most for loop, where as the bottom loop is the inner-most. Going from bottom to top, loops are allocated to the innermost memories of the memory hierarchy for each operand.

Energy and latency breakdown

The energy and latency breakdown of a list of CostModelEvaluation objects can be plotted using the bar_plot_cost_model_evaluations_breakdown function:

from zigzag.utils import pickle_load
from zigzag.visualization.results.plot_cme import bar_plot_cost_model_evaluations_breakdown

pickle_filename = f"outputs/list_of_cmes.pickle"
with open(pickle_filename, "rb") as fp:
    cmes = pickle.load(fp)

bar_plot_cost_model_evaluations_breakdown(cmes, save_path="outputs/plot_breakdown.png")

This will produce a bar chart, for example: