Option for different markers per kernel name

Signed-off-by: coleramos425 <colramos@amd.com> [ROCm/rocprofiler-compute commit: 27fa97aad9]
2023-02-27 16:18:42 -06:00
@@ -715,7 +715,7 @@ def main():
            # Setup prerequisits for roofline
            roof_setup(args, my_parser, VER)
            # Generate roofline
-            roofline_only(args.path, args.device, args.sort, args.mem_level, args.verbose)
+            roofline_only(args.path, args.device, args.sort, args.mem_level, args.kernel_names, args.verbose)

        # Profile only
        else:
@@ -212,7 +212,7 @@ def run_cli(args, runs):
        )


-def roofline_only(path_to_dir, dev_id, sort_type, mem_level, verbose):
+def roofline_only(path_to_dir, dev_id, sort_type, mem_level, kernel_names, verbose):
    import pandas as pd
    from collections import OrderedDict

@@ -235,6 +235,7 @@ def roofline_only(path_to_dir, dev_id, sort_type, mem_level, verbose):
        dev_id,  # [Optional] Specify device id to collect roofline info from
        sort_type,  # [Optional] Sort AI by top kernels or dispatches
        mem_level,  # [Optional] Toggle particular level(s) of memory hierarchy
+        kernel_names, # [Optional] Toggle overlay of kernel names in plot
        True,  # [Optional] Generate a standalone roofline analysis
    )

@@ -25,6 +25,7 @@
 from omniperf_analyze.utils import roofline_calc

 import time
+import sys
 import numpy as np
 from dash import html, dash_table

@@ -32,6 +33,8 @@ from dash import dcc
 import plotly.graph_objects as go


+SYMBOLS = [0,1,2,3,4,5,13,17,18,20]
+
 def to_int(a):
    if str(type(a)) == "<class 'NoneType'>":
        return np.nan
@@ -120,6 +123,7 @@ def generate_plots(roof_info, ai_data, mem_level, is_standalone, verbose, fig=No
                y=ai_data["ai_l1"][1],
                name="ai_l1",
                mode="markers",
+                marker_symbol=SYMBOLS,
            )
        )
        fig.add_trace(
@@ -128,6 +132,7 @@ def generate_plots(roof_info, ai_data, mem_level, is_standalone, verbose, fig=No
                y=ai_data["ai_l2"][1],
                name="ai_l2",
                mode="markers",
+                marker_symbol=SYMBOLS,
            )
        )
        fig.add_trace(
@@ -136,6 +141,7 @@ def generate_plots(roof_info, ai_data, mem_level, is_standalone, verbose, fig=No
                y=ai_data["ai_hbm"][1],
                name="ai_hbm",
                mode="markers",
+                marker_symbol=SYMBOLS,
            )
        )

@@ -158,8 +164,13 @@ def get_roofline(
    dev_id=None,
    sort_type="kernels",
    mem_level="ALL",
+    kernel_names=False,
    is_standalone=False,
 ):
+    if kernel_names and (not is_standalone):
+        print("ERROR: --roof-only is required for --kernel-names")
+        sys.exit(1)
+
    # Roofline settings
    fp32_details = {
        "path": path_to_dir,
@@ -190,6 +201,18 @@ def get_roofline(
    ml_combo_fig = generate_plots(
        int8_details, ai_data, mem_level, is_standalone, verbose, fp16_fig
    )
+    legend = go.Figure(
+        go.Scatter(
+            mode="markers",
+            x=[0]*10,
+            y=ai_data["kernelNames"],
+            marker_symbol=SYMBOLS,
+            marker_size=15,
+        )
+    )
+    legend.update_layout(title="Kernel Names and Markers",
+                  margin=dict(b=0,r=0), xaxis_range=[-1,1], xaxis_side="top", height=400, width=1000)
+    legend.update_xaxes(dtick=1)

    if is_standalone:
        dev_id = "ALL" if dev_id == -1 else str(dev_id)
@@ -198,12 +221,18 @@ def get_roofline(
        ml_combo_fig.write_image(
            path_to_dir + "/empirRoof_gpu-{}_fp8_fp16.pdf".format(dev_id)
        )
+        legend.write_image(
+            path_to_dir + "/kernelName_legend.pdf"
+        )
        time.sleep(1)
        # Re-save to remove loading MathJax pop up
        fp32_fig.write_image(path_to_dir + "/empirRoof_gpu-{}_fp32.pdf".format(dev_id))
        ml_combo_fig.write_image(
            path_to_dir + "/empirRoof_gpu-{}_fp8_fp16.pdf".format(dev_id)
        )
+        legend.write_image(
+            path_to_dir + "/kernelName_legend.pdf"
+        )
        print("Empirical Roofline PDFs saved!")
    else:
        return html.Section(
@@ -44,6 +44,8 @@ FONT_WEIGHT = "bold"

 SUPPORTED_SOC = ["mi200"]

+TOP_N = 10
+

 ################################################
 # Helper funcs
@@ -428,9 +430,11 @@ def plot_application(sortType, ret_df, verbose):
    # print("Top 5 intensities ('{}')...".format(roof_details["sort"]))
    intensities = {"ai_l1": [], "ai_l2": [], "ai_hbm": []}
    curr_perf = []
+    kernelNames = []
    i = 0
    # Create list of top 5 intensities
-    while i <= 9 and i != len(myList):
+    while i < TOP_N and i != len(myList):
+        kernelNames.append(myList[i].KernelName)
        intensities["ai_l1"].append(
            myList[i].total_flops / myList[i].L1cache_data
        ) if myList[i].L1cache_data else intensities["ai_l1"].append(0)
@@ -470,6 +474,9 @@ def plot_application(sortType, ret_df, verbose):
        intensityPoints[i].append(x)
        intensityPoints[i].append(y)

+    # Add an entry for kernel names
+    intensityPoints["kernelNames"] = kernelNames
+
    return intensityPoints


@@ -234,6 +234,13 @@ def parse(my_parser):
        type=int,
        help="\t\t\tGPU device ID. (DEFAULT: ALL)",
    )
+    roofline_group.add_argument(
+        "--kernel-names",
+        required=False,
+        default=False,
+        action="store_true",
+        help="\t\t\tInclude kernel names in roofline plot.",
+    )
    # roofline_group.add_argument('-w', '--workgroups', required=False, default=-1, type=int, help="\t\t\tNumber of kernel workgroups (DEFAULT: 1024)")
    # roofline_group.add_argument('--wsize', required=False, default=-1, type=int, help="\t\t\tWorkgroup size (DEFAULT: 256)")
    # roofline_group.add_argument('--dataset', required=False, default = -1, type=int, help="\t\t\tDataset size (DEFAULT: 536M)")