Merge pull request #112 from AMDResearch/dev
Prep v1.0.8-pr2 release
[ROCm/rocprofiler-compute commit: 047868c209]
This commit is contained in:
Cole Ramos
GitHub
@@ -0,0 +1,14 @@
|
||||
FROM --platform=linux/amd64 rockylinux:8
|
||||
|
||||
RUN yum -y install epel-release
|
||||
RUN yum -y install dnf-plugins-core
|
||||
RUN yum -y install cmake
|
||||
RUN yum -y install gcc-c++
|
||||
RUN yum -y install python39
|
||||
|
||||
# Enable rocm package repo
|
||||
COPY rocm.repo /etc/yum.repos.d
|
||||
RUN yum config-manager --set-enabled powertools
|
||||
|
||||
# Desired ROCm packages
|
||||
RUN yum -y install rocm-dev5.1.3
|
||||
@@ -0,0 +1,13 @@
|
||||
[ROCm-5.1.3]
|
||||
name=ROCm5.1.3
|
||||
baseurl=https://repo.radeon.com/rocm/centos8/5.1.3/main
|
||||
enabled=1
|
||||
gpgcheck=1
|
||||
gpgkey=https://repo.radeon.com/rocm/rocm.gpg.key
|
||||
|
||||
[amdgpu]
|
||||
name=amdgpu
|
||||
baseurl=https://repo.radeon.com/amdgpu/latest/rhel/8.5/main/x86_64
|
||||
enabled=1
|
||||
gpgcheck=1
|
||||
gpgkey=https://repo.radeon.com/rocm/rocm.gpg.key
|
||||
@@ -0,0 +1,12 @@
|
||||
FROM --platform=linux/amd64 registry.suse.com/suse/sle15:15.3
|
||||
|
||||
RUN zypper -n --gpg-auto-import-keys addrepo https://download.opensuse.org/repositories/devel:/languages:/perl/SLE_15_SP3/devel:languages:perl.repo
|
||||
|
||||
COPY amdgpu.repo /etc/zypp/repos.d/amdgpu.repo
|
||||
|
||||
RUN zypper -n --gpg-auto-import-keys install rocm-dev5.1.3
|
||||
|
||||
RUN zypper -n install gcc-c++ && \
|
||||
zypper -n install make && \
|
||||
zypper -n install git
|
||||
|
||||
@@ -0,0 +1,13 @@
|
||||
[ROCm-5.1.3]
|
||||
name=ROCm5.1.3
|
||||
baseurl=https://repo.radeon.com/rocm/zyp/5.1.3/main
|
||||
enabled=1
|
||||
gpgcheck=0
|
||||
gpgkey=https://repo.radeon.com/rocm/rocm.gpg.key
|
||||
|
||||
[amdgpu]
|
||||
name=amdgpu
|
||||
baseurl=https://repo.radeon.com/amdgpu/latest/sle/15.3/main/x86_64
|
||||
enabled=1
|
||||
gpgcheck=0
|
||||
gpgkey=https://repo.radeon.com/rocm/rocm.gpg.key
|
||||
@@ -0,0 +1,15 @@
|
||||
FROM --platform=linux/amd64 ubuntu:20.04
|
||||
WORKDIR /home
|
||||
|
||||
USER root
|
||||
|
||||
ENV DEBIAN_FRONTEND noninteractive
|
||||
ENV TZ "US/Chicago"
|
||||
|
||||
RUN apt-get update
|
||||
RUN apt-get install -y wget
|
||||
RUN wget https://repo.radeon.com/amdgpu-install/5.3/ubuntu/focal/amdgpu-install_5.3.50300-1_all.deb
|
||||
RUN apt-get install -y ./amdgpu-install_5.3.50300-1_all.deb
|
||||
RUN amdgpu-install --usecase=rocm -y --accept-eula
|
||||
|
||||
RUN apt-get install -y build-essential
|
||||
@@ -26,6 +26,8 @@
|
||||
|
||||
import sys
|
||||
import os
|
||||
import io
|
||||
import selectors
|
||||
import argparse
|
||||
import subprocess
|
||||
import glob
|
||||
@@ -57,6 +59,48 @@ from common import getVersion
|
||||
def run_subprocess(cmd):
|
||||
subprocess.run(cmd, check=True)
|
||||
|
||||
def capture_subprocess_output(subprocess_args):
|
||||
# Start subprocess
|
||||
# bufsize = 1 means output is line buffered
|
||||
# universal_newlines = True is required for line buffering
|
||||
process = subprocess.Popen(subprocess_args,
|
||||
bufsize=1,
|
||||
stdout=subprocess.PIPE,
|
||||
stderr=subprocess.STDOUT,
|
||||
universal_newlines=True)
|
||||
|
||||
# Create callback function for process output
|
||||
buf = io.StringIO()
|
||||
def handle_output(stream, mask):
|
||||
# Because the process' output is line buffered, there's only ever one
|
||||
# line to read when this function is called
|
||||
line = stream.readline()
|
||||
buf.write(line)
|
||||
sys.stdout.write(line)
|
||||
|
||||
# Register callback for an "available for read" event from subprocess' stdout stream
|
||||
selector = selectors.DefaultSelector()
|
||||
selector.register(process.stdout, selectors.EVENT_READ, handle_output)
|
||||
|
||||
# Loop until subprocess is terminated
|
||||
while process.poll() is None:
|
||||
# Wait for events and handle them with their registered callbacks
|
||||
events = selector.select()
|
||||
for key, mask in events:
|
||||
callback = key.data
|
||||
callback(key.fileobj, mask)
|
||||
|
||||
# Get process return code
|
||||
return_code = process.wait()
|
||||
selector.close()
|
||||
|
||||
success = (return_code == 0)
|
||||
|
||||
# Store buffered output
|
||||
output = buf.getvalue()
|
||||
buf.close()
|
||||
|
||||
return (success, output)
|
||||
|
||||
def resolve_rocprof():
|
||||
# ROCPROF INFO
|
||||
@@ -282,9 +326,13 @@ def detect_roofline():
|
||||
elif rhel_distro == "platform:el8":
|
||||
# Must be a valid RHEL machine
|
||||
distro = rhel_distro
|
||||
elif sles_distro == "15.3":
|
||||
elif (
|
||||
(type(sles_distro) == str and len(sles_distro) >= 3) and # confirm string and len
|
||||
sles_distro[:2] == "15" and int(sles_distro[3]) >= 3 # SLES15 and SP >= 3
|
||||
):
|
||||
# Must be a valid SLES machine
|
||||
distro = sles_distro
|
||||
# Use SP3 binary for all forward compatible service pack versions
|
||||
distro = "15.3"
|
||||
elif ubuntu_distro == "20.04":
|
||||
# Must be a valid Ubuntu machine
|
||||
distro = ubuntu_distro
|
||||
@@ -337,16 +385,20 @@ def characterize_app(args, VER):
|
||||
print("Target: ", args.target)
|
||||
print("Command: ", args.remaining)
|
||||
print("Kernel Selection: ", args.kernel)
|
||||
print("Dispatch Selection: ", args.dispatch, "\n")
|
||||
print("Dispatch Selection: ", args.dispatch)
|
||||
|
||||
perfmon_dir = str(OMNIPERF_HOME) + "/perfmon_pub"
|
||||
print("permon dir is ", os.path.abspath(perfmon_dir))
|
||||
app_cmd = args.remaining
|
||||
workload_dir =args.path
|
||||
workload_dir = args.path
|
||||
|
||||
# Perfmon filtering
|
||||
pmc_filter(workload_dir, perfmon_dir, args.target)
|
||||
|
||||
# Set up a log file
|
||||
log = open(workload_dir + "/log.txt", "w")
|
||||
print("Log: ", workload_dir + "/log.txt\n")
|
||||
|
||||
# Workload profiling
|
||||
for fname in glob.glob(workload_dir + "/perfmon/*.txt"):
|
||||
# Kernel filtering (in-place replacement)
|
||||
@@ -376,7 +428,11 @@ def characterize_app(args, VER):
|
||||
if args.use_rocscope == True:
|
||||
run_rocscope(args, fname)
|
||||
else:
|
||||
run_prof(fname, workload_dir, perfmon_dir, app_cmd, args.target, args.verbose)
|
||||
run_prof(fname, workload_dir, perfmon_dir, app_cmd, args.target, log, args.verbose)
|
||||
|
||||
# Close log
|
||||
log.close()
|
||||
|
||||
# run again with timestamps
|
||||
run_subprocess(
|
||||
[
|
||||
@@ -398,7 +454,6 @@ def characterize_app(args, VER):
|
||||
# Profiling Helpers
|
||||
################################################
|
||||
|
||||
|
||||
def run_rocscope(args, fname):
|
||||
# profile the app
|
||||
if args.use_rocscope == True:
|
||||
@@ -428,7 +483,7 @@ def run_rocscope(args, fname):
|
||||
sys.exit(1)
|
||||
|
||||
|
||||
def run_prof(fname, workload_dir, perfmon_dir, cmd, target, verbose):
|
||||
def run_prof(fname, workload_dir, perfmon_dir, cmd, target, log_file, verbose):
|
||||
global rocprof_cmd
|
||||
|
||||
fbase = os.path.splitext(os.path.basename(fname))[0]
|
||||
@@ -439,7 +494,7 @@ def run_prof(fname, workload_dir, perfmon_dir, cmd, target, verbose):
|
||||
# profile the app (run w/ custom config files for mi100)
|
||||
if target == "mi100":
|
||||
print("RUNNING WITH CUSTOM METRICS")
|
||||
run_subprocess(
|
||||
success, output = capture_subprocess_output(
|
||||
[
|
||||
rocprof_cmd,
|
||||
"-i",
|
||||
@@ -454,7 +509,7 @@ def run_prof(fname, workload_dir, perfmon_dir, cmd, target, verbose):
|
||||
]
|
||||
)
|
||||
else:
|
||||
run_subprocess(
|
||||
success, output = capture_subprocess_output(
|
||||
[
|
||||
rocprof_cmd,
|
||||
"-i",
|
||||
@@ -466,6 +521,9 @@ def run_prof(fname, workload_dir, perfmon_dir, cmd, target, verbose):
|
||||
'"' + cmd + '"',
|
||||
]
|
||||
)
|
||||
# Write output to log
|
||||
log_file.write(output)
|
||||
|
||||
|
||||
|
||||
def omniperf_profile(args, VER):
|
||||
@@ -482,9 +540,9 @@ def omniperf_profile(args, VER):
|
||||
print("Dispatch Selection: ", args.dispatch)
|
||||
|
||||
if args.ipblocks == None:
|
||||
print("IP Blocks: All", "\n")
|
||||
print("IP Blocks: All")
|
||||
else:
|
||||
print("IP Blocks: ", args.ipblocks, "\n")
|
||||
print("IP Blocks: ", args.ipblocks)
|
||||
|
||||
# Set up directories
|
||||
workload_dir = args.path + "/" + args.name + "/" + args.target
|
||||
@@ -493,6 +551,10 @@ def omniperf_profile(args, VER):
|
||||
# Perfmon filtering
|
||||
perfmon_filter(workload_dir, perfmon_dir, args)
|
||||
|
||||
# Set up a log file
|
||||
log = open(workload_dir + "/log.txt", "w")
|
||||
print("Log: ", workload_dir + "/log.txt\n")
|
||||
|
||||
if not args.lucky == None and args.lucky == True:
|
||||
print("You're feeling lucky - only profiling top N kernels")
|
||||
# look for whether workload_dir exists - create if not
|
||||
@@ -591,7 +653,7 @@ def omniperf_profile(args, VER):
|
||||
if args.use_rocscope == True:
|
||||
run_rocscope(args, fname)
|
||||
else:
|
||||
run_prof(fname, workload_dir, perfmon_dir, args.remaining, args.target, args.verbose)
|
||||
run_prof(fname, workload_dir, perfmon_dir, args.remaining, args.target, log, args.verbose)
|
||||
|
||||
# run again with timestamps
|
||||
run_subprocess(
|
||||
@@ -606,7 +668,6 @@ def omniperf_profile(args, VER):
|
||||
'"' + args.remaining + '"',
|
||||
]
|
||||
)
|
||||
|
||||
# Update pmc_perf.csv timestamps
|
||||
replace_timestamps(workload_dir)
|
||||
|
||||
@@ -648,6 +709,8 @@ def omniperf_profile(args, VER):
|
||||
str(args.device),
|
||||
]
|
||||
)
|
||||
# Close log
|
||||
log.close()
|
||||
|
||||
|
||||
################################################
|
||||
|
||||
@@ -213,7 +213,9 @@ ul#nav li {
|
||||
font-size: 14px;
|
||||
text-align: left;
|
||||
}
|
||||
|
||||
.VirtualizedSelectOption {
|
||||
overflow: hidden;
|
||||
}
|
||||
#nav .nav-right {
|
||||
float: right;
|
||||
}
|
||||
@@ -246,6 +248,7 @@ button.report:hover {
|
||||
font-size: 16px;
|
||||
}
|
||||
|
||||
/* Menu Button */
|
||||
.dropdown button {
|
||||
display: inline-block;
|
||||
font: 16px 'opensans-bold', sans-serif;
|
||||
@@ -396,6 +399,18 @@ button.report:hover {
|
||||
#memchart a, #memchart a:visited { color: #fff; }
|
||||
#memchart a:hover, #memchart a:focus { color: #11ABB0; }
|
||||
|
||||
/* ------------------------------------------------------------------ */
|
||||
/* Popup Section
|
||||
/* ------------------------------------------------------------------ */
|
||||
#popup{
|
||||
text-align: center;
|
||||
border: 3px solid #F06000;
|
||||
margin: auto;
|
||||
width: 50%;
|
||||
padding: 10px;
|
||||
color: white;
|
||||
}
|
||||
|
||||
/* ------------------------------------------------------------------ */
|
||||
/* Top Stat Section
|
||||
/* ------------------------------------------------------------------ */
|
||||
|
||||
@@ -22,7 +22,7 @@ Panel Config:
|
||||
Access Rate:
|
||||
value: AVG(((200 * SQ_ACTIVE_INST_LDS) / (GRBM_GUI_ACTIVE * $numCU)))
|
||||
tips:
|
||||
Bandwith (Pct-of-Peak):
|
||||
Bandwidth (Pct-of-Peak):
|
||||
value: AVG((((((SQ_LDS_IDX_ACTIVE - SQ_LDS_BANK_CONFLICT) * 4) * TO_INT($L2Banks))
|
||||
/ (EndNs - BeginNs)) / (($sclk * $numCU) * 0.00128)))
|
||||
tips:
|
||||
|
||||
@@ -22,7 +22,7 @@ Panel Config:
|
||||
Access Rate:
|
||||
value: AVG(((200 * SQ_ACTIVE_INST_LDS) / (GRBM_GUI_ACTIVE * $numCU)))
|
||||
tips:
|
||||
Bandwith (Pct-of-Peak):
|
||||
Bandwidth (Pct-of-Peak):
|
||||
value: AVG((((((SQ_LDS_IDX_ACTIVE - SQ_LDS_BANK_CONFLICT) * 4) * TO_INT($L2Banks))
|
||||
/ (EndNs - BeginNs)) / (($sclk * $numCU) * 0.00128)))
|
||||
tips:
|
||||
|
||||
@@ -22,7 +22,7 @@ Panel Config:
|
||||
Access Rate:
|
||||
value: AVG(((200 * SQ_ACTIVE_INST_LDS) / (GRBM_GUI_ACTIVE * $numCU)))
|
||||
tips:
|
||||
Bandwith (Pct-of-Peak):
|
||||
Bandwidth (Pct-of-Peak):
|
||||
value: AVG((((((SQ_LDS_IDX_ACTIVE - SQ_LDS_BANK_CONFLICT) * 4) * TO_INT($L2Banks))
|
||||
/ (EndNs - BeginNs)) / (($sclk * $numCU) * 0.00128)))
|
||||
tips:
|
||||
|
||||
@@ -25,6 +25,7 @@
|
||||
from selectors import EpollSelector
|
||||
import sys
|
||||
import copy
|
||||
import os.path
|
||||
import pandas as pd
|
||||
from dash.dash_table import FormatTemplate
|
||||
from dash.dash_table.Format import Format, Scheme, Symbol
|
||||
@@ -342,6 +343,8 @@ def build_table_chart(
|
||||
style_cell_conditional=[
|
||||
{"if": {"column_id": display_columns[0]}, "textAlign": "left"}
|
||||
],
|
||||
# style cell
|
||||
style_cell={"maxWidth": "500px"},
|
||||
# display style
|
||||
style_header={
|
||||
"backgroundColor": "rgb(30, 30, 30)",
|
||||
@@ -350,7 +353,12 @@ def build_table_chart(
|
||||
}
|
||||
if IS_DARK
|
||||
else {},
|
||||
style_data={"backgroundColor": "rgb(50, 50, 50)", "color": "white"}
|
||||
style_data={
|
||||
"backgroundColor": "rgb(50, 50, 50)",
|
||||
"color": "white",
|
||||
"whiteSpace": "normal",
|
||||
"height": "auto",
|
||||
}
|
||||
if IS_DARK
|
||||
else {},
|
||||
style_data_conditional=[
|
||||
@@ -466,13 +474,15 @@ def build_layout(
|
||||
get_memchart(panel_configs[1900]["data source"], base_data[base_run])
|
||||
)
|
||||
# append roofline section
|
||||
div_children.append(
|
||||
get_roofline(
|
||||
path_to_dir,
|
||||
parser.apply_filters(base_data[base_run], is_gui, debug),
|
||||
verbose,
|
||||
has_roofline = os.path.isfile(path_to_dir + "/roofline.csv")
|
||||
if has_roofline:
|
||||
div_children.append(
|
||||
get_roofline(
|
||||
path_to_dir,
|
||||
parser.apply_filters(base_data[base_run], is_gui, debug),
|
||||
verbose,
|
||||
)
|
||||
)
|
||||
)
|
||||
# Iterate over each section as defined in panel configs
|
||||
for panel_id, panel in panel_configs.items():
|
||||
title = str(panel_id // 100) + ". " + panel["title"]
|
||||
@@ -590,4 +600,17 @@ def build_layout(
|
||||
)
|
||||
)
|
||||
|
||||
# Display pop-up message if no filters are applied
|
||||
if not (disp_filt or kernel_filter or gcd_filter):
|
||||
div_children.append(
|
||||
html.Section(
|
||||
id="popup",
|
||||
children=[
|
||||
html.Div(
|
||||
children="To dive deeper, use the top drop down menus to isolate particular kernel(s) or dispatch(s). You will then see the web page update with additional low-level metrics specific to the filter you've applied.",
|
||||
),
|
||||
],
|
||||
)
|
||||
)
|
||||
|
||||
return div_children
|
||||
|
||||
@@ -39,6 +39,13 @@ def list_unique(orig_list, is_numeric):
|
||||
return unique_list
|
||||
|
||||
|
||||
def create_span(input):
|
||||
elmt = {}
|
||||
elmt["label"] = (html.Span(str(input), title=str(input)),)
|
||||
elmt["value"] = str(input)
|
||||
return elmt
|
||||
|
||||
|
||||
def get_header(raw_pmc, input_filters, kernel_names):
|
||||
return html.Header(
|
||||
id="home",
|
||||
@@ -163,40 +170,6 @@ def get_header(raw_pmc, input_filters, kernel_names):
|
||||
)
|
||||
],
|
||||
),
|
||||
html.Li(
|
||||
className="filter",
|
||||
children=[
|
||||
html.Div(
|
||||
children=[
|
||||
html.A(
|
||||
className="smoothscroll",
|
||||
children=["Kernels:"],
|
||||
),
|
||||
dcc.Dropdown(
|
||||
list_unique(
|
||||
list(
|
||||
map(
|
||||
str,
|
||||
raw_pmc[
|
||||
schema.pmc_perf_file_prefix
|
||||
]["KernelName"],
|
||||
)
|
||||
),
|
||||
False,
|
||||
), # list avail kernel names
|
||||
id="kernel-filt",
|
||||
multi=True,
|
||||
value=kernel_names,
|
||||
placeholder="ALL",
|
||||
style={
|
||||
"width": "600px", # TODO: Change these widths to % rather than fixed value
|
||||
"height": "34px",
|
||||
},
|
||||
),
|
||||
]
|
||||
)
|
||||
],
|
||||
),
|
||||
html.Li(
|
||||
className="filter",
|
||||
children=[
|
||||
@@ -261,6 +234,45 @@ def get_header(raw_pmc, input_filters, kernel_names):
|
||||
)
|
||||
],
|
||||
),
|
||||
html.Li(
|
||||
className="filter",
|
||||
children=[
|
||||
html.Div(
|
||||
children=[
|
||||
html.A(
|
||||
className="smoothscroll",
|
||||
children=["Kernels:"],
|
||||
),
|
||||
dcc.Dropdown(
|
||||
list(
|
||||
map(
|
||||
create_span,
|
||||
list_unique(
|
||||
list(
|
||||
map(
|
||||
str,
|
||||
raw_pmc[
|
||||
schema.pmc_perf_file_prefix
|
||||
]["KernelName"],
|
||||
)
|
||||
),
|
||||
False,
|
||||
), # list avail kernel names
|
||||
)
|
||||
),
|
||||
id="kernel-filt",
|
||||
multi=True,
|
||||
value=kernel_names,
|
||||
optionHeight=150,
|
||||
placeholder="ALL",
|
||||
style={
|
||||
"width": "600px", # TODO: Change these widths to % rather than fixed value
|
||||
},
|
||||
),
|
||||
]
|
||||
)
|
||||
],
|
||||
),
|
||||
html.Div(
|
||||
className="nav-right",
|
||||
children=[
|
||||
|
||||
@@ -235,7 +235,7 @@ def get_roofline(
|
||||
|
||||
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)
|
||||
path_to_dir + "/empirRoof_gpu-{}_int8_fp16.pdf".format(dev_id)
|
||||
)
|
||||
if kernel_names:
|
||||
# only save a legend if kernel_names option is toggled
|
||||
@@ -244,7 +244,7 @@ def get_roofline(
|
||||
# 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)
|
||||
path_to_dir + "/empirRoof_gpu-{}_int8_fp16.pdf".format(dev_id)
|
||||
)
|
||||
if kernel_names:
|
||||
legend.write_image(path_to_dir + "/kernelName_legend.pdf")
|
||||
|
||||
Reference in New Issue
Block a user