From a100a322fe414021a1e5878b910a84e9df37df61 Mon Sep 17 00:00:00 2001
From: Joshua Bakita <jbakita@cs.unc.edu>
Date: Fri, 12 Jun 2020 01:32:55 +0000
Subject: Add inital testing infrastructure

Works for three-way WSS/cache size/execution time comparisons
---
 dis/original/WSS_DOCS.md                  |  83 ++++++++++++++++++++++
 dis/original/clean.sh                     |   1 +
 dis/original/gen_input.py                 | 113 ++++++++++++++++++++++++++++++
 dis/original/inputs/WSSS                  |  12 ++++
 dis/original/inputs/WSSS_maxstride4mb     |  16 +++++
 dis/original/inputs/caches                |   7 ++
 dis/original/inputs/caches_maxstride2ways |  11 +++
 dis/original/postproc.sh                  |   3 +
 dis/original/run_dis.sh                   |  81 +++++++++++++++++++++
 dis/original/setup_mem_and_global.sh      |  13 ++++
 10 files changed, 340 insertions(+)
 create mode 100644 dis/original/WSS_DOCS.md
 create mode 100755 dis/original/clean.sh
 create mode 100755 dis/original/gen_input.py
 create mode 100644 dis/original/inputs/WSSS
 create mode 100644 dis/original/inputs/WSSS_maxstride4mb
 create mode 100644 dis/original/inputs/caches
 create mode 100644 dis/original/inputs/caches_maxstride2ways
 create mode 100755 dis/original/postproc.sh
 create mode 100755 dis/original/run_dis.sh
 create mode 100755 dis/original/setup_mem_and_global.sh

diff --git a/dis/original/WSS_DOCS.md b/dis/original/WSS_DOCS.md
new file mode 100644
index 0000000..da5e066
--- /dev/null
+++ b/dis/original/WSS_DOCS.md
@@ -0,0 +1,83 @@
+# Documentation Mapping DIS Stressmark Parameters to WSS
+
+## Field
+1 allocation in main()
+f = 1st input param
+
+sizeof(unsigned char) * f
+
+## Matrix
+6 allocations in main(), 7 allocations in biConjugateGradient()
+
+*Allocations in main()*
+dim = 2nd input param
+numberNonzero = 3nd input param
+
+sizeof(double) * (dim^2+3dim+numberNonzero) + sizeof(int) * (2dim+1+numberNonzero)
+
+*Allocations in biConjugateGradient()*
+sizeof(double) * 7dim
+
+## Neighborhood
+1 allocation in createImage, 2 allocations in neighborhoodCalculation
+
+*Allocations in createImage()*
+dimension = 3rd input param
+
+sizeof(short int) * dimension^2
+
+*Allocations in neighborhoodCalculation()*
+bitDepth = 2nd param
+
+sizeof(int) * (2^(bitDepth + 1) - 1)
+
+## Pointer
+n = 5th input param
+f = 1st input param
+
+sizeof(unsigned int) * 4n + sizeof(int) * f
+
+## Transitive
+n = 1st input param
+
+sizeof(unsigned int) * 2n^2
+
+## Update
+f = 1st input param
+
+sizeof(int) * f
+
+## Testplan
+*Problem!* Larger WSS = more computations
+Use testcase #1 for non-specified parameters
+Below math computed for x86_64
+- Test WSS at powers of 2: 16 KiB, 32, 64, 128, 256, 512, 1MiB, 2, 4, 8, 16, 32
+- For each WSS, measure cache allocation of 0, 1, 2, 4, 8, 16
+
+### Field
+Just vary first param
+
+f = WSS
+
+### Matrix
+0.3 - 16% number nonzero
+- Fixed at 8%
+Just vary dim (matrix size)
+
+sizeof(double) * (dim^2+10dim+numberNonzero) + sizeof(int) * (2dim+1+numberNonzero) = WSS
+
+
+### Neighborhood
+8 or 15 bit depth
+- Fix at 12?
+Just vary dim (image size)
+
+### Pointer
+10 for n
+Just vary f
+
+### Transitive
+Just vary n
+
+### Update
+Just vary f
diff --git a/dis/original/clean.sh b/dis/original/clean.sh
new file mode 100755
index 0000000..7c58295
--- /dev/null
+++ b/dis/original/clean.sh
@@ -0,0 +1 @@
+tail -n +2 $1 | tr "-" " " | sed "s/L3:0=0000;1=0000;2=0000;3=0000/0/" | sed "s/L3:0=0000;1=0000;2=0000;3=0001/1/" | sed "s/L3:0=0000;1=0000;2=0000;3=0003/2/" | sed "s/L3:0=0000;1=0000;2=0000;3=0007/3/" | sed "s/L3:0=0000;1=0000;2=0000;3=000f/4/" | sed "s/L3:0=0000;1=0000;2=0000;3=003f/6/" | sed "s/L3:0=0000;1=0000;2=0000;3=00ff/8/" | sed "s/L3:0=0000;1=0000;2=0000;3=03ff/10/" | sed "s/L3:0=0000;1=0000;2=0000;3=0fff/12/" | sed "s/L3:0=0000;1=0000;2=0000;3=3fff/14/" | sed "s/L3:0=0000;1=0000;2=0000;3=ffff/16/" > $1.clean
diff --git a/dis/original/gen_input.py b/dis/original/gen_input.py
new file mode 100755
index 0000000..c7821b0
--- /dev/null
+++ b/dis/original/gen_input.py
@@ -0,0 +1,113 @@
+#!/usr/bin/python3
+#####
+# Copyright 2020 Joshua Bakita
+#
+# This program generates input data for the DIS benchmark suite on stdout
+# given a requested working set size.
+#####
+
+
+from ctypes import sizeof, c_double, c_int, c_short
+from math import sqrt, floor
+import sys # For argv and stderr
+
+USAGE = """Usage: {} <benchmark> <template> <WSS in bytes>"""
+
+# Check input
+if (len(sys.argv) < 4):
+    print(USAGE.format(sys.argv[0]), file=sys.stderr)
+    exit(1);
+
+# Don't try to understand the logic in these functions, see WSS_DOCS.md
+def setup_field(params, wss):
+    params[0] = wss
+    return params
+
+def setup_matrix(params, wss):
+    nnZR = 0.08 # 8% seems average
+    # This formula is out of a solver
+    si = sizeof(c_int)
+    sd = sizeof(c_double)
+    d = (sqrt((si**2) * (-(nnZR-1)) - si*sd*(nnZR-9) + si*wss*nnZR + sd*(25*sd+wss*nnZR+wss)) - si - 5*sd) / (si*nnZR + sd*nnZR + sd)
+    params[1] = floor(d);
+    params[2] = floor(d*d*nnZR);
+    if params[1] <= 0 or params[2] <= 0:
+        raise Exception("WSS too small for matrix benchmark!")
+    return params
+
+def setup_neighborhood(params, wss):
+    bitDepth = 8
+    bitDepthAlloc = sizeof(c_int) * (2**(bitDepth + 1) - 1)
+    dim = sqrt((wss - bitDepthAlloc) / sizeof(c_short))
+    params[1] = bitDepth
+    params[2] = floor(dim)
+    if params[1] <= 0 or params[2] <= 0:
+        raise Exception("WSS too small for neighborhood benchmark!")
+    # Cap maximum line thinkness to the image size
+    params[5] = min(params[2]-1, int(params[5]))
+    # Cap line lengths to the image size
+    params[6] = min(params[2]-1, int(params[6]))
+    params[7] = min(params[2]-1, int(params[7]))
+    return params
+
+def setup_pointer(params, wss):
+    n = 10;
+    f = (wss - sizeof(c_int) * 4 * n) / sizeof(c_int)
+    params[0] = floor(f)
+    params[4] = floor(n)
+    if params[0] <= 0 or params[4] <= 0:
+        raise Exception("WSS too small for pointer benchmark!")
+    return params
+
+def setup_transitive(params, wss):
+    n = sqrt(wss / (sizeof(c_int) * 2))
+    params[0] = floor(n)
+    # Fix edges at 50%
+    params[1] = floor(params[0] * 0.5)
+    if params[0] <= 0:
+        raise Exception("WSS too small for transitive benchmark!")
+    return params
+
+def setup_update(params, wss):
+    f = wss / sizeof(c_int)
+    params[0] = floor(f)
+    if params[0] <= 0:
+        raise Exception("WSS too small for update benchmark!")
+    # Don't do more than 100M hops (keeps time array feasible)
+    params[2] = min(100000000, int(params[2]))
+    # Enforce size requirements
+    params[4] = min(params[0]-1, int(params[4]))
+    params[5] = min(params[0]-1, int(params[5]))
+    params[6] = min(params[0]-1, int(params[6]))
+    return params
+
+def setup_random_walk(params, wss):
+    params[0] = wss
+    return params
+
+BENCH_TO_PARAMS = {"field":setup_field, "matrix":setup_matrix, "neighborhood":setup_neighborhood, "pointer":setup_pointer, "transitive":setup_transitive, "update":setup_update, "random_walk":setup_random_walk}
+
+# Main logic
+benchmark_name = sys.argv[1]
+if benchmark_name not in BENCH_TO_PARAMS.keys():
+    print("Invalid benchmark name.", file=sys.stderr)
+    exit(2)
+
+wss = int(sys.argv[3])
+if wss <= 0:
+    print("Invalid working set size", file=sys.stderr)
+    exit(3)
+
+with open(sys.argv[2], "r") as template:
+    # We expect the initialization params to all be on the first line
+    params = template.readline().split()
+    mutated_params = BENCH_TO_PARAMS[benchmark_name](params, wss);
+    print(" ".join(map(lambda x: str(x), mutated_params)))
+    print(" ".join(map(lambda x: str(x), mutated_params)), file=sys.stderr)
+    if benchmark_name == "pointer":
+        # Clone the data format used in the template
+        for i in range(0,10):
+            print("10 " + str(mutated_params[0]-10) + " " + str(mutated_params[0]-10))
+    else:
+        print(template.read())
+    
diff --git a/dis/original/inputs/WSSS b/dis/original/inputs/WSSS
new file mode 100644
index 0000000..8836023
--- /dev/null
+++ b/dis/original/inputs/WSSS
@@ -0,0 +1,12 @@
+16384
+32768
+65536
+131072
+262144
+524288
+1048576
+2097152
+4194304
+8388608
+16777216
+33554432
diff --git a/dis/original/inputs/WSSS_maxstride4mb b/dis/original/inputs/WSSS_maxstride4mb
new file mode 100644
index 0000000..f402c01
--- /dev/null
+++ b/dis/original/inputs/WSSS_maxstride4mb
@@ -0,0 +1,16 @@
+16384
+32768
+65536
+131072
+262144
+524288
+1048576
+2097152
+4194304
+8388608
+12582912
+16777216
+20971520
+25165824
+29360120
+33554432
diff --git a/dis/original/inputs/caches b/dis/original/inputs/caches
new file mode 100644
index 0000000..f2293da
--- /dev/null
+++ b/dis/original/inputs/caches
@@ -0,0 +1,7 @@
+L3:0=0000;1=0000;2=0000;3=0000
+L3:0=0000;1=0000;2=0000;3=0001
+L3:0=0000;1=0000;2=0000;3=0003
+L3:0=0000;1=0000;2=0000;3=0007
+L3:0=0000;1=0000;2=0000;3=000f
+L3:0=0000;1=0000;2=0000;3=00ff
+L3:0=0000;1=0000;2=0000;3=ffff
diff --git a/dis/original/inputs/caches_maxstride2ways b/dis/original/inputs/caches_maxstride2ways
new file mode 100644
index 0000000..d5346b1
--- /dev/null
+++ b/dis/original/inputs/caches_maxstride2ways
@@ -0,0 +1,11 @@
+L3:0=0000;1=0000;2=0000;3=0000
+L3:0=0000;1=0000;2=0000;3=0001
+L3:0=0000;1=0000;2=0000;3=0003
+L3:0=0000;1=0000;2=0000;3=0007
+L3:0=0000;1=0000;2=0000;3=000f
+L3:0=0000;1=0000;2=0000;3=003f
+L3:0=0000;1=0000;2=0000;3=00ff
+L3:0=0000;1=0000;2=0000;3=03ff
+L3:0=0000;1=0000;2=0000;3=0fff
+L3:0=0000;1=0000;2=0000;3=3fff
+L3:0=0000;1=0000;2=0000;3=ffff
diff --git a/dis/original/postproc.sh b/dis/original/postproc.sh
new file mode 100755
index 0000000..c5dd1c9
--- /dev/null
+++ b/dis/original/postproc.sh
@@ -0,0 +1,3 @@
+#!/bin/bash
+./summarize.py $1.txt > $1-pp.txt
+./clean.sh $1-pp.txt
diff --git a/dis/original/run_dis.sh b/dis/original/run_dis.sh
new file mode 100755
index 0000000..3e82bfb
--- /dev/null
+++ b/dis/original/run_dis.sh
@@ -0,0 +1,81 @@
+#!/bin/bash
+
+core=$1
+maxJobs=$2
+runID=$3
+benchmark=${4,}
+template_input=inputs/$4/in1
+wss_settings=inputs/WSSS
+cache_settings=inputs/caches
+
+if [ $# -lt 4 ]; then
+    echo "Usage $0 <core ID> <number of iterations> <run ID> <benchmark> [template input] [DIS WSS file] [DIS cache file]"
+    exit
+fi
+
+if [ $# -gt 4 ]; then
+    echo "Using alternate input template from $5"
+    template_input=$5
+fi
+
+if [ $# -gt 5 ]; then
+    echo "Using alternate WSS settings from $6"
+    wss_settings=$6
+fi
+
+if [ $# -gt 6 ]; then
+    echo "Using alternate cache settings from $7"
+    cache_settings=$7
+fi
+
+echo "Making sure that binary is up to date..."
+make $benchmark
+echo "Done. Disabling real-time throttling..."
+
+# Turn off rt throttling
+echo -1 > /proc/sys/kernel/sched_rt_runtime_us
+echo "Done. Redirecting all interrupts to core 0..."
+
+# TODO: Make this cleaner
+# Redirect all interrupts to core 0
+i=0
+for IRQ in /proc/irq/*
+do
+    # Skip default_smp_affinity
+    if [ -d $IRQ ]; then
+        irqList[$i]=$(cat $IRQ/smp_affinity_list)
+        echo 0 > $IRQ/smp_affinity_list
+    fi
+    i=$(( $i + 1 ))
+done
+echo "Done. Beginning benchmarks..."
+
+# Setup cache control group
+mount -t resctrl resctrl /sys/fs/resctrl
+mkdir /sys/fs/resctrl/benchmarks
+sleep 1 # Wait a second for the group to initialize
+echo $core > /sys/fs/resctrl/benchmarks/cpus_list
+
+# Execute the benchmark for each WSS and cache config
+while read j; do
+    echo $j > /sys/fs/resctrl/benchmarks/schemata
+    while read i; do
+        if grep -q "#define LITMUS 1" ../../baseline/source/extra.h; then
+            echo "Using LITMUS-RT!"
+            ./gen_input.py $benchmark $template_input $i | ./$benchmark $benchmark-$i-$j $maxJobs $core $runID 1
+        else
+            ./gen_input.py $benchmark $template_input $i | chrt -r 97 taskset -c $core ./$benchmark $benchmark-$i-$j $maxJobs $core $runID 1
+        fi
+    done < $wss_settings
+done < $cache_settings
+
+# Put IRQs back as they were
+i=0
+for IRQ in /proc/irq/*
+do
+    if [ -d $IRQ ]; then
+        echo ${irqList[$i]} > $IRQ/smp_affinity_list
+    fi
+    i=$(( $i + 1 ))
+done
+
diff --git a/dis/original/setup_mem_and_global.sh b/dis/original/setup_mem_and_global.sh
new file mode 100755
index 0000000..56d6219
--- /dev/null
+++ b/dis/original/setup_mem_and_global.sh
@@ -0,0 +1,13 @@
+# Setup group
+mkdir -p /sys/fs/resctrl/benchmarks
+echo 15 > /sys/fs/resctrl/benchmarks/cpus_list
+# Remove mem from global domain
+echo "L3:0=ffff;1=ffff;2=ffff;3=0000" > /sys/fs/resctrl/schemata
+# Remove bandwidth from global domain
+echo "MB:0=2;1=2;2=2;3=2" > /sys/fs/resctrl/schemata
+echo "MB:0=2048;1=2048;2=2048;3=2048" > /sys/fs/resctrl/benchmarks/schemata
+echo "=== Global Config ==="
+cat /sys/fs/resctrl/schemata
+echo "=== Core 15 Config ==="
+cat /sys/fs/resctrl/benchmarks/schemata
+
-- 
cgit v1.2.2