scripts to compile and run experiments

sims/vcs/scripts/compile_designs.sh

@@ -0,0 +1,19 @@
+#!/bin/bash
+
+set -e
+
+CURRENT_DIR="${PWD##*/}"
+if [[ "$CURRENT_DIR" != "vcs" ]]; then
+    echo "Error: This script must be run from chipyard/sims/vcs."
+    exit 1
+fi
+
+source ./scripts/env.sh
+
+echo -e "\nCompiling volta & ampere designs"
+make CONFIG=VirgoFP16Config
+make CONFIG=VirgoFP16Config debug
+echo -e "\nCompiling hopper & virgo designs"
+make CONFIG=VirgoHopperConfig
+make CONFIG=VirgoHopperConfig debug
+echo -e "\nCompilation completed"

sims/vcs/scripts/env.sh

@@ -0,0 +1 @@
+export KERNELS_PATH="/scratch/yrh/vxsw/kernels"

sims/vcs/scripts/pprint

@@ -0,0 +1,163 @@
+#!/usr/bin/env python3
+
+import sys
+import signal
+import re
+import time
+
+PRINT_BUF = 0x20000 / 4
+
+def translator(line):
+    if 'core-req-wr' in line:
+        # Check rs1_data's last element condition
+        rs1_data_start = line.find('addr={') + len('addr={')
+        rs1_data_end = line.find('}', rs1_data_start)
+        rs1_data_elts = line[rs1_data_start:rs1_data_end].split(', ')
+
+        byteen_start = line.find('byteen={') + len('byteen={')
+        byteen_end = line.find('}', byteen_start)
+        byteen_elts = line[byteen_start:byteen_end].split(', ')
+
+        rs2_data_start = line.find('data={') + len('data={')
+        rs2_data_end = line.find('}', rs2_data_start)
+        rs2_data_elts = line[rs2_data_start:rs2_data_end].split(', ')
+
+        # print(rs1_data_last_element)
+        for rs1, rs2, byteen in zip(rs1_data_elts, rs2_data_elts, byteen_elts):
+            if int(rs1, 16) >> 18 == 0xff0:
+                offset = (int(rs1, 16) - PRINT_BUF) % 65536
+                if offset < 0 or offset >= 1024:
+                    continue
+                else:
+                    offset = offset % 16384
+                # Extract rs2_data's last element
+
+                hex_value = rs2[2:]  # Remove the '0x' prefix
+                if "x" in hex_value:
+                    continue
+                byteen_int = int(byteen, 16)
+                hex_value = "0" * (8 - len(hex_value)) + hex_value
+                bytes_object = bytes.fromhex(hex_value) # .replace(b"\x00", b"")
+
+                masked_bytes_list = []
+                    
+                assert(len(bytes_object) == 4)
+                
+                for i, byte in enumerate(bytes_object[::-1]):
+                    if byteen_int & (1 << i):
+                        masked_bytes_list.append(byte)
+
+                reversed_bytes = bytes(masked_bytes_list)
+                # print(reversed_bytes.decode('utf-8', errors="ignore"))
+                try:
+                    return reversed_bytes.decode('ascii', errors="ignore")
+                except UnicodeDecodeError:
+                    return ""
+
+def timestamp_parser(line):
+    match = re.match(r"^\s*(\d+):", line)
+    if match:
+        return match.group(1)
+    else:
+        return ""
+
+sim_started = False
+sim_ended = False
+
+def signal_handler(sig, frame):
+    if sim_started:
+        print("\033[B")
+    sys.exit(0)
+
+def main():
+    signal.signal(signal.SIGINT, signal_handler)
+
+    curr_timestamp = -1
+    prev_timestamp = -1
+    curr_clock = time.time()
+    prev_clock = time.time()
+    re_start_num = re.compile(r"^\s*[0-9]+:")
+
+    ts_countdown = 100
+    global sim_started, sim_ended
+    perf_counters = False
+    hang_detector = 0
+
+    # if (len(sys.argv) > 1) and (sys.argv[1] == "started"):
+    #     sim_started = True
+    run_label = sys.argv[1]
+    lineno = int(sys.argv[2])
+
+    print("\033[2J\033[H")
+
+    for line in sys.stdin:
+        line = line.rstrip('\n')
+
+        if "Chronologic VCS simulator" in line:
+            sim_started = True
+
+        sim_nontrace = re.match(re_start_num, line) is None
+
+        if "====================CORE" in line:
+            perf_counters = True
+            if hang_detector >= 8:
+                pass
+                # print("\n\033[3mpossible hang detected\033[0m\n")
+
+        if "has no more active warps" in line:
+            sim_ended = True
+
+        if (not sim_started):
+            print(line)
+            continue
+        
+        if sim_ended:
+            if "has no more active warps" not in line:
+                sim_ended = False
+            else:
+                continue
+        
+        if sim_started and (not sim_ended):
+            if sim_nontrace:
+                if not perf_counters:
+                    print(line)
+                elif line.startswith("dcache stores:"):
+                    perf_counters = False
+                    hang_detector += 1
+                continue
+            else:
+                hang_detector = 0
+
+        if ts_countdown == 0:
+            match = re.match(r"^\s*(\d+):", line)
+            if match:
+                prev_clock = curr_clock
+                prev_timestamp = curr_timestamp
+                curr_clock = time.time()
+                curr_timestamp = match.group(1)
+                speed_in_hz = (int(curr_timestamp) - int(prev_timestamp)) / (curr_clock - prev_clock) if curr_clock - prev_clock > 0 else 0
+
+                # Save cursor position
+                print("\033[s", end='')
+                # Move cursor to top-left corner, clear line, bold, timestamp, unbold
+                print("\033[" + str(lineno) + "H\033[2K\033[1m" + run_label,
+                      "[TIME]", curr_timestamp, "[SPEED]", int(speed_in_hz), "\033[0m", end='')
+                # Restore cursor position
+                print("\033[u", end='', flush=True)
+
+                ts_countdown = 200
+            else:
+                ts_countdown = 10
+
+        ts_countdown -= 1
+        translated_line = translator(line)
+
+        if translated_line:
+            print(translated_line, end='')
+            sys.stdout.flush()
+
+    print("")
+
+if __name__ == '__main__':
+    main()
+

sims/vcs/scripts/run_sims.sh

@@ -0,0 +1,61 @@
+#!/bin/bash
+
+set -e
+
+CURRENT_DIR="${PWD##*/}"
+if [[ "$CURRENT_DIR" != "vcs" ]]; then
+    echo "Error: This script must be run from chipyard/sims/vcs."
+    exit 1
+fi
+
+if [[ -z "${TMUX}" ]]; then
+    echo "Error: you must be in a tmux session to run simulations."
+fi
+
+source ./scripts/env.sh
+
+lineno=0
+
+start_run() {
+    echo "kickoff elf $KERNELS_PATH/$2 on config $1"
+    lineno=$((lineno+1))
+    make CONFIG=$1 BINARY="$KERNELS_PATH/$2" LOADMEM=1 run-binary"$4" 2>&1 | ./scripts/pprint "$3" $lineno &
+}
+
+check_exists() {
+    if ! [ -f "$1" ]; then
+        echo "Error: looked for file $1 that does not exist."
+        exit 1
+    fi
+}
+
+check_exists "simv-chipyard.harness-VirgoFP16Config"
+check_exists "simv-chipyard.harness-VirgoHopperConfig"
+check_exists "simv-chipyard.harness-VirgoFP16Config-debug"
+check_exists "simv-chipyard.harness-VirgoHopperConfig-debug"
+
+# sanity check that the kernels have been compiled
+check_exists "$KERNELS_PATH/sgemm_tcore/kernel.radiance.gemm.tcore.volta.dim256.elf"
+check_exists "$KERNELS_PATH/sgemm_tcore/kernel.radiance.gemm.tcore.hopper.dim512.elf"
+check_exists "$KERNELS_PATH/sgemm_gemmini_dma/kernel.radiance.gemm.virgo.hopper.dim1024.elf"
+
+echo "Simulations will be started in parallel in 5 seconds. Please do not Ctrl+C as it kills all subprocesses."
+
+sleep 5
+
+suffix="-debug"
+
+# dims=(256 512 1024)
+dims=(256 512)
+for dim in "${dims[@]}"; do
+    echo "$element"
+    start_run VirgoFP16Config sgemm_tcore/kernel.radiance.gemm.tcore.volta.dim${dim}.elf          "volta${dim} " "${suffix}"
+    start_run VirgoFP16Config sgemm_tcore/kernel.radiance.gemm.tcore.ampere.dim${dim}.elf         "ampere${dim}" "${suffix}"
+    start_run VirgoHopperConfig sgemm_tcore/kernel.radiance.gemm.tcore.hopper.dim${dim}.elf       "hopper${dim}" "${suffix}"
+    start_run VirgoHopperConfig sgemm_gemmini_dma/kernel.radiance.gemm.virgo.hopper.dim${dim}.elf "virgo${dim} " "${suffix}"
+    suffix=""
+done
+
+wait
+
+echo "All simulations have finished!"

sims/vcs/scripts/sram_util.sh

@@ -0,0 +1,42 @@
+#!/bin/bash
+
+# Define file names and corresponding n values
+declare -A files_and_n_values
+files_and_n_values=(
+    ["output/chipyard.harness.TestHarness.VirgoFP16Config/kernel.radiance.gemm.tcore.volta.dim256.fsdb"]=256
+    ["output/chipyard.harness.TestHarness.VirgoFP16Config/kernel.radiance.gemm.tcore.volta.dim512.fsdb"]=512
+    ["output/chipyard.harness.TestHarness.VirgoFP16Config/kernel.radiance.gemm.tcore.volta.dim1024.fsdb"]=1024
+    ["output/chipyard.harness.TestHarness.VirgoFP16Config/kernel.radiance.gemm.tcore.ampere.dim256.fsdb"]=256
+    ["output/chipyard.harness.TestHarness.VirgoFP16Config/kernel.radiance.gemm.tcore.ampere.dim512.fsdb"]=512
+    ["output/chipyard.harness.TestHarness.VirgoFP16Config/kernel.radiance.gemm.tcore.ampere.dim1024.fsdb"]=1024
+    ["output/chipyard.harness.TestHarness.VirgoHopperConfig/kernel.radiance.gemm.tcore.hopper.dim256.fsdb"]=256
+    ["output/chipyard.harness.TestHarness.VirgoHopperConfig/kernel.radiance.gemm.tcore.hopper.dim512.fsdb"]=512
+    ["output/chipyard.harness.TestHarness.VirgoHopperConfig/kernel.radiance.gemm.tcore.hopper.dim1024.fsdb"]=1024
+    ["output/chipyard.harness.TestHarness.VirgoHopperConfig/kernel.radiance.gemm.virgo.hopper.dim256.fsdb"]=256
+    ["output/chipyard.harness.TestHarness.VirgoHopperConfig/kernel.radiance.gemm.virgo.hopper.dim512.fsdb"]=512
+    ["output/chipyard.harness.TestHarness.VirgoHopperConfig/kernel.radiance.gemm.virgo.hopper.dim1024.fsdb"]=1024
+)
+
+for fsdb_file in "${!files_and_n_values[@]}"; do
+    n=${files_and_n_values[$fsdb_file]}
+
+    echo "parsing sharedmem reads for file $fsdb_file"
+
+    # Run fsdbreport command
+    fsdbreport "$fsdb_file" -s "/TestDriver/testHarness/chiptop0/system/cluster_prci_domain/element_reset_domain_element/shared_mem/smemReadCounter" -of d -nolog -o /tmp/smem_activity.log
+
+    # Extract last line and parse the second number
+    last_line=$(tail -n 1 /tmp/smem_activity.log)
+    reads=$(echo "$last_line" | awk '{print $2}')
+
+    # Clean up temp file
+    rm -f /tmp/smem_activity.log
+
+    echo "reads: $reads"
+
+    # Calculate final value
+    result=$(echo "scale=6; $reads / ($n * $n / 64)" | bc)
+
+    echo "multiple of input data size: $result"
+done
+