|
| 1 | +#!/bin/bash |
| 2 | +# CK Docker Skill - Build and test composable_kernel in Docker with ROCm support |
| 3 | + |
| 4 | +set -e |
| 5 | + |
| 6 | +# Find project root (where .git directory is) |
| 7 | +SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)" |
| 8 | +PROJECT_ROOT="$(cd "${SCRIPT_DIR}/../.." && pwd)" |
| 9 | + |
| 10 | +# Detect git branch and sanitize for docker naming (replace / and special chars with _) |
| 11 | +GIT_BRANCH=$(cd "${PROJECT_ROOT}" && git rev-parse --abbrev-ref HEAD 2>/dev/null | tr '/' '_' | tr -cd 'a-zA-Z0-9_-') |
| 12 | + |
| 13 | +# Default container name: ck_<username>_<branch> |
| 14 | +DEFAULT_NAME="ck_${USER}_${GIT_BRANCH}" |
| 15 | +CONTAINER_NAME="${CK_CONTAINER_NAME:-${DEFAULT_NAME}}" |
| 16 | + |
| 17 | +# Help message |
| 18 | +show_help() { |
| 19 | + cat << EOF |
| 20 | +CK Docker Skill - Build and test composable_kernel in Docker |
| 21 | +
|
| 22 | +Usage: ck-docker <command> [options] |
| 23 | +
|
| 24 | +Commands: |
| 25 | + start [name] Start Docker container |
| 26 | + build [target] [--name] Build target |
| 27 | + test <test> [options] Run test |
| 28 | + shell [name] Open shell in container |
| 29 | + status [name] Check container status |
| 30 | + stop [name] Stop and remove container |
| 31 | + rebuild-cmake [name] Reconfigure CMake from scratch |
| 32 | +
|
| 33 | +Examples: |
| 34 | + ck-docker start |
| 35 | + ck-docker build test_amdgcn_mma |
| 36 | + ck-docker test test_amdgcn_mma --gtest_filter=*Fp16* |
| 37 | + ck-docker shell |
| 38 | +
|
| 39 | +Environment: |
| 40 | + CK_CONTAINER_NAME - Override default container name (default: ck_<username>_<branch>) |
| 41 | +EOF |
| 42 | +} |
| 43 | + |
| 44 | +# Detect GPU target |
| 45 | +detect_gpu() { |
| 46 | + local container=$1 |
| 47 | + docker exec ${container} bash -c " |
| 48 | + rocminfo 2>/dev/null | grep -oP 'gfx[0-9a-z]+' | head -1 || echo 'gfx950' |
| 49 | + " | tr -d '\r\n' |
| 50 | +} |
| 51 | + |
| 52 | +# Start container |
| 53 | +cmd_start() { |
| 54 | + local name="${1:-${CONTAINER_NAME}}" |
| 55 | + |
| 56 | + if docker ps -a -f name=${name} | grep -q ${name}; then |
| 57 | + if docker ps -f name=${name} | grep -q ${name}; then |
| 58 | + echo "Container '${name}' is already running" |
| 59 | + return 0 |
| 60 | + else |
| 61 | + echo "Starting existing container '${name}'..." |
| 62 | + docker start ${name} |
| 63 | + echo "Container started" |
| 64 | + return 0 |
| 65 | + fi |
| 66 | + fi |
| 67 | + |
| 68 | + echo "Creating new Docker container '${name}'..." |
| 69 | + docker run -d \ |
| 70 | + --name ${name} \ |
| 71 | + --device=/dev/kfd --device=/dev/dri \ |
| 72 | + --security-opt seccomp=unconfined \ |
| 73 | + --group-add video \ |
| 74 | + -v "${PROJECT_ROOT}":/workspace \ |
| 75 | + -w /workspace \ |
| 76 | + rocm/composable_kernel:ck_ub24.04_rocm7.0.1 \ |
| 77 | + tail -f /dev/null |
| 78 | + |
| 79 | + echo "Container '${name}' started successfully" |
| 80 | + docker exec ${name} bash -c "echo 'Working directory:' && pwd" |
| 81 | +} |
| 82 | + |
| 83 | +# Build target |
| 84 | +cmd_build() { |
| 85 | + local target="" |
| 86 | + local name="${CONTAINER_NAME}" |
| 87 | + |
| 88 | + while [[ $# -gt 0 ]]; do |
| 89 | + case $1 in |
| 90 | + --name) |
| 91 | + name="$2" |
| 92 | + shift 2 |
| 93 | + ;; |
| 94 | + *) |
| 95 | + target="$1" |
| 96 | + shift |
| 97 | + ;; |
| 98 | + esac |
| 99 | + done |
| 100 | + |
| 101 | + if ! docker ps -f name=${name} | grep -q ${name}; then |
| 102 | + echo "Container '${name}' not running. Starting..." |
| 103 | + cmd_start ${name} |
| 104 | + fi |
| 105 | + |
| 106 | + if ! docker exec ${name} test -f /workspace/build/build.ninja 2>/dev/null; then |
| 107 | + echo "Detecting GPU target..." |
| 108 | + local gpu_target=$(detect_gpu ${name}) |
| 109 | + |
| 110 | + echo "Configuring build with CMake for GPU target: ${gpu_target}" |
| 111 | + docker exec ${name} bash -c " |
| 112 | + cd /workspace && |
| 113 | + rm -rf build && |
| 114 | + mkdir build && |
| 115 | + cd build && |
| 116 | + cmake .. -GNinja \ |
| 117 | + -DGPU_TARGETS=${gpu_target} \ |
| 118 | + -DCMAKE_BUILD_TYPE=Release \ |
| 119 | + -DCMAKE_CXX_COMPILER=/opt/rocm/llvm/bin/clang++ \ |
| 120 | + -DBUILD_TESTING=ON 2>&1 | tail -30 |
| 121 | + " |
| 122 | + fi |
| 123 | + |
| 124 | + if [ -z "$target" ]; then |
| 125 | + echo "Building all configured targets..." |
| 126 | + else |
| 127 | + echo "Building target: ${target}" |
| 128 | + fi |
| 129 | + |
| 130 | + docker exec ${name} bash -c " |
| 131 | + cd /workspace/build && |
| 132 | + ninja ${target} 2>&1 |
| 133 | + " |
| 134 | + |
| 135 | + echo "Build complete" |
| 136 | +} |
| 137 | + |
| 138 | +# Run test |
| 139 | +cmd_test() { |
| 140 | + local test_name="" |
| 141 | + local name="${CONTAINER_NAME}" |
| 142 | + local test_options="" |
| 143 | + |
| 144 | + while [[ $# -gt 0 ]]; do |
| 145 | + case $1 in |
| 146 | + --name) |
| 147 | + name="$2" |
| 148 | + shift 2 |
| 149 | + ;; |
| 150 | + --gtest_*|--help) |
| 151 | + test_options="${test_options} $1" |
| 152 | + shift |
| 153 | + ;; |
| 154 | + *) |
| 155 | + if [ -z "$test_name" ]; then |
| 156 | + test_name="$1" |
| 157 | + else |
| 158 | + test_options="${test_options} $1" |
| 159 | + fi |
| 160 | + shift |
| 161 | + ;; |
| 162 | + esac |
| 163 | + done |
| 164 | + |
| 165 | + if [ -z "$test_name" ]; then |
| 166 | + echo "Error: test_name required" |
| 167 | + echo "Usage: ck-docker test <test_name> [--name container_name] [gtest_options]" |
| 168 | + return 1 |
| 169 | + fi |
| 170 | + |
| 171 | + if ! docker ps -f name=${name} | grep -q ${name}; then |
| 172 | + echo "Error: Container '${name}' not running" |
| 173 | + echo "Start it with: ck-docker start --name ${name}" |
| 174 | + return 1 |
| 175 | + fi |
| 176 | + |
| 177 | + if ! docker exec ${name} test -f "/workspace/build/bin/${test_name}" 2>/dev/null; then |
| 178 | + echo "Test executable not found. Building ${test_name}..." |
| 179 | + cmd_build ${test_name} --name ${name} |
| 180 | + fi |
| 181 | + |
| 182 | + echo "Running: ${test_name} ${test_options}" |
| 183 | + echo "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━" |
| 184 | + docker exec ${name} bash -c " |
| 185 | + cd /workspace/build && |
| 186 | + ./bin/${test_name} ${test_options} |
| 187 | + " |
| 188 | +} |
| 189 | + |
| 190 | +# Shell |
| 191 | +cmd_shell() { |
| 192 | + local name="${1:-${CONTAINER_NAME}}" |
| 193 | + |
| 194 | + if ! docker ps -f name=${name} | grep -q ${name}; then |
| 195 | + echo "Container '${name}' not running. Starting..." |
| 196 | + cmd_start ${name} |
| 197 | + fi |
| 198 | + |
| 199 | + echo "Opening shell in '${name}' (type 'exit' to leave)..." |
| 200 | + docker exec -it ${name} bash |
| 201 | +} |
| 202 | + |
| 203 | +# Status |
| 204 | +cmd_status() { |
| 205 | + local name="${1:-}" |
| 206 | + |
| 207 | + if [ -z "$name" ]; then |
| 208 | + echo "Composable Kernel Docker Containers:" |
| 209 | + echo "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━" |
| 210 | + docker ps -a --filter "ancestor=rocm/composable_kernel:ck_ub24.04_rocm7.0.1" \ |
| 211 | + --format "table {{.Names}}\t{{.Status}}\t{{.CreatedAt}}" || echo "No containers found" |
| 212 | + else |
| 213 | + if docker ps -f name=${name} | grep -q ${name}; then |
| 214 | + echo "Container '${name}' is RUNNING" |
| 215 | + docker ps -f name=${name} --format "table {{.Names}}\t{{.Status}}\t{{.Image}}" |
| 216 | + echo "" |
| 217 | + echo "GPU Information:" |
| 218 | + docker exec ${name} bash -c "rocm-smi --showproductname 2>/dev/null | head -10 || echo 'No GPU detected'" |
| 219 | + elif docker ps -a -f name=${name} | grep -q ${name}; then |
| 220 | + echo "Container '${name}' exists but is STOPPED" |
| 221 | + echo "Start with: ck-docker start ${name}" |
| 222 | + else |
| 223 | + echo "Container '${name}' does NOT exist" |
| 224 | + echo "Create with: ck-docker start ${name}" |
| 225 | + fi |
| 226 | + fi |
| 227 | +} |
| 228 | + |
| 229 | +# Stop |
| 230 | +cmd_stop() { |
| 231 | + local name="${1:-${CONTAINER_NAME}}" |
| 232 | + |
| 233 | + if docker ps -a -f name=${name} | grep -q ${name}; then |
| 234 | + echo "Stopping and removing container '${name}'..." |
| 235 | + docker stop ${name} 2>/dev/null || true |
| 236 | + docker rm ${name} 2>/dev/null || true |
| 237 | + echo "Container stopped and removed" |
| 238 | + else |
| 239 | + echo "Container '${name}' does not exist" |
| 240 | + fi |
| 241 | +} |
| 242 | + |
| 243 | +# Rebuild CMake |
| 244 | +cmd_rebuild_cmake() { |
| 245 | + local name="${1:-${CONTAINER_NAME}}" |
| 246 | + |
| 247 | + if ! docker ps -f name=${name} | grep -q ${name}; then |
| 248 | + echo "Container '${name}' not running. Starting..." |
| 249 | + cmd_start ${name} |
| 250 | + fi |
| 251 | + |
| 252 | + echo "Detecting GPU target..." |
| 253 | + local gpu_target=$(detect_gpu ${name}) |
| 254 | + |
| 255 | + echo "Reconfiguring CMake from scratch in '${name}' for GPU target: ${gpu_target}" |
| 256 | + docker exec ${name} bash -c " |
| 257 | + cd /workspace && |
| 258 | + rm -rf build && |
| 259 | + mkdir build && |
| 260 | + cd build && |
| 261 | + cmake .. -GNinja \ |
| 262 | + -DGPU_TARGETS=${gpu_target} \ |
| 263 | + -DCMAKE_BUILD_TYPE=Release \ |
| 264 | + -DCMAKE_CXX_COMPILER=/opt/rocm/llvm/bin/clang++ \ |
| 265 | + -DBUILD_TESTING=ON 2>&1 | tail -30 |
| 266 | + " |
| 267 | + echo "CMake configuration complete for ${gpu_target}" |
| 268 | +} |
| 269 | + |
| 270 | +# Main command dispatcher |
| 271 | +case "${1:-}" in |
| 272 | + start) |
| 273 | + shift |
| 274 | + cmd_start "$@" |
| 275 | + ;; |
| 276 | + build) |
| 277 | + shift |
| 278 | + cmd_build "$@" |
| 279 | + ;; |
| 280 | + test) |
| 281 | + shift |
| 282 | + cmd_test "$@" |
| 283 | + ;; |
| 284 | + shell) |
| 285 | + shift |
| 286 | + cmd_shell "$@" |
| 287 | + ;; |
| 288 | + status) |
| 289 | + shift |
| 290 | + cmd_status "$@" |
| 291 | + ;; |
| 292 | + stop) |
| 293 | + shift |
| 294 | + cmd_stop "$@" |
| 295 | + ;; |
| 296 | + rebuild-cmake) |
| 297 | + shift |
| 298 | + cmd_rebuild_cmake "$@" |
| 299 | + ;; |
| 300 | + help|--help|-h) |
| 301 | + show_help |
| 302 | + ;; |
| 303 | + *) |
| 304 | + echo "Unknown command: ${1:-}" |
| 305 | + echo "" |
| 306 | + show_help |
| 307 | + exit 1 |
| 308 | + ;; |
| 309 | +esac |
0 commit comments