Add multi-platform XDP CI builds (Alveo, Edge, VE2)

ci/csr/config.defaults.sh

@@ -0,0 +1,35 @@
+# Defaults for XDP VE2 CSR board testing on XCO.
+# Override any value via environment before calling the ci/csr scripts.
+
+: "${CSR_ROOT:=/wrk/csr/snigupta/xdp_ci}"
+: "${CSR_SPEC:=/proj/testcases/xtc/HEAD/auto/ymls/config/VAI_6_3_GEN2_REGRESSION.yml}"
+: "${CSR_DAY:=sunday}"
+: "${CSR_SUITE:=VE2_XDP}"
+: "${CSR_TESTMODE:=hw}"
+: "${CSR_TQL_REPO:=/proj/testcases/xtc/HEAD/auto/tql/vaiml/}"
+: "${CSR_TA:=/proj/xbuilds/HEAD_INT_flexml_verified_vitis/installs}"
+: "${CSR_CREATE_SPRITE_RUN:=/proj/testcases/xtc/HEAD/auto/scripts/createSpriteRun/createSpriteRun.sh}"
+
+# VE2_XDP is commented out in VAI_6_2 runs.yml; it is active under eachday in 6.3.
+: "${CSR_SUPER_SUITE:=VAI_6_3_GEN2_REGRESSION}"
+
+: "${VE2_RPM_DEPLOY_SUBDIR:=cortexa72_cortexa53}"
+: "${VE2_RPM_GLOB:=xrt*.rpm}"
+: "${VE2_DNF_PACKAGES:=xrt-20261*.rpm xrt-dbg*.rpm xrt-src*.rpm}"
+
+: "${JENKINS_URL:=http://acasops:8080}"
+: "${JENKINS_JOB:=csr_submit_job}"
+: "${JENKINS_BOARD_SELECT:=VE2}"
+: "${JENKINS_COMPILE_JOBS_ONLY:=False}"
+: "${JENKINS_LNX_TA:=DEFAULT}"
+
+: "${CSR_WAIT_TIMEOUT_SEC:=14400}"
+: "${XDP_CSR_SUBMIT_MODE:=local}"
+
+# Trial phase: run one XDP design instead of the full VE2_XDP suite (~15 tests).
+# Set empty to run all tests in the merged TQL.
+: "${XDP_CSR_SINGLE_TEST:=ResNet18_MLTimeline}"
+
+# Local RDI trial: native FlexML compile avoids docker.sock on farm hosts.
+# Set to 1 for production docker compile (suite default).
+: "${XDP_CSR_USE_DOCKER:=0}"

ci/csr/generate_csr_run.sh

@@ -0,0 +1,74 @@
+#!/bin/bash
+# Generate CSR sprite files for a VE2_XDP hw run.
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+# shellcheck source=config.defaults.sh
+source "${SCRIPT_DIR}/config.defaults.sh"
+
+usage() {
+  cat <<'EOF'
+Usage: generate_csr_run.sh --sprite-dir DIR
+
+Creates CSR merged TQL and launch scripts under --sprite-dir.
+EOF
+}
+
+SPRITE_DIR=""
+
+while [ "$#" -gt 0 ]; do
+  case "$1" in
+    --sprite-dir)
+      SPRITE_DIR="$2"
+      shift 2
+      ;;
+    -h|--help)
+      usage
+      exit 0
+      ;;
+    *)
+      echo "Unknown argument: $1" >&2
+      usage >&2
+      exit 1
+      ;;
+  esac
+done
+
+if [ -z "${SPRITE_DIR}" ]; then
+  echo "Missing required --sprite-dir" >&2
+  usage >&2
+  exit 1
+fi
+
+mkdir -p "${SPRITE_DIR}/results"
+chmod a+w "${SPRITE_DIR}" "${SPRITE_DIR}/results"
+chmod a+w "$(dirname "${SPRITE_DIR}")" 2>/dev/null || true
+
+# createSpriteRunLaunch sets up RDI/XTC. Do not source g_profile here: under bash
+# nounset it fails when LD_LIBRARY_PATH is unset (common in GHA runner env).
+umask 022
+
+cd "${SPRITE_DIR}"
+
+"${CSR_CREATE_SPRITE_RUN}" \
+  --day "${CSR_DAY}" \
+  --spec "${CSR_SPEC}" \
+  --results "${SPRITE_DIR}/results" \
+  --flexml \
+  --tql-repo "${CSR_TQL_REPO}" \
+  --xtc-no-lut \
+  --testsuites "${CSR_SUITE}" \
+  --testmode "${CSR_TESTMODE}" \
+  --ta "${CSR_TA}"
+
+RDI_SCRIPT="${SPRITE_DIR}/${CSR_SUPER_SUITE}_${CSR_DAY}_rdi.sh"
+if [ ! -s "${RDI_SCRIPT}" ]; then
+  echo "CSR generation did not produce a non-empty RDI script: ${RDI_SCRIPT}" >&2
+  ls -la "${SPRITE_DIR}" >&2 || true
+  exit 1
+fi
+
+chmod -R a+w "${SPRITE_DIR}" 2>/dev/null || true
+echo "Generated CSR run in ${SPRITE_DIR}"
+ls -1 "${SPRITE_DIR}"/*"${CSR_SUITE}"*.tql
+echo "Launch script: ${RDI_SCRIPT}"

ci/csr/limit_merged_tql.sh

@@ -0,0 +1,97 @@
+#!/bin/bash
+# Limit merged CSR TQL to a single testcase directory (trial/smoke runs).
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+# shellcheck source=config.defaults.sh
+source "${SCRIPT_DIR}/config.defaults.sh"
+
+usage() {
+  cat <<'EOF'
+Usage: limit_merged_tql.sh --sprite-dir DIR [--test NAME]
+
+Environment:
+  XDP_CSR_SINGLE_TEST   Test subdirectory name under XDP_Suites/test_repo
+                        (default: ResNet18_MLTimeline). Set empty to run all tests.
+EOF
+}
+
+SPRITE_DIR=""
+SINGLE_TEST="${XDP_CSR_SINGLE_TEST:-ResNet18_MLTimeline}"
+
+while [ "$#" -gt 0 ]; do
+  case "$1" in
+    --sprite-dir)
+      SPRITE_DIR="$2"
+      shift 2
+      ;;
+    --test)
+      SINGLE_TEST="$2"
+      shift 2
+      ;;
+    -h|--help)
+      usage
+      exit 0
+      ;;
+    *)
+      echo "Unknown argument: $1" >&2
+      usage >&2
+      exit 1
+      ;;
+  esac
+done
+
+if [ -z "${SPRITE_DIR}" ]; then
+  echo "Missing required --sprite-dir" >&2
+  exit 1
+fi
+
+if [ -z "${SINGLE_TEST}" ]; then
+  echo "XDP_CSR_SINGLE_TEST unset; running full suite TQL unchanged"
+  exit 0
+fi
+
+shopt -s nullglob
+TQL_FILES=( "${SPRITE_DIR}"/*"${CSR_SUITE}"*.tql )
+shopt -u nullglob
+
+if [ "${#TQL_FILES[@]}" -eq 0 ]; then
+  echo "No merged TQL for suite ${CSR_SUITE} under ${SPRITE_DIR}" >&2
+  exit 1
+fi
+
+TQL_FILE="${TQL_FILES[0]}"
+
+python3 - <<'PY' "${TQL_FILE}" "${SINGLE_TEST}"
+import re
+import sys
+
+tql_path, single_test = sys.argv[1:3]
+text = open(tql_path, encoding="utf-8").read()
+
+# Split trailing GLOBAL APPLY blocks (e.g. XRT RPM patch) from query blocks.
+global_idx = text.find("\nGLOBAL APPLY")
+if global_idx == -1:
+    body, tail = text, ""
+else:
+    body, tail = text[:global_idx], text[global_idx:]
+
+# Each testcase query starts at a line beginning with FROM.
+parts = re.split(r"(?m)^(?=FROM )", body)
+header = parts[0] if parts and not parts[0].lstrip().startswith("FROM ") else ""
+queries = [p for p in parts if p.lstrip().startswith("FROM ")]
+
+kept = [q for q in queries if single_test in q.splitlines()[0]]
+if not kept:
+    available = [q.splitlines()[0].strip() for q in queries]
+    raise SystemExit(
+        f"No FROM block matches test '{single_test}'.\n"
+        f"Available FROM lines:\n  " + "\n  ".join(available)
+    )
+
+new_body = header + "".join(kept)
+open(tql_path, "w", encoding="utf-8").write(new_body + tail)
+print(f"Limited {tql_path} to 1 test: {single_test}")
+print(f"  kept FROM: {kept[0].splitlines()[0].strip()}")
+print(f"  removed {len(queries) - len(kept)} other FROM block(s)")
+PY

ci/csr/patch_merged_tql.sh

@@ -0,0 +1,87 @@
+#!/bin/bash
+# Patch CSR merged TQL to copy custom XRT RPMs and install them on the board.
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+# shellcheck source=config.defaults.sh
+source "${SCRIPT_DIR}/config.defaults.sh"
+
+usage() {
+  cat <<'EOF'
+Usage: patch_merged_tql.sh --sprite-dir DIR --rpm-dir DIR [--suite SUITE]
+
+Appends a GLOBAL APPLY block to the merged VE2_XDP TQL in the sprite directory.
+EOF
+}
+
+SPRITE_DIR=""
+RPM_DIR=""
+SUITE="${CSR_SUITE}"
+
+while [ "$#" -gt 0 ]; do
+  case "$1" in
+    --sprite-dir)
+      SPRITE_DIR="$2"
+      shift 2
+      ;;
+    --rpm-dir)
+      RPM_DIR="$2"
+      shift 2
+      ;;
+    --suite)
+      SUITE="$2"
+      shift 2
+      ;;
+    -h|--help)
+      usage
+      exit 0
+      ;;
+    *)
+      echo "Unknown argument: $1" >&2
+      usage >&2
+      exit 1
+      ;;
+  esac
+done
+
+if [ -z "${SPRITE_DIR}" ] || [ -z "${RPM_DIR}" ]; then
+  echo "Missing required --sprite-dir and/or --rpm-dir" >&2
+  usage >&2
+  exit 1
+fi
+
+if [ ! -d "${RPM_DIR}" ] || [ -z "$(find "${RPM_DIR}" -maxdepth 1 -name '*.rpm' -print -quit)" ]; then
+  echo "RPM directory is empty or missing: ${RPM_DIR}" >&2
+  exit 1
+fi
+
+shopt -s nullglob
+TQL_FILES=( "${SPRITE_DIR}"/*"${SUITE}"*.tql )
+shopt -u nullglob
+
+if [ "${#TQL_FILES[@]}" -eq 0 ]; then
+  echo "No merged TQL matching suite ${SUITE} under ${SPRITE_DIR}" >&2
+  ls -la "${SPRITE_DIR}" >&2 || true
+  exit 1
+fi
+
+TQL_FILE="${TQL_FILES[0]}"
+RPM_BASENAME="$(basename "${RPM_DIR}")"
+MARKER="# XDP CI: install custom XRT RPMs on Telluride before board run"
+
+if grep -qF "${MARKER}" "${TQL_FILE}"; then
+  echo "TQL already patched: ${TQL_FILE}"
+  exit 0
+fi
+
+cat >> "${TQL_FILE}" <<EOF
+
+${MARKER}
+GLOBAL APPLY * MODIFY
+    copy += ['${RPM_DIR}'],
+    tasks.board.pre_exec = 'cd ${RPM_BASENAME} && dnf --disablerepo="*" -y install ${VE2_DNF_PACKAGES}';
+EOF
+
+echo "Patched ${TQL_FILE}"
+echo "  copy += ['${RPM_DIR}']"
+echo "  tasks.board.pre_exec = cd ${RPM_BASENAME} && dnf ... ${VE2_DNF_PACKAGES}"

ci/csr/patch_merged_tql_no_docker.sh

@@ -0,0 +1,83 @@
+#!/bin/bash
+# Patch merged TQL to compile vaiml natively (no Docker) for local RDI trial runs.
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+# shellcheck source=config.defaults.sh
+source "${SCRIPT_DIR}/config.defaults.sh"
+
+usage() {
+  cat <<'EOF'
+Usage: patch_merged_tql_no_docker.sh --sprite-dir DIR [--suite SUITE]
+
+Appends a GLOBAL APPLY block that sets user.use_docker = 0 and drops the
+(vai_docker) constraint from tasks.vaiml.select_resource.
+
+Set XDP_CSR_USE_DOCKER=1 to skip this patch (production docker compile path).
+EOF
+}
+
+SPRITE_DIR=""
+SUITE="${CSR_SUITE}"
+
+while [ "$#" -gt 0 ]; do
+  case "$1" in
+    --sprite-dir)
+      SPRITE_DIR="$2"
+      shift 2
+      ;;
+    --suite)
+      SUITE="$2"
+      shift 2
+      ;;
+    -h|--help)
+      usage
+      exit 0
+      ;;
+    *)
+      echo "Unknown argument: $1" >&2
+      usage >&2
+      exit 1
+      ;;
+  esac
+done
+
+if [ -z "${SPRITE_DIR}" ]; then
+  echo "Missing required --sprite-dir" >&2
+  usage >&2
+  exit 1
+fi
+
+if [ "${XDP_CSR_USE_DOCKER:-0}" = "1" ]; then
+  echo "XDP_CSR_USE_DOCKER=1; keeping suite default docker compile (no patch)"
+  exit 0
+fi
+
+shopt -s nullglob
+TQL_FILES=( "${SPRITE_DIR}"/*"${SUITE}"*.tql )
+shopt -u nullglob
+
+if [ "${#TQL_FILES[@]}" -eq 0 ]; then
+  echo "No merged TQL matching suite ${SUITE} under ${SPRITE_DIR}" >&2
+  exit 1
+fi
+
+TQL_FILE="${TQL_FILES[0]}"
+MARKER="# XDP CI: native vaiml compile (user.use_docker = 0)"
+
+if grep -qF "${MARKER}" "${TQL_FILE}"; then
+  echo "TQL already patched for no-docker compile: ${TQL_FILE}"
+  exit 0
+fi
+
+cat >> "${TQL_FILE}" <<'EOF'
+
+# XDP CI: native vaiml compile (user.use_docker = 0)
+GLOBAL APPLY * MODIFY
+    user.use_docker = 0,
+    tasks.vaiml.select_resource = "(ostype == ubuntu2204 || ostype == ubuntu22044 || ostype == ubuntu24041)";
+EOF
+
+echo "Patched ${TQL_FILE} for native vaiml compile"
+echo "  user.use_docker = 0"
+echo "  tasks.vaiml.select_resource = (ubuntu hosts, no vai_docker tag)"