#!/usr/bin/env bash
# Shared runtime setup for CPU torch TN/MPS runs.
#
# This makes AOCL BLIS use the multithreaded library when available, which is
# required for complex64 tensordot/cgemm to actually use all cores on this host.

QIBOTN_BLIS_MT="${QIBOTN_BLIS_MT:-/home/aocc/aocl/5.2.0/aocc/lib_LP64/libblis-mt.so.5}"

export BLIS_NUM_THREADS="${BLIS_NUM_THREADS:-${OMP_NUM_THREADS:-1}}"

if [[ -f "$QIBOTN_BLIS_MT" ]]; then
  case ":${LD_PRELOAD:-}:" in
    *":$QIBOTN_BLIS_MT:"*)
      ;;
    *)
      export LD_PRELOAD="${LD_PRELOAD:+$LD_PRELOAD:}$QIBOTN_BLIS_MT"
      ;;
  esac
fi

export OMP_PROC_BIND="${OMP_PROC_BIND:-close}"
export OMP_PLACES="${OMP_PLACES:-cores}"