Work around CCE 19.0.0 compiler bugs for Cray+OpenACC builds (#1286)

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-authored-by: Spencer Bryngelson <sbryngelson@login10.frontier.olcf.ornl.gov>

Author: 3 people

.github/scripts/run_monitored_slurm_job.sh

@@ -0,0 +1,37 @@
+#!/bin/bash
+# Run monitor_slurm_job.sh and recover if the monitor is killed (e.g. SIGKILL
+# from the runner OS) before the SLURM job completes.  When the monitor exits
+# non-zero, sacct is used to verify the job's actual final state; if the SLURM
+# job succeeded we exit 0 so the CI step is not falsely marked as failed.
+#
+# Usage: run_monitored_slurm_job.sh <job_id> <output_file>
+
+set -euo pipefail
+
+if [ $# -ne 2 ]; then
+    echo "Usage: $0 <job_id> <output_file>"
+    exit 1
+fi
+
+job_id="$1"
+output_file="$2"
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+
+monitor_exit=0
+bash "$SCRIPT_DIR/monitor_slurm_job.sh" "$job_id" "$output_file" || monitor_exit=$?
+
+if [ "$monitor_exit" -ne 0 ]; then
+    echo "Monitor exited with code $monitor_exit; re-checking SLURM job $job_id final state..."
+    # Give the SLURM epilog time to finalize if the job just finished
+    sleep 30
+    final_state=$(sacct -j "$job_id" -n -X -P -o State 2>/dev/null | head -n1 | cut -d'|' -f1 | tr -d ' ' || echo "UNKNOWN")
+    final_exit=$(sacct -j "$job_id" --format=ExitCode --noheader --parsable2 2>/dev/null | head -n1 | tr -d ' ' || echo "")
+    echo "Final SLURM state=$final_state exit=$final_exit"
+    if [ "$final_state" = "COMPLETED" ] && [ "$final_exit" = "0:0" ]; then
+        echo "SLURM job $job_id completed successfully despite monitor failure — continuing."
+    else
+        echo "ERROR: SLURM job $job_id did not complete successfully (state=$final_state exit=$final_exit)"
+        exit 1
+    fi
+fi

.github/workflows/frontier/build.sh

@@ -20,10 +20,7 @@ build_opts="$gpu_opts"
 
 . ./mfc.sh load -c $compiler_flag -m $([ "$job_device" = "gpu" ] && echo "g" || echo "c")
 
-# Only set up build cache for test suite, not benchmarks
-if [ "$run_bench" != "bench" ]; then
-    source .github/scripts/setup-build-cache.sh "$cluster_name" "$job_device" "$job_interface"
-fi
+rm -rf build
 
 source .github/scripts/retry-build.sh
 if [ "$run_bench" == "bench" ]; then

.github/workflows/frontier/submit.sh

@@ -45,10 +45,10 @@ fi
 
 # Select SBATCH params based on job type
 if [ "$job_type" = "bench" ]; then
-    sbatch_account="#SBATCH -A ENG160"
-    sbatch_time="#SBATCH -t 05:59:00"
-    sbatch_partition="#SBATCH -p extended"
-    sbatch_extra=""
+    sbatch_account="#SBATCH -A CFD154"
+    sbatch_time="#SBATCH -t 01:59:00"
+    sbatch_partition="#SBATCH -p batch"
+    sbatch_extra="#SBATCH --qos=normal"
 else
     sbatch_account="#SBATCH -A CFD154"
     sbatch_time="#SBATCH -t 01:59:00"
@@ -102,5 +102,4 @@ fi
 
 echo "Submitted batch job $job_id"
 
-# Use resilient monitoring instead of sbatch -W
-bash "$SCRIPT_DIR/../../scripts/monitor_slurm_job.sh" "$job_id" "$output_file"
+bash "$SCRIPT_DIR/../../scripts/run_monitored_slurm_job.sh" "$job_id" "$output_file"

.github/workflows/phoenix/bench.sh

@@ -2,7 +2,7 @@
 
 source .github/scripts/bench-preamble.sh
 
-tmpbuild=/storage/scratch1/6/sbryngelson3/mytmp_build
+tmpbuild=/storage/project/r-sbryngelson3-0/sbryngelson3/mytmp_build
 currentdir=$tmpbuild/run-$(( RANDOM % 900 ))
 mkdir -p $tmpbuild
 mkdir -p $currentdir
@@ -15,6 +15,8 @@ else
     bench_opts="--mem 1"
 fi
 
+rm -rf build
+
 source .github/scripts/retry-build.sh
 RETRY_CLEAN_CMD="./mfc.sh clean" retry_build ./mfc.sh build -j $(nproc) $build_opts || exit 1
 

.github/workflows/phoenix/submit.sh

@@ -94,6 +94,5 @@ fi
 
 echo "Submitted batch job $job_id"
 
-# Use resilient monitoring instead of sbatch -W
 SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
-bash "$SCRIPT_DIR/../../scripts/monitor_slurm_job.sh" "$job_id" "$output_file"
+bash "$SCRIPT_DIR/../../scripts/run_monitored_slurm_job.sh" "$job_id" "$output_file"

.github/workflows/phoenix/test.sh

@@ -3,8 +3,7 @@
 source .github/scripts/gpu-opts.sh
 build_opts="$gpu_opts"
 
-# Set up persistent build cache
-source .github/scripts/setup-build-cache.sh phoenix "$job_device" "$job_interface"
+rm -rf build
 
 # Build with retry; smoke-test cached binaries to catch architecture mismatches
 # (SIGILL from binaries compiled on a different compute node).

.github/workflows/test.yml

@@ -98,6 +98,12 @@ jobs:
       - name: Clone
         uses: actions/checkout@v4
 
+      - name: Restore Build Cache
+        uses: actions/cache@v4
+        with:
+          path: build
+          key: mfc-build-${{ matrix.os }}-${{ matrix.mpi }}-${{ matrix.debug }}-${{ matrix.precision }}-${{ matrix.intel }}-${{ hashFiles('CMakeLists.txt', 'toolchain/dependencies/**', 'toolchain/cmake/**', 'src/**/*.fpp', 'src/**/*.f90') }}
+
       - name: Setup MacOS
         if:   matrix.os == 'macos'
         run:  |
@@ -131,32 +137,20 @@ jobs:
           printenv | sort > /tmp/env_after
           diff /tmp/env_before /tmp/env_after | grep '^>' | sed 's/^> //' >> $GITHUB_ENV
 
-      - name: Get system info for cache key
-        id: sys-info
-        run: |
-          {
-            uname -m
-            cat /proc/cpuinfo 2>/dev/null | grep 'model name' | head -1 || sysctl -n machdep.cpu.brand_string 2>/dev/null || true
-            if command -v ifx &>/dev/null; then ifx --version 2>/dev/null | head -1; else ${FC:-gfortran} --version 2>/dev/null | head -1 || true; fi
-            ${CC:-gcc} --version 2>/dev/null | head -1 || true
-          } | (sha256sum 2>/dev/null || shasum -a 256) | cut -c1-16 > /tmp/sys-hash
-          echo "sys-hash=$(cat /tmp/sys-hash)" >> "$GITHUB_OUTPUT"
-
-      - name: Restore Build Cache
-        uses: actions/cache@v4
+      - name: Set up Python 3.14
+        uses: actions/setup-python@v5
         with:
-          path: build
-          key: mfc-build-${{ matrix.os }}-${{ matrix.mpi }}-${{ matrix.debug }}-${{ matrix.precision }}-${{ matrix.intel }}-${{ steps.sys-info.outputs.sys-hash }}-${{ hashFiles('CMakeLists.txt', 'toolchain/dependencies/**', 'toolchain/cmake/**', 'src/**/*.fpp', 'src/**/*.f90') }}
+          python-version: '3.14'
 
       - name: Build
         run:  |
-          /bin/bash mfc.sh test -v --dry-run -j "$(nproc)" --${{ matrix.debug }} --${{ matrix.mpi }} $PRECISION $TEST_ALL
+          /bin/bash mfc.sh test -v --dry-run -j $(nproc) --${{ matrix.debug }} --${{ matrix.mpi }} --${{ matrix.precision }} $TEST_ALL
         env:
           TEST_ALL: ${{ matrix.mpi == 'mpi' && '--test-all' || '' }}
-          PRECISION: ${{ matrix.precision != '' && format('--{0}', matrix.precision) || '' }}
 
       - name: Test
-        run: bash .github/scripts/run-tests-with-retry.sh -v --max-attempts 3 -j "$(nproc)" $TEST_ALL $TEST_PCT
+        run:  |
+          /bin/bash mfc.sh test -v --max-attempts 3 -j $(nproc) $TEST_ALL $TEST_PCT
         env:
           TEST_ALL: ${{ matrix.mpi == 'mpi' && '--test-all' || '' }}
           TEST_PCT: ${{ matrix.debug == 'debug' && '-% 20' || '' }}
@@ -186,7 +180,7 @@ jobs:
             cluster_name: 'Georgia Tech | Phoenix'
             device: 'cpu'
             interface: 'none'
-          # Frontier (ORNL) — build on login node, GPU tests sharded for batch partition
+          # Frontier (ORNL) — CCE
           - runner:       'frontier'
             cluster:      'frontier'
             cluster_name: 'Oak Ridge | Frontier'
@@ -243,21 +237,30 @@ jobs:
       - name: Clone
         uses: actions/checkout@v4
         with:
-          clean: false
+          clean: true
 
       - name: Build
         if:   matrix.cluster != 'phoenix'
         uses: nick-fields/retry@ce71cc2ab81d554ebbe88c79ab5975992d79ba08 # v3
         with:
-          max_attempts: 3
+          max_attempts: 2
           retry_wait_seconds: 60
           timeout_minutes: 60
           command: bash .github/workflows/${{ matrix.cluster }}/build.sh ${{ matrix.device }} ${{ matrix.interface }}
-          on_retry_command: ./mfc.sh clean
+          on_retry_command: rm -rf build
 
       - name: Test
         run:  bash .github/workflows/${{ matrix.cluster }}/submit.sh .github/workflows/${{ matrix.cluster }}/test.sh ${{ matrix.device }} ${{ matrix.interface }} ${{ matrix.shard }}
 
+      - name: Cancel SLURM Jobs
+        if: cancelled()
+        run: |
+          find . -name "*.slurm_job_id" | while read -r f; do
+            job_id=$(cat "$f")
+            echo "Cancelling SLURM job $job_id"
+            scancel "$job_id" 2>/dev/null || true
+          done
+
       - name: Compute Log Slug
         if:   always()
         id:   log
@@ -321,25 +324,28 @@ jobs:
       - name: Clone
         uses: actions/checkout@v4
         with:
-          clean: false
+          clean: true
 
       - name: Pre-Build (SLURM)
         if:   matrix.cluster == 'phoenix'
         run:  bash .github/workflows/phoenix/submit.sh .github/scripts/prebuild-case-optimization.sh ${{ matrix.device }} ${{ matrix.interface }}
 
       - name: Pre-Build (login node)
         if:   matrix.cluster != 'phoenix'
-        uses: nick-fields/retry@ce71cc2ab81d554ebbe88c79ab5975992d79ba08 # v3
-        with:
-          max_attempts: 3
-          retry_wait_seconds: 60
-          timeout_minutes: 120
-          command: bash .github/scripts/prebuild-case-optimization.sh ${{ matrix.cluster }} ${{ matrix.device }} ${{ matrix.interface }}
-          on_retry_command: ./mfc.sh clean
+        run:  bash .github/scripts/prebuild-case-optimization.sh ${{ matrix.cluster }} ${{ matrix.device }} ${{ matrix.interface }}
 
       - name: Run Case-Optimization Tests
         run:  bash .github/workflows/${{ matrix.cluster }}/submit.sh .github/scripts/run_case_optimization.sh ${{ matrix.device }} ${{ matrix.interface }}
 
+      - name: Cancel SLURM Jobs
+        if: cancelled()
+        run: |
+          find . -name "*.slurm_job_id" | while read -r f; do
+            job_id=$(cat "$f")
+            echo "Cancelling SLURM job $job_id"
+            scancel "$job_id" 2>/dev/null || true
+          done
+
       - name: Print Logs
         if:   always()
         run:  |

.gitignore

@@ -105,4 +105,8 @@ benchmarks/*.png
 *.avi
 
 **isolation_rules/
-**.supercode/
+**.supercode/
+# CCE stress-test log directories (local testing artifacts)
+cce_*/
+cce_*.log
+run_cce_*.sh

CMakeLists.txt

@@ -224,13 +224,24 @@ endif()
 
 if (CMAKE_BUILD_TYPE STREQUAL "Release")
     # Processor tuning: Check if we can target the host's native CPU's ISA.
-    CHECK_FORTRAN_COMPILER_FLAG("-march=native" SUPPORTS_MARCH_NATIVE)
-    if (SUPPORTS_MARCH_NATIVE)
-        add_compile_options($<$<COMPILE_LANGUAGE:Fortran>:-march=native>)
-    else()
-    	CHECK_FORTRAN_COMPILER_FLAG("-mcpu=native" SUPPORTS_MCPU_NATIVE)
-        if (SUPPORTS_MCPU_NATIVE)
-            add_compile_options($<$<COMPILE_LANGUAGE:Fortran>:-mcpu=native>)
+    # Skip for gcov builds — -march=native on newer CPUs (e.g. Granite Rapids)
+    # can emit instructions the system assembler doesn't support.
+    if (NOT MFC_GCov)
+        CHECK_FORTRAN_COMPILER_FLAG("-march=native" SUPPORTS_MARCH_NATIVE)
+        if (SUPPORTS_MARCH_NATIVE)
+            add_compile_options($<$<COMPILE_LANGUAGE:Fortran>:-march=native>)
+            # Disable AVX-512 FP16: gfortran >=12 emits vmovw instructions on
+            # Granite Rapids CPUs, but binutils <2.38 cannot assemble them.
+            # FP16 is unused in MFC's double-precision computations.
+            CHECK_FORTRAN_COMPILER_FLAG("-mno-avx512fp16" SUPPORTS_MNO_AVX512FP16)
+            if (SUPPORTS_MNO_AVX512FP16)
+                add_compile_options($<$<COMPILE_LANGUAGE:Fortran>:-mno-avx512fp16>)
+            endif()
+        else()
+            CHECK_FORTRAN_COMPILER_FLAG("-mcpu=native" SUPPORTS_MCPU_NATIVE)
+            if (SUPPORTS_MCPU_NATIVE)
+                add_compile_options($<$<COMPILE_LANGUAGE:Fortran>:-mcpu=native>)
+            endif()
         endif()
     endif()
 
@@ -397,6 +408,7 @@ HANDLE_SOURCES(simulation ON)
 HANDLE_SOURCES(post_process ON)
 HANDLE_SOURCES(syscheck OFF)
 
+
 # MFC_SETUP_TARGET: Given a target (herein <target>), this macro creates a new
 # executable <target> with the appropriate sources, compiler definitions, and
 # linked libraries (assuming HANDLE_SOURCES was called on <target>).
@@ -633,6 +645,23 @@ if (MFC_SIMULATION)
     MFC_SETUP_TARGET(TARGET  simulation
                      SOURCES "${simulation_SRCs}"
                      MPI FFTW OpenACC OpenMP)
+    # CCE 19.0.0 IPA workaround: two files trigger IPA crashes:
+    #   m_bubbles_EL: castIsValid assertion (InstCombine/foldIntegerTypedPHI)
+    #   m_phase_change: bring_routine_resident SIGSEGV
+    # Disabling IPA per-file avoids the crashes while preserving IPA for
+    # the rest of simulation (needed for thermochem INLINEALWAYS inlining).
+    # Applied to Cray+OpenACC and Cray CPU, but NOT Cray+OpenMP: on OpenMP,
+    # m_thermochem uses !DIR$ INLINEALWAYS (requires IPA), so disabling IPA
+    # for these files breaks thermochem on-device calls. On OpenACC the
+    # pyrometheus patch emits !$acc routine seq instead (no IPA needed).
+    # See PR #1286.
+    if (CMAKE_Fortran_COMPILER_ID STREQUAL "Cray" AND NOT MFC_OpenMP)
+        set_source_files_properties(
+            "${CMAKE_BINARY_DIR}/fypp/simulation/m_bubbles_EL.fpp.f90"
+            "${CMAKE_BINARY_DIR}/fypp/simulation/m_phase_change.fpp.f90"
+            PROPERTIES COMPILE_OPTIONS "-Oipa0"
+        )
+    endif()
 endif()
 
 if (MFC_POST_PROCESS)

src/common/include/parallel_macros.fpp

@@ -48,18 +48,46 @@
 
 #:enddef
 
-#:def GPU_ROUTINE(function_name=None, parallelism=None, nohost=False, cray_inline=False, extraAccArgs=None, extraOmpArgs=None)
+#:def GPU_ROUTINE(function_name=None, parallelism=None, nohost=False, cray_inline=False, cray_noinline=False, extraAccArgs=None, extraOmpArgs=None)
     #:assert isinstance(cray_inline, bool)
+    #:assert isinstance(cray_noinline, bool)
+    #:assert not (cray_inline and cray_noinline), "cray_inline and cray_noinline are mutually exclusive"
     #:set acc_directive = ACC_ROUTINE(function_name=function_name, parallelism=parallelism, nohost=nohost, extraAccArgs=extraAccArgs)
     #:set omp_directive = OMP_ROUTINE(function_name=function_name, nohost=nohost, extraOmpArgs=extraOmpArgs)
 
-    #:if cray_inline == True
+    #:if cray_noinline == True
+        #:if not isinstance(function_name, str)
+            #:stop "When using cray_noinline, function name must be given and given as a string"
+        #:endif
+        #:set cray_noinline_directive = ('!DIR$ NOINLINE ' + function_name).strip('\n')
+#ifdef _CRAYFTN
+#if MFC_OpenACC
+        $:acc_directive
+#elif MFC_OpenMP
+        $:omp_directive
+#else
+        $:cray_noinline_directive
+#endif
+        #! On non-Cray CPU builds (no _CRAYFTN, no MFC_OpenACC, no MFC_OpenMP), nothing is
+        #! emitted — intentional, since !DIR$ NOINLINE is a Cray-specific directive.
+#elif MFC_OpenACC
+        $:acc_directive
+#elif MFC_OpenMP
+        $:omp_directive
+#endif
+    #:elif cray_inline == True
         #:if not isinstance(function_name, str)
             #:stop "When inlining for Cray Compiler, function name must be given and given as a string"
         #:endif
         #:set cray_directive = ('!DIR$ INLINEALWAYS ' + function_name).strip('\n')
 #ifdef _CRAYFTN
+#if MFC_OpenACC
+        $:acc_directive
+#elif MFC_OpenMP
+        $:omp_directive
+#else
         $:cray_directive
+#endif
 #elif MFC_OpenACC
         $:acc_directive
 #elif MFC_OpenMP