Merge branch 'ibverb-support' into add-simple-packet

.github/workflows/main.yml

@@ -3,7 +3,7 @@ name: "Build and Test"
 "on": [push, pull_request]
 jobs:
   pre_build:
-    runs-on: ubuntu-20.04
+    runs-on: ubuntu-latest
     outputs:
       should_skip: ${{ steps.skip_check.outputs.should_skip }}
     steps:
@@ -20,10 +20,8 @@ jobs:
     strategy:
       matrix:
         os: [self-hosted, ubuntu-latest, macos-latest]
-        python-version: ['3.8', '3.10']
+        python-version: ['3.8', '3.10', '3.12']
         include:
-          - os: ubuntu-20.04
-            python-version: '2.7'
           - os: ubuntu-20.04
             python-version: '3.6'
           - os: macos-latest
@@ -55,7 +53,7 @@ jobs:
             gnu-sed \
             hwloc \
             pkg-config
-      - uses: actions/setup-python@v4.3.0
+      - uses: actions/setup-python@v5.0.0
         with:
           python-version: ${{ matrix.python-version }}
       - name: "Software Install - Python"
@@ -68,9 +66,10 @@ jobs:
                pint \
                graphviz \
                ctypesgen==1.0.2 \
+               pylint \
                coverage
       - name: "Software Install - Python, part 2"
-        if: ${{ matrix.os == 'self-hosted' && matrix.python-version != '2.7' }}
+        if: ${{ matrix.os == 'self-hosted' }}
         # Setting CPLUS_INCLUDE_PATH helps pycuda find the right
         # Python header files <pyconfig.h> to use with its embedded
         # subset of Boost.
@@ -102,14 +101,27 @@ jobs:
           coverage run --source=bifrost.ring,bifrost,bifrost.pipeline \
             -m unittest discover
           coverage xml
+      - name: "Test, part 2"
+        if: ${{ matrix.os == 'self-hosted' }}
+        env:
+          LD_LIBRARY_PATH: /usr/local/lib:${{ env.LD_LIBRARY_PATH }}
+        run: |
+          cd testbench
+          python generate_test_data.py
+          coverage run --source=bifrost.ring,bifrost,bifrost.pipeline test_file_read_write.py
+          coverage run --source=bifrost.ring,bifrost,bifrost.pipeline test_fft.py
+          coverage run --source=bifrost.ring,bifrost,bifrost.pipeline your_first_block.py
+          python download_breakthrough_listen_data.py -y
+          coverage run --source=bifrost.ring,bifrost,bifrost.pipeline test_fdmt.py ./testdata/pulsars/blc0_guppi_57407_61054_PSR_J1840%2B5640_0004.fil
+          coverage xml
       - name: "Upload Coverage"
         env:
           UNITTEST_OS: ${{ matrix.os }}
           UNITTEST_PY: ${{ matrix.python-version }}
         if: ${{ matrix.os == 'self-hosted' && matrix.python-version == '3.8' }}
         uses: codecov/codecov-action@v2
         with:
-          directory: ./test/
+          files: ./test/coverage.xml, ./testbench/coverage.xml
           env_vars: UNITTEST_OS,UNITTEST_PY
           fail_ci_if_error: false
           verbose: true

.gitignore

@@ -2,6 +2,7 @@ doc
 version.py
 /python/bifrost/version/__init__.py
 *_generated.py
+*_typehints.py
 .log*.txt
 test/data/
 *.bin

config/cuda.m4

@@ -27,6 +27,9 @@ AC_DEFUN([AX_CHECK_CUDA],
   AC_SUBST([CUDA_HAVE_CXX11], [0])
   AC_SUBST([GPU_MIN_ARCH], [0])
   AC_SUBST([GPU_MAX_ARCH], [0])
+  AC_SUBST([GPU_SHAREDMEM], [0])
+  AC_SUBST([GPU_PASCAL_MANAGEDMEM], [0])
+  AC_SUBST([GPU_EXP_PINNED_ALLOC], [1])
   if test "$enable_cuda" != "no"; then
     AC_SUBST([HAVE_CUDA], [1])
     
@@ -36,14 +39,18 @@ AC_DEFUN([AX_CHECK_CUDA],
   fi
 
   if test "$HAVE_CUDA" = "1"; then
-    AC_MSG_CHECKING([for a working CUDA installation])
+    AC_MSG_CHECKING([for a working CUDA 10+ installation])
     
     CXXFLAGS_save="$CXXFLAGS"
     LDFLAGS_save="$LDFLAGS"
     NVCCLIBS_save="$NVCCLIBS"
     
     ac_compile='$NVCC -c $NVCCFLAGS conftest.$ac_ext >&5'
-    AC_COMPILE_IFELSE([
+    LDFLAGS="-L$CUDA_HOME/lib64 -L$CUDA_HOME/lib"
+    NVCCLIBS="$LIBS -lcuda -lcudart"
+
+    ac_link='$NVCC -o conftest$ac_exeext $NVCCFLAGS $LDFLAGS $LIBS conftest.$ac_ext >&5'
+    AC_LINK_IFELSE([
       AC_LANG_PROGRAM([[
           #include <cuda.h>
           #include <cuda_runtime.h>]],
@@ -234,6 +241,62 @@ AC_DEFUN([AX_CHECK_CUDA],
     ar_max_valid=$(echo $ar_valid | ${SED} -e 's/.* //g;' )
     AC_SUBST([GPU_MAX_ARCH], [$ar_max_valid])
 
+    AC_ARG_WITH([shared_mem],
+           [AS_HELP_STRING([--with-shared-mem=N],
+                           [default GPU shared memory per block in bytes (default=detect)])],
+           [],
+           [with_shared_mem='auto'])
+    if test "$with_gpu_archs" = "auto"; then
+      AC_MSG_CHECKING([for minimum shared memory per block])
+
+      CXXFLAGS_save="$CXXFLAGS"
+      LDFLAGS_save="$LDFLAGS"
+      NVCCLIBS_save="$NVCCLIBS"
+      
+      LDFLAGS="-L$CUDA_HOME/lib64 -L$CUDA_HOME/lib"
+      NVCCLIBS="-lcuda -lcudart"
+      ac_run='$NVCC -o conftest$ac_ext $LDFLAGS $LIBS conftest.$ac_ext>&5'
+      AC_RUN_IFELSE([
+        AC_LANG_PROGRAM([[
+            #include <cuda.h>
+            #include <cuda_runtime.h>
+            #include <iostream>
+            #include <fstream>
+            #include <set>]],
+            [[
+            std::set<int> smem;
+            int smemSize;
+            int deviceCount = 0;
+            cudaGetDeviceCount(&deviceCount);
+            if( deviceCount == 0 ) {
+              return 1;
+            }
+            for(int dev=0; dev<deviceCount; dev++) {
+              cudaSetDevice(dev);
+              cudaDeviceGetAttribute(&smemSize, cudaDevAttrMaxSharedMemoryPerBlock, dev);
+              if( smem.count(smemSize) == 0 ) {
+                smem.insert(smemSize);
+              }
+            }
+            std::ofstream fh;
+            fh.open("confsmem.out");
+            if( smem.empty() ) {
+              fh << 0;
+            } else {
+              fh << *smem.begin();
+            }
+            fh.close();]])],
+            [AC_SUBST([GPU_SHAREDMEM], [`cat confsmem.out`])
+             AC_MSG_RESULT([$GPU_SHAREDMEM B])],
+            [AC_MSG_ERROR(failed to determine a value)])
+
+      CXXFLAGS="$CXXFLAGS_save"
+      LDFLAGS="$LDFLAGS_save"
+      NVCCLIBS="$NVCCLIBS_save"
+    else
+      AC_SUBST([GPU_SHAREDMEM], [$with_shared_mem])
+    fi
+    
     AC_MSG_CHECKING([for Pascal-style CUDA managed memory])
     cm_invalid=$( echo $GPU_ARCHS | ${SED} -e 's/\b[[1-5]][[0-9]]\b/PRE/g;' )
     if ! echo $cm_invalid | ${GREP} -q PRE; then