Merge branch 'master' into cuda_12.4_dep

.github/workflows/main.yml

@@ -20,7 +20,7 @@ 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: '3.6'
@@ -39,6 +39,7 @@ jobs:
             exuberant-ctags \
             gfortran \
             git \
+            libhwloc-dev \
             libopenblas-dev \
             pkg-config \
             software-properties-common
@@ -50,6 +51,7 @@ jobs:
             ctags-exuberant \
             gawk \
             gnu-sed \
+            hwloc \
             pkg-config
       - uses: actions/setup-python@v5.0.0
         with:
@@ -110,8 +112,6 @@ jobs:
           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_guppi.py
-          coverage run --source=bifrost.ring,bifrost,bifrost.pipeline test_guppi_reader.py
           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"

Makefile.in

@@ -7,8 +7,12 @@ DAT_DIR = share
 SRC_DIR = src
 
 HAVE_PYTHON = @HAVE_PYTHON@
+
 HAVE_DOCKER = @HAVE_DOCKER@
 
+CAN_BUILD_CXX_DOCS = @HAVE_CXX_DOCS@
+CAN_BUILD_PYTHON_DOCS = @HAVE_PYTHON_DOCS@
+
 BIFROST_PYTHON_DIR = python
 
 all: libbifrost python
@@ -34,7 +38,7 @@ endif
 test:
 	#$(MAKE) -C $(SRC_DIR) test
 ifeq ($(HAVE_PYTHON),1)
-	cd test && ./download_test_data.sh ; python -m unittest discover
+	cd test && ./download_test_data.sh ; @PYTHON@ -m unittest discover -v
 endif
 .PHONY: test
 clean:
@@ -76,8 +80,13 @@ ifeq ($(HAVE_PYTHON),1)
 endif
 .PHONY: uninstall
 
-doc: $(INC_DIR)/bifrost/*.h Doxyfile
+doc: $(INC_DIR)/bifrost/*.h Doxyfile docs/source/*.rst docs/source/*.py
+ifeq ($(CAN_BUILD_CXX_DOCS),1)
 	@DX_DOXYGEN@ Doxyfile
+endif
+ifeq ($(CAN_BUILD_PYTHON_DOCS),1)
+	$(MAKE) -C docs singlehtml
+endif
 .PHONY: doc
 
 python: libbifrost

README.md

@@ -95,10 +95,20 @@ print "All done"
 **For a quick demo which you can run in-browser without installation,
 go to the following [link](https://colab.research.google.com/github/ledatelescope/bifrost/blob/master/BifrostDemo.ipynb).**
 
+### CUDA
+
+CUDA is available at https://developer.nvidia.com/cuda-downloads. You can check the
+["Getting Started guide"](http://ledatelescope.github.io/bifrost/Getting-started-guide.html) 
+in the docs to see which versions of the CUDA toolkit have been confirmed to work with Bifrost. 
+
 ### C Dependencies
 
+If using Ubuntu or another Debian-based linux distribution:
+
     $ sudo apt-get install exuberant-ctags
 
+Otherwise check https://ctags.sourceforge.net/ for install instructions.
+
 ### Python Dependencies
 
  * numpy
@@ -171,8 +181,17 @@ your machine.
 
 ### Building the Docs from Scratch
 
-Install sphinx and breathe using pip, and also install Doxygen.
+Install breathe using pip:
+
+    $ sudo pip install breathe sphinx
+
+Also install Doxygen using your package manager. 
+In Ubuntu, for example:
+
+    $ sudo apt-get install doxygen
 
+Otherwise check https://www.doxygen.nl/ for Doxygen install instructions.
+
 Doxygen documentation can be generated by running:
 
     $ make doc

config/cuda.m4

@@ -14,6 +14,11 @@ AC_DEFUN([AX_CHECK_CUDA],
                 [enable_cuda=no],
                 [enable_cuda=yes])
   
+  NVCCLIBS=""
+  ac_compile_save="$ac_compile"
+  ac_link_save="$ac_link"
+  ac_run_save="$ac_run"
+  
   AC_SUBST([HAVE_CUDA], [0])
   AC_SUBST([CUDA_VERSION], [0])
   AC_SUBST([CUDA_HAVE_CXX20], [0])
@@ -38,12 +43,12 @@ AC_DEFUN([AX_CHECK_CUDA],
     
     CXXFLAGS_save="$CXXFLAGS"
     LDFLAGS_save="$LDFLAGS"
-    LIBS_save="$LIBS"
+    NVCCLIBS_save="$NVCCLIBS"
     ac_ext_save="$ac_ext"
     
     ac_compile='$NVCC -c $NVCCFLAGS conftest.$ac_ext >&5'
     LDFLAGS="-L$CUDA_HOME/lib64 -L$CUDA_HOME/lib"
-    LIBS="$LIBS -lcuda -lcudart"
+    NVCCLIBS="$LIBS -lcuda -lcudart"
     ac_ext="cu"
 
     ac_link='$NVCC -o conftest$ac_exeext $NVCCFLAGS $LDFLAGS $LIBS conftest.$ac_ext >&5'
@@ -52,19 +57,31 @@ AC_DEFUN([AX_CHECK_CUDA],
           #include <cuda.h>
           #include <cuda_runtime.h>]],
           [[cudaMalloc(0, 0);]])],
-        [CUDA_VERSION=$( ${NVCC} --version | ${GREP} -Po -e "release.*," | cut -d,  -f1 | cut -d\  -f2 )
-         CUDA_MAJOR=$( echo "${CUDA_VERSION}" | cut -d. -f1 )
-         if test "${CUDA_MAJOR}" -ge 10; then
-           AC_MSG_RESULT(yes - v$CUDA_VERSION)
-         else
-           AC_MSG_RESULT(no - found v$CUDA_VERSION)
-         fi],
-        [AC_MSG_RESULT(no - build failure)
-         AC_SUBST([HAVE_CUDA], [0])])
+        [],
+        [AC_SUBST([HAVE_CUDA], [0])])
+    
+    if test "$HAVE_CUDA" = "1"; then
+      LDFLAGS="-L$CUDA_HOME/lib64 -L$CUDA_HOME/lib"
+      NVCCLIBS="$NVCCLIBS -lcuda -lcudart"
+
+      ac_link='$NVCC -o conftest$ac_exeext $NVCCFLAGS $LDFLAGS $LIBS $NVCCLIBS conftest.$ac_ext >&5'
+      AC_LINK_IFELSE([
+        AC_LANG_PROGRAM([[
+            #include <cuda.h>
+            #include <cuda_runtime.h>]],
+            [[cudaMalloc(0, 0);]])],
+          [CUDA_VERSION=$( ${NVCC} --version | ${GREP} -Po -e "release.*," | cut -d,  -f1 | cut -d\  -f2 )
+           AC_MSG_RESULT(yes - v$CUDA_VERSION)],
+          [AC_MSG_RESULT(no)
+           AC_SUBST([HAVE_CUDA], [0])])
+    else
+      AC_MSG_RESULT(no)
+      AC_SUBST([HAVE_CUDA], [0])
+    fi
     
     CXXFLAGS="$CXXFLAGS_save"
     LDFLAGS="$LDFLAGS_save"
-    LIBS="$LIBS_save"
+    NVCCLIBS="$NVCCLIBS_save"
     ac_ext="$ac_ext_save"
   fi
   
@@ -134,7 +151,7 @@ AC_DEFUN([AX_CHECK_CUDA],
     CXXFLAGS="$CXXFLAGS -DBF_CUDA_ENABLED=1"
     NVCCFLAGS="$NVCCFLAGS -DBF_CUDA_ENABLED=1"
     LDFLAGS="$LDFLAGS -L$CUDA_HOME/lib64 -L$CUDA_HOME/lib"
-    LIBS="$LIBS -lcuda -lcudart -lnvrtc -lcublas -lcudadevrt -L. -lcufft_static_pruned -lculibos -lnvToolsExt"
+    NVCCLIBS="$NVCCLIBS -lcuda -lcudart -lnvrtc -lcublas -lcudadevrt -L. -lcufft_static_pruned -lculibos -lnvToolsExt"
   fi
   
   AC_ARG_WITH([gpu_archs],
@@ -153,13 +170,13 @@ AC_DEFUN([AX_CHECK_CUDA],
 
       CXXFLAGS_save="$CXXFLAGS"
       LDFLAGS_save="$LDFLAGS"
-      LIBS_save="$LIBS"
+      NVCCLIBS_save="$NVCCLIBS"
       ac_ext_save="$ac_ext"
       
       LDFLAGS="-L$CUDA_HOME/lib64 -L$CUDA_HOME/lib"
-      LIBS="-lcuda -lcudart"
+      NVCCLIBS="-lcuda -lcudart"
       ax_ext="cu"
-      ac_run='$NVCC -o conftest$ac_ext $LDFLAGS $LIBS conftest.$ac_ext>&5'
+      ac_run='$NVCC -o conftest$ac_ext $LDFLAGS $LIBS $NVCCLIBS conftest.$ac_ext>&5'
       AC_RUN_IFELSE([
         AC_LANG_PROGRAM([[
             #include <cuda.h>
@@ -209,7 +226,7 @@ AC_DEFUN([AX_CHECK_CUDA],
 
       CXXFLAGS="$CXXFLAGS_save"
       LDFLAGS="$LDFLAGS_save"
-      LIBS="$LIBS_save"
+      NVCCLIBS="$NVCCLIBS_save"
       ac_ext="$ac_ext_save"
     else
       AC_SUBST([GPU_ARCHS], [$with_gpu_archs])
@@ -240,11 +257,11 @@ AC_DEFUN([AX_CHECK_CUDA],
 
       CXXFLAGS_save="$CXXFLAGS"
       LDFLAGS_save="$LDFLAGS"
-      LIBS_save="$LIBS"
+      NVCCLIBS_save="$NVCCLIBS"
       ac_ext_save="$ac_ext"
       
       LDFLAGS="-L$CUDA_HOME/lib64 -L$CUDA_HOME/lib"
-      LIBS="-lcuda -lcudart"
+      NVCCLIBS="-lcuda -lcudart"
       ac_ext="cu"
       ac_run='$NVCC -o conftest$ac_ext $LDFLAGS $LIBS conftest.$ac_ext>&5'
       AC_RUN_IFELSE([
@@ -283,7 +300,7 @@ AC_DEFUN([AX_CHECK_CUDA],
 
       CXXFLAGS="$CXXFLAGS_save"
       LDFLAGS="$LDFLAGS_save"
-      LIBS="$LIBS_save"
+      NVCCLIBS="$NVCCLIBS_save"
       ac_ext="$ac_ext_save"
     else
       AC_SUBST([GPU_SHAREDMEM], [$with_shared_mem])
@@ -302,13 +319,13 @@ AC_DEFUN([AX_CHECK_CUDA],
     AC_MSG_CHECKING([for thrust pinned allocated support])
     CXXFLAGS_save="$CXXFLAGS"
     LDFLAGS_save="$LDFLAGS"
-    LIBS_save="$LIBS"
+    NVCCLIBS_save="$NVCCLIBS"
     ac_ext_save="$ac_ext"
     
     LDFLAGS="-L$CUDA_HOME/lib64 -L$CUDA_HOME/lib"
-    LIBS="-lcuda -lcudart"
+    NVCCLIBS="-lcuda -lcudart"
     ac_ext="cu"
-    ac_run='$NVCC -o conftest$ac_ext $LDFLAGS $LIBS conftest.$ac_ext>&5'
+    ac_run='$NVCC -o conftest$ac_ext $LDFLAGS $LIBS $NVCCLIBS conftest.$ac_ext>&5'
     AC_RUN_IFELSE([
       AC_LANG_PROGRAM([[
           #include <cuda.h>
@@ -322,7 +339,14 @@ AC_DEFUN([AX_CHECK_CUDA],
 
     CXXFLAGS="$CXXFLAGS_save"
     LDFLAGS="$LDFLAGS_save"
-    LIBS="$LIBS_save"
+    NVCCLIBS="$NVCCLIBS_save"
     ac_ext="$ac_ext_save"
+  else
+     AC_SUBST([GPU_PASCAL_MANAGEDMEM], [0])
+     AC_SUBST([GPU_EXP_PINNED_ALLOC], [1])
   fi
+  
+  ac_compile="$ac_compile_save"
+  ac_link="$ac_link_save"
+  ac_run="$ac_run_save"
 ])

config/intrinsics.m4

@@ -0,0 +1,121 @@
+#
+# SSE
+#
+
+AC_DEFUN([AX_CHECK_SSE],
+[
+  AC_PROVIDE([AX_CHECK_SSE])
+  AC_ARG_ENABLE([sse],
+                [AS_HELP_STRING([--disable-sse],
+                                [disable SSE support (default=no)])],
+                [enable_sse=no],
+                [enable_sse=yes])
+
+  AC_SUBST([HAVE_SSE], [0])
+  
+  if test "$enable_sse" = "yes"; then
+    AC_MSG_CHECKING([for SSE support via '-msse'])
+    
+    CXXFLAGS_temp="$CXXFLAGS -msse"
+    
+    ac_run="$CXX -o conftest$ac_ext $CXXFLAGS_temp conftest.$ac_ext>&5"
+    AC_RUN_IFELSE([
+      AC_LANG_PROGRAM([[
+        #include <xmmintrin.h>]],
+        [[
+        __m128 x = _mm_set1_ps(1.0f);
+        x = _mm_add_ps(x, x);
+        return _mm_cvtss_f32(x) != 2.0f;]])], 
+    [
+     CXXFLAGS="$CXXFLAGS -msse"
+     AC_SUBST([HAVE_SSE], [1])
+     AC_MSG_RESULT([yes])
+    ], [
+     AC_MSG_RESULT([no])
+    ])
+  fi
+])
+
+
+
+#
+# AVX
+#
+
+AC_DEFUN([AX_CHECK_AVX],
+[
+  AC_PROVIDE([AX_CHECK_AVX])
+  AC_ARG_ENABLE([avx],
+                [AS_HELP_STRING([--disable-avx],
+                                [disable AVX support (default=no)])],
+                [enable_avx=no],
+                [enable_avx=yes])
+  
+  AC_SUBST([HAVE_AVX], [0])
+  
+  if test "$enable_avx" = "yes"; then
+    AC_MSG_CHECKING([for AVX support via '-mavx'])
+    
+    CXXFLAGS_temp="$CXXFLAGS -mavx"
+    ac_run_save="$ac_run"
+    
+    ac_run="$CXX -o conftest$ac_ext $CXXFLAGS_temp conftest.$ac_ext>&5"
+    AC_RUN_IFELSE([
+      AC_LANG_PROGRAM([[
+        #include <immintrin.h>]],
+        [[
+        __m256d x = _mm256_set1_pd(1.0);
+        x = _mm256_add_pd(x, x);
+        return _mm256_cvtsd_f64(x) != 2.0;]])],
+    [
+     CXXFLAGS="$CXXFLAGS -mavx"
+     AC_SUBST([HAVE_AVX], [1])
+     AC_MSG_RESULT([yes])
+    ], [
+     AC_MSG_RESULT([no])
+    ])
+    
+    ac_run="$ac_run_save"
+  fi
+])
+
+#
+# AVX512
+#
+
+AC_DEFUN([AX_CHECK_AVX512],
+[
+  AC_PROVIDE([AX_CHECK_AVX512])
+  AC_ARG_ENABLE([avx512],
+                [AS_HELP_STRING([--disable-avx512],
+                                [disable AVX512 support (default=no)])],
+                [enable_avx512=no],
+                [enable_avx512=yes])
+  
+  AC_SUBST([HAVE_AVX512], [0])
+  
+  if test "$enable_avx512" = "yes"; then
+    AC_MSG_CHECKING([for AVX-512 support via '-mavx512f'])
+    
+    CXXFLAGS_temp="$CXXFLAGS -mavx512f"
+    ac_run_save="$ac_run"
+    
+    ac_run="$CXX -o conftest$ac_ext $CXXFLAGS_temp conftest.$ac_ext>&5"
+    AC_RUN_IFELSE([
+      AC_LANG_PROGRAM([[
+        #include <immintrin.h>]],
+        [[
+        __m512d x = _mm512_set1_pd(1.0);
+        x = _mm512_add_pd(x, x);
+        return _mm512_cvtsd_f64(x) != 2.0;]])], 
+    [
+     CXXFLAGS="$CXXFLAGS -mavx512f"
+     AC_SUBST([HAVE_AVX512], [1])
+     AC_MSG_RESULT([yes])
+    ], [
+     AC_MSG_RESULT([no])
+    ])
+    
+    ac_run="$ac_run_save"
+  fi
+])