Adding fatbin to shared libs and fixing conda paths in cpu build

build.sh

@@ -60,7 +60,7 @@ SINGLEGPU=""
 NVTX=OFF
 CLEAN=0
 DISABLE_DEPRECATION_WARNINGS=ON
-CMAKE_TARGET=""
+CMAKE_TARGET="install;"
 
 # Set defaults for vars that may not have been defined externally
 #  FIXME: if INSTALL_PREFIX is not set, check PREFIX, then check
@@ -180,16 +180,16 @@ if (( ${NUMARGS} == 0 )) || hasArg libraft || hasArg docs; then
 
   if (( ${NUMARGS} == 0 )) || hasArg libraft; then
       # Run all c++ targets at once
-      if hasArg --compile-nn || hasArg --compile-dist || hasArg --compile-libs; then
+#      if hasArg --compile-nn || hasArg --compile-dist || hasArg --compile-libs; then
         if ! hasArg --nogtest; then
           CMAKE_TARGET="test_raft;${CMAKE_TARGET}"
         fi
 
         echo "-- Compiling targets: ${CMAKE_TARGET}"
         cmake --build  ${CPP_RAFT_BUILD_DIR} -j${PARALLEL_LEVEL} ${VERBOSE_FLAG} --target ${CMAKE_TARGET}
-      else
-        cmake --build  ${CPP_RAFT_BUILD_DIR} -j${PARALLEL_LEVEL} ${VERBOSE_FLAG}
-      fi
+#      else
+#        cmake --build  ${CPP_RAFT_BUILD_DIR} -j${PARALLEL_LEVEL} ${VERBOSE_FLAG}
+#      fi
   fi
 
 fi

ci/cpu/build.sh

@@ -70,7 +70,7 @@ conda config --set ssl_verify False
 # install. This should eliminate a mismatch between different CUDA versions on
 # cpu vs. gpu builds that is problematic with CUDA 11.5 Enhanced Compat.
 if [ "$BUILD_LIBRAFT" == '1' ]; then
-  BUILD_PYRAFT=1
+  BUILD_RAFT=1
   # If we are doing CUDA + Python builds, libraft package is located at ${CONDA_BLD_DIR}
   CONDA_LOCAL_CHANNEL="${CONDA_BLD_DIR}"
 else
@@ -86,21 +86,21 @@ fi
 if [ "$BUILD_LIBRAFT" == '1' ]; then
   gpuci_logger "Building conda packages for libraft-nn, libraft-distance, and libraft-headers"
   if [[ -z "$PROJECT_FLASH" || "$PROJECT_FLASH" == "0" ]]; then
-    gpuci_conda_retry build --no-build-id --croot ${CONDA_BLD_DIR} conda/recipes/libraft-nn
-    gpuci_conda_retry build --no-build-id --croot ${CONDA_BLD_DIR} conda/recipes/libraft-distance
-    gpuci_conda_retry build --no-build-id --croot ${CONDA_BLD_DIR} conda/recipes/libraft-headers
+    gpuci_conda_retry build --no-build-id --croot ${CONDA_BLD_DIR} conda/recipes/libraft_headers
+    gpuci_conda_retry build --no-build-id --croot ${CONDA_BLD_DIR} conda/recipes/libraft_nn
+    gpuci_conda_retry build --no-build-id --croot ${CONDA_BLD_DIR} conda/recipes/libraft_distance
   else
-    gpuci_conda_retry build --no-build-id --croot ${CONDA_BLD_DIR} --dirty --no-remove-work-dir conda/recipes/libraft-nn
-    gpuci_conda_retry build --no-build-id --croot ${CONDA_BLD_DIR} --dirty --no-remove-work-dir conda/recipes/libraft-distance
-    gpuci_conda_retry build --no-build-id --croot ${CONDA_BLD_DIR} --dirty --no-remove-work-dir conda/recipes/libraft-headers
+    gpuci_conda_retry build --no-build-id --croot ${CONDA_BLD_DIR} --dirty --no-remove-work-dir conda/recipes/libraft_headers
+    gpuci_conda_retry build --no-build-id --croot ${CONDA_BLD_DIR} --dirty --no-remove-work-dir conda/recipes/libraft_nn
+    gpuci_conda_retry build --no-build-id --croot ${CONDA_BLD_DIR} --dirty --no-remove-work-dir conda/recipes/libraft_distance
     mkdir -p ${CONDA_BLD_DIR}/libraft
     mv ${CONDA_BLD_DIR}/work ${CONDA_BLD_DIR}/libraft/work
   fi
 else
   gpuci_logger "SKIPPING build of conda packages for libraft-nn, libraft-distance and libraft-headers"
 fi
 
-if [ "$BUILD_raft" == "1" ]; then
+if [ "$BUILD_RAFT" == "1" ]; then
   gpuci_logger "Building conda packages for pyraft"
   if [[ -z "$PROJECT_FLASH" || "$PROJECT_FLASH" == "0" ]]; then
     gpuci_conda_retry build --no-build-id --croot ${CONDA_BLD_DIR} conda/recipes/pyraft --python=$PYTHON
@@ -118,4 +118,4 @@ fi
 ################################################################################
 
 gpuci_logger "Upload conda packages"
-source ci/cpu/upload.sh
+source ci/cpu/upload.sh

ci/cpu/prebuild.sh

@@ -14,7 +14,7 @@
 
 if [[ -z "$PROJECT_FLASH" || "$PROJECT_FLASH" == "0" ]]; then
     #If project flash is not activate, always build both
-    export BUILD_PYRAFT=1
+    export BUILD_RAFT=1
     export BUILD_LIBRAFT=1
 fi
 

ci/cpu/upload.sh

@@ -61,7 +61,7 @@ if [[ "$BUILD_LIBRAFT" == "1" && "$UPLOAD_LIBRAFT" == "1" ]]; then
   gpuci_retry anaconda -t ${MY_UPLOAD_KEY} upload -u ${CONDA_USERNAME:-rapidsai} ${LABEL_OPTION} --skip-existing ${LIBRAFT_HEADERS_FILE} --no-progress
 fi
 
-if [[ "$BUILD_PYRAFT" == "1" ]]; then
+if [[ "$BUILD_RAFT" == "1" ]]; then
   test -e ${PYRAFT_FILE}
   echo "Upload pyraft"
   echo ${PYRAFT_FILE}

conda/recipes/libraft_distance/meta.yaml

@@ -36,7 +36,7 @@ requirements:
   build:
     - cmake>=3.20.1
   host:
-    - libraft-headers {{
+    - libraft-headers {{ version }}
     - nccl>=2.9.9
     - cudatoolkit {{ cuda_version }}.*
     - ucx-py {{ ucx_py_version }}

conda/recipes/pyraft/meta.yaml

@@ -29,14 +29,14 @@ requirements:
     - python x.x
     - setuptools
     - cython>=0.29,<0.30
-    - libraft-headers={{ version }}
+    - libraft-headers {{ version }}
     - cudatoolkit {{ cuda_version }}.*
     - ucx-py {{ ucx_py_version }}
     - ucx-proc=*=gpu
   run:
     - python x.x
     - dask-cuda {{ minor_version }}
-    - libraft-headers={{ version }}
+    - libraft-headers {{ version }}
     - cupy>=7.8.0,<10.0.0a0
     - nccl>=2.9.9
     - ucx-py {{ ucx_py_version }}

cpp/CMakeLists.txt

@@ -142,6 +142,16 @@ target_link_libraries(raft INTERFACE
 target_compile_definitions(raft INTERFACE $<$<BOOL:${NVTX}>:NVTX_ENABLED>)
 target_compile_features(raft INTERFACE cxx_std_17 $<BUILD_INTERFACE:cuda_std_17>)
 
+if(RAFT_COMPILE_LIBRARIES)
+  file(WRITE "${CMAKE_CURRENT_BINARY_DIR}/fatbin.ld"
+[=[
+SECTIONS
+{
+.nvFatBinSegment : { *(.nvFatBinSegment) }
+.nv_fatbin : { *(.nv_fatbin) }
+}
+]=])
+endif()
 ##############################################################################
 # - raft_distance ------------------------------------------------------------
 add_library(raft_distance INTERFACE)
@@ -180,6 +190,9 @@ if(RAFT_COMPILE_LIBRARIES)
   target_compile_definitions(raft_distance_lib
           INTERFACE "RAFT_DISTANCE_COMPILED")
 
+  # ensure CUDA symbols aren't relocated to the middle of the debug build binaries
+  target_link_options(raft_distance_lib PRIVATE "${CMAKE_CURRENT_BINARY_DIR}/fatbin.ld")
+
 endif()
 
 target_link_libraries(raft_distance INTERFACE raft::raft
@@ -214,6 +227,9 @@ if(RAFT_COMPILE_LIBRARIES)
           PRIVATE "$<$<COMPILE_LANGUAGE:CXX>:${RAFT_CXX_FLAGS}>"
           "$<$<COMPILE_LANGUAGE:CUDA>:${RAFT_CUDA_FLAGS}>"
           )
+  # ensure CUDA symbols aren't relocated to the middle of the debug build binaries
+  target_link_options(raft_nn_lib PRIVATE "${CMAKE_CURRENT_BINARY_DIR}/fatbin.ld")
+
   target_compile_definitions(raft_nn_lib
           INTERFACE "RAFT_NN_COMPILED")