Cleanup remaining QMC_CUDA

QMCPACK · Mar 6, 2023 · 629ba45 · 629ba45
1 parent c2435c6
commit 629ba45
Show file tree

Hide file tree

Showing 16 changed files with 41 additions and 618 deletions.
diff --git a/CMake/ctest_script.cmake b/CMake/ctest_script.cmake
@@ -174,10 +174,6 @@ if(DEFINED RMG_BIN)
   set(CTEST_OPTIONS "${CTEST_OPTIONS};-DRMG_BIN='${RMG_BIN}'")
 endif()
 
-if(DEFINED QMC_CUDA)
-  set(CTEST_OPTIONS "${CTEST_OPTIONS};-DQMC_CUDA=${QMC_CUDA}")
-endif()
-
 if(DEFINED ENABLE_CUDA)
   set(CTEST_OPTIONS "${CTEST_OPTIONS};-DENABLE_CUDA=${ENABLE_CUDA}")
 endif()

diff --git a/CMake/macros.cmake b/CMake/macros.cmake
@@ -192,9 +192,9 @@ function(
       APPEND
       PROPERTY LABELS "QMCPACK")
 
-    if(QMC_CUDA
-       OR ENABLE_CUDA
+    if(ENABLE_CUDA
        OR ENABLE_ROCM
+       OR ENABLE_SYCL
        OR ENABLE_OFFLOAD)
       set_tests_properties(${TESTNAME} PROPERTIES RESOURCE_LOCK exclusively_owned_gpus)
     endif()

diff --git a/CMake/test_labels.cmake b/CMake/test_labels.cmake
@@ -1,12 +1,12 @@
 function(ADD_TEST_LABELS TEST_NAME TEST_LABELS)
   set(TEST_LABELS_TEMP "")
-  set(TEST_LABELS_UNIQUE_NAME TEST_LABELS_${TEST_NAME}_${QMC_CUDA}_${QMC_COMPLEX}_${QMC_MIXED_PRECISION})
+  set(TEST_LABELS_UNIQUE_NAME TEST_LABELS_${TEST_NAME}_${QMC_COMPLEX}_${QMC_MIXED_PRECISION})
   if(DEFINED ${TEST_LABELS_UNIQUE_NAME})
     set(TEST_LABELS_TEMP ${${TEST_LABELS_UNIQUE_NAME}})
   else()
     set(SUCCESS FALSE)
     execute_process(
-      COMMAND ${Python3_EXECUTABLE} ${qmcpack_SOURCE_DIR}/tests/scripts/test_labels.py ${TEST_NAME} ${QMC_CUDA}
+      COMMAND ${Python3_EXECUTABLE} ${qmcpack_SOURCE_DIR}/tests/scripts/test_labels.py ${TEST_NAME}
               ${QMC_COMPLEX} ${QMC_MIXED_PRECISION}
       OUTPUT_VARIABLE TEST_LABELS_TEMP
       RESULT_VARIABLE SUCCESS)

diff --git a/CMake/unit_test.cmake b/CMake/unit_test.cmake
@@ -21,9 +21,9 @@ function(ADD_UNIT_TEST TESTNAME PROCS THREADS TEST_BINARY)
     set_tests_properties(${TESTNAME} PROPERTIES PROCESSORS ${TOT_PROCS} ENVIRONMENT OMP_NUM_THREADS=${THREADS}
                                                 PROCESSOR_AFFINITY TRUE)
 
-    if(QMC_CUDA
-       OR ENABLE_CUDA
+    if(ENABLE_CUDA
        OR ENABLE_ROCM
+       OR ENABLE_SYCL
        OR ENABLE_OFFLOAD)
       set_tests_properties(${TESTNAME} PROPERTIES RESOURCE_LOCK exclusively_owned_gpus)
     endif()

diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -52,11 +52,7 @@ include(CMakeDependentOption)
 option(QMC_MPI "Enable/disable MPI" ON)
 option(QMC_OMP "Enable/disable OpenMP" ON)
 option(QMC_COMPLEX "Build for complex binary" OFF)
-option(QMC_CUDA "Build with GPU support through CUDA" OFF)
-option(
-  ENABLE_CUDA
-  "Build with the second generation of GPU support through CUDA (production quality for AFQMC, experimental for real space)"
-  OFF)
+option(ENABLE_CUDA "Build with GPU support through CUDA" OFF)
 option(QMC_CUDA2HIP "Map all CUDA kernels and library calls to HIP" OFF)
 if(QMC_CUDA2HIP OR ENABLE_HIP)
   set(ENABLE_ROCM ON) # option(ENABLE_ROCM) will be no-op

diff --git a/README.md b/README.md
@@ -153,28 +153,26 @@ make -j 8
                                         CMAKE_CXX_FLAGS_RELWITHDEBINFO
 ```
 
- * Key QMC build options
-
-```
-     QMC_CUDA            Enable legacy CUDA code path for NVIDIA GPU acceleration (1:yes, 0:no)
-     QMC_COMPLEX         Build the complex (general twist/k-point) version (1:yes, 0:no)
-     QMC_MIXED_PRECISION Build the mixed precision (mixing double/float) version
-                         (1:yes (GPU default), 0:no (CPU default)).
-                         The CPU support is experimental.
-                         Use float and double for base and full precision.
-                         The GPU support is quite mature.
-                         Use always double for host side base and full precision
-                         and use float and double for CUDA base and full precision.
-     ENABLE_CUDA         ON/OFF(default). Enable CUDA code path for NVIDIA GPU acceleration.
-                         Production quality for AFQMC. Pre-production quality for real-space.
-                         Use CMAKE_CUDA_ARCHITECTURES, default 70, to set the actual GPU architecture.
-     ENABLE_OFFLOAD      ON/OFF(default). Experimental feature. Enable OpenMP target offload for GPU acceleration.
-     ENABLE_TIMERS       ON(default)/OFF. Enable fine-grained timers. Timers are on by default but at level coarse
-                         to avoid potential slowdown in tiny systems.
-                         For systems beyond tiny sizes (100+ electrons) there is no risk.
+ * Key QMCPACK build options
+
 ```
+    QMC_COMPLEX           ON/OFF(default). Build the complex (general twist/k-point) version.
+    QMC_MIXED_PRECISION   ON/OFF(default). Build the mixed precision (mixing double/float) version
+                          Mixed precision calculations can be signifiantly faster but should be
+                          carefully checked validated against full double precision runs,
+                          particularly for large electron counts.
+    ENABLE_OFFLOAD        ON/OFF(default). Enable OpenMP target offload for GPU acceleration.
+    ENABLE_CUDA           ON/OFF(default). Enable CUDA code path for NVIDIA GPU acceleration.
+                          Production quality for AFQMC and real-space performance portable implementation.
+                          Use CMAKE_CUDA_ARCHITECTURES, default 70, to select the actual GPU architecture.
+    QMC_CUDA2HIP          ON/OFF(default). Map all CUDA kernels and library calls to HIP and use ROCm libraries.
+                          Set both ENABLE_CUDA and QMC_CUDA2HIP ON to target AMD GPUs.
+                          Use CMAKE_HIP_ARCHITECTURES, default gfx906, to select the actual GPU architecture.
+    ENABLE_SYCL           ON/OFF(default). Enable SYCL code path. Only support Intel GPUs and OneAPI compilers.
 
- * Additional QMC options
+```
+
+ * Additional QMCPACK options
 
 ```
      QE_BIN              Location of Quantum Espresso binaries including pw2qmcpack.x
@@ -187,6 +185,9 @@ make -j 8
                             saving default use of symbolic links for test files. Useful
                             if the build is on a separate filesystem from the source, as
                             required on some HPC systems.
+     ENABLE_TIMERS       ON(default)/OFF. Enable fine-grained timers. Timers are on by default but at level coarse
+                         to avoid potential slowdown in tiny systems.
+                         For systems beyond tiny sizes (100+ electrons) there is no risk.
 ```
 
   * libxml2 related

diff --git a/config/build_olcf_andes.sh b/config/build_olcf_andes.sh
@@ -15,8 +15,6 @@ export LAPACK_LIBS="$BLAS_LIBS $OLCF_NETLIB_LAPACK_ROOT/lib64/liblapack.a"
 
 declare -A builds=( ["cpu"]="-DBUILD_PPCONVERT=1" \
                     ["complex_cpu"]="-DQMC_COMPLEX=1" \
-#		    ["legacy_gpu"]="-DQMC_CUDA=1 " \
-#		    ["complex_legacy_gpu"]="-DQMC_CUDA=1 -DQMC_COMPLEX=1 " \
 		  )
 
 mkdir bin_andes

diff --git a/config/build_olcf_summit.sh b/config/build_olcf_summit.sh
diff --git a/config/load_olcf_summit_modules.sh b/config/load_olcf_summit_modules.sh
diff --git a/docs/hamiltonianobservable.rst b/docs/hamiltonianobservable.rst
@@ -112,8 +112,6 @@ Many pair potentials are supported.  Though only the most commonly used pair pot
   +------------------+---------+-----------------------------------------------+
   |                  | mpc     | Model periodic Coulomb interaction/correction |
   +------------------+---------+-----------------------------------------------+
-  |                  | cpp     | Core polarization potential                   |
-  +------------------+---------+-----------------------------------------------+
   |                  | skpot   | *Unknown*                                     |
   +------------------+---------+-----------------------------------------------+
 
@@ -148,16 +146,11 @@ Additional information:
    of the classical/quantum ``particleset``.
 
 -  Only ``Coulomb, pseudo``, and ``mpc`` are described in detail in the
-   following subsections. The older or less-used types (``cpp, skpot``)
+   following subsections. The older or less-used types (``skpot``)
    are not covered.
 
--  Available only if ``QMC_CUDA`` is not defined: ``skpot``.
-
 -  Available only if ``OHMMS_DIM==3``: ``mpc, vhxc, pseudo``.
 
--  Available only if ``OHMMS_DIM==3`` and ``QMC_CUDA`` is not defined:
-   ``cpp``.
-
 Coulomb potentials
 ~~~~~~~~~~~~~~~~~~
 

diff --git a/docs/installation.rst b/docs/installation.rst
@@ -282,19 +282,18 @@ the path to the source directory.
 
   ::
 
-    QMC_COMPLEX           Build the complex (general twist/k-point) version (1:yes, 0:no)
-    QMC_MIXED_PRECISION   Build the mixed precision (mixing double/float) version
-                          (1:yes (QMC_CUDA=1 default), 0:no (QMC_CUDA=0 default)).
+    QMC_COMPLEX           ON/OFF(default). Build the complex (general twist/k-point) version.
+    QMC_MIXED_PRECISION   ON/OFF(default). Build the mixed precision (mixing double/float) version
                           Mixed precision calculations can be signifiantly faster but should be
                           carefully checked validated against full double precision runs,
                           particularly for large electron counts.
     ENABLE_OFFLOAD        ON/OFF(default). Enable OpenMP target offload for GPU acceleration.
-    QMC_CUDA              Enable legacy CUDA code path for NVIDIA GPU acceleration (1:yes, 0:no)
     ENABLE_CUDA           ON/OFF(default). Enable CUDA code path for NVIDIA GPU acceleration.
                           Production quality for AFQMC and real-space performance portable implementation.
-                          Use CMAKE_CUDA_ARCHITECTURES, default 70, to set the actual GPU architecture.
-    QMC_CUDA2HIP          ON/OFF(default). To be set ON, it requires either QMC_CUDA or ENABLE_CUDA to be ON.
-                          Compile CUDA source code as HIP and use ROCm libraries for AMD GPUs.
+                          Use CMAKE_CUDA_ARCHITECTURES, default 70, to select the actual GPU architecture.
+    QMC_CUDA2HIP          ON/OFF(default). Map all CUDA kernels and library calls to HIP and use ROCm libraries.
+                          Set both ENABLE_CUDA and QMC_CUDA2HIP ON to target AMD GPUs.
+                          Use CMAKE_HIP_ARCHITECTURES, default gfx906, to select the actual GPU architecture.
     ENABLE_SYCL           ON/OFF(default). Enable SYCL code path. Only support Intel GPUs and OneAPI compilers.
 
 - General build options

diff --git a/examples/molecules/H2O/CMakeLists.txt b/examples/molecules/H2O/CMakeLists.txt
@@ -1,6 +1,6 @@
 # Test example
 
-if(NOT QMC_CUDA AND NOT QMC_COMPLEX)
+if(NOT QMC_COMPLEX)
 
   list(APPEND H2O_SCALARS "totenergy" "-17.26138 .006") # total energy
 

diff --git a/examples/molecules/He/CMakeLists.txt b/examples/molecules/He/CMakeLists.txt
@@ -6,7 +6,7 @@ else()
   set(MP_SUCCESS TRUE)
 endif()
 
-if(NOT QMC_CUDA AND NOT QMC_COMPLEX)
+if(NOT QMC_COMPLEX)
 
   list(APPEND HE_SIMPLE_VMC_SCALARS "totenergy" "-2.83 .006") # total energy
 

diff --git a/src/qmcpack.settings b/src/qmcpack.settings
@@ -13,7 +13,7 @@ BUILD_FCIQMC           = @BUILD_FCIQMC@
 ENABLE_OFFLOAD         = @ENABLE_OFFLOAD@
 ENABLE_CUDA            = @ENABLE_CUDA@
 QMC_CUDA2HIP           = @QMC_CUDA2HIP@
-ENABLE_CUDA            = @ENABLE_SYCL@
+ENABLE_SYCL            = @ENABLE_SYCL@
 ENABLE_PHDF5           = @ENABLE_PHDF5@
 ENABLE_MASS            = @ENABLE_MASS@
 ENABLE_STACKTRACE      = @ENABLE_STACKTRACE@ 

diff --git a/tests/scripts/test_labels.py b/tests/scripts/test_labels.py
@@ -476,12 +476,11 @@ def check_positive_label_sets(positive_label_sets):
 
 # extract test name and build flags from args
 try:
-    full_test,qmc_cuda,qmc_complex,qmc_mixed = sys.argv[1:]
-    qmc_cuda    = qmc_cuda=='1'
+    full_test,qmc_complex,qmc_mixed = sys.argv[1:]
     qmc_complex = qmc_complex=='1'
     qmc_mixed   = qmc_mixed=='1'
-    cpu   = not qmc_cuda
-    gpu   = qmc_cuda
+    cpu   = True
+    gpu   = False # was used for QMC_CUDA which has been removed from the codebase.
     aos   = False
     soa   = True
     real  = not qmc_complex