Require CUDA 12.2+

Author: jakirkham

ci/build_cpp.sh

@@ -26,42 +26,10 @@ export RAPIDS_ARTIFACTS_DIR
 # populates `RATTLER_CHANNELS` array and `RATTLER_ARGS` array
 source rapids-rattler-channel-string
 
-# Construct the extra variants according to the architecture
-if [[ "$(arch)" == "x86_64" ]]; then
-    cat > variants.yaml << EOF
-    c_compiler_version:
-      - 14
-
-    cxx_compiler_version:
-      - 14
-
-    cuda_version:
-      - ${RAPIDS_CUDA_VERSION%.*}
-EOF
-else
-    cat > variants.yaml << EOF
-    zip_keys:
-    - [c_compiler_version, cxx_compiler_version, cuda_version]
-
-    c_compiler_version:
-    - 12
-    - 14
-
-    cxx_compiler_version:
-    - 12
-    - 14
-
-    cuda_version:
-    - 12.1 # The last version to not support cufile
-    - ${RAPIDS_CUDA_VERSION%.*}
-EOF
-fi
-
 # --no-build-id allows for caching with `sccache`
 # more info is available at
 # https://rattler.build/latest/tips_and_tricks/#using-sccache-or-ccache-with-rattler-build
 rattler-build build --recipe conda/recipes/libkvikio \
-                    --variant-config variants.yaml \
                     "${RATTLER_ARGS[@]}" \
                     "${RATTLER_CHANNELS[@]}"
 

conda/recipes/libkvikio/recipe.yaml

@@ -5,12 +5,6 @@ schema_version: 1
 context:
   version: ${{ env.get("RAPIDS_PACKAGE_VERSION") }}
   minor_version: ${{ (version | split("."))[:2] | join(".") }}
-  # We need to support three cases:
-  # 1. Linux x86_64, which always uses libcufile
-  # 2. Linux aarch64 with CUDA >= 12.2, which uses libcufile
-  # 3. Linux aarch64 with CUDA < 12.2, which does not use libcufile
-  # Each case has different cuda-version constraints as expressed below
-  should_use_cufile: ${{ x86_64 or (aarch64 and cuda_version >= "12.2") }}
   # When reverting, instances of cuda_key_string can be replaced with cuda_major
   cuda_key_string: ${{ cuda_version | replace(".", "_") }}
   #cuda_version: ${{ (env.get("RAPIDS_CUDA_VERSION") | split("."))[:2] | join(".") }}
@@ -67,9 +61,7 @@ cache:
     host:
       - cuda-version =${{ cuda_version }}
       - libcurl ==${{ libcurl_version }}
-      - if: should_use_cufile
-        then:
-        - libcufile-dev
+      - libcufile-dev
       - libnuma
 
 outputs:
@@ -94,24 +86,16 @@ outputs:
         - cuda-version =${{ cuda_version }}
         - libcurl ==${{ libcurl_version }}
       run:
-        - if: x86_64 or (aarch64 and cuda_version >= "13.0")
+        - if: cuda_version >= "13.0"
           then:
           - ${{ pin_compatible("cuda-version", upper_bound="x", lower_bound="x") }}
           else:
-          - if: aarch64 and cuda_version >= "12.2"
-            then:
-            - ${{ pin_compatible("cuda-version", upper_bound="x", lower_bound="12.2.0a0") }}
-            else:
-            - ${{ pin_compatible("cuda-version", upper_bound="12.2.0a0", lower_bound="12.0") }}
-        - if: should_use_cufile
-          then:
-          - libcufile-dev
+          - ${{ pin_compatible("cuda-version", upper_bound="x", lower_bound="12.2.0a0") }}
+        - libcufile-dev
       ignore_run_exports:
         by_name:
           - cuda-version
-          - if: should_use_cufile
-            then:
-            - libcufile
+          - libcufile
     tests:
       - script:
           - test -f $PREFIX/include/kvikio/file_handle.hpp
@@ -139,28 +123,20 @@ outputs:
         - cuda-version =${{ cuda_version }}
         - cuda-cudart-dev
         - libcurl ==${{ libcurl_version }}
-        - if: should_use_cufile
-          then:
-          - libcufile-dev
+        - libcufile-dev
       run:
-        - if: x86_64
+        - if: cuda_version >= "13.0"
           then:
           - ${{ pin_compatible("cuda-version", upper_bound="x", lower_bound="x") }}
           else:
-          - if: aarch64 and cuda_version >= "12.2"
-            then:
-            - ${{ pin_compatible("cuda-version", upper_bound="x", lower_bound="12.2.0a0") }}
-            else:
-            - ${{ pin_compatible("cuda-version", upper_bound="12.2.0a0", lower_bound="12.0") }}
+          - ${{ pin_compatible("cuda-version", upper_bound="x", lower_bound="12.2.0a0") }}
         - cuda-cudart
       ignore_run_exports:
         by_name:
           - cuda-cudart
           - cuda-version
           - libnuma
-          - if: should_use_cufile
-            then:
-            - libcufile
+          - libcufile
     about:
       homepage: ${{ load_from_file("python/libkvikio/pyproject.toml").project.urls.Homepage }}
       license: ${{ load_from_file("python/libkvikio/pyproject.toml").project.license.text }}

cpp/include/kvikio/file_handle.hpp

@@ -283,9 +283,6 @@ class FileHandle {
    * This is an asynchronous version of `.read()`, which will be executed in sequence
    * for the specified stream.
    *
-   * When running CUDA v12.1 or older, this function falls back to use `.read()` after
-   * `stream` has been synchronized.
-   *
    * The arguments have the same meaning as in `.read()` but some of them are deferred.
    * That is, the values pointed to by `size_p`, `file_offset_p` and `devPtr_offset_p`
    * will not be evaluated until execution time. Notice, this behavior can be changed
@@ -324,9 +321,6 @@ class FileHandle {
    * This is an asynchronous version of `.read()`, which will be executed in sequence
    * for the specified stream.
    *
-   * When running CUDA v12.1 or older, this function falls back to use `.read()` after
-   * `stream` has been synchronized.
-   *
    * The arguments have the same meaning as in `.read()` but returns a `StreamFuture` object
    * that the caller must keep alive until all data has been read from disk. One way to do this,
    * is by calling `StreamFuture.check_bytes_done()`, which will synchronize the associated stream
@@ -355,9 +349,6 @@ class FileHandle {
    * This is an asynchronous version of `.write()`, which will be executed in sequence
    * for the specified stream.
    *
-   * When running CUDA v12.1 or older, this function falls back to use `.read()` after
-   * `stream` has been synchronized.
-   *
    * The arguments have the same meaning as in `.write()` but some of them are deferred.
    * That is, the values pointed to by `size_p`, `file_offset_p` and `devPtr_offset_p`
    * will not be evaluated until execution time. Notice, this behavior can be changed
@@ -397,9 +388,6 @@ class FileHandle {
    * This is an asynchronous version of `.write()`, which will be executed in sequence
    * for the specified stream.
    *
-   * When running CUDA v12.1 or older, this function falls back to use `.read()` after
-   * `stream` has been synchronized.
-   *
    * The arguments have the same meaning as in `.write()` but returns a `StreamFuture` object
    * that the caller must keep alive until all data has been written to disk. One way to do this,
    * is by calling `StreamFuture.check_bytes_done()`, which will synchronize the associated stream

cpp/src/batch.cpp

@@ -15,8 +15,6 @@
 
 namespace kvikio {
 
-#ifdef KVIKIO_CUFILE_BATCH_API_FOUND
-
 BatchHandle::BatchHandle(int max_num_events) : _initialized{true}, _max_num_events{max_num_events}
 {
   CUFILE_TRY(cuFileAPI::instance().BatchIOSetUp(&_handle, max_num_events));
@@ -78,28 +76,4 @@ std::vector<CUfileIOEvents_t> BatchHandle::status(unsigned min_nr,
 
 void BatchHandle::cancel() { CUFILE_TRY(cuFileAPI::instance().BatchIOCancel(_handle)); }
 
-#else
-
-BatchHandle::BatchHandle(int max_num_events)
-{
-  KVIKIO_FAIL("BatchHandle requires cuFile's batch API, please build with CUDA v12.1+");
-}
-
-bool BatchHandle::closed() const noexcept { return true; }
-
-void BatchHandle::close() noexcept {}
-
-void BatchHandle::submit(std::vector<BatchOp> const& operations) {}
-
-std::vector<CUfileIOEvents_t> BatchHandle::status(unsigned min_nr,
-                                                  unsigned max_nr,
-                                                  struct timespec* timeout)
-{
-  return std::vector<CUfileIOEvents_t>{};
-}
-
-void BatchHandle::cancel() {}
-
-#endif
-
 }  // namespace kvikio

dependencies.yaml

@@ -180,10 +180,6 @@ dependencies:
     specific:
       - output_types: conda
         matrices:
-          - matrix:
-              cuda: "12.0"
-            packages:
-              - cuda-version=12.0
           - matrix:
               cuda: "12.2"
             packages:

java/pom.xml

@@ -24,7 +24,7 @@ SPDX-License-Identifier: Apache-2.0
     <maven.compiler.source>11</maven.compiler.source>
     <maven.compiler.target>11</maven.compiler.target>
     <junit.version>5.4.2</junit.version>
-    <jcuda.version>12.0.0</jcuda.version>
+    <jcuda.version>12.2.2</jcuda.version>
     <cmake.version>3.31.5-b1</cmake.version>
   </properties>