Merge branch 'branch-22.02' into fea-nvtx-ranges

rapidsai · Dec 16, 2021 · 029c259 · 029c259
2 parents 409c049 + d2e5b7a
commit 029c259
Show file tree

Hide file tree

Showing 144 changed files with 239 additions and 254 deletions.
diff --git a/cpp/.clang-format b/cpp/.clang-format
@@ -72,15 +72,6 @@ ForEachMacros:
   - Q_FOREACH
   - BOOST_FOREACH
 IncludeBlocks: Preserve
-IncludeCategories:
-  - Regex:           '^<ext/.*\.h>'
-    Priority:        2
-  - Regex:           '^<.*\.h>'
-    Priority:        1
-  - Regex:           '^<.*'
-    Priority:        2
-  - Regex:           '.*'
-    Priority:        3
 IncludeIsMainRegex: '([-_](test|unittest))?$'
 IndentCaseLabels: true
 IndentPPDirectives: None

diff --git a/cpp/include/raft/comms/helper.hpp b/cpp/include/raft/comms/helper.hpp
@@ -20,9 +20,9 @@
 #include <raft/handle.hpp>
 #include <raft/mr/device/buffer.hpp>
 
+#include <iostream>
 #include <nccl.h>
 #include <ucp/api/ucp.h>
-#include <iostream>
 
 namespace raft {
 namespace comms {

diff --git a/cpp/include/raft/comms/mpi_comms.hpp b/cpp/include/raft/comms/mpi_comms.hpp
@@ -26,9 +26,9 @@
 #include <mpi.h>
 #include <nccl.h>
 
-#include <raft/cudart_utils.h>
 #include <raft/comms/comms.hpp>
 #include <raft/comms/util.hpp>
+#include <raft/cudart_utils.h>
 #include <raft/error.hpp>
 #include <raft/handle.hpp>
 

diff --git a/cpp/include/raft/comms/std_comms.hpp b/cpp/include/raft/comms/std_comms.hpp
@@ -24,29 +24,29 @@
 
 #include <raft/error.hpp>
 
-#include <raft/cudart_utils.h>
 #include <raft/comms/util.hpp>
+#include <raft/cudart_utils.h>
 
 #include <cuda_runtime.h>
 
+#include <raft/comms/ucp_helper.hpp>
 #include <ucp/api/ucp.h>
 #include <ucp/api/ucp_def.h>
-#include <raft/comms/ucp_helper.hpp>
 
 #include <nccl.h>
 
 #include <unordered_map>
 #include <unordered_set>
 #include <utility>
 
-#include <stdlib.h>
-#include <time.h>
 #include <algorithm>
 #include <chrono>
 #include <cstdio>
 #include <exception>
 #include <memory>
+#include <stdlib.h>
 #include <thread>
+#include <time.h>
 
 namespace raft {
 namespace comms {

diff --git a/cpp/include/raft/cudart_utils.h b/cpp/include/raft/cudart_utils.h
@@ -22,9 +22,9 @@
 
 #include <cuda_runtime.h>
 
-#include <execinfo.h>
 #include <chrono>
 #include <cstdio>
+#include <execinfo.h>
 #include <iomanip>
 #include <iostream>
 #include <mutex>

diff --git a/cpp/include/raft/distance/detail/distance.cuh b/cpp/include/raft/distance/detail/distance.cuh
@@ -17,7 +17,6 @@
 #pragma once
 
 #include <cuda_runtime_api.h>
-#include <raft/linalg/distance_type.h>
 #include <raft/cuda_utils.cuh>
 #include <raft/distance/detail/canberra.cuh>
 #include <raft/distance/detail/chebyshev.cuh>
@@ -31,6 +30,7 @@
 #include <raft/distance/detail/l1.cuh>
 #include <raft/distance/detail/minkowski.cuh>
 #include <raft/distance/detail/russell_rao.cuh>
+#include <raft/linalg/distance_type.h>
 #include <rmm/device_uvector.hpp>
 
 namespace raft {

diff --git a/cpp/include/raft/distance/detail/fused_l2_nn.cuh b/cpp/include/raft/distance/detail/fused_l2_nn.cuh
@@ -16,12 +16,12 @@
 
 #pragma once
 
-#include <stdint.h>
 #include <cub/cub.cuh>
 #include <limits>
 #include <raft/cuda_utils.cuh>
 #include <raft/distance/detail/pairwise_distance_base.cuh>
 #include <raft/linalg/contractions.cuh>
+#include <stdint.h>
 
 namespace raft {
 namespace distance {

diff --git a/cpp/include/raft/distance/detail/pairwise_distance_base.cuh b/cpp/include/raft/distance/detail/pairwise_distance_base.cuh
@@ -14,8 +14,8 @@
  * limitations under the License.
  */
 #pragma once
-#include <raft/cudart_utils.h>
 #include <raft/cuda_utils.cuh>
+#include <raft/cudart_utils.h>
 #include <raft/linalg/contractions.cuh>
 #include <raft/linalg/norm.cuh>
 #include <raft/vectorized.cuh>

diff --git a/cpp/include/raft/distance/distance.hpp b/cpp/include/raft/distance/distance.hpp
@@ -16,9 +16,9 @@
 
 #pragma once
 
-#include <raft/linalg/distance_type.h>
 #include <raft/distance/detail/distance.cuh>
 #include <raft/handle.hpp>
+#include <raft/linalg/distance_type.h>
 #include <rmm/device_uvector.hpp>
 
 namespace raft {

diff --git a/cpp/include/raft/distance/fused_l2_nn.hpp b/cpp/include/raft/distance/fused_l2_nn.hpp
@@ -16,12 +16,12 @@
 
 #pragma once
 
-#include <stdint.h>
 #include <cub/cub.cuh>
 #include <limits>
 #include <raft/cuda_utils.cuh>
 #include <raft/distance/detail/fused_l2_nn.cuh>
 #include <raft/handle.hpp>
+#include <stdint.h>
 
 namespace raft {
 namespace distance {

diff --git a/cpp/include/raft/handle.hpp b/cpp/include/raft/handle.hpp
@@ -32,13 +32,13 @@
 ///@todo: enable once we have migrated cuml-comms layer too
 //#include <common/cuml_comms_int.hpp>
 
+#include "cudart_utils.h"
+#include <raft/comms/comms.hpp>
 #include <raft/linalg/cublas_wrappers.h>
 #include <raft/linalg/cusolver_wrappers.h>
 #include <raft/sparse/cusparse_wrappers.h>
-#include <raft/comms/comms.hpp>
 #include <rmm/cuda_stream_pool.hpp>
 #include <rmm/exec_policy.hpp>
-#include "cudart_utils.h"
 
 namespace raft {
 

diff --git a/cpp/include/raft/label/classlabels.cuh b/cpp/include/raft/label/classlabels.cuh
@@ -18,8 +18,8 @@
 
 #include <cub/cub.cuh>
 
-#include <raft/cudart_utils.h>
 #include <raft/cuda_utils.cuh>
+#include <raft/cudart_utils.h>
 #include <raft/linalg/unary_op.cuh>
 #include <rmm/device_scalar.hpp>
 #include <rmm/device_uvector.hpp>

diff --git a/cpp/include/raft/label/merge_labels.cuh b/cpp/include/raft/label/merge_labels.cuh
@@ -16,12 +16,12 @@
 
 #pragma once
 
-#include <math.h>
 #include <limits>
+#include <math.h>
 
+#include <raft/cuda_utils.cuh>
 #include <raft/cudart_utils.h>
 #include <raft/linalg/init.h>
-#include <raft/cuda_utils.cuh>
 
 namespace raft {
 namespace label {

diff --git a/cpp/include/raft/linalg/cholesky_r1_update.cuh b/cpp/include/raft/linalg/cholesky_r1_update.cuh
@@ -16,11 +16,11 @@
 
 #pragma once
 
-#include <raft/linalg/cublas_wrappers.h>
-#include <raft/linalg/cusolver_wrappers.h>
 #include <raft/cuda_utils.cuh>
 #include <raft/handle.hpp>
 #include <raft/linalg/binary_op.cuh>
+#include <raft/linalg/cublas_wrappers.h>
+#include <raft/linalg/cusolver_wrappers.h>
 
 namespace raft {
 namespace linalg {

diff --git a/cpp/include/raft/linalg/eig.cuh b/cpp/include/raft/linalg/eig.cuh
@@ -17,10 +17,10 @@
 #pragma once
 
 #include <cuda_runtime_api.h>
-#include <raft/cudart_utils.h>
-#include <raft/linalg/cusolver_wrappers.h>
 #include <raft/cuda_utils.cuh>
+#include <raft/cudart_utils.h>
 #include <raft/handle.hpp>
+#include <raft/linalg/cusolver_wrappers.h>
 #include <raft/matrix/matrix.hpp>
 #include <rmm/device_scalar.hpp>
 #include <rmm/device_uvector.hpp>

diff --git a/cpp/include/raft/linalg/gemm.cuh b/cpp/include/raft/linalg/gemm.cuh
@@ -17,9 +17,9 @@
 #pragma once
 
 #include <cublas_v2.h>
-#include <raft/linalg/cublas_wrappers.h>
 #include <raft/cuda_utils.cuh>
 #include <raft/handle.hpp>
+#include <raft/linalg/cublas_wrappers.h>
 
 namespace raft {
 namespace linalg {

diff --git a/cpp/include/raft/linalg/gemv.h b/cpp/include/raft/linalg/gemv.h
@@ -17,8 +17,8 @@
 #pragma once
 
 #include <cublas_v2.h>
-#include <raft/linalg/cublas_wrappers.h>
 #include <raft/cuda_utils.cuh>
+#include <raft/linalg/cublas_wrappers.h>
 
 #include <raft/handle.hpp>
 

diff --git a/cpp/include/raft/linalg/init.h b/cpp/include/raft/linalg/init.h
@@ -16,10 +16,10 @@
 
 #pragma once
 
+#include <rmm/exec_policy.hpp>
 #include <thrust/copy.h>
 #include <thrust/device_ptr.h>
 #include <thrust/iterator/counting_iterator.h>
-#include <rmm/exec_policy.hpp>
 
 namespace raft {
 namespace linalg {

diff --git a/cpp/include/raft/linalg/lanczos.hpp b/cpp/include/raft/linalg/lanczos.hpp
@@ -26,8 +26,8 @@
 #include <curand.h>
 
 #include <raft/cudart_utils.h>
-#include <raft/linalg/cublas_wrappers.h>
 #include <raft/handle.hpp>
+#include <raft/linalg/cublas_wrappers.h>
 #include <raft/spectral/lapack.hpp>
 #include <raft/spectral/matrix_wrappers.hpp>
 #include <raft/spectral/warn_dbg.hpp>

diff --git a/cpp/include/raft/linalg/reduce.cuh b/cpp/include/raft/linalg/reduce.cuh
@@ -16,9 +16,9 @@
 
 #pragma once
 
-#include <raft/cuda_utils.cuh>
 #include "coalesced_reduction.cuh"
 #include "strided_reduction.cuh"
+#include <raft/cuda_utils.cuh>
 
 namespace raft {
 namespace linalg {

diff --git a/cpp/include/raft/linalg/strided_reduction.cuh b/cpp/include/raft/linalg/strided_reduction.cuh
@@ -16,10 +16,10 @@
 
 #pragma once
 
+#include "unary_op.cuh"
 #include <cub/cub.cuh>
 #include <raft/cuda_utils.cuh>
 #include <type_traits>
-#include "unary_op.cuh"
 
 namespace raft {
 namespace linalg {

diff --git a/cpp/include/raft/linalg/subtract.cuh b/cpp/include/raft/linalg/subtract.cuh
@@ -16,9 +16,9 @@
 
 #pragma once
 
-#include <raft/cuda_utils.cuh>
 #include "binary_op.cuh"
 #include "unary_op.cuh"
+#include <raft/cuda_utils.cuh>
 
 namespace raft {
 namespace linalg {

diff --git a/cpp/include/raft/linalg/svd.cuh b/cpp/include/raft/linalg/svd.cuh
@@ -16,19 +16,19 @@
 
 #pragma once
 
-#include <raft/cudart_utils.h>
-#include <raft/linalg/cublas_wrappers.h>
-#include <raft/linalg/cusolver_wrappers.h>
+#include "eig.cuh"
+#include "gemm.cuh"
+#include "transpose.h"
 #include <raft/common/nvtx.hpp>
 #include <raft/cuda_utils.cuh>
+#include <raft/cudart_utils.h>
 #include <raft/handle.hpp>
+#include <raft/linalg/cublas_wrappers.h>
+#include <raft/linalg/cusolver_wrappers.h>
 #include <raft/matrix/math.hpp>
 #include <raft/matrix/matrix.hpp>
 #include <rmm/device_scalar.hpp>
 #include <rmm/device_uvector.hpp>
-#include "eig.cuh"
-#include "gemm.cuh"
-#include "transpose.h"
 
 namespace raft {
 namespace linalg {

diff --git a/cpp/include/raft/linalg/transpose.h b/cpp/include/raft/linalg/transpose.h
@@ -16,8 +16,8 @@
 
 #pragma once
 
-#include <raft/linalg/cublas_wrappers.h>
 #include <raft/handle.hpp>
+#include <raft/linalg/cublas_wrappers.h>
 #include <rmm/exec_policy.hpp>
 
 namespace raft {

diff --git a/cpp/include/raft/linalg/unary_op.cuh b/cpp/include/raft/linalg/unary_op.cuh
@@ -16,8 +16,8 @@
 
 #pragma once
 
-#include <raft/cudart_utils.h>
 #include <raft/cuda_utils.cuh>
+#include <raft/cudart_utils.h>
 #include <raft/vectorized.cuh>
 
 namespace raft {

diff --git a/cpp/include/raft/matrix/detail/math.cuh b/cpp/include/raft/matrix/detail/math.cuh
@@ -114,4 +114,4 @@ void signFlip(math_t* inout, int n_rows, int n_cols, cudaStream_t stream)
 
 }  // end namespace detail
 }  // end namespace matrix
-}  // end namespace raft
+}  // end namespace raft
diff --git a/cpp/include/raft/matrix/math.hpp b/cpp/include/raft/matrix/math.hpp
@@ -240,17 +240,7 @@ void reciprocal(math_t* in,
     d_dest,
     d_src,
     len,
-    [=] __device__(math_t a) {
-      if (setzero) {
-        if (abs(a) <= thres) {
-          return math_t(0);
-        } else {
-          return scalar / a;
-        }
-      } else {
-        return scalar / a;
-      }
-    },
+    [=] __device__(math_t a) { return setzero && (abs(a) <= thres) ? math_t{0} : scalar / a; },
     stream);
 }
 

diff --git a/cpp/include/raft/matrix/matrix.hpp b/cpp/include/raft/matrix/matrix.hpp
@@ -18,13 +18,13 @@
 
 #include "detail/matrix.cuh"
 
+#include <algorithm>
+#include <cstddef>
 #include <cuda_runtime.h>
 #include <cusolverDn.h>
 #include <raft/cudart_utils.h>
-#include <raft/linalg/cublas_wrappers.h>
-#include <algorithm>
-#include <cstddef>
 #include <raft/handle.hpp>
+#include <raft/linalg/cublas_wrappers.h>
 
 namespace raft {
 namespace matrix {

diff --git a/cpp/include/raft/mr/device/buffer.hpp b/cpp/include/raft/mr/device/buffer.hpp
@@ -16,9 +16,9 @@
 
 #pragma once
 
+#include "allocator.hpp"
 #include <memory>
 #include <raft/mr/buffer_base.hpp>
-#include "allocator.hpp"
 
 namespace raft {
 namespace mr {