Add autotuning candidate filtering controls. Add MPI backend disablement option to autotune options struct.

docs/api/f_api.rst

@@ -70,6 +70,7 @@ _________________________________
   :f cudecompAutotuneGridMode grid_mode: which communication (transpose/halo) to use to autotune process grid (default: CUDECOMP_AUTOTUNE_GRID_TRANSPOSE)
   :f cudecompDataType dtype: datatype to use during autotuning (default: CUDECOMP_DOUBLE)
   :f logical allow_uneven_distributions: flag to control whether autotuning allows process grids that result in uneven distributions of elements across processes (default: true)
+  :f logical disable_mpi_backends: flag to disable MPI backend options during autotuning (default: false)
   :f logical disable_nccl_backends: flag to disable NCCL backend options during autotuning (default: false)
   :f logical disable_nvshmem_backends: flag to disable NVSHMEM backend options during autotuning (default: false)
   :f real(c_double) skip_threshold: threshold used to skip testing slow configurations; skip configuration if :code:`skip_threshold * t > t_best`, where :code:`t` is the duration of the first timed trial for the configuration and :code:`t_best` is the average trial time of the current best configuration (default: 0.0)

docs/autotuning.rst

@@ -117,19 +117,22 @@ The :code:`dtype` entry in the options struct controls which data type cuDecomp
 
     options%dtype = CUDECOMP_DOUBLE
 
-The :code:`disable_nccl_backends` and :code:`disable_nvshmem_backends` entries are boolean flags controlling whether
-the autotuner will test transpose and halo communication backends using the NCCL or NVSHMEM libraries respectively.
-By default, these flags are set to false and NCCL and NVSHMEM backends are enabled.
+The :code:`disable_mpi_backends`, :code:`disable_nccl_backends`, and :code:`disable_nvshmem_backends` entries are
+boolean flags controlling whether the autotuner will test transpose and halo communication backends using the MPI,
+NCCL, or NVSHMEM libraries respectively. These exclusions are applied after any backend environment-variable filter.
+By default, these flags are set to false and all three backend families are enabled.
 
 .. tabs::
 
   .. code-tab:: c++
 
+    options.disable_mpi_backends = false;
     options.disable_nccl_backends = false;
     options.disable_nvshmem_backends = false;
 
   .. code-tab:: fortran
 
+    options%disable_mpi_backends = .false.
     options%disable_nccl_backends = .false.
     options%disable_nvshmem_backends = .false.
 

docs/env_vars.rst

@@ -104,3 +104,43 @@ CUDECOMP_USE_COL_MAJOR_RANK_ORDER
 When :code:`rank_order` is left at :code:`CUDECOMP_RANK_ORDER_DEFAULT`, this environment variable is still honored for
 backward compatibility. Setting this variable to :code:`1` enables column-major rank assignment; otherwise rank assignment
 is row-major. Explicit :code:`rank_order` settings take precedence over this environment variable.
+
+CUDECOMP_AUTOTUNE_TRANSPOSE_BACKENDS
+------------------------------------
+(since v0.7.0)
+
+:code:`CUDECOMP_AUTOTUNE_TRANSPOSE_BACKENDS` restricts the candidates used when transpose communication backends are
+autotuned. Its value is a comma-separated inclusion list of uppercase backend names. Prefixing the complete value with
+:code:`^` changes it to an exclusion list. Accepted names are :code:`MPI_P2P`, :code:`MPI_P2P_PL`, :code:`MPI_A2A`,
+:code:`NCCL`, :code:`NCCL_PL`, :code:`NVSHMEM`, :code:`NVSHMEM_PL`, and :code:`NVSHMEM_SM`.
+
+Default setting is unset (no additional backend filtering).
+
+CUDECOMP_AUTOTUNE_HALO_BACKENDS
+-------------------------------
+(since v0.7.0)
+
+:code:`CUDECOMP_AUTOTUNE_HALO_BACKENDS` restricts the candidates used when halo communication backends are autotuned.
+Its value is a comma-separated inclusion list of uppercase backend names. Prefixing the complete value with :code:`^`
+changes it to an exclusion list. Accepted names are :code:`MPI`, :code:`MPI_BLOCKING`, :code:`NCCL`, :code:`NVSHMEM`,
+and :code:`NVSHMEM_BLOCKING`.
+
+Default setting is unset (no additional backend filtering).
+
+CUDECOMP_AUTOTUNE_P_ROW_RANGE
+------------------------------
+(since v0.7.0)
+
+:code:`CUDECOMP_AUTOTUNE_P_ROW_RANGE` restricts process-grid autotuning to row dimensions in the inclusive range
+:code:`<min>,<max>`.
+
+Default setting is unset (no row-dimension restriction).
+
+CUDECOMP_AUTOTUNE_P_COL_RANGE
+------------------------------
+(since v0.7.0)
+
+:code:`CUDECOMP_AUTOTUNE_P_COL_RANGE` restricts process-grid autotuning to column dimensions in the inclusive range
+:code:`<min>,<max>`.
+
+Default setting is unset (no column-dimension restriction).

include/cudecomp.h

@@ -1,5 +1,5 @@
 /*
- * SPDX-FileCopyrightText: Copyright (c) 2022-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+ * SPDX-FileCopyrightText: Copyright (c) 2022-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
  * SPDX-License-Identifier: Apache-2.0
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
@@ -158,6 +158,7 @@ typedef struct {
   cudecompDataType_t dtype;             ///< datatype to use during autotuning (default: CUDECOMP_DOUBLE)
   bool allow_uneven_decompositions; ///< flag to control whether autotuning allows process grids that result in uneven
                                     ///< distributions of elements across processes (default: true)
+  bool disable_mpi_backends;        ///< flag to disable MPI backend options during autotuning (default: false)
   bool disable_nccl_backends;       ///< flag to disable NCCL backend options during autotuning (default: false)
   bool disable_nvshmem_backends;    ///< flag to disable NVSHMEM backend options during autotuning (default: false)
   double skip_threshold;            ///< threshold used to skip testing slow configurations; skip configuration

include/internal/autotune.h

@@ -1,5 +1,5 @@
 /*
- * SPDX-FileCopyrightText: Copyright (c) 2022-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+ * SPDX-FileCopyrightText: Copyright (c) 2022-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
  * SPDX-License-Identifier: Apache-2.0
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
@@ -18,9 +18,19 @@
 #ifndef CUDECOMP_AUTOTUNE_H
 #define CUDECOMP_AUTOTUNE_H
 
+#include <array>
+#include <cstdint>
+#include <vector>
+
 #include "cudecomp.h"
 
 namespace cudecomp {
+std::vector<cudecompTransposeCommBackend_t>
+getAutotuneTransposeBackendCandidates(const cudecompGridDescAutotuneOptions_t* options);
+std::vector<cudecompHaloCommBackend_t>
+getAutotuneHaloBackendCandidates(const cudecompGridDescAutotuneOptions_t* options);
+std::vector<std::array<int32_t, 2>> getAutotunePdimCandidates(int nranks, cudecompRankOrder_t rank_order);
+
 void autotuneTransposeBackend(cudecompHandle_t handle, cudecompGridDesc_t grid_desc,
                               const cudecompGridDescAutotuneOptions_t* options);
 void autotuneHaloBackend(cudecompHandle_t handle, cudecompGridDesc_t grid_desc,