BUILD: Add stub include path for gdaki (#10971)

* BULID: Add gdaki header to stub include path

* BUILD: add stub into package path

---------

Co-authored-by: Zihao Zhao <[email protected]>

Author: jeynmann

Makefile.am

@@ -41,7 +41,8 @@ SUBDIRS = \
 	src/ucm \
 	src/ucs \
 	src/uct \
-	src/ucp
+	src/ucp \
+	stubs
 
 if HAVE_UCG
 SUBDIRS += $(UCG_SUBDIR)

config/m4/cuda.m4

@@ -232,4 +232,10 @@ AS_IF([test "x$cuda_checked" != "xyes"],
         UCX_CUDA_CHECK_NVCC
    ]) # "x$cuda_checked" != "xyes"
 
+   # Set stubs include path for pkg-config
+   # e.g. UCX Device API headers include <uct/ib/mlx5/gdaki/gdaki.cuh>
+   # from a separate package. Stubs provide fallback when not installed.
+   STUB_CFLAGS="-I\${includedir}/stubs"
+   AC_SUBST([STUB_CFLAGS])
+
 ]) # UCX_CHECK_CUDA

configure.ac

@@ -367,6 +367,7 @@ AC_SUBST([build_bindings], [${build_bindings}])
 # Final output
 #
 AC_CONFIG_FILES([Makefile
+                 stubs/Makefile
                  docs/doxygen/header.tex
                  src/uct/api/version.h
                  ])

stubs/Makefile.am

@@ -0,0 +1,13 @@
+#
+# Copyright (c) NVIDIA CORPORATION & AFFILIATES, 2025. ALL RIGHTS RESERVED.
+# See file LICENSE for terms.
+#
+
+# Stub headers: provide fallback if required package not installed.
+
+stubincludedir = $(includedir)/stubs
+
+# install gdaki stub headers in case of ucx-ib-mlx5-gda not installed.
+nobase_stubinclude_HEADERS = \
+	uct/ib/mlx5/gdaki/gdaki.cuh \
+	uct/ib/mlx5/gdaki/gdaki_dev.h

stubs/uct/ib/mlx5/gdaki/gdaki.cuh

@@ -0,0 +1,100 @@
+/**
+ * Copyright (c) NVIDIA CORPORATION & AFFILIATES, 2025. ALL RIGHTS RESERVED.
+ * See file LICENSE for terms.
+ */
+
+#ifndef UCT_GDAKI_CUH_H
+#define UCT_GDAKI_CUH_H
+
+/*
+ * Stub implementation for GDAKI (GPU Direct Async Kernel Interface).
+ * This file provides stub functions when DOCA GPUNetIO package is not available.
+ * If DOCA GPUNetIO is installed, the real implementation from that package
+ * will be used instead (via higher priority include path).
+ */
+
+#include "gdaki_dev.h"
+#include <ucs/sys/device_code.h>
+#include <ucs/type/status.h>
+
+/**
+ * Stub implementation: put_single operation
+ * Returns UCS_ERR_UNSUPPORTED as GDAKI is not available
+ */
+template<ucs_device_level_t level>
+UCS_F_DEVICE ucs_status_t uct_rc_mlx5_gda_ep_put_single(
+        uct_device_ep_h tl_ep, const uct_device_mem_element_t *tl_mem_elem,
+        const void *address, uint64_t remote_address, size_t length,
+        uint64_t flags, uct_device_completion_t *comp)
+{
+    return UCS_ERR_UNSUPPORTED;
+}
+
+/**
+ * Stub implementation: atomic_add operation
+ * Returns UCS_ERR_UNSUPPORTED as GDAKI is not available
+ */
+template<ucs_device_level_t level>
+UCS_F_DEVICE ucs_status_t uct_rc_mlx5_gda_ep_atomic_add(
+        uct_device_ep_h tl_ep, const uct_device_mem_element_t *tl_mem_elem,
+        uint64_t value, uint64_t remote_address, uint64_t flags,
+        uct_device_completion_t *comp)
+{
+    return UCS_ERR_UNSUPPORTED;
+}
+
+/**
+ * Stub implementation: put_multi operation
+ * Returns UCS_ERR_UNSUPPORTED as GDAKI is not available
+ */
+template<ucs_device_level_t level>
+UCS_F_DEVICE ucs_status_t uct_rc_mlx5_gda_ep_put_multi(
+        uct_device_ep_h tl_ep, const uct_device_mem_element_t *tl_mem_list,
+        unsigned mem_list_count, void *const *addresses,
+        const uint64_t *remote_addresses, const size_t *lengths,
+        uint64_t counter_inc_value, uint64_t counter_remote_address,
+        uint64_t flags, uct_device_completion_t *tl_comp)
+{
+    return UCS_ERR_UNSUPPORTED;
+}
+
+/**
+ * Stub implementation: put_multi_partial operation
+ * Returns UCS_ERR_UNSUPPORTED as GDAKI is not available
+ */
+template<ucs_device_level_t level>
+UCS_F_DEVICE ucs_status_t uct_rc_mlx5_gda_ep_put_multi_partial(
+        uct_device_ep_h tl_ep, const uct_device_mem_element_t *tl_mem_list,
+        const unsigned *mem_list_indices, unsigned mem_list_count,
+        void *const *addresses, const uint64_t *remote_addresses,
+        const size_t *local_offsets, const size_t *remote_offsets,
+        const size_t *lengths, unsigned counter_index,
+        uint64_t counter_inc_value, uint64_t counter_remote_address,
+        uint64_t flags, uct_device_completion_t *tl_comp)
+{
+    return UCS_ERR_UNSUPPORTED;
+}
+
+/**
+ * Stub implementation: endpoint progress
+ * No-op as GDAKI is not available
+ */
+template<ucs_device_level_t level>
+UCS_F_DEVICE void uct_rc_mlx5_gda_ep_progress(uct_device_ep_h tl_ep)
+{
+    /* No-op stub */
+}
+
+/**
+ * Stub implementation: check completion
+ * Returns UCS_ERR_UNSUPPORTED as GDAKI is not available
+ */
+template<ucs_device_level_t level>
+UCS_F_DEVICE ucs_status_t uct_rc_mlx5_gda_ep_check_completion(
+        uct_device_ep_h tl_ep, uct_device_completion_t *tl_comp)
+{
+    return UCS_ERR_UNSUPPORTED;
+}
+
+#endif /* UCT_GDAKI_CUH_H */
+

stubs/uct/ib/mlx5/gdaki/gdaki_dev.h

@@ -0,0 +1,14 @@
+/**
+ * Copyright (c) NVIDIA CORPORATION & AFFILIATES, 2025. ALL RIGHTS RESERVED.
+ * See file LICENSE for terms.
+ */
+
+#ifndef UCT_GDAKI_DEV_H
+#define UCT_GDAKI_DEV_H
+
+#include <uct/api/device/uct_device_types.h>
+
+typedef struct {
+} uct_rc_gda_completion_t;
+
+#endif /* UCT_GDAKI_DEV_H */

ucx.pc.in

@@ -13,7 +13,7 @@ includedir = @includedir@
 Name: @PACKAGE@
 Description: Unified Communication X Library
 Version: @VERSION@
-Cflags: -I${includedir}
+Cflags: -I${includedir} @STUB_CFLAGS@
 Libs: -L${libdir} -lucp
 Libs.private: -Wl,--undefined=ucp_global_init
 Requires: @PACKAGE@-uct, @PACKAGE@-ucs

ucx.spec.in

@@ -166,6 +166,7 @@ rm -f %{buildroot}%{_libdir}/ucx/lib*.so
 
 %files devel
 %{_includedir}/uc*
+%{_includedir}/stubs
 %{_libdir}/lib*.so
 %{_libdir}/pkgconfig/ucx.pc
 %{_libdir}/pkgconfig/ucx-uct.pc