Added Intel Gaudi GDR support for UCX

AUTHORS

@@ -15,9 +15,12 @@ Artem Ryabov <[email protected]>
 Artemy Kovalyov <[email protected]>
 Arun Chandran <[email protected]>
 Aurelien Bouteiller <[email protected]>
+Benson Muite <[email protected]>
 Bin Lei <[email protected]>
 Boris Karasev <[email protected]>
 Brad Benton <[email protected]>
+Brian Sheng <[email protected]>
+Bruno Faccini <[email protected]>
 Changcheng Liu <[email protected]>
 Colin Hirsch <[email protected]>
 Corey J. Nolet <[email protected]>
@@ -29,17 +32,20 @@ Dmitrii Gabor <[email protected]>
 Dmitry Gladkov <[email protected]>
 Doug Jacobsen <[email protected]>
 Edgar Gabriel <[email protected]>
+Elad Guttel <[email protected]>
 Elad Persiko <[email protected]>
 Eugene Voronov <[email protected]>
 Evgeny Leksikov <[email protected]>
 Fabian Ruhland <[email protected]>
+Felix Abecassis <[email protected]>
 Gilbert Lee <[email protected]>
 Gilles Gouaillardet <[email protected]>
 Gonzalo Brito Gadeschi <[email protected]>
 Graham Lopez <[email protected]>
 Guy Shattah <[email protected]>
 Hessam Mirsadeghi <[email protected]>
 Hiroyuki Sato <[email protected]>
+Honggang Li <[email protected]>
 Howard Pritchard <[email protected]>
 Huaxiang Fan <[email protected]>
 Hui Zhou <[email protected]>
@@ -48,13 +54,16 @@ Ilia Yastrebov <[email protected]>
 Ilya Nelkenbaum <[email protected]>
 Itay Alroy <[email protected]>
 Ivan Kochin <[email protected]>
+JKLiang9714 <[email protected]>
 Jakir Kham <[email protected]>
+Jan Ciesko <[email protected]>
 Jason Gunthorpe <[email protected]>
 Jeff Daily <[email protected]>
 Jianxin Xiong <[email protected]>
 John Snyder <[email protected]>
 Jonas Zhou <[email protected]>
 Joseph Schuchart <[email protected]>
+Kaidrikov Evgeny <[email protected]>
 Keisuke Fukuda <[email protected]>
 Ken Raffenetti <[email protected]>
 Khaled Hamidouche <[email protected]>
@@ -66,33 +75,40 @@ Manjunath Gorentla Venkata <[email protected]>
 Marek Schimara <[email protected]>
 Mark Allen <[email protected]>
 Matthew Baker <[email protected]>
+Matthias Diener <[email protected]>
 Michael Braverman <[email protected]>
 Michal Shalev <[email protected]>
 Mike Dubman <[email protected]>
 Mikhail Brinskii <[email protected]>
+Mikhail Brinskiy <[email protected]>
 Min Fang <[email protected]>
 Nathan Hjelm <[email protected]>
 Netanel Yosephian <[email protected]>
+Nysal Jan K.A <[email protected]>
 Ofir Farjon <[email protected]>
 Olly Perks <[email protected]>
 Ovidiu Mara <[email protected]>
 Pak Lui <[email protected]>
+Pasha (Pavel) Shamis <[email protected]>
 Pavan Balaji <[email protected]>
 Pavel Shamis (Pasha) <[email protected]>
 Peter Andreas Entschev <[email protected]>
 Peter Rudenko <[email protected]>
 Peter-Jan Gootzen <[email protected]>
 Qiang Yu <[email protected]>
 Raul Akhmetshin <[email protected]>
+Rob Bradford <[email protected]>
 Robert Dietrich <[email protected]>
 Rohit Zambre <[email protected]>
 Roie Danino <[email protected]>
 Romain Pereira <[email protected]>
 Sam James <[email protected]>
+Saravanan Vajravel <[email protected]>
 Sasha Kotchubievsky <[email protected]>
 Scott Saulters <[email protected]>
 Sergey Lebedev <[email protected]>
 Sergey Oblomov <[email protected]>
+Sergey Oblomov [email protected] () <[email protected]>
 Sergey Shalnov <[email protected]>
 Serguei Sagalovitch <[email protected]>
 Shachar Hasson <[email protected]>
@@ -104,17 +120,32 @@ Stephen Richmond <[email protected]>
 Swen Boehm <[email protected]>
 Thomas Vegas <[email protected]>
 Tony Curtis <[email protected]>
+Tooraj Taraz <[email protected]>
 Tzafrir Cohen <[email protected]>
 Valentin Petrov <[email protected]>
+Vasily Philipov <[email protected]>
 Wenbin Lu <[email protected]>
+Xiang Gao <[email protected]>
 Xin Zhao <[email protected]>
 Xu Yifeng <[email protected]>
 Yiltan Hassan Temucin <[email protected]>
 Yossi Itigin <[email protected]>
 Yuriy Shestakov <[email protected]>
+Zhongkai Zhang <[email protected]>
 Zhu Yanjun <[email protected]>
 Zihao Zhao <[email protected]>
+akolliasAMD <[email protected]>
+akshay-venkatesh <[email protected]>
+chenyidu <[email protected]>
+dmitrygx <[email protected]>
+esoha-nvidia <[email protected]>
+lyu <[email protected]>
 lzhang2 <[email protected]>
+nileshnegi <[email protected]>
+panda1100 <[email protected]>
+razor1991 <[email protected]>
+root <[email protected]>
+shasson5 <[email protected]>
 
 In addition we would like to acknowledge the following members of UCX community
 for their participation in annual face-to-face meeting, design discussions, and

config/m4/gaudi.m4

@@ -0,0 +1,71 @@
+#
+# Copyright (c) Intel Corporation, 2025. ALL RIGHTS RESERVED.
+# See file LICENSE for terms.
+#
+
+AC_DEFUN([UCX_CHECK_GAUDI],[
+
+AS_IF([test "x$gaudi_checked" != "xyes"],
+   [
+    AC_ARG_WITH([gaudi],
+                [AS_HELP_STRING([--with-gaudi=(DIR)], [Enable the use of GAUDI (default is guess).])],
+                [], [with_gaudi=guess])
+
+    AS_IF([test "x$with_gaudi" = "xno"],
+        [
+         gaudi_happy="no"
+        ],
+        [
+         save_CPPFLAGS="$CPPFLAGS"
+         save_LDFLAGS="$LDFLAGS"
+         save_LIBS="$LIBS"
+
+         GAUDI_CPPFLAGS=""
+         GAUDI_LDFLAGS=""
+         GAUDI_LIBS=""
+
+         AS_IF([test ! -z "$with_gaudi" -a "x$with_gaudi" != "xyes" -a "x$with_gaudi" != "xguess"],
+               [ucx_check_gaudi_dir="$with_gaudi"
+                ucx_check_gaudi_libdir="$with_gaudi/lib/habanalabs"
+                GAUDI_CPPFLAGS="-I$with_gaudi/include/habanalabs -I/usr/include/drm -I/usr/include/libdrm"
+                GAUDI_LDFLAGS="-L$ucx_check_gaudi_libdir"],
+               [GAUDI_CPPFLAGS="-I/usr/include/habanalabs -I/usr/include/drm -I/usr/include/libdrm"
+                GAUDI_LDFLAGS="-L/usr/lib/habanalabs"])
+
+         AS_IF([test ! -z "$with_gaudi_libdir" -a "x$with_gaudi_libdir" != "xyes"],
+               [ucx_check_gaudi_libdir="$with_gaudi_libdir"
+                GAUDI_LDFLAGS="-L$ucx_check_gaudi_libdir"])
+
+         CPPFLAGS="$CPPFLAGS $GAUDI_CPPFLAGS"
+         LDFLAGS="$LDFLAGS $GAUDI_LDFLAGS"
+
+         # Check gaudi header files
+         AC_CHECK_HEADERS([hlthunk.h],
+                          [gaudi_happy="yes"], [gaudi_happy="no"])
+
+         # Check gaudi libraries
+         AS_IF([test "x$gaudi_happy" = "xyes"],
+               [AC_CHECK_LIB([hl-thunk], [hlthunk_open],
+                             [GAUDI_LIBS="$GAUDI_LIBS -lhl-thunk -lscal -lSynapse -lSynapseMme"], [gaudi_happy="no"])])
+
+         CPPFLAGS="$save_CPPFLAGS"
+         LDFLAGS="$save_LDFLAGS"
+         LIBS="$save_LIBS"
+
+         AS_IF([test "x$gaudi_happy" = "xyes"],
+               [AC_SUBST([GAUDI_CPPFLAGS], ["$GAUDI_CPPFLAGS"])
+                AC_SUBST([GAUDI_LDFLAGS], ["$GAUDI_LDFLAGS"])
+                AC_SUBST([GAUDI_LIBS], ["$GAUDI_LIBS"])
+                AC_DEFINE([HAVE_GAUDI], 1, [Enable GAUDI support])],
+               [AS_IF([test "x$with_gaudi" != "xguess"],
+                      [AC_MSG_ERROR([GAUDI support is requested but gaudi packages cannot be found])],
+                      [AC_MSG_WARN([GAUDI not found])])])
+
+        ]) # "x$with_gaudi" = "xno"
+
+        gaudi_checked=yes
+        AM_CONDITIONAL([HAVE_GAUDI], [test "x$gaudi_happy" != xno])
+
+   ]) # "x$gaudi_checked" != "xyes"
+
+]) # UCX_CHECK_GAUDI

configure.ac

@@ -231,6 +231,7 @@ AS_IF([test "x$with_docs_only" = xyes],
      AM_CONDITIONAL([HAVE_GCOV], [false])
      AM_CONDITIONAL([HAVE_LCOV], [false])
      AM_CONDITIONAL([HAVE_ZE], [false])
+     AM_CONDITIONAL([HAVE_GAUDI], [false])
     ],
     [
      AM_CONDITIONAL([DOCS_ONLY], [false])
@@ -247,6 +248,7 @@ AS_IF([test "x$with_docs_only" = xyes],
      m4_include([config/m4/gdrcopy.m4])
      m4_include([config/m4/mad.m4])
      m4_include([config/m4/lcov.m4])
+     m4_include([config/m4/gaudi.m4])
      m4_include([src/ucm/configure.m4])
      m4_include([src/ucs/configure.m4])
      m4_include([src/uct/configure.m4])

debian/ucx-gaudi.install

@@ -0,0 +1,2 @@
+usr/lib/ucx/libucx_perftest_gaudi.*
+usr/lib/ucx/libuct_gaudi.*

src/ucs/Makefile.am

@@ -90,6 +90,11 @@ nobase_dist_libucs_la_HEADERS = \
 	arch/rv64/global_opts.h \
 	arch/global_opts.h
 
+if HAVE_GAUDI
+    nobase_dist_libucs_la_HEADERS += \
+	sys/topo/gaudi/topo.h
+endif
+
 noinst_HEADERS = \
 	arch/aarch64/cpu.h \
 	arch/generic/cpu.h \
@@ -220,9 +225,14 @@ libucs_la_SOURCES = \
 	vfs/base/vfs_obj.c \
 	vfs/base/vfs_cb.c
 
+if HAVE_GAUDI
+    libucs_la_SOURCES += \
+	sys/topo/gaudi/topo.c
+endif
+
 if HAVE_AARCH64_THUNDERX2
 libucs_la_SOURCES += \
-        arch/aarch64/memcpy_thunderx2.S
+    arch/aarch64/memcpy_thunderx2.S
 endif
 
 if HAVE_STATS

src/ucs/memory/memory_type.c

@@ -24,6 +24,7 @@ const char *ucs_memory_type_names[] = {
     [UCS_MEMORY_TYPE_ZE_HOST]      = "ze-host",
     [UCS_MEMORY_TYPE_ZE_DEVICE]    = "ze-device",
     [UCS_MEMORY_TYPE_ZE_MANAGED]   = "ze-managed",
+    [UCS_MEMORY_TYPE_GAUDI]        = "gaudi",
     [UCS_MEMORY_TYPE_LAST]         = "unknown",
     [UCS_MEMORY_TYPE_LAST + 1]     = NULL
 };
@@ -38,5 +39,6 @@ const char *ucs_memory_type_descs[] = {
     [UCS_MEMORY_TYPE_ZE_HOST]      = "Intel/Ze USM host memory",
     [UCS_MEMORY_TYPE_ZE_DEVICE]    = "Intel/Ze GPU memory",
     [UCS_MEMORY_TYPE_ZE_MANAGED]   = "Intel/Ze GPU managed memory",
+    [UCS_MEMORY_TYPE_GAUDI]        = "HabanaLabs Gaudi memory",
     [UCS_MEMORY_TYPE_LAST]         = "unknown"
 };

src/ucs/memory/memory_type.h

@@ -45,6 +45,7 @@ typedef enum ucs_memory_type {
     UCS_MEMORY_TYPE_ZE_HOST,       /**< Intel ZE memory (USM host) */
     UCS_MEMORY_TYPE_ZE_DEVICE,     /**< Intel ZE memory (USM device) */
     UCS_MEMORY_TYPE_ZE_MANAGED,    /**< Intel ZE managed memory (USM shared) */
+    UCS_MEMORY_TYPE_GAUDI,         /**< HabanaLabs Gaudi memory */
     UCS_MEMORY_TYPE_LAST,
     UCS_MEMORY_TYPE_UNKNOWN = UCS_MEMORY_TYPE_LAST
 } ucs_memory_type_t;