CL/DOCA_UROM: Add CL_DOCA_UROM

nsarka · Jun 24, 2024 · 1ae5756 · 1ae5756
1 parent 5e4b986
commit 1ae5756
Show file tree

Hide file tree

Showing 25 changed files with 6,329 additions and 6 deletions.
diff --git a/.github/workflows/codestyle.yaml b/.github/workflows/codestyle.yaml
@@ -37,7 +37,7 @@ jobs:
             fi
           fi
           H1="CODESTYLE|REVIEW|CORE|UTIL|TEST|API|DOCS|TOOLS|BUILD|MC|EC|SCHEDULE|TOPO"
-          H2="CI|CL/|TL/|MC/|EC/|UCP|SHM|NCCL|SHARP|BASIC|HIER|CUDA|CPU|EE|RCCL|ROCM|SELF|MLX5"
+          H2="CI|CL/|TL/|MC/|EC/|UCP|SHM|NCCL|SHARP|BASIC|HIER|DOCA_UROM|CUDA|CPU|EE|RCCL|ROCM|SELF|MLX5"
           if ! echo $msg | grep -qP '^Merge |^'"(($H1)|($H2))"'+: \w'
           then
             echo "Wrong header"

diff --git a/Makefile.am b/Makefile.am
@@ -7,6 +7,7 @@
 if !DOCS_ONLY
 SUBDIRS =      \
 	src        \
+	contrib    \
 	tools/info \
 	cmake
 

diff --git a/config/m4/doca_urom.m4 b/config/m4/doca_urom.m4
@@ -0,0 +1,75 @@
+#
+# Copyright (c) 2024, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# See file LICENSE for terms.
+#
+
+AC_DEFUN([CHECK_DOCA_UROM],[
+AS_IF([test "x$doca_urom_checked" != "xyes"],[
+    doca_urom_happy="no"
+    AC_ARG_WITH([doca_urom],
+            [AS_HELP_STRING([--with-doca_urom=(DIR)], [Enable the use of DOCA_UROM (default is guess).])],
+            [], [with_doca_urom=guess])
+    AS_IF([test "x$with_doca_urom" != "xno"],
+    [
+        save_CPPFLAGS="$CPPFLAGS"
+        save_LDFLAGS="$LDFLAGS"
+        AS_IF([test ! -z "$with_doca_urom" -a "x$with_doca_urom" != "xyes" -a "x$with_doca_urom" != "xguess"],
+        [
+            AS_IF([test ! -d $with_doca_urom],
+                  [AC_MSG_ERROR([Provided "--with-doca_urom=${with_doca_urom}" location does not exist])])
+            check_doca_urom_dir="$with_doca_urom"
+            check_doca_urom_libdir="$with_doca_urom/lib64"
+            CPPFLAGS="-I$with_doca_urom/include $UCS_CPPFLAGS $save_CPPFLAGS"
+            LDFLAGS="-L$check_doca_urom_libdir $save_LDFLAGS"
+        ])
+        AS_IF([test ! -z "$with_doca_urom_libdir" -a "x$with_doca_urom_libdir" != "xyes"],
+        [
+            check_doca_urom_libdir="$with_doca_urom_libdir"
+            LDFLAGS="-L$check_doca_urom_libdir $save_LDFLAGS"
+        ])
+        AC_CHECK_HEADERS([doca_urom.h],
+        [
+            AC_CHECK_LIB([doca_urom], [doca_urom_service_create],
+            [
+                doca_urom_happy="yes"
+            ],
+            [
+                echo "CPPFLAGS: $CPPFLAGS"
+                doca_urom_happy="no"
+            ], [-ldoca_common -ldoca_argp -ldoca_urom])
+        ],
+        [
+            doca_urom_happy="no"
+        ])
+        AS_IF([test "x$doca_urom_happy" = "xyes"],
+        [
+            AS_IF([test "x$check_doca_urom_dir" != "x"],
+            [
+                AC_MSG_RESULT([DOCA_UROM dir: $check_doca_urom_dir])
+                AC_SUBST(DOCA_UROM_CPPFLAGS, "-I$check_doca_urom_dir/include/ $doca_urom_old_headers")
+            ])
+            AS_IF([test "x$check_doca_urom_libdir" != "x"],
+            [
+                AC_SUBST(DOCA_UROM_LDFLAGS, "-L$check_doca_urom_libdir")
+            ])
+            AC_SUBST(DOCA_UROM_LIBADD, "-ldoca_common -ldoca_argp -ldoca_urom")
+            AC_DEFINE([HAVE_DOCA_UROM], 1, [Enable DOCA_UROM support])
+        ],
+        [
+            AS_IF([test "x$with_doca_urom" != "xguess"],
+            [
+                AC_MSG_ERROR([DOCA_UROM support is requested but DOCA_UROM packages cannot be found! $CPPFLAGS $LDFLAGS])
+            ],
+            [
+                AC_MSG_WARN([DOCA_UROM not found])
+            ])
+        ])
+        CPPFLAGS="$save_CPPFLAGS"
+        LDFLAGS="$save_LDFLAGS"
+    ],
+    [
+        AC_MSG_WARN([DOCA_UROM was explicitly disabled])
+    ])
+    doca_urom_checked=yes
+    AM_CONDITIONAL([HAVE_DOCA_UROM], [test "x$doca_urom_happy" != xno])
+])])
diff --git a/configure.ac b/configure.ac
@@ -162,6 +162,7 @@ AS_IF([test "x$with_docs_only" = xyes],
      AM_CONDITIONAL([HAVE_IBVERBS],[false])
      AM_CONDITIONAL([HAVE_RDMACM],[false])
      AM_CONDITIONAL([HAVE_MLX5DV],[false])
+     AM_CONDITIONAL([HAVE_DOCA_UROM], [false])
     ],
     [
      AM_CONDITIONAL([DOCS_ONLY], [false])
@@ -172,6 +173,7 @@ AS_IF([test "x$with_docs_only" = xyes],
      m4_include([config/m4/cuda.m4])
      m4_include([config/m4/nccl.m4])
      m4_include([config/m4/rocm.m4])
+     m4_include([config/m4/doca_urom.m4])
      m4_include([config/m4/rccl.m4])
      m4_include([config/m4/sharp.m4])
      m4_include([config/m4/mpi.m4])
@@ -205,6 +207,9 @@ AS_IF([test "x$with_docs_only" = xyes],
          mc_modules="${mc_modules}:rocm"
      fi
 
+     CHECK_DOCA_UROM
+     AC_MSG_RESULT([DOCA_UROM support: $doca_urom_happy])
+
      CHECK_GTEST
      AC_MSG_RESULT([GTEST support: $gtest_happy])
 
@@ -224,11 +229,13 @@ LDFLAGS="$LDFLAGS $UCS_LDFLAGS $UCS_LIBADD"
 CHECK_TL_COLL_PLUGINS
 AC_CONFIG_FILES([
                  Makefile
+                 contrib/Makefile
                  src/Makefile
                  src/ucc/api/ucc_version.h
                  src/core/ucc_version.c
                  src/components/cl/basic/Makefile
                  src/components/cl/hier/Makefile
+                 src/components/cl/doca_urom/Makefile
                  src/components/mc/cpu/Makefile
                  src/components/mc/cuda/Makefile
                  src/components/ec/cpu/Makefile
@@ -265,6 +272,7 @@ AC_MSG_NOTICE([      C++ compiler:   ${CXX} ${CXXFLAGS} ${BASE_CXXFLAGS}])
 AS_IF([test "x$cuda_happy" = "xyes"],[
 AC_MSG_NOTICE([     NVCC gencodes:   ${NVCC_ARCH}])
 ])
+AC_MSG_NOTICE([ DOCA UROM enabled:   ${doca_urom_happy}])
 AC_MSG_NOTICE([          Perftest:   ${mpi_enable}])
 AC_MSG_NOTICE([             Gtest:   ${gtest_enable}])
 AC_MSG_NOTICE([        MC modules:   <$(echo ${mc_modules}|tr ':' ' ') >])

diff --git a/contrib/Makefile.am b/contrib/Makefile.am
@@ -0,0 +1,22 @@
+#
+# Copyright (c) 2024, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+#
+
+if HAVE_DOCA_UROM
+
+sources =                                     \
+	doca_urom_ucc_plugin/common/urom_ucc.h    \
+	doca_urom_ucc_plugin/dpu/worker_ucc_p2p.c \
+	doca_urom_ucc_plugin/dpu/worker_ucc.h     \
+	doca_urom_ucc_plugin/dpu/worker_ucc.c
+
+plugindir = $(moduledir)/doca_plugins
+
+plugin_LTLIBRARIES         = libucc_doca_urom_plugin.la
+libucc_doca_urom_plugin_la_SOURCES  = $(sources)
+libucc_doca_urom_plugin_la_CPPFLAGS = $(AM_CPPFLAGS) $(BASE_CPPFLAGS) $(UCX_CPPFLAGS) $(DOCA_UROM_CPPFLAGS)
+libucc_doca_urom_plugin_la_CFLAGS   = $(BASE_CFLAGS)
+libucc_doca_urom_plugin_la_LDFLAGS  = -version-info $(SOVERSION) --as-needed $(DOCA_UROM_LDFLAGS)
+libucc_doca_urom_plugin_la_LIBADD   = $(UCX_LIBADD) $(DOCA_UROM_LIBADD) $(UCC_TOP_BUILDDIR)/src/libucc.la
+
+endif
diff --git a/contrib/doca_urom_ucc_plugin/common/urom_ucc.h b/contrib/doca_urom_ucc_plugin/common/urom_ucc.h
@@ -0,0 +1,171 @@
+/*
+ * Copyright (c) 2024 NVIDIA CORPORATION & AFFILIATES, ALL RIGHTS RESERVED.
+ *
+ * This software product is a proprietary product of NVIDIA CORPORATION &
+ * AFFILIATES (the "Company") and all right, title, and interest in and to the
+ * software product, including all associated intellectual property rights, are
+ * and shall remain exclusively with the Company.
+ *
+ * This software product is governed by the End User License Agreement
+ * provided with the software product.
+ *
+ */
+
+#ifndef UROM_UCC_H_
+#define UROM_UCC_H_
+
+#include <ucp/api/ucp.h>
+#include <ucc/api/ucc.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/* UCC serializing next raw, iter points to the offset place and returns the buffer start */
+#define urom_ucc_serialize_next_raw(_iter, _type, _offset) \
+	({ \
+		_type *_result = (_type *)(*(_iter)); \
+		*(_iter) = UCS_PTR_BYTE_OFFSET(*(_iter), _offset); \
+		_result; \
+	})
+
+/* UCC command types */
+enum urom_worker_ucc_cmd_type {
+	UROM_WORKER_CMD_UCC_LIB_CREATE,			 /* UCC library create command */
+	UROM_WORKER_CMD_UCC_LIB_DESTROY,		 /* UCC library destroy command */
+	UROM_WORKER_CMD_UCC_CONTEXT_CREATE,		 /* UCC context create command */
+	UROM_WORKER_CMD_UCC_CONTEXT_DESTROY,		 /* UCC context destroy command */
+	UROM_WORKER_CMD_UCC_TEAM_CREATE,		 /* UCC team create command */
+	UROM_WORKER_CMD_UCC_COLL,			 /* UCC collective create command */
+	UROM_WORKER_CMD_UCC_CREATE_PASSIVE_DATA_CHANNEL, /* UCC passive data channel command */
+};
+
+/*
+ * UCC library create command structure
+ *
+ * Input parameters for creating the library handle. The semantics of the parameters are defined by ucc.h
+ * On successful completion of urom_worker_cmd_ucc_lib_create,
+ * The UROM worker will generate a notification on the notification queue. This
+ * notification has reference to local library handle on the worker. The
+ * implementation can choose to create shadow handles or safely pack the
+ * library handle on the BlueCC worker to the AEU.
+ */
+struct urom_worker_cmd_ucc_lib_create {
+	void *params; /* UCC library parameters */
+};
+
+/* UCC context create command structure */
+struct urom_worker_cmd_ucc_context_create {
+	union {
+		int64_t start;	/* The started index */
+		int64_t *array; /* Set stride to <= 0 if array is used */
+	};
+	int64_t stride; /* Set number of strides */
+	int64_t size;	/* Set stride size */
+	void *base_va;	/* Shared buffer address */
+	uint64_t len;	/* Buffer length */
+};
+
+/* UCC passive data channel command structure */
+struct urom_worker_cmd_ucc_pass_dc {
+	void *ucp_addr;	 /* UCP worker address on host */
+	size_t addr_len; /* UCP worker address length */
+};
+
+/* UCC context destroy command structure */
+struct urom_worker_cmd_ucc_context_destroy {
+	void *context_h; /* UCC context pointer */
+};
+
+/* UCC team create command structure */
+struct urom_worker_cmd_ucc_team_create {
+	int64_t start;	 /* Team start index */
+	int64_t stride;	 /* Number of strides */
+	int64_t size;	 /* Stride size */
+	void *context_h; /* UCC context */
+};
+
+/* UCC team destroy command structure */
+struct urom_worker_cmd_ucc_team_destroy {
+	void *team; /* UCC team to destroy */
+};
+
+/* UCC collective command structure */
+struct urom_worker_cmd_ucc_coll {
+	void *coll_args;	 /* Collective arguments */
+	void *team;		 /* UCC team */
+	int use_xgvmi;		 /* If operation uses XGVMI */
+	void *work_buffer;	 /* Work buffer */
+	size_t work_buffer_size; /* Buffer size */
+	size_t team_size;	 /* Team size */
+};
+
+/* UROM UCC worker command structure */
+struct urom_worker_ucc_cmd {
+	uint64_t cmd_type;	/* Type of command as defined by urom_worker_ucc_cmd_type */
+	uint64_t dpu_worker_id; /* DPU worker id as part of the team */
+	union {
+		struct urom_worker_cmd_ucc_lib_create lib_create_cmd;		/* Lib create command */
+		struct urom_worker_cmd_ucc_context_create context_create_cmd;	/* Context create command */
+		struct urom_worker_cmd_ucc_context_destroy context_destroy_cmd; /* Context destroy command */
+		struct urom_worker_cmd_ucc_team_create team_create_cmd;		/* Team create command */
+		struct urom_worker_cmd_ucc_team_destroy team_destroy_cmd;	/* Team destroy command */
+		struct urom_worker_cmd_ucc_coll coll_cmd;			/* UCC collective command */
+		struct urom_worker_cmd_ucc_pass_dc pass_dc_create_cmd;		/* Passive data channel command */
+	};
+};
+
+/* UCC notification types */
+enum urom_worker_ucc_notify_type {
+	UROM_WORKER_NOTIFY_UCC_LIB_CREATE_COMPLETE,	      /* Create UCC library on DPU notification */
+	UROM_WORKER_NOTIFY_UCC_LIB_DESTROY_COMPLETE,	      /* Destroy UCC library on DPU notification */
+	UROM_WORKER_NOTIFY_UCC_CONTEXT_CREATE_COMPLETE,	      /* Create UCC context on DPU notification */
+	UROM_WORKER_NOTIFY_UCC_CONTEXT_DESTROY_COMPLETE,      /* Destroy UCC context on DPU notification */
+	UROM_WORKER_NOTIFY_UCC_TEAM_CREATE_COMPLETE,	      /* Create UCC team on DPU notification */
+	UROM_WORKER_NOTIFY_UCC_COLLECTIVE_COMPLETE,	      /* UCC collective completion notification */
+	UROM_WORKER_NOTIFY_UCC_PASSIVE_DATA_CHANNEL_COMPLETE, /* UCC data channel completion notification */
+};
+
+/* UCC context create notification structure */
+struct urom_worker_ucc_notify_context_create {
+	void *context; /* Pointer to UCC context */
+};
+
+/* UCC team create notification structure */
+struct urom_worker_ucc_notify_team_create {
+	void *team; /* Pointer to UCC team */
+};
+
+/* UCC collective notification structure */
+struct urom_worker_ucc_notify_collective {
+	ucc_status_t status; /* UCC collective status */
+};
+
+/* UCC passive data channel notification structure */
+struct urom_worker_ucc_notify_pass_dc {
+	ucc_status_t status; /* UCC  data channel status */
+};
+
+/* UROM UCC worker notification structure */
+struct urom_worker_notify_ucc {
+	uint64_t notify_type;	/* Notify type as defined by urom_worker_ucc_notify_type */
+	uint64_t dpu_worker_id; /* DPU worker id */
+	union {
+		struct urom_worker_ucc_notify_context_create context_create_nqe; /* Context create notification */
+		struct urom_worker_ucc_notify_team_create team_create_nqe;	 /* Team create notification */
+		struct urom_worker_ucc_notify_collective coll_nqe;		 /* Collective notification */
+		struct urom_worker_ucc_notify_pass_dc pass_dc_nqe;		 /* Passive data channel notification */
+	};
+};
+
+typedef struct ucc_worker_key_buf {
+	size_t src_len;
+	size_t dst_len;
+	char rkeys[1024];
+} ucc_worker_key_buf;
+
+#ifdef __cplusplus
+} /* extern "C" */
+#endif
+
+#endif /* UROM_UCC_H_ */