[mlir][mpi] Lowering Mpi To LLVM #127053
[mlir][mpi] Lowering Mpi To LLVM #127053
Conversation
fschlimb
requested review from
MaheshRavishankar,
matthias-springer,
AntonLydike and
River707
|
@llvm/pr-subscribers-mlir Author: Frank Schlimbach (fschlimb) ChangesAs agreed by @AntonLydike this replaces #95524.
This adds the following on top of #95524
cc @mofeing @wsmoses @tobiasgrosser @hhkit Patch is 44.08 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/127053.diff 10 Files Affected:
diff --git a/mlir/include/mlir/Conversion/MPIToLLVM/MPIToLLVM.h b/mlir/include/mlir/Conversion/MPIToLLVM/MPIToLLVM.h
new file mode 100644
index 0000000000000..8d2698aa91c7c
--- /dev/null
+++ b/mlir/include/mlir/Conversion/MPIToLLVM/MPIToLLVM.h
@@ -0,0 +1,31 @@
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef MLIR_CONVERSION_MPITOLLVM_H
+#define MLIR_CONVERSION_MPITOLLVM_H
+
+#include "mlir/IR/DialectRegistry.h"
+
+namespace mlir {
+
+class LLVMTypeConverter;
+class RewritePatternSet;
+
+#define GEN_PASS_DECL_MPITOLLVMCONVERSIONPASS
+#include "mlir/Conversion/Passes.h.inc"
+
+namespace mpi {
+
+void populateMPIToLLVMConversionPatterns(LLVMTypeConverter &converter,
+ RewritePatternSet &patterns);
+
+void registerConvertMPIToLLVMInterface(DialectRegistry ®istry);
+
+} // namespace mpi
+} // namespace mlir
+
+#endif // MLIR_CONVERSION_MPITOLLVM_H
diff --git a/mlir/include/mlir/Dialect/MPI/IR/MPI.td b/mlir/include/mlir/Dialect/MPI/IR/MPI.td
index 7c84443e5520d..df0cf9d518faf 100644
--- a/mlir/include/mlir/Dialect/MPI/IR/MPI.td
+++ b/mlir/include/mlir/Dialect/MPI/IR/MPI.td
@@ -42,104 +42,104 @@ def MPI_Dialect : Dialect {
// Error classes enum:
//===----------------------------------------------------------------------===//
-def MPI_CodeSuccess : I32EnumAttrCase<"MPI_SUCCESS", 0, "MPI_SUCCESS">;
-def MPI_CodeErrAccess : I32EnumAttrCase<"MPI_ERR_ACCESS", 1, "MPI_ERR_ACCESS">;
-def MPI_CodeErrAmode : I32EnumAttrCase<"MPI_ERR_AMODE", 2, "MPI_ERR_AMODE">;
-def MPI_CodeErrArg : I32EnumAttrCase<"MPI_ERR_ARG", 3, "MPI_ERR_ARG">;
-def MPI_CodeErrAssert : I32EnumAttrCase<"MPI_ERR_ASSERT", 4, "MPI_ERR_ASSERT">;
+def MPI_CodeSuccess : I32EnumAttrCase<"_MPI_SUCCESS", 0, "MPI_SUCCESS">;
+def MPI_CodeErrAccess : I32EnumAttrCase<"_MPI_ERR_ACCESS", 1, "MPI_ERR_ACCESS">;
+def MPI_CodeErrAmode : I32EnumAttrCase<"_MPI_ERR_AMODE", 2, "MPI_ERR_AMODE">;
+def MPI_CodeErrArg : I32EnumAttrCase<"_MPI_ERR_ARG", 3, "MPI_ERR_ARG">;
+def MPI_CodeErrAssert : I32EnumAttrCase<"_MPI_ERR_ASSERT", 4, "MPI_ERR_ASSERT">;
def MPI_CodeErrBadFile
- : I32EnumAttrCase<"MPI_ERR_BAD_FILE", 5, "MPI_ERR_BAD_FILE">;
-def MPI_CodeErrBase : I32EnumAttrCase<"MPI_ERR_BASE", 6, "MPI_ERR_BASE">;
-def MPI_CodeErrBuffer : I32EnumAttrCase<"MPI_ERR_BUFFER", 7, "MPI_ERR_BUFFER">;
-def MPI_CodeErrComm : I32EnumAttrCase<"MPI_ERR_COMM", 8, "MPI_ERR_COMM">;
+ : I32EnumAttrCase<"_MPI_ERR_BAD_FILE", 5, "MPI_ERR_BAD_FILE">;
+def MPI_CodeErrBase : I32EnumAttrCase<"_MPI_ERR_BASE", 6, "MPI_ERR_BASE">;
+def MPI_CodeErrBuffer : I32EnumAttrCase<"_MPI_ERR_BUFFER", 7, "MPI_ERR_BUFFER">;
+def MPI_CodeErrComm : I32EnumAttrCase<"_MPI_ERR_COMM", 8, "MPI_ERR_COMM">;
def MPI_CodeErrConversion
- : I32EnumAttrCase<"MPI_ERR_CONVERSION", 9, "MPI_ERR_CONVERSION">;
-def MPI_CodeErrCount : I32EnumAttrCase<"MPI_ERR_COUNT", 10, "MPI_ERR_COUNT">;
-def MPI_CodeErrDims : I32EnumAttrCase<"MPI_ERR_DIMS", 11, "MPI_ERR_DIMS">;
-def MPI_CodeErrDisp : I32EnumAttrCase<"MPI_ERR_DISP", 12, "MPI_ERR_DISP">;
+ : I32EnumAttrCase<"_MPI_ERR_CONVERSION", 9, "MPI_ERR_CONVERSION">;
+def MPI_CodeErrCount : I32EnumAttrCase<"_MPI_ERR_COUNT", 10, "MPI_ERR_COUNT">;
+def MPI_CodeErrDims : I32EnumAttrCase<"_MPI_ERR_DIMS", 11, "MPI_ERR_DIMS">;
+def MPI_CodeErrDisp : I32EnumAttrCase<"_MPI_ERR_DISP", 12, "MPI_ERR_DISP">;
def MPI_CodeErrDupDatarep
- : I32EnumAttrCase<"MPI_ERR_DUP_DATAREP", 13, "MPI_ERR_DUP_DATAREP">;
+ : I32EnumAttrCase<"_MPI_ERR_DUP_DATAREP", 13, "MPI_ERR_DUP_DATAREP">;
def MPI_CodeErrErrhandler
- : I32EnumAttrCase<"MPI_ERR_ERRHANDLER", 14, "MPI_ERR_ERRHANDLER">;
-def MPI_CodeErrFile : I32EnumAttrCase<"MPI_ERR_FILE", 15, "MPI_ERR_FILE">;
+ : I32EnumAttrCase<"_MPI_ERR_ERRHANDLER", 14, "MPI_ERR_ERRHANDLER">;
+def MPI_CodeErrFile : I32EnumAttrCase<"_MPI_ERR_FILE", 15, "MPI_ERR_FILE">;
def MPI_CodeErrFileExists
- : I32EnumAttrCase<"MPI_ERR_FILE_EXISTS", 16, "MPI_ERR_FILE_EXISTS">;
+ : I32EnumAttrCase<"_MPI_ERR_FILE_EXISTS", 16, "MPI_ERR_FILE_EXISTS">;
def MPI_CodeErrFileInUse
- : I32EnumAttrCase<"MPI_ERR_FILE_IN_USE", 17, "MPI_ERR_FILE_IN_USE">;
-def MPI_CodeErrGroup : I32EnumAttrCase<"MPI_ERR_GROUP", 18, "MPI_ERR_GROUP">;
-def MPI_CodeErrInfo : I32EnumAttrCase<"MPI_ERR_INFO", 19, "MPI_ERR_INFO">;
+ : I32EnumAttrCase<"_MPI_ERR_FILE_IN_USE", 17, "MPI_ERR_FILE_IN_USE">;
+def MPI_CodeErrGroup : I32EnumAttrCase<"_MPI_ERR_GROUP", 18, "MPI_ERR_GROUP">;
+def MPI_CodeErrInfo : I32EnumAttrCase<"_MPI_ERR_INFO", 19, "MPI_ERR_INFO">;
def MPI_CodeErrInfoKey
- : I32EnumAttrCase<"MPI_ERR_INFO_KEY", 20, "MPI_ERR_INFO_KEY">;
+ : I32EnumAttrCase<"_MPI_ERR_INFO_KEY", 20, "MPI_ERR_INFO_KEY">;
def MPI_CodeErrInfoNokey
- : I32EnumAttrCase<"MPI_ERR_INFO_NOKEY", 21, "MPI_ERR_INFO_NOKEY">;
+ : I32EnumAttrCase<"_MPI_ERR_INFO_NOKEY", 21, "MPI_ERR_INFO_NOKEY">;
def MPI_CodeErrInfoValue
- : I32EnumAttrCase<"MPI_ERR_INFO_VALUE", 22, "MPI_ERR_INFO_VALUE">;
+ : I32EnumAttrCase<"_MPI_ERR_INFO_VALUE", 22, "MPI_ERR_INFO_VALUE">;
def MPI_CodeErrInStatus
- : I32EnumAttrCase<"MPI_ERR_IN_STATUS", 23, "MPI_ERR_IN_STATUS">;
-def MPI_CodeErrIntern : I32EnumAttrCase<"MPI_ERR_INTERN", 24, "MPI_ERR_INTERN">;
-def MPI_CodeErrIo : I32EnumAttrCase<"MPI_ERR_IO", 25, "MPI_ERR_IO">;
-def MPI_CodeErrKeyval : I32EnumAttrCase<"MPI_ERR_KEYVAL", 26, "MPI_ERR_KEYVAL">;
+ : I32EnumAttrCase<"_MPI_ERR_IN_STATUS", 23, "MPI_ERR_IN_STATUS">;
+def MPI_CodeErrIntern : I32EnumAttrCase<"_MPI_ERR_INTERN", 24, "MPI_ERR_INTERN">;
+def MPI_CodeErrIo : I32EnumAttrCase<"_MPI_ERR_IO", 25, "MPI_ERR_IO">;
+def MPI_CodeErrKeyval : I32EnumAttrCase<"_MPI_ERR_KEYVAL", 26, "MPI_ERR_KEYVAL">;
def MPI_CodeErrLocktype
- : I32EnumAttrCase<"MPI_ERR_LOCKTYPE", 27, "MPI_ERR_LOCKTYPE">;
-def MPI_CodeErrName : I32EnumAttrCase<"MPI_ERR_NAME", 28, "MPI_ERR_NAME">;
-def MPI_CodeErrNoMem : I32EnumAttrCase<"MPI_ERR_NO_MEM", 29, "MPI_ERR_NO_MEM">;
+ : I32EnumAttrCase<"_MPI_ERR_LOCKTYPE", 27, "MPI_ERR_LOCKTYPE">;
+def MPI_CodeErrName : I32EnumAttrCase<"_MPI_ERR_NAME", 28, "MPI_ERR_NAME">;
+def MPI_CodeErrNoMem : I32EnumAttrCase<"_MPI_ERR_NO_MEM", 29, "MPI_ERR_NO_MEM">;
def MPI_CodeErrNoSpace
- : I32EnumAttrCase<"MPI_ERR_NO_SPACE", 30, "MPI_ERR_NO_SPACE">;
+ : I32EnumAttrCase<"_MPI_ERR_NO_SPACE", 30, "MPI_ERR_NO_SPACE">;
def MPI_CodeErrNoSuchFile
- : I32EnumAttrCase<"MPI_ERR_NO_SUCH_FILE", 31, "MPI_ERR_NO_SUCH_FILE">;
+ : I32EnumAttrCase<"_MPI_ERR_NO_SUCH_FILE", 31, "MPI_ERR_NO_SUCH_FILE">;
def MPI_CodeErrNotSame
- : I32EnumAttrCase<"MPI_ERR_NOT_SAME", 32, "MPI_ERR_NOT_SAME">;
-def MPI_CodeErrOp : I32EnumAttrCase<"MPI_ERR_OP", 33, "MPI_ERR_OP">;
-def MPI_CodeErrOther : I32EnumAttrCase<"MPI_ERR_OTHER", 34, "MPI_ERR_OTHER">;
+ : I32EnumAttrCase<"_MPI_ERR_NOT_SAME", 32, "MPI_ERR_NOT_SAME">;
+def MPI_CodeErrOp : I32EnumAttrCase<"_MPI_ERR_OP", 33, "MPI_ERR_OP">;
+def MPI_CodeErrOther : I32EnumAttrCase<"_MPI_ERR_OTHER", 34, "MPI_ERR_OTHER">;
def MPI_CodeErrPending
- : I32EnumAttrCase<"MPI_ERR_PENDING", 35, "MPI_ERR_PENDING">;
-def MPI_CodeErrPort : I32EnumAttrCase<"MPI_ERR_PORT", 36, "MPI_ERR_PORT">;
+ : I32EnumAttrCase<"_MPI_ERR_PENDING", 35, "MPI_ERR_PENDING">;
+def MPI_CodeErrPort : I32EnumAttrCase<"_MPI_ERR_PORT", 36, "MPI_ERR_PORT">;
def MPI_CodeErrProcAborted
- : I32EnumAttrCase<"MPI_ERR_PROC_ABORTED", 37, "MPI_ERR_PROC_ABORTED">;
-def MPI_CodeErrQuota : I32EnumAttrCase<"MPI_ERR_QUOTA", 38, "MPI_ERR_QUOTA">;
-def MPI_CodeErrRank : I32EnumAttrCase<"MPI_ERR_RANK", 39, "MPI_ERR_RANK">;
+ : I32EnumAttrCase<"_MPI_ERR_PROC_ABORTED", 37, "MPI_ERR_PROC_ABORTED">;
+def MPI_CodeErrQuota : I32EnumAttrCase<"_MPI_ERR_QUOTA", 38, "MPI_ERR_QUOTA">;
+def MPI_CodeErrRank : I32EnumAttrCase<"_MPI_ERR_RANK", 39, "MPI_ERR_RANK">;
def MPI_CodeErrReadOnly
- : I32EnumAttrCase<"MPI_ERR_READ_ONLY", 40, "MPI_ERR_READ_ONLY">;
+ : I32EnumAttrCase<"_MPI_ERR_READ_ONLY", 40, "MPI_ERR_READ_ONLY">;
def MPI_CodeErrRequest
- : I32EnumAttrCase<"MPI_ERR_REQUEST", 41, "MPI_ERR_REQUEST">;
+ : I32EnumAttrCase<"_MPI_ERR_REQUEST", 41, "MPI_ERR_REQUEST">;
def MPI_CodeErrRmaAttach
- : I32EnumAttrCase<"MPI_ERR_RMA_ATTACH", 42, "MPI_ERR_RMA_ATTACH">;
+ : I32EnumAttrCase<"_MPI_ERR_RMA_ATTACH", 42, "MPI_ERR_RMA_ATTACH">;
def MPI_CodeErrRmaConflict
- : I32EnumAttrCase<"MPI_ERR_RMA_CONFLICT", 43, "MPI_ERR_RMA_CONFLICT">;
+ : I32EnumAttrCase<"_MPI_ERR_RMA_CONFLICT", 43, "MPI_ERR_RMA_CONFLICT">;
def MPI_CodeErrRmaFlavor
- : I32EnumAttrCase<"MPI_ERR_RMA_FLAVOR", 44, "MPI_ERR_RMA_FLAVOR">;
+ : I32EnumAttrCase<"_MPI_ERR_RMA_FLAVOR", 44, "MPI_ERR_RMA_FLAVOR">;
def MPI_CodeErrRmaRange
- : I32EnumAttrCase<"MPI_ERR_RMA_RANGE", 45, "MPI_ERR_RMA_RANGE">;
+ : I32EnumAttrCase<"_MPI_ERR_RMA_RANGE", 45, "MPI_ERR_RMA_RANGE">;
def MPI_CodeErrRmaShared
- : I32EnumAttrCase<"MPI_ERR_RMA_SHARED", 46, "MPI_ERR_RMA_SHARED">;
+ : I32EnumAttrCase<"_MPI_ERR_RMA_SHARED", 46, "MPI_ERR_RMA_SHARED">;
def MPI_CodeErrRmaSync
- : I32EnumAttrCase<"MPI_ERR_RMA_SYNC", 47, "MPI_ERR_RMA_SYNC">;
-def MPI_CodeErrRoot : I32EnumAttrCase<"MPI_ERR_ROOT", 48, "MPI_ERR_ROOT">;
+ : I32EnumAttrCase<"_MPI_ERR_RMA_SYNC", 47, "MPI_ERR_RMA_SYNC">;
+def MPI_CodeErrRoot : I32EnumAttrCase<"_MPI_ERR_ROOT", 48, "MPI_ERR_ROOT">;
def MPI_CodeErrService
- : I32EnumAttrCase<"MPI_ERR_SERVICE", 49, "MPI_ERR_SERVICE">;
+ : I32EnumAttrCase<"_MPI_ERR_SERVICE", 49, "MPI_ERR_SERVICE">;
def MPI_CodeErrSession
- : I32EnumAttrCase<"MPI_ERR_SESSION", 50, "MPI_ERR_SESSION">;
-def MPI_CodeErrSize : I32EnumAttrCase<"MPI_ERR_SIZE", 51, "MPI_ERR_SIZE">;
-def MPI_CodeErrSpawn : I32EnumAttrCase<"MPI_ERR_SPAWN", 52, "MPI_ERR_SPAWN">;
-def MPI_CodeErrTag : I32EnumAttrCase<"MPI_ERR_TAG", 53, "MPI_ERR_TAG">;
+ : I32EnumAttrCase<"_MPI_ERR_SESSION", 50, "MPI_ERR_SESSION">;
+def MPI_CodeErrSize : I32EnumAttrCase<"_MPI_ERR_SIZE", 51, "MPI_ERR_SIZE">;
+def MPI_CodeErrSpawn : I32EnumAttrCase<"_MPI_ERR_SPAWN", 52, "MPI_ERR_SPAWN">;
+def MPI_CodeErrTag : I32EnumAttrCase<"_MPI_ERR_TAG", 53, "MPI_ERR_TAG">;
def MPI_CodeErrTopology
- : I32EnumAttrCase<"MPI_ERR_TOPOLOGY", 54, "MPI_ERR_TOPOLOGY">;
+ : I32EnumAttrCase<"_MPI_ERR_TOPOLOGY", 54, "MPI_ERR_TOPOLOGY">;
def MPI_CodeErrTruncate
- : I32EnumAttrCase<"MPI_ERR_TRUNCATE", 55, "MPI_ERR_TRUNCATE">;
-def MPI_CodeErrType : I32EnumAttrCase<"MPI_ERR_TYPE", 56, "MPI_ERR_TYPE">;
+ : I32EnumAttrCase<"_MPI_ERR_TRUNCATE", 55, "MPI_ERR_TRUNCATE">;
+def MPI_CodeErrType : I32EnumAttrCase<"_MPI_ERR_TYPE", 56, "MPI_ERR_TYPE">;
def MPI_CodeErrUnknown
- : I32EnumAttrCase<"MPI_ERR_UNKNOWN", 57, "MPI_ERR_UNKNOWN">;
+ : I32EnumAttrCase<"_MPI_ERR_UNKNOWN", 57, "MPI_ERR_UNKNOWN">;
def MPI_CodeErrUnsupportedDatarep
- : I32EnumAttrCase<"MPI_ERR_UNSUPPORTED_DATAREP", 58,
+ : I32EnumAttrCase<"_MPI_ERR_UNSUPPORTED_DATAREP", 58,
"MPI_ERR_UNSUPPORTED_DATAREP">;
def MPI_CodeErrUnsupportedOperation
- : I32EnumAttrCase<"MPI_ERR_UNSUPPORTED_OPERATION", 59,
+ : I32EnumAttrCase<"_MPI_ERR_UNSUPPORTED_OPERATION", 59,
"MPI_ERR_UNSUPPORTED_OPERATION">;
def MPI_CodeErrValueTooLarge
- : I32EnumAttrCase<"MPI_ERR_VALUE_TOO_LARGE", 60, "MPI_ERR_VALUE_TOO_LARGE">;
-def MPI_CodeErrWin : I32EnumAttrCase<"MPI_ERR_WIN", 61, "MPI_ERR_WIN">;
+ : I32EnumAttrCase<"_MPI_ERR_VALUE_TOO_LARGE", 60, "MPI_ERR_VALUE_TOO_LARGE">;
+def MPI_CodeErrWin : I32EnumAttrCase<"_MPI_ERR_WIN", 61, "MPI_ERR_WIN">;
def MPI_CodeErrLastcode
- : I32EnumAttrCase<"MPI_ERR_LASTCODE", 62, "MPI_ERR_LASTCODE">;
+ : I32EnumAttrCase<"_MPI_ERR_LASTCODE", 62, "MPI_ERR_LASTCODE">;
def MPI_ErrorClassEnum
: I32EnumAttr<"MPI_ErrorClassEnum", "MPI error class name", [
@@ -215,20 +215,20 @@ def MPI_ErrorClassAttr : EnumAttr<MPI_Dialect, MPI_ErrorClassEnum, "errclass"> {
let assemblyFormat = "`<` $value `>`";
}
-def MPI_OpNull : I32EnumAttrCase<"MPI_OP_NULL", 0, "MPI_OP_NULL">;
-def MPI_OpMax : I32EnumAttrCase<"MPI_MAX", 1, "MPI_MAX">;
-def MPI_OpMin : I32EnumAttrCase<"MPI_MIN", 2, "MPI_MIN">;
-def MPI_OpSum : I32EnumAttrCase<"MPI_SUM", 3, "MPI_SUM">;
-def MPI_OpProd : I32EnumAttrCase<"MPI_PROD", 4, "MPI_PROD">;
-def MPI_OpLand : I32EnumAttrCase<"MPI_LAND", 5, "MPI_LAND">;
-def MPI_OpBand : I32EnumAttrCase<"MPI_BAND", 6, "MPI_BAND">;
-def MPI_OpLor : I32EnumAttrCase<"MPI_LOR", 7, "MPI_LOR">;
-def MPI_OpBor : I32EnumAttrCase<"MPI_BOR", 8, "MPI_BOR">;
-def MPI_OpLxor : I32EnumAttrCase<"MPI_LXOR", 9, "MPI_LXOR">;
-def MPI_OpBxor : I32EnumAttrCase<"MPI_BXOR", 10, "MPI_BXOR">;
-def MPI_OpMinloc : I32EnumAttrCase<"MPI_MINLOC", 11, "MPI_MINLOC">;
-def MPI_OpMaxloc : I32EnumAttrCase<"MPI_MAXLOC", 12, "MPI_MAXLOC">;
-def MPI_OpReplace : I32EnumAttrCase<"MPI_REPLACE", 13, "MPI_REPLACE">;
+def MPI_OpNull : I32EnumAttrCase<"_MPI_OP_NULL", 0, "MPI_OP_NULL">;
+def MPI_OpMax : I32EnumAttrCase<"_MPI_MAX", 1, "MPI_MAX">;
+def MPI_OpMin : I32EnumAttrCase<"_MPI_MIN", 2, "MPI_MIN">;
+def MPI_OpSum : I32EnumAttrCase<"_MPI_SUM", 3, "MPI_SUM">;
+def MPI_OpProd : I32EnumAttrCase<"_MPI_PROD", 4, "MPI_PROD">;
+def MPI_OpLand : I32EnumAttrCase<"_MPI_LAND", 5, "MPI_LAND">;
+def MPI_OpBand : I32EnumAttrCase<"_MPI_BAND", 6, "MPI_BAND">;
+def MPI_OpLor : I32EnumAttrCase<"_MPI_LOR", 7, "MPI_LOR">;
+def MPI_OpBor : I32EnumAttrCase<"_MPI_BOR", 8, "MPI_BOR">;
+def MPI_OpLxor : I32EnumAttrCase<"_MPI_LXOR", 9, "MPI_LXOR">;
+def MPI_OpBxor : I32EnumAttrCase<"_MPI_BXOR", 10, "MPI_BXOR">;
+def MPI_OpMinloc : I32EnumAttrCase<"_MPI_MINLOC", 11, "MPI_MINLOC">;
+def MPI_OpMaxloc : I32EnumAttrCase<"_MPI_MAXLOC", 12, "MPI_MAXLOC">;
+def MPI_OpReplace : I32EnumAttrCase<"_MPI_REPLACE", 13, "MPI_REPLACE">;
def MPI_OpClassEnum : I32EnumAttr<"MPI_OpClassEnum", "MPI operation class", [
MPI_OpNull,
diff --git a/mlir/include/mlir/Dialect/MPI/IR/MPIOps.td b/mlir/include/mlir/Dialect/MPI/IR/MPIOps.td
index 284ba72af9768..db28bd09678f8 100644
--- a/mlir/include/mlir/Dialect/MPI/IR/MPIOps.td
+++ b/mlir/include/mlir/Dialect/MPI/IR/MPIOps.td
@@ -102,13 +102,13 @@ def MPI_SendOp : MPI_Op<"send", []> {
let arguments = (
ins AnyMemRef : $ref,
I32 : $tag,
- I32 : $rank
+ I32 : $dest
);
let results = (outs Optional<MPI_Retval>:$retval);
- let assemblyFormat = "`(` $ref `,` $tag `,` $rank `)` attr-dict `:` "
- "type($ref) `,` type($tag) `,` type($rank)"
+ let assemblyFormat = "`(` $ref `,` $tag `,` $dest `)` attr-dict `:` "
+ "type($ref) `,` type($tag) `,` type($dest)"
"(`->` type($retval)^)?";
let hasCanonicalizer = 1;
}
@@ -154,11 +154,11 @@ def MPI_ISendOp : MPI_Op<"isend", []> {
//===----------------------------------------------------------------------===//
def MPI_RecvOp : MPI_Op<"recv", []> {
- let summary = "Equivalent to `MPI_Recv(ptr, size, dtype, dest, tag, "
+ let summary = "Equivalent to `MPI_Recv(ptr, size, dtype, source, tag, "
"MPI_COMM_WORLD, MPI_STATUS_IGNORE)`";
let description = [{
MPI_Recv performs a blocking receive of `size` elements of type `dtype`
- from rank `dest`. The `tag` value and communicator enables the library to
+ from rank `source`. The `tag` value and communicator enables the library to
determine the matching of multiple sends and receives between the same
ranks.
@@ -172,13 +172,13 @@ def MPI_RecvOp : MPI_Op<"recv", []> {
let arguments = (
ins AnyMemRef : $ref,
- I32 : $tag, I32 : $rank
+ I32 : $tag, I32 : $source
);
let results = (outs Optional<MPI_Retval>:$retval);
- let assemblyFormat = "`(` $ref `,` $tag `,` $rank `)` attr-dict `:`"
- "type($ref) `,` type($tag) `,` type($rank)"
+ let assemblyFormat = "`(` $ref `,` $tag `,` $source `)` attr-dict `:` "
+ "type($ref) `,` type($tag) `,` type($source)"
"(`->` type($retval)^)?";
let hasCanonicalizer = 1;
}
diff --git a/mlir/include/mlir/Dialect/MPI/IR/MPITypes.td b/mlir/include/mlir/Dialect/MPI/IR/MPITypes.td
index fafea0eac8bb7..a55d30e778e22 100644
--- a/mlir/include/mlir/Dialect/MPI/IR/MPITypes.td
+++ b/mlir/include/mlir/Dialect/MPI/IR/MPITypes.td
@@ -30,7 +30,7 @@ class MPI_Type<string name, string typeMnemonic, list<Trait> traits = []>
//===----------------------------------------------------------------------===//
def MPI_Retval : MPI_Type<"Retval", "retval"> {
- let summary = "MPI function call return value";
+ let summary = "MPI function call return value (!mpi.retval)";
let description = [{
This type represents a return value from an MPI function call.
This value can be MPI_SUCCESS, MPI_ERR_IN_STATUS, or any error code.
diff --git a/mlir/include/mlir/InitAllExtensions.h b/mlir/include/mlir/InitAllExtensions.h
index 887db344ed88b..6ab23ff86b3c6 100644
--- a/mlir/include/mlir/InitAllExtensions.h
+++ b/mlir/include/mlir/InitAllExtensions.h
@@ -14,6 +14,7 @@
#ifndef MLIR_INITALLEXTENSIONS_H_
#define MLIR_INITALLEXTENSIONS_H_
+#include "Conversion/MPIToLLVM/MPIToLLVM.h"
#include "mlir/Conversion/ArithToLLVM/ArithToLLVM.h"
#include "mlir/Conversion/ComplexToLLVM/ComplexToLLVM.h"
#include "mlir/Conversion/ControlFlowToLLVM/ControlFlowToLLVM.h"
@@ -70,6 +71,7 @@ inline void registerAllExtensions(DialectRegistry ®istry) {
registerConvertFuncToLLVMInterface(registry);
index::registerConvertIndexToLLVMInterface(registry);
registerConvertMathToLLVMInterface(registry);
+ mpi::registerConvertMPIToLLVMInterface(registry);
registerConvertMemRefToLLVMInterface(registry);
registerConvertNVVMToLLVMInterface(registry);
registerConvertOpenMPToLLVMInterface(registry);
diff --git a/mlir/lib/Conversion/CMakeLists.txt b/mlir/lib/Conversion/CMakeLists.txt
index 0bd08ec6333e6..3dc7472584cf9 100644
--- a/mlir/lib/Conversion/CMakeLists.txt
+++ b/mlir/lib/Conversion/CMakeLists.txt
@@ -41,6 +41,7 @@ add_subdirectory(MemRefToEmitC)
add_subdirectory(MemRefToLLVM)
add_subdirectory(MemRefToSPIRV)
add_subdirectory(MeshToMPI)
+add_subdirectory(MPIToLLVM)
add_subdirectory(NVGPUToNVVM)
add_subdirectory(NVVMToLLVM)
add_subdirectory(OpenACCToSCF)
diff --git a/mlir/lib/Conversion/MPIToLLVM/CMakeLists.txt b/mlir/lib/Conversion/MPIToLLVM/CMakeLists.txt
new file mode 100644
index 0000000000000..17df603ff5686
--- /dev/null
+++ b/mlir/lib/Conversion/MPIToLLVM/CMakeLists.txt
@@ -0,0 +1,34 @@
+find_path(MPI_C_HEADER_DIR mpi.h
+ PATHS $ENV{I_MPI_ROOT}/include
+ $ENV{MPI_HOME}/include
+ $ENV{MPI_ROOT}/include)
+if(MPI_C_HEADER_DIR)
+ # cmake_path(REMOVE_FILENAME MPI_C_HEADER_DIR)
+ message(STATUS "found MPI_C_HEADER_DIR: ${MPI_C_HEADER_DIR}")
+else()
+ message(WARNING "MPI not found, disabling MLIRMPIToLLVM conversion")
+ return()
+endif()
+
+add_mlir_conversion_library(MLIRMPIToLLVM
+ MPIToLLVM.cpp
+
+ ADDITIONAL_HEADER_DIRS
+ ${MLIR_MAIN_INCLUDE_DIR}/mlir/Conversion/MPIToLLVM
+
+ DEPENDS
+ MLIRConversionPassIncGen
+
+ LINK_COMPONENTS
+ Core
+
+ LINK_LIBS PUBLIC
+ MLIRLLVMCommonConversion
+ MLIRLLVMDialect
+ MLIRMPIDialect
+)
+target_include_directories(
+ MLIRMPIToLLVM
+ PRIVATE
+ ${MPI_C_HEADER_DIR}
+)
\ No newline at end of file
diff --git a/mlir/lib/Conversion/MPIToLLVM/MPIImplTraits.h b/mlir/lib/Conversion/MPIToLLVM/MPIImplTraits.h
new file mode 100644
index 0000000000000..09811e1cb7c61
--- /dev/null
+++ b/mlir/lib/Conversion/MPIToLLVM/MPIImplTraits.h
@@ -0,0 +1,151 @@
+#define MPICH_SKIP_MPICXX 1
+#define OMPI_SKIP_MPICXX 1
+#include "mlir/Dialect/LLVMIR/LLVMDialect.h"
+#include "mlir/Dialect/MPI/IR/MPI.h"
+#include "mlir/Transforms/DialectConversion.h"
+#include <mpi.h>
+
+namespace {
+
+// when lowerring the mpi dialect to functions calls certain details
+// differ between various MPI implementations. This class will provide
+// these depending on the MPI implementation that got included.
+struct MPIImplTraits {
+ // get/create MPI_COMM_WORLD as a mlir::Value
+ static mlir::Value getCommWorld(mlir::ModuleOp &moduleOp,
+ const mlir::Location loc,
+ mlir::ConversionPatternRewriter &rewriter);
+ // get/create MPI datatype as a mlir::Value which corresponds to the given
+ // mlir::Type
+ static mlir::Value getDataType(mlir::ModuleOp &moduleOp,
+ cons...
[truncated]
|
Next MPI standard stabilizes the ABI, but I guess we still need to support earlier vendor-ABIs of earlier MPI implementations right? cc @JBlaschke @hhkit |
Yes, we can move to that once that is broadly available. |
There was a problem hiding this comment.
❤️ Really cool work! Thanks @fschlimb! Super excited to see this progress!
| // CHECK-NEXT: [[v28:%.*]] = llvm.mlir. | ||
| // CHECK-NEXT: [[v29:%.*]] = llvm.mlir. |
There was a problem hiding this comment.
these appear to be missing something?
There was a problem hiding this comment.
Yes, the constants used in there depend on the MPI implementation. This makes sure the test passes for any MPI.
There was a problem hiding this comment.
Very nice work and sorry for the late review.
Unfortunately, there are some issues with the dependencies on MPI headers and packages. I highly doubt that this is an acceptable thing to depend on in the build system. Note that test might break when someone has another header around that depends different values for the used constants.
I suggest to restructure the pass to take all the necessary values as configuration parameters. In the worst case, this could even be a json file, if the list of options is growing. I'm not really sure who could consult on this.
| #define GEN_PASS_DECL_MPITOLLVMCONVERSIONPASS | ||
| #include "mlir/Conversion/Passes.h.inc" |
There was a problem hiding this comment.
I don't understand. Why is this required when there is no actual pass that backs this. This is an interface based pass extension, no?
There was a problem hiding this comment.
Left-over. Can be removed.
| // TODO: this was copied from GPUOpsLowering.cpp:288 | ||
| // is this okay, or should this be moved to some common file? |
There was a problem hiding this comment.
This should be moved into Conversion/LLVMCommon, I presume
There was a problem hiding this comment.
This might be problematic to include here given the different licence. I'm not sure how this should be handled, tbh.
There was a problem hiding this comment.
The MPICH license is compatible because it is fully permissive. Should anyone see an issue with this, I can write it myself and make up my own values. We'd lose compatibility with MPICH/Intel MPI, though.
| matchAndRewrite(mpi::InitOp op, OpAdaptor adaptor, | ||
| ConversionPatternRewriter &rewriter) const override { | ||
| // get loc | ||
| auto loc = op.getLoc(); |
There was a problem hiding this comment.
Nit: Don't use auto when the type isn't exceptionally complex or given on the RHS.
There was a problem hiding this comment.
guidance is: "Don’t “almost always” use auto, but do use auto with initializers like cast(...) or other places where the type is already obvious from the context."
I guess it lies in the eye of the beholder, but to me the type is clearly obvious from the context.
There was a problem hiding this comment.
Many places in MLIR adapted a very clear stance on this. I'm aware that this rule is quite flexible, which makes it a bad one.
There was a problem hiding this comment.
Anyway, I changed as you suggested.
| auto loc = op.getLoc(); | ||
| auto context = rewriter.getContext(); | ||
| auto i32 = rewriter.getI32Type(); |
| func.func @mpi_test(%arg0: memref<100xf32>) { | ||
| // CHECK: [[varg0:%.*]]: !llvm.ptr, [[varg1:%.*]]: !llvm.ptr, [[varg2:%.*]]: i64, [[varg3:%.*]]: i64, [[varg4:%.*]]: i64 | ||
| // CHECK: [[v0:%.*]] = llvm.mlir.poison : !llvm.struct<(ptr, ptr, i64, array<1 x i64>, array<1 x i64>)> | ||
| // CHECK-NEXT: [[v1:%.*]] = llvm.insertvalue [[varg0]], [[v0]][0] : !llvm.struct<(ptr, ptr, i64, array<1 x i64>, array<1 x i64>)> |
There was a problem hiding this comment.
Nit: Only use CHECK-NEXT when necessary.
| %4 = mpi.retval_check %retval = <MPI_SUCCESS> : i1 | ||
|
|
||
| %5 = mpi.error_class %0 : !mpi.retval |
There was a problem hiding this comment.
Should there be checks for these as well?
There was a problem hiding this comment.
We currently do not handle these. Can add a check nevertheless.
| @@ -0,0 +1,84 @@ | |||
| // RUN: mlir-opt -convert-to-llvm %s | FileCheck %s | |||
|
|
|||
| module { | |||
There was a problem hiding this comment.
Nit: The module is not required.
Yes, there is no straight-forward solution for this right now. We could depend on yet another (external) library which unifies the interfaces. To me that's not really tempting for getting a few defines. Why would configuration parameters change anything about the issue you are mentioning? Whatever was configured can or cannot conflict later, no matter what. Notice that later is not when the compiler is built, but when the compiler is executed. If you are concerned about other passes/dialects using a different MPI in the same build, we could move finding the header-file to an upper level cmake file. Frankly, I don't see this as a problem. As mentioned above, as soon as the new MPI standard defines a stable ABI this problem is gone anyway. I suggest keeping this like this for the time being. |
|
Thanks a lot @Dinistro for your valuable comments and questions. |
There was a problem hiding this comment.
Why would configuration parameters change anything about the issue you are mentioning? Whatever was configured can or cannot conflict later, no matter what. Notice that later is not when the compiler is built, but when the compiler is executed. If you are concerned about other passes/dialects using a different MPI in the same build, we could move finding the header-file to an upper level cmake file. Frankly, I don't see this as a problem.
I was referring to lowering pass options that can be passed into the compiler, i.e., compiling to a specific MPI implementation/draft instead of tying the supported MPI implementation to the configuration time of MLIR itself.
In any case, I don't feel confident to judge if this is an acceptable workaround or not. Thus, I'm explicitly requesting someone from core to weight in. CC @joker-eph @matthias-springer @ftynse
There was a problem hiding this comment.
Internal headers like this are used very infrequently, that's why I raised it.
| #ifdef FOUND_MPI_C_HEADER | ||
| #include <mpi.h> | ||
| #else // not FOUND_MPI_C_HEADER | ||
| #include "mpi_fallback.h" | ||
| #endif // FOUND_MPI_C_HEADER |
There was a problem hiding this comment.
Do we have a clear plan of adapting this general format? I understand that this can be annoying to implement, but not doing it now might result in more work down the line.
I know of multiple companies that distribute MLIR in binary form, mine included. For now, we will not rely on the MPI dialect, but this might become an issue at some point. Specifically, setting up the build containers to contain the correct MPI version we want to target will be problematic.
| matchAndRewrite(mpi::InitOp op, OpAdaptor adaptor, | ||
| ConversionPatternRewriter &rewriter) const override { | ||
| // get loc | ||
| auto loc = op.getLoc(); |
There was a problem hiding this comment.
Many places in MLIR adapted a very clear stance on this. I'm aware that this rule is quite flexible, which makes it a bad one.
There was a problem hiding this comment.
Please adapt the name to follow the established format.
|
|
||
| // CHECK: mpi.retval_check | ||
| %4 = mpi.retval_check %retval = <MPI_SUCCESS> : i1 | ||
| // CEHCK: mpi.error_class |
There was a problem hiding this comment.
| // CEHCK: mpi.error_class | |
| // CHECK: mpi.error_class |
| // CHECK: llvm.call @MPI_Finalize() : () -> i32 | ||
| %3 = mpi.finalize : !mpi.retval | ||
|
|
||
| // CHECK: mpi.retval_check |
There was a problem hiding this comment.
Nit: Maybe add a TODO that states that these are not yet lowered and will be in the future.
There was a problem hiding this comment.
Removed the ops, they are unnecessary.
Instead of linking in the MPI header file that is decided on by the pre-processor, could we make the MPI library a run-time choice? Instead of using a header file, I suggest we hardcode the ABI based on our knowledge. We can use a little python script to extract the ABI we need, but requiring the mpi.h header might not be a good idea. In the end, 90% of this PR already hardcodes the MPI API without relying on the MPI headers. I feel the cost of adding some ABI knowledge is small and the benefit of being independent of headers and in-line with MLIR-style are notable. |
Yes, as the "fallback" shows, currently there is not much we need to know. We can do the same for OpenMPI. Of course this raises the same license questions as for MPICH. I am convinced both licenses are sufficiently permissive. Adding "runtime" option is of course more work (we must consider various versions of MPI implementations). Again, I do not think the current approach is a super special case in MLIR/LLVM. Other passe/dialects/features are also tied to specific HW or SW at build time. Any objections to addressing this in a separate PR? |
IIUC, the current lowering pass changes behavior depending on the MPI version that was around during build time. If that is really the case, then the test should not work properly depending on the build setting, no? |
The tests are written in a way, that the differences between OpenMPI and MPICH* are hidden. This is mostly about different constant values. @AntonLydike also found places where details of checked operations are missing (and only the operation name itself gets verified). |
|
I thought a bit about the runtime option. I see only two ways of doing this in a reasonable way:
module attributes { mpi.dlti = #dlti.map<"MPI:Implemention" = "Intel"> } {
%0 = mpi.....
}Any thoughts on this? |
|
Nice! |
|
✅ With the latest revision this PR passed the C/C++ code formatter. |
fixing MPI_Send/Recv signature: added size fixing MPIR ops.mlir tests MPI: fixing enum names and merge conflicts separate MPIImplTraits.h
MPIImplTraits for OpenMPI
…ing review comments
Co-authored-by: Christian Ulmann <[email protected]>
|
@Dinistro @tobiasgrosser I just (force-)pushed a solution which selects the MPI implementation at runtime. As suggested above, it uses DLTI. Hope this addresses your concerns. |
There was a problem hiding this comment.
Very nice, thanks for changing this to use the DLTI. I mainly added nit comments, related to things from the style guide: https://mlir.llvm.org/getting_started/DeveloperGuide/
Considering the new structure, it might be more elegant to rely on virtual dispatch to implement the different lowering strategies. This use case very nicely fits the strategy pattern, I believe.
This would ensure that you only need to check the DLTI once and can thereafter rely on the virtual dispatch to delegate to the proper implementation.
Co-authored-by: Christian Ulmann <[email protected]>
Yes, I have been thinking about this as well. Virtual dispatch requires an object. The easy lifetime of that object is each Yes, the current solution is not perfect. We can go on like this forever, or we declare this as good enough as a first step and together work on continuous improvements in future PRs. I vote for the latter. |
|
Thanks a lot for your review and suggestions, @Dinistro ! |
| // | ||
| // Copyright (C) by Argonne National Laboratory | ||
| // See COPYRIGHT in top-level directory | ||
| // of MPICH source repository. | ||
| // |
There was a problem hiding this comment.
I guess this is no longer needed?
| /// Instantiate a new MPIImplTraits object according to the DLTI attribute | ||
| /// on the given module. |
There was a problem hiding this comment.
Maybe clarify that the fallback is MPICH
Co-authored-by: Christian Ulmann <[email protected]>
|
Excellent. Thanks again @AntonLydike, @Dinistro, @tobiasgrosser and @mofeing for your comments! |
|
@fschlimb With this PR, I am getting: Is there any way you could take a look? My cmake configuration is |
|
@kazutakahirata Thanks so much! |
As agreed by @AntonLydike this replaces #95524.
This adds the following on top of #95524
cc @mofeing @wsmoses @tobiasgrosser @hhkit @Groverkss @sjw36 @joker-eph @zero9178