[Mlir-commits] [mlir] [mlir][mesh, MPI] Mesh2mpi (PR #104566)
Frank Schlimbach
llvmlistbot at llvm.org
Tue Aug 20 04:48:36 PDT 2024
================
@@ -0,0 +1,225 @@
+//===- MeshToMPI.cpp - Mesh to MPI dialect conversion -----------------===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements a translation of Mesh communication ops tp MPI ops.
+//
+//===----------------------------------------------------------------------===//
+
+#include "mlir/Conversion/MeshToMPI/MeshToMPI.h"
+
+#include "mlir/Dialect/Arith/IR/Arith.h"
+#include "mlir/Dialect/MPI/IR/MPI.h"
+#include "mlir/Dialect/MemRef/IR/MemRef.h"
+#include "mlir/Dialect/Mesh/IR/MeshOps.h"
+#include "mlir/Dialect/SCF/IR/SCF.h"
+#include "mlir/Dialect/Utils/StaticValueUtils.h"
+#include "mlir/IR/Builders.h"
+#include "mlir/IR/BuiltinAttributes.h"
+#include "mlir/IR/BuiltinTypes.h"
+#include "mlir/IR/PatternMatch.h"
+#include "mlir/Transforms/GreedyPatternRewriteDriver.h"
+
+#define DEBUG_TYPE "mesh-to-mpi"
+#define DBGS() (llvm::dbgs() << "[" DEBUG_TYPE "]: ")
+
+namespace mlir {
+#define GEN_PASS_DEF_CONVERTMESHTOMPIPASS
+#include "mlir/Conversion/Passes.h.inc"
+} // namespace mlir
+
+using namespace mlir;
+using namespace mlir::mesh;
+
+namespace {
+
+// This pattern converts the mesh.update_halo operation to MPI calls
+struct ConvertUpdateHaloOp
+ : public mlir::OpRewritePattern<mlir::mesh::UpdateHaloOp> {
+ using OpRewritePattern::OpRewritePattern;
+
+ mlir::LogicalResult
+ matchAndRewrite(mlir::mesh::UpdateHaloOp op,
+ mlir::PatternRewriter &rewriter) const override {
+ // Halos are exchanged as 2 blocks per dimension (one for each side: down
+ // and up). It is assumed that the last dim in a default memref is
+ // contiguous, hence iteration starts with the complete halo on the first
+ // dim which should be contiguous (unless the source is not). The size of
+ // the exchanged data will decrease when iterating over dimensions. That's
+ // good because the halos of last dim will be most fragmented.
+ // memref.subview is used to read and write the halo data from and to the
+ // local data. subviews and halos have dynamic and static values, so
+ // OpFoldResults are used whenever possible.
+
+ SymbolTableCollection symbolTableCollection;
+ auto loc = op.getLoc();
+
+ // convert a OpFoldResult into a Value
+ auto toValue = [&rewriter, &loc](OpFoldResult &v) {
+ return v.is<Value>()
+ ? v.get<Value>()
+ : rewriter.create<::mlir::arith::ConstantOp>(
+ loc,
+ rewriter.getIndexAttr(
+ cast<IntegerAttr>(v.get<Attribute>()).getInt()));
+ };
+
+ auto array = op.getInput();
+ auto rank = array.getType().getRank();
+ auto mesh = op.getMesh();
+ auto meshOp = getMesh(op, symbolTableCollection);
+ auto haloSizes = getMixedValues(op.getStaticHaloSizes(),
+ op.getDynamicHaloSizes(), rewriter);
+ // subviews need Index values
+ for (auto &sz : haloSizes) {
+ if (sz.is<Value>()) {
+ sz = rewriter
+ .create<arith::IndexCastOp>(loc, rewriter.getIndexType(),
+ sz.get<Value>())
+ .getResult();
+ }
+ }
+
+ // most of the offset/size/stride data is the same for all dims
+ SmallVector<OpFoldResult> offsets(rank, rewriter.getIndexAttr(0));
+ SmallVector<OpFoldResult> strides(rank, rewriter.getIndexAttr(1));
+ SmallVector<OpFoldResult> shape(rank);
+ // we need the actual shape to compute offsets and sizes
+ for (auto [i, s] : llvm::enumerate(array.getType().getShape())) {
+ if (ShapedType::isDynamic(s)) {
+ shape[i] = rewriter.create<memref::DimOp>(loc, array, s).getResult();
+ } else {
+ shape[i] = rewriter.getIndexAttr(s);
+ }
+ }
+
+ auto tagAttr = rewriter.getI32IntegerAttr(91); // we just pick something
+ auto tag = rewriter.create<::mlir::arith::ConstantOp>(loc, tagAttr);
+ auto zeroAttr = rewriter.getI32IntegerAttr(0); // for detecting v<0
+ auto zero = rewriter.create<::mlir::arith::ConstantOp>(loc, zeroAttr);
+ SmallVector<Type> indexResultTypes(meshOp.getShape().size(),
+ rewriter.getIndexType());
+ auto myMultiIndex =
+ rewriter.create<ProcessMultiIndexOp>(loc, indexResultTypes, mesh)
+ .getResult();
+ // halo sizes are provided for split dimensions only
+ auto currHaloDim = 0;
+
+ for (auto [dim, splitAxes] : llvm::enumerate(op.getSplitAxes())) {
+ if (splitAxes.empty()) {
+ continue;
+ }
+ // Get the linearized ids of the neighbors (down and up) for the
+ // given split
+ auto tmp = rewriter
+ .create<NeighborsLinearIndicesOp>(loc, mesh, myMultiIndex,
+ splitAxes)
+ .getResults();
+ // MPI operates on i32...
+ Value neighbourIDs[2] = {rewriter.create<arith::IndexCastOp>(
+ loc, rewriter.getI32Type(), tmp[0]),
+ rewriter.create<arith::IndexCastOp>(
+ loc, rewriter.getI32Type(), tmp[1])};
+ // store for later
+ auto orgDimSize = shape[dim];
+ // this dim's offset to the start of the upper halo
+ auto upperOffset = rewriter.create<arith::SubIOp>(
+ loc, toValue(shape[dim]), toValue(haloSizes[currHaloDim * 2 + 1]));
+
+ // Make sure we send/recv in a way that does not lead to a dead-lock.
+ // The current approach is by far not optimal, this should be at least
+ // be a red-black pattern or using MPI_sendrecv.
+ // Also, buffers should be re-used.
+ // Still using temporary contiguous buffers for MPI communication...
+ // Still yielding a "serialized" communication pattern...
+ auto genSendRecv = [&](auto dim, bool upperHalo) {
+ auto orgOffset = offsets[dim];
+ shape[dim] = upperHalo ? haloSizes[currHaloDim * 2 + 1]
+ : haloSizes[currHaloDim * 2];
+ // Check if we need to send and/or receive
+ // Processes on the mesh borders have only one neighbor
+ auto to = upperHalo ? neighbourIDs[1] : neighbourIDs[0];
+ auto from = upperHalo ? neighbourIDs[0] : neighbourIDs[1];
+ auto hasFrom = rewriter.create<arith::CmpIOp>(
+ loc, arith::CmpIPredicate::sge, from, zero);
+ auto hasTo = rewriter.create<arith::CmpIOp>(
+ loc, arith::CmpIPredicate::sge, to, zero);
+ auto buffer = rewriter.create<memref::AllocOp>(
+ loc, shape, array.getType().getElementType());
+ // if has neighbor: copy halo data from array to buffer and send
+ rewriter.create<scf::IfOp>(
+ loc, hasTo, [&](OpBuilder &builder, Location loc) {
+ offsets[dim] = upperHalo ? OpFoldResult(builder.getIndexAttr(0))
+ : OpFoldResult(upperOffset);
+ auto subview = builder.create<memref::SubViewOp>(
+ loc, array, offsets, shape, strides);
+ builder.create<memref::CopyOp>(loc, subview, buffer);
+ builder.create<mpi::SendOp>(loc, TypeRange{}, buffer, tag, to);
+ builder.create<scf::YieldOp>(loc);
+ });
+ // if has neighbor: receive halo data into buffer and copy to array
+ rewriter.create<scf::IfOp>(
+ loc, hasFrom, [&](OpBuilder &builder, Location loc) {
+ offsets[dim] = upperHalo ? OpFoldResult(upperOffset)
+ : OpFoldResult(builder.getIndexAttr(0));
+ builder.create<mpi::RecvOp>(loc, TypeRange{}, buffer, tag, from);
+ auto subview = builder.create<memref::SubViewOp>(
+ loc, array, offsets, shape, strides);
+ builder.create<memref::CopyOp>(loc, buffer, subview);
+ builder.create<scf::YieldOp>(loc);
+ });
+ rewriter.create<memref::DeallocOp>(loc, buffer);
+ offsets[dim] = orgOffset;
+ };
+
+ genSendRecv(dim, false);
+ genSendRecv(dim, true);
----------------
fschlimb wrote:
Good catch. It actually updates the corners: the lower dims exchange the data including the halos of the upper dims. The logic is currently not quite right. It needs to exchange the halos in descending dim order (right now it does the opposite).
Will fix it.
https://github.com/llvm/llvm-project/pull/104566
More information about the Mlir-commits
mailing list