Skip to content

[mlir] Add sub-byte type emulation support for memref.collapse_shape #89962

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 3 commits into from
Apr 26, 2024
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
32 changes: 29 additions & 3 deletions mlir/lib/Dialect/MemRef/Transforms/EmulateNarrowType.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -13,7 +13,6 @@
#include "mlir/Dialect/Arith/Transforms/Passes.h"
#include "mlir/Dialect/Arith/Utils/Utils.h"
#include "mlir/Dialect/MemRef/IR/MemRef.h"
#include "mlir/Dialect/MemRef/Transforms/Passes.h"
#include "mlir/Dialect/MemRef/Transforms/Transforms.h"
#include "mlir/Dialect/MemRef/Utils/MemRefUtils.h"
#include "mlir/Dialect/Vector/IR/VectorOps.h"
Expand All @@ -24,7 +23,6 @@
#include "mlir/Support/MathExtras.h"
#include "mlir/Transforms/DialectConversion.h"
#include "llvm/Support/FormatVariadic.h"
#include "llvm/Support/MathExtras.h"
#include <cassert>
#include <type_traits>

Expand Down Expand Up @@ -430,6 +428,33 @@ struct ConvertMemRefSubview final : OpConversionPattern<memref::SubViewOp> {
}
};

//===----------------------------------------------------------------------===//
// ConvertMemRefCollapseShape
//===----------------------------------------------------------------------===//

/// Emulating a `memref.collapse_shape` becomes a no-op after emulation given
/// that we flatten memrefs to a single dimension as part of the emulation and
/// there is no dimension to collapse any further.
struct ConvertMemRefCollapseShape final
: OpConversionPattern<memref::CollapseShapeOp> {
using OpConversionPattern::OpConversionPattern;

LogicalResult
matchAndRewrite(memref::CollapseShapeOp collapseShapeOp, OpAdaptor adaptor,
ConversionPatternRewriter &rewriter) const override {
Value srcVal = adaptor.getSrc();
auto newTy = dyn_cast<MemRefType>(srcVal.getType());
if (!newTy)
return failure();

if (newTy.getRank() != 1)
return failure();

rewriter.replaceOp(collapseShapeOp, srcVal);
return success();
}
};

} // end anonymous namespace

//===----------------------------------------------------------------------===//
Expand All @@ -442,7 +467,8 @@ void memref::populateMemRefNarrowTypeEmulationPatterns(

// Populate `memref.*` conversion patterns.
patterns.add<ConvertMemRefAllocation<memref::AllocOp>,
ConvertMemRefAllocation<memref::AllocaOp>, ConvertMemRefLoad,
ConvertMemRefAllocation<memref::AllocaOp>,
ConvertMemRefCollapseShape, ConvertMemRefLoad,
ConvertMemrefStore, ConvertMemRefAssumeAlignment,
ConvertMemRefSubview, ConvertMemRefReinterpretCast>(
typeConverter, patterns.getContext());
Expand Down
20 changes: 20 additions & 0 deletions mlir/test/Dialect/MemRef/emulate-narrow-type.mlir
Original file line number Diff line number Diff line change
Expand Up @@ -430,3 +430,23 @@ func.func @rank_zero_memref_store(%arg0: i4) -> () {
// CHECK32: %[[EXTUI:.+]] = arith.extui %[[ARG0]] : i4 to i32
// CHECK32: %[[WRITE_RMW:.+]] = memref.atomic_rmw assign %[[EXTUI]], %[[ALLOC]][] : (i32, memref<i32>) -> i32
// CHECK32: return

// -----

func.func @memref_collapse_shape_i4(%idx0 : index, %idx1 : index) -> i4 {
%arr = memref.alloc() : memref<32x8x128xi4>
%collapse = memref.collapse_shape %arr[[0, 1], [2]] : memref<32x8x128xi4> into memref<256x128xi4>
%1 = memref.load %collapse[%idx0, %idx1] : memref<256x128xi4>
return %1 : i4
}

// CHECK-LABEL: func.func @memref_collapse_shape_i4(
// CHECK: %[[ALLOC:.*]] = memref.alloc() : memref<16384xi8>
// CHECK-NOT: memref.collapse_shape
// CHECK: memref.load %[[ALLOC]][%{{.*}}] : memref<16384xi8>

// CHECK32-LABEL: func.func @memref_collapse_shape_i4(
// CHECK32: %[[ALLOC:.*]] = memref.alloc() : memref<4096xi32>
// CHECK32-NOT: memref.collapse_shape
// CHECK32: memref.load %[[ALLOC]][%{{.*}}] : memref<4096xi32>