[llvm-branch-commits] [mlir] [mlir][MemRef] Use specialized index ops to fold expand/collapse_shape (PR #138930)
https://github.com/MaheshRavishankar approved this pull request. The changes look good to me. It isn't strictly required, by given that book h of us work on the same downstream project, does this pass with the said downstream project. But this looks good to me https://github.com/llvm/llvm-project/pull/138930 ___ llvm-branch-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [mlir] [mlir][MemRef] Use specialized index ops to fold expand/collapse_shape (PR #138930)
krzysz00 wrote: Ping https://github.com/llvm/llvm-project/pull/138930 ___ llvm-branch-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [mlir] [mlir][MemRef] Use specialized index ops to fold expand/collapse_shape (PR #138930)
llvmbot wrote:
@llvm/pr-subscribers-mlir-memref
Author: Krzysztof Drewniak (krzysz00)
Changes
This PR updates the FoldMemRefAliasOps to use `affine.linearize_index` and
`affine.delinearize_index` to perform the index computations needed to fold a
`memref.expand_shape` or `memref.collapse_shape` into its consumers,
respectively.
This also loosens some limitations of the pass:
1. The existing `output_shape` argument to `memref.expand_shape` is now used,
eliminating the need to re-infer this shape or call `memref.dim`.
2. Because we're using `affine.delinearize_index`, the restriction that each
group in a `memref.collapse_shape` can only have one dynamic dimension is
removed.
---
Patch is 31.32 KiB, truncated to 20.00 KiB below, full version:
https://github.com/llvm/llvm-project/pull/138930.diff
3 Files Affected:
- (modified) mlir/include/mlir/Dialect/MemRef/IR/MemRefOps.td (+10-4)
- (modified) mlir/lib/Dialect/MemRef/Transforms/FoldMemRefAliasOps.cpp
(+49-120)
- (modified) mlir/test/Dialect/MemRef/fold-memref-alias-ops.mlir (+64-65)
``diff
diff --git a/mlir/include/mlir/Dialect/MemRef/IR/MemRefOps.td
b/mlir/include/mlir/Dialect/MemRef/IR/MemRefOps.td
index d6d8161d3117b..f34b5b46cab50 100644
--- a/mlir/include/mlir/Dialect/MemRef/IR/MemRefOps.td
+++ b/mlir/include/mlir/Dialect/MemRef/IR/MemRefOps.td
@@ -1342,14 +1342,14 @@ def MemRef_ReinterpretCastOp
according to specified offsets, sizes, and strides.
```mlir
-%result1 = memref.reinterpret_cast %arg0 to
+%result1 = memref.reinterpret_cast %arg0 to
offset: [9],
sizes: [4, 4],
strides: [16, 2]
: memref<8x8xf32, strided<[8, 1], offset: 0>> to
memref<4x4xf32, strided<[16, 2], offset: 9>>
-%result2 = memref.reinterpret_cast %result1 to
+%result2 = memref.reinterpret_cast %result1 to
offset: [0],
sizes: [2, 2],
strides: [4, 2]
@@ -1755,6 +1755,12 @@ def MemRef_ExpandShapeOp :
MemRef_ReassociativeReshapeOp<"expand_shape", [
OpBuilder &b, Location loc, MemRefType expandedType,
ArrayRef reassociation,
ArrayRef inputShape);
+
+// Return a vector with all the static and dynamic values in the output
shape.
+SmallVector getMixedOutputShape() {
+ OpBuilder builder(getContext());
+ return ::mlir::getMixedValues(getStaticOutputShape(), getOutputShape(),
builder);
+}
}];
let hasVerifier = 1;
@@ -1873,7 +1879,7 @@ def MemRef_StoreOp : MemRef_Op<"store",
let summary = "store operation";
let description = [{
The `store` op stores an element into a memref at the specified indices.
-
+
The number of indices must match the rank of the memref. The indices must
be in-bounds: `0 <= idx < dim_size`
@@ -2025,7 +2031,7 @@ def SubViewOp :
MemRef_OpWithOffsetSizesAndStrides<"subview", [
Unlike the `reinterpret_cast`, the values are relative to the strided
memref of the input (`%result1` in this case) and not its
underlying memory.
-
+
Example 2:
```mlir
diff --git a/mlir/lib/Dialect/MemRef/Transforms/FoldMemRefAliasOps.cpp
b/mlir/lib/Dialect/MemRef/Transforms/FoldMemRefAliasOps.cpp
index e4fb3f9bb87ed..2acb90613e5d1 100644
--- a/mlir/lib/Dialect/MemRef/Transforms/FoldMemRefAliasOps.cpp
+++ b/mlir/lib/Dialect/MemRef/Transforms/FoldMemRefAliasOps.cpp
@@ -59,92 +59,28 @@ using namespace mlir;
///
/// %2 = load %0[6 * i1 + i2, %i3] :
/// memref<12x42xf32>
-static LogicalResult
-resolveSourceIndicesExpandShape(Location loc, PatternRewriter &rewriter,
-memref::ExpandShapeOp expandShapeOp,
-ValueRange indices,
-SmallVectorImpl &sourceIndices) {
- // Record the rewriter context for constructing ops later.
- MLIRContext *ctx = rewriter.getContext();
-
- // Capture expand_shape's input dimensions as `SmallVector`.
- // This is done for the purpose of inferring the output shape via
- // `inferExpandOutputShape` which will in turn be used for suffix product
- // calculation later.
- SmallVector srcShape;
- MemRefType srcType = expandShapeOp.getSrcType();
-
- for (int64_t i = 0, e = srcType.getRank(); i < e; ++i) {
-if (srcType.isDynamicDim(i)) {
- srcShape.push_back(
- rewriter.create(loc, expandShapeOp.getSrc(), i)
- .getResult());
-} else {
- srcShape.push_back(rewriter.getIndexAttr(srcType.getShape()[i]));
-}
- }
-
- auto outputShape = inferExpandShapeOutputShape(
- rewriter, loc, expandShapeOp.getResultType(),
- expandShapeOp.getReassociationIndices(), srcShape);
- if (!outputShape.has_value())
-return failure();
+static LogicalResult resolveSourceIndicesExpandShape(
+Location loc, PatternRewriter &rewriter,
+memref::ExpandShapeOp expandShapeOp, ValueRange indices,
+SmallVectorImpl &sourceIndices, bool startsInbounds) {
+ SmallVector destShape = expandShapeOp.getMixedOu
[llvm-branch-commits] [mlir] [mlir][MemRef] Use specialized index ops to fold expand/collapse_shape (PR #138930)
llvmbot wrote:
@llvm/pr-subscribers-mlir
Author: Krzysztof Drewniak (krzysz00)
Changes
This PR updates the FoldMemRefAliasOps to use `affine.linearize_index` and
`affine.delinearize_index` to perform the index computations needed to fold a
`memref.expand_shape` or `memref.collapse_shape` into its consumers,
respectively.
This also loosens some limitations of the pass:
1. The existing `output_shape` argument to `memref.expand_shape` is now used,
eliminating the need to re-infer this shape or call `memref.dim`.
2. Because we're using `affine.delinearize_index`, the restriction that each
group in a `memref.collapse_shape` can only have one dynamic dimension is
removed.
---
Patch is 31.32 KiB, truncated to 20.00 KiB below, full version:
https://github.com/llvm/llvm-project/pull/138930.diff
3 Files Affected:
- (modified) mlir/include/mlir/Dialect/MemRef/IR/MemRefOps.td (+10-4)
- (modified) mlir/lib/Dialect/MemRef/Transforms/FoldMemRefAliasOps.cpp
(+49-120)
- (modified) mlir/test/Dialect/MemRef/fold-memref-alias-ops.mlir (+64-65)
``diff
diff --git a/mlir/include/mlir/Dialect/MemRef/IR/MemRefOps.td
b/mlir/include/mlir/Dialect/MemRef/IR/MemRefOps.td
index d6d8161d3117b..f34b5b46cab50 100644
--- a/mlir/include/mlir/Dialect/MemRef/IR/MemRefOps.td
+++ b/mlir/include/mlir/Dialect/MemRef/IR/MemRefOps.td
@@ -1342,14 +1342,14 @@ def MemRef_ReinterpretCastOp
according to specified offsets, sizes, and strides.
```mlir
-%result1 = memref.reinterpret_cast %arg0 to
+%result1 = memref.reinterpret_cast %arg0 to
offset: [9],
sizes: [4, 4],
strides: [16, 2]
: memref<8x8xf32, strided<[8, 1], offset: 0>> to
memref<4x4xf32, strided<[16, 2], offset: 9>>
-%result2 = memref.reinterpret_cast %result1 to
+%result2 = memref.reinterpret_cast %result1 to
offset: [0],
sizes: [2, 2],
strides: [4, 2]
@@ -1755,6 +1755,12 @@ def MemRef_ExpandShapeOp :
MemRef_ReassociativeReshapeOp<"expand_shape", [
OpBuilder &b, Location loc, MemRefType expandedType,
ArrayRef reassociation,
ArrayRef inputShape);
+
+// Return a vector with all the static and dynamic values in the output
shape.
+SmallVector getMixedOutputShape() {
+ OpBuilder builder(getContext());
+ return ::mlir::getMixedValues(getStaticOutputShape(), getOutputShape(),
builder);
+}
}];
let hasVerifier = 1;
@@ -1873,7 +1879,7 @@ def MemRef_StoreOp : MemRef_Op<"store",
let summary = "store operation";
let description = [{
The `store` op stores an element into a memref at the specified indices.
-
+
The number of indices must match the rank of the memref. The indices must
be in-bounds: `0 <= idx < dim_size`
@@ -2025,7 +2031,7 @@ def SubViewOp :
MemRef_OpWithOffsetSizesAndStrides<"subview", [
Unlike the `reinterpret_cast`, the values are relative to the strided
memref of the input (`%result1` in this case) and not its
underlying memory.
-
+
Example 2:
```mlir
diff --git a/mlir/lib/Dialect/MemRef/Transforms/FoldMemRefAliasOps.cpp
b/mlir/lib/Dialect/MemRef/Transforms/FoldMemRefAliasOps.cpp
index e4fb3f9bb87ed..2acb90613e5d1 100644
--- a/mlir/lib/Dialect/MemRef/Transforms/FoldMemRefAliasOps.cpp
+++ b/mlir/lib/Dialect/MemRef/Transforms/FoldMemRefAliasOps.cpp
@@ -59,92 +59,28 @@ using namespace mlir;
///
/// %2 = load %0[6 * i1 + i2, %i3] :
/// memref<12x42xf32>
-static LogicalResult
-resolveSourceIndicesExpandShape(Location loc, PatternRewriter &rewriter,
-memref::ExpandShapeOp expandShapeOp,
-ValueRange indices,
-SmallVectorImpl &sourceIndices) {
- // Record the rewriter context for constructing ops later.
- MLIRContext *ctx = rewriter.getContext();
-
- // Capture expand_shape's input dimensions as `SmallVector`.
- // This is done for the purpose of inferring the output shape via
- // `inferExpandOutputShape` which will in turn be used for suffix product
- // calculation later.
- SmallVector srcShape;
- MemRefType srcType = expandShapeOp.getSrcType();
-
- for (int64_t i = 0, e = srcType.getRank(); i < e; ++i) {
-if (srcType.isDynamicDim(i)) {
- srcShape.push_back(
- rewriter.create(loc, expandShapeOp.getSrc(), i)
- .getResult());
-} else {
- srcShape.push_back(rewriter.getIndexAttr(srcType.getShape()[i]));
-}
- }
-
- auto outputShape = inferExpandShapeOutputShape(
- rewriter, loc, expandShapeOp.getResultType(),
- expandShapeOp.getReassociationIndices(), srcShape);
- if (!outputShape.has_value())
-return failure();
+static LogicalResult resolveSourceIndicesExpandShape(
+Location loc, PatternRewriter &rewriter,
+memref::ExpandShapeOp expandShapeOp, ValueRange indices,
+SmallVectorImpl &sourceIndices, bool startsInbounds) {
+ SmallVector destShape = expandShapeOp.getMixedOutputSha
[llvm-branch-commits] [mlir] [mlir][MemRef] Use specialized index ops to fold expand/collapse_shape (PR #138930)
krzysz00 wrote: PRs this depends on: * #138929 https://github.com/llvm/llvm-project/pull/138930 ___ llvm-branch-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [mlir] [mlir][MemRef] Use specialized index ops to fold expand/collapse_shape (PR #138930)
https://github.com/krzysz00 created
https://github.com/llvm/llvm-project/pull/138930
This PR updates the FoldMemRefAliasOps to use `affine.linearize_index` and
`affine.delinearize_index` to perform the index computations needed to fold a
`memref.expand_shape` or `memref.collapse_shape` into its consumers,
respectively.
This also loosens some limitations of the pass:
1. The existing `output_shape` argument to `memref.expand_shape` is now used,
eliminating the need to re-infer this shape or call `memref.dim`.
2. Because we're using `affine.delinearize_index`, the restriction that each
group in a `memref.collapse_shape` can only have one dynamic dimension is
removed.
Rate limit · GitHub
body {
background-color: #f6f8fa;
color: #24292e;
font-family: -apple-system,BlinkMacSystemFont,Segoe
UI,Helvetica,Arial,sans-serif,Apple Color Emoji,Segoe UI Emoji,Segoe UI Symbol;
font-size: 14px;
line-height: 1.5;
margin: 0;
}
.container { margin: 50px auto; max-width: 600px; text-align: center;
padding: 0 24px; }
a { color: #0366d6; text-decoration: none; }
a:hover { text-decoration: underline; }
h1 { line-height: 60px; font-size: 48px; font-weight: 300; margin: 0px;
text-shadow: 0 1px 0 #fff; }
p { color: rgba(0, 0, 0, 0.5); margin: 20px 0 40px; }
ul { list-style: none; margin: 25px 0; padding: 0; }
li { display: table-cell; font-weight: bold; width: 1%; }
.logo { display: inline-block; margin-top: 35px; }
.logo-img-2x { display: none; }
@media
only screen and (-webkit-min-device-pixel-ratio: 2),
only screen and ( min--moz-device-pixel-ratio: 2),
only screen and ( -o-min-device-pixel-ratio: 2/1),
only screen and (min-device-pixel-ratio: 2),
only screen and (min-resolution: 192dpi),
only screen and (min-resolution: 2dppx) {
.logo-img-1x { display: none; }
.logo-img-2x { display: inline-block; }
}
#suggestions {
margin-top: 35px;
color: #ccc;
}
#suggestions a {
color: #66;
font-weight: 200;
font-size: 14px;
margin: 0 10px;
}
Whoa there!
You have exceeded a secondary rate limit.
Please wait a few minutes before you try again;
in some cases this may take up to an hour.
https://support.github.com/contact";>Contact Support —
https://githubstatus.com";>GitHub Status —
https://twitter.com/githubstatus";>@githubstatus
___
llvm-branch-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
