[mlir][sparse][tensor] replace bufferization with empty tensor #66450
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Rationale: A bufferization.alloc_tensor can be directly replaced with tensor.empty since these are more or less semantically equivalent. The latter is considered a bit more "pure" with respect to SSA semantics.
aartbik
requested review from
yinying-lisa-li,
matthias-springer and
PeimingLiu
|
@llvm/pr-subscribers-mlir-sparse @llvm/pr-subscribers-mlir ChangesRationale: A bufferization.alloc_tensor can be directly replaced with tensor.empty since these are more or less semantically equivalent. The latter is considered a bit more "pure" with respect to SSA semantics.-- Patch is 78.42 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/66450.diff 48 Files Affected:
diff --git a/mlir/test/Integration/Dialect/SparseTensor/CPU/dense_output.mlir b/mlir/test/Integration/Dialect/SparseTensor/CPU/dense_output.mlir
index 89bf215a2c7788b..4ef8b29ee4e1a84 100644
--- a/mlir/test/Integration/Dialect/SparseTensor/CPU/dense_output.mlir
+++ b/mlir/test/Integration/Dialect/SparseTensor/CPU/dense_output.mlir
@@ -71,7 +71,7 @@ module {
%c2 = arith.constant 2.0 : f64
%d0 = tensor.dim %arga, %c0 : tensor<?x?xf64, #SparseMatrix>
%d1 = tensor.dim %arga, %c1 : tensor<?x?xf64, #SparseMatrix>
- %init = bufferization.alloc_tensor(%d0, %d1) : tensor<?x?xf64, #DenseMatrix>
+ %init = tensor.empty(%d0, %d1) : tensor<?x?xf64, #DenseMatrix>
%0 = linalg.generic #trait_assign
ins(%arga: tensor<?x?xf64, #SparseMatrix>)
outs(%init: tensor<?x?xf64, #DenseMatrix>) {
diff --git a/mlir/test/Integration/Dialect/SparseTensor/CPU/dense_output_bf16.mlir b/mlir/test/Integration/Dialect/SparseTensor/CPU/dense_output_bf16.mlir
index 420d3d8c6232744..317c7af990f78c4 100644
--- a/mlir/test/Integration/Dialect/SparseTensor/CPU/dense_output_bf16.mlir
+++ b/mlir/test/Integration/Dialect/SparseTensor/CPU/dense_output_bf16.mlir
@@ -48,7 +48,7 @@ module {
%argb: tensor<?xbf16, #SparseVector>) -> tensor<?xbf16, #DenseVector> {
%c = arith.constant 0 : index
%d = tensor.dim %arga, %c : tensor<?xbf16, #SparseVector>
- %xv = bufferization.alloc_tensor (%d) : tensor<?xbf16, #DenseVector>
+ %xv = tensor.empty (%d) : tensor<?xbf16, #DenseVector>
%0 = linalg.generic #trait_vec_op
ins(%arga, %argb: tensor<?xbf16, #SparseVector>, tensor<?xbf16, #SparseVector>)
outs(%xv: tensor<?xbf16, #DenseVector>) {
diff --git a/mlir/test/Integration/Dialect/SparseTensor/CPU/dense_output_f16.mlir b/mlir/test/Integration/Dialect/SparseTensor/CPU/dense_output_f16.mlir
index 96ea972bd6b5f0e..7c8510d8fbabc92 100644
--- a/mlir/test/Integration/Dialect/SparseTensor/CPU/dense_output_f16.mlir
+++ b/mlir/test/Integration/Dialect/SparseTensor/CPU/dense_output_f16.mlir
@@ -49,7 +49,7 @@ module {
%argb: tensor<?xf16, #SparseVector>) -> tensor<?xf16, #DenseVector> {
%c = arith.constant 0 : index
%d = tensor.dim %arga, %c : tensor<?xf16, #SparseVector>
- %xv = bufferization.alloc_tensor (%d) : tensor<?xf16, #DenseVector>
+ %xv = tensor.empty (%d) : tensor<?xf16, #DenseVector>
%0 = linalg.generic #trait_vec_op
ins(%arga, %argb: tensor<?xf16, #SparseVector>, tensor<?xf16, #SparseVector>)
outs(%xv: tensor<?xf16, #DenseVector>) {
diff --git a/mlir/test/Integration/Dialect/SparseTensor/CPU/dual_sparse_conv_2d.mlir b/mlir/test/Integration/Dialect/SparseTensor/CPU/dual_sparse_conv_2d.mlir
index 0488f5186a4a77d..6cf99cf45997d43 100644
--- a/mlir/test/Integration/Dialect/SparseTensor/CPU/dual_sparse_conv_2d.mlir
+++ b/mlir/test/Integration/Dialect/SparseTensor/CPU/dual_sparse_conv_2d.mlir
@@ -51,7 +51,7 @@ module {
func.func @conv2d_all_sparse_DCSR(%input: tensor<8x8xi32, #DCSR>,
%filter: tensor<3x3xi32, #DCSR>) -> tensor<6x6xi32, #DCSR> {
- %s = bufferization.alloc_tensor() : tensor<6x6xi32, #DCSR>
+ %s = tensor.empty() : tensor<6x6xi32, #DCSR>
%0 = linalg.conv_2d
ins (%input, %filter: tensor<8x8xi32, #DCSR>, tensor<3x3xi32, #DCSR>)
outs (%s: tensor<6x6xi32, #DCSR>) -> tensor<6x6xi32, #DCSR>
@@ -60,7 +60,7 @@ module {
func.func @conv2d_all_sparse_CSR(%input: tensor<8x8xi32, #CSR>,
%filter: tensor<3x3xi32, #CSR>) -> tensor<6x6xi32, #CSR> {
- %s = bufferization.alloc_tensor() : tensor<6x6xi32, #CSR>
+ %s = tensor.empty() : tensor<6x6xi32, #CSR>
%0 = linalg.conv_2d
ins (%input, %filter: tensor<8x8xi32, #CSR>, tensor<3x3xi32, #CSR>)
outs (%s: tensor<6x6xi32, #CSR>) -> tensor<6x6xi32, #CSR>
@@ -69,7 +69,7 @@ module {
func.func @conv2d_all_sparse_CD(%input: tensor<8x8xi32, #CDR>,
%filter: tensor<3x3xi32, #CDR>) -> tensor<6x6xi32, #CDR> {
- %s = bufferization.alloc_tensor() : tensor<6x6xi32, #CDR>
+ %s = tensor.empty() : tensor<6x6xi32, #CDR>
%0 = linalg.conv_2d
ins (%input, %filter: tensor<8x8xi32, #CDR>, tensor<3x3xi32, #CDR>)
outs (%s: tensor<6x6xi32, #CDR>) -> tensor<6x6xi32, #CDR>
@@ -78,7 +78,7 @@ module {
func.func @conv2d_all_sparse_CSC(%input: tensor<8x8xi32, #CSC>,
%filter: tensor<3x3xi32, #CSC>) -> tensor<6x6xi32, #CSC> {
- %s = bufferization.alloc_tensor() : tensor<6x6xi32, #CSC>
+ %s = tensor.empty() : tensor<6x6xi32, #CSC>
%0 = linalg.conv_2d
ins (%input, %filter: tensor<8x8xi32, #CSC>, tensor<3x3xi32, #CSC>)
outs (%s: tensor<6x6xi32, #CSC>) -> tensor<6x6xi32, #CSC>
diff --git a/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_abs.mlir b/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_abs.mlir
index 584906034d2d20e..71054e456e49475 100644
--- a/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_abs.mlir
+++ b/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_abs.mlir
@@ -46,7 +46,7 @@ module {
-> tensor<?xf64, #SparseVector> {
%c0 = arith.constant 0 : index
%d = tensor.dim %arg0, %c0 : tensor<?xf64, #SparseVector>
- %xin = bufferization.alloc_tensor(%d) : tensor<?xf64, #SparseVector>
+ %xin = tensor.empty(%d) : tensor<?xf64, #SparseVector>
%0 = linalg.generic #trait_op
ins(%arg0: tensor<?xf64, #SparseVector>)
outs(%xin: tensor<?xf64, #SparseVector>) {
@@ -61,7 +61,7 @@ module {
-> tensor<?xi32, #SparseVector> {
%c0 = arith.constant 0 : index
%d = tensor.dim %arg0, %c0 : tensor<?xi32, #SparseVector>
- %xin = bufferization.alloc_tensor(%d) : tensor<?xi32, #SparseVector>
+ %xin = tensor.empty(%d) : tensor<?xi32, #SparseVector>
%0 = linalg.generic #trait_op
ins(%arg0: tensor<?xi32, #SparseVector>)
outs(%xin: tensor<?xi32, #SparseVector>) {
diff --git a/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_binary.mlir b/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_binary.mlir
index 917f8a4838f4de5..826bf0da0ec81f3 100644
--- a/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_binary.mlir
+++ b/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_binary.mlir
@@ -73,7 +73,7 @@ module {
%argb: tensor<?xi32, #SparseVector>) -> tensor<?xi32, #SparseVector> {
%c = arith.constant 0 : index
%d = tensor.dim %arga, %c : tensor<?xi32, #SparseVector>
- %xv = bufferization.alloc_tensor(%d) : tensor<?xi32, #SparseVector>
+ %xv = tensor.empty(%d) : tensor<?xi32, #SparseVector>
%0 = linalg.generic #trait_vec_op
ins(%arga, %argb: tensor<?xi32, #SparseVector>, tensor<?xi32, #SparseVector>)
outs(%xv: tensor<?xi32, #SparseVector>) {
@@ -97,7 +97,7 @@ module {
%argb: tensor<?xf64>) -> tensor<?xf64, #SparseVector> {
%c = arith.constant 0 : index
%d = tensor.dim %arga, %c : tensor<?xf64, #SparseVector>
- %xv = bufferization.alloc_tensor(%d) : tensor<?xf64, #SparseVector>
+ %xv = tensor.empty(%d) : tensor<?xf64, #SparseVector>
%0 = linalg.generic #trait_vec_op
ins(%arga, %argb: tensor<?xf64, #SparseVector>, tensor<?xf64>)
outs(%xv: tensor<?xf64, #SparseVector>) {
@@ -121,7 +121,7 @@ module {
%argb: tensor<?xf64, #SparseVector>) -> tensor<?xf64, #SparseVector> {
%c = arith.constant 0 : index
%d = tensor.dim %arga, %c : tensor<?xf64, #SparseVector>
- %xv = bufferization.alloc_tensor(%d) : tensor<?xf64, #SparseVector>
+ %xv = tensor.empty(%d) : tensor<?xf64, #SparseVector>
%0 = linalg.generic #trait_vec_op
ins(%arga, %argb: tensor<?xf64, #SparseVector>, tensor<?xf64, #SparseVector>)
outs(%xv: tensor<?xf64, #SparseVector>) {
@@ -139,7 +139,7 @@ module {
func.func @vector_index(%arga: tensor<?xf64, #SparseVector>) -> tensor<?xi32, #SparseVector> {
%c = arith.constant 0 : index
%d = tensor.dim %arga, %c : tensor<?xf64, #SparseVector>
- %xv = bufferization.alloc_tensor(%d) : tensor<?xi32, #SparseVector>
+ %xv = tensor.empty(%d) : tensor<?xi32, #SparseVector>
%0 = linalg.generic #trait_vec_scale
ins(%arga: tensor<?xf64, #SparseVector>)
outs(%xv: tensor<?xi32, #SparseVector>) {
@@ -166,7 +166,7 @@ module {
%c1 = arith.constant 1 : index
%d0 = tensor.dim %arga, %c0 : tensor<?x?xf64, #DCSR>
%d1 = tensor.dim %arga, %c1 : tensor<?x?xf64, #DCSR>
- %xv = bufferization.alloc_tensor(%d0, %d1) : tensor<?x?xf64, #DCSR>
+ %xv = tensor.empty(%d0, %d1) : tensor<?x?xf64, #DCSR>
%0 = linalg.generic #trait_mat_op
ins(%arga, %argb: tensor<?x?xf64, #DCSR>, tensor<?x?xf64, #DCSR>)
outs(%xv: tensor<?x?xf64, #DCSR>) {
@@ -191,7 +191,7 @@ module {
// Tensor addition (use semi-ring binary operation).
func.func @add_tensor_1(%A: tensor<4x4xf64, #DCSR>,
%B: tensor<4x4xf64, #DCSR>) -> tensor<4x4xf64, #DCSR> {
- %C = bufferization.alloc_tensor() : tensor<4x4xf64, #DCSR>
+ %C = tensor.empty() : tensor<4x4xf64, #DCSR>
%0 = linalg.generic #trait_mat_op
ins(%A, %B: tensor<4x4xf64, #DCSR>,
tensor<4x4xf64, #DCSR>)
@@ -213,7 +213,7 @@ module {
// Same as @add_tensor_1, but use sparse_tensor.yield instead of identity to yield value.
func.func @add_tensor_2(%A: tensor<4x4xf64, #DCSR>,
%B: tensor<4x4xf64, #DCSR>) -> tensor<4x4xf64, #DCSR> {
- %C = bufferization.alloc_tensor() : tensor<4x4xf64, #DCSR>
+ %C = tensor.empty() : tensor<4x4xf64, #DCSR>
%0 = linalg.generic #trait_mat_op
ins(%A, %B: tensor<4x4xf64, #DCSR>,
tensor<4x4xf64, #DCSR>)
@@ -241,7 +241,7 @@ module {
// Performs triangular add/sub operation (using semi-ring binary op).
func.func @triangular(%A: tensor<4x4xf64, #DCSR>,
%B: tensor<4x4xf64, #DCSR>) -> tensor<4x4xf64, #DCSR> {
- %C = bufferization.alloc_tensor() : tensor<4x4xf64, #DCSR>
+ %C = tensor.empty() : tensor<4x4xf64, #DCSR>
%0 = linalg.generic #trait_mat_op
ins(%A, %B: tensor<4x4xf64, #DCSR>,
tensor<4x4xf64, #DCSR>)
@@ -274,7 +274,7 @@ module {
// Perform sub operation (using semi-ring binary op) with a constant threshold.
func.func @sub_with_thres(%A: tensor<4x4xf64, #DCSR>,
%B: tensor<4x4xf64, #DCSR>) -> tensor<4x4xf64, #DCSR> {
- %C = bufferization.alloc_tensor() : tensor<4x4xf64, #DCSR>
+ %C = tensor.empty() : tensor<4x4xf64, #DCSR>
// Defines out-block constant bounds.
%thres_out_up = arith.constant 2.0 : f64
%thres_out_lo = arith.constant -2.0 : f64
@@ -323,7 +323,7 @@ module {
// Performs isEqual only on intersecting elements.
func.func @intersect_equal(%A: tensor<4x4xf64, #DCSR>,
%B: tensor<4x4xf64, #DCSR>) -> tensor<4x4xi8, #DCSR> {
- %C = bufferization.alloc_tensor() : tensor<4x4xi8, #DCSR>
+ %C = tensor.empty() : tensor<4x4xi8, #DCSR>
%0 = linalg.generic #trait_mat_op
ins(%A, %B: tensor<4x4xf64, #DCSR>,
tensor<4x4xf64, #DCSR>)
@@ -346,7 +346,7 @@ module {
// Keeps values on left, negate value on right, ignore value when overlapping.
func.func @only_left_right(%A: tensor<4x4xf64, #DCSR>,
%B: tensor<4x4xf64, #DCSR>) -> tensor<4x4xf64, #DCSR> {
- %C = bufferization.alloc_tensor() : tensor<4x4xf64, #DCSR>
+ %C = tensor.empty() : tensor<4x4xf64, #DCSR>
%0 = linalg.generic #trait_mat_op
ins(%A, %B: tensor<4x4xf64, #DCSR>,
tensor<4x4xf64, #DCSR>)
diff --git a/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_cmp.mlir b/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_cmp.mlir
index f6c72581153bfac..87ab88b8d9de99c 100644
--- a/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_cmp.mlir
+++ b/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_cmp.mlir
@@ -66,7 +66,7 @@ module {
func.func @cmp_lhs_sparse(%arga: tensor<4x4xf64, #DCSR>,
%argb: tensor<4x4xf64>) -> tensor<4x4xi8, #DCSR> {
- %argx = bufferization.alloc_tensor() : tensor<4x4xi8, #DCSR>
+ %argx = tensor.empty() : tensor<4x4xi8, #DCSR>
%0 = linalg.generic #trait
ins(%arga, %argb: tensor<4x4xf64, #DCSR>, tensor<4x4xf64>)
outs(%argx: tensor<4x4xi8, #DCSR>) {
@@ -80,7 +80,7 @@ module {
func.func @cmp_all_sparse(%arga: tensor<4x4xf64, #DCSR>,
%argb: tensor<4x4xf64, #DCSR>) -> tensor<4x4xi8, #DCSR> {
- %argx = bufferization.alloc_tensor() : tensor<4x4xi8, #DCSR>
+ %argx = tensor.empty() : tensor<4x4xi8, #DCSR>
%0 = linalg.generic #trait
ins(%arga, %argb: tensor<4x4xf64, #DCSR>, tensor<4x4xf64, #DCSR>)
outs(%argx: tensor<4x4xi8, #DCSR>) {
diff --git a/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_codegen_dim.mlir b/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_codegen_dim.mlir
index 3203473f68b324d..45ea95d1a6f36fd 100644
--- a/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_codegen_dim.mlir
+++ b/mlir/test/Integration/Dialect/SparseTensor/CPU/sparse_codegen_dim.mlir
@@ -43,8 +43,8 @@ module {
%c1 = arith.constant 1 : index
%c2 = arith.constant 2 : index
%c3 = arith.constant 3 : index
- %t1 = bufferization.alloc_tensor() : tensor<4x5xf64, #DCSR>
- %t2 = bufferization.alloc_tensor(%c2, %c3) : tensor<?x?xf64, #DCSR&g...
|
PeimingLiu
approved these changes
Sep 15, 2023
ZijunZhaoCCK
pushed a commit
to ZijunZhaoCCK/llvm-project
that referenced
this pull request
Sep 19, 2023
…66450) Rationale: A bufferization.alloc_tensor can be directly replaced with tensor.empty since these are more or less semantically equivalent. The latter is considered a bit more "pure" with respect to SSA semantics.
zahiraam
pushed a commit
to tahonermann/llvm-project
that referenced
this pull request
Oct 24, 2023
…66450) Rationale: A bufferization.alloc_tensor can be directly replaced with tensor.empty since these are more or less semantically equivalent. The latter is considered a bit more "pure" with respect to SSA semantics.
zahiraam
pushed a commit
to tahonermann/llvm-project
that referenced
this pull request
Oct 24, 2023
…66450) Rationale: A bufferization.alloc_tensor can be directly replaced with tensor.empty since these are more or less semantically equivalent. The latter is considered a bit more "pure" with respect to SSA semantics.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Rationale:
A bufferization.alloc_tensor can be directly replaced
with tensor.empty since these are more or less semantically
equivalent. The latter is considered a bit more "pure"
with respect to SSA semantics.