This is an archive of the discontinued LLVM Phabricator instance.

Differential D89265

[mlir][Linalg] Temporarily circumvent TensorConstant bufferize bug
ClosedPublic

Authored by nicolasvasilache on Oct 12 2020, 1:22 PM.

Details

Reviewers
mehdi_amini
aartbik
Commits
rG81ead8a53525: [mlir][Linalg] Temporarily circumvent TensorConstant bufferize bug
Summary

The TensorConstantOp bufferize conversion pattern has a bug that
makes it incorrect in the case of vectors whose alignment is not
the natural alignment. Circumvent it temporarily by using a power of 2.

Diff Detail

Event Timeline

nicolasvasilache created this revision.Oct 12 2020, 1:22 PM
Herald added a reviewer: aartbik. · View Herald TranscriptOct 12 2020, 1:22 PM
Herald added a project: Restricted Project. · View Herald Transcript
Herald added subscribers: rdzhabarov, tatianashp, msifontes and 13 others. · View Herald Transcript
nicolasvasilache requested review of this revision.Oct 12 2020, 1:22 PM
Herald added subscribers: limo1996, stephenneuendorffer. · View Herald TranscriptOct 12 2020, 1:22 PM
aartbik accepted this revision.Oct 12 2020, 1:23 PM
This revision is now accepted and ready to land.Oct 12 2020, 1:23 PM
This revision was landed with ongoing or failed builds.Oct 12 2020, 1:25 PM
Closed by commit rG81ead8a53525: [mlir][Linalg] Temporarily circumvent TensorConstant bufferize bug (authored by nicolasvasilache). · Explain Why
This revision was automatically updated to reflect the committed changes.
nicolasvasilache added a commit: rG81ead8a53525: [mlir][Linalg] Temporarily circumvent TensorConstant bufferize bug.
Harbormaster completed remote builds in B74846: Diff 297678.Oct 12 2020, 1:40 PM

Revision Contents

PathSize
mlir/
integration_test/
Dialect/
Linalg/
CPU/
11 lines

Diff 297679

mlir/integration_test/Dialect/Linalg/CPU/test-tensor-matmul.mlir

// RUN: mlir-opt %s -convert-linalg-on-tensors-to-buffers -convert-linalg-to-loops -convert-linalg-to-llvm -convert-std-to-llvm | \ // RUN: mlir-opt %s -convert-linalg-on-tensors-to-buffers -convert-linalg-to-loops -convert-linalg-to-llvm -convert-std-to-llvm | \
// RUN: mlir-cpu-runner -e main -entry-point-result=void \ // RUN: mlir-cpu-runner -e main -entry-point-result=void \
// RUN: -shared-libs=%mlir_integration_test_dir/libmlir_runner_utils%shlibext \ // RUN: -shared-libs=%mlir_integration_test_dir/libmlir_runner_utils%shlibext \
// RUN: | FileCheck %s // RUN: | FileCheck %s
func @main() { func @main() {
%A = constant dense<[[1.0, 2.0, 3.0], [4.0, 5.0, 6.0]]> : tensor<2x3xf32> %A = constant dense<[[1.0, 2.0], [4.0, 5.0]]> : tensor<2x2xf32>
%B = constant dense<[[1.0, 2.0, 3.0, 4.0], %B = constant dense<[[1.0, 2.0, 3.0, 4.0],
[5.0, 6.0, 7.0, 8.0], [5.0, 6.0, 7.0, 8.0]]> : tensor<2x4xf32>
[9.0, 10.0, 11.0, 12.0]]> : tensor<3x4xf32>
%C = constant dense<1000.0> : tensor<2x4xf32> %C = constant dense<1000.0> : tensor<2x4xf32>
%D = linalg.matmul ins(%A, %B: tensor<2x3xf32>, tensor<3x4xf32>) %D = linalg.matmul ins(%A, %B: tensor<2x2xf32>, tensor<2x4xf32>)
init(%C: tensor<2x4xf32>) -> tensor<2x4xf32> init(%C: tensor<2x4xf32>) -> tensor<2x4xf32>
%unranked = tensor_cast %D : tensor<2x4xf32> to tensor<*xf32> %unranked = tensor_cast %D : tensor<2x4xf32> to tensor<*xf32>
call @print_memref_f32(%unranked) : (tensor<*xf32>) -> () call @print_memref_f32(%unranked) : (tensor<*xf32>) -> ()
// CHECK: Unranked Memref base@ = {{0x[-9a-f]*}} // CHECK: Unranked Memref base@ = {{0x[-9a-f]*}}
// CHECK-SAME: rank = 2 offset = 0 sizes = [2, 4] strides = [4, 1] data = // CHECK-SAME: rank = 2 offset = 0 sizes = [2, 4] strides = [4, 1] data =
// CHECK-NEXT: [1038, 1044, 1050, 1056] // CHECK-NEXT: [1011, 1014, 1017, 1020]
// CHECK-NEXT: [1065, 1074, 1083, 1092] // CHECK-NEXT: [1029, 1038, 1047, 1056]
return return
} }
func @print_memref_f32(%ptr : tensor<*xf32>) func @print_memref_f32(%ptr : tensor<*xf32>)