Home
last modified time | relevance | path

Searched full:convolution (Results 1 – 25 of 47) sorted by relevance

12

/llvm-project/mlir/test/Dialect/Linalg/
H A Dconv-interface-invalid.mlir32 // expected-error @+1 {{unexpected input index map for convolution}}
51 // expected-error @+1 {{unexpected input index map for convolution}}
106 // Convolution op illegal if a loop dimension is used to access
110 // expected-error @+1 {{unexpected loop dimension for convolution op}}
127 // Convolution op illegal if a loop dimension is used only in the output.
130 // expected-error @+1 {{unexpected loop dimension for convolution op}}
148 // Convolution op illegal if a loop dimension is used only in the filter.
151 // expected-error @+1 {{unexpected loop dimension for convolution op}}
169 // Convolution op illegal if a loop dimension is used only in the input.
172 // expected-error @+1 {{unexpected loop dimension for convolution op}}
[all …]
H A Dvectorize-conv-masked-and-scalable.mlir50 /// Convolution
112 /// Convolution
174 /// Convolution
/llvm-project/mlir/include/mlir/Dialect/Linalg/IR/
H A DLinalgInterfaces.h74 /// convolution dimension.
87 /// dimension candidates that form a convolution subcomputation within
88 /// `linalgOp`. The LHS is assumed to be the convolution input while the
105 /// This allows e.g. detecting that some convolution is embedded within
186 /// convolution.
199 /// Returns the error message corresponding to the convolution checking return
H A DLinalgNamedStructuredOps.yaml2199 Performs 1-D convolution with no channels.
2266 Performs 2-D convolution with no channels.
2335 Performs 3-D convolution with no channels.
2407 Performs 1-D convolution.
2491 Performs 1-D convolution.
2579 Performs 2-D convolution.
2677 Performs 2-D convolution.
2775 Performs 2-D convolution with zero point offsets.
2912 Performs 2-D convolution with zero point offsets.
3049 Performs 2-D convolution wit
[all...]
/llvm-project/mlir/lib/Dialect/Linalg/Transforms/
H A DTransposeConv2D.cpp1 //===- TransposeConv2D.cpp - Convolution transposition -------------------===//
28 /// Convolution converter that applies the following rewrite:
54 // convolution this will be known statically as [1, 2, 3, 0]. in transposeConv2DHelper()
100 // It is possible the convolution doesn't define any results and its in transposeConv2DHelper()
H A DVectorization.cpp55 /// Try to vectorize `convOp` as a convolution.
88 // convolution. in extractConvInputSlices()
99 // for channeled convolution. in extractConvInputSlices()
121 // non-chanelled convolution] @ [kw]. in extractConvFilterSlices()
137 // Extract res slice: {wSizeStep} @ [w] for non-channeled convolution. in extractConvResultSlices()
146 // convolution. in extractConvResultSlices()
164 // Write back res slice: {wSizeStep} @ [w] for non-channeled convolution. in insertConvResultSlices()
173 // convolution. This does not depend on kw. in insertConvResultSlices()
517 W, // Corresponds to non-channeled 1D convolution operation. in getCombinerOpKind()
1821 // Support dynamic shapes in 1D depthwise convolution, bu in vectorizeLinalgOpPrecondition()
[all...]
/llvm-project/mlir/docs/Tutorials/transform/
H A DChH.md12 ## Channeled Convolution
36 We will consider the 2D channeled convolution example extracted from Halide
54 // Declarations of "mathematical functions" for convolution and relu.
64 // Core convolution with the result initialized to the bias value.
85 // Convolution proper. While Linalg has named operations for 2D convolutions,
325 Practically, this corresponds to fusing the convolution initialization and
330 * first the main convolution update is fused into ReLU that uses it and has
332 * then the bias initialization is fused into the convolution+relu loop nest.
349 loops from the convolution operation. However, these are reduction loops and it
363 This transformation materializes the desired loops around the convolution
[all …]
/llvm-project/mlir/python/mlir/dialects/linalg/opdsl/ops/
H A Dcore_named_ops.py664 """Performs 1-D convolution with no channels.
680 """Performs 2-D convolution with no channels.
698 """Performs 3-D convolution with no channels.
718 """Performs 1-D convolution.
738 """Performs 1-D convolution.
762 """Performs 2-D convolution.
786 """Performs 2-D convolution.
812 """Performs 2-D convolution with zero point offsets.
842 """Performs 2-D convolution with zero point offsets.
872 """Performs 2-D convolution wit
[all...]
/llvm-project/mlir/include/mlir/Dialect/Linalg/TransformOps/
H A DLinalgMatchOps.td184 "Checks if an operation has convolution-like dimensions and returns them";
186 Checks if the structured payload op has convolution-like dimensions as
200 Additionally this will match stride and dilation information for the convolution:
201 - 'strides' are the static strides per convolution window dimension;
202 - 'dilations' are the static dilations per convolution window dimension.
210 Succeeds if the operation has the convolution-like dimensions, produces a
/llvm-project/mlir/lib/Analysis/Presburger/
H A DUtils.cpp548 /// coefficients, by taking the convolution.
551 // The length of the convolution is the sum of the lengths in multiplyPolynomials()
562 std::vector<Fraction> convolution; in multiplyPolynomials()
563 convolution.reserve(len); in multiplyPolynomials() local
568 convolution.emplace_back(sum); in multiplyPolynomials()
570 return convolution; in multiplyPolynomials()
/llvm-project/llvm/test/Transforms/InstCombine/X86/
H A Dshufflemask-undef-inseltpoison.ll14 …%struct.Convolution = type { %struct.IColor4, %struct.ImagingColorScale, i16, i16, [0 x i32], ptr,…
21 …%struct.ImagingSubset = type { %struct.Convolution, %struct.Convolution, %struct.Convolution, %str…
H A Dshufflemask-undef.ll14 …%struct.Convolution = type { %struct.IColor4, %struct.ImagingColorScale, i16, i16, [0 x i32], ptr,…
21 …%struct.ImagingSubset = type { %struct.Convolution, %struct.Convolution, %struct.Convolution, %str…
/llvm-project/llvm/test/Transforms/LoopStrengthReduce/AArch64/
H A Dlsr-ldp.ll5 define void @convolution(ptr %src0, ptr %src1, i64 %stride_xm, i64 %stride_xp, ptr %dst, i32 %w) {
6 ; CHECK-LABEL: convolution:
/llvm-project/mlir/lib/Conversion/TosaToLinalg/
H A DTosaToLinalgNamed.cpp223 // Creates a map to collapse the last dimension of the Depthwise convolution op in createDepthwiseConvCollapseMap()
312 // For 2D convolutions, we need to check if the target convolution op in matchAndRewrite()
319 // convolution operation. in matchAndRewrite()
341 // convolution operation. Conv2D has a 1-1 mapping in linalg so better to in matchAndRewrite()
363 // Extract the attributes for convolution. in matchAndRewrite()
367 // Create the convolution op. in matchAndRewrite()
485 // Extract the attributes for convolution. in matchAndRewrite()
489 // Create the convolution op. in matchAndRewrite()
/llvm-project/mlir/lib/Dialect/Tosa/Transforms/
H A DTosaDecomposeTransposeConv.cpp49 // the x/y direction to make it a regular convolution. This is much simpler in createOpAndInfer()
119 // the x/y direction to make it a regular convolution. This is much simpler in matchAndRewrite()
230 // Perform the convolution using the zero bias. in matchAndRewrite()
259 // Factor striding out of the convolution result. in matchAndRewrite()
H A DTosaDecomposeConv2D.cpp10 // (1) Convert a 1x1 Convolution to a Reshape->FC->Reshape
/llvm-project/mlir/include/mlir/Dialect/Linalg/Transforms/
H A DTransforms.h1261 /// A convolution operation can be written as a matrix-matrix multiplication by
1285 /// and output (N, Ho, Wo, D) the convolution is the following matrix-matrix
1291 /// the final operation of the sequence that replaces the original convolution.
1304 /// reduction among the input channels so each convolution can be a
1404 /// Rewrites 2-D convolution ops with size-1 window dimensions into 1-D
1405 /// convolution ops.
1425 /// Rewrites 2-D depthwise convolution ops with size-1 (w, kw) or (h, kh)
1426 /// dimensions into 1-D depthwise convolution ops.
1706 /// Populates patterns to decompose high-D convolution ops into low-D ones.
1707 /// This is a step in progressive lowering for convolution op
[all...]
/llvm-project/mlir/lib/Dialect/Linalg/IR/
H A DLinalgInterfaces.cpp571 /// Walk the indexing expressions for input of a convolution operation to verify in getDimExprOrMulExprDimPos()
583 // Stores the dual mapping between LHS and RHS of convolution exprs. in getDimExprOrMulExprDimPos()
698 /// Classifies dimensions in the `linalgOp` used by a convolution in inferConvolutionDimsImpl()
701 /// at least convolved dimension pair (output image + filter loop). Convolution in inferConvolutionDimsImpl()
794 /// dimension candidates that form a convolution subcomputation within in isConvolutionInterfaceImpl()
795 /// `linalgOp`. The LHS is assumed to be the convolution input while the in isConvolutionInterfaceImpl()
812 /// This allows e.g. detecting that some convolution is embedded within in isConvolutionInterfaceImpl()
975 assert(succeeded(res) && "unexpected failure to infer convolution dims");
994 return "unexpected loop dimension for convolution op"; in createLoopRanges()
/llvm-project/mlir/test/Examples/transform/ChH/
H A Dfull.mlir8 // Fixed-size tensor types to be used in convolution.
18 // Function containing the convolution. Note that its arguments and results are
47 // Convolution proper. While Linalg has named operations for 2D convolutions,
194 // The loop reordering requested for the convolution operation requires
/llvm-project/mlir/test/Integration/Dialect/Linalg/CPU/ArmSVE/
H A D1d-depthwise-conv.mlir32 // Run the convolution
/llvm-project/mlir/test/python/dialects/linalg/opdsl/
H A Demit_convolution.py34 # Convolution indexing maps.
H A Demit_misc.py11 # fill, matmul, convolution, or pooling tests. The features include:
/llvm-project/mlir/unittests/Analysis/Presburger/
H A DUtilsTest.cpp79 TEST(UtilsTest, convolution) { in TEST() argument
/llvm-project/mlir/include/mlir/Dialect/Tosa/IR/
H A DTosaUtilOps.td36 multiple quantized operations (mul, convolution, rescale, matmul, pooling).
/llvm-project/mlir/test/Integration/Dialect/SparseTensor/CPU/
H A Dsparse_filter_conv2d.mlir36 // An example of a 2D convolution with a sparse filter.

12