- 19 12月, 2019 40 次提交
-
-
由 Gunhan Gulsoy 提交于
PiperOrigin-RevId: 286256041 Change-Id: Ie61edb82e440deda698d21b7655ecde550c97ca2
-
由 TensorFlower Gardener 提交于
PiperOrigin-RevId: 286253056 Change-Id: I9928d21effe92bed1a9f131a47d9e172cd970330
-
由 Daniel Situnayake 提交于
PiperOrigin-RevId: 286252818 Change-Id: Id74db3b9ed867c59b2d32cfe70079445f60ca96c
-
由 Feng Liu 提交于
The optimization pass might introduce constant in the tf dialect due to the constant folding for fully connected op. These constants should be converted to standard / tfl constant instead so the quantization can work. PiperOrigin-RevId: 286251602 Change-Id: Icc6c8512f7709b719cbe0409bfe14afcb90cd571
-
由 Prakalp Srivastava 提交于
'outfeed_config' is imported as a default valued string attribute. Custom support during export is required because of the shape argument required by the XlaBuilder API to set outfeed_shape field of HLO instruction. PiperOrigin-RevId: 286245978 Change-Id: I15040903f41c00c9080e090033a06580ade47fdc
-
由 Sean Silva 提交于
This makes it easier to narrow down on ops that are preventing inlining. PiperOrigin-RevId: 286243868 Change-Id: I7184463b74ff89fff65db1c0610d5295fda3cb16
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 286242467 Change-Id: I9b023f6e979a18978d3cc10cca2706134bb8611e
-
由 T.J. Alumbaugh 提交于
PiperOrigin-RevId: 286241942 Change-Id: Ie1320c17f6a50468a03dad2664a1c8645e09f3ce
-
由 A. Unique TensorFlower 提交于
https://gitlab.com/libeigen/eigen/commit/7252163335f56f23fcc7381c1efdea47161005fa PiperOrigin-RevId: 286240066 Change-Id: I55b71da4cb6148bf2f6ec3f8daffd4cabe34a522
-
由 Andy Ly 提交于
This is based on the current lowering/evaluation at https://github.com/tensorflow/tensorflow/blob/f4d29bb3635d2a7a9bb1f557d486de7fdaa3b101/tensorflow/compiler/tf2xla/kernels/concat_op.cc#L121 where for constant concat_dim and shape input tensors, an exclusive cumulative sum is performed over the concat_dim dimension of all shape input tensors. All values in other dimensions are zero'd out. PiperOrigin-RevId: 286237744 Change-Id: I49adf851d0dc5f2e6d52cac946b08b120f5e1302
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 286237503 Change-Id: Ib77a42b8f22f142d15ef6f42135852a3c8188d53
-
由 Mark Daoust 提交于
Converted it to a notebook. Expanded content. PiperOrigin-RevId: 286235401 Change-Id: I7419a59b34ee7ad56b2a29732c99204ae8cb5f64
-
由 River Riddle 提交于
Move the specializations of VectorTransferRewriter::matchAndRewrite back into the anonymous namespace. This appeases the GCC bug related to specializations in a different namespace. PiperOrigin-RevId: 286234667 Change-Id: I2f56a3f0e8bf7116c4721da11bc3a7b2ef756349
-
由 Marcel Koester 提交于
Added test cases for the newly added LLVM operations and lowering features. Closes #300 COPYBARA_INTEGRATE_REVIEW=https://github.com/tensorflow/mlir/pull/300 from dfki-jugr:std_to_llvm da6168bbc1a369ae2e99ad3881fdddd82f075dd4 PiperOrigin-RevId: 286231169 Change-Id: I364bda2adad4054052a3a53b4cb1c3de1102864f
-
由 Aart Bik 提交于
Examples: vector.print %f : f32 vector.print %x : vector<4xf32> vector.print %y : vector<3x4xf32> vector.print %z : vector<2x3x4xf32> LLVM lowering replaces these with fully unrolled calls into a small runtime support library that provides some basic printing operations (single value, opening closing bracket, comma, newline). PiperOrigin-RevId: 286230325 Change-Id: I4aec72975a3cb9612ab0aa44984c44aa014aae95
-
由 Andy Ly 提交于
ConvertMaxPoolOp, ConvertRangeOp and ConvertSigmoidOp were inserted twice to the patterns list. PiperOrigin-RevId: 286228460 Change-Id: I5ddaaa22d97ddc2414e2858d20d47402429bffa0
-
由 Dong Lin 提交于
PiperOrigin-RevId: 286228455 Change-Id: Id85aef40d98edfce4a93b4e9ab2eda304f54b865
-
由 River Riddle 提交于
See: https://llvm.org/docs/CodingStandards.html#include-iostream-is-forbidden PiperOrigin-RevId: 286226467 Change-Id: I2dfa26209902ecf695991a40b90a3f5dabc724da
-
由 River Riddle 提交于
This enables providing a default implementation of an interface method. This method is defined on the Trait that is attached to the operation, and thus has all of the same constraints and properties as any other interface method. This allows for interface authors to provide a conservative default implementation for certain methods, without requiring that all users explicitly define it. The default implementation can be specified via the argument directly after the interface method body: StaticInterfaceMethod< /*desc=*/"Returns whether two array of types are compatible result types for an op.", /*retTy=*/"bool", /*methodName=*/"isCompatibleReturnTypes", /*args=*/(ins "ArrayRef<Type>":$lhs, "ArrayRef<Type>":$rhs), /*methodBody=*/[{ return ConcreteOp::isCompatibleReturnTypes(lhs, rhs); }], /*defaultImplementation=*/[{ /// Returns whether two arrays are equal as strongest check for /// compatibility by default. return lhs == rhs; }] PiperOrigin-RevId: 286226054 Change-Id: I275d34d7a410d86813ca42291b0561bb2892cb18
-
由 Suharsh Sivakumar 提交于
PiperOrigin-RevId: 286225461 Change-Id: Iaabb7ed02e34bd36ed19cf774e79872255714356
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 286225031 Change-Id: I866ae69a24a47517aba285c40dbd9f2ecd615026
-
由 Prakalp Srivastava 提交于
PiperOrigin-RevId: 286224973 Change-Id: If77849b23b0ae49188df7ceb464908a8515b49ce
-
由 Jacques Pienaar 提交于
'```mlir' is used to indicate the code block is MLIR code/should use MLIR syntax highlighting, while '{.mlir}' was a markdown extension that used a style file to color the background differently of the code block. The background color extension was a custom one that we can retire given we have syntax highlighting. Also change '```td' to '```tablegen' to match chroma syntax highlighting designation. PiperOrigin-RevId: 286222976 Change-Id: I051210e78f5dcffb9b81cb521e59b94d72f208df
-
由 River Riddle 提交于
* Fixes use of anonymous namespace for static methods. * Uses explicit qualifiers(mlir::) instead of wrapping the definition with the namespace. PiperOrigin-RevId: 286222654 Change-Id: Ieffa72e3fefe50a6786347e736f120160e322bec
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 286221343 Change-Id: I04d49e928242de0df7eb8fe0b941221532c21368
-
由 Huanming Fang 提交于
PiperOrigin-RevId: 286216085 Change-Id: Id9bdb9167f2a0f0947c077ac6962a5cd3481b228
-
由 Uday Bondhugula 提交于
Introduce affine.prefetch: op to prefetch using a multi-dimensional subscript on a memref; similar to affine.load but has no effect on semantics, but only on performance. Provide lowering through std.prefetch, llvm.prefetch and map to llvm's prefetch instrinsic. All attributes reflected through the lowering - locality hint, rw, and instr/data cache. affine.prefetch %0[%i, %j + 5], false, 3, true : memref<400x400xi32> Signed-off-by: NUday Bondhugula <uday@polymagelabs.com> Closes #225 COPYBARA_INTEGRATE_REVIEW=https://github.com/tensorflow/mlir/pull/225 from bondhugula:prefetch 4c3b4e93bc64d9a5719504e6d6e1657818a2ead0 PiperOrigin-RevId: 286212997 Change-Id: I53484a6922b8f0fc52b610d0d29d6031112edfdb
-
由 Sergei Lebedev 提交于
PiperOrigin-RevId: 286209278 Change-Id: I66a292995dfefe2affc4dca098d87193342691f2
-
由 TensorFlower Gardener 提交于
PiperOrigin-RevId: 286208472 Change-Id: I7c8b0ce73988280b8b2cce0d7bb8494a190c570b
-
由 Guangda Lai 提交于
PiperOrigin-RevId: 286208460 Change-Id: I183a7887bf3a5ba32bf60e43c6c13fb7a1992c0e
-
由 River Riddle 提交于
Aside from being cleaner, this also makes the codebase more consistent. PiperOrigin-RevId: 286206974 Change-Id: I2eb3b84bfa317f1f3e3f04aeabe65e8e752ebb4a
-
由 Jacques Pienaar 提交于
LookUp only can return one error state (not found) which is already communicated by a nullptr & Status doesn't add additional value here. Communicating that only 1 error type is possible allows callers to handle that without needing to check error case first. Status return may still make sense for other OpRegistryInstances (e.g., remote op registries), so this doesn't update that and simply forwards the call. There could be a follow up to unify those though. PiperOrigin-RevId: 286205682 Change-Id: Ic7917e21457e2547111db4ff90a82fbb5d3efcfc
-
由 TensorFlower Gardener 提交于
PiperOrigin-RevId: 286205298 Change-Id: I9aed62e561f1ad23028bcfef42d41b8d478e09b1
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 286204492 Change-Id: I55b9bee40f5b1a98b813ec65f033560c163d6ca5
-
由 Martin Wicke 提交于
The change omits necessary checks to is_initialized(), which could lead to data races. PiperOrigin-RevId: 286203946 Change-Id: I678d05ccc7c5220e2d30111a853fd20f505fe933
-
由 A. Unique TensorFlower 提交于
The definition of the function template LLVM::ModuleTranslation::lookupValues has been located in a source file. As long as it has been the only file that actually called into the function, this did not cause any problem. However, it creates linking issues if the function is used from other translation units. PiperOrigin-RevId: 286203078 Change-Id: Ia593be5a572736d694e31a4e23634f8de2f2506b
-
由 Jacques Pienaar 提交于
This was munging up the example with the text. PiperOrigin-RevId: 286201762 Change-Id: If4260a51f687f7bbd5bdd93b52622380c2dd6f3a
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 286199768 Change-Id: I354bc27b3a083c2903364d155bf601dd370204e4
-
由 Derek Murray 提交于
Most paths to IndicesValid() implicitly assume that the indices are in the standard, row-major, order. For example, SparseToDenseOp makes this assumption, and it accounts for the majority of time spent in IndicesValid(). In addition, this change caches the `InlinedVector<int64>::data()` call implied by indexing into `SparseTensor::shape_`, which avoids a branch every time the vector is accessed. This change improves the throughput of the sparse_to_dense_op microbenchmark, which stresses IndicesValid(), by between 1.1% (BM_SparseToDense/977k) and 15% (BM_SparseToDense/3M). PiperOrigin-RevId: 286196888 Change-Id: I79b3075b8325f5b7345c134de6c4a3028a6d33e2
-
由 Chris Jones 提交于
PiperOrigin-RevId: 286192006 Change-Id: I9ce963a82048a7951e9134a9235f5c484db73df6
-