Lines Matching +full:non +full:- +full:batch
3 **Check the continuously-tested version of MLIR files under
4 [mlir/test/Examples/transform/Ch4](https://github.com/llvm/llvm-project/tree/main/mlir/test/Example…
23 becomes possible to set up a higher-level match and rewrite infrastructure
30 1](Ch1.md/#chaining-transformations-with-handles), reproduced below for
38 -> tensor<512x512xf32> {
39 // Matrix-matrix multiplication.
42 outs(%output: tensor<512x512xf32>) -> tensor<512x512xf32>
47 outs(%output : tensor<512x512xf32>) -> tensor<512x512xf32>
53 outs(%output : tensor<512x512xf32>) -> tensor<512x512xf32>
61 additional arguments, `bind-first-extra-to-ops=linalg.matmul
62 bind-second-extra-to-ops=linalg.elemwise_binary`, to provide initial
83 : (!transform.any_op) -> !transform.any_op
85 : (!transform.any_op) -> !transform.any_op
87 : (!transform.any_op) -> ()
89 : (!transform.any_op) -> ()
99 %entry: !transform.any_op {transform.readonly}) -> !transform.any_op {
105 %entry: !transform.any_op {transform.readonly}) -> !transform.any_op {
127 This script can be executed using the non-test interpreter pass running on the
128 root operation of the translation unit without additional flags: `mlir-opt
129 --transform-interpreter`. It will emit corresponding remarks at
132 by passing `-debug-only=transform-matcher` to `mlir-opt` or a derived tool. It
139 …transform-matcher] matching %0 = linalg.matmul ins(%arg0, %arg1 : tensor<512x512xf32>, tensor<512x…
140 [transform-matcher] matcher match_elemwise failed: wrong operation name
166 // successful. It starts matching from the last operation in the use-def chain
170 -> (!transform.any_op, !transform.any_op, !transform.any_op) {
180 : (!transform.any_op) -> !transform.any_op
187 : (!transform.any_op) -> !transform.any_op
209 : (!transform.any_op) -> (!transform.any_op, !transform.any_op, !transform.any_op)
213 : (!transform.any_op) -> ()
215 : (!transform.any_op) -> ()
263 $op `:` functional-type($op, results) attr-dict-with-keyword $body
270 operands it will match, has an associated single-block region containing the
289 // expected to modify payload, but use the tri-state result to signal failure or
306 for (OpOperand &operand : payload->getOpOperands()) {
334 // Report failure-to-match for debugging purposes and stop matching this
356 yieldedMappings, getBody().front().getTerminator()->getOperands(),
399 -> (!transform.any_op, !transform.any_op, !transform.any_op,
410 : (!transform.any_op) -> (!transform.param<i32>, !transform.any_op) {
414 : (!transform.any_value) -> !transform.any_op
424 : (!transform.any_op) -> !transform.any_op
444 to implement a simple high-level pattern rewriting approach within the transform
445 dialect (for advanced or lower-level pattern rewriting, consider PDL(L) or C++
460 @match_matmul_elemwise -> @print_matmul_elemwise
461 : (!transform.any_op) -> !transform.any_op
474 mostly rely on operation names and use-def chains. Alternative implementations
484 operation name-based matcher no longer applies. Yet such a representation is
486 transformation, e.g., where a higher-dimensional contraction is decomposed into
514 %candidate: !transform.any_op {transform.readonly}) -> !transform.any_op {
520 : (!transform.any_op) -> !transform.param<i64>
521 %c3 = transform.param.constant 3 : i64 -> !transform.param<i64>
526 : (!transform.any_op) -> !transform.param<i64>
527 %c2 = transform.param.constant 2 : i64 -> !transform.param<i64>
533 : (!transform.any_op) -> !transform.param<i64>
534 %c1 = transform.param.constant 1 : i64 -> !transform.param<i64>
546 %batch, %lhs, %rhs, %reduction =
549 -> (!transform.param<i64>, !transform.param<i64>, !transform.param<i64>,
553 // There is one of lhs, rhs and reduction dimensions and zero batch
555 %n_batch = transform.num_associations %batch
556 : (!transform.param<i64>) -> !transform.param<i64>
558 : (!transform.param<i64>) -> !transform.param<i64>
560 : (!transform.param<i64>) -> !transform.param<i64>
562 : (!transform.param<i64>) -> !transform.param<i64>
563 %c0 = transform.param.constant 0 : i64 -> !transform.param<i64>
574 While this example leverages the contraction-specific matchers that have a
575 rather non-trivial C++ implementation, the transform dialect is sufficiently