[Mlir-commits] [mlir] [mlir] add transform tutorial chapter for Halide conv mapping (PR #66386)
Oleksandr Alex Zinenko
llvmlistbot at llvm.org
Thu Sep 14 07:54:59 PDT 2023
https://github.com/ftynse updated https://github.com/llvm/llvm-project/pull/66386:
>From 72d9126bd0501c3fc08b66598d283e22304aa498 Mon Sep 17 00:00:00 2001
From: Alex Zinenko <zinenko at google.com>
Date: Thu, 14 Sep 2023 14:45:25 +0000
Subject: [PATCH] [mlir] add transform tutorial chapter for Halide conv mapping
This chapter demonstrates how one can replicate Halide DSL
transformations using transform dialect operations transforming payload
expressed using Linalg. This was a part of the live tutorial presented
at EuroLLVM 2023.
---
mlir/docs/Tutorials/transform/ChH.md | 691 +++++++++++++++++++++
mlir/docs/Tutorials/transform/_index.md | 1 +
mlir/test/Examples/transform/ChH/full.mlir | 393 ++++++++++++
3 files changed, 1085 insertions(+)
create mode 100644 mlir/docs/Tutorials/transform/ChH.md
create mode 100644 mlir/test/Examples/transform/ChH/full.mlir
diff --git a/mlir/docs/Tutorials/transform/ChH.md b/mlir/docs/Tutorials/transform/ChH.md
new file mode 100644
index 000000000000000..7f4d85e1e84a506
--- /dev/null
+++ b/mlir/docs/Tutorials/transform/ChH.md
@@ -0,0 +1,691 @@
+# Chapter H: Reproducing Halide Schedule
+
+This chapter demonstrates how a schedule from the [Halide
+DSL](http://halide-lang.org) can be implemented using transform dialect for
+structured ops.
+
+Note that the IR below is pseudo-code with types removed for brevity. It may
+also get out of sync with the current syntax. Always refer to the source code in
+[mlir/examples/transform/ChH](https://github.com/llvm/llvm-project/tree/main/mlir/test/Examples/transform/ChH)
+as the source of truth.
+
+## Channeled Convolution
+
+The Transform dialect provides a substrate for implementing “transformation
+directive” domain-specific languages (DSLs) in MLIR. Such a DSL, at least in its
+scheduling part, can target the operations in the Transform dialect that are
+later applied by the compiler. Sets of transform operations, or even new
+dialects leveraging the same interfaces and infrastructure, can be added to
+support a specific DSL for a particular scheduling model. In this chapter, we
+will revisit the Halide DSL that has (re)popularized separate specification of
+schedules originally for image processing programs.
+
+Two approaches Halide to the Transform dialect are possible:
+
+* Create a new dialect that corresponds to the computational part of Halide
+ DSL, and define a set of transformations wrapped into Transform dialect
+ operations, that correspond to the scheduling part of the DSL.
+* Map the Halide abstractions to the existing MLIR abstractions, for both
+ parts of the DSL.
+
+We will consider the latter approach as the computational part of the DSL easily
+maps to the structured ops in the Linalg dialect. This also gives us the
+opportunity to discuss how Linalg transformations on the so-called structured
+operations are similar to or different from the existing transformations.
+
+We will consider the 2D channeled convolution example extracted from Halide
+[application
+examples](https://github.com/halide/Halide/tree/294f80c49bf3bb8582446613c25fcce03b82bcd8/apps/conv_layer).
+
+```cpp
+// Sizes of the problem.
+const int N = 5, CI = 128, CO = 128, W = 100, H = 80;
+
+// Sized inputs. Note that the order of dimensions is
+// inverted in Halide with respect to C++, so the last dimension
+// in the list (N for input, CI for filter) is the least
+// frequently varying. The C++ equivalent is input[N][H+2][W+2][CI].
+Buffer<float, 4> input({CI, W+2, H+2, N}, "input");
+Buffer<float, 4> filter({CO, 3, 3, CI}, "filter");
+Buffer<float, 1> bias(std::vector<int>{CO}, "bias");
+
+// ... data initialization happens here ...
+
+// Declarations of "mathematical functions" for convolution and relu.
+Func conv("conv"), relu("relu");
+
+// Iterators/subscripts.
+Var x("x"), y("y"), c("c"), n("n");
+
+// 3D reduction domain (channels and 2 window dimensions),
+// dimensions are later referred to as r.x, r.y, r.z.
+RDom r(0, CI, 0, 3, 0, 3);
+
+// Core convolution with the result initialized to the bias value.
+// Note that the order of iterators is inverted in Halide DSL,
+// i.e. `n` corresponds to the lest frequently-varying (outermost) dimension
+// here and below.
+conv(c, x, y, n) = bias(c);
+conv(c, x, y, n) += filter(c, r.y, r.z, r.x) * input(r.x, x + r.y, y + r.z, n);
+
+// ReLU rectification, an elementwise operation.
+relu(c, x, y, n) = max(0, conv(c, x, y, n));
+```
+
+This can be almost directly converted to Linalg dialect operating on tensors,
+which is conceptually closer to the “mathematical function” abstraction and is
+where the majority of transformations are available.
+
+```mlir
+// Bias. Using a named Linalg operation for brevity.
+%bias_init = tensor.empty() : !toutput
+%biased = linalg.broadcast ins(%bias : !tbias)
+ outs(%bias_init : !toutput) dimensions = [0, 1, 2]
+
+// Convolution proper. While Linalg has named operations for 2D convolutions,
+// the one in the Halide example has an uncommon order of filter dimensions
+// and is not supported. It also takes the fitler as first argument. This
+// code recreates it faithfully using the generic form.
+%convolved = linalg.generic {
+ iterator_types = ["parallel", "parallel", "parallel", "parallel",
+ "reduction", "reduction", "reduction"],
+ indexing_maps = [
+ affine_map<(n, y, x, c, rz, ry, rx) -> (rx, rz, ry, c)>,
+ affine_map<(n, y, x, c, rz, ry, rx) -> (n, y+rz, x+ry, rx)>,
+ affine_map<(n, y, x, c, rz, ry, rx) -> (n, y, x, c)>
+ ]
+} ins(%filter, %input: !tfilter, !tinput)
+ outs(%biased : !toutput) {
+^bb0(%in: f32, %f: f32, %b: f32):
+ // Note the fastmath attributes that allow operations to be recombined into
+ // %0 = math.fma %in, %f, %b : f32
+ // later on and to reorder reductions.
+ %m1 = arith.mulf %in, %f {fastmath = #arith.fastmath<fast>} : f32
+ %0 = arith.addf %b, %m1 {fastmath = #arith.fastmath<fast>} : f32
+ linalg.yield %0 : f32
+} -> !toutput
+
+// ReLU is just a max(0, x).
+%c0 = arith.constant 0.0 : f32
+%relued = linalg.generic {
+ iterator_types = ["parallel", "parallel", "parallel", "parallel"],
+ indexing_maps = [
+ affine_map<(d0, d1, d2, d3) -> ()>,
+ affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>,
+ affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>
+ ]
+} ins(%c0, %convolved : f32, !toutput)
+ outs(%output : !toutput) {
+^bb0(%cst: f32, %in: f32, %out: f32):
+ %0 = llvm.intr.maxnum(%cst, %in) : (f32, f32) -> f32
+ linalg.yield %0 : f32
+} -> !toutput
+```
+
+In Halide, a function such as `conv` may consist of two parts: a “functional”
+initialization computation and an in-place update for reductions. This is
+expressed as two C++ statements in the embedded DSL, but internally is
+represented in a single object. Linalg doesn’t have such a capability to the
+initialization and the update are represented as two distinct Linalg operations
+that are not connected to each other. Furthermore, the `x`, `y`, `c`, `n`
+variables in Halide DSL correspond to implicit loops iterating over the
+corresponding objects, which implies that functions sharing these variables in
+their definitions also share the corresponding loops. In other words, the loop
+equivalent of the Halide definition starts in a fully-fused form. The Linalg
+model is the opposite with each structured operation corresponding to its own
+loop nest, resulting in a fully-distributed form. This will affect how the
+schedule is constructed later on.
+
+The loop structure for Halide computation resembles the following (adapted from
+debug dump with `HL_DEBUG_CODEGEN=1`)
+
+```python
+for n
+ for y
+ for x
+ for c
+ conv[n, y, x, c] = bias[c]
+ for rz
+ for ry
+ for rx
+ conv[n, y, x, c] += filter[rx, rz, ry, c] * input[n, y+rz, x+ry, rx]
+ relu[n, y, x, c] = max(0, conv[n, y, x, c])
+```
+
+The loop structure for the Linalg computation is as follows (obtained by
+`mlir-opt --linalg-generalize-named-ops --empty-tensor-to-alloc-tensor
+--one-shot-bufferize --convert-linalg-to-loops`)
+
+```python
+for n
+ for y
+ for x
+ for c
+ init[n, y, x, c] = bias[c]
+for n
+ for y
+ for x
+ for c
+ for rz
+ for ry
+ for rx
+ conv[n, y, x, c] += filter[rx, rz, ry, c] * input[n, y+rz, x+ry, rx]
+for n
+ for y
+ for x
+ for c
+ relu[n, y, x, c] = max(0, conv[n, y, x, c])
+
+```
+
+## Mapping Halide Scheduling Primitives to Linalg Structured Transforms
+
+The complete Halide schedule listed in the example is as follows
+
+```cpp
+Var co, ci, xo, xi;
+relu.split(c, co, ci, vec * tile_w)
+ .split(x, xo, xi, tile_h)
+ .reorder(ci, xi, xo, y, n, co)
+ .vectorize(ci, vec)
+ .unroll(ci)
+ .unroll(xi)
+ .parallel(y)
+ .parallel(n)
+ .parallel(co);
+
+conv.compute_at(relu, xo)
+ .vectorize(c, vec)
+ .unroll(c)
+ .unroll(x)
+ .unroll(y)
+ .update()
+ .reorder(c, x, y, r.x, r.y, r.z, n)
+ .vectorize(c, vec)
+ .unroll(c)
+ .unroll(x)
+ .unroll(y)
+ .unroll(r.x, 2);
+```
+
+We will consider only the case without parallelization to avoid the difference
+in parallel runtimes generated by Halide and used by MLIR. This schedule
+corresponds to a sequence of loop manipulations, unrolling and vectorization.
+The following directives are present and can be mapped to transformations on
+Linalg as described below.
+
+* `split` decomposes a loop dimension into two immediately nested loops with
+ the inner loop having at most the given number of iterations. This can be
+ understood as loop _strip-mining_ or a degenerate case of tiling a single
+ dimension using any of `linalg.tile_` transform ops. We will be using
+ `transform.structured.tile_to_forall_op` as this kind of loop is best
+ supported by bufferization and can also be turned into a parallel loop later
+ on. Unlike Halide, this doesn’t add new dimensions to the original
+ operation, but rather creates a loop around it and rewrites the operation
+ itself to operate on a subset of the original data.
+* `reorder` rearranges the loops arbitrarily. In Linalg representation, loops
+ are implicit and are intended to remain so as long as possible to target
+ microkernels. The order of implicit loops in a `linalg.generic` operation
+ can be changed by using `transform.structured.interchange`, but this does
+ not apply to named operations that need to be “generalized” first by calling
+ `transform.structured.generalize`. However, this can only reorder implicit
+ dimensions and not the explicit loops materialized by tiling operations that
+ can no longer be “folded” into the original operation. Instead, we can
+ leverage this behavior by materializing loops directly in the desired order
+ by “tiling” to size 1.
+* `vectorize` indicates that the given dimension should be vectorized with the
+ given factor; if the loop extent is larger than the factor, the loop is
+ effectively split into two parts and the inner one is vectorized. On the
+ contrary, structured Linalg op vectorization applies as a global
+ transformation to all suitable operations at, e.g., a function scope via
+ `transform.structured.vectorize`. It relies on MLIR’s support for
+ multidimensional vectors to directly map multidimensional tensors, which are
+ later decomposed into operations on smaller hardware-compatible vectors
+ during lowering.
+* `unroll` performs loop unrolling, fully or up to the given factor. It is
+ equivalent to `transform.loop.unroll`.
+* `compute_at` indicates that the value of the function must be computed
+ within the given loop that will be produced for another function; depending
+ on the relation between loops surrounding functions, this corresponds to
+ either a loop distribution or a producer/consumer fusion. Given that the
+ Linalg representation starts in the fully distributed form, it can be
+ represented as a sequence of `transform.structured.fuse_into_containing_op`
+ that operates on `forall` loops materialized by tiling beforehand.
+
+
+## Recreating the Loop Structure
+
+The three first transformation directives for `relu` in the Halide schedule aim
+at producing the following loop structure.
+
+```python
+for co
+ for n
+ for y
+ for xo
+ for xi
+ for ci
+ relu[n, y, xo*tile_h + xi, co*tile_w*vec + ci] = ...
+```
+
+Note that the outer part of the `c` gets hoisted from all of the surrounding
+loops. The implicit loop order for the operation is `n, y, x, c`, so the `co`
+loop needs to be materialized first in order to achieve the desired reordering.
+The remaining dimensions can be materialized as loops in one transformation.
+
+```mlir
+ // [n y x c]
+ %co, %relu2 = transform.structured.tile_to_forall_op %relu
+ tile_sizes [0, 0, 0, 64]
+ %n_y_xo, %relu3 = transform.structured.tile_to_forall_op %relu2
+ tile_sizes [1, 1, 5, 0]
+```
+
+This will result in the following loops being created in the IR with the nested
+elementwise operation operating on a smaller subset of original data via
+implicit loops.
+
+```mlir
+scf.forall (%co) in (2) {
+ scf.forall (%n, %y, %xo) in (5, 80, 20) {
+ tensor.extract_slice
+ // Implicit dimensions [ni=0:1, y=0:1, xi=0:5, ci=0:64]
+ %relued = linalg.elemwise_binary { fun = #linalg.binary_fn<max_signed> } // ...
+ scf.forall.in_parallel {
+ tensor.parallel_insert_slice // ...
+ }
+ }
+}
+```
+
+The following loop restructuring transformations are `compute_at` and `reorder`
+on the `conv` function that need to happen before loops are destroyed by
+unrolling and vectorization. They intend to produce the final desired loop
+structure.
+
+```python
+for co
+ for n
+ for y
+ for xo
+ for xi
+ for ci
+ conv[n, y, x*tile_h + xi, co*tile_w*vec + ci] = ...
+ for rz
+ for ry
+ for rx
+ for xi
+ for ci
+ conv[n, y, x*tile_h + xi, co*tile_w*vec + ci] += ...
+ for xi
+ for ci
+ relu[n, y, xo*tile_h + xi, co*tile_w*vec + ci] = ...
+```
+
+Practically, this corresponds to fusing the convolution initialization and
+update into the `co, n, y, xo` loops materialized by tiling earlier. Structured
+op transformation set supports fusing the producer of a value into its consumer,
+so fusion happens in two stages:
+
+* first the main convolution update is fused into ReLU that uses it and has
+ loops materialized;
+* then the bias initialization is fused into the convolution+relu loop nest.
+
+Each stage consists of two transformations fusing the computational operation
+into the outer loop, then the inner loop.
+
+```mlir
+%conv2, %co2 = transform.structured.fuse_into_containing_op %conv into %co
+%conv3, %n_y_xo2 = transform.structured.fuse_into_containing_op %conv2
+ into %n_y_xo
+
+%bias2, %co3 = transform.structured.fuse_into_containing_op %bias into %co2
+%bias3, %n_y_xo3 = transform.structured.fuse_into_containing_op %bias2
+ into %n_y_xo2
+```
+
+To complete the structure, we need to put the `rz, ry, rx` loops outside the
+“tile” loops `xi, ci`. This can be achieved materializing the corresponding
+loops from the convolution operation. However, these are reduction loops and it
+wouldn’t be valid to materialize them as intrinsically parallel “forall” loops.
+Instead, we use the dedicated “reduction tiling” transformation and produce
+sequential `scf.for` loops. (`scf.forall` loops can also express parallel
+reductions, but the corresponding transformation doesn’t handle reductions along
+more than one dimension at the moment of writing.)
+
+```mlir
+%rz_ry_rx, %red_fill, %conv4, %comb
+ = transform.structured.tile_reduction_using_scf %conv3
+// n y x c rz ry rx
+ by tile_sizes=[0, 0, 0, 0, 1, 1, 1]
+```
+
+This transformation materializes the desired loops around the convolution
+operation. It is also more capable than merely producing (reduction) loops: the
+transformed code performs `tile_size` partial reductions of `N / tile_size`
+elements, potentially in parallel by changing the dimension kind of the
+structured operation inside the loop, and then performs a final reduction of
+these partial results by producing a new “combiner” structured operation after
+the loops. In our case, `tile_size = 1` along all dimensions, so the reduction
+is entirely performed by the generated loops. The combiner structured operation
+is still produced and adds up the reduction result with the initial value. This
+changes the order of floating point operations (so would reduction tiling with
+non-unit size) and may affect the final result due to non-commutativity of these
+operations, but is explicitly allowed by `fastmath` flags. Halide also emits
+LLVM IR with full `fastmath` flags.
+
+Finally, we need to produce innermost loops `xi` and `ci` that are still not
+explicit. As our next step is going to be vectorization along `ci`, we need to
+take into account the way it operates on MLIR structured operations: rather than
+selecting a specific vector size and loop/dimension to vectorize, it directly
+substitutes multidimensional vector types for tensor types and updates the
+operations accordingly. Therefore, our tensor type should not become trivial,
+i.e. size-1, and retain a `vector_size` sized dimension along the desired axis,
+`ci`. This can be achieved by tiling with `vector_size` as tile size in that
+dimension:
+
+```mlir
+// n y xi ci
+%1, %c5 = transform.structured.tile_to_forall_op %conv4 tile_sizes [0, 0, 1, 16]
+%2, %b4 = transform.structured.tile_to_forall_op %bias3 tile_sizes [0, 0, 1, 16]
+%3, %r4 = transform.structured.tile_to_forall_op %relu3 tile_sizes [0, 0, 1, 16]
+%4, %c2 = transform.structured.tile_to_forall_op %comb tile_sizes [0, 0, 1, 16]
+```
+
+Note that the combiner operation produced by reduction tiling is also tiled here.
+
+
+## Explicit Loop Unrolling
+
+The remaining unhandled loop transformation is unrolling. Specifically,
+unrolling is requested for the innermost loops that form the 4x5 tile of
+16-element vector operations to ensure a contiguous sequence of `vfma`
+instructions using 20 512-bit vector registers as accumulators. Unrolling
+additional loops,, `unroll(y)` and `unroll(r.x, 2)`, is requested in the
+schedule but _has no practical effect_. That is, the code, and all intermediate
+representations, produced by Halide with these directives removed is _strictly
+identical_ to the code with the full schedule. Therefore, we will only unroll
+the corresponding loops corresponding to `xi` and `ci` dimensions that actually
+get unrolled by Halide.
+
+As tiling in the transform dialect produces handles to the loops materialized by
+tiling, unrolling those loops is just a matter of chaining the corresponding
+transformation. Note that the inner loop must be unrolled first as unrolling the
+outer loop will invalidate the handles to the inner loop.
+
+```mlir
+transform.loop.unroll %bias_ci {factor = 4}
+transform.loop.unroll %bias_xi {factor = 5}
+transform.loop.unroll %conv_ci {factor = 4}
+transform.loop.unroll %conv_xi {factor = 5}
+transform.loop.unroll %relu_ci {factor = 4}
+transform.loop.unroll %relu_xi {factor = 5}
+transform.loop.unroll %comb_ci {factor = 4}
+transform.loop.unroll %comb_xi {factor = 5}
+```
+
+## Vectorization
+
+These transformations produced the desired loop structure and we are now ready
+to vectorize. Before proceeding it is desirable to simplify the code as tiling
+and fusion may have produced a lot of operations computing tensor subsets and
+loop ranges, some of which may be duplicated or excessively complex.
+Simplification involving canonicalization, common subexpression elimination,
+loop invariant code motion and various rewrite patterns can be applied directly
+from the transform dialect. Furthermore, an arbitrary combination of rewrite
+patterns can be applied _in one sweep_ to a given scope, a functionality that
+_cannot be achieved with conventional compiler passes_ that apply each group of
+patterns separately (at least without creating a new pass for each combination
+of pattern groups).
+
+```mlir
+%f00 = transform.structured.match ops{["func.func"]} in %arg0
+transform.apply_patterns to %f00 {
+ transform.apply_patterns.canonicalization
+ transform.apply_patterns.linalg.tiling_canonicalization
+}
+transform.apply_cse to %f00
+
+%all_loops = transform.structured.match interface{LoopLikeInterface} in %arg0
+transform.apply_licm to %all_loops
+```
+
+One final simplification is necessary to produce good vectorized code.
+Tiling-by-one as a way of materializing loops produced structured (`linalg`)
+operations processing 4D types where only one dimension isn’t unit-sized, e.g.,
+`tensor<1x1x1x16xf32>` where 16 is the vector size corresponding to AVX512,
+as structured tiling doesn’t modify the rank of the operation in order to
+preserve the original structure. Even though the core computation is the same,
+the produced code may end up more complicated than necessary, in particular when
+decomposing multidimensional vectors into single-dimensional vectors supported
+by hardware. Such unit dimensions can be explicitly folded away using the
+corresponding pattern set before vectorization.
+
+```mlir
+transform.apply_patterns to %f00 {
+ transform.apply_patterns.linalg.fold_unit_extent_dims_via_reshapes
+}
+
+%fv = transform.structured.vectorize %f00
+```
+
+This produces the desired code performing arithmetic operations on
+`vector<16xf32>` types that can be easily lowered to AVX512 instructions by the
+downstream compiler. Vectorization may have created new opportunities for code
+simplification, in particular combining tensor subsetting and vector slicing
+operations. Another round of simplification can be applied post vectorization.
+
+```mlir
+transform.apply_patterns to %fv {
+ transform.apply_patterns.canonicalization
+ transform.apply_patterns.tensor.fold_tensor_subset_ops_into_vector_transfers
+}
+transform.apply_cse to %fv
+transform.structured.hoist_redundant_vector_transfers %fv
+```
+
+## Lowering to LLVM and The Bufferization Hurdle
+
+With the loop restructuring done, the program now needs to be converted to the
+executable form. The first step in doing so is _bufferization_, the process that
+associates a memory buffer with every tensor in the payload IR. MLIR’s one-shot
+bufferization is directly available as a transform operation.
+
+```mlir
+%arg1 = transform.bufferization.one_shot_bufferize %arg0 {
+ bufferize_function_boundaries = true,
+ function_boundary_type_conversion = 1 : i32 }
+```
+
+In this particular case, the transformed IR could be directly bufferized. This
+is not always the case in general as some operations, in particular
+`tensor.empty` may not be bufferizable. Such operations need to be removed
+before running the bufferization, which can often be achieved by sufficient
+fusion (as in our case), or by running dedicated transformations
+`transform.bufferization.eliminate_empty_tensors` that removes the
+`tensor.empty` operations only serving for defining the size of a computation or
+`transform.bufferization.empty_tensor_to_alloc_tensor` that materializes a new
+temporary buffer for empty tensors to be used as local caches.
+
+```mlir
+// Apply general canonicalization and CSE to each function after
+// bufferization as new simplification opportunities may have appeared.
+%fb = transform.structured.match ops{["func.func"]} in %arg1
+transform.apply_patterns to %fb {
+ transform.apply_patterns.canonicalization
+}
+transform.apply_cse to %fb
+
+// Lower complex, multidimensional vector operations into simpler
+// primitives. This particular selection of the pattern groups corresponds
+// to vector dialect operations present in the payload IR at this stage.
+// Many of these groups can be parameterized to use different strategies or
+// lower-level primitives offering performance trade-offs. In this case, we
+// are selecting the simplest strategies.
+transform.apply_patterns to %fb {
+ transform.apply_patterns.vector.lower_contraction
+ lowering_strategy = parallelarith
+ transform.apply_patterns.vector.lower_transfer
+ max_transfer_rank = 1
+ transform.apply_patterns.vector.lower_transpose
+ lowering_strategy = eltwise
+ transform.apply_patterns.vector.lower_shape_cast
+}
+
+// These patterns apply in a separate sweep to avoid transfer-to-scf
+// patterns overlap with lower-transfer patterns as they apply to the same
+// kind of operations. These patterns may produce local allocations to act
+// as temporary caches deep inside loops, which could lead to catastrophic
+// performance. Such allocations are moved onto the stack and hoisted from
+// all the surrounding loops.
+transform.apply_patterns to %fb {
+ transform.apply_patterns.vector.transfer_to_scf
+ transform.apply_patterns.memref.alloc_to_alloca
+ }
+transform.bufferization.buffer_loop_hoisting %fb
+
+// A final round of cleanups additionally includes patterns to simplify
+// buffer aliasing operations that may have been introduced during
+// bufferization and could result in excessively complex address
+// computation.
+transform.apply_patterns to %fb {
+ transform.apply_patterns.memref.fold_memref_alias_ops
+ transform.apply_patterns.canonicalization
+}
+transform.apply_cse to %fb
+```
+
+Due to its inter-procedural nature, one-bufferization processes the entire
+payload module and thus invalidates all previously created handles. Therefore,
+it is typically a late step in the transformation sequence where precise
+targeting of transformation is no longer required. The following transformations
+are typically module- or function-wide rewrites that are often pattern-based
+lowerings. This part of the sequence can be seen as a pass pipeline specified
+directly in the transform dialect, with pattern-based lowering passes
+constructed _on-the-fly_ from named groups of patterns.
+
+The resulting IR can be further completely lowered to the LLVM dialect, then to
+LLVM IR and processed by the LLVM compiler to produce an executable or JITted.
+
+The generated code runs in ~420ms on an Intel processor with Skylake-SP
+microarchitecture clocked at 2.0GHz. Given that the computation performs
+$5*80*100*128*(2*3*3*128 + 2) ~= 5.9 * 10^9$ floating point operations, it
+reaches ~14 GFlops. With 2 FMA units available, the single-core performance of
+the test processor is 64 GFlops $(16 * 2 * 2 * 10^9$, where 16 is the vector
+width), so only 22% of the theoretical peak is achieved.
+
+The code produced by Halide runs in ~120ms on the same processor, a 3.5x
+improvement and 77% of peak. Let us analyze the generated assembly to understand
+the source of the difference. The main computational effort is expected to
+happen around floating point multiplications and additions in the convolution.
+In both cases, the assembly features AVX512 `vfma231ps` instructions operating
+on `%zmm` 512-bit vector registers. In the MLIR-generated code, they are
+interspersed with memory accesses loading _two _of the `fma` operands before
+each operation and leading to increased latency.
+
+```asm
+vmovups -192(%r10), %zmm0
+vbroadcastss -1536(%rdi,%r9), %zmm1
+vmovups 112(%rsp), %zmm2
+vfmadd231ps %zmm1, %zmm0, %zmm2 # zmm2 = (zmm0 * zmm1) + zmm2
+vmovups %ymm2, 112(%rsp)
+vextractf64x4 $1, %zmm2, 144(%rsp)
+// 19 more blocks of either
+// (a) vmovups,vbroadcast,vfma(z,z),vextract,
+// (b) vbroadcast,vfma(z,mem),vextract
+```
+
+The Halide-generated code however features compact blocks of `vfma231ps` and
+`vbroadcastss` loading one of the operands while the other two are resident in
+registers and loaded before `fma`.
+
+```asm
+vbroadcastss -1536(%rsi,%rbx), %zmm25
+vmovups -192(%rdi), %zmm26
+vmovups -128(%rdi), %zmm27
+vmovups -64(%rdi), %zmm28
+vmovups (%rdi), %zmm29
+vfmadd231ps %zmm25, %zmm26, %zmm24 # zmm24 = (zmm26 * zmm25) + zmm24
+vfmadd231ps %zmm25, %zmm27, %zmm23 # zmm23 = (zmm27 * zmm25) + zmm23
+vfmadd231ps %zmm25, %zmm28, %zmm22 # zmm22 = (zmm28 * zmm25) + zmm22
+vfmadd231ps %zmm25, %zmm29, %zmm21 # zmm21 = (zmm29 * zmm25) + zmm21
+vbroadcastss -1024(%rsi,%rbx), %zmm25
+vfmadd231ps %zmm25, %zmm26, %zmm20 # zmm20 = (zmm26 * zmm25) + zmm20
+vfmadd231ps %zmm25, %zmm27, %zmm19 # zmm19 = (zmm27 * zmm25) + zmm19
+vfmadd231ps %zmm25, %zmm28, %zmm18 # zmm18 = (zmm28 * zmm25) + zmm18
+vfmadd231ps %zmm25, %zmm29, %zmm17 # zmm17 = (zmm29 * zmm25) + zmm17
+vbroadcastss -512(%rsi,%rbx), %zmm25
+
+// 3 more blocks of 4 vfmadd231 followed by a vbroadcast
+```
+
+Inspecting the progressive intermediate representations produced by MLIR, one
+can observe the load(transfer)/fma interspersing at all levels starting after
+schedule application. The repeated tensor subsetting operations, that are later
+transformed into vector transfer operations, and vector memory loads, are
+produced by loop unrolling that was explicitly requested in the schedule! The
+issue is the single-assignment model of tensors (and vectors) that results in
+long and complex chains of access and update operations that become so long that
+the lower-level transformations and the downstream compiler can no longer
+simplify them. In fact, unrolling loops early in the transformation sequence can
+lead to all sorts of compiler-performance related problems (including the
+compiler failing to perform some optimizations due to excessive code length) in
+the process.
+
+It is therefore desirable to perform loop unrolling at a later stage,
+specifically after bufferization and relevant simplification. However,
+bufferization invalidates all loop handles including to loops that we are
+willing to unroll. This hurdle can be overcome by matching the payload IR
+operations after bufferization to produce new handles. We will first change the
+kind of loops produced in the schedule from `scf.for` to `scf.forall` to have
+less operations to match by using `transform.structured.tile_to_forall_op`
+instead of `transform.structured.tile` when tiling with sizes `[0, 0, 1, 16]`.
+Then we can match all `scf.forall` operations in the payload IR and transform
+them into single-iterator `scf.for` loops _after bufferization_.
+
+```mlir
+%foralls = transform.structured.match ops{["scf.forall"]} in %arg1
+%xi_bias, %ci_bias = transform.loop.forall_to_for %xi_ci_bias
+%xi_conv, %ci_conv = transform.loop.forall_to_for %xi_ci_conv
+%xi_relu, %ci_relu = transform.loop.forall_to_for %xi_ci_relu
+%xi_comb, %ci_comb = transform.loop.forall_to_for %xi_ci_comb
+```
+
+We can then move our loop unrolling transformations later in the transformation
+sequence as desired. Compiling this new version to assembly produces exactly the
+same core computation around `vfmadd231ps` as Halide’s version, which only
+differs slightly in allocated registers. Unsurprisingly, this version runs
+roughly in 120ms on the same machine.
+
+
+## Multi-Dimensional Vectors to the Rescue
+
+While we managed to produce similar code to Halide in the previous section, we
+did so by rematching generated loops after bufferization, which partially defies
+the purpose of using handles to chain transformations in the Transform dialect.
+Luckily, this step is not really necessary. It only served as an exercise in
+producing the desired loop structure.
+
+Multidimensional structured operations on vectors are lowered to target-specific
+vectors by unrolling and splitting. For example, an elementwise arithmetic
+operation on `vector<5x64xf32>` is replaced with 5 operations on
+`vector<64xf32>` and additional vector value manipulations to recreate the
+required type at the MLIR level. Each of these operations is then split into 4
+operations on `vector<16xf32>` at the LLVM level where the information about
+the target vector width becomes available. Collectively, this has exactly the
+same effect as first materializing the 5x4 loop nest, and then fully unrolling
+these loops. Therefore, the last stage of tiling, re-matching and unrolling can
+be removed from the schedule.
+
+The resulting assembly has all `vbroadcast` grouped together before `vfmadd231`
+but otherwise has a similar structure. This grouping is due to each
+multi-dimensional vector operation being “unrolled” separately. When executed,
+it runs in ~110ms, a slight improvement of 8% over both the previous version and
+Halide, and reaches ~53.7 GFlop/s or 84% of peak single-core performance. The
+improvement is largely due to the intermediate representation being shorter and
+simpler in presence of large-vector operations, which allowed for more
+aggressive address computation and load placement optimization.
+
+The final transformation strategy is checked into the repository at
+[mlir/examples/transform/ChH/full.mlir](
+https://github.com/llvm/llvm-project/tree/main/mlir/test/Examples/transform/ChH/full.mlir).
diff --git a/mlir/docs/Tutorials/transform/_index.md b/mlir/docs/Tutorials/transform/_index.md
index 821177aa6cbda7e..3afb9c51da8b9fe 100644
--- a/mlir/docs/Tutorials/transform/_index.md
+++ b/mlir/docs/Tutorials/transform/_index.md
@@ -26,6 +26,7 @@ The tutorial is divided into the following chapters.
- [Chapter #1](Ch1.md): Combining Existing Transformations
- [Chapter #2](Ch2.md): Adding a Simple New Transformation Operation
- [Chapter #3](Ch3.md): More than Simple Transform Operations
+- [Chapter H](ChH.md): Reproducing Halide Schedule
The code corresponding to this tutorial is located under
`mlir/Examples/transform` and the corresponding tests in
diff --git a/mlir/test/Examples/transform/ChH/full.mlir b/mlir/test/Examples/transform/ChH/full.mlir
new file mode 100644
index 000000000000000..7e3a89fa63c4cb5
--- /dev/null
+++ b/mlir/test/Examples/transform/ChH/full.mlir
@@ -0,0 +1,393 @@
+// RUN: mlir-opt %s --test-transform-dialect-interpreter \
+// RUN: --test-transform-dialect-erase-schedule \
+// RUN: --math-uplift-to-fma \
+// RUN: --test-lower-to-llvm |\
+// RUN: FileCheck %s
+
+// Fixed-size tensor types to be used in convolution.
+// Named sizes are: N=5 OH=80 OW=100 F=C=128 KH=KW=3.
+// Input is NHWC.
+// Filter is CHWF.
+// Ouptut is NHWF.
+!tinput = tensor<5x82x102x128xf32>
+!tfilter = tensor<128x3x3x128xf32>
+!tbias = tensor<128xf32>
+!toutput = tensor<5x80x100x128xf32>
+
+// Function containing the convolution. Note that its arguments and results are
+// tensors annotated with attributes from the `bufferization` dialect. These
+// attributes hint the bufferization pass to assume buffers can be directly
+// used for these tensors without reshaping.
+func.func @conv(
+ %input: !tinput {bufferization.writable = false,
+ bufferization.access = "read",
+ bufferization.buffer_layout =
+ affine_map<(d0,d1,d2,d3)->(d0,d1,d2,d3)>},
+ %filter: !tfilter {bufferization.writable = false,
+ bufferization.access = "read",
+ bufferization.buffer_layout =
+ affine_map<(d0,d1,d2,d3)->(d0,d1,d2,d3)>},
+ %bias: !tbias {bufferization.writable = false,
+ bufferization.access = "read",
+ bufferization.buffer_layout = affine_map<(d0)->(d0)>},
+ %output: !toutput {bufferization.writable = true,
+ bufferization.buffer_layout =
+ affine_map<(d0,d1,d2,d3)->(d0,d1,d2,d3)>,
+ bufferization.access = "write"}) -> !toutput
+ // This requests a C-compatible interface to be emitted for the function
+ // when translating to LLVM IR.
+ attributes { llvm.emit_c_interface }
+{
+ // Bias. Using a named Linalg operation for brevity.
+ %bias_init = tensor.empty() : !toutput
+ %biased = linalg.broadcast ins(%bias : !tbias)
+ outs(%bias_init : !toutput) dimensions = [0, 1, 2]
+
+ // Convolution proper. While Linalg has named operations for 2D convolutions,
+ // the one in the Halide example has an uncommon order of filter dimensions
+ // and is not supported. It also takes the fitler as first argument. This
+ // code recreates it faithfully using the generic form.
+ %convolved = linalg.generic {
+ iterator_types = ["parallel", "parallel", "parallel", "parallel",
+ "reduction", "reduction", "reduction"],
+ indexing_maps = [
+ affine_map<(n, y, x, c, rz, ry, rx) -> (rx, rz, ry, c)>,
+ affine_map<(n, y, x, c, rz, ry, rx) -> (n, y+rz, x+ry, rx)>,
+ affine_map<(n, y, x, c, rz, ry, rx) -> (n, y, x, c)>
+ ]
+ } ins(%filter, %input: !tfilter, !tinput) outs(%biased : !toutput) {
+ ^bb0(%in: f32, %f: f32, %b: f32):
+ // Note the fastmath attributes that allow operations to be recombined into
+ // %0 = math.fma %in, %f, %b : f32
+ // later on and to reorder reductions.
+ %m1 = arith.mulf %in, %f {fastmath = #arith.fastmath<fast>} : f32
+ %0 = arith.addf %b, %m1 {fastmath = #arith.fastmath<fast>} : f32
+ linalg.yield %0 : f32
+ } -> !toutput
+
+ // ReLU is just a max(0, x).
+ %c0 = arith.constant 0.0 : f32
+ %relued = linalg.generic {
+ iterator_types = ["parallel", "parallel", "parallel", "parallel"],
+ indexing_maps = [
+ affine_map<(d0, d1, d2, d3) -> ()>,
+ affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>,
+ affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>
+ ]
+ } ins(%c0, %convolved : f32, !toutput)
+ outs(%output : !toutput) {
+ ^bb0(%cst: f32, %in: f32, %out: f32):
+ %0 = llvm.intr.maxnum(%cst, %in) : (f32, f32) -> f32
+ linalg.yield %0 : f32
+ } -> !toutput
+
+ return %relued : !toutput
+}
+
+// Module containing the transformation script to be applied. The attribute
+// is required to correctly verify the use of named (macro-like) sequences.
+module attributes { transform.with_named_sequence } {
+ // Apply transformations in a sequence to recreate the following Halide
+ // schedule:
+ //
+ // Var co, ci, xo, xi;
+ // relu.split(c, co, ci, vec * tile_w)
+ // .split(x, xo, xi, tile_h)
+ // .reorder(ci, xi, xo, y, n, co)
+ // .vectorize(ci, vec)
+ // .unroll(ci)
+ // .unroll(xi);
+ // conv.compute_at(relu, xo)
+ // .vectorize(c, vec)
+ // .unroll(c)
+ // .unroll(x)
+ // .unroll(y)
+ // .update()
+ // .reorder(c, x, y, r.x, r.y, r.z, n)
+ // .vectorize(c, vec)
+ // .unroll(c)
+ // .unroll(x)
+ // .unroll(y)
+ // .unroll(r.x, 2);
+ //
+ // where tile_w = 4, tile_h = 5, vec = 16. Note that unroll(y) and unroll(r.x)
+ // have no effect on the Halide IR as of 294f80c49bf3bb8582446613c25fcce03b82.
+ // Also note that the order of dimensions in Halide is inverted, e.g., co and
+ // n are the outermost loops in the respective reorder directives.
+ transform.sequence failures(propagate) {
+ // This argument will point to the top-level module.
+ ^bb0(%arg0: !transform.any_op):
+
+ // 1. Find the operations we are going to transform usnig their names. This
+ // is a simplistic approach that works when there are few operations in the
+ // IR to be transformed. More complex scenarios should rely on operations
+ // with `transform.match` prefix that are out of scope for this chapter.
+ %bias = transform.structured.match ops{["linalg.broadcast"]} in %arg0
+ : (!transform.any_op) -> !transform.any_op
+ %generics = transform.structured.match ops{["linalg.generic"]} in %arg0
+ : (!transform.any_op) -> !transform.any_op
+ %conv, %relu = transform.split_handle %generics
+ : (!transform.any_op) -> (!transform.any_op, !transform.any_op)
+
+ // 2. Initial tiling to start producing the loop structure. Note that the
+ // linalg.generic operation has the implicit loop order (n, y, x, c). Since
+ // the desired order of dimensions is (co, n, y, xo, xi, ci), we first tile
+ // only the c dimension to materialize the outermost co loop, and then tile
+ // the other dimensions since they are already in the expected order. Tiling
+ // by 1 produces the loop that iterates along the entire dimension. Tiling
+ // by 0 does not produce a loop. The size 64 is chosen as tiling by 4*16
+ // where 16 is the AVX512 vector length. Note that structured tiling doesn't
+ // remove the dimensions that became trivial (unit size) so the resulting
+ // sturucture is technically (co, no=n, yo=y, xo, [ni=1, yi=1, xi, ci])
+ // where brackets indicate implicit loops of the `linalg.generic` operation
+ // inside the loops produced by tiling.
+ //
+ // [n y x c]
+ %co, %relu2 = transform.structured.tile_to_forall_op %relu
+ tile_sizes [0, 0, 0, 64]
+ : (!transform.any_op) -> (!transform.any_op, !transform.any_op)
+ %n_y_xo, %relu3 = transform.structured.tile_to_forall_op %relu2
+ tile_sizes [1, 1, 5, 0]
+ : (!transform.any_op) -> (!transform.any_op, !transform.any_op)
+
+ // Compute_at is actually fusion into the given loop (given that we start
+ // with totally fissioned form, Halide starts with a fused form by reusing
+ // the loop iterators).
+ %conv2, %co2 = transform.structured.fuse_into_containing_op %conv into %co
+ : (!transform.any_op, !transform.any_op)
+ -> (!transform.any_op, !transform.any_op)
+ %conv3, %n_y_xo2 = transform.structured.fuse_into_containing_op %conv2
+ into %n_y_xo
+ : (!transform.any_op, !transform.any_op)
+ -> (!transform.any_op, !transform.any_op)
+
+ // Also fuse the bias that we represent as a separate operation and Halide
+ // represents as the "pure" (as opposed to "update") part of the conv
+ // expression. Note that fusion consumes both handles and produces new
+ // handles for chaining purposes.
+ %bias2, %co3 = transform.structured.fuse_into_containing_op %bias into %co2
+ : (!transform.any_op, !transform.any_op)
+ -> (!transform.any_op, !transform.any_op)
+ %bias3, %n_y_xo3 = transform.structured.fuse_into_containing_op %bias2
+ into %n_y_xo2
+ : (!transform.any_op, !transform.any_op)
+ -> (!transform.any_op, !transform.any_op)
+
+ // Clean up the result of fusion, which mechanically duplicates the producer
+ // operation in the consumer loop without removing the original operation.
+ // The original operation is now "dead": it has no uses and no side effects
+ // so it can be removed by dead-code elimination (DCE) that runs as part of
+ // pattern rewriting. The transform dialect allows to apply a combination
+ // of named pattern sets, exposed as operations, in one sweep to an
+ // isolated-from-above container payload operation. Note that we don't
+ // actually need any patterns for DCE to run, just trigger the rewriting.
+ //
+ // This step is optional. The transformation can continue without it and
+ // produce the same final IR, but makes it easier to manually examine the
+ // intermediate stages.
+ %f00 = transform.structured.match ops{["func.func"]} in %arg0
+ : (!transform.any_op) -> !transform.any_op
+ transform.apply_patterns to %f00 {
+ } : !transform.any_op
+
+ // The loop reordering requested for the convolution operation requires
+ // putting reduction loops (r.z, r.y. r.x) before the "inner" loops xi, ci.
+ // The "inner" loops are still implicit as part of the linalg.generic
+ // operation, and we need to materialize reduction loops around it by tiling
+ // with size 1. Since we are producing reduction loops, we indicate that we
+ // are tiling a reduction and request a sequential `scf.for` loops (parallel
+ // reductions are supported by `scf.forall`, but we don't need those here).
+ //
+ // This transform operation is more capable than merely producing
+ // (reduction) loops: the transformed code performs `tile_size` partial
+ // reductions of `N / tile_size` elements, potentially in parallel by
+ // changing the dimension kind of the structured operation inside the loop,
+ // and then performs a final reduction of these partial results by producing
+ // a new “combiner” structured operation after the loops. In our case,
+ // tile_size = 1 along all dimensions, so the reduction is entirely
+ // performed by the generated loops. The combiner structured operation is
+ // still produced and adds up the reduction result with the initial value.
+ %rz_ry_rx, %red_fill, %conv4, %combining
+ = transform.structured.tile_reduction_using_scf %conv3 by
+ // n y x c rz ry rx
+ tile_sizes=[0, 0, 0, 0, 1, 1, 1]
+ : (!transform.any_op)
+ -> (!transform.any_op, !transform.any_op, !transform.any_op,
+ !transform.any_op)
+
+ // At this point, the inner Linalg operations have implicit iteration spaces
+ // of 5x64 size, with some additional unit-size dimensions. Completely
+ // replicating Halide schedule would require materializing the loops with
+ // 5 and 4 iterations, respectively, unrolling those loops and marking the
+ // remaining 16-point iteration space for vectorization.
+ //
+ // This is unnecessary in MLIR that supports multi-dimensional vectors,
+ // which will be decomposed into target-specific sizes during the lowering.
+ // Therefore, this schedule stops here.
+
+ // Transform the named broadcast operation used for bias into the generic
+ // form before vectorization to prevent special cases from kicking in.
+ transform.structured.generalize %bias3
+ : (!transform.any_op) -> !transform.any_op
+
+ // Use the named macro to perform most of the lowering.
+ transform.include @lower failures(propagate) (%arg0)
+ : (!transform.any_op) -> ()
+ transform.yield
+ }
+
+ // Named sequence of transformations is a macro-like object that can be
+ // included from another place in the transform dialect, but doesn't allow for
+ // recursion. This can be reused in other scenarios.
+ transform.named_sequence @lower(
+ %arg0: !transform.any_op {transform.consumed}) {
+ %f00 = transform.structured.match ops{["func.func"]} in %arg0
+ : (!transform.any_op) -> !transform.any_op
+
+ // Simplify the code as tiling and fusion may have produced a lot of
+ // operations computing tensor subsets and loop ranges, some of which may be
+ // duplicated or excessively complex. Simplification involving
+ // canonicalization, common subexpression elimination, loop invariant code
+ // motion and various rewrite patterns can be applied directly from the
+ // transform dialect. Furthermore, an arbitrary combination of rewrite
+ // patterns can be applied in one sweep to a given scope, a functionality
+ // that cannot be achieved with conventional compiler passes that apply each
+ // group of patterns separately (at least without creating a new pass for
+ // each combination of pattern groups).
+ transform.apply_patterns to %f00 {
+ transform.apply_patterns.canonicalization
+ transform.apply_patterns.linalg.tiling_canonicalization
+ } : !transform.any_op
+ transform.apply_cse to %f00 : !transform.any_op
+ %all_loops = transform.structured.match interface{LoopLikeInterface}
+ in %arg0
+ : (!transform.any_op) -> !transform.any_op
+ transform.apply_licm to %all_loops : !transform.any_op
+
+ // Tiling-by-one as a way of materializing loops produced operations
+ // processing 4+D types where only a handful of dimension isn’t unit-sized,
+ // e.g., tensor<1x1x1x5x64xf32> where 5 and 64 are tile sizes. Remove such
+ // unit dimensions before vectorization, for clarity.
+ transform.apply_patterns to %f00 {
+ transform.apply_patterns.linalg.fold_unit_extent_dims_via_reshapes
+ } : !transform.any_op
+
+ // Vectorize the remaining non-unit dimensions in structured operations.
+ // This essentially rewrites operations on `tensor<5x64xf32>` into
+ // opreations on `vector<5x64xf32>`. Further lowering in MLIR and LLVM will
+ // decompose this into a sequence of operations on single-dimensional
+ // vectors of the platform-relevant size, e.g., `vector<16xf32>` for AVX512.
+ // High-level vector primitives, such as `vector.transpose` and
+ // `vector.broadcast` can be introduced at this stage. They will be later
+ // lowered to sequences of lower-level primitives such as `vector.shuffle`
+ // depending on the selected lowering strategy.
+ %fv = transform.structured.vectorize %f00
+ : (!transform.any_op) -> !transform.any_op
+
+ // Vectorization may have created new opportunities for cleanups. In
+ // particular, tensor subsetting operations can be composed with vector
+ // operations, and vector transfer (multi-dimensional load/store) operations
+ // can be recombined and hoisted out of loops.
+ transform.apply_patterns to %fv {
+ transform.apply_patterns.canonicalization
+ transform.apply_patterns.tensor.fold_tensor_subset_ops_into_vector_transfers
+ } : !transform.any_op
+ transform.apply_cse to %fv : !transform.any_op
+ transform.structured.hoist_redundant_vector_transfers %fv
+ : (!transform.any_op) -> !transform.any_op
+
+ // Apply bufferization that rewrites the remaining operations on tensors
+ // as operations on structured buffer (memref) types, including the function
+ // API. MLIR bufferization uses destination-passing style meaning that a
+ // buffer is shared between one of the operation's operands and its result.
+ //
+ // Since bufferization rewrites function signatures, it is applied as a
+ // module-wise transformation. Therefore, it invalidates all previously
+ // defined handles. Bufferization is usually a late step in the
+ // transformation process, so invalidation is not an issue. However, if
+ // other transformations, such as loop unrolling, are required after
+ // bufferization, new handles should be produced using the match operations.
+ %arg1 = transform.bufferization.one_shot_bufferize %arg0 {
+ bufferize_function_boundaries = true,
+ function_boundary_type_conversion = 1 : i32 }
+ : (!transform.any_op) -> !transform.any_op
+
+ // Apply general canonicalization and CSE to each function after
+ // bufferization as new simplification opportunities may have appeared.
+ %fb = transform.structured.match ops{["func.func"]} in %arg1
+ : (!transform.any_op) -> !transform.any_op
+ transform.apply_patterns to %fb {
+ transform.apply_patterns.canonicalization
+ } : !transform.any_op
+ transform.apply_cse to %fb : !transform.any_op
+
+ // Lower complex, multidimensional vector operations into simpler
+ // primitives. This particular selection of the pattern groups corresponds
+ // to vector dialect operations present in the payload IR at this stage.
+ // Many of these groups can be parameterized to use different strategies or
+ // lower-level primitives offering performance trade-offs. In this case, we
+ // are selecting the simplest strategies.
+ transform.apply_patterns to %fb {
+ transform.apply_patterns.vector.lower_contraction
+ lowering_strategy = parallelarith
+ transform.apply_patterns.vector.lower_transfer
+ max_transfer_rank = 1
+ transform.apply_patterns.vector.lower_transpose
+ lowering_strategy = eltwise
+ transform.apply_patterns.vector.lower_shape_cast
+ } : !transform.any_op
+
+ // These patterns apply in a separate sweep to avoid transfer-to-scf
+ // patterns overlap with lower-transfer patterns as they apply to the same
+ // kind of operations. These patterns may produce local allocations to act
+ // as temporary caches deep inside loops, which could lead to catastrophic
+ // performance. Such allocations are moved onto the stack and hoisted from
+ // all the surrounding loops.
+ transform.apply_patterns to %fb {
+ transform.apply_patterns.vector.transfer_to_scf
+ transform.apply_patterns.memref.alloc_to_alloca
+ } : !transform.any_op
+ transform.bufferization.buffer_loop_hoisting %fb : !transform.any_op
+
+ // A final round of cleanups additionally includes patterns to simplify
+ // buffer aliasing operations that may have been introduced during
+ // bufferization and could result in excessively complex address
+ // computation.
+ transform.apply_patterns to %fb {
+ transform.apply_patterns.memref.fold_memref_alias_ops
+ transform.apply_patterns.canonicalization
+ } : !transform.any_op
+ transform.apply_cse to %fb : !transform.any_op
+
+ transform.yield
+ }
+}
+
+// The core computation, at the LLVM dialect level, must correspond to five
+// immediately adjacent fma on vector<64xf32>.
+
+// CHECK: %[[R0:.+]] = llvm.mlir.undef : !llvm.array<5 x vector<64xf32>>
+// CHECK-NEXT: %[[LINE0:.+]] = llvm.extractvalue %[[V:.+]][0] : !llvm.array<5 x vector<64xf32>>
+// CHECK-NEXT: %[[FMA0:.+]] = llvm.intr.fma(%{{.*}}, %{{.*}}, %[[LINE0]])
+// CHECK-SAME: -> vector<64xf32>
+// CHECK-NEXT: %[[R1:.+]] = llvm.insertvalue %[[FMA0]], %[[R0]][0]
+
+// CHECK-NEXT: %[[LINE1:.+]] = llvm.extractvalue %[[V:.+]][1] : !llvm.array<5 x vector<64xf32>>
+// CHECK-NEXT: %[[FMA1:.+]] = llvm.intr.fma(%{{.*}}, %{{.*}}, %[[LINE1]])
+// CHECK-SAME: -> vector<64xf32>
+// CHECK-NEXT: %[[R2:.+]] = llvm.insertvalue %[[FMA1]], %[[R1]][1]
+
+// CHECK-NEXT: %[[LINE2:.+]] = llvm.extractvalue %[[V:.+]][2] : !llvm.array<5 x vector<64xf32>>
+// CHECK-NEXT: %[[FMA2:.+]] = llvm.intr.fma(%{{.*}}, %{{.*}}, %[[LINE2]])
+// CHECK-SAME: -> vector<64xf32>
+// CHECK-NEXT: %[[R3:.+]] = llvm.insertvalue %[[FMA2]], %[[R2]][2]
+
+// CHECK-NEXT: %[[LINE3:.+]] = llvm.extractvalue %[[V:.+]][3] : !llvm.array<5 x vector<64xf32>>
+// CHECK-NEXT: %[[FMA3:.+]] = llvm.intr.fma(%{{.*}}, %{{.*}}, %[[LINE3]])
+// CHECK-SAME: -> vector<64xf32>
+// CHECK-NEXT: %[[R4:.+]] = llvm.insertvalue %[[FMA3]], %[[R3]][3]
+
+// CHECK-NEXT: %[[LINE4:.+]] = llvm.extractvalue %[[V:.+]][4] : !llvm.array<5 x vector<64xf32>>
+// CHECK-NEXT: %[[FMA4:.+]] = llvm.intr.fma(%{{.*}}, %{{.*}}, %[[LINE4]])
+// CHECK-SAME: -> vector<64xf32>
+// CHECK-NEXT: %[[R5:.+]] = llvm.insertvalue %[[FMA4]], %[[R4]][4]
More information about the Mlir-commits
mailing list