[llvm-branch-commits] [mlir] [MLIR][AArch64] Add integration test for lowering of `vector.contract` to Neon FEAT_I8MM (PR #144699)

[Momchil Velikov via llvm-branch-commits]

================
@@ -0,0 +1,336 @@
+// REQUIRES: arm-emulator
+
+// DEFINE: %{compile} = mlir-opt %s \
+// DEFINE:   --convert-vector-to-scf --convert-scf-to-cf  --convert-vector-to-llvm='enable-arm-neon enable-arm-i8mm' \
+// DEFINE:   --expand-strided-metadata --convert-to-llvm --finalize-memref-to-llvm  \
+// DEFINE:   --lower-affine --convert-arith-to-llvm --reconcile-unrealized-casts \
+// DEFINE: -o %t
+
+// DEFINE: %{entry_point} = main
+
+// DEFINE: %{run} = %mcr_aarch64_cmd %t -e %{entry_point} -entry-point-result=void  --march=aarch64 --mattr="+neon,+i8mm" \
+// DEFINE:    -shared-libs=%mlir_runner_utils,%mlir_c_runner_utils,%native_mlir_arm_runner_utils
+
+// RUN: rm -f %t && %{compile} && FileCheck %s --input-file=%t -check-prefix CHECK-IR && %{run} | FileCheck %s
+
+#packed_maps = [
+  affine_map<(m, n, k) -> (m, k)>,
+  affine_map<(m, n, k) -> (n, k)>,
+  affine_map<(m, n, k) -> (m, n)>
+]
+
+//
+// Test the lowering of `vector.contract` using the `LowerContractionToNeonI8MMPattern`
+//
+// The operation that the `vector.contract` in this test performs is matrix
+// multiplication with accumulate
+//     OUT = ACC + LHS * RHS
+// of two 8-bit integer matrices LHS and RHS, and a 32-bit integer matrix ACC
+// into a 32-bit integer matrix OUT. The LHS and RHS can be sign- or zero- extended,
+// this test covers all the possible variants.
+//
+// Tested are calculations as well as that the relevant `ArmNeon` dialect
+// operations ('arm_neon.smmla`, arm_neon.ummla`, etc) are emitted.
+//
+// That pattern above handles (therefore this test prepares) input/output vectors with
+// specific shapes:
+//   * LHS:      vector<MxKxi8>
+//   * RHS:      vector<NxKxi8>
+//   * ACC, OUT: vector<MxNxi32>
+// where the M and N are even and K is divisible by 8.
+// Note that the RHS is transposed.
+// This data layout makes it efficient to load data into SIMD
+// registers in the layout expected by FEAT_I8MM instructions.
+// Such a `vector.contract` is representative of the code we aim to generate
+// by vectorisation of `linalg.mmt4d`.
+//
+// In this specific test we use M == 4, N == 4, and K == 8.
----------------
momchil-velikov wrote:

Ah, yes it is. Comment fixed.

https://github.com/llvm/llvm-project/pull/144699