[Mlir-commits] [mlir] [mlir][x86vector] Shuffle BF16 vector.contract output for Flat layout. (PR #174590)
Arun Thangamani
llvmlistbot at llvm.org
Wed Feb 11 07:43:52 PST 2026
================
@@ -27,10 +28,83 @@ using namespace mlir::x86vector;
namespace {
+// Returns true if the A or B matrix vector is packed (shuffled) to
+// VNNI layout, already.
+static bool isNonUnitDimOperandShuffled(Value nonUnitDimOperand) {
+ if (Operation *defOp = nonUnitDimOperand.getDefiningOp()) {
+ if (isa<vector::ShuffleOp>(defOp)) {
----------------
arun-thmn wrote:
Thanks, Removed.
https://github.com/llvm/llvm-project/pull/174590
More information about the Mlir-commits
mailing list