[mlir][linalg] Fix a bug in lower_pack when there are no padding values.

[hanchung]

delete unused IR from tests

[chelini]

Looks good to me, thanks.

[hanchung]

rebase

mlir/lib/Dialect/Linalg/TransformOps/LinalgTransformOps.cpp

static FailureOr<LowerPackResult> lowerPack(RewriterBase &rewriter,
   applyPermutationToVector(stripMinedShape, lastDimsToInsertPositionsPerm);
 
   // 4. Pad the source of packOp to a shape we can expand into stripMinedShape.
   RankedTensorType collapsed = tensor::CollapseShapeOp::inferCollapsedType(
       RankedTensorType::Builder(packedTensorType).setShape(stripMinedShape),
       packingMetadata.reassociations);
   Value paddingValue = packOp.getPaddingValue();
   if (!paddingValue) {
-    rewriter.create<arith::ConstantOp>(
+    paddingValue = rewriter.create<arith::ConstantOp>(
         loc, rewriter.getZeroAttr(getElementTypeOrSelf(collapsed)));
   }
   auto padOp =
       tensor::createPadHighOp(collapsed, packOp.getSource(), paddingValue,
                               /*nofold=*/false, loc, rewriter);
 
   LLVM_DEBUG(
       DBGSNL(); DBGSNL(); llvm::interleaveComma(packingMetadata.insertPositions,

mlir/test/Dialect/Linalg/transform-lower-pack.mlir

^bb1(%module_op: !pdl.operation):
   %pack = transform.structured.match ops{["tensor.pack"]} in %module_op 
     : (!pdl.operation) -> !transform.op<"tensor.pack">
   transform.structured.lower_pack %pack : (!transform.op<"tensor.pack">) 
     -> (!transform.op<"tensor.pad">, !transform.op<"tensor.expand_shape">, !transform.op<"linalg.transpose">)
 }
 
 // -----
 
+  // CHECK-LABEL: func.func @pack(
+func.func @pack(%arg0: tensor<128x8xf32>, %arg1: tensor<8x8x16x1xf32>) -> tensor<8x8x16x1xf32> {
+  %cst_0 = arith.constant 0.0 : f32
+
+  // tensor.pack is lowered to tensor.pad + tensor.expand_shape + linalg.transpose
+  //      CHECK: %[[C0:.*]] = arith.constant 0 : index
+  //      CHECK: tensor.pad {{.*}} low[%[[C0]], %[[C0]]]
+  //      CHECK:   : tensor<128x8xf32> to tensor<128x8xf32>
+  //      CHECK: tensor.expand_shape %{{.*}} [{{.*}}[0, 1], [2, 3]]
+  // CHECK-SAME:   : tensor<128x8xf32> into tensor<8x16x8x1xf32>
+  //      CHECK: linalg.transpose
+  // CHECK-SAME:   ins(%{{.*}} : tensor<8x16x8x1xf32>)
+  // CHECK-SAME:   outs(%{{.*}} : tensor<8x8x16x1xf32>)
+  // CHECK-SAME:   permutation = [0, 2, 1, 3]
+
+  %pack = tensor.pack %arg0 inner_dims_pos = [0, 1] inner_tiles = [16, 1] into %arg1
+    : tensor<128x8xf32> -> tensor<8x8x16x1xf32>
+
+  return %pack : tensor<8x8x16x1xf32>
+}
+
+transform.sequence failures(propagate) {
+^bb1(%module_op: !pdl.operation):
+  %pack = transform.structured.match ops{["tensor.pack"]} in %module_op
+    : (!pdl.operation) -> !transform.op<"tensor.pack">
+  transform.structured.lower_pack %pack : (!transform.op<"tensor.pack">)
+    -> (!transform.op<"tensor.pad">, !transform.op<"tensor.expand_shape">, !transform.op<"linalg.transpose">)
+}
+
+// -----
+
 // CHECK-LABEL: func.func @unpack(
 func.func @unpack(%arg0: tensor<17x2x16x16x32x8xf32>, %arg1: tensor<129x47x16x16xf32>) -> tensor<129x47x16x16xf32> {
   %cst_0 = arith.constant 0.0 : f32
 
   //      CHECK: tensor.empty() : tensor<17x8x2x32x16x16xf32>
   //      CHECK: linalg.transpose
   // CHECK-SAME:    ins(%{{.*}} : tensor<17x2x16x16x32x8xf32>)
   // CHECK-SAME:   outs(%{{.*}} : tensor<17x8x2x32x16x16xf32>)