[Mlir-commits] [mlir] [mlir][TOSA] Fix linalg lowering of depthwise conv2d (PR #130282)

[Luke Hutton]

================
@@ -812,6 +813,30 @@ func.func @depthwise_conv2d_dyn_w_h(%arg0: tensor<2x?x?x3xf32>, %arg1: tensor<3x
 
 // -----
 
+// CHECK: #[[$MAP0:.*]] = affine_map<(d0, d1, d2, d3) -> (d3)>
+// CHECK: #[[$MAP1:.*]] = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>
+
+// CHECK-LABEL: @depthwise_int_conv_zero_zp
+func.func @depthwise_int_conv_zero_zp(%arg0 : tensor<1x7x5x3xi8>, %arg1 : tensor<3x1x3x11xi8>, %arg2 : tensor<33xi32>) -> () {
+  // CHECK: [[INIT:%.+]] = tensor.empty()
+  // CHECK: [[CST0:%.+]] = arith.constant 0
+  // CHECK: [[FILL:%.+]] = linalg.fill ins([[CST0]]{{.*}}outs([[INIT]]
+  // CHECK: [[OUT:%.+]] = tensor.empty()
+  // CHECK: [[DEPTH:%.+]] = linalg.depthwise_conv_2d_nhwc_hwcm {dilations = dense<1> : tensor<2xi64>, strides = dense<1> : tensor<2xi64>} ins(%arg0, %arg1 : tensor<1x7x5x3xi8>, tensor<3x1x3x11xi8>) outs([[FILL]] : tensor<1x5x5x3x11xi32>)
+  // CHECK: [[COLLAPSED:%.+]] = tensor.collapse_shape [[DEPTH]] {{\[}}[0], [1], [2], [3, 4]]
+  // CHECK: [[BIAS:%.+]] = linalg.generic {indexing_maps = [#[[$MAP0]], #[[$MAP1]], #[[$MAP1]]], iterator_types = ["parallel", "parallel", "parallel", "parallel"]} ins(%arg2, [[COLLAPSED]] : tensor<33xi32>, tensor<1x5x5x33xi32>) outs([[OUT]] : tensor<1x5x5x33xi32>) {
+  // CHECK: ^bb0(%[[ARG3:[0-9a-zA-Z_]+]]: i32, %[[ARG4:[0-9a-zA-Z_]+]]: i32, %[[ARG5:[0-9a-zA-Z_]+]]: i32):
+  // CHECK:   [[ADD:%.+]] = arith.addi %[[ARG3]], %[[ARG4]] : i32
+  // CHECK:   linalg.yield [[ADD]] : i32
+  // CHECK: } -> tensor<1x5x5x33xi32>
+  %input_zp = "tosa.const"() <{value = dense<0> : tensor<1xi8>}> : () -> tensor<1xi8>
----------------
lhutton1 wrote:

I suspect `value` needs updating to `values` after https://github.com/llvm/llvm-project/pull/129943

https://github.com/llvm/llvm-project/pull/130282