Bump open source LLVM revision to 378b1e60809df7cc72436093f160b7ed228dad5a
Adjust code for refactorings that happened in MLIR.
PiperOrigin-RevId: 301272688
diff --git a/iree/compiler/Translation/CodegenPasses/XLAToLinalgOnBuffers.cpp b/iree/compiler/Translation/CodegenPasses/XLAToLinalgOnBuffers.cpp
index 001101f..f4b632a 100644
--- a/iree/compiler/Translation/CodegenPasses/XLAToLinalgOnBuffers.cpp
+++ b/iree/compiler/Translation/CodegenPasses/XLAToLinalgOnBuffers.cpp
@@ -175,7 +175,8 @@
auto dilationArg = ArrayAttr::get(dilation, op.getContext());
return rewriter.create<linalg::ConvOp>(op.getLoc(), args[1], args[0],
- results[0], stridesArg, dilationArg);
+ results[0], stridesArg, dilationArg,
+ op.paddingAttr());
}
// ----------------------------------------------------------------------------
diff --git a/iree/compiler/Translation/CodegenPasses/test/conv.mlir b/iree/compiler/Translation/CodegenPasses/test/conv.mlir
index 36e0c3c..b4d4516 100644
--- a/iree/compiler/Translation/CodegenPasses/test/conv.mlir
+++ b/iree/compiler/Translation/CodegenPasses/test/conv.mlir
@@ -4,7 +4,11 @@
func @conv(%arg0: memref<3x5x5x3xf32>, %arg1: memref<2x2x3x4xf32>, %arg2: memref<3x5x5x4xf32>) {
%0 = iree.load_input(%arg0 : memref<3x5x5x3xf32>) : tensor<3x5x5x3xf32>
%1 = iree.load_input(%arg1 : memref<2x2x3x4xf32>) : tensor<2x2x3x4xf32>
- // CHECK: linalg.conv(%arg0, %arg1, %arg2) {dilations = [1, 2], strides = [2, 1]}
+ // CHECK: linalg.conv(%arg0, %arg1, %arg2) {
+ // CHECK-SAME: dilations = [1, 2]
+ // CHECK-SAME: padding = dense<[
+ // CHECK-SAME: [0, 1], [0, 1]]> : tensor<2x2xi64>
+ // CHECK-SAME: strides = [2, 1]}
%2 = "xla_hlo.conv"(%1, %0) {
batch_group_count = 1 : i64,
dimension_numbers = {
