Integrate llvm-project @f1595ecfdce5387e41826fd72ff930a1a39ae398 (#18897

) - bump llvm-project to llvm/llvm-project@f1595ec - revert iree-org/llvm-project@1004865 due to compiler failures in VectorDistribute. Tracked in #18955. - bump stablehlo to openxla/stablehlo@c32f7c2 - bump torch-mlir to llvm/torch-mlir@8b0bf2e - revert llvm/torch-mlir@55ff110 due to new onnx failures. Tracked in #18961 --------- Signed-off-by: Max Dawkins <max.dawkins@gmail.com>
iree-org · Oct 31, 2024 · df2b8a4 · df2b8a4
1 parent b2b2d00
commit df2b8a4
Show file tree

Hide file tree

Showing 13 changed files with 53 additions and 61 deletions.
diff --git a/compiler/plugins/input/StableHLO/Conversion/StableHLOToIREEInputDialects.cpp b/compiler/plugins/input/StableHLO/Conversion/StableHLOToIREEInputDialects.cpp
@@ -424,11 +424,11 @@ struct GenericTypeConvert final : ConversionPattern {
   }
 };
 
-std::optional<Value> scalarToTensor(OpBuilder &builder, Type /*type*/,
-                                    ValueRange inputs, Location loc) {
+Value scalarToTensor(OpBuilder &builder, Type /*type*/, ValueRange inputs,
+                     Location loc) {
   assert(inputs.size() == 1);
   if (isa<ShapedType>(inputs.front().getType())) {
-    return std::nullopt;
+    return Value();
   }
   return builder
       .create<tensor::FromElementsOp>(

diff --git a/compiler/plugins/input/StableHLO/Conversion/StableHLOToLinalgExt.cpp b/compiler/plugins/input/StableHLO/Conversion/StableHLOToLinalgExt.cpp
@@ -65,13 +65,13 @@ Type convertShapedToSignless(ShapedType shapedType) {
   return shapedType;
 }
 
-std::optional<Value> materializeCast(OpBuilder &builder, Type toType,
-                                     ValueRange inputs, Location loc) {
+Value materializeCast(OpBuilder &builder, Type toType, ValueRange inputs,
+                      Location loc) {
   assert(inputs.size() == 1 && "too many inputs to type conversion");
   Value fromValue = inputs[0];
   auto fromType = dyn_cast<RankedTensorType>(fromValue.getType());
   if (!fromType)
-    return std::nullopt;
+    return Value();
 
   if (auto intFromType = dyn_cast<IntegerType>(fromType.getElementType())) {
     Type castType = getElementTypeOrSelf(toType);

diff --git a/compiler/plugins/input/StableHLO/Conversion/TypeConversion.cpp b/compiler/plugins/input/StableHLO/Conversion/TypeConversion.cpp
@@ -31,36 +31,35 @@ Type convertShapedType(ShapedType shapedType) {
   return shapedType;
 }
 
-std::optional<Value> materializeCastFromIllegal(OpBuilder &builder, Type type,
-                                                ValueRange inputs,
-                                                Location loc) {
+Value materializeCastFromIllegal(OpBuilder &builder, Type type,
+                                 ValueRange inputs, Location loc) {
   Type fromType = getElementTypeOrSelf(inputs[0].getType());
   Type toType = getElementTypeOrSelf(type);
   if ((!fromType.isSignedInteger() && !fromType.isUnsignedInteger()) ||
       !toType.isSignlessInteger())
-    return std::nullopt;
+    return Value();
   // Use unrealized conversion casts to do signful->signless conversions.
   return builder.create<UnrealizedConversionCastOp>(loc, type, inputs[0])
       ->getResult(0);
 }
 
-std::optional<Value> materializeCastToIllegal(OpBuilder &builder, Type type,
-                                              ValueRange inputs, Location loc) {
+Value materializeCastToIllegal(OpBuilder &builder, Type type, ValueRange inputs,
+                               Location loc) {
   Type fromType = getElementTypeOrSelf(inputs[0].getType());
   Type toType = getElementTypeOrSelf(type);
   if (!fromType.isSignlessInteger() ||
       (!toType.isSignedInteger() && !toType.isUnsignedInteger()))
-    return std::nullopt;
+    return Value();
   // Use unrealized conversion casts to do signless->signful conversions.
   return builder.create<UnrealizedConversionCastOp>(loc, type, inputs[0])
       ->getResult(0);
 }
 
-std::optional<Value> scalarToTensor(OpBuilder &builder, Type type,
-                                    ValueRange inputs, Location loc) {
+Value scalarToTensor(OpBuilder &builder, Type type, ValueRange inputs,
+                     Location loc) {
   assert(inputs.size() == 1);
   if (llvm::isa<ShapedType>(inputs.front().getType())) {
-    return std::nullopt;
+    return Value();
   }
   auto tensor =
       builder

diff --git a/compiler/src/iree/compiler/Codegen/Common/ConvertBf16ToUInt16Buffers.cpp b/compiler/src/iree/compiler/Codegen/Common/ConvertBf16ToUInt16Buffers.cpp
@@ -229,9 +229,8 @@ struct ConvertMemRefStore final : OpConversionPattern<memref::StoreOp> {
 // Helper functions
 //===----------------------------------------------------------------------===//
 
-std::optional<Value> materializeArithBitcast(OpBuilder &builder, Type resultTy,
-                                             mlir::ValueRange inputs,
-                                             mlir::Location loc) {
+Value materializeArithBitcast(OpBuilder &builder, Type resultTy,
+                              mlir::ValueRange inputs, mlir::Location loc) {
   return builder.create<arith::BitcastOp>(loc, resultTy, inputs);
 }
 

diff --git a/compiler/src/iree/compiler/Codegen/Common/test/tile_and_distribute_to_workgroups.mlir b/compiler/src/iree/compiler/Codegen/Common/test/tile_and_distribute_to_workgroups.mlir
@@ -2088,9 +2088,7 @@ hal.executable private @dynamic_unpack {
 
 // -----
 
-#pipeline_layout = #hal.pipeline.layout<constants = 4, bindings = [
-  #hal.pipeline.binding<storage_buffer>,
-  #hal.pipeline.binding<storage_buffer>,
+#pipeline_layout = #hal.pipeline.layout<constants = 6, bindings = [
   #hal.pipeline.binding<storage_buffer>,
   #hal.pipeline.binding<storage_buffer>
 ]>
@@ -2111,10 +2109,14 @@ hal.executable private @dynamic_unpack_dynamic_tile {
         %cl_1 = hal.interface.constant.load layout(#pipeline_layout) ordinal(1) : i32
         %cl_2 = hal.interface.constant.load layout(#pipeline_layout) ordinal(2) : i32
         %cl_3 = hal.interface.constant.load layout(#pipeline_layout) ordinal(3) : i32
+        %cl_4 = hal.interface.constant.load layout(#pipeline_layout) ordinal(4) : i32
+        %cl_5 = hal.interface.constant.load layout(#pipeline_layout) ordinal(5) : i32
         %0 = arith.index_castui %cl_0 : i32 to index
         %1 = arith.index_castui %cl_1 : i32 to index
         %2 = arith.index_castui %cl_2 : i32 to index
         %3 = arith.index_castui %cl_3 : i32 to index
+        %tile0 = arith.index_castui %cl_3 : i32 to index
+        %tile1 = arith.index_castui %cl_3 : i32 to index
         %4 = flow.dispatch.workload.ordinal %0, 0 : index
         %5 = flow.dispatch.workload.ordinal %1, 1 : index
         %6 = flow.dispatch.workload.ordinal %2, 2 : index
@@ -2123,7 +2125,7 @@ hal.executable private @dynamic_unpack_dynamic_tile {
         %9 = hal.interface.binding.subspan layout(#pipeline_layout) binding(1) alignment(64) offset(%c131072) : !flow.dispatch.tensor<writeonly:tensor<?x?xi32>>{%6, %7}
         %10 = flow.dispatch.tensor.load %8, offsets = [0, 0, 0, 0], sizes = [%4, %5, %c32, %c16], strides = [1, 1, 1, 1] : !flow.dispatch.tensor<readonly:tensor<?x?x?x?xi32>>{%4, %5, %c32, %c16} -> tensor<?x?x?x?xi32>
         %11 = tensor.empty(%6, %7) : tensor<?x?xi32>
-        %12 = tensor.unpack %10 inner_dims_pos = [0, 1] inner_tiles = [%c32, %c16] into %11
+        %12 = tensor.unpack %10 inner_dims_pos = [0, 1] inner_tiles = [%tile0, %tile1] into %11
           {lowering_config = #iree_codegen.lowering_config<tile_sizes = [[64, 64]]>}
           : tensor<?x?x?x?xi32> -> tensor<?x?xi32>
         flow.dispatch.tensor.store %12, %9, offsets = [0, 0], sizes = [%6, %7], strides = [1, 1] : tensor<?x?xi32> -> !flow.dispatch.tensor<writeonly:tensor<?x?xi32>>{%6, %7}

diff --git a/compiler/src/iree/compiler/Dialect/Stream/Transforms/ConvertToStream.cpp b/compiler/src/iree/compiler/Dialect/Stream/Transforms/ConvertToStream.cpp
@@ -261,25 +261,24 @@ struct ConvertToStreamPass final
           resultTypes.push_back(indexType);
           return success();
         });
-    typeConverter.addArgumentMaterialization(
-        [](OpBuilder &builder, TensorType resultType, ValueRange inputs,
-           Location loc) -> std::optional<Value> {
-          assert(inputs.size() >= 2);
-          auto resourceValue = inputs[0];
-          auto resourceSize = inputs[1];
-          assert(inputs.size() == 2 &&
-                 "expecting 2 operands (resource + size)");
-          Value cast = builder
-                           .create<IREE::Stream::AsyncTransferOp>(
-                               loc, resourceValue.getType(), resourceValue,
-                               resourceSize, resourceSize,
-                               /*source_affinity=*/nullptr,
-                               /*result_affinity=*/nullptr)
-                           .getResult();
-          return builder
-              .create<UnrealizedConversionCastOp>(loc, resultType, cast)
-              .getResult(0);
-        });
+    typeConverter.addArgumentMaterialization([](OpBuilder &builder,
+                                                TensorType resultType,
+                                                ValueRange inputs,
+                                                Location loc) -> Value {
+      assert(inputs.size() >= 2);
+      auto resourceValue = inputs[0];
+      auto resourceSize = inputs[1];
+      assert(inputs.size() == 2 && "expecting 2 operands (resource + size)");
+      Value cast = builder
+                       .create<IREE::Stream::AsyncTransferOp>(
+                           loc, resourceValue.getType(), resourceValue,
+                           resourceSize, resourceSize,
+                           /*source_affinity=*/nullptr,
+                           /*result_affinity=*/nullptr)
+                       .getResult();
+      return builder.create<UnrealizedConversionCastOp>(loc, resultType, cast)
+          .getResult(0);
+    });
 
     populateUtilConversionPatterns(context, conversionTarget, typeConverter,
                                    patterns);

diff --git a/compiler/src/iree/compiler/Dialect/Util/Transforms/TestConversion.cpp b/compiler/src/iree/compiler/Dialect/Util/Transforms/TestConversion.cpp
@@ -20,10 +20,8 @@ namespace mlir::iree_compiler::IREE::Util {
 
 namespace {
 
-static std::optional<Value> buildUnrealizedConversionCastOp(OpBuilder &builder,
-                                                            Type toType,
-                                                            ValueRange inputs,
-                                                            Location loc) {
+static Value buildUnrealizedConversionCastOp(OpBuilder &builder, Type toType,
+                                             ValueRange inputs, Location loc) {
   return builder.create<UnrealizedConversionCastOp>(loc, toType, inputs)
       .getResult(0);
 }

diff --git a/tests/external/iree-test-suites/onnx_ops/onnx_ops_cpu_llvm_sync.json b/tests/external/iree-test-suites/onnx_ops/onnx_ops_cpu_llvm_sync.json
@@ -149,7 +149,6 @@
     "onnx/node/generated/test_lppool_2d_same_lower",
     "onnx/node/generated/test_lppool_2d_same_upper",
     "onnx/node/generated/test_maxpool_2d_ceil_output_size_reduce_by_one",
-    "onnx/node/generated/test_maxpool_2d_precomputed_same_upper",
     "onnx/node/generated/test_maxpool_2d_same_lower",
     "onnx/node/generated/test_maxpool_2d_same_upper",
     "onnx/node/generated/test_maxpool_with_argmax_2d_precomputed_strides",
@@ -380,6 +379,7 @@
     "onnx/node/generated/test_sce_none_weights_expanded",
     "onnx/node/generated/test_sce_none_weights_log_prob",
     "onnx/node/generated/test_sce_none_weights_log_prob_expanded",
+    "onnx/node/generated/test_shape_clip_start",
     "onnx/node/generated/test_shape_end_negative_1",
     "onnx/node/generated/test_slice",
     "onnx/node/generated/test_slice_default_steps",
@@ -438,9 +438,8 @@
     "onnx/node/generated/test_reduce_min_empty_set",
     "onnx/node/generated/test_reduce_sum_empty_set_non_reduced_axis_zero",
     "onnx/node/generated/test_resize_downsample_scales_linear_align_corners",
-    "onnx/node/generated/test_scan_sum",
     "onnx/node/generated/test_scan9_sum",
-    "onnx/node/generated/test_shape_clip_start",
+    "onnx/node/generated/test_scan_sum",
     "onnx/node/generated/test_shape_end_1",
     "onnx/node/generated/test_shape_start_1",
     "onnx/node/generated/test_shape_start_1_end_2",

diff --git a/tests/external/iree-test-suites/onnx_ops/onnx_ops_gpu_rocm_rdna3.json b/tests/external/iree-test-suites/onnx_ops/onnx_ops_gpu_rocm_rdna3.json
@@ -153,7 +153,6 @@
     "onnx/node/generated/test_lppool_2d_same_lower",
     "onnx/node/generated/test_lppool_2d_same_upper",
     "onnx/node/generated/test_maxpool_2d_ceil_output_size_reduce_by_one",
-    "onnx/node/generated/test_maxpool_2d_precomputed_same_upper",
     "onnx/node/generated/test_maxpool_2d_same_lower",
     "onnx/node/generated/test_maxpool_2d_same_upper",
     "onnx/node/generated/test_maxpool_with_argmax_2d_precomputed_strides",
@@ -381,6 +380,7 @@
     "onnx/node/generated/test_sce_none_weights_expanded",
     "onnx/node/generated/test_sce_none_weights_log_prob",
     "onnx/node/generated/test_sce_none_weights_log_prob_expanded",
+    "onnx/node/generated/test_shape_clip_start",
     "onnx/node/generated/test_shape_end_negative_1",
     "onnx/node/generated/test_slice",
     "onnx/node/generated/test_slice_default_steps",
@@ -488,11 +488,10 @@
     "onnx/node/generated/test_reduce_sum_square_default_axes_keepdims_random",
     "onnx/node/generated/test_reduce_sum_square_default_axes_keepdims_random_expanded",
     "onnx/node/generated/test_resize_downsample_scales_linear_align_corners",
-    "onnx/node/generated/test_scan_sum",
     "onnx/node/generated/test_scan9_sum",
+    "onnx/node/generated/test_scan_sum",
     "onnx/node/generated/test_shape",
     "onnx/node/generated/test_shape_clip_end",
-    "onnx/node/generated/test_shape_clip_start",
     "onnx/node/generated/test_shape_end_1",
     "onnx/node/generated/test_shape_example",
     "onnx/node/generated/test_shape_start_1",

diff --git a/tests/external/iree-test-suites/onnx_ops/onnx_ops_gpu_vulkan.json b/tests/external/iree-test-suites/onnx_ops/onnx_ops_gpu_vulkan.json
@@ -195,9 +195,6 @@
     "onnx/node/generated/test_max_uint16",
     "onnx/node/generated/test_max_uint8",
     "onnx/node/generated/test_maxpool_2d_ceil_output_size_reduce_by_one",
-    "onnx/node/generated/test_maxpool_2d_precomputed_same_upper",
-    "onnx/node/generated/test_maxpool_2d_same_lower",
-    "onnx/node/generated/test_maxpool_2d_same_upper",
     "onnx/node/generated/test_maxpool_with_argmax_2d_precomputed_strides",
     "onnx/node/generated/test_maxunpool_export_with_output_shape",
     "onnx/node/generated/test_maxunpool_export_without_output_shape",
@@ -447,6 +444,7 @@
     "onnx/node/generated/test_sce_none_weights_expanded",
     "onnx/node/generated/test_sce_none_weights_log_prob",
     "onnx/node/generated/test_sce_none_weights_log_prob_expanded",
+    "onnx/node/generated/test_shape_clip_start",
     "onnx/node/generated/test_shape_end_negative_1",
     "onnx/node/generated/test_slice",
     "onnx/node/generated/test_slice_default_steps",
@@ -533,8 +531,8 @@
     "onnx/node/generated/test_dynamicquantizelinear_expanded",
     "onnx/node/generated/test_einsum_batch_diagonal",
     "onnx/node/generated/test_einsum_batch_matmul",
-    "onnx/node/generated/test_einsum_transpose",
     "onnx/node/generated/test_einsum_sum",
+    "onnx/node/generated/test_einsum_transpose",
     "onnx/node/generated/test_eyelike_with_dtype",
     "onnx/node/generated/test_isinf_float16",
     "onnx/node/generated/test_isnan_float16",
@@ -595,9 +593,8 @@
     "onnx/node/generated/test_reduce_sum_square_default_axes_keepdims_example_expanded",
     "onnx/node/generated/test_reduce_sum_square_default_axes_keepdims_random",
     "onnx/node/generated/test_reduce_sum_square_default_axes_keepdims_random_expanded",
-    "onnx/node/generated/test_scan_sum",
     "onnx/node/generated/test_scan9_sum",
-    "onnx/node/generated/test_shape_clip_start",
+    "onnx/node/generated/test_scan_sum",
     "onnx/node/generated/test_shape_end_1",
     "onnx/node/generated/test_shape_start_1",
     "onnx/node/generated/test_shape_start_1_end_2",

diff --git a/third_party/llvm-project b/third_party/llvm-project
diff --git a/third_party/stablehlo b/third_party/stablehlo
diff --git a/third_party/torch-mlir b/third_party/torch-mlir
+17 −0		BUILD.bazel
+2 −2		WORKSPACE.bazel
+1 −1		build_tools/llvm_version.txt
+1 −0		docs/generated/stablehlo_linalg_passes.md
+7 −0		docs/generated/stablehlo_passes.md
+1 −0		docs/generated/stablehlo_tosa_passes.md
+6 −2		docs/spec.md
+199 −0		rfcs/20241001-microscaling-formats.md
+48 −5		stablehlo/conversions/linalg/tests/pointwise.mlir
+47 −45		stablehlo/conversions/linalg/transforms/StablehloToLinalgPointwise.cpp
+9 −10		stablehlo/conversions/linalg/transforms/TypeConversion.cpp
+3 −2		stablehlo/dialect/Base.td
+2 −2		stablehlo/dialect/Version.cpp
+1 −1		stablehlo/dialect/Version.h
+49 −1		stablehlo/dialect/VhloBytecode.cpp
+1 −0		stablehlo/dialect/VhloDialect.td
+24 −0		stablehlo/dialect/VhloTypes.cpp
+12 −0		stablehlo/dialect/VhloTypes.td
+41 −45		stablehlo/reference/Tensor.cpp
+6 −4		stablehlo/reference/Types.cpp
+1 −1		stablehlo/testdata/igamma_float64_20_20_float64_20_20_chlo.mlir
+1 −1		stablehlo/testdata/igammac_float64_20_20_float64_20_20_chlo.mlir
+16 −0		stablehlo/tests/interpret/api_input_arguments.mlir
+32 −0		stablehlo/tests/interpret/constant.mlir
+40 −8		stablehlo/tests/ops_stablehlo.mlir
+53 −53		stablehlo/tests/ops_stablehlo_quantized.mlir
+4 −0		stablehlo/tests/ops_stablehlo_roundtrip.mlir
+156 −0		stablehlo/tests/transforms/stablehlo_aggressive_folder.mlir
+550 −526		stablehlo/tests/transforms/stablehlo_aggressive_simplification.mlir
+2,936 −0		stablehlo/tests/vhlo/stablehlo_legalize_to_vhlo.1_8_0.mlir
+ −		stablehlo/tests/vhlo/stablehlo_legalize_to_vhlo.1_8_0.mlir.bc
+32 −0		stablehlo/tests/vhlo/stablehlo_legalize_to_vhlo.mlir
+35 −0		stablehlo/tests/vhlo/vhlo_to_version_downgrade_invalid.1_7_0.mlir
+15 −0		stablehlo/tests/vhlo/vhlo_to_version_downgrade_patch.mlir
+41 −2		stablehlo/tools/StablehloTranslateMain.cpp
+7 −2		stablehlo/transforms/CMakeLists.txt
+31 −2		stablehlo/transforms/PassUtils.cpp
+27 −12		stablehlo/transforms/PassUtils.h
+5 −0		stablehlo/transforms/Passes.h
+2 −0		stablehlo/transforms/Passes.td
+190 −7		stablehlo/transforms/StablehloAggressiveFolder.cpp
+98 −492		stablehlo/transforms/StablehloAggressiveSimplification.cpp
+281 −0		stablehlo/transforms/StablehloAggressiveSimplificationPatterns.td
+7 −0		stablehlo/transforms/VhloToVersion.cpp
+1 −1		externals/llvm-project
+1 −1		externals/stablehlo
+141 −0		include/torch-mlir/Dialect/Torch/IR/GeneratedTorchOps.td
+29 −3		lib/Conversion/TorchOnnxToTorch/DefaultDomainGtoP.cpp
+31 −28		lib/Conversion/TorchToLinalg/Linear.cpp
+48 −35		lib/Conversion/TorchToTosa/TorchToTosa.cpp
+152 −14		lib/Dialect/Torch/IR/TorchOps.cpp
+79 −0		lib/Dialect/Torch/Transforms/AbstractInterpLibrary.cpp
+132 −0		lib/Dialect/Torch/Transforms/DecomposeComplexOps.cpp
+2 −0		lib/Dialect/Torch/Transforms/LowerToBackendContract.cpp
+43 −43		lib/Dialect/TorchConversion/Transforms/BackendTypeConversion.cpp
+39 −42		projects/pt1/e2e_testing/xfail_sets.py
+49 −1		projects/pt1/python/torch_mlir/jit_ir_importer/build_tools/abstract_interp_lib_gen.py
+15 −4		projects/pt1/python/torch_mlir/jit_ir_importer/build_tools/torch_ods_gen.py
+161 −0		projects/pt1/python/torch_mlir_e2e_test/test_suite/backprop.py
+55 −39		projects/pt1/python/torch_mlir_e2e_test/test_suite/conv.py
+82 −0		projects/pt1/python/torch_mlir_e2e_test/test_suite/elementwise.py
+75 −55		projects/pt1/python/torch_mlir_e2e_test/test_suite/matmul.py
+1 −1		python/torch_mlir/fx.py
+1 −1		pytorch-hash.txt
+1 −1		pytorch-requirements.txt
+2 −2		test/CAPI/torch.c
+80 −0		test/Conversion/TorchOnnxToTorch/simple_ops_g_to_p.mlir
+2 −6		test/Conversion/TorchToLinalg/convolution.mlir
+32 −0		test/Conversion/TorchToTosa/basic.mlir
+94 −0		test/Dialect/Torch/canonicalize.mlir
+1 −1		torchvision-requirements.txt