Resolve merge conflicts for sync_20260505

Author: Harsha Havanur Shamsundara

tensorflow/compiler/mlir/tensorflow/tests/BUILD

@@ -17,17 +17,13 @@ glob_lit_tests(
         "layout_optimization_to_nhwc.mlir": "medium",
     },
     tags_override = {
-<<<<<<< HEAD
-        "tf-reduce-identity.mlir": ["no_windows"],
-=======
         "optimize.mlir": ["no_rocm"],
         "tf_optimize.mlir": ["no_rocm"],
         # TODO(b/495279482): Remove no_oss once the compiler crash is fixed.
         "tf-reduce-identity.mlir": [
             "no_windows",
             "no_oss",
         ],
->>>>>>> upstream/master
     },
     test_file_exts = ["mlir"],
 )

tensorflow/core/kernels/image/image_ops_gpu.cu.cc

@@ -28,16 +28,10 @@ namespace functor {
 
 // Explicit instantiation of the GPU functor.
 typedef Eigen::GpuDevice GPUDevice;
-<<<<<<< HEAD
-template class FillProjectiveTransform<GPUDevice, uint8>;
-template class FillProjectiveTransform<GPUDevice, int32>;
-template class FillProjectiveTransform<GPUDevice, int64>;
-=======
 
 template class FillProjectiveTransform<GPUDevice, uint8_t>;
 template class FillProjectiveTransform<GPUDevice, int32_t>;
 template class FillProjectiveTransform<GPUDevice, int64_t>;
->>>>>>> upstream/master
 template class FillProjectiveTransform<GPUDevice, Eigen::half>;
 template class FillProjectiveTransform<GPUDevice, float>;
 template class FillProjectiveTransform<GPUDevice, double>;

tensorflow/core/kernels/linalg/matrix_solve_op.cc

@@ -335,22 +335,12 @@ class MatrixSolveOpGpu : public AsyncOpKernel {
                                pivots_mat.data(), transposed_rhs_ptrs_base, n,
                                &host_info, batch_size),
           done);
-<<<<<<< HEAD
-
-      OP_REQUIRES_ASYNC(
-          context, host_info == 0,
-          errors::InvalidArgument("The ", -host_info,
-                                  "'th argument to cublas*getrsBatched had "
-                                  "an illegal value."),
-          done);
-=======
       OP_REQUIRES_ASYNC(context, host_info == 0,
                         absl::InvalidArgumentError(absl::StrCat(
                             "The ", -host_info,
                             "'th argument to cublas*getrsBatched had "
                             "an illegal value.")),
                         done);
->>>>>>> upstream/master
     } else {
       dev_info.push_back(solver->GetDeviceLapackInfo(batch_size, "getrs"));
       for (int batch = 0; batch < batch_size; ++batch) {

third_party/xla/xla/backends/gpu/codegen/triton/triton_gemm_fusion_test.cc

@@ -490,13 +490,8 @@ ENTRY entry {
 
   const HloFusionInstruction* fusion1 = Cast<HloFusionInstruction>(
       module1_and_metadata.computation->FusionInstruction());
-<<<<<<< HEAD
-    EXPECT_THAT(
-      TritonWrapper("test_fn", fusion1, se::GpuComputeCapability{cc},
-=======
   EXPECT_THAT(
       TritonWrapper("test_fn", *fusion1, se::GpuComputeCapability{cc},
->>>>>>> upstream/master
                     device_info, module1_and_metadata.block_level_parameters,
                     target_triple, data_layout, llvm_ctx, mlir_context_),
       absl_testing::StatusIs(

third_party/xla/xla/backends/gpu/collectives/gpu_clique_key.h

@@ -54,26 +54,7 @@ namespace xla::gpu {
 // clique have a well defined total execution order, enforced with events.
 TSL_LIB_GTL_DEFINE_INT_TYPE(CommunicationId, uint64_t);
 
-<<<<<<< HEAD
-bool IsP2PStreamKind(AsyncStreamKind stream_kind);
-
-inline constexpr int64_t kAsyncStreamTotal =
-    static_cast<int64_t>(AsyncStreamKind::ASYNC_STREAM_KIND_MEMCPYP2P) + 1;
-
-// Strongly-typed wrapper to represent collective stream ID.
-TSL_LIB_GTL_DEFINE_INT_TYPE(CollectiveStreamId, uint64_t);
-
-// Assigns a unique ID to a stream for asynchronous or synchronous execution.
-// These IDs can be used, for example, to look up the NCCL communicator.
-CollectiveStreamId GetCollectiveStreamId(
-    bool is_async, CollectiveStreamId stream_id = CollectiveStreamId(1),
-    AsyncStreamKind stream_kind =
-        AsyncStreamKind::ASYNC_STREAM_KIND_COLLECTIVE);
-
 // StrJoin for device groups that shortens long list of devices for readbility.
-=======
-// StrJoin for device groups that shortens long list of devices for readability.
->>>>>>> upstream/master
 std::string HumanReadableDeviceGroups(
     absl::Span<const std::vector<GlobalDeviceId>> device_groups,
     absl::string_view separator = ",", size_t first = 2, size_t last = 1);

third_party/xla/xla/backends/gpu/runtime/gpublas_lt_matmul_thunk.cc

@@ -195,14 +195,6 @@ absl::Status CublasLtMatmulThunk::ExecuteOnStreamInternal(
     workspace = allocs.GetDeviceAddress(workspace_->slice);
   }
 
-<<<<<<< HEAD
-
-  return plan->ExecuteOnStream(
-      stream, allocs.GetDeviceAddress(a_.slice),
-      allocs.GetDeviceAddress(b_.slice), allocs.GetDeviceAddress(c_.slice),
-      allocs.GetDeviceAddress(d_.slice), bias, aux, a_scale, b_scale, c_scale,
-      d_scale, d_amax, workspace);
-=======
   if (is_grouped()) {
     if (!group_sizes_.has_value()) {
       return absl::InternalError(
@@ -225,7 +217,6 @@ absl::Status CublasLtMatmulThunk::ExecuteOnStreamInternal(
         allocs.GetDeviceAddress(d_.slice), bias, aux, a_scale, b_scale, c_scale,
         d_scale, d_amax, workspace);
   }
->>>>>>> upstream/master
 }
 
 absl::StatusOr<se::gpu::BlasLt::MatmulPlan*>

third_party/xla/xla/pjrt/triton_rocm.cc

@@ -118,15 +118,6 @@ absl::StatusOr<std::string> LLVMToHSACO(mlir::ModuleOp module,
   }
 
   xla::DebugOptions debug_opts = xla::DefaultDebugOptionsIgnoringFlags();
-<<<<<<< HEAD
-  auto llvm_opts = xla::gpu::amdgpu::GetAMDGPUBackendOptions(debug_opts);
-  xla::llvm_ir::LLVMCommandLineOptionsLock llvm_lock(llvm_opts);
-  TF_ASSIGN_OR_RETURN(auto hsaco_file_result,
-                      xla::gpu::amdgpu::CompileToHsacoAndReturnFilePath(
-                          llvm_module.get(), gpu_version, debug_opts, false,
-                          llvm_lock));
-  return hsaco_file_result.hsaco_path;
-=======
   TF_ASSIGN_OR_RETURN(auto compile_result,
                       gpu::amdgpu::CompileToHsaco(llvm_module.get(),
                                                   gpu_version, debug_opts, ""));
@@ -148,7 +139,6 @@ absl::StatusOr<std::string> LLVMToHSACO(mlir::ModuleOp module,
   ofs.write(reinterpret_cast<const char*>(compile_result.hsaco.data()),
             compile_result.hsaco.size());
   return temp_file_path;
->>>>>>> upstream/master
 }
 
 }  // namespace

third_party/xla/xla/python/ifrt/ir/tests/ifrt_lower_atom_program_metadata_to_xla-gspmd_partitioned_compile_override.mlir

@@ -69,28 +69,7 @@ module @arg_unspecified_sharding attributes {ifrt.num_devices = 2, ifrt.compile_
 
 // -----
 
-<<<<<<< HEAD
-// CHECK-LABEL: @arg_unspecified_sharding_sdy_partitioned
-module @arg_unspecified_sharding_sdy_partitioned attributes {ifrt.num_devices = 2, ifrt.is_sdy_partitioned} {
-  // CHECK: %arg0: tensor<2x2xi32>
-  // CHECK-SAME: {
-  // CHECK-NOT:    mhlo.sharding
-  // CHECK-DAG:    ifrt.sharding = #ifrt.sharding_param<2x1 to [0] on 2>
-  // CHECK-SAME: }
-  // CHECK: %arg1: tensor<2x2xi32> {ifrt.sharding = #ifrt.sharding_unspecified})
-  func.func @main(
-      %arg0: tensor<2x2xi32> {
-        ifrt.sharding=#ifrt.sharding_param<2x1 to [0] on 2>},
-      %arg1: tensor<2x2xi32> {ifrt.sharding=#ifrt.sharding_unspecified}) {
-    return
-  }
-}
-
-// -----
-
-=======
 // CHECK: #sp = #ifrt.sharding_param<2x1 to [0] on 2>
->>>>>>> upstream/master
 // CHECK-LABEL: @result_metadata
 module @result_metadata attributes {ifrt.num_devices = 2, ifrt.compile_options_key = "test_override"} {
   // CHECK: -> (tensor<2x2xi32>
@@ -111,30 +90,7 @@ module @result_metadata attributes {ifrt.num_devices = 2, ifrt.compile_options_k
 
 // -----
 
-<<<<<<< HEAD
-// CHECK-LABEL: @result_metadata_sdy_partitioned
-module @result_metadata_sdy_partitioned attributes {ifrt.num_devices = 2, ifrt.is_sdy_partitioned} {
-  // CHECK: -> (tensor<2x2xi32>
-  // CHECK-SAME: {
-  // CHECK-NOT:    mhlo.sharding
-  // CHECK-DAG:    ifrt.sharding = #ifrt.sharding_param<2x1 to [0] on 2>
-  // CHECK-DAG:    ifrt.memory_kind = "device"
-  // CHECK-DAG:    mhlo.memory_kind = "device"
-  // CHECK-SAME: }
-  func.func @main()
-      -> (tensor<2x2xi32> {
-        ifrt.sharding=#ifrt.sharding_param<2x1 to [0] on 2>,
-        ifrt.memory_kind = "device"}) {
-    %0 = mhlo.constant dense<1> : tensor<2x2xi32>
-    return %0 : tensor<2x2xi32>
-  }
-}
-
-// -----
-
-=======
 // CHECK: #sp = #ifrt.sharding_param<2x1 to [0] on 2>
->>>>>>> upstream/master
 // CHECK-LABEL: @result_unspecified_sharding
 module @result_unspecified_sharding attributes {ifrt.num_devices = 2, ifrt.compile_options_key = "test_override"} {
   // CHECK: -> (tensor<2x2xi32>