Reverts 22f9c10

PiperOrigin-RevId: 726090280

xla/service/gpu/BUILD

@@ -1556,7 +1556,6 @@ cc_library(
         "//xla/service/llvm_ir:llvm_command_line_options",
         "//xla/service/llvm_ir:llvm_util",
         "//xla/service/spmd:collective_permute_motion",
-        "//xla/service/spmd:schedule_aware_collective_ops_cse",
         "//xla/service:all_reduce_promotion",
         "//xla/service:all_reduce_reassociate",
         "//xla/service:all_reduce_simplifier",

xla/service/gpu/gpu_compiler.cc

@@ -246,7 +246,6 @@ limitations under the License.
 #include "xla/service/select_and_scatter_expander.h"
 #include "xla/service/sharding_remover.h"
 #include "xla/service/slow_operation_alarm.h"
-#include "xla/service/spmd/schedule_aware_collective_ops_cse.h"
 #include "xla/service/topk_rewriter.h"
 #include "xla/service/transpose_folding.h"
 #include "xla/service/while_loop_all_reduce_code_motion.h"
@@ -859,10 +858,6 @@ absl::Status RunCollectiveOptimizationPasses(
       /*enable_reduce_scatter=*/debug_options
           .xla_gpu_enable_while_loop_reduce_scatter_code_motion());
 
-  collectives_pipeline.AddPass<ScheduleAwareCollectiveOpsCSE>(
-      /*distance_threshold=*/100,
-      /*for_replicas=*/false);
-
   // Moves collectives' subsequent quantization before the collective to
   // minimize data transfers.
   collectives_pipeline.AddPass<CollectiveQuantizer>();

xla/tools/hlo_opt/BUILD

@@ -63,7 +63,6 @@ cc_library(
         "//xla/service/gpu/transforms:collective_permute_valid_iteration_annotator",
         "//xla/service/gpu/transforms:scatter_expander",
         "//xla/service/gpu/transforms:scatter_slice_simplifier",
-        "//xla/service/spmd:schedule_aware_collective_ops_cse",
         "//xla/service/spmd/shardy:shardy_xla_pass",
         "//xla/stream_executor:platform",
         "//xla/stream_executor:stream_executor_h",
@@ -217,7 +216,6 @@ lit_test_suite(
             "tests/gpu_hlo.hlo",
             "tests/gpu_hlo_backend.hlo",
             "tests/gpu_hlo_buffers.hlo",
-            "tests/gpu_hlo_collective_cse.hlo",
             "tests/gpu_hlo_llvm.hlo",
             "tests/gpu_hlo_pass.hlo",
             "tests/gpu_hlo_ptx.hlo",

xla/tools/hlo_opt/compiled_opt_lib.cc

@@ -49,7 +49,6 @@ limitations under the License.
 #include "xla/service/scatter_simplifier.h"
 #include "xla/service/select_and_scatter_expander.h"
 #include "xla/service/sharding_remover.h"
-#include "xla/service/spmd/schedule_aware_collective_ops_cse.h"
 #include "xla/service/spmd/shardy/shardy_xla_pass.h"
 #include "xla/service/topk_rewriter.h"
 #include "xla/service/triangular_solve_expander.h"
@@ -165,7 +164,6 @@ void CompiledOptProvider::RegisterSharedHardwareSpecificPasses() {
   RegisterPass<ScatterDeterminismExpander>();
   RegisterPass<ScatterSimplifier>();
   RegisterPass<ScatterSliceSimplifier>();
-  RegisterPass<ScheduleAwareCollectiveOpsCSE>(100, false);
   RegisterPass<SelectAndScatterExpander>();
   RegisterPass<ShardingRemover>();
   RegisterPass<TopkDecomposer>();