Revert "[BACKEND] Add barrier after assert op to avoid race condition" (triton-lang#5047)

Reverts triton-lang#5035

Author: ThomasRaoux

lib/Conversion/TritonGPUToLLVM/AssertOpToLLVM.cpp

@@ -35,11 +35,6 @@ struct AssertOpConversion : public ConvertOpToLLVMPattern<triton::AssertOp> {
       }
     }
     llAssert(op, condition, adaptor.getMessage(), rewriter);
-    // Add a barrier to avoid a race condition in case an assert is followed by
-    // an op that may trap if the assert condition is true. Since the tensor in
-    // those two operations may have different layout we need to make sure all
-    // the threads are done executing the assert before going to the next op.
-    barrier();
     rewriter.eraseOp(op);
     return success();
   }

test/Conversion/tritongpu_to_llvm.mlir

@@ -1906,8 +1906,6 @@ module attributes {"triton_gpu.num-ctas" = 1 : i32, "triton_gpu.num-warps" = 4 :
 // CHECK-DAG: llvm.mlir.global internal constant @assertFunc_0("unknown\00") {addr_space = 0 : i32}
 // CHECK-DAG: llvm.mlir.global internal constant @assertFile_0("inner_call\00") {addr_space = 0 : i32}
 // CHECK-DAG: llvm.mlir.global internal constant @assertMessage_0("assert text\00") {addr_space = 0 : i32}
-// CHECK: llvm.call @__assertfail
-// CHECK: nvvm.barrier0
 module attributes {"triton_gpu.num-ctas" = 1 : i32, "triton_gpu.num-warps" = 4 : i32, triton_gpu.target = "cuda:90", "triton_gpu.threads-per-warp" = 32 : i32} {
   tt.func public @add_kernel(%arg0: tensor<1xi1, #blocked>) {
     tt.assert %arg0, "assert text" : tensor<1xi1, #blocked> loc(#loc5)