diff --git a/src/amd/compiler/aco_insert_NOPs.cpp b/src/amd/compiler/aco_insert_NOPs.cpp
index 5bbf3f39a00..ad7942501e1 100644
--- a/src/amd/compiler/aco_insert_NOPs.cpp
+++ b/src/amd/compiler/aco_insert_NOPs.cpp
@@ -1388,6 +1388,12 @@ handle_instruction_gfx11(State& state, NOP_ctx_gfx11& ctx, aco_ptr<Instruction>&
 {
    Builder bld(state.program, &new_instructions);
 
+   /* Due to a hazard, an s_nop is needed before "s_sendmsg sendmsg_dealloc_vgprs". */
+   if (instr->opcode == aco_opcode::s_sendmsg && instr->salu().imm == sendmsg_dealloc_vgprs &&
+       (new_instructions.empty() || new_instructions.back()->opcode != aco_opcode::s_nop)) {
+      bld.sopp(aco_opcode::s_nop, 0);
+   }
+
    /* VcmpxPermlaneHazard
     * Handle any permlane following a VOPC instruction writing exec, insert v_mov between them.
     */
diff --git a/src/amd/compiler/aco_ir.cpp b/src/amd/compiler/aco_ir.cpp
index 022dd60e87c..19aa6e6e2eb 100644
--- a/src/amd/compiler/aco_ir.cpp
+++ b/src/amd/compiler/aco_ir.cpp
@@ -1555,8 +1555,6 @@ dealloc_vgprs(Program* program)
    Builder bld(program);
    if (!block.instructions.empty() && block.instructions.back()->opcode == aco_opcode::s_endpgm) {
       bld.reset(&block.instructions, block.instructions.begin() + (block.instructions.size() - 1));
-      /* Due to a hazard, an s_nop is needed before "s_sendmsg sendmsg_dealloc_vgprs". */
-      bld.sopp(aco_opcode::s_nop, 0);
       bld.sopp(aco_opcode::s_sendmsg, sendmsg_dealloc_vgprs);
    }
 
diff --git a/src/amd/compiler/aco_lower_to_hw_instr.cpp b/src/amd/compiler/aco_lower_to_hw_instr.cpp
index 9d7c895f195..f0b2ac191de 100644
--- a/src/amd/compiler/aco_lower_to_hw_instr.cpp
+++ b/src/amd/compiler/aco_lower_to_hw_instr.cpp
@@ -2397,10 +2397,8 @@ lower_to_hw_instr(Program* program)
                   if (program->stage == fragment_fs)
                      bld.exp(aco_opcode::exp, Operand(v1), Operand(v1), Operand(v1), Operand(v1), 0,
                              target, false, true, true);
-                  if (should_dealloc_vgprs) {
-                     bld.sopp(aco_opcode::s_nop, 0);
+                  if (should_dealloc_vgprs)
                      bld.sopp(aco_opcode::s_sendmsg, sendmsg_dealloc_vgprs);
-                  }
                   bld.sopp(aco_opcode::s_endpgm);
 
                   bld.reset(&ctx.instructions);
diff --git a/src/amd/compiler/tests/test_to_hw_instr.cpp b/src/amd/compiler/tests/test_to_hw_instr.cpp
index 8e26695f632..574209f396d 100644
--- a/src/amd/compiler/tests/test_to_hw_instr.cpp
+++ b/src/amd/compiler/tests/test_to_hw_instr.cpp
@@ -209,7 +209,6 @@ BEGIN_TEST(to_hw_instr.swap_subdword)
       bld.pseudo(aco_opcode::p_parallelcopy, Definition(v128_hi, v2b), Definition(v129_lo, v2b),
                  Operand(v129_lo, v2b), Operand(v128_hi, v2b));
 
-      //~gfx11! s_nop
       //~gfx11! s_sendmsg sendmsg(dealloc_vgprs)
       //~gfx(8|9|11)! s_endpgm
 
@@ -361,7 +360,6 @@ BEGIN_TEST(to_hw_instr.subdword_constant)
       bld.pseudo(aco_opcode::p_unit_test, Operand::c32(17u));
       bld.pseudo(aco_opcode::p_parallelcopy, Definition(v0_lo, v1b), Operand::zero(1));
 
-      //~gfx11! s_nop
       //~gfx11! s_sendmsg sendmsg(dealloc_vgprs)
       //! s_endpgm
 
@@ -493,7 +491,6 @@ BEGIN_TEST(to_hw_instr.extract)
 
          finish_to_hw_instr_test();
 
-         //~gfx11_.*! s_nop
          //~gfx11_.*! s_sendmsg sendmsg(dealloc_vgprs)
          //! s_endpgm
       }
@@ -612,7 +609,6 @@ BEGIN_TEST(to_hw_instr.insert)
 
       finish_to_hw_instr_test();
 
-      //~gfx11! s_nop
       //~gfx11! s_sendmsg sendmsg(dealloc_vgprs)
       //! s_endpgm
    }
@@ -764,7 +760,6 @@ BEGIN_TEST(to_hw_instr.pack2x16_constant)
       bld.pseudo(aco_opcode::p_parallelcopy, Definition(v0_lo, v2b), Definition(v0_hi, v2b),
                  Operand::zero(2), Operand(v1_lo, v2b));
 
-      //~gfx11! s_nop
       //~gfx11! s_sendmsg sendmsg(dealloc_vgprs)
       //! s_endpgm