diff --git a/.pick_status.json b/.pick_status.json
index 5208dd7b037..3285f76b9a5 100644
--- a/.pick_status.json
+++ b/.pick_status.json
@@ -4054,7 +4054,7 @@
         "description": "aco: fix p_bpermute_gfx6 with input at non-zero byte",
         "nominated": true,
         "nomination_type": 0,
-        "resolution": 0,
+        "resolution": 1,
         "main_sha": null,
         "because_sha": null,
         "notes": null
diff --git a/src/amd/compiler/aco_lower_to_hw_instr.cpp b/src/amd/compiler/aco_lower_to_hw_instr.cpp
index ea43c3f0cef..efa27872629 100644
--- a/src/amd/compiler/aco_lower_to_hw_instr.cpp
+++ b/src/amd/compiler/aco_lower_to_hw_instr.cpp
@@ -1109,6 +1109,15 @@ emit_gfx6_bpermute(Program* program, aco_ptr<Instruction>& instr, Builder& bld)
       /* Restore original EXEC */
       bld.sop1(aco_opcode::s_mov_b64, Definition(exec, s2), Operand(temp_exec.physReg(), s2));
    }
+
+   /* RA assumes that the result is always in the low part of the register, so we have to shift,
+    * if it's not there already.
+    */
+   if (input.physReg().byte()) {
+      unsigned right_shift = input.physReg().byte() * 8;
+      bld.vop2(aco_opcode::v_lshrrev_b32, dst, Operand::c32(right_shift),
+               Operand(dst.physReg(), v1));
+   }
 }
 
 struct copy_operation {