aco: fix p_bpermute_gfx6 with input at non-zero byte

Same as the other bpermute pseudo instructions.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24693>
(cherry picked from commit 85957dd6e5)

.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

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 {