aco: implement nir_op_isign on SI/CI

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>

diff --git a/src/amd/compiler/aco_instruction_selection.cpp b/src/amd/compiler/aco_instruction_selection.cpp
index f63daa31bcae42d8ca6b741c3701ff4cc7efe7ef..2cffa19b29b13af304c857584b976df6e50ce63f 100644 (file)
--- a/src/amd/compiler/aco_instruction_selection.cpp
+++ b/src/amd/compiler/aco_instruction_selection.cpp
@@ -835,8 +835,13 @@ void visit_alu_instr(isel_context *ctx, nir_alu_instr *instr)
          bld.sop2(aco_opcode::s_add_i32, Definition(dst), bld.def(s1, scc), gtz, tmp);
       } else if (dst.regClass() == s2) {
          Temp neg = bld.sop2(aco_opcode::s_ashr_i64, bld.def(s2), bld.def(s1, scc), src, Operand(63u));
-         Temp neqz = bld.sopc(aco_opcode::s_cmp_lg_u64, bld.def(s1, scc), src, Operand(0u));
-         bld.sop2(aco_opcode::s_or_b64, Definition(dst), bld.def(s1, scc), neg, neqz);
+         Temp neqz;
+         if (ctx->program->chip_class >= GFX8)
+            neqz = bld.sopc(aco_opcode::s_cmp_lg_u64, bld.def(s1, scc), src, Operand(0u));
+         else
+            neqz = bld.sop2(aco_opcode::s_or_b64, bld.def(s2), bld.def(s1, scc), src, Operand(0u)).def(1).getTemp();
+         /* SCC gets zero-extended to 64 bit */
+         bld.sop2(aco_opcode::s_or_b64, Definition(dst), bld.def(s1, scc), neg, bld.scc(neqz));
       } else if (dst.regClass() == v1) {
          Temp tmp = bld.vop2(aco_opcode::v_ashrrev_i32, bld.def(v1), Operand(31u), src);
          Temp gtz = bld.vopc(aco_opcode::v_cmp_ge_i32, bld.hint_vcc(bld.def(bld.lm)), Operand(0u), src);