aco: improve code for 32-bit isign

No shader-db changes on Navi.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4667>

diff --git a/src/amd/compiler/aco_instruction_selection.cpp b/src/amd/compiler/aco_instruction_selection.cpp
index a76f1bb25af72e4785d6accf687dc6fe27da10dc..acf8b36912aa8ba5bd851bc30857d1e102b64560 100644 (file)
--- a/src/amd/compiler/aco_instruction_selection.cpp
+++ b/src/amd/compiler/aco_instruction_selection.cpp
@@ -1130,9 +1130,8 @@ void visit_alu_instr(isel_context *ctx, nir_alu_instr *instr)
    case nir_op_isign: {
       Temp src = get_alu_src(ctx, instr->src[0]);
       if (dst.regClass() == s1) {
-         Temp tmp = bld.sop2(aco_opcode::s_ashr_i32, bld.def(s1), bld.def(s1, scc), src, Operand(31u));
-         Temp gtz = bld.sopc(aco_opcode::s_cmp_gt_i32, bld.def(s1, scc), src, Operand(0u));
-         bld.sop2(aco_opcode::s_add_i32, Definition(dst), bld.def(s1, scc), gtz, tmp);
+         Temp tmp = bld.sop2(aco_opcode::s_max_i32, bld.def(s1), bld.def(s1, scc), src, Operand((uint32_t)-1));
+         bld.sop2(aco_opcode::s_min_i32, Definition(dst), bld.def(s1, scc), tmp, Operand(1u));
       } 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;
@@ -1143,9 +1142,7 @@ void visit_alu_instr(isel_context *ctx, nir_alu_instr *instr)
          /* 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);
-         bld.vop2(aco_opcode::v_cndmask_b32, Definition(dst), Operand(1u), tmp, gtz);
+         bld.vop3(aco_opcode::v_med3_i32, Definition(dst), Operand((uint32_t)-1), src, Operand(1u));
       } else if (dst.regClass() == v2) {
          Temp upper = emit_extract_vector(ctx, src, 1, v1);
          Temp neg = bld.vop2(aco_opcode::v_ashrrev_i32, bld.def(v1), Operand(31u), upper);