aco: implement 64-bit sgpr swaps

In our pipeline-db, helps almost exclusively Detroit: Become Human.

Totals from 6726 (5.36% of 125503) affected shaders:
CodeSize: 74680952 -> 74102228 (-0.77%)
Instrs: 14551507 -> 14406001 (-1.00%)
Cycles: 1748272436 -> 1690173104 (-3.32%)
VMEM: 964671 -> 964058 (-0.06%)
Copies: 1993312 -> 1847806 (-7.30%)

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4469>

diff --git a/src/amd/compiler/aco_lower_to_hw_instr.cpp b/src/amd/compiler/aco_lower_to_hw_instr.cpp
index 4a9cc9c9c622f7ba595300e1b69ad8b41ca64a19..da381dc7ced5c73884425d5c59bb9653ee60bf37 100644 (file)
--- a/src/amd/compiler/aco_lower_to_hw_instr.cpp
+++ b/src/amd/compiler/aco_lower_to_hw_instr.cpp
@@ -818,7 +818,7 @@ void do_swap(lower_context *ctx, Builder& bld, const copy_operation& copy, bool
    for (; offset < copy.bytes;) {
       Definition def;
       Operand op;
-      split_copy(offset, &def, &op, copy, true, 4);
+      split_copy(offset, &def, &op, copy, true, 8);
 
       assert(op.regClass() == def.regClass());
       Operand def_as_op = Operand(def.physReg(), def.regClass());
@@ -852,6 +852,15 @@ void do_swap(lower_context *ctx, Builder& bld, const copy_operation& copy, bool
             bld.sop2(aco_opcode::s_xor_b32, op_as_def, Definition(scc, s1), op, def_as_op);
          }
          ctx->program->statistics[statistic_copies] += 3;
+      } else if (def.regClass() == s2) {
+         if (preserve_scc)
+            bld.sop1(aco_opcode::s_mov_b32, Definition(pi->scratch_sgpr, s1), Operand(scc, s1));
+         bld.sop2(aco_opcode::s_xor_b64, op_as_def, Definition(scc, s1), op, def_as_op);
+         bld.sop2(aco_opcode::s_xor_b64, def, Definition(scc, s1), op, def_as_op);
+         bld.sop2(aco_opcode::s_xor_b64, op_as_def, Definition(scc, s1), op, def_as_op);
+         if (preserve_scc)
+            bld.sopc(aco_opcode::s_cmp_lg_i32, Definition(scc, s1), Operand(pi->scratch_sgpr, s1), Operand(0u));
+         ctx->program->statistics[statistic_copies] += 3;
       } else if (ctx->program->chip_class >= GFX9 && def.bytes() == 2 && def.physReg().reg() == op.physReg().reg()) {
          aco_ptr<VOP3P_instruction> vop3p{create_instruction<VOP3P_instruction>(aco_opcode::v_pk_add_u16, Format::VOP3P, 2, 1)};
          vop3p->operands[0] = Operand(PhysReg{op.physReg().reg()}, v1);