aco: implement 64-bit ineg

We currently lower them, but nir_opt_algebraic() can add new ones because
lower_sub=true.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>

diff --git a/src/amd/compiler/aco_instruction_selection.cpp b/src/amd/compiler/aco_instruction_selection.cpp
index d52043f3c0d4a81f3a6e3adfb64d84d9f9dbf078..7405b1142f9758cef353fb32906d1d6a933966a8 100644 (file)
--- a/src/amd/compiler/aco_instruction_selection.cpp
+++ b/src/amd/compiler/aco_instruction_selection.cpp
@@ -689,6 +689,22 @@ void visit_alu_instr(isel_context *ctx, nir_alu_instr *instr)
          bld.vsub32(Definition(dst), Operand(0u), Operand(src));
       } else if (dst.regClass() == s1) {
          bld.sop2(aco_opcode::s_mul_i32, Definition(dst), Operand((uint32_t) -1), src);
+      } else if (dst.size() == 2) {
+         Temp src0 = bld.tmp(dst.type(), 1);
+         Temp src1 = bld.tmp(dst.type(), 1);
+         bld.pseudo(aco_opcode::p_split_vector, Definition(src0), Definition(src1), src);
+
+         if (dst.regClass() == s2) {
+            Temp carry = bld.tmp(s1);
+            Temp dst0 = bld.sop2(aco_opcode::s_sub_u32, bld.def(s1), bld.scc(Definition(carry)), Operand(0u), src0);
+            Temp dst1 = bld.sop2(aco_opcode::s_subb_u32, bld.def(s1), bld.def(s1, scc), Operand(0u), src1, carry);
+            bld.pseudo(aco_opcode::p_create_vector, Definition(dst), dst0, dst1);
+         } else {
+            Temp lower = bld.tmp(v1);
+            Temp borrow = bld.vsub32(Definition(lower), Operand(0u), src0, true).def(1).getTemp();
+            Temp upper = bld.vsub32(bld.def(v1), Operand(0u), src1, false, borrow);
+            bld.pseudo(aco_opcode::p_create_vector, Definition(dst), lower, upper);
+         }
       } else {
          fprintf(stderr, "Unimplemented NIR instr bit size: ");
          nir_print_instr(&instr->instr, stderr);

diff --git a/src/amd/compiler/aco_instruction_selection_setup.cpp b/src/amd/compiler/aco_instruction_selection_setup.cpp
index f77d12d23648a0cae66dfaedb14e32cfe157be7f..3a276035ba547f548c3effbd8b08bdfa0ae1c0c5 100644 (file)
--- a/src/amd/compiler/aco_instruction_selection_setup.cpp
+++ b/src/amd/compiler/aco_instruction_selection_setup.cpp
@@ -1314,8 +1314,7 @@ setup_isel_context(Program* program,
                                                       nir_lower_divmod64 |
                                                       nir_lower_logic64 |
                                                       nir_lower_minmax64 |
-                                                      nir_lower_iabs64 |
-                                                      nir_lower_ineg64));
+                                                      nir_lower_iabs64));
 
       nir_opt_idiv_const(nir, 32);
       nir_lower_idiv(nir); // TODO: use the LLVM path once !1239 is merged