aco: implement 16-bit nir_op_fabs/nir_op_fneg

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4452>

diff --git a/src/amd/compiler/aco_instruction_selection.cpp b/src/amd/compiler/aco_instruction_selection.cpp
index 11dda97e2c71af1fa90e76cef8b0540592c99c91..f454f48e0548331a2585d5d8b3cbf819dc6d772b 100644 (file)
--- a/src/amd/compiler/aco_instruction_selection.cpp
+++ b/src/amd/compiler/aco_instruction_selection.cpp
@@ -1754,11 +1754,14 @@ void visit_alu_instr(isel_context *ctx, nir_alu_instr *instr)
    }
    case nir_op_fneg: {
       Temp src = get_alu_src(ctx, instr->src[0]);
-      if (dst.size() == 1) {
+      if (dst.regClass() == v2b) {
+         Temp tmp = bld.vop2(aco_opcode::v_xor_b32, bld.def(v1), Operand(0x8000u), as_vgpr(ctx, src));
+         bld.pseudo(aco_opcode::p_split_vector, Definition(dst), bld.def(v2b), tmp);
+      } else if (dst.regClass() == v1) {
          if (ctx->block->fp_mode.must_flush_denorms32)
             src = bld.vop2(aco_opcode::v_mul_f32, bld.def(v1), Operand(0x3f800000u), as_vgpr(ctx, src));
          bld.vop2(aco_opcode::v_xor_b32, Definition(dst), Operand(0x80000000u), as_vgpr(ctx, src));
-      } else if (dst.size() == 2) {
+      } else if (dst.regClass() == v2) {
          if (ctx->block->fp_mode.must_flush_denorms16_64)
             src = bld.vop3(aco_opcode::v_mul_f64, bld.def(v2), Operand(0x3FF0000000000000lu), as_vgpr(ctx, src));
          Temp upper = bld.tmp(v1), lower = bld.tmp(v1);
@@ -1774,11 +1777,14 @@ void visit_alu_instr(isel_context *ctx, nir_alu_instr *instr)
    }
    case nir_op_fabs: {
       Temp src = get_alu_src(ctx, instr->src[0]);
-      if (dst.size() == 1) {
+      if (dst.regClass() == v2b) {
+         Temp tmp = bld.vop2(aco_opcode::v_and_b32, bld.def(v1), Operand(0x7FFFu), as_vgpr(ctx, src));
+         bld.pseudo(aco_opcode::p_split_vector, Definition(dst), bld.def(v2b), tmp);
+      } else if (dst.regClass() == v1) {
          if (ctx->block->fp_mode.must_flush_denorms32)
             src = bld.vop2(aco_opcode::v_mul_f32, bld.def(v1), Operand(0x3f800000u), as_vgpr(ctx, src));
          bld.vop2(aco_opcode::v_and_b32, Definition(dst), Operand(0x7FFFFFFFu), as_vgpr(ctx, src));
-      } else if (dst.size() == 2) {
+      } else if (dst.regClass() == v2) {
          if (ctx->block->fp_mode.must_flush_denorms16_64)
             src = bld.vop3(aco_opcode::v_mul_f64, bld.def(v2), Operand(0x3FF0000000000000lu), as_vgpr(ctx, src));
          Temp upper = bld.tmp(v1), lower = bld.tmp(v1);