diff --git a/.pick_status.json b/.pick_status.json index e02d0a747fd..44dbf34ddd6 100644 --- a/.pick_status.json +++ b/.pick_status.json @@ -5125,7 +5125,7 @@ "description": "aco: implement missing nir_op_unpack_half_2x16_split_{x,y}_flush_to_zero", "nominated": true, "nomination_type": 1, - "resolution": 0, + "resolution": 1, "master_sha": null, "because_sha": "56d9bcdded8f3eb7bd45262ce013ef1809d8edb1" }, diff --git a/src/amd/compiler/aco_instruction_selection.cpp b/src/amd/compiler/aco_instruction_selection.cpp index 0f151cb481e..f98499fcc3b 100644 --- a/src/amd/compiler/aco_instruction_selection.cpp +++ b/src/amd/compiler/aco_instruction_selection.cpp @@ -2857,8 +2857,10 @@ void visit_alu_instr(isel_context *ctx, nir_alu_instr *instr) } break; } + case nir_op_unpack_half_2x16_split_x_flush_to_zero: case nir_op_unpack_half_2x16_split_x: { if (dst.regClass() == v1) { + assert(ctx->block->fp_mode.must_flush_denorms16_64 == (instr->op == nir_op_unpack_half_2x16_split_x_flush_to_zero)); bld.vop1(aco_opcode::v_cvt_f32_f16, Definition(dst), get_alu_src(ctx, instr->src[0])); } else { fprintf(stderr, "Unimplemented NIR instr bit size: "); @@ -2867,8 +2869,10 @@ void visit_alu_instr(isel_context *ctx, nir_alu_instr *instr) } break; } + case nir_op_unpack_half_2x16_split_y_flush_to_zero: case nir_op_unpack_half_2x16_split_y: { if (dst.regClass() == v1) { + assert(ctx->block->fp_mode.must_flush_denorms16_64 == (instr->op == nir_op_unpack_half_2x16_split_y_flush_to_zero)); /* TODO: use SDWA here */ bld.vop1(aco_opcode::v_cvt_f32_f16, Definition(dst), bld.vop2(aco_opcode::v_lshrrev_b32, bld.def(v1), Operand(16u), as_vgpr(ctx, get_alu_src(ctx, instr->src[0]))));