From 66d5bfb09ab067469ec747aa7745f77c302e2df6 Mon Sep 17 00:00:00 2001 From: Samuel Pitoiset Date: Fri, 3 Apr 2020 11:47:54 +0200 Subject: [PATCH] aco: implement 16-bit nir_op_ffloor/nir_op_fceil MIME-Version: 1.0 Content-Type: text/plain; charset=utf8 Content-Transfer-Encoding: 8bit Signed-off-by: Samuel Pitoiset Reviewed-by: Daniel Schürmann Part-of: --- src/amd/compiler/aco_instruction_selection.cpp | 20 +++++++++++++------- 1 file changed, 13 insertions(+), 7 deletions(-) diff --git a/src/amd/compiler/aco_instruction_selection.cpp b/src/amd/compiler/aco_instruction_selection.cpp index 0befc68..3d4750c 100644 --- a/src/amd/compiler/aco_instruction_selection.cpp +++ b/src/amd/compiler/aco_instruction_selection.cpp @@ -1880,10 +1880,14 @@ void visit_alu_instr(isel_context *ctx, nir_alu_instr *instr) break; } case nir_op_ffloor: { - if (dst.size() == 1) { + Temp src = get_alu_src(ctx, instr->src[0]); + if (dst.regClass() == v2b) { + Temp tmp = bld.vop1(aco_opcode::v_floor_f16, bld.def(v1), src); + bld.pseudo(aco_opcode::p_split_vector, Definition(dst), bld.def(v2b), tmp); + } else if (dst.regClass() == v1) { emit_vop1_instruction(ctx, instr, aco_opcode::v_floor_f32, dst); - } else if (dst.size() == 2) { - emit_floor_f64(ctx, bld, Definition(dst), get_alu_src(ctx, instr->src[0])); + } else if (dst.regClass() == v2) { + emit_floor_f64(ctx, bld, Definition(dst), src); } else { fprintf(stderr, "Unimplemented NIR instr bit size: "); nir_print_instr(&instr->instr, stderr); @@ -1892,15 +1896,17 @@ void visit_alu_instr(isel_context *ctx, nir_alu_instr *instr) break; } case nir_op_fceil: { - if (dst.size() == 1) { + Temp src0 = get_alu_src(ctx, instr->src[0]); + if (dst.regClass() == v2b) { + Temp tmp = bld.vop1(aco_opcode::v_ceil_f16, bld.def(v1), src0); + bld.pseudo(aco_opcode::p_split_vector, Definition(dst), bld.def(v2b), tmp); + } else if (dst.regClass() == v1) { emit_vop1_instruction(ctx, instr, aco_opcode::v_ceil_f32, dst); - } else if (dst.size() == 2) { + } else if (dst.regClass() == v2) { if (ctx->options->chip_class >= GFX7) { emit_vop1_instruction(ctx, instr, aco_opcode::v_ceil_f64, dst); } else { /* GFX6 doesn't support V_CEIL_F64, lower it. */ - Temp src0 = get_alu_src(ctx, instr->src[0]); - /* trunc = trunc(src0) * if (src0 > 0.0 && src0 != trunc) * trunc += 1.0 -- 2.7.4