1 /*
2 * Copyright © 2018 Intel Corporation
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21 * IN THE SOFTWARE.
22 */
23
24 #include "intel_nir.h"
25 #include "compiler/nir/nir_builder.h"
26
27 static void
split_conversion(nir_builder * b,nir_alu_instr * alu,nir_alu_type src_type,nir_alu_type tmp_type,nir_alu_type dst_type)28 split_conversion(nir_builder *b, nir_alu_instr *alu, nir_alu_type src_type,
29 nir_alu_type tmp_type, nir_alu_type dst_type)
30 {
31 b->cursor = nir_before_instr(&alu->instr);
32 nir_def *src = nir_ssa_for_alu_src(b, alu, 0);
33 nir_def *tmp = nir_type_convert(b, src, src_type, tmp_type, nir_rounding_mode_undef);
34 nir_def *res = nir_type_convert(b, tmp, tmp_type, dst_type, nir_rounding_mode_undef);
35 nir_def_rewrite_uses(&alu->def, res);
36 nir_instr_remove(&alu->instr);
37 }
38
39 static bool
lower_alu_instr(nir_builder * b,nir_alu_instr * alu)40 lower_alu_instr(nir_builder *b, nir_alu_instr *alu)
41 {
42 unsigned src_bit_size = nir_src_bit_size(alu->src[0].src);
43 nir_alu_type src_type = nir_op_infos[alu->op].input_types[0];
44 nir_alu_type src_full_type = (nir_alu_type) (src_type | src_bit_size);
45
46 unsigned dst_bit_size = alu->def.bit_size;
47 nir_alu_type dst_full_type = nir_op_infos[alu->op].output_type;
48 nir_alu_type dst_type = nir_alu_type_get_base_type(dst_full_type);
49
50 /* BDW PRM, vol02, Command Reference Instructions, mov - MOVE:
51 *
52 * "There is no direct conversion from HF to DF or DF to HF.
53 * Use two instructions and F (Float) as an intermediate type.
54 *
55 * There is no direct conversion from HF to Q/UQ or Q/UQ to HF.
56 * Use two instructions and F (Float) or a word integer type
57 * or a DWord integer type as an intermediate type."
58 *
59 * It is important that the intermediate conversion happens through a
60 * 32-bit float type so we don't lose range when we convert from
61 * a 64-bit integer.
62 */
63 unsigned int64_types = nir_type_int64 | nir_type_uint64;
64 if ((src_full_type == nir_type_float16 && (dst_full_type & int64_types)) ||
65 ((src_full_type & int64_types) && dst_full_type == nir_type_float16)) {
66 split_conversion(b, alu, src_type, nir_type_float | 32,
67 dst_type | dst_bit_size);
68 return true;
69 }
70
71 /* SKL PRM, vol 02a, Command Reference: Instructions, Move:
72 *
73 * "There is no direct conversion from B/UB to DF or DF to B/UB. Use
74 * two instructions and a word or DWord intermediate type."
75 *
76 * "There is no direct conversion from B/UB to Q/UQ or Q/UQ to B/UB.
77 * Use two instructions and a word or DWord intermediate integer
78 * type."
79 *
80 * It is important that we use a 32-bit integer matching the sign of the
81 * destination as the intermediate type so we avoid any chance of rtne
82 * rounding happening before the conversion to integer (which is expected
83 * to round towards zero) in double to byte conversions.
84 */
85 if ((src_bit_size == 8 && dst_bit_size == 64) ||
86 (src_bit_size == 64 && dst_bit_size == 8)) {
87 split_conversion(b, alu, src_type, dst_type | 32, dst_type | dst_bit_size);
88 return true;
89 }
90
91 return false;
92 }
93
94 static bool
lower_instr(nir_builder * b,nir_instr * instr,UNUSED void * cb_data)95 lower_instr(nir_builder *b, nir_instr *instr, UNUSED void *cb_data)
96 {
97 if (instr->type != nir_instr_type_alu)
98 return false;
99
100 nir_alu_instr *alu = nir_instr_as_alu(instr);
101
102 if (!nir_op_infos[alu->op].is_conversion)
103 return false;
104
105 return lower_alu_instr(b, alu);
106 }
107
108 bool
intel_nir_lower_conversions(nir_shader * shader)109 intel_nir_lower_conversions(nir_shader *shader)
110 {
111 return nir_shader_instructions_pass(shader, lower_instr,
112 nir_metadata_block_index |
113 nir_metadata_dominance,
114 NULL);
115 }
116