1 /*
2 * Copyright © 2018 Intel Corporation
3 * Copyright © 2019 Vasily Khoruzhick <[email protected]>
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * Software is furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
14 * Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
21 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
22 * IN THE SOFTWARE.
23 */
24
25 #include "nir.h"
26 #include "nir_builder.h"
27
28 static bool
assert_ssa_def_is_not_int(nir_def * def,void * arg)29 assert_ssa_def_is_not_int(nir_def *def, void *arg)
30 {
31 ASSERTED BITSET_WORD *int_types = arg;
32 assert(!BITSET_TEST(int_types, def->index));
33 return true;
34 }
35
36 static bool
instr_has_only_trivial_swizzles(nir_alu_instr * alu)37 instr_has_only_trivial_swizzles(nir_alu_instr *alu)
38 {
39 const nir_op_info *info = &nir_op_infos[alu->op];
40
41 for (unsigned i = 0; i < info->num_inputs; i++) {
42 for (unsigned chan = 0; chan < alu->def.num_components; chan++) {
43 if (alu->src[i].swizzle[chan] != chan)
44 return false;
45 }
46 }
47 return true;
48 }
49
50 /* Recognize the y = x - ffract(x) patterns from lowered ffloor.
51 * It only works for the simple case when no swizzling is involved.
52 */
53 static bool
check_for_lowered_ffloor(nir_alu_instr * fadd)54 check_for_lowered_ffloor(nir_alu_instr *fadd)
55 {
56 if (!instr_has_only_trivial_swizzles(fadd))
57 return false;
58
59 nir_alu_instr *fneg = NULL;
60 nir_src x;
61 for (unsigned i = 0; i < 2; i++) {
62 nir_alu_instr *fadd_src_alu = nir_src_as_alu_instr(fadd->src[i].src);
63 if (fadd_src_alu && fadd_src_alu->op == nir_op_fneg) {
64 fneg = fadd_src_alu;
65 x = fadd->src[1 - i].src;
66 }
67 }
68
69 if (!fneg || !instr_has_only_trivial_swizzles(fneg))
70 return false;
71
72 nir_alu_instr *ffract = nir_src_as_alu_instr(fneg->src[0].src);
73 if (ffract && ffract->op == nir_op_ffract &&
74 nir_srcs_equal(ffract->src[0].src, x) &&
75 instr_has_only_trivial_swizzles(ffract))
76 return true;
77
78 return false;
79 }
80
81 static bool
lower_alu_instr(nir_builder * b,nir_alu_instr * alu)82 lower_alu_instr(nir_builder *b, nir_alu_instr *alu)
83 {
84 const nir_op_info *info = &nir_op_infos[alu->op];
85
86 bool is_bool_only = alu->def.bit_size == 1;
87 for (unsigned i = 0; i < info->num_inputs; i++) {
88 if (alu->src[i].src.ssa->bit_size != 1)
89 is_bool_only = false;
90 }
91
92 if (is_bool_only) {
93 /* avoid lowering integers ops are used for booleans (ieq,ine,etc) */
94 return false;
95 }
96
97 b->cursor = nir_before_instr(&alu->instr);
98
99 /* Replacement SSA value */
100 nir_def *rep = NULL;
101 switch (alu->op) {
102 case nir_op_mov:
103 case nir_op_vec2:
104 case nir_op_vec3:
105 case nir_op_vec4:
106 case nir_op_bcsel:
107 /* These we expect to have integers but the opcode doesn't change */
108 break;
109
110 case nir_op_b2i32:
111 alu->op = nir_op_b2f32;
112 break;
113 case nir_op_i2f32:
114 alu->op = nir_op_mov;
115 break;
116 case nir_op_u2f32:
117 alu->op = nir_op_mov;
118 break;
119
120 case nir_op_f2i32: {
121 alu->op = nir_op_ftrunc;
122
123 /* If the source was already integer, then we did't need to truncate and
124 * can switch it to a mov that can be copy-propagated away.
125 */
126 nir_alu_instr *src_alu = nir_src_as_alu_instr(alu->src[0].src);
127 if (src_alu) {
128 switch (src_alu->op) {
129 /* Check for the y = x - ffract(x) patterns from lowered ffloor. */
130 case nir_op_fadd:
131 if (check_for_lowered_ffloor(src_alu))
132 alu->op = nir_op_mov;
133 break;
134 case nir_op_fround_even:
135 case nir_op_fceil:
136 case nir_op_ftrunc:
137 case nir_op_ffloor:
138 alu->op = nir_op_mov;
139 break;
140 default:
141 break;
142 }
143 }
144 break;
145 }
146
147 case nir_op_f2u32:
148 alu->op = nir_op_ffloor;
149 break;
150
151 case nir_op_ilt:
152 alu->op = nir_op_flt;
153 break;
154 case nir_op_ige:
155 alu->op = nir_op_fge;
156 break;
157 case nir_op_ieq:
158 alu->op = nir_op_feq;
159 break;
160 case nir_op_ine:
161 alu->op = nir_op_fneu;
162 break;
163 case nir_op_ult:
164 alu->op = nir_op_flt;
165 break;
166 case nir_op_uge:
167 alu->op = nir_op_fge;
168 break;
169
170 case nir_op_iadd:
171 alu->op = nir_op_fadd;
172 break;
173 case nir_op_isub:
174 alu->op = nir_op_fsub;
175 break;
176 case nir_op_imul:
177 alu->op = nir_op_fmul;
178 break;
179
180 case nir_op_idiv: {
181 nir_def *x = nir_ssa_for_alu_src(b, alu, 0);
182 nir_def *y = nir_ssa_for_alu_src(b, alu, 1);
183
184 /* Hand-lower fdiv, since lower_int_to_float is after nir_opt_algebraic. */
185 if (b->shader->options->lower_fdiv) {
186 rep = nir_ftrunc(b, nir_fmul(b, x, nir_frcp(b, y)));
187 } else {
188 rep = nir_ftrunc(b, nir_fdiv(b, x, y));
189 }
190 break;
191 }
192
193 case nir_op_iabs:
194 alu->op = nir_op_fabs;
195 break;
196 case nir_op_ineg:
197 alu->op = nir_op_fneg;
198 break;
199 case nir_op_imax:
200 alu->op = nir_op_fmax;
201 break;
202 case nir_op_imin:
203 alu->op = nir_op_fmin;
204 break;
205 case nir_op_umax:
206 alu->op = nir_op_fmax;
207 break;
208 case nir_op_umin:
209 alu->op = nir_op_fmin;
210 break;
211
212 case nir_op_ball_iequal2:
213 alu->op = nir_op_ball_fequal2;
214 break;
215 case nir_op_ball_iequal3:
216 alu->op = nir_op_ball_fequal3;
217 break;
218 case nir_op_ball_iequal4:
219 alu->op = nir_op_ball_fequal4;
220 break;
221 case nir_op_bany_inequal2:
222 alu->op = nir_op_bany_fnequal2;
223 break;
224 case nir_op_bany_inequal3:
225 alu->op = nir_op_bany_fnequal3;
226 break;
227 case nir_op_bany_inequal4:
228 alu->op = nir_op_bany_fnequal4;
229 break;
230
231 case nir_op_i32csel_gt:
232 alu->op = nir_op_fcsel_gt;
233 break;
234 case nir_op_i32csel_ge:
235 alu->op = nir_op_fcsel_ge;
236 break;
237
238 default:
239 assert(nir_alu_type_get_base_type(info->output_type) != nir_type_int &&
240 nir_alu_type_get_base_type(info->output_type) != nir_type_uint);
241 for (unsigned i = 0; i < info->num_inputs; i++) {
242 assert(nir_alu_type_get_base_type(info->input_types[i]) != nir_type_int &&
243 nir_alu_type_get_base_type(info->input_types[i]) != nir_type_uint);
244 }
245 return false;
246 }
247
248 if (rep) {
249 /* We've emitted a replacement instruction */
250 nir_def_replace(&alu->def, rep);
251 }
252
253 return true;
254 }
255
256 static bool
nir_lower_int_to_float_impl(nir_function_impl * impl)257 nir_lower_int_to_float_impl(nir_function_impl *impl)
258 {
259 bool progress = false;
260 BITSET_WORD *float_types = NULL, *int_types = NULL;
261
262 nir_builder b = nir_builder_create(impl);
263
264 nir_index_ssa_defs(impl);
265 float_types = calloc(BITSET_WORDS(impl->ssa_alloc),
266 sizeof(BITSET_WORD));
267 int_types = calloc(BITSET_WORDS(impl->ssa_alloc),
268 sizeof(BITSET_WORD));
269 nir_gather_types(impl, float_types, int_types);
270
271 nir_foreach_block(block, impl) {
272 nir_foreach_instr_safe(instr, block) {
273 switch (instr->type) {
274 case nir_instr_type_alu:
275 progress |= lower_alu_instr(&b, nir_instr_as_alu(instr));
276 break;
277
278 case nir_instr_type_load_const: {
279 nir_load_const_instr *load = nir_instr_as_load_const(instr);
280 if (load->def.bit_size != 1 && BITSET_TEST(int_types, load->def.index)) {
281 for (unsigned i = 0; i < load->def.num_components; i++)
282 load->value[i].f32 = load->value[i].i32;
283 }
284 break;
285 }
286
287 case nir_instr_type_intrinsic:
288 case nir_instr_type_undef:
289 case nir_instr_type_phi:
290 case nir_instr_type_tex:
291 break;
292
293 default:
294 nir_foreach_def(instr, assert_ssa_def_is_not_int, (void *)int_types);
295 break;
296 }
297 }
298 }
299
300 if (progress) {
301 nir_metadata_preserve(impl, nir_metadata_control_flow);
302 } else {
303 nir_metadata_preserve(impl, nir_metadata_all);
304 }
305
306 free(float_types);
307 free(int_types);
308
309 return progress;
310 }
311
312 bool
nir_lower_int_to_float(nir_shader * shader)313 nir_lower_int_to_float(nir_shader *shader)
314 {
315 bool progress = false;
316
317 nir_foreach_function_impl(impl, shader) {
318 if (nir_lower_int_to_float_impl(impl))
319 progress = true;
320 }
321
322 return progress;
323 }
324