• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright © 2014-2015 Broadcom
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 "nir.h"
25 #include "nir_builder.h"
26 
27 /** @file nir_lower_alu_to_scalar.c
28  *
29  * Replaces nir_alu_instr operations with more than one channel used in the
30  * arguments with individual per-channel operations.
31  */
32 
33 static void
nir_alu_ssa_dest_init(nir_alu_instr * instr,unsigned num_components,unsigned bit_size)34 nir_alu_ssa_dest_init(nir_alu_instr *instr, unsigned num_components,
35                       unsigned bit_size)
36 {
37    nir_ssa_dest_init(&instr->instr, &instr->dest.dest, num_components,
38                      bit_size, NULL);
39    instr->dest.write_mask = (1 << num_components) - 1;
40 }
41 
42 static void
lower_reduction(nir_alu_instr * instr,nir_op chan_op,nir_op merge_op,nir_builder * builder)43 lower_reduction(nir_alu_instr *instr, nir_op chan_op, nir_op merge_op,
44                 nir_builder *builder)
45 {
46    unsigned num_components = nir_op_infos[instr->op].input_sizes[0];
47 
48    nir_ssa_def *last = NULL;
49    for (unsigned i = 0; i < num_components; i++) {
50       nir_alu_instr *chan = nir_alu_instr_create(builder->shader, chan_op);
51       nir_alu_ssa_dest_init(chan, 1, instr->dest.dest.ssa.bit_size);
52       nir_alu_src_copy(&chan->src[0], &instr->src[0], chan);
53       chan->src[0].swizzle[0] = chan->src[0].swizzle[i];
54       if (nir_op_infos[chan_op].num_inputs > 1) {
55          assert(nir_op_infos[chan_op].num_inputs == 2);
56          nir_alu_src_copy(&chan->src[1], &instr->src[1], chan);
57          chan->src[1].swizzle[0] = chan->src[1].swizzle[i];
58       }
59       chan->exact = instr->exact;
60 
61       nir_builder_instr_insert(builder, &chan->instr);
62 
63       if (i == 0) {
64          last = &chan->dest.dest.ssa;
65       } else {
66          last = nir_build_alu(builder, merge_op,
67                               last, &chan->dest.dest.ssa, NULL, NULL);
68       }
69    }
70 
71    assert(instr->dest.write_mask == 1);
72    nir_ssa_def_rewrite_uses(&instr->dest.dest.ssa, nir_src_for_ssa(last));
73    nir_instr_remove(&instr->instr);
74 }
75 
76 static bool
lower_alu_instr_scalar(nir_alu_instr * instr,nir_builder * b)77 lower_alu_instr_scalar(nir_alu_instr *instr, nir_builder *b)
78 {
79    unsigned num_src = nir_op_infos[instr->op].num_inputs;
80    unsigned i, chan;
81 
82    assert(instr->dest.dest.is_ssa);
83    assert(instr->dest.write_mask != 0);
84 
85    b->cursor = nir_before_instr(&instr->instr);
86    b->exact = instr->exact;
87 
88 #define LOWER_REDUCTION(name, chan, merge) \
89    case name##2: \
90    case name##3: \
91    case name##4: \
92       lower_reduction(instr, chan, merge, b); \
93       return true;
94 
95    switch (instr->op) {
96    case nir_op_vec4:
97    case nir_op_vec3:
98    case nir_op_vec2:
99       /* We don't need to scalarize these ops, they're the ones generated to
100        * group up outputs into a value that can be SSAed.
101        */
102       return false;
103 
104    case nir_op_pack_half_2x16:
105       if (!b->shader->options->lower_pack_half_2x16)
106          return false;
107 
108       nir_ssa_def *val =
109          nir_pack_half_2x16_split(b, nir_channel(b, instr->src[0].src.ssa,
110                                                  instr->src[0].swizzle[0]),
111                                      nir_channel(b, instr->src[0].src.ssa,
112                                                  instr->src[0].swizzle[1]));
113 
114       nir_ssa_def_rewrite_uses(&instr->dest.dest.ssa, nir_src_for_ssa(val));
115       nir_instr_remove(&instr->instr);
116       return true;
117 
118    case nir_op_unpack_unorm_4x8:
119    case nir_op_unpack_snorm_4x8:
120    case nir_op_unpack_unorm_2x16:
121    case nir_op_unpack_snorm_2x16:
122       /* There is no scalar version of these ops, unless we were to break it
123        * down to bitshifts and math (which is definitely not intended).
124        */
125       return false;
126 
127    case nir_op_unpack_half_2x16: {
128       if (!b->shader->options->lower_unpack_half_2x16)
129          return false;
130 
131       nir_ssa_def *comps[2];
132       comps[0] = nir_unpack_half_2x16_split_x(b, instr->src[0].src.ssa);
133       comps[1] = nir_unpack_half_2x16_split_y(b, instr->src[0].src.ssa);
134       nir_ssa_def *vec = nir_vec(b, comps, 2);
135 
136       nir_ssa_def_rewrite_uses(&instr->dest.dest.ssa, nir_src_for_ssa(vec));
137       nir_instr_remove(&instr->instr);
138       return true;
139    }
140 
141    case nir_op_pack_uvec2_to_uint: {
142       assert(b->shader->options->lower_pack_snorm_2x16 ||
143              b->shader->options->lower_pack_unorm_2x16);
144 
145       nir_ssa_def *word =
146          nir_extract_u16(b, instr->src[0].src.ssa, nir_imm_int(b, 0));
147       nir_ssa_def *val =
148          nir_ior(b, nir_ishl(b, nir_channel(b, word, 1), nir_imm_int(b, 16)),
149                                 nir_channel(b, word, 0));
150 
151       nir_ssa_def_rewrite_uses(&instr->dest.dest.ssa, nir_src_for_ssa(val));
152       nir_instr_remove(&instr->instr);
153       break;
154    }
155 
156    case nir_op_pack_uvec4_to_uint: {
157       assert(b->shader->options->lower_pack_snorm_4x8 ||
158              b->shader->options->lower_pack_unorm_4x8);
159 
160       nir_ssa_def *byte =
161          nir_extract_u8(b, instr->src[0].src.ssa, nir_imm_int(b, 0));
162       nir_ssa_def *val =
163          nir_ior(b, nir_ior(b, nir_ishl(b, nir_channel(b, byte, 3), nir_imm_int(b, 24)),
164                                nir_ishl(b, nir_channel(b, byte, 2), nir_imm_int(b, 16))),
165                     nir_ior(b, nir_ishl(b, nir_channel(b, byte, 1), nir_imm_int(b, 8)),
166                                nir_channel(b, byte, 0)));
167 
168       nir_ssa_def_rewrite_uses(&instr->dest.dest.ssa, nir_src_for_ssa(val));
169       nir_instr_remove(&instr->instr);
170       break;
171    }
172 
173    case nir_op_fdph: {
174       nir_ssa_def *sum[4];
175       for (unsigned i = 0; i < 3; i++) {
176          sum[i] = nir_fmul(b, nir_channel(b, instr->src[0].src.ssa,
177                                           instr->src[0].swizzle[i]),
178                               nir_channel(b, instr->src[1].src.ssa,
179                                           instr->src[1].swizzle[i]));
180       }
181       sum[3] = nir_channel(b, instr->src[1].src.ssa, instr->src[1].swizzle[3]);
182 
183       nir_ssa_def *val = nir_fadd(b, nir_fadd(b, sum[0], sum[1]),
184                                      nir_fadd(b, sum[2], sum[3]));
185 
186       nir_ssa_def_rewrite_uses(&instr->dest.dest.ssa, nir_src_for_ssa(val));
187       nir_instr_remove(&instr->instr);
188       return true;
189    }
190 
191    case nir_op_unpack_64_2x32:
192       return false;
193 
194       LOWER_REDUCTION(nir_op_fdot, nir_op_fmul, nir_op_fadd);
195       LOWER_REDUCTION(nir_op_ball_fequal, nir_op_feq, nir_op_iand);
196       LOWER_REDUCTION(nir_op_ball_iequal, nir_op_ieq, nir_op_iand);
197       LOWER_REDUCTION(nir_op_bany_fnequal, nir_op_fne, nir_op_ior);
198       LOWER_REDUCTION(nir_op_bany_inequal, nir_op_ine, nir_op_ior);
199       LOWER_REDUCTION(nir_op_fall_equal, nir_op_seq, nir_op_fand);
200       LOWER_REDUCTION(nir_op_fany_nequal, nir_op_sne, nir_op_for);
201 
202    default:
203       break;
204    }
205 
206    if (instr->dest.dest.ssa.num_components == 1)
207       return false;
208 
209    unsigned num_components = instr->dest.dest.ssa.num_components;
210    nir_ssa_def *comps[] = { NULL, NULL, NULL, NULL };
211 
212    for (chan = 0; chan < 4; chan++) {
213       if (!(instr->dest.write_mask & (1 << chan)))
214          continue;
215 
216       nir_alu_instr *lower = nir_alu_instr_create(b->shader, instr->op);
217       for (i = 0; i < num_src; i++) {
218          /* We only handle same-size-as-dest (input_sizes[] == 0) or scalar
219           * args (input_sizes[] == 1).
220           */
221          assert(nir_op_infos[instr->op].input_sizes[i] < 2);
222          unsigned src_chan = (nir_op_infos[instr->op].input_sizes[i] == 1 ?
223                               0 : chan);
224 
225          nir_alu_src_copy(&lower->src[i], &instr->src[i], lower);
226          for (int j = 0; j < 4; j++)
227             lower->src[i].swizzle[j] = instr->src[i].swizzle[src_chan];
228       }
229 
230       nir_alu_ssa_dest_init(lower, 1, instr->dest.dest.ssa.bit_size);
231       lower->dest.saturate = instr->dest.saturate;
232       comps[chan] = &lower->dest.dest.ssa;
233       lower->exact = instr->exact;
234 
235       nir_builder_instr_insert(b, &lower->instr);
236    }
237 
238    nir_ssa_def *vec = nir_vec(b, comps, num_components);
239 
240    nir_ssa_def_rewrite_uses(&instr->dest.dest.ssa, nir_src_for_ssa(vec));
241 
242    nir_instr_remove(&instr->instr);
243    return true;
244 }
245 
246 static bool
nir_lower_alu_to_scalar_impl(nir_function_impl * impl)247 nir_lower_alu_to_scalar_impl(nir_function_impl *impl)
248 {
249    nir_builder builder;
250    nir_builder_init(&builder, impl);
251    bool progress = false;
252 
253    nir_foreach_block(block, impl) {
254       nir_foreach_instr_safe(instr, block) {
255          if (instr->type == nir_instr_type_alu) {
256             progress = lower_alu_instr_scalar(nir_instr_as_alu(instr),
257                                               &builder) || progress;
258          }
259       }
260    }
261 
262    nir_metadata_preserve(impl, nir_metadata_block_index |
263                                nir_metadata_dominance);
264 
265    return progress;
266 }
267 
268 bool
nir_lower_alu_to_scalar(nir_shader * shader)269 nir_lower_alu_to_scalar(nir_shader *shader)
270 {
271    bool progress = false;
272 
273    nir_foreach_function(function, shader) {
274       if (function->impl)
275          progress = nir_lower_alu_to_scalar_impl(function->impl) || progress;
276    }
277 
278    return progress;
279 }
280