summaryrefslogtreecommitdiff
path: root/src/gallium/drivers/vc4/vc4_nir_lower_io.c
blob: b7969a562a6bc20302a56ac5be2a7dee3ebd253f (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
/*
 * Copyright © 2015 Broadcom
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice (including the next
 * paragraph) shall be included in all copies or substantial portions of the
 * Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
 * IN THE SOFTWARE.
 */

#include "vc4_qir.h"
#include "compiler/nir/nir_builder.h"
#include "util/u_format.h"

/**
 * Walks the NIR generated by TGSI-to-NIR or GLSL-to-NIR to lower its io
 * intrinsics into something amenable to the VC4 architecture.
 *
 * Currently, it splits VS inputs and uniforms into scalars, drops any
 * non-position outputs in coordinate shaders, and fixes up the addressing on
 * indirect uniform loads.  FS input and VS output scalarization is handled by
 * nir_lower_io_to_scalar().
 */

static void
replace_intrinsic_with_vec(nir_builder *b, nir_intrinsic_instr *intr,
                           nir_ssa_def **comps)
{

        /* Batch things back together into a vector.  This will get split by
         * the later ALU scalarization pass.
         */
        nir_ssa_def *vec = nir_vec(b, comps, intr->num_components);

        /* Replace the old intrinsic with a reference to our reconstructed
         * vector.
         */
        nir_ssa_def_rewrite_uses(&intr->dest.ssa, nir_src_for_ssa(vec));
        nir_instr_remove(&intr->instr);
}

static nir_ssa_def *
vc4_nir_unpack_8i(nir_builder *b, nir_ssa_def *src, unsigned chan)
{
        return nir_ubitfield_extract(b,
                                     src,
                                     nir_imm_int(b, 8 * chan),
                                     nir_imm_int(b, 8));
}

/** Returns the 16 bit field as a sign-extended 32-bit value. */
static nir_ssa_def *
vc4_nir_unpack_16i(nir_builder *b, nir_ssa_def *src, unsigned chan)
{
        return nir_ibitfield_extract(b,
                                     src,
                                     nir_imm_int(b, 16 * chan),
                                     nir_imm_int(b, 16));
}

/** Returns the 16 bit field as an unsigned 32 bit value. */
static nir_ssa_def *
vc4_nir_unpack_16u(nir_builder *b, nir_ssa_def *src, unsigned chan)
{
        if (chan == 0) {
                return nir_iand(b, src, nir_imm_int(b, 0xffff));
        } else {
                return nir_ushr(b, src, nir_imm_int(b, 16));
        }
}

static nir_ssa_def *
vc4_nir_unpack_8f(nir_builder *b, nir_ssa_def *src, unsigned chan)
{
        return nir_channel(b, nir_unpack_unorm_4x8(b, src), chan);
}

static nir_ssa_def *
vc4_nir_get_vattr_channel_vpm(struct vc4_compile *c,
                              nir_builder *b,
                              nir_ssa_def **vpm_reads,
                              uint8_t swiz,
                              const struct util_format_description *desc)
{
        const struct util_format_channel_description *chan =
                &desc->channel[swiz];
        nir_ssa_def *temp;

        if (swiz > PIPE_SWIZZLE_W) {
                return vc4_nir_get_swizzled_channel(b, vpm_reads, swiz);
        } else if (chan->size == 32 && chan->type == UTIL_FORMAT_TYPE_FLOAT) {
                return vc4_nir_get_swizzled_channel(b, vpm_reads, swiz);
        } else if (chan->size == 32 && chan->type == UTIL_FORMAT_TYPE_SIGNED) {
                if (chan->normalized) {
                        return nir_fmul(b,
                                        nir_i2f32(b, vpm_reads[swiz]),
                                        nir_imm_float(b,
                                                      1.0 / 0x7fffffff));
                } else {
                        return nir_i2f32(b, vpm_reads[swiz]);
                }
        } else if (chan->size == 8 &&
                   (chan->type == UTIL_FORMAT_TYPE_UNSIGNED ||
                    chan->type == UTIL_FORMAT_TYPE_SIGNED)) {
                nir_ssa_def *vpm = vpm_reads[0];
                if (chan->type == UTIL_FORMAT_TYPE_SIGNED) {
                        temp = nir_ixor(b, vpm, nir_imm_int(b, 0x80808080));
                        if (chan->normalized) {
                                return nir_fsub(b, nir_fmul(b,
                                                            vc4_nir_unpack_8f(b, temp, swiz),
                                                            nir_imm_float(b, 2.0)),
                                                nir_imm_float(b, 1.0));
                        } else {
                                return nir_fadd(b,
                                                nir_i2f32(b,
                                                          vc4_nir_unpack_8i(b, temp,
                                                                            swiz)),
                                                nir_imm_float(b, -128.0));
                        }
                } else {
                        if (chan->normalized) {
                                return vc4_nir_unpack_8f(b, vpm, swiz);
                        } else {
                                return nir_i2f32(b, vc4_nir_unpack_8i(b, vpm, swiz));
                        }
                }
        } else if (chan->size == 16 &&
                   (chan->type == UTIL_FORMAT_TYPE_UNSIGNED ||
                    chan->type == UTIL_FORMAT_TYPE_SIGNED)) {
                nir_ssa_def *vpm = vpm_reads[swiz / 2];

                /* Note that UNPACK_16F eats a half float, not ints, so we use
                 * UNPACK_16_I for all of these.
                 */
                if (chan->type == UTIL_FORMAT_TYPE_SIGNED) {
                        temp = nir_i2f32(b, vc4_nir_unpack_16i(b, vpm, swiz & 1));
                        if (chan->normalized) {
                                return nir_fmul(b, temp,
                                                nir_imm_float(b, 1/32768.0f));
                        } else {
                                return temp;
                        }
                } else {
                        temp = nir_i2f32(b, vc4_nir_unpack_16u(b, vpm, swiz & 1));
                        if (chan->normalized) {
                                return nir_fmul(b, temp,
                                                nir_imm_float(b, 1 / 65535.0));
                        } else {
                                return temp;
                        }
                }
        } else {
                return NULL;
        }
}

static void
vc4_nir_lower_vertex_attr(struct vc4_compile *c, nir_builder *b,
                          nir_intrinsic_instr *intr)
{
        b->cursor = nir_before_instr(&intr->instr);

        int attr = nir_intrinsic_base(intr);
        enum pipe_format format = c->vs_key->attr_formats[attr];
        uint32_t attr_size = util_format_get_blocksize(format);

        /* We only accept direct outputs and TGSI only ever gives them to us
         * with an offset value of 0.
         */
        assert(nir_src_as_const_value(intr->src[0]) &&
               nir_src_as_const_value(intr->src[0])->u32[0] == 0);

        /* Generate dword loads for the VPM values (Since these intrinsics may
         * be reordered, the actual reads will be generated at the top of the
         * shader by ntq_setup_inputs().
         */
        nir_ssa_def *vpm_reads[4];
        for (int i = 0; i < align(attr_size, 4) / 4; i++) {
                nir_intrinsic_instr *intr_comp =
                        nir_intrinsic_instr_create(c->s,
                                                   nir_intrinsic_load_input);
                intr_comp->num_components = 1;
                nir_intrinsic_set_base(intr_comp, nir_intrinsic_base(intr));
                nir_intrinsic_set_component(intr_comp, i);
                intr_comp->src[0] = nir_src_for_ssa(nir_imm_int(b, 0));
                nir_ssa_dest_init(&intr_comp->instr, &intr_comp->dest, 1, 32, NULL);
                nir_builder_instr_insert(b, &intr_comp->instr);

                vpm_reads[i] = &intr_comp->dest.ssa;
        }

        bool format_warned = false;
        const struct util_format_description *desc =
                util_format_description(format);

        nir_ssa_def *dests[4];
        for (int i = 0; i < intr->num_components; i++) {
                uint8_t swiz = desc->swizzle[i];
                dests[i] = vc4_nir_get_vattr_channel_vpm(c, b, vpm_reads, swiz,
                                                         desc);

                if (!dests[i]) {
                        if (!format_warned) {
                                fprintf(stderr,
                                        "vtx element %d unsupported type: %s\n",
                                        attr, util_format_name(format));
                                format_warned = true;
                        }
                        dests[i] = nir_imm_float(b, 0.0);
                }
        }

        replace_intrinsic_with_vec(b, intr, dests);
}

static bool
is_point_sprite(struct vc4_compile *c, nir_variable *var)
{
        if (var->data.location < VARYING_SLOT_VAR0 ||
            var->data.location > VARYING_SLOT_VAR31)
                return false;

        return (c->fs_key->point_sprite_mask &
                (1 << (var->data.location - VARYING_SLOT_VAR0)));
}

static void
vc4_nir_lower_fs_input(struct vc4_compile *c, nir_builder *b,
                       nir_intrinsic_instr *intr)
{
        b->cursor = nir_after_instr(&intr->instr);

        if (nir_intrinsic_base(intr) >= VC4_NIR_TLB_COLOR_READ_INPUT &&
            nir_intrinsic_base(intr) < (VC4_NIR_TLB_COLOR_READ_INPUT +
                                        VC4_MAX_SAMPLES)) {
                /* This doesn't need any lowering. */
                return;
        }

        nir_variable *input_var = NULL;
        nir_foreach_variable(var, &c->s->inputs) {
                if (var->data.driver_location == nir_intrinsic_base(intr)) {
                        input_var = var;
                        break;
                }
        }
        assert(input_var);

        int comp = nir_intrinsic_component(intr);

        /* Lower away point coordinates, and fix up PNTC. */
        if (is_point_sprite(c, input_var) ||
            input_var->data.location == VARYING_SLOT_PNTC) {
                assert(intr->num_components == 1);

                nir_ssa_def *result = &intr->dest.ssa;

                switch (comp) {
                case 0:
                case 1:
                        /* If we're not rendering points, we need to set a
                         * defined value for the input that would come from
                         * PNTC.
                         */
                        if (!c->fs_key->is_points)
                                result = nir_imm_float(b, 0.0);
                        break;
                case 2:
                        result = nir_imm_float(b, 0.0);
                        break;
                case 3:
                        result = nir_imm_float(b, 1.0);
                        break;
                }

                if (c->fs_key->point_coord_upper_left && comp == 1)
                        result = nir_fsub(b, nir_imm_float(b, 1.0), result);

                if (result != &intr->dest.ssa) {
                        nir_ssa_def_rewrite_uses_after(&intr->dest.ssa,
                                                       nir_src_for_ssa(result),
                                                       result->parent_instr);
                }
        }
}

static void
vc4_nir_lower_output(struct vc4_compile *c, nir_builder *b,
                     nir_intrinsic_instr *intr)
{
        nir_variable *output_var = NULL;
        nir_foreach_variable(var, &c->s->outputs) {
                if (var->data.driver_location == nir_intrinsic_base(intr)) {
                        output_var = var;
                        break;
                }
        }
        assert(output_var);

        if (c->stage == QSTAGE_COORD &&
            output_var->data.location != VARYING_SLOT_POS &&
            output_var->data.location != VARYING_SLOT_PSIZ) {
                nir_instr_remove(&intr->instr);
                return;
        }
}

static void
vc4_nir_lower_uniform(struct vc4_compile *c, nir_builder *b,
                      nir_intrinsic_instr *intr)
{
        b->cursor = nir_before_instr(&intr->instr);

        /* Generate scalar loads equivalent to the original vector. */
        nir_ssa_def *dests[4];
        for (unsigned i = 0; i < intr->num_components; i++) {
                nir_intrinsic_instr *intr_comp =
                        nir_intrinsic_instr_create(c->s, intr->intrinsic);
                intr_comp->num_components = 1;
                nir_ssa_dest_init(&intr_comp->instr, &intr_comp->dest, 1, 32, NULL);

                /* Convert the uniform offset to bytes.  If it happens
                 * to be a constant, constant-folding will clean up
                 * the shift for us.
                 */
                nir_intrinsic_set_base(intr_comp,
                                       nir_intrinsic_base(intr) * 16 +
                                       i * 4);

                intr_comp->src[0] =
                        nir_src_for_ssa(nir_ishl(b, intr->src[0].ssa,
                                                 nir_imm_int(b, 4)));

                dests[i] = &intr_comp->dest.ssa;

                nir_builder_instr_insert(b, &intr_comp->instr);
        }

        replace_intrinsic_with_vec(b, intr, dests);
}

static void
vc4_nir_lower_io_instr(struct vc4_compile *c, nir_builder *b,
                       struct nir_instr *instr)
{
        if (instr->type != nir_instr_type_intrinsic)
                return;
        nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr);

        switch (intr->intrinsic) {
        case nir_intrinsic_load_input:
                if (c->stage == QSTAGE_FRAG)
                        vc4_nir_lower_fs_input(c, b, intr);
                else
                        vc4_nir_lower_vertex_attr(c, b, intr);
                break;

        case nir_intrinsic_store_output:
                vc4_nir_lower_output(c, b, intr);
                break;

        case nir_intrinsic_load_uniform:
                vc4_nir_lower_uniform(c, b, intr);
                break;

        case nir_intrinsic_load_user_clip_plane:
        default:
                break;
        }
}

static bool
vc4_nir_lower_io_impl(struct vc4_compile *c, nir_function_impl *impl)
{
        nir_builder b;
        nir_builder_init(&b, impl);

        nir_foreach_block(block, impl) {
                nir_foreach_instr_safe(instr, block)
                        vc4_nir_lower_io_instr(c, &b, instr);
        }

        nir_metadata_preserve(impl, nir_metadata_block_index |
                              nir_metadata_dominance);

        return true;
}

void
vc4_nir_lower_io(nir_shader *s, struct vc4_compile *c)
{
        nir_foreach_function(function, s) {
                if (function->impl)
                        vc4_nir_lower_io_impl(c, function->impl);
        }
}