v3d33_tex.c revision 01e04c3f 
1/*
2 * Copyright © 2016-2018 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 "v3d_compiler.h"
25
26/* We don't do any address packing. */
27#define __gen_user_data void
28#define __gen_address_type uint32_t
29#define __gen_address_offset(reloc) (*reloc)
30#define __gen_emit_reloc(cl, reloc)
31#include "cle/v3d_packet_v33_pack.h"
32
33void
34v3d33_vir_emit_tex(struct v3d_compile *c, nir_tex_instr *instr)
35{
36        unsigned unit = instr->texture_index;
37
38        struct V3D33_TEXTURE_UNIFORM_PARAMETER_0_CFG_MODE1 p0_unpacked = {
39                V3D33_TEXTURE_UNIFORM_PARAMETER_0_CFG_MODE1_header,
40
41                .fetch_sample_mode = instr->op == nir_texop_txf,
42        };
43
44        struct V3D33_TEXTURE_UNIFORM_PARAMETER_1_CFG_MODE1 p1_unpacked = {
45        };
46
47        switch (instr->sampler_dim) {
48        case GLSL_SAMPLER_DIM_1D:
49                if (instr->is_array)
50                        p0_unpacked.lookup_type = TEXTURE_1D_ARRAY;
51                else
52                        p0_unpacked.lookup_type = TEXTURE_1D;
53                break;
54        case GLSL_SAMPLER_DIM_2D:
55        case GLSL_SAMPLER_DIM_RECT:
56                if (instr->is_array)
57                        p0_unpacked.lookup_type = TEXTURE_2D_ARRAY;
58                else
59                        p0_unpacked.lookup_type = TEXTURE_2D;
60                break;
61        case GLSL_SAMPLER_DIM_3D:
62                p0_unpacked.lookup_type = TEXTURE_3D;
63                break;
64        case GLSL_SAMPLER_DIM_CUBE:
65                p0_unpacked.lookup_type = TEXTURE_CUBE_MAP;
66                break;
67        default:
68                unreachable("Bad sampler type");
69        }
70
71        struct qreg coords[5];
72        int next_coord = 0;
73        for (unsigned i = 0; i < instr->num_srcs; i++) {
74                switch (instr->src[i].src_type) {
75                case nir_tex_src_coord:
76                        for (int j = 0; j < instr->coord_components; j++) {
77                                coords[next_coord++] =
78                                        ntq_get_src(c, instr->src[i].src, j);
79                        }
80                        if (instr->coord_components < 2)
81                                coords[next_coord++] = vir_uniform_f(c, 0.5);
82                        break;
83                case nir_tex_src_bias:
84                        coords[next_coord++] =
85                                ntq_get_src(c, instr->src[i].src, 0);
86
87                        p0_unpacked.bias_supplied = true;
88                        break;
89                case nir_tex_src_lod:
90                        coords[next_coord++] =
91                                vir_FADD(c,
92                                         ntq_get_src(c, instr->src[i].src, 0),
93                                         vir_uniform(c, QUNIFORM_TEXTURE_FIRST_LEVEL,
94                                                     unit));
95
96                        if (instr->op != nir_texop_txf &&
97                            instr->op != nir_texop_tg4) {
98                                p0_unpacked.disable_autolod_use_bias_only = true;
99                        }
100                        break;
101                case nir_tex_src_comparator:
102                        coords[next_coord++] =
103                                ntq_get_src(c, instr->src[i].src, 0);
104
105                        p0_unpacked.shadow = true;
106                        break;
107
108                case nir_tex_src_offset: {
109                        nir_const_value *offset =
110                                nir_src_as_const_value(instr->src[i].src);
111                        p0_unpacked.texel_offset_for_s_coordinate =
112                                offset->i32[0];
113
114                        if (instr->coord_components >= 2)
115                                p0_unpacked.texel_offset_for_t_coordinate =
116                                        offset->i32[1];
117
118                        if (instr->coord_components >= 3)
119                                p0_unpacked.texel_offset_for_r_coordinate =
120                                        offset->i32[2];
121                        break;
122                }
123
124                default:
125                        unreachable("unknown texture source");
126                }
127        }
128
129        bool return_16 = (c->key->tex[unit].return_size == 16 ||
130                          p0_unpacked.shadow);
131
132        /* Limit the number of channels returned to both how many the NIR
133         * instruction writes and how many the instruction could produce.
134         */
135        uint32_t instr_return_channels = nir_tex_instr_dest_size(instr);
136        if (return_16)
137                instr_return_channels = (instr_return_channels + 1) / 2;
138
139        p1_unpacked.return_words_of_texture_data =
140                (1 << MIN2(instr_return_channels,
141                           c->key->tex[unit].return_channels)) - 1;
142
143        uint32_t p0_packed;
144        V3D33_TEXTURE_UNIFORM_PARAMETER_0_CFG_MODE1_pack(NULL,
145                                                         (uint8_t *)&p0_packed,
146                                                         &p0_unpacked);
147
148        uint32_t p1_packed;
149        V3D33_TEXTURE_UNIFORM_PARAMETER_1_CFG_MODE1_pack(NULL,
150                                                         (uint8_t *)&p1_packed,
151                                                         &p1_unpacked);
152        /* Load unit number into the address field, which will be be used by
153         * the driver to decide which texture to put in the actual address
154         * field.
155         */
156        p1_packed |= unit << 5;
157
158        /* There is no native support for GL texture rectangle coordinates, so
159         * we have to rescale from ([0, width], [0, height]) to ([0, 1], [0,
160         * 1]).
161         */
162        if (instr->sampler_dim == GLSL_SAMPLER_DIM_RECT) {
163                coords[0] = vir_FMUL(c, coords[0],
164                                     vir_uniform(c, QUNIFORM_TEXRECT_SCALE_X,
165                                                 unit));
166                coords[1] = vir_FMUL(c, coords[1],
167                                     vir_uniform(c, QUNIFORM_TEXRECT_SCALE_Y,
168                                                 unit));
169        }
170
171        struct qreg texture_u[] = {
172                vir_uniform(c, QUNIFORM_TEXTURE_CONFIG_P0_0 + unit, p0_packed),
173                vir_uniform(c, QUNIFORM_TEXTURE_CONFIG_P1, p1_packed),
174        };
175        uint32_t next_texture_u = 0;
176
177        for (int i = 0; i < next_coord; i++) {
178                struct qreg dst;
179
180                if (i == next_coord - 1)
181                        dst = vir_reg(QFILE_MAGIC, V3D_QPU_WADDR_TMUL);
182                else
183                        dst = vir_reg(QFILE_MAGIC, V3D_QPU_WADDR_TMU);
184
185                struct qinst *tmu = vir_MOV_dest(c, dst, coords[i]);
186
187                if (i < 2) {
188                        tmu->has_implicit_uniform = true;
189                        tmu->src[vir_get_implicit_uniform_src(tmu)] =
190                                texture_u[next_texture_u++];
191                }
192        }
193
194        vir_emit_thrsw(c);
195
196        struct qreg return_values[4];
197        for (int i = 0; i < 4; i++) {
198                /* Swizzling .zw of an RG texture should give undefined
199                 * results, not crash the compiler.
200                 */
201                if (p1_unpacked.return_words_of_texture_data & (1 << i))
202                        return_values[i] = vir_LDTMU(c);
203                else
204                        return_values[i] = c->undef;
205        }
206
207        for (int i = 0; i < nir_tex_instr_dest_size(instr); i++) {
208                struct qreg chan;
209
210                if (return_16) {
211                        STATIC_ASSERT(PIPE_SWIZZLE_X == 0);
212                        chan = return_values[i / 2];
213
214                        if (nir_alu_type_get_base_type(instr->dest_type) ==
215                            nir_type_float) {
216                                enum v3d_qpu_input_unpack unpack;
217                                if (i & 1)
218                                        unpack = V3D_QPU_UNPACK_H;
219                                else
220                                        unpack = V3D_QPU_UNPACK_L;
221
222                                chan = vir_FMOV(c, chan);
223                                vir_set_unpack(c->defs[chan.index], 0, unpack);
224                        } else {
225                                /* If we're unpacking the low field, shift it
226                                 * up to the top first.
227                                 */
228                                if ((i & 1) == 0) {
229                                        chan = vir_SHL(c, chan,
230                                                       vir_uniform_ui(c, 16));
231                                }
232
233                                /* Do proper sign extension to a 32-bit int. */
234                                if (nir_alu_type_get_base_type(instr->dest_type) ==
235                                    nir_type_int) {
236                                        chan = vir_ASR(c, chan,
237                                                       vir_uniform_ui(c, 16));
238                                } else {
239                                        chan = vir_SHR(c, chan,
240                                                       vir_uniform_ui(c, 16));
241                                }
242                        }
243                } else {
244                        chan = vir_MOV(c, return_values[i]);
245                }
246                ntq_store_dest(c, &instr->dest, i, chan);
247        }
248}
249