broadcom/vc5: Add compiler support for V3D 4.x texturing.
[mesa.git] / src / broadcom / compiler / v3d40_tex.c
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_v41_pack.h"
32
33 static void
34 vir_TMU_WRITE(struct v3d_compile *c, enum v3d_qpu_waddr waddr, struct qreg val)
35 {
36 vir_MOV_dest(c, vir_reg(QFILE_MAGIC, waddr), val);
37 }
38
39 static void
40 vir_WRTMUC(struct v3d_compile *c, enum quniform_contents contents, uint32_t data)
41 {
42 struct qinst *inst = vir_NOP(c);
43 inst->qpu.sig.wrtmuc = true;
44 inst->has_implicit_uniform = true;
45 inst->src[0] = vir_uniform(c, contents, data);
46 }
47
48 void
49 v3d40_vir_emit_tex(struct v3d_compile *c, nir_tex_instr *instr)
50 {
51 unsigned unit = instr->texture_index;
52
53 struct V3D41_TMU_CONFIG_PARAMETER_0 p0_unpacked = {
54 };
55
56 struct V3D41_TMU_CONFIG_PARAMETER_1 p1_unpacked = {
57 .output_type_32_bit = (c->key->tex[unit].return_size == 32 &&
58 !instr->is_shadow),
59
60 .unnormalized_coordinates = (instr->sampler_dim ==
61 GLSL_SAMPLER_DIM_RECT),
62 };
63
64 struct V3D41_TMU_CONFIG_PARAMETER_2 p2_unpacked = {
65 .op = V3D_TMU_OP_REGULAR,
66
67 .gather_mode = instr->op == nir_texop_tg4,
68 .gather_component = instr->component,
69
70 .coefficient_mode = instr->op == nir_texop_txd,
71 };
72
73 int non_array_components = instr->coord_components - instr->is_array;
74 struct qreg s;
75
76 for (unsigned i = 0; i < instr->num_srcs; i++) {
77 switch (instr->src[i].src_type) {
78 case nir_tex_src_coord:
79 /* S triggers the lookup, so save it for the end. */
80 s = ntq_get_src(c, instr->src[i].src, 0);
81
82 if (non_array_components > 1) {
83 vir_TMU_WRITE(c, V3D_QPU_WADDR_TMUT,
84 ntq_get_src(c, instr->src[i].src,
85 1));
86 }
87 if (non_array_components > 2) {
88 vir_TMU_WRITE(c, V3D_QPU_WADDR_TMUR,
89 ntq_get_src(c, instr->src[i].src,
90 2));
91 }
92
93 if (instr->is_array) {
94 vir_TMU_WRITE(c, V3D_QPU_WADDR_TMUI,
95 ntq_get_src(c, instr->src[i].src,
96 instr->coord_components - 1));
97 }
98 break;
99
100 case nir_tex_src_bias:
101 vir_TMU_WRITE(c, V3D_QPU_WADDR_TMUB,
102 ntq_get_src(c, instr->src[i].src, 0));
103 break;
104
105 case nir_tex_src_lod:
106 vir_TMU_WRITE(c, V3D_QPU_WADDR_TMUB,
107 ntq_get_src(c, instr->src[i].src, 0));
108
109 if (instr->op != nir_texop_txf &&
110 instr->op != nir_texop_tg4) {
111 p2_unpacked.disable_autolod = true;
112 }
113 break;
114
115 case nir_tex_src_comparator:
116 vir_TMU_WRITE(c, V3D_QPU_WADDR_TMUDREF,
117 ntq_get_src(c, instr->src[i].src, 0));
118 break;
119
120 case nir_tex_src_offset: {
121 nir_const_value *offset =
122 nir_src_as_const_value(instr->src[i].src);
123
124 p2_unpacked.offset_s = offset->i32[0];
125 if (instr->coord_components >= 2)
126 p2_unpacked.offset_t = offset->i32[1];
127 if (instr->coord_components >= 3)
128 p2_unpacked.offset_r = offset->i32[2];
129 break;
130 }
131
132 default:
133 unreachable("unknown texture source");
134 }
135 }
136
137 /* Limit the number of channels returned to both how many the NIR
138 * instruction writes and how many the instruction could produce.
139 */
140 uint32_t instr_return_channels = nir_tex_instr_dest_size(instr);
141 if (!p1_unpacked.output_type_32_bit)
142 instr_return_channels = (instr_return_channels + 1) / 2;
143
144 p0_unpacked.return_words_of_texture_data =
145 (1 << MIN2(instr_return_channels,
146 c->key->tex[unit].return_channels)) - 1;
147
148 uint32_t p0_packed;
149 V3D41_TMU_CONFIG_PARAMETER_0_pack(NULL,
150 (uint8_t *)&p0_packed,
151 &p0_unpacked);
152
153 uint32_t p1_packed;
154 V3D41_TMU_CONFIG_PARAMETER_1_pack(NULL,
155 (uint8_t *)&p1_packed,
156 &p1_unpacked);
157
158 uint32_t p2_packed;
159 V3D41_TMU_CONFIG_PARAMETER_2_pack(NULL,
160 (uint8_t *)&p2_packed,
161 &p2_unpacked);
162
163 /* Load unit number into the high bits of the texture or sampler
164 * address field, which will be be used by the driver to decide which
165 * texture to put in the actual address field.
166 */
167 p0_packed |= unit << 24;
168 p1_packed |= unit << 24;
169
170 vir_WRTMUC(c, QUNIFORM_TMU_CONFIG_P0, p0_packed);
171 vir_WRTMUC(c, QUNIFORM_TMU_CONFIG_P1, p1_packed);
172 vir_WRTMUC(c, QUNIFORM_CONSTANT, p2_packed);
173
174 if (instr->op == nir_texop_txf) {
175 assert(instr->sampler_dim != GLSL_SAMPLER_DIM_CUBE);
176 vir_TMU_WRITE(c, V3D_QPU_WADDR_TMUSF, s);
177 } else if (instr->sampler_dim == GLSL_SAMPLER_DIM_CUBE) {
178 vir_TMU_WRITE(c, V3D_QPU_WADDR_TMUSCM, s);
179 } else {
180 vir_TMU_WRITE(c, V3D_QPU_WADDR_TMUS, s);
181 }
182
183 vir_emit_thrsw(c);
184
185 struct qreg return_values[4];
186 for (int i = 0; i < 4; i++) {
187 /* Swizzling .zw of an RG texture should give undefined
188 * results, not crash the compiler.
189 */
190 if (p0_unpacked.return_words_of_texture_data & (1 << i))
191 return_values[i] = vir_LDTMU(c);
192 else
193 return_values[i] = c->undef;
194 }
195
196 for (int i = 0; i < nir_tex_instr_dest_size(instr); i++) {
197 struct qreg chan;
198
199 if (!p1_unpacked.output_type_32_bit) {
200 STATIC_ASSERT(PIPE_SWIZZLE_X == 0);
201 chan = return_values[i / 2];
202
203 if (nir_alu_type_get_base_type(instr->dest_type) ==
204 nir_type_float) {
205 enum v3d_qpu_input_unpack unpack;
206 if (i & 1)
207 unpack = V3D_QPU_UNPACK_H;
208 else
209 unpack = V3D_QPU_UNPACK_L;
210
211 chan = vir_FMOV(c, chan);
212 vir_set_unpack(c->defs[chan.index], 0, unpack);
213 } else {
214 /* If we're unpacking the low field, shift it
215 * up to the top first.
216 */
217 if ((i & 1) == 0) {
218 chan = vir_SHL(c, chan,
219 vir_uniform_ui(c, 16));
220 }
221
222 /* Do proper sign extension to a 32-bit int. */
223 if (nir_alu_type_get_base_type(instr->dest_type) ==
224 nir_type_int) {
225 chan = vir_ASR(c, chan,
226 vir_uniform_ui(c, 16));
227 } else {
228 chan = vir_SHR(c, chan,
229 vir_uniform_ui(c, 16));
230 }
231 }
232 } else {
233 chan = vir_MOV(c, return_values[i]);
234 }
235 ntq_store_dest(c, &instr->dest, i, chan);
236 }
237 }