1 /*
2 * Copyright (c) 2019 Zodiac Inflight Innovations
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, sub license,
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
12 * next paragraph) shall be included in all copies or substantial portions
13 * of the 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 NON-INFRINGEMENT. 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
21 * DEALINGS IN THE SOFTWARE.
22 *
23 * Authors:
24 * Jonathan Marek <[email protected]>
25 */
26
27 #include "etnaviv_nir.h"
28
29 /* io related lowering
30 * run after lower_int_to_float because it adds i2f/f2i ops
31 */
32 void
etna_lower_io(nir_shader * shader,struct etna_shader_variant * v)33 etna_lower_io(nir_shader *shader, struct etna_shader_variant *v)
34 {
35 nir_foreach_function_impl(impl, shader) {
36 nir_builder b = nir_builder_create(impl);
37
38 nir_foreach_block(block, impl) {
39 nir_foreach_instr_safe(instr, block) {
40 if (instr->type == nir_instr_type_intrinsic) {
41 nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr);
42
43 switch (intr->intrinsic) {
44 case nir_intrinsic_load_front_face: {
45 /* HW front_face is 0.0/1.0, not 0/~0u for bool
46 * lower with a comparison with 0
47 */
48 intr->def.bit_size = 32;
49
50 b.cursor = nir_after_instr(instr);
51
52 nir_def *ssa = nir_ine_imm(&b, &intr->def, 0);
53 if (v->key.front_ccw)
54 nir_instr_as_alu(ssa->parent_instr)->op = nir_op_ieq;
55
56 nir_def_rewrite_uses_after(&intr->def,
57 ssa,
58 ssa->parent_instr);
59 } break;
60 case nir_intrinsic_store_deref: {
61 nir_deref_instr *deref = nir_src_as_deref(intr->src[0]);
62 if (shader->info.stage != MESA_SHADER_FRAGMENT || !v->key.frag_rb_swap)
63 break;
64
65 assert(deref->deref_type == nir_deref_type_var);
66
67 if (deref->var->data.location != FRAG_RESULT_COLOR &&
68 deref->var->data.location != FRAG_RESULT_DATA0)
69 break;
70
71 b.cursor = nir_before_instr(instr);
72
73 nir_def *ssa = nir_mov(&b, intr->src[1].ssa);
74 nir_alu_instr *alu = nir_instr_as_alu(ssa->parent_instr);
75 alu->src[0].swizzle[0] = 2;
76 alu->src[0].swizzle[2] = 0;
77 nir_src_rewrite(&intr->src[1], ssa);
78 } break;
79 case nir_intrinsic_load_vertex_id:
80 case nir_intrinsic_load_instance_id:
81 /* detect use of vertex_id/instance_id */
82 v->vs_id_in_reg = v->infile.num_reg;
83 break;
84 default:
85 break;
86 }
87 }
88
89 if (instr->type != nir_instr_type_tex)
90 continue;
91
92 nir_tex_instr *tex = nir_instr_as_tex(instr);
93 nir_src *coord = NULL;
94 nir_src *src1 = NULL;
95 unsigned src1_idx;
96
97 assert(tex->sampler_index == tex->texture_index);
98
99 for (unsigned i = 0; i < tex->num_srcs; i++) {
100 switch (tex->src[i].src_type) {
101 case nir_tex_src_coord:
102 coord = &tex->src[i].src;
103 break;
104 case nir_tex_src_bias:
105 case nir_tex_src_lod:
106 assert(!src1);
107 src1 = &tex->src[i].src;
108 src1_idx = i;
109 break;
110 case nir_tex_src_ddx:
111 case nir_tex_src_ddy:
112 case nir_tex_src_comparator:
113 break;
114 default:
115 assert(0);
116 break;
117 }
118 }
119
120 /* pre HALTI5 needs texture sources in a single source */
121
122 if (!src1 || v->shader->info->halti >= 5)
123 continue;
124
125 assert(coord && src1 && tex->coord_components < 4);
126
127 nir_alu_instr *vec = nir_alu_instr_create(shader, nir_op_vec4);
128 for (unsigned i = 0; i < tex->coord_components; i++) {
129 vec->src[i].src = nir_src_for_ssa(coord->ssa);
130 vec->src[i].swizzle[0] = i;
131 }
132 for (unsigned i = tex->coord_components; i < 4; i++)
133 vec->src[i].src = nir_src_for_ssa(src1->ssa);
134
135 nir_def_init(&vec->instr, &vec->def, 4, 32);
136
137 nir_tex_instr_remove_src(tex, src1_idx);
138 nir_src_rewrite(coord, &vec->def);
139 tex->coord_components = 4;
140
141 nir_instr_insert_before(&tex->instr, &vec->instr);
142 }
143 }
144 }
145 }
146
147 static void
etna_lower_alu_impl(nir_function_impl * impl,bool has_new_transcendentals)148 etna_lower_alu_impl(nir_function_impl *impl, bool has_new_transcendentals)
149 {
150 nir_shader *shader = impl->function->shader;
151
152 nir_builder b = nir_builder_create(impl);
153
154 /* in a seperate loop so we can apply the multiple-uniform logic to the new fmul */
155 nir_foreach_block(block, impl) {
156 nir_foreach_instr_safe(instr, block) {
157 if (instr->type != nir_instr_type_alu)
158 continue;
159
160 nir_alu_instr *alu = nir_instr_as_alu(instr);
161 /* multiply sin/cos src by constant
162 * TODO: do this earlier (but it breaks const_prop opt)
163 */
164 if (alu->op == nir_op_fsin || alu->op == nir_op_fcos) {
165 b.cursor = nir_before_instr(instr);
166
167 nir_def *imm = has_new_transcendentals ?
168 nir_imm_float(&b, 1.0 / M_PI) :
169 nir_imm_float(&b, 2.0 / M_PI);
170
171 nir_src_rewrite(&alu->src[0].src,
172 nir_fmul(&b, alu->src[0].src.ssa, imm));
173 }
174
175 /* change transcendental ops to vec2 and insert vec1 mul for the result
176 * TODO: do this earlier (but it breaks with optimizations)
177 */
178 if (has_new_transcendentals && (
179 alu->op == nir_op_fdiv || alu->op == nir_op_flog2 ||
180 alu->op == nir_op_fsin || alu->op == nir_op_fcos)) {
181 nir_def *ssa = &alu->def;
182
183 assert(ssa->num_components == 1);
184
185 nir_alu_instr *mul = nir_alu_instr_create(shader, nir_op_fmul);
186 mul->src[0].src = mul->src[1].src = nir_src_for_ssa(ssa);
187 mul->src[1].swizzle[0] = 1;
188
189 nir_def_init(&mul->instr, &mul->def, 1, 32);
190
191 alu->src[0].swizzle[1] = 0;
192 ssa->num_components = 2;
193
194 nir_instr_insert_after(instr, &mul->instr);
195
196 nir_def_rewrite_uses_after(ssa, &mul->def,
197 &mul->instr);
198 }
199 }
200 }
201 }
202
203 void
etna_lower_alu(nir_shader * shader,bool has_new_transcendentals)204 etna_lower_alu(nir_shader *shader, bool has_new_transcendentals)
205 {
206 nir_foreach_function_impl(impl, shader) {
207 etna_lower_alu_impl(impl, has_new_transcendentals);
208 }
209 }
210