1 /*
2 * Copyright © 2016 Intel Corporation
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 "compiler/nir/nir_builder.h"
25 #include "brw_nir.h"
26
27 /**
28 * Prior to Haswell, the hardware can't natively support GL_FIXED or
29 * 2_10_10_10_REV vertex formats. This pass inserts extra shader code
30 * to produce the correct values.
31 */
32
33 struct attr_wa_state {
34 nir_builder builder;
35 bool impl_progress;
36 const uint8_t *wa_flags;
37 };
38
39 static bool
apply_attr_wa_block(nir_block * block,struct attr_wa_state * state)40 apply_attr_wa_block(nir_block *block, struct attr_wa_state *state)
41 {
42 nir_builder *b = &state->builder;
43
44 nir_foreach_instr_safe(instr, block) {
45 if (instr->type != nir_instr_type_intrinsic)
46 continue;
47
48 nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
49 if (intrin->intrinsic != nir_intrinsic_load_input)
50 continue;
51
52 uint8_t wa_flags = state->wa_flags[intrin->const_index[0]];
53 if (wa_flags == 0)
54 continue;
55
56 b->cursor = nir_after_instr(instr);
57
58 nir_ssa_def *val = &intrin->dest.ssa;
59
60 /* Do GL_FIXED rescaling for GLES2.0. Our GL_FIXED attributes
61 * come in as floating point conversions of the integer values.
62 */
63 if (wa_flags & BRW_ATTRIB_WA_COMPONENT_MASK) {
64 nir_ssa_def *scaled =
65 nir_fmul(b, val, nir_imm_float(b, 1.0f / 65536.0f));
66 nir_ssa_def *comps[4];
67 for (int i = 0; i < val->num_components; i++) {
68 bool rescale = i < (wa_flags & BRW_ATTRIB_WA_COMPONENT_MASK);
69 comps[i] = nir_channel(b, rescale ? scaled : val, i);
70 }
71 val = nir_vec(b, comps, val->num_components);
72 }
73
74 /* Do sign recovery for 2101010 formats if required. */
75 if (wa_flags & BRW_ATTRIB_WA_SIGN) {
76 /* sign recovery shift: <22, 22, 22, 30> */
77 nir_ssa_def *shift = nir_imm_ivec4(b, 22, 22, 22, 30);
78 val = nir_ishr(b, nir_ishl(b, val, shift), shift);
79 }
80
81 /* Apply BGRA swizzle if required. */
82 if (wa_flags & BRW_ATTRIB_WA_BGRA) {
83 val = nir_swizzle(b, val, (unsigned[4]){2,1,0,3}, 4, true);
84 }
85
86 if (wa_flags & BRW_ATTRIB_WA_NORMALIZE) {
87 /* ES 3.0 has different rules for converting signed normalized
88 * fixed-point numbers than desktop GL.
89 */
90 if (wa_flags & BRW_ATTRIB_WA_SIGN) {
91 /* According to equation 2.2 of the ES 3.0 specification,
92 * signed normalization conversion is done by:
93 *
94 * f = c / (2^(b-1)-1)
95 *
96 * OpenGL 4.2+ uses this equation as well. Since most contexts
97 * promote to the new higher version, and this is what Haswell+
98 * hardware does anyway, we just always use this formula.
99 */
100 nir_ssa_def *es3_normalize_factor =
101 nir_imm_vec4(b, 1.0f / ((1 << 9) - 1), 1.0f / ((1 << 9) - 1),
102 1.0f / ((1 << 9) - 1), 1.0f / ((1 << 1) - 1));
103 val = nir_fmax(b,
104 nir_fmul(b, nir_i2f32(b, val), es3_normalize_factor),
105 nir_imm_float(b, -1.0f));
106 } else {
107 /* The following equation is from the OpenGL 3.2 specification:
108 *
109 * 2.1 unsigned normalization
110 * f = c/(2^n-1)
111 */
112 nir_ssa_def *normalize_factor =
113 nir_imm_vec4(b, 1.0f / ((1 << 10) - 1), 1.0f / ((1 << 10) - 1),
114 1.0f / ((1 << 10) - 1), 1.0f / ((1 << 2) - 1));
115
116 val = nir_fmul(b, nir_u2f32(b, val), normalize_factor);
117 }
118 }
119
120 if (wa_flags & BRW_ATTRIB_WA_SCALE) {
121 val = (wa_flags & BRW_ATTRIB_WA_SIGN) ? nir_i2f32(b, val)
122 : nir_u2f32(b, val);
123 }
124
125 nir_ssa_def_rewrite_uses_after(&intrin->dest.ssa, nir_src_for_ssa(val),
126 val->parent_instr);
127 state->impl_progress = true;
128 }
129
130 return true;
131 }
132
133 bool
brw_nir_apply_attribute_workarounds(nir_shader * shader,const uint8_t * attrib_wa_flags)134 brw_nir_apply_attribute_workarounds(nir_shader *shader,
135 const uint8_t *attrib_wa_flags)
136 {
137 bool progress = false;
138 struct attr_wa_state state = {
139 .wa_flags = attrib_wa_flags,
140 };
141
142 nir_foreach_function(func, shader) {
143 if (!func->impl)
144 continue;
145
146 nir_builder_init(&state.builder, func->impl);
147 state.impl_progress = false;
148
149 nir_foreach_block(block, func->impl) {
150 apply_attr_wa_block(block, &state);
151 }
152
153 if (state.impl_progress) {
154 nir_metadata_preserve(func->impl, nir_metadata_block_index |
155 nir_metadata_dominance);
156 progress = true;
157 }
158 }
159
160 return progress;
161 }
162