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
|
/*
* Copyright © 2020 Intel Corporation
*
* 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 "anv_nir.h"
#include "nir_builder.h"
static bool
lower_ubo_load_instr(nir_builder *b, nir_intrinsic_instr *load,
UNUSED void *_data)
{
if (load->intrinsic != nir_intrinsic_load_global_constant_offset &&
load->intrinsic != nir_intrinsic_load_global_constant_bounded)
return false;
b->cursor = nir_before_instr(&load->instr);
nir_def *base_addr = load->src[0].ssa;
nir_def *bound = NULL;
if (load->intrinsic == nir_intrinsic_load_global_constant_bounded)
bound = load->src[2].ssa;
unsigned bit_size = load->def.bit_size;
assert(bit_size >= 8 && bit_size % 8 == 0);
unsigned byte_size = bit_size / 8;
nir_def *val;
if (!nir_src_is_divergent(&load->src[0]) && nir_src_is_const(load->src[1])) {
uint32_t offset = nir_src_as_uint(load->src[1]);
/* Things should be component-aligned. */
assert(offset % byte_size == 0);
assert(ANV_UBO_ALIGNMENT == 64);
unsigned suboffset = offset % 64;
unsigned aligned_offset = offset - suboffset;
/* Load two just in case we go over a 64B boundary */
nir_def *data[2];
for (unsigned i = 0; i < 2; i++) {
nir_def *addr = nir_iadd_imm(b, base_addr, aligned_offset + i * 64);
data[i] = nir_load_global_constant_uniform_block_intel(
b, 16, 32, addr,
.access = nir_intrinsic_access(load),
.align_mul = 64);
}
if (bound) {
nir_def* offsets =
nir_imm_uvec8(b, aligned_offset, aligned_offset + 16,
aligned_offset + 32, aligned_offset + 48,
aligned_offset + 64, aligned_offset + 80,
aligned_offset + 96, aligned_offset + 112);
nir_def* mask =
nir_bcsel(b, nir_ilt(b, offsets, bound),
nir_imm_int(b, 0xFFFFFFFF),
nir_imm_int(b, 0x00000000));
for (unsigned i = 0; i < 2; i++) {
/* We prepared a mask where every 1 bit of mask covers 4 bits of the
* UBO block we've loaded, when we apply it we'll sign extend each
* byte of the mask to a dword to get the final bitfield, this can
* be optimized because Intel HW allows instructions to mix several
* types and perform the sign extensions implicitly.
*/
data[i] =
nir_iand(b,
nir_i2iN(b, nir_extract_bits(b, &mask, 1, i * 128, 16, 8), 32),
data[i]);
}
}
val = nir_extract_bits(b, data, 2, suboffset * 8,
load->num_components, bit_size);
} else {
nir_def *offset = load->src[1].ssa;
nir_def *addr = nir_iadd(b, base_addr, nir_u2u64(b, offset));
if (bound) {
nir_def *zero = nir_imm_zero(b, load->num_components, bit_size);
unsigned load_size = byte_size * load->num_components;
nir_def *in_bounds =
nir_ilt(b, nir_iadd_imm(b, offset, load_size - 1), bound);
nir_push_if(b, in_bounds);
nir_def *load_val =
nir_load_global_constant(b, load->def.num_components,
load->def.bit_size, addr,
.access = nir_intrinsic_access(load),
.align_mul = nir_intrinsic_align_mul(load),
.align_offset = nir_intrinsic_align_offset(load));
nir_pop_if(b, NULL);
val = nir_if_phi(b, load_val, zero);
} else {
val = nir_load_global_constant(b, load->def.num_components,
load->def.bit_size, addr,
.access = nir_intrinsic_access(load),
.align_mul = nir_intrinsic_align_mul(load),
.align_offset = nir_intrinsic_align_offset(load));
}
}
nir_def_replace(&load->def, val);
return true;
}
bool
anv_nir_lower_ubo_loads(nir_shader *shader)
{
return nir_shader_intrinsics_pass(shader, lower_ubo_load_instr,
nir_metadata_none,
NULL);
}
|
