xref: /aosp_15_r20/external/mesa3d/src/panfrost/compiler/bi_lower_divergent_indirects.c (revision 6104692788411f58d303aa86923a9ff6ecaded22)
1 /*
2  * Copyright (C) 2021 Collabora, Ltd.
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 FROM,
20  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21  * SOFTWARE.
22  */
23 
24 #include "compiler/nir/nir_builder.h"
25 #include "compiler.h"
26 
27 /* Divergent attribute access is undefined behaviour. To avoid divergence,
28  * lower to an if-chain like:
29  *
30  *   value = 0;
31  *   if (lane == 0)
32  *      value = ld()
33  *   else if (lane == 1)
34  *      value = ld()
35  *   ...
36  *   else if (lane == MAX_LANE)
37  *      value = ld()
38  */
39 
40 static bool
bi_lower_divergent_indirects_impl(nir_builder * b,nir_intrinsic_instr * intr,void * data)41 bi_lower_divergent_indirects_impl(nir_builder *b, nir_intrinsic_instr *intr,
42                                   void *data)
43 {
44    gl_shader_stage stage = b->shader->info.stage;
45    nir_src *offset;
46 
47    /* Not all indirect access needs this workaround */
48    switch (intr->intrinsic) {
49    case nir_intrinsic_load_input:
50    case nir_intrinsic_load_interpolated_input:
51       /* Attributes and varyings */
52       offset = nir_get_io_offset_src(intr);
53       break;
54 
55    case nir_intrinsic_store_output:
56       /* Varyings only */
57       if (stage == MESA_SHADER_FRAGMENT)
58          return false;
59 
60       offset = nir_get_io_offset_src(intr);
61       break;
62 
63    case nir_intrinsic_image_texel_address:
64    case nir_intrinsic_image_load:
65    case nir_intrinsic_image_store:
66       /* Any image access */
67       offset = &intr->src[0];
68       break;
69    default:
70       return false;
71    }
72 
73    if (!nir_src_is_divergent(*offset))
74       return false;
75 
76    /* This indirect does need it */
77 
78    b->cursor = nir_before_instr(&intr->instr);
79    nir_def *lane = nir_load_subgroup_invocation(b);
80    unsigned *lanes = data;
81 
82    /* Write zero in a funny way to bypass lower_load_const_to_scalar */
83    bool has_dest = nir_intrinsic_infos[intr->intrinsic].has_dest;
84    unsigned size = has_dest ? intr->def.bit_size : 32;
85    nir_def *zero = has_dest ? nir_imm_zero(b, 1, size) : NULL;
86    nir_def *zeroes[4] = {zero, zero, zero, zero};
87    nir_def *res =
88       has_dest ? nir_vec(b, zeroes, intr->def.num_components) : NULL;
89 
90    for (unsigned i = 0; i < (*lanes); ++i) {
91       nir_push_if(b, nir_ieq_imm(b, lane, i));
92 
93       nir_instr *c = nir_instr_clone(b->shader, &intr->instr);
94       nir_intrinsic_instr *c_intr = nir_instr_as_intrinsic(c);
95       nir_builder_instr_insert(b, c);
96       nir_pop_if(b, NULL);
97 
98       if (has_dest) {
99          nir_def *c_ssa = &c_intr->def;
100          res = nir_if_phi(b, c_ssa, res);
101       }
102    }
103 
104    if (has_dest)
105       nir_def_rewrite_uses(&intr->def, res);
106 
107    nir_instr_remove(&intr->instr);
108    return true;
109 }
110 
111 bool
bi_lower_divergent_indirects(nir_shader * shader,unsigned lanes)112 bi_lower_divergent_indirects(nir_shader *shader, unsigned lanes)
113 {
114    return nir_shader_intrinsics_pass(shader, bi_lower_divergent_indirects_impl,
115                                      nir_metadata_none, &lanes);
116 }
117