1 /*
2 * Copyright © 2014 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 /**
25 * @file vc4_opt_small_immediates.c
26 *
27 * Turns references to small constant uniform values into small immediates
28 * fields.
29 */
30
31 #include "vc4_qir.h"
32 #include "vc4_qpu.h"
33
34 static bool debug;
35
36 bool
qir_opt_small_immediates(struct vc4_compile * c)37 qir_opt_small_immediates(struct vc4_compile *c)
38 {
39 bool progress = false;
40
41 qir_for_each_inst_inorder(inst, c) {
42 /* The small immediate value sits in the raddr B field, so we
43 * can't have 2 small immediates in one instruction (unless
44 * they're the same value, but that should be optimized away
45 * elsewhere).
46 */
47 bool uses_small_imm = false;
48 for (int i = 0; i < qir_get_nsrc(inst); i++) {
49 if (inst->src[i].file == QFILE_SMALL_IMM)
50 uses_small_imm = true;
51 }
52 if (uses_small_imm)
53 continue;
54
55 /* Don't propagate small immediates into the top-end bounds
56 * checking for indirect UBO loads. The kernel doesn't parse
57 * small immediates and rejects the shader in this case. UBO
58 * loads are much more expensive than the uniform load, and
59 * indirect UBO regions are usually much larger than a small
60 * immediate, so it's not worth updating the kernel to allow
61 * optimizing it.
62 */
63 if (inst->op == QOP_MIN_NOIMM)
64 continue;
65
66 for (int i = 0; i < qir_get_nsrc(inst); i++) {
67 struct qreg src = qir_follow_movs(c, inst->src[i]);
68
69 if (src.file != QFILE_UNIF ||
70 src.pack ||
71 c->uniform_contents[src.index] !=
72 QUNIFORM_CONSTANT) {
73 continue;
74 }
75
76 if (qir_is_tex(inst) &&
77 i == qir_get_tex_uniform_src(inst)) {
78 /* No turning the implicit uniform read into
79 * an immediate.
80 */
81 continue;
82 }
83
84 uint32_t imm = c->uniform_data[src.index];
85 uint32_t small_imm = qpu_encode_small_immediate(imm);
86 if (small_imm == ~0)
87 continue;
88
89 if (debug) {
90 fprintf(stderr, "opt_small_immediate() from: ");
91 qir_dump_inst(c, inst);
92 fprintf(stderr, "\n");
93 }
94 inst->src[i].file = QFILE_SMALL_IMM;
95 inst->src[i].index = imm;
96 if (debug) {
97 fprintf(stderr, "to: ");
98 qir_dump_inst(c, inst);
99 fprintf(stderr, "\n");
100 }
101 progress = true;
102 break;
103 }
104 }
105
106 return progress;
107 }
108