2 * Copyright © 2014 Broadcom
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:
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
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
24 #include "util/u_memory.h"
25 #include "util/u_simple_list.h"
26 #include "util/ralloc.h"
34 bool has_side_effects
;
37 static const struct qir_op_info qir_op_info
[] = {
38 [QOP_MOV
] = { "mov", 1, 1 },
39 [QOP_FADD
] = { "fadd", 1, 2 },
40 [QOP_FSUB
] = { "fsub", 1, 2 },
41 [QOP_FMUL
] = { "fmul", 1, 2 },
42 [QOP_MUL24
] = { "mul24", 1, 2 },
43 [QOP_FMIN
] = { "fmin", 1, 2 },
44 [QOP_FMAX
] = { "fmax", 1, 2 },
45 [QOP_FMINABS
] = { "fminabs", 1, 2 },
46 [QOP_FMAXABS
] = { "fmaxabs", 1, 2 },
47 [QOP_FTOI
] = { "ftoi", 1, 1 },
48 [QOP_ITOF
] = { "itof", 1, 1 },
49 [QOP_ADD
] = { "add", 1, 2 },
50 [QOP_SUB
] = { "sub", 1, 2 },
51 [QOP_SHR
] = { "shr", 1, 2 },
52 [QOP_ASR
] = { "asr", 1, 2 },
53 [QOP_SHL
] = { "shl", 1, 2 },
54 [QOP_MIN
] = { "min", 1, 2 },
55 [QOP_MAX
] = { "max", 1, 2 },
56 [QOP_AND
] = { "and", 1, 2 },
57 [QOP_OR
] = { "or", 1, 2 },
58 [QOP_XOR
] = { "xor", 1, 2 },
59 [QOP_NOT
] = { "not", 1, 1 },
61 [QOP_SF
] = { "sf", 0, 1 },
62 [QOP_SEL_X_0_NS
] = { "fsel_x_0_ns", 1, 1 },
63 [QOP_SEL_X_0_NC
] = { "fsel_x_0_nc", 1, 1 },
64 [QOP_SEL_X_0_ZS
] = { "fsel_x_0_zs", 1, 1 },
65 [QOP_SEL_X_0_ZC
] = { "fsel_x_0_zc", 1, 1 },
66 [QOP_SEL_X_Y_NS
] = { "fsel_x_y_ns", 1, 2 },
67 [QOP_SEL_X_Y_NC
] = { "fsel_x_y_nc", 1, 2 },
68 [QOP_SEL_X_Y_ZS
] = { "fsel_x_y_zs", 1, 2 },
69 [QOP_SEL_X_Y_ZC
] = { "fsel_x_y_zc", 1, 2 },
71 [QOP_RCP
] = { "rcp", 1, 1 },
72 [QOP_RSQ
] = { "rsq", 1, 1 },
73 [QOP_EXP2
] = { "exp2", 1, 2 },
74 [QOP_LOG2
] = { "log2", 1, 2 },
75 [QOP_PACK_COLORS
] = { "pack_colors", 1, 4 },
76 [QOP_PACK_SCALED
] = { "pack_scaled", 1, 2 },
77 [QOP_VPM_WRITE
] = { "vpm_write", 0, 1, true },
78 [QOP_VPM_READ
] = { "vpm_read", 0, 1, true },
79 [QOP_TLB_DISCARD_SETUP
] = { "discard", 0, 1, true },
80 [QOP_TLB_STENCIL_SETUP
] = { "tlb_stencil_setup", 0, 1, true },
81 [QOP_TLB_Z_WRITE
] = { "tlb_z", 0, 1, true },
82 [QOP_TLB_COLOR_WRITE
] = { "tlb_color", 0, 1, true },
83 [QOP_TLB_COLOR_READ
] = { "tlb_color_read", 1, 0 },
84 [QOP_VARY_ADD_C
] = { "vary_add_c", 1, 1 },
86 [QOP_FRAG_X
] = { "frag_x", 1, 0 },
87 [QOP_FRAG_Y
] = { "frag_y", 1, 0 },
88 [QOP_FRAG_Z
] = { "frag_z", 1, 0 },
89 [QOP_FRAG_W
] = { "frag_w", 1, 0 },
90 [QOP_FRAG_REV_FLAG
] = { "frag_rev_flag", 1, 0 },
92 [QOP_TEX_S
] = { "tex_s", 0, 2 },
93 [QOP_TEX_T
] = { "tex_t", 0, 2 },
94 [QOP_TEX_R
] = { "tex_r", 0, 2 },
95 [QOP_TEX_B
] = { "tex_b", 0, 2 },
96 [QOP_TEX_DIRECT
] = { "tex_direct", 0, 2 },
97 [QOP_TEX_RESULT
] = { "tex_result", 1, 0, true },
98 [QOP_R4_UNPACK_A
] = { "r4_unpack_a", 1, 1 },
99 [QOP_R4_UNPACK_B
] = { "r4_unpack_b", 1, 1 },
100 [QOP_R4_UNPACK_C
] = { "r4_unpack_c", 1, 1 },
101 [QOP_R4_UNPACK_D
] = { "r4_unpack_d", 1, 1 },
102 [QOP_UNPACK_8A_F
] = { "unpack_8a_f", 1, 1 },
103 [QOP_UNPACK_8B_F
] = { "unpack_8b_f", 1, 1 },
104 [QOP_UNPACK_8C_F
] = { "unpack_8c_f", 1, 1 },
105 [QOP_UNPACK_8D_F
] = { "unpack_8d_f", 1, 1 },
106 [QOP_UNPACK_16A_F
] = { "unpack_16a_f", 1, 1 },
107 [QOP_UNPACK_16B_F
] = { "unpack_16b_f", 1, 1 },
108 [QOP_UNPACK_8A_I
] = { "unpack_8a_i", 1, 1 },
109 [QOP_UNPACK_8B_I
] = { "unpack_8b_i", 1, 1 },
110 [QOP_UNPACK_8C_I
] = { "unpack_8c_i", 1, 1 },
111 [QOP_UNPACK_8D_I
] = { "unpack_8d_i", 1, 1 },
112 [QOP_UNPACK_16A_I
] = { "unpack_16a_i", 1, 1 },
113 [QOP_UNPACK_16B_I
] = { "unpack_16b_i", 1, 1 },
117 qir_get_op_name(enum qop qop
)
119 if (qop
< ARRAY_SIZE(qir_op_info
) && qir_op_info
[qop
].name
)
120 return qir_op_info
[qop
].name
;
126 qir_get_op_nsrc(enum qop qop
)
128 if (qop
< ARRAY_SIZE(qir_op_info
) && qir_op_info
[qop
].name
)
129 return qir_op_info
[qop
].nsrc
;
135 * Returns whether the instruction has any side effects that must be
139 qir_has_side_effects(struct vc4_compile
*c
, struct qinst
*inst
)
141 /* We can dead-code eliminate varyings, because we only tell the VS
142 * about the live ones at the end. But we have to preserve the
143 * point/line coordinates reads, because they're generated by
144 * fixed-function hardware.
146 for (int i
= 0; i
< qir_get_op_nsrc(inst
->op
); i
++) {
147 if (inst
->src
[i
].file
== QFILE_VARY
&&
148 c
->input_semantics
[inst
->src
[i
].index
].semantic
== 0xff) {
153 return qir_op_info
[inst
->op
].has_side_effects
;
157 qir_depends_on_flags(struct qinst
*inst
)
175 qir_writes_r4(struct qinst
*inst
)
179 case QOP_TLB_COLOR_READ
:
191 qir_reads_r4(struct qinst
*inst
)
194 case QOP_R4_UNPACK_A
:
195 case QOP_R4_UNPACK_B
:
196 case QOP_R4_UNPACK_C
:
197 case QOP_R4_UNPACK_D
:
205 qir_print_reg(struct vc4_compile
*c
, struct qreg reg
)
207 const char *files
[] = {
213 if (reg
.file
== QFILE_NULL
)
214 fprintf(stderr
, "null");
216 fprintf(stderr
, "%s%d", files
[reg
.file
], reg
.index
);
218 if (reg
.file
== QFILE_UNIF
&&
219 c
->uniform_contents
[reg
.index
] == QUNIFORM_CONSTANT
) {
220 fprintf(stderr
, " (0x%08x / %f)",
221 c
->uniform_data
[reg
.index
],
222 uif(c
->uniform_data
[reg
.index
]));
227 qir_dump_inst(struct vc4_compile
*c
, struct qinst
*inst
)
229 fprintf(stderr
, "%s ", qir_get_op_name(inst
->op
));
231 qir_print_reg(c
, inst
->dst
);
232 for (int i
= 0; i
< qir_get_op_nsrc(inst
->op
); i
++) {
233 fprintf(stderr
, ", ");
234 qir_print_reg(c
, inst
->src
[i
]);
239 qir_dump(struct vc4_compile
*c
)
241 struct simple_node
*node
;
243 foreach(node
, &c
->instructions
) {
244 struct qinst
*inst
= (struct qinst
*)node
;
245 qir_dump_inst(c
, inst
);
246 fprintf(stderr
, "\n");
251 qir_get_temp(struct vc4_compile
*c
)
255 reg
.file
= QFILE_TEMP
;
256 reg
.index
= c
->num_temps
++;
262 qir_inst(enum qop op
, struct qreg dst
, struct qreg src0
, struct qreg src1
)
264 struct qinst
*inst
= CALLOC_STRUCT(qinst
);
268 inst
->src
= calloc(2, sizeof(inst
->src
[0]));
276 qir_inst4(enum qop op
, struct qreg dst
,
282 struct qinst
*inst
= CALLOC_STRUCT(qinst
);
286 inst
->src
= calloc(4, sizeof(*inst
->src
));
296 qir_emit(struct vc4_compile
*c
, struct qinst
*inst
)
298 insert_at_tail(&c
->instructions
, &inst
->link
);
302 qir_reg_equals(struct qreg a
, struct qreg b
)
304 return a
.file
== b
.file
&& a
.index
== b
.index
;
308 qir_compile_init(void)
310 struct vc4_compile
*c
= rzalloc(NULL
, struct vc4_compile
);
312 make_empty_list(&c
->instructions
);
314 c
->output_position_index
= -1;
315 c
->output_clipvertex_index
= -1;
316 c
->output_color_index
= -1;
317 c
->output_point_size_index
= -1;
323 qir_remove_instruction(struct qinst
*qinst
)
325 remove_from_list(&qinst
->link
);
331 qir_compile_destroy(struct vc4_compile
*c
)
333 while (!is_empty_list(&c
->instructions
)) {
334 struct qinst
*qinst
=
335 (struct qinst
*)first_elem(&c
->instructions
);
336 qir_remove_instruction(qinst
);
343 qir_get_stage_name(enum qstage stage
)
345 static const char *names
[] = {
346 [QSTAGE_FRAG
] = "FS",
347 [QSTAGE_VERT
] = "VS",
348 [QSTAGE_COORD
] = "CS",
354 #define OPTPASS(func) \
356 bool stage_progress = func(c); \
357 if (stage_progress) { \
359 if (print_opt_debug) { \
361 "QIR opt pass %2d: %s progress\n", \
368 qir_optimize(struct vc4_compile
*c
)
370 bool print_opt_debug
= false;
374 bool progress
= false;
376 OPTPASS(qir_opt_algebraic
);
377 OPTPASS(qir_opt_cse
);
378 OPTPASS(qir_opt_copy_propagation
);
379 OPTPASS(qir_opt_dead_code
);