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
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
|
/*
* Copyright © 2014 Broadcom
*
* 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 "vc4_qpu.h"
static void
fail_instr(uint64_t inst, const char *msg)
{
fprintf(stderr, "vc4_qpu_validate: %s: ", msg);
vc4_qpu_disasm(&inst, 1);
fprintf(stderr, "\n");
abort();
}
static bool
writes_reg(uint64_t inst, uint32_t w)
{
return (QPU_GET_FIELD(inst, QPU_WADDR_ADD) == w ||
QPU_GET_FIELD(inst, QPU_WADDR_MUL) == w);
}
static bool
_reads_reg(uint64_t inst, uint32_t r, bool ignore_a, bool ignore_b)
{
struct {
uint32_t mux, addr;
} src_regs[] = {
{ QPU_GET_FIELD(inst, QPU_ADD_A) },
{ QPU_GET_FIELD(inst, QPU_ADD_B) },
{ QPU_GET_FIELD(inst, QPU_MUL_A) },
{ QPU_GET_FIELD(inst, QPU_MUL_B) },
};
/* Branches only reference raddr_a (no mux), and we don't use that
* feature of branching.
*/
if (QPU_GET_FIELD(inst, QPU_SIG) == QPU_SIG_BRANCH)
return false;
for (int i = 0; i < ARRAY_SIZE(src_regs); i++) {
if (!ignore_a &&
src_regs[i].mux == QPU_MUX_A &&
(QPU_GET_FIELD(inst, QPU_RADDR_A) == r))
return true;
if (!ignore_b &&
QPU_GET_FIELD(inst, QPU_SIG) != QPU_SIG_SMALL_IMM &&
src_regs[i].mux == QPU_MUX_B &&
(QPU_GET_FIELD(inst, QPU_RADDR_B) == r))
return true;
}
return false;
}
static bool
reads_reg(uint64_t inst, uint32_t r)
{
return _reads_reg(inst, r, false, false);
}
static bool
reads_a_reg(uint64_t inst, uint32_t r)
{
return _reads_reg(inst, r, false, true);
}
static bool
reads_b_reg(uint64_t inst, uint32_t r)
{
return _reads_reg(inst, r, true, false);
}
static bool
writes_sfu(uint64_t inst)
{
return (writes_reg(inst, QPU_W_SFU_RECIP) ||
writes_reg(inst, QPU_W_SFU_RECIPSQRT) ||
writes_reg(inst, QPU_W_SFU_EXP) ||
writes_reg(inst, QPU_W_SFU_LOG));
}
/**
* Checks for the instruction restrictions from page 37 ("Summary of
* Instruction Restrictions").
*/
void
vc4_qpu_validate(uint64_t *insts, uint32_t num_inst)
{
bool scoreboard_locked = false;
/* We don't want to do validation in release builds, but we want to
* keep compiling the validation code to make sure it doesn't get
* broken.
*/
#ifndef DEBUG
return;
#endif
for (int i = 0; i < num_inst; i++) {
uint64_t inst = insts[i];
if (QPU_GET_FIELD(inst, QPU_SIG) != QPU_SIG_PROG_END) {
if (qpu_inst_is_tlb(inst))
scoreboard_locked = true;
continue;
}
/* "The Thread End instruction must not write to either physical
* regfile A or B."
*/
if (QPU_GET_FIELD(inst, QPU_WADDR_ADD) < 32 ||
QPU_GET_FIELD(inst, QPU_WADDR_MUL) < 32) {
fail_instr(inst, "write to phys reg in thread end");
}
/* Can't trigger an implicit wait on scoreboard in the program
* end instruction.
*/
if (qpu_inst_is_tlb(inst) && !scoreboard_locked)
fail_instr(inst, "implicit sb wait in program end");
/* Two delay slots will be executed. */
assert(i + 2 <= num_inst);
for (int j = i; j < i + 2; j++) {
/* "The last three instructions of any program
* (Thread End plus the following two delay-slot
* instructions) must not do varyings read, uniforms
* read or any kind of VPM, VDR, or VDW read or
* write."
*/
if (writes_reg(insts[j], QPU_W_VPM) ||
reads_reg(insts[j], QPU_R_VARY) ||
reads_reg(insts[j], QPU_R_UNIF) ||
reads_reg(insts[j], QPU_R_VPM)) {
fail_instr(insts[j], "last 3 instructions "
"using fixed functions");
}
/* "The Thread End instruction and the following two
* delay slot instructions must not write or read
* address 14 in either regfile A or B."
*/
if (writes_reg(insts[j], 14) ||
reads_reg(insts[j], 14)) {
fail_instr(insts[j], "last 3 instructions "
"must not use r14");
}
}
/* "The final program instruction (the second delay slot
* instruction) must not do a TLB Z write."
*/
if (writes_reg(insts[i + 2], QPU_W_TLB_Z)) {
fail_instr(insts[i + 2], "final instruction doing "
"Z write");
}
}
/* "A scoreboard wait must not occur in the first two instructions of
* a fragment shader. This is either the explicit Wait for Scoreboard
* signal or an implicit wait with the first tile-buffer read or
* write instruction."
*/
for (int i = 0; i < 2; i++) {
uint64_t inst = insts[i];
if (qpu_inst_is_tlb(inst))
fail_instr(inst, "sb wait in first two insts");
}
/* "If TMU_NOSWAP is written, the write must be three instructions
* before the first TMU write instruction. For example, if
* TMU_NOSWAP is written in the first shader instruction, the first
* TMU write cannot occur before the 4th shader instruction."
*/
int last_tmu_noswap = -10;
for (int i = 0; i < num_inst; i++) {
uint64_t inst = insts[i];
if ((i - last_tmu_noswap) <= 3 &&
(writes_reg(inst, QPU_W_TMU0_S) ||
writes_reg(inst, QPU_W_TMU1_S))) {
fail_instr(inst, "TMU write too soon after TMU_NOSWAP");
}
if (writes_reg(inst, QPU_W_TMU_NOSWAP))
last_tmu_noswap = i;
}
/* "An instruction must not read from a location in physical regfile A
* or B that was written to by the previous instruction."
*/
for (int i = 0; i < num_inst - 1; i++) {
uint64_t inst = insts[i];
uint32_t add_waddr = QPU_GET_FIELD(inst, QPU_WADDR_ADD);
uint32_t mul_waddr = QPU_GET_FIELD(inst, QPU_WADDR_MUL);
uint32_t waddr_a, waddr_b;
if (inst & QPU_WS) {
waddr_b = add_waddr;
waddr_a = mul_waddr;
} else {
waddr_a = add_waddr;
waddr_b = mul_waddr;
}
if ((waddr_a < 32 && reads_a_reg(insts[i + 1], waddr_a)) ||
(waddr_b < 32 && reads_b_reg(insts[i + 1], waddr_b))) {
fail_instr(insts[i + 1],
"Reads physical reg too soon after write");
}
}
/* "After an SFU lookup instruction, accumulator r4 must not be read
* in the following two instructions. Any other instruction that
* results in r4 being written (that is, TMU read, TLB read, SFU
* lookup) cannot occur in the two instructions following an SFU
* lookup."
*/
int last_sfu_inst = -10;
for (int i = 0; i < num_inst - 1; i++) {
uint64_t inst = insts[i];
uint32_t sig = QPU_GET_FIELD(inst, QPU_SIG);
if (i - last_sfu_inst <= 2 &&
(writes_sfu(inst) ||
sig == QPU_SIG_LOAD_TMU0 ||
sig == QPU_SIG_LOAD_TMU1 ||
sig == QPU_SIG_COLOR_LOAD)) {
fail_instr(inst, "R4 write too soon after SFU write");
}
if (writes_sfu(inst))
last_sfu_inst = i;
}
for (int i = 0; i < num_inst - 1; i++) {
uint64_t inst = insts[i];
if (QPU_GET_FIELD(inst, QPU_SIG) == QPU_SIG_SMALL_IMM &&
QPU_GET_FIELD(inst, QPU_SMALL_IMM) >=
QPU_SMALL_IMM_MUL_ROT) {
uint32_t mux_a = QPU_GET_FIELD(inst, QPU_MUL_A);
uint32_t mux_b = QPU_GET_FIELD(inst, QPU_MUL_B);
/* "The full horizontal vector rotate is only
* available when both of the mul ALU input arguments
* are taken from accumulators r0-r3."
*/
if (mux_a > QPU_MUX_R3 || mux_b > QPU_MUX_R3) {
fail_instr(inst,
"MUL rotate using non-accumulator "
"input");
}
if (QPU_GET_FIELD(inst, QPU_SMALL_IMM) ==
QPU_SMALL_IMM_MUL_ROT) {
/* "An instruction that does a vector rotate
* by r5 must not immediately follow an
* instruction that writes to r5."
*/
if (writes_reg(insts[i - 1], QPU_W_ACC5)) {
fail_instr(inst,
"vector rotate by r5 "
"immediately after r5 write");
}
}
/* "An instruction that does a vector rotate must not
* immediately follow an instruction that writes to the
* accumulator that is being rotated."
*/
if (writes_reg(insts[i - 1], QPU_W_ACC0 + mux_a) ||
writes_reg(insts[i - 1], QPU_W_ACC0 + mux_b)) {
fail_instr(inst,
"vector rotate of value "
"written in previous instruction");
}
}
}
/* "An instruction that does a vector rotate must not immediately
* follow an instruction that writes to the accumulator that is being
* rotated.
*
* XXX: TODO.
*/
/* "After an instruction that does a TLB Z write, the multisample mask
* must not be read as an instruction input argument in the following
* two instruction. The TLB Z write instruction can, however, be
* followed immediately by a TLB color write."
*/
for (int i = 0; i < num_inst - 1; i++) {
uint64_t inst = insts[i];
if (writes_reg(inst, QPU_W_TLB_Z) &&
(reads_a_reg(insts[i + 1], QPU_R_MS_REV_FLAGS) ||
reads_a_reg(insts[i + 2], QPU_R_MS_REV_FLAGS))) {
fail_instr(inst, "TLB Z write followed by MS mask read");
}
}
/*
* "A single instruction can only perform a maximum of one of the
* following closely coupled peripheral accesses in a single
* instruction: TMU write, TMU read, TLB write, TLB read, TLB
* combined color read and write, SFU write, Mutex read or Semaphore
* access."
*/
for (int i = 0; i < num_inst - 1; i++) {
uint64_t inst = insts[i];
if (qpu_num_sf_accesses(inst) > 1)
fail_instr(inst, "Single instruction writes SFU twice");
}
/* "The uniform base pointer can be written (from SIMD element 0) by
* the processor to reset the stream, there must be at least two
* nonuniform-accessing instructions following a pointer change
* before uniforms can be accessed once more."
*/
int last_unif_pointer_update = -3;
for (int i = 0; i < num_inst; i++) {
uint64_t inst = insts[i];
uint32_t waddr_add = QPU_GET_FIELD(inst, QPU_WADDR_ADD);
uint32_t waddr_mul = QPU_GET_FIELD(inst, QPU_WADDR_MUL);
if (reads_reg(inst, QPU_R_UNIF) &&
i - last_unif_pointer_update <= 2) {
fail_instr(inst,
"uniform read too soon after pointer update");
}
if (waddr_add == QPU_W_UNIFORMS_ADDRESS ||
waddr_mul == QPU_W_UNIFORMS_ADDRESS)
last_unif_pointer_update = i;
}
}
|
