vc4: Fix assertion about SFU versus texturing.
[mesa.git] / src / gallium / drivers / vc4 / vc4_qpu_validate.c
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 #include "vc4_qpu.h"
25
26 static bool
27 writes_reg(uint64_t inst, uint32_t w)
28 {
29 return (QPU_GET_FIELD(inst, QPU_WADDR_ADD) == w ||
30 QPU_GET_FIELD(inst, QPU_WADDR_MUL) == w);
31 }
32
33 static bool
34 _reads_reg(uint64_t inst, uint32_t r, bool ignore_a, bool ignore_b)
35 {
36 struct {
37 uint32_t mux, addr;
38 } src_regs[] = {
39 { QPU_GET_FIELD(inst, QPU_ADD_A) },
40 { QPU_GET_FIELD(inst, QPU_ADD_B) },
41 { QPU_GET_FIELD(inst, QPU_MUL_A) },
42 { QPU_GET_FIELD(inst, QPU_MUL_B) },
43 };
44
45 for (int i = 0; i < ARRAY_SIZE(src_regs); i++) {
46 if (!ignore_a &&
47 src_regs[i].mux == QPU_MUX_A &&
48 (QPU_GET_FIELD(inst, QPU_RADDR_A) == r))
49 return true;
50
51 if (!ignore_b &&
52 src_regs[i].mux == QPU_MUX_B &&
53 (QPU_GET_FIELD(inst, QPU_RADDR_B) == r))
54 return true;
55 }
56
57 return false;
58 }
59
60 static bool
61 reads_reg(uint64_t inst, uint32_t r)
62 {
63 return _reads_reg(inst, r, false, false);
64 }
65
66 static bool
67 reads_a_reg(uint64_t inst, uint32_t r)
68 {
69 return _reads_reg(inst, r, false, true);
70 }
71
72 static bool
73 reads_b_reg(uint64_t inst, uint32_t r)
74 {
75 return _reads_reg(inst, r, true, false);
76 }
77
78 static bool
79 writes_sfu(uint64_t inst)
80 {
81 return (writes_reg(inst, QPU_W_SFU_RECIP) ||
82 writes_reg(inst, QPU_W_SFU_RECIPSQRT) ||
83 writes_reg(inst, QPU_W_SFU_EXP) ||
84 writes_reg(inst, QPU_W_SFU_LOG));
85 }
86
87 /**
88 * Checks for the instruction restrictions from page 37 ("Summary of
89 * Instruction Restrictions").
90 */
91 void
92 vc4_qpu_validate(uint64_t *insts, uint32_t num_inst)
93 {
94 bool scoreboard_locked = false;
95
96 for (int i = 0; i < num_inst; i++) {
97 uint64_t inst = insts[i];
98
99 if (QPU_GET_FIELD(inst, QPU_SIG) != QPU_SIG_PROG_END) {
100 if (qpu_inst_is_tlb(inst))
101 scoreboard_locked = true;
102
103 continue;
104 }
105
106 /* "The Thread End instruction must not write to either physical
107 * regfile A or B."
108 */
109 assert(QPU_GET_FIELD(inst, QPU_WADDR_ADD) >= 32);
110 assert(QPU_GET_FIELD(inst, QPU_WADDR_MUL) >= 32);
111
112 /* Can't trigger an implicit wait on scoreboard in the program
113 * end instruction.
114 */
115 assert(!qpu_inst_is_tlb(inst) || scoreboard_locked);
116
117 /* Two delay slots will be executed. */
118 assert(i + 2 <= num_inst);
119
120 for (int j = i; j < i + 2; j++) {
121 /* "The last three instructions of any program
122 * (Thread End plus the following two delay-slot
123 * instructions) must not do varyings read, uniforms
124 * read or any kind of VPM, VDR, or VDW read or
125 * write."
126 */
127 assert(!writes_reg(insts[j], QPU_W_VPM));
128 assert(!reads_reg(insts[j], QPU_R_VARY));
129 assert(!reads_reg(insts[j], QPU_R_UNIF));
130 assert(!reads_reg(insts[j], QPU_R_VPM));
131
132 /* "The Thread End instruction and the following two
133 * delay slot instructions must not write or read
134 * address 14 in either regfile A or B."
135 */
136 assert(!writes_reg(insts[j], 14));
137 assert(!reads_reg(insts[j], 14));
138
139 }
140
141 /* "The final program instruction (the second delay slot
142 * instruction) must not do a TLB Z write."
143 */
144 assert(!writes_reg(insts[i + 2], QPU_W_TLB_Z));
145 }
146
147 /* "A scoreboard wait must not occur in the first two instructions of
148 * a fragment shader. This is either the explicit Wait for Scoreboard
149 * signal or an implicit wait with the first tile-buffer read or
150 * write instruction."
151 */
152 for (int i = 0; i < 2; i++) {
153 uint64_t inst = insts[i];
154
155 assert(!qpu_inst_is_tlb(inst));
156 }
157
158 /* "If TMU_NOSWAP is written, the write must be three instructions
159 * before the first TMU write instruction. For example, if
160 * TMU_NOSWAP is written in the first shader instruction, the first
161 * TMU write cannot occur before the 4th shader instruction."
162 */
163 int last_tmu_noswap = -10;
164 for (int i = 0; i < num_inst; i++) {
165 uint64_t inst = insts[i];
166
167 assert((i - last_tmu_noswap) > 3 ||
168 (!writes_reg(inst, QPU_W_TMU0_S) &&
169 !writes_reg(inst, QPU_W_TMU1_S)));
170
171 if (writes_reg(inst, QPU_W_TMU_NOSWAP))
172 last_tmu_noswap = i;
173 }
174
175 /* "An instruction must not read from a location in physical regfile A
176 * or B that was written to by the previous instruction."
177 */
178 for (int i = 0; i < num_inst - 1; i++) {
179 uint64_t inst = insts[i];
180 uint32_t add_waddr = QPU_GET_FIELD(inst, QPU_WADDR_ADD);
181 uint32_t mul_waddr = QPU_GET_FIELD(inst, QPU_WADDR_MUL);
182 uint32_t waddr_a, waddr_b;
183
184 if (inst & QPU_WS) {
185 waddr_b = add_waddr;
186 waddr_a = mul_waddr;
187 } else {
188 waddr_a = add_waddr;
189 waddr_b = mul_waddr;
190 }
191
192 assert(waddr_a >= 32 || !reads_a_reg(insts[i + 1], waddr_a));
193 assert(waddr_b >= 32 || !reads_b_reg(insts[i + 1], waddr_b));
194 }
195
196 /* "After an SFU lookup instruction, accumulator r4 must not be read
197 * in the following two instructions. Any other instruction that
198 * results in r4 being written (that is, TMU read, TLB read, SFU
199 * lookup) cannot occur in the two instructions following an SFU
200 * lookup."
201 */
202 int last_sfu_inst = -10;
203 for (int i = 0; i < num_inst - 1; i++) {
204 uint64_t inst = insts[i];
205 uint32_t sig = QPU_GET_FIELD(inst, QPU_SIG);
206
207 assert(i - last_sfu_inst > 2 ||
208 (!writes_sfu(inst) &&
209 sig != QPU_SIG_LOAD_TMU0 &&
210 sig != QPU_SIG_LOAD_TMU1 &&
211 sig != QPU_SIG_COLOR_LOAD));
212
213 if (writes_sfu(inst))
214 last_sfu_inst = i;
215 }
216
217 int last_r5_write = -10;
218 for (int i = 0; i < num_inst - 1; i++) {
219 uint64_t inst = insts[i];
220
221 /* "An instruction that does a vector rotate by r5 must not
222 * immediately follow an instruction that writes to r5."
223 */
224 assert(last_r5_write != i - 1 ||
225 QPU_GET_FIELD(inst, QPU_SIG) != QPU_SIG_SMALL_IMM ||
226 QPU_GET_FIELD(inst, QPU_SMALL_IMM) != 48);
227 }
228
229 /* "An instruction that does a vector rotate must not immediately
230 * follow an instruction that writes to the accumulator that is being
231 * rotated.
232 *
233 * XXX: TODO.
234 */
235
236 /* "After an instruction that does a TLB Z write, the multisample mask
237 * must not be read as an instruction input argument in the following
238 * two instruction. The TLB Z write instruction can, however, be
239 * followed immediately by a TLB color write."
240 */
241 for (int i = 0; i < num_inst - 1; i++) {
242 uint64_t inst = insts[i];
243 if (writes_reg(inst, QPU_W_TLB_Z)) {
244 assert(!reads_a_reg(insts[i + 1], QPU_R_MS_REV_FLAGS));
245 assert(!reads_a_reg(insts[i + 2], QPU_R_MS_REV_FLAGS));
246 }
247 }
248
249 /*
250 * "A single instruction can only perform a maximum of one of the
251 * following closely coupled peripheral accesses in a single
252 * instruction: TMU write, TMU read, TLB write, TLB read, TLB
253 * combined color read and write, SFU write, Mutex read or Semaphore
254 * access."
255 */
256 for (int i = 0; i < num_inst - 1; i++) {
257 uint64_t inst = insts[i];
258 int accesses = 0;
259 static const uint32_t specials[] = {
260 QPU_W_TLB_COLOR_MS,
261 QPU_W_TLB_COLOR_ALL,
262 QPU_W_TLB_Z,
263 QPU_W_TMU0_S,
264 QPU_W_TMU0_T,
265 QPU_W_TMU0_R,
266 QPU_W_TMU0_B,
267 QPU_W_TMU1_S,
268 QPU_W_TMU1_T,
269 QPU_W_TMU1_R,
270 QPU_W_TMU1_B,
271 QPU_W_SFU_RECIP,
272 QPU_W_SFU_RECIPSQRT,
273 QPU_W_SFU_EXP,
274 QPU_W_SFU_LOG,
275 };
276
277 for (int j = 0; j < ARRAY_SIZE(specials); j++) {
278 if (writes_reg(inst, specials[j]))
279 accesses++;
280 }
281
282 if (reads_reg(inst, QPU_R_MUTEX_ACQUIRE))
283 accesses++;
284
285 /* XXX: semaphore, combined color read/write? */
286 switch (QPU_GET_FIELD(inst, QPU_SIG)) {
287 case QPU_SIG_COLOR_LOAD:
288 case QPU_SIG_COLOR_LOAD_END:
289 case QPU_SIG_LOAD_TMU0:
290 case QPU_SIG_LOAD_TMU1:
291 accesses++;
292 }
293
294 assert(accesses <= 1);
295 }
296 }