1 /*
2 * Copyright © 2016-2017 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 "broadcom/common/v3d_device_info.h"
25 #include "v3d_compiler.h"
26
27 /* Prints a human-readable description of the uniform reference. */
28 void
vir_dump_uniform(enum quniform_contents contents,uint32_t data)29 vir_dump_uniform(enum quniform_contents contents,
30 uint32_t data)
31 {
32 static const char *quniform_names[] = {
33 [QUNIFORM_LINE_WIDTH] = "line_width",
34 [QUNIFORM_AA_LINE_WIDTH] = "aa_line_width",
35 [QUNIFORM_VIEWPORT_X_SCALE] = "vp_x_scale",
36 [QUNIFORM_VIEWPORT_Y_SCALE] = "vp_y_scale",
37 [QUNIFORM_VIEWPORT_Z_OFFSET] = "vp_z_offset",
38 [QUNIFORM_VIEWPORT_Z_SCALE] = "vp_z_scale",
39 [QUNIFORM_SHARED_OFFSET] = "shared_offset",
40 };
41
42 switch (contents) {
43 case QUNIFORM_CONSTANT:
44 fprintf(stderr, "0x%08x / %f", data, uif(data));
45 break;
46
47 case QUNIFORM_UNIFORM:
48 fprintf(stderr, "push[%d]", data);
49 break;
50
51 case QUNIFORM_TEXTURE_CONFIG_P1:
52 fprintf(stderr, "tex[%d].p1", data);
53 break;
54
55 case QUNIFORM_TMU_CONFIG_P0:
56 fprintf(stderr, "tex[%d].p0 | 0x%x",
57 v3d_unit_data_get_unit(data),
58 v3d_unit_data_get_offset(data));
59 break;
60
61 case QUNIFORM_TMU_CONFIG_P1:
62 fprintf(stderr, "tex[%d].p1 | 0x%x",
63 v3d_unit_data_get_unit(data),
64 v3d_unit_data_get_offset(data));
65 break;
66
67 case QUNIFORM_IMAGE_TMU_CONFIG_P0:
68 fprintf(stderr, "img[%d].p0 | 0x%x",
69 v3d_unit_data_get_unit(data),
70 v3d_unit_data_get_offset(data));
71 break;
72
73 case QUNIFORM_TEXTURE_WIDTH:
74 fprintf(stderr, "tex[%d].width", data);
75 break;
76 case QUNIFORM_TEXTURE_HEIGHT:
77 fprintf(stderr, "tex[%d].height", data);
78 break;
79 case QUNIFORM_TEXTURE_DEPTH:
80 fprintf(stderr, "tex[%d].depth", data);
81 break;
82 case QUNIFORM_TEXTURE_ARRAY_SIZE:
83 fprintf(stderr, "tex[%d].array_size", data);
84 break;
85 case QUNIFORM_TEXTURE_LEVELS:
86 fprintf(stderr, "tex[%d].levels", data);
87 break;
88
89 case QUNIFORM_IMAGE_WIDTH:
90 fprintf(stderr, "img[%d].width", data);
91 break;
92 case QUNIFORM_IMAGE_HEIGHT:
93 fprintf(stderr, "img[%d].height", data);
94 break;
95 case QUNIFORM_IMAGE_DEPTH:
96 fprintf(stderr, "img[%d].depth", data);
97 break;
98 case QUNIFORM_IMAGE_ARRAY_SIZE:
99 fprintf(stderr, "img[%d].array_size", data);
100 break;
101
102 case QUNIFORM_SPILL_OFFSET:
103 fprintf(stderr, "spill_offset");
104 break;
105
106 case QUNIFORM_SPILL_SIZE_PER_THREAD:
107 fprintf(stderr, "spill_size_per_thread");
108 break;
109
110 case QUNIFORM_UBO_ADDR:
111 fprintf(stderr, "ubo[%d]+0x%x",
112 v3d_unit_data_get_unit(data),
113 v3d_unit_data_get_offset(data));
114 break;
115
116 case QUNIFORM_SSBO_OFFSET:
117 fprintf(stderr, "ssbo[%d]", data);
118 break;
119
120 case QUNIFORM_GET_SSBO_SIZE:
121 fprintf(stderr, "ssbo_size[%d]", data);
122 break;
123
124 case QUNIFORM_GET_UBO_SIZE:
125 fprintf(stderr, "ubo_size[%d]", data);
126 break;
127
128 case QUNIFORM_NUM_WORK_GROUPS:
129 fprintf(stderr, "num_wg.%c", data < 3 ? "xyz"[data] : '?');
130 break;
131
132 default:
133 if (quniform_contents_is_texture_p0(contents)) {
134 fprintf(stderr, "tex[%d].p0: 0x%08x",
135 contents - QUNIFORM_TEXTURE_CONFIG_P0_0,
136 data);
137 } else if (contents < ARRAY_SIZE(quniform_names) &&
138 quniform_names[contents]) {
139 fprintf(stderr, "%s",
140 quniform_names[contents]);
141 } else {
142 fprintf(stderr, "%d / 0x%08x", contents, data);
143 }
144 }
145 }
146
147 static void
vir_print_reg(struct v3d_compile * c,const struct qinst * inst,struct qreg reg)148 vir_print_reg(struct v3d_compile *c, const struct qinst *inst,
149 struct qreg reg)
150 {
151 switch (reg.file) {
152
153 case QFILE_NULL:
154 fprintf(stderr, "null");
155 break;
156
157 case QFILE_LOAD_IMM:
158 fprintf(stderr, "0x%08x (%f)", reg.index, uif(reg.index));
159 break;
160
161 case QFILE_REG:
162 fprintf(stderr, "rf%d", reg.index);
163 break;
164
165 case QFILE_MAGIC:
166 fprintf(stderr, "%s",
167 v3d_qpu_magic_waddr_name(c->devinfo, reg.index));
168 break;
169
170 case QFILE_SMALL_IMM: {
171 uint32_t unpacked;
172 bool ok = v3d_qpu_small_imm_unpack(c->devinfo,
173 inst->qpu.raddr_b,
174 &unpacked);
175 assert(ok); (void) ok;
176
177 int8_t *p = (int8_t *)&inst->qpu.raddr_b;
178 if (*p >= -16 && *p <= 15)
179 fprintf(stderr, "%d", unpacked);
180 else
181 fprintf(stderr, "%f", uif(unpacked));
182 break;
183 }
184
185 case QFILE_VPM:
186 fprintf(stderr, "vpm%d.%d",
187 reg.index / 4, reg.index % 4);
188 break;
189
190 case QFILE_TEMP:
191 fprintf(stderr, "t%d", reg.index);
192 break;
193 }
194 }
195
196 static void
vir_dump_sig_addr(const struct v3d_device_info * devinfo,const struct v3d_qpu_instr * instr)197 vir_dump_sig_addr(const struct v3d_device_info *devinfo,
198 const struct v3d_qpu_instr *instr)
199 {
200 if (devinfo->ver < 41)
201 return;
202
203 if (!instr->sig_magic)
204 fprintf(stderr, ".rf%d", instr->sig_addr);
205 else {
206 const char *name =
207 v3d_qpu_magic_waddr_name(devinfo, instr->sig_addr);
208 if (name)
209 fprintf(stderr, ".%s", name);
210 else
211 fprintf(stderr, ".UNKNOWN%d", instr->sig_addr);
212 }
213 }
214
215 static void
vir_dump_sig(struct v3d_compile * c,struct qinst * inst)216 vir_dump_sig(struct v3d_compile *c, struct qinst *inst)
217 {
218 struct v3d_qpu_sig *sig = &inst->qpu.sig;
219
220 if (sig->thrsw)
221 fprintf(stderr, "; thrsw");
222 if (sig->ldvary) {
223 fprintf(stderr, "; ldvary");
224 vir_dump_sig_addr(c->devinfo, &inst->qpu);
225 }
226 if (sig->ldvpm)
227 fprintf(stderr, "; ldvpm");
228 if (sig->ldtmu) {
229 fprintf(stderr, "; ldtmu");
230 vir_dump_sig_addr(c->devinfo, &inst->qpu);
231 }
232 if (sig->ldtlb) {
233 fprintf(stderr, "; ldtlb");
234 vir_dump_sig_addr(c->devinfo, &inst->qpu);
235 }
236 if (sig->ldtlbu) {
237 fprintf(stderr, "; ldtlbu");
238 vir_dump_sig_addr(c->devinfo, &inst->qpu);
239 }
240 if (sig->ldunif)
241 fprintf(stderr, "; ldunif");
242 if (sig->ldunifrf) {
243 fprintf(stderr, "; ldunifrf");
244 vir_dump_sig_addr(c->devinfo, &inst->qpu);
245 }
246 if (sig->ldunifa)
247 fprintf(stderr, "; ldunifa");
248 if (sig->ldunifarf) {
249 fprintf(stderr, "; ldunifarf");
250 vir_dump_sig_addr(c->devinfo, &inst->qpu);
251 }
252 if (sig->wrtmuc)
253 fprintf(stderr, "; wrtmuc");
254 }
255
256 static void
vir_dump_alu(struct v3d_compile * c,struct qinst * inst)257 vir_dump_alu(struct v3d_compile *c, struct qinst *inst)
258 {
259 struct v3d_qpu_instr *instr = &inst->qpu;
260 int nsrc = vir_get_nsrc(inst);
261 enum v3d_qpu_input_unpack unpack[2];
262
263 if (inst->qpu.alu.add.op != V3D_QPU_A_NOP) {
264 fprintf(stderr, "%s", v3d_qpu_add_op_name(instr->alu.add.op));
265 fprintf(stderr, "%s", v3d_qpu_cond_name(instr->flags.ac));
266 fprintf(stderr, "%s", v3d_qpu_pf_name(instr->flags.apf));
267 fprintf(stderr, "%s", v3d_qpu_uf_name(instr->flags.auf));
268 fprintf(stderr, " ");
269
270 vir_print_reg(c, inst, inst->dst);
271 fprintf(stderr, "%s", v3d_qpu_pack_name(instr->alu.add.output_pack));
272
273 unpack[0] = instr->alu.add.a_unpack;
274 unpack[1] = instr->alu.add.b_unpack;
275 } else {
276 fprintf(stderr, "%s", v3d_qpu_mul_op_name(instr->alu.mul.op));
277 fprintf(stderr, "%s", v3d_qpu_cond_name(instr->flags.mc));
278 fprintf(stderr, "%s", v3d_qpu_pf_name(instr->flags.mpf));
279 fprintf(stderr, "%s", v3d_qpu_uf_name(instr->flags.muf));
280 fprintf(stderr, " ");
281
282 vir_print_reg(c, inst, inst->dst);
283 fprintf(stderr, "%s", v3d_qpu_pack_name(instr->alu.mul.output_pack));
284
285 unpack[0] = instr->alu.mul.a_unpack;
286 unpack[1] = instr->alu.mul.b_unpack;
287 }
288
289 for (int i = 0; i < nsrc; i++) {
290 fprintf(stderr, ", ");
291 vir_print_reg(c, inst, inst->src[i]);
292 fprintf(stderr, "%s", v3d_qpu_unpack_name(unpack[i]));
293 }
294
295 vir_dump_sig(c, inst);
296 }
297
298 void
vir_dump_inst(struct v3d_compile * c,struct qinst * inst)299 vir_dump_inst(struct v3d_compile *c, struct qinst *inst)
300 {
301 struct v3d_qpu_instr *instr = &inst->qpu;
302
303 switch (inst->qpu.type) {
304 case V3D_QPU_INSTR_TYPE_ALU:
305 vir_dump_alu(c, inst);
306 break;
307 case V3D_QPU_INSTR_TYPE_BRANCH:
308 fprintf(stderr, "b");
309 if (instr->branch.ub)
310 fprintf(stderr, "u");
311
312 fprintf(stderr, "%s",
313 v3d_qpu_branch_cond_name(instr->branch.cond));
314 fprintf(stderr, "%s", v3d_qpu_msfign_name(instr->branch.msfign));
315
316 switch (instr->branch.bdi) {
317 case V3D_QPU_BRANCH_DEST_ABS:
318 fprintf(stderr, " zero_addr+0x%08x", instr->branch.offset);
319 break;
320
321 case V3D_QPU_BRANCH_DEST_REL:
322 fprintf(stderr, " %d", instr->branch.offset);
323 break;
324
325 case V3D_QPU_BRANCH_DEST_LINK_REG:
326 fprintf(stderr, " lri");
327 break;
328
329 case V3D_QPU_BRANCH_DEST_REGFILE:
330 fprintf(stderr, " rf%d", instr->branch.raddr_a);
331 break;
332 }
333
334 if (instr->branch.ub) {
335 switch (instr->branch.bdu) {
336 case V3D_QPU_BRANCH_DEST_ABS:
337 fprintf(stderr, ", a:unif");
338 break;
339
340 case V3D_QPU_BRANCH_DEST_REL:
341 fprintf(stderr, ", r:unif");
342 break;
343
344 case V3D_QPU_BRANCH_DEST_LINK_REG:
345 fprintf(stderr, ", lri");
346 break;
347
348 case V3D_QPU_BRANCH_DEST_REGFILE:
349 fprintf(stderr, ", rf%d", instr->branch.raddr_a);
350 break;
351 }
352 }
353 break;
354 }
355
356 if (vir_has_uniform(inst)) {
357 fprintf(stderr, " (");
358 vir_dump_uniform(c->uniform_contents[inst->uniform],
359 c->uniform_data[inst->uniform]);
360 fprintf(stderr, ")");
361 }
362 }
363
364 void
vir_dump(struct v3d_compile * c)365 vir_dump(struct v3d_compile *c)
366 {
367 int ip = 0;
368 int pressure = 0;
369
370 vir_for_each_block(block, c) {
371 fprintf(stderr, "BLOCK %d:\n", block->index);
372 vir_for_each_inst(inst, block) {
373 if (c->live_intervals_valid) {
374 for (int i = 0; i < c->num_temps; i++) {
375 if (c->temp_start[i] == ip)
376 pressure++;
377 }
378
379 fprintf(stderr, "P%4d ", pressure);
380
381 bool first = true;
382
383 for (int i = 0; i < c->num_temps; i++) {
384 if (c->temp_start[i] != ip)
385 continue;
386
387 if (first) {
388 first = false;
389 } else {
390 fprintf(stderr, ", ");
391 }
392 if (BITSET_TEST(c->spillable, i))
393 fprintf(stderr, "S%4d", i);
394 else
395 fprintf(stderr, "U%4d", i);
396 }
397
398 if (first)
399 fprintf(stderr, " ");
400 else
401 fprintf(stderr, " ");
402 }
403
404 if (c->live_intervals_valid) {
405 bool first = true;
406
407 for (int i = 0; i < c->num_temps; i++) {
408 if (c->temp_end[i] != ip)
409 continue;
410
411 if (first) {
412 first = false;
413 } else {
414 fprintf(stderr, ", ");
415 }
416 fprintf(stderr, "E%4d", i);
417 pressure--;
418 }
419
420 if (first)
421 fprintf(stderr, " ");
422 else
423 fprintf(stderr, " ");
424 }
425
426 vir_dump_inst(c, inst);
427 fprintf(stderr, "\n");
428 ip++;
429 }
430 if (block->successors[1]) {
431 fprintf(stderr, "-> BLOCK %d, %d\n",
432 block->successors[0]->index,
433 block->successors[1]->index);
434 } else if (block->successors[0]) {
435 fprintf(stderr, "-> BLOCK %d\n",
436 block->successors[0]->index);
437 }
438 }
439 }
440