v3d: Implement a small immediates optimization, based on VC4's.
[mesa.git] / src / broadcom / compiler / vir_dump.c
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 static void
28 vir_print_reg(struct v3d_compile *c, const struct qinst *inst,
29 struct qreg reg)
30 {
31 static const char *files[] = {
32 [QFILE_TEMP] = "t",
33 [QFILE_UNIF] = "u",
34 [QFILE_TLB] = "tlb",
35 [QFILE_TLBU] = "tlbu",
36 };
37 static const char *quniform_names[] = {
38 [QUNIFORM_VIEWPORT_X_SCALE] = "vp_x_scale",
39 [QUNIFORM_VIEWPORT_Y_SCALE] = "vp_y_scale",
40 [QUNIFORM_VIEWPORT_Z_OFFSET] = "vp_z_offset",
41 [QUNIFORM_VIEWPORT_Z_SCALE] = "vp_z_scale",
42 };
43
44 switch (reg.file) {
45
46 case QFILE_NULL:
47 fprintf(stderr, "null");
48 break;
49
50 case QFILE_LOAD_IMM:
51 fprintf(stderr, "0x%08x (%f)", reg.index, uif(reg.index));
52 break;
53
54 case QFILE_REG:
55 fprintf(stderr, "rf%d", reg.index);
56 break;
57
58 case QFILE_MAGIC:
59 fprintf(stderr, "%s", v3d_qpu_magic_waddr_name(reg.index));
60 break;
61
62 case QFILE_SMALL_IMM: {
63 uint32_t unpacked;
64 bool ok = v3d_qpu_small_imm_unpack(c->devinfo,
65 inst->qpu.raddr_b,
66 &unpacked);
67 assert(ok); (void) ok;
68
69 if ((int)inst->qpu.raddr_b >= -16 &&
70 (int)inst->qpu.raddr_b <= 15)
71 fprintf(stderr, "%d", unpacked);
72 else
73 fprintf(stderr, "%f", uif(unpacked));
74 break;
75 }
76
77 case QFILE_VPM:
78 fprintf(stderr, "vpm%d.%d",
79 reg.index / 4, reg.index % 4);
80 break;
81
82 case QFILE_TLB:
83 case QFILE_TLBU:
84 fprintf(stderr, "%s", files[reg.file]);
85 break;
86
87 case QFILE_UNIF: {
88 enum quniform_contents contents = c->uniform_contents[reg.index];
89
90 fprintf(stderr, "%s%d", files[reg.file], reg.index);
91
92 switch (contents) {
93 case QUNIFORM_CONSTANT:
94 fprintf(stderr, " (0x%08x / %f)",
95 c->uniform_data[reg.index],
96 uif(c->uniform_data[reg.index]));
97 break;
98
99 case QUNIFORM_UNIFORM:
100 fprintf(stderr, " (push[%d])",
101 c->uniform_data[reg.index]);
102 break;
103
104 case QUNIFORM_TEXTURE_CONFIG_P1:
105 fprintf(stderr, " (tex[%d].p1)",
106 c->uniform_data[reg.index]);
107 break;
108
109 case QUNIFORM_TEXTURE_WIDTH:
110 fprintf(stderr, " (tex[%d].width)",
111 c->uniform_data[reg.index]);
112 break;
113 case QUNIFORM_TEXTURE_HEIGHT:
114 fprintf(stderr, " (tex[%d].height)",
115 c->uniform_data[reg.index]);
116 break;
117 case QUNIFORM_TEXTURE_DEPTH:
118 fprintf(stderr, " (tex[%d].depth)",
119 c->uniform_data[reg.index]);
120 break;
121 case QUNIFORM_TEXTURE_ARRAY_SIZE:
122 fprintf(stderr, " (tex[%d].array_size)",
123 c->uniform_data[reg.index]);
124 break;
125 case QUNIFORM_TEXTURE_LEVELS:
126 fprintf(stderr, " (tex[%d].levels)",
127 c->uniform_data[reg.index]);
128 break;
129
130 case QUNIFORM_UBO_ADDR:
131 fprintf(stderr, " (ubo[%d])",
132 c->uniform_data[reg.index]);
133 break;
134
135 default:
136 if (quniform_contents_is_texture_p0(contents)) {
137 fprintf(stderr, " (tex[%d].p0: 0x%08x)",
138 contents - QUNIFORM_TEXTURE_CONFIG_P0_0,
139 c->uniform_data[reg.index]);
140 } else if (contents < ARRAY_SIZE(quniform_names)) {
141 fprintf(stderr, " (%s)",
142 quniform_names[contents]);
143 } else {
144 fprintf(stderr, " (%d / 0x%08x)", contents,
145 c->uniform_data[reg.index]);
146 }
147 }
148
149 break;
150 }
151
152 default:
153 fprintf(stderr, "%s%d", files[reg.file], reg.index);
154 break;
155 }
156 }
157
158 static void
159 vir_dump_sig_addr(const struct v3d_device_info *devinfo,
160 const struct v3d_qpu_instr *instr)
161 {
162 if (devinfo->ver < 41)
163 return;
164
165 if (!instr->sig_magic)
166 fprintf(stderr, ".rf%d", instr->sig_addr);
167 else {
168 const char *name = v3d_qpu_magic_waddr_name(instr->sig_addr);
169 if (name)
170 fprintf(stderr, ".%s", name);
171 else
172 fprintf(stderr, ".UNKNOWN%d", instr->sig_addr);
173 }
174 }
175
176 static void
177 vir_dump_sig(struct v3d_compile *c, struct qinst *inst)
178 {
179 struct v3d_qpu_sig *sig = &inst->qpu.sig;
180
181 if (sig->thrsw)
182 fprintf(stderr, "; thrsw");
183 if (sig->ldvary) {
184 fprintf(stderr, "; ldvary");
185 vir_dump_sig_addr(c->devinfo, &inst->qpu);
186 }
187 if (sig->ldvpm)
188 fprintf(stderr, "; ldvpm");
189 if (sig->ldtmu) {
190 fprintf(stderr, "; ldtmu");
191 vir_dump_sig_addr(c->devinfo, &inst->qpu);
192 }
193 if (sig->ldtlb) {
194 fprintf(stderr, "; ldtlb");
195 vir_dump_sig_addr(c->devinfo, &inst->qpu);
196 }
197 if (sig->ldtlbu) {
198 fprintf(stderr, "; ldtlbu");
199 vir_dump_sig_addr(c->devinfo, &inst->qpu);
200 }
201 if (sig->ldunif)
202 fprintf(stderr, "; ldunif");
203 if (sig->ldunifrf) {
204 fprintf(stderr, "; ldunifrf");
205 vir_dump_sig_addr(c->devinfo, &inst->qpu);
206 }
207 if (sig->ldunifa)
208 fprintf(stderr, "; ldunifa");
209 if (sig->ldunifarf) {
210 fprintf(stderr, "; ldunifarf");
211 vir_dump_sig_addr(c->devinfo, &inst->qpu);
212 }
213 if (sig->wrtmuc)
214 fprintf(stderr, "; wrtmuc");
215 }
216
217 static void
218 vir_dump_alu(struct v3d_compile *c, struct qinst *inst)
219 {
220 struct v3d_qpu_instr *instr = &inst->qpu;
221 int nsrc = vir_get_non_sideband_nsrc(inst);
222 int sideband_nsrc = vir_get_nsrc(inst);
223 enum v3d_qpu_input_unpack unpack[2];
224
225 if (inst->qpu.alu.add.op != V3D_QPU_A_NOP) {
226 fprintf(stderr, "%s", v3d_qpu_add_op_name(instr->alu.add.op));
227 fprintf(stderr, "%s", v3d_qpu_cond_name(instr->flags.ac));
228 fprintf(stderr, "%s", v3d_qpu_pf_name(instr->flags.apf));
229 fprintf(stderr, "%s", v3d_qpu_uf_name(instr->flags.auf));
230 fprintf(stderr, " ");
231
232 vir_print_reg(c, inst, inst->dst);
233 fprintf(stderr, "%s", v3d_qpu_pack_name(instr->alu.add.output_pack));
234
235 unpack[0] = instr->alu.add.a_unpack;
236 unpack[1] = instr->alu.add.b_unpack;
237 } else {
238 fprintf(stderr, "%s", v3d_qpu_mul_op_name(instr->alu.mul.op));
239 fprintf(stderr, "%s", v3d_qpu_cond_name(instr->flags.mc));
240 fprintf(stderr, "%s", v3d_qpu_pf_name(instr->flags.mpf));
241 fprintf(stderr, "%s", v3d_qpu_uf_name(instr->flags.muf));
242 fprintf(stderr, " ");
243
244 vir_print_reg(c, inst, inst->dst);
245 fprintf(stderr, "%s", v3d_qpu_pack_name(instr->alu.mul.output_pack));
246
247 unpack[0] = instr->alu.mul.a_unpack;
248 unpack[1] = instr->alu.mul.b_unpack;
249 }
250
251 for (int i = 0; i < sideband_nsrc; i++) {
252 fprintf(stderr, ", ");
253 vir_print_reg(c, inst, inst->src[i]);
254 if (i < nsrc)
255 fprintf(stderr, "%s", v3d_qpu_unpack_name(unpack[i]));
256 }
257
258 vir_dump_sig(c, inst);
259 }
260
261 void
262 vir_dump_inst(struct v3d_compile *c, struct qinst *inst)
263 {
264 struct v3d_qpu_instr *instr = &inst->qpu;
265
266 switch (inst->qpu.type) {
267 case V3D_QPU_INSTR_TYPE_ALU:
268 vir_dump_alu(c, inst);
269 break;
270 case V3D_QPU_INSTR_TYPE_BRANCH:
271 fprintf(stderr, "b");
272 if (instr->branch.ub)
273 fprintf(stderr, "u");
274
275 fprintf(stderr, "%s",
276 v3d_qpu_branch_cond_name(instr->branch.cond));
277 fprintf(stderr, "%s", v3d_qpu_msfign_name(instr->branch.msfign));
278
279 switch (instr->branch.bdi) {
280 case V3D_QPU_BRANCH_DEST_ABS:
281 fprintf(stderr, " zero_addr+0x%08x", instr->branch.offset);
282 break;
283
284 case V3D_QPU_BRANCH_DEST_REL:
285 fprintf(stderr, " %d", instr->branch.offset);
286 break;
287
288 case V3D_QPU_BRANCH_DEST_LINK_REG:
289 fprintf(stderr, " lri");
290 break;
291
292 case V3D_QPU_BRANCH_DEST_REGFILE:
293 fprintf(stderr, " rf%d", instr->branch.raddr_a);
294 break;
295 }
296
297 if (instr->branch.ub) {
298 switch (instr->branch.bdu) {
299 case V3D_QPU_BRANCH_DEST_ABS:
300 fprintf(stderr, ", a:unif");
301 break;
302
303 case V3D_QPU_BRANCH_DEST_REL:
304 fprintf(stderr, ", r:unif");
305 break;
306
307 case V3D_QPU_BRANCH_DEST_LINK_REG:
308 fprintf(stderr, ", lri");
309 break;
310
311 case V3D_QPU_BRANCH_DEST_REGFILE:
312 fprintf(stderr, ", rf%d", instr->branch.raddr_a);
313 break;
314 }
315 }
316
317 if (vir_has_implicit_uniform(inst)) {
318 fprintf(stderr, " ");
319 vir_print_reg(c, inst, inst->src[vir_get_implicit_uniform_src(inst)]);
320 }
321
322 break;
323 }
324 }
325
326 void
327 vir_dump(struct v3d_compile *c)
328 {
329 int ip = 0;
330
331 vir_for_each_block(block, c) {
332 fprintf(stderr, "BLOCK %d:\n", block->index);
333 vir_for_each_inst(inst, block) {
334 if (c->live_intervals_valid) {
335 bool first = true;
336
337 for (int i = 0; i < c->num_temps; i++) {
338 if (c->temp_start[i] != ip)
339 continue;
340
341 if (first) {
342 first = false;
343 } else {
344 fprintf(stderr, ", ");
345 }
346 fprintf(stderr, "S%4d", i);
347 }
348
349 if (first)
350 fprintf(stderr, " ");
351 else
352 fprintf(stderr, " ");
353 }
354
355 if (c->live_intervals_valid) {
356 bool first = true;
357
358 for (int i = 0; i < c->num_temps; i++) {
359 if (c->temp_end[i] != ip)
360 continue;
361
362 if (first) {
363 first = false;
364 } else {
365 fprintf(stderr, ", ");
366 }
367 fprintf(stderr, "E%4d", i);
368 }
369
370 if (first)
371 fprintf(stderr, " ");
372 else
373 fprintf(stderr, " ");
374 }
375
376 vir_dump_inst(c, inst);
377 fprintf(stderr, "\n");
378 ip++;
379 }
380 if (block->successors[1]) {
381 fprintf(stderr, "-> BLOCK %d, %d\n",
382 block->successors[0]->index,
383 block->successors[1]->index);
384 } else if (block->successors[0]) {
385 fprintf(stderr, "-> BLOCK %d\n",
386 block->successors[0]->index);
387 }
388 }
389 }