src/mesa/drivers/dri/i965/test_fs_saturate_propagation.cpp

   1 /*
   2  * Copyright © 2015 Intel Corporation
   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 <gtest/gtest.h>
  25 #include "brw_fs.h"
  26 #include "brw_cfg.h"
  27 #include "program/program.h"
  28
  29 class saturate_propagation_test : public ::testing::Test {
  30    virtual void SetUp();
  31
  32 public:
  33    struct brw_context *brw;
  34    struct gl_context *ctx;
  35    struct brw_wm_prog_data *prog_data;
  36    struct gl_shader_program *shader_prog;
  37    struct brw_fragment_program *fp;
  38    fs_visitor *v;
  39 };
  40
  41 class saturate_propagation_fs_visitor : public fs_visitor
  42 {
  43 public:
  44    saturate_propagation_fs_visitor(struct brw_context *brw,
  45                                    struct brw_wm_prog_data *prog_data,
  46                                    struct gl_shader_program *shader_prog)
  47       : fs_visitor(brw, NULL, NULL, prog_data, shader_prog, NULL, 8) {}
  48 };
  49
  50
  51 void saturate_propagation_test::SetUp()
  52 {
  53    brw = (struct brw_context *)calloc(1, sizeof(*brw));
  54    ctx = &brw->ctx;
  55
  56    fp = ralloc(NULL, struct brw_fragment_program);
  57    prog_data = ralloc(NULL, struct brw_wm_prog_data);
  58    shader_prog = ralloc(NULL, struct gl_shader_program);
  59
  60    v = new saturate_propagation_fs_visitor(brw, prog_data, shader_prog);
  61
  62    _mesa_init_fragment_program(ctx, &fp->program, GL_FRAGMENT_SHADER, 0);
  63
  64    brw->gen = 4;
  65 }
  66
  67 static fs_inst *
  68 instruction(bblock_t *block, int num)
  69 {
  70    fs_inst *inst = (fs_inst *)block->start();
  71    for (int i = 0; i < num; i++) {
  72       inst = (fs_inst *)inst->next;
  73    }
  74    return inst;
  75 }
  76
  77 static bool
  78 saturate_propagation(fs_visitor *v)
  79 {
  80    const bool print = false;
  81
  82    if (print) {
  83       fprintf(stderr, "= Before =\n");
  84       v->cfg->dump(v);
  85    }
  86
  87    bool ret = v->opt_saturate_propagation();
  88
  89    if (print) {
  90       fprintf(stderr, "\n= After =\n");
  91       v->cfg->dump(v);
  92    }
  93
  94    return ret;
  95 }
  96
  97 TEST_F(saturate_propagation_test, basic)
  98 {
  99    fs_reg dst0 = v->vgrf(glsl_type::float_type);
 100    fs_reg dst1 = v->vgrf(glsl_type::float_type);
 101    fs_reg src0 = v->vgrf(glsl_type::float_type);
 102    fs_reg src1 = v->vgrf(glsl_type::float_type);
 103    v->emit(BRW_OPCODE_ADD, dst0, src0, src1);
 104    v->emit(BRW_OPCODE_MOV, dst1, dst0)
 105       ->saturate = true;
 106
 107    /* = Before =
 108     *
 109     * 0: add(8)        dst0  src0  src1
 110     * 1: mov.sat(8)    dst1  dst0
 111     *
 112     * = After =
 113     * 0: add.sat(8)    dst0  src0  src1
 114     * 1: mov(8)        dst1  dst0
 115     */
 116
 117    v->calculate_cfg();
 118    bblock_t *block0 = v->cfg->blocks[0];
 119
 120    EXPECT_EQ(0, block0->start_ip);
 121    EXPECT_EQ(1, block0->end_ip);
 122
 123    EXPECT_TRUE(saturate_propagation(v));
 124    EXPECT_EQ(0, block0->start_ip);
 125    EXPECT_EQ(1, block0->end_ip);
 126    EXPECT_EQ(BRW_OPCODE_ADD, instruction(block0, 0)->opcode);
 127    EXPECT_TRUE(instruction(block0, 0)->saturate);
 128    EXPECT_EQ(BRW_OPCODE_MOV, instruction(block0, 1)->opcode);
 129    EXPECT_FALSE(instruction(block0, 1)->saturate);
 130 }
 131
 132 TEST_F(saturate_propagation_test, other_non_saturated_use)
 133 {
 134    fs_reg dst0 = v->vgrf(glsl_type::float_type);
 135    fs_reg dst1 = v->vgrf(glsl_type::float_type);
 136    fs_reg dst2 = v->vgrf(glsl_type::float_type);
 137    fs_reg src0 = v->vgrf(glsl_type::float_type);
 138    fs_reg src1 = v->vgrf(glsl_type::float_type);
 139    v->emit(BRW_OPCODE_ADD, dst0, src0, src1);
 140    v->emit(BRW_OPCODE_MOV, dst1, dst0)
 141       ->saturate = true;
 142    v->emit(BRW_OPCODE_ADD, dst2, dst0, src0);
 143
 144    /* = Before =
 145     *
 146     * 0: add(8)        dst0  src0  src1
 147     * 1: mov.sat(8)    dst1  dst0
 148     * 2: add(8)        dst2  dst0  src0
 149     *
 150     * = After =
 151     * (no changes)
 152     */
 153
 154    v->calculate_cfg();
 155    bblock_t *block0 = v->cfg->blocks[0];
 156
 157    EXPECT_EQ(0, block0->start_ip);
 158    EXPECT_EQ(2, block0->end_ip);
 159
 160    EXPECT_FALSE(saturate_propagation(v));
 161    EXPECT_EQ(0, block0->start_ip);
 162    EXPECT_EQ(2, block0->end_ip);
 163    EXPECT_EQ(BRW_OPCODE_ADD, instruction(block0, 0)->opcode);
 164    EXPECT_FALSE(instruction(block0, 0)->saturate);
 165    EXPECT_EQ(BRW_OPCODE_MOV, instruction(block0, 1)->opcode);
 166    EXPECT_TRUE(instruction(block0, 1)->saturate);
 167    EXPECT_EQ(BRW_OPCODE_ADD, instruction(block0, 2)->opcode);
 168 }
 169
 170 TEST_F(saturate_propagation_test, predicated_instruction)
 171 {
 172    fs_reg dst0 = v->vgrf(glsl_type::float_type);
 173    fs_reg dst1 = v->vgrf(glsl_type::float_type);
 174    fs_reg src0 = v->vgrf(glsl_type::float_type);
 175    fs_reg src1 = v->vgrf(glsl_type::float_type);
 176    v->emit(BRW_OPCODE_ADD, dst0, src0, src1)
 177       ->predicate = BRW_PREDICATE_NORMAL;
 178    v->emit(BRW_OPCODE_MOV, dst1, dst0)
 179       ->saturate = true;
 180
 181    /* = Before =
 182     *
 183     * 0: (+f0) add(8)  dst0  src0  src1
 184     * 1: mov.sat(8)    dst1  dst0
 185     *
 186     * = After =
 187     * (no changes)
 188     */
 189
 190    v->calculate_cfg();
 191    bblock_t *block0 = v->cfg->blocks[0];
 192
 193    EXPECT_EQ(0, block0->start_ip);
 194    EXPECT_EQ(1, block0->end_ip);
 195
 196    EXPECT_FALSE(saturate_propagation(v));
 197    EXPECT_EQ(0, block0->start_ip);
 198    EXPECT_EQ(1, block0->end_ip);
 199    EXPECT_EQ(BRW_OPCODE_ADD, instruction(block0, 0)->opcode);
 200    EXPECT_FALSE(instruction(block0, 0)->saturate);
 201    EXPECT_EQ(BRW_OPCODE_MOV, instruction(block0, 1)->opcode);
 202    EXPECT_TRUE(instruction(block0, 1)->saturate);
 203 }
 204
 205 TEST_F(saturate_propagation_test, neg_mov_sat)
 206 {
 207    fs_reg dst0 = v->vgrf(glsl_type::float_type);
 208    fs_reg dst1 = v->vgrf(glsl_type::float_type);
 209    fs_reg src0 = v->vgrf(glsl_type::float_type);
 210    fs_reg src1 = v->vgrf(glsl_type::float_type);
 211    v->emit(BRW_OPCODE_ADD, dst0, src0, src1);
 212    dst0.negate = true;
 213    v->emit(BRW_OPCODE_MOV, dst1, dst0)
 214       ->saturate = true;
 215
 216    /* = Before =
 217     *
 218     * 0: add(8)        dst0  src0  src1
 219     * 1: mov.sat(8)    dst1  -dst0
 220     *
 221     * = After =
 222     * (no changes)
 223     */
 224
 225    v->calculate_cfg();
 226    bblock_t *block0 = v->cfg->blocks[0];
 227
 228    EXPECT_EQ(0, block0->start_ip);
 229    EXPECT_EQ(1, block0->end_ip);
 230
 231    EXPECT_FALSE(saturate_propagation(v));
 232    EXPECT_EQ(0, block0->start_ip);
 233    EXPECT_EQ(1, block0->end_ip);
 234    EXPECT_EQ(BRW_OPCODE_ADD, instruction(block0, 0)->opcode);
 235    EXPECT_FALSE(instruction(block0, 0)->saturate);
 236    EXPECT_EQ(BRW_OPCODE_MOV, instruction(block0, 1)->opcode);
 237    EXPECT_TRUE(instruction(block0, 1)->saturate);
 238 }
 239
 240 TEST_F(saturate_propagation_test, abs_mov_sat)
 241 {
 242    fs_reg dst0 = v->vgrf(glsl_type::float_type);
 243    fs_reg dst1 = v->vgrf(glsl_type::float_type);
 244    fs_reg src0 = v->vgrf(glsl_type::float_type);
 245    fs_reg src1 = v->vgrf(glsl_type::float_type);
 246    v->emit(BRW_OPCODE_ADD, dst0, src0, src1);
 247    dst0.abs = true;
 248    v->emit(BRW_OPCODE_MOV, dst1, dst0)
 249       ->saturate = true;
 250
 251    /* = Before =
 252     *
 253     * 0: add(8)        dst0  src0  src1
 254     * 1: mov.sat(8)    dst1  (abs)dst0
 255     *
 256     * = After =
 257     * (no changes)
 258     */
 259
 260    v->calculate_cfg();
 261    bblock_t *block0 = v->cfg->blocks[0];
 262
 263    EXPECT_EQ(0, block0->start_ip);
 264    EXPECT_EQ(1, block0->end_ip);
 265
 266    EXPECT_FALSE(saturate_propagation(v));
 267    EXPECT_EQ(0, block0->start_ip);
 268    EXPECT_EQ(1, block0->end_ip);
 269    EXPECT_EQ(BRW_OPCODE_ADD, instruction(block0, 0)->opcode);
 270    EXPECT_FALSE(instruction(block0, 0)->saturate);
 271    EXPECT_EQ(BRW_OPCODE_MOV, instruction(block0, 1)->opcode);
 272    EXPECT_TRUE(instruction(block0, 1)->saturate);
 273 }
 274
 275 TEST_F(saturate_propagation_test, producer_saturates)
 276 {
 277    fs_reg dst0 = v->vgrf(glsl_type::float_type);
 278    fs_reg dst1 = v->vgrf(glsl_type::float_type);
 279    fs_reg dst2 = v->vgrf(glsl_type::float_type);
 280    fs_reg src0 = v->vgrf(glsl_type::float_type);
 281    fs_reg src1 = v->vgrf(glsl_type::float_type);
 282    v->emit(BRW_OPCODE_ADD, dst0, src0, src1)
 283       ->saturate = true;
 284    v->emit(BRW_OPCODE_MOV, dst1, dst0)
 285       ->saturate = true;
 286    v->emit(BRW_OPCODE_MOV, dst2, dst0);
 287
 288    /* = Before =
 289     *
 290     * 0: add.sat(8)    dst0  src0  src1
 291     * 1: mov.sat(8)    dst1  dst0
 292     * 2: mov(8)        dst2  dst0
 293     *
 294     * = After =
 295     * 0: add.sat(8)    dst0  src0  src1
 296     * 1: mov(8)        dst1  dst0
 297     * 2: mov(8)        dst2  dst0
 298     */
 299
 300    v->calculate_cfg();
 301    bblock_t *block0 = v->cfg->blocks[0];
 302
 303    EXPECT_EQ(0, block0->start_ip);
 304    EXPECT_EQ(2, block0->end_ip);
 305
 306    EXPECT_TRUE(saturate_propagation(v));
 307    EXPECT_EQ(0, block0->start_ip);
 308    EXPECT_EQ(2, block0->end_ip);
 309    EXPECT_EQ(BRW_OPCODE_ADD, instruction(block0, 0)->opcode);
 310    EXPECT_TRUE(instruction(block0, 0)->saturate);
 311    EXPECT_EQ(BRW_OPCODE_MOV, instruction(block0, 1)->opcode);
 312    EXPECT_FALSE(instruction(block0, 1)->saturate);
 313 }
 314
 315 TEST_F(saturate_propagation_test, intervening_saturating_copy)
 316 {
 317    fs_reg dst0 = v->vgrf(glsl_type::float_type);
 318    fs_reg dst1 = v->vgrf(glsl_type::float_type);
 319    fs_reg dst2 = v->vgrf(glsl_type::float_type);
 320    fs_reg src0 = v->vgrf(glsl_type::float_type);
 321    fs_reg src1 = v->vgrf(glsl_type::float_type);
 322    v->emit(BRW_OPCODE_ADD, dst0, src0, src1);
 323    v->emit(BRW_OPCODE_MOV, dst1, dst0)
 324       ->saturate = true;
 325    v->emit(BRW_OPCODE_MOV, dst2, dst0)
 326       ->saturate = true;
 327
 328    /* = Before =
 329     *
 330     * 0: add(8)    dst0  src0  src1
 331     * 1: mov.sat(8)    dst1  dst0
 332     * 2: mov.sat(8)    dst2  dst0
 333     *
 334     * = After =
 335     * 0: add.sat(8)    dst0  src0  src1
 336     * 1: mov(8)        dst1  dst0
 337     * 2: mov(8)        dst2  dst0
 338     */
 339
 340    v->calculate_cfg();
 341    bblock_t *block0 = v->cfg->blocks[0];
 342
 343    EXPECT_EQ(0, block0->start_ip);
 344    EXPECT_EQ(2, block0->end_ip);
 345
 346    EXPECT_TRUE(saturate_propagation(v));
 347    EXPECT_EQ(0, block0->start_ip);
 348    EXPECT_EQ(2, block0->end_ip);
 349    EXPECT_EQ(BRW_OPCODE_ADD, instruction(block0, 0)->opcode);
 350    EXPECT_TRUE(instruction(block0, 0)->saturate);
 351    EXPECT_EQ(BRW_OPCODE_MOV, instruction(block0, 1)->opcode);
 352    EXPECT_FALSE(instruction(block0, 1)->saturate);
 353    EXPECT_EQ(BRW_OPCODE_MOV, instruction(block0, 2)->opcode);
 354    EXPECT_FALSE(instruction(block0, 2)->saturate);
 355 }
 356
 357 TEST_F(saturate_propagation_test, intervening_dest_write)
 358 {
 359    fs_reg dst0 = v->vgrf(glsl_type::vec4_type);
 360    fs_reg dst1 = v->vgrf(glsl_type::float_type);
 361    fs_reg src0 = v->vgrf(glsl_type::float_type);
 362    fs_reg src1 = v->vgrf(glsl_type::float_type);
 363    fs_reg src2 = v->vgrf(glsl_type::vec2_type);
 364    v->emit(BRW_OPCODE_ADD, offset(dst0, 2), src0, src1);
 365    v->emit(SHADER_OPCODE_TEX, dst0, src2)
 366       ->regs_written = 4;
 367    v->emit(BRW_OPCODE_MOV, dst1, offset(dst0, 2))
 368       ->saturate = true;
 369
 370    /* = Before =
 371     *
 372     * 0: add(8)        dst0+2  src0    src1
 373     * 1: tex(8) rlen 4 dst0+0  src2
 374     * 2: mov.sat(8)    dst1    dst0+2
 375     *
 376     * = After =
 377     * (no changes)
 378     */
 379
 380    v->calculate_cfg();
 381    bblock_t *block0 = v->cfg->blocks[0];
 382
 383    EXPECT_EQ(0, block0->start_ip);
 384    EXPECT_EQ(2, block0->end_ip);
 385
 386    EXPECT_FALSE(saturate_propagation(v));
 387    EXPECT_EQ(0, block0->start_ip);
 388    EXPECT_EQ(2, block0->end_ip);
 389    EXPECT_EQ(BRW_OPCODE_ADD, instruction(block0, 0)->opcode);
 390    EXPECT_FALSE(instruction(block0, 0)->saturate);
 391    EXPECT_EQ(SHADER_OPCODE_TEX, instruction(block0, 1)->opcode);
 392    EXPECT_FALSE(instruction(block0, 0)->saturate);
 393    EXPECT_EQ(BRW_OPCODE_MOV, instruction(block0, 2)->opcode);
 394    EXPECT_TRUE(instruction(block0, 2)->saturate);
 395 }
 396
 397 TEST_F(saturate_propagation_test, mul_neg_mov_sat_mov_sat)
 398 {
 399    fs_reg dst0 = v->vgrf(glsl_type::float_type);
 400    fs_reg dst1 = v->vgrf(glsl_type::float_type);
 401    fs_reg dst2 = v->vgrf(glsl_type::float_type);
 402    fs_reg src0 = v->vgrf(glsl_type::float_type);
 403    fs_reg src1 = v->vgrf(glsl_type::float_type);
 404    v->emit(BRW_OPCODE_MUL, dst0, src0, src1);
 405    dst0.negate = true;
 406    v->emit(BRW_OPCODE_MOV, dst1, dst0)
 407       ->saturate = true;
 408    dst0.negate = false;
 409    v->emit(BRW_OPCODE_MOV, dst2, dst0)
 410       ->saturate = true;
 411
 412    /* = Before =
 413     *
 414     * 0: mul(8)        dst0  src0  src1
 415     * 1: mov.sat(8)    dst1  -dst0
 416     * 2: mov.sat(8)    dst2  dst0
 417     *
 418     * = After =
 419     * (no changes)
 420     */
 421
 422    v->calculate_cfg();
 423    bblock_t *block0 = v->cfg->blocks[0];
 424
 425    EXPECT_EQ(0, block0->start_ip);
 426    EXPECT_EQ(2, block0->end_ip);
 427
 428    EXPECT_FALSE(saturate_propagation(v));
 429    EXPECT_EQ(0, block0->start_ip);
 430    EXPECT_EQ(2, block0->end_ip);
 431    EXPECT_EQ(BRW_OPCODE_MUL, instruction(block0, 0)->opcode);
 432    EXPECT_FALSE(instruction(block0, 0)->saturate);
 433    EXPECT_FALSE(instruction(block0, 0)->src[1].negate);
 434    EXPECT_EQ(BRW_OPCODE_MOV, instruction(block0, 1)->opcode);
 435    EXPECT_TRUE(instruction(block0, 1)->saturate);
 436    EXPECT_TRUE(instruction(block0, 1)->src[0].negate);
 437    EXPECT_EQ(BRW_OPCODE_MOV, instruction(block0, 2)->opcode);
 438    EXPECT_TRUE(instruction(block0, 2)->saturate);
 439 }