1 /* 2 * Copyright 2015 Advanced Micro Devices, Inc. 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 shall be included in 12 * all copies or substantial portions of the Software. 13 * 14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR 18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, 19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR 20 * OTHER DEALINGS IN THE SOFTWARE. 21 * 22 */ 23 24 #include <stdio.h> 25 #include <inttypes.h> 26 27 #include "CUnit/Basic.h" 28 29 #include "util_math.h" 30 31 #include "amdgpu_test.h" 32 #include "amdgpu_drm.h" 33 #include "amdgpu_internal.h" 34 35 #include "vce_ib.h" 36 #include "frame.h" 37 38 #define IB_SIZE 4096 39 #define MAX_RESOURCES 16 40 #define FW_53_0_03 ((53 << 24) | (0 << 16) | (03 << 8)) 41 42 struct amdgpu_vce_bo { 43 amdgpu_bo_handle handle; 44 amdgpu_va_handle va_handle; 45 uint64_t addr; 46 uint64_t size; 47 uint8_t *ptr; 48 }; 49 50 struct amdgpu_vce_encode { 51 unsigned width; 52 unsigned height; 53 struct amdgpu_vce_bo vbuf; 54 struct amdgpu_vce_bo bs[2]; 55 struct amdgpu_vce_bo fb[2]; 56 struct amdgpu_vce_bo cpb; 57 unsigned ib_len; 58 bool two_instance; 59 struct amdgpu_vce_bo mvrefbuf; 60 struct amdgpu_vce_bo mvb; 61 unsigned mvbuf_size; 62 }; 63 64 static amdgpu_device_handle device_handle; 65 static uint32_t major_version; 66 static uint32_t minor_version; 67 static uint32_t family_id; 68 static uint32_t vce_harvest_config; 69 static uint32_t chip_rev; 70 static uint32_t chip_id; 71 static uint32_t ids_flags; 72 static bool is_mv_supported = true; 73 74 static amdgpu_context_handle context_handle; 75 static amdgpu_bo_handle ib_handle; 76 static amdgpu_va_handle ib_va_handle; 77 static uint64_t ib_mc_address; 78 static uint32_t *ib_cpu; 79 80 static struct amdgpu_vce_encode enc; 81 static amdgpu_bo_handle resources[MAX_RESOURCES]; 82 static unsigned num_resources; 83 84 static void amdgpu_cs_vce_create(void); 85 static void amdgpu_cs_vce_encode(void); 86 static void amdgpu_cs_vce_encode_mv(void); 87 static void amdgpu_cs_vce_destroy(void); 88 89 CU_TestInfo vce_tests[] = { 90 { "VCE create", amdgpu_cs_vce_create }, 91 { "VCE encode", amdgpu_cs_vce_encode }, 92 { "VCE MV dump", amdgpu_cs_vce_encode_mv }, 93 { "VCE destroy", amdgpu_cs_vce_destroy }, 94 CU_TEST_INFO_NULL, 95 }; 96 97 CU_BOOL suite_vce_tests_enable(void) 98 { 99 uint32_t version, feature; 100 CU_BOOL ret_mv = CU_FALSE; 101 102 if (amdgpu_device_initialize(drm_amdgpu[0], &major_version, 103 &minor_version, &device_handle)) 104 return CU_FALSE; 105 106 family_id = device_handle->info.family_id; 107 chip_rev = device_handle->info.chip_rev; 108 chip_id = device_handle->info.chip_external_rev; 109 ids_flags = device_handle->info.ids_flags; 110 111 amdgpu_query_firmware_version(device_handle, AMDGPU_INFO_FW_VCE, 0, 112 0, &version, &feature); 113 114 if (amdgpu_device_deinitialize(device_handle)) 115 return CU_FALSE; 116 117 if (family_id >= AMDGPU_FAMILY_RV || family_id == AMDGPU_FAMILY_SI || 118 asic_is_gfx_pipe_removed(family_id, chip_id, chip_rev)) { 119 printf("\n\nThe ASIC NOT support VCE, suite disabled\n"); 120 return CU_FALSE; 121 } 122 123 if (!(chip_id == (chip_rev + 0x3C) || /* FIJI */ 124 chip_id == (chip_rev + 0x50) || /* Polaris 10*/ 125 chip_id == (chip_rev + 0x5A) || /* Polaris 11*/ 126 chip_id == (chip_rev + 0x64) || /* Polaris 12*/ 127 (family_id >= AMDGPU_FAMILY_AI && !ids_flags))) /* dGPU > Polaris */ 128 printf("\n\nThe ASIC NOT support VCE MV, suite disabled\n"); 129 else if (FW_53_0_03 > version) 130 printf("\n\nThe ASIC FW version NOT support VCE MV, suite disabled\n"); 131 else 132 ret_mv = CU_TRUE; 133 134 if (ret_mv == CU_FALSE) { 135 amdgpu_set_test_active("VCE Tests", "VCE MV dump", ret_mv); 136 is_mv_supported = false; 137 } 138 139 return CU_TRUE; 140 } 141 142 int suite_vce_tests_init(void) 143 { 144 int r; 145 146 r = amdgpu_device_initialize(drm_amdgpu[0], &major_version, 147 &minor_version, &device_handle); 148 if (r) { 149 if ((r == -EACCES) && (errno == EACCES)) 150 printf("\n\nError:%s. " 151 "Hint:Try to run this test program as root.", 152 strerror(errno)); 153 154 return CUE_SINIT_FAILED; 155 } 156 157 family_id = device_handle->info.family_id; 158 vce_harvest_config = device_handle->info.vce_harvest_config; 159 160 r = amdgpu_cs_ctx_create(device_handle, &context_handle); 161 if (r) 162 return CUE_SINIT_FAILED; 163 164 r = amdgpu_bo_alloc_and_map(device_handle, IB_SIZE, 4096, 165 AMDGPU_GEM_DOMAIN_GTT, 0, 166 &ib_handle, (void**)&ib_cpu, 167 &ib_mc_address, &ib_va_handle); 168 if (r) 169 return CUE_SINIT_FAILED; 170 171 memset(&enc, 0, sizeof(struct amdgpu_vce_encode)); 172 173 return CUE_SUCCESS; 174 } 175 176 int suite_vce_tests_clean(void) 177 { 178 int r; 179 180 r = amdgpu_bo_unmap_and_free(ib_handle, ib_va_handle, 181 ib_mc_address, IB_SIZE); 182 if (r) 183 return CUE_SCLEAN_FAILED; 184 185 r = amdgpu_cs_ctx_free(context_handle); 186 if (r) 187 return CUE_SCLEAN_FAILED; 188 189 r = amdgpu_device_deinitialize(device_handle); 190 if (r) 191 return CUE_SCLEAN_FAILED; 192 193 return CUE_SUCCESS; 194 } 195 196 static int submit(unsigned ndw, unsigned ip) 197 { 198 struct amdgpu_cs_request ibs_request = {0}; 199 struct amdgpu_cs_ib_info ib_info = {0}; 200 struct amdgpu_cs_fence fence_status = {0}; 201 uint32_t expired; 202 int r; 203 204 ib_info.ib_mc_address = ib_mc_address; 205 ib_info.size = ndw; 206 207 ibs_request.ip_type = ip; 208 209 r = amdgpu_bo_list_create(device_handle, num_resources, resources, 210 NULL, &ibs_request.resources); 211 if (r) 212 return r; 213 214 ibs_request.number_of_ibs = 1; 215 ibs_request.ibs = &ib_info; 216 ibs_request.fence_info.handle = NULL; 217 218 r = amdgpu_cs_submit(context_handle, 0, &ibs_request, 1); 219 if (r) 220 return r; 221 222 r = amdgpu_bo_list_destroy(ibs_request.resources); 223 if (r) 224 return r; 225 226 fence_status.context = context_handle; 227 fence_status.ip_type = ip; 228 fence_status.fence = ibs_request.seq_no; 229 230 r = amdgpu_cs_query_fence_status(&fence_status, 231 AMDGPU_TIMEOUT_INFINITE, 232 0, &expired); 233 if (r) 234 return r; 235 236 return 0; 237 } 238 239 static void alloc_resource(struct amdgpu_vce_bo *vce_bo, unsigned size, unsigned domain) 240 { 241 struct amdgpu_bo_alloc_request req = {0}; 242 amdgpu_bo_handle buf_handle; 243 amdgpu_va_handle va_handle; 244 uint64_t va = 0; 245 int r; 246 247 req.alloc_size = ALIGN(size, 4096); 248 req.preferred_heap = domain; 249 r = amdgpu_bo_alloc(device_handle, &req, &buf_handle); 250 CU_ASSERT_EQUAL(r, 0); 251 r = amdgpu_va_range_alloc(device_handle, 252 amdgpu_gpu_va_range_general, 253 req.alloc_size, 1, 0, &va, 254 &va_handle, 0); 255 CU_ASSERT_EQUAL(r, 0); 256 r = amdgpu_bo_va_op(buf_handle, 0, req.alloc_size, va, 0, 257 AMDGPU_VA_OP_MAP); 258 CU_ASSERT_EQUAL(r, 0); 259 vce_bo->addr = va; 260 vce_bo->handle = buf_handle; 261 vce_bo->size = req.alloc_size; 262 vce_bo->va_handle = va_handle; 263 r = amdgpu_bo_cpu_map(vce_bo->handle, (void **)&vce_bo->ptr); 264 CU_ASSERT_EQUAL(r, 0); 265 memset(vce_bo->ptr, 0, size); 266 r = amdgpu_bo_cpu_unmap(vce_bo->handle); 267 CU_ASSERT_EQUAL(r, 0); 268 } 269 270 static void free_resource(struct amdgpu_vce_bo *vce_bo) 271 { 272 int r; 273 274 r = amdgpu_bo_va_op(vce_bo->handle, 0, vce_bo->size, 275 vce_bo->addr, 0, AMDGPU_VA_OP_UNMAP); 276 CU_ASSERT_EQUAL(r, 0); 277 278 r = amdgpu_va_range_free(vce_bo->va_handle); 279 CU_ASSERT_EQUAL(r, 0); 280 281 r = amdgpu_bo_free(vce_bo->handle); 282 CU_ASSERT_EQUAL(r, 0); 283 memset(vce_bo, 0, sizeof(*vce_bo)); 284 } 285 286 static void amdgpu_cs_vce_create(void) 287 { 288 unsigned align = (family_id >= AMDGPU_FAMILY_AI) ? 256 : 16; 289 int len, r; 290 291 enc.width = vce_create[6]; 292 enc.height = vce_create[7]; 293 294 num_resources = 0; 295 alloc_resource(&enc.fb[0], 4096, AMDGPU_GEM_DOMAIN_GTT); 296 resources[num_resources++] = enc.fb[0].handle; 297 resources[num_resources++] = ib_handle; 298 299 len = 0; 300 memcpy(ib_cpu, vce_session, sizeof(vce_session)); 301 len += sizeof(vce_session) / 4; 302 memcpy((ib_cpu + len), vce_taskinfo, sizeof(vce_taskinfo)); 303 len += sizeof(vce_taskinfo) / 4; 304 memcpy((ib_cpu + len), vce_create, sizeof(vce_create)); 305 ib_cpu[len + 8] = ALIGN(enc.width, align); 306 ib_cpu[len + 9] = ALIGN(enc.width, align); 307 if (is_mv_supported == true) {/* disableTwoInstance */ 308 if (family_id >= AMDGPU_FAMILY_AI) 309 ib_cpu[len + 11] = 0x01000001; 310 else 311 ib_cpu[len + 11] = 0x01000201; 312 } 313 len += sizeof(vce_create) / 4; 314 memcpy((ib_cpu + len), vce_feedback, sizeof(vce_feedback)); 315 ib_cpu[len + 2] = enc.fb[0].addr >> 32; 316 ib_cpu[len + 3] = enc.fb[0].addr; 317 len += sizeof(vce_feedback) / 4; 318 319 r = submit(len, AMDGPU_HW_IP_VCE); 320 CU_ASSERT_EQUAL(r, 0); 321 322 free_resource(&enc.fb[0]); 323 } 324 325 static void amdgpu_cs_vce_config(void) 326 { 327 int len = 0, r; 328 329 memcpy((ib_cpu + len), vce_session, sizeof(vce_session)); 330 len += sizeof(vce_session) / 4; 331 memcpy((ib_cpu + len), vce_taskinfo, sizeof(vce_taskinfo)); 332 ib_cpu[len + 3] = 2; 333 ib_cpu[len + 6] = 0xffffffff; 334 len += sizeof(vce_taskinfo) / 4; 335 memcpy((ib_cpu + len), vce_rate_ctrl, sizeof(vce_rate_ctrl)); 336 len += sizeof(vce_rate_ctrl) / 4; 337 memcpy((ib_cpu + len), vce_config_ext, sizeof(vce_config_ext)); 338 len += sizeof(vce_config_ext) / 4; 339 memcpy((ib_cpu + len), vce_motion_est, sizeof(vce_motion_est)); 340 len += sizeof(vce_motion_est) / 4; 341 memcpy((ib_cpu + len), vce_rdo, sizeof(vce_rdo)); 342 len += sizeof(vce_rdo) / 4; 343 memcpy((ib_cpu + len), vce_pic_ctrl, sizeof(vce_pic_ctrl)); 344 if (is_mv_supported == true) 345 ib_cpu[len + 27] = 0x00000001; /* encSliceMode */ 346 len += sizeof(vce_pic_ctrl) / 4; 347 348 r = submit(len, AMDGPU_HW_IP_VCE); 349 CU_ASSERT_EQUAL(r, 0); 350 } 351 352 static void amdgpu_cs_vce_encode_idr(struct amdgpu_vce_encode *enc) 353 { 354 355 uint64_t luma_offset, chroma_offset; 356 unsigned align = (family_id >= AMDGPU_FAMILY_AI) ? 256 : 16; 357 unsigned luma_size = ALIGN(enc->width, align) * ALIGN(enc->height, 16); 358 int len = 0, i, r; 359 360 luma_offset = enc->vbuf.addr; 361 chroma_offset = luma_offset + luma_size; 362 363 memcpy((ib_cpu + len), vce_session, sizeof(vce_session)); 364 len += sizeof(vce_session) / 4; 365 memcpy((ib_cpu + len), vce_taskinfo, sizeof(vce_taskinfo)); 366 len += sizeof(vce_taskinfo) / 4; 367 memcpy((ib_cpu + len), vce_bs_buffer, sizeof(vce_bs_buffer)); 368 ib_cpu[len + 2] = enc->bs[0].addr >> 32; 369 ib_cpu[len + 3] = enc->bs[0].addr; 370 len += sizeof(vce_bs_buffer) / 4; 371 memcpy((ib_cpu + len), vce_context_buffer, sizeof(vce_context_buffer)); 372 ib_cpu[len + 2] = enc->cpb.addr >> 32; 373 ib_cpu[len + 3] = enc->cpb.addr; 374 len += sizeof(vce_context_buffer) / 4; 375 memcpy((ib_cpu + len), vce_aux_buffer, sizeof(vce_aux_buffer)); 376 for (i = 0; i < 8; ++i) 377 ib_cpu[len + 2 + i] = luma_size * 1.5 * (i + 2); 378 for (i = 0; i < 8; ++i) 379 ib_cpu[len + 10 + i] = luma_size * 1.5; 380 len += sizeof(vce_aux_buffer) / 4; 381 memcpy((ib_cpu + len), vce_feedback, sizeof(vce_feedback)); 382 ib_cpu[len + 2] = enc->fb[0].addr >> 32; 383 ib_cpu[len + 3] = enc->fb[0].addr; 384 len += sizeof(vce_feedback) / 4; 385 memcpy((ib_cpu + len), vce_encode, sizeof(vce_encode)); 386 ib_cpu[len + 9] = luma_offset >> 32; 387 ib_cpu[len + 10] = luma_offset; 388 ib_cpu[len + 11] = chroma_offset >> 32; 389 ib_cpu[len + 12] = chroma_offset; 390 ib_cpu[len + 14] = ALIGN(enc->width, align); 391 ib_cpu[len + 15] = ALIGN(enc->width, align); 392 ib_cpu[len + 73] = luma_size * 1.5; 393 ib_cpu[len + 74] = luma_size * 2.5; 394 len += sizeof(vce_encode) / 4; 395 enc->ib_len = len; 396 if (!enc->two_instance) { 397 r = submit(len, AMDGPU_HW_IP_VCE); 398 CU_ASSERT_EQUAL(r, 0); 399 } 400 } 401 402 static void amdgpu_cs_vce_encode_p(struct amdgpu_vce_encode *enc) 403 { 404 uint64_t luma_offset, chroma_offset; 405 int len, i, r; 406 unsigned align = (family_id >= AMDGPU_FAMILY_AI) ? 256 : 16; 407 unsigned luma_size = ALIGN(enc->width, align) * ALIGN(enc->height, 16); 408 409 len = (enc->two_instance) ? enc->ib_len : 0; 410 luma_offset = enc->vbuf.addr; 411 chroma_offset = luma_offset + luma_size; 412 413 if (!enc->two_instance) { 414 memcpy((ib_cpu + len), vce_session, sizeof(vce_session)); 415 len += sizeof(vce_session) / 4; 416 } 417 memcpy((ib_cpu + len), vce_taskinfo, sizeof(vce_taskinfo)); 418 len += sizeof(vce_taskinfo) / 4; 419 memcpy((ib_cpu + len), vce_bs_buffer, sizeof(vce_bs_buffer)); 420 ib_cpu[len + 2] = enc->bs[1].addr >> 32; 421 ib_cpu[len + 3] = enc->bs[1].addr; 422 len += sizeof(vce_bs_buffer) / 4; 423 memcpy((ib_cpu + len), vce_context_buffer, sizeof(vce_context_buffer)); 424 ib_cpu[len + 2] = enc->cpb.addr >> 32; 425 ib_cpu[len + 3] = enc->cpb.addr; 426 len += sizeof(vce_context_buffer) / 4; 427 memcpy((ib_cpu + len), vce_aux_buffer, sizeof(vce_aux_buffer)); 428 for (i = 0; i < 8; ++i) 429 ib_cpu[len + 2 + i] = luma_size * 1.5 * (i + 2); 430 for (i = 0; i < 8; ++i) 431 ib_cpu[len + 10 + i] = luma_size * 1.5; 432 len += sizeof(vce_aux_buffer) / 4; 433 memcpy((ib_cpu + len), vce_feedback, sizeof(vce_feedback)); 434 ib_cpu[len + 2] = enc->fb[1].addr >> 32; 435 ib_cpu[len + 3] = enc->fb[1].addr; 436 len += sizeof(vce_feedback) / 4; 437 memcpy((ib_cpu + len), vce_encode, sizeof(vce_encode)); 438 ib_cpu[len + 2] = 0; 439 ib_cpu[len + 9] = luma_offset >> 32; 440 ib_cpu[len + 10] = luma_offset; 441 ib_cpu[len + 11] = chroma_offset >> 32; 442 ib_cpu[len + 12] = chroma_offset; 443 ib_cpu[len + 14] = ALIGN(enc->width, align); 444 ib_cpu[len + 15] = ALIGN(enc->width, align); 445 ib_cpu[len + 18] = 0; 446 ib_cpu[len + 19] = 0; 447 ib_cpu[len + 56] = 3; 448 ib_cpu[len + 57] = 0; 449 ib_cpu[len + 58] = 0; 450 ib_cpu[len + 59] = luma_size * 1.5; 451 ib_cpu[len + 60] = luma_size * 2.5; 452 ib_cpu[len + 73] = 0; 453 ib_cpu[len + 74] = luma_size; 454 ib_cpu[len + 81] = 1; 455 ib_cpu[len + 82] = 1; 456 len += sizeof(vce_encode) / 4; 457 458 r = submit(len, AMDGPU_HW_IP_VCE); 459 CU_ASSERT_EQUAL(r, 0); 460 } 461 462 static void check_result(struct amdgpu_vce_encode *enc) 463 { 464 uint64_t sum; 465 uint32_t s[2] = {180325, 15946}; 466 uint32_t *ptr, size; 467 int i, j, r; 468 469 for (i = 0; i < 2; ++i) { 470 r = amdgpu_bo_cpu_map(enc->fb[i].handle, (void **)&enc->fb[i].ptr); 471 CU_ASSERT_EQUAL(r, 0); 472 ptr = (uint32_t *)enc->fb[i].ptr; 473 size = ptr[4] - ptr[9]; 474 r = amdgpu_bo_cpu_unmap(enc->fb[i].handle); 475 CU_ASSERT_EQUAL(r, 0); 476 r = amdgpu_bo_cpu_map(enc->bs[i].handle, (void **)&enc->bs[i].ptr); 477 CU_ASSERT_EQUAL(r, 0); 478 for (j = 0, sum = 0; j < size; ++j) 479 sum += enc->bs[i].ptr[j]; 480 CU_ASSERT_EQUAL(sum, s[i]); 481 r = amdgpu_bo_cpu_unmap(enc->bs[i].handle); 482 CU_ASSERT_EQUAL(r, 0); 483 } 484 } 485 486 static void amdgpu_cs_vce_encode(void) 487 { 488 uint32_t vbuf_size, bs_size = 0x154000, cpb_size; 489 unsigned align = (family_id >= AMDGPU_FAMILY_AI) ? 256 : 16; 490 int i, r; 491 492 vbuf_size = ALIGN(enc.width, align) * ALIGN(enc.height, 16) * 1.5; 493 cpb_size = vbuf_size * 10; 494 num_resources = 0; 495 alloc_resource(&enc.fb[0], 4096, AMDGPU_GEM_DOMAIN_GTT); 496 resources[num_resources++] = enc.fb[0].handle; 497 alloc_resource(&enc.fb[1], 4096, AMDGPU_GEM_DOMAIN_GTT); 498 resources[num_resources++] = enc.fb[1].handle; 499 alloc_resource(&enc.bs[0], bs_size, AMDGPU_GEM_DOMAIN_GTT); 500 resources[num_resources++] = enc.bs[0].handle; 501 alloc_resource(&enc.bs[1], bs_size, AMDGPU_GEM_DOMAIN_GTT); 502 resources[num_resources++] = enc.bs[1].handle; 503 alloc_resource(&enc.vbuf, vbuf_size, AMDGPU_GEM_DOMAIN_VRAM); 504 resources[num_resources++] = enc.vbuf.handle; 505 alloc_resource(&enc.cpb, cpb_size, AMDGPU_GEM_DOMAIN_VRAM); 506 resources[num_resources++] = enc.cpb.handle; 507 resources[num_resources++] = ib_handle; 508 509 r = amdgpu_bo_cpu_map(enc.vbuf.handle, (void **)&enc.vbuf.ptr); 510 CU_ASSERT_EQUAL(r, 0); 511 512 memset(enc.vbuf.ptr, 0, vbuf_size); 513 for (i = 0; i < enc.height; ++i) { 514 memcpy(enc.vbuf.ptr, (frame + i * enc.width), enc.width); 515 enc.vbuf.ptr += ALIGN(enc.width, align); 516 } 517 for (i = 0; i < enc.height / 2; ++i) { 518 memcpy(enc.vbuf.ptr, ((frame + enc.height * enc.width) + i * enc.width), enc.width); 519 enc.vbuf.ptr += ALIGN(enc.width, align); 520 } 521 522 r = amdgpu_bo_cpu_unmap(enc.vbuf.handle); 523 CU_ASSERT_EQUAL(r, 0); 524 525 amdgpu_cs_vce_config(); 526 527 if (family_id >= AMDGPU_FAMILY_VI) { 528 vce_taskinfo[3] = 3; 529 amdgpu_cs_vce_encode_idr(&enc); 530 amdgpu_cs_vce_encode_p(&enc); 531 check_result(&enc); 532 533 /* two pipes */ 534 vce_encode[16] = 0; 535 amdgpu_cs_vce_encode_idr(&enc); 536 amdgpu_cs_vce_encode_p(&enc); 537 check_result(&enc); 538 539 /* two instances */ 540 if (vce_harvest_config == 0) { 541 enc.two_instance = true; 542 vce_taskinfo[2] = 0x83; 543 vce_taskinfo[4] = 1; 544 amdgpu_cs_vce_encode_idr(&enc); 545 vce_taskinfo[2] = 0xffffffff; 546 vce_taskinfo[4] = 2; 547 amdgpu_cs_vce_encode_p(&enc); 548 check_result(&enc); 549 } 550 } else { 551 vce_taskinfo[3] = 3; 552 vce_encode[16] = 0; 553 amdgpu_cs_vce_encode_idr(&enc); 554 amdgpu_cs_vce_encode_p(&enc); 555 check_result(&enc); 556 } 557 558 free_resource(&enc.fb[0]); 559 free_resource(&enc.fb[1]); 560 free_resource(&enc.bs[0]); 561 free_resource(&enc.bs[1]); 562 free_resource(&enc.vbuf); 563 free_resource(&enc.cpb); 564 } 565 566 static void amdgpu_cs_vce_mv(struct amdgpu_vce_encode *enc) 567 { 568 uint64_t luma_offset, chroma_offset; 569 uint64_t mv_ref_luma_offset; 570 unsigned align = (family_id >= AMDGPU_FAMILY_AI) ? 256 : 16; 571 unsigned luma_size = ALIGN(enc->width, align) * ALIGN(enc->height, 16); 572 int len = 0, i, r; 573 574 luma_offset = enc->vbuf.addr; 575 chroma_offset = luma_offset + luma_size; 576 mv_ref_luma_offset = enc->mvrefbuf.addr; 577 578 memcpy((ib_cpu + len), vce_session, sizeof(vce_session)); 579 len += sizeof(vce_session) / 4; 580 memcpy((ib_cpu + len), vce_taskinfo, sizeof(vce_taskinfo)); 581 len += sizeof(vce_taskinfo) / 4; 582 memcpy((ib_cpu + len), vce_bs_buffer, sizeof(vce_bs_buffer)); 583 ib_cpu[len + 2] = enc->bs[0].addr >> 32; 584 ib_cpu[len + 3] = enc->bs[0].addr; 585 len += sizeof(vce_bs_buffer) / 4; 586 memcpy((ib_cpu + len), vce_context_buffer, sizeof(vce_context_buffer)); 587 ib_cpu[len + 2] = enc->cpb.addr >> 32; 588 ib_cpu[len + 3] = enc->cpb.addr; 589 len += sizeof(vce_context_buffer) / 4; 590 memcpy((ib_cpu + len), vce_aux_buffer, sizeof(vce_aux_buffer)); 591 for (i = 0; i < 8; ++i) 592 ib_cpu[len + 2 + i] = luma_size * 1.5 * (i + 2); 593 for (i = 0; i < 8; ++i) 594 ib_cpu[len + 10 + i] = luma_size * 1.5; 595 len += sizeof(vce_aux_buffer) / 4; 596 memcpy((ib_cpu + len), vce_feedback, sizeof(vce_feedback)); 597 ib_cpu[len + 2] = enc->fb[0].addr >> 32; 598 ib_cpu[len + 3] = enc->fb[0].addr; 599 len += sizeof(vce_feedback) / 4; 600 memcpy((ib_cpu + len), vce_mv_buffer, sizeof(vce_mv_buffer)); 601 ib_cpu[len + 2] = mv_ref_luma_offset >> 32; 602 ib_cpu[len + 3] = mv_ref_luma_offset; 603 ib_cpu[len + 4] = ALIGN(enc->width, align); 604 ib_cpu[len + 5] = ALIGN(enc->width, align); 605 ib_cpu[len + 6] = luma_size; 606 ib_cpu[len + 7] = enc->mvb.addr >> 32; 607 ib_cpu[len + 8] = enc->mvb.addr; 608 len += sizeof(vce_mv_buffer) / 4; 609 memcpy((ib_cpu + len), vce_encode, sizeof(vce_encode)); 610 ib_cpu[len + 2] = 0; 611 ib_cpu[len + 3] = 0; 612 ib_cpu[len + 4] = 0x154000; 613 ib_cpu[len + 9] = luma_offset >> 32; 614 ib_cpu[len + 10] = luma_offset; 615 ib_cpu[len + 11] = chroma_offset >> 32; 616 ib_cpu[len + 12] = chroma_offset; 617 ib_cpu[len + 13] = ALIGN(enc->height, 16);; 618 ib_cpu[len + 14] = ALIGN(enc->width, align); 619 ib_cpu[len + 15] = ALIGN(enc->width, align); 620 /* encDisableMBOffloading-encDisableTwoPipeMode-encInputPicArrayMode-encInputPicAddrMode */ 621 ib_cpu[len + 16] = 0x01010000; 622 ib_cpu[len + 18] = 0; /* encPicType */ 623 ib_cpu[len + 19] = 0; /* encIdrFlag */ 624 ib_cpu[len + 20] = 0; /* encIdrPicId */ 625 ib_cpu[len + 21] = 0; /* encMGSKeyPic */ 626 ib_cpu[len + 22] = 0; /* encReferenceFlag */ 627 ib_cpu[len + 23] = 0; /* encTemporalLayerIndex */ 628 ib_cpu[len + 55] = 0; /* pictureStructure */ 629 ib_cpu[len + 56] = 0; /* encPicType -ref[0] */ 630 ib_cpu[len + 61] = 0; /* pictureStructure */ 631 ib_cpu[len + 62] = 0; /* encPicType -ref[1] */ 632 ib_cpu[len + 67] = 0; /* pictureStructure */ 633 ib_cpu[len + 68] = 0; /* encPicType -ref1 */ 634 ib_cpu[len + 81] = 1; /* frameNumber */ 635 ib_cpu[len + 82] = 2; /* pictureOrderCount */ 636 ib_cpu[len + 83] = 0xffffffff; /* numIPicRemainInRCGOP */ 637 ib_cpu[len + 84] = 0xffffffff; /* numPPicRemainInRCGOP */ 638 ib_cpu[len + 85] = 0xffffffff; /* numBPicRemainInRCGOP */ 639 ib_cpu[len + 86] = 0xffffffff; /* numIRPicRemainInRCGOP */ 640 ib_cpu[len + 87] = 0; /* remainedIntraRefreshPictures */ 641 len += sizeof(vce_encode) / 4; 642 643 enc->ib_len = len; 644 r = submit(len, AMDGPU_HW_IP_VCE); 645 CU_ASSERT_EQUAL(r, 0); 646 } 647 648 static void check_mv_result(struct amdgpu_vce_encode *enc) 649 { 650 uint64_t sum; 651 uint32_t s = 140790; 652 int j, r; 653 654 r = amdgpu_bo_cpu_map(enc->fb[0].handle, (void **)&enc->fb[0].ptr); 655 CU_ASSERT_EQUAL(r, 0); 656 r = amdgpu_bo_cpu_unmap(enc->fb[0].handle); 657 CU_ASSERT_EQUAL(r, 0); 658 r = amdgpu_bo_cpu_map(enc->mvb.handle, (void **)&enc->mvb.ptr); 659 CU_ASSERT_EQUAL(r, 0); 660 for (j = 0, sum = 0; j < enc->mvbuf_size; ++j) 661 sum += enc->mvb.ptr[j]; 662 CU_ASSERT_EQUAL(sum, s); 663 r = amdgpu_bo_cpu_unmap(enc->mvb.handle); 664 CU_ASSERT_EQUAL(r, 0); 665 } 666 667 static void amdgpu_cs_vce_encode_mv(void) 668 { 669 uint32_t vbuf_size, bs_size = 0x154000, cpb_size; 670 unsigned align = (family_id >= AMDGPU_FAMILY_AI) ? 256 : 16; 671 int i, r; 672 673 vbuf_size = ALIGN(enc.width, align) * ALIGN(enc.height, 16) * 1.5; 674 enc.mvbuf_size = ALIGN(enc.width, 16) * ALIGN(enc.height, 16) / 8; 675 cpb_size = vbuf_size * 10; 676 num_resources = 0; 677 alloc_resource(&enc.fb[0], 4096, AMDGPU_GEM_DOMAIN_GTT); 678 resources[num_resources++] = enc.fb[0].handle; 679 alloc_resource(&enc.bs[0], bs_size, AMDGPU_GEM_DOMAIN_GTT); 680 resources[num_resources++] = enc.bs[0].handle; 681 alloc_resource(&enc.mvb, enc.mvbuf_size, AMDGPU_GEM_DOMAIN_GTT); 682 resources[num_resources++] = enc.mvb.handle; 683 alloc_resource(&enc.vbuf, vbuf_size, AMDGPU_GEM_DOMAIN_VRAM); 684 resources[num_resources++] = enc.vbuf.handle; 685 alloc_resource(&enc.mvrefbuf, vbuf_size, AMDGPU_GEM_DOMAIN_VRAM); 686 resources[num_resources++] = enc.mvrefbuf.handle; 687 alloc_resource(&enc.cpb, cpb_size, AMDGPU_GEM_DOMAIN_VRAM); 688 resources[num_resources++] = enc.cpb.handle; 689 resources[num_resources++] = ib_handle; 690 691 r = amdgpu_bo_cpu_map(enc.vbuf.handle, (void **)&enc.vbuf.ptr); 692 CU_ASSERT_EQUAL(r, 0); 693 694 memset(enc.vbuf.ptr, 0, vbuf_size); 695 for (i = 0; i < enc.height; ++i) { 696 memcpy(enc.vbuf.ptr, (frame + i * enc.width), enc.width); 697 enc.vbuf.ptr += ALIGN(enc.width, align); 698 } 699 for (i = 0; i < enc.height / 2; ++i) { 700 memcpy(enc.vbuf.ptr, ((frame + enc.height * enc.width) + i * enc.width), enc.width); 701 enc.vbuf.ptr += ALIGN(enc.width, align); 702 } 703 704 r = amdgpu_bo_cpu_unmap(enc.vbuf.handle); 705 CU_ASSERT_EQUAL(r, 0); 706 707 r = amdgpu_bo_cpu_map(enc.mvrefbuf.handle, (void **)&enc.mvrefbuf.ptr); 708 CU_ASSERT_EQUAL(r, 0); 709 710 memset(enc.mvrefbuf.ptr, 0, vbuf_size); 711 for (i = 0; i < enc.height; ++i) { 712 memcpy(enc.mvrefbuf.ptr, (frame + (enc.height - i -1) * enc.width), enc.width); 713 enc.mvrefbuf.ptr += ALIGN(enc.width, align); 714 } 715 for (i = 0; i < enc.height / 2; ++i) { 716 memcpy(enc.mvrefbuf.ptr, 717 ((frame + enc.height * enc.width) + (enc.height / 2 - i -1) * enc.width), enc.width); 718 enc.mvrefbuf.ptr += ALIGN(enc.width, align); 719 } 720 721 r = amdgpu_bo_cpu_unmap(enc.mvrefbuf.handle); 722 CU_ASSERT_EQUAL(r, 0); 723 724 amdgpu_cs_vce_config(); 725 726 vce_taskinfo[3] = 3; 727 amdgpu_cs_vce_mv(&enc); 728 check_mv_result(&enc); 729 730 free_resource(&enc.fb[0]); 731 free_resource(&enc.bs[0]); 732 free_resource(&enc.vbuf); 733 free_resource(&enc.cpb); 734 free_resource(&enc.mvrefbuf); 735 free_resource(&enc.mvb); 736 } 737 738 static void amdgpu_cs_vce_destroy(void) 739 { 740 int len, r; 741 742 num_resources = 0; 743 alloc_resource(&enc.fb[0], 4096, AMDGPU_GEM_DOMAIN_GTT); 744 resources[num_resources++] = enc.fb[0].handle; 745 resources[num_resources++] = ib_handle; 746 747 len = 0; 748 memcpy(ib_cpu, vce_session, sizeof(vce_session)); 749 len += sizeof(vce_session) / 4; 750 memcpy((ib_cpu + len), vce_taskinfo, sizeof(vce_taskinfo)); 751 ib_cpu[len + 3] = 1; 752 len += sizeof(vce_taskinfo) / 4; 753 memcpy((ib_cpu + len), vce_feedback, sizeof(vce_feedback)); 754 ib_cpu[len + 2] = enc.fb[0].addr >> 32; 755 ib_cpu[len + 3] = enc.fb[0].addr; 756 len += sizeof(vce_feedback) / 4; 757 memcpy((ib_cpu + len), vce_destroy, sizeof(vce_destroy)); 758 len += sizeof(vce_destroy) / 4; 759 760 r = submit(len, AMDGPU_HW_IP_VCE); 761 CU_ASSERT_EQUAL(r, 0); 762 763 free_resource(&enc.fb[0]); 764 } 765