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 #ifdef HAVE_CONFIG_H
25 #include "config.h"
26 #endif
27 
28 #include <stdio.h>
29 #include <inttypes.h>
30 
31 #include "CUnit/Basic.h"
32 
33 #include "util_math.h"
34 
35 #include "amdgpu_test.h"
36 #include "amdgpu_drm.h"
37 #include "amdgpu_internal.h"
38 
39 #include "vce_ib.h"
40 #include "frame.h"
41 
42 #define IB_SIZE		4096
43 #define MAX_RESOURCES	16
44 
45 struct amdgpu_vce_bo {
46 	amdgpu_bo_handle handle;
47 	amdgpu_va_handle va_handle;
48 	uint64_t addr;
49 	uint64_t size;
50 	uint8_t *ptr;
51 };
52 
53 struct amdgpu_vce_encode {
54 	unsigned width;
55 	unsigned height;
56 	struct amdgpu_vce_bo vbuf;
57 	struct amdgpu_vce_bo bs[2];
58 	struct amdgpu_vce_bo fb[2];
59 	struct amdgpu_vce_bo cpb;
60 	unsigned ib_len;
61 	bool two_instance;
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 
69 static amdgpu_context_handle context_handle;
70 static amdgpu_bo_handle ib_handle;
71 static amdgpu_va_handle ib_va_handle;
72 static uint64_t ib_mc_address;
73 static uint32_t *ib_cpu;
74 
75 static struct amdgpu_vce_encode enc;
76 static amdgpu_bo_handle resources[MAX_RESOURCES];
77 static unsigned num_resources;
78 
79 static void amdgpu_cs_vce_create(void);
80 static void amdgpu_cs_vce_encode(void);
81 static void amdgpu_cs_vce_destroy(void);
82 
83 CU_TestInfo vce_tests[] = {
84 	{ "VCE create",  amdgpu_cs_vce_create },
85 	{ "VCE encode",  amdgpu_cs_vce_encode },
86 	{ "VCE destroy",  amdgpu_cs_vce_destroy },
87 	CU_TEST_INFO_NULL,
88 };
89 
suite_vce_tests_init(void)90 int suite_vce_tests_init(void)
91 {
92 	int r;
93 
94 	r = amdgpu_device_initialize(drm_amdgpu[0], &major_version,
95 				     &minor_version, &device_handle);
96 	if (r)
97 		return CUE_SINIT_FAILED;
98 
99 	family_id = device_handle->info.family_id;
100 
101 	r = amdgpu_cs_ctx_create(device_handle, &context_handle);
102 	if (r)
103 		return CUE_SINIT_FAILED;
104 
105 	r = amdgpu_bo_alloc_and_map(device_handle, IB_SIZE, 4096,
106 				    AMDGPU_GEM_DOMAIN_GTT, 0,
107 				    &ib_handle, (void**)&ib_cpu,
108 				    &ib_mc_address, &ib_va_handle);
109 	if (r)
110 		return CUE_SINIT_FAILED;
111 
112 	memset(&enc, 0, sizeof(struct amdgpu_vce_encode));
113 
114 	return CUE_SUCCESS;
115 }
116 
suite_vce_tests_clean(void)117 int suite_vce_tests_clean(void)
118 {
119 	int r;
120 
121 	r = amdgpu_bo_unmap_and_free(ib_handle, ib_va_handle,
122 				     ib_mc_address, IB_SIZE);
123 	if (r)
124 		return CUE_SCLEAN_FAILED;
125 
126 	r = amdgpu_cs_ctx_free(context_handle);
127 	if (r)
128 		return CUE_SCLEAN_FAILED;
129 
130 	r = amdgpu_device_deinitialize(device_handle);
131 	if (r)
132 		return CUE_SCLEAN_FAILED;
133 
134 	return CUE_SUCCESS;
135 }
136 
submit(unsigned ndw,unsigned ip)137 static int submit(unsigned ndw, unsigned ip)
138 {
139 	struct amdgpu_cs_request ibs_request = {0};
140 	struct amdgpu_cs_ib_info ib_info = {0};
141 	struct amdgpu_cs_fence fence_status = {0};
142 	uint32_t expired;
143 	int r;
144 
145 	ib_info.ib_mc_address = ib_mc_address;
146 	ib_info.size = ndw;
147 
148 	ibs_request.ip_type = ip;
149 
150 	r = amdgpu_bo_list_create(device_handle, num_resources, resources,
151 				  NULL, &ibs_request.resources);
152 	if (r)
153 		return r;
154 
155 	ibs_request.number_of_ibs = 1;
156 	ibs_request.ibs = &ib_info;
157 	ibs_request.fence_info.handle = NULL;
158 
159 	r = amdgpu_cs_submit(context_handle, 0, &ibs_request, 1);
160 	if (r)
161 		return r;
162 
163 	r = amdgpu_bo_list_destroy(ibs_request.resources);
164 	if (r)
165 		return r;
166 
167 	fence_status.context = context_handle;
168 	fence_status.ip_type = ip;
169 	fence_status.fence = ibs_request.seq_no;
170 
171 	r = amdgpu_cs_query_fence_status(&fence_status,
172 					 AMDGPU_TIMEOUT_INFINITE,
173 					 0, &expired);
174 	if (r)
175 		return r;
176 
177 	return 0;
178 }
179 
alloc_resource(struct amdgpu_vce_bo * vce_bo,unsigned size,unsigned domain)180 static void alloc_resource(struct amdgpu_vce_bo *vce_bo, unsigned size, unsigned domain)
181 {
182 	struct amdgpu_bo_alloc_request req = {0};
183 	amdgpu_bo_handle buf_handle;
184 	amdgpu_va_handle va_handle;
185 	uint64_t va = 0;
186 	int r;
187 
188 	req.alloc_size = ALIGN(size, 4096);
189 	req.preferred_heap = domain;
190 	r = amdgpu_bo_alloc(device_handle, &req, &buf_handle);
191 	CU_ASSERT_EQUAL(r, 0);
192 	r = amdgpu_va_range_alloc(device_handle,
193 				  amdgpu_gpu_va_range_general,
194 				  req.alloc_size, 1, 0, &va,
195 				  &va_handle, 0);
196 	CU_ASSERT_EQUAL(r, 0);
197 	r = amdgpu_bo_va_op(buf_handle, 0, req.alloc_size, va, 0,
198 			    AMDGPU_VA_OP_MAP);
199 	CU_ASSERT_EQUAL(r, 0);
200 	vce_bo->addr = va;
201 	vce_bo->handle = buf_handle;
202 	vce_bo->size = req.alloc_size;
203 	vce_bo->va_handle = va_handle;
204 	r = amdgpu_bo_cpu_map(vce_bo->handle, (void **)&vce_bo->ptr);
205 	CU_ASSERT_EQUAL(r, 0);
206 	memset(vce_bo->ptr, 0, size);
207 	r = amdgpu_bo_cpu_unmap(vce_bo->handle);
208 	CU_ASSERT_EQUAL(r, 0);
209 }
210 
free_resource(struct amdgpu_vce_bo * vce_bo)211 static void free_resource(struct amdgpu_vce_bo *vce_bo)
212 {
213 	int r;
214 
215 	r = amdgpu_bo_va_op(vce_bo->handle, 0, vce_bo->size,
216 			    vce_bo->addr, 0, AMDGPU_VA_OP_UNMAP);
217 	CU_ASSERT_EQUAL(r, 0);
218 
219 	r = amdgpu_va_range_free(vce_bo->va_handle);
220 	CU_ASSERT_EQUAL(r, 0);
221 
222 	r = amdgpu_bo_free(vce_bo->handle);
223 	CU_ASSERT_EQUAL(r, 0);
224 	memset(vce_bo, 0, sizeof(*vce_bo));
225 }
226 
amdgpu_cs_vce_create(void)227 static void amdgpu_cs_vce_create(void)
228 {
229 	int len, r;
230 
231 	enc.width = vce_create[6];
232 	enc.height = vce_create[7];
233 
234 	num_resources  = 0;
235 	alloc_resource(&enc.fb[0], 4096, AMDGPU_GEM_DOMAIN_GTT);
236 	resources[num_resources++] = enc.fb[0].handle;
237 	resources[num_resources++] = ib_handle;
238 
239 	len = 0;
240 	memcpy(ib_cpu, vce_session, sizeof(vce_session));
241 	len += sizeof(vce_session) / 4;
242 	memcpy((ib_cpu + len), vce_taskinfo, sizeof(vce_taskinfo));
243 	len += sizeof(vce_taskinfo) / 4;
244 	memcpy((ib_cpu + len), vce_create, sizeof(vce_create));
245 	len += sizeof(vce_create) / 4;
246 	memcpy((ib_cpu + len), vce_feedback, sizeof(vce_feedback));
247 	ib_cpu[len + 2] = enc.fb[0].addr >> 32;
248 	ib_cpu[len + 3] = enc.fb[0].addr;
249 	len += sizeof(vce_feedback) / 4;
250 
251 	r = submit(len, AMDGPU_HW_IP_VCE);
252 	CU_ASSERT_EQUAL(r, 0);
253 
254 	free_resource(&enc.fb[0]);
255 }
256 
amdgpu_cs_vce_config(void)257 static void amdgpu_cs_vce_config(void)
258 {
259 	int len = 0, r;
260 
261 	memcpy((ib_cpu + len), vce_session, sizeof(vce_session));
262 	len += sizeof(vce_session) / 4;
263 	memcpy((ib_cpu + len), vce_taskinfo, sizeof(vce_taskinfo));
264 	ib_cpu[len + 3] = 2;
265 	ib_cpu[len + 6] = 0xffffffff;
266 	len += sizeof(vce_taskinfo) / 4;
267 	memcpy((ib_cpu + len), vce_rate_ctrl, sizeof(vce_rate_ctrl));
268 	len += sizeof(vce_rate_ctrl) / 4;
269 	memcpy((ib_cpu + len), vce_config_ext, sizeof(vce_config_ext));
270 	len += sizeof(vce_config_ext) / 4;
271 	memcpy((ib_cpu + len), vce_motion_est, sizeof(vce_motion_est));
272 	len += sizeof(vce_motion_est) / 4;
273 	memcpy((ib_cpu + len), vce_rdo, sizeof(vce_rdo));
274 	len += sizeof(vce_rdo) / 4;
275 	memcpy((ib_cpu + len), vce_pic_ctrl, sizeof(vce_pic_ctrl));
276 	len += sizeof(vce_pic_ctrl) / 4;
277 
278 	r = submit(len, AMDGPU_HW_IP_VCE);
279 	CU_ASSERT_EQUAL(r, 0);
280 }
281 
amdgpu_cs_vce_encode_idr(struct amdgpu_vce_encode * enc)282 static  void amdgpu_cs_vce_encode_idr(struct amdgpu_vce_encode *enc)
283 {
284 
285 	uint64_t luma_offset, chroma_offset;
286 	int len = 0, r;
287 
288 	luma_offset = enc->vbuf.addr;
289 	chroma_offset = luma_offset + enc->width * enc->height;
290 
291 	memcpy((ib_cpu + len), vce_session, sizeof(vce_session));
292 	len += sizeof(vce_session) / 4;
293 	memcpy((ib_cpu + len), vce_taskinfo, sizeof(vce_taskinfo));
294 	len += sizeof(vce_taskinfo) / 4;
295 	memcpy((ib_cpu + len), vce_bs_buffer, sizeof(vce_bs_buffer));
296 	ib_cpu[len + 2] = enc->bs[0].addr >> 32;
297 	ib_cpu[len + 3] = enc->bs[0].addr;
298 	len += sizeof(vce_bs_buffer) / 4;
299 	memcpy((ib_cpu + len), vce_context_buffer, sizeof(vce_context_buffer));
300 	ib_cpu[len + 2] = enc->cpb.addr >> 32;
301 	ib_cpu[len + 3] = enc->cpb.addr;
302 	len += sizeof(vce_context_buffer) / 4;
303 	memcpy((ib_cpu + len), vce_aux_buffer, sizeof(vce_aux_buffer));
304 	len += sizeof(vce_aux_buffer) / 4;
305 	memcpy((ib_cpu + len), vce_feedback, sizeof(vce_feedback));
306 	ib_cpu[len + 2] = enc->fb[0].addr >> 32;
307 	ib_cpu[len + 3] = enc->fb[0].addr;
308 	len += sizeof(vce_feedback) / 4;
309 	memcpy((ib_cpu + len), vce_encode, sizeof(vce_encode));
310 	ib_cpu[len + 9] = luma_offset >> 32;
311 	ib_cpu[len + 10] = luma_offset;
312 	ib_cpu[len + 11] = chroma_offset >> 32;
313 	ib_cpu[len + 12] = chroma_offset;
314 	ib_cpu[len + 73] = 0x7800;
315 	ib_cpu[len + 74] = 0x7800 + 0x5000;
316 	len += sizeof(vce_encode) / 4;
317 	enc->ib_len = len;
318 	if (!enc->two_instance) {
319 		r = submit(len, AMDGPU_HW_IP_VCE);
320 		CU_ASSERT_EQUAL(r, 0);
321 	}
322 }
323 
amdgpu_cs_vce_encode_p(struct amdgpu_vce_encode * enc)324 static void amdgpu_cs_vce_encode_p(struct amdgpu_vce_encode *enc)
325 {
326 	uint64_t luma_offset, chroma_offset;
327 	int len, r;
328 
329 	len = (enc->two_instance) ? enc->ib_len : 0;
330 	luma_offset = enc->vbuf.addr;
331 	chroma_offset = luma_offset + enc->width * enc->height;
332 
333 	if (!enc->two_instance) {
334 		memcpy((ib_cpu + len), vce_session, sizeof(vce_session));
335 		len += sizeof(vce_session) / 4;
336 	}
337 	memcpy((ib_cpu + len), vce_taskinfo, sizeof(vce_taskinfo));
338 	len += sizeof(vce_taskinfo) / 4;
339 	memcpy((ib_cpu + len), vce_bs_buffer, sizeof(vce_bs_buffer));
340 	ib_cpu[len + 2] = enc->bs[1].addr >> 32;
341 	ib_cpu[len + 3] = enc->bs[1].addr;
342 	len += sizeof(vce_bs_buffer) / 4;
343 	memcpy((ib_cpu + len), vce_context_buffer, sizeof(vce_context_buffer));
344 	ib_cpu[len + 2] = enc->cpb.addr >> 32;
345 	ib_cpu[len + 3] = enc->cpb.addr;
346 	len += sizeof(vce_context_buffer) / 4;
347 	memcpy((ib_cpu + len), vce_aux_buffer, sizeof(vce_aux_buffer));
348 	len += sizeof(vce_aux_buffer) / 4;
349 	memcpy((ib_cpu + len), vce_feedback, sizeof(vce_feedback));
350 	ib_cpu[len + 2] = enc->fb[1].addr >> 32;
351 	ib_cpu[len + 3] = enc->fb[1].addr;
352 	len += sizeof(vce_feedback) / 4;
353 	memcpy((ib_cpu + len), vce_encode, sizeof(vce_encode));
354 	ib_cpu[len + 2] = 0;
355 	ib_cpu[len + 9] = luma_offset >> 32;
356 	ib_cpu[len + 10] = luma_offset;
357 	ib_cpu[len + 11] = chroma_offset >> 32;
358 	ib_cpu[len + 12] = chroma_offset;
359 	ib_cpu[len + 18] = 0;
360 	ib_cpu[len + 19] = 0;
361 	ib_cpu[len + 56] = 3;
362 	ib_cpu[len + 57] = 0;
363 	ib_cpu[len + 58] = 0;
364 	ib_cpu[len + 59] = 0x7800;
365 	ib_cpu[len + 60] = 0x7800 + 0x5000;
366 	ib_cpu[len + 73] = 0;
367 	ib_cpu[len + 74] = 0x5000;
368 	ib_cpu[len + 81] = 1;
369 	ib_cpu[len + 82] = 1;
370 	len += sizeof(vce_encode) / 4;
371 
372 	r = submit(len, AMDGPU_HW_IP_VCE);
373 	CU_ASSERT_EQUAL(r, 0);
374 }
375 
check_result(struct amdgpu_vce_encode * enc)376 static void check_result(struct amdgpu_vce_encode *enc)
377 {
378 	uint64_t sum;
379 	uint32_t s[2] = {180325, 15946};
380 	uint32_t *ptr, size;
381 	int i, j, r;
382 
383 	for (i = 0; i < 2; ++i) {
384 		r = amdgpu_bo_cpu_map(enc->fb[i].handle, (void **)&enc->fb[i].ptr);
385 		CU_ASSERT_EQUAL(r, 0);
386 		ptr = (uint32_t *)enc->fb[i].ptr;
387 		size = ptr[4] - ptr[9];
388 		r = amdgpu_bo_cpu_unmap(enc->fb[i].handle);
389 		CU_ASSERT_EQUAL(r, 0);
390 		r = amdgpu_bo_cpu_map(enc->bs[i].handle, (void **)&enc->bs[i].ptr);
391 		CU_ASSERT_EQUAL(r, 0);
392 		for (j = 0, sum = 0; j < size; ++j)
393 			sum += enc->bs[i].ptr[j];
394 		CU_ASSERT_EQUAL(sum, s[i]);
395 		r = amdgpu_bo_cpu_unmap(enc->bs[i].handle);
396 		CU_ASSERT_EQUAL(r, 0);
397 	}
398 }
399 
amdgpu_cs_vce_encode(void)400 static void amdgpu_cs_vce_encode(void)
401 {
402 	uint32_t vbuf_size, bs_size = 0x154000, cpb_size;
403 	int r;
404 
405 	vbuf_size = enc.width * enc.height * 1.5;
406 	cpb_size = vbuf_size * 10;
407 	num_resources = 0;
408 	alloc_resource(&enc.fb[0], 4096, AMDGPU_GEM_DOMAIN_GTT);
409 	resources[num_resources++] = enc.fb[0].handle;
410 	alloc_resource(&enc.fb[1], 4096, AMDGPU_GEM_DOMAIN_GTT);
411 	resources[num_resources++] = enc.fb[1].handle;
412 	alloc_resource(&enc.bs[0], bs_size, AMDGPU_GEM_DOMAIN_GTT);
413 	resources[num_resources++] = enc.bs[0].handle;
414 	alloc_resource(&enc.bs[1], bs_size, AMDGPU_GEM_DOMAIN_GTT);
415 	resources[num_resources++] = enc.bs[1].handle;
416 	alloc_resource(&enc.vbuf, vbuf_size, AMDGPU_GEM_DOMAIN_VRAM);
417 	resources[num_resources++] = enc.vbuf.handle;
418 	alloc_resource(&enc.cpb, cpb_size, AMDGPU_GEM_DOMAIN_VRAM);
419 	resources[num_resources++] = enc.cpb.handle;
420 	resources[num_resources++] = ib_handle;
421 
422 	r = amdgpu_bo_cpu_map(enc.vbuf.handle, (void **)&enc.vbuf.ptr);
423 	CU_ASSERT_EQUAL(r, 0);
424 	memcpy(enc.vbuf.ptr, frame, sizeof(frame));
425 	r = amdgpu_bo_cpu_unmap(enc.vbuf.handle);
426 	CU_ASSERT_EQUAL(r, 0);
427 
428 	amdgpu_cs_vce_config();
429 
430 	if (family_id >= AMDGPU_FAMILY_VI) {
431 		vce_taskinfo[3] = 3;
432 		amdgpu_cs_vce_encode_idr(&enc);
433 		amdgpu_cs_vce_encode_p(&enc);
434 		check_result(&enc);
435 
436 		/* two pipes */
437 		vce_encode[16] = 0;
438 		amdgpu_cs_vce_encode_idr(&enc);
439 		amdgpu_cs_vce_encode_p(&enc);
440 		check_result(&enc);
441 
442 		/* two instances */
443 		enc.two_instance = true;
444 		vce_taskinfo[2] = 0x83;
445 		vce_taskinfo[4] = 1;
446 		amdgpu_cs_vce_encode_idr(&enc);
447 		vce_taskinfo[2] = 0xffffffff;
448 		vce_taskinfo[4] = 2;
449 		amdgpu_cs_vce_encode_p(&enc);
450 		check_result(&enc);
451 	} else {
452 		vce_taskinfo[3] = 3;
453 		vce_encode[16] = 0;
454 		amdgpu_cs_vce_encode_idr(&enc);
455 		amdgpu_cs_vce_encode_p(&enc);
456 		check_result(&enc);
457 	}
458 
459 	free_resource(&enc.fb[0]);
460 	free_resource(&enc.fb[1]);
461 	free_resource(&enc.bs[0]);
462 	free_resource(&enc.bs[1]);
463 	free_resource(&enc.vbuf);
464 	free_resource(&enc.cpb);
465 }
466 
amdgpu_cs_vce_destroy(void)467 static void amdgpu_cs_vce_destroy(void)
468 {
469 	int len, r;
470 
471 	num_resources  = 0;
472 	alloc_resource(&enc.fb[0], 4096, AMDGPU_GEM_DOMAIN_GTT);
473 	resources[num_resources++] = enc.fb[0].handle;
474 	resources[num_resources++] = ib_handle;
475 
476 	len = 0;
477 	memcpy(ib_cpu, vce_session, sizeof(vce_session));
478 	len += sizeof(vce_session) / 4;
479 	memcpy((ib_cpu + len), vce_taskinfo, sizeof(vce_taskinfo));
480 	ib_cpu[len + 3] = 1;
481 	len += sizeof(vce_taskinfo) / 4;
482 	memcpy((ib_cpu + len), vce_feedback, sizeof(vce_feedback));
483 	ib_cpu[len + 2] = enc.fb[0].addr >> 32;
484 	ib_cpu[len + 3] = enc.fb[0].addr;
485 	len += sizeof(vce_feedback) / 4;
486 	memcpy((ib_cpu + len), vce_destroy, sizeof(vce_destroy));
487 	len += sizeof(vce_destroy) / 4;
488 
489 	r = submit(len, AMDGPU_HW_IP_VCE);
490 	CU_ASSERT_EQUAL(r, 0);
491 
492 	free_resource(&enc.fb[0]);
493 }
494