vce_tests.c revision 3f012e29
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
45struct 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
53struct 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
64static amdgpu_device_handle device_handle;
65static uint32_t major_version;
66static uint32_t minor_version;
67static uint32_t family_id;
68static uint32_t vce_harvest_config;
69
70static amdgpu_context_handle context_handle;
71static amdgpu_bo_handle ib_handle;
72static amdgpu_va_handle ib_va_handle;
73static uint64_t ib_mc_address;
74static uint32_t *ib_cpu;
75
76static struct amdgpu_vce_encode enc;
77static amdgpu_bo_handle resources[MAX_RESOURCES];
78static unsigned num_resources;
79
80static void amdgpu_cs_vce_create(void);
81static void amdgpu_cs_vce_encode(void);
82static void amdgpu_cs_vce_destroy(void);
83
84CU_TestInfo vce_tests[] = {
85	{ "VCE create",  amdgpu_cs_vce_create },
86	{ "VCE encode",  amdgpu_cs_vce_encode },
87	{ "VCE destroy",  amdgpu_cs_vce_destroy },
88	CU_TEST_INFO_NULL,
89};
90
91int suite_vce_tests_init(void)
92{
93	int r;
94
95	r = amdgpu_device_initialize(drm_amdgpu[0], &major_version,
96				     &minor_version, &device_handle);
97	if (r)
98		return CUE_SINIT_FAILED;
99
100	family_id = device_handle->info.family_id;
101	vce_harvest_config = device_handle->info.vce_harvest_config;
102
103	r = amdgpu_cs_ctx_create(device_handle, &context_handle);
104	if (r)
105		return CUE_SINIT_FAILED;
106
107	r = amdgpu_bo_alloc_and_map(device_handle, IB_SIZE, 4096,
108				    AMDGPU_GEM_DOMAIN_GTT, 0,
109				    &ib_handle, (void**)&ib_cpu,
110				    &ib_mc_address, &ib_va_handle);
111	if (r)
112		return CUE_SINIT_FAILED;
113
114	memset(&enc, 0, sizeof(struct amdgpu_vce_encode));
115
116	return CUE_SUCCESS;
117}
118
119int suite_vce_tests_clean(void)
120{
121	int r;
122
123	r = amdgpu_bo_unmap_and_free(ib_handle, ib_va_handle,
124				     ib_mc_address, IB_SIZE);
125	if (r)
126		return CUE_SCLEAN_FAILED;
127
128	r = amdgpu_cs_ctx_free(context_handle);
129	if (r)
130		return CUE_SCLEAN_FAILED;
131
132	r = amdgpu_device_deinitialize(device_handle);
133	if (r)
134		return CUE_SCLEAN_FAILED;
135
136	return CUE_SUCCESS;
137}
138
139static int submit(unsigned ndw, unsigned ip)
140{
141	struct amdgpu_cs_request ibs_request = {0};
142	struct amdgpu_cs_ib_info ib_info = {0};
143	struct amdgpu_cs_fence fence_status = {0};
144	uint32_t expired;
145	int r;
146
147	ib_info.ib_mc_address = ib_mc_address;
148	ib_info.size = ndw;
149
150	ibs_request.ip_type = ip;
151
152	r = amdgpu_bo_list_create(device_handle, num_resources, resources,
153				  NULL, &ibs_request.resources);
154	if (r)
155		return r;
156
157	ibs_request.number_of_ibs = 1;
158	ibs_request.ibs = &ib_info;
159	ibs_request.fence_info.handle = NULL;
160
161	r = amdgpu_cs_submit(context_handle, 0, &ibs_request, 1);
162	if (r)
163		return r;
164
165	r = amdgpu_bo_list_destroy(ibs_request.resources);
166	if (r)
167		return r;
168
169	fence_status.context = context_handle;
170	fence_status.ip_type = ip;
171	fence_status.fence = ibs_request.seq_no;
172
173	r = amdgpu_cs_query_fence_status(&fence_status,
174					 AMDGPU_TIMEOUT_INFINITE,
175					 0, &expired);
176	if (r)
177		return r;
178
179	return 0;
180}
181
182static void alloc_resource(struct amdgpu_vce_bo *vce_bo, unsigned size, unsigned domain)
183{
184	struct amdgpu_bo_alloc_request req = {0};
185	amdgpu_bo_handle buf_handle;
186	amdgpu_va_handle va_handle;
187	uint64_t va = 0;
188	int r;
189
190	req.alloc_size = ALIGN(size, 4096);
191	req.preferred_heap = domain;
192	r = amdgpu_bo_alloc(device_handle, &req, &buf_handle);
193	CU_ASSERT_EQUAL(r, 0);
194	r = amdgpu_va_range_alloc(device_handle,
195				  amdgpu_gpu_va_range_general,
196				  req.alloc_size, 1, 0, &va,
197				  &va_handle, 0);
198	CU_ASSERT_EQUAL(r, 0);
199	r = amdgpu_bo_va_op(buf_handle, 0, req.alloc_size, va, 0,
200			    AMDGPU_VA_OP_MAP);
201	CU_ASSERT_EQUAL(r, 0);
202	vce_bo->addr = va;
203	vce_bo->handle = buf_handle;
204	vce_bo->size = req.alloc_size;
205	vce_bo->va_handle = va_handle;
206	r = amdgpu_bo_cpu_map(vce_bo->handle, (void **)&vce_bo->ptr);
207	CU_ASSERT_EQUAL(r, 0);
208	memset(vce_bo->ptr, 0, size);
209	r = amdgpu_bo_cpu_unmap(vce_bo->handle);
210	CU_ASSERT_EQUAL(r, 0);
211}
212
213static void free_resource(struct amdgpu_vce_bo *vce_bo)
214{
215	int r;
216
217	r = amdgpu_bo_va_op(vce_bo->handle, 0, vce_bo->size,
218			    vce_bo->addr, 0, AMDGPU_VA_OP_UNMAP);
219	CU_ASSERT_EQUAL(r, 0);
220
221	r = amdgpu_va_range_free(vce_bo->va_handle);
222	CU_ASSERT_EQUAL(r, 0);
223
224	r = amdgpu_bo_free(vce_bo->handle);
225	CU_ASSERT_EQUAL(r, 0);
226	memset(vce_bo, 0, sizeof(*vce_bo));
227}
228
229static void amdgpu_cs_vce_create(void)
230{
231	int len, r;
232
233	enc.width = vce_create[6];
234	enc.height = vce_create[7];
235
236	num_resources  = 0;
237	alloc_resource(&enc.fb[0], 4096, AMDGPU_GEM_DOMAIN_GTT);
238	resources[num_resources++] = enc.fb[0].handle;
239	resources[num_resources++] = ib_handle;
240
241	len = 0;
242	memcpy(ib_cpu, vce_session, sizeof(vce_session));
243	len += sizeof(vce_session) / 4;
244	memcpy((ib_cpu + len), vce_taskinfo, sizeof(vce_taskinfo));
245	len += sizeof(vce_taskinfo) / 4;
246	memcpy((ib_cpu + len), vce_create, sizeof(vce_create));
247	len += sizeof(vce_create) / 4;
248	memcpy((ib_cpu + len), vce_feedback, sizeof(vce_feedback));
249	ib_cpu[len + 2] = enc.fb[0].addr >> 32;
250	ib_cpu[len + 3] = enc.fb[0].addr;
251	len += sizeof(vce_feedback) / 4;
252
253	r = submit(len, AMDGPU_HW_IP_VCE);
254	CU_ASSERT_EQUAL(r, 0);
255
256	free_resource(&enc.fb[0]);
257}
258
259static void amdgpu_cs_vce_config(void)
260{
261	int len = 0, r;
262
263	memcpy((ib_cpu + len), vce_session, sizeof(vce_session));
264	len += sizeof(vce_session) / 4;
265	memcpy((ib_cpu + len), vce_taskinfo, sizeof(vce_taskinfo));
266	ib_cpu[len + 3] = 2;
267	ib_cpu[len + 6] = 0xffffffff;
268	len += sizeof(vce_taskinfo) / 4;
269	memcpy((ib_cpu + len), vce_rate_ctrl, sizeof(vce_rate_ctrl));
270	len += sizeof(vce_rate_ctrl) / 4;
271	memcpy((ib_cpu + len), vce_config_ext, sizeof(vce_config_ext));
272	len += sizeof(vce_config_ext) / 4;
273	memcpy((ib_cpu + len), vce_motion_est, sizeof(vce_motion_est));
274	len += sizeof(vce_motion_est) / 4;
275	memcpy((ib_cpu + len), vce_rdo, sizeof(vce_rdo));
276	len += sizeof(vce_rdo) / 4;
277	memcpy((ib_cpu + len), vce_pic_ctrl, sizeof(vce_pic_ctrl));
278	len += sizeof(vce_pic_ctrl) / 4;
279
280	r = submit(len, AMDGPU_HW_IP_VCE);
281	CU_ASSERT_EQUAL(r, 0);
282}
283
284static  void amdgpu_cs_vce_encode_idr(struct amdgpu_vce_encode *enc)
285{
286
287	uint64_t luma_offset, chroma_offset;
288	int len = 0, r;
289
290	luma_offset = enc->vbuf.addr;
291	chroma_offset = luma_offset + enc->width * enc->height;
292
293	memcpy((ib_cpu + len), vce_session, sizeof(vce_session));
294	len += sizeof(vce_session) / 4;
295	memcpy((ib_cpu + len), vce_taskinfo, sizeof(vce_taskinfo));
296	len += sizeof(vce_taskinfo) / 4;
297	memcpy((ib_cpu + len), vce_bs_buffer, sizeof(vce_bs_buffer));
298	ib_cpu[len + 2] = enc->bs[0].addr >> 32;
299	ib_cpu[len + 3] = enc->bs[0].addr;
300	len += sizeof(vce_bs_buffer) / 4;
301	memcpy((ib_cpu + len), vce_context_buffer, sizeof(vce_context_buffer));
302	ib_cpu[len + 2] = enc->cpb.addr >> 32;
303	ib_cpu[len + 3] = enc->cpb.addr;
304	len += sizeof(vce_context_buffer) / 4;
305	memcpy((ib_cpu + len), vce_aux_buffer, sizeof(vce_aux_buffer));
306	len += sizeof(vce_aux_buffer) / 4;
307	memcpy((ib_cpu + len), vce_feedback, sizeof(vce_feedback));
308	ib_cpu[len + 2] = enc->fb[0].addr >> 32;
309	ib_cpu[len + 3] = enc->fb[0].addr;
310	len += sizeof(vce_feedback) / 4;
311	memcpy((ib_cpu + len), vce_encode, sizeof(vce_encode));
312	ib_cpu[len + 9] = luma_offset >> 32;
313	ib_cpu[len + 10] = luma_offset;
314	ib_cpu[len + 11] = chroma_offset >> 32;
315	ib_cpu[len + 12] = chroma_offset;
316	ib_cpu[len + 73] = 0x7800;
317	ib_cpu[len + 74] = 0x7800 + 0x5000;
318	len += sizeof(vce_encode) / 4;
319	enc->ib_len = len;
320	if (!enc->two_instance) {
321		r = submit(len, AMDGPU_HW_IP_VCE);
322		CU_ASSERT_EQUAL(r, 0);
323	}
324}
325
326static void amdgpu_cs_vce_encode_p(struct amdgpu_vce_encode *enc)
327{
328	uint64_t luma_offset, chroma_offset;
329	int len, r;
330
331	len = (enc->two_instance) ? enc->ib_len : 0;
332	luma_offset = enc->vbuf.addr;
333	chroma_offset = luma_offset + enc->width * enc->height;
334
335	if (!enc->two_instance) {
336		memcpy((ib_cpu + len), vce_session, sizeof(vce_session));
337		len += sizeof(vce_session) / 4;
338	}
339	memcpy((ib_cpu + len), vce_taskinfo, sizeof(vce_taskinfo));
340	len += sizeof(vce_taskinfo) / 4;
341	memcpy((ib_cpu + len), vce_bs_buffer, sizeof(vce_bs_buffer));
342	ib_cpu[len + 2] = enc->bs[1].addr >> 32;
343	ib_cpu[len + 3] = enc->bs[1].addr;
344	len += sizeof(vce_bs_buffer) / 4;
345	memcpy((ib_cpu + len), vce_context_buffer, sizeof(vce_context_buffer));
346	ib_cpu[len + 2] = enc->cpb.addr >> 32;
347	ib_cpu[len + 3] = enc->cpb.addr;
348	len += sizeof(vce_context_buffer) / 4;
349	memcpy((ib_cpu + len), vce_aux_buffer, sizeof(vce_aux_buffer));
350	len += sizeof(vce_aux_buffer) / 4;
351	memcpy((ib_cpu + len), vce_feedback, sizeof(vce_feedback));
352	ib_cpu[len + 2] = enc->fb[1].addr >> 32;
353	ib_cpu[len + 3] = enc->fb[1].addr;
354	len += sizeof(vce_feedback) / 4;
355	memcpy((ib_cpu + len), vce_encode, sizeof(vce_encode));
356	ib_cpu[len + 2] = 0;
357	ib_cpu[len + 9] = luma_offset >> 32;
358	ib_cpu[len + 10] = luma_offset;
359	ib_cpu[len + 11] = chroma_offset >> 32;
360	ib_cpu[len + 12] = chroma_offset;
361	ib_cpu[len + 18] = 0;
362	ib_cpu[len + 19] = 0;
363	ib_cpu[len + 56] = 3;
364	ib_cpu[len + 57] = 0;
365	ib_cpu[len + 58] = 0;
366	ib_cpu[len + 59] = 0x7800;
367	ib_cpu[len + 60] = 0x7800 + 0x5000;
368	ib_cpu[len + 73] = 0;
369	ib_cpu[len + 74] = 0x5000;
370	ib_cpu[len + 81] = 1;
371	ib_cpu[len + 82] = 1;
372	len += sizeof(vce_encode) / 4;
373
374	r = submit(len, AMDGPU_HW_IP_VCE);
375	CU_ASSERT_EQUAL(r, 0);
376}
377
378static void check_result(struct amdgpu_vce_encode *enc)
379{
380	uint64_t sum;
381	uint32_t s[2] = {180325, 15946};
382	uint32_t *ptr, size;
383	int i, j, r;
384
385	for (i = 0; i < 2; ++i) {
386		r = amdgpu_bo_cpu_map(enc->fb[i].handle, (void **)&enc->fb[i].ptr);
387		CU_ASSERT_EQUAL(r, 0);
388		ptr = (uint32_t *)enc->fb[i].ptr;
389		size = ptr[4] - ptr[9];
390		r = amdgpu_bo_cpu_unmap(enc->fb[i].handle);
391		CU_ASSERT_EQUAL(r, 0);
392		r = amdgpu_bo_cpu_map(enc->bs[i].handle, (void **)&enc->bs[i].ptr);
393		CU_ASSERT_EQUAL(r, 0);
394		for (j = 0, sum = 0; j < size; ++j)
395			sum += enc->bs[i].ptr[j];
396		CU_ASSERT_EQUAL(sum, s[i]);
397		r = amdgpu_bo_cpu_unmap(enc->bs[i].handle);
398		CU_ASSERT_EQUAL(r, 0);
399	}
400}
401
402static void amdgpu_cs_vce_encode(void)
403{
404	uint32_t vbuf_size, bs_size = 0x154000, cpb_size;
405	int r;
406
407	vbuf_size = enc.width * enc.height * 1.5;
408	cpb_size = vbuf_size * 10;
409	num_resources = 0;
410	alloc_resource(&enc.fb[0], 4096, AMDGPU_GEM_DOMAIN_GTT);
411	resources[num_resources++] = enc.fb[0].handle;
412	alloc_resource(&enc.fb[1], 4096, AMDGPU_GEM_DOMAIN_GTT);
413	resources[num_resources++] = enc.fb[1].handle;
414	alloc_resource(&enc.bs[0], bs_size, AMDGPU_GEM_DOMAIN_GTT);
415	resources[num_resources++] = enc.bs[0].handle;
416	alloc_resource(&enc.bs[1], bs_size, AMDGPU_GEM_DOMAIN_GTT);
417	resources[num_resources++] = enc.bs[1].handle;
418	alloc_resource(&enc.vbuf, vbuf_size, AMDGPU_GEM_DOMAIN_VRAM);
419	resources[num_resources++] = enc.vbuf.handle;
420	alloc_resource(&enc.cpb, cpb_size, AMDGPU_GEM_DOMAIN_VRAM);
421	resources[num_resources++] = enc.cpb.handle;
422	resources[num_resources++] = ib_handle;
423
424	r = amdgpu_bo_cpu_map(enc.vbuf.handle, (void **)&enc.vbuf.ptr);
425	CU_ASSERT_EQUAL(r, 0);
426	memcpy(enc.vbuf.ptr, frame, sizeof(frame));
427	r = amdgpu_bo_cpu_unmap(enc.vbuf.handle);
428	CU_ASSERT_EQUAL(r, 0);
429
430	amdgpu_cs_vce_config();
431
432	if (family_id >= AMDGPU_FAMILY_VI) {
433		vce_taskinfo[3] = 3;
434		amdgpu_cs_vce_encode_idr(&enc);
435		amdgpu_cs_vce_encode_p(&enc);
436		check_result(&enc);
437
438		/* two pipes */
439		vce_encode[16] = 0;
440		amdgpu_cs_vce_encode_idr(&enc);
441		amdgpu_cs_vce_encode_p(&enc);
442		check_result(&enc);
443
444		/* two instances */
445		if (vce_harvest_config == 0) {
446			enc.two_instance = true;
447			vce_taskinfo[2] = 0x83;
448			vce_taskinfo[4] = 1;
449			amdgpu_cs_vce_encode_idr(&enc);
450			vce_taskinfo[2] = 0xffffffff;
451			vce_taskinfo[4] = 2;
452			amdgpu_cs_vce_encode_p(&enc);
453			check_result(&enc);
454		}
455	} else {
456		vce_taskinfo[3] = 3;
457		vce_encode[16] = 0;
458		amdgpu_cs_vce_encode_idr(&enc);
459		amdgpu_cs_vce_encode_p(&enc);
460		check_result(&enc);
461	}
462
463	free_resource(&enc.fb[0]);
464	free_resource(&enc.fb[1]);
465	free_resource(&enc.bs[0]);
466	free_resource(&enc.bs[1]);
467	free_resource(&enc.vbuf);
468	free_resource(&enc.cpb);
469}
470
471static void amdgpu_cs_vce_destroy(void)
472{
473	int len, r;
474
475	num_resources  = 0;
476	alloc_resource(&enc.fb[0], 4096, AMDGPU_GEM_DOMAIN_GTT);
477	resources[num_resources++] = enc.fb[0].handle;
478	resources[num_resources++] = ib_handle;
479
480	len = 0;
481	memcpy(ib_cpu, vce_session, sizeof(vce_session));
482	len += sizeof(vce_session) / 4;
483	memcpy((ib_cpu + len), vce_taskinfo, sizeof(vce_taskinfo));
484	ib_cpu[len + 3] = 1;
485	len += sizeof(vce_taskinfo) / 4;
486	memcpy((ib_cpu + len), vce_feedback, sizeof(vce_feedback));
487	ib_cpu[len + 2] = enc.fb[0].addr >> 32;
488	ib_cpu[len + 3] = enc.fb[0].addr;
489	len += sizeof(vce_feedback) / 4;
490	memcpy((ib_cpu + len), vce_destroy, sizeof(vce_destroy));
491	len += sizeof(vce_destroy) / 4;
492
493	r = submit(len, AMDGPU_HW_IP_VCE);
494	CU_ASSERT_EQUAL(r, 0);
495
496	free_resource(&enc.fb[0]);
497}
498