• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright 2017 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 #include "decode_messages.h"
35 
36 #define IB_SIZE		4096
37 #define MAX_RESOURCES	16
38 
39 struct amdgpu_vcn_bo {
40 	amdgpu_bo_handle handle;
41 	amdgpu_va_handle va_handle;
42 	uint64_t addr;
43 	uint64_t size;
44 	uint8_t *ptr;
45 };
46 
47 struct amdgpu_vcn_reg {
48 	uint32_t data0;
49 	uint32_t data1;
50 	uint32_t cmd;
51 	uint32_t nop;
52 	uint32_t cntl;
53 };
54 
55 static amdgpu_device_handle device_handle;
56 static uint32_t major_version;
57 static uint32_t minor_version;
58 static uint32_t family_id;
59 static uint32_t chip_rev;
60 static uint32_t chip_id;
61 static uint32_t asic_id;
62 static uint32_t chip_rev;
63 static uint32_t chip_id;
64 
65 static amdgpu_context_handle context_handle;
66 static amdgpu_bo_handle ib_handle;
67 static amdgpu_va_handle ib_va_handle;
68 static uint64_t ib_mc_address;
69 static uint32_t *ib_cpu;
70 
71 static amdgpu_bo_handle resources[MAX_RESOURCES];
72 static unsigned num_resources;
73 
74 static uint8_t vcn_reg_index;
75 static struct amdgpu_vcn_reg reg[] = {
76 	{0x81c4, 0x81c5, 0x81c3, 0x81ff, 0x81c6},
77 	{0x504, 0x505, 0x503, 0x53f, 0x506},
78 	{0x10, 0x11, 0xf, 0x29, 0x26d},
79 };
80 
81 static void amdgpu_cs_vcn_dec_create(void);
82 static void amdgpu_cs_vcn_dec_decode(void);
83 static void amdgpu_cs_vcn_dec_destroy(void);
84 
85 static void amdgpu_cs_vcn_enc_create(void);
86 static void amdgpu_cs_vcn_enc_encode(void);
87 static void amdgpu_cs_vcn_enc_destroy(void);
88 
89 CU_TestInfo vcn_tests[] = {
90 
91 	{ "VCN DEC create",  amdgpu_cs_vcn_dec_create },
92 	{ "VCN DEC decode",  amdgpu_cs_vcn_dec_decode },
93 	{ "VCN DEC destroy",  amdgpu_cs_vcn_dec_destroy },
94 
95 	{ "VCN ENC create",  amdgpu_cs_vcn_enc_create },
96 	{ "VCN ENC decode",  amdgpu_cs_vcn_enc_encode },
97 	{ "VCN ENC destroy",  amdgpu_cs_vcn_enc_destroy },
98 	CU_TEST_INFO_NULL,
99 };
100 
suite_vcn_tests_enable(void)101 CU_BOOL suite_vcn_tests_enable(void)
102 {
103 	struct drm_amdgpu_info_hw_ip info;
104 	int r;
105 
106 	if (amdgpu_device_initialize(drm_amdgpu[0], &major_version,
107 				   &minor_version, &device_handle))
108 		return CU_FALSE;
109 
110 	family_id = device_handle->info.family_id;
111 	asic_id = device_handle->info.asic_id;
112 	chip_rev = device_handle->info.chip_rev;
113 	chip_id = device_handle->info.chip_external_rev;
114 
115 	r = amdgpu_query_hw_ip_info(device_handle, AMDGPU_HW_IP_VCN_DEC, 0, &info);
116 
117 	if (amdgpu_device_deinitialize(device_handle))
118 			return CU_FALSE;
119 
120 	if (r != 0 || !info.available_rings ||
121 	    (family_id < AMDGPU_FAMILY_RV &&
122 	     (family_id == AMDGPU_FAMILY_AI &&
123 	      (chip_id - chip_rev) < 0x32))) {  /* Arcturus */
124 		printf("\n\nThe ASIC NOT support VCN, suite disabled\n");
125 		return CU_FALSE;
126 	}
127 
128 	if (family_id == AMDGPU_FAMILY_AI) {
129 		amdgpu_set_test_active("VCN Tests", "VCN ENC create", CU_FALSE);
130 		amdgpu_set_test_active("VCN Tests", "VCN ENC decode", CU_FALSE);
131 		amdgpu_set_test_active("VCN Tests", "VCN ENC destroy", CU_FALSE);
132 	}
133 
134 	if (info.hw_ip_version_major == 1)
135 		vcn_reg_index = 0;
136 	else if (info.hw_ip_version_major == 2)
137 		vcn_reg_index = 1;
138 	else if ((info.hw_ip_version_major == 2 && info.hw_ip_version_minor >= 5) ||
139 		  info.hw_ip_version_major == 3)
140 		vcn_reg_index = 2;
141 	else
142 		return CU_FALSE;
143 
144 	return CU_TRUE;
145 }
146 
suite_vcn_tests_init(void)147 int suite_vcn_tests_init(void)
148 {
149 	int r;
150 
151 	r = amdgpu_device_initialize(drm_amdgpu[0], &major_version,
152 				     &minor_version, &device_handle);
153 	if (r)
154 		return CUE_SINIT_FAILED;
155 
156 	family_id = device_handle->info.family_id;
157 
158 	r = amdgpu_cs_ctx_create(device_handle, &context_handle);
159 	if (r)
160 		return CUE_SINIT_FAILED;
161 
162 	r = amdgpu_bo_alloc_and_map(device_handle, IB_SIZE, 4096,
163 				    AMDGPU_GEM_DOMAIN_GTT, 0,
164 				    &ib_handle, (void**)&ib_cpu,
165 				    &ib_mc_address, &ib_va_handle);
166 	if (r)
167 		return CUE_SINIT_FAILED;
168 
169 	return CUE_SUCCESS;
170 }
171 
suite_vcn_tests_clean(void)172 int suite_vcn_tests_clean(void)
173 {
174 	int r;
175 
176 	r = amdgpu_bo_unmap_and_free(ib_handle, ib_va_handle,
177 			     ib_mc_address, IB_SIZE);
178 	if (r)
179 		return CUE_SCLEAN_FAILED;
180 
181 	r = amdgpu_cs_ctx_free(context_handle);
182 	if (r)
183 		return CUE_SCLEAN_FAILED;
184 
185 	r = amdgpu_device_deinitialize(device_handle);
186 	if (r)
187 		return CUE_SCLEAN_FAILED;
188 
189 	return CUE_SUCCESS;
190 }
191 
submit(unsigned ndw,unsigned ip)192 static int submit(unsigned ndw, unsigned ip)
193 {
194 	struct amdgpu_cs_request ibs_request = {0};
195 	struct amdgpu_cs_ib_info ib_info = {0};
196 	struct amdgpu_cs_fence fence_status = {0};
197 	uint32_t expired;
198 	int r;
199 
200 	ib_info.ib_mc_address = ib_mc_address;
201 	ib_info.size = ndw;
202 
203 	ibs_request.ip_type = ip;
204 
205 	r = amdgpu_bo_list_create(device_handle, num_resources, resources,
206 				  NULL, &ibs_request.resources);
207 	if (r)
208 		return r;
209 
210 	ibs_request.number_of_ibs = 1;
211 	ibs_request.ibs = &ib_info;
212 	ibs_request.fence_info.handle = NULL;
213 
214 	r = amdgpu_cs_submit(context_handle, 0, &ibs_request, 1);
215 	if (r)
216 		return r;
217 
218 	r = amdgpu_bo_list_destroy(ibs_request.resources);
219 	if (r)
220 		return r;
221 
222 	fence_status.context = context_handle;
223 	fence_status.ip_type = ip;
224 	fence_status.fence = ibs_request.seq_no;
225 
226 	r = amdgpu_cs_query_fence_status(&fence_status,
227 					 AMDGPU_TIMEOUT_INFINITE,
228 					 0, &expired);
229 	if (r)
230 		return r;
231 
232 	return 0;
233 }
234 
alloc_resource(struct amdgpu_vcn_bo * vcn_bo,unsigned size,unsigned domain)235 static void alloc_resource(struct amdgpu_vcn_bo *vcn_bo,
236 			unsigned size, unsigned domain)
237 {
238 	struct amdgpu_bo_alloc_request req = {0};
239 	amdgpu_bo_handle buf_handle;
240 	amdgpu_va_handle va_handle;
241 	uint64_t va = 0;
242 	int r;
243 
244 	req.alloc_size = ALIGN(size, 4096);
245 	req.preferred_heap = domain;
246 	r = amdgpu_bo_alloc(device_handle, &req, &buf_handle);
247 	CU_ASSERT_EQUAL(r, 0);
248 	r = amdgpu_va_range_alloc(device_handle,
249 				  amdgpu_gpu_va_range_general,
250 				  req.alloc_size, 1, 0, &va,
251 				  &va_handle, 0);
252 	CU_ASSERT_EQUAL(r, 0);
253 	r = amdgpu_bo_va_op(buf_handle, 0, req.alloc_size, va, 0,
254 			    AMDGPU_VA_OP_MAP);
255 	CU_ASSERT_EQUAL(r, 0);
256 	vcn_bo->addr = va;
257 	vcn_bo->handle = buf_handle;
258 	vcn_bo->size = req.alloc_size;
259 	vcn_bo->va_handle = va_handle;
260 	r = amdgpu_bo_cpu_map(vcn_bo->handle, (void **)&vcn_bo->ptr);
261 	CU_ASSERT_EQUAL(r, 0);
262 	memset(vcn_bo->ptr, 0, size);
263 	r = amdgpu_bo_cpu_unmap(vcn_bo->handle);
264 	CU_ASSERT_EQUAL(r, 0);
265 }
266 
free_resource(struct amdgpu_vcn_bo * vcn_bo)267 static void free_resource(struct amdgpu_vcn_bo *vcn_bo)
268 {
269 	int r;
270 
271 	r = amdgpu_bo_va_op(vcn_bo->handle, 0, vcn_bo->size,
272 			    vcn_bo->addr, 0, AMDGPU_VA_OP_UNMAP);
273 	CU_ASSERT_EQUAL(r, 0);
274 
275 	r = amdgpu_va_range_free(vcn_bo->va_handle);
276 	CU_ASSERT_EQUAL(r, 0);
277 
278 	r = amdgpu_bo_free(vcn_bo->handle);
279 	CU_ASSERT_EQUAL(r, 0);
280 	memset(vcn_bo, 0, sizeof(*vcn_bo));
281 }
282 
vcn_dec_cmd(uint64_t addr,unsigned cmd,int * idx)283 static void vcn_dec_cmd(uint64_t addr, unsigned cmd, int *idx)
284 {
285 	ib_cpu[(*idx)++] = reg[vcn_reg_index].data0;
286 	ib_cpu[(*idx)++] = addr;
287 	ib_cpu[(*idx)++] = reg[vcn_reg_index].data1;
288 	ib_cpu[(*idx)++] = addr >> 32;
289 	ib_cpu[(*idx)++] = reg[vcn_reg_index].cmd;
290 	ib_cpu[(*idx)++] = cmd << 1;
291 }
292 
amdgpu_cs_vcn_dec_create(void)293 static void amdgpu_cs_vcn_dec_create(void)
294 {
295 	struct amdgpu_vcn_bo msg_buf;
296 	int len, r;
297 
298 	num_resources  = 0;
299 	alloc_resource(&msg_buf, 4096, AMDGPU_GEM_DOMAIN_GTT);
300 	resources[num_resources++] = msg_buf.handle;
301 	resources[num_resources++] = ib_handle;
302 
303 	r = amdgpu_bo_cpu_map(msg_buf.handle, (void **)&msg_buf.ptr);
304 	CU_ASSERT_EQUAL(r, 0);
305 
306 	memset(msg_buf.ptr, 0, 4096);
307 	memcpy(msg_buf.ptr, vcn_dec_create_msg, sizeof(vcn_dec_create_msg));
308 
309 	len = 0;
310 	ib_cpu[len++] = reg[vcn_reg_index].data0;
311 	ib_cpu[len++] = msg_buf.addr;
312 	ib_cpu[len++] = reg[vcn_reg_index].data1;
313 	ib_cpu[len++] = msg_buf.addr >> 32;
314 	ib_cpu[len++] = reg[vcn_reg_index].cmd;
315 	ib_cpu[len++] = 0;
316 	for (; len % 16; ) {
317 		ib_cpu[len++] = reg[vcn_reg_index].nop;
318 		ib_cpu[len++] = 0;
319 	}
320 
321 	r = submit(len, AMDGPU_HW_IP_VCN_DEC);
322 	CU_ASSERT_EQUAL(r, 0);
323 
324 	free_resource(&msg_buf);
325 }
326 
amdgpu_cs_vcn_dec_decode(void)327 static void amdgpu_cs_vcn_dec_decode(void)
328 {
329 	const unsigned dpb_size = 15923584, dt_size = 737280;
330 	uint64_t msg_addr, fb_addr, bs_addr, dpb_addr, ctx_addr, dt_addr, it_addr, sum;
331 	struct amdgpu_vcn_bo dec_buf;
332 	int size, len, i, r;
333 	uint8_t *dec;
334 
335 	size = 4*1024; /* msg */
336 	size += 4*1024; /* fb */
337 	size += 4096; /*it_scaling_table*/
338 	size += ALIGN(sizeof(uvd_bitstream), 4*1024);
339 	size += ALIGN(dpb_size, 4*1024);
340 	size += ALIGN(dt_size, 4*1024);
341 
342 	num_resources  = 0;
343 	alloc_resource(&dec_buf, size, AMDGPU_GEM_DOMAIN_GTT);
344 	resources[num_resources++] = dec_buf.handle;
345 	resources[num_resources++] = ib_handle;
346 
347 	r = amdgpu_bo_cpu_map(dec_buf.handle, (void **)&dec_buf.ptr);
348 	dec = dec_buf.ptr;
349 
350 	CU_ASSERT_EQUAL(r, 0);
351 	memset(dec_buf.ptr, 0, size);
352 	memcpy(dec_buf.ptr, vcn_dec_decode_msg, sizeof(vcn_dec_decode_msg));
353 	memcpy(dec_buf.ptr + sizeof(vcn_dec_decode_msg),
354 			avc_decode_msg, sizeof(avc_decode_msg));
355 
356 	dec += 4*1024;
357 	memcpy(dec, feedback_msg, sizeof(feedback_msg));
358 	dec += 4*1024;
359 	memcpy(dec, uvd_it_scaling_table, sizeof(uvd_it_scaling_table));
360 
361 	dec += 4*1024;
362 	memcpy(dec, uvd_bitstream, sizeof(uvd_bitstream));
363 
364 	dec += ALIGN(sizeof(uvd_bitstream), 4*1024);
365 
366 	dec += ALIGN(dpb_size, 4*1024);
367 
368 	msg_addr = dec_buf.addr;
369 	fb_addr = msg_addr + 4*1024;
370 	it_addr = fb_addr + 4*1024;
371 	bs_addr = it_addr + 4*1024;
372 	dpb_addr = ALIGN(bs_addr + sizeof(uvd_bitstream), 4*1024);
373 	ctx_addr = ALIGN(dpb_addr + 0x006B9400, 4*1024);
374 	dt_addr = ALIGN(dpb_addr + dpb_size, 4*1024);
375 
376 	len = 0;
377 	vcn_dec_cmd(msg_addr, 0x0, &len);
378 	vcn_dec_cmd(dpb_addr, 0x1, &len);
379 	vcn_dec_cmd(dt_addr, 0x2, &len);
380 	vcn_dec_cmd(fb_addr, 0x3, &len);
381 	vcn_dec_cmd(bs_addr, 0x100, &len);
382 	vcn_dec_cmd(it_addr, 0x204, &len);
383 	vcn_dec_cmd(ctx_addr, 0x206, &len);
384 
385 	ib_cpu[len++] = reg[vcn_reg_index].cntl;
386 	ib_cpu[len++] = 0x1;
387 	for (; len % 16; ) {
388 		ib_cpu[len++] = reg[vcn_reg_index].nop;
389 		ib_cpu[len++] = 0;
390 	}
391 
392 	r = submit(len, AMDGPU_HW_IP_VCN_DEC);
393 	CU_ASSERT_EQUAL(r, 0);
394 
395 	for (i = 0, sum = 0; i < dt_size; ++i)
396 		sum += dec[i];
397 
398 	CU_ASSERT_EQUAL(sum, SUM_DECODE);
399 
400 	free_resource(&dec_buf);
401 }
402 
amdgpu_cs_vcn_dec_destroy(void)403 static void amdgpu_cs_vcn_dec_destroy(void)
404 {
405 	struct amdgpu_vcn_bo msg_buf;
406 	int len, r;
407 
408 	num_resources  = 0;
409 	alloc_resource(&msg_buf, 1024, AMDGPU_GEM_DOMAIN_GTT);
410 	resources[num_resources++] = msg_buf.handle;
411 	resources[num_resources++] = ib_handle;
412 
413 	r = amdgpu_bo_cpu_map(msg_buf.handle, (void **)&msg_buf.ptr);
414 	CU_ASSERT_EQUAL(r, 0);
415 
416 	memset(msg_buf.ptr, 0, 1024);
417 	memcpy(msg_buf.ptr, vcn_dec_destroy_msg, sizeof(vcn_dec_destroy_msg));
418 
419 	len = 0;
420 	ib_cpu[len++] = reg[vcn_reg_index].data0;
421 	ib_cpu[len++] = msg_buf.addr;
422 	ib_cpu[len++] = reg[vcn_reg_index].data1;
423 	ib_cpu[len++] = msg_buf.addr >> 32;
424 	ib_cpu[len++] = reg[vcn_reg_index].cmd;
425 	ib_cpu[len++] = 0;
426 	for (; len % 16; ) {
427 		ib_cpu[len++] = reg[vcn_reg_index].nop;
428 		ib_cpu[len++] = 0;
429 	}
430 
431 	r = submit(len, AMDGPU_HW_IP_VCN_DEC);
432 	CU_ASSERT_EQUAL(r, 0);
433 
434 	free_resource(&msg_buf);
435 }
436 
amdgpu_cs_vcn_enc_create(void)437 static void amdgpu_cs_vcn_enc_create(void)
438 {
439 	/* TODO */
440 }
441 
amdgpu_cs_vcn_enc_encode(void)442 static void amdgpu_cs_vcn_enc_encode(void)
443 {
444 	/* TODO */
445 }
446 
amdgpu_cs_vcn_enc_destroy(void)447 static void amdgpu_cs_vcn_enc_destroy(void)
448 {
449 	/* TODO */
450 }
451