1 /**************************************************************************
3 * Copyright 2013 Advanced Micro Devices, Inc.
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21 * IN NO EVENT SHALL THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR
22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 **************************************************************************/
30 * Christian König <christian.koenig@amd.com>
36 #include "pipe/p_video_codec.h"
38 #include "util/u_video.h"
39 #include "util/u_memory.h"
41 #include "vl/vl_video_buffer.h"
43 #include "r600_pipe_common.h"
44 #include "radeon_video.h"
45 #include "radeon_vce.h"
47 static const unsigned profiles
[7] = { 66, 77, 88, 100, 110, 122, 244 };
49 static void session(struct rvce_encoder
*enc
)
51 RVCE_BEGIN(0x00000001); // session cmd
52 RVCE_CS(enc
->stream_handle
);
56 static void task_info(struct rvce_encoder
*enc
, uint32_t taskOperation
)
58 RVCE_BEGIN(0x00000002); // task info
59 RVCE_CS(0xffffffff); // offsetOfNextTaskInfo
60 RVCE_CS(taskOperation
); // taskOperation
61 RVCE_CS(0x00000000); // referencePictureDependency
62 RVCE_CS(0x00000000); // collocateFlagDependency
63 RVCE_CS(0x00000000); // feedbackIndex
64 RVCE_CS(0x00000000); // videoBitstreamRingIndex
68 static void feedback(struct rvce_encoder
*enc
)
70 RVCE_BEGIN(0x05000005); // feedback buffer
71 RVCE_WRITE(enc
->fb
->res
->cs_buf
, enc
->fb
->res
->domains
); // feedbackRingAddressHi
72 RVCE_CS(0x00000000); // feedbackRingAddressLo
73 RVCE_CS(0x00000001); // feedbackRingSize
77 static void create(struct rvce_encoder
*enc
)
79 task_info(enc
, 0x00000000);
81 RVCE_BEGIN(0x01000001); // create cmd
82 RVCE_CS(0x00000000); // encUseCircularBuffer
83 RVCE_CS(profiles
[enc
->base
.profile
-
84 PIPE_VIDEO_PROFILE_MPEG4_AVC_BASELINE
]); // encProfile
85 RVCE_CS(enc
->base
.level
); // encLevel
86 RVCE_CS(0x00000000); // encPicStructRestriction
87 RVCE_CS(enc
->base
.width
); // encImageWidth
88 RVCE_CS(enc
->base
.height
); // encImageHeight
89 RVCE_CS(enc
->luma
->level
[0].pitch_bytes
); // encRefPicLumaPitch
90 RVCE_CS(enc
->chroma
->level
[0].pitch_bytes
); // encRefPicChromaPitch
91 RVCE_CS(align(enc
->luma
->npix_y
, 16) / 8); // encRefYHeightInQw
92 RVCE_CS(0x00000000); // encRefPic(Addr|Array)Mode, encPicStructRestriction, disableRDO
96 static void rate_control(struct rvce_encoder
*enc
)
98 RVCE_BEGIN(0x04000005); // rate control
99 RVCE_CS(enc
->pic
.rate_ctrl
.rate_ctrl_method
); // encRateControlMethod
100 RVCE_CS(enc
->pic
.rate_ctrl
.target_bitrate
); // encRateControlTargetBitRate
101 RVCE_CS(enc
->pic
.rate_ctrl
.peak_bitrate
); // encRateControlPeakBitRate
102 RVCE_CS(enc
->pic
.rate_ctrl
.frame_rate_num
); // encRateControlFrameRateNum
103 RVCE_CS(0x00000000); // encGOPSize
104 RVCE_CS(enc
->pic
.quant_i_frames
); // encQP_I
105 RVCE_CS(enc
->pic
.quant_p_frames
); // encQP_P
106 RVCE_CS(enc
->pic
.quant_b_frames
); // encQP_B
107 RVCE_CS(enc
->pic
.rate_ctrl
.vbv_buffer_size
); // encVBVBufferSize
108 RVCE_CS(enc
->pic
.rate_ctrl
.frame_rate_den
); // encRateControlFrameRateDen
109 RVCE_CS(0x00000000); // encVBVBufferLevel
110 RVCE_CS(0x00000000); // encMaxAUSize
111 RVCE_CS(0x00000000); // encQPInitialMode
112 RVCE_CS(enc
->pic
.rate_ctrl
.target_bits_picture
); // encTargetBitsPerPicture
113 RVCE_CS(enc
->pic
.rate_ctrl
.peak_bits_picture_integer
); // encPeakBitsPerPictureInteger
114 RVCE_CS(enc
->pic
.rate_ctrl
.peak_bits_picture_fraction
); // encPeakBitsPerPictureFractional
115 RVCE_CS(0x00000000); // encMinQP
116 RVCE_CS(0x00000033); // encMaxQP
117 RVCE_CS(0x00000000); // encSkipFrameEnable
118 RVCE_CS(0x00000000); // encFillerDataEnable
119 RVCE_CS(0x00000000); // encEnforceHRD
120 RVCE_CS(0x00000000); // encBPicsDeltaQP
121 RVCE_CS(0x00000000); // encReferenceBPicsDeltaQP
122 RVCE_CS(0x00000000); // encRateControlReInitDisable
126 static void config_extension(struct rvce_encoder
*enc
)
128 RVCE_BEGIN(0x04000001); // config extension
129 RVCE_CS(0x00000003); // encEnablePerfLogging
133 static void pic_control(struct rvce_encoder
*enc
)
135 unsigned encNumMBsPerSlice
;
137 encNumMBsPerSlice
= align(enc
->base
.width
, 16) / 16;
138 encNumMBsPerSlice
*= align(enc
->base
.height
, 16) / 16;
140 RVCE_BEGIN(0x04000002); // pic control
141 RVCE_CS(0x00000000); // encUseConstrainedIntraPred
142 RVCE_CS(0x00000000); // encCABACEnable
143 RVCE_CS(0x00000000); // encCABACIDC
144 RVCE_CS(0x00000000); // encLoopFilterDisable
145 RVCE_CS(0x00000000); // encLFBetaOffset
146 RVCE_CS(0x00000000); // encLFAlphaC0Offset
147 RVCE_CS(0x00000000); // encCropLeftOffset
148 RVCE_CS((align(enc
->base
.width
, 16) - enc
->base
.width
) >> 1); // encCropRightOffset
149 RVCE_CS(0x00000000); // encCropTopOffset
150 RVCE_CS((align(enc
->base
.height
, 16) - enc
->base
.height
) >> 1); // encCropBottomOffset
151 RVCE_CS(encNumMBsPerSlice
); // encNumMBsPerSlice
152 RVCE_CS(0x00000000); // encIntraRefreshNumMBsPerSlot
153 RVCE_CS(0x00000000); // encForceIntraRefresh
154 RVCE_CS(0x00000000); // encForceIMBPeriod
155 RVCE_CS(0x00000000); // encPicOrderCntType
156 RVCE_CS(0x00000000); // log2_max_pic_order_cnt_lsb_minus4
157 RVCE_CS(0x00000000); // encSPSID
158 RVCE_CS(0x00000000); // encPPSID
159 RVCE_CS(0x00000040); // encConstraintSetFlags
160 RVCE_CS(MAX2(enc
->base
.max_references
, 1) - 1); // encBPicPattern
161 RVCE_CS(0x00000000); // weightPredModeBPicture
162 RVCE_CS(MIN2(enc
->base
.max_references
, 2)); // encNumberOfReferenceFrames
163 RVCE_CS(enc
->base
.max_references
+ 1); // encMaxNumRefFrames
164 RVCE_CS(0x00000001); // encNumDefaultActiveRefL0
165 RVCE_CS(0x00000001); // encNumDefaultActiveRefL1
166 RVCE_CS(0x00000000); // encSliceMode
167 RVCE_CS(0x00000000); // encMaxSliceSize
171 static void motion_estimation(struct rvce_encoder
*enc
)
173 RVCE_BEGIN(0x04000007); // motion estimation
174 RVCE_CS(0x00000001); // encIMEDecimationSearch
175 RVCE_CS(0x00000001); // motionEstHalfPixel
176 RVCE_CS(0x00000000); // motionEstQuarterPixel
177 RVCE_CS(0x00000000); // disableFavorPMVPoint
178 RVCE_CS(0x00000000); // forceZeroPointCenter
179 RVCE_CS(0x00000000); // LSMVert
180 RVCE_CS(0x00000010); // encSearchRangeX
181 RVCE_CS(0x00000010); // encSearchRangeY
182 RVCE_CS(0x00000010); // encSearch1RangeX
183 RVCE_CS(0x00000010); // encSearch1RangeY
184 RVCE_CS(0x00000000); // disable16x16Frame1
185 RVCE_CS(0x00000000); // disableSATD
186 RVCE_CS(0x00000000); // enableAMD
187 RVCE_CS(0x000000fe); // encDisableSubMode
188 RVCE_CS(0x00000000); // encIMESkipX
189 RVCE_CS(0x00000000); // encIMESkipY
190 RVCE_CS(0x00000000); // encEnImeOverwDisSubm
191 RVCE_CS(0x00000000); // encImeOverwDisSubmNo
192 RVCE_CS(0x00000001); // encIME2SearchRangeX
193 RVCE_CS(0x00000001); // encIME2SearchRangeY
194 RVCE_CS(0x00000000); // parallelModeSpeedupEnable
195 RVCE_CS(0x00000000); // fme0_encDisableSubMode
196 RVCE_CS(0x00000000); // fme1_encDisableSubMode
197 RVCE_CS(0x00000000); // imeSWSpeedupEnable
201 static void rdo(struct rvce_encoder
*enc
)
203 RVCE_BEGIN(0x04000008); // rdo
204 RVCE_CS(0x00000000); // encDisableTbePredIFrame
205 RVCE_CS(0x00000000); // encDisableTbePredPFrame
206 RVCE_CS(0x00000000); // useFmeInterpolY
207 RVCE_CS(0x00000000); // useFmeInterpolUV
208 RVCE_CS(0x00000000); // useFmeIntrapolY
209 RVCE_CS(0x00000000); // useFmeIntrapolUV
210 RVCE_CS(0x00000000); // useFmeInterpolY_1
211 RVCE_CS(0x00000000); // useFmeInterpolUV_1
212 RVCE_CS(0x00000000); // useFmeIntrapolY_1
213 RVCE_CS(0x00000000); // useFmeIntrapolUV_1
214 RVCE_CS(0x00000000); // enc16x16CostAdj
215 RVCE_CS(0x00000000); // encSkipCostAdj
216 RVCE_CS(0x00000000); // encForce16x16skip
217 RVCE_CS(0x00000000); // encDisableThresholdCalcA
218 RVCE_CS(0x00000000); // encLumaCoeffCost
219 RVCE_CS(0x00000000); // encLumaMBCoeffCost
220 RVCE_CS(0x00000000); // encChromaCoeffCost
224 static void vui(struct rvce_encoder
*enc
)
228 RVCE_BEGIN(0x04000009); // vui
229 RVCE_CS(0x00000000); //aspectRatioInfoPresentFlag
230 RVCE_CS(0x00000000); //aspectRatioInfo.aspectRatioIdc
231 RVCE_CS(0x00000000); //aspectRatioInfo.sarWidth
232 RVCE_CS(0x00000000); //aspectRatioInfo.sarHeight
233 RVCE_CS(0x00000000); //overscanInfoPresentFlag
234 RVCE_CS(0x00000000); //overScanInfo.overscanAppropFlag
235 RVCE_CS(0x00000000); //videoSignalTypePresentFlag
236 RVCE_CS(0x00000005); //videoSignalTypeInfo.videoFormat
237 RVCE_CS(0x00000000); //videoSignalTypeInfo.videoFullRangeFlag
238 RVCE_CS(0x00000000); //videoSignalTypeInfo.colorDescriptionPresentFlag
239 RVCE_CS(0x00000002); //videoSignalTypeInfo.colorPrim
240 RVCE_CS(0x00000002); //videoSignalTypeInfo.transferChar
241 RVCE_CS(0x00000002); //videoSignalTypeInfo.matrixCoef
242 RVCE_CS(0x00000000); //chromaLocInfoPresentFlag
243 RVCE_CS(0x00000000); //chromaLocInfo.chromaLocTop
244 RVCE_CS(0x00000000); //chromaLocInfo.chromaLocBottom
245 RVCE_CS(0x00000001); //timingInfoPresentFlag
246 RVCE_CS(enc
->pic
.rate_ctrl
.frame_rate_den
); //timingInfo.numUnitsInTick
247 RVCE_CS(enc
->pic
.rate_ctrl
.frame_rate_num
* 2); //timingInfo.timeScale;
248 RVCE_CS(0x00000001); //timingInfo.fixedFrameRateFlag
249 RVCE_CS(0x00000000); //nalHRDParametersPresentFlag
250 RVCE_CS(0x00000000); //hrdParam.cpbCntMinus1
251 RVCE_CS(0x00000004); //hrdParam.bitRateScale
252 RVCE_CS(0x00000006); //hrdParam.cpbSizeScale
253 for (i
= 0; i
< 32; i
++) {
254 RVCE_CS(0x00000000); //hrdParam.bitRateValueMinus
255 RVCE_CS(0x00000000); //hrdParam.cpbSizeValueMinus
256 RVCE_CS(0x00000000); //hrdParam.cbrFlag
258 RVCE_CS(0x00000017); //hrdParam.initialCpbRemovalDelayLengthMinus1
259 RVCE_CS(0x00000017); //hrdParam.cpbRemovalDelayLengthMinus1
260 RVCE_CS(0x00000017); //hrdParam.dpbOutputDelayLengthMinus1
261 RVCE_CS(0x00000018); //hrdParam.timeOffsetLength
262 RVCE_CS(0x00000000); //lowDelayHRDFlag
263 RVCE_CS(0x00000000); //picStructPresentFlag
264 RVCE_CS(0x00000000); //bitstreamRestrictionPresentFlag
265 RVCE_CS(0x00000001); //bitstreamRestrictions.motionVectorsOverPicBoundariesFlag
266 RVCE_CS(0x00000002); //bitstreamRestrictions.maxBytesPerPicDenom
267 RVCE_CS(0x00000001); //bitstreamRestrictions.maxBitsPerMbDenom
268 RVCE_CS(0x00000010); //bitstreamRestrictions.log2MaxMvLengthHori
269 RVCE_CS(0x00000010); //bitstreamRestrictions.log2MaxMvLengthVert
270 RVCE_CS(0x00000003); //bitstreamRestrictions.numReorderFrames
271 RVCE_CS(0x00000003); //bitstreamRestrictions.maxDecFrameBuffering
275 static void encode(struct rvce_encoder
*enc
)
278 unsigned luma_offset
, chroma_offset
;
280 task_info(enc
, 0x00000003);
282 RVCE_BEGIN(0x05000001); // context buffer
283 RVCE_READWRITE(enc
->cpb
.res
->cs_buf
, enc
->cpb
.res
->domains
); // encodeContextAddressHi
284 RVCE_CS(0x00000000); // encodeContextAddressLo
287 RVCE_BEGIN(0x05000004); // video bitstream buffer
288 RVCE_WRITE(enc
->bs_handle
, RADEON_DOMAIN_GTT
); // videoBitstreamRingAddressHi
289 RVCE_CS(0x00000000); // videoBitstreamRingAddressLo
290 RVCE_CS(enc
->bs_size
); // videoBitstreamRingSize
293 RVCE_BEGIN(0x03000001); // encode
294 RVCE_CS(0x00000000); // insertHeaders
295 RVCE_CS(0x00000000); // pictureStructure
296 RVCE_CS(enc
->bs_size
); // allowedMaxBitstreamSize
297 RVCE_CS(0x00000000); // forceRefreshMap
298 RVCE_CS(0x00000000); // insertAUD
299 RVCE_CS(0x00000000); // endOfSequence
300 RVCE_CS(0x00000000); // endOfStream
301 RVCE_READ(enc
->handle
, RADEON_DOMAIN_VRAM
); // inputPictureLumaAddressHi
302 RVCE_CS(enc
->luma
->level
[0].offset
); // inputPictureLumaAddressLo
303 RVCE_READ(enc
->handle
, RADEON_DOMAIN_VRAM
); // inputPictureChromaAddressHi
304 RVCE_CS(enc
->chroma
->level
[0].offset
); // inputPictureChromaAddressLo
305 RVCE_CS(align(enc
->luma
->npix_y
, 16)); // encInputFrameYPitch
306 RVCE_CS(enc
->luma
->level
[0].pitch_bytes
); // encInputPicLumaPitch
307 RVCE_CS(enc
->chroma
->level
[0].pitch_bytes
); // encInputPicChromaPitch
308 RVCE_CS(0x00000000); // encInputPic(Addr|Array)Mode
309 RVCE_CS(0x00000000); // encInputPicTileConfig
310 RVCE_CS(enc
->pic
.picture_type
); // encPicType
311 RVCE_CS(enc
->pic
.picture_type
== PIPE_H264_ENC_PICTURE_TYPE_IDR
); // encIdrFlag
312 RVCE_CS(0x00000000); // encIdrPicId
313 RVCE_CS(0x00000000); // encMGSKeyPic
314 RVCE_CS(!enc
->pic
.not_referenced
); // encReferenceFlag
315 RVCE_CS(0x00000000); // encTemporalLayerIndex
316 RVCE_CS(0x00000000); // num_ref_idx_active_override_flag
317 RVCE_CS(0x00000000); // num_ref_idx_l0_active_minus1
318 RVCE_CS(0x00000000); // num_ref_idx_l1_active_minus1
320 i
= enc
->pic
.frame_num
- enc
->pic
.ref_idx_l0
;
321 if (i
> 1 && enc
->pic
.picture_type
== PIPE_H264_ENC_PICTURE_TYPE_P
) {
322 RVCE_CS(0x00000001); // encRefListModificationOp
323 RVCE_CS(i
- 1); // encRefListModificationNum
325 RVCE_CS(0x00000000); // encRefListModificationOp
326 RVCE_CS(0x00000000); // encRefListModificationNum
329 for (i
= 0; i
< 3; ++i
) {
330 RVCE_CS(0x00000000); // encRefListModificationOp
331 RVCE_CS(0x00000000); // encRefListModificationNum
333 for (i
= 0; i
< 4; ++i
) {
334 RVCE_CS(0x00000000); // encDecodedPictureMarkingOp
335 RVCE_CS(0x00000000); // encDecodedPictureMarkingNum
336 RVCE_CS(0x00000000); // encDecodedPictureMarkingIdx
337 RVCE_CS(0x00000000); // encDecodedRefBasePictureMarkingOp
338 RVCE_CS(0x00000000); // encDecodedRefBasePictureMarkingNum
341 // encReferencePictureL0[0]
342 RVCE_CS(0x00000000); // pictureStructure
343 if(enc
->pic
.picture_type
== PIPE_H264_ENC_PICTURE_TYPE_P
||
344 enc
->pic
.picture_type
== PIPE_H264_ENC_PICTURE_TYPE_B
) {
345 struct rvce_cpb_slot
*l0
= l0_slot(enc
);
346 rvce_frame_offset(enc
, l0
, &luma_offset
, &chroma_offset
);
347 RVCE_CS(l0
->picture_type
); // encPicType
348 RVCE_CS(l0
->frame_num
); // frameNumber
349 RVCE_CS(l0
->pic_order_cnt
); // pictureOrderCount
350 RVCE_CS(luma_offset
); // lumaOffset
351 RVCE_CS(chroma_offset
); // chromaOffset
353 RVCE_CS(0x00000000); // encPicType
354 RVCE_CS(0x00000000); // frameNumber
355 RVCE_CS(0x00000000); // pictureOrderCount
356 RVCE_CS(0xffffffff); // lumaOffset
357 RVCE_CS(0xffffffff); // chromaOffset
360 // encReferencePictureL0[1]
361 RVCE_CS(0x00000000); // pictureStructure
362 RVCE_CS(0x00000000); // encPicType
363 RVCE_CS(0x00000000); // frameNumber
364 RVCE_CS(0x00000000); // pictureOrderCount
365 RVCE_CS(0xffffffff); // lumaOffset
366 RVCE_CS(0xffffffff); // chromaOffset
368 // encReferencePictureL1[0]
369 RVCE_CS(0x00000000); // pictureStructure
370 if(enc
->pic
.picture_type
== PIPE_H264_ENC_PICTURE_TYPE_B
) {
371 struct rvce_cpb_slot
*l1
= l1_slot(enc
);
372 rvce_frame_offset(enc
, l1
, &luma_offset
, &chroma_offset
);
373 RVCE_CS(l1
->picture_type
); // encPicType
374 RVCE_CS(l1
->frame_num
); // frameNumber
375 RVCE_CS(l1
->pic_order_cnt
); // pictureOrderCount
376 RVCE_CS(luma_offset
); // lumaOffset
377 RVCE_CS(chroma_offset
); // chromaOffset
379 RVCE_CS(0x00000000); // encPicType
380 RVCE_CS(0x00000000); // frameNumber
381 RVCE_CS(0x00000000); // pictureOrderCount
382 RVCE_CS(0xffffffff); // lumaOffset
383 RVCE_CS(0xffffffff); // chromaOffset
386 rvce_frame_offset(enc
, current_slot(enc
), &luma_offset
, &chroma_offset
);
387 RVCE_CS(luma_offset
); // encReconstructedLumaOffset
388 RVCE_CS(chroma_offset
); // encReconstructedChromaOffset
389 RVCE_CS(0x00000000); // encColocBufferOffset
390 RVCE_CS(0x00000000); // encReconstructedRefBasePictureLumaOffset
391 RVCE_CS(0x00000000); // encReconstructedRefBasePictureChromaOffset
392 RVCE_CS(0x00000000); // encReferenceRefBasePictureLumaOffset
393 RVCE_CS(0x00000000); // encReferenceRefBasePictureChromaOffset
394 RVCE_CS(0x00000000); // pictureCount
395 RVCE_CS(enc
->pic
.frame_num
); // frameNumber
396 RVCE_CS(enc
->pic
.pic_order_cnt
); // pictureOrderCount
397 RVCE_CS(0x00000000); // numIPicRemainInRCGOP
398 RVCE_CS(0x00000000); // numPPicRemainInRCGOP
399 RVCE_CS(0x00000000); // numBPicRemainInRCGOP
400 RVCE_CS(0x00000000); // numIRPicRemainInRCGOP
401 RVCE_CS(0x00000000); // enableIntraRefresh
405 static void destroy(struct rvce_encoder
*enc
)
407 task_info(enc
, 0x00000001);
409 RVCE_BEGIN(0x02000001); // destroy
413 void radeon_vce_40_2_2_init(struct rvce_encoder
*enc
)
415 enc
->session
= session
;
416 enc
->create
= create
;
417 enc
->feedback
= feedback
;
418 enc
->rate_control
= rate_control
;
419 enc
->config_extension
= config_extension
;
420 enc
->pic_control
= pic_control
;
421 enc
->motion_estimation
= motion_estimation
;
424 enc
->encode
= encode
;
425 enc
->destroy
= destroy
;