src/gallium/drivers/radeon/radeon_vce_40_2_2.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2013 Advanced Micro Devices, Inc.
   4  * All Rights Reserved.
   5  *
   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:
  13  *
  14  * The above copyright notice and this permission notice (including the
  15  * next paragraph) shall be included in all copies or substantial portions
  16  * of the Software.
  17  *
  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.
  25  *
  26  **************************************************************************/
  27
  28 /*
  29  * Authors:
  30  *      Christian König <christian.koenig@amd.com>
  31  *
  32  */
  33
  34 #include <stdio.h>
  35
  36 #include "pipe/p_video_codec.h"
  37
  38 #include "util/u_video.h"
  39 #include "util/u_memory.h"
  40
  41 #include "vl/vl_video_buffer.h"
  42
  43 #include "../../winsys/radeon/drm/radeon_winsys.h"
  44 #include "radeon_video.h"
  45 #include "radeon_vce.h"
  46
  47 static const unsigned profiles[7] = { 66, 77, 88, 100, 110, 122, 244 };
  48
  49 static struct rvce_cpb_slot *current_slot(struct rvce_encoder *enc)
  50 {
  51         return LIST_ENTRY(struct rvce_cpb_slot, enc->cpb_slots.prev, list);
  52 }
  53
  54 static struct rvce_cpb_slot *l0_slot(struct rvce_encoder *enc)
  55 {
  56         return LIST_ENTRY(struct rvce_cpb_slot, enc->cpb_slots.next, list);
  57 }
  58
  59 static struct rvce_cpb_slot *l1_slot(struct rvce_encoder *enc)
  60 {
  61         return LIST_ENTRY(struct rvce_cpb_slot, enc->cpb_slots.next->next, list);
  62 }
  63
  64 static void frame_offset(struct rvce_encoder *enc, struct rvce_cpb_slot *slot,
  65                          unsigned *luma_offset, unsigned *chroma_offset)
  66 {
  67         unsigned pitch = align(enc->luma->level[0].pitch_bytes, 128);
  68         unsigned vpitch = align(enc->luma->npix_y, 16);
  69         unsigned fsize = pitch * (vpitch + vpitch / 2);
  70
  71         *luma_offset = slot->index * fsize;
  72         *chroma_offset = *luma_offset + pitch * vpitch;
  73 }
  74
  75 static void session(struct rvce_encoder *enc)
  76 {
  77         RVCE_BEGIN(0x00000001); // session cmd
  78         RVCE_CS(enc->stream_handle);
  79         RVCE_END();
  80 }
  81
  82 static void task_info(struct rvce_encoder *enc, uint32_t taskOperation)
  83 {
  84         RVCE_BEGIN(0x00000002); // task info
  85         RVCE_CS(0xffffffff); // offsetOfNextTaskInfo
  86         RVCE_CS(taskOperation); // taskOperation
  87         RVCE_CS(0x00000000); // referencePictureDependency
  88         RVCE_CS(0x00000000); // collocateFlagDependency
  89         RVCE_CS(0x00000000); // feedbackIndex
  90         RVCE_CS(0x00000000); // videoBitstreamRingIndex
  91         RVCE_END();
  92 }
  93
  94 static void feedback(struct rvce_encoder *enc)
  95 {
  96         RVCE_BEGIN(0x05000005); // feedback buffer
  97         RVCE_WRITE(enc->fb->cs_handle, enc->fb->domain); // feedbackRingAddressHi
  98         RVCE_CS(0x00000000); // feedbackRingAddressLo
  99         RVCE_CS(0x00000001); // feedbackRingSize
 100         RVCE_END();
 101 }
 102
 103 static void create(struct rvce_encoder *enc)
 104 {
 105         task_info(enc, 0x00000000);
 106
 107         RVCE_BEGIN(0x01000001); // create cmd
 108         RVCE_CS(0x00000000); // encUseCircularBuffer
 109         RVCE_CS(profiles[enc->base.profile -
 110                 PIPE_VIDEO_PROFILE_MPEG4_AVC_BASELINE]); // encProfile
 111         RVCE_CS(enc->base.level); // encLevel
 112         RVCE_CS(0x00000000); // encPicStructRestriction
 113         RVCE_CS(enc->base.width); // encImageWidth
 114         RVCE_CS(enc->base.height); // encImageHeight
 115         RVCE_CS(enc->luma->level[0].pitch_bytes); // encRefPicLumaPitch
 116         RVCE_CS(enc->chroma->level[0].pitch_bytes); // encRefPicChromaPitch
 117         RVCE_CS(align(enc->luma->npix_y, 16) / 8); // encRefYHeightInQw
 118         RVCE_CS(0x00000000); // encRefPic(Addr|Array)Mode, encPicStructRestriction, disableRDO
 119         RVCE_END();
 120 }
 121
 122 static void rate_control(struct rvce_encoder *enc)
 123 {
 124         RVCE_BEGIN(0x04000005); // rate control
 125         RVCE_CS(enc->pic.rate_ctrl.rate_ctrl_method); // encRateControlMethod
 126         RVCE_CS(enc->pic.rate_ctrl.target_bitrate); // encRateControlTargetBitRate
 127         RVCE_CS(enc->pic.rate_ctrl.peak_bitrate); // encRateControlPeakBitRate
 128         RVCE_CS(enc->pic.rate_ctrl.frame_rate_num); // encRateControlFrameRateNum
 129         RVCE_CS(0x00000000); // encGOPSize
 130         RVCE_CS(enc->pic.quant_i_frames); // encQP_I
 131         RVCE_CS(enc->pic.quant_p_frames); // encQP_P
 132         RVCE_CS(enc->pic.quant_b_frames); // encQP_B
 133         RVCE_CS(enc->pic.rate_ctrl.vbv_buffer_size); // encVBVBufferSize
 134         RVCE_CS(enc->pic.rate_ctrl.frame_rate_den); // encRateControlFrameRateDen
 135         RVCE_CS(0x00000000); // encVBVBufferLevel
 136         RVCE_CS(0x00000000); // encMaxAUSize
 137         RVCE_CS(0x00000000); // encQPInitialMode
 138         RVCE_CS(enc->pic.rate_ctrl.target_bits_picture); // encTargetBitsPerPicture
 139         RVCE_CS(enc->pic.rate_ctrl.peak_bits_picture_integer); // encPeakBitsPerPictureInteger
 140         RVCE_CS(enc->pic.rate_ctrl.peak_bits_picture_fraction); // encPeakBitsPerPictureFractional
 141         RVCE_CS(0x00000000); // encMinQP
 142         RVCE_CS(0x00000033); // encMaxQP
 143         RVCE_CS(0x00000000); // encSkipFrameEnable
 144         RVCE_CS(0x00000000); // encFillerDataEnable
 145         RVCE_CS(0x00000000); // encEnforceHRD
 146         RVCE_CS(0x00000000); // encBPicsDeltaQP
 147         RVCE_CS(0x00000000); // encReferenceBPicsDeltaQP
 148         RVCE_CS(0x00000000); // encRateControlReInitDisable
 149         RVCE_END();
 150 }
 151
 152 static void config_extension(struct rvce_encoder *enc)
 153 {
 154         RVCE_BEGIN(0x04000001); // config extension
 155         RVCE_CS(0x00000003); // encEnablePerfLogging
 156         RVCE_END();
 157 }
 158
 159 static void pic_control(struct rvce_encoder *enc)
 160 {
 161         unsigned encNumMBsPerSlice;
 162
 163         encNumMBsPerSlice = align(enc->base.width, 16) / 16;
 164         encNumMBsPerSlice *= align(enc->base.height, 16) / 16;
 165
 166         RVCE_BEGIN(0x04000002); // pic control
 167         RVCE_CS(0x00000000); // encUseConstrainedIntraPred
 168         RVCE_CS(0x00000000); // encCABACEnable
 169         RVCE_CS(0x00000000); // encCABACIDC
 170         RVCE_CS(0x00000000); // encLoopFilterDisable
 171         RVCE_CS(0x00000000); // encLFBetaOffset
 172         RVCE_CS(0x00000000); // encLFAlphaC0Offset
 173         RVCE_CS(0x00000000); // encCropLeftOffset
 174         RVCE_CS((align(enc->base.width, 16) - enc->base.width) >> 1); // encCropRightOffset
 175         RVCE_CS(0x00000000); // encCropTopOffset
 176         RVCE_CS((align(enc->base.height, 16) - enc->base.height) >> 1); // encCropBottomOffset
 177         RVCE_CS(encNumMBsPerSlice); // encNumMBsPerSlice
 178         RVCE_CS(0x00000000); // encIntraRefreshNumMBsPerSlot
 179         RVCE_CS(0x00000000); // encForceIntraRefresh
 180         RVCE_CS(0x00000000); // encForceIMBPeriod
 181         RVCE_CS(0x00000000); // encPicOrderCntType
 182         RVCE_CS(0x00000000); // log2_max_pic_order_cnt_lsb_minus4
 183         RVCE_CS(0x00000000); // encSPSID
 184         RVCE_CS(0x00000000); // encPPSID
 185         RVCE_CS(0x00000040); // encConstraintSetFlags
 186         RVCE_CS(MAX2(enc->base.max_references, 1) - 1); // encBPicPattern
 187         RVCE_CS(0x00000000); // weightPredModeBPicture
 188         RVCE_CS(MIN2(enc->base.max_references, 2)); // encNumberOfReferenceFrames
 189         RVCE_CS(enc->base.max_references + 1); // encMaxNumRefFrames
 190         RVCE_CS(0x00000001); // encNumDefaultActiveRefL0
 191         RVCE_CS(0x00000001); // encNumDefaultActiveRefL1
 192         RVCE_CS(0x00000000); // encSliceMode
 193         RVCE_CS(0x00000000); // encMaxSliceSize
 194         RVCE_END();
 195 }
 196
 197 static void motion_estimation(struct rvce_encoder *enc)
 198 {
 199         RVCE_BEGIN(0x04000007); // motion estimation
 200         RVCE_CS(0x00000001); // encIMEDecimationSearch
 201         RVCE_CS(0x00000001); // motionEstHalfPixel
 202         RVCE_CS(0x00000000); // motionEstQuarterPixel
 203         RVCE_CS(0x00000000); // disableFavorPMVPoint
 204         RVCE_CS(0x00000000); // forceZeroPointCenter
 205         RVCE_CS(0x00000000); // LSMVert
 206         RVCE_CS(0x00000010); // encSearchRangeX
 207         RVCE_CS(0x00000010); // encSearchRangeY
 208         RVCE_CS(0x00000010); // encSearch1RangeX
 209         RVCE_CS(0x00000010); // encSearch1RangeY
 210         RVCE_CS(0x00000000); // disable16x16Frame1
 211         RVCE_CS(0x00000000); // disableSATD
 212         RVCE_CS(0x00000000); // enableAMD
 213         RVCE_CS(0x000000fe); // encDisableSubMode
 214         RVCE_CS(0x00000000); // encIMESkipX
 215         RVCE_CS(0x00000000); // encIMESkipY
 216         RVCE_CS(0x00000000); // encEnImeOverwDisSubm
 217         RVCE_CS(0x00000000); // encImeOverwDisSubmNo
 218         RVCE_CS(0x00000001); // encIME2SearchRangeX
 219         RVCE_CS(0x00000001); // encIME2SearchRangeY
 220         RVCE_CS(0x00000000); // parallelModeSpeedupEnable
 221         RVCE_CS(0x00000000); // fme0_encDisableSubMode
 222         RVCE_CS(0x00000000); // fme1_encDisableSubMode
 223         RVCE_CS(0x00000000); // imeSWSpeedupEnable
 224         RVCE_END();
 225 }
 226
 227 static void rdo(struct rvce_encoder *enc)
 228 {
 229         RVCE_BEGIN(0x04000008); // rdo
 230         RVCE_CS(0x00000000); // encDisableTbePredIFrame
 231         RVCE_CS(0x00000000); // encDisableTbePredPFrame
 232         RVCE_CS(0x00000000); // useFmeInterpolY
 233         RVCE_CS(0x00000000); // useFmeInterpolUV
 234         RVCE_CS(0x00000000); // useFmeIntrapolY
 235         RVCE_CS(0x00000000); // useFmeIntrapolUV
 236         RVCE_CS(0x00000000); // useFmeInterpolY_1
 237         RVCE_CS(0x00000000); // useFmeInterpolUV_1
 238         RVCE_CS(0x00000000); // useFmeIntrapolY_1
 239         RVCE_CS(0x00000000); // useFmeIntrapolUV_1
 240         RVCE_CS(0x00000000); // enc16x16CostAdj
 241         RVCE_CS(0x00000000); // encSkipCostAdj
 242         RVCE_CS(0x00000000); // encForce16x16skip
 243         RVCE_CS(0x00000000); // encDisableThresholdCalcA
 244         RVCE_CS(0x00000000); // encLumaCoeffCost
 245         RVCE_CS(0x00000000); // encLumaMBCoeffCost
 246         RVCE_CS(0x00000000); // encChromaCoeffCost
 247         RVCE_END();
 248 }
 249
 250 static void encode(struct rvce_encoder *enc)
 251 {
 252         int i;
 253         unsigned luma_offset, chroma_offset;
 254
 255         task_info(enc, 0x00000003);
 256
 257         RVCE_BEGIN(0x05000001); // context buffer
 258         RVCE_READWRITE(enc->cpb.cs_handle, enc->cpb.domain); // encodeContextAddressHi
 259         RVCE_CS(0x00000000); // encodeContextAddressLo
 260         RVCE_END();
 261
 262         RVCE_BEGIN(0x05000004); // video bitstream buffer
 263         RVCE_WRITE(enc->bs_handle, RADEON_DOMAIN_GTT); // videoBitstreamRingAddressHi
 264         RVCE_CS(0x00000000); // videoBitstreamRingAddressLo
 265         RVCE_CS(enc->bs_size); // videoBitstreamRingSize
 266         RVCE_END();
 267
 268         RVCE_BEGIN(0x03000001); // encode
 269         RVCE_CS(0x00000000); // insertHeaders
 270         RVCE_CS(0x00000000); // pictureStructure
 271         RVCE_CS(enc->bs_size); // allowedMaxBitstreamSize
 272         RVCE_CS(0x00000000); // forceRefreshMap
 273         RVCE_CS(0x00000000); // insertAUD
 274         RVCE_CS(0x00000000); // endOfSequence
 275         RVCE_CS(0x00000000); // endOfStream
 276         RVCE_READ(enc->handle, RADEON_DOMAIN_VRAM); // inputPictureLumaAddressHi
 277         RVCE_CS(enc->luma->level[0].offset); // inputPictureLumaAddressLo
 278         RVCE_READ(enc->handle, RADEON_DOMAIN_VRAM); // inputPictureChromaAddressHi
 279         RVCE_CS(enc->chroma->level[0].offset); // inputPictureChromaAddressLo
 280         RVCE_CS(align(enc->luma->npix_y, 16)); // encInputFrameYPitch
 281         RVCE_CS(enc->luma->level[0].pitch_bytes); // encInputPicLumaPitch
 282         RVCE_CS(enc->chroma->level[0].pitch_bytes); // encInputPicChromaPitch
 283         RVCE_CS(0x00000000); // encInputPic(Addr|Array)Mode
 284         RVCE_CS(0x00000000); // encInputPicTileConfig
 285         RVCE_CS(enc->pic.picture_type); // encPicType
 286         RVCE_CS(enc->pic.picture_type == PIPE_H264_ENC_PICTURE_TYPE_IDR); // encIdrFlag
 287         RVCE_CS(0x00000000); // encIdrPicId
 288         RVCE_CS(0x00000000); // encMGSKeyPic
 289         RVCE_CS(!enc->pic.not_referenced); // encReferenceFlag
 290         RVCE_CS(0x00000000); // encTemporalLayerIndex
 291         RVCE_CS(0x00000000); // num_ref_idx_active_override_flag
 292         RVCE_CS(0x00000000); // num_ref_idx_l0_active_minus1
 293         RVCE_CS(0x00000000); // num_ref_idx_l1_active_minus1
 294
 295         i = enc->pic.frame_num - enc->pic.ref_idx_l0;
 296         if (i > 1 && enc->pic.picture_type == PIPE_H264_ENC_PICTURE_TYPE_P) {
 297                 RVCE_CS(0x00000001); // encRefListModificationOp
 298                 RVCE_CS(i - 1);      // encRefListModificationNum
 299         } else {
 300                 RVCE_CS(0x00000000); // encRefListModificationOp
 301                 RVCE_CS(0x00000000); // encRefListModificationNum
 302         }
 303
 304         for (i = 0; i < 3; ++i) {
 305                 RVCE_CS(0x00000000); // encRefListModificationOp
 306                 RVCE_CS(0x00000000); // encRefListModificationNum
 307         }
 308         for (i = 0; i < 4; ++i) {
 309                 RVCE_CS(0x00000000); // encDecodedPictureMarkingOp
 310                 RVCE_CS(0x00000000); // encDecodedPictureMarkingNum
 311                 RVCE_CS(0x00000000); // encDecodedPictureMarkingIdx
 312                 RVCE_CS(0x00000000); // encDecodedRefBasePictureMarkingOp
 313                 RVCE_CS(0x00000000); // encDecodedRefBasePictureMarkingNum
 314         }
 315
 316         // encReferencePictureL0[0]
 317         RVCE_CS(0x00000000); // pictureStructure
 318         if(enc->pic.picture_type == PIPE_H264_ENC_PICTURE_TYPE_P ||
 319            enc->pic.picture_type == PIPE_H264_ENC_PICTURE_TYPE_B) {
 320                 struct rvce_cpb_slot *l0 = l0_slot(enc);
 321                 frame_offset(enc, l0, &luma_offset, &chroma_offset);
 322                 RVCE_CS(l0->picture_type); // encPicType
 323                 RVCE_CS(l0->frame_num); // frameNumber
 324                 RVCE_CS(l0->pic_order_cnt); // pictureOrderCount
 325                 RVCE_CS(luma_offset); // lumaOffset
 326                 RVCE_CS(chroma_offset); // chromaOffset
 327         } else {
 328                 RVCE_CS(0x00000000); // encPicType
 329                 RVCE_CS(0x00000000); // frameNumber
 330                 RVCE_CS(0x00000000); // pictureOrderCount
 331                 RVCE_CS(0xffffffff); // lumaOffset
 332                 RVCE_CS(0xffffffff); // chromaOffset
 333         }
 334
 335         // encReferencePictureL0[1]
 336         RVCE_CS(0x00000000); // pictureStructure
 337         RVCE_CS(0x00000000); // encPicType
 338         RVCE_CS(0x00000000); // frameNumber
 339         RVCE_CS(0x00000000); // pictureOrderCount
 340         RVCE_CS(0xffffffff); // lumaOffset
 341         RVCE_CS(0xffffffff); // chromaOffset
 342
 343         // encReferencePictureL1[0]
 344         RVCE_CS(0x00000000); // pictureStructure
 345         if(enc->pic.picture_type == PIPE_H264_ENC_PICTURE_TYPE_B) {
 346                 struct rvce_cpb_slot *l1 = l1_slot(enc);
 347                 frame_offset(enc, l1, &luma_offset, &chroma_offset);
 348                 RVCE_CS(l1->picture_type); // encPicType
 349                 RVCE_CS(l1->frame_num); // frameNumber
 350                 RVCE_CS(l1->pic_order_cnt); // pictureOrderCount
 351                 RVCE_CS(luma_offset); // lumaOffset
 352                 RVCE_CS(chroma_offset); // chromaOffset
 353         } else {
 354                 RVCE_CS(0x00000000); // encPicType
 355                 RVCE_CS(0x00000000); // frameNumber
 356                 RVCE_CS(0x00000000); // pictureOrderCount
 357                 RVCE_CS(0xffffffff); // lumaOffset
 358                 RVCE_CS(0xffffffff); // chromaOffset
 359         }
 360
 361         frame_offset(enc, current_slot(enc), &luma_offset, &chroma_offset);
 362         RVCE_CS(luma_offset); // encReconstructedLumaOffset
 363         RVCE_CS(chroma_offset); // encReconstructedChromaOffset
 364         RVCE_CS(0x00000000); // encColocBufferOffset
 365         RVCE_CS(0x00000000); // encReconstructedRefBasePictureLumaOffset
 366         RVCE_CS(0x00000000); // encReconstructedRefBasePictureChromaOffset
 367         RVCE_CS(0x00000000); // encReferenceRefBasePictureLumaOffset
 368         RVCE_CS(0x00000000); // encReferenceRefBasePictureChromaOffset
 369         RVCE_CS(0x00000000); // pictureCount
 370         RVCE_CS(enc->pic.frame_num); // frameNumber
 371         RVCE_CS(enc->pic.pic_order_cnt); // pictureOrderCount
 372         RVCE_CS(0x00000000); // numIPicRemainInRCGOP
 373         RVCE_CS(0x00000000); // numPPicRemainInRCGOP
 374         RVCE_CS(0x00000000); // numBPicRemainInRCGOP
 375         RVCE_CS(0x00000000); // numIRPicRemainInRCGOP
 376         RVCE_CS(0x00000000); // enableIntraRefresh
 377         RVCE_END();
 378 }
 379
 380 static void destroy(struct rvce_encoder *enc)
 381 {
 382         task_info(enc, 0x00000001);
 383
 384         RVCE_BEGIN(0x02000001); // destroy
 385         RVCE_END();
 386 }
 387
 388 void radeon_vce_40_2_2_init(struct rvce_encoder *enc)
 389 {
 390         enc->session = session;
 391         enc->create = create;
 392         enc->feedback = feedback;
 393         enc->rate_control = rate_control;
 394         enc->config_extension = config_extension;
 395         enc->pic_control = pic_control;
 396         enc->motion_estimation = motion_estimation;
 397         enc->rdo = rdo;
 398         enc->encode = encode;
 399         enc->destroy = destroy;
 400 }