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 void session(struct rvce_encoder *enc)
  48 {
  49         RVCE_BEGIN(0x00000001); // session cmd
  50         RVCE_CS(enc->stream_handle);
  51         RVCE_END();
  52 }
  53
  54 static void task_info(struct rvce_encoder *enc, uint32_t taskOperation)
  55 {
  56         RVCE_BEGIN(0x00000002); // task info
  57         RVCE_CS(0xffffffff); // offsetOfNextTaskInfo
  58         RVCE_CS(taskOperation); // taskOperation
  59         RVCE_CS(0x00000000); // referencePictureDependency
  60         RVCE_CS(0x00000000); // collocateFlagDependency
  61         RVCE_CS(0x00000000); // feedbackIndex
  62         RVCE_CS(0x00000000); // videoBitstreamRingIndex
  63         RVCE_END();
  64 }
  65
  66 static void feedback(struct rvce_encoder *enc)
  67 {
  68         RVCE_BEGIN(0x05000005); // feedback buffer
  69         RVCE_WRITE(enc->fb->cs_handle, enc->fb->domain); // feedbackRingAddressHi
  70         RVCE_CS(0x00000000); // feedbackRingAddressLo
  71         RVCE_CS(0x00000001); // feedbackRingSize
  72         RVCE_END();
  73 }
  74
  75 static void create(struct rvce_encoder *enc)
  76 {
  77         task_info(enc, 0x00000000);
  78
  79         RVCE_BEGIN(0x01000001); // create cmd
  80         RVCE_CS(0x00000000); // encUseCircularBuffer
  81         RVCE_CS(0x00000041); // encProfile
  82         RVCE_CS(0x0000000a); // encLevel
  83         RVCE_CS(0x00000000); // encPicStructRestriction
  84         RVCE_CS(enc->base.width); // encImageWidth
  85         RVCE_CS(enc->base.height); // encImageHeight
  86         RVCE_CS(enc->luma->level[0].pitch_bytes); // encRefPicLumaPitch
  87         RVCE_CS(enc->chroma->level[0].pitch_bytes); // encRefPicChromaPitch
  88         RVCE_CS(align(enc->luma->npix_y, 16) / 8); // encRefYHeightInQw
  89         RVCE_CS(0x00000000); // encRefPic(Addr|Array)Mode, encPicStructRestriction, disableRDO
  90         RVCE_END();
  91 }
  92
  93 static void rate_control(struct rvce_encoder *enc)
  94 {
  95         RVCE_BEGIN(0x04000005); // rate control
  96         RVCE_CS(enc->pic.rate_ctrl.rate_ctrl_method); // encRateControlMethod
  97         RVCE_CS(enc->pic.rate_ctrl.target_bitrate); // encRateControlTargetBitRate
  98         RVCE_CS(enc->pic.rate_ctrl.peak_bitrate); // encRateControlPeakBitRate
  99         RVCE_CS(enc->pic.rate_ctrl.frame_rate_num); // encRateControlFrameRateNum
 100         RVCE_CS(0x00000000); // encGOPSize
 101         RVCE_CS(enc->pic.quant_i_frames); // encQP_I
 102         RVCE_CS(enc->pic.quant_p_frames); // encQP_P
 103         RVCE_CS(enc->pic.quant_b_frames); // encQP_B
 104         RVCE_CS(enc->pic.rate_ctrl.vbv_buffer_size); // encVBVBufferSize
 105         RVCE_CS(enc->pic.rate_ctrl.frame_rate_den); // encRateControlFrameRateDen
 106         RVCE_CS(0x00000000); // encVBVBufferLevel
 107         RVCE_CS(0x00000000); // encMaxAUSize
 108         RVCE_CS(0x00000000); // encQPInitialMode
 109         RVCE_CS(enc->pic.rate_ctrl.target_bits_picture); // encTargetBitsPerPicture
 110         RVCE_CS(enc->pic.rate_ctrl.peak_bits_picture_integer); // encPeakBitsPerPictureInteger
 111         RVCE_CS(enc->pic.rate_ctrl.peak_bits_picture_fraction); // encPeakBitsPerPictureFractional
 112         RVCE_CS(0x00000000); // encMinQP
 113         RVCE_CS(0x00000033); // encMaxQP
 114         RVCE_CS(0x00000000); // encSkipFrameEnable
 115         RVCE_CS(0x00000000); // encFillerDataEnable
 116         RVCE_CS(0x00000000); // encEnforceHRD
 117         RVCE_CS(0x00000000); // encBPicsDeltaQP
 118         RVCE_CS(0x00000000); // encReferenceBPicsDeltaQP
 119         RVCE_CS(0x00000000); // encRateControlReInitDisable
 120         RVCE_END();
 121 }
 122
 123 static void config_extension(struct rvce_encoder *enc)
 124 {
 125         RVCE_BEGIN(0x04000001); // config extension
 126         RVCE_CS(0x00000003); // encEnablePerfLogging
 127         RVCE_END();
 128 }
 129
 130 static void pic_control(struct rvce_encoder *enc)
 131 {
 132         unsigned encNumMBsPerSlice;
 133
 134         encNumMBsPerSlice = align(enc->base.width, 16) / 16;
 135         encNumMBsPerSlice *= align(enc->base.height, 16) / 16;
 136
 137         RVCE_BEGIN(0x04000002); // pic control
 138         RVCE_CS(0x00000000); // encUseConstrainedIntraPred
 139         RVCE_CS(0x00000000); // encCABACEnable
 140         RVCE_CS(0x00000000); // encCABACIDC
 141         RVCE_CS(0x00000000); // encLoopFilterDisable
 142         RVCE_CS(0x00000000); // encLFBetaOffset
 143         RVCE_CS(0x00000000); // encLFAlphaC0Offset
 144         RVCE_CS(0x00000000); // encCropLeftOffset
 145         RVCE_CS((align(enc->base.width, 16) - enc->base.width) >> 1); // encCropRightOffset
 146         RVCE_CS(0x00000000); // encCropTopOffset
 147         RVCE_CS((align(enc->base.height, 16) - enc->base.height) >> 1); // encCropBottomOffset
 148         RVCE_CS(encNumMBsPerSlice); // encNumMBsPerSlice
 149         RVCE_CS(0x00000000); // encIntraRefreshNumMBsPerSlot
 150         RVCE_CS(0x00000000); // encForceIntraRefresh
 151         RVCE_CS(0x00000000); // encForceIMBPeriod
 152         RVCE_CS(0x00000000); // encPicOrderCntType
 153         RVCE_CS(0x00000000); // log2_max_pic_order_cnt_lsb_minus4
 154         RVCE_CS(0x00000000); // encSPSID
 155         RVCE_CS(0x00000000); // encPPSID
 156         RVCE_CS(0x00000040); // encConstraintSetFlags
 157         RVCE_CS(0x00000000); // encBPicPattern
 158         RVCE_CS(0x00000000); // weightPredModeBPicture
 159         RVCE_CS(0x00000001); // encNumberOfReferenceFrames
 160         RVCE_CS(0x00000001); // encMaxNumRefFrames
 161         RVCE_CS(0x00000000); // encNumDefaultActiveRefL0
 162         RVCE_CS(0x00000000); // encNumDefaultActiveRefL1
 163         RVCE_CS(0x00000000); // encSliceMode
 164         RVCE_CS(0x00000000); // encMaxSliceSize
 165         RVCE_END();
 166 }
 167
 168 static void motion_estimation(struct rvce_encoder *enc)
 169 {
 170         RVCE_BEGIN(0x04000007); // motion estimation
 171         RVCE_CS(0x00000001); // encIMEDecimationSearch
 172         RVCE_CS(0x00000001); // motionEstHalfPixel
 173         RVCE_CS(0x00000000); // motionEstQuarterPixel
 174         RVCE_CS(0x00000000); // disableFavorPMVPoint
 175         RVCE_CS(0x00000000); // forceZeroPointCenter
 176         RVCE_CS(0x00000000); // LSMVert
 177         RVCE_CS(0x00000010); // encSearchRangeX
 178         RVCE_CS(0x00000010); // encSearchRangeY
 179         RVCE_CS(0x00000010); // encSearch1RangeX
 180         RVCE_CS(0x00000010); // encSearch1RangeY
 181         RVCE_CS(0x00000000); // disable16x16Frame1
 182         RVCE_CS(0x00000000); // disableSATD
 183         RVCE_CS(0x00000000); // enableAMD
 184         RVCE_CS(0x000000fe); // encDisableSubMode
 185         RVCE_CS(0x00000000); // encIMESkipX
 186         RVCE_CS(0x00000000); // encIMESkipY
 187         RVCE_CS(0x00000000); // encEnImeOverwDisSubm
 188         RVCE_CS(0x00000000); // encImeOverwDisSubmNo
 189         RVCE_CS(0x00000001); // encIME2SearchRangeX
 190         RVCE_CS(0x00000001); // encIME2SearchRangeY
 191         RVCE_CS(0x00000000); // parallelModeSpeedupEnable
 192         RVCE_CS(0x00000000); // fme0_encDisableSubMode
 193         RVCE_CS(0x00000000); // fme1_encDisableSubMode
 194         RVCE_CS(0x00000000); // imeSWSpeedupEnable
 195         RVCE_END();
 196 }
 197
 198 static void rdo(struct rvce_encoder *enc)
 199 {
 200         RVCE_BEGIN(0x04000008); // rdo
 201         RVCE_CS(0x00000000); // encDisableTbePredIFrame
 202         RVCE_CS(0x00000000); // encDisableTbePredPFrame
 203         RVCE_CS(0x00000000); // useFmeInterpolY
 204         RVCE_CS(0x00000000); // useFmeInterpolUV
 205         RVCE_CS(0x00000000); // useFmeIntrapolY
 206         RVCE_CS(0x00000000); // useFmeIntrapolUV
 207         RVCE_CS(0x00000000); // useFmeInterpolY_1
 208         RVCE_CS(0x00000000); // useFmeInterpolUV_1
 209         RVCE_CS(0x00000000); // useFmeIntrapolY_1
 210         RVCE_CS(0x00000000); // useFmeIntrapolUV_1
 211         RVCE_CS(0x00000000); // enc16x16CostAdj
 212         RVCE_CS(0x00000000); // encSkipCostAdj
 213         RVCE_CS(0x00000000); // encForce16x16skip
 214         RVCE_CS(0x00000000); // encDisableThresholdCalcA
 215         RVCE_CS(0x00000000); // encLumaCoeffCost
 216         RVCE_CS(0x00000000); // encLumaMBCoeffCost
 217         RVCE_CS(0x00000000); // encChromaCoeffCost
 218         RVCE_END();
 219 }
 220
 221 static void frame_offset(struct rvce_encoder *enc, unsigned frame_num,
 222                          unsigned *luma_offset, unsigned *chroma_offset)
 223 {
 224         unsigned pitch = align(enc->luma->level[0].pitch_bytes, 128);
 225         unsigned vpitch = align(enc->luma->npix_y, 16);
 226         unsigned fsize = pitch * (vpitch + vpitch / 2);
 227
 228         *luma_offset = (frame_num % RVCE_NUM_CPB_FRAMES) * fsize;
 229         *chroma_offset = *luma_offset + pitch * vpitch;
 230 }
 231
 232 static void encode(struct rvce_encoder *enc)
 233 {
 234         int i;
 235         unsigned luma_offset, chroma_offset;
 236
 237         task_info(enc, 0x00000003);
 238
 239         RVCE_BEGIN(0x05000001); // context buffer
 240         RVCE_READWRITE(enc->cpb.cs_handle, enc->cpb.domain); // encodeContextAddressHi
 241         RVCE_CS(0x00000000); // encodeContextAddressLo
 242         RVCE_END();
 243
 244         RVCE_BEGIN(0x05000004); // video bitstream buffer
 245         RVCE_WRITE(enc->bs_handle, RADEON_DOMAIN_GTT); // videoBitstreamRingAddressHi
 246         RVCE_CS(0x00000000); // videoBitstreamRingAddressLo
 247         RVCE_CS(enc->bs_size); // videoBitstreamRingSize
 248         RVCE_END();
 249
 250         RVCE_BEGIN(0x03000001); // encode
 251         RVCE_CS(0x00000000); // insertHeaders
 252         RVCE_CS(0x00000000); // pictureStructure
 253         RVCE_CS(enc->bs_size); // allowedMaxBitstreamSize
 254         RVCE_CS(0x00000000); // forceRefreshMap
 255         RVCE_CS(0x00000000); // insertAUD
 256         RVCE_CS(0x00000000); // endOfSequence
 257         RVCE_CS(0x00000000); // endOfStream
 258         RVCE_READ(enc->handle, RADEON_DOMAIN_VRAM); // inputPictureLumaAddressHi
 259         RVCE_CS(enc->luma->level[0].offset); // inputPictureLumaAddressLo
 260         RVCE_READ(enc->handle, RADEON_DOMAIN_VRAM); // inputPictureChromaAddressHi
 261         RVCE_CS(enc->chroma->level[0].offset); // inputPictureChromaAddressLo
 262         RVCE_CS(align(enc->luma->npix_y, 16)); // encInputFrameYPitch
 263         RVCE_CS(enc->luma->level[0].pitch_bytes); // encInputPicLumaPitch
 264         RVCE_CS(enc->chroma->level[0].pitch_bytes); // encInputPicChromaPitch
 265         RVCE_CS(0x00000000); // encInputPic(Addr|Array)Mode
 266         RVCE_CS(0x00000000); // encInputPicTileConfig
 267         RVCE_CS(enc->pic.picture_type); // encPicType
 268         RVCE_CS(enc->pic.picture_type == 3); // encIdrFlag
 269         RVCE_CS(0x00000000); // encIdrPicId
 270         RVCE_CS(0x00000000); // encMGSKeyPic
 271         RVCE_CS(0x00000001); // encReferenceFlag
 272         RVCE_CS(0x00000000); // encTemporalLayerIndex
 273         RVCE_CS(0x00000000); // num_ref_idx_active_override_flag
 274         RVCE_CS(0x00000000); // num_ref_idx_l0_active_minus1
 275         RVCE_CS(0x00000000); // num_ref_idx_l1_active_minus1
 276         for (i = 0; i < 4; ++i) {
 277                 RVCE_CS(0x00000000); // encRefListModificationOp
 278                 RVCE_CS(0x00000000); // encRefListModificationNum
 279         }
 280         for (i = 0; i < 4; ++i) {
 281                 RVCE_CS(0x00000000); // encDecodedPictureMarkingOp
 282                 RVCE_CS(0x00000000); // encDecodedPictureMarkingNum
 283                 RVCE_CS(0x00000000); // encDecodedPictureMarkingIdx
 284         }
 285         for (i = 0; i < 4; ++i) {
 286                 RVCE_CS(0x00000000); // encDecodedRefBasePictureMarkingOp
 287                 RVCE_CS(0x00000000); // encDecodedRefBasePictureMarkingNum
 288         }
 289
 290         RVCE_CS(0x00000000); // pictureStructure
 291
 292         if (enc->pic.picture_type == PIPE_H264_ENC_PICTURE_TYPE_IDR) {
 293                 RVCE_CS(0x00000000); // encPicType
 294                 RVCE_CS(0x00000000); // frameNumber
 295                 RVCE_CS(0x00000000); // pictureOrderCount
 296                 RVCE_CS(0xffffffff); // lumaOffset
 297                 RVCE_CS(0xffffffff); // chromaOffset
 298         }
 299         else if(enc->pic.picture_type == PIPE_H264_ENC_PICTURE_TYPE_P) {
 300                 frame_offset(enc, enc->pic.frame_num - 1, &luma_offset, &chroma_offset);
 301                 RVCE_CS(0x00000000); // encPicType
 302                 RVCE_CS(enc->pic.frame_num - 1); // frameNumber
 303                 RVCE_CS(enc->pic.frame_num - 1); // pictureOrderCount
 304                 RVCE_CS(luma_offset); // lumaOffset
 305                 RVCE_CS(chroma_offset); // chromaOffset
 306         }
 307         for (i = 0; i < 2; ++i) {
 308                 RVCE_CS(0x00000000); // pictureStructure
 309                 RVCE_CS(0x00000000); // encPicType
 310                 RVCE_CS(0x00000000); // frameNumber
 311                 RVCE_CS(0x00000000); // pictureOrderCount
 312                 RVCE_CS(0xffffffff); // lumaOffset
 313                 RVCE_CS(0xffffffff); // chromaOffset
 314         }
 315
 316         frame_offset(enc, enc->pic.frame_num, &luma_offset, &chroma_offset);
 317         RVCE_CS(luma_offset); // encReconstructedLumaOffset
 318         RVCE_CS(chroma_offset); // encReconstructedChromaOffset
 319         RVCE_CS(0x00000000); // encColocBufferOffset
 320         RVCE_CS(0x00000000); // encReconstructedRefBasePictureLumaOffset
 321         RVCE_CS(0x00000000); // encReconstructedRefBasePictureChromaOffset
 322         RVCE_CS(0x00000000); // encReferenceRefBasePictureLumaOffset
 323         RVCE_CS(0x00000000); // encReferenceRefBasePictureChromaOffset
 324         RVCE_CS(0x00000000); // pictureCount
 325         RVCE_CS(0x00000000); // frameNumber
 326         RVCE_CS(0x00000000); // pictureOrderCount
 327         RVCE_CS(0x00000000); // numIPicRemainInRCGOP
 328         RVCE_CS(0x00000000); // numPPicRemainInRCGOP
 329         RVCE_CS(0x00000000); // numBPicRemainInRCGOP
 330         RVCE_CS(0x00000000); // numIRPicRemainInRCGOP
 331         RVCE_CS(0x00000000); // enableIntraRefresh
 332         RVCE_END();
 333 }
 334
 335 static void destroy(struct rvce_encoder *enc)
 336 {
 337         task_info(enc, 0x00000001);
 338
 339         RVCE_BEGIN(0x02000001); // destroy
 340         RVCE_END();
 341 }
 342
 343 void radeon_vce_40_2_2_init(struct rvce_encoder *enc)
 344 {
 345         enc->session = session;
 346         enc->create = create;
 347         enc->feedback = feedback;
 348         enc->rate_control = rate_control;
 349         enc->config_extension = config_extension;
 350         enc->pic_control = pic_control;
 351         enc->motion_estimation = motion_estimation;
 352         enc->rdo = rdo;
 353         enc->encode = encode;
 354         enc->destroy = destroy;
 355 }