src/gallium/drivers/nouveau/nouveau_vp3_video.h

   1 /*
   2  * Copyright 2011-2013 Maarten Lankhorst
   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 AUTHORS OR COPYRIGHT HOLDERS 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 #include <libdrm/nouveau.h>
  24
  25 #include "pipe/p_defines.h"
  26 #include "vl/vl_video_buffer.h"
  27 #include "util/u_video.h"
  28
  29 struct nouveau_vp3_video_buffer {
  30    struct pipe_video_buffer base;
  31    unsigned num_planes, valid_ref;
  32    struct pipe_resource *resources[VL_NUM_COMPONENTS];
  33    struct pipe_sampler_view *sampler_view_planes[VL_NUM_COMPONENTS];
  34    struct pipe_sampler_view *sampler_view_components[VL_NUM_COMPONENTS];
  35    struct pipe_surface *surfaces[VL_NUM_COMPONENTS * 2];
  36 };
  37
  38 #define SLICE_SIZE 0x200
  39 #define VP_OFFSET 0x200
  40 #define COMM_OFFSET 0x500
  41
  42 #define NOUVEAU_VP3_DEBUG_FENCE 0
  43
  44 #if NOUVEAU_VP3_DEBUG_FENCE
  45 # define NOUVEAU_VP3_VIDEO_QDEPTH 1
  46 #else
  47 # define NOUVEAU_VP3_VIDEO_QDEPTH 2
  48 #endif
  49
  50 #define SUBC_BSP(m) dec->bsp_idx, (m)
  51 #define SUBC_VP(m) dec->vp_idx, (m)
  52 #define SUBC_PPP(m) dec->ppp_idx, (m)
  53
  54 union pipe_desc {
  55    struct pipe_picture_desc *base;
  56    struct pipe_mpeg12_picture_desc *mpeg12;
  57    struct pipe_mpeg4_picture_desc *mpeg4;
  58    struct pipe_vc1_picture_desc *vc1;
  59    struct pipe_h264_picture_desc *h264;
  60 };
  61
  62 struct nouveau_vp3_decoder {
  63    struct pipe_video_codec base;
  64    struct nouveau_client *client;
  65    struct nouveau_object *channel[3], *bsp, *vp, *ppp;
  66    struct nouveau_pushbuf *pushbuf[3];
  67
  68 #if NOUVEAU_VP3_DEBUG_FENCE
  69    /* dump fence and comm, as needed.. */
  70    unsigned *fence_map;
  71    struct comm *comm;
  72
  73    struct nouveau_bo *fence_bo;
  74 #endif
  75
  76    struct nouveau_bo *fw_bo, *bitplane_bo;
  77
  78    // array size max_references + 2, contains unpostprocessed images
  79    // added at the end of ref_bo is a tmp array
  80    // tmp is an array for h264, with each member being used for a ref frame or current
  81    // target.. size = (((mb(w)*((mb(h)+1)&~1))+3)>>2)<<8 * (max_references+1)
  82    // for other codecs, it simply seems that size = w*h is enough
  83    // unsure what it's supposed to contain..
  84    struct nouveau_bo *ref_bo;
  85
  86    struct nouveau_bo *inter_bo[2];
  87
  88    struct nouveau_bo *bsp_bo[NOUVEAU_VP3_VIDEO_QDEPTH];
  89
  90    // bo's used by each cycle:
  91
  92    // bsp_bo: contains raw bitstream data and parameters for BSP and VP.
  93    // inter_bo: contains data shared between BSP and VP
  94    // ref_bo: reference image data, used by PPP and VP
  95    // bitplane_bo: contain bitplane data (similar to ref_bo), used by BSP only
  96    // fw_bo: used by VP only.
  97
  98    // Needed amount of copies in optimal case:
  99    // 2 copies of inter_bo, VP would process the last inter_bo, while BSP is
 100    // writing out a new set.
 101    // NOUVEAU_VP3_VIDEO_QDEPTH copies of bsp_bo. We don't want to block the
 102    // pipeline ever, and give shaders a chance to run as well.
 103
 104    struct {
 105       struct nouveau_vp3_video_buffer *vidbuf;
 106       unsigned last_used;
 107       unsigned field_pic_flag : 1;
 108       unsigned decoded_top : 1;
 109       unsigned decoded_bottom : 1;
 110    } refs[17];
 111    unsigned fence_seq, fw_sizes, last_frame_num, tmp_stride, ref_stride;
 112
 113    unsigned bsp_idx, vp_idx, ppp_idx;
 114 };
 115
 116 struct comm {
 117         uint32_t bsp_cur_index; // 000
 118         uint32_t byte_ofs; // 004
 119         uint32_t status[0x10]; // 008
 120         uint32_t pos[0x10]; // 048
 121         uint8_t pad[0x100 - 0x88]; // 0a0 bool comm_encrypted
 122
 123         uint32_t pvp_cur_index; // 100
 124         uint32_t acked_byte_ofs; // 104
 125         uint32_t status_vp[0x10]; // 108
 126         uint16_t mb_y[0x10]; //148
 127         uint32_t pvp_stage; // 168 0xeeXX
 128         uint16_t parse_endpos_index; // 16c
 129         uint16_t irq_index; // 16e
 130         uint8_t  irq_470[0x10]; // 170
 131         uint32_t irq_pos[0x10]; // 180
 132         uint32_t parse_endpos[0x10]; // 1c0
 133 };
 134
 135 static INLINE uint32_t nouveau_vp3_video_align(uint32_t h)
 136 {
 137    return ((h+0x3f)&~0x3f);
 138 };
 139
 140 static INLINE uint32_t mb(uint32_t coord)
 141 {
 142    return (coord + 0xf)>>4;
 143 }
 144
 145 static INLINE uint32_t mb_half(uint32_t coord)
 146 {
 147    return (coord + 0x1f)>>5;
 148 }
 149
 150 static INLINE uint64_t
 151 nouveau_vp3_video_addr(struct nouveau_vp3_decoder *dec, struct nouveau_vp3_video_buffer *target)
 152 {
 153    uint64_t ret;
 154    if (target)
 155       ret = dec->ref_stride * target->valid_ref;
 156    else
 157       ret = dec->ref_stride * (dec->base.max_references+1);
 158    return dec->ref_bo->offset + ret;
 159 }
 160
 161 static INLINE void
 162 nouveau_vp3_ycbcr_offsets(struct nouveau_vp3_decoder *dec, uint32_t *y2,
 163                           uint32_t *cbcr, uint32_t *cbcr2)
 164 {
 165    uint32_t w = mb(dec->base.width), size;
 166    *y2 = mb_half(dec->base.height)*w;
 167    *cbcr = *y2 * 2;
 168    *cbcr2 = *cbcr + w * (nouveau_vp3_video_align(dec->base.height)>>6);
 169
 170    /* The check here should never fail because it means a bug
 171     * in the code rather than a bug in hardware..
 172     */
 173    size = (2 * (*cbcr2 - *cbcr) + *cbcr) << 8;
 174    if (size > dec->ref_stride) {
 175       debug_printf("Overshot ref_stride (%u) with size %u and ofs (%u,%u,%u)\n",
 176                    dec->ref_stride, size, *y2<<8, *cbcr<<8, *cbcr2<<8);
 177       *y2 = *cbcr = *cbcr2 = 0;
 178       assert(size <= dec->ref_stride);
 179    }
 180 }
 181
 182 static INLINE void
 183 nouveau_vp3_inter_sizes(struct nouveau_vp3_decoder *dec, uint32_t slice_count,
 184                         uint32_t *slice_size, uint32_t *bucket_size,
 185                         uint32_t *ring_size)
 186 {
 187    *slice_size = (SLICE_SIZE * slice_count)>>8;
 188    if (u_reduce_video_profile(dec->base.profile) == PIPE_VIDEO_FORMAT_MPEG12)
 189       *bucket_size = 0;
 190    else
 191       *bucket_size = mb(dec->base.width) * 3;
 192    *ring_size = (dec->inter_bo[0]->size >> 8) - *bucket_size - *slice_size;
 193 }
 194
 195 struct pipe_video_buffer *
 196 nouveau_vp3_video_buffer_create(struct pipe_context *pipe,
 197                                 const struct pipe_video_buffer *templat,
 198                                 int flags);
 199
 200 void
 201 nouveau_vp3_decoder_init_common(struct pipe_video_codec *decoder);
 202
 203 int
 204 nouveau_vp3_load_firmware(struct nouveau_vp3_decoder *dec,
 205                           enum pipe_video_profile profile,
 206                           unsigned chipset);
 207
 208 uint32_t
 209 nouveau_vp3_bsp(struct nouveau_vp3_decoder *dec,  union pipe_desc desc,
 210                 struct nouveau_vp3_video_buffer *target,
 211                 unsigned comm_seq, unsigned num_buffers,
 212                 const void *const *data, const unsigned *num_bytes);
 213
 214 void
 215 nouveau_vp3_vp_caps(struct nouveau_vp3_decoder *dec, union pipe_desc desc,
 216                     struct nouveau_vp3_video_buffer *target, unsigned comm_seq,
 217                     unsigned *caps, unsigned *is_ref,
 218                     struct nouveau_vp3_video_buffer *refs[16]);
 219
 220 int
 221 nouveau_vp3_screen_get_video_param(struct pipe_screen *pscreen,
 222                                    enum pipe_video_profile profile,
 223                                    enum pipe_video_entrypoint entrypoint,
 224                                    enum pipe_video_cap param);
 225
 226 boolean
 227 nouveau_vp3_screen_video_supported(struct pipe_screen *screen,
 228                                    enum pipe_format format,
 229                                    enum pipe_video_profile profile,
 230                                    enum pipe_video_entrypoint entrypoint);