Attachment 117910 Details for Bug 89969 – WIP - nouveau: add support for chunk decoding

[patch] WIP - nouveau: add support for chunk decoding

0001-nouveau-add-support-for-chunk-decoding.patch (text/plain), 20.85 KB, created by Julien Isorce on 2015-08-25 16:05:14 UTC

(hide)

Description:

Filename:

MIME Type:

Creator: Julien Isorce

Created: 2015-08-25 16:05:14 UTC

Size: 20.85 KB

patch

obsolete

>From 20ca2cf458a7e620f314f11ee563e58a736442e8 Mon Sep 17 00:00:00 2001
>From: Julien Isorce <j.isorce@samsung.com>
>Date: Fri, 17 Jul 2015 11:00:07 +0100
>Subject: [PATCH] nouveau: add support for chunk decoding
>
>---
> src/gallium/drivers/nouveau/nouveau_vp3_video.h    |  15 ++
> .../drivers/nouveau/nouveau_vp3_video_bsp.c        | 188 ++++++++++++-------
> src/gallium/drivers/nouveau/nvc0/nvc0_video.c      |  65 +++++--
> src/gallium/drivers/nouveau/nvc0/nvc0_video.h      |  13 ++
> src/gallium/drivers/nouveau/nvc0/nvc0_video_bsp.c  | 206 +++++++++++++++++++++
> 5 files changed, 404 insertions(+), 83 deletions(-)
>
>diff --git a/src/gallium/drivers/nouveau/nouveau_vp3_video.h b/src/gallium/drivers/nouveau/nouveau_vp3_video.h
>index 33e3bef..7800cef 100644
>--- a/src/gallium/drivers/nouveau/nouveau_vp3_video.h
>+++ b/src/gallium/drivers/nouveau/nouveau_vp3_video.h
>@@ -89,6 +89,10 @@ struct nouveau_vp3_decoder {
> 
>    struct nouveau_bo *bsp_bo[NOUVEAU_VP3_VIDEO_QDEPTH];
> 
>+   char *bsp_ptr;
>+   unsigned bsp_size;
>+   unsigned int nb_slices;
>+
>    // bo's used by each cycle:
> 
>    // bsp_bo: contains raw bitstream data and parameters for BSP and VP.
>@@ -208,6 +212,17 @@ nouveau_vp3_load_firmware(struct nouveau_vp3_decoder *dec,
>                           enum pipe_video_profile profile,
>                           unsigned chipset);
> 
>+
>+void
>+nouveau_vp3_bsp_begin(struct nouveau_vp3_decoder *dec);
>+
>+void
>+nouveau_vp3_bsp_next(struct nouveau_vp3_decoder *dec, unsigned num_buffers,
>+                     const void *const *data, const unsigned *num_bytes);
>+
>+uint32_t
>+nouveau_vp3_bsp_end(struct nouveau_vp3_decoder *dec, union pipe_desc desc);
>+
> uint32_t
> nouveau_vp3_bsp(struct nouveau_vp3_decoder *dec,  union pipe_desc desc,
>                 struct nouveau_vp3_video_buffer *target,
>diff --git a/src/gallium/drivers/nouveau/nouveau_vp3_video_bsp.c b/src/gallium/drivers/nouveau/nouveau_vp3_video_bsp.c
>index 6d968c1..d5e2734 100644
>--- a/src/gallium/drivers/nouveau/nouveau_vp3_video_bsp.c
>+++ b/src/gallium/drivers/nouveau/nouveau_vp3_video_bsp.c
>@@ -22,6 +22,7 @@
> 
> #include "nouveau_vp3_video.h"
> 
>+
> struct strparm_bsp {
> 	uint32_t w0[4]; // bits 0-23 length, bits 24-31 addr_hi
> 	uint32_t w1[4]; // bit 8-24 addr_lo
>@@ -126,7 +127,10 @@ nouveau_vp3_fill_picparm_mpeg12_bsp(struct nouveau_vp3_decoder *dec,
>    pic_bsp->intra_vlc_format = desc->intra_vlc_format;
>    pic_bsp->pad = 0;
>    for (i = 0; i < 4; ++i)
>-      pic_bsp->f_code[i/2][i%2] = desc->f_code[i/2][i%2] + 1; // FU
>+     pic_bsp->f_code[i/2][i%2] = desc->f_code[i/2][i%2] + 1; // FU
>+
>+   if (desc->num_slices == 0)
>+     desc->num_slices = dec->nb_slices;
> 
>    return (desc->num_slices << 4) | (dec->base.profile != PIPE_VIDEO_PROFILE_MPEG1);
> }
>@@ -230,81 +234,127 @@ nouveau_vp3_fill_picparm_h264_bsp(struct nouveau_vp3_decoder *dec,
>    return caps | 3;
> }
> 
>-uint32_t
>-nouveau_vp3_bsp(struct nouveau_vp3_decoder *dec,  union pipe_desc desc,
>-                struct nouveau_vp3_video_buffer *target,
>-                unsigned comm_seq, unsigned num_buffers,
>-                const void *const *data, const unsigned *num_bytes)
>+void
>+nouveau_vp3_bsp_begin(struct nouveau_vp3_decoder *dec)
> {
>-   enum pipe_video_format codec = u_reduce_video_profile(dec->base.profile);
>+    uint32_t comm_seq = dec->fence_seq;
>+    struct nouveau_bo *bsp_bo = dec->bsp_bo[comm_seq % NOUVEAU_VP3_VIDEO_QDEPTH];
>+    struct strparm_bsp *str_bsp = NULL;
>+
>+    unsigned ret = 0;
>+
>+    dec->bsp_ptr = NULL;
>+
>+    ret = nouveau_bo_map(bsp_bo, NOUVEAU_BO_WR, dec->client);
>+    if (ret) {
>+       debug_printf("map failed: %i %s\n", ret, strerror(-ret));
>+       assert(ret == 0);
>+    }
>+
>+    dec->bsp_ptr = bsp_bo->map;
>+    dec->bsp_size = NOUVEAU_VP3_BSP_RESERVED_SIZE;
>+    dec->nb_slices = 0;
>+
>+    dec->bsp_ptr += 0x100;
>+
>+    str_bsp = (struct strparm_bsp *)dec->bsp_ptr;
>+    memset(str_bsp, 0, 0x80);
>+    str_bsp->w0[0] = 16;
>+    str_bsp->w1[0] = 0x1;
>+    dec->bsp_ptr += 0x100;
>+    /* Reserved for picparm_vp */
>+    dec->bsp_ptr += 0x300;
>+    /* Reserved for comm */
>+ #if !NOUVEAU_VP3_DEBUG_FENCE
>+    memset(dec->bsp_ptr, 0, 0x200);
>+ #endif
>+    dec->bsp_ptr += 0x200;
>+}
>+
>+void
>+nouveau_vp3_bsp_next(struct nouveau_vp3_decoder *dec, unsigned num_buffers,
>+                     const void *const *data, const unsigned *num_bytes)
>+{
>+   uint32_t comm_seq = dec->fence_seq;
>    struct nouveau_bo *bsp_bo = dec->bsp_bo[comm_seq % NOUVEAU_VP3_VIDEO_QDEPTH];
>-   char *bsp;
>-   uint32_t endmarker, caps;
>-   struct strparm_bsp *str_bsp;
>-   int i;
>+   char *bsp_origin = bsp_bo->map;
>+   struct strparm_bsp *str_bsp = NULL;
>+   int i = 0;
> 
>-   bsp = bsp_bo->map;
>-   /*
>-    * 0x000..0x100: picparm_bsp
>-    * 0x200..0x500: picparm_vp
>-    * 0x500..0x700: comm
>-    * 0x700..onward: raw bitstream
>-    */
>-
>-   switch (codec){
>-   case PIPE_VIDEO_FORMAT_MPEG12:
>-      endmarker = 0xb7010000;
>-      caps = nouveau_vp3_fill_picparm_mpeg12_bsp(dec, desc.mpeg12, bsp);
>-      break;
>-   case PIPE_VIDEO_FORMAT_MPEG4:
>-      endmarker = 0xb1010000;
>-      caps = nouveau_vp3_fill_picparm_mpeg4_bsp(dec, desc.mpeg4, bsp);
>-      break;
>-   case PIPE_VIDEO_FORMAT_VC1: {
>-      endmarker = 0x0a010000;
>-      caps = nouveau_vp3_fill_picparm_vc1_bsp(dec, desc.vc1, bsp);
>-      break;
>-   }
>-   case PIPE_VIDEO_FORMAT_MPEG4_AVC: {
>-      endmarker = 0x0b010000;
>-      caps = nouveau_vp3_fill_picparm_h264_bsp(dec, desc.h264, bsp);
>-      break;
>-   }
>-   default: assert(0); return -1;
>-   }
>+   ++dec->nb_slices;
>+
>+   bsp_origin += 0x100;
>+   str_bsp = (struct strparm_bsp *)bsp_origin;
> 
>-   caps |= 0 << 16; // reset struct comm if flag is set
>-   caps |= 1 << 17; // enable watchdog
>-   caps |= 0 << 18; // do not report error to VP, so it can continue decoding what we have
>-   caps |= 0 << 19; // if enabled, use crypto crap?
>-   bsp += 0x100;
>-
>-   str_bsp = (struct strparm_bsp *)bsp;
>-   memset(str_bsp, 0, 0x80);
>-   str_bsp->w0[0] = 16;
>-   str_bsp->w1[0] = 0x1;
>-   bsp += 0x100;
>-   /* Reserved for picparm_vp */
>-   bsp += 0x300;
>-   /* Reserved for comm */
>-#if !NOUVEAU_VP3_DEBUG_FENCE
>-   memset(bsp, 0, 0x200);
>-#endif
>-   bsp += 0x200;
>    for (i = 0; i < num_buffers; ++i) {
>-      memcpy(bsp, data[i], num_bytes[i]);
>-      bsp += num_bytes[i];
>+      memcpy(dec->bsp_ptr, data[i], num_bytes[i]);
>+      dec->bsp_ptr += num_bytes[i];
>       str_bsp->w0[0] += num_bytes[i];
>    }
>+}
>+
>+uint32_t
>+nouveau_vp3_bsp_end(struct nouveau_vp3_decoder *dec, union pipe_desc desc)
>+{
>+    uint32_t comm_seq = dec->fence_seq;
>+    struct nouveau_bo *bsp_bo = dec->bsp_bo[comm_seq % NOUVEAU_VP3_VIDEO_QDEPTH];
>+    enum pipe_video_format codec = u_reduce_video_profile(dec->base.profile);
>+    uint32_t caps = 0;
>+    uint32_t endmarker = 0;
>+
>+    /*
>+     * 0x000..0x100: picparm_bsp
>+     * 0x200..0x500: picparm_vp
>+     * 0x500..0x700: comm
>+     * 0x700..onward: raw bitstream
>+     */
>+
>+    switch (codec){
>+    case PIPE_VIDEO_FORMAT_MPEG12:
>+       caps = nouveau_vp3_fill_picparm_mpeg12_bsp(dec, desc.mpeg12, bsp_bo->map);
>+       endmarker = 0xb7010000;
>+       break;
>+    case PIPE_VIDEO_FORMAT_MPEG4:
>+       caps = nouveau_vp3_fill_picparm_mpeg4_bsp(dec, desc.mpeg4, bsp_bo->map);
>+       endmarker = 0xb1010000;
>+       break;
>+    case PIPE_VIDEO_FORMAT_VC1:
>+       caps = nouveau_vp3_fill_picparm_vc1_bsp(dec, desc.vc1, bsp_bo->map);
>+       endmarker = 0x0a010000;
>+       break;
>+    case PIPE_VIDEO_FORMAT_MPEG4_AVC:
>+       caps = nouveau_vp3_fill_picparm_h264_bsp(dec, desc.h264, bsp_bo->map);
>+       endmarker = 0x0b010000;
>+       break;
>+    default: assert(0); return -1;
>+
>+        caps |= 0 << 16; // reset struct comm if flag is set
>+        caps |= 1 << 17; // enable watchdog
>+        caps |= 0 << 18; // do not report error to VP, so it can continue decoding what we have
>+        caps |= 0 << 19; // if enabled, use crypto crap?
>+    }
> 
>-   /* Append end sequence */
>-   *(uint32_t *)bsp = endmarker;
>-   bsp += 4;
>-   *(uint32_t *)bsp = 0x00000000;
>-   bsp += 4;
>-   *(uint32_t *)bsp = endmarker;
>-   bsp += 4;
>-   *(uint32_t *)bsp = 0x00000000;
>+    /* Append end sequence */
>+    *(uint32_t *)dec->bsp_ptr = endmarker;
>+    dec->bsp_ptr += 4;
>+    *(uint32_t *)dec->bsp_ptr = 0x00000000;
>+    dec->bsp_ptr += 4;
>+    *(uint32_t *)dec->bsp_ptr  = endmarker;
>+    dec->bsp_ptr += 4;
>+    *(uint32_t *)dec->bsp_ptr = 0x00000000;
>+
>+    return caps;
>+}
>+
>+uint32_t
>+nouveau_vp3_bsp(struct nouveau_vp3_decoder *dec,  union pipe_desc desc,
>+                struct nouveau_vp3_video_buffer *target,
>+                unsigned comm_seq, unsigned num_buffers,
>+                const void *const *data, const unsigned *num_bytes)
>+{
>+  nouveau_vp3_bsp_begin(dec);
>+  nouveau_vp3_bsp_next(dec, num_buffers, data, num_bytes);
>+  uint32_t caps = nouveau_vp3_bsp_end(dec, desc);
> 
>-   return caps;
>+  return caps;
> }
>diff --git a/src/gallium/drivers/nouveau/nvc0/nvc0_video.c b/src/gallium/drivers/nouveau/nvc0/nvc0_video.c
>index 48ffac1..1816a27 100644
>--- a/src/gallium/drivers/nouveau/nvc0/nvc0_video.c
>+++ b/src/gallium/drivers/nouveau/nvc0/nvc0_video.c
>@@ -26,6 +26,22 @@
> #include "util/u_format.h"
> 
> static void
>+nvc0_decoder_begin_frame(struct pipe_video_codec *decoder,
>+                         struct pipe_video_buffer *target,
>+                         struct pipe_picture_desc *picture)
>+{
>+    struct nouveau_vp3_decoder *dec = (struct nouveau_vp3_decoder *)decoder;
>+
>+    assert(dec);
>+    assert(target);
>+    assert(target->base.buffer_format == PIPE_FORMAT_NV12);
>+
>+    ++dec->fence_seq;
>+
>+    nouveau_vp3_bsp_begin(dec);
>+}
>+
>+static void
> nvc0_decoder_decode_bitstream(struct pipe_video_codec *decoder,
>                               struct pipe_video_buffer *video_target,
>                               struct pipe_picture_desc *picture,
>@@ -34,26 +50,45 @@ nvc0_decoder_decode_bitstream(struct pipe_video_codec *decoder,
>                               const unsigned *num_bytes)
> {
>    struct nouveau_vp3_decoder *dec = (struct nouveau_vp3_decoder *)decoder;
>-   struct nouveau_vp3_video_buffer *target = (struct nouveau_vp3_video_buffer *)video_target;
>-   uint32_t comm_seq = ++dec->fence_seq;
>-   union pipe_desc desc;
>+   uint32_t comm_seq = dec->fence_seq;
>+   unsigned ret = 0;
> 
>-   unsigned vp_caps, is_ref, ret;
>-   struct nouveau_vp3_video_buffer *refs[16] = {};
>+   assert(dec);
> 
>-   desc.base = picture;
>+   ret = nvc0_decoder_bsp_next_chunk(dec, comm_seq, num_buffers, data, num_bytes);
> 
>-   assert(target->base.buffer_format == PIPE_FORMAT_NV12);
>+   assert(ret == 2);
>+}
> 
>-   ret = nvc0_decoder_bsp(dec, desc, target, comm_seq,
>-                          num_buffers, data, num_bytes,
>-                          &vp_caps, &is_ref, refs);
>+static void
>+nvc0_decoder_end_frame(struct pipe_video_codec *decoder,
>+                       struct pipe_video_buffer *video_target,
>+                       struct pipe_picture_desc *picture)
>+{
>+    struct nouveau_vp3_decoder *dec = (struct nouveau_vp3_decoder *)decoder;
>+    struct nouveau_vp3_video_buffer *target = (struct nouveau_vp3_video_buffer *)video_target;
>+    uint32_t comm_seq = dec->fence_seq;
>+    uint32_t caps = 0;
>+    union pipe_desc desc;
>+    unsigned vp_caps = 0;
>+    unsigned is_ref = 0;
>+    unsigned ret = 0;
> 
>-   /* did we decode bitstream correctly? */
>-   assert(ret == 2);
>+    struct nouveau_vp3_video_buffer *refs[16] = {};
>+
>+    assert(dec);
>+
>+    desc.base = picture;
>+
>+    caps = nouveau_vp3_bsp_end(dec, desc);
>+
>+    ret = nvc0_decoder_bsp_end_frame(dec, desc, target, comm_seq, caps, &vp_caps, &is_ref, refs);
>+
>+    /* did we decode bitstream correctly? */
>+    assert(ret == 2);
> 
>-   nvc0_decoder_vp(dec, desc, target, comm_seq, vp_caps, is_ref, refs);
>-   nvc0_decoder_ppp(dec, desc, target, comm_seq);
>+    nvc0_decoder_vp(dec, desc, target, comm_seq, vp_caps, is_ref, refs);
>+    nvc0_decoder_ppp(dec, desc, target, comm_seq);
> }
> 
> struct pipe_video_codec *
>@@ -164,7 +199,9 @@ nvc0_create_decoder(struct pipe_context *context,
>    PUSH_DATA (push[2], dec->ppp->handle);
> 
>    dec->base.context = context;
>+   dec->base.begin_frame = nvc0_decoder_begin_frame;
>    dec->base.decode_bitstream = nvc0_decoder_decode_bitstream;
>+   dec->base.end_frame = nvc0_decoder_end_frame;
> 
>    for (i = 0; i < NOUVEAU_VP3_VIDEO_QDEPTH && !ret; ++i)
>       ret = nouveau_bo_new(screen->device, NOUVEAU_BO_VRAM,
>diff --git a/src/gallium/drivers/nouveau/nvc0/nvc0_video.h b/src/gallium/drivers/nouveau/nvc0/nvc0_video.h
>index 9ee0280..f5ab8d4 100644
>--- a/src/gallium/drivers/nouveau/nvc0/nvc0_video.h
>+++ b/src/gallium/drivers/nouveau/nvc0/nvc0_video.h
>@@ -29,6 +29,19 @@
> 
> #include "util/u_video.h"
> 
>+
>+extern unsigned
>+nvc0_decoder_bsp_next_chunk(struct nouveau_vp3_decoder *dec,
>+                            unsigned comm_seq, unsigned num_buffers,
>+                             const void *const *data, const unsigned *num_bytes);
>+
>+extern unsigned
>+nvc0_decoder_bsp_end_frame(struct nouveau_vp3_decoder *dec, union pipe_desc desc,
>+                           struct nouveau_vp3_video_buffer *target,
>+                           unsigned comm_seq,
>+                           uint32_t caps, unsigned *vp_caps, unsigned *is_ref,
>+                           struct nouveau_vp3_video_buffer *refs[16]);
>+
> extern unsigned
> nvc0_decoder_bsp(struct nouveau_vp3_decoder *dec, union pipe_desc desc,
>                  struct nouveau_vp3_video_buffer *target,
>diff --git a/src/gallium/drivers/nouveau/nvc0/nvc0_video_bsp.c b/src/gallium/drivers/nouveau/nvc0/nvc0_video_bsp.c
>index 9139bc1..786d2a7 100644
>--- a/src/gallium/drivers/nouveau/nvc0/nvc0_video_bsp.c
>+++ b/src/gallium/drivers/nouveau/nvc0/nvc0_video_bsp.c
>@@ -31,6 +31,212 @@ static void dump_comm_bsp(struct comm *comm)
> }
> #endif
> 
>+
>+unsigned
>+nvc0_decoder_bsp_next_chunk(struct nouveau_vp3_decoder *dec,
>+                            unsigned comm_seq, unsigned num_buffers,
>+                            const void *const *data, const unsigned *num_bytes)
>+{
>+    struct nouveau_bo *bsp_bo = dec->bsp_bo[comm_seq % NOUVEAU_VP3_VIDEO_QDEPTH];
>+    struct nouveau_bo *inter_bo = dec->inter_bo[comm_seq & 1];
>+    uint32_t bsp_size = 0;
>+    uint32_t i = 0;
>+    unsigned ret = 0;
>+
>+    bsp_size = dec->bsp_size;
>+    for (i = 0; i < num_buffers; i++)
>+       bsp_size += num_bytes[i];
>+    bsp_size += 256; /* the 4 end markers */
>+
>+    if (!bsp_bo || bsp_size > bsp_bo->size) {
>+       union nouveau_bo_config cfg;
>+       struct nouveau_bo *tmp_bo = NULL;
>+
>+       cfg.nvc0.tile_mode = 0x10;
>+       cfg.nvc0.memtype = 0xfe;
>+
>+       /* round up to the nearest mb */
>+       bsp_size += (1 << 20) - 1;
>+       bsp_size &= ~((1 << 20) - 1);
>+
>+       printf(" *** TODOD need to resize nouveau_bo \n");
>+
>+       ret = nouveau_bo_new(dec->bitplane_bo->device, NOUVEAU_BO_VRAM, 0, bsp_size, &cfg, &tmp_bo);
>+       if (ret) {
>+          debug_printf("reallocating bsp %u -> %u failed with %i\n",
>+                       bsp_bo ? (unsigned)bsp_bo->size : 0, bsp_size, ret);
>+          return -1;
>+       }
>+
>+       /*ret = nouveau_bo_map(tmp_bo, NOUVEAU_BO_WR, dec->client);
>+       if (ret) {
>+          debug_printf("map failed: %i %s\n", ret, strerror(-ret));
>+          return -1;
>+       }
>+       nouveau_bo_map(bsp_bo, NOUVEAU_BO_RD, dec->client);
>+       memcpy(tmp_bo->map, bsp_bo->map, bsp_size);*/
>+       // TODO map again ?
>+
>+       nouveau_bo_ref(NULL, &bsp_bo);
>+       dec->bsp_bo[comm_seq % NOUVEAU_VP3_VIDEO_QDEPTH] = bsp_bo = tmp_bo;
>+    }
>+
>+    if (!inter_bo || bsp_bo->size * 4 > inter_bo->size) {
>+       union nouveau_bo_config cfg;
>+       struct nouveau_bo *tmp_bo = NULL;
>+
>+       cfg.nvc0.tile_mode = 0x10;
>+       cfg.nvc0.memtype = 0xfe;
>+
>+       ret = nouveau_bo_new(dec->bitplane_bo->device, NOUVEAU_BO_VRAM, 0, bsp_bo->size * 4, &cfg, &tmp_bo);
>+       if (ret) {
>+          debug_printf("reallocating inter %u -> %u failed with %i\n",
>+                       inter_bo ? (unsigned)inter_bo->size : 0, (unsigned)bsp_bo->size * 4, ret);
>+          return -1;
>+       }
>+
>+       /*ret = nouveau_bo_map(tmp_bo, NOUVEAU_BO_WR, dec->client);
>+       if (ret) {
>+          debug_printf("map failed: %i %s\n", ret, strerror(-ret));
>+          return -1;
>+       }
>+       nouveau_bo_map(inter_bo, NOUVEAU_BO_RD, dec->client);
>+       memcpy(tmp_bo->map, inter_bo->map, bsp_bo->size * 4);*/
>+       // TODO map again ?
>+
>+       nouveau_bo_ref(NULL, &inter_bo);
>+       dec->inter_bo[comm_seq & 1] = inter_bo = tmp_bo;
>+    }
>+
>+    dec->bsp_size = bsp_size - 256;
>+
>+    // move this to resize
>+    /*ret = nouveau_bo_map(bsp_bo, NOUVEAU_BO_WR, dec->client);
>+    if (ret) {
>+       debug_printf("map failed: %i %s\n", ret, strerror(-ret));
>+       return -1;
>+    }*/
>+
>+    nouveau_vp3_bsp_next(dec, num_buffers, data, num_bytes);
>+
>+    return 2;
>+}
>+
>+
>+unsigned
>+nvc0_decoder_bsp_end_frame(struct nouveau_vp3_decoder *dec, union pipe_desc desc,
>+                           struct nouveau_vp3_video_buffer *target,
>+                           unsigned comm_seq,
>+                           uint32_t caps, unsigned *vp_caps, unsigned *is_ref,
>+                           struct nouveau_vp3_video_buffer *refs[16])
>+{
>+    struct nouveau_pushbuf *push = dec->pushbuf[0];
>+    enum pipe_video_format codec = u_reduce_video_profile(dec->base.profile);
>+    uint32_t bsp_addr, comm_addr, inter_addr;
>+    uint32_t slice_size, bucket_size, ring_size;
>+    struct nouveau_bo *bsp_bo = dec->bsp_bo[comm_seq % NOUVEAU_VP3_VIDEO_QDEPTH];
>+    struct nouveau_bo *inter_bo = dec->inter_bo[comm_seq & 1];
>+    unsigned fence_extra = 0;
>+    struct nouveau_pushbuf_refn bo_refs[] = {
>+       { bsp_bo, NOUVEAU_BO_RD | NOUVEAU_BO_VRAM },
>+       { inter_bo, NOUVEAU_BO_WR | NOUVEAU_BO_VRAM },
>+ #if NOUVEAU_VP3_DEBUG_FENCE
>+       { dec->fence_bo, NOUVEAU_BO_WR | NOUVEAU_BO_GART },
>+ #endif
>+       { dec->bitplane_bo, NOUVEAU_BO_RDWR | NOUVEAU_BO_VRAM },
>+    };
>+    int num_refs = sizeof(bo_refs)/sizeof(*bo_refs);
>+
>+    if (!dec->bitplane_bo)
>+       num_refs--;
>+
>+ #if NOUVEAU_VP3_DEBUG_FENCE
>+    fence_extra = 4;
>+ #endif
>+
>+    bo_refs[0].bo = dec->bsp_bo[comm_seq % NOUVEAU_VP3_VIDEO_QDEPTH];
>+    bo_refs[1].bo = dec->inter_bo[comm_seq & 1];
>+
>+    nouveau_vp3_vp_caps(dec, desc, target, comm_seq, vp_caps, is_ref, refs);
>+
>+    nouveau_pushbuf_space(push, 6 + (codec == PIPE_VIDEO_FORMAT_MPEG4_AVC ? 9 : 7) + fence_extra + 2, num_refs, 0);
>+    nouveau_pushbuf_refn(push, bo_refs, num_refs);
>+
>+    bsp_addr = bsp_bo->offset >> 8;
>+    inter_addr = inter_bo->offset >> 8;
>+
>+ #if NOUVEAU_VP3_DEBUG_FENCE
>+    memset(dec->comm, 0, 0x200);
>+    comm_addr = (dec->fence_bo->offset + COMM_OFFSET) >> 8;
>+ #else
>+    comm_addr = bsp_addr + (COMM_OFFSET>>8);
>+ #endif
>+
>+    BEGIN_NVC0(push, SUBC_BSP(0x700), 5);
>+    PUSH_DATA (push, caps); // 700 cmd
>+    PUSH_DATA (push, bsp_addr + 1); // 704 strparm_bsp
>+    PUSH_DATA (push, bsp_addr + 7); // 708 str addr
>+    PUSH_DATA (push, comm_addr); // 70c comm
>+    PUSH_DATA (push, comm_seq); // 710 seq
>+
>+    if (codec != PIPE_VIDEO_FORMAT_MPEG4_AVC) {
>+       u32 bitplane_addr;
>+
>+       bitplane_addr = dec->bitplane_bo->offset >> 8;
>+
>+       nouveau_vp3_inter_sizes(dec, 1, &slice_size, &bucket_size, &ring_size);
>+       BEGIN_NVC0(push, SUBC_BSP(0x400), 6);
>+       PUSH_DATA (push, bsp_addr); // 400 picparm addr
>+       PUSH_DATA (push, inter_addr); // 404 interparm addr
>+       PUSH_DATA (push, inter_addr + slice_size + bucket_size); // 408 interdata addr
>+       PUSH_DATA (push, ring_size << 8); // 40c interdata_size
>+       PUSH_DATA (push, bitplane_addr); // 410 BITPLANE_DATA
>+       PUSH_DATA (push, 0x400); // 414 BITPLANE_DATA_SIZE
>+    } else {
>+       nouveau_vp3_inter_sizes(dec, desc.h264->slice_count, &slice_size, &bucket_size, &ring_size);
>+       BEGIN_NVC0(push, SUBC_BSP(0x400), 8);
>+       PUSH_DATA (push, bsp_addr); // 400 picparm addr
>+       PUSH_DATA (push, inter_addr); // 404 interparm addr
>+       PUSH_DATA (push, slice_size << 8); // 408 interparm size?
>+       PUSH_DATA (push, inter_addr + slice_size + bucket_size); // 40c interdata addr
>+       PUSH_DATA (push, ring_size << 8); // 410 interdata size
>+       PUSH_DATA (push, inter_addr + slice_size); // 414 bucket?
>+       PUSH_DATA (push, bucket_size << 8); // 418 bucket size? unshifted..
>+       PUSH_DATA (push, 0); // 41c targets
>+       // TODO: Double check 414 / 418 with nvidia trace
>+    }
>+
>+ #if NOUVEAU_VP3_DEBUG_FENCE
>+    BEGIN_NVC0(push, SUBC_BSP(0x240), 3);
>+    PUSH_DATAh(push, dec->fence_bo->offset);
>+    PUSH_DATA (push, dec->fence_bo->offset);
>+    PUSH_DATA (push, dec->fence_seq);
>+
>+    BEGIN_NVC0(push, SUBC_BSP(0x300), 1);
>+    PUSH_DATA (push, 1);
>+    PUSH_KICK (push);
>+
>+    {
>+       unsigned spin = 0;
>+       do {
>+          usleep(100);
>+          if ((spin++ & 0xff) == 0xff) {
>+             debug_printf("b%u: %u\n", dec->fence_seq, dec->fence_map[0]);
>+             dump_comm_bsp(dec->comm);
>+          }
>+       } while (dec->fence_seq > dec->fence_map[0]);
>+    }
>+
>+    dump_comm_bsp(dec->comm);
>+    return dec->comm->status[comm_seq & 0xf];
>+ #else
>+    BEGIN_NVC0(push, SUBC_BSP(0x300), 1);
>+    PUSH_DATA (push, 0);
>+    PUSH_KICK (push);
>+    return 2;
>+ #endif
>+}
>+
> unsigned
> nvc0_decoder_bsp(struct nouveau_vp3_decoder *dec, union pipe_desc desc,
>                  struct nouveau_vp3_video_buffer *target,
>-- 
>1.9.1
>

Actions: View | Splinter Review

Attachments on bug 89969: 114987 | 117214 | 117215 | 117358 | 117399 | 117910 | 117928 | 117929 | 117930 | 117931 | 117932 | 117934 | 117936 | 117937 | 117938 | 117939

Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.