nvc0: add support for st/va

- split nvc0_decoder_bsp in begin/next/end
- preserve content buffer when calling nvc0_decoder_bsp_next
- implement pipe_video_codec::begin_frame/end_frame

https://bugs.freedesktop.org/show_bug.cgi?id=89969

Signed-off-by: Julien Isorce <[email protected]>
Reviewed-by: Ilia Mirkin <[email protected]>

3 files changed, 133 insertions, 50 deletions

diff --git a/src/gallium/drivers/nouveau/nvc0/nvc0_video.c b/src/gallium/drivers/nouveau/nvc0/nvc0_video.c
index 5a946ca6314..a9fd1d20942 100644
--- a/src/gallium/drivers/nouveau/nvc0/nvc0_video.c
+++ b/src/gallium/drivers/nouveau/nvc0/nvc0_video.c
@@ -26,6 +26,24 @@
 #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;
+   uint32_t comm_seq = ++dec->fence_seq;
+   unsigned ret = 0;
+
+   assert(dec);
+   assert(target);
+   assert(target->buffer_format == PIPE_FORMAT_NV12);
+
+   ret = nvc0_decoder_bsp_begin(dec, comm_seq);
+
+   assert(ret == 2);
+}
+
+static void
 nvc0_decoder_decode_bitstream(struct pipe_video_codec *decoder,
                               struct pipe_video_buffer *video_target,
                               struct pipe_picture_desc *picture,
@@ -34,8 +52,24 @@ nvc0_decoder_decode_bitstream(struct pipe_video_codec *decoder,
                               const unsigned *num_bytes)
 {
    struct nouveau_vp3_decoder *dec = (struct nouveau_vp3_decoder *)decoder;
+   uint32_t comm_seq = dec->fence_seq;
+   unsigned ret = 0;
+
+   assert(decoder);
+
+   ret = nvc0_decoder_bsp_next(dec, comm_seq, num_buffers, data, num_bytes);
+
+   assert(ret == 2);
+}
+
+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 comm_seq = dec->fence_seq;
    union pipe_desc desc;
 
    unsigned vp_caps, is_ref, ret;
@@ -43,11 +77,7 @@ nvc0_decoder_decode_bitstream(struct pipe_video_codec *decoder,
 
    desc.base = picture;
 
-   assert(target->base.buffer_format == PIPE_FORMAT_NV12);
-
-   ret = nvc0_decoder_bsp(dec, desc, target, comm_seq,
-                          num_buffers, data, num_bytes,
-                          &vp_caps, &is_ref, refs);
+   ret = nvc0_decoder_bsp_end(dec, desc, target, comm_seq, &vp_caps, &is_ref, refs);
 
    /* did we decode bitstream correctly? */
    assert(ret == 2);
@@ -164,7 +194,9 @@ 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 9ee0280f8ea..cf3c942355b 100644
--- a/src/gallium/drivers/nouveau/nvc0/nvc0_video.h
+++ b/src/gallium/drivers/nouveau/nvc0/nvc0_video.h
@@ -30,12 +30,18 @@
 #include "util/u_video.h"
 
 extern unsigned
-nvc0_decoder_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,
-                 unsigned *vp_caps, unsigned *is_ref,
-                 struct nouveau_vp3_video_buffer *refs[16]);
+nvc0_decoder_bsp_begin(struct nouveau_vp3_decoder *dec, unsigned comm_seq);
+
+extern unsigned
+nvc0_decoder_bsp_next(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(struct nouveau_vp3_decoder *dec, union pipe_desc desc,
+                     struct nouveau_vp3_video_buffer *target,
+                     unsigned comm_seq, unsigned *vp_caps, unsigned *is_ref,
+                     struct nouveau_vp3_video_buffer *refs[16]);
 
 extern void
 nvc0_decoder_vp(struct nouveau_vp3_decoder *dec, union pipe_desc desc,
diff --git a/src/gallium/drivers/nouveau/nvc0/nvc0_video_bsp.c b/src/gallium/drivers/nouveau/nvc0/nvc0_video_bsp.c
index 6cedeaf9f27..c53f946a762 100644
--- a/src/gallium/drivers/nouveau/nvc0/nvc0_video_bsp.c
+++ b/src/gallium/drivers/nouveau/nvc0/nvc0_video_bsp.c
@@ -32,40 +32,34 @@ static void dump_comm_bsp(struct comm *comm)
 #endif
 
 unsigned
-nvc0_decoder_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,
-                 unsigned *vp_caps, unsigned *is_ref,
-                 struct nouveau_vp3_video_buffer *refs[16])
+nvc0_decoder_bsp_begin(struct nouveau_vp3_decoder *dec, unsigned comm_seq)
 {
-   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, bsp_size;
-   uint32_t caps, i;
-   int ret;
    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 = ARRAY_SIZE(bo_refs);
+   unsigned ret = 0;
 
-   if (!dec->bitplane_bo)
-      num_refs--;
+   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;
+   }
 
-#if NOUVEAU_VP3_DEBUG_FENCE
-   fence_extra = 4;
-#endif
+   nouveau_vp3_bsp_begin(dec);
+
+   return 2;
+}
+
+unsigned
+nvc0_decoder_bsp_next(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 = NOUVEAU_VP3_BSP_RESERVED_SIZE;
+   bsp_size = dec->bsp_ptr - (char *)bsp_bo->map;
    for (i = 0; i < num_buffers; i++)
       bsp_size += num_bytes[i];
    bsp_size += 256; /* the 4 end markers */
@@ -87,8 +81,23 @@ nvc0_decoder_bsp(struct nouveau_vp3_decoder *dec, union pipe_desc desc,
                       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;
+      }
+
+      /* Preserve previous buffer. */
+      /* TODO: offload this copy to the GPU, as otherwise we're reading and
+       * writing to VRAM. */
+      memcpy(tmp_bo->map, bsp_bo->map, bsp_bo->size);
+
+      /* update position to current chunk */
+      dec->bsp_ptr = tmp_bo->map + (dec->bsp_ptr - (char *)bsp_bo->map);
+
       nouveau_bo_ref(NULL, &bsp_bo);
-      bo_refs[0].bo = dec->bsp_bo[comm_seq % NOUVEAU_VP3_VIDEO_QDEPTH] = bsp_bo = tmp_bo;
+      dec->bsp_bo[comm_seq % NOUVEAU_VP3_VIDEO_QDEPTH] = bsp_bo = tmp_bo;
    }
 
    if (!inter_bo || bsp_bo->size * 4 > inter_bo->size) {
@@ -104,18 +113,54 @@ nvc0_decoder_bsp(struct nouveau_vp3_decoder *dec, union pipe_desc desc,
                       inter_bo ? (unsigned)inter_bo->size : 0, (unsigned)bsp_bo->size * 4, ret);
          return -1;
       }
-      nouveau_bo_ref(NULL, &inter_bo);
-      bo_refs[1].bo = dec->inter_bo[comm_seq & 1] = inter_bo = tmp_bo;
-   }
 
-   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;
+      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_ref(NULL, &inter_bo);
+      dec->inter_bo[comm_seq & 1] = inter_bo = tmp_bo;
    }
 
-   nouveau_vp3_bsp_begin(dec);
    nouveau_vp3_bsp_next(dec, num_buffers, data, num_bytes);
+
+   return 2;
+}
+
+
+unsigned
+nvc0_decoder_bsp_end(struct nouveau_vp3_decoder *dec, union pipe_desc desc,
+                     struct nouveau_vp3_video_buffer *target, unsigned comm_seq,
+                     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;
+   uint32_t caps;
+   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 = ARRAY_SIZE(bo_refs);
+
+   if (!dec->bitplane_bo)
+      num_refs--;
+
+#if NOUVEAU_VP3_DEBUG_FENCE
+   fence_extra = 4;
+#endif
+
    caps = nouveau_vp3_bsp_end(dec, desc);
 
    nouveau_vp3_vp_caps(dec, desc, target, comm_seq, vp_caps, is_ref, refs);