[Mesa-dev] [PATCH 1/8] nouveau: extract memcpy loop from nouveau_vp3_bsp

[Julien Isorce]

This patch add 3 functions to split nouveau_vp3_bsp:

nouveau_vp3_bsp_begin
nouveau_vp3_bsp_next
nouveau_vp3_bsp_end

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

Signed-off-by: Julien Isorce <j.isorce at samsung.com>
---
 src/gallium/drivers/nouveau/nouveau_vp3_video.h    |  15 +++
 .../drivers/nouveau/nouveau_vp3_video_bsp.c        | 105 +++++++++++++++++++++
 2 files changed, 120 insertions(+)

diff --git a/src/gallium/drivers/nouveau/nouveau_vp3_video.h b/src/gallium/drivers/nouveau/nouveau_vp3_video.h
index 33e3bef..16563de 100644
--- a/src/gallium/drivers/nouveau/nouveau_vp3_video.h
+++ b/src/gallium/drivers/nouveau/nouveau_vp3_video.h
@@ -114,6 +114,11 @@ struct nouveau_vp3_decoder {
    unsigned fence_seq, fw_sizes, last_frame_num, tmp_stride, ref_stride;
 
    unsigned bsp_idx, vp_idx, ppp_idx;
+
+   // To manage chunck decoding.
+   char *bsp_ptr;
+   unsigned bsp_size;
+   unsigned int nb_slices;
 };
 
 struct comm {
@@ -215,6 +220,16 @@ nouveau_vp3_bsp(struct nouveau_vp3_decoder *dec,  union pipe_desc desc,
                 const void *const *data, const unsigned *num_bytes);
 
 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);
+
+void
 nouveau_vp3_vp_caps(struct nouveau_vp3_decoder *dec, union pipe_desc desc,
                     struct nouveau_vp3_video_buffer *target, unsigned comm_seq,
                     unsigned *caps, unsigned *is_ref,
diff --git a/src/gallium/drivers/nouveau/nouveau_vp3_video_bsp.c b/src/gallium/drivers/nouveau/nouveau_vp3_video_bsp.c
index 6d968c1..10dd23b 100644
--- a/src/gallium/drivers/nouveau/nouveau_vp3_video_bsp.c
+++ b/src/gallium/drivers/nouveau/nouveau_vp3_video_bsp.c
@@ -308,3 +308,108 @@ nouveau_vp3_bsp(struct nouveau_vp3_decoder *dec,  union pipe_desc desc,
 
    return caps;
 }
+
+void
+nouveau_vp3_bsp_begin(struct nouveau_vp3_decoder *dec)
+{
+    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;
+
+    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_origin = bsp_bo->map;
+    struct strparm_bsp *str_bsp = NULL;
+    int i = 0;
+
+    ++dec->nb_slices;
+
+    bsp_origin += 0x100;
+    str_bsp = (struct strparm_bsp *)bsp_origin;
+
+    for (i = 0; i < num_buffers; ++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;
+    char *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;
+    }
+
+    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 *)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;
+}
-- 
1.9.1