Nouveau - Jul 2013 - [PATCH 2/2] nv50: allow forcing PMPEG use, for ease of testing

It seems as though at least the NV44 expects the same format as VP2
does. But NV96 expects the format as currently done.

Signed-off-by: Ilia Mirkin <imirkin at alum.mit.edu>
---
 src/gallium/drivers/nouveau/nouveau_video.c | 43 +++++++++++++++++++++++++----
 src/gallium/drivers/nouveau/nouveau_video.h |  2 ++
 src/gallium/drivers/nv30/nv30_context.c     |  2 ++
 src/gallium/drivers/nv30/nv30_screen.c      |  1 +
 4 files changed, 43 insertions(+), 5 deletions(-)

diff --git a/src/gallium/drivers/nouveau/nouveau_video.c
b/src/gallium/drivers/nouveau/nouveau_video.c
index 9357508..62cb8d5 100644
--- a/src/gallium/drivers/nouveau/nouveau_video.c
+++ b/src/gallium/drivers/nouveau/nouveau_video.c
@@ -101,7 +101,32 @@ nouveau_vpe_fini(struct nouveau_decoder *dec) {
 }
 
 static INLINE void
-nouveau_vpe_mb_dct_blocks(struct nouveau_decoder *dec, const struct
pipe_mpeg12_macroblock *mb)
+nouveau_vpe_mb_dct_blocks_31(struct nouveau_decoder *dec, const struct
pipe_mpeg12_macroblock *mb)
+{
+   int cbb;
+   unsigned cbp = mb->coded_block_pattern;
+   short *db = mb->blocks;
+   for (cbb = 0x20; cbb > 0; cbb >>= 1) {
+      if (cbb & cbp) {
+         int i, found = 0;
+         for (i = 0; i < 64; ++i) {
+            if (!db[i]) continue;
+            dec->data[dec->data_pos++] = (db[i] << 16) | (i * 2);
+            found = 1;
+         }
+         if (found)
+            dec->data[dec->data_pos - 1] |= 1;
+         else
+            dec->data[dec->data_pos++] = 1;
+         db += 64;
+      } else if (mb->macroblock_type & PIPE_MPEG12_MB_TYPE_INTRA) {
+         dec->data[dec->data_pos++] = 1;
+      }
+   }
+}
+
+static INLINE void
+nouveau_vpe_mb_dct_blocks_84(struct nouveau_decoder *dec, const struct
pipe_mpeg12_macroblock *mb)
 {
    int cbb;
    unsigned cbp = mb->coded_block_pattern;
@@ -455,10 +480,14 @@ nouveau_decoder_decode_macroblock(struct
pipe_video_decoder *decoder,
          nouveau_vpe_mb_mv_header(dec, mb, FALSE);
          nouveau_vpe_mb_dct_header(dec, mb, FALSE);
       }
-      if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT)
-         nouveau_vpe_mb_dct_blocks(dec, mb);
-      else
+      if (dec->base.entrypoint <= PIPE_VIDEO_ENTRYPOINT_IDCT) {
+         if (dec->is8274)
+            nouveau_vpe_mb_dct_blocks_84(dec, mb);
+         else
+            nouveau_vpe_mb_dct_blocks_31(dec, mb);
+      } else {
          nouveau_vpe_mb_data_blocks(dec, mb);
+      }
    }
 }
 
@@ -528,6 +557,8 @@ nouveau_create_decoder(struct pipe_context *context,
       goto vl;
    if (screen->device->chipset >= 0x98 &&
screen->device->chipset != 0xa0)
       goto vl;
+   if (screen->device->chipset < 0x31)
+      goto vl;
 
    dec = CALLOC_STRUCT(nouveau_decoder);
    if (!dec)
@@ -577,6 +608,7 @@ nouveau_create_decoder(struct pipe_context *context,
    dec->base.end_frame = nouveau_decoder_end_frame;
    dec->base.flush = nouveau_decoder_flush;
    dec->screen = screen;
+   dec->is8274 = is8274;
 
    ret = nouveau_bo_new(dec->screen->device, NOUVEAU_BO_GART |
NOUVEAU_BO_MAP,
                         0, 1024 * 1024, NULL, &dec->cmd_bo);
@@ -793,7 +825,8 @@ nouveau_video_buffer_create(struct pipe_context *pipe,
     * and it only supports the NV12 format
     */
    if (templat->buffer_format != PIPE_FORMAT_NV12 ||
getenv("XVMC_VL") ||
-       (screen->device->chipset >= 0x98 &&
screen->device->chipset != 0xa0))
+       (screen->device->chipset >= 0x98 &&
screen->device->chipset != 0xa0) ||
+       screen->device->chipset < 0x31)
       return vl_video_buffer_create(pipe, templat);
 
    assert(templat->chroma_format == PIPE_VIDEO_CHROMA_FORMAT_420);
diff --git a/src/gallium/drivers/nouveau/nouveau_video.h
b/src/gallium/drivers/nouveau/nouveau_video.h
index 1d6ced0..6306ac3 100644
--- a/src/gallium/drivers/nouveau/nouveau_video.h
+++ b/src/gallium/drivers/nouveau/nouveau_video.h
@@ -26,6 +26,8 @@ struct nouveau_decoder {
    struct nouveau_object *mpeg;
    struct nouveau_bo *cmd_bo, *data_bo, *fence_bo;
 
+   bool is8274;
+
    unsigned *fence_map;
    unsigned fence_seq;
 
diff --git a/src/gallium/drivers/nv30/nv30_context.c
b/src/gallium/drivers/nv30/nv30_context.c
index bd05042..e872c02 100644
--- a/src/gallium/drivers/nv30/nv30_context.c
+++ b/src/gallium/drivers/nv30/nv30_context.c
@@ -257,5 +257,7 @@ nv30_context_create(struct pipe_screen *pscreen, void *priv)
       return NULL;
    }
 
+   nouveau_context_init_vdec(&nv30->base);
+
    return pipe;
 }
diff --git a/src/gallium/drivers/nv30/nv30_screen.c
b/src/gallium/drivers/nv30/nv30_screen.c
index 3d55d6f..40e8b5f 100644
--- a/src/gallium/drivers/nv30/nv30_screen.c
+++ b/src/gallium/drivers/nv30/nv30_screen.c
@@ -377,6 +377,7 @@ nv30_screen_create(struct nouveau_device *dev)
    pscreen->context_create = nv30_context_create;
    pscreen->is_format_supported = nv30_screen_is_format_supported;
    nv30_resource_screen_init(pscreen);
+   nouveau_screen_init_vdec(&screen->base);
 
    screen->base.fence.emit = nv30_screen_fence_emit;
    screen->base.fence.update = nv30_screen_fence_update;
-- 
1.8.1.5

This also allows people who don't want to install the binary blobs
required for VP2 to still get MPEG decoding.

Signed-off-by: Ilia Mirkin <imirkin at alum.mit.edu>
---
 src/gallium/drivers/nv50/nv50_context.c | 2 +-
 src/gallium/drivers/nv50/nv50_screen.c  | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/src/gallium/drivers/nv50/nv50_context.c
b/src/gallium/drivers/nv50/nv50_context.c
index 79a0473..3654a41 100644
--- a/src/gallium/drivers/nv50/nv50_context.c
+++ b/src/gallium/drivers/nv50/nv50_context.c
@@ -258,7 +258,7 @@ nv50_create(struct pipe_screen *pscreen, void *priv)
    draw_set_rasterize_stage(nv50->draw, nv50_draw_render_stage(nv50));
 #endif
 
-   if (screen->base.device->chipset < 0x84) {
+   if (screen->base.device->chipset < 0x84 ||
getenv("XVMC_PMPEG")) {
       /* PMPEG */
       nouveau_context_init_vdec(&nv50->base);
    } else if (screen->base.device->chipset < 0x98 ||
diff --git a/src/gallium/drivers/nv50/nv50_screen.c
b/src/gallium/drivers/nv50/nv50_screen.c
index 2951eb4..959366e 100644
--- a/src/gallium/drivers/nv50/nv50_screen.c
+++ b/src/gallium/drivers/nv50/nv50_screen.c
@@ -647,7 +647,7 @@ nv50_screen_create(struct nouveau_device *dev)
 
    nv50_screen_init_resource_functions(pscreen);
 
-   if (screen->base.device->chipset < 0x84) {
+   if (screen->base.device->chipset < 0x84 ||
getenv("XVMC_PMPEG")) {
       /* PMPEG */
       nouveau_screen_init_vdec(&screen->base);
    } else if (screen->base.device->chipset < 0x98 ||
-- 
1.8.1.5

This also allows people who don't want to install the binary blobs
required for VP2 to still get MPEG decoding.

Signed-off-by: Ilia Mirkin <imirkin at alum.mit.edu>
---

v1 -> v2:

  - switched to using debug_get_bool_option and NOUVEAU prefix, for which
    there is some precedent

 src/gallium/drivers/nv50/nv50_context.c | 3 ++-
 src/gallium/drivers/nv50/nv50_screen.c  | 3 ++-
 2 files changed, 4 insertions(+), 2 deletions(-)

diff --git a/src/gallium/drivers/nv50/nv50_context.c
b/src/gallium/drivers/nv50/nv50_context.c
index 79a0473..185d241 100644
--- a/src/gallium/drivers/nv50/nv50_context.c
+++ b/src/gallium/drivers/nv50/nv50_context.c
@@ -258,7 +258,8 @@ nv50_create(struct pipe_screen *pscreen, void *priv)
    draw_set_rasterize_stage(nv50->draw, nv50_draw_render_stage(nv50));
 #endif
 
-   if (screen->base.device->chipset < 0x84) {
+   if (screen->base.device->chipset < 0x84 ||
+       debug_get_bool_option("NOUVEAU_PMPEG", FALSE)) {
       /* PMPEG */
       nouveau_context_init_vdec(&nv50->base);
    } else if (screen->base.device->chipset < 0x98 ||
diff --git a/src/gallium/drivers/nv50/nv50_screen.c
b/src/gallium/drivers/nv50/nv50_screen.c
index 2951eb4..0cbee5d 100644
--- a/src/gallium/drivers/nv50/nv50_screen.c
+++ b/src/gallium/drivers/nv50/nv50_screen.c
@@ -647,7 +647,8 @@ nv50_screen_create(struct nouveau_device *dev)
 
    nv50_screen_init_resource_functions(pscreen);
 
-   if (screen->base.device->chipset < 0x84) {
+   if (screen->base.device->chipset < 0x84 ||
+       debug_get_bool_option("NOUVEAU_PMPEG", FALSE)) {
       /* PMPEG */
       nouveau_screen_init_vdec(&screen->base);
    } else if (screen->base.device->chipset < 0x98 ||
-- 
1.8.1.5