radeonsi: Add support for DPB resize

This patch will add support for dpb resize when low to high resolution
change/ svc use-cases.

With DPB tier1 type,vp9 svc decoder use cases are failed. This
Change will fix this[VCN1/VCN2].

Signed-off-by: SureshGuttula <suresh.guttula@amd.com>
Reviewed-by: Leo Liu <leo.liu@amd.com>
Reviewed-by: Boyuan Zhang <boyuan.zhang@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21548>

src/amd/common/ac_vcn_dec.h

@@ -160,6 +160,11 @@
 
 #define RDECODE_VP9_PROBS_DATA_SIZE                         2304
 
+/* *** decode flags *** */
+#define RDECODE_FLAGS_USE_DYNAMIC_DPB_MASK                  0x00000001
+#define RDECODE_FLAGS_USE_PAL_MASK                          0x00000008
+#define RDECODE_FLAGS_DPB_RESIZE_MASK                       0x00000100
+
 #define mmUVD_JPEG_CNTL                                     0x0200
 #define mmUVD_JPEG_CNTL_BASE_IDX                            1
 #define mmUVD_JPEG_RB_BASE                                  0x0201

src/gallium/drivers/radeonsi/radeon_uvd.c

@@ -1059,7 +1059,7 @@ static void ruvd_decode_bitstream(struct pipe_video_codec *decoder,
 
       if (new_size > buf->res->buf->size) {
          dec->ws->buffer_unmap(dec->ws, buf->res->buf);
-         if (!si_vid_resize_buffer(dec->screen, &dec->cs, buf, new_size)) {
+         if (!si_vid_resize_buffer(dec->screen, &dec->cs, buf, new_size, NULL)) {
             RVID_ERR("Can't resize bitstream buffer!");
             return;
          }

src/gallium/drivers/radeonsi/radeon_vcn_dec.c

@@ -1958,7 +1958,7 @@ static struct pb_buffer *rvcn_dec_message_decode(struct radeon_decoder *dec,
    rvcn_dec_message_drm_t *drm = NULL;
    rvcn_dec_message_dynamic_dpb_t *dynamic_dpb = NULL;
    rvcn_dec_message_dynamic_dpb_t2_t *dynamic_dpb_t2 = NULL;
-
+   bool dpb_resize = false;
    header = dec->msg;
    sizes += sizeof(rvcn_dec_message_header_t);
 
@@ -2044,9 +2044,9 @@ static struct pb_buffer *rvcn_dec_message_decode(struct radeon_decoder *dec,
 
    decode->bsd_size = align(dec->bs_size, 128);
 
-   if (!dec->dpb.res && dec->dpb_type != DPB_DYNAMIC_TIER_2) {
+   if (dec->dpb_type != DPB_DYNAMIC_TIER_2) {
       bool r;
-      if (dec->dpb_size) {
+      if (!dec->dpb.res && dec->dpb_size) {
          if (encrypted) {
             r = si_vid_create_tmz_buffer(dec->screen, &dec->dpb, dec->dpb_size, PIPE_USAGE_DEFAULT);
          } else {
@@ -2058,6 +2058,25 @@ static struct pb_buffer *rvcn_dec_message_decode(struct radeon_decoder *dec,
             return NULL;
          }
          si_vid_clear_buffer(dec->base.context, &dec->dpb);
+      } else if (dec->dpb_type == DPB_DYNAMIC_TIER_1 && dec->dpb.res &&
+                 (dec->max_width < dec->base.width || dec->max_height < dec->base.height)) {
+         struct rvid_buf_offset_info buf_offset_info;
+
+         buf_offset_info.num_units = (NUM_VP9_REFS + 1);
+         buf_offset_info.old_offset = (align(dec->max_width, dec->db_alignment) *
+            align(dec->max_height, dec->db_alignment) * 3 / 2);
+         buf_offset_info.new_offset = (align(dec->base.width, dec->db_alignment) *
+            align(dec->base.height, dec->db_alignment) * 3 / 2);
+
+         dec->dpb_size = calc_dpb_size(dec);
+         r = si_vid_resize_buffer(dec->screen, &dec->cs, &dec->dpb, dec->dpb_size, &buf_offset_info);
+         if (!r) {
+            RVID_ERR("Can't resize dpb.\n");
+            return NULL;
+	 }
+         dec->max_width = dec->base.width;
+         dec->max_height = dec->base.height;
+         dpb_resize = true;
       }
    }
 
@@ -2202,10 +2221,14 @@ static struct pb_buffer *rvcn_dec_message_decode(struct radeon_decoder *dec,
    }
 
    if (dec->dpb_type == DPB_DYNAMIC_TIER_1) {
-      decode->decode_flags = 1;
+      decode->decode_flags |= (RDECODE_FLAGS_USE_DYNAMIC_DPB_MASK | RDECODE_FLAGS_USE_PAL_MASK);
+      // Add decode flag for RESIZE_DPB ,when we do resize
+      if (dpb_resize == true)
+        decode->decode_flags |= RDECODE_FLAGS_DPB_RESIZE_MASK;
+
       dynamic_dpb->dpbArraySize = NUM_VP9_REFS + 1;
-      dynamic_dpb->dpbLumaPitch = align(decode->width_in_samples, dec->db_alignment);
-      dynamic_dpb->dpbLumaAlignedHeight = align(decode->height_in_samples, dec->db_alignment);
+      dynamic_dpb->dpbLumaPitch = align(dec->max_width, dec->db_alignment);
+      dynamic_dpb->dpbLumaAlignedHeight = align(dec->max_height, dec->db_alignment);
       dynamic_dpb->dpbLumaAlignedSize =
          dynamic_dpb->dpbLumaPitch * dynamic_dpb->dpbLumaAlignedHeight;
       dynamic_dpb->dpbChromaPitch = dynamic_dpb->dpbLumaPitch >> 1;
@@ -2846,7 +2869,7 @@ static void radeon_dec_decode_bitstream(struct pipe_video_codec *decoder,
       if (new_size > buf->res->buf->size) {
          dec->ws->buffer_unmap(dec->ws, buf->res->buf);
          dec->bs_ptr = NULL;
-         if (!si_vid_resize_buffer(dec->screen, &dec->cs, buf, new_size)) {
+         if (!si_vid_resize_buffer(dec->screen, &dec->cs, buf, new_size, NULL)) {
             RVID_ERR("Can't resize bitstream buffer!");
             return;
          }
@@ -3021,7 +3044,8 @@ struct pipe_video_codec *radeon_create_decoder(struct pipe_context *context,
    dec->base.context = context;
    dec->base.width = width;
    dec->base.height = height;
-
+   dec->max_width = width;
+   dec->max_height = height;
    dec->base.destroy = radeon_dec_destroy;
    dec->base.begin_frame = radeon_dec_begin_frame;
    dec->base.decode_macroblock = radeon_dec_decode_macroblock;

src/gallium/drivers/radeonsi/radeon_vcn_dec.h

@@ -51,6 +51,8 @@ struct radeon_decoder {
    unsigned dpb_size;
    unsigned last_width;
    unsigned last_height;
+   unsigned max_width;
+   unsigned max_height;
    unsigned addr_gfx_mode;
 
    struct pipe_screen *screen;

src/gallium/drivers/radeonsi/radeon_video.c

@@ -87,7 +87,8 @@ void si_vid_destroy_buffer(struct rvid_buffer *buffer)
 
 /* reallocate a buffer, preserving its content */
 bool si_vid_resize_buffer(struct pipe_screen *screen, struct radeon_cmdbuf *cs,
-                          struct rvid_buffer *new_buf, unsigned new_size)
+                          struct rvid_buffer *new_buf, unsigned new_size,
+                          struct rvid_buf_offset_info *buf_ofst_info)
 {
    struct si_screen *sscreen = (struct si_screen *)screen;
    struct radeon_winsys *ws = sscreen->ws;
@@ -106,11 +107,20 @@ bool si_vid_resize_buffer(struct pipe_screen *screen, struct radeon_cmdbuf *cs,
    if (!dst)
       goto error;
 
-   memcpy(dst, src, bytes);
-   if (new_size > bytes) {
-      new_size -= bytes;
-      dst += bytes;
+   if (buf_ofst_info) {
       memset(dst, 0, new_size);
+      for(int i =0; i < buf_ofst_info->num_units; i++) {
+          memcpy(dst, src, buf_ofst_info->old_offset);
+          dst += buf_ofst_info->new_offset;
+          src += buf_ofst_info->old_offset;
+      }
+   } else {
+      memcpy(dst, src, bytes);
+      if (new_size > bytes) {
+         new_size -= bytes;
+         dst += bytes;
+         memset(dst, 0, new_size);
+      }
    }
    ws->buffer_unmap(ws, new_buf->res->buf);
    ws->buffer_unmap(ws, old_buf.res->buf);

src/gallium/drivers/radeonsi/radeon_video.h

@@ -42,6 +42,13 @@ struct rvid_buffer {
    struct si_resource *res;
 };
 
+/* video buffer offset info representation */
+struct rvid_buf_offset_info {
+   unsigned num_units;
+   unsigned old_offset;
+   unsigned new_offset;
+};
+
 /* generate an stream handle */
 unsigned si_vid_alloc_stream_handle(void);
 
@@ -58,7 +65,8 @@ void si_vid_destroy_buffer(struct rvid_buffer *buffer);
 
 /* reallocate a buffer, preserving its content */
 bool si_vid_resize_buffer(struct pipe_screen *screen, struct radeon_cmdbuf *cs,
-                          struct rvid_buffer *new_buf, unsigned new_size);
+                          struct rvid_buffer *new_buf, unsigned new_size,
+                          struct rvid_buf_offset_info *buf_ofst_info);
 
 /* clear the buffer with zeros */
 void si_vid_clear_buffer(struct pipe_context *context, struct rvid_buffer *buffer);