radeonsi/vcn: support unified queue in vcn4

- use unified queue only in vcn4
 - implement signature and engine-info ib headers
   in vcn4
 - implemented unified queue functions

Reviewed-by: Boyuan Zhang <Boyuan.Zhang@amd.com>
Reviewed-by: Mihai Preda <mhpreda@gmail.com>
Signed-off-by: Ruijing Dong <ruijing.dong@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16911>

src/gallium/drivers/radeonsi/radeon_vcn_dec.c

@@ -28,7 +28,6 @@
 #include "radeon_vcn_dec.h"
 
 #include "pipe/p_video_codec.h"
-#include "radeon_video.h"
 #include "radeonsi/si_pipe.h"
 #include "util/u_memory.h"
 #include "util/u_video.h"
@@ -385,7 +384,7 @@ static rvcn_dec_message_hevc_t get_h265_msg(struct radeon_decoder *dec,
    result.curr_poc = pic->CurrPicOrderCntVal;
 
    for (i = 0; i < ARRAY_SIZE(dec->render_pic_list); i++) {
-      for (j = 0; 
+      for (j = 0;
            (pic->ref[j] != NULL) && (j < ARRAY_SIZE(dec->render_pic_list));
            j++) {
          if (dec->render_pic_list[i] == pic->ref[j])
@@ -2124,9 +2123,18 @@ static void rvcn_dec_message_feedback(struct radeon_decoder *dec)
    header->num_buffers = 0;
 }
 
+static void rvcn_dec_sq_tail(struct radeon_decoder *dec)
+{
+   if (dec->vcn_dec_sw_ring == false)
+      return;
+
+   rvcn_sq_tail(&dec->cs, &dec->sq);
+}
 /* flush IB to the hardware */
 static int flush(struct radeon_decoder *dec, unsigned flags)
 {
+   rvcn_dec_sq_tail(dec);
+
    return dec->ws->cs_flush(&dec->cs, flags, NULL);
 }
 
@@ -2155,6 +2163,7 @@ static void send_cmd(struct radeon_decoder *dec, unsigned cmd, struct pb_buffer
    }
 
    if (!dec->cs.current.cdw) {
+      rvcn_sq_header(&dec->cs, &dec->sq, false);
       rvcn_decode_ib_package_t *ib_header =
          (rvcn_decode_ib_package_t *)&(dec->cs.current.buf[dec->cs.current.cdw]);
 
@@ -2770,8 +2779,13 @@ struct pipe_video_codec *radeon_create_decoder(struct pipe_context *context,
    dec->ws = ws;
 
    if (u_reduce_video_profile(templ->profile) != PIPE_VIDEO_FORMAT_JPEG &&
-       sctx->gfx_level >= GFX11)
+       sctx->gfx_level >= GFX11) {
       dec->vcn_dec_sw_ring = true;
+      ring = AMD_IP_VCN_UNIFIED;
+   }
+
+   dec->sq.ib_total_size_in_dw = NULL;
+   dec->sq.ib_checksum = NULL;
 
    if (!ws->cs_create(&dec->cs, sctx->ctx, ring, NULL, NULL, false)) {
       RVID_ERR("Can't get command submission context.\n");

src/gallium/drivers/radeonsi/radeon_vcn_enc.h

@@ -595,6 +595,10 @@ struct radeon_encoder {
    void (*input_format)(struct radeon_encoder *enc);
    void (*output_format)(struct radeon_encoder *enc);
    void (*efc_params)(struct radeon_encoder *enc);
+   /* mq is used for preversing multiple queue ibs */
+   void (*mq_begin)(struct radeon_encoder *enc);
+   void (*mq_encode)(struct radeon_encoder *enc);
+   void (*mq_destroy)(struct radeon_encoder *enc);
 
    unsigned stream_handle;
 

src/gallium/drivers/radeonsi/radeon_vcn_enc_4_0.c

@@ -32,12 +32,32 @@
 #include "util/u_video.h"
 
 #include "si_pipe.h"
-#include "radeon_video.h"
 #include "radeon_vcn_enc.h"
 
 #define RENCODE_FW_INTERFACE_MAJOR_VERSION   1
 #define RENCODE_FW_INTERFACE_MINOR_VERSION   0
 
+static void radeon_enc_sq_begin(struct radeon_encoder *enc)
+{
+   rvcn_sq_header(&enc->cs, &enc->sq, true);
+   enc->mq_begin(enc);
+   rvcn_sq_tail(&enc->cs, &enc->sq);
+}
+
+static void radeon_enc_sq_encode(struct radeon_encoder *enc)
+{
+   rvcn_sq_header(&enc->cs, &enc->sq, true);
+   enc->mq_encode(enc);
+   rvcn_sq_tail(&enc->cs, &enc->sq);
+}
+
+static void radeon_enc_sq_destroy(struct radeon_encoder *enc)
+{
+   rvcn_sq_header(&enc->cs, &enc->sq, true);
+   enc->mq_destroy(enc);
+   rvcn_sq_tail(&enc->cs, &enc->sq);
+}
+
 static void radeon_enc_session_init(struct radeon_encoder *enc)
 {
    if (u_reduce_video_profile(enc->base.profile) == PIPE_VIDEO_FORMAT_MPEG4_AVC) {
@@ -129,6 +149,12 @@ void radeon_enc_4_0_init(struct radeon_encoder *enc)
 
    enc->session_init = radeon_enc_session_init;
    enc->ctx = radeon_enc_ctx;
+   enc->mq_begin = enc->begin;
+   enc->mq_encode = enc->encode;
+   enc->mq_destroy = enc->destroy;
+   enc->begin = radeon_enc_sq_begin;
+   enc->encode = radeon_enc_sq_encode;
+   enc->destroy = radeon_enc_sq_destroy;
 
    enc->enc_pic.session_info.interface_version =
       ((RENCODE_FW_INTERFACE_MAJOR_VERSION << RENCODE_IF_MAJOR_VERSION_SHIFT) |