radeonsi/vcn: add support for picture crop on JPEG 4.0.3

set the crop region and enable the feature if requested

Signed-off-by: Sathishkumar S <sathishkumar.sundararaju@amd.com>
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Reviewed-by: Leo Liu <leo.liu@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22158>

src/amd/common/ac_vcn_dec.h

@@ -1195,6 +1195,10 @@ struct jpeg_params {
    unsigned dt_luma_top_offset;
    unsigned dt_chroma_top_offset;
    unsigned dt_chromav_top_offset;
+   uint16_t crop_x;
+   uint16_t crop_y;
+   uint16_t crop_width;
+   uint16_t crop_height;
 };
 
 #define RDECODE_VCN1_GPCOM_VCPU_CMD   0x2070c

src/gallium/drivers/radeonsi/radeon_vcn_dec.c

@@ -2958,12 +2958,21 @@ static void radeon_dec_jpeg_end_frame(struct pipe_video_codec *decoder, struct p
                                  struct pipe_picture_desc *picture)
 {
    struct radeon_decoder *dec = (struct radeon_decoder *)decoder;
+   struct pipe_mjpeg_picture_desc *pic = (struct pipe_mjpeg_picture_desc *)picture;
 
    assert(decoder);
 
    if (!dec->bs_ptr)
       return;
 
+   dec->jpg.crop_x = ROUND_DOWN_TO(pic->picture_parameter.crop_x, VL_MACROBLOCK_WIDTH);
+   dec->jpg.crop_y = ROUND_DOWN_TO(pic->picture_parameter.crop_y, VL_MACROBLOCK_HEIGHT);
+   dec->jpg.crop_width = align(pic->picture_parameter.crop_width, VL_MACROBLOCK_WIDTH);
+   dec->jpg.crop_height = align(pic->picture_parameter.crop_height, VL_MACROBLOCK_HEIGHT);
+   if (dec->jpg.crop_x + dec->jpg.crop_width > pic->picture_parameter.picture_width)
+      dec->jpg.crop_width = 0;
+   if (dec->jpg.crop_y + dec->jpg.crop_height > pic->picture_parameter.picture_height)
+      dec->jpg.crop_height = 0;
    dec->send_cmd(dec, target, picture);
    dec->ws->cs_flush(&dec->jcs[dec->cb_idx], PIPE_FLUSH_ASYNC, NULL);
    next_buffer(dec);

src/gallium/drivers/radeonsi/radeon_vcn_dec_jpeg.c

@@ -246,6 +246,7 @@ static void send_cmd_target_direct(struct radeon_decoder *dec, struct pb_buffer
                                    unsigned usage, enum radeon_bo_domain domain)
 {
    uint64_t addr;
+   uint32_t val;
 
    set_reg_jpeg(dec, dec->jpg_reg.jpeg_pitch, COND0, TYPE0, (dec->jpg.dt_pitch >> 4));
    set_reg_jpeg(dec, dec->jpg_reg.jpeg_uv_pitch, COND0, TYPE0, ((dec->jpg.dt_uv_pitch * 2) >> 4));
@@ -276,6 +277,12 @@ static void send_cmd_target_direct(struct radeon_decoder *dec, struct pb_buffer
       set_reg_jpeg(dec, dec->jpg_reg.jpeg_luma_base0_0, COND0, TYPE0, dec->jpg.dt_luma_top_offset);
       set_reg_jpeg(dec, dec->jpg_reg.jpeg_chroma_base0_0, COND0, TYPE0, dec->jpg.dt_chroma_top_offset);
       set_reg_jpeg(dec, dec->jpg_reg.jpeg_chromav_base0_0, COND0, TYPE0, dec->jpg.dt_chromav_top_offset);
+      if (dec->jpg.crop_width && dec->jpg.crop_height) {
+         set_reg_jpeg(dec, vcnipUVD_JPEG_ROI_CROP_POS_START, COND0, TYPE0,
+                      ((dec->jpg.crop_y << 16) | dec->jpg.crop_x));
+         set_reg_jpeg(dec, vcnipUVD_JPEG_ROI_CROP_POS_STRIDE, COND0, TYPE0,
+                      ((dec->jpg.crop_height << 16) | dec->jpg.crop_width));
+      }
    }
    set_reg_jpeg(dec, dec->jpg_reg.jpeg_tier_cntl2, COND0, 0, 0);
 
@@ -288,7 +295,12 @@ static void send_cmd_target_direct(struct radeon_decoder *dec, struct pb_buffer
    set_reg_jpeg(dec, dec->jpg_reg.jpeg_int_en, COND0, TYPE0, 0xFFFFFFFE);
 
    // start engine command
-   set_reg_jpeg(dec, dec->jpg_reg.jpeg_cntl, COND0, TYPE0, 0x6);
+   val = 0x6;
+   if (dec->jpg_reg.version == RDECODE_JPEG_REG_VER_V3) {
+      if (dec->jpg.crop_width && dec->jpg.crop_height)
+         val = val | (0x3 << 24);
+   }
+   set_reg_jpeg(dec, dec->jpg_reg.jpeg_cntl, COND0, TYPE0, val);
 
    // wait for job completion, wait for job JBSI fetch done
    set_reg_jpeg(dec, dec->jpg_reg.jrbc_ib_ref_data, COND0, TYPE0, (dec->jpg.bsd_size >> 2));
@@ -299,6 +311,14 @@ static void send_cmd_target_direct(struct radeon_decoder *dec, struct pb_buffer
    set_reg_jpeg(dec, dec->jpg_reg.jrbc_ib_ref_data, COND0, TYPE0, 0xFFFFFFFF);
    set_reg_jpeg(dec, dec->jpg_reg.jpeg_outbuf_wptr, COND3, TYPE3, 0x00000001);
 
+   if (dec->jpg_reg.version == RDECODE_JPEG_REG_VER_V3) {
+      val = 0;
+      if (dec->jpg.crop_width && dec->jpg.crop_height)
+         val = val | (0x1 << 19);
+      set_reg_jpeg(dec, dec->jpg_reg.jrbc_ib_ref_data, COND0, TYPE0, 0);
+      set_reg_jpeg(dec, vcnipUVD_JPEG_INT_STAT, COND3, TYPE3, val);
+   }
+
    // stop engine
    set_reg_jpeg(dec, dec->jpg_reg.jpeg_cntl, COND0, TYPE0, 0x4);
 }

src/gallium/include/pipe/p_video_state.h

@@ -793,6 +793,10 @@ struct pipe_mjpeg_picture_desc
       } components[255];
 
       uint8_t num_components;
+      uint16_t crop_x;
+      uint16_t crop_y;
+      uint16_t crop_width;
+      uint16_t crop_height;
    } picture_parameter;
 
    struct