radeonsi/vcn: enable swizzle mode in encoding ref frames.

swizzle mode in ref frames could potentially
improve encoding performance, the main reason
is just because linear mapping is the worst mode
for reference frames comparing to block level
mapping.

Reviewed-by: Boyuan Zhang <Boyuan.Zhang@amd.com>
Signed-off-by: Ruijing Dong <ruijing.dong@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22585>

diff --git a/src/gallium/drivers/radeonsi/radeon_vcn_enc.h b/src/gallium/drivers/radeonsi/radeon_vcn_enc.h
index 5b0646d..d032df5 100644 (file)
--- a/src/gallium/drivers/radeonsi/radeon_vcn_enc.h
+++ b/src/gallium/drivers/radeonsi/radeon_vcn_enc.h
@@ -162,6 +162,8 @@
 
 #define RENCODE_REC_SWIZZLE_MODE_LINEAR                                             0
 #define RENCODE_REC_SWIZZLE_MODE_256B_S                                             1
+#define RENCODE_REC_SWIZZLE_MODE_256B_D                                             2
+#define RENCODE_REC_SWIZZLE_MODE_8x8_1D_THIN_12_24BPP                               0x10000001
 
 #define RENCODE_VIDEO_BITSTREAM_BUFFER_MODE_LINEAR                                  0
 #define RENCODE_VIDEO_BITSTREAM_BUFFER_MODE_CIRCULAR                                1

diff --git a/src/gallium/drivers/radeonsi/radeon_vcn_enc_2_0.c b/src/gallium/drivers/radeonsi/radeon_vcn_enc_2_0.c
index 3d285fd..202e59d 100644 (file)
--- a/src/gallium/drivers/radeonsi/radeon_vcn_enc_2_0.c
+++ b/src/gallium/drivers/radeonsi/radeon_vcn_enc_2_0.c
@@ -461,9 +461,18 @@ static void radeon_enc_output_format(struct radeon_encoder *enc)
    RADEON_ENC_END();
 }
 
+static uint32_t radeon_enc_ref_swizzle_mode(struct radeon_encoder *enc)
+{
+   /* return RENCODE_REC_SWIZZLE_MODE_LINEAR; for debugging purpose */
+   if (enc->enc_pic.bit_depth_luma_minus8 != 0)
+      return RENCODE_REC_SWIZZLE_MODE_8x8_1D_THIN_12_24BPP;
+   else
+      return RENCODE_REC_SWIZZLE_MODE_256B_S;
+}
+
 static void radeon_enc_ctx(struct radeon_encoder *enc)
 {
-   enc->enc_pic.ctx_buf.swizzle_mode = 0;
+   enc->enc_pic.ctx_buf.swizzle_mode = radeon_enc_ref_swizzle_mode(enc);
    enc->enc_pic.ctx_buf.two_pass_search_center_map_offset = 0;
 
    RADEON_ENC_BEGIN(enc->cmd.ctx);

diff --git a/src/gallium/drivers/radeonsi/radeon_vcn_enc_3_0.c b/src/gallium/drivers/radeonsi/radeon_vcn_enc_3_0.c
index cd78079..e2f7fbd 100644 (file)
--- a/src/gallium/drivers/radeonsi/radeon_vcn_enc_3_0.c
+++ b/src/gallium/drivers/radeonsi/radeon_vcn_enc_3_0.c
@@ -169,9 +169,18 @@ static void radeon_enc_nalu_pps_hevc(struct radeon_encoder *enc)
    RADEON_ENC_END();
 }
 
+static uint32_t radeon_enc_ref_swizzle_mode(struct radeon_encoder *enc)
+{
+   /* return RENCODE_REC_SWIZZLE_MODE_LINEAR; for debugging purpose */
+   if (enc->enc_pic.bit_depth_luma_minus8 != 0)
+      return RENCODE_REC_SWIZZLE_MODE_8x8_1D_THIN_12_24BPP;
+   else
+      return RENCODE_REC_SWIZZLE_MODE_256B_S;
+}
+
 static void radeon_enc_ctx(struct radeon_encoder *enc)
 {
-   enc->enc_pic.ctx_buf.swizzle_mode = 0;
+   enc->enc_pic.ctx_buf.swizzle_mode = radeon_enc_ref_swizzle_mode(enc);
    enc->enc_pic.ctx_buf.two_pass_search_center_map_offset = 0;
 
    RADEON_ENC_BEGIN(enc->cmd.ctx);

diff --git a/src/gallium/drivers/radeonsi/radeon_vcn_enc_4_0.c b/src/gallium/drivers/radeonsi/radeon_vcn_enc_4_0.c
index 4015d0a..52b5b86 100644 (file)
--- a/src/gallium/drivers/radeonsi/radeon_vcn_enc_4_0.c
+++ b/src/gallium/drivers/radeonsi/radeon_vcn_enc_4_0.c
@@ -60,9 +60,18 @@ static void radeon_enc_sq_destroy(struct radeon_encoder *enc)
    rvcn_sq_tail(&enc->cs, &enc->sq);
 }
 
+static uint32_t radeon_enc_ref_swizzle_mode(struct radeon_encoder *enc)
+{
+   /* return RENCODE_REC_SWIZZLE_MODE_LINEAR; for debugging purpose */
+   if (enc->enc_pic.bit_depth_luma_minus8 != 0)
+      return RENCODE_REC_SWIZZLE_MODE_8x8_1D_THIN_12_24BPP;
+   else
+      return RENCODE_REC_SWIZZLE_MODE_256B_D;
+}
+
 static void radeon_enc_ctx(struct radeon_encoder *enc)
 {
-   enc->enc_pic.ctx_buf.swizzle_mode = 0;
+   enc->enc_pic.ctx_buf.swizzle_mode = radeon_enc_ref_swizzle_mode(enc);
    enc->enc_pic.ctx_buf.two_pass_search_center_map_offset = 0;
    enc->enc_pic.ctx_buf.colloc_buffer_offset = enc->dpb_size;