[Libva] [PATCH 03/14] Encoding: VME shader reads mbmv_cost from cost_table surface instead of constant buffer on Haswell
Pengfei Qu
Pengfei.Qu at intel.com
Thu Jun 30 02:33:00 UTC 2016
From: Zhao Yakui <yakui.zhao at intel.com>
This is to do the prepartion of enhanced features.
Signed-off-by: Zhao Yakui <yakui.zhao at intel.com>
Signed-off-by: pjl <ceciliapeng at intel.com>
Signed-off-by: Pengfei Qu <Pengfei.Qu at intel.com>
---
src/gen75_vme.c | 14 +++++++++++++-
src/shaders/vme/inter_bframe_haswell.asm | 24 +++++++++++++++++++++++-
src/shaders/vme/inter_bframe_haswell.g75b | 23 ++++++++++++++---------
src/shaders/vme/inter_frame_haswell.asm | 24 ++++++++++++++++++++++--
src/shaders/vme/inter_frame_haswell.g75b | 11 ++++++++---
src/shaders/vme/intra_frame_haswell.asm | 23 ++++++++++++++++++++++-
src/shaders/vme/intra_frame_haswell.g75b | 7 ++++++-
src/shaders/vme/vme75.inc | 4 ++++
8 files changed, 112 insertions(+), 18 deletions(-)
diff --git a/src/gen75_vme.c b/src/gen75_vme.c
index 8434e6f..b2e7812 100644
--- a/src/gen75_vme.c
+++ b/src/gen75_vme.c
@@ -491,6 +491,16 @@ gen75_vme_fill_vme_batchbuffer(VADriverContextP ctx,
int mb_x = 0, mb_y = 0;
int i, s;
unsigned int *command_ptr;
+ struct gen6_mfc_context *mfc_context = encoder_context->mfc_context;
+ VAEncPictureParameterBufferH264 *pic_param = (VAEncPictureParameterBufferH264 *)encode_state->pic_param_ext->buffer;
+ VAEncSliceParameterBufferH264 *slice_param = (VAEncSliceParameterBufferH264 *)encode_state->slice_params_ext[0]->buffer;
+ int qp;
+ int slice_type = intel_avc_enc_slice_type_fixup(slice_param->slice_type);
+
+ if (encoder_context->rate_control_mode == VA_RC_CQP)
+ qp = pic_param->pic_init_qp + slice_param->slice_qp_delta;
+ else
+ qp = mfc_context->bit_rate_control_context[slice_type].QpPrimeY;
dri_bo_map(vme_context->vme_batchbuffer.bo, 1);
command_ptr = vme_context->vme_batchbuffer.bo->virtual;
@@ -528,7 +538,7 @@ gen75_vme_fill_vme_batchbuffer(VADriverContextP ctx,
if ((i == mb_width) && slice_mb_x) {
mb_intra_ub &= ~(INTRA_PRED_AVAIL_FLAG_D);
}
- *command_ptr++ = (CMD_MEDIA_OBJECT | (8 - 2));
+ *command_ptr++ = (CMD_MEDIA_OBJECT | (9 - 2));
*command_ptr++ = kernel;
*command_ptr++ = 0;
*command_ptr++ = 0;
@@ -538,6 +548,8 @@ gen75_vme_fill_vme_batchbuffer(VADriverContextP ctx,
/*inline data */
*command_ptr++ = (mb_width << 16 | mb_y << 8 | mb_x);
*command_ptr++ = ((encoder_context->quality_level << 24) | (1 << 16) | transform_8x8_mode_flag | (mb_intra_ub << 8));
+ /* qp occupies one byte */
+ *command_ptr++ = qp;
i += 1;
}
diff --git a/src/shaders/vme/inter_bframe_haswell.asm b/src/shaders/vme/inter_bframe_haswell.asm
index ff91487..dbc15c5 100644
--- a/src/shaders/vme/inter_bframe_haswell.asm
+++ b/src/shaders/vme/inter_bframe_haswell.asm
@@ -485,9 +485,31 @@ __mb_hwdep_end:
asr (4) mb_ref_win.0<1>:w mb_mvp_ref.0<4,4,1>:w 2:w {align1};
add (4) mb_ref_win.8<1>:w mb_ref_win.0<4,4,1>:w 3:w {align1};
and (4) mb_ref_win.16<1>:uw mb_ref_win.8<4,4,1>:uw 0xFFFC:uw {align1};
+
+mov (8) msg_reg0.0<1>:ud 0:ud {align1};
+mov (1) msg_reg0.20<1>:UB thread_id_ub {align1}; /* dispatch id */
+mov (1) tmp_reg0.0<1>:UD qp_ub<0,1,0>:ub {align1};
+mul (1) msg_reg0.8<1>:ud tmp_reg0.0<1>:ud 2:ud {align1};
+
+send (16)
+ msg_ind
+ vme_cost_wb.0<1>:ud
+ null
+ data_port(
+ OBR_CACHE_TYPE,
+ OBR_MESSAGE_TYPE,
+ OBR_CONTROL_2,
+ BIND_IDX_COST,
+ OBR_WRITE_COMMIT_CATEGORY,
+ OBR_HEADER_PRESENT
+ )
+ mlen 1
+ rlen 1
+ {align1};
+
/* m2, get the MV/Mb cost passed from constant buffer when
spawning thread by MEDIA_OBJECT */
-mov (8) vme_m2<1>:UD r1.0<8,8,1>:UD {align1};
+mov (8) vme_m2<1>:UD vme_cost_wb.0<8,8,1>:UD {align1};
mov (8) vme_msg_2<1>:UD vme_m2.0<8,8,1>:UD {align1};
diff --git a/src/shaders/vme/inter_bframe_haswell.g75b b/src/shaders/vme/inter_bframe_haswell.g75b
index cabef20..113eac0 100644
--- a/src/shaders/vme/inter_bframe_haswell.g75b
+++ b/src/shaders/vme/inter_bframe_haswell.g75b
@@ -64,7 +64,7 @@
{ 0x00000001, 0x2fa40021, 0x00000b80, 0x00000000 },
{ 0x00000001, 0x2fa80061, 0x00000000, 0x00000001 },
{ 0x00000040, 0x2fe00c01, 0x00001400, 0x00000020 },
- { 0x00000020, 0x34001c00, 0x00001400, 0x000013e0 },
+ { 0x00000020, 0x34001c00, 0x00001400, 0x00001430 },
{ 0x00000001, 0x2aa00129, 0x00000fe4, 0x00000000 },
{ 0x01000010, 0x20002d28, 0x00000aa0, 0x00000000 },
{ 0x00010001, 0x2af001e9, 0x00000000, 0x00010001 },
@@ -100,7 +100,7 @@
{ 0x00000001, 0x2fa40021, 0x00000b80, 0x00000000 },
{ 0x00000001, 0x2fa80061, 0x00000000, 0x00000002 },
{ 0x00000040, 0x2fe00c01, 0x00001400, 0x00000020 },
- { 0x00000020, 0x34001c00, 0x00001400, 0x000011a0 },
+ { 0x00000020, 0x34001c00, 0x00001400, 0x000011f0 },
{ 0x00000001, 0x2aa00129, 0x00000fe4, 0x00000000 },
{ 0x01000010, 0x20002d28, 0x00000aa0, 0x00000000 },
{ 0x00010001, 0x2b1001e9, 0x00000000, 0x00010001 },
@@ -135,7 +135,7 @@
{ 0x00000001, 0x2fa40021, 0x00000b80, 0x00000000 },
{ 0x00000001, 0x2fa80061, 0x00000000, 0x00000002 },
{ 0x00000040, 0x2fe00c01, 0x00001400, 0x00000020 },
- { 0x00000020, 0x34001c00, 0x00001400, 0x00000f70 },
+ { 0x00000020, 0x34001c00, 0x00001400, 0x00000fc0 },
{ 0x00000001, 0x2aa00129, 0x00000fe4, 0x00000000 },
{ 0x01000010, 0x20002d28, 0x00000aa0, 0x00000000 },
{ 0x00010001, 0x2b3001e9, 0x00000000, 0x00010001 },
@@ -170,7 +170,7 @@
{ 0x00000001, 0x2fa40021, 0x00000b80, 0x00000000 },
{ 0x00000001, 0x2fa80061, 0x00000000, 0x00000003 },
{ 0x00000040, 0x2fe00c01, 0x00001400, 0x00000020 },
- { 0x00000020, 0x34001c00, 0x00001400, 0x00000d40 },
+ { 0x00000020, 0x34001c00, 0x00001400, 0x00000d90 },
{ 0x00000001, 0x2aa00129, 0x00000fe4, 0x00000000 },
{ 0x00000001, 0x2b3201ed, 0x00000000, 0x00010001 },
{ 0x01000010, 0x20002d28, 0x00000aa0, 0x00000000 },
@@ -211,13 +211,13 @@
{ 0x00000001, 0x2fa401ad, 0x00000b04, 0x00000000 },
{ 0x00000001, 0x2fa801ad, 0x00000b24, 0x00000000 },
{ 0x00000040, 0x2fe00c01, 0x00001400, 0x00000020 },
- { 0x00000020, 0x34001c00, 0x00001400, 0x000009b0 },
+ { 0x00000020, 0x34001c00, 0x00001400, 0x00000a00 },
{ 0x00000001, 0x2ac001ad, 0x00000fe4, 0x00000000 },
{ 0x00000001, 0x2fa001ad, 0x00000ae6, 0x00000000 },
{ 0x00000001, 0x2fa401ad, 0x00000b06, 0x00000000 },
{ 0x00000001, 0x2fa801ad, 0x00000b26, 0x00000000 },
{ 0x00000040, 0x2fe00c01, 0x00001400, 0x00000020 },
- { 0x00000020, 0x34001c00, 0x00001400, 0x00000950 },
+ { 0x00000020, 0x34001c00, 0x00001400, 0x000009a0 },
{ 0x00000001, 0x2ac201ad, 0x00000fe4, 0x00000000 },
{ 0x00600001, 0x24000061, 0x00000000, 0x00000000 },
{ 0x01000010, 0x20003da4, 0x00200af6, 0x00000000 },
@@ -236,18 +236,23 @@
{ 0x00000001, 0x2fa401ad, 0x00000b08, 0x00000000 },
{ 0x00000001, 0x2fa801ad, 0x00000b28, 0x00000000 },
{ 0x00000040, 0x2fe00c01, 0x00001400, 0x00000020 },
- { 0x00000020, 0x34001c00, 0x00001400, 0x00000820 },
+ { 0x00000020, 0x34001c00, 0x00001400, 0x00000870 },
{ 0x00000001, 0x2ac401ad, 0x00000fe4, 0x00000000 },
{ 0x00000001, 0x2fa001ad, 0x00000aea, 0x00000000 },
{ 0x00000001, 0x2fa401ad, 0x00000b0a, 0x00000000 },
{ 0x00000001, 0x2fa801ad, 0x00000b2a, 0x00000000 },
{ 0x00000040, 0x2fe00c01, 0x00001400, 0x00000020 },
- { 0x00000020, 0x34001c00, 0x00001400, 0x000007c0 },
+ { 0x00000020, 0x34001c00, 0x00001400, 0x00000810 },
{ 0x00000001, 0x2ac601ad, 0x00000fe4, 0x00000000 },
{ 0x0040000c, 0x2a803dad, 0x00690ac0, 0x00020002 },
{ 0x00400040, 0x2a883dad, 0x00690a80, 0x00030003 },
{ 0x00400005, 0x2a902d29, 0x00690a88, 0xfffcfffc },
- { 0x00600001, 0x25600021, 0x008d0020, 0x00000000 },
+ { 0x00600001, 0x28000061, 0x00000000, 0x00000000 },
+ { 0x00000001, 0x28140231, 0x00000014, 0x00000000 },
+ { 0x00000001, 0x24000221, 0x000000a8, 0x00000000 },
+ { 0x00000041, 0x28080c21, 0x00200400, 0x00000002 },
+ { 0x0a800031, 0x25801ca1, 0x00000800, 0x02180208 },
+ { 0x00600001, 0x25600021, 0x008d0580, 0x00000000 },
{ 0x00600001, 0x28400021, 0x008d0560, 0x00000000 },
{ 0x00600001, 0x28600061, 0x00000000, 0x00000000 },
{ 0x00000001, 0x23800061, 0x00000000, 0x00000000 },
diff --git a/src/shaders/vme/inter_frame_haswell.asm b/src/shaders/vme/inter_frame_haswell.asm
index 8a829f6..7f08d3e 100644
--- a/src/shaders/vme/inter_frame_haswell.asm
+++ b/src/shaders/vme/inter_frame_haswell.asm
@@ -387,10 +387,30 @@ __mb_hwdep_end:
asr (2) mb_ref_win.0<1>:w mb_mvp_ref.0<2,2,1>:w 2:w {align1};
add (2) mb_ref_win.8<1>:w mb_ref_win.0<2,2,1>:w 3:w {align1};
and (2) mb_ref_win.16<1>:uw mb_ref_win.8<2,2,1>:uw 0xFFFC:uw {align1};
-
+
+mov (8) msg_reg0.0<1>:ud 0:ud {align1};
+mov (1) msg_reg0.20<1>:UB thread_id_ub {align1}; /* dispatch id */
+mov (1) tmp_reg0.0<1>:UD qp_ub<0,1,0>:ub {align1};
+mul (1) msg_reg0.8<1>:ud tmp_reg0.0<1>:ud 2:ud {align1};
+
+send (16)
+ msg_ind
+ vme_cost_wb.0<1>:ud
+ null
+ data_port(
+ OBR_CACHE_TYPE,
+ OBR_MESSAGE_TYPE,
+ OBR_CONTROL_2,
+ BIND_IDX_COST,
+ OBR_WRITE_COMMIT_CATEGORY,
+ OBR_HEADER_PRESENT
+ )
+ mlen 1
+ rlen 1
+ {align1};
/* m2, get the MV/Mb cost passed from constant buffer when
spawning thread by MEDIA_OBJECT */
-mov (8) vme_m2<1>:UD r1.0<8,8,1>:UD {align1};
+mov (8) vme_m2<1>:UD vme_cost_wb.0<8,8,1>:UD {align1};
mov (8) vme_msg_2<1>:UD vme_m2.0<8,8,1>:UD {align1};
diff --git a/src/shaders/vme/inter_frame_haswell.g75b b/src/shaders/vme/inter_frame_haswell.g75b
index 1a60c51..5aa520e 100644
--- a/src/shaders/vme/inter_frame_haswell.g75b
+++ b/src/shaders/vme/inter_frame_haswell.g75b
@@ -145,18 +145,23 @@
{ 0x00000001, 0x2fa401ad, 0x00000b04, 0x00000000 },
{ 0x00000001, 0x2fa801ad, 0x00000b24, 0x00000000 },
{ 0x00000040, 0x2fe00c01, 0x00001400, 0x00000020 },
- { 0x00000020, 0x34001c00, 0x00001400, 0x00000930 },
+ { 0x00000020, 0x34001c00, 0x00001400, 0x00000980 },
{ 0x00000001, 0x2ac001ad, 0x00000fe4, 0x00000000 },
{ 0x00000001, 0x2fa001ad, 0x00000ae6, 0x00000000 },
{ 0x00000001, 0x2fa401ad, 0x00000b06, 0x00000000 },
{ 0x00000001, 0x2fa801ad, 0x00000b26, 0x00000000 },
{ 0x00000040, 0x2fe00c01, 0x00001400, 0x00000020 },
- { 0x00000020, 0x34001c00, 0x00001400, 0x000008d0 },
+ { 0x00000020, 0x34001c00, 0x00001400, 0x00000920 },
{ 0x00000001, 0x2ac201ad, 0x00000fe4, 0x00000000 },
{ 0x0020000c, 0x2a803dad, 0x00450ac0, 0x00020002 },
{ 0x00200040, 0x2a883dad, 0x00450a80, 0x00030003 },
{ 0x00200005, 0x2a902d29, 0x00450a88, 0xfffcfffc },
- { 0x00600001, 0x25600021, 0x008d0020, 0x00000000 },
+ { 0x00600001, 0x28000061, 0x00000000, 0x00000000 },
+ { 0x00000001, 0x28140231, 0x00000014, 0x00000000 },
+ { 0x00000001, 0x24000221, 0x000000a8, 0x00000000 },
+ { 0x00000041, 0x28080c21, 0x00200400, 0x00000002 },
+ { 0x0a800031, 0x25801ca1, 0x00000800, 0x02180208 },
+ { 0x00600001, 0x25600021, 0x008d0580, 0x00000000 },
{ 0x00600001, 0x28400021, 0x008d0560, 0x00000000 },
{ 0x00600001, 0x28600061, 0x00000000, 0x00000000 },
{ 0x00000001, 0x23800061, 0x00000000, 0x00000000 },
diff --git a/src/shaders/vme/intra_frame_haswell.asm b/src/shaders/vme/intra_frame_haswell.asm
index c4cb76c..ac4a6ba 100644
--- a/src/shaders/vme/intra_frame_haswell.asm
+++ b/src/shaders/vme/intra_frame_haswell.asm
@@ -89,9 +89,30 @@ mov (1) read1_header.8<1>:UD BLOCK_8X4 {align1};
mov (8) msg_reg0.0<1>:UD read1_header.0<8,8,1>:UD {align1};
send (8) msg_ind CHROMA_COL<1>:UB null read(BIND_IDX_CBCR, 0, 0, 4) mlen 1 rlen 1 {align1};
+mov (8) msg_reg0.0<1>:ud 0:ud {align1};
+mov (1) msg_reg0.20<1>:UB thread_id_ub {align1}; /* dispatch id */
+mov (1) tmp_reg0.0<1>:UD qp_ub<0,1,0>:ub {align1};
+mul (1) msg_reg0.8<1>:ud tmp_reg0.0<1>:ud 2:ud {align1};
+
+send (16)
+ msg_ind
+ vme_cost_wb.0<1>:ud
+ null
+ data_port(
+ OBR_CACHE_TYPE,
+ OBR_MESSAGE_TYPE,
+ OBR_CONTROL_2,
+ BIND_IDX_COST,
+ OBR_WRITE_COMMIT_CATEGORY,
+ OBR_HEADER_PRESENT
+ )
+ mlen 1
+ rlen 1
+ {align1};
+
/* m2, get the MV/Mb cost passed by constant buffer
when creating EU thread by MEDIA_OBJECT */
-mov (8) vme_msg_2<1>:UD r1.0<8,8,1>:UD {align1};
+mov (8) vme_msg_2<1>:UD vme_cost_wb<8,8,1>:UD {align1};
/* m3 */
mov (8) vme_msg_3<1>:UD 0x0:UD {align1};
diff --git a/src/shaders/vme/intra_frame_haswell.g75b b/src/shaders/vme/intra_frame_haswell.g75b
index 4bb8ad9..88866d3 100644
--- a/src/shaders/vme/intra_frame_haswell.g75b
+++ b/src/shaders/vme/intra_frame_haswell.g75b
@@ -33,7 +33,12 @@
{ 0x00000001, 0x242800e1, 0x00000000, 0x00070003 },
{ 0x00600001, 0x28000021, 0x008d0420, 0x00000000 },
{ 0x04600031, 0x26201cb1, 0x00000800, 0x02190006 },
- { 0x00600001, 0x28400021, 0x008d0020, 0x00000000 },
+ { 0x00600001, 0x28000061, 0x00000000, 0x00000000 },
+ { 0x00000001, 0x28140231, 0x00000014, 0x00000000 },
+ { 0x00000001, 0x24000221, 0x000000a8, 0x00000000 },
+ { 0x00000041, 0x28080c21, 0x00200400, 0x00000002 },
+ { 0x0a800031, 0x25801ca1, 0x00000800, 0x02180208 },
+ { 0x00600001, 0x28400021, 0x008d0580, 0x00000000 },
{ 0x00600001, 0x28600061, 0x00000000, 0x00000000 },
{ 0x00000001, 0x23800061, 0x00000000, 0x00000000 },
{ 0x00000005, 0x23840c21, 0x00000384, 0xff000000 },
diff --git a/src/shaders/vme/vme75.inc b/src/shaders/vme/vme75.inc
index 75c461e..fa7aa7e 100644
--- a/src/shaders/vme/vme75.inc
+++ b/src/shaders/vme/vme75.inc
@@ -56,6 +56,7 @@ define(`BIND_IDX_VME_REF0', `1')
define(`BIND_IDX_VME_REF1', `2')
define(`BIND_IDX_OUTPUT', `3')
define(`BIND_IDX_INEP', `4')
+define(`BIND_IDX_COST', `8')
define(`SUB_PEL_MODE_INTEGER', `0x00000000')
define(`SUB_PEL_MODE_HALF', `0x00001000')
@@ -153,6 +154,8 @@ define(`input_mb_intra_ub', `inline_reg0.5')
define(`num_macroblocks', `inline_reg0.6')
define(`quality_level_ub', `inline_reg0.7')
+define(`qp_ub', `inline_reg0.8')
+
/*
* GRF 6~11 -- reserved
*/
@@ -235,6 +238,7 @@ define(`tmp_uw6', `tmp_rega.12')
define(`tmp_uw7', `tmp_rega.14')
define(`vme_m2', `r43')
+define(`vme_cost_wb', `r44')
/*
* MRF registers
*/
--
2.7.4
More information about the Libva
mailing list