[Libva] [PATCH 4/7] Enabling SNB avc encoding configuration
Zhao, Yakui
yakui.zhao at intel.com
Mon Jul 28 17:49:37 PDT 2014
On Mon, 2014-07-28 at 00:54 -0600, Li, Zhong wrote:
> On 07/28/2014 02:17 PM, Zhao, Yakui wrote:
> > On Thu, 2014-07-24 at 03:47 -0600, Zhong Li wrote:
> >> Signed-off-by: Zhong Li <zhong.li at intel.com>
> >> ---
> >> src/gen6_vme.c | 50 ++++++++++++++++++++++++++-------------
> >> src/shaders/vme/inter_frame.asm | 25 ++++++++++++++++++--
> >> src/shaders/vme/inter_frame.g6b | 9 +++++++
> >> src/shaders/vme/vme.inc | 5 ++++
> >> 4 files changed, 71 insertions(+), 18 deletions(-)
> >>
> >> diff --git a/src/gen6_vme.c b/src/gen6_vme.c
> >> index 8f0006f..2e02591 100644
> >> --- a/src/gen6_vme.c
> >> +++ b/src/gen6_vme.c
> >> @@ -384,21 +384,38 @@ static VAStatus gen6_vme_vme_state_setup(VADriverContextP ctx,
> >> dri_bo_map(vme_context->vme_state.bo, 1);
> >> assert(vme_context->vme_state.bo->virtual);
> >> vme_state_message = (unsigned int *)vme_context->vme_state.bo->virtual;
> >> -
> >> - vme_state_message[0] = 0x01010101;
> >> - vme_state_message[1] = 0x10010101;
> >> - vme_state_message[2] = 0x0F0F0F0F;
> >> - vme_state_message[3] = 0x100F0F0F;
> >> - vme_state_message[4] = 0x01010101;
> >> - vme_state_message[5] = 0x10010101;
> >> - vme_state_message[6] = 0x0F0F0F0F;
> >> - vme_state_message[7] = 0x100F0F0F;
> >> - vme_state_message[8] = 0x01010101;
> >> - vme_state_message[9] = 0x10010101;
> >> - vme_state_message[10] = 0x0F0F0F0F;
> >> - vme_state_message[11] = 0x000F0F0F;
> >> - vme_state_message[12] = 0x00;
> >> - vme_state_message[13] = 0x00;
> >> +
> >> + if (encoder_context->quality_level != ENCODER_LOW_QUALITY) {
> >> + vme_state_message[0] = 0x01010101;
> >> + vme_state_message[1] = 0x10010101;
> >> + vme_state_message[2] = 0x0F0F0F0F;
> >> + vme_state_message[3] = 0x100F0F0F;
> >> + vme_state_message[4] = 0x01010101;
> >> + vme_state_message[5] = 0x10010101;
> >> + vme_state_message[6] = 0x0F0F0F0F;
> >> + vme_state_message[7] = 0x100F0F0F;
> >> + vme_state_message[8] = 0x01010101;
> >> + vme_state_message[9] = 0x10010101;
> >> + vme_state_message[10] = 0x0F0F0F0F;
> >> + vme_state_message[11] = 0x000F0F0F;
> >> + vme_state_message[12] = 0x00;
> >> + vme_state_message[13] = 0x00;
> >> + } else {
> >> + vme_state_message[0] = 0x10010101;
> >> + vme_state_message[1] = 0x100F0F0F;
> >> + vme_state_message[2] = 0x10010101;
> >> + vme_state_message[3] = 0x000F0F0F;
> >> + vme_state_message[4] = 0;
> >> + vme_state_message[5] = 0;
> >> + vme_state_message[6] = 0;
> >> + vme_state_message[7] = 0;
> >> + vme_state_message[8] = 0;
> >> + vme_state_message[9] = 0;
> >> + vme_state_message[10] = 0;
> >> + vme_state_message[11] = 0;
> >> + vme_state_message[12] = 0;
> >> + vme_state_message[13] = 0;
> >> + }
> >>
> >> vme_state_message[14] = 0x4a4a;
> >> vme_state_message[15] = 0x0;
> >> @@ -452,7 +469,7 @@ gen6_vme_fill_vme_batchbuffer(VADriverContextP ctx,
> >> number_mb_cmds = slice_mb_number - i;
> >> }
> >>
> >> - *command_ptr++ = (CMD_MEDIA_OBJECT | (8 - 2));
> >> + *command_ptr++ = (CMD_MEDIA_OBJECT | (9 - 2));
> >> *command_ptr++ = kernel;
> >> *command_ptr++ = 0;
> >> *command_ptr++ = 0;
> >> @@ -462,6 +479,7 @@ gen6_vme_fill_vme_batchbuffer(VADriverContextP ctx,
> >> /*inline data */
> >> *command_ptr++ = (mb_width << 16 | mb_y << 8 | mb_x);
> >> *command_ptr++ = (number_mb_cmds << 16 | transform_8x8_mode_flag | ((i==0) << 1));
> >> + *command_ptr++ = encoder_context->quality_level;
> > It will be better that snb also uses the same inline field as that on
> > Haswell/Ivybridge.
> >
> One difference between SNB and HSW/BDW is that
> number_mb_cmds(`num_macroblocks', `inline_reg0.6') is useful for SNB,
> and it is UW data type,
> That why I define `quality_level_ub' to be `inline_reg0.8' instead of
> `inline_reg0.7`.
> For HSW/BDW, num_macroblocks is useless, I define `quality_level_ub' to
> be `inline_reg0.7' because I want to save reg space.
Understand your concern. If so, can we update the inline field for
quality_level for HSW/BDW like that on Snb?
Of course it is also OK to keep the current style.
Thanks.
Yakui
> >>
> >> i += number_mb_cmds;
> >> }
> >> diff --git a/src/shaders/vme/inter_frame.asm b/src/shaders/vme/inter_frame.asm
> >> index e1b6e68..7c5cfd4 100644
> >> --- a/src/shaders/vme/inter_frame.asm
> >> +++ b/src/shaders/vme/inter_frame.asm
> >> @@ -35,7 +35,11 @@ mov (1) read1_header.8<1>:UD BLOCK_4X16 {align1};
> >> mov (1) read1_header.20<1>:UB thread_id_ub {align1}; /* dispatch id */
> >>
> >> shl (2) vme_m0.8<1>:UW orig_xy_ub<2,2,1>:UB 4:UW {align1}; /* Source = (x, y) * 16 */
> >> -
> >> +
> >> +cmp.z.f0.0 (1) null<1>:uw quality_level_ub<0,1,0>:ub LOW_QUALITY_LEVEL:uw {align1};
> >> +(f0.0) jmpi (1) __low_quality_search;
> >> +
> >> +__high_quality_search:
> >> #ifdef DEV_SNB
> >> shl (2) vme_m0.0<1>:UW orig_xy_ub<2,2,1>:UB 4:UW {align1};
> >> add (1) vme_m0.0<1>:W vme_m0.0<0,1,0>:W -16:W {align1}; /* Reference = (x-16,y-12)-(x+32,y+24) */
> >> @@ -47,8 +51,25 @@ mov (1) vme_m0.2<1>:W -12:W {align1} ;
> >>
> >> mov (1) vme_m0.12<1>:UD SEARCH_CTRL_SINGLE + INTER_PART_MASK + INTER_SAD_HAAR + SUB_PEL_MODE_QUARTER:UD {align1}; /* 16x16 Source, 1/4 pixel, harr */
> >> mov (1) vme_m0.20<1>:UB thread_id_ub {align1}; /* dispatch id */
> >> -mov (1) vme_m0.22<1>:UW REF_REGION_SIZE {align1}; /* Reference Width&Height, 32x32 */
> >> +mov (1) vme_m0.22<1>:UW REF_REGION_SIZE {align1}; /* Reference Width&Height, 48x40 */
> >> +jmpi __vme_msg1;
> >> +
> >> +
> >> +__low_quality_search:
> >> +#ifdef DEV_SNB
> >> +shl (2) vme_m0.0<1>:UW orig_xy_ub<2,2,1>:UB 4:UW {align1};
> >> +add (1) vme_m0.0<1>:W vme_m0.0<0,1,0>:W -8:W {align1};
> >> +add (1) vme_m0.2<1>:W vme_m0.2<0,1,0>:W -8:W {align1};
> >> +#else
> >> +mov (1) vme_m0.0<1>:W -8:W {align1} ;
> >> +mov (1) vme_m0.2<1>:W -8:W {align1} ;
> >> +#endif
> >> +
> >> +mov (1) vme_m0.12<1>:UD SEARCH_CTRL_SINGLE + INTER_PART_MASK + INTER_SAD_HAAR + SUB_PEL_MODE_HALF:UD {align1}; /* 16x16 Source, 1/2 pixel, harr */
> >> +mov (1) vme_m0.20<1>:UB thread_id_ub {align1}; /* dispatch id */
> >> +mov (1) vme_m0.22<1>:UW MIN_REF_REGION_SIZE {align1}; /* Reference Width&Height, 32x32 */
> >>
> >> +__vme_msg1:
> >> mov (1) vme_m1.0<1>:UD ADAPTIVE_SEARCH_ENABLE:ud {align1} ;
> >> mov (1) vme_m1.4<1>:UD FB_PRUNING_ENABLE:UD {align1};
> >> /* MV num is passed by constant buffer. R4.28 */
> >> diff --git a/src/shaders/vme/inter_frame.g6b b/src/shaders/vme/inter_frame.g6b
> >> index ca251bb..bc7cd43 100644
> >> --- a/src/shaders/vme/inter_frame.g6b
> >> +++ b/src/shaders/vme/inter_frame.g6b
> >> @@ -11,12 +11,21 @@
> >> { 0x00000001, 0x242800e1, 0x00000000, 0x000f0003 },
> >> { 0x00000001, 0x24340231, 0x00000014, 0x00000000 },
> >> { 0x00200009, 0x24482e29, 0x004500a0, 0x00040004 },
> >> + { 0x01000010, 0x20002e28, 0x000000a8, 0x00020002 },
> >> + { 0x00010020, 0x34001c00, 0x00001400, 0x0000000e },
> >> { 0x00200009, 0x24402e29, 0x004500a0, 0x00040004 },
> >> { 0x00000040, 0x24403dad, 0x00000440, 0xfff0fff0 },
> >> { 0x00000040, 0x24423dad, 0x00000442, 0xfff4fff4 },
> >> { 0x00000001, 0x244c0061, 0x00000000, 0x00203000 },
> >> { 0x00000001, 0x24540231, 0x00000014, 0x00000000 },
> >> { 0x00000001, 0x24560169, 0x00000000, 0x28302830 },
> >> + { 0x00000020, 0x34001c00, 0x00001400, 0x0000000c },
> >> + { 0x00200009, 0x24402e29, 0x004500a0, 0x00040004 },
> >> + { 0x00000040, 0x24403dad, 0x00000440, 0xfff8fff8 },
> >> + { 0x00000040, 0x24423dad, 0x00000442, 0xfff8fff8 },
> >> + { 0x00000001, 0x244c0061, 0x00000000, 0x00201000 },
> >> + { 0x00000001, 0x24540231, 0x00000014, 0x00000000 },
> >> + { 0x00000001, 0x24560169, 0x00000000, 0x20202020 },
> >> { 0x00000001, 0x24600061, 0x00000000, 0x00000002 },
> >> { 0x00000001, 0x24640061, 0x00000000, 0x40000000 },
> >> { 0x00000001, 0x24640231, 0x0000009c, 0x00000000 },
> >> diff --git a/src/shaders/vme/vme.inc b/src/shaders/vme/vme.inc
> >> index dd7e1bb..8a39363 100644
> >> --- a/src/shaders/vme/vme.inc
> >> +++ b/src/shaders/vme/vme.inc
> >> @@ -54,6 +54,7 @@ define(`SEARCH_CTRL_DUAL_RECORD', `0x00000300')
> >> define(`SEARCH_CTRL_DUAL_REFERENCE', `0x00000700')
> >>
> >> define(`REF_REGION_SIZE', `0x2830:UW')
> >> +define(`MIN_REF_REGION_SIZE', `0x2020:UW')
> >>
> >> define(`BI_SUB_MB_PART_MASK', `0x0c000000')
> >> define(`MAX_NUM_MV', `0x00000020')
> >> @@ -140,6 +141,7 @@ define(`orig_y_ub', `inline_reg0.1')
> >> define(`transform_8x8_ub', `inline_reg0.4')
> >> define(`slice_edge_ub', `inline_reg0.4')
> >> define(`num_macroblocks', `inline_reg0.6')
> >> +define(`quality_level_ub', `inline_reg0.8')
> >>
> >> /*
> >> * GRF 6~11 -- reserved
> >> @@ -277,3 +279,6 @@ define(`vme_msg_4', `msg_reg4')
> >>
> >> #endif
> >>
> >> +define(`DEFAULT_QUALITY_LEVEL', `0x00')
> >> +define(`HIGH_QUALITY_LEVEL', `0x01')
> >> +define(`LOW_QUALITY_LEVEL', `0x02')
> >> --
> >> 1.7.9.5
> >>
> >> _______________________________________________
> >> Libva mailing list
> >> Libva at lists.freedesktop.org
> >> http://lists.freedesktop.org/mailman/listinfo/libva
> >
> >
> >
>
More information about the Libva
mailing list