[Libva] [PATCH v2 2/8] Use the same shader to configure quality level
Zhong Li
zhong.li at intel.com
Thu Aug 7 21:05:37 PDT 2014
Using the same shader to configure gen7 avc quality level,
it is benefit to make code clear and reduce profile limitation.
Signed-off-by: Zhong Li <zhong.li at intel.com>
---
src/gen7_vme.c | 73 +++++------------------------------
src/shaders/vme/inter_frame_ivb.asm | 22 ++++++++++-
src/shaders/vme/inter_frame_ivb.g7b | 16 ++++++--
src/shaders/vme/vme7.inc | 8 +++-
4 files changed, 49 insertions(+), 70 deletions(-)
diff --git a/src/gen7_vme.c b/src/gen7_vme.c
index 135a5ad..dc15445 100644
--- a/src/gen7_vme.c
+++ b/src/gen7_vme.c
@@ -63,8 +63,6 @@ enum VIDEO_CODING_TYPE{
enum AVC_VME_KERNEL_TYPE{
AVC_VME_INTRA_SHADER = 0,
AVC_VME_INTER_SHADER,
- AVC_VME_OLD_INTRA_SHADER,
- AVC_VME_OLD_INTER_SHADER,
AVC_VME_BATCHBUFFER,
AVC_VME_BINTER_SHADER,
AVC_VME_KERNEL_SUM
@@ -85,14 +83,6 @@ static const uint32_t gen7_vme_inter_frame[][4] = {
#include "shaders/vme/inter_frame_ivb.g7b"
};
-static const uint32_t gen7_vme_old_intra_frame[][4] = {
-#include "shaders/vme_old/intra_frame.g7b"
-};
-
-static const uint32_t gen7_vme_old_inter_frame[][4] = {
-#include "shaders/vme_old/inter_frame.g7b"
-};
-
static const uint32_t gen7_vme_batchbuffer[][4] = {
#include "shaders/vme/batchbuffer.g7b"
};
@@ -117,20 +107,6 @@ static struct i965_kernel gen7_vme_kernels[] = {
NULL
},
{
- "AVC VME Old Intra Frame",
- AVC_VME_OLD_INTRA_SHADER,
- gen7_vme_old_intra_frame,
- sizeof(gen7_vme_old_intra_frame),
- NULL
- },
- {
- "AVC VME Old Inter Frame",
- AVC_VME_OLD_INTER_SHADER,
- gen7_vme_old_inter_frame,
- sizeof(gen7_vme_old_inter_frame),
- NULL
- },
- {
"AVC VME BATCHBUFFER",
AVC_VME_BATCHBUFFER,
gen7_vme_batchbuffer,
@@ -383,36 +359,6 @@ static VAStatus gen7_vme_constant_setup(VADriverContextP ctx,
return VA_STATUS_SUCCESS;
}
-static VAStatus
-gen7_vme_vme_state_setup(VADriverContextP ctx,
- struct encode_state *encode_state,
- int is_intra,
- struct intel_encoder_context *encoder_context)
-{
- struct gen6_vme_context *vme_context = encoder_context->vme_context;
- unsigned int *vme_state_message;
- int i;
-
- //building VME state message
- 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] = 0x10010101;
- vme_state_message[1] = 0x100F0F0F;
- vme_state_message[2] = 0x10010101;
- vme_state_message[3] = 0x000F0F0F;
- for(i = 4; i < 14; i++) {
- vme_state_message[i] = 0x00000000;
- }
-
- for(i = 14; i < 32; i++) {
- vme_state_message[i] = 0x00000000;
- }
-
- dri_bo_unmap( vme_context->vme_state.bo);
- return VA_STATUS_SUCCESS;
-}
static VAStatus gen7_vme_avc_state_setup(VADriverContextP ctx,
struct encode_state *encode_state,
@@ -424,6 +370,7 @@ static VAStatus gen7_vme_avc_state_setup(VADriverContextP ctx,
unsigned int *mb_cost_table;
int i;
VAEncSliceParameterBufferH264 *slice_param = (VAEncSliceParameterBufferH264 *)encode_state->slice_params_ext[0]->buffer;
+ unsigned int is_low_quality = (encoder_context->quality_level == ENCODER_LOW_QUALITY);
mb_cost_table = (unsigned int *)vme_context->vme_state_message;
//building VME state message
@@ -431,8 +378,9 @@ static VAStatus gen7_vme_avc_state_setup(VADriverContextP ctx,
assert(vme_context->vme_state.bo->virtual);
vme_state_message = (unsigned int *)vme_context->vme_state.bo->virtual;
- if ((slice_param->slice_type == SLICE_TYPE_P) ||
- (slice_param->slice_type == SLICE_TYPE_SP)) {
+ if (((slice_param->slice_type == SLICE_TYPE_P) ||
+ (slice_param->slice_type == SLICE_TYPE_SP) &&
+ !is_low_quality)) {
vme_state_message[0] = 0x01010101;
vme_state_message[1] = 0x10010101;
vme_state_message[2] = 0x0F0F0F0F;
@@ -598,7 +546,7 @@ gen7_vme_fill_vme_batchbuffer(VADriverContextP ctx,
/*inline data */
*command_ptr++ = (mb_width << 16 | mb_y << 8 | mb_x);
- *command_ptr++ = ( (1 << 16) | transform_8x8_mode_flag | (mb_intra_ub << 8));
+ *command_ptr++ = ((encoder_context->quality_level << 24) | (1 << 16) | transform_8x8_mode_flag | (mb_intra_ub << 8));
i += 1;
}
@@ -668,14 +616,14 @@ static void gen7_vme_pipeline_programing(VADriverContextP ctx,
if ((pSliceParameter->slice_type == SLICE_TYPE_I) ||
(pSliceParameter->slice_type == SLICE_TYPE_I)) {
- kernel_shader = (is_low_quality ? AVC_VME_OLD_INTRA_SHADER : AVC_VME_INTRA_SHADER);
+ kernel_shader = AVC_VME_INTRA_SHADER;
} else if ((pSliceParameter->slice_type == SLICE_TYPE_P) ||
(pSliceParameter->slice_type == SLICE_TYPE_SP)) {
- kernel_shader = (is_low_quality ? AVC_VME_OLD_INTER_SHADER : AVC_VME_INTER_SHADER);
+ kernel_shader = AVC_VME_INTER_SHADER;
} else {
kernel_shader = AVC_VME_BINTER_SHADER;
if (!allow_hwscore)
- kernel_shader = (is_low_quality ? AVC_VME_OLD_INTER_SHADER : AVC_VME_INTER_SHADER);
+ kernel_shader = AVC_VME_INTER_SHADER;
}
if (allow_hwscore)
@@ -727,10 +675,7 @@ static VAStatus gen7_vme_prepare(VADriverContextP ctx,
gen7_vme_surface_setup(ctx, encode_state, is_intra, encoder_context);
gen7_vme_interface_setup(ctx, encode_state, encoder_context);
gen7_vme_constant_setup(ctx, encode_state, encoder_context);
- if (encoder_context->quality_level == ENCODER_LOW_QUALITY)
- gen7_vme_vme_state_setup(ctx, encode_state, is_intra, encoder_context);
- else
- gen7_vme_avc_state_setup(ctx, encode_state, is_intra, encoder_context);
+ gen7_vme_avc_state_setup(ctx, encode_state, is_intra, encoder_context);
/*Programing media pipeline*/
gen7_vme_pipeline_programing(ctx, encode_state, encoder_context);
diff --git a/src/shaders/vme/inter_frame_ivb.asm b/src/shaders/vme/inter_frame_ivb.asm
index 3c08851..46f2b4b 100644
--- a/src/shaders/vme/inter_frame_ivb.asm
+++ b/src/shaders/vme/inter_frame_ivb.asm
@@ -391,12 +391,14 @@ mov (1) mb_intra_struct_ub<1>:UB input_mb_intra_ub<0,1,0>:UB {align1};
/* M0 */
/* IME search */
+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:
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_m0.22<1>:UW REF_REGION_SIZE {align1}; /* Reference Width&Height, 48x40 */
-mov (1) vme_m0.0<1>:UD vme_m0.8<0,1,0>:UD {align1};
-
mov (1) vme_m0.0<1>:W -16:W {align1};
mov (1) vme_m0.2<1>:W -12:W {align1};
@@ -405,6 +407,22 @@ and.z.f0.0 (1) null:uw input_mb_intra_ub<0,1,0>:ub INTRA_PRED_AVAIL_FLAG_AE:uw
and.z.f0.0 (1) null:uw input_mb_intra_ub<0,1,0>:ub INTRA_PRED_AVAIL_FLAG_B:uw {align1};
(f0.0) add (1) vme_m0.2<1>:w vme_m0.2<0,1,0>:w 8:w {align1};
+jmpi __vme_msg;
+
+__low_quality_search:
+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_m0.22<1>:UW MIN_REF_REGION_SIZE {align1}; /* Reference Width&Height, 32x32 */
+
+mov (1) vme_m0.0<1>:W -8:W {align1};
+mov (1) vme_m0.2<1>:W -8:W {align1};
+
+and.z.f0.0 (1) null:uw input_mb_intra_ub<0,1,0>:ub INTRA_PRED_AVAIL_FLAG_AE:uw {align1};
+(f0.0) add (1) vme_m0.0<1>:w vme_m0.0<0,1,0>:w 4:w {align1};
+and.z.f0.0 (1) null:uw input_mb_intra_ub<0,1,0>:ub INTRA_PRED_AVAIL_FLAG_B:uw {align1};
+(f0.0) add (1) vme_m0.2<1>:w vme_m0.2<0,1,0>:w 4:w {align1};
+
+__vme_msg:
mov (1) vme_m0.4<1>:UD vme_m0.0<0,1,0>:UD {align1};
add (2) vme_m0.0<1>:w vme_m0.0<2,2,1>:w mb_ref_win.16<2,2,1>:w {align1};
add (2) vme_m0.4<1>:w vme_m0.4<2,2,1>:w mb_ref_win.16<2,2,1>:w {align1};
diff --git a/src/shaders/vme/inter_frame_ivb.g7b b/src/shaders/vme/inter_frame_ivb.g7b
index e4db6ea..7ed38c5 100644
--- a/src/shaders/vme/inter_frame_ivb.g7b
+++ b/src/shaders/vme/inter_frame_ivb.g7b
@@ -141,13 +141,13 @@
{ 0x00000001, 0x2fa401ad, 0x00000b04, 0x00000000 },
{ 0x00000001, 0x2fa801ad, 0x00000b24, 0x00000000 },
{ 0x00000040, 0x2fe00c01, 0x00001400, 0x00000020 },
- { 0x00000020, 0x34001c00, 0x00001400, 0x000000bc },
+ { 0x00000020, 0x34001c00, 0x00001400, 0x000000d0 },
{ 0x00000001, 0x2ac001ad, 0x00000fe4, 0x00000000 },
{ 0x00000001, 0x2fa001ad, 0x00000ae6, 0x00000000 },
{ 0x00000001, 0x2fa401ad, 0x00000b06, 0x00000000 },
{ 0x00000001, 0x2fa801ad, 0x00000b26, 0x00000000 },
{ 0x00000040, 0x2fe00c01, 0x00001400, 0x00000020 },
- { 0x00000020, 0x34001c00, 0x00001400, 0x000000b0 },
+ { 0x00000020, 0x34001c00, 0x00001400, 0x000000c4 },
{ 0x00000001, 0x2ac201ad, 0x00000fe4, 0x00000000 },
{ 0x0020000c, 0x2a803dad, 0x00450ac0, 0x00020002 },
{ 0x00200040, 0x2a883dad, 0x00450a80, 0x00030003 },
@@ -163,15 +163,25 @@
{ 0x01000005, 0x20002e28, 0x000000a4, 0x00010001 },
{ 0x00010001, 0x247c0171, 0x00000000, 0x00020002 },
{ 0x00000001, 0x247d0231, 0x000000a5, 0x00000000 },
+ { 0x01000010, 0x20002e28, 0x000000a7, 0x00020002 },
+ { 0x00010020, 0x34001c00, 0x00001400, 0x00000012 },
{ 0x00000001, 0x244c0061, 0x00000000, 0x00203000 },
{ 0x00000001, 0x24560169, 0x00000000, 0x28302830 },
- { 0x00000001, 0x24400021, 0x00000448, 0x00000000 },
{ 0x00000001, 0x244001ed, 0x00000000, 0xfff0fff0 },
{ 0x00000001, 0x244201ed, 0x00000000, 0xfff4fff4 },
{ 0x01000005, 0x20002e28, 0x000000a5, 0x00600060 },
{ 0x00010040, 0x24403dad, 0x00000440, 0x000c000c },
{ 0x01000005, 0x20002e28, 0x000000a5, 0x00100010 },
{ 0x00010040, 0x24423dad, 0x00000442, 0x00080008 },
+ { 0x00000020, 0x34001c00, 0x00001400, 0x00000010 },
+ { 0x00000001, 0x244c0061, 0x00000000, 0x00201000 },
+ { 0x00000001, 0x24560169, 0x00000000, 0x20202020 },
+ { 0x00000001, 0x244001ed, 0x00000000, 0xfff8fff8 },
+ { 0x00000001, 0x244201ed, 0x00000000, 0xfff8fff8 },
+ { 0x01000005, 0x20002e28, 0x000000a5, 0x00600060 },
+ { 0x00010040, 0x24403dad, 0x00000440, 0x00040004 },
+ { 0x01000005, 0x20002e28, 0x000000a5, 0x00100010 },
+ { 0x00010040, 0x24423dad, 0x00000442, 0x00040004 },
{ 0x00000001, 0x24440021, 0x00000440, 0x00000000 },
{ 0x00200040, 0x244035ad, 0x00450440, 0x00450a90 },
{ 0x00200040, 0x244435ad, 0x00450444, 0x00450a90 },
diff --git a/src/shaders/vme/vme7.inc b/src/shaders/vme/vme7.inc
index 3fa99b7..e9d5864 100644
--- a/src/shaders/vme/vme7.inc
+++ b/src/shaders/vme/vme7.inc
@@ -54,6 +54,8 @@ 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(`DREF_REGION_SIZE', `0x2020:UW')
define(`BI_SUB_MB_PART_MASK', `0x0c000000')
define(`MAX_NUM_MV', `0x00000020')
@@ -132,6 +134,7 @@ define(`orig_y_ub', `inline_reg0.1')
define(`transform_8x8_ub', `inline_reg0.4')
define(`input_mb_intra_ub', `inline_reg0.5')
define(`num_macroblocks', `inline_reg0.6')
+define(`quality_level_ub', `inline_reg0.7')
/*
* GRF 6~11 -- reserved
@@ -291,7 +294,6 @@ define(`mb_mv3', `r96')
define(`mb_ref', `r97')
define(`mb_ref_win', `r84')
-define(`DREF_REGION_SIZE', `0x2020:UW')
define(`PRED_L0', `0x0':uw)
define(`PRED_L1', `0x1':uw)
define(`PRED_BI', `0x2':uw)
@@ -317,3 +319,7 @@ define(`INTER_8X16MODE', `0x02')
define(`OBR_MESSAGE_FENCE', `7')
define(`OBR_MF_NOCOMMIT', `0')
define(`OBR_MF_COMMIT', `0x20')
+
+define(`DEFAULT_QUALITY_LEVEL', `0x01')
+define(`HIGH_QUALITY_LEVEL', `DEFAULT_QUALITY_LEVEL')
+define(`LOW_QUALITY_LEVEL', `0x02')
--
1.7.9.5
More information about the Libva
mailing list