Signed-off-by: Zhong Li <[email protected]> --- src/shaders/vme/Makefile.am | 6 +- src/shaders/vme/vme8.inc | 5 + src/shaders/vme/vp8_intra_frame_gen9.asm | 187 +++++++++++++++++++++++++++++++ src/shaders/vme/vp8_intra_frame_gen9.g9a | 2 + src/shaders/vme/vp8_intra_frame_gen9.g9b | 73 ++++++++++++ 5 files changed, 270 insertions(+), 3 deletions(-) create mode 100644 src/shaders/vme/vp8_intra_frame_gen9.asm create mode 100644 src/shaders/vme/vp8_intra_frame_gen9.g9a create mode 100644 src/shaders/vme/vp8_intra_frame_gen9.g9b
diff --git a/src/shaders/vme/Makefile.am b/src/shaders/vme/Makefile.am index 435e4ae..123378f 100644 --- a/src/shaders/vme/Makefile.am +++ b/src/shaders/vme/Makefile.am @@ -2,7 +2,7 @@ VME_CORE = batchbuffer.asm intra_frame.asm inter_frame.asm VME7_CORE = batchbuffer.asm intra_frame_ivb.asm inter_frame_ivb.asm inter_bframe_ivb.asm mpeg2_inter_ivb.asm VME75_CORE = batchbuffer.asm intra_frame_haswell.asm inter_frame_haswell.asm inter_bframe_haswell.asm mpeg2_inter_haswell.asm VME8_CORE = intra_frame_gen8.asm inter_frame_gen8.asm inter_bframe_gen8.asm -VME9_CORE = $(VME8_CORE) +VME9_CORE = $(VME8_CORE) vp8_intra_frame_gen9.asm INTEL_G6B = batchbuffer.g6b intra_frame.g6b inter_frame.g6b INTEL_G6A = batchbuffer.g6a intra_frame.g6a inter_frame.g6a @@ -26,8 +26,8 @@ INTEL_GEN8_INC = vme8.inc vme75_mpeg2.inc INTEL_GEN8_ASM = $(INTEL_G8A:%.g8a=%.gen8.asm) -INTEL_G9B = intra_frame_gen9.g9b inter_frame_gen9.g9b inter_bframe_gen9.g9b mpeg2_inter_gen9.g9b -INTEL_G9A = $(INTEL_G8A) +INTEL_G9B = intra_frame_gen9.g9b inter_frame_gen9.g9b inter_bframe_gen9.g9b mpeg2_inter_gen9.g9b vp8_intra_frame_gen9.g9b +INTEL_G9A = intra_frame_gen9.g9a inter_frame_gen9.g9a inter_bframe_gen9.g9a mpeg2_inter_gen9.g9a vp8_intra_frame_gen9.g9a INTEL_GEN9_INC = $(INTEL_GEN8_INC) INTEL_GEN9_ASM = $(INTEL_G9A:%.g9a=%.gen9.asm) diff --git a/src/shaders/vme/vme8.inc b/src/shaders/vme/vme8.inc index 5b6f469..46cefe4 100644 --- a/src/shaders/vme/vme8.inc +++ b/src/shaders/vme/vme8.inc @@ -30,6 +30,9 @@ define(`LUMA_INTRA_16x16_DISABLE', `0x1') define(`LUMA_INTRA_8x8_DISABLE', `0x2') define(`LUMA_INTRA_4x4_DISABLE', `0x4') +define(`SUB_PART_8x4_DISABLE', `0x20') +define(`SUB_PART_4x8_DISABLE', `0x40') + define(`INTRA_PRED_AVAIL_FLAG_AE', `0x60') define(`INTRA_PRED_AVAIL_FLAG_B', `0x10') define(`INTRA_PRED_AVAIL_FLAG_C', `0x8') @@ -72,6 +75,8 @@ define(`START_CENTER', `0x30000000') define(`ADAPTIVE_SEARCH_ENABLE', `0x00000002') define(`INTRA_PREDICTORE_MODE', `0x11111111:UD') +define(`INTRA_PLANAR_MODE_MASK', `0x10001000:UD') + define(`INTER_VME_OUTPUT_IN_OWS', `10') define(`INTER_VME_OUTPUT_MV_IN_OWS', `8') diff --git a/src/shaders/vme/vp8_intra_frame_gen9.asm b/src/shaders/vme/vp8_intra_frame_gen9.asm new file mode 100644 index 0000000..0ca35be --- /dev/null +++ b/src/shaders/vme/vp8_intra_frame_gen9.asm @@ -0,0 +1,187 @@ +/* + * Copyright © <2014>, Intel Corporation. + * + * This program is licensed under the terms and conditions of the + * Eclipse Public License (EPL), version 1.0. The full text of the EPL is at + * http://www.opensource.org/licenses/eclipse-1.0.php. + * + */ +// Modual name: VP8_intra_frame_gen9.asm +// +// Make intra predition estimation for VP8 intra frame on Gen9 +// + +// +// Now, begin source code.... +// + +/* + * __START + */ +__INTRA_START: +mov (16) tmp_reg0.0<1>:UD 0x0:UD {align1}; +mov (16) tmp_reg2.0<1>:UD 0x0:UD {align1}; +mov (16) tmp_reg4.0<1>:UD 0x0:UD {align1} ; +mov (16) tmp_reg6.0<1>:UD 0x0:UD {align1} ; + +shl (2) read0_header.0<1>:D orig_xy_ub<2,2,1>:UB 4:UW {align1}; /* (x, y) * 16 */ +add (1) read0_header.0<1>:D read0_header.0<0,1,0>:D -8:W {align1}; /* X offset */ +add (1) read0_header.4<1>:D read0_header.4<0,1,0>:D -1:W {align1}; /* Y offset */ +mov (1) read0_header.8<1>:UD BLOCK_32X1 {align1}; +mov (1) read0_header.20<1>:UB thread_id_ub {align1}; /* dispatch id */ + +shl (2) read1_header.0<1>:D orig_xy_ub<2,2,1>:UB 4:UW {align1}; /* (x, y) * 16 */ +add (1) read1_header.0<1>:D read1_header.0<0,1,0>:D -4:W {align1}; /* X offset */ +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}; /* (x, y) * 16 */ +mov (1) vme_m0.20<1>:UB thread_id_ub {align1}; /* dispatch id */ + +mul (1) obw_m0.8<1>:UD w_in_mb_uw<0,1,0>:UW orig_y_ub<0,1,0>:UB {align1}; +add (1) obw_m0.8<1>:UD obw_m0.8<0,1,0>:UD orig_x_ub<0,1,0>:UB {align1}; +mul (1) obw_m0.8<1>:UD obw_m0.8<0,1,0>:UD 0x02:UD {align1}; +mov (1) obw_m0.20<1>:UB thread_id_ub {align1}; /* dispatch id */ + +/* + * Media Read Message -- fetch Luma neighbor edge pixels + */ +/* ROW */ +mov (8) msg_reg0.0<1>:UD read0_header.0<8,8,1>:UD {align1}; +send (8) msg_ind INEP_ROW<1>:UB null read(BIND_IDX_INEP, 0, 0, 4) mlen 1 rlen 1 {align1}; + +/* COL */ +mov (8) msg_reg0.0<1>:UD read1_header.0<8,8,1>:UD {align1}; +send (8) msg_ind INEP_COL0<1>:UB null read(BIND_IDX_INEP, 0, 0, 4) mlen 1 rlen 2 {align1}; + +/* + * Media Read Message -- fetch Chroma neighbor edge pixels + */ +/* ROW */ +shl (2) read0_header.0<1>:D orig_xy_ub<2,2,1>:UB 3:UW {align1}; /* x * 16 , y * 8 */ +mul (1) read0_header.0<1>:D read0_header.0<0,1,0>:D 2:W {align1}; +add (1) read0_header.0<1>:D read0_header.0<0,1,0>:D -8:W {align1}; /* X offset */ +add (1) read0_header.4<1>:D read0_header.4<0,1,0>:D -1:W {align1}; /* Y offset */ +mov (8) msg_reg0.0<1>:UD read0_header.0<8,8,1>:UD {align1}; +send (8) msg_ind CHROMA_ROW<1>:UB null read(BIND_IDX_CBCR, 0, 0, 4) mlen 1 rlen 1 {align1}; + +/* COL */ +shl (2) read1_header.0<1>:D orig_xy_ub<2,2,1>:UB 3:UW {align1}; /* x * 16, y * 8 */ +mul (1) read1_header.0<1>:D read1_header.0<0,1,0>:D 2:W {align1}; +add (1) read1_header.0<1>:D read1_header.0<0,1,0>:D -4:W {align1}; /* X offset */ +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}; + +/* 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}; + +/* m3. This is changed for FWD/BWD cost center */ +mov (8) vme_msg_3<1>:UD 0x0:UD {align1}; + +/* m4.*/ +mov (8) vme_msg_4<1>:ud 0x0:ud {align1}; + +/* m5 */ +mov (1) INEP_ROW.0<1>:UD 0x0:UD {align1}; +and (1) INEP_ROW.4<1>:UD INEP_ROW.4<0,1,0>:UD 0xFF000000:UD {align1}; +mov (8) vme_msg_5<1>:UD INEP_ROW.0<8,8,1>:UD {align1}; + +mov (1) tmp_reg0.0<1>:UB INTRA_PLANAR_MODE_MASK {align1}; /* vp8 don't support planar intra mode */ +mov (1) tmp_reg0.1<1>:UB LUMA_CHROMA_MODE {align1}; /* Intra type: Luma + Chroma */ + +/* Intra mode mask && Intra compute type */ +mov (1) vme_msg_5.4<1>:UW tmp_reg0.0<0,1,0>:UW {align1}; + +/* m6 */ +mov (8) vme_msg_6<1>:UD 0x0:UD {align1}; +mov (16) vme_msg_6.0<1>:UB INEP_COL0.3<32,8,4>:UB {align1}; +mov (1) vme_msg_6.16<1>:UD INTRA_PREDICTORE_MODE {align1}; + +/* the penalty for Intra mode */ +mov (1) vme_msg_6.28<1>:UD 0x010101:UD {align1}; +mov (1) vme_msg_6.20<1>:UW CHROMA_ROW.6<0,1,0>:UW {align1}; + + +/* m7 */ + +mov (4) vme_msg_7.16<1>:UD CHROMA_ROW.8<4,4,1>:UD {align1}; +mov (8) vme_msg_7.0<1>:UW CHROMA_COL.2<16,8,2>:UW {align1}; + +/* + * VME message + */ + +/* m1 */ +mov (1) intra_flag<1>:UW 0x0:UW {align1}; +mov (1) intra_part_mask_ub<1>:UB LUMA_INTRA_8x8_DISABLE {align1}; /* vp8 don't support intra_8x8 mode*/ + +/* assign MB intra struct from the thread payload*/ +mov (1) mb_intra_struct_ub<1>:UB input_mb_intra_ub<0,1,0>:UB {align1}; + +/* Disable DC HAAR component when calculating HARR SATD block */ +mov (1) tmp_reg0.0<1>:UW DC_HARR_DISABLE:UW {align1}; +mov (1) vme_m1.30<1>:UB tmp_reg0.0<0,1,0>:UB {align1}; + +mov (8) vme_msg_1<1>:UD vme_m1.0<8,8,1>:UD {align1}; + +/* m0 */ +add (1) vme_m0.12<1>:UD vme_m0.12<0,1,0>:ud INTRA_SAD_HAAR:UD {align1};/* 16x16 Source, Intra_harr */ +mov (1) vme_m0.15<1>:UB SUB_PART_8x4_DISABLE + SUB_PART_4x8_DISABLE {align1}; /* vp8 don't support 8x4 and 4x8 partion */ +mov (8) vme_msg_0<1>:UD vme_m0.0<8,8,1>:UD {align1}; + +/* after verification it will be passed by using payload */ +send (8) + vme_msg_ind + vme_wb<1>:UD + null + cre( + BIND_IDX_VME, + VME_SIC_MESSAGE_TYPE + ) + mlen sic_vme_msg_length + rlen vme_wb_length + {align1}; +/* + * Oword Block Write message + */ +mov (8) msg_reg0.0<1>:UD obw_m0<8,8,1>:UD {align1}; + +mov (1) msg_reg1.0<1>:UD vme_wb.0<0,1,0>:UD {align1}; +mov (1) msg_reg1.4<1>:UD vme_wb.16<0,1,0>:UD {align1}; +mov (1) msg_reg1.8<1>:UD vme_wb.20<0,1,0>:UD {align1}; +mov (1) msg_reg1.12<1>:UD vme_wb.24<0,1,0>:UD {align1}; + +/* Distortion, Intra (17-16), */ +mov (1) msg_reg1.16<1>:UW vme_wb.12<0,1,0>:UW {align1}; + +mov (1) msg_reg1.20<1>:UD vme_wb.8<0,1,0>:UD {align1}; +/* VME clock counts */ +mov (1) msg_reg1.24<1>:UD vme_wb.28<0,1,0>:UD {align1}; + +mov (1) msg_reg1.28<1>:UD obw_m0.8<0,1,0>:UD {align1}; + +/* bind index 3, write 2 oword (32bytes), msg type: 8(OWord Block Write) */ +send (16) + msg_ind + obw_wb + null + data_port( + OBW_CACHE_TYPE, + OBW_MESSAGE_TYPE, + OBW_CONTROL_2, + OBW_BIND_IDX, + OBW_WRITE_COMMIT_CATEGORY, + OBW_HEADER_PRESENT + ) + mlen 2 + rlen obw_wb_length + {align1}; + +__EXIT: +/* + * kill thread + */ +mov (8) ts_msg_reg0<1>:UD r0<8,8,1>:UD {align1}; +send (16) ts_msg_ind acc0<1>UW null thread_spawner(0, 0, 1) mlen 1 rlen 0 {align1 EOT}; diff --git a/src/shaders/vme/vp8_intra_frame_gen9.g9a b/src/shaders/vme/vp8_intra_frame_gen9.g9a new file mode 100644 index 0000000..db0b9ff --- /dev/null +++ b/src/shaders/vme/vp8_intra_frame_gen9.g9a @@ -0,0 +1,2 @@ +#include "vme8.inc" +#include "vp8_intra_frame_gen9.asm" diff --git a/src/shaders/vme/vp8_intra_frame_gen9.g9b b/src/shaders/vme/vp8_intra_frame_gen9.g9b new file mode 100644 index 0000000..c4f9baf --- /dev/null +++ b/src/shaders/vme/vp8_intra_frame_gen9.g9b @@ -0,0 +1,73 @@ + { 0x00800001, 0x24000608, 0x00000000, 0x00000000 }, + { 0x00800001, 0x24400608, 0x00000000, 0x00000000 }, + { 0x00800001, 0x24800608, 0x00000000, 0x00000000 }, + { 0x00800001, 0x24c00608, 0x00000000, 0x00000000 }, + { 0x00200009, 0x24002228, 0x164500a0, 0x00040004 }, + { 0x00000040, 0x24000a28, 0x1e000400, 0xfff8fff8 }, + { 0x00000040, 0x24040a28, 0x1e000404, 0xffffffff }, + { 0x00000001, 0x24080e08, 0x08000000, 0x0000001f }, + { 0x00000001, 0x24142288, 0x00000014, 0x00000000 }, + { 0x00200009, 0x24202228, 0x164500a0, 0x00040004 }, + { 0x00000040, 0x24200a28, 0x1e000420, 0xfffcfffc }, + { 0x00000001, 0x24280e08, 0x08000000, 0x000f0003 }, + { 0x00000001, 0x24342288, 0x00000014, 0x00000000 }, + { 0x00200009, 0x24482248, 0x164500a0, 0x00040004 }, + { 0x00000001, 0x24542288, 0x00000014, 0x00000000 }, + { 0x00000041, 0x24881208, 0x220000a2, 0x000000a1 }, + { 0x00000040, 0x24880208, 0x22000488, 0x000000a0 }, + { 0x00000041, 0x24880208, 0x06000488, 0x00000002 }, + { 0x00000001, 0x24942288, 0x00000014, 0x00000000 }, + { 0x00600001, 0x28000208, 0x008d0400, 0x00000000 }, + { 0x0c600031, 0x23800a88, 0x06000800, 0x02190004 }, + { 0x00600001, 0x28000208, 0x008d0420, 0x00000000 }, + { 0x0c600031, 0x23a00a88, 0x06000800, 0x02290004 }, + { 0x00200009, 0x24002228, 0x164500a0, 0x00030003 }, + { 0x00000041, 0x24000a28, 0x1e000400, 0x00020002 }, + { 0x00000040, 0x24000a28, 0x1e000400, 0xfff8fff8 }, + { 0x00000040, 0x24040a28, 0x1e000404, 0xffffffff }, + { 0x00600001, 0x28000208, 0x008d0400, 0x00000000 }, + { 0x0c600031, 0x26000a88, 0x06000800, 0x02190006 }, + { 0x00200009, 0x24202228, 0x164500a0, 0x00030003 }, + { 0x00000041, 0x24200a28, 0x1e000420, 0x00020002 }, + { 0x00000040, 0x24200a28, 0x1e000420, 0xfffcfffc }, + { 0x00000001, 0x24280e08, 0x08000000, 0x00070003 }, + { 0x00600001, 0x28000208, 0x008d0420, 0x00000000 }, + { 0x0c600031, 0x26200a88, 0x06000800, 0x02190006 }, + { 0x00600001, 0x28400208, 0x008d0020, 0x00000000 }, + { 0x00600001, 0x28600608, 0x00000000, 0x00000000 }, + { 0x00600001, 0x28800608, 0x00000000, 0x00000000 }, + { 0x00000001, 0x23800608, 0x00000000, 0x00000000 }, + { 0x00000005, 0x23840208, 0x06000384, 0xff000000 }, + { 0x00600001, 0x28a00208, 0x008d0380, 0x00000000 }, + { 0x00000001, 0x24000688, 0x00000000, 0x10001000 }, + { 0x00000001, 0x24010e88, 0x08000000, 0x00000000 }, + { 0x00000001, 0x28a41248, 0x00000400, 0x00000000 }, + { 0x00600001, 0x28c00608, 0x00000000, 0x00000000 }, + { 0x00800001, 0x28c02288, 0x00cf03a3, 0x00000000 }, + { 0x00000001, 0x28d00608, 0x00000000, 0x11111111 }, + { 0x00000001, 0x28dc0608, 0x00000000, 0x00010101 }, + { 0x00000001, 0x28d41248, 0x00000606, 0x00000000 }, + { 0x00400001, 0x28f00208, 0x00690608, 0x00000000 }, + { 0x00600001, 0x28e01248, 0x00ae0622, 0x00000000 }, + { 0x00000001, 0x247c1648, 0x10000000, 0x00000000 }, + { 0x00000001, 0x247c0e88, 0x08000000, 0x00000002 }, + { 0x00000001, 0x247d2288, 0x000000a5, 0x00000000 }, + { 0x00000001, 0x24001648, 0x10000000, 0x00200020 }, + { 0x00000001, 0x247e2288, 0x00000400, 0x00000000 }, + { 0x00600001, 0x28200208, 0x008d0460, 0x00000000 }, + { 0x00000040, 0x244c0208, 0x0600044c, 0x00800000 }, + { 0x00000001, 0x244f0e88, 0x08000000, 0x00000060 }, + { 0x00600001, 0x28000208, 0x008d0440, 0x00000000 }, + { 0x0d600031, 0x21800a08, 0x06000800, 0x10782000 }, + { 0x00600001, 0x28000208, 0x008d0480, 0x00000000 }, + { 0x00000001, 0x28200208, 0x00000180, 0x00000000 }, + { 0x00000001, 0x28240208, 0x00000190, 0x00000000 }, + { 0x00000001, 0x28280208, 0x00000194, 0x00000000 }, + { 0x00000001, 0x282c0208, 0x00000198, 0x00000000 }, + { 0x00000001, 0x28301248, 0x0000018c, 0x00000000 }, + { 0x00000001, 0x28340208, 0x00000188, 0x00000000 }, + { 0x00000001, 0x28380208, 0x0000019c, 0x00000000 }, + { 0x00000001, 0x283c0208, 0x00000488, 0x00000000 }, + { 0x0a800031, 0x20000a60, 0x06000800, 0x040a0203 }, + { 0x00600001, 0x2e000208, 0x008d0000, 0x00000000 }, + { 0x07800031, 0x24000a40, 0x06000e00, 0x82000010 }, -- 1.9.1 _______________________________________________ Libva mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/libva
