<div style="line-height:1.7;color:#000000;font-size:14px;font-family:arial"><div>more: your code is right, just washing cycles to combin data, for 16xN, we can use movu to reduce instruction on same things</div><pre><br>At 2014-11-19 20:21:01,"Divya Manivannan" <divya@multicorewareinc.com> wrote:
># HG changeset patch
># User Divya Manivannan <divya@multicorewareinc.com>
># Date 1416399611 -19800
># Wed Nov 19 17:50:11 2014 +0530
># Node ID b9ad1bc1610c0941eff5df6c5c0eb6a42ac9fdca
># Parent ce962434be14a13338286ce6f824e2dfbd127eb2
>asm: luma_vpp[16x16] in avx2 for 32 bit: improve 2141c->1488c
>
>diff -r ce962434be14 -r b9ad1bc1610c source/common/x86/ipfilter8.asm
>--- a/source/common/x86/ipfilter8.asm Wed Nov 19 17:07:02 2014 +0530
>+++ b/source/common/x86/ipfilter8.asm Wed Nov 19 17:50:11 2014 +0530
>@@ -3882,6 +3882,80 @@
> ;-------------------------------------------------------------------------------------------------------------
> FILTER_VER_LUMA_12xN 12, 16, ps
>
>+%macro PROCESS_LUMA_AVX2_16_W8_8R 0
>+ movq xm1, [r0] ; m1 = row 0
>+ movq xm2, [r0 + r1] ; m2 = row 1
>+ punpcklbw xm1, xm2 ; m1 = [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
>+ lea r6, [r0 + r1 * 2]
>+ movq xm3, [r6] ; m3 = row 2
>+ punpcklbw xm2, xm3 ; m2 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10]
>+ vinserti128 m5, m1, xm2, 1 ; m5 = [27 17 26 16 25 15 24 14 23 13 22 12 21 11 20 10] - [17 07 16 06 15 05 14 04 13 03 12 02 11 01 10 00]
>+ pmaddubsw m5, [r5]
>+ movq xm4, [r6 + r1] ; m4 = row 3
>+ punpcklbw xm3, xm4 ; m3 = [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
>+ lea r6, [r6 + r1 * 2]
>+ movq xm1, [r6] ; m1 = row 4
>+ punpcklbw xm4, xm1 ; m4 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30]
>+ vinserti128 m2, m3, xm4, 1 ; m2 = [47 37 46 36 45 35 44 34 43 33 42 32 41 31 40 30] - [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
>+ pmaddubsw m0, m2, [r5 + 32]
>+ paddw m5, m0
>+ pmaddubsw m2, [r5]
>+ movq xm3, [r6 + r1] ; m3 = row 5
>+ punpcklbw xm1, xm3 ; m1 = [57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40]
>+ lea r6, [r6 + r1 * 2]
>+ movq xm4, [r6] ; m4 = row 6
>+ punpcklbw xm3, xm4 ; m3 = [67 57 66 56 65 55 64 54 63 53 62 52 61 51 60 50]
>+ vinserti128 m1, m1, xm3, 1 ; m1 = [67 57 66 56 65 55 64 54 63 53 62 52 61 51 60 50] - [57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40]
>+ pmaddubsw m3, m1, [r5 + 64]
>+ paddw m5, m3
>+ pmaddubsw m0, m1, [r5 + 32]
>+ paddw m2, m0
>+ pmaddubsw m1, [r5]
>+ movq xm3, [r6 + r1] ; m3 = row 7
>+ punpcklbw xm4, xm3 ; m4 = [77 67 76 66 75 65 74 64 73 63 72 62 71 61 70 60]
>+ lea r6, [r6 + r1 * 2]
>+ movq xm0, [r6] ; m0 = row 8
>+ punpcklbw xm3, xm0 ; m3 = [87 77 86 76 85 75 84 74 83 73 82 72 81 71 80 70]
>+ vinserti128 m4, m4, xm3, 1 ; m4 = [87 77 86 76 85 75 84 74 83 73 82 72 81 71 80 70] - [77 67 76 66 75 65 74 64 73 63 72 62 71 61 70 60]
>+ pmaddubsw m3, m4, [r5 + 96]
>+ paddw m5, m3
>+ pmaddubsw m3, m4, [r5 + 64]
>+ paddw m2, m3
>+ pmaddubsw m3, m4, [r5 + 32]
>+ paddw m1, m3
>+ pmaddubsw m4, [r5]
>+ movq xm3, [r6 + r1] ; m3 = row 9
>+ punpcklbw xm0, xm3 ; m0 = [97 87 96 86 95 85 94 84 93 83 92 82 91 81 90 80]
>+ lea r6, [r6 + r1 * 2]
>+ movq xm6, [r6] ; m6 = row 10
>+ punpcklbw xm3, xm6 ; m3 = [A7 97 A6 96 A5 95 A4 94 A3 93 A2 92 A1 91 A0 90]
>+ vinserti128 m0, m0, xm3, 1 ; m0 = [A7 97 A6 96 A5 95 A4 94 A3 93 A2 92 A1 91 A0 90] - [97 87 96 86 95 85 94 84 93 83 92 82 91 81 90 80]
>+ pmaddubsw m3, m0, [r5 + 96]
>+ paddw m2, m3
>+ pmaddubsw m3, m0, [r5 + 64]
>+ paddw m1, m3
>+ pmaddubsw m0, [r5 + 32]
>+ paddw m4, m0
>+
>+ movq xm3, [r6 + r1] ; m3 = row 11
>+ punpcklbw xm6, xm3 ; m6 = [B7 A7 B6 A6 B5 A5 B4 A4 B3 A3 B2 A2 B1 A1 B0 A0]
>+ lea r6, [r6 + r1 * 2]
>+ movq xm0, [r6] ; m0 = row 12
>+ punpcklbw xm3, xm0 ; m3 = [C7 B7 C6 B6 C5 B5 C4 B4 C3 B3 C2 B2 C1 B1 C0 B0]
>+ vinserti128 m6, m6, xm3, 1 ; m6 = [C7 B7 C6 B6 C5 B5 C4 B4 C3 B3 C2 B2 C1 B1 C0 B0] - [B7 A7 B6 A6 B5 A5 B4 A4 B3 A3 B2 A2 B1 A1 B0 A0]
>+ pmaddubsw m3, m6, [r5 + 96]
>+ paddw m1, m3
>+ pmaddubsw m6, [r5 + 64]
>+ paddw m4, m6
>+ movq xm3, [r6 + r1] ; m3 = row 13
>+ punpcklbw xm0, xm3 ; m0 = [D7 C7 D6 C6 D5 C5 D4 C4 D3 C3 D2 C2 D1 C1 D0 C0]
>+ movq xm6, [r6 + r1 * 2] ; m6 = row 14
>+ punpcklbw xm3, xm6 ; m3 = [E7 D7 E6 D6 E5 D5 E4 D4 E3 D3 E2 D2 E1 D1 E0 D0]
>+ vinserti128 m0, m0, xm3, 1 ; m0 = [E7 D7 E6 D6 E5 D5 E4 D4 E3 D3 E2 D2 E1 D1 E0 D0] - [D7 C7 D6 C6 D5 C5 D4 C4 D3 C3 D2 C2 D1 C1 D0 C0]
>+ pmaddubsw m0, [r5 + 96]
>+ paddw m4, m0
>+%endmacro
>+
> INIT_YMM avx2
> %if ARCH_X86_64 == 1
> cglobal interp_8tap_vert_pp_16x16, 4, 7, 15
>@@ -4171,6 +4245,58 @@
> movu [r2 + r3 * 2], xm0
> movu [r2 + r6], xm1
> RET
>+
>+%else
>+cglobal interp_8tap_vert_pp_16x16, 4, 7, 7, 0-gprsize
>+ mov r4d, r4m
>+ shl r4d, 7
>+ lea r5, [r1 * 3]
>+ sub r0, r5
>+
>+%ifdef PIC
>+ lea r5, [tab_LumaCoeffVer_32]
>+ add r5, r4
>+%else
>+ lea r5, [tab_LumaCoeffVer_32 + r4]
>+%endif
>+
>+ mov word [rsp], 2
>+
>+.loopH:
>+ mov r4d, 2
>+.loopW:
>+ PROCESS_LUMA_AVX2_16_W8_8R
>+ lea r6, [r2]
>+ mova m3, [pw_512]
>+ pmulhrsw m5, m3 ; m5 = word: row 0, row 1
>+ pmulhrsw m2, m3 ; m2 = word: row 2, row 3
>+ pmulhrsw m1, m3 ; m1 = word: row 4, row 5
>+ pmulhrsw m4, m3 ; m4 = word: row 6, row 7
>+ packuswb m5, m2
>+ packuswb m1, m4
>+ vextracti128 xm2, m5, 1
>+ vextracti128 xm4, m1, 1
>+ movq [r6], xm5
>+ movq [r6 + r3], xm2
>+ lea r6, [r6 + r3 * 2]
>+ movhps [r6], xm5
>+ movhps [r6 + r3], xm2
>+ lea r6, [r6 + r3 * 2]
>+ movq [r6], xm1
>+ movq [r6 + r3], xm4
>+ lea r6, [r6 + r3 * 2]
>+ movhps [r6], xm1
>+ movhps [r6 + r3], xm4
>+
>+ add r0, 8
>+ add r2, 8
>+ dec r4d
>+ jnz .loopW
>+ lea r2, [r2 + r3 * 8 - 16]
>+ lea r0, [r0 + r1 * 8 - 16]
>+ dec word [rsp]
>+ jnz .loopH
>+ RET
> %endif
>
> ;-------------------------------------------------------------------------------------------------------------
>_______________________________________________
>x265-devel mailing list
>x265-devel@videolan.org
>https://mailman.videolan.org/listinfo/x265-devel
</pre></div>