[x265] [PATCH] asm: luma_vpp[16x32, 16x64] in avx2: improve 3875c->2463c, 7499c->4894c

[chen]

At 2014-11-24 13:04:35,"Divya Manivannan" <divya at multicorewareinc.com> wrote:
># HG changeset patch
># User Divya Manivannan <divya at multicorewareinc.com>
># Date 1416805422 -19800
>#      Mon Nov 24 10:33:42 2014 +0530
># Node ID 9dfa1886c8802073a7d56db85c896a641c54d98a
># Parent  3c6f703f94eae26ac221c6cfd2ae850fb5b6f94b
>asm: luma_vpp[16x32, 16x64] in avx2: improve 3875c->2463c, 7499c->4894c
>
>diff -r 3c6f703f94ea -r 9dfa1886c880 source/common/x86/asm-primitives.cpp
>--- a/source/common/x86/asm-primitives.cpp	Fri Nov 21 11:31:28 2014 +0530
>+++ b/source/common/x86/asm-primitives.cpp	Mon Nov 24 10:33:42 2014 +0530
>@@ -1802,6 +1802,8 @@
>         p.luma_vpp[LUMA_16x8] = x265_interp_8tap_vert_pp_16x8_avx2;
>         p.luma_vpp[LUMA_16x12] = x265_interp_8tap_vert_pp_16x12_avx2;
>         p.luma_vpp[LUMA_16x16] = x265_interp_8tap_vert_pp_16x16_avx2;
>+        p.luma_vpp[LUMA_16x32] = x265_interp_8tap_vert_pp_16x32_avx2;
>+        p.luma_vpp[LUMA_16x64] = x265_interp_8tap_vert_pp_16x64_avx2;
> #endif
>         p.luma_hpp[LUMA_4x4] = x265_interp_8tap_horiz_pp_4x4_avx2;
>         p.luma_vpp[LUMA_4x4] = x265_interp_8tap_vert_pp_4x4_avx2;
>diff -r 3c6f703f94ea -r 9dfa1886c880 source/common/x86/ipfilter8.asm
>--- a/source/common/x86/ipfilter8.asm	Fri Nov 21 11:31:28 2014 +0530
>+++ b/source/common/x86/ipfilter8.asm	Mon Nov 24 10:33:42 2014 +0530
>@@ -4753,6 +4753,309 @@
>     RET
> %endif
> 
>+%macro FILTER_VER_LUMA_AVX2_16xN 2
>+INIT_YMM avx2
>+%if ARCH_X86_64 == 1
>+cglobal interp_8tap_vert_pp_%1x%2, 4, 9, 15
>+    mov             r4d, r4m
>+    shl             r4d, 7
>+
>+%ifdef PIC
>+    lea             r5, [tab_LumaCoeffVer_32]
>+    add             r5, r4
>+%else
>+    lea             r5, [tab_LumaCoeffVer_32 + r4]
>+%endif
>+
>+    lea             r4, [r1 * 3]
>+    sub             r0, r4
>+    lea             r6, [r3 * 3]
>+    lea             r7, [r1 * 4]
>+    mova            m14, [pw_512]
>+    mov             r8w, %2 / 16

I will verify code tomorrow morning, here have a little issue.
r8w is right, but more prefix byte, I suggest use r8d
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.videolan.org/pipermail/x265-devel/attachments/20141124/03f45a37/attachment.html>