[x265] [PATCH 4 of 5] asm: filter_vsp[6x8], filter_vss[6x8] in avx2: 863c->653c, 823c->524c

Divya Manivannan divya at multicorewareinc.com
Thu Mar 19 06:13:56 CET 2015


# HG changeset patch
# User Divya Manivannan <divya at multicorewareinc.com>
# Date 1426740297 -19800
#      Thu Mar 19 10:14:57 2015 +0530
# Node ID ad53f152fce599c1801304a0fd1ed0c5992f834f
# Parent  db0cad2b0212b7876a62090f95390427e5cc5dcd
asm: filter_vsp[6x8], filter_vss[6x8] in avx2: 863c->653c, 823c->524c

diff -r db0cad2b0212 -r ad53f152fce5 source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Thu Mar 19 10:09:10 2015 +0530
+++ b/source/common/x86/asm-primitives.cpp	Thu Mar 19 10:14:57 2015 +0530
@@ -1695,6 +1695,7 @@
         p.chroma[X265_CSP_I420].pu[CHROMA_420_4x2].filter_vsp = x265_interp_4tap_vert_sp_4x2_avx2;
         p.chroma[X265_CSP_I420].pu[CHROMA_420_4x8].filter_vsp = x265_interp_4tap_vert_sp_4x8_avx2;
         p.chroma[X265_CSP_I420].pu[CHROMA_420_4x16].filter_vsp = x265_interp_4tap_vert_sp_4x16_avx2;
+        p.chroma[X265_CSP_I420].pu[CHROMA_420_6x8].filter_vsp = x265_interp_4tap_vert_sp_6x8_avx2;
         p.chroma[X265_CSP_I420].pu[CHROMA_420_8x2].filter_vsp = x265_interp_4tap_vert_sp_8x2_avx2;
         p.chroma[X265_CSP_I420].pu[CHROMA_420_8x4].filter_vsp = x265_interp_4tap_vert_sp_8x4_avx2;
         p.chroma[X265_CSP_I420].pu[CHROMA_420_8x6].filter_vsp = x265_interp_4tap_vert_sp_8x6_avx2;
@@ -1719,6 +1720,7 @@
         p.chroma[X265_CSP_I420].pu[CHROMA_420_4x2].filter_vss = x265_interp_4tap_vert_ss_4x2_avx2;
         p.chroma[X265_CSP_I420].pu[CHROMA_420_4x8].filter_vss = x265_interp_4tap_vert_ss_4x8_avx2;
         p.chroma[X265_CSP_I420].pu[CHROMA_420_4x16].filter_vss = x265_interp_4tap_vert_ss_4x16_avx2;
+        p.chroma[X265_CSP_I420].pu[CHROMA_420_6x8].filter_vss = x265_interp_4tap_vert_ss_6x8_avx2;
         p.chroma[X265_CSP_I420].pu[CHROMA_420_8x2].filter_vss = x265_interp_4tap_vert_ss_8x2_avx2;
         p.chroma[X265_CSP_I420].pu[CHROMA_420_8x4].filter_vss = x265_interp_4tap_vert_ss_8x4_avx2;
         p.chroma[X265_CSP_I420].pu[CHROMA_420_8x6].filter_vss = x265_interp_4tap_vert_ss_8x6_avx2;
diff -r db0cad2b0212 -r ad53f152fce5 source/common/x86/ipfilter8.asm
--- a/source/common/x86/ipfilter8.asm	Thu Mar 19 10:09:10 2015 +0530
+++ b/source/common/x86/ipfilter8.asm	Thu Mar 19 10:14:57 2015 +0530
@@ -15427,6 +15427,194 @@
 FILTER_VER_CHROMA_S_AVX2_2x8 sp
 FILTER_VER_CHROMA_S_AVX2_2x8 ss
 
+%macro FILTER_VER_CHROMA_S_AVX2_6x8 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_6x8, 4, 6, 8
+    mov             r4d, r4m
+    shl             r4d, 6
+    add             r1d, r1d
+
+%ifdef PIC
+    lea             r5, [pw_ChromaCoeffV]
+    add             r5, r4
+%else
+    lea             r5, [pw_ChromaCoeffV + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+%ifidn %1,sp
+    mova            m7, [pd_526336]
+%else
+    add             r3d, r3d
+%endif
+
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhwd       xm2, xm0, xm1
+    punpcklwd       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddwd         m0, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhwd       xm3, xm1, xm2
+    punpcklwd       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddwd         m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhwd       xm4, xm2, xm3
+    punpcklwd       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddwd         m4, m2, [r5 + 1 * mmsize]
+    pmaddwd         m2, [r5]
+    paddd           m0, m4
+    lea             r0, [r0 + r1 * 4]
+    movu            xm4, [r0]                       ; m4 = row 4
+    punpckhwd       xm5, xm3, xm4
+    punpcklwd       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddwd         m5, m3, [r5 + 1 * mmsize]
+    pmaddwd         m3, [r5]
+    paddd           m1, m5
+%ifidn %1,sp
+    paddd           m0, m7
+    paddd           m1, m7
+    psrad           m0, 12
+    psrad           m1, 12
+%else
+    psrad           m0, 6
+    psrad           m1, 6
+%endif
+    packssdw        m0, m1
+
+    movu            xm5, [r0 + r1]                  ; m5 = row 5
+    punpckhwd       xm6, xm4, xm5
+    punpcklwd       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddwd         m6, m4, [r5 + 1 * mmsize]
+    paddd           m2, m6
+    pmaddwd         m4, [r5]
+    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
+    punpckhwd       xm1, xm5, xm6
+    punpcklwd       xm5, xm6
+    vinserti128     m5, m5, xm1, 1
+    pmaddwd         m1, m5, [r5 + 1 * mmsize]
+    pmaddwd         m5, [r5]
+    paddd           m3, m1
+%ifidn %1,sp
+    paddd           m2, m7
+    paddd           m3, m7
+    psrad           m2, 12
+    psrad           m3, 12
+%else
+    psrad           m2, 6
+    psrad           m3, 6
+%endif
+    packssdw        m2, m3
+
+    movu            xm1, [r0 + r4]                  ; m1 = row 7
+    punpckhwd       xm3, xm6, xm1
+    punpcklwd       xm6, xm1
+    vinserti128     m6, m6, xm3, 1
+    pmaddwd         m3, m6, [r5 + 1 * mmsize]
+    pmaddwd         m6, [r5]
+    paddd           m4, m3
+
+    lea             r4, [r3 * 3]
+%ifidn %1,sp
+    packuswb        m0, m2
+    vextracti128    xm2, m0, 1
+    movd            [r2], xm0
+    pextrw          [r2 + 4], xm2, 0
+    pextrd          [r2 + r3], xm0, 1
+    pextrw          [r2 + r3 + 4], xm2, 2
+    pextrd          [r2 + r3 * 2], xm0, 2
+    pextrw          [r2 + r3 * 2 + 4], xm2, 4
+    pextrd          [r2 + r4], xm0, 3
+    pextrw          [r2 + r4 + 4], xm2, 6
+%else
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
+    movq            [r2 + r3 * 2], xm2
+    movhps          [r2 + r4], xm2
+    vextracti128    xm0, m0, 1
+    vextracti128    xm3, m2, 1
+    movd            [r2 + 8], xm0
+    pextrd          [r2 + r3 + 8], xm0, 2
+    movd            [r2 + r3 * 2 + 8], xm3
+    pextrd          [r2 + r4 + 8], xm3, 2
+%endif
+    lea             r2, [r2 + r3 * 4]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm0, [r0]                       ; m0 = row 8
+    punpckhwd       xm2, xm1, xm0
+    punpcklwd       xm1, xm0
+    vinserti128     m1, m1, xm2, 1
+    pmaddwd         m2, m1, [r5 + 1 * mmsize]
+    pmaddwd         m1, [r5]
+    paddd           m5, m2
+%ifidn %1,sp
+    paddd           m4, m7
+    paddd           m5, m7
+    psrad           m4, 12
+    psrad           m5, 12
+%else
+    psrad           m4, 6
+    psrad           m5, 6
+%endif
+    packssdw        m4, m5
+
+    movu            xm2, [r0 + r1]                  ; m2 = row 9
+    punpckhwd       xm5, xm0, xm2
+    punpcklwd       xm0, xm2
+    vinserti128     m0, m0, xm5, 1
+    pmaddwd         m0, [r5 + 1 * mmsize]
+    paddd           m6, m0
+    movu            xm5, [r0 + r1 * 2]              ; m5 = row 10
+    punpckhwd       xm0, xm2, xm5
+    punpcklwd       xm2, xm5
+    vinserti128     m2, m2, xm0, 1
+    pmaddwd         m2, [r5 + 1 * mmsize]
+    paddd           m1, m2
+
+%ifidn %1,sp
+    paddd           m6, m7
+    paddd           m1, m7
+    psrad           m6, 12
+    psrad           m1, 12
+%else
+    psrad           m6, 6
+    psrad           m1, 6
+%endif
+    packssdw        m6, m1
+%ifidn %1,sp
+    packuswb        m4, m6
+    vextracti128    xm6, m4, 1
+    movd            [r2], xm4
+    pextrw          [r2 + 4], xm6, 0
+    pextrd          [r2 + r3], xm4, 1
+    pextrw          [r2 + r3 + 4], xm6, 2
+    pextrd          [r2 + r3 * 2], xm4, 2
+    pextrw          [r2 + r3 * 2 + 4], xm6, 4
+    pextrd          [r2 + r4], xm4, 3
+    pextrw          [r2 + r4 + 4], xm6, 6
+%else
+    movq            [r2], xm4
+    movhps          [r2 + r3], xm4
+    movq            [r2 + r3 * 2], xm6
+    movhps          [r2 + r4], xm6
+    vextracti128    xm5, m4, 1
+    vextracti128    xm1, m6, 1
+    movd            [r2 + 8], xm5
+    pextrd          [r2 + r3 + 8], xm5, 2
+    movd            [r2 + r3 * 2 + 8], xm1
+    pextrd          [r2 + r4 + 8], xm1, 2
+%endif
+    RET
+%endmacro
+
+FILTER_VER_CHROMA_S_AVX2_6x8 sp
+FILTER_VER_CHROMA_S_AVX2_6x8 ss
+
 ;---------------------------------------------------------------------------------------------------------------------
 ; void interp_4tap_vertical_ss_%1x%2(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
 ;---------------------------------------------------------------------------------------------------------------------


More information about the x265-devel mailing list