[x265] [PATCH] asm: filter_vsp[8x8], filter_vss[8x8] in avx2: 887c->525c, 828c->524c

Divya Manivannan divya at multicorewareinc.com
Thu Mar 12 09:32:45 CET 2015


# HG changeset patch
# User Divya Manivannan <divya at multicorewareinc.com>
# Date 1426149138 -19800
#      Thu Mar 12 14:02:18 2015 +0530
# Node ID 698ee043fd4f294b15c4e8d1bbf94ab44cd806eb
# Parent  ed3549b49cc488315da7d4709d6932e7244e5b33
asm: filter_vsp[8x8], filter_vss[8x8] in avx2: 887c->525c, 828c->524c

diff -r ed3549b49cc4 -r 698ee043fd4f source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Thu Mar 12 11:32:16 2015 +0530
+++ b/source/common/x86/asm-primitives.cpp	Thu Mar 12 14:02:18 2015 +0530
@@ -1623,8 +1623,10 @@
         p.chroma[X265_CSP_I420].pu[CHROMA_420_32x8].filter_vps = x265_interp_4tap_vert_ps_32x8_avx2;
 
         p.chroma[X265_CSP_I420].pu[CHROMA_420_4x4].filter_vsp = x265_interp_4tap_vert_sp_4x4_avx2;
+        p.chroma[X265_CSP_I420].pu[CHROMA_420_8x8].filter_vsp = x265_interp_4tap_vert_sp_8x8_avx2;
 
         p.chroma[X265_CSP_I420].pu[CHROMA_420_4x4].filter_vss = x265_interp_4tap_vert_ss_4x4_avx2;
+        p.chroma[X265_CSP_I420].pu[CHROMA_420_8x8].filter_vss = x265_interp_4tap_vert_ss_8x8_avx2;
     }
 #endif
 }
diff -r ed3549b49cc4 -r 698ee043fd4f source/common/x86/ipfilter8.asm
--- a/source/common/x86/ipfilter8.asm	Thu Mar 12 11:32:16 2015 +0530
+++ b/source/common/x86/ipfilter8.asm	Thu Mar 12 14:02:18 2015 +0530
@@ -11723,6 +11723,185 @@
 FILTER_VER_CHROMA_S_AVX2_4x4 sp
 FILTER_VER_CHROMA_S_AVX2_4x4 ss
 
+%macro FILTER_VER_CHROMA_S_AVX2_8x8 1
+INIT_YMM avx2
+cglobal interp_4tap_vert_%1_8x8, 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
+    mova            m3, [interp8_hps_shuf]
+    vpermd          m0, m3, m0
+    vextracti128    xm2, m0, 1
+    movq            [r2], xm0
+    movhps          [r2 + r3], xm0
+    movq            [r2 + r3 * 2], xm2
+    movhps          [r2 + r4], xm2
+%else
+    vpermq          m0, m0, 11011000b
+    vpermq          m2, m2, 11011000b
+    movu            [r2], xm0
+    vextracti128    xm0, m0, 1
+    vextracti128    xm3, m2, 1
+    movu            [r2 + r3], xm0
+    movu            [r2 + r3 * 2], xm2
+    movu            [r2 + r4], xm3
+%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
+    vpermd          m4, m3, m4
+    vextracti128    xm6, m4, 1
+    movq            [r2], xm4
+    movhps          [r2 + r3], xm4
+    movq            [r2 + r3 * 2], xm6
+    movhps          [r2 + r4], xm6
+%else
+    vpermq          m4, m4, 11011000b
+    vpermq          m6, m6, 11011000b
+    vextracti128    xm5, m4, 1
+    vextracti128    xm1, m6, 1
+    movu            [r2], xm4
+    movu            [r2 + r3], xm5
+    movu            [r2 + r3 * 2], xm6
+    movu            [r2 + r4], xm1
+%endif
+    RET
+%endmacro
+
+FILTER_VER_CHROMA_S_AVX2_8x8 sp
+FILTER_VER_CHROMA_S_AVX2_8x8 ss
+
 ;---------------------------------------------------------------------------------------------------------------------
 ; void interp_4tap_vertical_ss_%1x%2(int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)
 ;---------------------------------------------------------------------------------------------------------------------
diff -r ed3549b49cc4 -r 698ee043fd4f source/common/x86/ipfilter8.h
--- a/source/common/x86/ipfilter8.h	Thu Mar 12 11:32:16 2015 +0530
+++ b/source/common/x86/ipfilter8.h	Thu Mar 12 14:02:18 2015 +0530
@@ -576,6 +576,7 @@
 CHROMA_420_FILTERS(_avx2);
 CHROMA_420_SP_FILTERS(_sse2);
 CHROMA_420_SP_FILTERS_SSE4(_sse4);
+CHROMA_420_SP_FILTERS(_avx2);
 CHROMA_420_SP_FILTERS_SSE4(_avx2);
 CHROMA_420_SS_FILTERS(_sse2);
 CHROMA_420_SS_FILTERS_SSE4(_sse4);


More information about the x265-devel mailing list