[x265] [PATCH Review only] asm: chroma_vpp[8x8] for color space i420 in avx2: improve 338c->269c

Divya Manivannan divya at multicorewareinc.com
Fri Nov 28 09:39:13 CET 2014


# HG changeset patch
# User Divya Manivannan <divya at multicorewareinc.com>
# Date 1417163641 -19800
#      Fri Nov 28 14:04:01 2014 +0530
# Node ID 1b61b5eb52e67feda773f3cd5b367b23d3d6774c
# Parent  d32249002258d65092674c7782ea022ae17b75d4
asm: chroma_vpp[8x8] for color space i420 in avx2: improve 338c->269c

diff -r d32249002258 -r 1b61b5eb52e6 source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Fri Nov 28 11:10:05 2014 +0530
+++ b/source/common/x86/asm-primitives.cpp	Fri Nov 28 14:04:01 2014 +0530
@@ -1847,6 +1847,8 @@
         p.luma_vpp[LUMA_4x4] = x265_interp_8tap_vert_pp_4x4_avx2;
         p.luma_vpp[LUMA_8x4] = x265_interp_8tap_vert_pp_8x4_avx2;
         p.luma_vpp[LUMA_8x8] = x265_interp_8tap_vert_pp_8x8_avx2;
+
+        p.chroma[X265_CSP_I420].filter_vpp[CHROMA_8x8] = x265_interp_4tap_vert_pp_8x8_avx2;
     }
 #endif // if HIGH_BIT_DEPTH
 }
diff -r d32249002258 -r 1b61b5eb52e6 source/common/x86/ipfilter8.asm
--- a/source/common/x86/ipfilter8.asm	Fri Nov 28 11:10:05 2014 +0530
+++ b/source/common/x86/ipfilter8.asm	Fri Nov 28 14:04:01 2014 +0530
@@ -143,6 +143,31 @@
                      times 16 db 58, -10
                      times 16 db 4, -1
 
+ALIGN 32
+tab_ChromaCoeffVer_32: times 16 db 0, 64
+                       times 16 db 0, 0
+
+                       times 16 db -2, 58
+                       times 16 db 10, -2
+
+                       times 16 db -4, 54
+                       times 16 db 16, -2
+
+                       times 16 db -6, 46
+                       times 16 db 28, -4
+
+                       times 16 db -4, 36
+                       times 16 db 36, -4
+
+                       times 16 db -4, 28
+                       times 16 db 46, -6
+
+                       times 16 db -2, 16
+                       times 16 db 54, -4
+
+                       times 16 db -2, 10
+                       times 16 db 58, -2
+
 tab_c_64_n64:   times 8 db 64, -64
 
 const interp4_shuf, times 2 db 0, 1, 8, 9, 4, 5, 12, 13, 2, 3, 10, 11, 6, 7, 14, 15
@@ -1901,6 +1926,85 @@
 
 RET
 
+%macro PROCESS_CHROMA_AVX2_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]
+    movq            xm3, [r0 + r1 * 2]              ; 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, [r0 + r4]                  ; m4 = row 3
+    punpcklbw       xm3, xm4                        ; m3 = [37 27 36 26 35 25 34 24 33 23 32 22 31 21 30 20]
+    lea             r0, [r0 + r1 * 4]
+    movq            xm1, [r0]                       ; 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 + 1 * mmsize]
+    paddw           m5, m0
+    pmaddubsw       m2, [r5]
+    movq            xm3, [r0 + r1]                  ; m3 = row 5
+    punpcklbw       xm1, xm3                        ; m1 = [57 47 56 46 55 45 54 44 53 43 52 42 51 41 50 40]
+    movq            xm4, [r0 + r1 * 2]              ; 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       m0, m1, [r5 + 1 * mmsize]
+    paddw           m2, m0
+    pmaddubsw       m1, [r5]
+    movq            xm3, [r0 + r4]                  ; m3 = row 7
+    punpcklbw       xm4, xm3                        ; m4 = [77 67 76 66 75 65 74 64 73 63 72 62 71 61 70 60]
+    lea             r0, [r0 + r1 * 4]
+    movq            xm0, [r0]                       ; 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 + 1 * mmsize]
+    paddw           m1, m3
+    pmaddubsw       m4, [r5]
+    movq            xm3, [r0 + r1]                  ; m3 = row 9
+    punpcklbw       xm0, xm3                        ; m0 = [97 87 96 86 95 85 94 84 93 83 92 82 91 81 90 80]
+    movq            xm6, [r0 + r1 * 2]              ; 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       m0, [r5 + 1 * mmsize]
+    paddw           m4, m0
+%endmacro
+
+INIT_YMM avx2
+cglobal interp_4tap_vert_pp_8x8, 4, 6, 7
+    mov             r4d, r4m
+    shl             r4d, 6
+
+%ifdef PIC
+    lea             r5, [tab_ChromaCoeffVer_32]
+    add             r5, r4
+%else
+    lea             r5, [tab_ChromaCoeffVer_32 + r4]
+%endif
+
+    lea             r4, [r1 * 3]
+    sub             r0, r1
+    PROCESS_CHROMA_AVX2_W8_8R
+    lea             r4, [r3 * 3]
+    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            [r2], xm5
+    movq            [r2 + r3], xm2
+    movhps          [r2 + r3 * 2], xm5
+    movhps          [r2 + r4], xm2
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm1
+    movq            [r2 + r3], xm4
+    movhps          [r2 + r3 * 2], xm1
+    movhps          [r2 + r4], xm4
+    RET
+
 ;-----------------------------------------------------------------------------
 ; void interp_4tap_vert_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
 ;-----------------------------------------------------------------------------
diff -r d32249002258 -r 1b61b5eb52e6 source/common/x86/ipfilter8.h
--- a/source/common/x86/ipfilter8.h	Fri Nov 28 11:10:05 2014 +0530
+++ b/source/common/x86/ipfilter8.h	Fri Nov 28 14:04:01 2014 +0530
@@ -573,6 +573,7 @@
     SETUP_CHROMA_SS_FUNC_DEF(16, 64, cpu);
 
 CHROMA_FILTERS(_sse4);
+CHROMA_FILTERS(_avx2);
 CHROMA_SP_FILTERS(_sse2);
 CHROMA_SP_FILTERS_SSE4(_sse4);
 CHROMA_SS_FILTERS(_sse2);


More information about the x265-devel mailing list