[x265-commits] [x265] x265: remove validateFanout

Deepthi Nandakumar deepthi at multicorewareinc.com
Tue Dec 2 19:37:04 CET 2014


details:   http://hg.videolan.org/x265/rev/d9f835ddd112
branches:  
changeset: 8925:d9f835ddd112
user:      Deepthi Nandakumar <deepthi at multicorewareinc.com>
date:      Sun Nov 30 22:17:46 2014 +0530
description:
x265: remove validateFanout
Subject: [x265] asm: luma_vpp[48x64] in avx2: improve 21298c->14696c

details:   http://hg.videolan.org/x265/rev/984271a3aae9
branches:  
changeset: 8926:984271a3aae9
user:      Divya Manivannan <divya at multicorewareinc.com>
date:      Thu Nov 27 09:10:47 2014 +0530
description:
asm: luma_vpp[48x64] in avx2: improve 21298c->14696c
Subject: [x265] asm: luma_vpp[24x32] in avx2: improve 5637c->3695c

details:   http://hg.videolan.org/x265/rev/e1ca311bbb5b
branches:  
changeset: 8927:e1ca311bbb5b
user:      Divya Manivannan <divya at multicorewareinc.com>
date:      Thu Nov 27 09:36:00 2014 +0530
description:
asm: luma_vpp[24x32] in avx2: improve 5637c->3695c
Subject: [x265] asm: luma_vpp[12x16] in avx2: improve 1977c->1418c

details:   http://hg.videolan.org/x265/rev/e280ce2e5076
branches:  
changeset: 8928:e280ce2e5076
user:      Divya Manivannan <divya at multicorewareinc.com>
date:      Thu Nov 27 10:40:13 2014 +0530
description:
asm: luma_vpp[12x16] in avx2: improve 1977c->1418c
Subject: [x265] asm: luma_vpp[8x16, 8x32] in avx2: improve 1139c->774c, 1968c->1452c

details:   http://hg.videolan.org/x265/rev/5ee693e4b5fa
branches:  
changeset: 8929:5ee693e4b5fa
user:      Divya Manivannan <divya at multicorewareinc.com>
date:      Thu Nov 27 11:19:16 2014 +0530
description:
asm: luma_vpp[8x16, 8x32] in avx2: improve 1139c->774c, 1968c->1452c
Subject: [x265] encoder: free csv file name

details:   http://hg.videolan.org/x265/rev/bde1753de250
branches:  
changeset: 8930:bde1753de250
user:      Deepthi Nandakumar <deepthi at multicorewareinc.com>
date:      Mon Dec 01 16:59:57 2014 +0530
description:
encoder: free csv file name

Since strdup is used uniformly for filenames, csvfn cannot be const.
Subject: [x265] x265: add ratetol to command line help

details:   http://hg.videolan.org/x265/rev/f636a0aadd68
branches:  
changeset: 8931:f636a0aadd68
user:      Deepthi Nandakumar <deepthi at multicorewareinc.com>
date:      Tue Dec 02 12:14:32 2014 +0530
description:
x265: add ratetol to command line help
Subject: [x265] Merge

details:   http://hg.videolan.org/x265/rev/de54cffaecf2
branches:  
changeset: 8932:de54cffaecf2
user:      Steve Borho <steve at borho.org>
date:      Tue Dec 02 12:36:41 2014 -0600
description:
Merge

diffstat:

 source/CMakeLists.txt                |    2 +-
 source/common/x86/asm-primitives.cpp |    9 +
 source/common/x86/ipfilter8.asm      |  577 +++++++++++++++++++++++++++++++++++
 source/encoder/encoder.cpp           |    2 +-
 source/x265.cpp                      |   87 +-----
 source/x265.h                        |    2 +-
 6 files changed, 590 insertions(+), 89 deletions(-)

diffs (truncated from 801 to 300 lines):

diff -r 90401d77a05d -r de54cffaecf2 source/CMakeLists.txt
--- a/source/CMakeLists.txt	Thu Nov 27 10:12:03 2014 +0900
+++ b/source/CMakeLists.txt	Tue Dec 02 12:36:41 2014 -0600
@@ -21,7 +21,7 @@ include(CheckSymbolExists)
 include(CheckCXXCompilerFlag)
 
 # X265_BUILD must be incremented each time the public API is changed
-set(X265_BUILD 38)
+set(X265_BUILD 39)
 configure_file("${PROJECT_SOURCE_DIR}/x265.def.in"
                "${PROJECT_BINARY_DIR}/x265.def")
 configure_file("${PROJECT_SOURCE_DIR}/x265_config.h.in"
diff -r 90401d77a05d -r de54cffaecf2 source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Thu Nov 27 10:12:03 2014 +0900
+++ b/source/common/x86/asm-primitives.cpp	Tue Dec 02 12:36:41 2014 -0600
@@ -1822,6 +1822,8 @@ void Setup_Assembly_Primitives(EncoderPr
         p.transpose[BLOCK_32x32] = x265_transpose32_avx2;
         p.transpose[BLOCK_64x64] = x265_transpose64_avx2;
 
+        p.luma_vpp[LUMA_12x16] = x265_interp_8tap_vert_pp_12x16_avx2;
+
         p.luma_vpp[LUMA_16x4] = x265_interp_8tap_vert_pp_16x4_avx2;
         p.luma_vpp[LUMA_16x8] = x265_interp_8tap_vert_pp_16x8_avx2;
         p.luma_vpp[LUMA_16x12] = x265_interp_8tap_vert_pp_16x12_avx2;
@@ -1829,12 +1831,16 @@ void Setup_Assembly_Primitives(EncoderPr
         p.luma_vpp[LUMA_16x32] = x265_interp_8tap_vert_pp_16x32_avx2;
         p.luma_vpp[LUMA_16x64] = x265_interp_8tap_vert_pp_16x64_avx2;
 
+        p.luma_vpp[LUMA_24x32] = x265_interp_8tap_vert_pp_24x32_avx2;
+
         p.luma_vpp[LUMA_32x8] = x265_interp_8tap_vert_pp_32x8_avx2;
         p.luma_vpp[LUMA_32x16] = x265_interp_8tap_vert_pp_32x16_avx2;
         p.luma_vpp[LUMA_32x24] = x265_interp_8tap_vert_pp_32x24_avx2;
         p.luma_vpp[LUMA_32x32] = x265_interp_8tap_vert_pp_32x32_avx2;
         p.luma_vpp[LUMA_32x64] = x265_interp_8tap_vert_pp_32x64_avx2;
 
+        p.luma_vpp[LUMA_48x64] = x265_interp_8tap_vert_pp_48x64_avx2;
+
         p.luma_vpp[LUMA_64x16] = x265_interp_8tap_vert_pp_64x16_avx2;
         p.luma_vpp[LUMA_64x32] = x265_interp_8tap_vert_pp_64x32_avx2;
         p.luma_vpp[LUMA_64x48] = x265_interp_8tap_vert_pp_64x48_avx2;
@@ -1868,8 +1874,11 @@ void Setup_Assembly_Primitives(EncoderPr
         p.luma_hpp[LUMA_48x64] = x265_interp_8tap_horiz_pp_48x64_avx2;
 
         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.luma_vpp[LUMA_8x16] = x265_interp_8tap_vert_pp_8x16_avx2;
+        p.luma_vpp[LUMA_8x32] = x265_interp_8tap_vert_pp_8x32_avx2;
     }
 #endif // if HIGH_BIT_DEPTH
 }
diff -r 90401d77a05d -r de54cffaecf2 source/common/x86/ipfilter8.asm
--- a/source/common/x86/ipfilter8.asm	Thu Nov 27 10:12:03 2014 +0900
+++ b/source/common/x86/ipfilter8.asm	Tue Dec 02 12:36:41 2014 -0600
@@ -4250,6 +4250,52 @@ cglobal interp_8tap_vert_%3_%1x%2, 5, 7,
     RET
 %endmacro
 
+%macro FILTER_VER_LUMA_AVX2_8xN 2
+INIT_YMM avx2
+cglobal interp_8tap_vert_pp_%1x%2, 4, 7, 8, 0-gprsize
+    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, [r1 * 4]
+    mov             word [rsp], %2 / 8
+    mova            m7, [pw_512]
+
+.loop:
+    PROCESS_LUMA_AVX2_W8_8R
+    pmulhrsw        m5, m7                          ; m5 = word: row 0, row 1
+    pmulhrsw        m2, m7                          ; m2 = word: row 2, row 3
+    pmulhrsw        m1, m7                          ; m1 = word: row 4, row 5
+    pmulhrsw        m4, m7                          ; 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
+    lea             r2, [r2 + r3 * 2]
+    movhps          [r2], xm5
+    movhps          [r2 + r3], xm2
+    lea             r2, [r2 + r3 * 2]
+    movq            [r2], xm1
+    movq            [r2 + r3], xm4
+    lea             r2, [r2 + r3 * 2]
+    movhps          [r2], xm1
+    movhps          [r2 + r3], xm4
+    lea             r2, [r2 + r3 * 2]
+    sub             r0, r6
+    dec             word [rsp]
+    jnz             .loop
+    RET
+%endmacro
+
 INIT_YMM avx2
 cglobal interp_8tap_vert_pp_8x8, 4, 6, 7
     mov             r4d, r4m
@@ -4327,11 +4373,13 @@ FILTER_VER_LUMA_8xN 8, 8, pp
 ; void interp_8tap_vert_pp_8x16(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
 ;-------------------------------------------------------------------------------------------------------------
 FILTER_VER_LUMA_8xN 8, 16, pp
+FILTER_VER_LUMA_AVX2_8xN 8, 16
 
 ;-------------------------------------------------------------------------------------------------------------
 ; void interp_8tap_vert_pp_8x32(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
 ;-------------------------------------------------------------------------------------------------------------
 FILTER_VER_LUMA_8xN 8, 32, pp
+FILTER_VER_LUMA_AVX2_8xN 8, 32
 
 ;-------------------------------------------------------------------------------------------------------------
 ; void interp_8tap_vert_ps_8x4(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
@@ -4469,6 +4517,313 @@ FILTER_VER_LUMA_12xN 12, 16, ps
 
 INIT_YMM avx2
 %if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_pp_12x16, 4, 7, 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]
+    mova            m14, [pw_512]
+
+    movu            xm0, [r0]                       ; m0 = row 0
+    movu            xm1, [r0 + r1]                  ; m1 = row 1
+    punpckhbw       xm2, xm0, xm1
+    punpcklbw       xm0, xm1
+    vinserti128     m0, m0, xm2, 1
+    pmaddubsw       m0, [r5]
+    movu            xm2, [r0 + r1 * 2]              ; m2 = row 2
+    punpckhbw       xm3, xm1, xm2
+    punpcklbw       xm1, xm2
+    vinserti128     m1, m1, xm3, 1
+    pmaddubsw       m1, [r5]
+    movu            xm3, [r0 + r4]                  ; m3 = row 3
+    punpckhbw       xm4, xm2, xm3
+    punpcklbw       xm2, xm3
+    vinserti128     m2, m2, xm4, 1
+    pmaddubsw       m4, m2, [r5 + 1 * mmsize]
+    paddw           m0, m4
+    pmaddubsw       m2, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm4, [r0]                       ; m4 = row 4
+    punpckhbw       xm5, xm3, xm4
+    punpcklbw       xm3, xm4
+    vinserti128     m3, m3, xm5, 1
+    pmaddubsw       m5, m3, [r5 + 1 * mmsize]
+    paddw           m1, m5
+    pmaddubsw       m3, [r5]
+    movu            xm5, [r0 + r1]                  ; m5 = row 5
+    punpckhbw       xm6, xm4, xm5
+    punpcklbw       xm4, xm5
+    vinserti128     m4, m4, xm6, 1
+    pmaddubsw       m6, m4, [r5 + 2 * mmsize]
+    paddw           m0, m6
+    pmaddubsw       m6, m4, [r5 + 1 * mmsize]
+    paddw           m2, m6
+    pmaddubsw       m4, [r5]
+    movu            xm6, [r0 + r1 * 2]              ; m6 = row 6
+    punpckhbw       xm7, xm5, xm6
+    punpcklbw       xm5, xm6
+    vinserti128     m5, m5, xm7, 1
+    pmaddubsw       m7, m5, [r5 + 2 * mmsize]
+    paddw           m1, m7
+    pmaddubsw       m7, m5, [r5 + 1 * mmsize]
+    paddw           m3, m7
+    pmaddubsw       m5, [r5]
+    movu            xm7, [r0 + r4]                  ; m7 = row 7
+    punpckhbw       xm8, xm6, xm7
+    punpcklbw       xm6, xm7
+    vinserti128     m6, m6, xm8, 1
+    pmaddubsw       m8, m6, [r5 + 3 * mmsize]
+    paddw           m0, m8
+    pmaddubsw       m8, m6, [r5 + 2 * mmsize]
+    paddw           m2, m8
+    pmaddubsw       m8, m6, [r5 + 1 * mmsize]
+    paddw           m4, m8
+    pmaddubsw       m6, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm8, [r0]                       ; m8 = row 8
+    punpckhbw       xm9, xm7, xm8
+    punpcklbw       xm7, xm8
+    vinserti128     m7, m7, xm9, 1
+    pmaddubsw       m9, m7, [r5 + 3 * mmsize]
+    paddw           m1, m9
+    pmaddubsw       m9, m7, [r5 + 2 * mmsize]
+    paddw           m3, m9
+    pmaddubsw       m9, m7, [r5 + 1 * mmsize]
+    paddw           m5, m9
+    pmaddubsw       m7, [r5]
+    movu            xm9, [r0 + r1]                  ; m9 = row 9
+    punpckhbw       xm10, xm8, xm9
+    punpcklbw       xm8, xm9
+    vinserti128     m8, m8, xm10, 1
+    pmaddubsw       m10, m8, [r5 + 3 * mmsize]
+    paddw           m2, m10
+    pmaddubsw       m10, m8, [r5 + 2 * mmsize]
+    paddw           m4, m10
+    pmaddubsw       m10, m8, [r5 + 1 * mmsize]
+    paddw           m6, m10
+    pmaddubsw       m8, [r5]
+    movu            xm10, [r0 + r1 * 2]             ; m10 = row 10
+    punpckhbw       xm11, xm9, xm10
+    punpcklbw       xm9, xm10
+    vinserti128     m9, m9, xm11, 1
+    pmaddubsw       m11, m9, [r5 + 3 * mmsize]
+    paddw           m3, m11
+    pmaddubsw       m11, m9, [r5 + 2 * mmsize]
+    paddw           m5, m11
+    pmaddubsw       m11, m9, [r5 + 1 * mmsize]
+    paddw           m7, m11
+    pmaddubsw       m9, [r5]
+    movu            xm11, [r0 + r4]                 ; m11 = row 11
+    punpckhbw       xm12, xm10, xm11
+    punpcklbw       xm10, xm11
+    vinserti128     m10, m10, xm12, 1
+    pmaddubsw       m12, m10, [r5 + 3 * mmsize]
+    paddw           m4, m12
+    pmaddubsw       m12, m10, [r5 + 2 * mmsize]
+    paddw           m6, m12
+    pmaddubsw       m12, m10, [r5 + 1 * mmsize]
+    paddw           m8, m12
+    pmaddubsw       m10, [r5]
+    lea             r0, [r0 + r1 * 4]
+    movu            xm12, [r0]                      ; m12 = row 12
+    punpckhbw       xm13, xm11, xm12
+    punpcklbw       xm11, xm12
+    vinserti128     m11, m11, xm13, 1
+    pmaddubsw       m13, m11, [r5 + 3 * mmsize]
+    paddw           m5, m13
+    pmaddubsw       m13, m11, [r5 + 2 * mmsize]
+    paddw           m7, m13
+    pmaddubsw       m13, m11, [r5 + 1 * mmsize]
+    paddw           m9, m13
+    pmaddubsw       m11, [r5]
+
+    pmulhrsw        m0, m14                         ; m0 = word: row 0
+    pmulhrsw        m1, m14                         ; m1 = word: row 1
+    pmulhrsw        m2, m14                         ; m2 = word: row 2
+    pmulhrsw        m3, m14                         ; m3 = word: row 3
+    pmulhrsw        m4, m14                         ; m4 = word: row 4
+    pmulhrsw        m5, m14                         ; m5 = word: row 5
+    packuswb        m0, m1
+    packuswb        m2, m3
+    packuswb        m4, m5
+    vpermq          m0, m0, 11011000b
+    vpermq          m2, m2, 11011000b
+    vpermq          m4, m4, 11011000b
+    vextracti128    xm1, m0, 1
+    vextracti128    xm3, m2, 1
+    vextracti128    xm5, m4, 1
+    movq            [r2], xm0
+    pextrd          [r2 + 8], xm0, 2
+    movq            [r2 + r3], xm1
+    pextrd          [r2 + r3 + 8], xm1, 2
+    movq            [r2 + r3 * 2], xm2
+    pextrd          [r2 + r3 * 2 + 8], xm2, 2
+    movq            [r2 + r6], xm3
+    pextrd          [r2 + r6 + 8], xm3, 2
+    lea             r2, [r2 + r3 * 4]
+    movq            [r2], xm4
+    pextrd          [r2 + 8], xm4, 2
+    movq            [r2 + r3], xm5
+    pextrd          [r2 + r3 + 8], xm5, 2
+
+    movu            xm13, [r0 + r1]                 ; m13 = row 13
+    punpckhbw       xm0, xm12, xm13
+    punpcklbw       xm12, xm13
+    vinserti128     m12, m12, xm0, 1
+    pmaddubsw       m0, m12, [r5 + 3 * mmsize]
+    paddw           m6, m0
+    pmaddubsw       m0, m12, [r5 + 2 * mmsize]
+    paddw           m8, m0
+    pmaddubsw       m0, m12, [r5 + 1 * mmsize]
+    paddw           m10, m0
+    pmaddubsw       m12, [r5]
+    movu            xm0, [r0 + r1 * 2]              ; m0 = row 14
+    punpckhbw       xm1, xm13, xm0
+    punpcklbw       xm13, xm0
+    vinserti128     m13, m13, xm1, 1


More information about the x265-commits mailing list