[x265] [PATCH] asm: intrapred dc4 sse2 high bit

[chen]

x265_intra_pred_dc4_sse2 undefined
movzx is my fault, the value is word, so MOVD get twice value, the movzx clear high part



At 2015-02-25 04:54:05,dtyx265 at gmail.com wrote:
># HG changeset patch
># User David T Yuen <dtyx265 at gmail.com>
># Date 1424811103 28800
># Node ID 1a703601f7c8b85f1e6e680caa281b9edada89ab
># Parent  8528fea56a4202c871544621d5806ee686e671fc
>asm: intrapred dc4 sse2 high bit
>
>This replaces c code for systems using ssse3 to sse2 processors
>The code is backported from intrapred dc4 sse4 high bit
>
>./test/TestBench --testbench intrapred
>Testing only harnesses that match name <intrapred>
>Using random seed 54ECE3AB 16bpp
>Testing primitives: SSE2
>Testing primitives: SSE3
>
>Test performance improvement with full optimizations
>== intrapred primitives ==
>intra_dc_4x4[f=0]	1.20x 	 215.41   	 258.21
>intra_dc_4x4[f=1]	1.12x 	 420.52   	 470.23
>
>diff -r 8528fea56a42 -r 1a703601f7c8 source/common/x86/asm-primitives.cpp
>--- a/source/common/x86/asm-primitives.cpp	Tue Feb 24 12:39:46 2015 -0800
>+++ b/source/common/x86/asm-primitives.cpp	Tue Feb 24 12:51:43 2015 -0800
>@@ -863,6 +863,8 @@
>         ALL_LUMA_TU_S(calcresidual, getResidual, sse2);
>         ALL_LUMA_TU_S(transpose, transpose, sse2);
> 
>+        p.cu[BLOCK_4x4].intra_pred[DC_IDX] = x265_intra_pred_dc4_sse2;
>+
>         p.cu[BLOCK_4x4].sse_ss = x265_pixel_ssd_ss_4x4_mmx2;
>         ALL_LUMA_CU(sse_ss, pixel_ssd_ss, sse2);
> 
>diff -r 8528fea56a42 -r 1a703601f7c8 source/common/x86/intrapred16.asm
>--- a/source/common/x86/intrapred16.asm	Tue Feb 24 12:39:46 2015 -0800
>+++ b/source/common/x86/intrapred16.asm	Tue Feb 24 12:51:43 2015 -0800
>@@ -99,6 +99,73 @@
> ;-----------------------------------------------------------------------------------
> ; void intra_pred_dc(pixel* dst, intptr_t dstStride, pixel* above, int, int filter)
> ;-----------------------------------------------------------------------------------
>+INIT_XMM sse2
>+cglobal intra_pred_dc4, 5,6,2
>+    lea         r3,             [r2 + 18]
>+    add         r2,             2
>+
>+    movh        m0,             [r3]           ; sumAbove
>+    movh        m1,             [r2]           ; sumLeft
>+
>+    paddw       m0,             m1
>+    pshufd      m1,             m0, 0xFD
>+    paddw       m0,             m1
>+    pshuflw     m1,             m0, 1
>+    paddw       m0,             m1
>+
>+    test        r4d,            r4d
>+
>+    paddw       m0,             [pw_4]
>+    psraw       m0, 3
>+    movd        r4d,            m0             ; r4d = dc_val
>+    movzx       r4d,            r4w
>+    pshuflw     m0,             m0, 0          ; m0 = word [dc_val ...]
>+
>+    ; store DC 4x4
>+    movh        [r0],           m0
>+    movh        [r0 + r1 * 2],  m0
>+    movh        [r0 + r1 * 4],  m0
>+    lea         r5,             [r0 + r1 * 4]
>+    movh        [r5 + r1 * 2],  m0
>+
>+    ; do DC filter
>+    jz          .end
>+    lea         r5d,            [r4d * 2 + 2]  ; r5d = DC * 2 + 2
>+    add         r4d,            r5d            ; r4d = DC * 3 + 2
>+    movd        m0,             r4d
>+    pshuflw     m0,             m0, 0          ; m0 = pixDCx3
>+
>+    ; filter top
>+    movh        m1,             [r2]
>+    paddw       m1,             m0
>+    psraw       m1,             2
>+    movh        [r0],           m1             ; overwrite top-left pixel, we will update it later
>+
>+    ; filter top-left
>+    movzx       r4d, word       [r3]
>+    add         r5d,            r4d
>+    movzx       r4d, word       [r2]
>+    add         r4d,            r5d
>+    shr         r4d,            2
>+    mov         [r0],           r4w
>+
>+    ; filter left
>+    lea         r0,             [r0 + r1 * 2]
>+    movu        m1,             [r3 + 2]
>+    paddw       m1,             m0
>+    psraw       m1,             2
>+    movd        r3d,            m1
>+    mov         [r0],           r3w
>+    shr         r3d,            16
>+    mov         [r0 + r1 * 2],  r3w
>+    pextrw      r3d,  m1, 2
>+    mov         [r0 + r1 * 4],  r3w
>+.end:
>+    RET
>+
>+;-----------------------------------------------------------------------------------
>+; void intra_pred_dc(pixel* dst, intptr_t dstStride, pixel* above, int, int filter)
>+;-----------------------------------------------------------------------------------
> INIT_XMM sse4
> cglobal intra_pred_dc4, 5,6,2
>     lea         r3,             [r2 + 18]
>_______________________________________________
>x265-devel mailing list
>x265-devel at videolan.org
>https://mailman.videolan.org/listinfo/x265-devel
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.videolan.org/pipermail/x265-devel/attachments/20150225/cd214f2e/attachment-0001.html>