<div dir="ltr"><div>Hi min,</div><div>Thanks for the feedback. </div>Testbench passes an input array to this function called 'short_test_buff' which is initialized to min value -1 << 12 and max value 1 << 12 which are 13 bit and 16 bit values rspv.<div>My best guess for dynamic range is 16+16+3+4 = 39 . So, I think we should use QWORD. </div></div><div class="gmail_extra"><br clear="all"><div><div class="gmail_signature"><div dir="ltr"><div><div dir="ltr"><div><div><span style="color:rgb(56,118,29)"><br></span></div><div><span style="color:rgb(56,118,29)">Thank you<br></span></div><span style="color:rgb(56,118,29)">Regards<br></span></div><span style="color:rgb(56,118,29)">Ramya</span><br></div></div></div></div></div>
<br><div class="gmail_quote">On Mon, Sep 21, 2015 at 9:50 PM, chen <span dir="ltr"><<a href="mailto:chenm003@163.com" target="_blank">chenm003@163.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div style="line-height:1.7;color:#000000;font-size:14px;font-family:arial"><div><br></div><pre><div><div class="h5"><br>At 2015-09-18 12:30:06,<a href="mailto:ramya@multicorewareinc.com" target="_blank">ramya@multicorewareinc.com</a> wrote:
># HG changeset patch
># User Ramya Sriraman <<a href="mailto:ramya@multicorewareinc.com" target="_blank">ramya@multicorewareinc.com</a>>
># Date 1442550588 -19800
># Fri Sep 18 09:59:48 2015 +0530
># Node ID 2cca9810882147d5aece67e22403d5d40f768024
># Parent 8db83511da0b11b7347adea081269e3591029841
>asm: fix sse_ss for [16x16], [32x32] and [64x64] sse2 12bpp
>
>diff -r 8db83511da0b -r 2cca98108821 source/common/x86/asm-primitives.cpp
>--- a/source/common/x86/asm-primitives.cpp Mon Sep 14 09:28:07 2015 +0530
>+++ b/source/common/x86/asm-primitives.cpp Fri Sep 18 09:59:48 2015 +0530
>@@ -1002,10 +1002,12 @@
> p.chroma[X265_CSP_I422].cu[BLOCK_422_8x16].sse_pp = (pixel_sse_t)PFX(pixel_ssd_ss_8x16_sse2);
> p.chroma[X265_CSP_I422].cu[BLOCK_422_16x32].sse_pp = (pixel_sse_t)PFX(pixel_ssd_ss_16x32_sse2);
> p.chroma[X265_CSP_I422].cu[BLOCK_422_32x64].sse_pp = (pixel_sse_t)PFX(pixel_ssd_ss_32x64_sse2);
>-#if X265_DEPTH <= 10
>- <a href="http://p.cu" target="_blank">p.cu</a>[BLOCK_4x4].sse_ss = PFX(pixel_ssd_ss_4x4_mmx2);
>- ALL_LUMA_CU(sse_ss, pixel_ssd_ss, sse2);
>-#endif
>+ <a href="http://p.cu" target="_blank">p.cu</a>[BLOCK_4x4].sse_ss = (pixel_sse_ss_t)PFX(pixel_ssd_ss_4x4_mmx2);
>+ <a href="http://p.cu" target="_blank">p.cu</a>[BLOCK_8x8].sse_ss = (pixel_sse_ss_t)PFX(pixel_ssd_ss_8x8_sse2);
>+ <a href="http://p.cu" target="_blank">p.cu</a>[BLOCK_16x16].sse_ss = (pixel_sse_ss_t)PFX(pixel_ssd_ss_16x16_sse2);
>+ <a href="http://p.cu" target="_blank">p.cu</a>[BLOCK_32x32].sse_ss = (pixel_sse_ss_t)PFX(pixel_ssd_ss_32x32_sse2);
>+ <a href="http://p.cu" target="_blank">p.cu</a>[BLOCK_64x64].sse_ss = (pixel_sse_ss_t)PFX(pixel_ssd_ss_64x64_sse2);
>+
> <a href="http://p.cu" target="_blank">p.cu</a>[BLOCK_4x4].dct = PFX(dct4_sse2);
> <a href="http://p.cu" target="_blank">p.cu</a>[BLOCK_8x8].dct = PFX(dct8_sse2);
> <a href="http://p.cu" target="_blank">p.cu</a>[BLOCK_4x4].idct = PFX(idct4_sse2);
>diff -r 8db83511da0b -r 2cca98108821 source/common/x86/ssd-a.asm
>--- a/source/common/x86/ssd-a.asm Mon Sep 14 09:28:07 2015 +0530
>+++ b/source/common/x86/ssd-a.asm Fri Sep 18 09:59:48 2015 +0530
>@@ -181,6 +181,197 @@
> RET
> %endmacro
>
>+;Function to find ssd for 8x16 block, sse2, 12 bit depth
>+;Defined sepeartely to be called from SSD_ONE_SS_16 macro
>+INIT_XMM sse2
>+cglobal ssd_ss_8x16
>+ pxor m8, m8
>+ mov r4d, 4
>+.loop:
>+ movu m0, [r0]
>+ movu m1, [r0 + mmsize]
>+ movu m2, [r0 + r1]
>+ movu m3, [r0 + r1+ mmsize]
>+ movu m4, [r2]
>+ movu m5, [r2 + mmsize]
>+ movu m6, [r2 + r3]
>+ movu m7, [r2 + r3 + mmsize]
>+ psubw m0, m4
>+ psubw m1, m5
>+ psubw m2, m6
>+ psubw m3, m7
>+ lea r0, [r0 + 2 * r1]
>+ lea r2, [r2 + 2 * r3]
>+ pmaddwd m0, m0
>+ pmaddwd m1, m1
>+ pmaddwd m2, m2
>+ pmaddwd m3, m3
>+ paddd m2, m3
>+ paddd m0, m1
>+ paddd m0, m2<br></div></div>paddd m8, m2 <-- this style broken dependency link, may faster</pre><span class=""><pre>>+ paddd m8, m0
>+ dec r4d
>+ jnz .loop
>+
>+ mova m4, m8
>+ pxor m5, m5
>+ punpckldq m8, m5
>+ punpckhdq m4, m5
>+ paddq m4, m8
>+ movhlps m5, m4
>+ paddq m4, m5
>+ paddq m9, m4</pre></span><pre>in this case, dynamic range is</pre><pre>12+12 + 3+4 = 31, so we didn't need QWORD operators (3+4 because 8x16)</pre><pre>another problem, this function as common function to share with other functions, so we didn't need sum all elements into one QWORD, we can do it in last stage.</pre><pre>same comment in below code</pre><pre><br></pre></div><br>_______________________________________________<br>
x265-devel mailing list<br>
<a href="mailto:x265-devel@videolan.org">x265-devel@videolan.org</a><br>
<a href="https://mailman.videolan.org/listinfo/x265-devel" rel="noreferrer" target="_blank">https://mailman.videolan.org/listinfo/x265-devel</a><br>
<br></blockquote></div><br></div>