[x265] [PATCH] asm-ssse3: filter_p2s[12x16](10.27x), filter_p2s[24x32](10.30x),
chen
chenm003 at 163.com
Thu Mar 5 22:03:56 CET 2015
At 2015-03-05 20:07:14,rajesh at multicorewareinc.com wrote:
># HG changeset patch
># User Rajesh Paulraj<rajesh at multicorewareinc.com>
># Date 1425556197 -19800
># Thu Mar 05 17:19:57 2015 +0530
># Node ID 38cb572c2927eee8039464ec462b874c0da20871
># Parent ea9bdb10353fcb06cea1045ba0186c22c448df63
>asm-ssse3: filter_p2s[12x16](10.27x), filter_p2s[24x32](10.30x),
> filter_p2s[48x64](9.60x)
>
>diff -r ea9bdb10353f -r 38cb572c2927 source/common/x86/asm-primitives.cpp
>--- a/source/common/x86/asm-primitives.cpp Wed Mar 04 13:20:55 2015 +0530
>+++ b/source/common/x86/asm-primitives.cpp Thu Mar 05 17:19:57 2015 +0530
>@@ -1278,7 +1278,9 @@
> p.pu[LUMA_64x32].filter_p2s = x265_pixelToShort_64x32_ssse3;
> p.pu[LUMA_64x48].filter_p2s = x265_pixelToShort_64x48_ssse3;
> p.pu[LUMA_64x64].filter_p2s = x265_pixelToShort_64x64_ssse3;
>-
>+ p.pu[LUMA_12x16].filter_p2s = x265_pixelToShort_12x16_ssse3;
>+ p.pu[LUMA_24x32].filter_p2s = x265_pixelToShort_24x32_ssse3;
>+ p.pu[LUMA_48x64].filter_p2s = x265_pixelToShort_48x64_ssse3;
> p.chroma[X265_CSP_I420].p2s = x265_chroma_p2s_ssse3;
> p.chroma[X265_CSP_I422].p2s = x265_chroma_p2s_ssse3;
>
>diff -r ea9bdb10353f -r 38cb572c2927 source/common/x86/ipfilter8.asm
>--- a/source/common/x86/ipfilter8.asm Wed Mar 04 13:20:55 2015 +0530
>+++ b/source/common/x86/ipfilter8.asm Thu Mar 05 17:19:57 2015 +0530
>@@ -5579,8 +5579,9 @@
> FILTER_V4_W16n_H2 64, 48
> FILTER_V4_W16n_H2 48, 64
> FILTER_V4_W16n_H2 64, 16
>-;-----------------------------------------------------------------------------
>-; void pixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int width, int height)
>+
>+;-----------------------------------------------------------------------------
>+; void pixelToShort(pixel *src, intptr_t srcStride, int16_t *dst)
> ;-----------------------------------------------------------------------------
> %macro PIXEL_WH_4xN 2
> INIT_XMM ssse3
>@@ -5641,9 +5642,8 @@
> PIXEL_WH_4xN 4, 4
> PIXEL_WH_4xN 4, 8
> PIXEL_WH_4xN 4, 16
>-
>-;-----------------------------------------------------------------------------
>-; void pixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int width, int height)
>+;-----------------------------------------------------------------------------
>+; void pixelToShort(pixel *src, intptr_t srcStride, int16_t *dst)
> ;-----------------------------------------------------------------------------
> %macro PIXEL_WH_8xN 2
> INIT_XMM ssse3
>@@ -5703,10 +5703,8 @@
> PIXEL_WH_8xN 8, 4
> PIXEL_WH_8xN 8, 16
> PIXEL_WH_8xN 8, 32
>-
>-
>-;-----------------------------------------------------------------------------
>-; void pixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int width, int height)
>+;-----------------------------------------------------------------------------
>+; void pixelToShort(pixel *src, intptr_t srcStride, int16_t *dst)
> ;-----------------------------------------------------------------------------
> %macro PIXEL_WH_16xN 2
> INIT_XMM ssse3
>@@ -5768,9 +5766,8 @@
> PIXEL_WH_16xN 16, 12
> PIXEL_WH_16xN 16, 32
> PIXEL_WH_16xN 16, 64
>-
>-;-----------------------------------------------------------------------------
>-; void pixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int width, int height)
>+;-----------------------------------------------------------------------------
>+; void pixelToShort(pixel *src, intptr_t srcStride, int16_t *dst)
> ;-----------------------------------------------------------------------------
> %macro PIXEL_WH_32xN 2
> INIT_XMM ssse3
>@@ -5831,9 +5828,8 @@
> PIXEL_WH_32xN 32, 16
> PIXEL_WH_32xN 32, 24
> PIXEL_WH_32xN 32, 64
>-
>-;-----------------------------------------------------------------------------
>-; void pixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int width, int height)
>+;-----------------------------------------------------------------------------
>+; void pixelToShort(pixel *src, intptr_t srcStride, int16_t *dst)
> ;-----------------------------------------------------------------------------
> %macro PIXEL_WH_64xN 2
> INIT_XMM ssse3
>@@ -5893,6 +5889,165 @@
> PIXEL_WH_64xN 64, 16
> PIXEL_WH_64xN 64, 32
> PIXEL_WH_64xN 64, 48
>+;-----------------------------------------------------------------------------
>+; void pixelToShort(pixel *src, intptr_t srcStride, int16_t *dst)
>+;-----------------------------------------------------------------------------
>+INIT_XMM ssse3
>+cglobal pixelToShort_12x16, 3, 4, 6
this func is right now
>+
>+ ; load constant
>+ mova m4, [pb_128]
>+ mova m5, [tab_c_64_n64]
>+
>+%rep 4
>+
>+ movu m0, [r0]
>+ punpcklbw m1, m0, m4
>+ punpckhbw m0, m4
>+ pmaddubsw m0, m5
>+ pmaddubsw m1, m5
>+
>+ movu m2, [r0 + r1]
>+ punpcklbw m3, m2, m4
>+ punpckhbw m2, m4
>+ pmaddubsw m2, m5
>+ pmaddubsw m3, m5
>+
>+ movu [r2 + FENC_STRIDE * 0], m1
>+ movu [r2 + FENC_STRIDE * 2], m3
>+
>+ movh [r2 + FENC_STRIDE * 0 + 16], m0
>+ movh [r2 + FENC_STRIDE * 2 + 16], m2
>+
>+ movu m0, [r0 + r1 * 2]
>+ punpcklbw m1, m0, m4
>+ punpckhbw m0, m4
>+ pmaddubsw m0, m5
>+ pmaddubsw m1, m5
>+
>+ lea r3, [r0 + r1 * 2]
>+ movu m2, [r3 + r1]
>+ punpcklbw m3, m2, m4
>+ punpckhbw m2, m4
>+ pmaddubsw m2, m5
>+ pmaddubsw m3, m5
>+
>+ movu [r2 + FENC_STRIDE * 4], m1
>+ movu [r2 + FENC_STRIDE * 6], m3
>+
>+ movh [r2 + FENC_STRIDE * 4 + 16], m0
>+ movh [r2 + FENC_STRIDE * 6 + 16], m2
>+
>+ lea r0, [r0 + r1 * 4]
>+ add r2, FENC_STRIDE * 8
>+%endrep
>+ RET
>+
>+;-----------------------------------------------------------------------------
>+; void pixelToShort(pixel *src, intptr_t srcStride, int16_t *dst)
>+;-----------------------------------------------------------------------------
>+INIT_XMM ssse3
>+cglobal pixelToShort_24x32, 3, 6, 6
>+
>+ ; load height
>+ mov r3d, 32
>+
>+ ; load constant
>+ mova m4, [pb_128]
>+ mova m5, [tab_c_64_n64]
>+
>+.loopH:
>+
>+ xor r4d, r4d
it is fixed count loop and loop body is small, so we may use %rep here
>+.loopW:
>+ lea r5, [r0 + r4]
>+
>+ movh m0, [r5]
>+ punpcklbw m0, m4
>+ pmaddubsw m0, m5
>+
>+ movh m1, [r5 + r1]
>+ punpcklbw m1, m4
>+ pmaddubsw m1, m5
>+
>+ movh m2, [r5 + r1 * 2]
>+ punpcklbw m2, m4
>+ pmaddubsw m2, m5
>+
>+ lea r5, [r5 + r1 * 2]
>+ movh m3, [r5 + r1]
>+ punpcklbw m3, m4
>+ pmaddubsw m3, m5
>+
>+ add r4, 8
>+ cmp r4, 24
>+ movu [r2 + r4 * 2 + FENC_STRIDE * 0 - 16], m0
>+ movu [r2 + r4 * 2 + FENC_STRIDE * 2 - 16], m1
>+ movu [r2 + r4 * 2 + FENC_STRIDE * 4 - 16], m2
>+ movu [r2 + r4 * 2 + FENC_STRIDE * 6 - 16], m3
>+ je .nextH
>+ jnz .loopW
>+
>+.nextH:
>+ lea r0, [r0 + r1 * 4]
>+ add r2, FENC_STRIDE * 8
>+
>+ sub r3d, 4
>+ jnz .loopH
>+ RET
>+
>+;-----------------------------------------------------------------------------
>+; void pixelToShort(pixel *src, intptr_t srcStride, int16_t *dst)
>+;-----------------------------------------------------------------------------
>+INIT_XMM ssse3
>+cglobal pixelToShort_48x64, 3, 6, 6
>+
>+ ; load height
>+ mov r3d, 64
>+
>+ ; load constant
>+ mova m4, [pb_128]
>+ mova m5, [tab_c_64_n64]
>+
>+.loopH:
>+
>+ xor r4d, r4d
same
>+.loopW:
>+ lea r5, [r0 + r4]
>+
>+ movh m0, [r5]
>+ punpcklbw m0, m4
>+ pmaddubsw m0, m5
>+
>+ movh m1, [r5 + r1]
>+ punpcklbw m1, m4
>+ pmaddubsw m1, m5
>+
>+ movh m2, [r5 + r1 * 2]
>+ punpcklbw m2, m4
>+ pmaddubsw m2, m5
>+
>+ lea r5, [r5 + r1 * 2]
>+ movh m3, [r5 + r1]
>+ punpcklbw m3, m4
>+ pmaddubsw m3, m5
>+
>+ add r4, 8
>+ cmp r4, 48
>+ movu [r2 + r4 * 2 + FENC_STRIDE * 0 - 16], m0
>+ movu [r2 + r4 * 2 + FENC_STRIDE * 2 - 16], m1
>+ movu [r2 + r4 * 2 + FENC_STRIDE * 4 - 16], m2
>+ movu [r2 + r4 * 2 + FENC_STRIDE * 6 - 16], m3
>+ je .nextH
>+ jnz .loopW
>+
>+.nextH:
>+ lea r0, [r0 + r1 * 4]
>+ add r2, FENC_STRIDE * 8
>+
>+ sub r3d, 4
>+ jnz .loopH
>+ RET
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.videolan.org/pipermail/x265-devel/attachments/20150306/54d19119/attachment-0001.html>
More information about the x265-devel
mailing list