[x265] [PATCH] asm: ssse3 8bpp code for convert_p2s[12x16](9.82x), convert_p2s[24x32](13.61x)

rajesh at multicorewareinc.com rajesh at multicorewareinc.com
Mon Apr 6 07:10:40 CEST 2015


# HG changeset patch
# User Rajesh Paulraj<rajesh at multicorewareinc.com>
# Date 1428296987 -19800
#      Mon Apr 06 10:39:47 2015 +0530
# Node ID 6e9471155dec62e653db0a96df14e62df53041bd
# Parent  ebe5e57c4b45b45338035a1009b64585f21d66d5
asm: ssse3 8bpp code for convert_p2s[12x16](9.82x), convert_p2s[24x32](13.61x)
                         convert_p2s[48x64](11.12x)

diff -r ebe5e57c4b45 -r 6e9471155dec source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Sat Apr 04 15:11:39 2015 -0500
+++ b/source/common/x86/asm-primitives.cpp	Mon Apr 06 10:39:47 2015 +0530
@@ -1332,6 +1332,9 @@
         p.pu[LUMA_64x32].convert_p2s = x265_filterPixelToShort_64x32_ssse3;
         p.pu[LUMA_64x48].convert_p2s = x265_filterPixelToShort_64x48_ssse3;
         p.pu[LUMA_64x64].convert_p2s = x265_filterPixelToShort_64x64_ssse3;
+        p.pu[LUMA_12x16].convert_p2s = x265_filterPixelToShort_12x16_ssse3;
+        p.pu[LUMA_24x32].convert_p2s = x265_filterPixelToShort_24x32_ssse3;
+        p.pu[LUMA_48x64].convert_p2s = x265_filterPixelToShort_48x64_ssse3;
 
         p.chroma[X265_CSP_I420].pu[CHROMA_420_8x2].p2s = x265_filterPixelToShort_8x2_ssse3;
         p.chroma[X265_CSP_I420].pu[CHROMA_420_8x4].p2s = x265_filterPixelToShort_8x4_ssse3;
diff -r ebe5e57c4b45 -r 6e9471155dec source/common/x86/ipfilter8.asm
--- a/source/common/x86/ipfilter8.asm	Sat Apr 04 15:11:39 2015 -0500
+++ b/source/common/x86/ipfilter8.asm	Mon Apr 06 10:39:47 2015 +0530
@@ -8248,6 +8248,274 @@
 P2S_H_64xN 32
 P2S_H_64xN 48
 
+;-----------------------------------------------------------------------------
+; void filterPixelToShort(pixel src, intptr_t srcStride, int16_t dst, int16_t dstStride)
+;-----------------------------------------------------------------------------
+INIT_XMM ssse3
+cglobal filterPixelToShort_12x16, 3, 7, 6
+    mov         r3d, r3m
+    add         r3d, r3d
+    lea         r4, [r1 * 3]
+    lea         r6, [r3 * 3]
+    mov         r5d, 4
+
+    ; load constant
+    mova        m4, [pb_128]
+    mova        m5, [tab_c_64_n64]
+
+.loop:
+    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 + r3 * 0], m1
+    movu        [r2 + r3 * 1], m3
+
+    movh        [r2 + r3 * 0 + 16], m0
+    movh        [r2 + r3 * 1 + 16], m2
+
+    movu        m0, [r0 + r1 * 2]
+    punpcklbw   m1, m0, m4
+    punpckhbw   m0, m4
+    pmaddubsw   m0, m5
+    pmaddubsw   m1, m5
+
+    movu        m2, [r0 + r4]
+    punpcklbw   m3, m2, m4
+    punpckhbw   m2, m4
+    pmaddubsw   m2, m5
+    pmaddubsw   m3, m5
+
+    movu        [r2 + r3 * 2], m1
+    movu        [r2 + r6], m3
+
+    movh        [r2 + r3 * 2 + 16], m0
+    movh        [r2 + r6 + 16], m2
+
+    lea         r0, [r0 + r1 * 4]
+    lea         r2, [r2 + r3 * 4]
+
+    dec         r5d
+    jnz         .loop
+    RET
+
+;-----------------------------------------------------------------------------
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
+;-----------------------------------------------------------------------------
+INIT_XMM ssse3
+cglobal filterPixelToShort_24x32, 3, 7, 5
+    mov         r3d, r3m
+    add         r3d, r3d
+    lea         r4, [r1 * 3]
+    lea         r5, [r3 * 3]
+    mov         r6d, 8
+
+    ; load constant
+    mova        m3, [pb_128]
+    mova        m4, [tab_c_64_n64]
+
+.loop:
+    movu        m0, [r0]
+    punpcklbw   m1, m0, m3
+    punpckhbw   m0, m3
+    pmaddubsw   m0, m4
+    pmaddubsw   m1, m4
+
+    movu        m2, [r0 + 16]
+    punpcklbw   m2, m3
+    pmaddubsw   m2, m4
+
+    movu        [r2 +  r3 * 0], m1
+    movu        [r2 +  r3 * 0 + 16], m0
+    movu        [r2 +  r3 * 0 + 32], m2
+
+    movu        m0, [r0 + r1]
+    punpcklbw   m1, m0, m3
+    punpckhbw   m0, m3
+    pmaddubsw   m0, m4
+    pmaddubsw   m1, m4
+
+    movu        m2, [r0 + r1 + 16]
+    punpcklbw   m2, m3
+    pmaddubsw   m2, m4
+
+    movu        [r2 +  r3 * 1], m1
+    movu        [r2 +  r3 * 1 + 16], m0
+    movu        [r2 +  r3 * 1 + 32], m2
+
+    movu        m0, [r0 + r1 * 2]
+    punpcklbw   m1, m0, m3
+    punpckhbw   m0, m3
+    pmaddubsw   m0, m4
+    pmaddubsw   m1, m4
+
+    movu        m2, [r0 + r1 * 2 + 16]
+    punpcklbw   m2, m3
+    pmaddubsw   m2, m4
+
+    movu        [r2 +  r3 * 2], m1
+    movu        [r2 +  r3 * 2 + 16], m0
+    movu        [r2 +  r3 * 2 + 32], m2
+
+    movu        m0, [r0 + r4]
+    punpcklbw   m1, m0, m3
+    punpckhbw   m0, m3
+    pmaddubsw   m0, m4
+    pmaddubsw   m1, m4
+
+    movu        m2, [r0 + r4 + 16]
+    punpcklbw   m2, m3
+    pmaddubsw   m2, m4
+    movu        [r2 +  r5], m1
+    movu        [r2 +  r5 + 16], m0
+    movu        [r2 +  r5 + 32], m2
+
+    lea         r0, [r0 + r1 * 4]
+    lea         r2, [r2 + r3 * 4]
+
+    dec         r6d
+    jnz         .loop
+    RET
+
+;-----------------------------------------------------------------------------
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
+;-----------------------------------------------------------------------------
+INIT_XMM ssse3
+cglobal filterPixelToShort_48x64, 3, 7, 4
+    mov         r3d, r3m
+    add         r3d, r3d
+    lea         r4, [r1 * 3]
+    lea         r5, [r3 * 3]
+    mov         r6d, 16
+
+    ; load constant
+    mova        m2, [pb_128]
+    mova        m3, [tab_c_64_n64]
+
+.loop:
+    movu        m0, [r0]
+    punpcklbw   m1, m0, m2
+    punpckhbw   m0, m2
+    pmaddubsw   m0, m3
+    pmaddubsw   m1, m3
+
+    movu        [r2 +  r3 * 0], m1
+    movu        [r2 +  r3 * 0 + 16], m0
+
+    movu        m0, [r0 + 16]
+    punpcklbw   m1, m0, m2
+    punpckhbw   m0, m2
+    pmaddubsw   m0, m3
+    pmaddubsw   m1, m3
+
+    movu        [r2 +  r3 * 0 + 32], m1
+    movu        [r2 +  r3 * 0 + 48], m0
+
+    movu        m0, [r0 + 32]
+    punpcklbw   m1, m0, m2
+    punpckhbw   m0, m2
+    pmaddubsw   m0, m3
+    pmaddubsw   m1, m3
+
+    movu        [r2 +  r3 * 0 + 64], m1
+    movu        [r2 +  r3 * 0 + 80], m0
+
+    movu        m0, [r0 + r1]
+    punpcklbw   m1, m0, m2
+    punpckhbw   m0, m2
+    pmaddubsw   m0, m3
+    pmaddubsw   m1, m3
+
+    movu        [r2 +  r3 * 1], m1
+    movu        [r2 +  r3 * 1 + 16], m0
+
+    movu        m0, [r0 + r1 + 16]
+    punpcklbw   m1, m0, m2
+    punpckhbw   m0, m2
+    pmaddubsw   m0, m3
+    pmaddubsw   m1, m3
+
+    movu        [r2 +  r3 * 1 + 32], m1
+    movu        [r2 +  r3 * 1 + 48], m0
+
+    movu        m0, [r0 + r1 + 32]
+    punpcklbw   m1, m0, m2
+    punpckhbw   m0, m2
+    pmaddubsw   m0, m3
+    pmaddubsw   m1, m3
+
+    movu        [r2 +  r3 * 1 + 64], m1
+    movu        [r2 +  r3 * 1 + 80], m0
+
+    movu        m0, [r0 + r1 * 2]
+    punpcklbw   m1, m0, m2
+    punpckhbw   m0, m2
+    pmaddubsw   m0, m3
+    pmaddubsw   m1, m3
+
+    movu        [r2 +  r3 * 2], m1
+    movu        [r2 +  r3 * 2 + 16], m0
+
+    movu        m0, [r0 + r1 * 2 + 16]
+    punpcklbw   m1, m0, m2
+    punpckhbw   m0, m2
+    pmaddubsw   m0, m3
+    pmaddubsw   m1, m3
+
+    movu        [r2 +  r3 * 2 + 32], m1
+    movu        [r2 +  r3 * 2 + 48], m0
+
+    movu        m0, [r0 + r1 * 2 + 32]
+    punpcklbw   m1, m0, m2
+    punpckhbw   m0, m2
+    pmaddubsw   m0, m3
+    pmaddubsw   m1, m3
+
+    movu        [r2 +  r3 * 2 + 64], m1
+    movu        [r2 +  r3 * 2 + 80], m0
+
+    movu        m0, [r0 + r4]
+    punpcklbw   m1, m0, m2
+    punpckhbw   m0, m2
+    pmaddubsw   m0, m3
+    pmaddubsw   m1, m3
+
+    movu        [r2 +  r5], m1
+    movu        [r2 +  r5 + 16], m0
+
+    movu        m0, [r0 + r4 + 16]
+    punpcklbw   m1, m0, m2
+    punpckhbw   m0, m2
+    pmaddubsw   m0, m3
+    pmaddubsw   m1, m3
+
+    movu        [r2 +  r5 + 32], m1
+    movu        [r2 +  r5 + 48], m0
+
+    movu        m0, [r0 + r4 + 32]
+    punpcklbw   m1, m0, m2
+    punpckhbw   m0, m2
+    pmaddubsw   m0, m3
+    pmaddubsw   m1, m3
+
+    movu        [r2 +  r5 + 64], m1
+    movu        [r2 +  r5 + 80], m0
+
+    lea         r0, [r0 + r1 * 4]
+    lea         r2, [r2 + r3 * 4]
+
+    dec         r6d
+    jnz         .loop
+    RET
+
 %macro PROCESS_LUMA_W4_4R 0
     movd        m0, [r0]
     movd        m1, [r0 + r1]
diff -r ebe5e57c4b45 -r 6e9471155dec source/common/x86/ipfilter8.h
--- a/source/common/x86/ipfilter8.h	Sat Apr 04 15:11:39 2015 -0500
+++ b/source/common/x86/ipfilter8.h	Mon Apr 06 10:39:47 2015 +0530
@@ -654,6 +654,9 @@
 void x265_filterPixelToShort_64x32_ssse3(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride);
 void x265_filterPixelToShort_64x48_ssse3(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride);
 void x265_filterPixelToShort_64x64_ssse3(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride);
+void x265_filterPixelToShort_12x16_ssse3(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride);
+void x265_filterPixelToShort_24x32_ssse3(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride);
+void x265_filterPixelToShort_48x64_ssse3(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride);
 #undef LUMA_FILTERS
 #undef LUMA_SP_FILTERS
 #undef LUMA_SS_FILTERS


More information about the x265-devel mailing list