[x265] [PATCH] asm: ssse3 8bpp code for convert_p2s[8xN], convert_p2s[16xN]
rajesh at multicorewareinc.com
rajesh at multicorewareinc.com
Wed Apr 1 15:11:32 CEST 2015
# HG changeset patch
# User Rajesh Paulraj<rajesh at multicorewareinc.com>
# Date 1427891004 -19800
# Wed Apr 01 17:53:24 2015 +0530
# Node ID 99659be11e50befca0acca6f859615b46f66343a
# Parent 5b77c75f87af668910020b52d3cd0bd88cdf5de9
asm: ssse3 8bpp code for convert_p2s[8xN],convert_p2s[16xN]
convert_p2s[8x4](4.15x), convert_p2s[8x8](4.87x), convert_p2s[8x16](5.57x),
convert_p2s[8x32](5.71x), convert_p2s[16x4](9.48x),convert_p2s[16x8](11.68x),
convert_p2s[16x12](12.47x), convert_p2s[16x16](12.77x),convert_p2s[16x32](13.26x),
convert_p2s[16x64](12.68x)
diff -r 5b77c75f87af -r 99659be11e50 source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp Wed Apr 01 17:45:18 2015 +0530
+++ b/source/common/x86/asm-primitives.cpp Wed Apr 01 17:53:24 2015 +0530
@@ -1247,6 +1247,17 @@
p.cu[BLOCK_4x4].sse_pp = x265_pixel_ssd_4x4_ssse3;
p.chroma[X265_CSP_I422].cu[BLOCK_422_4x8].sse_pp = x265_pixel_ssd_4x8_ssse3;
+ p.pu[LUMA_8x4].convert_p2s = x265_filterPixelToShort_8x4_ssse3;
+ p.pu[LUMA_8x8].convert_p2s = x265_filterPixelToShort_8x8_ssse3;
+ p.pu[LUMA_8x16].convert_p2s = x265_filterPixelToShort_8x16_ssse3;
+ p.pu[LUMA_8x32].convert_p2s = x265_filterPixelToShort_8x32_ssse3;
+ p.pu[LUMA_16x4].convert_p2s = x265_filterPixelToShort_16x4_ssse3;
+ p.pu[LUMA_16x8].convert_p2s = x265_filterPixelToShort_16x8_ssse3;
+ p.pu[LUMA_16x12].convert_p2s = x265_filterPixelToShort_16x12_ssse3;
+ p.pu[LUMA_16x16].convert_p2s = x265_filterPixelToShort_16x16_ssse3;
+ p.pu[LUMA_16x32].convert_p2s = x265_filterPixelToShort_16x32_ssse3;
+ p.pu[LUMA_16x64].convert_p2s = x265_filterPixelToShort_16x64_ssse3;
+
p.dst4x4 = x265_dst4_ssse3;
p.cu[BLOCK_8x8].idct = x265_idct8_ssse3;
diff -r 5b77c75f87af -r 99659be11e50 source/common/x86/ipfilter8.asm
--- a/source/common/x86/ipfilter8.asm Wed Apr 01 17:45:18 2015 +0530
+++ b/source/common/x86/ipfilter8.asm Wed Apr 01 17:53:24 2015 +0530
@@ -7784,6 +7784,135 @@
P2S_H_4xN 8
P2S_H_4xN 16
+;-----------------------------------------------------------------------------
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
+;-----------------------------------------------------------------------------
+%macro P2S_H_8xN 1
+INIT_XMM ssse3
+cglobal filterPixelToShort_8x%1, 3, 7, 6
+ mov r3d, r3m
+ add r3d, r3d
+ lea r5, [r1 * 3]
+ lea r6, [r3 * 3]
+
+ ; load height
+ mov r4d, %1/4
+
+ ; load constant
+ mova m4, [pb_128]
+ mova m5, [tab_c_64_n64]
+
+.loop
+ movh m0, [r0]
+ punpcklbw m0, m4
+ pmaddubsw m0, m5
+
+ movh m1, [r0 + r1]
+ punpcklbw m1, m4
+ pmaddubsw m1, m5
+
+ movh m2, [r0 + r1 * 2]
+ punpcklbw m2, m4
+ pmaddubsw m2, m5
+
+ movh m3, [r0 + r5]
+ punpcklbw m3, m4
+ pmaddubsw m3, m5
+
+ movu [r2 + r3 * 0], m0
+ movu [r2 + r3 * 1], m1
+ movu [r2 + r3 * 2], m2
+ movu [r2 + r6 ], m3
+
+ lea r0, [r0 + r1 * 4]
+ lea r2, [r2 + r3 * 4]
+
+ dec r4d
+ jnz .loop
+ RET
+%endmacro
+P2S_H_8xN 8
+P2S_H_8xN 4
+P2S_H_8xN 16
+P2S_H_8xN 32
+
+;-----------------------------------------------------------------------------
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
+;-----------------------------------------------------------------------------
+%macro P2S_H_16xN 1
+INIT_XMM ssse3
+cglobal filterPixelToShort_16x%1, 3, 7, 6
+ mov r3d, r3m
+ add r3d, r3d
+ lea r4, [r3 * 3]
+ lea r5, [r1 * 3]
+
+ ; load height
+ mov r6d, %1/4
+
+ ; load constant
+ mova m4, [pb_128]
+ mova m5, [tab_c_64_n64]
+
+.loop:
+ movh m0, [r0]
+ punpcklbw m0, m4
+ pmaddubsw m0, m5
+
+ movh m1, [r0 + r1]
+ punpcklbw m1, m4
+ pmaddubsw m1, m5
+
+ movh m2, [r0 + r1 * 2]
+ punpcklbw m2, m4
+ pmaddubsw m2, m5
+
+ movh m3, [r0 + r5]
+ punpcklbw m3, m4
+ pmaddubsw m3, m5
+
+ movu [r2 + r3 * 0], m0
+ movu [r2 + r3 * 1], m1
+ movu [r2 + r3 * 2], m2
+ movu [r2 + r4], m3
+
+ lea r0, [r0 + 8]
+
+ movh m0, [r0]
+ punpcklbw m0, m4
+ pmaddubsw m0, m5
+
+ movh m1, [r0 + r1]
+ punpcklbw m1, m4
+ pmaddubsw m1, m5
+
+ movh m2, [r0 + r1 * 2]
+ punpcklbw m2, m4
+ pmaddubsw m2, m5
+
+ movh m3, [r0 + r5]
+ punpcklbw m3, m4
+ pmaddubsw m3, m5
+
+ movu [r2 + r3 * 0 + 16], m0
+ movu [r2 + r3 * 1 + 16], m1
+ movu [r2 + r3 * 2 + 16], m2
+ movu [r2 + r4 + 16], m3
+
+ lea r0, [r0 + r1 * 4 - 8]
+ lea r2, [r2 + r3 * 4]
+
+ dec r6d
+ jnz .loop
+ RET
+%endmacro
+P2S_H_16xN 16
+P2S_H_16xN 4
+P2S_H_16xN 8
+P2S_H_16xN 12
+P2S_H_16xN 32
+P2S_H_16xN 64
+
%macro PROCESS_LUMA_W4_4R 0
movd m0, [r0]
movd m1, [r0 + r1]
diff -r 5b77c75f87af -r 99659be11e50 source/common/x86/ipfilter8.h
--- a/source/common/x86/ipfilter8.h Wed Apr 01 17:45:18 2015 +0530
+++ b/source/common/x86/ipfilter8.h Wed Apr 01 17:53:24 2015 +0530
@@ -625,6 +625,16 @@
void x265_filterPixelToShort_4x4_sse4(const pixel* src, intptr_t srcStride, int16_t* dst, int16_t dstStride);
void x265_filterPixelToShort_4x8_sse4(const pixel* src, intptr_t srcStride, int16_t* dst, int16_t dstStride);
void x265_filterPixelToShort_4x16_sse4(const pixel* src, intptr_t srcStride, int16_t* dst, int16_t dstStride);
+void x265_filterPixelToShort_8x4_ssse3(const pixel* src, intptr_t srcStride, int16_t* dst, int16_t dstStride);
+void x265_filterPixelToShort_8x8_ssse3(const pixel* src, intptr_t srcStride, int16_t* dst, int16_t dstStride);
+void x265_filterPixelToShort_8x16_ssse3(const pixel* src, intptr_t srcStride, int16_t* dst, int16_t dstStride);
+void x265_filterPixelToShort_8x32_ssse3(const pixel* src, intptr_t srcStride, int16_t* dst, int16_t dstStride);
+void x265_filterPixelToShort_16x4_ssse3(const pixel* src, intptr_t srcStride, int16_t* dst, int16_t dstStride);
+void x265_filterPixelToShort_16x8_ssse3(const pixel* src, intptr_t srcStride, int16_t* dst, int16_t dstStride);
+void x265_filterPixelToShort_16x12_ssse3(const pixel* src, intptr_t srcStride, int16_t* dst, int16_t dstStride);
+void x265_filterPixelToShort_16x16_ssse3(const pixel* src, intptr_t srcStride, int16_t* dst, int16_t dstStride);
+void x265_filterPixelToShort_16x32_ssse3(const pixel* src, intptr_t srcStride, int16_t* dst, int16_t dstStride);
+void x265_filterPixelToShort_16x64_ssse3(const pixel* src, intptr_t srcStride, int16_t* dst, int16_t dstStride);
#undef LUMA_FILTERS
#undef LUMA_SP_FILTERS
#undef LUMA_SS_FILTERS
More information about the x265-devel
mailing list