[x265] [PATCH] asm: luma_vpp[64x32, 64x48, 64x64] in avx2: improve 14150c->9810c, 21132c->14684c, 28663c->19616c
Divya Manivannan
divya at multicorewareinc.com
Wed Nov 26 06:34:04 CET 2014
- Previous message: [x265] [PATCH] Luma_hpp[16x4, 16x8, 16x12, 16x16 , 16x32, 16x64] avx2 asm code: improved 619c->458c, 1174c->812c, 1694c->1112c, 2291c->1535c, 4846c->3207c, 9294c->6104c
- Next message: [x265] [PATCH] asm: luma_vpp[64x32, 64x48, 64x64] in avx2: improve 14150c->9810c, 21132c->14684c, 28663c->19616c
- Messages sorted by:
[ date ]
[ thread ]
[ subject ]
[ author ]
# HG changeset patch
# User Divya Manivannan <divya at multicorewareinc.com>
# Date 1416979975 -19800
# Wed Nov 26 11:02:55 2014 +0530
# Node ID 879b986bbc706ecaf5eb93db5b5a381d9cea47e0
# Parent d57c28a3010b0d8c9b10575cd105902434efc981
asm: luma_vpp[64x32, 64x48, 64x64] in avx2: improve 14150c->9810c, 21132c->14684c, 28663c->19616c
diff -r d57c28a3010b -r 879b986bbc70 source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp Tue Nov 25 11:45:44 2014 +0530
+++ b/source/common/x86/asm-primitives.cpp Wed Nov 26 11:02:55 2014 +0530
@@ -1798,17 +1798,23 @@
p.transpose[BLOCK_16x16] = x265_transpose16_avx2;
p.transpose[BLOCK_32x32] = x265_transpose32_avx2;
p.transpose[BLOCK_64x64] = x265_transpose64_avx2;
+
p.luma_vpp[LUMA_16x4] = x265_interp_8tap_vert_pp_16x4_avx2;
p.luma_vpp[LUMA_16x8] = x265_interp_8tap_vert_pp_16x8_avx2;
p.luma_vpp[LUMA_16x12] = x265_interp_8tap_vert_pp_16x12_avx2;
p.luma_vpp[LUMA_16x16] = x265_interp_8tap_vert_pp_16x16_avx2;
p.luma_vpp[LUMA_16x32] = x265_interp_8tap_vert_pp_16x32_avx2;
p.luma_vpp[LUMA_16x64] = x265_interp_8tap_vert_pp_16x64_avx2;
+
p.luma_vpp[LUMA_32x8] = x265_interp_8tap_vert_pp_32x8_avx2;
p.luma_vpp[LUMA_32x16] = x265_interp_8tap_vert_pp_32x16_avx2;
p.luma_vpp[LUMA_32x24] = x265_interp_8tap_vert_pp_32x24_avx2;
p.luma_vpp[LUMA_32x32] = x265_interp_8tap_vert_pp_32x32_avx2;
p.luma_vpp[LUMA_32x64] = x265_interp_8tap_vert_pp_32x64_avx2;
+
+ p.luma_vpp[LUMA_64x32] = x265_interp_8tap_vert_pp_64x32_avx2;
+ p.luma_vpp[LUMA_64x48] = x265_interp_8tap_vert_pp_64x48_avx2;
+ p.luma_vpp[LUMA_64x64] = x265_interp_8tap_vert_pp_64x64_avx2;
#endif
p.luma_hpp[LUMA_4x4] = x265_interp_8tap_horiz_pp_4x4_avx2;
diff -r d57c28a3010b -r 879b986bbc70 source/common/x86/ipfilter8.asm
--- a/source/common/x86/ipfilter8.asm Tue Nov 25 11:45:44 2014 +0530
+++ b/source/common/x86/ipfilter8.asm Wed Nov 26 11:02:55 2014 +0530
@@ -5805,6 +5805,47 @@
RET
%endif
+%macro FILTER_VER_LUMA_AVX2_64xN 2
+INIT_YMM avx2
+%if ARCH_X86_64 == 1
+cglobal interp_8tap_vert_pp_%1x%2, 4, 12, 15
+ mov r4d, r4m
+ shl r4d, 7
+
+%ifdef PIC
+ lea r5, [tab_LumaCoeffVer_32]
+ add r5, r4
+%else
+ lea r5, [tab_LumaCoeffVer_32 + r4]
+%endif
+
+ lea r4, [r1 * 3]
+ sub r0, r4
+ lea r6, [r3 * 3]
+ lea r11, [r1 * 4]
+ mova m14, [pw_512]
+ mov r9d, %2 / 16
+.loopH:
+ mov r10d, %1 / 16
+.loopW:
+ PROCESS_LUMA_AVX2_W16_16R
+ add r2, 16
+ add r0, 16
+ dec r10d
+ jnz .loopW
+ sub r7, r11
+ lea r0, [r7 - 48]
+ lea r2, [r8 + r3 * 4 - 48]
+ dec r9d
+ jnz .loopH
+ RET
+%endif
+%endmacro
+
+FILTER_VER_LUMA_AVX2_64xN 64, 32
+FILTER_VER_LUMA_AVX2_64xN 64, 48
+FILTER_VER_LUMA_AVX2_64xN 64, 64
+
;-------------------------------------------------------------------------------------------------------------
; void interp_8tap_vert_%3_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
;-------------------------------------------------------------------------------------------------------------
- Previous message: [x265] [PATCH] Luma_hpp[16x4, 16x8, 16x12, 16x16 , 16x32, 16x64] avx2 asm code: improved 619c->458c, 1174c->812c, 1694c->1112c, 2291c->1535c, 4846c->3207c, 9294c->6104c
- Next message: [x265] [PATCH] asm: luma_vpp[64x32, 64x48, 64x64] in avx2: improve 14150c->9810c, 21132c->14684c, 28663c->19616c
- Messages sorted by:
[ date ]
[ thread ]
[ subject ]
[ author ]
More information about the x265-devel
mailing list