[x265] [PATCH 17 of 18] asm: intra_pred_ang4_17_sse2 16-bit
dtyx265 at gmail.com
dtyx265 at gmail.com
Fri Apr 3 18:24:48 CEST 2015
# HG changeset patch
# User David T Yuen <dtyx265 at gmail.com>
# Date 1428078042 25200
# Node ID e4b343abfa73af9b68f26283a9e75e161860d7b2
# Parent 97e3406e46539aefb8e1568fb49c452a63554887
asm: intra_pred_ang4_17_sse2 16-bit
This is backported from sse4 code and replaces c code.
./test/TestBench --testbench intrapred | grep "intra_ang_4x4\[17\]"
intra_ang_4x4[17] 3.80x 580.18 2202.73
transposed mode
./test/TestBench --testbench intrapred | grep "intra_ang_4x4\[19\]"
intra_ang_4x4[19] 3.39x 512.79 1735.90
diff -r 97e3406e4653 -r e4b343abfa73 source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp Fri Apr 03 09:09:04 2015 -0700
+++ b/source/common/x86/asm-primitives.cpp Fri Apr 03 09:20:42 2015 -0700
@@ -897,6 +897,8 @@
p.cu[BLOCK_4x4].intra_pred[14] = x265_intra_pred_ang4_14_sse2;
p.cu[BLOCK_4x4].intra_pred[15] = x265_intra_pred_ang4_15_sse2;
p.cu[BLOCK_4x4].intra_pred[16] = x265_intra_pred_ang4_16_sse2;
+ p.cu[BLOCK_4x4].intra_pred[17] = x265_intra_pred_ang4_17_sse2;
+ p.cu[BLOCK_4x4].intra_pred[19] = x265_intra_pred_ang4_17_sse2;
p.cu[BLOCK_4x4].intra_pred[20] = x265_intra_pred_ang4_16_sse2;
p.cu[BLOCK_4x4].intra_pred[21] = x265_intra_pred_ang4_15_sse2;
p.cu[BLOCK_4x4].intra_pred[22] = x265_intra_pred_ang4_14_sse2;
diff -r 97e3406e4653 -r e4b343abfa73 source/common/x86/intrapred16.asm
--- a/source/common/x86/intrapred16.asm Fri Apr 03 09:09:04 2015 -0700
+++ b/source/common/x86/intrapred16.asm Fri Apr 03 09:20:42 2015 -0700
@@ -1141,6 +1141,40 @@
mova m7, [r3 - 7 * 16] ; [12]
jmp mangle(private_prefix %+ _ %+ intra_pred_ang4_3 %+ SUFFIX %+ .do_filter4x4)
+cglobal intra_pred_ang4_17, 3,5,8
+ xor r4d, r4d
+ cmp r3m, byte 19
+ mov r3d, 16
+ jz .next
+ xchg r3d, r4d
+.next:
+ movd m4, [r2]
+ movh m5, [r2 + r3 + 2] ;[D x C B]
+ pshuflw m5, m5, 0x1F ;[B C D D]
+ punpcklqdq m5, m4 ;[x x x A B C D D]
+ psrldq m5, 2 ;[x x x x A B C D]
+ movhps m5, [r2 + r4 + 2]
+
+ mova m4, m5
+ psrldq m4, 2
+ punpcklwd m5, m4
+ mova m3, m4
+ psrldq m3, 2
+ punpcklwd m4, m3
+ mova m2, m3
+ psrldq m2, 2
+ punpcklwd m3, m2
+ mova m1, m2
+ psrldq m1, 2
+ punpcklwd m2, m1
+
+ lea r3, [ang_table + 14 * 16]
+ mova m0, [r3 - 8 * 16] ; [ 6]
+ mova m1, [r3 - 2 * 16] ; [12]
+ mova m6, [r3 + 4 * 16] ; [18]
+ mova m7, [r3 + 10 * 16] ; [24]
+ jmp mangle(private_prefix %+ _ %+ intra_pred_ang4_3 %+ SUFFIX %+ .do_filter4x4)
+
;-----------------------------------------------------------------------------------
; void intra_pred_dc(pixel* dst, intptr_t dstStride, pixel* above, int, int filter)
;-----------------------------------------------------------------------------------
More information about the x265-devel
mailing list