[x265] [PATCH 2 of 3] asm: intra_pred_ang32_25 improved by ~53% over SSE4
praveen at multicorewareinc.com
praveen at multicorewareinc.com
Fri Mar 27 06:11:08 CET 2015
# HG changeset patch
# User Praveen Tiwari <praveen at multicorewareinc.com>
# Date 1427360000 -19800
# Thu Mar 26 14:23:20 2015 +0530
# Branch stable
# Node ID 39c139322fde1f8c62545fd8bbed9cc8198e540c
# Parent 24bdb3e594556ca6e12ee9dae58100a6bd115d2a
asm: intra_pred_ang32_25 improved by ~53% over SSE4
AVX2:
intra_ang_32x32[25] 23.11x 1293.83 29904.12
SSE4:
intra_ang_32x32[25] 10.31x 2759.33 28451.26
diff -r 24bdb3e59455 -r 39c139322fde source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp Thu Mar 26 13:20:04 2015 +0530
+++ b/source/common/x86/asm-primitives.cpp Thu Mar 26 14:23:20 2015 +0530
@@ -1643,6 +1643,7 @@
p.cu[BLOCK_32x32].intra_pred[31] = x265_intra_pred_ang32_31_avx2;
p.cu[BLOCK_32x32].intra_pred[32] = x265_intra_pred_ang32_32_avx2;
p.cu[BLOCK_32x32].intra_pred[33] = x265_intra_pred_ang32_33_avx2;
+ p.cu[BLOCK_32x32].intra_pred[25] = x265_intra_pred_ang32_25_avx2;
// copy_sp primitives
p.cu[BLOCK_16x16].copy_sp = x265_blockcopy_sp_16x16_avx2;
diff -r 24bdb3e59455 -r 39c139322fde source/common/x86/intrapred.h
--- a/source/common/x86/intrapred.h Thu Mar 26 13:20:04 2015 +0530
+++ b/source/common/x86/intrapred.h Thu Mar 26 14:23:20 2015 +0530
@@ -213,6 +213,7 @@
void x265_intra_pred_ang32_31_avx2(pixel* dst, intptr_t dstStride, const pixel* srcPix, int dirMode, int bFilter);
void x265_intra_pred_ang32_32_avx2(pixel* dst, intptr_t dstStride, const pixel* srcPix, int dirMode, int bFilter);
void x265_intra_pred_ang32_33_avx2(pixel* dst, intptr_t dstStride, const pixel* srcPix, int dirMode, int bFilter);
+void x265_intra_pred_ang32_25_avx2(pixel* dst, intptr_t dstStride, const pixel* srcPix, int dirMode, int bFilter);
void x265_all_angs_pred_4x4_sse4(pixel *dest, pixel *refPix, pixel *filtPix, int bLuma);
void x265_all_angs_pred_8x8_sse4(pixel *dest, pixel *refPix, pixel *filtPix, int bLuma);
void x265_all_angs_pred_16x16_sse4(pixel *dest, pixel *refPix, pixel *filtPix, int bLuma);
diff -r 24bdb3e59455 -r 39c139322fde source/common/x86/intrapred8.asm
--- a/source/common/x86/intrapred8.asm Thu Mar 26 13:20:04 2015 +0530
+++ b/source/common/x86/intrapred8.asm Thu Mar 26 14:23:20 2015 +0530
@@ -407,6 +407,26 @@
db 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0
+
+ALIGN 32
+c_ang32_mode_25: db 2, 30, 2, 30, 2, 30, 2, 30, 2, 30, 2, 30, 2, 30, 2, 30, 4, 28, 4, 28, 4, 28, 4, 28, 4, 28, 4, 28, 4, 28, 4, 28
+ db 6, 26, 6, 26, 6, 26, 6, 26, 6, 26, 6, 26, 6, 26, 6, 26, 8, 24, 8, 24, 8, 24, 8, 24, 8, 24, 8, 24, 8, 24, 8, 24
+ db 10, 22, 10, 22, 10, 22, 10, 22, 10, 22, 10, 22, 10, 22, 10, 22, 12, 20, 12, 20, 12, 20, 12, 20, 12, 20, 12, 20, 12, 20, 12, 20
+ db 14, 18, 14, 18, 14, 18, 14, 18, 14, 18, 14, 18, 14, 18, 14, 18, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16
+ db 18, 14, 18, 14, 18, 14, 18, 14, 18, 14, 18, 14, 18, 14, 18, 14, 20, 12, 20, 12, 20, 12, 20, 12, 20, 12, 20, 12, 20, 12, 20, 12
+ db 22, 10, 22, 10, 22, 10, 22, 10, 22, 10, 22, 10, 22, 10, 22, 10, 24, 8, 24, 8, 24, 8, 24, 8, 24, 8, 24, 8, 24, 8, 24, 8
+ db 26, 6, 26, 6, 26, 6, 26, 6, 26, 6, 26, 6, 26, 6, 26, 6, 28, 4, 28, 4, 28, 4, 28, 4, 28, 4, 28, 4, 28, 4, 28, 4
+ db 30, 2, 30, 2, 30, 2, 30, 2, 30, 2, 30, 2, 30, 2, 30, 2, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0
+ db 2, 30, 2, 30, 2, 30, 2, 30, 2, 30, 2, 30, 2, 30, 2, 30, 4, 28, 4, 28, 4, 28, 4, 28, 4, 28, 4, 28, 4, 28, 4, 28
+ db 6, 26, 6, 26, 6, 26, 6, 26, 6, 26, 6, 26, 6, 26, 6, 26, 8, 24, 8, 24, 8, 24, 8, 24, 8, 24, 8, 24, 8, 24, 8, 24
+ db 10, 22, 10, 22, 10, 22, 10, 22, 10, 22, 10, 22, 10, 22, 10, 22, 12, 20, 12, 20, 12, 20, 12, 20, 12, 20, 12, 20, 12, 20, 12, 20
+ db 14, 18, 14, 18, 14, 18, 14, 18, 14, 18, 14, 18, 14, 18, 14, 18, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16
+ db 18, 14, 18, 14, 18, 14, 18, 14, 18, 14, 18, 14, 18, 14, 18, 14, 20, 12, 20, 12, 20, 12, 20, 12, 20, 12, 20, 12, 20, 12, 20, 12
+ db 22, 10, 22, 10, 22, 10, 22, 10, 22, 10, 22, 10, 22, 10, 22, 10, 24, 8, 24, 8, 24, 8, 24, 8, 24, 8, 24, 8, 24, 8, 24, 8
+ db 26, 6, 26, 6, 26, 6, 26, 6, 26, 6, 26, 6, 26, 6, 26, 6, 28, 4, 28, 4, 28, 4, 28, 4, 28, 4, 28, 4, 28, 4, 28, 4
+ db 30, 2, 30, 2, 30, 2, 30, 2, 30, 2, 30, 2, 30, 2, 30, 2, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0, 32, 0
+
+
ALIGN 32
;; (blkSize - 1 - x)
pw_planar4_0: dw 3, 2, 1, 0, 3, 2, 1, 0
@@ -14108,5 +14128,155 @@
vpermq m6, m6, 11011000b
movu [r0 + r3], m6
RET
+
+INIT_YMM avx2
+cglobal intra_pred_ang32_25, 3, 5, 11
+ mova m0, [pw_1024]
+ mova m1, [intra_pred_shuff_0_8]
+ lea r3, [3 * r1]
+ lea r4, [c_ang32_mode_25]
+
+ ;row [0, 1]
+ vbroadcasti128 m2, [r2 + 0]
+ pshufb m2, m1
+ vbroadcasti128 m3, [r2 + 8]
+ pshufb m3, m1
+ vbroadcasti128 m4, [r2 + 16]
+ pshufb m4, m1
+ vbroadcasti128 m5, [r2 + 24]
+ pshufb m5, m1
+
+ mova m10, [r4 + 0 * mmsize]
+
+ INTRA_PRED_ANG32_CAL_ROW
+ movu [r0], m7
+ movu [r0 + r1], m6
+
+ ;row[2, 3]
+ mova m10, [r4 + 1 * mmsize]
+
+ INTRA_PRED_ANG32_CAL_ROW
+ movu [r0 + 2 * r1], m7
+ movu [r0 + r3], m6
+
+ ;row[4, 5]
+ mova m10, [r4 + 2 * mmsize]
+ lea r0, [r0 + 4 * r1]
+
+ INTRA_PRED_ANG32_CAL_ROW
+ movu [r0], m7
+ movu [r0 + r1], m6
+
+ ;row[6, 7]
+ mova m10, [r4 + 3 * mmsize]
+
+ INTRA_PRED_ANG32_CAL_ROW
+ movu [r0 + 2 * r1], m7
+ movu [r0 + r3], m6
+
+ ;row[8, 9]
+ add r4, 4 * mmsize
+ lea r0, [r0 + 4 * r1]
+ mova m10, [r4 + 0 * mmsize]
+
+ INTRA_PRED_ANG32_CAL_ROW
+ movu [r0], m7
+ movu [r0 + r1], m6
+
+ ;row[10, 11]
+ mova m10, [r4 + 1 * mmsize]
+
+ INTRA_PRED_ANG32_CAL_ROW
+ movu [r0 + 2 * r1], m7
+ movu [r0 + r3], m6
+
+ ;row[12, 13]
+ mova m10, [r4 + 2 * mmsize]
+ lea r0, [r0 + 4 * r1]
+
+ INTRA_PRED_ANG32_CAL_ROW
+ movu [r0], m7
+ movu [r0 + r1], m6
+
+ ;row[14, 15]
+ mova m10, [r4 + 3 * mmsize]
+
+ INTRA_PRED_ANG32_CAL_ROW
+ movu [r0 + 2 * r1], m7
+ movu [r0 + r3], m6
+
+ ;row[16, 17]
+ movu xm2, [r2 - 1]
+ pinsrb xm2, [r2 + 80], 0
+ vinserti128 m2, m2, xm2, 1
+ pshufb m2, m1
+ vbroadcasti128 m3, [r2 + 7]
+ pshufb m3, m1
+ vbroadcasti128 m4, [r2 + 15]
+ pshufb m4, m1
+ vbroadcasti128 m5, [r2 + 23]
+ pshufb m5, m1
+
+ add r4, 4 * mmsize
+ lea r0, [r0 + 4 * r1]
+ mova m10, [r4 + 0 * mmsize]
+
+ INTRA_PRED_ANG32_CAL_ROW
+ movu [r0], m7
+ movu [r0 + r1], m6
+
+ ;row[18, 19]
+ mova m10, [r4 + 1 * mmsize]
+
+ INTRA_PRED_ANG32_CAL_ROW
+ movu [r0 + 2 * r1], m7
+ movu [r0 + r3], m6
+
+ ;row[20, 21]
+ mova m10, [r4 + 2 * mmsize]
+ lea r0, [r0 + 4 * r1]
+
+ INTRA_PRED_ANG32_CAL_ROW
+ movu [r0], m7
+ movu [r0 + r1], m6
+
+ ;row[22, 23]
+ mova m10, [r4 + 3 * mmsize]
+
+ INTRA_PRED_ANG32_CAL_ROW
+ movu [r0 + 2 * r1], m7
+ movu [r0 + r3], m6
+
+ ;row[24, 25]
+ add r4, 4 * mmsize
+ lea r0, [r0 + 4 * r1]
+ mova m10, [r4 + 0 * mmsize]
+
+ INTRA_PRED_ANG32_CAL_ROW
+ movu [r0], m7
+ movu [r0 + r1], m6
+
+ ;row[26, 27]
+ mova m10, [r4 + 1 * mmsize]
+
+ INTRA_PRED_ANG32_CAL_ROW
+ movu [r0 + 2 * r1], m7
+ movu [r0 + r3], m6
+
+ ;row[28, 29]
+ mova m10, [r4 + 2 * mmsize]
+ lea r0, [r0 + 4 * r1]
+
+ INTRA_PRED_ANG32_CAL_ROW
+ movu [r0], m7
+ movu [r0 + r1], m6
+
+ ;row[30, 31]
+ mova m10, [r4 + 3 * mmsize]
+
+ INTRA_PRED_ANG32_CAL_ROW
+ movu [r0 + 2 * r1], m7
+ movu [r0 + r3], m6
+ RET
%endif
More information about the x265-devel
mailing list