[x265] [PATCH] asm: new algorithm for intra_ang_16 mode 5, improved over 16% than previous avx2 code
rajesh at multicorewareinc.com
rajesh at multicorewareinc.com
Thu Aug 6 09:46:28 CEST 2015
# HG changeset patch
# User Rajesh Paulraj<rajesh at multicorewareinc.com>
# Date 1438847047 -19800
# Thu Aug 06 13:14:07 2015 +0530
# Node ID ac57cc4fedfc9fab8311a67a49e6c1c6259f6ec3
# Parent e5d57775bbef81e37bc028c27d61a0a20e64bc9e
asm: new algorithm for intra_ang_16 mode 5, improved over 16% than previous avx2 code
diff -r e5d57775bbef -r ac57cc4fedfc source/common/x86/intrapred8.asm
--- a/source/common/x86/intrapred8.asm Thu Aug 06 13:08:58 2015 +0530
+++ b/source/common/x86/intrapred8.asm Thu Aug 06 13:14:07 2015 +0530
@@ -259,17 +259,6 @@
db 18, 14, 18, 14, 18, 14, 18, 14, 18, 14, 18, 14, 18, 14, 18, 14, 1, 31, 1, 31, 1, 31, 1, 31, 1, 31, 1, 31, 1, 31, 1, 31
db 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16
-
-ALIGN 32
-c_ang16_mode_5: db 15, 17, 15, 17, 15, 17, 15, 17, 15, 17, 15, 17, 15, 17, 15, 17, 7, 25, 7, 25, 7, 25, 7, 25, 7, 25, 7, 25, 7, 25, 7, 25
- db 30, 2, 30, 2, 30, 2, 30, 2, 30, 2, 30, 2, 30, 2, 30, 2, 22, 10, 22, 10, 22, 10, 22, 10, 22, 10, 22, 10, 22, 10, 22, 10
- db 13, 19, 13, 19, 13, 19, 13, 19, 13, 19, 13, 19, 13, 19, 13, 19, 5, 27, 5, 27, 5, 27, 5, 27, 5, 27, 5, 27, 5, 27, 5, 27
- db 28, 4, 28, 4, 28, 4, 28, 4, 28, 4, 28, 4, 28, 4, 28, 4, 20, 12, 20, 12, 20, 12, 20, 12, 20, 12, 20, 12, 20, 12, 20, 12
- db 11, 21, 11, 21, 11, 21, 11, 21, 11, 21, 11, 21, 11, 21, 11, 21, 3, 29, 3, 29, 3, 29, 3, 29, 3, 29, 3, 29, 3, 29, 3, 29
- db 26, 6, 26, 6, 26, 6, 26, 6, 26, 6, 26, 6, 26, 6, 26, 6, 18, 14, 18, 14, 18, 14, 18, 14, 18, 14, 18, 14, 18, 14, 18, 14
- db 9, 23, 9, 23, 9, 23, 9, 23, 9, 23, 9, 23, 9, 23, 9, 23, 1, 31, 1, 31, 1, 31, 1, 31, 1, 31, 1, 31, 1, 31, 1, 31
- db 24, 8, 24, 8, 24, 8, 24, 8, 24, 8, 24, 8, 24, 8, 24, 8, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16
-
ALIGN 32
c_ang16_mode_24: db 5, 27, 5, 27, 5, 27, 5, 27, 5, 27, 5, 27, 5, 27, 5, 27, 10, 22, 10, 22, 10, 22, 10, 22, 10, 22, 10, 22, 10, 22, 10, 22
db 15, 17, 15, 17, 15, 17, 15, 17, 15, 17, 15, 17, 15, 17, 15, 17, 20, 12, 20, 12, 20, 12, 20, 12, 20, 12, 20, 12, 20, 12, 20, 12
@@ -13812,102 +13801,97 @@
call ang16_mode_4_32
RET
+
+cglobal ang16_mode_5
+ ; rows 0 to 7
+ movu m0, [r2 + 1] ; [32 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1]
+ movu m1, [r2 + 2] ; [33 32 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2]
+
+ punpckhbw m2, m0, m1 ; [33 32 32 31 31 30 30 29 29 28 28 27 27 26 26 25 17 16 16 15 15 14 14 13 13 12 12 11 11 10 10 9]
+ punpcklbw m0, m1 ; [25 24 24 23 23 22 22 21 21 20 20 19 19 18 18 17 9 8 8 7 7 6 6 5 5 4 4 3 3 2 2 1]
+ vextracti128 xm1, m0, 1
+ vperm2i128 m0, m0, m2, 0x20 ; [17 16 16 15 15 14 14 13 13 12 12 11 11 10 10 9 9 8 8 7 7 6 6 5 5 4 4 3 3 2 2 1]
+ vperm2i128 m2, m2, m1, 0x20 ; [25 24 24 23 23 22 22 21 21 20 20 19 19 18 18 17 17 16 16 15 15 14 14 13 13 12 12 11 11 10 10 9]
+
+ pmaddubsw m4, m0, [r3 + 1 * 32] ; [17]
+ pmulhrsw m4, m7
+
+ palignr m1, m2, m0, 2
+ pmaddubsw m5, m1, [r3 - 14 * 32] ; [2]
+ pmulhrsw m5, m7
+
+ palignr m3, m2, m0, 4
+ pmaddubsw m6, m1, [r3 + 3 * 32] ; [19]
+ pmulhrsw m6, m7
+ pmaddubsw m8, m3, [r3 - 12 * 32] ; [4]
+ pmulhrsw m8, m7
+ pmaddubsw m9, m3, [r3 + 5 * 32] ; [21]
+ pmulhrsw m9, m7
+
+ palignr m3, m2, m0, 6
+ pmaddubsw m10, m3, [r3 - 10 * 32] ; [6]
+ pmulhrsw m10, m7
+
+ palignr m1, m2, m0, 8
+ pmaddubsw m11, m3, [r3 + 7 * 32] ; [23]
+ pmulhrsw m11, m7
+ pmaddubsw m12, m1, [r3 - 8 * 32] ; [8]
+ pmulhrsw m12, m7
+
+ ; rows 8 to 15
+ pmaddubsw m3, m1, [r3 + 9 * 32] ; [25]
+ pmulhrsw m3, m7
+ packuswb m4, m3
+
+ palignr m1, m2, m0, 10
+ pmaddubsw m3, m1, [r3 - 6 * 32] ; [10]
+ pmulhrsw m3, m7
+ packuswb m5, m3
+
+ pmaddubsw m3, m1, [r3 + 11 * 32] ; [27]
+ pmulhrsw m3, m7
+ packuswb m6, m3
+
+ palignr m1, m2, m0, 12
+ pmaddubsw m3, m1, [r3 - 4 * 32] ; [12]
+ pmulhrsw m3, m7
+ packuswb m8, m3
+
+ pmaddubsw m3, m1, [r3 + 13 * 32] ; [29]
+ pmulhrsw m3, m7
+ packuswb m9, m3
+
+ palignr m1, m2, m0, 14
+ pmaddubsw m3, m1, [r3 - 2 * 32] ; [14]
+ pmulhrsw m3, m7
+ packuswb m10, m3
+
+ pmaddubsw m3, m1, [r3 + 15 * 32] ; [31]
+ pmulhrsw m3, m7
+ packuswb m11, m3
+
+ palignr m1, m2, m0, 16
+ pmaddubsw m1, [r3] ; [16]
+ pmulhrsw m1, m7
+ packuswb m12, m1
+
+ TRANSPOSE_STORE_8x32 4, 5, 6, 8, 9, 10, 11, 12, 0, 1, 2, 3
+ ret
+
+INIT_YMM avx2
+cglobal intra_pred_ang16_5, 3, 7, 13
+ add r2, 32
+ lea r3, [ang_table_avx2 + 16 * 32]
+ lea r5, [r1 * 3] ; r5 -> 3 * stride
+ lea r6, [r1 * 4] ; r6 -> 4 * stride
+ mova m7, [pw_1024]
+ clc
+
+ call ang16_mode_5
+ RET
%endif ; ARCH_X86_64
INIT_YMM avx2
-cglobal intra_pred_ang16_5, 3, 6, 12
- mova m11, [pw_1024]
- lea r5, [intra_pred_shuff_0_8]
-
- movu xm9, [r2 + 1 + 32]
- pshufb xm9, [r5]
- movu xm10, [r2 + 9 + 32]
- pshufb xm10, [r5]
-
- movu xm7, [r2 + 5 + 32]
- pshufb xm7, [r5]
- vinserti128 m9, m9, xm7, 1
-
- movu xm8, [r2 + 13 + 32]
- pshufb xm8, [r5]
- vinserti128 m10, m10, xm8, 1
-
- lea r3, [3 * r1]
- lea r4, [c_ang16_mode_5]
-
- INTRA_PRED_ANG16_CAL_ROW m0, m1, 0
-
- movu xm9, [r2 + 2 + 32]
- pshufb xm9, [r5]
- movu xm10, [r2 + 10 + 32]
- pshufb xm10, [r5]
-
- movu xm7, [r2 + 6 + 32]
- pshufb xm7, [r5]
- vinserti128 m9, m9, xm7, 1
-
- movu xm8, [r2 + 14 + 32]
- pshufb xm8, [r5]
- vinserti128 m10, m10, xm8, 1
-
- INTRA_PRED_ANG16_CAL_ROW m1, m2, 1
- INTRA_PRED_ANG16_CAL_ROW m2, m3, 2
-
- movu xm9, [r2 + 3 + 32]
- pshufb xm9, [r5]
- movu xm10, [r2 + 11 + 32]
- pshufb xm10, [r5]
-
- movu xm7, [r2 + 7 + 32]
- pshufb xm7, [r5]
- vinserti128 m9, m9, xm7, 1
-
- movu xm8, [r2 + 15 + 32]
- pshufb xm8, [r5]
- vinserti128 m10, m10, xm8, 1
-
- INTRA_PRED_ANG16_CAL_ROW m3, m4, 3
-
- add r4, 4 * mmsize
-
- INTRA_PRED_ANG16_CAL_ROW m4, m5, 0
-
- movu xm9, [r2 + 4 + 32]
- pshufb xm9, [r5]
- movu xm10, [r2 + 12 + 32]
- pshufb xm10, [r5]
-
- movu xm7, [r2 + 8 + 32]
- pshufb xm7, [r5]
- vinserti128 m9, m9, xm7, 1
-
- movu xm8, [r2 + 16 + 32]
- pshufb xm8, [r5]
- vinserti128 m10, m10, xm8, 1
-
- INTRA_PRED_ANG16_CAL_ROW m5, m6, 1
- INTRA_PRED_ANG16_CAL_ROW m6, m7, 2
-
- movu xm9, [r2 + 5 + 32]
- pshufb xm9, [r5]
- movu xm10, [r2 + 13 + 32]
- pshufb xm10, [r5]
-
- movu xm7, [r2 + 9 + 32]
- pshufb xm7, [r5]
- vinserti128 m9, m9, xm7, 1
-
- movu xm8, [r2 + 17 + 32]
- pshufb xm8, [r5]
- vinserti128 m10, m10, xm8, 1
-
- INTRA_PRED_ANG16_CAL_ROW m7, m8, 3
-
- ; transpose and store
- INTRA_PRED_TRANS_STORE_16x16
- RET
-
-INIT_YMM avx2
cglobal intra_pred_ang16_6, 3, 6, 12
mova m11, [pw_1024]
lea r5, [intra_pred_shuff_0_8]
More information about the x265-devel
mailing list