History

avcodec/la: add LSX optimization for h264 idct.

loongson_asm.S is LoongArch asm optimization helper. Add functions: ff_h264_idct_add_8_lsx ff_h264_idct8_add_8_lsx ff_h264_idct_dc_add_8_lsx ff_h264_idct8_dc_add_8_lsx ff_h264_idct_add16_8_lsx ff_h264_idct8_add4_8_lsx ff_h264_idct_add8_8_lsx ff_h264_idct_add8_422_8_lsx ff_h264_idct_add16_intra_8_lsx ff_h264_luma_dc_dequant_idct_8_lsx Replaced function(LSX is sufficient for these functions): ff_h264_idct_add_lasx ff_h264_idct4x4_addblk_dc_lasx ff_h264_idct_add16_lasx ff_h264_idct8_add4_lasx ff_h264_idct_add8_lasx ff_h264_idct_add8_422_lasx ff_h264_idct_add16_intra_lasx ff_h264_deq_idct_luma_dc_lasx Renamed functions: ff_h264_idct8_addblk_lasx ==> ff_h264_idct8_add_8_lasx ff_h264_idct8_dc_addblk_lasx ==> ff_h264_idct8_dc_add_8_lasx ./configure --disable-lasx ffmpeg -i 1_h264_1080p_30fps_3Mbps.mp4 -f rawvideo -y /dev/null -an before: 155fps after: 161fps

Shiyou Yin committed 3 years ago

Commit e1b6ecd1 parent 90fba279 files · +1848 −543Browse files

Showing the changes to libavcodec (9 of 9 files). Show all

Unified · Split

9 files
  1. libavcodec/loongarch/Makefile +2 −1
  2. libavcodec/loongarch/h264_deblock_lasx.c +1 −1
  3. libavcodec/loongarch/h264dsp_init_loongarch.c +24 −15
  4. libavcodec/loongarch/h264dsp_lasx.c +1 −1
  5. libavcodec/loongarch/h264dsp_loongarch.h +33 −27
  6. libavcodec/loongarch/h264idct.S +658 −0
  7. libavcodec/loongarch/h264idct_lasx.c +0 −498
  8. libavcodec/loongarch/h264idct_loongarch.c +184 −0
  9. libavcodec/loongarch/loongson_asm.S +945 −0
libavcodec/loongarch/Makefile+2−1
@@ -12,7 +12,6 @@ OBJS-$(CONFIG_HEVC_DECODER) += loongarch/hevcdsp_init_loongarch.o · 11 unmodified lines
1212 LASX-OBJS-$(CONFIG_H264CHROMA) += loongarch/h264chroma_lasx.o
1313 LASX-OBJS-$(CONFIG_H264QPEL) += loongarch/h264qpel_lasx.o
1414 LASX-OBJS-$(CONFIG_H264DSP) += loongarch/h264dsp_lasx.o \
15− loongarch/h264idct_lasx.o \
1615 loongarch/h264_deblock_lasx.o
1716 LASX-OBJS-$(CONFIG_H264PRED) += loongarch/h264_intrapred_lasx.o
1817 LASX-OBJS-$(CONFIG_VC1_DECODER) += loongarch/vc1dsp_lasx.o
@@ -31,3 +30,5 @@ LSX-OBJS-$(CONFIG_HEVC_DECODER) += loongarch/hevcdsp_lsx.o \ · 12 unmodified lines
3130 loongarch/hevc_mc_bi_lsx.o \
3231 loongarch/hevc_mc_uni_lsx.o \
3332 loongarch/hevc_mc_uniw_lsx.o
33+LSX-OBJS-$(CONFIG_H264DSP) += loongarch/h264idct.o \
34+ loongarch/h264idct_loongarch.o
libavcodec/loongarch/h264_deblock_lasx.c+1−1
@@ -20,7 +20,7 @@ · 19 unmodified lines
2020 */
2121
2222 #include "libavcodec/bit_depth_template.c"
23−#include "h264dsp_lasx.h"
23+#include "h264dsp_loongarch.h"
2424 #include "libavutil/loongarch/loongson_intrinsics.h"
2525
2626 #define H264_LOOP_FILTER_STRENGTH_ITERATION_LASX(edges, step, mask_mv, dir, \
libavcodec/loongarch/h264dsp_init_loongarch.c+24−15
@@ -21,13 +21,32 @@ · 20 unmodified lines
2121 */
2222
2323 #include "libavutil/loongarch/cpu.h"
24−#include "h264dsp_lasx.h"
24+#include "h264dsp_loongarch.h"
2525
2626 av_cold void ff_h264dsp_init_loongarch(H264DSPContext *c, const int bit_depth,
2727 const int chroma_format_idc)
2828 {
2929 int cpu_flags = av_get_cpu_flags();
3030
31+ if (have_lsx(cpu_flags)) {
32+ if (bit_depth == 8) {
33+ c->h264_idct_add = ff_h264_idct_add_8_lsx;
34+ c->h264_idct8_add = ff_h264_idct8_add_8_lsx;
35+ c->h264_idct_dc_add = ff_h264_idct_dc_add_8_lsx;
36+ c->h264_idct8_dc_add = ff_h264_idct8_dc_add_8_lsx;
37+
38+ if (chroma_format_idc <= 1)
39+ c->h264_idct_add8 = ff_h264_idct_add8_8_lsx;
40+ else
41+ c->h264_idct_add8 = ff_h264_idct_add8_422_8_lsx;
42+
43+ c->h264_idct_add16 = ff_h264_idct_add16_8_lsx;
44+ c->h264_idct8_add4 = ff_h264_idct8_add4_8_lsx;
45+ c->h264_luma_dc_dequant_idct = ff_h264_luma_dc_dequant_idct_8_lsx;
46+ c->h264_idct_add16intra = ff_h264_idct_add16_intra_8_lsx;
47+ }
48+ }
49+#if HAVE_LASX
3150 if (have_lasx(cpu_flags)) {
3251 if (chroma_format_idc <= 1)
3352 c->h264_loop_filter_strength = ff_h264_loop_filter_strength_lasx;
@@ -56,20 +75,10 @@ av_cold void ff_h264dsp_init_loongarch(H264DSPContext *c, const int bit_depth, · 22 unmodified lines
5675 c->biweight_h264_pixels_tab[1] = ff_biweight_h264_pixels8_8_lasx;
5776 c->biweight_h264_pixels_tab[2] = ff_biweight_h264_pixels4_8_lasx;
5877
59− c->h264_idct_add = ff_h264_idct_add_lasx;
60− c->h264_idct8_add = ff_h264_idct8_addblk_lasx;
61− c->h264_idct_dc_add = ff_h264_idct4x4_addblk_dc_lasx;
62− c->h264_idct8_dc_add = ff_h264_idct8_dc_addblk_lasx;
63− c->h264_idct_add16 = ff_h264_idct_add16_lasx;
64− c->h264_idct8_add4 = ff_h264_idct8_add4_lasx;
65−
66− if (chroma_format_idc <= 1)
67− c->h264_idct_add8 = ff_h264_idct_add8_lasx;
68− else
69− c->h264_idct_add8 = ff_h264_idct_add8_422_lasx;
70−
71− c->h264_idct_add16intra = ff_h264_idct_add16_intra_lasx;
72− c->h264_luma_dc_dequant_idct = ff_h264_deq_idct_luma_dc_lasx;
78+ c->h264_idct8_add = ff_h264_idct8_add_8_lasx;
79+ c->h264_idct8_dc_add = ff_h264_idct8_dc_add_8_lasx;
80+ c->h264_idct8_add4 = ff_h264_idct8_add4_8_lasx;
7381 }
7482 }
83+#endif // #if HAVE_LASX
7584 }
libavcodec/loongarch/h264dsp_lasx.c+1−1
@@ -23,7 +23,7 @@ · 22 unmodified lines
2323 */
2424
2525 #include "libavutil/loongarch/loongson_intrinsics.h"
26−#include "h264dsp_lasx.h"
26+#include "h264dsp_loongarch.h"
2727
2828 #define AVC_LPF_P1_OR_Q1(p0_or_q0_org_in, q0_or_p0_org_in, \
2929 p1_or_q1_org_in, p2_or_q2_org_in, \
libavcodec/loongarch/h264dsp_loongarch.hrenamed from libavcodec/loongarch/h264dsp_lasx.h+33−27
@@ -1,5 +1,5 @@
11 /*
2− * Copyright (c) 2021 Loongson Technology Corporation Limited
2+ * Copyright (c) 2023 Loongson Technology Corporation Limited
33 * Contributed by Shiyou Yin <yinshiyou-hf@loongson.cn>
44 * Xiwei Gu <guxiwei-hf@loongson.cn>
55 *
@@ -20,11 +20,34 @@ · 14 unmodified lines
2020 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
2121 */
2222
23−#ifndef AVCODEC_LOONGARCH_H264DSP_LASX_H
24−#define AVCODEC_LOONGARCH_H264DSP_LASX_H
23+#ifndef AVCODEC_LOONGARCH_H264DSP_LOONGARCH_H
24+#define AVCODEC_LOONGARCH_H264DSP_LOONGARCH_H
2525
2626 #include "libavcodec/h264dec.h"
27+#include "config.h"
2728
29+void ff_h264_idct_add_8_lsx(uint8_t *dst, int16_t *src, int dst_stride);
30+void ff_h264_idct8_add_8_lsx(uint8_t *dst, int16_t *src, int dst_stride);
31+void ff_h264_idct_dc_add_8_lsx(uint8_t *dst, int16_t *src, int dst_stride);
32+void ff_h264_idct8_dc_add_8_lsx(uint8_t *dst, int16_t *src, int dst_stride);
33+void ff_h264_luma_dc_dequant_idct_8_lsx(int16_t *_output, int16_t *_input, int qmul);
34+void ff_h264_idct_add16_8_lsx(uint8_t *dst, const int32_t *blk_offset,
35+ int16_t *block, int32_t dst_stride,
36+ const uint8_t nzc[15 * 8]);
37+void ff_h264_idct8_add4_8_lsx(uint8_t *dst, const int32_t *blk_offset,
38+ int16_t *block, int32_t dst_stride,
39+ const uint8_t nzc[15 * 8]);
40+void ff_h264_idct_add8_8_lsx(uint8_t **dst, const int32_t *blk_offset,
41+ int16_t *block, int32_t dst_stride,
42+ const uint8_t nzc[15 * 8]);
43+void ff_h264_idct_add8_422_8_lsx(uint8_t **dst, const int32_t *blk_offset,
44+ int16_t *block, int32_t dst_stride,
45+ const uint8_t nzc[15 * 8]);
46+void ff_h264_idct_add16_intra_8_lsx(uint8_t *dst, const int32_t *blk_offset,
47+ int16_t *block, int32_t dst_stride,
48+ const uint8_t nzc[15 * 8]);
49+
50+#if HAVE_LASX
2851 void ff_h264_h_lpf_luma_8_lasx(uint8_t *src, ptrdiff_t stride,
2952 int alpha, int beta, int8_t *tc0);
3053 void ff_h264_v_lpf_luma_8_lasx(uint8_t *src, ptrdiff_t stride,
@@ -65,33 +88,16 @@ void ff_weight_h264_pixels4_8_lasx(uint8_t *src, ptrdiff_t stride, · 34 unmodified lines
6588 void ff_h264_add_pixels4_8_lasx(uint8_t *_dst, int16_t *_src, int stride);
6689
6790 void ff_h264_add_pixels8_8_lasx(uint8_t *_dst, int16_t *_src, int stride);
68−void ff_h264_idct_add_lasx(uint8_t *dst, int16_t *src, int32_t dst_stride);
69−void ff_h264_idct8_addblk_lasx(uint8_t *dst, int16_t *src, int32_t dst_stride);
70−void ff_h264_idct4x4_addblk_dc_lasx(uint8_t *dst, int16_t *src,
71− int32_t dst_stride);
72−void ff_h264_idct8_dc_addblk_lasx(uint8_t *dst, int16_t *src,
91+void ff_h264_idct8_add_8_lasx(uint8_t *dst, int16_t *src, int32_t dst_stride);
92+void ff_h264_idct8_dc_add_8_lasx(uint8_t *dst, int16_t *src,
7393 int32_t dst_stride);
74−void ff_h264_idct_add16_lasx(uint8_t *dst, const int32_t *blk_offset,
75− int16_t *block, int32_t dst_stride,
76− const uint8_t nzc[15 * 8]);
77−void ff_h264_idct8_add4_lasx(uint8_t *dst, const int32_t *blk_offset,
78− int16_t *block, int32_t dst_stride,
79− const uint8_t nzc[15 * 8]);
80−void ff_h264_idct_add8_lasx(uint8_t **dst, const int32_t *blk_offset,
81− int16_t *block, int32_t dst_stride,
82− const uint8_t nzc[15 * 8]);
83−void ff_h264_idct_add8_422_lasx(uint8_t **dst, const int32_t *blk_offset,
84− int16_t *block, int32_t dst_stride,
85− const uint8_t nzc[15 * 8]);
86−void ff_h264_idct_add16_intra_lasx(uint8_t *dst, const int32_t *blk_offset,
87− int16_t *block, int32_t dst_stride,
88− const uint8_t nzc[15 * 8]);
89−void ff_h264_deq_idct_luma_dc_lasx(int16_t *dst, int16_t *src,
90− int32_t de_qval);
91−
94+void ff_h264_idct8_add4_8_lasx(uint8_t *dst, const int32_t *blk_offset,
95+ int16_t *block, int32_t dst_stride,
96+ const uint8_t nzc[15 * 8]);
9297 void ff_h264_loop_filter_strength_lasx(int16_t bS[2][4][4], uint8_t nnz[40],
9398 int8_t ref[2][40], int16_t mv[2][40][2],
9499 int bidir, int edges, int step,
95100 int mask_mv0, int mask_mv1, int field);
101+#endif // #if HAVE_LASX
96102
97−#endif // #ifndef AVCODEC_LOONGARCH_H264DSP_LASX_H
103+#endif // #ifndef AVCODEC_LOONGARCH_H264DSP_LOONGARCH_H
libavcodec/loongarch/h264idct.S+658−0
@@ -0,0 +1,658 @@
1+/*
2+ * Loongson LASX optimized h264idct
3+ *
4+ * Copyright (c) 2023 Loongson Technology Corporation Limited
5+ * Contributed by Shiyou Yin <yinshiyou-hf@loongson.cn>
6+ *
7+ * This file is part of FFmpeg.
8+ *
9+ * FFmpeg is free software; you can redistribute it and/or
10+ * modify it under the terms of the GNU Lesser General Public
11+ * License as published by the Free Software Foundation; either
12+ * version 2.1 of the License, or (at your option) any later version.
13+ *
14+ * FFmpeg is distributed in the hope that it will be useful,
15+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
16+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17+ * Lesser General Public License for more details.
18+ *
19+ * You should have received a copy of the GNU Lesser General Public
20+ * License along with FFmpeg; if not, write to the Free Software
21+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22+ */
23+
24+#include "loongson_asm.S"
25+
26+/*
27+ * #define FUNC2(a, b, c) FUNC3(a, b, c)
28+ * #define FUNCC(a) FUNC2(a, BIT_DEPTH, _c)
29+ * void FUNCC(ff_h264_idct_add)(uint8_t *_dst, int16_t *_block, int stride)
30+ * LSX optimization is enough for this function.
31+ */
32+function ff_h264_idct_add_8_lsx
33+ fld.d f0, a1, 0
34+ fld.d f1, a1, 8
35+ fld.d f2, a1, 16
36+ fld.d f3, a1, 24
37+ vxor.v vr7, vr7, vr7
38+ add.d t2, a2, a2
39+ add.d t3, t2, a2
40+ vst vr7, a1, 0
41+ vst vr7, a1, 16
42+
43+ vadd.h vr4, vr0, vr2
44+ vsub.h vr5, vr0, vr2
45+ vsrai.h vr6, vr1, 1
46+ vsrai.h vr7, vr3, 1
47+ vsub.h vr6, vr6, vr3
48+ vadd.h vr7, vr1, vr7
49+ LSX_BUTTERFLY_4_H vr4, vr5, vr6, vr7, vr0, vr1, vr2, vr3
50+ LSX_TRANSPOSE4x4_H vr0, vr1, vr2, vr3, vr0, vr1, vr2, vr3, vr4, vr5
51+ vadd.h vr4, vr0, vr2
52+ vsub.h vr5, vr0, vr2
53+ vsrai.h vr6, vr1, 1
54+ vsrai.h vr7, vr3, 1
55+ vsub.h vr6, vr6, vr3
56+ vadd.h vr7, vr1, vr7
57+ LSX_BUTTERFLY_4_H vr4, vr5, vr6, vr7, vr0, vr1, vr2, vr3
58+
59+ fld.s f4, a0, 0
60+ fldx.s f5, a0, a2
61+ fldx.s f6, a0, t2
62+ fldx.s f7, a0, t3
63+
64+ vsrari.h vr0, vr0, 6
65+ vsrari.h vr1, vr1, 6
66+ vsrari.h vr2, vr2, 6
67+ vsrari.h vr3, vr3, 6
68+
69+ vsllwil.hu.bu vr4, vr4, 0
70+ vsllwil.hu.bu vr5, vr5, 0
71+ vsllwil.hu.bu vr6, vr6, 0
72+ vsllwil.hu.bu vr7, vr7, 0
73+ vadd.h vr0, vr0, vr4
74+ vadd.h vr1, vr1, vr5
75+ vadd.h vr2, vr2, vr6
76+ vadd.h vr3, vr3, vr7
77+ vssrarni.bu.h vr1, vr0, 0
78+ vssrarni.bu.h vr3, vr2, 0
79+
80+ vbsrl.v vr0, vr1, 8
81+ vbsrl.v vr2, vr3, 8
82+ fst.s f1, a0, 0
83+ fstx.s f0, a0, a2
84+ fstx.s f3, a0, t2
85+ fstx.s f2, a0, t3
86+endfunc
87+
88+/*
89+ * #define FUNC2(a, b, c) FUNC3(a, b, c)
90+ * #define FUNCC(a) FUNC2(a, BIT_DEPTH, _c)
91+ * void FUNCC(ff_h264_idct8_add)(uint8_t *_dst, int16_t *_block, int stride)
92+ */
93+function ff_h264_idct8_add_8_lsx
94+ ld.h t0, a1, 0
95+ add.d t2, a2, a2
96+ add.d t3, t2, a2
97+ add.d t4, t3, a2
98+ add.d t5, t4, a2
99+ add.d t6, t5, a2
100+ add.d t7, t6, a2
101+ addi.w t0, t0, 32
102+ st.h t0, a1, 0
103+
104+ vld vr0, a1, 0
105+ vld vr1, a1, 16
106+ vld vr2, a1, 32
107+ vld vr3, a1, 48
108+ vld vr4, a1, 64
109+ vld vr5, a1, 80
110+ vld vr6, a1, 96
111+ vld vr7, a1, 112
112+ vxor.v vr8, vr8, vr8
113+ vst vr8, a1, 0
114+ vst vr8, a1, 16
115+ vst vr8, a1, 32
116+ vst vr8, a1, 48
117+ vst vr8, a1, 64
118+ vst vr8, a1, 80
119+ vst vr8, a1, 96
120+ vst vr8, a1, 112
121+
122+ vadd.h vr18, vr0, vr4
123+ vsub.h vr19, vr0, vr4
124+ vsrai.h vr20, vr2, 1
125+ vsrai.h vr21, vr6, 1
126+ vsub.h vr20, vr20, vr6
127+ vadd.h vr21, vr21, vr2
128+ LSX_BUTTERFLY_4_H vr18, vr19, vr20, vr21, vr10, vr12, vr14, vr16
129+ vsrai.h vr11, vr7, 1
130+ vsrai.h vr13, vr3, 1
131+ vsrai.h vr15, vr5, 1
132+ vsrai.h vr17, vr1, 1
133+ vsub.h vr11, vr5, vr11
134+ vsub.h vr13, vr7, vr13
135+ vadd.h vr15, vr7, vr15
136+ vadd.h vr17, vr5, vr17
137+ vsub.h vr11, vr11, vr7
138+ vsub.h vr13, vr13, vr3
139+ vadd.h vr15, vr15, vr5
140+ vadd.h vr17, vr17, vr1
141+ vsub.h vr11, vr11, vr3
142+ vadd.h vr13, vr13, vr1
143+ vsub.h vr15, vr15, vr1
144+ vadd.h vr17, vr17, vr3
145+ vsrai.h vr18, vr11, 2
146+ vsrai.h vr19, vr13, 2
147+ vsrai.h vr20, vr15, 2
148+ vsrai.h vr21, vr17, 2
149+ vadd.h vr11, vr11, vr21
150+ vadd.h vr13, vr13, vr20
151+ vsub.h vr15, vr19, vr15
152+ vsub.h vr17, vr17, vr18
153+ LSX_BUTTERFLY_8_H vr10, vr16, vr12, vr14, vr13, vr15, vr11, vr17, \
154+ vr0, vr3, vr1, vr2, vr5, vr6, vr4, vr7
155+
156+ LSX_TRANSPOSE8x8_H vr0, vr1, vr2, vr3, vr4, vr5, vr6, vr7, \
157+ vr0, vr1, vr2, vr3, vr4, vr5, vr6, vr7, \
158+ vr10, vr11, vr12, vr13, vr14, vr15, vr16, vr17
159+ vexth.w.h vr20, vr0
160+ vexth.w.h vr21, vr1
161+ vexth.w.h vr22, vr2
162+ vexth.w.h vr23, vr3
163+ vexth.w.h vr8, vr4
164+ vexth.w.h vr9, vr5
165+ vexth.w.h vr18, vr6
166+ vexth.w.h vr19, vr7
167+ vsllwil.w.h vr0, vr0, 0
168+ vsllwil.w.h vr1, vr1, 0
169+ vsllwil.w.h vr2, vr2, 0
170+ vsllwil.w.h vr3, vr3, 0
171+ vsllwil.w.h vr4, vr4, 0
172+ vsllwil.w.h vr5, vr5, 0
173+ vsllwil.w.h vr6, vr6, 0
174+ vsllwil.w.h vr7, vr7, 0
175+
176+ vadd.w vr11, vr0, vr4
177+ vsub.w vr13, vr0, vr4
178+ vsrai.w vr15, vr2, 1
179+ vsrai.w vr17, vr6, 1
180+ vsub.w vr15, vr15, vr6
181+ vadd.w vr17, vr17, vr2
182+ LSX_BUTTERFLY_4_W vr11, vr13, vr15, vr17, vr10, vr12, vr14, vr16
183+ vsrai.w vr11, vr7, 1
184+ vsrai.w vr13, vr3, 1
185+ vsrai.w vr15, vr5, 1
186+ vsrai.w vr17, vr1, 1
187+ vsub.w vr11, vr5, vr11
188+ vsub.w vr13, vr7, vr13
189+ vadd.w vr15, vr7, vr15
190+ vadd.w vr17, vr5, vr17
191+ vsub.w vr11, vr11, vr7
192+ vsub.w vr13, vr13, vr3
193+ vadd.w vr15, vr15, vr5
194+ vadd.w vr17, vr17, vr1
195+ vsub.w vr11, vr11, vr3
196+ vadd.w vr13, vr13, vr1
197+ vsub.w vr15, vr15, vr1
198+ vadd.w vr17, vr17, vr3
199+ vsrai.w vr0, vr11, 2
200+ vsrai.w vr1, vr13, 2
201+ vsrai.w vr2, vr15, 2
202+ vsrai.w vr3, vr17, 2
203+ vadd.w vr11, vr11, vr3
204+ vadd.w vr13, vr13, vr2
205+ vsub.w vr15, vr1, vr15
206+ vsub.w vr17, vr17, vr0
207+ LSX_BUTTERFLY_8_W vr10, vr12, vr14, vr16, vr11, vr13, vr15, vr17, \
208+ vr0, vr1, vr2, vr3, vr4, vr5, vr6, vr7
209+
210+ vadd.w vr11, vr20, vr8
211+ vsub.w vr13, vr20, vr8
212+ vsrai.w vr15, vr22, 1
213+ vsrai.w vr17, vr18, 1
214+ vsub.w vr15, vr15, vr18
215+ vadd.w vr17, vr17, vr22
216+ LSX_BUTTERFLY_4_W vr11, vr13, vr15, vr17, vr10, vr12, vr14, vr16
217+ vsrai.w vr11, vr19, 1
218+ vsrai.w vr13, vr23, 1
219+ vsrai.w vr15, vr9, 1
220+ vsrai.w vr17, vr21, 1
221+ vsub.w vr11, vr9, vr11
222+ vsub.w vr13, vr19, vr13
223+ vadd.w vr15, vr19, vr15
224+ vadd.w vr17, vr9, vr17
225+ vsub.w vr11, vr11, vr19
226+ vsub.w vr13, vr13, vr23
227+ vadd.w vr15, vr15, vr9
228+ vadd.w vr17, vr17, vr21
229+ vsub.w vr11, vr11, vr23
230+ vadd.w vr13, vr13, vr21
231+ vsub.w vr15, vr15, vr21
232+ vadd.w vr17, vr17, vr23
233+ vsrai.w vr20, vr11, 2
234+ vsrai.w vr21, vr13, 2
235+ vsrai.w vr22, vr15, 2
236+ vsrai.w vr23, vr17, 2
237+ vadd.w vr11, vr11, vr23
238+ vadd.w vr13, vr13, vr22
239+ vsub.w vr15, vr21, vr15
240+ vsub.w vr17, vr17, vr20
241+ LSX_BUTTERFLY_8_W vr10, vr12, vr14, vr16, vr11, vr13, vr15, vr17, \
242+ vr20, vr21, vr22, vr23, vr8, vr9, vr18, vr19
243+
244+ vld vr10, a0, 0
245+ vldx vr11, a0, a2
246+ vldx vr12, a0, t2
247+ vldx vr13, a0, t3
248+ vldx vr14, a0, t4
249+ vldx vr15, a0, t5
250+ vldx vr16, a0, t6
251+ vldx vr17, a0, t7
252+ vsrani.h.w vr20, vr0, 6
253+ vsrani.h.w vr21, vr1, 6
254+ vsrani.h.w vr22, vr2, 6
255+ vsrani.h.w vr23, vr3, 6
256+ vsrani.h.w vr8, vr4, 6
257+ vsrani.h.w vr9, vr5, 6
258+ vsrani.h.w vr18, vr6, 6
259+ vsrani.h.w vr19, vr7, 6
260+ vsllwil.hu.bu vr10, vr10, 0
261+ vsllwil.hu.bu vr11, vr11, 0
262+ vsllwil.hu.bu vr12, vr12, 0
263+ vsllwil.hu.bu vr13, vr13, 0
264+ vsllwil.hu.bu vr14, vr14, 0
265+ vsllwil.hu.bu vr15, vr15, 0
266+ vsllwil.hu.bu vr16, vr16, 0
267+ vsllwil.hu.bu vr17, vr17, 0
268+
269+ vadd.h vr0, vr20, vr10
270+ vadd.h vr1, vr21, vr11
271+ vadd.h vr2, vr22, vr12
272+ vadd.h vr3, vr23, vr13
273+ vadd.h vr4, vr8, vr14
274+ vadd.h vr5, vr9, vr15
275+ vadd.h vr6, vr18, vr16
276+ vadd.h vr7, vr19, vr17
277+ vssrarni.bu.h vr1, vr0, 0
278+ vssrarni.bu.h vr3, vr2, 0
279+ vssrarni.bu.h vr5, vr4, 0
280+ vssrarni.bu.h vr7, vr6, 0
281+ vbsrl.v vr0, vr1, 8
282+ vbsrl.v vr2, vr3, 8
283+ vbsrl.v vr4, vr5, 8
284+ vbsrl.v vr6, vr7, 8
285+ fst.d f1, a0, 0
286+ fstx.d f0, a0, a2
287+ fstx.d f3, a0, t2
288+ fstx.d f2, a0, t3
289+ fstx.d f5, a0, t4
290+ fstx.d f4, a0, t5
291+ fstx.d f7, a0, t6
292+ fstx.d f6, a0, t7
293+endfunc
294+
295+/*
296+ * #define FUNC2(a, b, c) FUNC3(a, b, c)
297+ * #define FUNCC(a) FUNC2(a, BIT_DEPTH, _c)
298+ * void FUNCC(ff_h264_idct8_add)(uint8_t *_dst, int16_t *_block, int stride)
299+ */
300+function ff_h264_idct8_add_8_lasx
301+ ld.h t0, a1, 0
302+ add.d t2, a2, a2
303+ add.d t3, t2, a2
304+ add.d t4, t3, a2
305+ add.d t5, t4, a2
306+ add.d t6, t5, a2
307+ add.d t7, t6, a2
308+ addi.w t0, t0, 32
309+ st.h t0, a1, 0
310+
311+ vld vr0, a1, 0
312+ vld vr1, a1, 16
313+ vld vr2, a1, 32
314+ vld vr3, a1, 48
315+ vld vr4, a1, 64
316+ vld vr5, a1, 80
317+ vld vr6, a1, 96
318+ vld vr7, a1, 112
319+ xvxor.v xr8, xr8, xr8
320+ xvst xr8, a1, 0
321+ xvst xr8, a1, 32
322+ xvst xr8, a1, 64
323+ xvst xr8, a1, 96
324+
325+ vadd.h vr18, vr0, vr4
326+ vsub.h vr19, vr0, vr4
327+ vsrai.h vr20, vr2, 1
328+ vsrai.h vr21, vr6, 1
329+ vsub.h vr20, vr20, vr6
330+ vadd.h vr21, vr21, vr2
331+ LSX_BUTTERFLY_4_H vr18, vr19, vr20, vr21, vr10, vr12, vr14, vr16
332+ vsrai.h vr11, vr7, 1
333+ vsrai.h vr13, vr3, 1
334+ vsrai.h vr15, vr5, 1
335+ vsrai.h vr17, vr1, 1
336+ vsub.h vr11, vr5, vr11
337+ vsub.h vr13, vr7, vr13
338+ vadd.h vr15, vr7, vr15
339+ vadd.h vr17, vr5, vr17
340+ vsub.h vr11, vr11, vr7
341+ vsub.h vr13, vr13, vr3
342+ vadd.h vr15, vr15, vr5
343+ vadd.h vr17, vr17, vr1
344+ vsub.h vr11, vr11, vr3
345+ vadd.h vr13, vr13, vr1
346+ vsub.h vr15, vr15, vr1
347+ vadd.h vr17, vr17, vr3
348+ vsrai.h vr18, vr11, 2
349+ vsrai.h vr19, vr13, 2
350+ vsrai.h vr20, vr15, 2
351+ vsrai.h vr21, vr17, 2
352+ vadd.h vr11, vr11, vr21
353+ vadd.h vr13, vr13, vr20
354+ vsub.h vr15, vr19, vr15
355+ vsub.h vr17, vr17, vr18
356+ LSX_BUTTERFLY_8_H vr10, vr16, vr12, vr14, vr13, vr15, vr11, vr17, \
357+ vr0, vr3, vr1, vr2, vr5, vr6, vr4, vr7
358+
359+ LSX_TRANSPOSE8x8_H vr0, vr1, vr2, vr3, vr4, vr5, vr6, vr7, \
360+ vr0, vr1, vr2, vr3, vr4, vr5, vr6, vr7, \
361+ vr10, vr11, vr12, vr13, vr14, vr15, vr16, vr17
362+ vext2xv.w.h xr0, xr0
363+ vext2xv.w.h xr1, xr1
364+ vext2xv.w.h xr2, xr2
365+ vext2xv.w.h xr3, xr3
366+ vext2xv.w.h xr4, xr4
367+ vext2xv.w.h xr5, xr5
368+ vext2xv.w.h xr6, xr6
369+ vext2xv.w.h xr7, xr7
370+
371+ xvadd.w xr11, xr0, xr4
372+ xvsub.w xr13, xr0, xr4
373+ xvsrai.w xr15, xr2, 1
374+ xvsrai.w xr17, xr6, 1
375+ xvsub.w xr15, xr15, xr6
376+ xvadd.w xr17, xr17, xr2
377+ LASX_BUTTERFLY_4_W xr11, xr13, xr15, xr17, xr10, xr12, xr14, xr16
378+ xvsrai.w xr11, xr7, 1
379+ xvsrai.w xr13, xr3, 1
380+ xvsrai.w xr15, xr5, 1
381+ xvsrai.w xr17, xr1, 1
382+ xvsub.w xr11, xr5, xr11
383+ xvsub.w xr13, xr7, xr13
384+ xvadd.w xr15, xr7, xr15
385+ xvadd.w xr17, xr5, xr17
386+ xvsub.w xr11, xr11, xr7
387+ xvsub.w xr13, xr13, xr3
388+ xvadd.w xr15, xr15, xr5
389+ xvadd.w xr17, xr17, xr1
390+ xvsub.w xr11, xr11, xr3
391+ xvadd.w xr13, xr13, xr1
392+ xvsub.w xr15, xr15, xr1
393+ xvadd.w xr17, xr17, xr3
394+ xvsrai.w xr0, xr11, 2
395+ xvsrai.w xr1, xr13, 2
396+ xvsrai.w xr2, xr15, 2
397+ xvsrai.w xr3, xr17, 2
398+ xvadd.w xr11, xr11, xr3
399+ xvadd.w xr13, xr13, xr2
400+ xvsub.w xr15, xr1, xr15
401+ xvsub.w xr17, xr17, xr0
402+ LASX_BUTTERFLY_8_W xr10, xr12, xr14, xr16, xr11, xr13, xr15, xr17, \
403+ xr0, xr1, xr2, xr3, xr4, xr5, xr6, xr7
404+
405+ vld vr10, a0, 0
406+ vldx vr11, a0, a2
407+ vldx vr12, a0, t2
408+ vldx vr13, a0, t3
409+ vldx vr14, a0, t4
410+ vldx vr15, a0, t5
411+ vldx vr16, a0, t6
412+ vldx vr17, a0, t7
413+ xvldi xr8, 0x806 //"xvldi.w xr8 6"
414+ xvsran.h.w xr0, xr0, xr8
415+ xvsran.h.w xr1, xr1, xr8
416+ xvsran.h.w xr2, xr2, xr8
417+ xvsran.h.w xr3, xr3, xr8
418+ xvsran.h.w xr4, xr4, xr8
419+ xvsran.h.w xr5, xr5, xr8
420+ xvsran.h.w xr6, xr6, xr8
421+ xvsran.h.w xr7, xr7, xr8
422+ xvpermi.d xr0, xr0, 0x08
423+ xvpermi.d xr1, xr1, 0x08
424+ xvpermi.d xr2, xr2, 0x08
425+ xvpermi.d xr3, xr3, 0x08
426+ xvpermi.d xr4, xr4, 0x08
427+ xvpermi.d xr5, xr5, 0x08
428+ xvpermi.d xr6, xr6, 0x08
429+ xvpermi.d xr7, xr7, 0x08
430+
431+ vsllwil.hu.bu vr10, vr10, 0
432+ vsllwil.hu.bu vr11, vr11, 0
433+ vsllwil.hu.bu vr12, vr12, 0
434+ vsllwil.hu.bu vr13, vr13, 0
435+ vsllwil.hu.bu vr14, vr14, 0
436+ vsllwil.hu.bu vr15, vr15, 0
437+ vsllwil.hu.bu vr16, vr16, 0
438+ vsllwil.hu.bu vr17, vr17, 0
439+
440+ vadd.h vr0, vr0, vr10
441+ vadd.h vr1, vr1, vr11
442+ vadd.h vr2, vr2, vr12
443+ vadd.h vr3, vr3, vr13
444+ vadd.h vr4, vr4, vr14
445+ vadd.h vr5, vr5, vr15
446+ vadd.h vr6, vr6, vr16
447+ vadd.h vr7, vr7, vr17
448+ vssrarni.bu.h vr1, vr0, 0
449+ vssrarni.bu.h vr3, vr2, 0
450+ vssrarni.bu.h vr5, vr4, 0
451+ vssrarni.bu.h vr7, vr6, 0
452+ vbsrl.v vr0, vr1, 8
453+ vbsrl.v vr2, vr3, 8
454+ vbsrl.v vr4, vr5, 8
455+ vbsrl.v vr6, vr7, 8
456+ fst.d f1, a0, 0
457+ fstx.d f0, a0, a2
458+ fstx.d f3, a0, t2
459+ fstx.d f2, a0, t3
460+ fstx.d f5, a0, t4
461+ fstx.d f4, a0, t5
462+ fstx.d f7, a0, t6
463+ fstx.d f6, a0, t7
464+endfunc
465+
466+/*
467+ * #define FUNC2(a, b, c) FUNC3(a, b, c)
468+ * #define FUNCC(a) FUNC2(a, BIT_DEPTH, _c)
469+ * void FUNCC(ff_h264_idct_dc_add)(uint8_t *_dst, int16_t *_block, int stride)
470+ * LSX optimization is enough for this function.
471+ */
472+function ff_h264_idct_dc_add_8_lsx
473+ vldrepl.h vr4, a1, 0
474+ add.d t2, a2, a2
475+ add.d t3, t2, a2
476+ fld.s f0, a0, 0
477+ fldx.s f1, a0, a2
478+ fldx.s f2, a0, t2
479+ fldx.s f3, a0, t3
480+ st.h zero, a1, 0
481+
482+ vsrari.h vr4, vr4, 6
483+ vilvl.w vr0, vr1, vr0
484+ vilvl.w vr1, vr3, vr2
485+ vsllwil.hu.bu vr0, vr0, 0
486+ vsllwil.hu.bu vr1, vr1, 0
487+ vadd.h vr0, vr0, vr4
488+ vadd.h vr1, vr1, vr4
489+ vssrarni.bu.h vr1, vr0, 0
490+
491+ vbsrl.v vr2, vr1, 4
492+ vbsrl.v vr3, vr1, 8
493+ vbsrl.v vr4, vr1, 12
494+ fst.s f1, a0, 0
495+ fstx.s f2, a0, a2
496+ fstx.s f3, a0, t2
497+ fstx.s f4, a0, t3
498+endfunc
499+
500+/*
501+ * #define FUNC2(a, b, c) FUNC3(a, b, c)
502+ * #define FUNCC(a) FUNC2(a, BIT_DEPTH, _c)
503+ * void FUNCC(ff_h264_idct8_dc_add)(uint8_t *_dst, int16_t *_block, int stride)
504+ */
505+function ff_h264_idct8_dc_add_8_lsx
506+ vldrepl.h vr8, a1, 0
507+ add.d t2, a2, a2
508+ add.d t3, t2, a2
509+ add.d t4, t3, a2
510+ add.d t5, t4, a2
511+ add.d t6, t5, a2
512+ add.d t7, t6, a2
513+
514+ fld.d f0, a0, 0
515+ fldx.d f1, a0, a2
516+ fldx.d f2, a0, t2
517+ fldx.d f3, a0, t3
518+ fldx.d f4, a0, t4
519+ fldx.d f5, a0, t5
520+ fldx.d f6, a0, t6
521+ fldx.d f7, a0, t7
522+ st.h zero, a1, 0
523+
524+ vsrari.h vr8, vr8, 6
525+ vsllwil.hu.bu vr0, vr0, 0
526+ vsllwil.hu.bu vr1, vr1, 0
527+ vsllwil.hu.bu vr2, vr2, 0
528+ vsllwil.hu.bu vr3, vr3, 0
529+ vsllwil.hu.bu vr4, vr4, 0
530+ vsllwil.hu.bu vr5, vr5, 0
531+ vsllwil.hu.bu vr6, vr6, 0
532+ vsllwil.hu.bu vr7, vr7, 0
533+ vadd.h vr0, vr0, vr8
534+ vadd.h vr1, vr1, vr8
535+ vadd.h vr2, vr2, vr8
536+ vadd.h vr3, vr3, vr8
537+ vadd.h vr4, vr4, vr8
538+ vadd.h vr5, vr5, vr8
539+ vadd.h vr6, vr6, vr8
540+ vadd.h vr7, vr7, vr8
541+ vssrarni.bu.h vr1, vr0, 0
542+ vssrarni.bu.h vr3, vr2, 0
543+ vssrarni.bu.h vr5, vr4, 0
544+ vssrarni.bu.h vr7, vr6, 0
545+
546+ vbsrl.v vr0, vr1, 8
547+ vbsrl.v vr2, vr3, 8
548+ vbsrl.v vr4, vr5, 8
549+ vbsrl.v vr6, vr7, 8
550+ fst.d f1, a0, 0
551+ fstx.d f0, a0, a2
552+ fstx.d f3, a0, t2
553+ fstx.d f2, a0, t3
554+ fstx.d f5, a0, t4
555+ fstx.d f4, a0, t5
556+ fstx.d f7, a0, t6
557+ fstx.d f6, a0, t7
558+endfunc
559+function ff_h264_idct8_dc_add_8_lasx
560+ xvldrepl.h xr8, a1, 0
561+ add.d t2, a2, a2
562+ add.d t3, t2, a2
563+ add.d t4, t3, a2
564+ add.d t5, t4, a2
565+ add.d t6, t5, a2
566+ add.d t7, t6, a2
567+
568+ fld.d f0, a0, 0
569+ fldx.d f1, a0, a2
570+ fldx.d f2, a0, t2
571+ fldx.d f3, a0, t3
572+ fldx.d f4, a0, t4
573+ fldx.d f5, a0, t5
574+ fldx.d f6, a0, t6
575+ fldx.d f7, a0, t7
576+ st.h zero, a1, 0
577+
578+ xvsrari.h xr8, xr8, 6
579+ xvpermi.q xr1, xr0, 0x20
580+ xvpermi.q xr3, xr2, 0x20
581+ xvpermi.q xr5, xr4, 0x20
582+ xvpermi.q xr7, xr6, 0x20
583+ xvsllwil.hu.bu xr1, xr1, 0
584+ xvsllwil.hu.bu xr3, xr3, 0
585+ xvsllwil.hu.bu xr5, xr5, 0
586+ xvsllwil.hu.bu xr7, xr7, 0
587+ xvadd.h xr1, xr1, xr8
588+ xvadd.h xr3, xr3, xr8
589+ xvadd.h xr5, xr5, xr8
590+ xvadd.h xr7, xr7, xr8
591+
592+ xvssrarni.bu.h xr3, xr1, 0
593+ xvssrarni.bu.h xr7, xr5, 0
594+
595+ xvpermi.q xr1, xr3, 0x11
596+ xvpermi.q xr5, xr7, 0x11
597+ xvbsrl.v xr0, xr1, 8
598+ xvbsrl.v xr2, xr3, 8
599+ xvbsrl.v xr4, xr5, 8
600+ xvbsrl.v xr6, xr7, 8
601+
602+ fst.d f3, a0, 0
603+ fstx.d f1, a0, a2
604+ fstx.d f2, a0, t2
605+ fstx.d f0, a0, t3
606+ fstx.d f7, a0, t4
607+ fstx.d f5, a0, t5
608+ fstx.d f6, a0, t6
609+ fstx.d f4, a0, t7
610+endfunc
611+
612+/**
613+ * IDCT transforms the 16 dc values and dequantizes them.
614+ * @param qmul quantization parameter
615+ * void FUNCC(ff_h264_luma_dc_dequant_idct)(int16_t *_output, int16_t *_input, int qmul){
616+ * LSX optimization is enough for this function.
617+ */
618+function ff_h264_luma_dc_dequant_idct_8_lsx
619+ vld vr0, a1, 0
620+ vld vr1, a1, 8
621+ vld vr2, a1, 16
622+ vld vr3, a1, 24
623+ vreplgr2vr.w vr8, a2
624+ LSX_TRANSPOSE4x4_H vr0, vr1, vr2, vr3, vr4, vr5, vr6, vr7, vr9, vr10
625+ LSX_BUTTERFLY_4_H vr4, vr6, vr7, vr5, vr0, vr3, vr2, vr1
626+ LSX_BUTTERFLY_4_H vr0, vr1, vr2, vr3, vr4, vr7, vr6, vr5
627+ LSX_TRANSPOSE4x4_H vr4, vr5, vr6, vr7, vr0, vr1, vr2, vr3, vr9, vr10
628+ LSX_BUTTERFLY_4_H vr0, vr1, vr3, vr2, vr4, vr7, vr6, vr5
629+ LSX_BUTTERFLY_4_H vr4, vr5, vr6, vr7, vr0, vr1, vr2, vr3
630+ vsllwil.w.h vr0, vr0, 0
631+ vsllwil.w.h vr1, vr1, 0
632+ vsllwil.w.h vr2, vr2, 0
633+ vsllwil.w.h vr3, vr3, 0
634+ vmul.w vr0, vr0, vr8
635+ vmul.w vr1, vr1, vr8
636+ vmul.w vr2, vr2, vr8
637+ vmul.w vr3, vr3, vr8
638+ vsrarni.h.w vr1, vr0, 8
639+ vsrarni.h.w vr3, vr2, 8
640+
641+ vstelm.h vr1, a0, 0, 0
642+ vstelm.h vr1, a0, 32, 4
643+ vstelm.h vr1, a0, 64, 1
644+ vstelm.h vr1, a0, 96, 5
645+ vstelm.h vr3, a0, 128, 0
646+ vstelm.h vr3, a0, 160, 4
647+ vstelm.h vr3, a0, 192, 1
648+ vstelm.h vr3, a0, 224, 5
649+ addi.d a0, a0, 256
650+ vstelm.h vr1, a0, 0, 2
651+ vstelm.h vr1, a0, 32, 6
652+ vstelm.h vr1, a0, 64, 3
653+ vstelm.h vr1, a0, 96, 7
654+ vstelm.h vr3, a0, 128, 2
655+ vstelm.h vr3, a0, 160, 6
656+ vstelm.h vr3, a0, 192, 3
657+ vstelm.h vr3, a0, 224, 7
658+endfunc
libavcodec/loongarch/h264idct_lasx.c+0−498
@@ -1,498 +0,0 @@
1−/*
2− * Loongson LASX optimized h264dsp
3− *
4− * Copyright (c) 2021 Loongson Technology Corporation Limited
5− * Contributed by Shiyou Yin <yinshiyou-hf@loongson.cn>
6− * Xiwei Gu <guxiwei-hf@loongson.cn>
7− *
8− * This file is part of FFmpeg.
9− *
10− * FFmpeg is free software; you can redistribute it and/or
11− * modify it under the terms of the GNU Lesser General Public
12− * License as published by the Free Software Foundation; either
13− * version 2.1 of the License, or (at your option) any later version.
14− *
15− * FFmpeg is distributed in the hope that it will be useful,
16− * but WITHOUT ANY WARRANTY; without even the implied warranty of
17− * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18− * Lesser General Public License for more details.
19− *
20− * You should have received a copy of the GNU Lesser General Public
21− * License along with FFmpeg; if not, write to the Free Software
22− * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23− */
24−
25−#include "libavutil/loongarch/loongson_intrinsics.h"
26−#include "h264dsp_lasx.h"
27−#include "libavcodec/bit_depth_template.c"
28−
29−#define AVC_ITRANS_H(in0, in1, in2, in3, out0, out1, out2, out3) \
30−{ \
31− __m256i tmp0_m, tmp1_m, tmp2_m, tmp3_m; \
32− \
33− tmp0_m = __lasx_xvadd_h(in0, in2); \
34− tmp1_m = __lasx_xvsub_h(in0, in2); \
35− tmp2_m = __lasx_xvsrai_h(in1, 1); \
36− tmp2_m = __lasx_xvsub_h(tmp2_m, in3); \
37− tmp3_m = __lasx_xvsrai_h(in3, 1); \
38− tmp3_m = __lasx_xvadd_h(in1, tmp3_m); \
39− \
40− LASX_BUTTERFLY_4_H(tmp0_m, tmp1_m, tmp2_m, tmp3_m, \
41− out0, out1, out2, out3); \
42−}
43−
44−void ff_h264_idct_add_lasx(uint8_t *dst, int16_t *src, int32_t dst_stride)
45−{
46− __m256i src0_m, src1_m, src2_m, src3_m;
47− __m256i dst0_m, dst1_m;
48− __m256i hres0, hres1, hres2, hres3, vres0, vres1, vres2, vres3;
49− __m256i inp0_m, inp1_m, res0_m, src1, src3;
50− __m256i src0 = __lasx_xvld(src, 0);
51− __m256i src2 = __lasx_xvld(src, 16);
52− __m256i zero = __lasx_xvldi(0);
53− int32_t dst_stride_2x = dst_stride << 1;
54− int32_t dst_stride_3x = dst_stride_2x + dst_stride;
55−
56− __lasx_xvst(zero, src, 0);
57− DUP2_ARG2(__lasx_xvilvh_d, src0, src0, src2, src2, src1, src3);
58− AVC_ITRANS_H(src0, src1, src2, src3, hres0, hres1, hres2, hres3);
59− LASX_TRANSPOSE4x4_H(hres0, hres1, hres2, hres3, hres0, hres1, hres2, hres3);
60− AVC_ITRANS_H(hres0, hres1, hres2, hres3, vres0, vres1, vres2, vres3);
61− DUP4_ARG2(__lasx_xvldx, dst, 0, dst, dst_stride, dst, dst_stride_2x,
62− dst, dst_stride_3x, src0_m, src1_m, src2_m, src3_m);
63− DUP4_ARG2(__lasx_xvld, dst, 0, dst + dst_stride, 0, dst + dst_stride_2x,
64− 0, dst + dst_stride_3x, 0, src0_m, src1_m, src2_m, src3_m);
65− DUP2_ARG2(__lasx_xvilvl_d, vres1, vres0, vres3, vres2, inp0_m, inp1_m);
66− inp0_m = __lasx_xvpermi_q(inp1_m, inp0_m, 0x20);
67− inp0_m = __lasx_xvsrari_h(inp0_m, 6);
68− DUP2_ARG2(__lasx_xvilvl_w, src1_m, src0_m, src3_m, src2_m, dst0_m, dst1_m);
69− dst0_m = __lasx_xvilvl_d(dst1_m, dst0_m);
70− res0_m = __lasx_vext2xv_hu_bu(dst0_m);
71− res0_m = __lasx_xvadd_h(res0_m, inp0_m);
72− res0_m = __lasx_xvclip255_h(res0_m);
73− dst0_m = __lasx_xvpickev_b(res0_m, res0_m);
74− __lasx_xvstelm_w(dst0_m, dst, 0, 0);
75− __lasx_xvstelm_w(dst0_m, dst + dst_stride, 0, 1);
76− __lasx_xvstelm_w(dst0_m, dst + dst_stride_2x, 0, 4);
77− __lasx_xvstelm_w(dst0_m, dst + dst_stride_3x, 0, 5);
78−}
79−
80−void ff_h264_idct8_addblk_lasx(uint8_t *dst, int16_t *src,
81− int32_t dst_stride)
82−{
83− __m256i src0, src1, src2, src3, src4, src5, src6, src7;
84− __m256i vec0, vec1, vec2, vec3;
85− __m256i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
86− __m256i res0, res1, res2, res3, res4, res5, res6, res7;
87− __m256i dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
88− __m256i zero = __lasx_xvldi(0);
89− int32_t dst_stride_2x = dst_stride << 1;
90− int32_t dst_stride_4x = dst_stride << 2;
91− int32_t dst_stride_3x = dst_stride_2x + dst_stride;
92−
93− src[0] += 32;
94− DUP4_ARG2(__lasx_xvld, src, 0, src, 16, src, 32, src, 48,
95− src0, src1, src2, src3);
96− DUP4_ARG2(__lasx_xvld, src, 64, src, 80, src, 96, src, 112,
97− src4, src5, src6, src7);
98− __lasx_xvst(zero, src, 0);
99− __lasx_xvst(zero, src, 32);
100− __lasx_xvst(zero, src, 64);
101− __lasx_xvst(zero, src, 96);
102−
103− vec0 = __lasx_xvadd_h(src0, src4);
104− vec1 = __lasx_xvsub_h(src0, src4);
105− vec2 = __lasx_xvsrai_h(src2, 1);
106− vec2 = __lasx_xvsub_h(vec2, src6);
107− vec3 = __lasx_xvsrai_h(src6, 1);
108− vec3 = __lasx_xvadd_h(src2, vec3);
109−
110− LASX_BUTTERFLY_4_H(vec0, vec1, vec2, vec3, tmp0, tmp1, tmp2, tmp3);
111−
112− vec0 = __lasx_xvsrai_h(src7, 1);
113− vec0 = __lasx_xvsub_h(src5, vec0);
114− vec0 = __lasx_xvsub_h(vec0, src3);
115− vec0 = __lasx_xvsub_h(vec0, src7);
116−
117− vec1 = __lasx_xvsrai_h(src3, 1);
118− vec1 = __lasx_xvsub_h(src1, vec1);
119− vec1 = __lasx_xvadd_h(vec1, src7);
120− vec1 = __lasx_xvsub_h(vec1, src3);
121−
122− vec2 = __lasx_xvsrai_h(src5, 1);
123− vec2 = __lasx_xvsub_h(vec2, src1);
124− vec2 = __lasx_xvadd_h(vec2, src7);
125− vec2 = __lasx_xvadd_h(vec2, src5);
126−
127− vec3 = __lasx_xvsrai_h(src1, 1);
128− vec3 = __lasx_xvadd_h(src3, vec3);
129− vec3 = __lasx_xvadd_h(vec3, src5);
130− vec3 = __lasx_xvadd_h(vec3, src1);
131−
132− tmp4 = __lasx_xvsrai_h(vec3, 2);
133− tmp4 = __lasx_xvadd_h(tmp4, vec0);
134− tmp5 = __lasx_xvsrai_h(vec2, 2);
135− tmp5 = __lasx_xvadd_h(tmp5, vec1);
136− tmp6 = __lasx_xvsrai_h(vec1, 2);
137− tmp6 = __lasx_xvsub_h(tmp6, vec2);
138− tmp7 = __lasx_xvsrai_h(vec0, 2);
139− tmp7 = __lasx_xvsub_h(vec3, tmp7);
140−
141− LASX_BUTTERFLY_8_H(tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7,
142− res0, res1, res2, res3, res4, res5, res6, res7);
143− LASX_TRANSPOSE8x8_H(res0, res1, res2, res3, res4, res5, res6, res7,
144− res0, res1, res2, res3, res4, res5, res6, res7);
145−
146− DUP4_ARG1(__lasx_vext2xv_w_h, res0, res1, res2, res3,
147− tmp0, tmp1, tmp2, tmp3);
148− DUP4_ARG1(__lasx_vext2xv_w_h, res4, res5, res6, res7,
149− tmp4, tmp5, tmp6, tmp7);
150− vec0 = __lasx_xvadd_w(tmp0, tmp4);
151− vec1 = __lasx_xvsub_w(tmp0, tmp4);
152−
153− vec2 = __lasx_xvsrai_w(tmp2, 1);
154− vec2 = __lasx_xvsub_w(vec2, tmp6);
155− vec3 = __lasx_xvsrai_w(tmp6, 1);
156− vec3 = __lasx_xvadd_w(vec3, tmp2);
157−
158− tmp0 = __lasx_xvadd_w(vec0, vec3);
159− tmp2 = __lasx_xvadd_w(vec1, vec2);
160− tmp4 = __lasx_xvsub_w(vec1, vec2);
161− tmp6 = __lasx_xvsub_w(vec0, vec3);
162−
163− vec0 = __lasx_xvsrai_w(tmp7, 1);
164− vec0 = __lasx_xvsub_w(tmp5, vec0);
165− vec0 = __lasx_xvsub_w(vec0, tmp3);
166− vec0 = __lasx_xvsub_w(vec0, tmp7);
167−
168− vec1 = __lasx_xvsrai_w(tmp3, 1);
169− vec1 = __lasx_xvsub_w(tmp1, vec1);
170− vec1 = __lasx_xvadd_w(vec1, tmp7);
171− vec1 = __lasx_xvsub_w(vec1, tmp3);
172−
173− vec2 = __lasx_xvsrai_w(tmp5, 1);
174− vec2 = __lasx_xvsub_w(vec2, tmp1);
175− vec2 = __lasx_xvadd_w(vec2, tmp7);
176− vec2 = __lasx_xvadd_w(vec2, tmp5);
177−
178− vec3 = __lasx_xvsrai_w(tmp1, 1);
179− vec3 = __lasx_xvadd_w(tmp3, vec3);
180− vec3 = __lasx_xvadd_w(vec3, tmp5);
181− vec3 = __lasx_xvadd_w(vec3, tmp1);
182−
183− tmp1 = __lasx_xvsrai_w(vec3, 2);
184− tmp1 = __lasx_xvadd_w(tmp1, vec0);
185− tmp3 = __lasx_xvsrai_w(vec2, 2);
186− tmp3 = __lasx_xvadd_w(tmp3, vec1);
187− tmp5 = __lasx_xvsrai_w(vec1, 2);
188− tmp5 = __lasx_xvsub_w(tmp5, vec2);
189− tmp7 = __lasx_xvsrai_w(vec0, 2);
190− tmp7 = __lasx_xvsub_w(vec3, tmp7);
191−
192− LASX_BUTTERFLY_4_W(tmp0, tmp2, tmp5, tmp7, res0, res1, res6, res7);
193− LASX_BUTTERFLY_4_W(tmp4, tmp6, tmp1, tmp3, res2, res3, res4, res5);
194−
195− DUP4_ARG2(__lasx_xvsrai_w, res0, 6, res1, 6, res2, 6, res3, 6,
196− res0, res1, res2, res3);
197− DUP4_ARG2(__lasx_xvsrai_w, res4, 6, res5, 6, res6, 6, res7, 6,
198− res4, res5, res6, res7);
199− DUP4_ARG2(__lasx_xvpickev_h, res1, res0, res3, res2, res5, res4, res7,
200− res6, res0, res1, res2, res3);
201− DUP4_ARG2(__lasx_xvpermi_d, res0, 0xd8, res1, 0xd8, res2, 0xd8, res3, 0xd8,
202− res0, res1, res2, res3);
203−
204− DUP4_ARG2(__lasx_xvldx, dst, 0, dst, dst_stride, dst, dst_stride_2x,
205− dst, dst_stride_3x, dst0, dst1, dst2, dst3);
206− dst += dst_stride_4x;
207− DUP4_ARG2(__lasx_xvldx, dst, 0, dst, dst_stride, dst, dst_stride_2x,
208− dst, dst_stride_3x, dst4, dst5, dst6, dst7);
209− dst -= dst_stride_4x;
210− DUP4_ARG2(__lasx_xvilvl_b, zero, dst0, zero, dst1, zero, dst2, zero, dst3,
211− dst0, dst1, dst2, dst3);
212− DUP4_ARG2(__lasx_xvilvl_b, zero, dst4, zero, dst5, zero, dst6, zero, dst7,
213− dst4, dst5, dst6, dst7);
214− DUP4_ARG3(__lasx_xvpermi_q, dst1, dst0, 0x20, dst3, dst2, 0x20, dst5,
215− dst4, 0x20, dst7, dst6, 0x20, dst0, dst1, dst2, dst3);
216− res0 = __lasx_xvadd_h(res0, dst0);
217− res1 = __lasx_xvadd_h(res1, dst1);
218− res2 = __lasx_xvadd_h(res2, dst2);
219− res3 = __lasx_xvadd_h(res3, dst3);
220− DUP4_ARG1(__lasx_xvclip255_h, res0, res1, res2, res3, res0, res1,
221− res2, res3);
222− DUP2_ARG2(__lasx_xvpickev_b, res1, res0, res3, res2, res0, res1);
223− __lasx_xvstelm_d(res0, dst, 0, 0);
224− __lasx_xvstelm_d(res0, dst + dst_stride, 0, 2);
225− __lasx_xvstelm_d(res0, dst + dst_stride_2x, 0, 1);
226− __lasx_xvstelm_d(res0, dst + dst_stride_3x, 0, 3);
227− dst += dst_stride_4x;
228− __lasx_xvstelm_d(res1, dst, 0, 0);
229− __lasx_xvstelm_d(res1, dst + dst_stride, 0, 2);
230− __lasx_xvstelm_d(res1, dst + dst_stride_2x, 0, 1);
231− __lasx_xvstelm_d(res1, dst + dst_stride_3x, 0, 3);
232−}
233−
234−void ff_h264_idct4x4_addblk_dc_lasx(uint8_t *dst, int16_t *src,
235− int32_t dst_stride)
236−{
237− const int16_t dc = (src[0] + 32) >> 6;
238− int32_t dst_stride_2x = dst_stride << 1;
239− int32_t dst_stride_3x = dst_stride_2x + dst_stride;
240− __m256i pred, out;
241− __m256i src0, src1, src2, src3;
242− __m256i input_dc = __lasx_xvreplgr2vr_h(dc);
243−
244− src[0] = 0;
245− DUP4_ARG2(__lasx_xvldx, dst, 0, dst, dst_stride, dst, dst_stride_2x,
246− dst, dst_stride_3x, src0, src1, src2, src3);
247− DUP2_ARG2(__lasx_xvilvl_w, src1, src0, src3, src2, src0, src1);
248−
249− pred = __lasx_xvpermi_q(src0, src1, 0x02);
250− pred = __lasx_xvaddw_h_h_bu(input_dc, pred);
251− pred = __lasx_xvclip255_h(pred);
252− out = __lasx_xvpickev_b(pred, pred);
253− __lasx_xvstelm_w(out, dst, 0, 0);
254− __lasx_xvstelm_w(out, dst + dst_stride, 0, 1);
255− __lasx_xvstelm_w(out, dst + dst_stride_2x, 0, 4);
256− __lasx_xvstelm_w(out, dst + dst_stride_3x, 0, 5);
257−}
258−
259−void ff_h264_idct8_dc_addblk_lasx(uint8_t *dst, int16_t *src,
260− int32_t dst_stride)
261−{
262− int32_t dc_val;
263− int32_t dst_stride_2x = dst_stride << 1;
264− int32_t dst_stride_4x = dst_stride << 2;
265− int32_t dst_stride_3x = dst_stride_2x + dst_stride;
266− __m256i dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
267− __m256i dc;
268−
269− dc_val = (src[0] + 32) >> 6;
270− dc = __lasx_xvreplgr2vr_h(dc_val);
271−
272− src[0] = 0;
273−
274− DUP4_ARG2(__lasx_xvldx, dst, 0, dst, dst_stride, dst, dst_stride_2x,
275− dst, dst_stride_3x, dst0, dst1, dst2, dst3);
276− dst += dst_stride_4x;
277− DUP4_ARG2(__lasx_xvldx, dst, 0, dst, dst_stride, dst, dst_stride_2x,
278− dst, dst_stride_3x, dst4, dst5, dst6, dst7);
279− dst -= dst_stride_4x;
280− DUP4_ARG1(__lasx_vext2xv_hu_bu, dst0, dst1, dst2, dst3,
281− dst0, dst1, dst2, dst3);
282− DUP4_ARG1(__lasx_vext2xv_hu_bu, dst4, dst5, dst6, dst7,
283− dst4, dst5, dst6, dst7);
284− DUP4_ARG3(__lasx_xvpermi_q, dst1, dst0, 0x20, dst3, dst2, 0x20, dst5,
285− dst4, 0x20, dst7, dst6, 0x20, dst0, dst1, dst2, dst3);
286− dst0 = __lasx_xvadd_h(dst0, dc);
287− dst1 = __lasx_xvadd_h(dst1, dc);
288− dst2 = __lasx_xvadd_h(dst2, dc);
289− dst3 = __lasx_xvadd_h(dst3, dc);
290− DUP4_ARG1(__lasx_xvclip255_h, dst0, dst1, dst2, dst3,
291− dst0, dst1, dst2, dst3);
292− DUP2_ARG2(__lasx_xvpickev_b, dst1, dst0, dst3, dst2, dst0, dst1);
293− __lasx_xvstelm_d(dst0, dst, 0, 0);
294− __lasx_xvstelm_d(dst0, dst + dst_stride, 0, 2);
295− __lasx_xvstelm_d(dst0, dst + dst_stride_2x, 0, 1);
296− __lasx_xvstelm_d(dst0, dst + dst_stride_3x, 0, 3);
297− dst += dst_stride_4x;
298− __lasx_xvstelm_d(dst1, dst, 0, 0);
299− __lasx_xvstelm_d(dst1, dst + dst_stride, 0, 2);
300− __lasx_xvstelm_d(dst1, dst + dst_stride_2x, 0, 1);
301− __lasx_xvstelm_d(dst1, dst + dst_stride_3x, 0, 3);
302−}
303−
304−void ff_h264_idct_add16_lasx(uint8_t *dst,
305− const int32_t *blk_offset,
306− int16_t *block, int32_t dst_stride,
307− const uint8_t nzc[15 * 8])
308−{
309− int32_t i;
310−
311− for (i = 0; i < 16; i++) {
312− int32_t nnz = nzc[scan8[i]];
313−
314− if (nnz) {
315− if (nnz == 1 && ((dctcoef *) block)[i * 16])
316− ff_h264_idct4x4_addblk_dc_lasx(dst + blk_offset[i],
317− block + i * 16 * sizeof(pixel),
318− dst_stride);
319− else
320− ff_h264_idct_add_lasx(dst + blk_offset[i],
321− block + i * 16 * sizeof(pixel),
322− dst_stride);
323− }
324− }
325−}
326−
327−void ff_h264_idct8_add4_lasx(uint8_t *dst, const int32_t *blk_offset,
328− int16_t *block, int32_t dst_stride,
329− const uint8_t nzc[15 * 8])
330−{
331− int32_t cnt;
332−
333− for (cnt = 0; cnt < 16; cnt += 4) {
334− int32_t nnz = nzc[scan8[cnt]];
335−
336− if (nnz) {
337− if (nnz == 1 && ((dctcoef *) block)[cnt * 16])
338− ff_h264_idct8_dc_addblk_lasx(dst + blk_offset[cnt],
339− block + cnt * 16 * sizeof(pixel),
340− dst_stride);
341− else
342− ff_h264_idct8_addblk_lasx(dst + blk_offset[cnt],
343− block + cnt * 16 * sizeof(pixel),
344− dst_stride);
345− }
346− }
347−}
348−
349−
350−void ff_h264_idct_add8_lasx(uint8_t **dst,
351− const int32_t *blk_offset,
352− int16_t *block, int32_t dst_stride,
353− const uint8_t nzc[15 * 8])
354−{
355− int32_t i;
356−
357− for (i = 16; i < 20; i++) {
358− if (nzc[scan8[i]])
359− ff_h264_idct_add_lasx(dst[0] + blk_offset[i],
360− block + i * 16 * sizeof(pixel),
361− dst_stride);
362− else if (((dctcoef *) block)[i * 16])
363− ff_h264_idct4x4_addblk_dc_lasx(dst[0] + blk_offset[i],
364− block + i * 16 * sizeof(pixel),
365− dst_stride);
366− }
367− for (i = 32; i < 36; i++) {
368− if (nzc[scan8[i]])
369− ff_h264_idct_add_lasx(dst[1] + blk_offset[i],
370− block + i * 16 * sizeof(pixel),
371− dst_stride);
372− else if (((dctcoef *) block)[i * 16])
373− ff_h264_idct4x4_addblk_dc_lasx(dst[1] + blk_offset[i],
374− block + i * 16 * sizeof(pixel),
375− dst_stride);
376− }
377−}
378−
379−void ff_h264_idct_add8_422_lasx(uint8_t **dst,
380− const int32_t *blk_offset,
381− int16_t *block, int32_t dst_stride,
382− const uint8_t nzc[15 * 8])
383−{
384− int32_t i;
385−
386− for (i = 16; i < 20; i++) {
387− if (nzc[scan8[i]])
388− ff_h264_idct_add_lasx(dst[0] + blk_offset[i],
389− block + i * 16 * sizeof(pixel),
390− dst_stride);
391− else if (((dctcoef *) block)[i * 16])
392− ff_h264_idct4x4_addblk_dc_lasx(dst[0] + blk_offset[i],
393− block + i * 16 * sizeof(pixel),
394− dst_stride);
395− }
396− for (i = 32; i < 36; i++) {
397− if (nzc[scan8[i]])
398− ff_h264_idct_add_lasx(dst[1] + blk_offset[i],
399− block + i * 16 * sizeof(pixel),
400− dst_stride);
401− else if (((dctcoef *) block)[i * 16])
402− ff_h264_idct4x4_addblk_dc_lasx(dst[1] + blk_offset[i],
403− block + i * 16 * sizeof(pixel),
404− dst_stride);
405− }
406− for (i = 20; i < 24; i++) {
407− if (nzc[scan8[i + 4]])
408− ff_h264_idct_add_lasx(dst[0] + blk_offset[i + 4],
409− block + i * 16 * sizeof(pixel),
410− dst_stride);
411− else if (((dctcoef *) block)[i * 16])
412− ff_h264_idct4x4_addblk_dc_lasx(dst[0] + blk_offset[i + 4],
413− block + i * 16 * sizeof(pixel),
414− dst_stride);
415− }
416− for (i = 36; i < 40; i++) {
417− if (nzc[scan8[i + 4]])
418− ff_h264_idct_add_lasx(dst[1] + blk_offset[i + 4],
419− block + i * 16 * sizeof(pixel),
420− dst_stride);
421− else if (((dctcoef *) block)[i * 16])
422− ff_h264_idct4x4_addblk_dc_lasx(dst[1] + blk_offset[i + 4],
423− block + i * 16 * sizeof(pixel),
424− dst_stride);
425− }
426−}
427−
428−void ff_h264_idct_add16_intra_lasx(uint8_t *dst,
429− const int32_t *blk_offset,
430− int16_t *block,
431− int32_t dst_stride,
432− const uint8_t nzc[15 * 8])
433−{
434− int32_t i;
435−
436− for (i = 0; i < 16; i++) {
437− if (nzc[scan8[i]])
438− ff_h264_idct_add_lasx(dst + blk_offset[i],
439− block + i * 16 * sizeof(pixel), dst_stride);
440− else if (((dctcoef *) block)[i * 16])
441− ff_h264_idct4x4_addblk_dc_lasx(dst + blk_offset[i],
442− block + i * 16 * sizeof(pixel),
443− dst_stride);
444− }
445−}
446−
447−void ff_h264_deq_idct_luma_dc_lasx(int16_t *dst, int16_t *src,
448− int32_t de_qval)
449−{
450−#define DC_DEST_STRIDE 16
451−
452− __m256i src0, src1, src2, src3;
453− __m256i vec0, vec1, vec2, vec3;
454− __m256i tmp0, tmp1, tmp2, tmp3;
455− __m256i hres0, hres1, hres2, hres3;
456− __m256i vres0, vres1, vres2, vres3;
457− __m256i de_q_vec = __lasx_xvreplgr2vr_w(de_qval);
458−
459− DUP4_ARG2(__lasx_xvld, src, 0, src, 8, src, 16, src, 24,
460− src0, src1, src2, src3);
461− LASX_TRANSPOSE4x4_H(src0, src1, src2, src3, tmp0, tmp1, tmp2, tmp3);
462− LASX_BUTTERFLY_4_H(tmp0, tmp2, tmp3, tmp1, vec0, vec3, vec2, vec1);
463− LASX_BUTTERFLY_4_H(vec0, vec1, vec2, vec3, hres0, hres3, hres2, hres1);
464− LASX_TRANSPOSE4x4_H(hres0, hres1, hres2, hres3,
465− hres0, hres1, hres2, hres3);
466− LASX_BUTTERFLY_4_H(hres0, hres1, hres3, hres2, vec0, vec3, vec2, vec1);
467− LASX_BUTTERFLY_4_H(vec0, vec1, vec2, vec3, vres0, vres1, vres2, vres3);
468− DUP4_ARG1(__lasx_vext2xv_w_h, vres0, vres1, vres2, vres3,
469− vres0, vres1, vres2, vres3);
470− DUP2_ARG3(__lasx_xvpermi_q, vres1, vres0, 0x20, vres3, vres2, 0x20,
471− vres0, vres1);
472−
473− vres0 = __lasx_xvmul_w(vres0, de_q_vec);
474− vres1 = __lasx_xvmul_w(vres1, de_q_vec);
475−
476− vres0 = __lasx_xvsrari_w(vres0, 8);
477− vres1 = __lasx_xvsrari_w(vres1, 8);
478− vec0 = __lasx_xvpickev_h(vres1, vres0);
479− vec0 = __lasx_xvpermi_d(vec0, 0xd8);
480− __lasx_xvstelm_h(vec0, dst + 0 * DC_DEST_STRIDE, 0, 0);
481− __lasx_xvstelm_h(vec0, dst + 2 * DC_DEST_STRIDE, 0, 1);
482− __lasx_xvstelm_h(vec0, dst + 8 * DC_DEST_STRIDE, 0, 2);
483− __lasx_xvstelm_h(vec0, dst + 10 * DC_DEST_STRIDE, 0, 3);
484− __lasx_xvstelm_h(vec0, dst + 1 * DC_DEST_STRIDE, 0, 4);
485− __lasx_xvstelm_h(vec0, dst + 3 * DC_DEST_STRIDE, 0, 5);
486− __lasx_xvstelm_h(vec0, dst + 9 * DC_DEST_STRIDE, 0, 6);
487− __lasx_xvstelm_h(vec0, dst + 11 * DC_DEST_STRIDE, 0, 7);
488− __lasx_xvstelm_h(vec0, dst + 4 * DC_DEST_STRIDE, 0, 8);
489− __lasx_xvstelm_h(vec0, dst + 6 * DC_DEST_STRIDE, 0, 9);
490− __lasx_xvstelm_h(vec0, dst + 12 * DC_DEST_STRIDE, 0, 10);
491− __lasx_xvstelm_h(vec0, dst + 14 * DC_DEST_STRIDE, 0, 11);
492− __lasx_xvstelm_h(vec0, dst + 5 * DC_DEST_STRIDE, 0, 12);
493− __lasx_xvstelm_h(vec0, dst + 7 * DC_DEST_STRIDE, 0, 13);
494− __lasx_xvstelm_h(vec0, dst + 13 * DC_DEST_STRIDE, 0, 14);
495− __lasx_xvstelm_h(vec0, dst + 15 * DC_DEST_STRIDE, 0, 15);
496−
497−#undef DC_DEST_STRIDE
498−}
libavcodec/loongarch/h264idct_loongarch.c+184−0
@@ -0,0 +1,184 @@
1+/*
2+ * Loongson LSX/LASX optimized h264idct
3+ *
4+ * Copyright (c) 2023 Loongson Technology Corporation Limited
5+ * Contributed by Shiyou Yin <yinshiyou-hf@loongson.cn>
6+ * Xiwei Gu <guxiwei-hf@loongson.cn>
7+ *
8+ * This file is part of FFmpeg.
9+ *
10+ * FFmpeg is free software; you can redistribute it and/or
11+ * modify it under the terms of the GNU Lesser General Public
12+ * License as published by the Free Software Foundation; either
13+ * version 2.1 of the License, or (at your option) any later version.
14+ *
15+ * FFmpeg is distributed in the hope that it will be useful,
16+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
17+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18+ * Lesser General Public License for more details.
19+ *
20+ * You should have received a copy of the GNU Lesser General Public
21+ * License along with FFmpeg; if not, write to the Free Software
22+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23+ */
24+
25+#include "h264dsp_loongarch.h"
26+#include "libavcodec/bit_depth_template.c"
27+
28+void ff_h264_idct_add16_8_lsx(uint8_t *dst, const int32_t *blk_offset,
29+ int16_t *block, int32_t dst_stride,
30+ const uint8_t nzc[15 * 8])
31+{
32+ int32_t i;
33+
34+ for (i = 0; i < 16; i++) {
35+ int32_t nnz = nzc[scan8[i]];
36+
37+ if (nnz == 1 && ((dctcoef *) block)[i * 16]) {
38+ ff_h264_idct_dc_add_8_lsx(dst + blk_offset[i],
39+ block + i * 16 * sizeof(pixel),
40+ dst_stride);
41+ } else if (nnz) {
42+ ff_h264_idct_add_8_lsx(dst + blk_offset[i],
43+ block + i * 16 * sizeof(pixel),
44+ dst_stride);
45+ }
46+ }
47+}
48+
49+void ff_h264_idct8_add4_8_lsx(uint8_t *dst, const int32_t *blk_offset,
50+ int16_t *block, int32_t dst_stride,
51+ const uint8_t nzc[15 * 8])
52+{
53+ int32_t cnt;
54+
55+ for (cnt = 0; cnt < 16; cnt += 4) {
56+ int32_t nnz = nzc[scan8[cnt]];
57+
58+ if (nnz == 1 && ((dctcoef *) block)[cnt * 16]) {
59+ ff_h264_idct8_dc_add_8_lsx(dst + blk_offset[cnt],
60+ block + cnt * 16 * sizeof(pixel),
61+ dst_stride);
62+ } else if (nnz) {
63+ ff_h264_idct8_add_8_lsx(dst + blk_offset[cnt],
64+ block + cnt * 16 * sizeof(pixel),
65+ dst_stride);
66+ }
67+ }
68+}
69+
70+#if HAVE_LASX
71+void ff_h264_idct8_add4_8_lasx(uint8_t *dst, const int32_t *blk_offset,
72+ int16_t *block, int32_t dst_stride,
73+ const uint8_t nzc[15 * 8])
74+{
75+ int32_t cnt;
76+
77+ for (cnt = 0; cnt < 16; cnt += 4) {
78+ int32_t nnz = nzc[scan8[cnt]];
79+
80+ if (nnz == 1 && ((dctcoef *) block)[cnt * 16]) {
81+ ff_h264_idct8_dc_add_8_lasx(dst + blk_offset[cnt],
82+ block + cnt * 16 * sizeof(pixel),
83+ dst_stride);
84+ } else if (nnz) {
85+ ff_h264_idct8_add_8_lasx(dst + blk_offset[cnt],
86+ block + cnt * 16 * sizeof(pixel),
87+ dst_stride);
88+ }
89+ }
90+}
91+#endif // #if HAVE_LASX
92+
93+void ff_h264_idct_add8_8_lsx(uint8_t **dst, const int32_t *blk_offset,
94+ int16_t *block, int32_t dst_stride,
95+ const uint8_t nzc[15 * 8])
96+{
97+ int32_t i;
98+
99+ for (i = 16; i < 20; i++) {
100+ if (nzc[scan8[i]])
101+ ff_h264_idct_add_8_lsx(dst[0] + blk_offset[i],
102+ block + i * 16 * sizeof(pixel),
103+ dst_stride);
104+ else if (((dctcoef *) block)[i * 16])
105+ ff_h264_idct_dc_add_8_lsx(dst[0] + blk_offset[i],
106+ block + i * 16 * sizeof(pixel),
107+ dst_stride);
108+ }
109+ for (i = 32; i < 36; i++) {
110+ if (nzc[scan8[i]])
111+ ff_h264_idct_add_8_lsx(dst[1] + blk_offset[i],
112+ block + i * 16 * sizeof(pixel),
113+ dst_stride);
114+ else if (((dctcoef *) block)[i * 16])
115+ ff_h264_idct_dc_add_8_lsx(dst[1] + blk_offset[i],
116+ block + i * 16 * sizeof(pixel),
117+ dst_stride);
118+ }
119+}
120+
121+void ff_h264_idct_add8_422_8_lsx(uint8_t **dst, const int32_t *blk_offset,
122+ int16_t *block, int32_t dst_stride,
123+ const uint8_t nzc[15 * 8])
124+{
125+ int32_t i;
126+
127+ for (i = 16; i < 20; i++) {
128+ if (nzc[scan8[i]])
129+ ff_h264_idct_add_8_lsx(dst[0] + blk_offset[i],
130+ block + i * 16 * sizeof(pixel),
131+ dst_stride);
132+ else if (((dctcoef *) block)[i * 16])
133+ ff_h264_idct_dc_add_8_lsx(dst[0] + blk_offset[i],
134+ block + i * 16 * sizeof(pixel),
135+ dst_stride);
136+ }
137+ for (i = 20; i < 24; i++) {
138+ if (nzc[scan8[i + 4]])
139+ ff_h264_idct_add_8_lsx(dst[0] + blk_offset[i + 4],
140+ block + i * 16 * sizeof(pixel),
141+ dst_stride);
142+ else if (((dctcoef *) block)[i * 16])
143+ ff_h264_idct_dc_add_8_lsx(dst[0] + blk_offset[i + 4],
144+ block + i * 16 * sizeof(pixel),
145+ dst_stride);
146+ }
147+ for (i = 32; i < 36; i++) {
148+ if (nzc[scan8[i]])
149+ ff_h264_idct_add_8_lsx(dst[1] + blk_offset[i],
150+ block + i * 16 * sizeof(pixel),
151+ dst_stride);
152+ else if (((dctcoef *) block)[i * 16])
153+ ff_h264_idct_dc_add_8_lsx(dst[1] + blk_offset[i],
154+ block + i * 16 * sizeof(pixel),
155+ dst_stride);
156+ }
157+ for (i = 36; i < 40; i++) {
158+ if (nzc[scan8[i + 4]])
159+ ff_h264_idct_add_8_lsx(dst[1] + blk_offset[i + 4],
160+ block + i * 16 * sizeof(pixel),
161+ dst_stride);
162+ else if (((dctcoef *) block)[i * 16])
163+ ff_h264_idct_dc_add_8_lsx(dst[1] + blk_offset[i + 4],
164+ block + i * 16 * sizeof(pixel),
165+ dst_stride);
166+ }
167+}
168+
169+void ff_h264_idct_add16_intra_8_lsx(uint8_t *dst, const int32_t *blk_offset,
170+ int16_t *block, int32_t dst_stride,
171+ const uint8_t nzc[15 * 8])
172+{
173+ int32_t i;
174+
175+ for (i = 0; i < 16; i++) {
176+ if (nzc[scan8[i]])
177+ ff_h264_idct_add_8_lsx(dst + blk_offset[i],
178+ block + i * 16 * sizeof(pixel), dst_stride);
179+ else if (((dctcoef *) block)[i * 16])
180+ ff_h264_idct_dc_add_8_lsx(dst + blk_offset[i],
181+ block + i * 16 * sizeof(pixel),
182+ dst_stride);
183+ }
184+}
libavcodec/loongarch/loongson_asm.S+945−0
@@ -0,0 +1,945 @@
1+/*
2+ * Loongson asm helper.
3+ *
4+ * Copyright (c) 2022 Loongson Technology Corporation Limited
5+ * Contributed by Gu Xiwei(guxiwei-hf@loongson.cn)
6+ * Shiyou Yin(yinshiyou-hf@loongson.cn)
7+ *
8+ * This file is part of FFmpeg.
9+ *
10+ * FFmpeg is free software; you can redistribute it and/or
11+ * modify it under the terms of the GNU Lesser General Public
12+ * License as published by the Free Software Foundation; either
13+ * version 2.1 of the License, or (at your option) any later version.
14+ *
15+ * FFmpeg is distributed in the hope that it will be useful,
16+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
17+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18+ * Lesser General Public License for more details.
19+ *
20+ * You should have received a copy of the GNU Lesser General Public
21+ * License along with FFmpeg; if not, write to the Free Software
22+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23+ */
24+
25+/**
26+ * MAJOR version: Macro usage changes.
27+ * MINOR version: Add new functions, or bug fixes.
28+ * MICRO version: Comment changes or implementation changes.
29+ */
30+#define LML_VERSION_MAJOR 0
31+#define LML_VERSION_MINOR 2
32+#define LML_VERSION_MICRO 0
33+
34+/*
35+ *============================================================================
36+ * macros for specific projetc, set them as needed.
37+ * Following LoongML macros for your reference.
38+ *============================================================================
39+ */
40+#define ASM_PREF
41+#define DEFAULT_ALIGN 5
42+
43+.macro function name, align=DEFAULT_ALIGN
44+.macro endfunc
45+ jirl $r0, $r1, 0x0
46+ .size ASM_PREF\name, . - ASM_PREF\name
47+ .purgem endfunc
48+.endm
49+.text ;
50+.align \align ;
51+.globl ASM_PREF\name ;
52+.type ASM_PREF\name, @function ;
53+ASM_PREF\name: ;
54+.endm
55+
56+/**
57+ * Attention: If align is not zero, the macro will use
58+ * t7 until the end of function
59+ */
60+.macro alloc_stack size, align=0
61+.if \align
62+ .macro clean_stack
63+ add.d sp, sp, t7
64+ .endm
65+ addi.d sp, sp, - \size
66+ andi.d t7, sp, \align - 1
67+ sub.d sp, sp, t7
68+ addi.d t7, t7, \size
69+.else
70+ .macro clean_stack
71+ addi.d sp, sp, \size
72+ .endm
73+ addi.d sp, sp, - \size
74+.endif
75+.endm
76+
77+.macro const name, align=DEFAULT_ALIGN
78+ .macro endconst
79+ .size \name, . - \name
80+ .purgem endconst
81+ .endm
82+.section .rodata
83+.align \align
84+\name:
85+.endm
86+
87+/*
88+ *============================================================================
89+ * LoongArch register alias
90+ *============================================================================
91+ */
92+
93+#define a0 $a0
94+#define a1 $a1
95+#define a2 $a2
96+#define a3 $a3
97+#define a4 $a4
98+#define a5 $a5
99+#define a6 $a6
100+#define a7 $a7
101+
102+#define t0 $t0
103+#define t1 $t1
104+#define t2 $t2
105+#define t3 $t3
106+#define t4 $t4
107+#define t5 $t5
108+#define t6 $t6
109+#define t7 $t7
110+#define t8 $t8
111+
112+#define s0 $s0
113+#define s1 $s1
114+#define s2 $s2
115+#define s3 $s3
116+#define s4 $s4
117+#define s5 $s5
118+#define s6 $s6
119+#define s7 $s7
120+#define s8 $s8
121+
122+#define zero $zero
123+#define sp $sp
124+#define ra $ra
125+
126+#define f0 $f0
127+#define f1 $f1
128+#define f2 $f2
129+#define f3 $f3
130+#define f4 $f4
131+#define f5 $f5
132+#define f6 $f6
133+#define f7 $f7
134+#define f8 $f8
135+#define f9 $f9
136+#define f10 $f10
137+#define f11 $f11
138+#define f12 $f12
139+#define f13 $f13
140+#define f14 $f14
141+#define f15 $f15
142+#define f16 $f16
143+#define f17 $f17
144+#define f18 $f18
145+#define f19 $f19
146+#define f20 $f20
147+#define f21 $f21
148+#define f22 $f22
149+#define f23 $f23
150+#define f24 $f24
151+#define f25 $f25
152+#define f26 $f26
153+#define f27 $f27
154+#define f28 $f28
155+#define f29 $f29
156+#define f30 $f30
157+#define f31 $f31
158+
159+#define vr0 $vr0
160+#define vr1 $vr1
161+#define vr2 $vr2
162+#define vr3 $vr3
163+#define vr4 $vr4
164+#define vr5 $vr5
165+#define vr6 $vr6
166+#define vr7 $vr7
167+#define vr8 $vr8
168+#define vr9 $vr9
169+#define vr10 $vr10
170+#define vr11 $vr11
171+#define vr12 $vr12
172+#define vr13 $vr13
173+#define vr14 $vr14
174+#define vr15 $vr15
175+#define vr16 $vr16
176+#define vr17 $vr17
177+#define vr18 $vr18
178+#define vr19 $vr19
179+#define vr20 $vr20
180+#define vr21 $vr21
181+#define vr22 $vr22
182+#define vr23 $vr23
183+#define vr24 $vr24
184+#define vr25 $vr25
185+#define vr26 $vr26
186+#define vr27 $vr27
187+#define vr28 $vr28
188+#define vr29 $vr29
189+#define vr30 $vr30
190+#define vr31 $vr31
191+
192+#define xr0 $xr0
193+#define xr1 $xr1
194+#define xr2 $xr2
195+#define xr3 $xr3
196+#define xr4 $xr4
197+#define xr5 $xr5
198+#define xr6 $xr6
199+#define xr7 $xr7
200+#define xr8 $xr8
201+#define xr9 $xr9
202+#define xr10 $xr10
203+#define xr11 $xr11
204+#define xr12 $xr12
205+#define xr13 $xr13
206+#define xr14 $xr14
207+#define xr15 $xr15
208+#define xr16 $xr16
209+#define xr17 $xr17
210+#define xr18 $xr18
211+#define xr19 $xr19
212+#define xr20 $xr20
213+#define xr21 $xr21
214+#define xr22 $xr22
215+#define xr23 $xr23
216+#define xr24 $xr24
217+#define xr25 $xr25
218+#define xr26 $xr26
219+#define xr27 $xr27
220+#define xr28 $xr28
221+#define xr29 $xr29
222+#define xr30 $xr30
223+#define xr31 $xr31
224+
225+/*
226+ *============================================================================
227+ * LSX/LASX synthesize instructions
228+ *============================================================================
229+ */
230+
231+/*
232+ * Description : Dot product of byte vector elements
233+ * Arguments : Inputs - vj, vk
234+ * Outputs - vd
235+ * Return Type - halfword
236+ */
237+.macro vdp2.h.bu vd, vj, vk
238+ vmulwev.h.bu \vd, \vj, \vk
239+ vmaddwod.h.bu \vd, \vj, \vk
240+.endm
241+
242+.macro vdp2.h.bu.b vd, vj, vk
243+ vmulwev.h.bu.b \vd, \vj, \vk
244+ vmaddwod.h.bu.b \vd, \vj, \vk
245+.endm
246+
247+.macro vdp2.w.h vd, vj, vk
248+ vmulwev.w.h \vd, \vj, \vk
249+ vmaddwod.w.h \vd, \vj, \vk
250+.endm
251+
252+.macro xvdp2.h.bu xd, xj, xk
253+ xvmulwev.h.bu \xd, \xj, \xk
254+ xvmaddwod.h.bu \xd, \xj, \xk
255+.endm
256+
257+.macro xvdp2.h.bu.b xd, xj, xk
258+ xvmulwev.h.bu.b \xd, \xj, \xk
259+ xvmaddwod.h.bu.b \xd, \xj, \xk
260+.endm
261+
262+.macro xvdp2.w.h xd, xj, xk
263+ xvmulwev.w.h \xd, \xj, \xk
264+ xvmaddwod.w.h \xd, \xj, \xk
265+.endm
266+
267+/*
268+ * Description : Dot product & addition of halfword vector elements
269+ * Arguments : Inputs - vj, vk
270+ * Outputs - vd
271+ * Return Type - twice size of input
272+ */
273+.macro vdp2add.h.bu vd, vj, vk
274+ vmaddwev.h.bu \vd, \vj, \vk
275+ vmaddwod.h.bu \vd, \vj, \vk
276+.endm
277+
278+.macro vdp2add.h.bu.b vd, vj, vk
279+ vmaddwev.h.bu.b \vd, \vj, \vk
280+ vmaddwod.h.bu.b \vd, \vj, \vk
281+.endm
282+
283+.macro vdp2add.w.h vd, vj, vk
284+ vmaddwev.w.h \vd, \vj, \vk
285+ vmaddwod.w.h \vd, \vj, \vk
286+.endm
287+
288+.macro xvdp2add.h.bu.b xd, xj, xk
289+ xvmaddwev.h.bu.b \xd, \xj, \xk
290+ xvmaddwod.h.bu.b \xd, \xj, \xk
291+.endm
292+
293+.macro xvdp2add.w.h xd, xj, xk
294+ xvmaddwev.w.h \xd, \xj, \xk
295+ xvmaddwod.w.h \xd, \xj, \xk
296+.endm
297+
298+/*
299+ * Description : Range each element of vector
300+ * clip: vj > vk ? vj : vk && vj < va ? vj : va
301+ * clip255: vj < 255 ? vj : 255 && vj > 0 ? vj : 0
302+ */
303+.macro vclip.h vd, vj, vk, va
304+ vmax.h \vd, \vj, \vk
305+ vmin.h \vd, \vd, \va
306+.endm
307+
308+.macro vclip255.w vd, vj
309+ vmaxi.w \vd, \vj, 0
310+ vsat.wu \vd, \vd, 7
311+.endm
312+
313+.macro vclip255.h vd, vj
314+ vmaxi.h \vd, \vj, 0
315+ vsat.hu \vd, \vd, 7
316+.endm
317+
318+.macro xvclip.h xd, xj, xk, xa
319+ xvmax.h \xd, \xj, \xk
320+ xvmin.h \xd, \xd, \xa
321+.endm
322+
323+.macro xvclip255.h xd, xj
324+ xvmaxi.h \xd, \xj, 0
325+ xvsat.hu \xd, \xd, 7
326+.endm
327+
328+.macro xvclip255.w xd, xj
329+ xvmaxi.w \xd, \xj, 0
330+ xvsat.wu \xd, \xd, 7
331+.endm
332+
333+/*
334+ * Description : Store elements of vector
335+ * vd : Data vector to be stroed
336+ * rk : Address of data storage
337+ * ra : Offset of address
338+ * si : Index of data in vd
339+ */
340+.macro vstelmx.b vd, rk, ra, si
341+ add.d \rk, \rk, \ra
342+ vstelm.b \vd, \rk, 0, \si
343+.endm
344+
345+.macro vstelmx.h vd, rk, ra, si
346+ add.d \rk, \rk, \ra
347+ vstelm.h \vd, \rk, 0, \si
348+.endm
349+
350+.macro vstelmx.w vd, rk, ra, si
351+ add.d \rk, \rk, \ra
352+ vstelm.w \vd, \rk, 0, \si
353+.endm
354+
355+.macro vstelmx.d vd, rk, ra, si
356+ add.d \rk, \rk, \ra
357+ vstelm.d \vd, \rk, 0, \si
358+.endm
359+
360+.macro vmov xd, xj
361+ vor.v \xd, \xj, \xj
362+.endm
363+
364+.macro xmov xd, xj
365+ xvor.v \xd, \xj, \xj
366+.endm
367+
368+.macro xvstelmx.d xd, rk, ra, si
369+ add.d \rk, \rk, \ra
370+ xvstelm.d \xd, \rk, 0, \si
371+.endm
372+
373+/*
374+ *============================================================================
375+ * LSX/LASX custom macros
376+ *============================================================================
377+ */
378+
379+/*
380+ * Load 4 float, double, V128, v256 elements with stride.
381+ */
382+.macro FLDS_LOADX_4 src, stride, stride2, stride3, out0, out1, out2, out3
383+ fld.s \out0, \src, 0
384+ fldx.s \out1, \src, \stride
385+ fldx.s \out2, \src, \stride2
386+ fldx.s \out3, \src, \stride3
387+.endm
388+
389+.macro FLDD_LOADX_4 src, stride, stride2, stride3, out0, out1, out2, out3
390+ fld.d \out0, \src, 0
391+ fldx.d \out1, \src, \stride
392+ fldx.d \out2, \src, \stride2
393+ fldx.d \out3, \src, \stride3
394+.endm
395+
396+.macro LSX_LOADX_4 src, stride, stride2, stride3, out0, out1, out2, out3
397+ vld \out0, \src, 0
398+ vldx \out1, \src, \stride
399+ vldx \out2, \src, \stride2
400+ vldx \out3, \src, \stride3
401+.endm
402+
403+.macro LASX_LOADX_4 src, stride, stride2, stride3, out0, out1, out2, out3
404+ xvld \out0, \src, 0
405+ xvldx \out1, \src, \stride
406+ xvldx \out2, \src, \stride2
407+ xvldx \out3, \src, \stride3
408+.endm
409+
410+/*
411+ * Description : Transpose 4x4 block with half-word elements in vectors
412+ * Arguments : Inputs - in0, in1, in2, in3
413+ * Outputs - out0, out1, out2, out3
414+ */
415+.macro LSX_TRANSPOSE4x4_H in0, in1, in2, in3, out0, out1, out2, out3, \
416+ tmp0, tmp1
417+ vilvl.h \tmp0, \in1, \in0
418+ vilvl.h \tmp1, \in3, \in2
419+ vilvl.w \out0, \tmp1, \tmp0
420+ vilvh.w \out2, \tmp1, \tmp0
421+ vilvh.d \out1, \out0, \out0
422+ vilvh.d \out3, \out0, \out2
423+.endm
424+
425+/*
426+ * Description : Transpose 4x4 block with word elements in vectors
427+ * Arguments : Inputs - in0, in1, in2, in3
428+ * Outputs - out0, out1, out2, out3
429+ * Details :
430+ * Example :
431+ * 1, 2, 3, 4 1, 5, 9,13
432+ * 5, 6, 7, 8 to 2, 6,10,14
433+ * 9,10,11,12 =====> 3, 7,11,15
434+ * 13,14,15,16 4, 8,12,16
435+ */
436+.macro LSX_TRANSPOSE4x4_W _in0, _in1, _in2, _in3, _out0, _out1, _out2, _out3, \
437+ _tmp0, _tmp1
438+
439+ vilvl.w \_tmp0, \_in1, \_in0
440+ vilvh.w \_out1, \_in1, \_in0
441+ vilvl.w \_tmp1, \_in3, \_in2
442+ vilvh.w \_out3, \_in3, \_in2
443+
444+ vilvl.d \_out0, \_tmp1, \_tmp0
445+ vilvl.d \_out2, \_out3, \_out1
446+ vilvh.d \_out3, \_out3, \_out1
447+ vilvh.d \_out1, \_tmp1, \_tmp0
448+.endm
449+
450+/*
451+ * Description : Transpose 8x8 block with half-word elements in vectors
452+ * Arguments : Inputs - in0, in1, in2, in3, in4, in5, in6, in7
453+ * Outputs - out0, out1, out2, out3, out4, out5, out6, out7
454+ */
455+.macro LSX_TRANSPOSE8x8_H in0, in1, in2, in3, in4, in5, in6, in7, out0, out1, \
456+ out2, out3, out4, out5, out6, out7, tmp0, tmp1, tmp2, \
457+ tmp3, tmp4, tmp5, tmp6, tmp7
458+ vilvl.h \tmp0, \in6, \in4
459+ vilvl.h \tmp1, \in7, \in5
460+ vilvl.h \tmp2, \in2, \in0
461+ vilvl.h \tmp3, \in3, \in1
462+
463+ vilvl.h \tmp4, \tmp1, \tmp0
464+ vilvh.h \tmp5, \tmp1, \tmp0
465+ vilvl.h \tmp6, \tmp3, \tmp2
466+ vilvh.h \tmp7, \tmp3, \tmp2
467+
468+ vilvh.h \tmp0, \in6, \in4
469+ vilvh.h \tmp1, \in7, \in5
470+ vilvh.h \tmp2, \in2, \in0
471+ vilvh.h \tmp3, \in3, \in1
472+
473+ vpickev.d \out0, \tmp4, \tmp6
474+ vpickod.d \out1, \tmp4, \tmp6
475+ vpickev.d \out2, \tmp5, \tmp7
476+ vpickod.d \out3, \tmp5, \tmp7
477+
478+ vilvl.h \tmp4, \tmp1, \tmp0
479+ vilvh.h \tmp5, \tmp1, \tmp0
480+ vilvl.h \tmp6, \tmp3, \tmp2
481+ vilvh.h \tmp7, \tmp3, \tmp2
482+
483+ vpickev.d \out4, \tmp4, \tmp6
484+ vpickod.d \out5, \tmp4, \tmp6
485+ vpickev.d \out6, \tmp5, \tmp7
486+ vpickod.d \out7, \tmp5, \tmp7
487+.endm
488+
489+/*
490+ * Description : Transpose 16x8 block with byte elements in vectors
491+ * Arguments : Inputs - in0, in1, in2, in3, in4, in5, in6, in7
492+ * Outputs - out0, out1, out2, out3, out4, out5, out6, out7
493+ */
494+.macro LASX_TRANSPOSE16X8_B in0, in1, in2, in3, in4, in5, in6, in7, \
495+ in8, in9, in10, in11, in12, in13, in14, in15, \
496+ out0, out1, out2, out3, out4, out5, out6, out7,\
497+ tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7
498+ xvilvl.b \tmp0, \in2, \in0
499+ xvilvl.b \tmp1, \in3, \in1
500+ xvilvl.b \tmp2, \in6, \in4
501+ xvilvl.b \tmp3, \in7, \in5
502+ xvilvl.b \tmp4, \in10, \in8
503+ xvilvl.b \tmp5, \in11, \in9
504+ xvilvl.b \tmp6, \in14, \in12
505+ xvilvl.b \tmp7, \in15, \in13
506+ xvilvl.b \out0, \tmp1, \tmp0
507+ xvilvh.b \out1, \tmp1, \tmp0
508+ xvilvl.b \out2, \tmp3, \tmp2
509+ xvilvh.b \out3, \tmp3, \tmp2
510+ xvilvl.b \out4, \tmp5, \tmp4
511+ xvilvh.b \out5, \tmp5, \tmp4
512+ xvilvl.b \out6, \tmp7, \tmp6
513+ xvilvh.b \out7, \tmp7, \tmp6
514+ xvilvl.w \tmp0, \out2, \out0
515+ xvilvh.w \tmp2, \out2, \out0
516+ xvilvl.w \tmp4, \out3, \out1
517+ xvilvh.w \tmp6, \out3, \out1
518+ xvilvl.w \tmp1, \out6, \out4
519+ xvilvh.w \tmp3, \out6, \out4
520+ xvilvl.w \tmp5, \out7, \out5
521+ xvilvh.w \tmp7, \out7, \out5
522+ xvilvl.d \out0, \tmp1, \tmp0
523+ xvilvh.d \out1, \tmp1, \tmp0
524+ xvilvl.d \out2, \tmp3, \tmp2
525+ xvilvh.d \out3, \tmp3, \tmp2
526+ xvilvl.d \out4, \tmp5, \tmp4
527+ xvilvh.d \out5, \tmp5, \tmp4
528+ xvilvl.d \out6, \tmp7, \tmp6
529+ xvilvh.d \out7, \tmp7, \tmp6
530+.endm
531+
532+/*
533+ * Description : Transpose 16x8 block with byte elements in vectors
534+ * Arguments : Inputs - in0, in1, in2, in3, in4, in5, in6, in7
535+ * Outputs - out0, out1, out2, out3, out4, out5, out6, out7
536+ */
537+.macro LSX_TRANSPOSE16X8_B in0, in1, in2, in3, in4, in5, in6, in7, \
538+ in8, in9, in10, in11, in12, in13, in14, in15, \
539+ out0, out1, out2, out3, out4, out5, out6, out7,\
540+ tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7
541+ vilvl.b \tmp0, \in2, \in0
542+ vilvl.b \tmp1, \in3, \in1
543+ vilvl.b \tmp2, \in6, \in4
544+ vilvl.b \tmp3, \in7, \in5
545+ vilvl.b \tmp4, \in10, \in8
546+ vilvl.b \tmp5, \in11, \in9
547+ vilvl.b \tmp6, \in14, \in12
548+ vilvl.b \tmp7, \in15, \in13
549+
550+ vilvl.b \out0, \tmp1, \tmp0
551+ vilvh.b \out1, \tmp1, \tmp0
552+ vilvl.b \out2, \tmp3, \tmp2
553+ vilvh.b \out3, \tmp3, \tmp2
554+ vilvl.b \out4, \tmp5, \tmp4
555+ vilvh.b \out5, \tmp5, \tmp4
556+ vilvl.b \out6, \tmp7, \tmp6
557+ vilvh.b \out7, \tmp7, \tmp6
558+ vilvl.w \tmp0, \out2, \out0
559+ vilvh.w \tmp2, \out2, \out0
560+ vilvl.w \tmp4, \out3, \out1
561+ vilvh.w \tmp6, \out3, \out1
562+ vilvl.w \tmp1, \out6, \out4
563+ vilvh.w \tmp3, \out6, \out4
564+ vilvl.w \tmp5, \out7, \out5
565+ vilvh.w \tmp7, \out7, \out5
566+ vilvl.d \out0, \tmp1, \tmp0
567+ vilvh.d \out1, \tmp1, \tmp0
568+ vilvl.d \out2, \tmp3, \tmp2
569+ vilvh.d \out3, \tmp3, \tmp2
570+ vilvl.d \out4, \tmp5, \tmp4
571+ vilvh.d \out5, \tmp5, \tmp4
572+ vilvl.d \out6, \tmp7, \tmp6
573+ vilvh.d \out7, \tmp7, \tmp6
574+.endm
575+
576+/*
577+ * Description : Transpose 4x4 block with half-word elements in vectors
578+ * Arguments : Inputs - in0, in1, in2, in3
579+ * Outputs - out0, out1, out2, out3
580+ */
581+.macro LASX_TRANSPOSE4x4_H in0, in1, in2, in3, out0, out1, out2, out3, \
582+ tmp0, tmp1
583+ xvilvl.h \tmp0, \in1, \in0
584+ xvilvl.h \tmp1, \in3, \in2
585+ xvilvl.w \out0, \tmp1, \tmp0
586+ xvilvh.w \out2, \tmp1, \tmp0
587+ xvilvh.d \out1, \out0, \out0
588+ xvilvh.d \out3, \out0, \out2
589+.endm
590+
591+/*
592+ * Description : Transpose 4x8 block with half-word elements in vectors
593+ * Arguments : Inputs - in0, in1, in2, in3
594+ * Outputs - out0, out1, out2, out3
595+ */
596+.macro LASX_TRANSPOSE4x8_H in0, in1, in2, in3, out0, out1, out2, out3, \
597+ tmp0, tmp1
598+ xvilvl.h \tmp0, \in2, \in0
599+ xvilvl.h \tmp1, \in3, \in1
600+ xvilvl.h \out2, \tmp1, \tmp0
601+ xvilvh.h \out3, \tmp1, \tmp0
602+
603+ xvilvl.d \out0, \out2, \out2
604+ xvilvh.d \out1, \out2, \out2
605+ xvilvl.d \out2, \out3, \out3
606+ xvilvh.d \out3, \out3, \out3
607+.endm
608+
609+/*
610+ * Description : Transpose 8x8 block with half-word elements in vectors
611+ * Arguments : Inputs - in0, in1, in2, in3, in4, in5, in6, in7
612+ * Outputs - out0, out1, out2, out3, out4, out5, out6, out7
613+ */
614+.macro LASX_TRANSPOSE8x8_H in0, in1, in2, in3, in4, in5, in6, in7, \
615+ out0, out1, out2, out3, out4, out5, out6, out7, \
616+ tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7
617+ xvilvl.h \tmp0, \in6, \in4
618+ xvilvl.h \tmp1, \in7, \in5
619+ xvilvl.h \tmp2, \in2, \in0
620+ xvilvl.h \tmp3, \in3, \in1
621+
622+ xvilvl.h \tmp4, \tmp1, \tmp0
623+ xvilvh.h \tmp5, \tmp1, \tmp0
624+ xvilvl.h \tmp6, \tmp3, \tmp2
625+ xvilvh.h \tmp7, \tmp3, \tmp2
626+
627+ xvilvh.h \tmp0, \in6, \in4
628+ xvilvh.h \tmp1, \in7, \in5
629+ xvilvh.h \tmp2, \in2, \in0
630+ xvilvh.h \tmp3, \in3, \in1
631+
632+ xvpickev.d \out0, \tmp4, \tmp6
633+ xvpickod.d \out1, \tmp4, \tmp6
634+ xvpickev.d \out2, \tmp5, \tmp7
635+ xvpickod.d \out3, \tmp5, \tmp7
636+
637+ xvilvl.h \tmp4, \tmp1, \tmp0
638+ xvilvh.h \tmp5, \tmp1, \tmp0
639+ xvilvl.h \tmp6, \tmp3, \tmp2
640+ xvilvh.h \tmp7, \tmp3, \tmp2
641+
642+ xvpickev.d \out4, \tmp4, \tmp6
643+ xvpickod.d \out5, \tmp4, \tmp6
644+ xvpickev.d \out6, \tmp5, \tmp7
645+ xvpickod.d \out7, \tmp5, \tmp7
646+.endm
647+
648+/*
649+ * Description : Transpose 2x4x4 block with half-word elements in vectors
650+ * Arguments : Inputs - in0, in1, in2, in3
651+ * Outputs - out0, out1, out2, out3
652+ */
653+.macro LASX_TRANSPOSE2x4x4_H in0, in1, in2, in3, out0, out1, out2, out3, \
654+ tmp0, tmp1, tmp2
655+ xvilvh.h \tmp1, \in0, \in1
656+ xvilvl.h \out1, \in0, \in1
657+ xvilvh.h \tmp0, \in2, \in3
658+ xvilvl.h \out3, \in2, \in3
659+
660+ xvilvh.w \tmp2, \out3, \out1
661+ xvilvl.w \out3, \out3, \out1
662+
663+ xvilvl.w \out2, \tmp0, \tmp1
664+ xvilvh.w \tmp1, \tmp0, \tmp1
665+
666+ xvilvh.d \out0, \out2, \out3
667+ xvilvl.d \out2, \out2, \out3
668+ xvilvh.d \out1, \tmp1, \tmp2
669+ xvilvl.d \out3, \tmp1, \tmp2
670+.endm
671+
672+/*
673+ * Description : Transpose 4x4 block with word elements in vectors
674+ * Arguments : Inputs - in0, in1, in2, in3
675+ * Outputs - out0, out1, out2, out3
676+ * Details :
677+ * Example :
678+ * 1, 2, 3, 4, 1, 2, 3, 4 1,5, 9,13, 1,5, 9,13
679+ * 5, 6, 7, 8, 5, 6, 7, 8 to 2,6,10,14, 2,6,10,14
680+ * 9,10,11,12, 9,10,11,12 =====> 3,7,11,15, 3,7,11,15
681+ * 13,14,15,16, 13,14,15,16 4,8,12,16, 4,8,12,16
682+ */
683+.macro LASX_TRANSPOSE4x4_W _in0, _in1, _in2, _in3, _out0, _out1, _out2, _out3, \
684+ _tmp0, _tmp1
685+
686+ xvilvl.w \_tmp0, \_in1, \_in0
687+ xvilvh.w \_out1, \_in1, \_in0
688+ xvilvl.w \_tmp1, \_in3, \_in2
689+ xvilvh.w \_out3, \_in3, \_in2
690+
691+ xvilvl.d \_out0, \_tmp1, \_tmp0
692+ xvilvl.d \_out2, \_out3, \_out1
693+ xvilvh.d \_out3, \_out3, \_out1
694+ xvilvh.d \_out1, \_tmp1, \_tmp0
695+.endm
696+
697+/*
698+ * Description : Transpose 8x8 block with word elements in vectors
699+ * Arguments : Inputs - _in0, _in1, _in2, _in3, _in4, _in5, _in6, _in7
700+ * Outputs - _out0, _out1, _out2, _out3, _out4, _out5, _out6,
701+ * _out7
702+ * Example : LASX_TRANSPOSE8x8_W
703+ * _in0 : 1,2,3,4,5,6,7,8
704+ * _in1 : 2,2,3,4,5,6,7,8
705+ * _in2 : 3,2,3,4,5,6,7,8
706+ * _in3 : 4,2,3,4,5,6,7,8
707+ * _in4 : 5,2,3,4,5,6,7,8
708+ * _in5 : 6,2,3,4,5,6,7,8
709+ * _in6 : 7,2,3,4,5,6,7,8
710+ * _in7 : 8,2,3,4,5,6,7,8
711+ *
712+ * _out0 : 1,2,3,4,5,6,7,8
713+ * _out1 : 2,2,2,2,2,2,2,2
714+ * _out2 : 3,3,3,3,3,3,3,3
715+ * _out3 : 4,4,4,4,4,4,4,4
716+ * _out4 : 5,5,5,5,5,5,5,5
717+ * _out5 : 6,6,6,6,6,6,6,6
718+ * _out6 : 7,7,7,7,7,7,7,7
719+ * _out7 : 8,8,8,8,8,8,8,8
720+ */
721+.macro LASX_TRANSPOSE8x8_W _in0, _in1, _in2, _in3, _in4, _in5, _in6, _in7,\
722+ _out0, _out1, _out2, _out3, _out4, _out5, _out6, _out7,\
723+ _tmp0, _tmp1, _tmp2, _tmp3
724+ xvilvl.w \_tmp0, \_in2, \_in0
725+ xvilvl.w \_tmp1, \_in3, \_in1
726+ xvilvh.w \_tmp2, \_in2, \_in0
727+ xvilvh.w \_tmp3, \_in3, \_in1
728+ xvilvl.w \_out0, \_tmp1, \_tmp0
729+ xvilvh.w \_out1, \_tmp1, \_tmp0
730+ xvilvl.w \_out2, \_tmp3, \_tmp2
731+ xvilvh.w \_out3, \_tmp3, \_tmp2
732+
733+ xvilvl.w \_tmp0, \_in6, \_in4
734+ xvilvl.w \_tmp1, \_in7, \_in5
735+ xvilvh.w \_tmp2, \_in6, \_in4
736+ xvilvh.w \_tmp3, \_in7, \_in5
737+ xvilvl.w \_out4, \_tmp1, \_tmp0
738+ xvilvh.w \_out5, \_tmp1, \_tmp0
739+ xvilvl.w \_out6, \_tmp3, \_tmp2
740+ xvilvh.w \_out7, \_tmp3, \_tmp2
741+
742+ xmov \_tmp0, \_out0
743+ xmov \_tmp1, \_out1
744+ xmov \_tmp2, \_out2
745+ xmov \_tmp3, \_out3
746+ xvpermi.q \_out0, \_out4, 0x02
747+ xvpermi.q \_out1, \_out5, 0x02
748+ xvpermi.q \_out2, \_out6, 0x02
749+ xvpermi.q \_out3, \_out7, 0x02
750+ xvpermi.q \_out4, \_tmp0, 0x31
751+ xvpermi.q \_out5, \_tmp1, 0x31
752+ xvpermi.q \_out6, \_tmp2, 0x31
753+ xvpermi.q \_out7, \_tmp3, 0x31
754+.endm
755+
756+/*
757+ * Description : Transpose 4x4 block with double-word elements in vectors
758+ * Arguments : Inputs - _in0, _in1, _in2, _in3
759+ * Outputs - _out0, _out1, _out2, _out3
760+ * Example : LASX_TRANSPOSE4x4_D
761+ * _in0 : 1,2,3,4
762+ * _in1 : 1,2,3,4
763+ * _in2 : 1,2,3,4
764+ * _in3 : 1,2,3,4
765+ *
766+ * _out0 : 1,1,1,1
767+ * _out1 : 2,2,2,2
768+ * _out2 : 3,3,3,3
769+ * _out3 : 4,4,4,4
770+ */
771+.macro LASX_TRANSPOSE4x4_D _in0, _in1, _in2, _in3, _out0, _out1, _out2, _out3, \
772+ _tmp0, _tmp1
773+ xvilvl.d \_tmp0, \_in1, \_in0
774+ xvilvh.d \_out1, \_in1, \_in0
775+ xvilvh.d \_tmp1, \_in3, \_in2
776+ xvilvl.d \_out2, \_in3, \_in2
777+
778+ xvor.v \_out0, \_tmp0, \_tmp0
779+ xvor.v \_out3, \_tmp1, \_tmp1
780+
781+ xvpermi.q \_out0, \_out2, 0x02
782+ xvpermi.q \_out2, \_tmp0, 0x31
783+ xvpermi.q \_out3, \_out1, 0x31
784+ xvpermi.q \_out1, \_tmp1, 0x02
785+.endm
786+
787+/*
788+ * Description : Butterfly of 4 input vectors
789+ * Arguments : Inputs - _in0, _in1, _in2, _in3
790+ * Outputs - _out0, _out1, _out2, _out3
791+ * Details : Butterfly operation
792+ * Example : LSX_BUTTERFLY_4
793+ * _out0 = _in0 + _in3;
794+ * _out1 = _in1 + _in2;
795+ * _out2 = _in1 - _in2;
796+ * _out3 = _in0 - _in3;
797+ */
798+.macro LSX_BUTTERFLY_4_B _in0, _in1, _in2, _in3, _out0, _out1, _out2, _out3
799+ vadd.b \_out0, \_in0, \_in3
800+ vadd.b \_out1, \_in1, \_in2
801+ vsub.b \_out2, \_in1, \_in2
802+ vsub.b \_out3, \_in0, \_in3
803+.endm
804+.macro LSX_BUTTERFLY_4_H _in0, _in1, _in2, _in3, _out0, _out1, _out2, _out3
805+ vadd.h \_out0, \_in0, \_in3
806+ vadd.h \_out1, \_in1, \_in2
807+ vsub.h \_out2, \_in1, \_in2
808+ vsub.h \_out3, \_in0, \_in3
809+.endm
810+.macro LSX_BUTTERFLY_4_W _in0, _in1, _in2, _in3, _out0, _out1, _out2, _out3
811+ vadd.w \_out0, \_in0, \_in3
812+ vadd.w \_out1, \_in1, \_in2
813+ vsub.w \_out2, \_in1, \_in2
814+ vsub.w \_out3, \_in0, \_in3
815+.endm
816+.macro LSX_BUTTERFLY_4_D _in0, _in1, _in2, _in3, _out0, _out1, _out2, _out3
817+ vadd.d \_out0, \_in0, \_in3
818+ vadd.d \_out1, \_in1, \_in2
819+ vsub.d \_out2, \_in1, \_in2
820+ vsub.d \_out3, \_in0, \_in3
821+.endm
822+
823+.macro LASX_BUTTERFLY_4_B _in0, _in1, _in2, _in3, _out0, _out1, _out2, _out3
824+ xvadd.b \_out0, \_in0, \_in3
825+ xvadd.b \_out1, \_in1, \_in2
826+ xvsub.b \_out2, \_in1, \_in2
827+ xvsub.b \_out3, \_in0, \_in3
828+.endm
829+.macro LASX_BUTTERFLY_4_H _in0, _in1, _in2, _in3, _out0, _out1, _out2, _out3
830+ xvadd.h \_out0, \_in0, \_in3
831+ xvadd.h \_out1, \_in1, \_in2
832+ xvsub.h \_out2, \_in1, \_in2
833+ xvsub.h \_out3, \_in0, \_in3
834+.endm
835+.macro LASX_BUTTERFLY_4_W _in0, _in1, _in2, _in3, _out0, _out1, _out2, _out3
836+ xvadd.w \_out0, \_in0, \_in3
837+ xvadd.w \_out1, \_in1, \_in2
838+ xvsub.w \_out2, \_in1, \_in2
839+ xvsub.w \_out3, \_in0, \_in3
840+.endm
841+.macro LASX_BUTTERFLY_4_D _in0, _in1, _in2, _in3, _out0, _out1, _out2, _out3
842+ xvadd.d \_out0, \_in0, \_in3
843+ xvadd.d \_out1, \_in1, \_in2
844+ xvsub.d \_out2, \_in1, \_in2
845+ xvsub.d \_out3, \_in0, \_in3
846+.endm
847+
848+/*
849+ * Description : Butterfly of 8 input vectors
850+ * Arguments : Inputs - _in0, _in1, _in2, _in3, ~
851+ * Outputs - _out0, _out1, _out2, _out3, ~
852+ * Details : Butterfly operation
853+ * Example : LASX_BUTTERFLY_8
854+ * _out0 = _in0 + _in7;
855+ * _out1 = _in1 + _in6;
856+ * _out2 = _in2 + _in5;
857+ * _out3 = _in3 + _in4;
858+ * _out4 = _in3 - _in4;
859+ * _out5 = _in2 - _in5;
860+ * _out6 = _in1 - _in6;
861+ * _out7 = _in0 - _in7;
862+ */
863+.macro LSX_BUTTERFLY_8_B _in0, _in1, _in2, _in3, _in4, _in5, _in6, _in7, \
864+ _out0, _out1, _out2, _out3, _out4, _out5, _out6, _out7
865+ vadd.b \_out0, \_in0, \_in7
866+ vadd.b \_out1, \_in1, \_in6
867+ vadd.b \_out2, \_in2, \_in5
868+ vadd.b \_out3, \_in3, \_in4
869+ vsub.b \_out4, \_in3, \_in4
870+ vsub.b \_out5, \_in2, \_in5
871+ vsub.b \_out6, \_in1, \_in6
872+ vsub.b \_out7, \_in0, \_in7
873+.endm
874+
875+.macro LSX_BUTTERFLY_8_H _in0, _in1, _in2, _in3, _in4, _in5, _in6, _in7, \
876+ _out0, _out1, _out2, _out3, _out4, _out5, _out6, _out7
877+ vadd.h \_out0, \_in0, \_in7
878+ vadd.h \_out1, \_in1, \_in6
879+ vadd.h \_out2, \_in2, \_in5
880+ vadd.h \_out3, \_in3, \_in4
881+ vsub.h \_out4, \_in3, \_in4
882+ vsub.h \_out5, \_in2, \_in5
883+ vsub.h \_out6, \_in1, \_in6
884+ vsub.h \_out7, \_in0, \_in7
885+.endm
886+
887+.macro LSX_BUTTERFLY_8_W _in0, _in1, _in2, _in3, _in4, _in5, _in6, _in7, \
888+ _out0, _out1, _out2, _out3, _out4, _out5, _out6, _out7
889+ vadd.w \_out0, \_in0, \_in7
890+ vadd.w \_out1, \_in1, \_in6
891+ vadd.w \_out2, \_in2, \_in5
892+ vadd.w \_out3, \_in3, \_in4
893+ vsub.w \_out4, \_in3, \_in4
894+ vsub.w \_out5, \_in2, \_in5
895+ vsub.w \_out6, \_in1, \_in6
896+ vsub.w \_out7, \_in0, \_in7
897+.endm
898+
899+.macro LSX_BUTTERFLY_8_D _in0, _in1, _in2, _in3, _in4, _in5, _in6, _in7, \
900+ _out0, _out1, _out2, _out3, _out4, _out5, _out6, _out7
901+ vadd.d \_out0, \_in0, \_in7
902+ vadd.d \_out1, \_in1, \_in6
903+ vadd.d \_out2, \_in2, \_in5
904+ vadd.d \_out3, \_in3, \_in4
905+ vsub.d \_out4, \_in3, \_in4
906+ vsub.d \_out5, \_in2, \_in5
907+ vsub.d \_out6, \_in1, \_in6
908+ vsub.d \_out7, \_in0, \_in7
909+.endm
910+
911+.macro LASX_BUTTERFLY_8_B _in0, _in1, _in2, _in3, _in4, _in5, _in6, _in7, \
912+ _out0, _out1, _out2, _out3, _out4, _out5, _out6, _out7
913+ xvadd.b \_out0, \_in0, \_in7
914+ xvadd.b \_out1, \_in1, \_in6
915+ xvadd.b \_out2, \_in2, \_in5
916+ xvadd.b \_out3, \_in3, \_in4
917+ xvsub.b \_out4, \_in3, \_in4
918+ xvsub.b \_out5, \_in2, \_in5
919+ xvsub.b \_out6, \_in1, \_in6
920+ xvsub.b \_out7, \_in0, \_in7
921+.endm
922+
923+.macro LASX_BUTTERFLY_8_H _in0, _in1, _in2, _in3, _in4, _in5, _in6, _in7, \
924+ _out0, _out1, _out2, _out3, _out4, _out5, _out6, _out7
925+ xvadd.h \_out0, \_in0, \_in7
926+ xvadd.h \_out1, \_in1, \_in6
927+ xvadd.h \_out2, \_in2, \_in5
928+ xvadd.h \_out3, \_in3, \_in4
929+ xvsub.h \_out4, \_in3, \_in4
930+ xvsub.h \_out5, \_in2, \_in5
931+ xvsub.h \_out6, \_in1, \_in6
932+ xvsub.h \_out7, \_in0, \_in7
933+.endm
934+
935+.macro LASX_BUTTERFLY_8_W _in0, _in1, _in2, _in3, _in4, _in5, _in6, _in7, \
936+ _out0, _out1, _out2, _out3, _out4, _out5, _out6, _out7
937+ xvadd.w \_out0, \_in0, \_in7
938+ xvadd.w \_out1, \_in1, \_in6
939+ xvadd.w \_out2, \_in2, \_in5
940+ xvadd.w \_out3, \_in3, \_in4
941+ xvsub.w \_out4, \_in3, \_in4
942+ xvsub.w \_out5, \_in2, \_in5
943+ xvsub.w \_out6, \_in1, \_in6
944+ xvsub.w \_out7, \_in0, \_in7
945+.endm

Keyboard shortcuts

General

Show keyboard shortcuts
?
Focus the current search
/

Issues and pull requests

Open filters
f

Issues, pull requests and Inbox

Select next item
j↓
Select previous item
k↑
Open selected item
Enter
Dismiss filters or clear selection
Esc

Repository code

Go to file
t