avcodec/mips: MSA (MIPS-SIMD-Arch) optimizations for idctdsp functions

This patch adds MSA (MIPS-SIMD-Arch) optimizations for idctdsp functions in new file idctdsp_msa.c and simple_idct_msa.c

Shivraj Patil committed 11 years ago

Commit d12f76f1 parent 9c957348 files · +857 −0Browse files

Unified · Split

8 files

libavcodec/idctdsp.c +2 −0
libavcodec/idctdsp.h +2 −0
libavcodec/mips/Makefile +3 −0
libavcodec/mips/idctdsp_init_mips.c +49 −0
libavcodec/mips/idctdsp_mips.h +42 −0
libavcodec/mips/idctdsp_msa.c +149 −0
libavcodec/mips/simple_idct_msa.c +573 −0
libavutil/mips/generic_macros_msa.h +37 −0

libavcodec/idctdsp.c+2−0

@@ -305,6 +305,8 @@ av_cold void ff_idctdsp_init(IDCTDSPContext c, AVCodecContext avctx) · 304 unmodified lines
305	305	ff_idctdsp_init_ppc(c, avctx, high_bit_depth);
306	306	if (ARCH_X86)
307	307	ff_idctdsp_init_x86(c, avctx, high_bit_depth);
	308	+ if (ARCH_MIPS)
	309	+ ff_idctdsp_init_mips(c, avctx, high_bit_depth);
308	310
309	311	ff_put_pixels_clamped = c->put_pixels_clamped;
310	312	ff_add_pixels_clamped = c->add_pixels_clamped;

libavcodec/idctdsp.h+2−0

@@ -108,5 +108,7 @@ void ff_idctdsp_init_ppc(IDCTDSPContext c, AVCodecContext avctx, · 107 unmodified lines
108	108	unsigned high_bit_depth);
109	109	void ff_idctdsp_init_x86(IDCTDSPContext c, AVCodecContext avctx,
110	110	unsigned high_bit_depth);
	111	+void ff_idctdsp_init_mips(IDCTDSPContext c, AVCodecContext avctx,
	112	+ unsigned high_bit_depth);
111	113
112	114	#endif /* AVCODEC_IDCTDSP_H */

libavcodec/mips/Makefile+3−0

@@ -29,6 +29,7 @@ OBJS-$(CONFIG_QPELDSP) += mips/qpeldsp_init_mips.o · 28 unmodified lines
29	29	OBJS-$(CONFIG_HPELDSP) += mips/hpeldsp_init_mips.o
30	30	OBJS-$(CONFIG_BLOCKDSP) += mips/blockdsp_init_mips.o
31	31	OBJS-$(CONFIG_PIXBLOCKDSP) += mips/pixblockdsp_init_mips.o
	32	+OBJS-$(CONFIG_IDCTDSP) += mips/idctdsp_init_mips.o
32	33	OBJS-$(CONFIG_MPEGVIDEO) += mips/mpegvideo_init_mips.o
33	34	OBJS-$(CONFIG_MPEGVIDEOENC) += mips/mpegvideoencdsp_init_mips.o
34	35	OBJS-$(CONFIG_ME_CMP) += mips/me_cmp_init_mips.o
@@ -50,6 +51,8 @@ MSA-OBJS-$(CONFIG_QPELDSP) += mips/qpeldsp_msa.o · 15 unmodified lines
50	51	MSA-OBJS-$(CONFIG_HPELDSP) += mips/hpeldsp_msa.o
51	52	MSA-OBJS-$(CONFIG_BLOCKDSP) += mips/blockdsp_msa.o
52	53	MSA-OBJS-$(CONFIG_PIXBLOCKDSP) += mips/pixblockdsp_msa.o
	54	+MSA-OBJS-$(CONFIG_IDCTDSP) += mips/idctdsp_msa.o \
	55	+ mips/simple_idct_msa.o
53	56	MSA-OBJS-$(CONFIG_MPEGVIDEO) += mips/mpegvideo_msa.o
54	57	MSA-OBJS-$(CONFIG_MPEGVIDEOENC) += mips/mpegvideoencdsp_msa.o
55	58	MSA-OBJS-$(CONFIG_ME_CMP) += mips/me_cmp_msa.o

libavcodec/mips/idctdsp_init_mips.c+49−0

@@ -0,0 +1,49 @@
	1	+/*
	2	+ * Copyright (c) 2015 Manojkumar Bhosale (Manojkumar.Bhosale@imgtec.com)
	3	+ *
	4	+ * This file is part of FFmpeg.
	5	+ *
	6	+ * FFmpeg is free software; you can redistribute it and/or
	7	+ * modify it under the terms of the GNU Lesser General Public
	8	+ * License as published by the Free Software Foundation; either
	9	+ * version 2.1 of the License, or (at your option) any later version.
	10	+ *
	11	+ * FFmpeg is distributed in the hope that it will be useful,
	12	+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
	13	+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
	14	+ * Lesser General Public License for more details.
	15	+ *
	16	+ * You should have received a copy of the GNU Lesser General Public
	17	+ * License along with FFmpeg; if not, write to the Free Software
	18	+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
	19	+ */
	20	+
	21	+#include "idctdsp_mips.h"
	22	+
	23	+#if HAVE_MSA
	24	+static av_cold void idctdsp_init_msa(IDCTDSPContext c, AVCodecContext avctx,
	25	+ unsigned high_bit_depth)
	26	+{
	27	+ if ((avctx->lowres != 1) && (avctx->lowres != 2) && (avctx->lowres != 3) &&
	28	+ (avctx->bits_per_raw_sample != 10) &&
	29	+ (avctx->bits_per_raw_sample != 12) &&
	30	+ (avctx->idct_algo == FF_IDCT_AUTO)) {
	31	+ c->idct_put = ff_simple_idct_put_msa;
	32	+ c->idct_add = ff_simple_idct_add_msa;
	33	+ c->idct = ff_simple_idct_msa;
	34	+ c->perm_type = FF_IDCT_PERM_NONE;
	35	+ }
	36	+
	37	+ c->put_pixels_clamped = ff_put_pixels_clamped_msa;
	38	+ c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_msa;
	39	+ c->add_pixels_clamped = ff_add_pixels_clamped_msa;
	40	+}
	41	+#endif // #if HAVE_MSA
	42	+
	43	+void ff_idctdsp_init_mips(IDCTDSPContext c, AVCodecContext avctx,
	44	+ unsigned high_bit_depth)
	45	+{
	46	+#if HAVE_MSA
	47	+ idctdsp_init_msa(c, avctx, high_bit_depth);
	48	+#endif // #if HAVE_MSA
	49	+}

libavcodec/mips/idctdsp_mips.h+42−0

@@ -0,0 +1,42 @@
	1	+/*
	2	+ * Copyright (c) 2015 Manojkumar Bhosale (Manojkumar.Bhosale@imgtec.com)
	3	+ *
	4	+ * This file is part of FFmpeg.
	5	+ *
	6	+ * FFmpeg is free software; you can redistribute it and/or
	7	+ * modify it under the terms of the GNU Lesser General Public
	8	+ * License as published by the Free Software Foundation; either
	9	+ * version 2.1 of the License, or (at your option) any later version.
	10	+ *
	11	+ * FFmpeg is distributed in the hope that it will be useful,
	12	+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
	13	+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
	14	+ * Lesser General Public License for more details.
	15	+ *
	16	+ * You should have received a copy of the GNU Lesser General Public
	17	+ * License along with FFmpeg; if not, write to the Free Software
	18	+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
	19	+ */
	20	+
	21	+#ifndef AVCODEC_MIPS_IDCTDSP_MIPS_H
	22	+#define AVCODEC_MIPS_IDCTDSP_MIPS_H
	23	+
	24	+#include "../mpegvideo.h"
	25	+
	26	+void ff_put_pixels_clamped_msa(const int16_t *block,
	27	+ uint8_t *av_restrict pixels,
	28	+ ptrdiff_t line_size);
	29	+void ff_put_signed_pixels_clamped_msa(const int16_t *block,
	30	+ uint8_t *av_restrict pixels,
	31	+ ptrdiff_t line_size);
	32	+void ff_add_pixels_clamped_msa(const int16_t *block,
	33	+ uint8_t *av_restrict pixels,
	34	+ ptrdiff_t line_size);
	35	+void ff_j_rev_dct_msa(int16_t *data);
	36	+void ff_jref_idct_put_msa(uint8_t dest, int32_t stride, int16_t block);
	37	+void ff_jref_idct_add_msa(uint8_t dest, int32_t stride, int16_t block);
	38	+void ff_simple_idct_msa(int16_t *block);
	39	+void ff_simple_idct_put_msa(uint8_t dest, int32_t stride_dst, int16_t block);
	40	+void ff_simple_idct_add_msa(uint8_t dest, int32_t stride_dst, int16_t block);
	41	+
	42	+#endif // #ifndef AVCODEC_MIPS_IDCTDSP_MIPS_H

libavcodec/mips/idctdsp_msa.c+149−0

@@ -0,0 +1,149 @@
	1	+/*
	2	+ * Copyright (c) 2015 Manojkumar Bhosale (Manojkumar.Bhosale@imgtec.com)
	3	+ *
	4	+ * This file is part of FFmpeg.
	5	+ *
	6	+ * FFmpeg is free software; you can redistribute it and/or
	7	+ * modify it under the terms of the GNU Lesser General Public
	8	+ * License as published by the Free Software Foundation; either
	9	+ * version 2.1 of the License, or (at your option) any later version.
	10	+ *
	11	+ * FFmpeg is distributed in the hope that it will be useful,
	12	+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
	13	+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
	14	+ * Lesser General Public License for more details.
	15	+ *
	16	+ * You should have received a copy of the GNU Lesser General Public
	17	+ * License along with FFmpeg; if not, write to the Free Software
	18	+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
	19	+ */
	20	+
	21	+#include "libavutil/mips/generic_macros_msa.h"
	22	+#include "idctdsp_mips.h"
	23	+
	24	+static void put_pixels_clamped_msa(const int16_t block, uint8_t pixels,
	25	+ int32_t stride)
	26	+{
	27	+ uint64_t in0_d, in1_d, in2_d, in3_d, in4_d, in5_d, in6_d, in7_d;
	28	+ v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
	29	+
	30	+ LD_SH8(block, 8, in0, in1, in2, in3, in4, in5, in6, in7);
	31	+ CLIP_SH4_0_255(in0, in1, in2, in3);
	32	+ CLIP_SH4_0_255(in4, in5, in6, in7);
	33	+ PCKEV_B4_SH(in0, in0, in1, in1, in2, in2, in3, in3, in0, in1, in2, in3);
	34	+ PCKEV_B4_SH(in4, in4, in5, in5, in6, in6, in7, in7, in4, in5, in6, in7);
	35	+
	36	+ in0_d = __msa_copy_u_d((v2i64) in0, 0);
	37	+ in1_d = __msa_copy_u_d((v2i64) in1, 0);
	38	+ in2_d = __msa_copy_u_d((v2i64) in2, 0);
	39	+ in3_d = __msa_copy_u_d((v2i64) in3, 0);
	40	+ in4_d = __msa_copy_u_d((v2i64) in4, 0);
	41	+ in5_d = __msa_copy_u_d((v2i64) in5, 0);
	42	+ in6_d = __msa_copy_u_d((v2i64) in6, 0);
	43	+ in7_d = __msa_copy_u_d((v2i64) in7, 0);
	44	+ SD4(in0_d, in1_d, in2_d, in3_d, pixels, stride);
	45	+ pixels += 4 * stride;
	46	+ SD4(in4_d, in5_d, in6_d, in7_d, pixels, stride);
	47	+}
	48	+
	49	+static void put_signed_pixels_clamped_msa(const int16_t block, uint8_t pixels,
	50	+ int32_t stride)
	51	+{
	52	+ uint64_t in0_d, in1_d, in2_d, in3_d, in4_d, in5_d, in6_d, in7_d;
	53	+ v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
	54	+
	55	+ LD_SH8(block, 8, in0, in1, in2, in3, in4, in5, in6, in7);
	56	+
	57	+ in0 += 128;
	58	+ in1 += 128;
	59	+ in2 += 128;
	60	+ in3 += 128;
	61	+ in4 += 128;
	62	+ in5 += 128;
	63	+ in6 += 128;
	64	+ in7 += 128;
	65	+
	66	+ CLIP_SH4_0_255(in0, in1, in2, in3);
	67	+ CLIP_SH4_0_255(in4, in5, in6, in7);
	68	+ PCKEV_B4_SH(in0, in0, in1, in1, in2, in2, in3, in3, in0, in1, in2, in3);
	69	+ PCKEV_B4_SH(in4, in4, in5, in5, in6, in6, in7, in7, in4, in5, in6, in7);
	70	+
	71	+ in0_d = __msa_copy_u_d((v2i64) in0, 0);
	72	+ in1_d = __msa_copy_u_d((v2i64) in1, 0);
	73	+ in2_d = __msa_copy_u_d((v2i64) in2, 0);
	74	+ in3_d = __msa_copy_u_d((v2i64) in3, 0);
	75	+ in4_d = __msa_copy_u_d((v2i64) in4, 0);
	76	+ in5_d = __msa_copy_u_d((v2i64) in5, 0);
	77	+ in6_d = __msa_copy_u_d((v2i64) in6, 0);
	78	+ in7_d = __msa_copy_u_d((v2i64) in7, 0);
	79	+ SD4(in0_d, in1_d, in2_d, in3_d, pixels, stride);
	80	+ pixels += 4 * stride;
	81	+ SD4(in4_d, in5_d, in6_d, in7_d, pixels, stride);
	82	+}
	83	+
	84	+static void add_pixels_clamped_msa(const int16_t block, uint8_t pixels,
	85	+ int32_t stride)
	86	+{
	87	+ uint64_t in0_d, in1_d, in2_d, in3_d, in4_d, in5_d, in6_d, in7_d;
	88	+ v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
	89	+ v16u8 pix_in0, pix_in1, pix_in2, pix_in3;
	90	+ v16u8 pix_in4, pix_in5, pix_in6, pix_in7;
	91	+ v8u16 pix0, pix1, pix2, pix3, pix4, pix5, pix6, pix7;
	92	+ v8i16 zero = { 0 };
	93	+
	94	+ LD_SH8(block, 8, in0, in1, in2, in3, in4, in5, in6, in7);
	95	+ LD_UB8(pixels, stride, pix_in0, pix_in1, pix_in2,
	96	+ pix_in3, pix_in4, pix_in5, pix_in6, pix_in7);
	97	+
	98	+ ILVR_B4_UH(zero, pix_in0, zero, pix_in1, zero, pix_in2, zero, pix_in3,
	99	+ pix0, pix1, pix2, pix3);
	100	+ ILVR_B4_UH(zero, pix_in4, zero, pix_in5, zero, pix_in6, zero, pix_in7,
	101	+ pix4, pix5, pix6, pix7);
	102	+
	103	+ in0 += (v8i16) pix0;
	104	+ in1 += (v8i16) pix1;
	105	+ in2 += (v8i16) pix2;
	106	+ in3 += (v8i16) pix3;
	107	+ in4 += (v8i16) pix4;
	108	+ in5 += (v8i16) pix5;
	109	+ in6 += (v8i16) pix6;
	110	+ in7 += (v8i16) pix7;
	111	+
	112	+ CLIP_SH4_0_255(in0, in1, in2, in3);
	113	+ CLIP_SH4_0_255(in4, in5, in6, in7);
	114	+ PCKEV_B4_SH(in0, in0, in1, in1, in2, in2, in3, in3, in0, in1, in2, in3);
	115	+ PCKEV_B4_SH(in4, in4, in5, in5, in6, in6, in7, in7, in4, in5, in6, in7);
	116	+
	117	+ in0_d = __msa_copy_u_d((v2i64) in0, 0);
	118	+ in1_d = __msa_copy_u_d((v2i64) in1, 0);
	119	+ in2_d = __msa_copy_u_d((v2i64) in2, 0);
	120	+ in3_d = __msa_copy_u_d((v2i64) in3, 0);
	121	+ in4_d = __msa_copy_u_d((v2i64) in4, 0);
	122	+ in5_d = __msa_copy_u_d((v2i64) in5, 0);
	123	+ in6_d = __msa_copy_u_d((v2i64) in6, 0);
	124	+ in7_d = __msa_copy_u_d((v2i64) in7, 0);
	125	+ SD4(in0_d, in1_d, in2_d, in3_d, pixels, stride);
	126	+ pixels += 4 * stride;
	127	+ SD4(in4_d, in5_d, in6_d, in7_d, pixels, stride);
	128	+}
	129	+
	130	+void ff_put_pixels_clamped_msa(const int16_t *block,
	131	+ uint8_t *av_restrict pixels,
	132	+ ptrdiff_t line_size)
	133	+{
	134	+ put_pixels_clamped_msa(block, pixels, line_size);
	135	+}
	136	+
	137	+void ff_put_signed_pixels_clamped_msa(const int16_t *block,
	138	+ uint8_t *av_restrict pixels,
	139	+ ptrdiff_t line_size)
	140	+{
	141	+ put_signed_pixels_clamped_msa(block, pixels, line_size);
	142	+}
	143	+
	144	+void ff_add_pixels_clamped_msa(const int16_t *block,
	145	+ uint8_t *av_restrict pixels,
	146	+ ptrdiff_t line_size)
	147	+{
	148	+ add_pixels_clamped_msa(block, pixels, line_size);
	149	+}

libavcodec/mips/simple_idct_msa.c+573−0

@@ -0,0 +1,573 @@
	1	+/*
	2	+ * Copyright (c) 2015 Parag Salasakar (Parag.Salasakar@imgtec.com)
	3	+ *
	4	+ * This file is part of FFmpeg.
	5	+ *
	6	+ * FFmpeg is free software; you can redistribute it and/or
	7	+ * modify it under the terms of the GNU Lesser General Public
	8	+ * License as published by the Free Software Foundation; either
	9	+ * version 2.1 of the License, or (at your option) any later version.
	10	+ *
	11	+ * FFmpeg is distributed in the hope that it will be useful,
	12	+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
	13	+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
	14	+ * Lesser General Public License for more details.
	15	+ *
	16	+ * You should have received a copy of the GNU Lesser General Public
	17	+ * License along with FFmpeg; if not, write to the Free Software
	18	+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
	19	+ */
	20	+
	21	+#include "libavutil/mips/generic_macros_msa.h"
	22	+#include "idctdsp_mips.h"
	23	+
	24	+static void simple_idct_msa(int16_t *block)
	25	+{
	26	+ int32_t const_val;
	27	+ v8i16 weights = { 0, 22725, 21407, 19266, 16383, 12873, 8867, 4520 };
	28	+ v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
	29	+ v8i16 w1, w3, w5, w7;
	30	+ v8i16 const0, const1, const2, const3, const4, const5, const6, const7;
	31	+ v4i32 temp0_r, temp1_r, temp2_r, temp3_r;
	32	+ v4i32 temp0_l, temp1_l, temp2_l, temp3_l;
	33	+ v4i32 a0_r, a1_r, a2_r, a3_r, a0_l, a1_l, a2_l, a3_l;
	34	+ v4i32 b0_r, b1_r, b2_r, b3_r, b0_l, b1_l, b2_l, b3_l;
	35	+ v4i32 w2, w4, w6;
	36	+ v8i16 select_vec, temp;
	37	+ v8i16 zero = { 0 };
	38	+ v4i32 const_val0 = __msa_ldi_w(1);
	39	+ v4i32 const_val1 = __msa_ldi_w(1);
	40	+
	41	+ LD_SH8(block, 8, in0, in1, in2, in3, in4, in5, in6, in7);
	42	+ const_val0 <<= 10;
	43	+ const_val = 16383 * ((1 << 19) / 16383);
	44	+ const_val1 = __msa_insert_w(const_val0, 0, const_val);
	45	+ const_val1 = __msa_splati_w(const_val1, 0);
	46	+ TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
	47	+ in0, in1, in2, in3, in4, in5, in6, in7);
	48	+ select_vec = in1 \| in2 \| in3 \| in4 \| in5 \| in6 \| in7;
	49	+ select_vec = __msa_clti_u_h((v8u16) select_vec, 1);
	50	+ UNPCK_SH_SW(in0, a0_r, a0_l);
	51	+ UNPCK_SH_SW(in2, temp3_r, temp3_l);
	52	+ temp = in0 << 3;
	53	+ w2 = (v4i32) __msa_splati_h(weights, 2);
	54	+ w2 = (v4i32) __msa_ilvr_h(zero, (v8i16) w2);
	55	+ w4 = (v4i32) __msa_splati_h(weights, 4);
	56	+ w4 = (v4i32) __msa_ilvr_h(zero, (v8i16) w4);
	57	+ w6 = (v4i32) __msa_splati_h(weights, 6);
	58	+ w6 = (v4i32) __msa_ilvr_h(zero, (v8i16) w6);
	59	+ MUL2(a0_r, w4, a0_l, w4, a0_r, a0_l);
	60	+ ADD2(a0_r, const_val0, a0_l, const_val0, temp0_r, temp0_l);
	61	+ MUL4(w2, temp3_r, w2, temp3_l, w6, temp3_r, w6, temp3_l,
	62	+ temp1_r, temp1_l, temp2_r, temp2_l);
	63	+ BUTTERFLY_8(temp0_r, temp0_l, temp0_r, temp0_l,
	64	+ temp2_l, temp2_r, temp1_l, temp1_r,
	65	+ a0_r, a0_l, a1_r, a1_l, a2_l, a2_r, a3_l, a3_r);
	66	+ UNPCK_SH_SW(in4, temp0_r, temp0_l);
	67	+ UNPCK_SH_SW(in6, temp3_r, temp3_l);
	68	+ MUL2(temp0_r, w4, temp0_l, w4, temp0_r, temp0_l);
	69	+ MUL4(w2, temp3_r, w2, temp3_l, w6, temp3_r, w6, temp3_l,
	70	+ temp2_r, temp2_l, temp1_r, temp1_l);
	71	+ ADD2(a0_r, temp0_r, a0_l, temp0_l, a0_r, a0_l);
	72	+ SUB4(a1_r, temp0_r, a1_l, temp0_l, a2_r, temp0_r, a2_l, temp0_l,
	73	+ a1_r, a1_l, a2_r, a2_l);
	74	+ ADD4(a3_r, temp0_r, a3_l, temp0_l, a0_r, temp1_r, a0_l, temp1_l,
	75	+ a3_r, a3_l, a0_r, a0_l);
	76	+ SUB2(a1_r, temp2_r, a1_l, temp2_l, a1_r, a1_l);
	77	+ ADD2(a2_r, temp2_r, a2_l, temp2_l, a2_r, a2_l);
	78	+ SUB2(a3_r, temp1_r, a3_l, temp1_l, a3_r, a3_l);
	79	+ ILVRL_H2_SW(in1, in3, b3_r, b3_l);
	80	+ SPLATI_H4_SH(weights, 1, 3, 5, 7, w1, w3, w5, w7);
	81	+ ILVRL_H2_SW(in5, in7, temp0_r, temp0_l);
	82	+ ILVR_H4_SH(w1, w3, w3, -w7, w5, -w1, w7, -w5,
	83	+ const0, const1, const2, const3);
	84	+ ILVR_H2_SH(w5, w7, w7, w3, const4, const6);
	85	+ const5 = __msa_ilvod_h(-w1, -w5);
	86	+ const7 = __msa_ilvod_h(w3, -w1);
	87	+ DOTP_SH4_SW(b3_r, b3_r, b3_r, b3_r, const0, const1, const2, const3,
	88	+ b0_r, b1_r, b2_r, b3_r);
	89	+ DPADD_SH4_SW(temp0_r, temp0_r, temp0_r, temp0_r,
	90	+ const4, const5, const6, const7, b0_r, b1_r, b2_r, b3_r);
	91	+ DOTP_SH4_SW(b3_l, b3_l, b3_l, b3_l, const0, const1, const2, const3,
	92	+ b0_l, b1_l, b2_l, b3_l);
	93	+ DPADD_SH4_SW(temp0_l, temp0_l, temp0_l, temp0_l,
	94	+ const4, const5, const6, const7, b0_l, b1_l, b2_l, b3_l);
	95	+ BUTTERFLY_16(a0_r, a0_l, a1_r, a1_l, a2_r, a2_l, a3_r, a3_l,
	96	+ b3_l, b3_r, b2_l, b2_r, b1_l, b1_r, b0_l, b0_r,
	97	+ temp0_r, temp0_l, temp1_r, temp1_l,
	98	+ temp2_r, temp2_l, temp3_r, temp3_l,
	99	+ a3_l, a3_r, a2_l, a2_r, a1_l, a1_r, a0_l, a0_r);
	100	+ SRA_4V(temp0_r, temp0_l, temp1_r, temp1_l, 11);
	101	+ SRA_4V(temp2_r, temp2_l, temp3_r, temp3_l, 11);
	102	+ PCKEV_H4_SW(temp0_l, temp0_r, temp1_l, temp1_r,
	103	+ temp2_l, temp2_r, temp3_l, temp3_r,
	104	+ temp0_r, temp1_r, temp2_r, temp3_r);
	105	+ in0 = (v8i16) __msa_bmnz_v((v16u8) temp0_r, (v16u8) temp,
	106	+ (v16u8) select_vec);
	107	+ in1 = (v8i16) __msa_bmnz_v((v16u8) temp1_r, (v16u8) temp,
	108	+ (v16u8) select_vec);
	109	+ in2 = (v8i16) __msa_bmnz_v((v16u8) temp2_r, (v16u8) temp,
	110	+ (v16u8) select_vec);
	111	+ in3 = (v8i16) __msa_bmnz_v((v16u8) temp3_r, (v16u8) temp,
	112	+ (v16u8) select_vec);
	113	+ SRA_4V(a3_r, a3_l, a2_r, a2_l, 11);
	114	+ SRA_4V(a1_r, a1_l, a0_r, a0_l, 11);
	115	+ PCKEV_H4_SW(a0_l, a0_r, a1_l, a1_r, a2_l, a2_r, a3_l, a3_r,
	116	+ a0_r, a1_r, a2_r, a3_r);
	117	+ in4 = (v8i16) __msa_bmnz_v((v16u8) a3_r, (v16u8) temp, (v16u8) select_vec);
	118	+ in5 = (v8i16) __msa_bmnz_v((v16u8) a2_r, (v16u8) temp, (v16u8) select_vec);
	119	+ in6 = (v8i16) __msa_bmnz_v((v16u8) a1_r, (v16u8) temp, (v16u8) select_vec);
	120	+ in7 = (v8i16) __msa_bmnz_v((v16u8) a0_r, (v16u8) temp, (v16u8) select_vec);
	121	+ TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
	122	+ in0, in1, in2, in3, in4, in5, in6, in7);
	123	+
	124	+ UNPCK_SH_SW(in0, a0_r, a0_l);
	125	+ UNPCK_SH_SW(in2, temp3_r, temp3_l);
	126	+ w2 = (v4i32) __msa_splati_h(weights, 2);
	127	+ w2 = (v4i32) __msa_ilvr_h(zero, (v8i16) w2);
	128	+ w4 = (v4i32) __msa_splati_h(weights, 4);
	129	+ w4 = (v4i32) __msa_ilvr_h(zero, (v8i16) w4);
	130	+ w6 = (v4i32) __msa_splati_h(weights, 6);
	131	+ w6 = (v4i32) __msa_ilvr_h(zero, (v8i16) w6);
	132	+ MUL2(a0_r, w4, a0_l, w4, a0_r, a0_l);
	133	+ ADD2(a0_r, const_val1, a0_l, const_val1, temp0_r, temp0_l);
	134	+ MUL4(w2, temp3_r, w2, temp3_l, w6, temp3_r, w6, temp3_l,
	135	+ temp1_r, temp1_l, temp2_r, temp2_l);
	136	+ BUTTERFLY_8(temp0_r, temp0_l, temp0_r, temp0_l,
	137	+ temp2_l, temp2_r, temp1_l, temp1_r,
	138	+ a0_r, a0_l, a1_r, a1_l, a2_l, a2_r, a3_l, a3_r);
	139	+ UNPCK_SH_SW(in4, temp0_r, temp0_l);
	140	+ UNPCK_SH_SW(in6, temp3_r, temp3_l);
	141	+ MUL2(temp0_r, w4, temp0_l, w4, temp0_r, temp0_l);
	142	+ MUL4(w2, temp3_r, w2, temp3_l, w6, temp3_r, w6, temp3_l,
	143	+ temp2_r, temp2_l, temp1_r, temp1_l);
	144	+ ADD2(a0_r, temp0_r, a0_l, temp0_l, a0_r, a0_l);
	145	+ SUB4(a1_r, temp0_r, a1_l, temp0_l, a2_r, temp0_r, a2_l, temp0_l,
	146	+ a1_r, a1_l, a2_r, a2_l);
	147	+ ADD4(a3_r, temp0_r, a3_l, temp0_l, a0_r, temp1_r, a0_l, temp1_l,
	148	+ a3_r, a3_l, a0_r, a0_l);
	149	+ SUB2(a1_r, temp2_r, a1_l, temp2_l, a1_r, a1_l);
	150	+ ADD2(a2_r, temp2_r, a2_l, temp2_l, a2_r, a2_l);
	151	+ SUB2(a3_r, temp1_r, a3_l, temp1_l, a3_r, a3_l);
	152	+ ILVRL_H2_SW(in1, in3, b3_r, b3_l);
	153	+ SPLATI_H4_SH(weights, 1, 3, 5, 7, w1, w3, w5, w7);
	154	+ ILVR_H4_SH(w1, w3, w3, -w7, w5, -w1, w7, -w5,
	155	+ const0, const1, const2, const3);
	156	+ DOTP_SH4_SW(b3_r, b3_r, b3_r, b3_r, const0, const1, const2, const3,
	157	+ b0_r, b1_r, b2_r, b3_r);
	158	+ DOTP_SH4_SW(b3_l, b3_l, b3_l, b3_l, const0, const1, const2, const3,
	159	+ b0_l, b1_l, b2_l, b3_l);
	160	+ ILVRL_H2_SW(in5, in7, temp0_r, temp0_l);
	161	+ ILVR_H2_SH(w5, w7, w7, w3, const4, const6);
	162	+ const5 = __msa_ilvod_h(-w1, -w5);
	163	+ const7 = __msa_ilvod_h(w3, -w1);
	164	+ DPADD_SH4_SW(temp0_r, temp0_r, temp0_r, temp0_r,
	165	+ const4, const5, const6, const7, b0_r, b1_r, b2_r, b3_r);
	166	+ DPADD_SH4_SW(temp0_l, temp0_l, temp0_l, temp0_l,
	167	+ const4, const5, const6, const7, b0_l, b1_l, b2_l, b3_l);
	168	+ BUTTERFLY_16(a0_r, a0_l, a1_r, a1_l, a2_r, a2_l, a3_r, a3_l,
	169	+ b3_l, b3_r, b2_l, b2_r, b1_l, b1_r, b0_l, b0_r,
	170	+ temp0_r, temp0_l, temp1_r, temp1_l,
	171	+ temp2_r, temp2_l, temp3_r, temp3_l,
	172	+ a3_l, a3_r, a2_l, a2_r, a1_l, a1_r, a0_l, a0_r);
	173	+ SRA_4V(temp0_r, temp0_l, temp1_r, temp1_l, 20);
	174	+ SRA_4V(temp2_r, temp2_l, temp3_r, temp3_l, 20);
	175	+ PCKEV_H4_SW(temp0_l, temp0_r, temp1_l, temp1_r, temp2_l, temp2_r,
	176	+ temp3_l, temp3_r, temp0_r, temp1_r, temp2_r, temp3_r);
	177	+ SRA_4V(a3_r, a3_l, a2_r, a2_l, 20);
	178	+ SRA_4V(a1_r, a1_l, a0_r, a0_l, 20);
	179	+ PCKEV_H4_SW(a0_l, a0_r, a1_l, a1_r, a2_l, a2_r, a3_l, a3_r,
	180	+ a0_r, a1_r, a2_r, a3_r);
	181	+ ST_SW8(temp0_r, temp1_r, temp2_r, temp3_r, a3_r, a2_r, a1_r, a0_r,
	182	+ block, 8);
	183	+}
	184	+
	185	+static void simple_idct_put_msa(uint8_t *dst, int32_t dst_stride,
	186	+ int16_t *block)
	187	+{
	188	+ int32_t const_val;
	189	+ uint64_t tmp0, tmp1, tmp2, tmp3;
	190	+ v8i16 weights = { 0, 22725, 21407, 19266, 16383, 12873, 8867, 4520 };
	191	+ v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
	192	+ v8i16 w1, w3, w5, w7;
	193	+ v8i16 const0, const1, const2, const3, const4, const5, const6, const7;
	194	+ v4i32 temp0_r, temp1_r, temp2_r, temp3_r;
	195	+ v4i32 temp0_l, temp1_l, temp2_l, temp3_l;
	196	+ v4i32 a0_r, a1_r, a2_r, a3_r, a0_l, a1_l, a2_l, a3_l;
	197	+ v4i32 b0_r, b1_r, b2_r, b3_r, b0_l, b1_l, b2_l, b3_l;
	198	+ v4i32 w2, w4, w6;
	199	+ v8i16 select_vec, temp;
	200	+ v8i16 zero = { 0 };
	201	+ v4i32 const_val0 = __msa_ldi_w(1);
	202	+ v4i32 const_val1 = __msa_ldi_w(1);
	203	+
	204	+ LD_SH8(block, 8, in0, in1, in2, in3, in4, in5, in6, in7);
	205	+ const_val0 <<= 10;
	206	+ const_val = 16383 * ((1 << 19) / 16383);
	207	+ const_val1 = __msa_insert_w(const_val0, 0, const_val);
	208	+ const_val1 = __msa_splati_w(const_val1, 0);
	209	+ TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
	210	+ in0, in1, in2, in3, in4, in5, in6, in7);
	211	+ select_vec = in1 \| in2 \| in3 \| in4 \| in5 \| in6 \| in7;
	212	+ select_vec = __msa_clti_u_h((v8u16) select_vec, 1);
	213	+ UNPCK_SH_SW(in0, a0_r, a0_l);
	214	+ UNPCK_SH_SW(in2, temp3_r, temp3_l);
	215	+ temp = in0 << 3;
	216	+ w2 = (v4i32) __msa_splati_h(weights, 2);
	217	+ w2 = (v4i32) __msa_ilvr_h(zero, (v8i16) w2);
	218	+ w4 = (v4i32) __msa_splati_h(weights, 4);
	219	+ w4 = (v4i32) __msa_ilvr_h(zero, (v8i16) w4);
	220	+ w6 = (v4i32) __msa_splati_h(weights, 6);
	221	+ w6 = (v4i32) __msa_ilvr_h(zero, (v8i16) w6);
	222	+ MUL2(a0_r, w4, a0_l, w4, a0_r, a0_l);
	223	+ ADD2(a0_r, const_val0, a0_l, const_val0, temp0_r, temp0_l);
	224	+ MUL2(w2, temp3_r, w2, temp3_l, temp1_r, temp1_l);
	225	+ MUL2(w6, temp3_r, w6, temp3_l, temp2_r, temp2_l);
	226	+ BUTTERFLY_8(temp0_r, temp0_l, temp0_r, temp0_l,
	227	+ temp2_l, temp2_r, temp1_l, temp1_r,
	228	+ a0_r, a0_l, a1_r, a1_l, a2_l, a2_r, a3_l, a3_r);
	229	+ UNPCK_SH_SW(in4, temp0_r, temp0_l);
	230	+ UNPCK_SH_SW(in6, temp3_r, temp3_l);
	231	+ MUL2(temp0_r, w4, temp0_l, w4, temp0_r, temp0_l);
	232	+ MUL2(w2, temp3_r, w2, temp3_l, temp2_r, temp2_l);
	233	+ MUL2(w6, temp3_r, w6, temp3_l, temp1_r, temp1_l);
	234	+ ADD2(a0_r, temp0_r, a0_l, temp0_l, a0_r, a0_l);
	235	+ SUB2(a1_r, temp0_r, a1_l, temp0_l, a1_r, a1_l);
	236	+ SUB2(a2_r, temp0_r, a2_l, temp0_l, a2_r, a2_l);
	237	+ ADD2(a3_r, temp0_r, a3_l, temp0_l, a3_r, a3_l);
	238	+ ADD2(a0_r, temp1_r, a0_l, temp1_l, a0_r, a0_l);
	239	+ SUB2(a1_r, temp2_r, a1_l, temp2_l, a1_r, a1_l);
	240	+ ADD2(a2_r, temp2_r, a2_l, temp2_l, a2_r, a2_l);
	241	+ SUB2(a3_r, temp1_r, a3_l, temp1_l, a3_r, a3_l);
	242	+ ILVRL_H2_SW(in1, in3, b3_r, b3_l);
	243	+ SPLATI_H4_SH(weights, 1, 3, 5, 7, w1, w3, w5, w7);
	244	+ ILVRL_H2_SW(in5, in7, temp0_r, temp0_l);
	245	+ ILVR_H4_SH(w1, w3, w3, -w7, w5, -w1, w7, -w5,
	246	+ const0, const1, const2, const3);
	247	+ ILVR_H2_SH(w5, w7, w7, w3, const4, const6);
	248	+ const5 = __msa_ilvod_h(-w1, -w5);
	249	+ const7 = __msa_ilvod_h(w3, -w1);
	250	+ DOTP_SH4_SW(b3_r, b3_r, b3_r, b3_r, const0, const1, const2, const3,
	251	+ b0_r, b1_r, b2_r, b3_r);
	252	+ DPADD_SH4_SW(temp0_r, temp0_r, temp0_r, temp0_r,
	253	+ const4, const5, const6, const7, b0_r, b1_r, b2_r, b3_r);
	254	+ DOTP_SH4_SW(b3_l, b3_l, b3_l, b3_l, const0, const1, const2, const3,
	255	+ b0_l, b1_l, b2_l, b3_l);
	256	+ DPADD_SH4_SW(temp0_l, temp0_l, temp0_l, temp0_l,
	257	+ const4, const5, const6, const7, b0_l, b1_l, b2_l, b3_l);
	258	+ BUTTERFLY_16(a0_r, a0_l, a1_r, a1_l, a2_r, a2_l, a3_r, a3_l,
	259	+ b3_l, b3_r, b2_l, b2_r, b1_l, b1_r, b0_l, b0_r,
	260	+ temp0_r, temp0_l, temp1_r, temp1_l,
	261	+ temp2_r, temp2_l, temp3_r, temp3_l,
	262	+ a3_l, a3_r, a2_l, a2_r, a1_l, a1_r, a0_l, a0_r);
	263	+ SRA_4V(temp0_r, temp0_l, temp1_r, temp1_l, 11);
	264	+ SRA_4V(temp2_r, temp2_l, temp3_r, temp3_l, 11);
	265	+ PCKEV_H4_SW(temp0_l, temp0_r, temp1_l, temp1_r,
	266	+ temp2_l, temp2_r, temp3_l, temp3_r,
	267	+ temp0_r, temp1_r, temp2_r, temp3_r);
	268	+ in0 = (v8i16) __msa_bmnz_v((v16u8) temp0_r, (v16u8) temp,
	269	+ (v16u8) select_vec);
	270	+ in1 = (v8i16) __msa_bmnz_v((v16u8) temp1_r, (v16u8) temp,
	271	+ (v16u8) select_vec);
	272	+ in2 = (v8i16) __msa_bmnz_v((v16u8) temp2_r, (v16u8) temp,
	273	+ (v16u8) select_vec);
	274	+ in3 = (v8i16) __msa_bmnz_v((v16u8) temp3_r, (v16u8) temp,
	275	+ (v16u8) select_vec);
	276	+ SRA_4V(a3_r, a3_l, a2_r, a2_l, 11);
	277	+ SRA_4V(a1_r, a1_l, a0_r, a0_l, 11);
	278	+ PCKEV_H4_SW(a0_l, a0_r, a1_l, a1_r, a2_l, a2_r, a3_l, a3_r,
	279	+ a0_r, a1_r, a2_r, a3_r);
	280	+ in4 = (v8i16) __msa_bmnz_v((v16u8) a3_r, (v16u8) temp, (v16u8) select_vec);
	281	+ in5 = (v8i16) __msa_bmnz_v((v16u8) a2_r, (v16u8) temp, (v16u8) select_vec);
	282	+ in6 = (v8i16) __msa_bmnz_v((v16u8) a1_r, (v16u8) temp, (v16u8) select_vec);
	283	+ in7 = (v8i16) __msa_bmnz_v((v16u8) a0_r, (v16u8) temp, (v16u8) select_vec);
	284	+ TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
	285	+ in0, in1, in2, in3, in4, in5, in6, in7);
	286	+ UNPCK_SH_SW(in0, a0_r, a0_l);
	287	+ UNPCK_SH_SW(in2, temp3_r, temp3_l);
	288	+ w2 = (v4i32) __msa_splati_h(weights, 2);
	289	+ w2 = (v4i32) __msa_ilvr_h(zero, (v8i16) w2);
	290	+ w4 = (v4i32) __msa_splati_h(weights, 4);
	291	+ w4 = (v4i32) __msa_ilvr_h(zero, (v8i16) w4);
	292	+ w6 = (v4i32) __msa_splati_h(weights, 6);
	293	+ w6 = (v4i32) __msa_ilvr_h(zero, (v8i16) w6);
	294	+ MUL2(a0_r, w4, a0_l, w4, a0_r, a0_l);
	295	+ ADD2(a0_r, const_val1, a0_l, const_val1, temp0_r, temp0_l);
	296	+ MUL2(w2, temp3_r, w2, temp3_l, temp1_r, temp1_l);
	297	+ MUL2(w6, temp3_r, w6, temp3_l, temp2_r, temp2_l);
	298	+ BUTTERFLY_8(temp0_r, temp0_l, temp0_r, temp0_l,
	299	+ temp2_l, temp2_r, temp1_l, temp1_r,
	300	+ a0_r, a0_l, a1_r, a1_l, a2_l, a2_r, a3_l, a3_r);
	301	+ UNPCK_SH_SW(in4, temp0_r, temp0_l);
	302	+ UNPCK_SH_SW(in6, temp3_r, temp3_l);
	303	+ MUL2(temp0_r, w4, temp0_l, w4, temp0_r, temp0_l);
	304	+ MUL2(w2, temp3_r, w2, temp3_l, temp2_r, temp2_l);
	305	+ MUL2(w6, temp3_r, w6, temp3_l, temp1_r, temp1_l);
	306	+ ADD2(a0_r, temp0_r, a0_l, temp0_l, a0_r, a0_l);
	307	+ SUB2(a1_r, temp0_r, a1_l, temp0_l, a1_r, a1_l);
	308	+ SUB2(a2_r, temp0_r, a2_l, temp0_l, a2_r, a2_l);
	309	+ ADD2(a3_r, temp0_r, a3_l, temp0_l, a3_r, a3_l);
	310	+ ADD2(a0_r, temp1_r, a0_l, temp1_l, a0_r, a0_l);
	311	+ SUB2(a1_r, temp2_r, a1_l, temp2_l, a1_r, a1_l);
	312	+ ADD2(a2_r, temp2_r, a2_l, temp2_l, a2_r, a2_l);
	313	+ SUB2(a3_r, temp1_r, a3_l, temp1_l, a3_r, a3_l);
	314	+ ILVRL_H2_SW(in1, in3, b3_r, b3_l);
	315	+ SPLATI_H4_SH(weights, 1, 3, 5, 7, w1, w3, w5, w7);
	316	+ ILVR_H4_SH(w1, w3, w3, -w7, w5, -w1, w7, -w5,
	317	+ const0, const1, const2, const3);
	318	+ DOTP_SH4_SW(b3_r, b3_r, b3_r, b3_r, const0, const1, const2, const3,
	319	+ b0_r, b1_r, b2_r, b3_r);
	320	+ DOTP_SH4_SW(b3_l, b3_l, b3_l, b3_l, const0, const1, const2, const3,
	321	+ b0_l, b1_l, b2_l, b3_l);
	322	+ ILVRL_H2_SW(in5, in7, temp0_r, temp0_l);
	323	+ ILVR_H2_SH(w5, w7, w7, w3, const4, const6);
	324	+ const5 = __msa_ilvod_h(-w1, -w5);
	325	+ const7 = __msa_ilvod_h(w3, -w1);
	326	+ DPADD_SH4_SW(temp0_r, temp0_r, temp0_r, temp0_r,
	327	+ const4, const5, const6, const7, b0_r, b1_r, b2_r, b3_r);
	328	+ DPADD_SH4_SW(temp0_l, temp0_l, temp0_l, temp0_l,
	329	+ const4, const5, const6, const7, b0_l, b1_l, b2_l, b3_l);
	330	+ BUTTERFLY_16(a0_r, a0_l, a1_r, a1_l, a2_r, a2_l, a3_r, a3_l,
	331	+ b3_l, b3_r, b2_l, b2_r, b1_l, b1_r, b0_l, b0_r,
	332	+ temp0_r, temp0_l, temp1_r, temp1_l,
	333	+ temp2_r, temp2_l, temp3_r, temp3_l,
	334	+ a3_l, a3_r, a2_l, a2_r, a1_l, a1_r, a0_l, a0_r);
	335	+ SRA_4V(temp0_r, temp0_l, temp1_r, temp1_l, 20);
	336	+ SRA_4V(temp2_r, temp2_l, temp3_r, temp3_l, 20);
	337	+ SRA_4V(a3_r, a3_l, a2_r, a2_l, 20);
	338	+ SRA_4V(a1_r, a1_l, a0_r, a0_l, 20);
	339	+ PCKEV_H4_SW(temp0_l, temp0_r, temp1_l, temp1_r, temp2_l, temp2_r,
	340	+ temp3_l, temp3_r, temp0_r, temp1_r, temp2_r, temp3_r);
	341	+ PCKEV_H4_SW(a0_l, a0_r, a1_l, a1_r, a2_l, a2_r, a3_l, a3_r,
	342	+ a0_r, a1_r, a2_r, a3_r);
	343	+ temp0_r = (v4i32) CLIP_SH_0_255(temp0_r);
	344	+ temp1_r = (v4i32) CLIP_SH_0_255(temp1_r);
	345	+ temp2_r = (v4i32) CLIP_SH_0_255(temp2_r);
	346	+ temp3_r = (v4i32) CLIP_SH_0_255(temp3_r);
	347	+ PCKEV_B4_SW(temp0_r, temp0_r, temp1_r, temp1_r,
	348	+ temp2_r, temp2_r, temp3_r, temp3_r,
	349	+ temp0_r, temp1_r, temp2_r, temp3_r);
	350	+ tmp0 = __msa_copy_u_d((v2i64) temp0_r, 1);
	351	+ tmp1 = __msa_copy_u_d((v2i64) temp1_r, 1);
	352	+ tmp2 = __msa_copy_u_d((v2i64) temp2_r, 1);
	353	+ tmp3 = __msa_copy_u_d((v2i64) temp3_r, 1);
	354	+ SD4(tmp0, tmp1, tmp2, tmp3, dst, dst_stride);
	355	+ dst += 4 * dst_stride;
	356	+ a0_r = (v4i32) CLIP_SH_0_255(a0_r);
	357	+ a1_r = (v4i32) CLIP_SH_0_255(a1_r);
	358	+ a2_r = (v4i32) CLIP_SH_0_255(a2_r);
	359	+ a3_r = (v4i32) CLIP_SH_0_255(a3_r);
	360	+ PCKEV_B4_SW(a0_r, a0_r, a1_r, a1_r,
	361	+ a2_r, a2_r, a3_r, a3_r, a0_r, a1_r, a2_r, a3_r);
	362	+ tmp3 = __msa_copy_u_d((v2i64) a0_r, 1);
	363	+ tmp2 = __msa_copy_u_d((v2i64) a1_r, 1);
	364	+ tmp1 = __msa_copy_u_d((v2i64) a2_r, 1);
	365	+ tmp0 = __msa_copy_u_d((v2i64) a3_r, 1);
	366	+ SD4(tmp0, tmp1, tmp2, tmp3, dst, dst_stride);
	367	+ dst += 4 * dst_stride;
	368	+}
	369	+
	370	+static void simple_idct_add_msa(uint8_t *dst, int32_t dst_stride,
	371	+ int16_t *block)
	372	+{
	373	+ int32_t const_val;
	374	+ uint64_t tmp0, tmp1, tmp2, tmp3;
	375	+ v8i16 weights = { 0, 22725, 21407, 19266, 16383, 12873, 8867, 4520 };
	376	+ v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
	377	+ v8i16 w1, w3, w5, w7;
	378	+ v8i16 const0, const1, const2, const3, const4, const5, const6, const7;
	379	+ v4i32 temp0_r, temp1_r, temp2_r, temp3_r;
	380	+ v4i32 temp4_r, temp5_r, temp6_r, temp7_r, temp8_r;
	381	+ v4i32 temp0_l, temp1_l, temp2_l, temp3_l;
	382	+ v4i32 temp4_l, temp5_l, temp6_l, temp7_l, temp8_l;
	383	+ v4i32 a0_r, a1_r, a2_r, a3_r, a0_l, a1_l, a2_l, a3_l;
	384	+ v4i32 b0_r, b1_r, b2_r, b3_r, b0_l, b1_l, b2_l, b3_l;
	385	+ v4i32 w2, w4, w6;
	386	+ v8i16 select_vec, temp;
	387	+ v8i16 zero = { 0 };
	388	+ v4i32 const_val0 = __msa_ldi_w(1);
	389	+ v4i32 const_val1 = __msa_ldi_w(1);
	390	+
	391	+ const_val0 <<= 10;
	392	+ const_val = 16383 * ((1 << 19) / 16383);
	393	+ const_val1 = __msa_insert_w(const_val0, 0, const_val);
	394	+ const_val1 = __msa_splati_w(const_val1, 0);
	395	+ LD_SH8(block, 8, in0, in1, in2, in3, in4, in5, in6, in7);
	396	+ TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
	397	+ in0, in1, in2, in3, in4, in5, in6, in7);
	398	+
	399	+ select_vec = in1 \| in2 \| in3 \| in4 \| in5 \| in6 \| in7;
	400	+ select_vec = __msa_clti_u_h((v8u16) select_vec, 1);
	401	+ UNPCK_SH_SW(in0, a0_r, a0_l);
	402	+ UNPCK_SH_SW(in2, temp3_r, temp3_l);
	403	+ ILVRL_H2_SW(in1, in3, b3_r, b3_l);
	404	+ UNPCK_SH_SW(in4, temp4_r, temp4_l);
	405	+ UNPCK_SH_SW(in6, temp7_r, temp7_l);
	406	+ ILVRL_H2_SW(in5, in7, temp8_r, temp8_l);
	407	+ temp = in0 << 3;
	408	+ SPLATI_H4_SH(weights, 1, 3, 5, 7, w1, w3, w5, w7);
	409	+ ILVR_H4_SH(w1, w3, w3, -w7, w5, -w1, w7, -w5,
	410	+ const0, const1, const2, const3);
	411	+ ILVR_H2_SH(w5, w7, w7, w3, const4, const6);
	412	+ const5 = __msa_ilvod_h(-w1, -w5);
	413	+ const7 = __msa_ilvod_h(w3, -w1);
	414	+ DOTP_SH4_SW(b3_r, b3_r, b3_r, b3_r, const0, const1, const2, const3,
	415	+ b0_r, b1_r, b2_r, b3_r);
	416	+ DPADD_SH4_SW(temp8_r, temp8_r, temp8_r, temp8_r,
	417	+ const4, const5, const6, const7, b0_r, b1_r, b2_r, b3_r);
	418	+ DOTP_SH4_SW(b3_l, b3_l, b3_l, b3_l, const0, const1, const2, const3,
	419	+ b0_l, b1_l, b2_l, b3_l);
	420	+ DPADD_SH4_SW(temp8_l, temp8_l, temp8_l, temp8_l,
	421	+ const4, const5, const6, const7, b0_l, b1_l, b2_l, b3_l);
	422	+ w2 = (v4i32) __msa_splati_h(weights, 2);
	423	+ w2 = (v4i32) __msa_ilvr_h(zero, (v8i16) w2);
	424	+ w4 = (v4i32) __msa_splati_h(weights, 4);
	425	+ w4 = (v4i32) __msa_ilvr_h(zero, (v8i16) w4);
	426	+ w6 = (v4i32) __msa_splati_h(weights, 6);
	427	+ w6 = (v4i32) __msa_ilvr_h(zero, (v8i16) w6);
	428	+ MUL2(a0_r, w4, a0_l, w4, a0_r, a0_l);
	429	+ ADD2(a0_r, const_val0, a0_l, const_val0, temp0_r, temp0_l);
	430	+ MUL2(w2, temp3_r, w2, temp3_l, temp1_r, temp1_l);
	431	+ MUL2(w6, temp3_r, w6, temp3_l, temp2_r, temp2_l);
	432	+ BUTTERFLY_8(temp0_r, temp0_l, temp0_r, temp0_l,
	433	+ temp2_l, temp2_r, temp1_l, temp1_r,
	434	+ a0_r, a0_l, a1_r, a1_l, a2_l, a2_r, a3_l, a3_r);
	435	+ MUL2(temp4_r, w4, temp4_l, w4, temp4_r, temp4_l);
	436	+ MUL2(temp7_r, w2, temp7_l, w2, temp6_r, temp6_l);
	437	+ MUL2(temp7_r, w6, temp7_l, w6, temp5_r, temp5_l);
	438	+ ADD2(a0_r, temp4_r, a0_l, temp4_l, a0_r, a0_l);
	439	+ SUB2(a1_r, temp4_r, a1_l, temp4_l, a1_r, a1_l);
	440	+ SUB2(a2_r, temp4_r, a2_l, temp4_l, a2_r, a2_l);
	441	+ ADD2(a3_r, temp4_r, a3_l, temp4_l, a3_r, a3_l);
	442	+ ADD2(a0_r, temp5_r, a0_l, temp5_l, a0_r, a0_l);
	443	+ SUB2(a1_r, temp6_r, a1_l, temp6_l, a1_r, a1_l);
	444	+ ADD2(a2_r, temp6_r, a2_l, temp6_l, a2_r, a2_l);
	445	+ SUB2(a3_r, temp5_r, a3_l, temp5_l, a3_r, a3_l);
	446	+ BUTTERFLY_16(a0_r, a0_l, a1_r, a1_l, a2_r, a2_l, a3_r, a3_l,
	447	+ b3_l, b3_r, b2_l, b2_r, b1_l, b1_r, b0_l, b0_r,
	448	+ temp0_r, temp0_l, temp1_r, temp1_l,
	449	+ temp2_r, temp2_l, temp3_r, temp3_l,
	450	+ a3_l, a3_r, a2_l, a2_r, a1_l, a1_r, a0_l, a0_r);
	451	+ SRA_4V(temp0_r, temp0_l, temp1_r, temp1_l, 11);
	452	+ SRA_4V(temp2_r, temp2_l, temp3_r, temp3_l, 11);
	453	+ PCKEV_H4_SW(temp0_l, temp0_r, temp1_l, temp1_r,
	454	+ temp2_l, temp2_r, temp3_l, temp3_r,
	455	+ temp0_r, temp1_r, temp2_r, temp3_r);
	456	+ in0 = (v8i16) __msa_bmnz_v((v16u8) temp0_r, (v16u8) temp,
	457	+ (v16u8) select_vec);
	458	+ in1 = (v8i16) __msa_bmnz_v((v16u8) temp1_r, (v16u8) temp,
	459	+ (v16u8) select_vec);
	460	+ in2 = (v8i16) __msa_bmnz_v((v16u8) temp2_r, (v16u8) temp,
	461	+ (v16u8) select_vec);
	462	+ in3 = (v8i16) __msa_bmnz_v((v16u8) temp3_r, (v16u8) temp,
	463	+ (v16u8) select_vec);
	464	+ SRA_4V(a3_r, a3_l, a2_r, a2_l, 11);
	465	+ SRA_4V(a1_r, a1_l, a0_r, a0_l, 11);
	466	+ PCKEV_H4_SW(a0_l, a0_r, a1_l, a1_r, a2_l, a2_r, a3_l, a3_r,
	467	+ a0_r, a1_r, a2_r, a3_r);
	468	+ in4 = (v8i16) __msa_bmnz_v((v16u8) a3_r, (v16u8) temp, (v16u8) select_vec);
	469	+ in5 = (v8i16) __msa_bmnz_v((v16u8) a2_r, (v16u8) temp, (v16u8) select_vec);
	470	+ in6 = (v8i16) __msa_bmnz_v((v16u8) a1_r, (v16u8) temp, (v16u8) select_vec);
	471	+ in7 = (v8i16) __msa_bmnz_v((v16u8) a0_r, (v16u8) temp, (v16u8) select_vec);
	472	+ TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
	473	+ in0, in1, in2, in3, in4, in5, in6, in7);
	474	+
	475	+ UNPCK_SH_SW(in0, a0_r, a0_l);
	476	+ UNPCK_SH_SW(in2, temp3_r, temp3_l);
	477	+ MUL2(a0_r, w4, a0_l, w4, a0_r, a0_l);
	478	+ ADD2(a0_r, const_val1, a0_l, const_val1, temp0_r, temp0_l);
	479	+ MUL2(w2, temp3_r, w2, temp3_l, temp1_r, temp1_l);
	480	+ MUL2(w6, temp3_r, w6, temp3_l, temp2_r, temp2_l);
	481	+ BUTTERFLY_8(temp0_r, temp0_l, temp0_r, temp0_l,
	482	+ temp2_l, temp2_r, temp1_l, temp1_r,
	483	+ a0_r, a0_l, a1_r, a1_l, a2_l, a2_r, a3_l, a3_r);
	484	+ UNPCK_SH_SW(in4, temp0_r, temp0_l);
	485	+ UNPCK_SH_SW(in6, temp3_r, temp3_l);
	486	+ MUL2(temp0_r, w4, temp0_l, w4, temp0_r, temp0_l);
	487	+ MUL2(w2, temp3_r, w2, temp3_l, temp2_r, temp2_l);
	488	+ MUL2(w6, temp3_r, w6, temp3_l, temp1_r, temp1_l);
	489	+ ADD2(a0_r, temp0_r, a0_l, temp0_l, a0_r, a0_l);
	490	+ SUB2(a1_r, temp0_r, a1_l, temp0_l, a1_r, a1_l);
	491	+ SUB2(a2_r, temp0_r, a2_l, temp0_l, a2_r, a2_l);
	492	+ ADD2(a3_r, temp0_r, a3_l, temp0_l, a3_r, a3_l);
	493	+ ADD2(a0_r, temp1_r, a0_l, temp1_l, a0_r, a0_l);
	494	+ SUB2(a1_r, temp2_r, a1_l, temp2_l, a1_r, a1_l);
	495	+ ADD2(a2_r, temp2_r, a2_l, temp2_l, a2_r, a2_l);
	496	+ SUB2(a3_r, temp1_r, a3_l, temp1_l, a3_r, a3_l);
	497	+ ILVRL_H2_SW(in1, in3, b3_r, b3_l);
	498	+ ILVRL_H2_SW(in5, in7, temp0_r, temp0_l);
	499	+ DOTP_SH4_SW(b3_r, b3_r, b3_r, b3_r, const0, const1, const2, const3,
	500	+ b0_r, b1_r, b2_r, b3_r);
	501	+ DOTP_SH4_SW(b3_l, b3_l, b3_l, b3_l, const0, const1, const2, const3,
	502	+ b0_l, b1_l, b2_l, b3_l);
	503	+ DPADD_SH4_SW(temp0_r, temp0_r, temp0_r, temp0_r,
	504	+ const4, const5, const6, const7, b0_r, b1_r, b2_r, b3_r);
	505	+ DPADD_SH4_SW(temp0_l, temp0_l, temp0_l, temp0_l,
	506	+ const4, const5, const6, const7, b0_l, b1_l, b2_l, b3_l);
	507	+ BUTTERFLY_16(a0_r, a0_l, a1_r, a1_l, a2_r, a2_l, a3_r, a3_l,
	508	+ b3_l, b3_r, b2_l, b2_r, b1_l, b1_r, b0_l, b0_r,
	509	+ temp0_r, temp0_l, temp1_r, temp1_l,
	510	+ temp2_r, temp2_l, temp3_r, temp3_l,
	511	+ a3_l, a3_r, a2_l, a2_r, a1_l, a1_r, a0_l, a0_r);
	512	+ SRA_4V(temp0_r, temp0_l, temp1_r, temp1_l, 20);
	513	+ SRA_4V(temp2_r, temp2_l, temp3_r, temp3_l, 20);
	514	+ LD_SH4(dst, dst_stride, in0, in1, in2, in3);
	515	+ PCKEV_H4_SW(temp0_l, temp0_r, temp1_l, temp1_r, temp2_l, temp2_r,
	516	+ temp3_l, temp3_r, temp0_r, temp1_r, temp2_r, temp3_r);
	517	+ ILVR_B4_SW(zero, in0, zero, in1, zero, in2, zero, in3,
	518	+ temp0_l, temp1_l, temp2_l, temp3_l);
	519	+ temp0_r = (v4i32) ((v8i16) (temp0_r) + (v8i16) (temp0_l));
	520	+ temp1_r = (v4i32) ((v8i16) (temp1_r) + (v8i16) (temp1_l));
	521	+ temp2_r = (v4i32) ((v8i16) (temp2_r) + (v8i16) (temp2_l));
	522	+ temp3_r = (v4i32) ((v8i16) (temp3_r) + (v8i16) (temp3_l));
	523	+ temp0_r = (v4i32) CLIP_SH_0_255(temp0_r);
	524	+ temp1_r = (v4i32) CLIP_SH_0_255(temp1_r);
	525	+ temp2_r = (v4i32) CLIP_SH_0_255(temp2_r);
	526	+ temp3_r = (v4i32) CLIP_SH_0_255(temp3_r);
	527	+ PCKEV_B4_SW(temp0_r, temp0_r, temp1_r, temp1_r,
	528	+ temp2_r, temp2_r, temp3_r, temp3_r,
	529	+ temp0_r, temp1_r, temp2_r, temp3_r);
	530	+ tmp0 = __msa_copy_u_d((v2i64) temp0_r, 1);
	531	+ tmp1 = __msa_copy_u_d((v2i64) temp1_r, 1);
	532	+ tmp2 = __msa_copy_u_d((v2i64) temp2_r, 1);
	533	+ tmp3 = __msa_copy_u_d((v2i64) temp3_r, 1);
	534	+ SD4(tmp0, tmp1, tmp2, tmp3, dst, dst_stride);
	535	+
	536	+ SRA_4V(a3_r, a3_l, a2_r, a2_l, 20);
	537	+ SRA_4V(a1_r, a1_l, a0_r, a0_l, 20);
	538	+ LD_SH4(dst + 4 * dst_stride, dst_stride, in4, in5, in6, in7);
	539	+ PCKEV_H4_SW(a0_l, a0_r, a1_l, a1_r, a2_l, a2_r, a3_l, a3_r,
	540	+ a0_r, a1_r, a2_r, a3_r);
	541	+ ILVR_B4_SW(zero, in4, zero, in5, zero, in6, zero, in7,
	542	+ a3_l, a2_l, a1_l, a0_l);
	543	+ a3_r = (v4i32) ((v8i16) (a3_r) + (v8i16) (a3_l));
	544	+ a2_r = (v4i32) ((v8i16) (a2_r) + (v8i16) (a2_l));
	545	+ a1_r = (v4i32) ((v8i16) (a1_r) + (v8i16) (a1_l));
	546	+ a0_r = (v4i32) ((v8i16) (a0_r) + (v8i16) (a0_l));
	547	+ a3_r = (v4i32) CLIP_SH_0_255(a3_r);
	548	+ a2_r = (v4i32) CLIP_SH_0_255(a2_r);
	549	+ a1_r = (v4i32) CLIP_SH_0_255(a1_r);
	550	+ a0_r = (v4i32) CLIP_SH_0_255(a0_r);
	551	+ PCKEV_B4_SW(a0_r, a0_r, a1_r, a1_r,
	552	+ a2_r, a2_r, a3_r, a3_r, a0_r, a1_r, a2_r, a3_r);
	553	+ tmp0 = __msa_copy_u_d((v2i64) a3_r, 1);
	554	+ tmp1 = __msa_copy_u_d((v2i64) a2_r, 1);
	555	+ tmp2 = __msa_copy_u_d((v2i64) a1_r, 1);
	556	+ tmp3 = __msa_copy_u_d((v2i64) a0_r, 1);
	557	+ SD4(tmp0, tmp1, tmp2, tmp3, dst + 4 * dst_stride, dst_stride);
	558	+}
	559	+
	560	+void ff_simple_idct_msa(int16_t *block)
	561	+{
	562	+ simple_idct_msa(block);
	563	+}
	564	+
	565	+void ff_simple_idct_put_msa(uint8_t dst, int32_t dst_stride, int16_t block)
	566	+{
	567	+ simple_idct_put_msa(dst, dst_stride, block);
	568	+}
	569	+
	570	+void ff_simple_idct_add_msa(uint8_t dst, int32_t dst_stride, int16_t block)
	571	+{
	572	+ simple_idct_add_msa(dst, dst_stride, block);
	573	+}

libavutil/mips/generic_macros_msa.h+37−0

@@ -507,6 +507,14 @@ · 506 unmodified lines
507	507	ST_SW(in0, (pdst)); \
508	508	ST_SW(in1, (pdst) + stride); \
509	509	}
	510	+#define ST_SW8(in0, in1, in2, in3, in4, in5, in6, in7, \
	511	+ pdst, stride) \
	512	+{ \
	513	+ ST_SW2(in0, in1, (pdst), stride); \
	514	+ ST_SW2(in2, in3, (pdst) + 2 * stride, stride); \
	515	+ ST_SW2(in4, in5, (pdst) + 4 * stride, stride); \
	516	+ ST_SW2(in6, in7, (pdst) + 6 * stride, stride); \
	517	+}
510	518
511	519	/* Description : Store as 2x4 byte block to destination memory from input vector
512	520	Arguments : Inputs - in, stidx, pdst, stride
@@ -2382,6 +2390,35 @@ · 1869 unmodified lines
2382	2390	out7 = in0 - in7; \
2383	2391	}
2384	2392
	2393	+/* Description : Butterfly of 16 input vectors
	2394	+ Arguments : Inputs - in0 ... in15
	2395	+ Outputs - out0 .. out15
	2396	+ Details : Butterfly operation
	2397	+*/
	2398	+#define BUTTERFLY_16(in0, in1, in2, in3, in4, in5, in6, in7, \
	2399	+ in8, in9, in10, in11, in12, in13, in14, in15, \
	2400	+ out0, out1, out2, out3, out4, out5, out6, out7, \
	2401	+ out8, out9, out10, out11, out12, out13, out14, out15) \
	2402	+{ \
	2403	+ out0 = in0 + in15; \
	2404	+ out1 = in1 + in14; \
	2405	+ out2 = in2 + in13; \
	2406	+ out3 = in3 + in12; \
	2407	+ out4 = in4 + in11; \
	2408	+ out5 = in5 + in10; \
	2409	+ out6 = in6 + in9; \
	2410	+ out7 = in7 + in8; \
	2411	+ \
	2412	+ out8 = in7 - in8; \
	2413	+ out9 = in6 - in9; \
	2414	+ out10 = in5 - in10; \
	2415	+ out11 = in4 - in11; \
	2416	+ out12 = in3 - in12; \
	2417	+ out13 = in2 - in13; \
	2418	+ out14 = in1 - in14; \
	2419	+ out15 = in0 - in15; \
	2420	+}
	2421	+
2385	2422	/* Description : Transposes input 4x4 byte block
2386	2423	Arguments : Inputs - in0, in1, in2, in3 (input 4x4 byte block)
2387	2424	Outputs - out0, out1, out2, out3 (output 4x4 byte block)

avcodec/mips: MSA (MIPS-SIMD-Arch) optimizations for idctdsp functions

Keyboard shortcuts

General

Issues and pull requests

Issues, pull requests and Inbox

Repository code