/******************************************************************************
* Copyright © 2018, VideoLAN and dav1d authors
* Copyright © 2020, Martin Storsjo
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
*
* 1. Redistributions of source code must retain the above copyright notice, this
* list of conditions and the following disclaimer.
*
* 2. Redistributions in binary form must reproduce the above copyright notice,
* this list of conditions and the following disclaimer in the documentation
* and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
* WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
* ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
* (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
* LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
* ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
* SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*****************************************************************************/
#include "src/arm/asm.S"
#include "util.S"
// The exported functions in this file have got the following signature:
// void itxfm_add(pixel *dst, ptrdiff_t dst_stride, coef *coeff, int eob,
// int bitdepth_max);
// Most of the functions use the following register layout:
// x0-x3 external parameters
// x4 function pointer to first transform
// x5 function pointer to second transform
// x6 output parameter for helper function
// x7 input parameter for helper function
// x8 input stride for helper function
// x9-x12 scratch variables for helper functions
// x13 pointer to list of eob thresholds
// x14 return pointer for helper function
// x15 return pointer for main function
// The SIMD registers most often use the following layout:
// v0-v1 multiplication coefficients
// v2-v7 scratch registers
// v8-v15 unused
// v16-v31 inputs/outputs of transforms
const idct_coeffs, align =4
// idct4
.int 2896, 2896*8*(1<<16), 1567, 3784
// idct8
.int 799, 4017, 3406, 2276
// idct16
.int 401, 4076, 3166, 2598
.int 1931, 3612, 3920, 1189
// idct32
.int 201, 4091, 3035, 2751
.int 1751, 3703, 3857, 1380
.int 995, 3973, 3513, 2106
.int 2440, 3290, 4052, 601
endconst
const idct64_coeffs, align =4
.int 101*8*(1<<16), 4095*8*(1<<16), 2967*8*(1<<16), -2824*8*(1<<16)
.int 1660*8*(1<<16), 3745*8*(1<<16), 3822*8*(1<<16), -1474*8*(1<<16)
.int 4076, 401, 4017, 799
.int 4036*8*(1<<16), -700*8*(1<<16), 2359*8*(1<<16), 3349*8*(1<<16)
.int 3461*8*(1<<16), -2191*8*(1<<16), 897*8*(1<<16), 3996*8*(1<<16)
.int -3166, -2598, -799, -4017
.int 501*8*(1<<16), 4065*8*(1<<16), 3229*8*(1<<16), -2520*8*(1<<16)
.int 2019*8*(1<<16), 3564*8*(1<<16), 3948*8*(1<<16), -1092*8*(1<<16)
.int 3612, 1931, 2276, 3406
.int 4085*8*(1<<16), -301*8*(1<<16), 2675*8*(1<<16), 3102*8*(1<<16)
.int 3659*8*(1<<16), -1842*8*(1<<16), 1285*8*(1<<16), 3889*8*(1<<16)
.int -3920, -1189, -3406, -2276
endconst
const iadst4_coeffs, align =4
.int 1321, 3803, 2482, 3344
endconst
const iadst8_coeffs, align =4
.int 4076, 401, 3612, 1931
.int 2598, 3166, 1189, 3920
// idct_coeffs
.int 2896, 0, 1567, 3784
endconst
const iadst16_coeffs, align =4
.int 4091, 201, 3973, 995
.int 3703, 1751, 3290, 2440
.int 2751, 3035, 2106, 3513
.int 1380, 3857, 601, 4052
endconst
.macro mul_mla d, s0, s1, c0, c1
mul \d\().4s, \s0\().4s, \c0
mla \d\().4s, \s1\().4s, \c1
.endm
.macro mul_mls d, s0, s1, c0, c1
mul \d\().4s, \s0\().4s, \c0
mls \d\().4s, \s1\().4s, \c1
.endm
.macro scale_input sz, c, r0, r1, r2 r3, r4, r5, r6, r7
sqrdmulh \r0\sz, \r0\sz, \c
sqrdmulh \r1\sz, \r1\sz, \c
sqrdmulh \r2\sz, \r2\sz, \c
sqrdmulh \r3\sz, \r3\sz, \c
.ifnb \r4
sqrdmulh \r4\sz, \r4\sz, \c
sqrdmulh \r5\sz, \r5\sz, \c
sqrdmulh \r6\sz, \r6\sz, \c
sqrdmulh \r7\sz, \r7\sz, \c
.endif
.endm
.macro smin_4s r0, r1, r2
smin \r0\().4s, \r1\().4s, \r2\().4s
.endm
.macro smax_4s r0, r1, r2
smax \r0\().4s, \r1\().4s, \r2\().4s
.endm
.macro load_add_store load, shift, addsrc, adddst, min, store, dst, src, shiftbits=4
.ifnb \load
ld1 {\load}, [\src], x1
.endif
.ifnb \shift
srshr \shift, \shift, #\shiftbits
.endif
.ifnb \addsrc
usqadd \adddst, \addsrc
.endif
.ifnb \min
smin \min, \min, v7.8h
.endif
.ifnb \store
st1 {\store}, [\dst], x1
.endif
.endm
.macro load_add_store_8x16 dst, src
mov \src, \dst
mvni v7.8h, #0xfc, lsl #8 // 0x3ff
load_add_store v2.8h, v16.8h, , , , , \dst, \src
load_add_store v3.8h, v17.8h, , , , , \dst, \src
load_add_store v4.8h, v18.8h, v16.8h, v2.8h, , , \dst, \src
load_add_store v5.8h, v19.8h, v17.8h, v3.8h, v2.8h, , \dst, \src
load_add_store v16.8h, v20.8h, v18.8h, v4.8h, v3.8h, v2.8h, \dst, \src
load_add_store v17.8h, v21.8h, v19.8h, v5.8h, v4.8h, v3.8h, \dst, \src
load_add_store v18.8h, v22.8h, v20.8h, v16.8h, v5.8h, v4.8h, \dst, \src
load_add_store v19.8h, v23.8h, v21.8h, v17.8h, v16.8h, v5.8h, \dst, \src
load_add_store v20.8h, v24.8h, v22.8h, v18.8h, v17.8h, v16.8h, \dst, \src
load_add_store v21.8h, v25.8h, v23.8h, v19.8h, v18.8h, v17.8h, \dst, \src
load_add_store v22.8h, v26.8h, v24.8h, v20.8h, v19.8h, v18.8h, \dst, \src
load_add_store v23.8h, v27.8h, v25.8h, v21.8h, v20.8h, v19.8h, \dst, \src
load_add_store v24.8h, v28.8h, v26.8h, v22.8h, v21.8h, v20.8h, \dst, \src
load_add_store v25.8h, v29.8h, v27.8h, v23.8h, v22.8h, v21.8h, \dst, \src
load_add_store v26.8h, v30.8h, v28.8h, v24.8h, v23.8h, v22.8h, \dst, \src
load_add_store v27.8h, v31.8h, v29.8h, v25.8h, v24.8h, v23.8h, \dst, \src
load_add_store , , v30.8h, v26.8h, v25.8h, v24.8h, \dst, \src
load_add_store , , v31.8h, v27.8h, v26.8h, v25.8h, \dst, \src
load_add_store , , , , v27.8h, v26.8h, \dst, \src
load_add_store , , , , , v27.8h, \dst, \src
.endm
.macro load_add_store_8x8 dst, src, shiftbits=4
mov \src, \dst
mvni v7.8h, #0xfc, lsl #8 // 0x3ff
load_add_store v2.8h, v16.8h, , , , , \dst, \src, \shiftbits
load_add_store v3.8h, v17.8h, , , , , \dst, \src, \shiftbits
load_add_store v4.8h, v18.8h, v16.8h, v2.8h, , , \dst, \src, \shiftbits
load_add_store v5.8h, v19.8h, v17.8h, v3.8h, v2.8h, , \dst, \src, \shiftbits
load_add_store v16.8h, v20.8h, v18.8h, v4.8h, v3.8h, v2.8h, \dst, \src, \shiftbits
load_add_store v17.8h, v21.8h, v19.8h, v5.8h, v4.8h, v3.8h, \dst, \src, \shiftbits
load_add_store v18.8h, v22.8h, v20.8h, v16.8h, v5.8h, v4.8h, \dst, \src, \shiftbits
load_add_store v19.8h, v23.8h, v21.8h, v17.8h, v16.8h, v5.8h, \dst, \src, \shiftbits
load_add_store , , v22.8h, v18.8h, v17.8h, v16.8h, \dst, \src, \shiftbits
load_add_store , , v23.8h, v19.8h, v18.8h, v17.8h, \dst, \src, \shiftbits
load_add_store , , , , v19.8h, v18.8h, \dst, \src, \shiftbits
load_add_store , , , , , v19.8h, \dst, \src, \shiftbits
.endm
.macro load_add_store_8x4 dst, src, shiftbits=4
mov \src, \dst
mvni v7.8h, #0xfc, lsl #8 // 0x3ff
load_add_store v2.8h, v16.8h, , , , , \dst, \src, \shiftbits
load_add_store v3.8h, v17.8h, , , , , \dst, \src, \shiftbits
load_add_store v4.8h, v18.8h, v16.8h, v2.8h, , , \dst, \src, \shiftbits
load_add_store v5.8h, v19.8h, v17.8h, v3.8h, v2.8h, , \dst, \src, \shiftbits
load_add_store , , v18.8h, v4.8h, v3.8h, v2.8h, \dst, \src, \shiftbits
load_add_store , , v19.8h, v5.8h, v4.8h, v3.8h, \dst, \src, \shiftbits
load_add_store , , , , v5.8h, v4.8h, \dst, \src, \shiftbits
load_add_store , , , , , v5.8h, \dst, \src, \shiftbits
.endm
.macro load_add_store4 load, inssrc, insdst, shift, addsrc, adddst, min, store, dst, src
.ifnb \load
ld1 {\load}[0], [\src], x1
.endif
.ifnb \inssrc
ins \insdst\().d[1], \inssrc\().d[0]
.endif
.ifnb \shift
srshr \shift, \shift, #4
.endif
.ifnb \load
ld1 {\load}[1], [\src], x1
.endif
.ifnb \addsrc
usqadd \adddst, \addsrc
.endif
.ifnb \store
st1 {\store}[0], [\dst], x1
.endif
.ifnb \min
smin \min, \min, v7.8h
.endif
.ifnb \store
st1 {\store}[1], [\dst], x1
.endif
.endm
.macro load_add_store_4x16 dst, src
mov \src, \dst
mvni v7.8h, #0xfc, lsl #8 // 0x3ff
load_add_store4 v0.d, v17, v16, , , , , , \dst, \src
load_add_store4 v1.d, v19, v18, , , , , , \dst, \src
load_add_store4 v2.d, v21, v20, v16.8h, , , , , \dst, \src
load_add_store4 v3.d, v23, v22, v18.8h, v16.8h, v0.8h, , , \dst, \src
load_add_store4 v17.d, v25, v24, v20.8h, v18.8h, v1.8h, v0.8h, , \dst, \src
load_add_store4 v19.d, v27, v26, v22.8h, v20.8h, v2.8h, v1.8h, v0.d, \dst, \src
load_add_store4 v21.d, v29, v28, v24.8h, v22.8h, v3.8h, v2.8h, v1.d, \dst, \src
load_add_store4 v23.d, v31, v30, v26.8h, v24.8h, v17.8h, v3.8h, v2.d, \dst, \src
load_add_store4 , , , v28.8h, v26.8h, v19.8h, v17.8h, v3.d, \dst, \src
load_add_store4 , , , v30.8h, v28.8h, v21.8h, v19.8h, v17.d, \dst, \src
load_add_store4 , , , , v30.8h, v23.8h, v21.8h, v19.d, \dst, \src
load_add_store4 , , , , , , v23.8h, v21.d, \dst, \src
load_add_store4 , , , , , , , v23.d, \dst, \src
.endm
.macro load_add_store_4x8 dst, src
mov \src, \dst
mvni v7.8h, #0xfc, lsl #8 // 0x3ff
load_add_store4 v0.d, v17, v16, , , , , , \dst, \src
load_add_store4 v1.d, v19, v18, , , , , , \dst, \src
load_add_store4 v2.d, v21, v20, v16.8h, , , , , \dst, \src
load_add_store4 v3.d, v23, v22, v18.8h, v16.8h, v0.8h, , , \dst, \src
load_add_store4 , , , v20.8h, v18.8h, v1.8h, v0.8h, , \dst, \src
load_add_store4 , , , v22.8h, v20.8h, v2.8h, v1.8h, v0.d, \dst, \src
load_add_store4 , , , , v22.8h, v3.8h, v2.8h, v1.d, \dst, \src
load_add_store4 , , , , , , v3.8h, v2.d, \dst, \src
load_add_store4 , , , , , , , v3.d, \dst, \src
.endm
.macro idct_dc w, h, shift
cbnz w3, 1f
movz w16, #2896*8, lsl #16
ld1r {v16.4s}, [x2]
dup v0.2s, w16
sqrdmulh v20.4s, v16.4s, v0.s[0]
str wzr, [x2]
.if (\w == 2*\h) || (2*\w == \h)
sqrdmulh v20.4s, v20.4s, v0.s[0]
.endif
.if \shift > 0
sqrshrn v16.4h, v20.4s, #\shift
sqrshrn2 v16.8h, v20.4s, #\shift
.else
sqxtn v16.4h, v20.4s
sqxtn2 v16.8h, v20.4s
.endif
sqrdmulh v16.8h, v16.8h, v0.h[1]
srshr v16.8h, v16.8h, #4
mov w4, #\h
b idct_dc_w\w\()_neon
1:
.endm
function idct_dc_w4_neon
mvni v31.8h, #0xfc, lsl #8 // 0x3ff
1:
ld1 {v0.d}[0], [x0], x1
ld1 {v0.d}[1], [x0], x1
ld1 {v1.d}[0], [x0], x1
subs w4, w4, #4
ld1 {v1.d}[1], [x0], x1
usqadd v0.8h, v16.8h
sub x0, x0, x1, lsl #2
usqadd v1.8h, v16.8h
smin v0.8h, v0.8h, v31.8h
st1 {v0.d}[0], [x0], x1
smin v1.8h, v1.8h, v31.8h
st1 {v0.d}[1], [x0], x1
st1 {v1.d}[0], [x0], x1
st1 {v1.d}[1], [x0], x1
b.gt 1b
ret
endfunc
function idct_dc_w8_neon
mvni v31.8h, #0xfc, lsl #8 // 0x3ff
1:
ld1 {v0.8h}, [x0], x1
subs w4, w4, #4
ld1 {v1.8h}, [x0], x1
usqadd v0.8h, v16.8h
ld1 {v2.8h}, [x0], x1
usqadd v1.8h, v16.8h
ld1 {v3.8h}, [x0], x1
usqadd v2.8h, v16.8h
usqadd v3.8h, v16.8h
sub x0, x0, x1, lsl #2
smin v0.8h, v0.8h, v31.8h
smin v1.8h, v1.8h, v31.8h
st1 {v0.8h}, [x0], x1
smin v2.8h, v2.8h, v31.8h
st1 {v1.8h}, [x0], x1
smin v3.8h, v3.8h, v31.8h
st1 {v2.8h}, [x0], x1
st1 {v3.8h}, [x0], x1
b.gt 1b
ret
endfunc
function idct_dc_w16_neon
mvni v31.8h, #0xfc, lsl #8 // 0x3ff
1:
ld1 {v0.8h, v1.8h}, [x0], x1
subs w4, w4, #2
ld1 {v2.8h, v3.8h}, [x0], x1
usqadd v0.8h, v16.8h
usqadd v1.8h, v16.8h
sub x0, x0, x1, lsl #1
usqadd v2.8h, v16.8h
usqadd v3.8h, v16.8h
smin v0.8h, v0.8h, v31.8h
smin v1.8h, v1.8h, v31.8h
smin v2.8h, v2.8h, v31.8h
st1 {v0.8h, v1.8h}, [x0], x1
smin v3.8h, v3.8h, v31.8h
st1 {v2.8h, v3.8h}, [x0], x1
b.gt 1b
ret
endfunc
function idct_dc_w32_neon
mvni v31.8h, #0xfc, lsl #8 // 0x3ff
1:
ld1 {v0.8h, v1.8h, v2.8h, v3.8h}, [x0]
subs w4, w4, #1
usqadd v0.8h, v16.8h
usqadd v1.8h, v16.8h
usqadd v2.8h, v16.8h
usqadd v3.8h, v16.8h
smin v0.8h, v0.8h, v31.8h
smin v1.8h, v1.8h, v31.8h
smin v2.8h, v2.8h, v31.8h
smin v3.8h, v3.8h, v31.8h
st1 {v0.8h, v1.8h, v2.8h, v3.8h}, [x0], x1
b.gt 1b
ret
endfunc
function idct_dc_w64_neon
mvni v31.8h, #0xfc, lsl #8 // 0x3ff
sub x1, x1, #64
1:
ld1 {v0.8h, v1.8h, v2.8h, v3.8h}, [x0], #64
subs w4, w4, #1
usqadd v0.8h, v16.8h
ld1 {v4.8h, v5.8h, v6.8h, v7.8h}, [x0]
usqadd v1.8h, v16.8h
sub x0, x0, #64
usqadd v2.8h, v16.8h
usqadd v3.8h, v16.8h
usqadd v4.8h, v16.8h
usqadd v5.8h, v16.8h
usqadd v6.8h, v16.8h
usqadd v7.8h, v16.8h
smin v0.8h, v0.8h, v31.8h
smin v1.8h, v1.8h, v31.8h
smin v2.8h, v2.8h, v31.8h
smin v3.8h, v3.8h, v31.8h
smin v4.8h, v4.8h, v31.8h
smin v5.8h, v5.8h, v31.8h
st1 {v0.8h, v1.8h, v2.8h, v3.8h}, [x0], #64
smin v6.8h, v6.8h, v31.8h
smin v7.8h, v7.8h, v31.8h
st1 {v4.8h, v5.8h, v6.8h, v7.8h}, [x0], x1
b.gt 1b
ret
endfunc
.macro iwht4
add v16.4s, v16.4s, v17.4s
sub v21.4s, v18.4s, v19.4s
sub v20.4s, v16.4s, v21.4s
sshr v20.4s, v20.4s, #1
sub v18.4s, v20.4s, v17.4s
sub v17.4s, v20.4s, v19.4s
add v19.4s, v21.4s, v18.4s
sub v16.4s, v16.4s, v17.4s
.endm
.macro idct_4 r0, r1, r2, r3
mul_mla v6, \r1, \r3, v0.s[3], v0.s[2]
mul_mla v2, \r0, \r2, v0.s[0], v0.s[0]
mul_mls v4, \r1, \r3, v0.s[2], v0.s[3]
mul_mls v3, \r0, \r2, v0.s[0], v0.s[0]
srshr v6.4s, v6.4s, #12
srshr v2.4s, v2.4s, #12
srshr v7.4s, v4.4s, #12
srshr v3.4s, v3.4s, #12
sqadd \r0\().4s, v2.4s, v6.4s
sqsub \r3\().4s, v2.4s, v6.4s
sqadd \r1\().4s, v3.4s, v7.4s
sqsub \r2\().4s, v3.4s, v7.4s
.endm
function inv_dct_4s_x4_neon
AARCH64_VALID_CALL_TARGET
movrel x16, idct_coeffs
ld1 {v0.4s}, [x16]
idct_4 v16, v17, v18, v19
ret
endfunc
.macro iadst_4x4 o0, o1, o2, o3
movrel x16, iadst4_coeffs
ld1 {v0.4s}, [x16]
sub v3.4s, v16.4s, v18.4s
mul v4.4s, v16.4s, v0.s[0]
mla v4.4s, v18.4s, v0.s[1]
mla v4.4s, v19.4s, v0.s[2]
mul v7.4s, v17.4s, v0.s[3]
add v3.4s, v3.4s, v19.4s
mul v5.4s, v16.4s, v0.s[2]
mls v5.4s, v18.4s, v0.s[0]
mls v5.4s, v19.4s, v0.s[1]
add \o3\().4s, v4.4s, v5.4s
mul \o2\().4s, v3.4s, v0.s[3]
add \o0\().4s, v4.4s, v7.4s
add \o1\().4s, v5.4s, v7.4s
sub \o3\().4s, \o3\().4s, v7.4s
srshr \o0\().4s, \o0\().4s, #12
srshr \o2\().4s, \o2\().4s, #12
srshr \o1\().4s, \o1\().4s, #12
srshr \o3\().4s, \o3\().4s, #12
.endm
function inv_adst_4s_x4_neon
AARCH64_VALID_CALL_TARGET
iadst_4x4 v16, v17, v18, v19
ret
endfunc
function inv_flipadst_4s_x4_neon
AARCH64_VALID_CALL_TARGET
iadst_4x4 v19, v18, v17, v16
ret
endfunc
function inv_identity_4s_x4_neon
AARCH64_VALID_CALL_TARGET
movz w16, #(5793-4096)*8, lsl #16
dup v0.2s, w16
sqrdmulh v4.4s, v16.4s, v0.s[0]
sqrdmulh v5.4s, v17.4s, v0.s[0]
sqrdmulh v6.4s, v18.4s, v0.s[0]
sqrdmulh v7.4s, v19.4s, v0.s[0]
sqadd v16.4s, v16.4s, v4.4s
sqadd v17.4s, v17.4s, v5.4s
sqadd v18.4s, v18.4s, v6.4s
sqadd v19.4s, v19.4s, v7.4s
ret
endfunc
function inv_txfm_add_wht_wht_4x4_16bpc_neon, export=1
mov x15, x30
movi v30.4s, #0
movi v31.4s, #0
ld1 {v16.4s,v17.4s,v18.4s,v19.4s}, [x2]
st1 {v30.4s, v31.4s}, [x2], #32
sshr v16.4s, v16.4s, #2
sshr v17.4s, v17.4s, #2
sshr v18.4s, v18.4s, #2
sshr v19.4s, v19.4s, #2
iwht4
st1 {v30.4s, v31.4s}, [x2], #32
transpose_4x4s v16, v17, v18, v19, v20, v21, v22, v23
iwht4
ld1 {v0.d}[0], [x0], x1
sqxtn v16.4h, v16.4s
ld1 {v0.d}[1], [x0], x1
sqxtn2 v16.8h, v17.4s
ld1 {v1.d}[0], [x0], x1
sqxtn v18.4h, v18.4s
ld1 {v1.d}[1], [x0], x1
sqxtn2 v18.8h, v19.4s
b L(itx_4x4_end)
endfunc
// HBD inv_txfm_add_4x4_neon deviates from the common pattern with registers
// x0-x4 external parameters
// x5 function pointer to first transform
// x6 function pointer to second transform
function inv_txfm_add_4x4_neon
movi v30.4s, #0
movi v31.4s, #0
ld1 {v16.4s,v17.4s,v18.4s,v19.4s}, [x2]
st1 {v30.4s, v31.4s}, [x2], #32
blr x5
st1 {v30.4s, v31.4s}, [x2], #32
sqxtn v16.4h, v16.4s
sqxtn v17.4h, v17.4s
sqxtn v18.4h, v18.4s
sqxtn v19.4h, v19.4s
transpose_4x4h v16, v17, v18, v19, v20, v21, v22, v23
blr x6
ld1 {v0.d}[0], [x0], x1
ld1 {v0.d}[1], [x0], x1
ins v16.d[1], v17.d[0]
ins v18.d[1], v19.d[0]
ld1 {v1.d}[0], [x0], x1
ld1 {v1.d}[1], [x0], x1
srshr v16.8h, v16.8h, #4
srshr v18.8h, v18.8h, #4
L(itx_4x4_end):
dup v31.8h, w4
sub x0, x0, x1, lsl #2
usqadd v0.8h, v16.8h
usqadd v1.8h, v18.8h
smin v0.8h, v0.8h, v31.8h
st1 {v0.d}[0], [x0], x1
smin v1.8h, v1.8h, v31.8h
st1 {v0.d}[1], [x0], x1
st1 {v1.d}[0], [x0], x1
st1 {v1.d}[1], [x0], x1
ret x15
endfunc
.macro def_fn_4x4 txfm1, txfm2
function inv_txfm_add_\txfm1\()_\txfm2\()_4x4_16bpc_neon, export=1
mov x15, x30
.ifc \txfm1\()_\txfm2, dct_dct
cbnz w3, 1f
movz w16, #2896*8, lsl #16
ld1r {v16.4s}, [x2]
dup v4.2s, w16
str wzr, [x2]
sqrdmulh v16.4s, v16.4s, v4.s[0]
ld1 {v0.d}[0], [x0], x1
sqxtn v20.4h, v16.4s
sqxtn2 v20.8h, v16.4s
ld1 {v0.d}[1], [x0], x1
sqrdmulh v20.8h, v20.8h, v4.h[1]
ld1 {v1.d}[0], [x0], x1
srshr v16.8h, v20.8h, #4
ld1 {v1.d}[1], [x0], x1
srshr v18.8h, v20.8h, #4
movi v30.8h, #0
b L(itx_4x4_end)
1:
.endif
adr x5, inv_\txfm1\()_4s_x4_neon
movrel x6, X(inv_\txfm2\()_4h_x4_neon)
b inv_txfm_add_4x4_neon
endfunc
.endm
def_fn_4x4 dct, dct
def_fn_4x4 identity, identity
def_fn_4x4 dct, adst
def_fn_4x4 dct, flipadst
def_fn_4x4 dct, identity
def_fn_4x4 adst, dct
def_fn_4x4 adst, adst
def_fn_4x4 adst, flipadst
def_fn_4x4 flipadst, dct
def_fn_4x4 flipadst, adst
def_fn_4x4 flipadst, flipadst
def_fn_4x4 identity, dct
def_fn_4x4 adst, identity
def_fn_4x4 flipadst, identity
def_fn_4x4 identity, adst
def_fn_4x4 identity, flipadst
.macro idct_8 r0, r1, r2, r3, r4, r5, r6, r7
idct_4 \r0, \r2, \r4, \r6
movi v5.4s, #1, msl #16 // row_clip_max = ~(~bdmax << 7), 0x1ffff
mvni v4.4s, #1, msl #16 // row_clip_min = (~bdmax << 7), 0xfffe0000
.irp r, \r0, \r2, \r4, \r6
smin_4s \r, \r, v5
.endr
.irp r, \r0, \r2, \r4, \r6
smax_4s \r, \r, v4
.endr
mul_mls v2, \r1, \r7, v1.s[0], v1.s[1] // -> t4a
mul_mla v3, \r1, \r7, v1.s[1], v1.s[0] // -> t7a
mul_mls v6, \r5, \r3, v1.s[2], v1.s[3] // -> t5a
mul_mla v7, \r5, \r3, v1.s[3], v1.s[2] // -> t6a
srshr \r1\().4s, v2.4s, #12 // t4a
srshr \r7\().4s, v3.4s, #12 // t7a
srshr \r3\().4s, v6.4s, #12 // t5a
srshr \r5\().4s, v7.4s, #12 // t6a
sqadd v2.4s, \r1\().4s, \r3\().4s // t4
sqsub \r1\().4s, \r1\().4s, \r3\().4s // t5a
sqadd v3.4s, \r7\().4s, \r5\().4s // t7
sqsub \r3\().4s, \r7\().4s, \r5\().4s // t6a
.irp r, v2, \r1, v3, \r3
smin_4s \r, \r, v5
.endr
.irp r, v2, \r1, v3, \r3
smax_4s \r, \r, v4
.endr
mul_mls v7, \r3, \r1, v0.s[0], v0.s[0] // -> t5
mul_mla v6, \r3, \r1, v0.s[0], v0.s[0] // -> t6
srshr v7.4s, v7.4s, #12 // t5
srshr v6.4s, v6.4s, #12 // t6
sqsub \r7\().4s, \r0\().4s, v3.4s // out7
sqadd \r0\().4s, \r0\().4s, v3.4s // out0
sqadd \r1\().4s, \r2\().4s, v6.4s // out1
sqsub v6.4s, \r2\().4s, v6.4s // out6
sqadd \r2\().4s, \r4\().4s, v7.4s // out2
sqsub \r5\().4s, \r4\().4s, v7.4s // out5
sqadd \r3\().4s, \r6\().4s, v2.4s // out3
sqsub \r4\().4s, \r6\().4s, v2.4s // out4
mov \r6\().16b, v6.16b // out6
.endm
function inv_dct_4s_x8_neon
AARCH64_VALID_CALL_TARGET
movrel x16, idct_coeffs
ld1 {v0.4s, v1.4s}, [x16]
idct_8 v16, v17, v18, v19, v20, v21, v22, v23
ret
endfunc
.macro iadst_8 o0, o1, o2, o3, o4, o5, o6, o7
movrel x16, iadst8_coeffs
ld1 {v0.4s, v1.4s}, [x16], #32
mul_mla v2, v23, v16, v0.s[0], v0.s[1]
mul_mls v4, v23, v16, v0.s[1], v0.s[0]
mul_mla v6, v21, v18, v0.s[2], v0.s[3]
srshr v16.4s, v2.4s, #12 // t0a
srshr v23.4s, v4.4s, #12 // t1a
mul_mls v2, v21, v18, v0.s[3], v0.s[2]
mul_mla v4, v19, v20, v1.s[0], v1.s[1]
srshr v18.4s, v6.4s, #12 // t2a
srshr v21.4s, v2.4s, #12 // t3a
mul_mls v6, v19, v20, v1.s[1], v1.s[0]
mul_mla v2, v17, v22, v1.s[2], v1.s[3]
srshr v20.4s, v4.4s, #12 // t4a
srshr v19.4s, v6.4s, #12 // t5a
mul_mls v4, v17, v22, v1.s[3], v1.s[2]
srshr v22.4s, v2.4s, #12 // t6a
srshr v17.4s, v4.4s, #12 // t7a
ld1 {v0.4s}, [x16]
movi v1.4s, #1, msl #16 // row_clip_max = ~(~bdmax << 7), 0x1ffff
sqadd v2.4s, v16.4s, v20.4s // t0
sqsub v3.4s, v16.4s, v20.4s // t4
mvni v20.4s, #1, msl #16 // row_clip_min = (~bdmax << 7), 0xfffe0000
sqadd v4.4s, v23.4s, v19.4s // t1
sqsub v5.4s, v23.4s, v19.4s // t5
sqadd v6.4s, v18.4s, v22.4s // t2
sqsub v7.4s, v18.4s, v22.4s // t6
sqadd v18.4s, v21.4s, v17.4s // t3
sqsub v19.4s, v21.4s, v17.4s // t7
.irp r, v2, v3, v4, v5, v6, v7, v18, v19
smin_4s \r, \r, v1
.endr
.irp r, v2, v3, v4, v5, v6, v7, v18, v19
smax_4s \r, \r, v20
.endr
mul_mla v16, v3, v5, v0.s[3], v0.s[2]
mul_mls v20, v3, v5, v0.s[2], v0.s[3]
mul_mls v22, v19, v7, v0.s[3], v0.s[2]
srshr v3.4s, v16.4s, #12 // t4a
srshr v5.4s, v20.4s, #12 // t5a
mul_mla v16, v19, v7, v0.s[2], v0.s[3]
srshr v7.4s, v22.4s, #12 // t6a
srshr v19.4s, v16.4s, #12 // t7a
sqadd \o0\().4s, v2.4s, v6.4s // out0
sqsub v2.4s, v2.4s, v6.4s // t2
sqadd \o7\().4s, v4.4s, v18.4s // out7
sqsub v4.4s, v4.4s, v18.4s // t3
mvni v18.4s, #1, msl #16 // row_clip_min = (~bdmax << 7), 0xfffe0000
sqadd \o1\().4s, v3.4s, v7.4s // out1
sqsub v3.4s, v3.4s, v7.4s // t6
sqadd \o6\().4s, v5.4s, v19.4s // out6
sqsub v5.4s, v5.4s, v19.4s // t7
// Not clipping the output registers, as they will be downshifted and
// narrowed afterwards anyway.
.irp r, v2, v4, v3, v5
smin_4s \r, \r, v1
.endr
.irp r, v2, v4, v3, v5
smax_4s \r, \r, v18
.endr
sqneg \o7\().4s, \o7\().4s // out7
sqneg \o1\().4s, \o1\().4s // out1
mul_mla v18, v2, v4, v0.s[0], v0.s[0] // -> out3 (v19 or v20)
mul_mls v6, v2, v4, v0.s[0], v0.s[0] // -> out4 (v20 or v19)
mul_mls v20, v3, v5, v0.s[0], v0.s[0] // -> out5 (v21 or v18)
srshr v2.4s, v18.4s, #12 // out3
mul_mla v18, v3, v5, v0.s[0], v0.s[0] // -> out2 (v18 or v21)
srshr v3.4s, v20.4s, #12 // out5
srshr \o2\().4s, v18.4s, #12 // out2 (v18 or v21)
srshr \o4\().4s, v6.4s, #12 // out4 (v20 or v19)
sqneg \o3\().4s, v2.4s // out3
sqneg \o5\().4s, v3.4s // out5
.endm
function inv_adst_4s_x8_neon
AARCH64_VALID_CALL_TARGET
iadst_8 v16, v17, v18, v19, v20, v21, v22, v23
ret
endfunc
function inv_flipadst_4s_x8_neon
AARCH64_VALID_CALL_TARGET
iadst_8 v23, v22, v21, v20, v19, v18, v17, v16
ret
endfunc
function inv_identity_4s_x8_neon
AARCH64_VALID_CALL_TARGET
sqshl v16.4s, v16.4s, #1
sqshl v17.4s, v17.4s, #1
sqshl v18.4s, v18.4s, #1
sqshl v19.4s, v19.4s, #1
sqshl v20.4s, v20.4s, #1
sqshl v21.4s, v21.4s, #1
sqshl v22.4s, v22.4s, #1
sqshl v23.4s, v23.4s, #1
ret
endfunc
function inv_txfm_add_8x8_neon
movi v31.4s, #0
cmp w3, w13
mov x11, #32
b.lt 1f
add x6, x2, #16
.irp i, v16.4s, v17.4s, v18.4s, v19.4s, v20.4s, v21.4s, v22.4s, v23.4s
ld1 {\i}, [x6]
st1 {v31.4s}, [x6], x11
.endr
blr x4
sqrshrn v24.4h, v16.4s, #1
sqrshrn v25.4h, v17.4s, #1
sqrshrn v26.4h, v18.4s, #1
sqrshrn v27.4h, v19.4s, #1
sqrshrn2 v24.8h, v20.4s, #1
sqrshrn2 v25.8h, v21.4s, #1
sqrshrn2 v26.8h, v22.4s, #1
sqrshrn2 v27.8h, v23.4s, #1
transpose_4x8h v24, v25, v26, v27, v2, v3, v4, v5
b 2f
1:
.irp i, v24.8h, v25.8h, v26.8h, v27.8h
movi \i, #0
.endr
2:
.irp i, v16.4s, v17.4s, v18.4s, v19.4s, v20.4s, v21.4s, v22.4s, v23.4s
ld1 {\i}, [x2]
st1 {v31.4s}, [x2], x11
.endr
blr x4
sqrshrn v16.4h, v16.4s, #1
sqrshrn v17.4h, v17.4s, #1
sqrshrn v18.4h, v18.4s, #1
sqrshrn v19.4h, v19.4s, #1
sqrshrn2 v16.8h, v20.4s, #1
sqrshrn2 v17.8h, v21.4s, #1
sqrshrn2 v18.8h, v22.4s, #1
sqrshrn2 v19.8h, v23.4s, #1
transpose_4x8h v16, v17, v18, v19, v20, v21, v22, v23
mov v20.16b, v24.16b
mov v21.16b, v25.16b
mov v22.16b, v26.16b
mov v23.16b, v27.16b
blr x5
load_add_store_8x8 x0, x7
ret x15
endfunc
.macro def_fn_8x8 txfm1, txfm2, eob_half
function inv_txfm_add_\txfm1\()_\txfm2\()_8x8_16bpc_neon, export=1
mov x15, x30
.ifc \txfm1\()_\txfm2, dct_dct
idct_dc 8, 8, 1
.endif
movrel x5, X(inv_\txfm2\()_8h_x8_neon)
mov w13, #\eob_half
adr x4, inv_\txfm1\()_4s_x8_neon
b inv_txfm_add_8x8_neon
endfunc
.endm
def_fn_8x8 dct, dct, 10
def_fn_8x8 identity, identity, 10
def_fn_8x8 dct, adst, 10
def_fn_8x8 dct, flipadst, 10
def_fn_8x8 dct, identity, 4
def_fn_8x8 adst, dct, 10
def_fn_8x8 adst, adst, 10
def_fn_8x8 adst, flipadst, 10
def_fn_8x8 flipadst, dct, 10
def_fn_8x8 flipadst, adst, 10
def_fn_8x8 flipadst, flipadst, 10
def_fn_8x8 identity, dct, 4
def_fn_8x8 adst, identity, 4
def_fn_8x8 flipadst, identity, 4
def_fn_8x8 identity, adst, 4
def_fn_8x8 identity, flipadst, 4
function inv_txfm_add_8x4_neon
movi v28.4s, #0
movi v29.4s, #0
movi v30.4s, #0
movi v31.4s, #0
ld1 {v16.4s,v17.4s,v18.4s,v19.4s}, [x2]
st1 {v28.4s,v29.4s,v30.4s,v31.4s}, [x2], #64
movz w16, #2896*8, lsl #16
dup v0.2s, w16
ld1 {v20.4s,v21.4s,v22.4s,v23.4s}, [x2]
st1 {v28.4s,v29.4s,v30.4s,v31.4s}, [x2]
scale_input .4s, v0.s[0], v16, v17, v18, v19, v20, v21, v22, v23
blr x4
sqxtn v16.4h, v16.4s
sqxtn v17.4h, v17.4s
sqxtn v18.4h, v18.4s
sqxtn v19.4h, v19.4s
sqxtn v20.4h, v20.4s
sqxtn v21.4h, v21.4s
sqxtn v22.4h, v22.4s
sqxtn v23.4h, v23.4s
transpose_4x4h v16, v17, v18, v19, v4, v5, v6, v7
transpose_4x4h v20, v21, v22, v23, v4, v5, v6, v7
ins v16.d[1], v20.d[0]
ins v17.d[1], v21.d[0]
ins v18.d[1], v22.d[0]
ins v19.d[1], v23.d[0]
blr x5
load_add_store_8x4 x0, x7
ret x15
endfunc
function inv_txfm_add_4x8_neon
movz w16, #2896*8, lsl #16
movi v31.4s, #0
dup v30.2s, w16
cmp w3, w13
mov x11, #32
b.lt 1f
add x6, x2, #16
.irp i, v16.4s, v17.4s, v18.4s, v19.4s
ld1 {\i}, [x6]
st1 {v31.4s}, [x6], x11
.endr
scale_input .4s, v30.s[0], v16, v17, v18, v19
blr x4
sqxtn v20.4h, v16.4s
sqxtn v21.4h, v17.4s
sqxtn v22.4h, v18.4s
sqxtn v23.4h, v19.4s
transpose_4x4h v20, v21, v22, v23, v4, v5, v6, v7
b 2f
1:
.irp i, v20, v21, v22, v23
movi \i\().4h, #0
.endr
2:
.irp i, v16.4s, v17.4s, v18.4s, v19.4s
ld1 {\i}, [x2]
st1 {v31.4s}, [x2], x11
.endr
scale_input .4s, v30.s[0], v16, v17, v18, v19
blr x4
sqxtn v16.4h, v16.4s
sqxtn v17.4h, v17.4s
sqxtn v18.4h, v18.4s
sqxtn v19.4h, v19.4s
transpose_4x4h v16, v17, v18, v19, v4, v5, v6, v7
blr x5
load_add_store_4x8 x0, x7
ret x15
endfunc
.macro def_fn_48 w, h, txfm1, txfm2, eob_half
function inv_txfm_add_\txfm1\()_\txfm2\()_\w\()x\h\()_16bpc_neon, export=1
mov x15, x30
.ifc \txfm1\()_\txfm2, dct_dct
idct_dc \w, \h, 0
.endif
adr x4, inv_\txfm1\()_4s_x\w\()_neon
.if \w == 4
mov w13, #\eob_half
.endif
movrel x5, X(inv_\txfm2\()_\w\()h_x\h\()_neon)
b inv_txfm_add_\w\()x\h\()_neon
endfunc
.endm
.macro def_fns_48 w, h
def_fn_48 \w, \h, dct, dct, 13
def_fn_48 \w, \h, identity, identity, 13
def_fn_48 \w, \h, dct, adst, 13
def_fn_48 \w, \h, dct, flipadst, 13
def_fn_48 \w, \h, dct, identity, 4
def_fn_48 \w, \h, adst, dct, 13
def_fn_48 \w, \h, adst, adst, 13
def_fn_48 \w, \h, adst, flipadst, 13
def_fn_48 \w, \h, flipadst, dct, 13
def_fn_48 \w, \h, flipadst, adst, 13
def_fn_48 \w, \h, flipadst, flipadst, 13
def_fn_48 \w, \h, identity, dct, 16
def_fn_48 \w, \h, adst, identity, 4
def_fn_48 \w, \h, flipadst, identity, 4
def_fn_48 \w, \h, identity, adst, 16
def_fn_48 \w, \h, identity, flipadst, 16
.endm
def_fns_48 4, 8
def_fns_48 8, 4
function inv_dct_4s_x16_neon
AARCH64_VALID_CALL_TARGET
movrel x16, idct_coeffs
ld1 {v0.4s, v1.4s}, [x16], #32
idct_8 v16, v18, v20, v22, v24, v26, v28, v30
// idct_8 leaves the row_clip_max/min constants in v5 and v4
.irp r, v16, v18, v20, v22, v24, v26, v28, v30
smin \r\().4s, \r\().4s, v5.4s
.endr
.irp r, v16, v18, v20, v22, v24, v26, v28, v30
smax \r\().4s, \r\().4s, v4.4s
.endr
ld1 {v0.4s, v1.4s}, [x16]
sub x16, x16, #32
mul_mls v2, v17, v31, v0.s[0], v0.s[1] // -> t8a
mul_mla v3, v17, v31, v0.s[1], v0.s[0] // -> t15a
mul_mls v6, v25, v23, v0.s[2], v0.s[3] // -> t9a
srshr v17.4s, v2.4s, #12 // t8a
srshr v31.4s, v3.4s, #12 // t15a
mul_mla v2, v25, v23, v0.s[3], v0.s[2] // -> t14a
mul_mls v3, v21, v27, v1.s[0], v1.s[1] // -> t10a
srshr v23.4s, v6.4s, #12 // t9a
srshr v25.4s, v2.4s, #12 // t14a
mul_mla v6, v21, v27, v1.s[1], v1.s[0] // -> t13a
mul_mls v2, v29, v19, v1.s[2], v1.s[3] // -> t11a
srshr v21.4s, v3.4s, #12 // t10a
srshr v27.4s, v6.4s, #12 // t13a
mul_mla v3, v29, v19, v1.s[3], v1.s[2] // -> t12a
srshr v19.4s, v2.4s, #12 // t11a
srshr v29.4s, v3.4s, #12 // t12a
ld1 {v0.4s}, [x16]
sqsub v2.4s, v17.4s, v23.4s // t9
sqadd v17.4s, v17.4s, v23.4s // t8
sqsub v3.4s, v31.4s, v25.4s // t14
sqadd v31.4s, v31.4s, v25.4s // t15
sqsub v23.4s, v19.4s, v21.4s // t10
sqadd v19.4s, v19.4s, v21.4s // t11
sqadd v25.4s, v29.4s, v27.4s // t12
sqsub v29.4s, v29.4s, v27.4s // t13
.irp r, v2, v17, v3, v31, v23, v19, v25, v29
smin \r\().4s, \r\().4s, v5.4s
.endr
.irp r, v2, v17, v3, v31, v23, v19, v25, v29
smax \r\().4s, \r\().4s, v4.4s
.endr
mul_mls v7, v3, v2, v0.s[2], v0.s[3] // -> t9a
mul_mla v6, v3, v2, v0.s[3], v0.s[2] // -> t14a
srshr v21.4s, v7.4s, #12 // t9a
srshr v27.4s, v6.4s, #12 // t14a
mul_mls v7, v29, v23, v0.s[2], v0.s[3] // -> t13a
mul_mla v6, v29, v23, v0.s[3], v0.s[2] // -> t10a
srshr v29.4s, v7.4s, #12 // t13a
neg v6.4s, v6.4s
srshr v23.4s, v6.4s, #12 // t10a
sqsub v2.4s, v17.4s, v19.4s // t11a
sqadd v17.4s, v17.4s, v19.4s // t8a
sqsub v3.4s, v31.4s, v25.4s // t12a
sqadd v31.4s, v31.4s, v25.4s // t15a
sqadd v19.4s, v21.4s, v23.4s // t9
sqsub v21.4s, v21.4s, v23.4s // t10
sqsub v25.4s, v27.4s, v29.4s // t13
sqadd v27.4s, v27.4s, v29.4s // t14
.irp r, v2, v17, v3, v31, v19, v21, v25, v27
smin \r\().4s, \r\().4s, v5.4s
.endr
.irp r, v2, v17, v3, v31, v19, v21, v25, v27
smax \r\().4s, \r\().4s, v4.4s
.endr
mul_mls v7, v3, v2, v0.s[0], v0.s[0] // -> t11
mul_mla v6, v3, v2, v0.s[0], v0.s[0] // -> t12
mul_mls v2, v25, v21, v0.s[0], v0.s[0] // -> t10a
srshr v7.4s, v7.4s, #12 // t11
srshr v6.4s, v6.4s, #12 // t12
mul_mla v3, v25, v21, v0.s[0], v0.s[0] // -> t13a
srshr v2.4s, v2.4s, #12 // t10a
srshr v3.4s, v3.4s, #12 // t13a
sqadd v1.4s, v16.4s, v31.4s // out0
sqsub v31.4s, v16.4s, v31.4s // out15
mov v16.16b, v1.16b
sqadd v23.4s, v30.4s, v17.4s // out7
sqsub v1.4s, v30.4s, v17.4s // out8
sqadd v17.4s, v18.4s, v27.4s // out1
sqsub v30.4s, v18.4s, v27.4s // out14
sqadd v18.4s, v20.4s, v3.4s // out2
sqsub v29.4s, v20.4s, v3.4s // out13
sqadd v3.4s, v28.4s, v19.4s // out6
sqsub v25.4s, v28.4s, v19.4s // out9
sqadd v19.4s, v22.4s, v6.4s // out3
sqsub v28.4s, v22.4s, v6.4s // out12
sqadd v20.4s, v24.4s, v7.4s // out4
sqsub v27.4s, v24.4s, v7.4s // out11
sqadd v21.4s, v26.4s, v2.4s // out5
sqsub v26.4s, v26.4s, v2.4s // out10
mov v24.16b, v1.16b
mov v22.16b, v3.16b
ret
endfunc
.macro iadst_16 o0, o1, o2, o3, o4, o5, o6, o7, o8, o9, o10, o11, o12, o13, o14, o15
movrel x16, iadst16_coeffs
ld1 {v0.4s, v1.4s}, [x16], #32
mul_mla v2, v31, v16, v0.s[0], v0.s[1] // -> t0
mul_mls v4, v31, v16, v0.s[1], v0.s[0] // -> t1
mul_mla v6, v29, v18, v0.s[2], v0.s[3] // -> t2
srshr v16.4s, v2.4s, #12 // t0
srshr v31.4s, v4.4s, #12 // t1
mul_mls v2, v29, v18, v0.s[3], v0.s[2] // -> t3
mul_mla v4, v27, v20, v1.s[0], v1.s[1] // -> t4
srshr v18.4s, v6.4s, #12 // t2
srshr v29.4s, v2.4s, #12 // t3
mul_mls v6, v27, v20, v1.s[1], v1.s[0] // -> t5
mul_mla v2, v25, v22, v1.s[2], v1.s[3] // -> t6
srshr v20.4s, v4.4s, #12 // t4
srshr v27.4s, v6.4s, #12 // t5
mul_mls v4, v25, v22, v1.s[3], v1.s[2] // -> t7
ld1 {v0.4s, v1.4s}, [x16]
movrel x16, idct_coeffs
mul_mla v6, v23, v24, v0.s[0], v0.s[1] // -> t8
srshr v22.4s, v2.4s, #12 // t6
srshr v25.4s, v4.4s, #12 // t7
mul_mls v2, v23, v24, v0.s[1], v0.s[0] // -> t9
mul_mla v4, v21, v26, v0.s[2], v0.s[3] // -> t10
srshr v23.4s, v6.4s, #12 // t8
srshr v24.4s, v2.4s, #12 // t9
mul_mls v6, v21, v26, v0.s[3], v0.s[2] // -> t11
mul_mla v2, v19, v28, v1.s[0], v1.s[1] // -> t12
srshr v21.4s, v4.4s, #12 // t10
srshr v26.4s, v6.4s, #12 // t11
mul_mls v4, v19, v28, v1.s[1], v1.s[0] // -> t13
mul_mla v6, v17, v30, v1.s[2], v1.s[3] // -> t14
srshr v19.4s, v2.4s, #12 // t12
srshr v28.4s, v4.4s, #12 // t13
mul_mls v2, v17, v30, v1.s[3], v1.s[2] // -> t15
srshr v17.4s, v6.4s, #12 // t14
srshr v30.4s, v2.4s, #12 // t15
ld1 {v0.4s, v1.4s}, [x16]
movi v5.4s, #1, msl #16 // row_clip_max = ~(~bdmax << 7), 0x1ffff
mvni v7.4s, #1, msl #16 // row_clip_min = (~bdmax << 7), 0xfffe0000
sqsub v2.4s, v16.4s, v23.4s // t8a
sqadd v16.4s, v16.4s, v23.4s // t0a
sqsub v3.4s, v31.4s, v24.4s // t9a
sqadd v31.4s, v31.4s, v24.4s // t1a
sqadd v23.4s, v18.4s, v21.4s // t2a
sqsub v18.4s, v18.4s, v21.4s // t10a
sqadd v24.4s, v29.4s, v26.4s // t3a
sqsub v29.4s, v29.4s, v26.4s // t11a
sqadd v21.4s, v20.4s, v19.4s // t4a
sqsub v20.4s, v20.4s, v19.4s // t12a
sqadd v26.4s, v27.4s, v28.4s // t5a
sqsub v27.4s, v27.4s, v28.4s // t13a
sqadd v19.4s, v22.4s, v17.4s // t6a
sqsub v22.4s, v22.4s, v17.4s // t14a
sqadd v28.4s, v25.4s, v30.4s // t7a
sqsub v25.4s, v25.4s, v30.4s // t15a
.irp r, v2, v16, v3, v31, v23, v18, v24, v29, v21, v20, v26, v27, v19, v22, v28, v25
smin_4s \r, \r, v5
.endr
.irp r, v2, v16, v3, v31, v23, v18, v24, v29, v21, v20, v26, v27, v19, v22, v28, v25
smax_4s \r, \r, v7
.endr
mul_mla v4, v2, v3, v1.s[1], v1.s[0] // -> t8
mul_mls v6, v2, v3, v1.s[0], v1.s[1] // -> t9
mul_mla v2, v18, v29, v1.s[3], v1.s[2] // -> t10
srshr v17.4s, v4.4s, #12 // t8
srshr v30.4s, v6.4s, #12 // t9
mul_mls v4, v18, v29, v1.s[2], v1.s[3] // -> t11
mul_mls v6, v27, v20, v1.s[1], v1.s[0] // -> t12
srshr v18.4s, v2.4s, #12 // t10
srshr v29.4s, v4.4s, #12 // t11
mul_mla v2, v27, v20, v1.s[0], v1.s[1] // -> t13
mul_mls v4, v25, v22, v1.s[3], v1.s[2] // -> t14
srshr v27.4s, v6.4s, #12 // t12
srshr v20.4s, v2.4s, #12 // t13
mul_mla v6, v25, v22, v1.s[2], v1.s[3] // -> t15
srshr v25.4s, v4.4s, #12 // t14
srshr v22.4s, v6.4s, #12 // t15
sqsub v2.4s, v16.4s, v21.4s // t4
sqadd v16.4s, v16.4s, v21.4s // t0
sqsub v3.4s, v31.4s, v26.4s // t5
sqadd v31.4s, v31.4s, v26.4s // t1
sqadd v21.4s, v23.4s, v19.4s // t2
sqsub v23.4s, v23.4s, v19.4s // t6
sqadd v26.4s, v24.4s, v28.4s // t3
sqsub v24.4s, v24.4s, v28.4s // t7
sqadd v19.4s, v17.4s, v27.4s // t8a
sqsub v17.4s, v17.4s, v27.4s // t12a
sqadd v28.4s, v30.4s, v20.4s // t9a
sqsub v30.4s, v30.4s, v20.4s // t13a
sqadd v27.4s, v18.4s, v25.4s // t10a
sqsub v18.4s, v18.4s, v25.4s // t14a
sqadd v20.4s, v29.4s, v22.4s // t11a
sqsub v29.4s, v29.4s, v22.4s // t15a
.irp r, v2, v16, v3, v31, v21, v23, v26, v24, v19, v17, v28, v30, v27, v18, v20, v29
smin_4s \r, \r, v5
.endr
.irp r, v2, v16, v3, v31, v21, v23, v26, v24, v19, v17, v28, v30, v27, v18, v20, v29
smax_4s \r, \r, v7
.endr
mul_mla v4, v2, v3, v0.s[3], v0.s[2] // -> t4a
mul_mls v6, v2, v3, v0.s[2], v0.s[3] // -> t5a
mul_mls v2, v24, v23, v0.s[3], v0.s[2] // -> t6a
srshr v22.4s, v4.4s, #12 // t4a
srshr v25.4s, v6.4s, #12 // t5a
mul_mla v4, v24, v23, v0.s[2], v0.s[3] // -> t7a
mul_mla v6, v17, v30, v0.s[3], v0.s[2] // -> t12
srshr v24.4s, v2.4s, #12 // t6a
srshr v23.4s, v4.4s, #12 // t7a
mul_mls v2, v17, v30, v0.s[2], v0.s[3] // -> t13
mul_mls v4, v29, v18, v0.s[3], v0.s[2] // -> t14
srshr v17.4s, v6.4s, #12 // t12
mul_mla v6, v29, v18, v0.s[2], v0.s[3] // -> t15
srshr v29.4s, v2.4s, #12 // t13
srshr v30.4s, v4.4s, #12 // t14
srshr v18.4s, v6.4s, #12 // t15
sqsub v2.4s, v16.4s, v21.4s // t2a
.ifc \o0, v16
sqadd \o0\().4s, v16.4s, v21.4s // out0
sqsub v21.4s, v31.4s, v26.4s // t3a
sqadd \o15\().4s, v31.4s, v26.4s // out15
.else
sqadd v4.4s, v16.4s, v21.4s // out0
sqsub v21.4s, v31.4s, v26.4s // t3a
sqadd \o15\().4s, v31.4s, v26.4s // out15
mov \o0\().16b, v4.16b
.endif
sqsub v3.4s, v29.4s, v18.4s // t15a
sqadd \o13\().4s, v29.4s, v18.4s // out13
sqadd \o2\().4s, v17.4s, v30.4s // out2
sqsub v26.4s, v17.4s, v30.4s // t14a
sqadd \o1\().4s, v19.4s, v27.4s // out1
sqsub v27.4s, v19.4s, v27.4s // t10
sqadd \o14\().4s, v28.4s, v20.4s // out14
sqsub v20.4s, v28.4s, v20.4s // t11
sqadd \o3\().4s, v22.4s, v24.4s // out3
sqsub v22.4s, v22.4s, v24.4s // t6
sqadd \o12\().4s, v25.4s, v23.4s // out12
sqsub v23.4s, v25.4s, v23.4s // t7
// Not clipping the output registers, as they will be downshifted and
// narrowed afterwards anyway.
.irp r, v2, v21, v3, v26, v27, v20, v22, v23
smin_4s \r, \r, v5
.endr
.irp r, v2, v21, v3, v26, v27, v20, v22, v23
smax_4s \r, \r, v7
.endr
sqneg \o15\().4s, \o15\().4s // out15
sqneg \o13\().4s, \o13\().4s // out13
sqneg \o1\().4s, \o1\().4s // out1
sqneg \o3\().4s, \o3\().4s // out3
mul_mls v24, v2, v21, v0.s[0], v0.s[0] // -> out8 (v24 or v23)
mul_mla v4, v2, v21, v0.s[0], v0.s[0] // -> out7 (v23 or v24)
mul_mla v6, v26, v3, v0.s[0], v0.s[0] // -> out5 (v21 or v26)
srshr v24.4s, v24.4s, #12 // out8
srshr v4.4s, v4.4s, #12 // out7
srshr v5.4s, v6.4s, #12 // out5
mul_mls v6, v26, v3, v0.s[0], v0.s[0] // -> out10 (v26 or v21)
mul_mla v2, v22, v23, v0.s[0], v0.s[0] // -> out4 (v20 or v27)
srshr v26.4s, v6.4s, #12 // out10
mul_mls v6, v22, v23, v0.s[0], v0.s[0] // -> out11 (v27 or v20)
mul_mla v22, v27, v20, v0.s[0], v0.s[0] // -> out6 (v22 or v25)
mul_mls v21, v27, v20, v0.s[0], v0.s[0] // -> out9 (v25 or v22)
srshr \o4\().4s, v2.4s, #12 // out4
srshr v6.4s, v6.4s, #12 // out11
srshr v7.4s, v21.4s, #12 // out9
srshr \o6\().4s, v22.4s, #12 // out6
.ifc \o8, v23
mov \o8\().16b, v24.16b
mov \o10\().16b, v26.16b
.endif
sqneg \o7\().4s, v4.4s // out7
sqneg \o5\().4s, v5.4s // out5
sqneg \o11\().4s, v6.4s // out11
sqneg \o9\().4s, v7.4s // out9
.endm
function inv_adst_4s_x16_neon
AARCH64_VALID_CALL_TARGET
iadst_16 v16, v17, v18, v19, v20, v21, v22, v23, v24, v25, v26, v27, v28, v29, v30, v31
ret
endfunc
function inv_flipadst_4s_x16_neon
AARCH64_VALID_CALL_TARGET
iadst_16 v31, v30, v29, v28, v27, v26, v25, v24, v23, v22, v21, v20, v19, v18, v17, v16
ret
endfunc
function inv_identity_4s_x16_neon
AARCH64_VALID_CALL_TARGET
movz w16, #2*(5793-4096)*8, lsl #16
dup v0.2s, w16
.irp i, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31
sqrdmulh v2.4s, v\i\().4s, v0.s[0]
sqadd v\i\().4s, v\i\().4s, v\i\().4s
sqadd v\i\().4s, v\i\().4s, v2.4s
.endr
ret
endfunc
.macro identity_4x16_shift1 c
.irp i, v16.4s, v17.4s, v18.4s, v19.4s, v20.4s, v21.4s, v22.4s, v23.4s, v24.4s, v25.4s, v26.4s, v27.4s, v28.4s, v29.4s, v30.4s, v31.4s
sqrdmulh v3.4s, \i, \c
srshr v3.4s, v3.4s, #1
sqadd \i, \i, v3.4s
.endr
.endm
.macro identity_4x16 c
.irp i, v16.4s, v17.4s, v18.4s, v19.4s, v20.4s, v21.4s, v22.4s, v23.4s, v24.4s, v25.4s, v26.4s, v27.4s, v28.4s, v29.4s, v30.4s, v31.4s
sqrdmulh v3.4s, \i, \c
sqadd \i, \i, \i
sqadd \i, \i, v3.4s
.endr
.endm
.macro def_horz_16 scale=0, shift=2, suffix
function inv_txfm_horz\suffix\()_16x4_neon
mov x14, x30
movi v7.4s, #0
.if \scale
movz w16, #2896*8, lsl #16
dup v0.2s, w16
.endif
.irp i, v16.4s, v17.4s, v18.4s, v19.4s, v20.4s, v21.4s, v22.4s, v23.4s, v24.4s, v25.4s, v26.4s, v27.4s, v28.4s, v29.4s, v30.4s, v31.4s
ld1 {\i}, [x7]
st1 {v7.4s}, [x7], x8
.endr
.if \scale
scale_input .4s, v0.s[0], v16, v17, v18, v19, v20, v21, v22, v23
scale_input .4s, v0.s[0], v24, v25, v26, v27, v28, v29, v30, v31
.endif
blr x4
sqrshrn v16.4h, v16.4s, #\shift
sqrshrn v17.4h, v17.4s, #\shift
sqrshrn v18.4h, v18.4s, #\shift
sqrshrn v19.4h, v19.4s, #\shift
sqrshrn2 v16.8h, v20.4s, #\shift
sqrshrn2 v17.8h, v21.4s, #\shift
sqrshrn2 v18.8h, v22.4s, #\shift
sqrshrn2 v19.8h, v23.4s, #\shift
sqrshrn v20.4h, v24.4s, #\shift
sqrshrn v21.4h, v25.4s, #\shift
sqrshrn v22.4h, v26.4s, #\shift
sqrshrn v23.4h, v27.4s, #\shift
sqrshrn2 v20.8h, v28.4s, #\shift
sqrshrn2 v21.8h, v29.4s, #\shift
sqrshrn2 v22.8h, v30.4s, #\shift
sqrshrn2 v23.8h, v31.4s, #\shift
.if \scale
b L(horz_16x4_epilog)
.else
L(horz_16x4_epilog):
transpose_4x8h v16, v17, v18, v19, v4, v5, v6, v7
transpose_4x8h v20, v21, v22, v23, v4, v5, v6, v7
.irp i, v16.8h, v20.8h, v17.8h, v21.8h, v18.8h, v22.8h, v19.8h, v23.8h
st1 {\i}, [x6], #16
.endr
ret x14
.endif
endfunc
.endm
def_horz_16 scale=1, shift=1, suffix=_scale
def_horz_16 scale=0, shift=2
function inv_txfm_add_vert_8x16_neon
mov x14, x30
.irp i, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31
ld1 {v\i\().8h}, [x7], x8
.endr
blr x5
load_add_store_8x16 x6, x7
ret x14
endfunc
function inv_txfm_add_16x16_neon
mov x15, x30
sub sp, sp, #512
ldrh w12, [x13], #2
.irp i, 0, 4, 8, 12
add x6, sp, #(\i*16*2)
.if \i > 0
mov w8, #(16 - \i)
cmp w3, w12
b.lt 1f
.if \i < 12
ldrh w12, [x13], #2
.endif
.endif
add x7, x2, #(\i*4)
mov x8, #16*4
bl inv_txfm_horz_16x4_neon
.endr
b 3f
1:
movi v4.8h, #0
movi v5.8h, #0
movi v6.8h, #0
movi v7.8h, #0
2:
subs w8, w8, #4
.rept 2
st1 {v4.8h, v5.8h, v6.8h, v7.8h}, [x6], #64
.endr
b.gt 2b
3:
.irp i, 0, 8
add x6, x0, #(\i*2)
add x7, sp, #(\i*2)
mov x8, #32
bl inv_txfm_add_vert_8x16_neon
.endr
add sp, sp, #512
ret x15
endfunc
const eob_16x16
.short 10, 36, 78, 256
endconst
const eob_16x16_identity
.short 4, 8, 12, 256
endconst
.macro def_fn_16x16 txfm1, txfm2
function inv_txfm_add_\txfm1\()_\txfm2\()_16x16_16bpc_neon, export=1
.ifc \txfm1\()_\txfm2, dct_dct
idct_dc 16, 16, 2
.endif
adr x4, inv_\txfm1\()_4s_x16_neon
movrel x5, X(inv_\txfm2\()_8h_x16_neon)
.ifc \txfm1, identity
.ifc \txfm2, identity
movrel x13, eob_16x16
.else
movrel x13, eob_16x16_identity
.endif
.else
.ifc \txfm2, identity
movrel x13, eob_16x16_identity
.else
movrel x13, eob_16x16
.endif
.endif
b inv_txfm_add_16x16_neon
endfunc
.endm
def_fn_16x16 dct, dct
def_fn_16x16 identity, identity
def_fn_16x16 dct, adst
def_fn_16x16 dct, flipadst
def_fn_16x16 dct, identity
def_fn_16x16 adst, dct
def_fn_16x16 adst, adst
def_fn_16x16 adst, flipadst
def_fn_16x16 flipadst, dct
def_fn_16x16 flipadst, adst
def_fn_16x16 flipadst, flipadst
def_fn_16x16 identity, dct
function inv_txfm_add_16x4_neon
mov x15, x30
movi v4.4s, #0
.irp i, v16.4s, v17.4s, v18.4s, v19.4s, v20.4s, v21.4s, v22.4s, v23.4s, v24.4s, v25.4s, v26.4s, v27.4s, v28.4s, v29.4s, v30.4s, v31.4s
ld1 {\i}, [x2]
st1 {v4.4s}, [x2], #16
.endr
blr x4
sqrshrn v16.4h, v16.4s, #1
sqrshrn v17.4h, v17.4s, #1
sqrshrn v18.4h, v18.4s, #1
sqrshrn v19.4h, v19.4s, #1
sqrshrn2 v16.8h, v20.4s, #1
sqrshrn2 v17.8h, v21.4s, #1
sqrshrn2 v18.8h, v22.4s, #1
sqrshrn2 v19.8h, v23.4s, #1
transpose_4x8h v16, v17, v18, v19, v2, v3, v4, v5
blr x5
mov x6, x0
load_add_store_8x4 x6, x7
sqrshrn v16.4h, v24.4s, #1
sqrshrn v17.4h, v25.4s, #1
sqrshrn v18.4h, v26.4s, #1
sqrshrn v19.4h, v27.4s, #1
sqrshrn2 v16.8h, v28.4s, #1
sqrshrn2 v17.8h, v29.4s, #1
sqrshrn2 v18.8h, v30.4s, #1
sqrshrn2 v19.8h, v31.4s, #1
transpose_4x8h v16, v17, v18, v19, v2, v3, v4, v5
blr x5
add x6, x0, #16
load_add_store_8x4 x6, x7
ret x15
endfunc
function inv_txfm_add_4x16_neon
ldrh w12, [x13, #4]
mov x15, x30
mov x11, #64
cmp w3, w12
ldrh w12, [x13, #2]
b.lt 1f
add x6, x2, #48
movi v2.4s, #0
.irp i, v16.4s, v17.4s, v18.4s, v19.4s
ld1 {\i}, [x6]
st1 {v2.4s}, [x6], x11
.endr
blr x4
sqrshrn v28.4h, v16.4s, #1
sqrshrn v29.4h, v17.4s, #1
sqrshrn v30.4h, v18.4s, #1
sqrshrn v31.4h, v19.4s, #1
transpose_4x4h v28, v29, v30, v31, v4, v5, v6, v7
b 2f
1:
.irp i, v28.4h, v29.4h, v30.4h, v31.4h
movi \i, #0
.endr
2:
cmp w3, w12
ldrh w12, [x13, #0]
b.lt 1f
add x6, x2, #32
movi v2.4s, #0
.irp i, v16.4s, v17.4s, v18.4s, v19.4s
ld1 {\i}, [x6]
st1 {v2.4s}, [x6], x11
.endr
blr x4
sqrshrn v24.4h, v16.4s, #1
sqrshrn v25.4h, v17.4s, #1
sqrshrn v26.4h, v18.4s, #1
sqrshrn v27.4h, v19.4s, #1
transpose_4x4h v24, v25, v26, v27, v4, v5, v6, v7
b 2f
1:
.irp i, v24.4h, v25.4h, v26.4h, v27.4h
movi \i, #0
.endr
2:
cmp w3, w12
b.lt 1f
add x6, x2, #16
movi v2.4s, #0
.irp i, v16.4s, v17.4s, v18.4s, v19.4s
ld1 {\i}, [x6]
st1 {v2.4s}, [x6], x11
.endr
blr x4
sqrshrn v20.4h, v16.4s, #1
sqrshrn v21.4h, v17.4s, #1
sqrshrn v22.4h, v18.4s, #1
sqrshrn v23.4h, v19.4s, #1
transpose_4x4h v20, v21, v22, v23, v4, v5, v6, v7
b 2f
1:
.irp i, v20.4h, v21.4h, v22.4h, v23.4h
movi \i, #0
.endr
2:
movi v2.4s, #0
.irp i, v16.4s, v17.4s, v18.4s, v19.4s
ld1 {\i}, [x2]
st1 {v2.4s}, [x2], x11
.endr
blr x4
sqrshrn v16.4h, v16.4s, #1
sqrshrn v17.4h, v17.4s, #1
sqrshrn v18.4h, v18.4s, #1
sqrshrn v19.4h, v19.4s, #1
transpose_4x8h v16, v17, v18, v19, v4, v5, v6, v7
blr x5
load_add_store_4x16 x0, x6
ret x15
endfunc
const eob_4x16
.short 13, 29, 45, 64
endconst
const eob_4x16_identity1
.short 16, 32, 48, 64
endconst
const eob_4x16_identity2
.short 4, 8, 12, 64
endconst
.macro def_fn_416 w, h, txfm1, txfm2
function inv_txfm_add_\txfm1\()_\txfm2\()_\w\()x\h\()_16bpc_neon, export=1
.ifc \txfm1\()_\txfm2, dct_dct
idct_dc \w, \h, 1
.endif
.if \w == 4
adr x4, inv_\txfm1\()_4s_x\w\()_neon
movrel x5, X(inv_\txfm2\()_4h_x\h\()_neon)
.ifc \txfm1, identity
.ifc \txfm2, identity
movrel x13, eob_4x16
.else
movrel x13, eob_4x16_identity1
.endif
.else
.ifc \txfm2, identity
movrel x13, eob_4x16_identity2
.else
movrel x13, eob_4x16
.endif
.endif
.else
adr x4, inv_\txfm1\()_4s_x\w\()_neon
movrel x5, X(inv_\txfm2\()_8h_x\h\()_neon)
.endif
b inv_txfm_add_\w\()x\h\()_neon
endfunc
.endm
.macro def_fns_416 w, h
def_fn_416 \w, \h, dct, dct
def_fn_416 \w, \h, identity, identity
def_fn_416 \w, \h, dct, adst
def_fn_416 \w, \h, dct, flipadst
def_fn_416 \w, \h, dct, identity
def_fn_416 \w, \h, adst, dct
def_fn_416 \w, \h, adst, adst
def_fn_416 \w, \h, adst, flipadst
def_fn_416 \w, \h, flipadst, dct
def_fn_416 \w, \h, flipadst, adst
def_fn_416 \w, \h, flipadst, flipadst
def_fn_416 \w, \h, identity, dct
def_fn_416 \w, \h, adst, identity
def_fn_416 \w, \h, flipadst, identity
def_fn_416 \w, \h, identity, adst
def_fn_416 \w, \h, identity, flipadst
.endm
def_fns_416 4, 16
def_fns_416 16, 4
function inv_txfm_add_16x8_neon
mov x15, x30
stp d8, d9, [sp, #-0x40]!
stp d10, d11, [sp, #0x10]
stp d12, d13, [sp, #0x20]
stp d14, d15, [sp, #0x30]
cmp w3, w13
mov x11, #32
b.lt 1f
movi v4.4s, #0
movz w16, #2896*8, lsl #16
dup v0.2s, w16
add x6, x2, #16
--> --------------------
--> maximum size reached
--> --------------------
Messung V0.5 C=100 H=94 G=96