ih264_weighted_pred_av8.s revision a2b49e5f0574dee76f81507f288143d83a4b7c1a
1//******************************************************************************
2//*
3//* Copyright (C) 2015 The Android Open Source Project
4//*
5//* Licensed under the Apache License, Version 2.0 (the "License");
6//* you may not use this file except in compliance with the License.
7//* You may obtain a copy of the License at:
8//*
9//* http://www.apache.org/licenses/LICENSE-2.0
10//*
11//* Unless required by applicable law or agreed to in writing, software
12//* distributed under the License is distributed on an "AS IS" BASIS,
13//* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14//* See the License for the specific language governing permissions and
15//* limitations under the License.
16//*
17//*****************************************************************************
18//* Originally developed and contributed by Ittiam Systems Pvt. Ltd, Bangalore
19//*/
20///**
21//******************************************************************************
22//* @file
23//*  ih264_weighted_pred_av8.s
24//*
25//* @brief
26//*  Contains function definitions for weighted prediction.
27//* Functions are coded using NEON intrinsics and can be compiled using ARM RVCT
28//*
29//* @author
30//*  Kaushik Senthoor R
31//*
32//* @par List of Functions:
33//*
34//*  - ih264_weighted_pred_luma_av8()
35//*  - ih264_weighted_pred_chroma_av8()
36//*
37//* @remarks
38//*  None
39//*
40//*******************************************************************************
41//*/
42//*******************************************************************************
43//* @function
44//*  ih264_weighted_pred_luma_av8()
45//*
46//* @brief
47//*  This routine performs the default weighted prediction as described in sec
48//* 8.4.2.3.2 titled "Weighted sample prediction process" for luma.
49//*
50//* @par Description:
51//*  This function gets a ht x wd block, calculates the weighted sample, rounds
52//* off, adds offset and stores it in the destination block.
53//*
54//* @param[in] puc_src:
55//*  UWORD8 Pointer to the buffer containing the input block.
56//*
57//* @param[out] puc_dst
58//*  UWORD8 pointer to the destination where the output block is stored.
59//*
60//* @param[in] src_strd
61//*  Stride of the input buffer
62//*
63//* @param[in] dst_strd
64//*  Stride of the destination buffer
65//*
66//* @param[in] log_WD
67//*  number of bits to be rounded off
68//*
69//* @param[in] wt
70//*  weight for the weighted prediction
71//*
72//* @param[in] ofst
73//*  offset used after rounding off
74//*
75//* @param[in] ht
76//*  integer height of the array
77//*
78//* @param[in] wd
79//*  integer width of the array
80//*
81//* @returns
82//*  None
83//*
84//* @remarks
85//*  (ht,wd) can be (4,4), (4,8), (8,4), (8,8), (8,16), (16,8) or (16,16).
86//*
87//*******************************************************************************
88//*/
89//void ih264_weighted_pred_luma_av8(UWORD8 *puc_src,
90//                                  UWORD8 *puc_dst,
91//                                  WORD32 src_strd,
92//                                  WORD32 dst_strd,
93//                                  UWORD8 log_WD,
94//                                  UWORD32 wt,
95//                                  UWORD16 ofst,
96//                                  UWORD8 ht,
97//                                  UWORD8 wd)
98//
99//**************Variables Vs Registers*****************************************
100//    x0      => puc_src
101//    x1      => puc_dst
102//    x2      => src_strd
103//    x3      => dst_strd
104//    [sp]    => log_WD (x4)
105//    [sp+4]  => wt     (x5)
106//   [sp+8]  => ofst   (x6)
107//    [sp+12] => ht     (x7)
108//    [sp+16] => wd     (x8)
109//
110.text
111.p2align 2
112.include "ih264_neon_macros.s"
113
114
115
116    .global ih264_weighted_pred_luma_av8
117
118ih264_weighted_pred_luma_av8:
119
120    // STMFD sp!, {x4-x9,x14}                //stack stores the values of the arguments
121    push_v_regs
122    stp       x19, x20, [sp, #-16]!
123    ldr       w8, [sp, #80]             //Load wd
124    sxtw      x8, w8
125
126    dup       v2.4h, w5                 //D2 = wt (16-bit)
127    sub       x20, x4, #0               //x9 = -log_WD
128    neg       x9, x20
129    dup       v3.8b, w6                 //D3 = ofst (8-bit)
130    cmp       w8, #16                   //check if wd is 16
131    dup       v0.8h, w9                 //Q0 = -log_WD (16-bit)
132    beq       loop_16                   //branch if wd is 16
133
134    cmp       w8, #8                    //check if wd is 8
135    beq       loop_8                    //branch if wd is 8
136
137loop_4:                                 //each iteration processes four rows
138
139    ld1       {v4.s}[0], [x0], x2       //load row 1 in source
140    ld1       {v4.s}[1], [x0], x2       //load row 2 in source
141    ld1       {v6.s}[0], [x0], x2       //load row 3 in source
142    ld1       {v6.s}[1], [x0], x2       //load row 4 in source
143
144    uxtl      v4.8h, v4.8b              //converting rows 1,2 to 16-bit
145    uxtl      v6.8h, v6.8b              //converting rows 3,4 to 16-bit
146
147    mul       v4.8h, v4.8h , v2.4h[0]   //weight mult. for rows 1,2
148    mul       v6.8h, v6.8h , v2.4h[0]   //weight mult. for rows 3,4
149
150    subs      w7, w7, #4                //decrement ht by 4
151    srshl     v4.8h, v4.8h , v0.8h      //rounds off the weighted samples from rows 1,2
152    srshl     v6.8h, v6.8h , v0.8h      //rounds off the weighted samples from rows 3,4
153
154    saddw     v4.8h, v4.8h , v3.8b      //adding offset for rows 1,2
155    saddw     v6.8h, v6.8h , v3.8b      //adding offset for rows 3,4
156
157    sqxtun    v4.8b, v4.8h              //saturating rows 1,2 to unsigned 8-bit
158    sqxtun    v6.8b, v6.8h              //saturating rows 3,4 to unsigned 8-bit
159
160    st1       {v4.s}[0], [x1], x3       //store row 1 in destination
161    st1       {v4.s}[1], [x1], x3       //store row 2 in destination
162    st1       {v6.s}[0], [x1], x3       //store row 3 in destination
163    st1       {v6.s}[1], [x1], x3       //store row 4 in destination
164
165    bgt       loop_4                    //if greater than 0 repeat the loop again
166
167    b         end_loops
168
169loop_8:                                 //each iteration processes four rows
170
171    ld1       {v4.8b}, [x0], x2         //load row 1 in source
172    ld1       {v6.8b}, [x0], x2         //load row 2 in source
173    ld1       {v8.8b}, [x0], x2         //load row 3 in source
174    uxtl      v4.8h, v4.8b              //converting row 1 to 16-bit
175    ld1       {v10.8b}, [x0], x2        //load row 4 in source
176    uxtl      v6.8h, v6.8b              //converting row 2 to 16-bit
177
178    uxtl      v8.8h, v8.8b              //converting row 3 to 16-bit
179    mul       v4.8h, v4.8h , v2.4h[0]   //weight mult. for row 1
180    uxtl      v10.8h, v10.8b            //converting row 4 to 16-bit
181    mul       v6.8h, v6.8h , v2.4h[0]   //weight mult. for row 2
182    mul       v8.8h, v8.8h , v2.4h[0]   //weight mult. for row 3
183    mul       v10.8h, v10.8h , v2.4h[0] //weight mult. for row 4
184
185    srshl     v4.8h, v4.8h , v0.8h      //rounds off the weighted samples from row 1
186    srshl     v6.8h, v6.8h , v0.8h      //rounds off the weighted samples from row 2
187    srshl     v8.8h, v8.8h , v0.8h      //rounds off the weighted samples from row 3
188    saddw     v4.8h, v4.8h , v3.8b      //adding offset for row 1
189    srshl     v10.8h, v10.8h , v0.8h    //rounds off the weighted samples from row 4
190    saddw     v6.8h, v6.8h , v3.8b      //adding offset for row 2
191
192    saddw     v8.8h, v8.8h , v3.8b      //adding offset for row 3
193    sqxtun    v4.8b, v4.8h              //saturating row 1 to unsigned 8-bit
194    saddw     v10.8h, v10.8h , v3.8b    //adding offset for row 4
195    sqxtun    v6.8b, v6.8h              //saturating row 2 to unsigned 8-bit
196    sqxtun    v8.8b, v8.8h              //saturating row 3 to unsigned 8-bit
197    sqxtun    v10.8b, v10.8h            //saturating row 4 to unsigned 8-bit
198
199    st1       {v4.8b}, [x1], x3         //store row 1 in destination
200    st1       {v6.8b}, [x1], x3         //store row 2 in destination
201    subs      w7, w7, #4                //decrement ht by 4
202    st1       {v8.8b}, [x1], x3         //store row 3 in destination
203    st1       {v10.8b}, [x1], x3        //store row 4 in destination
204
205    bgt       loop_8                    //if greater than 0 repeat the loop again
206
207    b         end_loops
208
209loop_16:                                //each iteration processes two rows
210
211    ld1       {v4.8b, v5.8b}, [x0], x2  //load row 1 in source
212    ld1       {v6.8b, v7.8b}, [x0], x2  //load row 2 in source
213    uxtl      v12.8h, v4.8b             //converting row 1L to 16-bit
214    ld1       {v8.8b, v9.8b}, [x0], x2  //load row 3 in source
215    uxtl      v14.8h, v5.8b             //converting row 1H to 16-bit
216    ld1       {v10.8b, v11.8b}, [x0], x2 //load row 4 in source
217    uxtl      v16.8h, v6.8b             //converting row 2L to 16-bit
218    mul       v12.8h, v12.8h , v2.4h[0] //weight mult. for row 1L
219    uxtl      v18.8h, v7.8b             //converting row 2H to 16-bit
220    mul       v14.8h, v14.8h , v2.4h[0] //weight mult. for row 1H
221    uxtl      v20.8h, v8.8b             //converting row 3L to 16-bit
222    mul       v16.8h, v16.8h , v2.4h[0] //weight mult. for row 2L
223    uxtl      v22.8h, v9.8b             //converting row 3H to 16-bit
224    mul       v18.8h, v18.8h , v2.4h[0] //weight mult. for row 2H
225    uxtl      v24.8h, v10.8b            //converting row 4L to 16-bit
226    mul       v20.8h, v20.8h , v2.4h[0] //weight mult. for row 3L
227    uxtl      v26.8h, v11.8b            //converting row 4H to 16-bit
228    mul       v22.8h, v22.8h , v2.4h[0] //weight mult. for row 3H
229    mul       v24.8h, v24.8h , v2.4h[0] //weight mult. for row 4L
230    srshl     v12.8h, v12.8h , v0.8h    //rounds off the weighted samples from row 1L
231    mul       v26.8h, v26.8h , v2.4h[0] //weight mult. for row 4H
232    srshl     v14.8h, v14.8h , v0.8h    //rounds off the weighted samples from row 1H
233    srshl     v16.8h, v16.8h , v0.8h    //rounds off the weighted samples from row 2L
234    saddw     v12.8h, v12.8h , v3.8b    //adding offset for row 1L
235    srshl     v18.8h, v18.8h , v0.8h    //rounds off the weighted samples from row 2H
236    saddw     v14.8h, v14.8h , v3.8b    //adding offset for row 1H
237    sqxtun    v4.8b, v12.8h             //saturating row 1L to unsigned 8-bit
238    srshl     v20.8h, v20.8h , v0.8h    //rounds off the weighted samples from row 3L
239    saddw     v16.8h, v16.8h , v3.8b    //adding offset for row 2L
240    sqxtun    v5.8b, v14.8h             //saturating row 1H to unsigned 8-bit
241    srshl     v22.8h, v22.8h , v0.8h    //rounds off the weighted samples from row 3H
242    saddw     v18.8h, v18.8h , v3.8b    //adding offset for row 2H
243    sqxtun    v6.8b, v16.8h             //saturating row 2L to unsigned 8-bit
244    srshl     v24.8h, v24.8h , v0.8h    //rounds off the weighted samples from row 4L
245    saddw     v20.8h, v20.8h , v3.8b    //adding offset for row 3L
246    sqxtun    v7.8b, v18.8h             //saturating row 2H to unsigned 8-bit
247    srshl     v26.8h, v26.8h , v0.8h    //rounds off the weighted samples from row 4H
248    saddw     v22.8h, v22.8h , v3.8b    //adding offset for row 3H
249    sqxtun    v8.8b, v20.8h             //saturating row 3L to unsigned 8-bit
250    saddw     v24.8h, v24.8h , v3.8b    //adding offset for row 4L
251    sqxtun    v9.8b, v22.8h             //saturating row 3H to unsigned 8-bit
252    saddw     v26.8h, v26.8h , v3.8b    //adding offset for row 4H
253    sqxtun    v10.8b, v24.8h            //saturating row 4L to unsigned 8-bit
254    st1       {v4.8b, v5.8b}, [x1], x3  //store row 1 in destination
255    sqxtun    v11.8b, v26.8h            //saturating row 4H to unsigned 8-bit
256    st1       {v6.8b, v7.8b}, [x1], x3  //store row 2 in destination
257    subs      w7, w7, #4                //decrement ht by 4
258    st1       {v8.8b, v9.8b}, [x1], x3  //store row 3 in destination
259    st1       {v10.8b, v11.8b}, [x1], x3 //store row 4 in destination
260
261    bgt       loop_16                   //if greater than 0 repeat the loop again
262
263end_loops:
264
265    // LDMFD sp!,{x4-x9,x15}                      //Reload the registers from sp
266    ldp       x19, x20, [sp], #16
267    pop_v_regs
268    ret
269
270
271//*******************************************************************************
272//* @function
273//*  ih264_weighted_pred_chroma_av8()
274//*
275//* @brief
276//*  This routine performs the default weighted prediction as described in sec
277//* 8.4.2.3.2 titled "Weighted sample prediction process" for chroma.
278//*
279//* @par Description:
280//*  This function gets a ht x wd block, calculates the weighted sample, rounds
281//* off, adds offset and stores it in the destination block for U and V.
282//*
283//* @param[in] puc_src:
284//*  UWORD8 Pointer to the buffer containing the input block.
285//*
286//* @param[out] puc_dst
287//*  UWORD8 pointer to the destination where the output block is stored.
288//*
289//* @param[in] src_strd
290//*  Stride of the input buffer
291//*
292//* @param[in] dst_strd
293//*  Stride of the destination buffer
294//*
295//* @param[in] log_WD
296//*  number of bits to be rounded off
297//*
298//* @param[in] wt
299//*  weights for the weighted prediction for U and V
300//*
301//* @param[in] ofst
302//*  offsets used after rounding off for U and V
303//*
304//* @param[in] ht
305//*  integer height of the array
306//*
307//* @param[in] wd
308//*  integer width of the array
309//*
310//* @returns
311//*  None
312//*
313//* @remarks
314//*  (ht,wd) can be (2,2), (2,4), (4,2), (4,4), (4,8), (8,4) or (8,8).
315//*
316//*******************************************************************************
317//*/
318//void ih264_weighted_pred_chroma_av8(UWORD8 *puc_src,
319//                                    UWORD8 *puc_dst,
320//                                    WORD32 src_strd,
321//                                    WORD32 dst_strd,
322//                                    UWORD8 log_WD,
323//                                    UWORD32 wt,
324//                                    UWORD16 ofst,
325//                                    UWORD8 ht,
326//                                    UWORD8 wd)
327//
328//**************Variables Vs Registers*****************************************
329//    x0      => puc_src
330//    x1      => puc_dst
331//    x2      => src_strd
332//    x3      => dst_strd
333//    [sp]    => log_WD (x4)
334//    [sp+4]  => wt     (x5)
335//   [sp+8]  => ofst   (x6)
336//    [sp+12] => ht     (x7)
337//    [sp+16] => wd     (x8)
338//
339
340
341
342
343    .global ih264_weighted_pred_chroma_av8
344
345ih264_weighted_pred_chroma_av8:
346
347    // STMFD sp!, {x4-x9,x14}                //stack stores the values of the arguments
348    push_v_regs
349    stp       x19, x20, [sp, #-16]!
350
351    ldr       w8, [sp, #80]             //Load wd
352    sxtw      x8, w8
353
354    sub       x20, x4, #0               //x9 = -log_WD
355    neg       x9, x20
356    dup       v2.4s, w5                 //Q1 = {wt_u (16-bit), wt_v (16-bit)}
357
358
359    dup       v4.4h, w6                 //D4 = {ofst_u (8-bit), ofst_v (8-bit)}
360    cmp       w8, #8                    //check if wd is 8
361    dup       v0.8h, w9                 //Q0 = -log_WD (16-bit)
362    beq       loop_8_uv                 //branch if wd is 8
363
364    cmp       w8, #4                    //check if ws is 4
365    beq       loop_4_uv                 //branch if wd is 4
366
367loop_2_uv:                              //each iteration processes two rows
368
369    ld1       {v6.s}[0], [x0], x2       //load row 1 in source
370    ld1       {v6.s}[1], [x0], x2       //load row 2 in source
371    uxtl      v6.8h, v6.8b              //converting rows 1,2 to 16-bit
372    mul       v6.8h, v6.8h , v2.8h      //weight mult. for rows 1,2
373    srshl     v6.8h, v6.8h , v0.8h      //rounds off the weighted samples from rows 1,2
374    saddw     v6.8h, v6.8h , v4.8b      //adding offset for rows 1,2
375    sqxtun    v6.8b, v6.8h              //saturating rows 1,2 to unsigned 8-bit
376    subs      w7, w7, #2                //decrement ht by 2
377    st1       {v6.s}[0], [x1], x3       //store row 1 in destination
378    st1       {v6.s}[1], [x1], x3       //store row 2 in destination
379    bgt       loop_2_uv                 //if greater than 0 repeat the loop again
380    b         end_loops_uv
381
382loop_4_uv:                              //each iteration processes two rows
383
384    ld1       {v6.8b}, [x0], x2         //load row 1 in source
385    ld1       {v8.8b}, [x0], x2         //load row 2 in source
386    uxtl      v6.8h, v6.8b              //converting row 1 to 16-bit
387    uxtl      v8.8h, v8.8b              //converting row 2 to 16-bit
388    mul       v6.8h, v6.8h , v2.8h      //weight mult. for row 1
389    mul       v8.8h, v8.8h , v2.8h      //weight mult. for row 2
390    subs      w7, w7, #2                //decrement ht by 2
391    srshl     v6.8h, v6.8h , v0.8h      //rounds off the weighted samples from row 1
392    srshl     v8.8h, v8.8h , v0.8h      //rounds off the weighted samples from row 2
393    saddw     v6.8h, v6.8h , v4.8b      //adding offset for row 1
394    saddw     v8.8h, v8.8h , v4.8b      //adding offset for row 2
395    sqxtun    v6.8b, v6.8h              //saturating row 1 to unsigned 8-bit
396    sqxtun    v8.8b, v8.8h              //saturating row 2 to unsigned 8-bit
397    st1       {v6.8b}, [x1], x3         //store row 1 in destination
398    st1       {v8.8b}, [x1], x3         //store row 2 in destination
399
400    bgt       loop_4_uv                 //if greater than 0 repeat the loop again
401
402    b         end_loops_uv
403
404loop_8_uv:                              //each iteration processes two rows
405
406    ld1       {v6.8b, v7.8b}, [x0], x2  //load row 1 in source
407    ld1       {v8.8b, v9.8b}, [x0], x2  //load row 2 in source
408    uxtl      v14.8h, v6.8b             //converting row 1L to 16-bit
409    ld1       {v10.8b, v11.8b}, [x0], x2 //load row 3 in source
410    uxtl      v16.8h, v7.8b             //converting row 1H to 16-bit
411    ld1       {v12.8b, v13.8b}, [x0], x2 //load row 4 in source
412
413    mul       v14.8h, v14.8h , v2.8h    //weight mult. for row 1L
414    uxtl      v18.8h, v8.8b             //converting row 2L to 16-bit
415    mul       v16.8h, v16.8h , v2.8h    //weight mult. for row 1H
416    uxtl      v20.8h, v9.8b             //converting row 2H to 16-bit
417    mul       v18.8h, v18.8h , v2.8h    //weight mult. for row 2L
418    uxtl      v22.8h, v10.8b            //converting row 3L to 16-bit
419    mul       v20.8h, v20.8h , v2.8h    //weight mult. for row 2H
420    uxtl      v24.8h, v11.8b            //converting row 3H to 16-bit
421    mul       v22.8h, v22.8h , v2.8h    //weight mult. for row 3L
422    uxtl      v26.8h, v12.8b            //converting row 4L to 16-bit
423    mul       v24.8h, v24.8h , v2.8h    //weight mult. for row 3H
424    uxtl      v28.8h, v13.8b            //converting row 4H to 16-bit
425
426    mul       v26.8h, v26.8h , v2.8h    //weight mult. for row 4L
427    srshl     v14.8h, v14.8h , v0.8h    //rounds off the weighted samples from row 1L
428    mul       v28.8h, v28.8h , v2.8h    //weight mult. for row 4H
429
430    srshl     v16.8h, v16.8h , v0.8h    //rounds off the weighted samples from row 1H
431    srshl     v18.8h, v18.8h , v0.8h    //rounds off the weighted samples from row 2L
432    saddw     v14.8h, v14.8h , v4.8b    //adding offset for row 1L
433    srshl     v20.8h, v20.8h , v0.8h    //rounds off the weighted samples from row 2H
434    saddw     v16.8h, v16.8h , v4.8b    //adding offset for row 1H
435    sqxtun    v6.8b, v14.8h             //saturating row 1L to unsigned 8-bit
436    srshl     v22.8h, v22.8h , v0.8h    //rounds off the weighted samples from row 3L
437    saddw     v18.8h, v18.8h , v4.8b    //adding offset for row 2L
438    sqxtun    v7.8b, v16.8h             //saturating row 1H to unsigned 8-bit
439    srshl     v24.8h, v24.8h , v0.8h    //rounds off the weighted samples from row 3H
440    saddw     v20.8h, v20.8h , v4.8b    //adding offset for row 2H
441    sqxtun    v8.8b, v18.8h             //saturating row 2L to unsigned 8-bit
442    srshl     v26.8h, v26.8h , v0.8h    //rounds off the weighted samples from row 4L
443    saddw     v22.8h, v22.8h , v4.8b    //adding offset for row 3L
444    sqxtun    v9.8b, v20.8h             //saturating row 2H to unsigned 8-bit
445    srshl     v28.8h, v28.8h , v0.8h    //rounds off the weighted samples from row 4H
446    saddw     v24.8h, v24.8h , v4.8b    //adding offset for row 3H
447
448    sqxtun    v10.8b, v22.8h            //saturating row 3L to unsigned 8-bit
449    saddw     v26.8h, v26.8h , v4.8b    //adding offset for row 4L
450    sqxtun    v11.8b, v24.8h            //saturating row 3H to unsigned 8-bit
451    saddw     v28.8h, v28.8h , v4.8b    //adding offset for row 4H
452
453    sqxtun    v12.8b, v26.8h            //saturating row 4L to unsigned 8-bit
454    st1       {v6.8b, v7.8b}, [x1], x3  //store row 1 in destination
455    sqxtun    v13.8b, v28.8h            //saturating row 4H to unsigned 8-bit
456    st1       {v8.8b, v9.8b}, [x1], x3  //store row 2 in destination
457    subs      w7, w7, #4                //decrement ht by 4
458    st1       {v10.8b, v11.8b}, [x1], x3 //store row 3 in destination
459    st1       {v12.8b, v13.8b}, [x1], x3 //store row 4 in destination
460
461    bgt       loop_8_uv                 //if greater than 0 repeat the loop again
462
463end_loops_uv:
464
465    // LDMFD sp!,{x4-x9,x15}                      //Reload the registers from sp
466    ldp       x19, x20, [sp], #16
467    pop_v_regs
468    ret
469
470
471
472