History

lavc/aarch64: new optimization for 8-bit hevc_epel_v

checkasm bench: put_hevc_epel_v4_8_c: 79.9 put_hevc_epel_v4_8_neon: 25.7 put_hevc_epel_v6_8_c: 151.4 put_hevc_epel_v6_8_neon: 46.4 put_hevc_epel_v8_8_c: 250.9 put_hevc_epel_v8_8_neon: 41.7 put_hevc_epel_v12_8_c: 542.7 put_hevc_epel_v12_8_neon: 108.7 put_hevc_epel_v16_8_c: 939.4 put_hevc_epel_v16_8_neon: 169.2 put_hevc_epel_v24_8_c: 2104.9 put_hevc_epel_v24_8_neon: 307.9 put_hevc_epel_v32_8_c: 3713.9 put_hevc_epel_v32_8_neon: 524.2 put_hevc_epel_v48_8_c: 8175.2 put_hevc_epel_v48_8_neon: 1197.2 put_hevc_epel_v64_8_c: 16049.4 put_hevc_epel_v64_8_neon: 2094.9

Logan Lyu committed 3 years ago

Commit 22c72911 parent 77286572 files · +228 −0Browse files

Showing the changes to libavcodec (2 of 2 files). Show all

Unified · Split

2 files
  1. libavcodec/aarch64/hevcdsp_epel_neon.S +223 −0
  2. libavcodec/aarch64/hevcdsp_init_aarch64.c +5 −0
libavcodec/aarch64/hevcdsp_epel_neon.S+223−0
@@ -243,6 +243,229 @@ function ff_hevc_put_hevc_pel_pixels64_8_neon, export=1 · 242 unmodified lines
243243 ret
244244 endfunc
245245
246+
247+function ff_hevc_put_hevc_epel_v4_8_neon, export=1
248+ load_epel_filterb x5, x4
249+ sub x1, x1, x2
250+ mov x10, #(MAX_PB_SIZE * 2)
251+ ldr s16, [x1]
252+ ldr s17, [x1, x2]
253+ add x1, x1, x2, lsl #1
254+ ld1 {v18.s}[0], [x1], x2
255+.macro calc src0, src1, src2, src3
256+ ld1 {\src3\().s}[0], [x1], x2
257+ movi v4.8h, #0
258+ calc_epelb v4, \src0, \src1, \src2, \src3
259+ subs w3, w3, #1
260+ st1 {v4.4h}, [x0], x10
261+.endm
262+1: calc_all4
263+.purgem calc
264+2: ret
265+endfunc
266+
267+function ff_hevc_put_hevc_epel_v6_8_neon, export=1
268+ load_epel_filterb x5, x4
269+ sub x1, x1, x2
270+ mov x10, #(MAX_PB_SIZE * 2 - 8)
271+ ldr d16, [x1]
272+ ldr d17, [x1, x2]
273+ add x1, x1, x2, lsl #1
274+ ld1 {v18.8b}, [x1], x2
275+.macro calc src0, src1, src2, src3
276+ ld1 {\src3\().8b}, [x1], x2
277+ movi v4.8h, #0
278+ calc_epelb v4, \src0, \src1, \src2, \src3
279+ st1 {v4.d}[0], [x0], #8
280+ subs w3, w3, #1
281+ st1 {v4.s}[2], [x0], x10
282+.endm
283+1: calc_all4
284+.purgem calc
285+2: ret
286+endfunc
287+
288+function ff_hevc_put_hevc_epel_v8_8_neon, export=1
289+ load_epel_filterb x5, x4
290+ sub x1, x1, x2
291+ mov x10, #(MAX_PB_SIZE * 2)
292+ ldr d16, [x1]
293+ ldr d17, [x1, x2]
294+ add x1, x1, x2, lsl #1
295+ ld1 {v18.8b}, [x1], x2
296+.macro calc src0, src1, src2, src3
297+ ld1 {\src3\().8b}, [x1], x2
298+ movi v4.8h, #0
299+ calc_epelb v4, \src0, \src1, \src2, \src3
300+ subs w3, w3, #1
301+ st1 {v4.8h}, [x0], x10
302+.endm
303+1: calc_all4
304+.purgem calc
305+2: ret
306+endfunc
307+
308+function ff_hevc_put_hevc_epel_v12_8_neon, export=1
309+ load_epel_filterb x5, x4
310+ sub x1, x1, x2
311+ mov x10, #(MAX_PB_SIZE * 2)
312+ ldr q16, [x1]
313+ ldr q17, [x1, x2]
314+ add x1, x1, x2, lsl #1
315+ ld1 {v18.16b}, [x1], x2
316+.macro calc src0, src1, src2, src3
317+ ld1 {\src3\().16b}, [x1], x2
318+ movi v4.8h, #0
319+ movi v5.8h, #0
320+ calc_epelb v4, \src0, \src1, \src2, \src3
321+ calc_epelb2 v5, \src0, \src1, \src2, \src3
322+ str q4, [x0]
323+ subs w3, w3, #1
324+ str d5, [x0, #16]
325+ add x0, x0, x10
326+.endm
327+1: calc_all4
328+.purgem calc
329+2: ret
330+endfunc
331+
332+function ff_hevc_put_hevc_epel_v16_8_neon, export=1
333+ load_epel_filterb x5, x4
334+ sub x1, x1, x2
335+ mov x10, #(MAX_PB_SIZE * 2)
336+ ldr q16, [x1]
337+ ldr q17, [x1, x2]
338+ add x1, x1, x2, lsl #1
339+ ld1 {v18.16b}, [x1], x2
340+.macro calc src0, src1, src2, src3
341+ ld1 {\src3\().16b}, [x1], x2
342+ movi v4.8h, #0
343+ movi v5.8h, #0
344+ calc_epelb v4, \src0, \src1, \src2, \src3
345+ calc_epelb2 v5, \src0, \src1, \src2, \src3
346+ subs w3, w3, #1
347+ st1 {v4.8h, v5.8h}, [x0], x10
348+.endm
349+1: calc_all4
350+.purgem calc
351+2: ret
352+endfunc
353+
354+function ff_hevc_put_hevc_epel_v24_8_neon, export=1
355+ load_epel_filterb x5, x4
356+ sub x1, x1, x2
357+ mov x10, #(MAX_PB_SIZE * 2)
358+ ld1 {v16.8b, v17.8b, v18.8b}, [x1], x2
359+ ld1 {v19.8b, v20.8b, v21.8b}, [x1], x2
360+ ld1 {v22.8b, v23.8b, v24.8b}, [x1], x2
361+.macro calc src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10, src11
362+ ld1 {\src9\().8b, \src10\().8b, \src11\().8b}, [x1], x2
363+ movi v4.8h, #0
364+ movi v5.8h, #0
365+ movi v6.8h, #0
366+ calc_epelb v4, \src0, \src3, \src6, \src9
367+ calc_epelb v5, \src1, \src4, \src7, \src10
368+ calc_epelb v6, \src2, \src5, \src8, \src11
369+ subs w3, w3, #1
370+ st1 {v4.8h-v6.8h}, [x0], x10
371+.endm
372+1: calc_all12
373+.purgem calc
374+2: ret
375+endfunc
376+
377+function ff_hevc_put_hevc_epel_v32_8_neon, export=1
378+ load_epel_filterb x5, x4
379+ sub x1, x1, x2
380+ mov x10, #(MAX_PB_SIZE * 2)
381+ ld1 {v16.16b, v17.16b}, [x1], x2
382+ ld1 {v18.16b, v19.16b}, [x1], x2
383+ ld1 {v20.16b, v21.16b}, [x1], x2
384+.macro calc src0, src1, src2, src3, src4, src5, src6, src7
385+ ld1 {\src6\().16b, \src7\().16b}, [x1], x2
386+ movi v4.8h, #0
387+ movi v5.8h, #0
388+ movi v6.8h, #0
389+ movi v7.8h, #0
390+ calc_epelb v4, \src0, \src2, \src4, \src6
391+ calc_epelb2 v5, \src0, \src2, \src4, \src6
392+ calc_epelb v6, \src1, \src3, \src5, \src7
393+ calc_epelb2 v7, \src1, \src3, \src5, \src7
394+ subs w3, w3, #1
395+ st1 {v4.8h-v7.8h}, [x0], x10
396+.endm
397+1: calc_all8
398+.purgem calc
399+2: ret
400+endfunc
401+
402+function ff_hevc_put_hevc_epel_v48_8_neon, export=1
403+ load_epel_filterb x5, x4
404+ sub x1, x1, x2
405+ mov x10, #64
406+ ld1 {v16.16b, v17.16b, v18.16b}, [x1], x2
407+ ld1 {v19.16b, v20.16b, v21.16b}, [x1], x2
408+ ld1 {v22.16b, v23.16b, v24.16b}, [x1], x2
409+.macro calc src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10, src11
410+ ld1 {\src9\().16b, \src10\().16b, \src11\().16b}, [x1], x2
411+ movi v4.8h, #0
412+ movi v5.8h, #0
413+ movi v6.8h, #0
414+ movi v7.8h, #0
415+ movi v28.8h, #0
416+ movi v29.8h, #0
417+ calc_epelb v4, \src0, \src3, \src6, \src9
418+ calc_epelb2 v5, \src0, \src3, \src6, \src9
419+ calc_epelb v6, \src1, \src4, \src7, \src10
420+ calc_epelb2 v7, \src1, \src4, \src7, \src10
421+ calc_epelb v28, \src2, \src5, \src8, \src11
422+ calc_epelb2 v29, \src2, \src5, \src8, \src11
423+ st1 {v4.8h-v7.8h}, [x0], #64
424+ subs w3, w3, #1
425+ st1 {v28.8h-v29.8h}, [x0], x10
426+.endm
427+1: calc_all12
428+.purgem calc
429+2: ret
430+endfunc
431+
432+function ff_hevc_put_hevc_epel_v64_8_neon, export=1
433+ load_epel_filterb x5, x4
434+ sub sp, sp, #32
435+ st1 {v8.8b-v11.8b}, [sp]
436+ sub x1, x1, x2
437+ ld1 {v16.16b, v17.16b, v18.16b, v19.16b}, [x1], x2
438+ ld1 {v20.16b, v21.16b, v22.16b, v23.16b}, [x1], x2
439+ ld1 {v24.16b, v25.16b, v26.16b, v27.16b}, [x1], x2
440+.macro calc src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10, src11, src12, src13, src14, src15
441+ ld1 {\src12\().16b-\src15\().16b}, [x1], x2
442+ movi v4.8h, #0
443+ movi v5.8h, #0
444+ movi v6.8h, #0
445+ movi v7.8h, #0
446+ movi v8.8h, #0
447+ movi v9.8h, #0
448+ movi v10.8h, #0
449+ movi v11.8h, #0
450+ calc_epelb v4, \src0, \src4, \src8, \src12
451+ calc_epelb2 v5, \src0, \src4, \src8, \src12
452+ calc_epelb v6, \src1, \src5, \src9, \src13
453+ calc_epelb2 v7, \src1, \src5, \src9, \src13
454+ calc_epelb v8, \src2, \src6, \src10, \src14
455+ calc_epelb2 v9, \src2, \src6, \src10, \src14
456+ calc_epelb v10, \src3, \src7, \src11, \src15
457+ calc_epelb2 v11, \src3, \src7, \src11, \src15
458+ st1 {v4.8h-v7.8h}, [x0], #64
459+ subs w3, w3, #1
460+ st1 {v8.8h-v11.8h}, [x0], #64
461+.endm
462+1: calc_all16
463+.purgem calc
464+2: ld1 {v8.8b-v11.8b}, [sp]
465+ add sp, sp, #32
466+ ret
467+endfunc
468+
246469 function ff_hevc_put_hevc_epel_uni_v4_8_neon, export=1
247470 load_epel_filterb x6, x5
248471 sub x2, x2, x3
libavcodec/aarch64/hevcdsp_init_aarch64.c+5−0
@@ -156,6 +156,10 @@ NEON8_FNPROTO(pel_pixels, (int16_t *dst, · 155 unmodified lines
156156 const uint8_t *src, ptrdiff_t srcstride,
157157 int height, intptr_t mx, intptr_t my, int width),);
158158
159+NEON8_FNPROTO(epel_v, (int16_t *dst,
160+ const uint8_t *src, ptrdiff_t srcstride,
161+ int height, intptr_t mx, intptr_t my, int width),);
162+
159163 NEON8_FNPROTO(pel_uni_pixels, (uint8_t *_dst, ptrdiff_t _dststride,
160164 const uint8_t *_src, ptrdiff_t _srcstride,
161165 int height, intptr_t mx, intptr_t my, int width),);
@@ -305,6 +309,7 @@ av_cold void ff_hevc_dsp_init_aarch64(HEVCDSPContext *c, const int bit_depth) · 143 unmodified lines
305309 c->put_hevc_qpel_bi[9][0][1] = ff_hevc_put_hevc_qpel_bi_h16_8_neon;
306310
307311 NEON8_FNASSIGN(c->put_hevc_epel, 0, 0, pel_pixels,);
312+ NEON8_FNASSIGN(c->put_hevc_epel, 1, 0, epel_v,);
308313 NEON8_FNASSIGN(c->put_hevc_qpel, 0, 0, pel_pixels,);
309314 NEON8_FNASSIGN(c->put_hevc_epel_uni, 0, 0, pel_uni_pixels,);
310315 NEON8_FNASSIGN(c->put_hevc_epel_uni, 1, 0, epel_uni_v,);

Keyboard shortcuts

General

Show keyboard shortcuts
?
Focus the current search
/

Issues and pull requests

Open filters
f

Issues, pull requests and Inbox

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

Repository code

Go to file
t