max/x264
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

pixel: Add neon ssd implementations for 10 bit

Browse Source 
Provide arm64 neon implementation for ssd functions for 10 bit
depth. Benchmarks are shown below.

ssd_4x4_c: 1466
ssd_4x4_neon: 240
ssd_4x8_c: 1918
ssd_4x8_neon: 482
ssd_4x16_c: 5258
ssd_4x16_neon: 1025
ssd_8x4_c: 1291
ssd_8x4_neon: 235
ssd_8x8_c: 2431
ssd_8x8_neon: 425
ssd_8x16_c: 4635
ssd_8x16_neon: 910
ssd_16x8_c: 4198
ssd_16x8_neon: 897
ssd_16x16_c: 8549
ssd_16x16_neon: 1907

Signed-off-by: Hubert Mazur <hum@semihalf.com>
This commit is contained in:
Hubert Mazur
2022-11-03 08:51:01 +00:00
committed by Anton Mitrofanov
parent 90b3391ee6
commit 8fd1e5f26d
2 changed files with 198 additions and 102 deletions
Download Patch File Download Diff File Expand all files Collapse all files
common/aarch64/pixel-a.S
+197 -101
View File
@@ -274,6 +274,99 @@ function pixel_asd8_neon, export=1
ret
endfunc
.macro SSD_START_4
ld1 {v16.s}[0], [x0], x1
ld1 {v17.s}[0], [x2], x3
usubl v2.8h, v16.8b, v17.8b
ld1 {v16.s}[0], [x0], x1
ld1 {v17.s}[0], [x2], x3
smull v0.4s, v2.4h, v2.4h
.endm
.macro SSD_4
usubl v2.8h, v16.8b, v17.8b
ld1 {v16.s}[0], [x0], x1
ld1 {v17.s}[0], [x2], x3
smlal v0.4s, v2.4h, v2.4h
.endm
.macro SSD_END_4
usubl v2.8h, v16.8b, v17.8b
smlal v0.4s, v2.4h, v2.4h
.endm
.macro SSD_START_8
ld1 {v16.8b}, [x0], x1
ld1 {v17.8b}, [x2], x3
usubl v2.8h, v16.8b, v17.8b
ld1 {v16.8b}, [x0], x1
smull v0.4s, v2.4h, v2.4h
ld1 {v17.8b}, [x2], x3
smlal2 v0.4s, v2.8h, v2.8h
.endm
.macro SSD_8
usubl v2.8h, v16.8b, v17.8b
ld1 {v16.8b}, [x0], x1
smlal v0.4s, v2.4h, v2.4h
ld1 {v17.8b}, [x2], x3
smlal2 v0.4s, v2.8h, v2.8h
.endm
.macro SSD_END_8
usubl v2.8h, v16.8b, v17.8b
smlal v0.4s, v2.4h, v2.4h
smlal2 v0.4s, v2.8h, v2.8h
.endm
.macro SSD_START_16
ld1 {v16.16b}, [x0], x1
ld1 {v17.16b}, [x2], x3
usubl v2.8h, v16.8b, v17.8b
usubl2 v3.8h, v16.16b, v17.16b
ld1 {v16.16b}, [x0], x1
smull v0.4s, v2.4h, v2.4h
smull2 v1.4s, v2.8h, v2.8h
ld1 {v17.16b}, [x2], x3
smlal v0.4s, v3.4h, v3.4h
smlal2 v1.4s, v3.8h, v3.8h
.endm
.macro SSD_16
usubl v2.8h, v16.8b, v17.8b
usubl2 v3.8h, v16.16b, v17.16b
ld1 {v16.16b}, [x0], x1
smlal v0.4s, v2.4h, v2.4h
smlal2 v1.4s, v2.8h, v2.8h
ld1 {v17.16b}, [x2], x3
smlal v0.4s, v3.4h, v3.4h
smlal2 v1.4s, v3.8h, v3.8h
.endm
.macro SSD_END_16
usubl v2.8h, v16.8b, v17.8b
usubl2 v3.8h, v16.16b, v17.16b
smlal v0.4s, v2.4h, v2.4h
smlal2 v1.4s, v2.8h, v2.8h
smlal v0.4s, v3.4h, v3.4h
smlal2 v1.4s, v3.8h, v3.8h
add v0.4s, v0.4s, v1.4s
.endm
.macro SSD_FUNC w h
function pixel_ssd_\w\()x\h\()_neon, export=1
SSD_START_\w
.rept \h-2
SSD_\w
.endr
SSD_END_\w
addv s0, v0.4s
mov w0, v0.s[0]
ret
endfunc
.endm
#else /* BIT_DEPTH == 8 */
.macro SAD_START_4
@@ -557,6 +650,102 @@ function pixel_asd8_neon, export=1
ret
endfunc
.macro SSD_START_4
ld1 {v16.d}[0], [x0], x1
ld1 {v17.d}[0], [x2], x3
sub v2.4h, v16.4h, v17.4h
ld1 {v16.d}[0], [x0], x1
ld1 {v17.d}[0], [x2], x3
smull v0.4s, v2.4h, v2.4h
.endm
.macro SSD_4
sub v2.4h, v16.4h, v17.4h
ld1 {v16.d}[0], [x0], x1
ld1 {v17.d}[0], [x2], x3
smlal v0.4s, v2.4h, v2.4h
.endm
.macro SSD_END_4
sub v2.4h, v16.4h, v17.4h
smlal v0.4s, v2.4h, v2.4h
.endm
.macro SSD_START_8
ld1 {v16.8h}, [x0], x1
ld1 {v17.8h}, [x2], x3
sub v2.8h, v16.8h, v17.8h
ld1 {v16.8h}, [x0], x1
ld1 {v17.8h}, [x2], x3
smull v0.4s, v2.4h, v2.4h
smull2 v20.4s, v2.8h, v2.8h
.endm
.macro SSD_8
sub v2.8h, v16.8h, v17.8h
ld1 {v16.8h}, [x0], x1
ld1 {v17.8h}, [x2], x3
smlal v0.4s, v2.4h, v2.4h
smlal2 v20.4s, v2.8h, v2.8h
.endm
.macro SSD_END_8
sub v2.8h, v16.8h, v17.8h
smlal v0.4s, v2.4h, v2.4h
smlal2 v20.4s, v2.8h, v2.8h
add v0.4s, v0.4s, v20.4s
.endm
.macro SSD_START_16
ld1 {v16.8h, v17.8h}, [x0], x1
ld1 {v18.8h, v19.8h}, [x2], x3
sub v2.8h, v16.8h, v18.8h
sub v3.8h, v17.8h, v19.8h
ld1 {v16.8h, v17.8h}, [x0], x1
smull v0.4s, v2.4h, v2.4h
smull2 v20.4s, v2.8h, v2.8h
ld1 {v18.8h, v19.8h}, [x2], x3
smlal v0.4s, v3.4h, v3.4h
smlal2 v20.4s, v3.8h, v3.8h
.endm
.macro SSD_16
sub v2.8h, v16.8h, v18.8h
sub v3.8h, v17.8h, v19.8h
ld1 {v16.8h, v17.8h}, [x0], x1
smlal v0.4s, v2.4h, v2.4h
smlal2 v20.4s, v2.8h, v2.8h
ld1 {v18.8h, v19.8h}, [x2], x3
smlal v0.4s, v3.4h, v3.4h
smlal2 v20.4s, v3.8h, v3.8h
.endm
.macro SSD_END_16
sub v2.8h, v16.8h, v18.8h
sub v3.8h, v17.8h, v19.8h
smlal v0.4s, v2.4h, v2.4h
smlal2 v20.4s, v2.8h, v2.8h
smlal v0.4s, v3.4h, v3.4h
smlal2 v20.4s, v3.8h, v3.8h
add v0.4s, v0.4s, v20.4s
.endm
.macro SSD_FUNC w h
function pixel_ssd_\w\()x\h\()_neon, export=1
lsl x1, x1, #1
lsl x3, x3, #1
SSD_START_\w
.rept \h-2
SSD_\w
.endr
SSD_END_\w
addv s0, v0.4s
fmov w0, s0
ret
endfunc
.endm
#endif /* BIT_DEPTH == 8 */
SAD_FUNC 4, 4
@@ -584,107 +773,14 @@ SAD_X_FUNC 4, 8, 16
SAD_X_FUNC 4, 16, 8
SAD_X_FUNC 4, 16, 16
.macro SSD_START_4
ld1 {v16.s}[0], [x0], x1
ld1 {v17.s}[0], [x2], x3
usubl v2.8h, v16.8b, v17.8b
ld1 {v16.s}[0], [x0], x1
ld1 {v17.s}[0], [x2], x3
smull v0.4s, v2.4h, v2.4h
.endm
.macro SSD_4
usubl v2.8h, v16.8b, v17.8b
ld1 {v16.s}[0], [x0], x1
ld1 {v17.s}[0], [x2], x3
smlal v0.4s, v2.4h, v2.4h
.endm
.macro SSD_END_4
usubl v2.8h, v16.8b, v17.8b
smlal v0.4s, v2.4h, v2.4h
.endm
.macro SSD_START_8
ld1 {v16.8b}, [x0], x1
ld1 {v17.8b}, [x2], x3
usubl v2.8h, v16.8b, v17.8b
ld1 {v16.8b}, [x0], x1
smull v0.4s, v2.4h, v2.4h
ld1 {v17.8b}, [x2], x3
smlal2 v0.4s, v2.8h, v2.8h
.endm
.macro SSD_8
usubl v2.8h, v16.8b, v17.8b
ld1 {v16.8b}, [x0], x1
smlal v0.4s, v2.4h, v2.4h
ld1 {v17.8b}, [x2], x3
smlal2 v0.4s, v2.8h, v2.8h
.endm
.macro SSD_END_8
usubl v2.8h, v16.8b, v17.8b
smlal v0.4s, v2.4h, v2.4h
smlal2 v0.4s, v2.8h, v2.8h
.endm
.macro SSD_START_16
ld1 {v16.16b}, [x0], x1
ld1 {v17.16b}, [x2], x3
usubl v2.8h, v16.8b, v17.8b
usubl2 v3.8h, v16.16b, v17.16b
ld1 {v16.16b}, [x0], x1
smull v0.4s, v2.4h, v2.4h
smull2 v1.4s, v2.8h, v2.8h
ld1 {v17.16b}, [x2], x3
smlal v0.4s, v3.4h, v3.4h
smlal2 v1.4s, v3.8h, v3.8h
.endm
.macro SSD_16
usubl v2.8h, v16.8b, v17.8b
usubl2 v3.8h, v16.16b, v17.16b
ld1 {v16.16b}, [x0], x1
smlal v0.4s, v2.4h, v2.4h
smlal2 v1.4s, v2.8h, v2.8h
ld1 {v17.16b}, [x2], x3
smlal v0.4s, v3.4h, v3.4h
smlal2 v1.4s, v3.8h, v3.8h
.endm
.macro SSD_END_16
usubl v2.8h, v16.8b, v17.8b
usubl2 v3.8h, v16.16b, v17.16b
smlal v0.4s, v2.4h, v2.4h
smlal2 v1.4s, v2.8h, v2.8h
smlal v0.4s, v3.4h, v3.4h
smlal2 v1.4s, v3.8h, v3.8h
add v0.4s, v0.4s, v1.4s
.endm
.macro SSD_FUNC w h
function pixel_ssd_\w\()x\h\()_neon, export=1
SSD_START_\w
.rept \h-2
SSD_\w
.endr
SSD_END_\w
addv s0, v0.4s
mov w0, v0.s[0]
ret
endfunc
.endm
SSD_FUNC 4, 4
SSD_FUNC 4, 8
SSD_FUNC 4, 16
SSD_FUNC 8, 4
SSD_FUNC 8, 8
SSD_FUNC 8, 16
SSD_FUNC 16, 8
SSD_FUNC 16, 16
SSD_FUNC 4, 4
SSD_FUNC 4, 8
SSD_FUNC 4, 16
SSD_FUNC 8, 4
SSD_FUNC 8, 8
SSD_FUNC 8, 16
SSD_FUNC 16, 8
SSD_FUNC 16, 16
function pixel_ssd_nv12_core_neon, export=1
common/pixel.c
+1 -1
View File
@@ -1061,7 +1061,7 @@ void x264_pixel_init( uint32_t cpu, x264_pixel_function_t *pixf )
INIT7( sad_x3, _neon);
pixf->vsad = x264_pixel_vsad_neon;
pixf->asd8 = x264_pixel_asd8_neon;
INIT8(ssd, _neon);
}
#endif // HAVE_AARCH64