ref: a913cc853e517c2a5a0f79cc72cd5df590d82317
dir: /processing/src/asm/downsample_bilinear.asm/
;*!
;* \copy
;* Copyright (c) 2009-2013, Cisco Systems
;* All rights reserved.
;*
;* Redistribution and use in source and binary forms, with or without
;* modification, are permitted provided that the following conditions
;* are met:
;*
;* * Redistributions of source code must retain the above copyright
;* notice, this list of conditions and the following disclaimer.
;*
;* * 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 HOLDER 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.
;*
;*
;* upsampling.asm
;*
;* Abstract
;* SIMD for pixel domain down sampling
;*
;* History
;* 10/22/2009 Created
;*
;*************************************************************************/
%include "asm_inc.asm"
BITS 32
;***********************************************************************
; Macros and other preprocessor constants
;***********************************************************************
;***********************************************************************
; Some constants
;***********************************************************************
;***********************************************************************
; Local Data (Read Only)
;***********************************************************************
SECTION .rodata align=16
;***********************************************************************
; Various memory constants (trigonometric values or rounding values)
;***********************************************************************
ALIGN 16
shufb_mask_low:
db 00h, 80h, 02h, 80h, 04h, 80h, 06h, 80h, 08h, 80h, 0ah, 80h, 0ch, 80h, 0eh, 80h
shufb_mask_high:
db 01h, 80h, 03h, 80h, 05h, 80h, 07h, 80h, 09h, 80h, 0bh, 80h, 0dh, 80h, 0fh, 80h
ALIGN 16
;***********************************************************************
; Code
;***********************************************************************
SECTION .text
WELS_EXTERN DyadicBilinearDownsamplerWidthx32_sse
;***********************************************************************
; void DyadicBilinearDownsamplerWidthx32_sse( unsigned char* pDst, const int iDstStride,
; unsigned char* pSrc, const int iSrcStride,
; const int iSrcWidth, const int iSrcHeight );
;***********************************************************************
ALIGN 16
DyadicBilinearDownsamplerWidthx32_sse:
push ebx
push edx
push esi
push edi
push ebp
mov edi, [esp+24] ; pDst
mov edx, [esp+28] ; iDstStride
mov esi, [esp+32] ; pSrc
mov ecx, [esp+36] ; iSrcStride
mov ebp, [esp+44] ; iSrcHeight
sar ebp, $1 ; iSrcHeight >> 1
.yloops:
mov eax, [esp+40] ; iSrcWidth
sar eax, $1 ; iSrcWidth >> 1
mov ebx, eax ; iDstWidth restored at ebx
sar eax, $4 ; (iSrcWidth >> 1) / 16 ; loop count = num_of_mb
neg ebx ; - (iSrcWidth >> 1)
; each loop = source bandwidth: 32 bytes
.xloops:
; 1st part horizonal loop: x16 bytes
; mem hi<- ->lo
;1st Line Src: mm0: d D c C b B a A mm1: h H g G f F e E
;2nd Line Src: mm2: l L k K j J i I mm3: p P o O n N m M
;=> target:
;: H G F E D C B A, P O N M L K J I
;: h g f e d c b a, p o n m l k j i
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
movq mm0, [esi] ; 1st pSrc line
movq mm1, [esi+8] ; 1st pSrc line + 8
movq mm2, [esi+ecx] ; 2nd pSrc line
movq mm3, [esi+ecx+8] ; 2nd pSrc line + 8
; to handle mm0, mm1, mm2, mm3
pshufw mm4, mm0, 0d8h ; d D b B c C a A ; 11011000 B
pshufw mm5, mm4, 04eh ; c C a A d D b B ; 01001110 B
punpcklbw mm4, mm5 ; d c D C b a B A
pshufw mm4, mm4, 0d8h ; d c b a D C B A ; 11011000 B: mm4
pshufw mm5, mm1, 0d8h ; h H f F g G e E ; 11011000 B
pshufw mm6, mm5, 04eh ; g G e E h H f F ; 01001110 B
punpcklbw mm5, mm6 ; h g H G f e F E
pshufw mm5, mm5, 0d8h ; h g f e H G F E ; 11011000 B: mm5
pshufw mm6, mm2, 0d8h ; l L j J k K i I ; 11011000 B
pshufw mm7, mm6, 04eh ; k K i I l L j J ; 01001110 B
punpcklbw mm6, mm7 ; l k L K j i J I
pshufw mm6, mm6, 0d8h ; l k j i L K J I ; 11011000 B: mm6
pshufw mm7, mm3, 0d8h ; p P n N o O m M ; 11011000 B
pshufw mm0, mm7, 04eh ; o O m M p P n N ; 01001110 B
punpcklbw mm7, mm0 ; p o P O n m N M
pshufw mm7, mm7, 0d8h ; p o n m P O N M ; 11011000 B: mm7
; to handle mm4, mm5, mm6, mm7
movq mm0, mm4 ;
punpckldq mm0, mm5 ; H G F E D C B A
punpckhdq mm4, mm5 ; h g f e d c b a
movq mm1, mm6
punpckldq mm1, mm7 ; P O N M L K J I
punpckhdq mm6, mm7 ; p o n m l k j i
; avg within MB horizon width (16 x 2 lines)
pavgb mm0, mm4 ; (A+a+1)>>1, .., (H+h+1)>>1, temp_row1
pavgb mm1, mm6 ; (I+i+1)>>1, .., (P+p+1)>>1, temp_row2
pavgb mm0, mm1 ; (temp_row1+temp_row2+1)>>1, pending here and wait another horizonal part done then write memory once
; 2nd part horizonal loop: x16 bytes
; mem hi<- ->lo
;1st Line Src: mm0: d D c C b B a A mm1: h H g G f F e E
;2nd Line Src: mm2: l L k K j J i I mm3: p P o O n N m M
;=> target:
;: H G F E D C B A, P O N M L K J I
;: h g f e d c b a, p o n m l k j i
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
movq mm1, [esi+16] ; 1st pSrc line + 16
movq mm2, [esi+24] ; 1st pSrc line + 24
movq mm3, [esi+ecx+16] ; 2nd pSrc line + 16
movq mm4, [esi+ecx+24] ; 2nd pSrc line + 24
; to handle mm1, mm2, mm3, mm4
pshufw mm5, mm1, 0d8h ; d D b B c C a A ; 11011000 B
pshufw mm6, mm5, 04eh ; c C a A d D b B ; 01001110 B
punpcklbw mm5, mm6 ; d c D C b a B A
pshufw mm5, mm5, 0d8h ; d c b a D C B A ; 11011000 B: mm5
pshufw mm6, mm2, 0d8h ; h H f F g G e E ; 11011000 B
pshufw mm7, mm6, 04eh ; g G e E h H f F ; 01001110 B
punpcklbw mm6, mm7 ; h g H G f e F E
pshufw mm6, mm6, 0d8h ; h g f e H G F E ; 11011000 B: mm6
pshufw mm7, mm3, 0d8h ; l L j J k K i I ; 11011000 B
pshufw mm1, mm7, 04eh ; k K i I l L j J ; 01001110 B
punpcklbw mm7, mm1 ; l k L K j i J I
pshufw mm7, mm7, 0d8h ; l k j i L K J I ; 11011000 B: mm7
pshufw mm1, mm4, 0d8h ; p P n N o O m M ; 11011000 B
pshufw mm2, mm1, 04eh ; o O m M p P n N ; 01001110 B
punpcklbw mm1, mm2 ; p o P O n m N M
pshufw mm1, mm1, 0d8h ; p o n m P O N M ; 11011000 B: mm1
; to handle mm5, mm6, mm7, mm1
movq mm2, mm5
punpckldq mm2, mm6 ; H G F E D C B A
punpckhdq mm5, mm6 ; h g f e d c b a
movq mm3, mm7
punpckldq mm3, mm1 ; P O N M L K J I
punpckhdq mm7, mm1 ; p o n m l k j i
; avg within MB horizon width (16 x 2 lines)
pavgb mm2, mm5 ; (A+a+1)>>1, .., (H+h+1)>>1, temp_row1
pavgb mm3, mm7 ; (I+i+1)>>1, .., (P+p+1)>>1, temp_row2
pavgb mm2, mm3 ; (temp_row1+temp_row2+1)>>1, done in another 2nd horizonal part
movq [edi ], mm0
movq [edi+8], mm2
; next SMB
lea esi, [esi+32]
lea edi, [edi+16]
dec eax
jg near .xloops
; next line
lea esi, [esi+2*ecx] ; next end of lines
lea esi, [esi+2*ebx] ; reset to base 0 [- 2 * iDstWidth]
lea edi, [edi+edx]
lea edi, [edi+ebx] ; reset to base 0 [- iDstWidth]
dec ebp
jg near .yloops
WELSEMMS
pop ebp
pop edi
pop esi
pop edx
pop ebx
ret
WELS_EXTERN DyadicBilinearDownsamplerWidthx16_sse
;***********************************************************************
; void DyadicBilinearDownsamplerWidthx16_sse( unsigned char* pDst, const int iDstStride,
; unsigned char* pSrc, const int iSrcStride,
; const int iSrcWidth, const int iSrcHeight );
;***********************************************************************
ALIGN 16
DyadicBilinearDownsamplerWidthx16_sse:
push ebx
push edx
push esi
push edi
push ebp
mov edi, [esp+24] ; pDst
mov edx, [esp+28] ; iDstStride
mov esi, [esp+32] ; pSrc
mov ecx, [esp+36] ; iSrcStride
mov ebp, [esp+44] ; iSrcHeight
sar ebp, $1 ; iSrcHeight >> 1
.yloops:
mov eax, [esp+40] ; iSrcWidth
sar eax, $1 ; iSrcWidth >> 1
mov ebx, eax ; iDstWidth restored at ebx
sar eax, $3 ; (iSrcWidth >> 1) / 8 ; loop count = num_of_mb
neg ebx ; - (iSrcWidth >> 1)
; each loop = source bandwidth: 16 bytes
.xloops:
; 1st part horizonal loop: x16 bytes
; mem hi<- ->lo
;1st Line Src: mm0: d D c C b B a A mm1: h H g G f F e E
;2nd Line Src: mm2: l L k K j J i I mm3: p P o O n N m M
;=> target:
;: H G F E D C B A, P O N M L K J I
;: h g f e d c b a, p o n m l k j i
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
movq mm0, [esi] ; 1st pSrc line
movq mm1, [esi+8] ; 1st pSrc line + 8
movq mm2, [esi+ecx] ; 2nd pSrc line
movq mm3, [esi+ecx+8] ; 2nd pSrc line + 8
; to handle mm0, mm1, mm2, mm3
pshufw mm4, mm0, 0d8h ; d D b B c C a A ; 11011000 B
pshufw mm5, mm4, 04eh ; c C a A d D b B ; 01001110 B
punpcklbw mm4, mm5 ; d c D C b a B A
pshufw mm4, mm4, 0d8h ; d c b a D C B A ; 11011000 B: mm4
pshufw mm5, mm1, 0d8h ; h H f F g G e E ; 11011000 B
pshufw mm6, mm5, 04eh ; g G e E h H f F ; 01001110 B
punpcklbw mm5, mm6 ; h g H G f e F E
pshufw mm5, mm5, 0d8h ; h g f e H G F E ; 11011000 B: mm5
pshufw mm6, mm2, 0d8h ; l L j J k K i I ; 11011000 B
pshufw mm7, mm6, 04eh ; k K i I l L j J ; 01001110 B
punpcklbw mm6, mm7 ; l k L K j i J I
pshufw mm6, mm6, 0d8h ; l k j i L K J I ; 11011000 B: mm6
pshufw mm7, mm3, 0d8h ; p P n N o O m M ; 11011000 B
pshufw mm0, mm7, 04eh ; o O m M p P n N ; 01001110 B
punpcklbw mm7, mm0 ; p o P O n m N M
pshufw mm7, mm7, 0d8h ; p o n m P O N M ; 11011000 B: mm7
; to handle mm4, mm5, mm6, mm7
movq mm0, mm4 ;
punpckldq mm0, mm5 ; H G F E D C B A
punpckhdq mm4, mm5 ; h g f e d c b a
movq mm1, mm6
punpckldq mm1, mm7 ; P O N M L K J I
punpckhdq mm6, mm7 ; p o n m l k j i
; avg within MB horizon width (16 x 2 lines)
pavgb mm0, mm4 ; (A+a+1)>>1, .., (H+h+1)>>1, temp_row1
pavgb mm1, mm6 ; (I+i+1)>>1, .., (P+p+1)>>1, temp_row2
pavgb mm0, mm1 ; (temp_row1+temp_row2+1)>>1, pending here and wait another horizonal part done then write memory once
movq [edi ], mm0
; next SMB
lea esi, [esi+16]
lea edi, [edi+8]
dec eax
jg near .xloops
; next line
lea esi, [esi+2*ecx] ; next end of lines
lea esi, [esi+2*ebx] ; reset to base 0 [- 2 * iDstWidth]
lea edi, [edi+edx]
lea edi, [edi+ebx] ; reset to base 0 [- iDstWidth]
dec ebp
jg near .yloops
WELSEMMS
pop ebp
pop edi
pop esi
pop edx
pop ebx
ret
WELS_EXTERN DyadicBilinearDownsamplerWidthx8_sse
;***********************************************************************
; void DyadicBilinearDownsamplerWidthx8_sse( unsigned char* pDst, const int iDstStride,
; unsigned char* pSrc, const int iSrcStride,
; const int iSrcWidth, const int iSrcHeight );
;***********************************************************************
ALIGN 16
DyadicBilinearDownsamplerWidthx8_sse:
push ebx
push edx
push esi
push edi
push ebp
mov edi, [esp+24] ; pDst
mov edx, [esp+28] ; iDstStride
mov esi, [esp+32] ; pSrc
mov ecx, [esp+36] ; iSrcStride
mov ebp, [esp+44] ; iSrcHeight
sar ebp, $1 ; iSrcHeight >> 1
.yloops:
mov eax, [esp+40] ; iSrcWidth
sar eax, $1 ; iSrcWidth >> 1
mov ebx, eax ; iDstWidth restored at ebx
sar eax, $2 ; (iSrcWidth >> 1) / 4 ; loop count = num_of_mb
neg ebx ; - (iSrcWidth >> 1)
; each loop = source bandwidth: 8 bytes
.xloops:
; 1st part horizonal loop: x8 bytes
; mem hi<- ->lo
;1st Line Src: mm0: d D c C b B a A
;2nd Line Src: mm1: h H g G f F e E
;=> target:
;: H G F E D C B A
;: h g f e d c b a
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
movq mm0, [esi] ; 1st pSrc line
movq mm1, [esi+ecx] ; 2nd pSrc line
; to handle mm0, mm1, mm2, mm3
pshufw mm2, mm0, 0d8h ; d D b B c C a A ; 11011000 B
pshufw mm3, mm2, 04eh ; c C a A d D b B ; 01001110 B
punpcklbw mm2, mm3 ; d c D C b a B A
pshufw mm2, mm2, 0d8h ; d c b a D C B A ; 11011000 B: mm4
pshufw mm4, mm1, 0d8h ; h H f F g G e E ; 11011000 B
pshufw mm5, mm4, 04eh ; g G e E h H f F ; 01001110 B
punpcklbw mm4, mm5 ; h g H G f e F E
pshufw mm4, mm4, 0d8h ; h g f e H G F E ; 11011000 B: mm5
; to handle mm2, mm4
movq mm0, mm2 ;
punpckldq mm0, mm4 ; H G F E D C B A
punpckhdq mm2, mm4 ; h g f e d c b a
; avg within MB horizon width (16 x 2 lines)
pavgb mm0, mm2 ; (H+h+1)>>1, .., (A+a+1)>>1, temp_row1, 2
pshufw mm1, mm0, 04eh ; 01001110 B
pavgb mm0, mm1 ; (temp_row1+temp_row2+1)>>1, pending here and wait another horizonal part done then write memory once
movd [edi], mm0
; next unit
lea esi, [esi+8]
lea edi, [edi+4]
dec eax
jg near .xloops
; next line
lea esi, [esi+2*ecx] ; next end of lines
lea esi, [esi+2*ebx] ; reset to base 0 [- 2 * iDstWidth]
lea edi, [edi+edx]
lea edi, [edi+ebx] ; reset to base 0 [- iDstWidth]
dec ebp
jg near .yloops
WELSEMMS
pop ebp
pop edi
pop esi
pop edx
pop ebx
ret
; got about 50% improvement over DyadicBilinearDownsamplerWidthx32_sse
WELS_EXTERN DyadicBilinearDownsamplerWidthx32_ssse3
;***********************************************************************
; void DyadicBilinearDownsamplerWidthx32_ssse3( unsigned char* pDst, const int iDstStride,
; unsigned char* pSrc, const int iSrcStride,
; const int iSrcWidth, const int iSrcHeight );
;***********************************************************************
ALIGN 16
DyadicBilinearDownsamplerWidthx32_ssse3:
push ebx
push edx
push esi
push edi
push ebp
mov edi, [esp+24] ; pDst
mov edx, [esp+28] ; iDstStride
mov esi, [esp+32] ; pSrc
mov ecx, [esp+36] ; iSrcStride
mov ebp, [esp+44] ; iSrcHeight
sar ebp, $1 ; iSrcHeight >> 1
movdqa xmm7, [shufb_mask_low] ; mask low
movdqa xmm6, [shufb_mask_high] ; mask high
.yloops:
mov eax, [esp+40] ; iSrcWidth
sar eax, $1 ; iSrcWidth >> 1
mov ebx, eax ; iDstWidth restored at ebx
sar eax, $4 ; (iSrcWidth >> 1) / 16 ; loop count = num_of_mb
neg ebx ; - (iSrcWidth >> 1)
; each loop = source bandwidth: 32 bytes
.xloops:
; 1st part horizonal loop: x16 bytes
; mem hi<- ->lo
;1st Line Src: xmm0: h H g G f F e E d D c C b B a A
; xmm1: p P o O n N m M l L k K j J i I
;2nd Line Src: xmm2: h H g G f F e E d D c C b B a A
; xmm3: p P o O n N m M l L k K j J i I
;=> target:
;: P O N M L K J I H G F E D C B A
;: p o n m l k j i h g f e d c b a
;: P .. A
;: p .. a
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
movdqa xmm0, [esi] ; 1st_src_line
movdqa xmm1, [esi+16] ; 1st_src_line + 16
movdqa xmm2, [esi+ecx] ; 2nd_src_line
movdqa xmm3, [esi+ecx+16] ; 2nd_src_line + 16
; packing & avg
movdqa xmm4, xmm0 ; h H g G f F e E d D c C b B a A
pshufb xmm0, xmm7 ; 0 H 0 G 0 F 0 E 0 D 0 C 0 B 0 A
pshufb xmm4, xmm6 ; 0 h 0 g 0 f 0 e 0 d 0 c 0 b 0 a
; another implementation for xmm4 high bits
; psubb xmm4, xmm0 ; h 0 g 0 f 0 e 0 d 0 c 0 b 0 a 0
; psrlw xmm4, 8 ; 0 h 0 g 0 f 0 e 0 d 0 c 0 b 0 a
pavgb xmm0, xmm4
movdqa xmm5, xmm1
pshufb xmm1, xmm7
pshufb xmm5, xmm6
; psubb xmm5, xmm1
; psrlw xmm5, 8
pavgb xmm1, xmm5
movdqa xmm4, xmm2
pshufb xmm2, xmm7
pshufb xmm4, xmm6
; psubb xmm4, xmm2
; psrlw xmm4, 8
pavgb xmm2, xmm4
movdqa xmm5, xmm3
pshufb xmm3, xmm7
pshufb xmm5, xmm6
; psubb xmm5, xmm3
; psrlw xmm5, 8
pavgb xmm3, xmm5
packuswb xmm0, xmm1
packuswb xmm2, xmm3
pavgb xmm0, xmm2
; write pDst
movdqa [edi], xmm0
; next SMB
lea esi, [esi+32]
lea edi, [edi+16]
dec eax
jg near .xloops
; next line
lea esi, [esi+2*ecx] ; next end of lines
lea esi, [esi+2*ebx] ; reset to base 0 [- 2 * iDstWidth]
lea edi, [edi+edx]
lea edi, [edi+ebx] ; reset to base 0 [- iDstWidth]
dec ebp
jg near .yloops
pop ebp
pop edi
pop esi
pop edx
pop ebx
ret
WELS_EXTERN DyadicBilinearDownsamplerWidthx16_ssse3
;***********************************************************************
; void DyadicBilinearDownsamplerWidthx16_ssse3( unsigned char* pDst, const int iDstStride,
; unsigned char* pSrc, const int iSrcStride,
; const int iSrcWidth, const int iSrcHeight );
;***********************************************************************
ALIGN 16
DyadicBilinearDownsamplerWidthx16_ssse3:
push ebx
push edx
push esi
push edi
push ebp
mov edi, [esp+24] ; pDst
mov edx, [esp+28] ; iDstStride
mov esi, [esp+32] ; pSrc
mov ecx, [esp+36] ; iSrcStride
mov ebp, [esp+44] ; iSrcHeight
sar ebp, $1 ; iSrcHeight >> 1
movdqa xmm7, [shufb_mask_low] ; mask low
movdqa xmm6, [shufb_mask_high] ; mask high
.yloops:
mov eax, [esp+40] ; iSrcWidth
sar eax, $1 ; iSrcWidth >> 1
mov ebx, eax ; iDstWidth restored at ebx
sar eax, $3 ; (iSrcWidth >> 1) / 8 ; loop count = num_of_mb
neg ebx ; - (iSrcWidth >> 1)
; each loop = source bandwidth: 16 bytes
.xloops:
; horizonal loop: x16 bytes by source
; mem hi<- ->lo
;1st line pSrc: xmm0: h H g G f F e E d D c C b B a A
;2nd line pSrc: xmm1: p P o O n N m M l L k K j J i I
;=> target:
;: H G F E D C B A, P O N M L K J I
;: h g f e d c b a, p o n m l k j i
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
movdqa xmm0, [esi] ; 1st_src_line
movdqa xmm1, [esi+ecx] ; 2nd_src_line
; packing & avg
movdqa xmm2, xmm0 ; h H g G f F e E d D c C b B a A
pshufb xmm0, xmm7 ; 0 H 0 G 0 F 0 E 0 D 0 C 0 B 0 A
pshufb xmm2, xmm6 ; 0 h 0 g 0 f 0 e 0 d 0 c 0 b 0 a
; another implementation for xmm2 high bits
; psubb xmm2, xmm0 ; h 0 g 0 f 0 e 0 d 0 c 0 b 0 a 0
; psrlw xmm2, 8 ; 0 h 0 g 0 f 0 e 0 d 0 c 0 b 0 a
pavgb xmm0, xmm2
movdqa xmm3, xmm1
pshufb xmm1, xmm7
pshufb xmm3, xmm6
; psubb xmm3, xmm1
; psrlw xmm3, 8
pavgb xmm1, xmm3
pavgb xmm0, xmm1
packuswb xmm0, xmm1
; write pDst
movq [edi], xmm0
; next SMB
lea esi, [esi+16]
lea edi, [edi+8]
dec eax
jg near .xloops
; next line
lea esi, [esi+2*ecx] ; next end of lines
lea esi, [esi+2*ebx] ; reset to base 0 [- 2 * iDstWidth]
lea edi, [edi+edx]
lea edi, [edi+ebx] ; reset to base 0 [- iDstWidth]
dec ebp
jg near .yloops
pop ebp
pop edi
pop esi
pop edx
pop ebx
ret
; got about 65% improvement over DyadicBilinearDownsamplerWidthx32_sse
WELS_EXTERN DyadicBilinearDownsamplerWidthx32_sse4
;***********************************************************************
; void DyadicBilinearDownsamplerWidthx32_sse4( unsigned char* pDst, const int iDstStride,
; unsigned char* pSrc, const int iSrcStride,
; const int iSrcWidth, const int iSrcHeight );
;***********************************************************************
ALIGN 16
DyadicBilinearDownsamplerWidthx32_sse4:
push ebx
push edx
push esi
push edi
push ebp
mov edi, [esp+24] ; pDst
mov edx, [esp+28] ; iDstStride
mov esi, [esp+32] ; pSrc
mov ecx, [esp+36] ; iSrcStride
mov ebp, [esp+44] ; iSrcHeight
sar ebp, $1 ; iSrcHeight >> 1
movdqa xmm7, [shufb_mask_low] ; mask low
movdqa xmm6, [shufb_mask_high] ; mask high
.yloops:
mov eax, [esp+40] ; iSrcWidth
sar eax, $1 ; iSrcWidth >> 1
mov ebx, eax ; iDstWidth restored at ebx
sar eax, $4 ; (iSrcWidth >> 1) / 16 ; loop count = num_of_mb
neg ebx ; - (iSrcWidth >> 1)
; each loop = source bandwidth: 32 bytes
.xloops:
; 1st part horizonal loop: x16 bytes
; mem hi<- ->lo
;1st Line Src: xmm0: h H g G f F e E d D c C b B a A
; xmm1: p P o O n N m M l L k K j J i I
;2nd Line Src: xmm2: h H g G f F e E d D c C b B a A
; xmm3: p P o O n N m M l L k K j J i I
;=> target:
;: P O N M L K J I H G F E D C B A
;: p o n m l k j i h g f e d c b a
;: P .. A
;: p .. a
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
movntdqa xmm0, [esi] ; 1st_src_line
movntdqa xmm1, [esi+16] ; 1st_src_line + 16
movntdqa xmm2, [esi+ecx] ; 2nd_src_line
movntdqa xmm3, [esi+ecx+16] ; 2nd_src_line + 16
; packing & avg
movdqa xmm4, xmm0 ; h H g G f F e E d D c C b B a A
pshufb xmm0, xmm7 ; 0 H 0 G 0 F 0 E 0 D 0 C 0 B 0 A
pshufb xmm4, xmm6 ; 0 h 0 g 0 f 0 e 0 d 0 c 0 b 0 a
; psubb xmm4, xmm0 ; h 0 g 0 f 0 e 0 d 0 c 0 b 0 a 0
; psrlw xmm4, 8 ; 0 h 0 g 0 f 0 e 0 d 0 c 0 b 0 a
pavgb xmm0, xmm4
movdqa xmm5, xmm1
pshufb xmm1, xmm7
pshufb xmm5, xmm6
; psubb xmm5, xmm1
; psrlw xmm5, 8
pavgb xmm1, xmm5
movdqa xmm4, xmm2
pshufb xmm2, xmm7
pshufb xmm4, xmm6
; psubb xmm4, xmm2
; psrlw xmm4, 8
pavgb xmm2, xmm4
movdqa xmm5, xmm3
pshufb xmm3, xmm7
pshufb xmm5, xmm6
; psubb xmm5, xmm3
; psrlw xmm5, 8
pavgb xmm3, xmm5
packuswb xmm0, xmm1
packuswb xmm2, xmm3
pavgb xmm0, xmm2
; write pDst
movdqa [edi], xmm0
; next SMB
lea esi, [esi+32]
lea edi, [edi+16]
dec eax
jg near .xloops
; next line
lea esi, [esi+2*ecx] ; next end of lines
lea esi, [esi+2*ebx] ; reset to base 0 [- 2 * iDstWidth]
lea edi, [edi+edx]
lea edi, [edi+ebx] ; reset to base 0 [- iDstWidth]
dec ebp
jg near .yloops
pop ebp
pop edi
pop esi
pop edx
pop ebx
ret
WELS_EXTERN DyadicBilinearDownsamplerWidthx16_sse4
;***********************************************************************
; void DyadicBilinearDownsamplerWidthx16_sse4( unsigned char* pDst, const int iDstStride,
; unsigned char* pSrc, const int iSrcStride,
; const int iSrcWidth, const int iSrcHeight );
;***********************************************************************
ALIGN 16
DyadicBilinearDownsamplerWidthx16_sse4:
push ebx
push edx
push esi
push edi
push ebp
mov edi, [esp+24] ; pDst
mov edx, [esp+28] ; iDstStride
mov esi, [esp+32] ; pSrc
mov ecx, [esp+36] ; iSrcStride
mov ebp, [esp+44] ; iSrcHeight
sar ebp, $1 ; iSrcHeight >> 1
movdqa xmm7, [shufb_mask_low] ; mask low
movdqa xmm6, [shufb_mask_high] ; mask high
.yloops:
mov eax, [esp+40] ; iSrcWidth
sar eax, $1 ; iSrcWidth >> 1
mov ebx, eax ; iDstWidth restored at ebx
sar eax, $3 ; (iSrcWidth >> 1) / 8 ; loop count = num_of_mb
neg ebx ; - (iSrcWidth >> 1)
; each loop = source bandwidth: 16 bytes
.xloops:
; horizonal loop: x16 bytes by source
; mem hi<- ->lo
;1st line pSrc: xmm0: h H g G f F e E d D c C b B a A
;2nd line pSrc: xmm1: p P o O n N m M l L k K j J i I
;=> target:
;: H G F E D C B A, P O N M L K J I
;: h g f e d c b a, p o n m l k j i
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
movntdqa xmm0, [esi] ; 1st_src_line
movntdqa xmm1, [esi+ecx] ; 2nd_src_line
; packing & avg
movdqa xmm2, xmm0 ; h H g G f F e E d D c C b B a A
pshufb xmm0, xmm7 ; 0 H 0 G 0 F 0 E 0 D 0 C 0 B 0 A
pshufb xmm2, xmm6 ; 0 h 0 g 0 f 0 e 0 d 0 c 0 b 0 a
; psubb xmm2, xmm0 ; h 0 g 0 f 0 e 0 d 0 c 0 b 0 a 0
; psrlw xmm2, 8 ; 0 h 0 g 0 f 0 e 0 d 0 c 0 b 0 a
pavgb xmm0, xmm2
movdqa xmm3, xmm1
pshufb xmm1, xmm7
pshufb xmm3, xmm6
; psubb xmm3, xmm1
; psrlw xmm3, 8
pavgb xmm1, xmm3
pavgb xmm0, xmm1
packuswb xmm0, xmm1
; write pDst
movq [edi], xmm0
; next SMB
lea esi, [esi+16]
lea edi, [edi+8]
dec eax
jg near .xloops
; next line
lea esi, [esi+2*ecx] ; next end of lines
lea esi, [esi+2*ebx] ; reset to base 0 [- 2 * iDstWidth]
lea edi, [edi+edx]
lea edi, [edi+ebx] ; reset to base 0 [- iDstWidth]
dec ebp
jg near .yloops
pop ebp
pop edi
pop esi
pop edx
pop ebx
ret
WELS_EXTERN GeneralBilinearAccurateDownsampler_sse2
;**************************************************************************************************************
;int GeneralBilinearAccurateDownsampler_sse2( unsigned char* pDst, const int iDstStride, const int iDstWidth, const int iDstHeight,
; unsigned char* pSrc, const int iSrcStride, const int iSrcWidth, const int iSrcHeight,
; unsigned int uiScaleX, unsigned int uiScaleY );
;{
;**************************************************************************************************************
ALIGN 16
GeneralBilinearAccurateDownsampler_sse2:
push ebp
push esi
push edi
push ebx
%define pushsize 16
%define localsize 28
%define pDstData esp + pushsize + localsize + 4
%define dwDstStride esp + pushsize + localsize + 8
%define dwDstWidth esp + pushsize + localsize + 12
%define dwDstHeight esp + pushsize + localsize + 16
%define pSrcData esp + pushsize + localsize + 20
%define dwSrcStride esp + pushsize + localsize + 24
%define dwSrcWidth esp + pushsize + localsize + 28
%define dwSrcHeight esp + pushsize + localsize + 32
%define scale esp + 0
%define uiScaleX esp + pushsize + localsize + 36
%define uiScaleY esp + pushsize + localsize + 40
%define tmpHeight esp + 12
%define yInverse esp + 16
%define xInverse esp + 20
%define dstStep esp + 24
sub esp, localsize
pxor xmm0, xmm0
mov edx, 32767
mov eax, [uiScaleX]
and eax, 32767
mov ebx, eax
neg ebx
and ebx, 32767
movd xmm1, eax ; uinc(uiScaleX mod 32767)
movd xmm2, ebx ; -uinc
psllq xmm1, 32
por xmm1, xmm2 ; 0 0 uinc -uinc (dword)
pshufd xmm7, xmm1, 01000100b ; xmm7: uinc -uinc uinc -uinc
mov eax, [uiScaleY]
and eax, 32767
mov ebx, eax
neg ebx
and ebx, 32767
movd xmm6, eax ; vinc(uiScaleY mod 32767)
movd xmm2, ebx ; -vinc
psllq xmm6, 32
por xmm6, xmm2 ; 0 0 vinc -vinc (dword)
pshufd xmm6, xmm6, 01010000b ; xmm6: vinc vinc -vinc -vinc
mov edx, 40003fffh
movd xmm5, edx
punpcklwd xmm5, xmm0 ; 16384 16383
pshufd xmm5, xmm5, 01000100b ; xmm5: 16384 16383 16384 16383
DOWNSAMPLE:
mov eax, [dwDstHeight]
mov edi, [pDstData]
mov edx, [dwDstStride]
mov ecx, [dwDstWidth]
sub edx, ecx
mov [dstStep], edx ; stride - width
dec eax
mov [tmpHeight], eax
mov eax, 16384
mov [yInverse], eax
pshufd xmm4, xmm5, 01010000b ; initial v to 16384 16384 16383 16383
HEIGHT:
mov eax, [yInverse]
mov esi, [pSrcData]
shr eax, 15
mul dword [dwSrcStride]
add esi, eax ; get current row address
mov ebp, esi
add ebp, [dwSrcStride]
mov eax, 16384
mov [xInverse], eax
mov ecx, [dwDstWidth]
dec ecx
movdqa xmm3, xmm5 ; initial u to 16384 16383 16384 16383
WIDTH:
mov eax, [xInverse]
shr eax, 15
movd xmm1, [esi+eax] ; xxxxxxba
movd xmm2, [ebp+eax] ; xxxxxxdc
pxor xmm0, xmm0
punpcklwd xmm1, xmm2 ; xxxxdcba
punpcklbw xmm1, xmm0 ; 0d0c0b0a
punpcklwd xmm1, xmm0 ; 000d000c000b000a
movdqa xmm2, xmm4 ; xmm2: vv(1-v)(1-v) tmpv
pmaddwd xmm2, xmm3 ; mul u(1-u)u(1-u) on xmm2
movdqa xmm0, xmm2
pmuludq xmm2, xmm1
psrlq xmm0, 32
psrlq xmm1, 32
pmuludq xmm0, xmm1
paddq xmm2, xmm0
pshufd xmm1, xmm2, 00001110b
paddq xmm2, xmm1
psrlq xmm2, 29
movd eax, xmm2
inc eax
shr eax, 1
mov [edi], al
inc edi
mov eax, [uiScaleX]
add [xInverse], eax
paddw xmm3, xmm7 ; inc u
psllw xmm3, 1
psrlw xmm3, 1
loop WIDTH
WIDTH_END:
mov eax, [xInverse]
shr eax, 15
mov cl, [esi+eax]
mov [edi], cl
inc edi
mov eax, [uiScaleY]
add [yInverse], eax
add edi, [dstStep]
paddw xmm4, xmm6 ; inc v
psllw xmm4, 1
psrlw xmm4, 1
dec dword [tmpHeight]
jg HEIGHT
LAST_ROW:
mov eax, [yInverse]
mov esi, [pSrcData]
shr eax, 15
mul dword [dwSrcStride]
add esi, eax ; get current row address
mov eax, 16384
mov [xInverse], eax
mov ecx, [dwDstWidth]
LAST_ROW_WIDTH:
mov eax, [xInverse]
shr eax, 15
mov al, [esi+eax]
mov [edi], al
inc edi
mov eax, [uiScaleX]
add [xInverse], eax
loop LAST_ROW_WIDTH
LAST_ROW_END:
add esp, localsize
pop ebx
pop edi
pop esi
pop ebp
%undef pushsize
%undef localsize
%undef pSrcData
%undef dwSrcWidth
%undef dwSrcHeight
%undef dwSrcStride
%undef pDstData
%undef dwDstWidth
%undef dwDstHeight
%undef dwDstStride
%undef scale
%undef uiScaleX
%undef uiScaleY
%undef tmpHeight
%undef yInverse
%undef xInverse
%undef dstStep
ret
WELS_EXTERN GeneralBilinearFastDownsampler_sse2
;**************************************************************************************************************
;int GeneralBilinearFastDownsampler_sse2( unsigned char* pDst, const int iDstStride, const int iDstWidth, const int iDstHeight,
; unsigned char* pSrc, const int iSrcStride, const int iSrcWidth, const int iSrcHeight,
; unsigned int uiScaleX, unsigned int uiScaleY );
;{
;**************************************************************************************************************
ALIGN 16
GeneralBilinearFastDownsampler_sse2:
push ebp
push esi
push edi
push ebx
%define pushsize 16
%define localsize 28
%define pDstData esp + pushsize + localsize + 4
%define dwDstStride esp + pushsize + localsize + 8
%define dwDstWidth esp + pushsize + localsize + 12
%define dwDstHeight esp + pushsize + localsize + 16
%define pSrcData esp + pushsize + localsize + 20
%define dwSrcStride esp + pushsize + localsize + 24
%define dwSrcWidth esp + pushsize + localsize + 28
%define dwSrcHeight esp + pushsize + localsize + 32
%define scale esp + 0
%define uiScaleX esp + pushsize + localsize + 36
%define uiScaleY esp + pushsize + localsize + 40
%define tmpHeight esp + 12
%define yInverse esp + 16
%define xInverse esp + 20
%define dstStep esp + 24
sub esp, localsize
pxor xmm0, xmm0
mov edx, 65535
mov eax, [uiScaleX]
and eax, edx
mov ebx, eax
neg ebx
and ebx, 65535
movd xmm1, eax ; uinc(uiScaleX mod 65536)
movd xmm2, ebx ; -uinc
psllq xmm1, 32
por xmm1, xmm2 ; 0 uinc 0 -uinc
pshuflw xmm7, xmm1, 10001000b ; xmm7: uinc -uinc uinc -uinc
mov eax, [uiScaleY]
and eax, 32767
mov ebx, eax
neg ebx
and ebx, 32767
movd xmm6, eax ; vinc(uiScaleY mod 32767)
movd xmm2, ebx ; -vinc
psllq xmm6, 32
por xmm6, xmm2 ; 0 vinc 0 -vinc
pshuflw xmm6, xmm6, 10100000b ; xmm6: vinc vinc -vinc -vinc
mov edx, 80007fffh ; 32768 32767
movd xmm5, edx
pshuflw xmm5, xmm5, 01000100b ; 32768 32767 32768 32767
mov ebx, 16384
FAST_DOWNSAMPLE:
mov eax, [dwDstHeight]
mov edi, [pDstData]
mov edx, [dwDstStride]
mov ecx, [dwDstWidth]
sub edx, ecx
mov [dstStep], edx ; stride - width
dec eax
mov [tmpHeight], eax
mov eax, 16384
mov [yInverse], eax
pshuflw xmm4, xmm5, 01010000b
psrlw xmm4, 1 ; initial v to 16384 16384 16383 16383
FAST_HEIGHT:
mov eax, [yInverse]
mov esi, [pSrcData]
shr eax, 15
mul dword [dwSrcStride]
add esi, eax ; get current row address
mov ebp, esi
add ebp, [dwSrcStride]
mov eax, 32768
mov [xInverse], eax
mov ecx, [dwDstWidth]
dec ecx
movdqa xmm3, xmm5 ; initial u to 32768 32767 32768 32767
FAST_WIDTH:
mov eax, [xInverse]
shr eax, 16
movd xmm1, [esi+eax] ; xxxxxxba
movd xmm2, [ebp+eax] ; xxxxxxdc
punpcklwd xmm1, xmm2 ; xxxxdcba
punpcklbw xmm1, xmm0 ; 0d0c0b0a
movdqa xmm2, xmm4 ; xmm2: vv(1-v)(1-v) tmpv
pmulhuw xmm2, xmm3 ; mul u(1-u)u(1-u) on xmm2
pmaddwd xmm2, xmm1
pshufd xmm1, xmm2, 00000001b
paddd xmm2, xmm1
movd xmm1, ebx
paddd xmm2, xmm1
psrld xmm2, 15
packuswb xmm2, xmm0
movd eax, xmm2
mov [edi], al
inc edi
mov eax, [uiScaleX]
add [xInverse], eax
paddw xmm3, xmm7 ; inc u
loop FAST_WIDTH
FAST_WIDTH_END:
mov eax, [xInverse]
shr eax, 16
mov cl, [esi+eax]
mov [edi], cl
inc edi
mov eax, [uiScaleY]
add [yInverse], eax
add edi, [dstStep]
paddw xmm4, xmm6 ; inc v
psllw xmm4, 1
psrlw xmm4, 1
dec dword [tmpHeight]
jg FAST_HEIGHT
FAST_LAST_ROW:
mov eax, [yInverse]
mov esi, [pSrcData]
shr eax, 15
mul dword [dwSrcStride]
add esi, eax ; get current row address
mov eax, 32768
mov [xInverse], eax
mov ecx, [dwDstWidth]
FAST_LAST_ROW_WIDTH:
mov eax, [xInverse]
shr eax, 16
mov al, [esi+eax]
mov [edi], al
inc edi
mov eax, [uiScaleX]
add [xInverse], eax
loop FAST_LAST_ROW_WIDTH
FAST_LAST_ROW_END:
add esp, localsize
pop ebx
pop edi
pop esi
pop ebp
%undef pushsize
%undef localsize
%undef pSrcData
%undef dwSrcWidth
%undef dwSrcHeight
%undef dwSrcStride
%undef pDstData
%undef dwDstWidth
%undef dwDstHeight
%undef dwDstStride
%undef scale
%undef uiScaleX
%undef uiScaleY
%undef tmpHeight
%undef yInverse
%undef xInverse
%undef dstStep
ret