ref: b397b397502d940a22a0fda35f1d787336dbdaf2
dir: /codec/encoder/core/x86/quant.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.
;*
;*
;* quant.asm
;*
;* Abstract
;* sse2 quantize inter-block
;*
;* History
;* 7/6/2009 Created
;*
;*
;*************************************************************************/
%include "asm_inc.asm"
SECTION .text
;************************************************
;NEW_QUANT
;************************************************
%macro SSE2_Quant8 5
MOVDQ %1, %5
pxor %2, %2
pcmpgtw %2, %1
pxor %1, %2
psubw %1, %2
paddusw %1, %3
pmulhuw %1, %4
pxor %1, %2
psubw %1, %2
MOVDQ %5, %1
%endmacro
%macro SSE2_QuantMax8 6
MOVDQ %1, %5
pxor %2, %2
pcmpgtw %2, %1
pxor %1, %2
psubw %1, %2
paddusw %1, %3
pmulhuw %1, %4
pmaxsw %6, %1
pxor %1, %2
psubw %1, %2
MOVDQ %5, %1
%endmacro
%define pDct esp + 4
%define ff esp + 8
%define mf esp + 12
%define max esp + 16
;***********************************************************************
; void WelsQuant4x4_sse2(int16_t *pDct, int16_t* ff, int16_t *mf);
;***********************************************************************
WELS_EXTERN WelsQuant4x4_sse2
%assign push_num 0
LOAD_3_PARA
movdqa xmm2, [r1]
movdqa xmm3, [r2]
SSE2_Quant8 xmm0, xmm1, xmm2, xmm3, [r0]
SSE2_Quant8 xmm0, xmm1, xmm2, xmm3, [r0 + 0x10]
ret
;***********************************************************************
;void WelsQuant4x4Dc_sse2(int16_t *pDct, const int16_t ff, int16_t mf);
;***********************************************************************
WELS_EXTERN WelsQuant4x4Dc_sse2
%assign push_num 0
LOAD_3_PARA
SIGN_EXTENSIONW r1, r1w
SIGN_EXTENSIONW r2, r2w
SSE2_Copy8Times xmm3, r2d
SSE2_Copy8Times xmm2, r1d
SSE2_Quant8 xmm0, xmm1, xmm2, xmm3, [r0]
SSE2_Quant8 xmm0, xmm1, xmm2, xmm3, [r0 + 0x10]
ret
;***********************************************************************
; void WelsQuantFour4x4_sse2(int16_t *pDct, int16_t* ff, int16_t *mf);
;***********************************************************************
WELS_EXTERN WelsQuantFour4x4_sse2
%assign push_num 0
LOAD_3_PARA
MOVDQ xmm2, [r1]
MOVDQ xmm3, [r2]
SSE2_Quant8 xmm0, xmm1, xmm2, xmm3, [r0]
SSE2_Quant8 xmm0, xmm1, xmm2, xmm3, [r0 + 0x10]
SSE2_Quant8 xmm0, xmm1, xmm2, xmm3, [r0 + 0x20]
SSE2_Quant8 xmm0, xmm1, xmm2, xmm3, [r0 + 0x30]
SSE2_Quant8 xmm0, xmm1, xmm2, xmm3, [r0 + 0x40]
SSE2_Quant8 xmm0, xmm1, xmm2, xmm3, [r0 + 0x50]
SSE2_Quant8 xmm0, xmm1, xmm2, xmm3, [r0 + 0x60]
SSE2_Quant8 xmm0, xmm1, xmm2, xmm3, [r0 + 0x70]
ret
;***********************************************************************
; void WelsQuantFour4x4Max_sse2(int16_t *pDct, int32_t* f, int16_t *mf, int16_t *max);
;***********************************************************************
WELS_EXTERN WelsQuantFour4x4Max_sse2
%assign push_num 0
LOAD_4_PARA
PUSH_XMM 8
MOVDQ xmm2, [r1]
MOVDQ xmm3, [r2]
pxor xmm4, xmm4
pxor xmm5, xmm5
pxor xmm6, xmm6
pxor xmm7, xmm7
SSE2_QuantMax8 xmm0, xmm1, xmm2, xmm3, [r0 ], xmm4
SSE2_QuantMax8 xmm0, xmm1, xmm2, xmm3, [r0 + 0x10], xmm4
SSE2_QuantMax8 xmm0, xmm1, xmm2, xmm3, [r0 + 0x20], xmm5
SSE2_QuantMax8 xmm0, xmm1, xmm2, xmm3, [r0 + 0x30], xmm5
SSE2_QuantMax8 xmm0, xmm1, xmm2, xmm3, [r0 + 0x40], xmm6
SSE2_QuantMax8 xmm0, xmm1, xmm2, xmm3, [r0 + 0x50], xmm6
SSE2_QuantMax8 xmm0, xmm1, xmm2, xmm3, [r0 + 0x60], xmm7
SSE2_QuantMax8 xmm0, xmm1, xmm2, xmm3, [r0 + 0x70], xmm7
SSE2_TransTwo4x4W xmm4, xmm5, xmm6, xmm7, xmm0
pmaxsw xmm0, xmm4
pmaxsw xmm0, xmm5
pmaxsw xmm0, xmm7
movdqa xmm1, xmm0
punpckhqdq xmm0, xmm1
pmaxsw xmm0, xmm1
movq [r3], xmm0
POP_XMM
LOAD_4_PARA_POP
ret
%macro MMX_Copy4Times 2
movd %1, %2
punpcklwd %1, %1
punpckldq %1, %1
%endmacro
SECTION .text
%macro MMX_Quant4 4
pxor %2, %2
pcmpgtw %2, %1
pxor %1, %2
psubw %1, %2
paddusw %1, %3
pmulhuw %1, %4
pxor %1, %2
psubw %1, %2
%endmacro
;***********************************************************************
;int32_t WelsHadamardQuant2x2_mmx(int16_t *rs, const int16_t ff, int16_t mf, int16_t * pDct, int16_t * block);
;***********************************************************************
WELS_EXTERN WelsHadamardQuant2x2_mmx
%assign push_num 0
LOAD_5_PARA
SIGN_EXTENSIONW r1, r1w
SIGN_EXTENSIONW r2, r2w
movd mm0, [r0]
movd mm1, [r0 + 0x20]
punpcklwd mm0, mm1
movd mm3, [r0 + 0x40]
movd mm1, [r0 + 0x60]
punpcklwd mm3, mm1
;hdm_2x2, mm0 = dct0 dct1, mm3 = dct2 dct3
movq mm5, mm3
paddw mm3, mm0
psubw mm0, mm5
punpcklwd mm3, mm0
movq mm1, mm3
psrlq mm1, 32
movq mm5, mm1
paddw mm1, mm3
psubw mm3, mm5
punpcklwd mm1, mm3
;quant_2x2_dc
MMX_Copy4Times mm3, r2d
MMX_Copy4Times mm2, r1d
MMX_Quant4 mm1, mm0, mm2, mm3
; store dct_2x2
movq [r3], mm1
movq [r4], mm1
; pNonZeroCount of dct_2x2
pcmpeqb mm2, mm2 ; mm2 = FF
pxor mm3, mm3
packsswb mm1, mm3
pcmpeqb mm1, mm3 ; set FF if equal, 0 if not equal
psubsb mm1, mm2 ; set 0 if equal, 1 if not equal
psadbw mm1, mm3 ;
mov r1w, 0
mov [r0], r1w
mov [r0 + 0x20], r1w
mov [r0 + 0x40], r1w
mov [r0 + 0x60], r1w
movd retrd, mm1
WELSEMMS
LOAD_5_PARA_POP
ret
;***********************************************************************
;int32_t WelsHadamardQuant2x2Skip_mmx(int16_t *pDct, int16_t ff, int16_t mf);
;***********************************************************************
WELS_EXTERN WelsHadamardQuant2x2Skip_mmx
%assign push_num 0
LOAD_3_PARA
SIGN_EXTENSIONW r1, r1w
SIGN_EXTENSIONW r2, r2w
movd mm0, [r0]
movd mm1, [r0 + 0x20]
punpcklwd mm0, mm1
movd mm3, [r0 + 0x40]
movd mm1, [r0 + 0x60]
punpcklwd mm3, mm1
;hdm_2x2, mm0 = dct0 dct1, mm3 = dct2 dct3
movq mm5, mm3
paddw mm3, mm0
psubw mm0, mm5
punpcklwd mm3, mm0
movq mm1, mm3
psrlq mm1, 32
movq mm5, mm1
paddw mm1, mm3
psubw mm3, mm5
punpcklwd mm1, mm3
;quant_2x2_dc
MMX_Copy4Times mm3, r2d
MMX_Copy4Times mm2, r1d
MMX_Quant4 mm1, mm0, mm2, mm3
; pNonZeroCount of dct_2x2
pcmpeqb mm2, mm2 ; mm2 = FF
pxor mm3, mm3
packsswb mm1, mm3
pcmpeqb mm1, mm3 ; set FF if equal, 0 if not equal
psubsb mm1, mm2 ; set 0 if equal, 1 if not equal
psadbw mm1, mm3 ;
movd retrd, mm1
WELSEMMS
ret
%macro SSE2_DeQuant8 3
MOVDQ %2, %1
pmullw %2, %3
MOVDQ %1, %2
%endmacro
;***********************************************************************
; void WelsDequant4x4_sse2(int16_t *pDct, const uint16_t* mf);
;***********************************************************************
WELS_EXTERN WelsDequant4x4_sse2
%assign push_num 0
LOAD_2_PARA
movdqa xmm1, [r1]
SSE2_DeQuant8 [r0 ], xmm0, xmm1
SSE2_DeQuant8 [r0 + 0x10], xmm0, xmm1
ret
;***********************************************************************
;void WelsDequantFour4x4_sse2(int16_t *pDct, const uint16_t* mf);
;***********************************************************************
WELS_EXTERN WelsDequantFour4x4_sse2
%assign push_num 0
LOAD_2_PARA
movdqa xmm1, [r1]
SSE2_DeQuant8 [r0 ], xmm0, xmm1
SSE2_DeQuant8 [r0+0x10 ], xmm0, xmm1
SSE2_DeQuant8 [r0+0x20 ], xmm0, xmm1
SSE2_DeQuant8 [r0+0x30 ], xmm0, xmm1
SSE2_DeQuant8 [r0+0x40 ], xmm0, xmm1
SSE2_DeQuant8 [r0+0x50 ], xmm0, xmm1
SSE2_DeQuant8 [r0+0x60 ], xmm0, xmm1
SSE2_DeQuant8 [r0+0x70 ], xmm0, xmm1
ret
;***********************************************************************
;void WelsDequantIHadamard4x4_sse2(int16_t *rs, const uint16_t mf);
;***********************************************************************
WELS_EXTERN WelsDequantIHadamard4x4_sse2
%assign push_num 0
LOAD_2_PARA
%ifndef X86_32
movzx r1, r1w
%endif
; WelsDequantLumaDc4x4
SSE2_Copy8Times xmm1, r1d
;psrlw xmm1, 2 ; for the (>>2) in ihdm
MOVDQ xmm0, [r0]
MOVDQ xmm2, [r0+0x10]
pmullw xmm0, xmm1
pmullw xmm2, xmm1
; ihdm_4x4
movdqa xmm1, xmm0
psrldq xmm1, 8
movdqa xmm3, xmm2
psrldq xmm3, 8
SSE2_SumSub xmm0, xmm3, xmm5 ; xmm0 = xmm0 - xmm3, xmm3 = xmm0 + xmm3
SSE2_SumSub xmm1, xmm2, xmm5 ; xmm1 = xmm1 - xmm2, xmm2 = xmm1 + xmm2
SSE2_SumSub xmm3, xmm2, xmm5 ; xmm3 = xmm3 - xmm2, xmm2 = xmm3 + xmm2
SSE2_SumSub xmm0, xmm1, xmm5 ; xmm0 = xmm0 - xmm1, xmm1 = xmm0 + xmm1
SSE2_TransTwo4x4W xmm2, xmm1, xmm3, xmm0, xmm4
SSE2_SumSub xmm2, xmm4, xmm5
SSE2_SumSub xmm1, xmm0, xmm5
SSE2_SumSub xmm4, xmm0, xmm5
SSE2_SumSub xmm2, xmm1, xmm5
SSE2_TransTwo4x4W xmm0, xmm1, xmm4, xmm2, xmm3
punpcklqdq xmm0, xmm1
MOVDQ [r0], xmm0
punpcklqdq xmm2, xmm3
MOVDQ [r0+16], xmm2
ret
%ifdef HAVE_AVX2
; data=%1 abs_out=%2 ff=%3 mf=%4 7FFFh=%5
%macro AVX2_Quant 5
vpabsw %2, %1
vpor %1, %1, %5 ; ensure non-zero before vpsignw
vpaddusw %2, %2, %3
vpmulhuw %2, %2, %4
vpsignw %1, %2, %1
%endmacro
;***********************************************************************
; void WelsQuant4x4_avx2(int16_t *pDct, int16_t* ff, int16_t *mf);
;***********************************************************************
WELS_EXTERN WelsQuant4x4_avx2
%assign push_num 0
LOAD_3_PARA
PUSH_XMM 5
vbroadcasti128 ymm0, [r1]
vbroadcasti128 ymm1, [r2]
WELS_DW32767_VEX ymm2
vmovdqu ymm3, [r0]
AVX2_Quant ymm3, ymm4, ymm0, ymm1, ymm2
vmovdqu [r0], ymm3
vzeroupper
POP_XMM
ret
;***********************************************************************
;void WelsQuant4x4Dc_avx2(int16_t *pDct, int16_t ff, int16_t mf);
;***********************************************************************
WELS_EXTERN WelsQuant4x4Dc_avx2
%assign push_num 0
LOAD_1_PARA
PUSH_XMM 5
%ifidni r1, arg2
vmovd xmm0, arg2d
vpbroadcastw ymm0, xmm0
%else
vpbroadcastw ymm0, arg2
%endif
%ifidni r2, arg3
vmovd xmm1, arg3d
vpbroadcastw ymm1, xmm1
%else
vpbroadcastw ymm1, arg3
%endif
WELS_DW32767_VEX ymm2
vmovdqu ymm3, [r0]
AVX2_Quant ymm3, ymm4, ymm0, ymm1, ymm2
vmovdqu [r0], ymm3
vzeroupper
POP_XMM
ret
;***********************************************************************
; void WelsQuantFour4x4_avx2(int16_t *pDct, int16_t* ff, int16_t *mf);
;***********************************************************************
WELS_EXTERN WelsQuantFour4x4_avx2
%assign push_num 0
LOAD_3_PARA
PUSH_XMM 6
vbroadcasti128 ymm0, [r1]
vbroadcasti128 ymm1, [r2]
WELS_DW32767_VEX ymm4
vmovdqu ymm3, [r0 + 0x00]
vmovdqu ymm5, [r0 + 0x20]
AVX2_Quant ymm3, ymm2, ymm0, ymm1, ymm4
vmovdqu [r0 + 0x00], ymm3
AVX2_Quant ymm5, ymm2, ymm0, ymm1, ymm4
vmovdqu [r0 + 0x20], ymm5
vmovdqu ymm3, [r0 + 0x40]
vmovdqu ymm5, [r0 + 0x60]
AVX2_Quant ymm3, ymm2, ymm0, ymm1, ymm4
vmovdqu [r0 + 0x40], ymm3
AVX2_Quant ymm5, ymm2, ymm0, ymm1, ymm4
vmovdqu [r0 + 0x60], ymm5
vzeroupper
POP_XMM
ret
;***********************************************************************
; void WelsQuantFour4x4Max_avx2(int16_t *pDct, int32_t* ff, int16_t *mf, int16_t *max);
;***********************************************************************
WELS_EXTERN WelsQuantFour4x4Max_avx2
%assign push_num 0
LOAD_4_PARA
PUSH_XMM 7
vbroadcasti128 ymm0, [r1]
vbroadcasti128 ymm1, [r2]
WELS_DW32767_VEX ymm6
vmovdqu ymm4, [r0 + 0x00]
vmovdqu ymm5, [r0 + 0x20]
AVX2_Quant ymm4, ymm2, ymm0, ymm1, ymm6
vmovdqu [r0 + 0x00], ymm4
AVX2_Quant ymm5, ymm3, ymm0, ymm1, ymm6
vmovdqu [r0 + 0x20], ymm5
vperm2i128 ymm4, ymm2, ymm3, 00100000b
vperm2i128 ymm3, ymm2, ymm3, 00110001b
vpmaxsw ymm2, ymm4, ymm3
vmovdqu ymm4, [r0 + 0x40]
vmovdqu ymm5, [r0 + 0x60]
AVX2_Quant ymm4, ymm3, ymm0, ymm1, ymm6
vmovdqu [r0 + 0x40], ymm4
AVX2_Quant ymm5, ymm4, ymm0, ymm1, ymm6
vmovdqu [r0 + 0x60], ymm5
vperm2i128 ymm5, ymm3, ymm4, 00100000b
vperm2i128 ymm4, ymm3, ymm4, 00110001b
vpmaxsw ymm3, ymm5, ymm4
vpxor ymm2, ymm2, ymm6 ; flip bits so as to enable use of vphminposuw to find max value.
vpxor ymm3, ymm3, ymm6 ; flip bits so as to enable use of vphminposuw to find max value.
vextracti128 xmm4, ymm2, 1
vextracti128 xmm5, ymm3, 1
vphminposuw xmm2, xmm2
vphminposuw xmm3, xmm3
vphminposuw xmm4, xmm4
vphminposuw xmm5, xmm5
vpunpcklwd xmm2, xmm2, xmm4
vpunpcklwd xmm3, xmm3, xmm5
vpunpckldq xmm2, xmm2, xmm3
vpxor xmm2, xmm2, xmm6 ; restore non-flipped values.
vmovq [r3], xmm2 ; store max values.
vzeroupper
POP_XMM
LOAD_4_PARA_POP
ret
%endif