generic-library/vpx
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity
b336dc6bff
vpx/vp8/encoder/x86/sad_sse4.asm

354 lines
9.9 KiB
NASM
Raw Normal View History

Full search SAD function optimization in SSE4.1 Use mpsadbw, and calculate 8 sad at once. Function list: vp8_sad16x16x8_sse4 vp8_sad16x8x8_sse4 vp8_sad8x16x8_sse4 vp8_sad8x8x8_sse4 vp8_sad4x4x8_sse4 (test clip: tulip) For best quality mode, this gave encoder a 5% performance boost. For good quality mode with speed=1, this gave encoder a 3% performance boost. Change-Id: I083b5a39d39144f88dcbccbef95da6498e490134
2010-10-27 14:45:24 +02:00
;
; Copyright (c) 2010 The WebM project authors. All Rights Reserved.
;
; Use of this source code is governed by a BSD-style license
; that can be found in the LICENSE file in the root of the source
; tree. An additional intellectual property rights grant can be found
; in the file PATENTS. All contributing project authors may
; be found in the AUTHORS file in the root of the source tree.
;
%include "vpx_ports/x86_abi_support.asm"
%macro PROCESS_16X2X8 1
%if %1
movdqa xmm0, XMMWORD PTR [rsi]
movq xmm1, MMWORD PTR [rdi]
movq xmm3, MMWORD PTR [rdi+8]
movq xmm2, MMWORD PTR [rdi+16]
punpcklqdq xmm1, xmm3
punpcklqdq xmm3, xmm2
movdqa xmm2, xmm1
mpsadbw xmm1, xmm0, 0x0
mpsadbw xmm2, xmm0, 0x5
psrldq xmm0, 8
movdqa xmm4, xmm3
mpsadbw xmm3, xmm0, 0x0
mpsadbw xmm4, xmm0, 0x5
paddw xmm1, xmm2
paddw xmm1, xmm3
paddw xmm1, xmm4
%else
movdqa xmm0, XMMWORD PTR [rsi]
movq xmm5, MMWORD PTR [rdi]
movq xmm3, MMWORD PTR [rdi+8]
movq xmm2, MMWORD PTR [rdi+16]
punpcklqdq xmm5, xmm3
punpcklqdq xmm3, xmm2
movdqa xmm2, xmm5
mpsadbw xmm5, xmm0, 0x0
mpsadbw xmm2, xmm0, 0x5
psrldq xmm0, 8
movdqa xmm4, xmm3
mpsadbw xmm3, xmm0, 0x0
mpsadbw xmm4, xmm0, 0x5
paddw xmm5, xmm2
paddw xmm5, xmm3
paddw xmm5, xmm4
paddw xmm1, xmm5
%endif
movdqa xmm0, XMMWORD PTR [rsi + rax]
movq xmm5, MMWORD PTR [rdi+ rdx]
movq xmm3, MMWORD PTR [rdi+ rdx+8]
movq xmm2, MMWORD PTR [rdi+ rdx+16]
punpcklqdq xmm5, xmm3
punpcklqdq xmm3, xmm2
lea rsi, [rsi+rax*2]
lea rdi, [rdi+rdx*2]
movdqa xmm2, xmm5
mpsadbw xmm5, xmm0, 0x0
mpsadbw xmm2, xmm0, 0x5
psrldq xmm0, 8
movdqa xmm4, xmm3
mpsadbw xmm3, xmm0, 0x0
mpsadbw xmm4, xmm0, 0x5
paddw xmm5, xmm2
paddw xmm5, xmm3
paddw xmm5, xmm4
paddw xmm1, xmm5
%endmacro
%macro PROCESS_8X2X8 1
%if %1
movq xmm0, MMWORD PTR [rsi]
movq xmm1, MMWORD PTR [rdi]
movq xmm3, MMWORD PTR [rdi+8]
punpcklqdq xmm1, xmm3
movdqa xmm2, xmm1
mpsadbw xmm1, xmm0, 0x0
mpsadbw xmm2, xmm0, 0x5
paddw xmm1, xmm2
%else
movq xmm0, MMWORD PTR [rsi]
movq xmm5, MMWORD PTR [rdi]
movq xmm3, MMWORD PTR [rdi+8]
punpcklqdq xmm5, xmm3
movdqa xmm2, xmm5
mpsadbw xmm5, xmm0, 0x0
mpsadbw xmm2, xmm0, 0x5
paddw xmm5, xmm2
paddw xmm1, xmm5
%endif
movq xmm0, MMWORD PTR [rsi + rax]
movq xmm5, MMWORD PTR [rdi+ rdx]
movq xmm3, MMWORD PTR [rdi+ rdx+8]
punpcklqdq xmm5, xmm3
lea rsi, [rsi+rax*2]
lea rdi, [rdi+rdx*2]
movdqa xmm2, xmm5
mpsadbw xmm5, xmm0, 0x0
mpsadbw xmm2, xmm0, 0x5
paddw xmm5, xmm2
paddw xmm1, xmm5
%endmacro
%macro PROCESS_4X2X8 1
%if %1
movd xmm0, [rsi]
movq xmm1, MMWORD PTR [rdi]
movq xmm3, MMWORD PTR [rdi+8]
punpcklqdq xmm1, xmm3
mpsadbw xmm1, xmm0, 0x0
%else
movd xmm0, [rsi]
movq xmm5, MMWORD PTR [rdi]
movq xmm3, MMWORD PTR [rdi+8]
punpcklqdq xmm5, xmm3
mpsadbw xmm5, xmm0, 0x0
paddw xmm1, xmm5
%endif
movd xmm0, [rsi + rax]
movq xmm5, MMWORD PTR [rdi+ rdx]
movq xmm3, MMWORD PTR [rdi+ rdx+8]
punpcklqdq xmm5, xmm3
lea rsi, [rsi+rax*2]
lea rdi, [rdi+rdx*2]
mpsadbw xmm5, xmm0, 0x0
paddw xmm1, xmm5
%endmacro
;void vp8_sad16x16x8_sse4(
; const unsigned char *src_ptr,
; int src_stride,
; const unsigned char *ref_ptr,
; int ref_stride,
; unsigned short *sad_array);
global sym(vp8_sad16x16x8_sse4)
sym(vp8_sad16x16x8_sse4):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 5
push rsi
push rdi
; end prolog
mov rsi, arg(0) ;src_ptr
mov rdi, arg(2) ;ref_ptr
movsxd rax, dword ptr arg(1) ;src_stride
movsxd rdx, dword ptr arg(3) ;ref_stride
PROCESS_16X2X8 1
PROCESS_16X2X8 0
PROCESS_16X2X8 0
PROCESS_16X2X8 0
PROCESS_16X2X8 0
PROCESS_16X2X8 0
PROCESS_16X2X8 0
PROCESS_16X2X8 0
mov rdi, arg(4) ;Results
Align SAD output array to be 16-byte aligned Use aligned store. Change-Id: Icab4c0c53da811d0c52bb7e8134927f249ba2499
2011-03-11 14:24:23 +01:00
movdqa XMMWORD PTR [rdi], xmm1
Full search SAD function optimization in SSE4.1 Use mpsadbw, and calculate 8 sad at once. Function list: vp8_sad16x16x8_sse4 vp8_sad16x8x8_sse4 vp8_sad8x16x8_sse4 vp8_sad8x8x8_sse4 vp8_sad4x4x8_sse4 (test clip: tulip) For best quality mode, this gave encoder a 5% performance boost. For good quality mode with speed=1, this gave encoder a 3% performance boost. Change-Id: I083b5a39d39144f88dcbccbef95da6498e490134
2010-10-27 14:45:24 +02:00
; begin epilog
pop rdi
pop rsi
UNSHADOW_ARGS
pop rbp
ret
;void vp8_sad16x8x8_sse4(
; const unsigned char *src_ptr,
; int src_stride,
; const unsigned char *ref_ptr,
; int ref_stride,
; unsigned short *sad_array
;);
global sym(vp8_sad16x8x8_sse4)
sym(vp8_sad16x8x8_sse4):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 5
push rsi
push rdi
; end prolog
mov rsi, arg(0) ;src_ptr
mov rdi, arg(2) ;ref_ptr
movsxd rax, dword ptr arg(1) ;src_stride
movsxd rdx, dword ptr arg(3) ;ref_stride
PROCESS_16X2X8 1
PROCESS_16X2X8 0
PROCESS_16X2X8 0
PROCESS_16X2X8 0
mov rdi, arg(4) ;Results
Align SAD output array to be 16-byte aligned Use aligned store. Change-Id: Icab4c0c53da811d0c52bb7e8134927f249ba2499
2011-03-11 14:24:23 +01:00
movdqa XMMWORD PTR [rdi], xmm1
Full search SAD function optimization in SSE4.1 Use mpsadbw, and calculate 8 sad at once. Function list: vp8_sad16x16x8_sse4 vp8_sad16x8x8_sse4 vp8_sad8x16x8_sse4 vp8_sad8x8x8_sse4 vp8_sad4x4x8_sse4 (test clip: tulip) For best quality mode, this gave encoder a 5% performance boost. For good quality mode with speed=1, this gave encoder a 3% performance boost. Change-Id: I083b5a39d39144f88dcbccbef95da6498e490134
2010-10-27 14:45:24 +02:00
; begin epilog
pop rdi
pop rsi
UNSHADOW_ARGS
pop rbp
ret
;void vp8_sad8x8x8_sse4(
; const unsigned char *src_ptr,
; int src_stride,
; const unsigned char *ref_ptr,
; int ref_stride,
; unsigned short *sad_array
;);
global sym(vp8_sad8x8x8_sse4)
sym(vp8_sad8x8x8_sse4):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 5
push rsi
push rdi
; end prolog
mov rsi, arg(0) ;src_ptr
mov rdi, arg(2) ;ref_ptr
movsxd rax, dword ptr arg(1) ;src_stride
movsxd rdx, dword ptr arg(3) ;ref_stride
PROCESS_8X2X8 1
PROCESS_8X2X8 0
PROCESS_8X2X8 0
PROCESS_8X2X8 0
mov rdi, arg(4) ;Results
Align SAD output array to be 16-byte aligned Use aligned store. Change-Id: Icab4c0c53da811d0c52bb7e8134927f249ba2499
2011-03-11 14:24:23 +01:00
movdqa XMMWORD PTR [rdi], xmm1
Full search SAD function optimization in SSE4.1 Use mpsadbw, and calculate 8 sad at once. Function list: vp8_sad16x16x8_sse4 vp8_sad16x8x8_sse4 vp8_sad8x16x8_sse4 vp8_sad8x8x8_sse4 vp8_sad4x4x8_sse4 (test clip: tulip) For best quality mode, this gave encoder a 5% performance boost. For good quality mode with speed=1, this gave encoder a 3% performance boost. Change-Id: I083b5a39d39144f88dcbccbef95da6498e490134
2010-10-27 14:45:24 +02:00
; begin epilog
pop rdi
pop rsi
UNSHADOW_ARGS
pop rbp
ret
;void vp8_sad8x16x8_sse4(
; const unsigned char *src_ptr,
; int src_stride,
; const unsigned char *ref_ptr,
; int ref_stride,
; unsigned short *sad_array
;);
global sym(vp8_sad8x16x8_sse4)
sym(vp8_sad8x16x8_sse4):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 5
push rsi
push rdi
; end prolog
mov rsi, arg(0) ;src_ptr
mov rdi, arg(2) ;ref_ptr
movsxd rax, dword ptr arg(1) ;src_stride
movsxd rdx, dword ptr arg(3) ;ref_stride
PROCESS_8X2X8 1
PROCESS_8X2X8 0
PROCESS_8X2X8 0
PROCESS_8X2X8 0
PROCESS_8X2X8 0
PROCESS_8X2X8 0
PROCESS_8X2X8 0
PROCESS_8X2X8 0
mov rdi, arg(4) ;Results
Align SAD output array to be 16-byte aligned Use aligned store. Change-Id: Icab4c0c53da811d0c52bb7e8134927f249ba2499
2011-03-11 14:24:23 +01:00
movdqa XMMWORD PTR [rdi], xmm1
Full search SAD function optimization in SSE4.1 Use mpsadbw, and calculate 8 sad at once. Function list: vp8_sad16x16x8_sse4 vp8_sad16x8x8_sse4 vp8_sad8x16x8_sse4 vp8_sad8x8x8_sse4 vp8_sad4x4x8_sse4 (test clip: tulip) For best quality mode, this gave encoder a 5% performance boost. For good quality mode with speed=1, this gave encoder a 3% performance boost. Change-Id: I083b5a39d39144f88dcbccbef95da6498e490134
2010-10-27 14:45:24 +02:00
; begin epilog
pop rdi
pop rsi
UNSHADOW_ARGS
pop rbp
ret
;void vp8_sad4x4x8_c(
; const unsigned char *src_ptr,
; int src_stride,
; const unsigned char *ref_ptr,
; int ref_stride,
; unsigned short *sad_array
;);
global sym(vp8_sad4x4x8_sse4)
sym(vp8_sad4x4x8_sse4):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 5
push rsi
push rdi
; end prolog
mov rsi, arg(0) ;src_ptr
mov rdi, arg(2) ;ref_ptr
movsxd rax, dword ptr arg(1) ;src_stride
movsxd rdx, dword ptr arg(3) ;ref_stride
PROCESS_4X2X8 1
PROCESS_4X2X8 0
mov rdi, arg(4) ;Results
Align SAD output array to be 16-byte aligned Use aligned store. Change-Id: Icab4c0c53da811d0c52bb7e8134927f249ba2499
2011-03-11 14:24:23 +01:00
movdqa XMMWORD PTR [rdi], xmm1
Full search SAD function optimization in SSE4.1 Use mpsadbw, and calculate 8 sad at once. Function list: vp8_sad16x16x8_sse4 vp8_sad16x8x8_sse4 vp8_sad8x16x8_sse4 vp8_sad8x8x8_sse4 vp8_sad4x4x8_sse4 (test clip: tulip) For best quality mode, this gave encoder a 5% performance boost. For good quality mode with speed=1, this gave encoder a 3% performance boost. Change-Id: I083b5a39d39144f88dcbccbef95da6498e490134
2010-10-27 14:45:24 +02:00
; begin epilog
pop rdi
pop rsi
UNSHADOW_ARGS
pop rbp
ret
Reference in New Issue Copy Permalink