generic-library/vpx
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity
a040eb37e4
vpx/vp8/common/x86/recon_sse2.asm

1054 lines
27 KiB
NASM
Raw Normal View History

Initial WebM release
2010-05-18 17:58:33 +02:00
;
Use WebM in copyright notice for consistency Changes 'The VP8 project' to 'The WebM project', for consistency with other webmproject.org repositories. Fixes issue #97. Change-Id: I37c13ed5fbdb9d334ceef71c6350e9febed9bbba
2010-09-09 14:16:39 +02:00
; Copyright (c) 2010 The WebM project authors. All Rights Reserved.
Initial WebM release
2010-05-18 17:58:33 +02:00
;
cosmetics: trim trailing whitespace When the license headers were updated, they accidentally contained trailing whitespace, so unfortunately we have to touch all the files again. Change-Id: I236c05fade06589e417179c0444cb39b09e4200d
2010-06-18 18:39:21 +02:00
; Use of this source code is governed by a BSD-style license
LICENSE: update with latest text Change-Id: Ieebea089095d9073b3a94932791099f614ce120c
2010-06-04 22:19:40 +02:00
; that can be found in the LICENSE file in the root of the source
; tree. An additional intellectual property rights grant can be found
cosmetics: trim trailing whitespace When the license headers were updated, they accidentally contained trailing whitespace, so unfortunately we have to touch all the files again. Change-Id: I236c05fade06589e417179c0444cb39b09e4200d
2010-06-18 18:39:21 +02:00
; in the file PATENTS. All contributing project authors may
LICENSE: update with latest text Change-Id: Ieebea089095d9073b3a94932791099f614ce120c
2010-06-04 22:19:40 +02:00
; be found in the AUTHORS file in the root of the source tree.
Initial WebM release
2010-05-18 17:58:33 +02:00
;
%include "vpx_ports/x86_abi_support.asm"
;void copy_mem16x16_sse2(
; unsigned char *src,
; int src_stride,
; unsigned char *dst,
; int dst_stride
; )
global sym(vp8_copy_mem16x16_sse2)
sym(vp8_copy_mem16x16_sse2):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 4
push rsi
push rdi
; end prolog
mov rsi, arg(0) ;src;
movdqu xmm0, [rsi]
movsxd rax, dword ptr arg(1) ;src_stride;
mov rdi, arg(2) ;dst;
movdqu xmm1, [rsi+rax]
movdqu xmm2, [rsi+rax*2]
movsxd rcx, dword ptr arg(3) ;dst_stride
lea rsi, [rsi+rax*2]
movdqa [rdi], xmm0
add rsi, rax
movdqa [rdi+rcx], xmm1
movdqa [rdi+rcx*2],xmm2
lea rdi, [rdi+rcx*2]
movdqu xmm3, [rsi]
add rdi, rcx
movdqu xmm4, [rsi+rax]
movdqu xmm5, [rsi+rax*2]
lea rsi, [rsi+rax*2]
movdqa [rdi], xmm3
add rsi, rax
movdqa [rdi+rcx], xmm4
movdqa [rdi+rcx*2],xmm5
lea rdi, [rdi+rcx*2]
movdqu xmm0, [rsi]
add rdi, rcx
movdqu xmm1, [rsi+rax]
movdqu xmm2, [rsi+rax*2]
lea rsi, [rsi+rax*2]
movdqa [rdi], xmm0
add rsi, rax
movdqa [rdi+rcx], xmm1
movdqa [rdi+rcx*2], xmm2
movdqu xmm3, [rsi]
movdqu xmm4, [rsi+rax]
lea rdi, [rdi+rcx*2]
add rdi, rcx
movdqu xmm5, [rsi+rax*2]
lea rsi, [rsi+rax*2]
movdqa [rdi], xmm3
add rsi, rax
movdqa [rdi+rcx], xmm4
movdqa [rdi+rcx*2],xmm5
movdqu xmm0, [rsi]
lea rdi, [rdi+rcx*2]
movdqu xmm1, [rsi+rax]
add rdi, rcx
movdqu xmm2, [rsi+rax*2]
lea rsi, [rsi+rax*2]
movdqa [rdi], xmm0
movdqa [rdi+rcx], xmm1
movdqa [rdi+rcx*2],xmm2
movdqu xmm3, [rsi+rax]
lea rdi, [rdi+rcx*2]
movdqa [rdi+rcx], xmm3
; begin epilog
pop rdi
pop rsi
UNSHADOW_ARGS
pop rbp
ret
SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}(). decoding before 10.425 10.432 10.423 =10.426 after: 10.405 10.416 10.398 =10.406, 0.2% faster encoding before 14.252 14.331 14.250 14.223 14.241 14.220 14.221 =14.248 after 14.095 14.090 14.085 14.095 14.064 14.081 14.089 =14.086, 1.1% faster Change-Id: I483d3d8f0deda8ad434cea76e16028380722aee2
2011-04-27 19:05:10 +02:00
;void vp8_intra_pred_uv_dc_mmx2(
; unsigned char *dst,
; int dst_stride
; unsigned char *src,
; int src_stride,
; )
global sym(vp8_intra_pred_uv_dc_mmx2)
sym(vp8_intra_pred_uv_dc_mmx2):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 4
push rsi
push rdi
; end prolog
; from top
mov rsi, arg(2) ;src;
movsxd rax, dword ptr arg(3) ;src_stride;
sub rsi, rax
pxor mm0, mm0
Use psadbw to get the sum of bytes in a line. Thanks Jason for pointing that out on #vp8. ;-). Change-Id: I5330a753e752a8704b78a409597472628e0b26a5
2011-04-27 22:49:21 +02:00
movq mm1, [rsi]
psadbw mm1, mm0
SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}(). decoding before 10.425 10.432 10.423 =10.426 after: 10.405 10.416 10.398 =10.406, 0.2% faster encoding before 14.252 14.331 14.250 14.223 14.241 14.220 14.221 =14.248 after 14.095 14.090 14.085 14.095 14.064 14.081 14.089 =14.086, 1.1% faster Change-Id: I483d3d8f0deda8ad434cea76e16028380722aee2
2011-04-27 19:05:10 +02:00
; from left
dec rsi
lea rdi, [rax*3]
movzx ecx, byte [rsi+rax]
movzx edx, byte [rsi+rax*2]
add ecx, edx
movzx edx, byte [rsi+rdi]
add ecx, edx
lea rsi, [rsi+rax*4]
movzx edx, byte [rsi]
add ecx, edx
movzx edx, byte [rsi+rax]
add ecx, edx
movzx edx, byte [rsi+rax*2]
add ecx, edx
movzx edx, byte [rsi+rdi]
add ecx, edx
movzx edx, byte [rsi+rax*4]
add ecx, edx
; add up
pextrw edx, mm1, 0x0
lea edx, [edx+ecx+8]
sar edx, 4
movd mm1, edx
pshufw mm1, mm1, 0x0
packuswb mm1, mm1
; write out
mov rdi, arg(0) ;dst;
movsxd rcx, dword ptr arg(1) ;dst_stride
lea rax, [rcx*3]
movq [rdi ], mm1
movq [rdi+rcx ], mm1
movq [rdi+rcx*2], mm1
movq [rdi+rax ], mm1
lea rdi, [rdi+rcx*4]
movq [rdi ], mm1
movq [rdi+rcx ], mm1
movq [rdi+rcx*2], mm1
movq [rdi+rax ], mm1
; begin epilog
pop rdi
pop rsi
UNSHADOW_ARGS
pop rbp
ret
;void vp8_intra_pred_uv_dctop_mmx2(
; unsigned char *dst,
; int dst_stride
; unsigned char *src,
; int src_stride,
; )
global sym(vp8_intra_pred_uv_dctop_mmx2)
sym(vp8_intra_pred_uv_dctop_mmx2):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 4
set up Global Offset Table in recon global values were being referenced, but the GOT was not being set up. as the GOT is only required for PIC, this issue wasn't caught in the default configuration. Change-Id: I8006e53776139362a76f2c80cf9d0f8458602b2f http://code.google.com/p/webm/issues/detail?id=328
2011-05-10 21:58:56 +02:00
GET_GOT rbx
SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}(). decoding before 10.425 10.432 10.423 =10.426 after: 10.405 10.416 10.398 =10.406, 0.2% faster encoding before 14.252 14.331 14.250 14.223 14.241 14.220 14.221 =14.248 after 14.095 14.090 14.085 14.095 14.064 14.081 14.089 =14.086, 1.1% faster Change-Id: I483d3d8f0deda8ad434cea76e16028380722aee2
2011-04-27 19:05:10 +02:00
push rsi
push rdi
; end prolog
; from top
mov rsi, arg(2) ;src;
movsxd rax, dword ptr arg(3) ;src_stride;
sub rsi, rax
pxor mm0, mm0
Use psadbw to get the sum of bytes in a line. Thanks Jason for pointing that out on #vp8. ;-). Change-Id: I5330a753e752a8704b78a409597472628e0b26a5
2011-04-27 22:49:21 +02:00
movq mm1, [rsi]
psadbw mm1, mm0
SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}(). decoding before 10.425 10.432 10.423 =10.426 after: 10.405 10.416 10.398 =10.406, 0.2% faster encoding before 14.252 14.331 14.250 14.223 14.241 14.220 14.221 =14.248 after 14.095 14.090 14.085 14.095 14.064 14.081 14.089 =14.086, 1.1% faster Change-Id: I483d3d8f0deda8ad434cea76e16028380722aee2
2011-04-27 19:05:10 +02:00
; add up
paddw mm1, [GLOBAL(dc_4)]
psraw mm1, 3
pshufw mm1, mm1, 0x0
packuswb mm1, mm1
; write out
mov rdi, arg(0) ;dst;
movsxd rcx, dword ptr arg(1) ;dst_stride
lea rax, [rcx*3]
movq [rdi ], mm1
movq [rdi+rcx ], mm1
movq [rdi+rcx*2], mm1
movq [rdi+rax ], mm1
lea rdi, [rdi+rcx*4]
movq [rdi ], mm1
movq [rdi+rcx ], mm1
movq [rdi+rcx*2], mm1
movq [rdi+rax ], mm1
; begin epilog
pop rdi
pop rsi
set up Global Offset Table in recon global values were being referenced, but the GOT was not being set up. as the GOT is only required for PIC, this issue wasn't caught in the default configuration. Change-Id: I8006e53776139362a76f2c80cf9d0f8458602b2f http://code.google.com/p/webm/issues/detail?id=328
2011-05-10 21:58:56 +02:00
RESTORE_GOT
SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}(). decoding before 10.425 10.432 10.423 =10.426 after: 10.405 10.416 10.398 =10.406, 0.2% faster encoding before 14.252 14.331 14.250 14.223 14.241 14.220 14.221 =14.248 after 14.095 14.090 14.085 14.095 14.064 14.081 14.089 =14.086, 1.1% faster Change-Id: I483d3d8f0deda8ad434cea76e16028380722aee2
2011-04-27 19:05:10 +02:00
UNSHADOW_ARGS
pop rbp
ret
;void vp8_intra_pred_uv_dcleft_mmx2(
; unsigned char *dst,
; int dst_stride
; unsigned char *src,
; int src_stride,
; )
global sym(vp8_intra_pred_uv_dcleft_mmx2)
sym(vp8_intra_pred_uv_dcleft_mmx2):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 4
push rsi
push rdi
; end prolog
; from left
mov rsi, arg(2) ;src;
movsxd rax, dword ptr arg(3) ;src_stride;
dec rsi
lea rdi, [rax*3]
movzx ecx, byte [rsi]
movzx edx, byte [rsi+rax]
add ecx, edx
movzx edx, byte [rsi+rax*2]
add ecx, edx
movzx edx, byte [rsi+rdi]
add ecx, edx
lea rsi, [rsi+rax*4]
movzx edx, byte [rsi]
add ecx, edx
movzx edx, byte [rsi+rax]
add ecx, edx
movzx edx, byte [rsi+rax*2]
add ecx, edx
movzx edx, byte [rsi+rdi]
lea edx, [ecx+edx+4]
; add up
shr edx, 3
movd mm1, edx
pshufw mm1, mm1, 0x0
packuswb mm1, mm1
; write out
mov rdi, arg(0) ;dst;
movsxd rcx, dword ptr arg(1) ;dst_stride
lea rax, [rcx*3]
movq [rdi ], mm1
movq [rdi+rcx ], mm1
movq [rdi+rcx*2], mm1
movq [rdi+rax ], mm1
lea rdi, [rdi+rcx*4]
movq [rdi ], mm1
movq [rdi+rcx ], mm1
movq [rdi+rcx*2], mm1
movq [rdi+rax ], mm1
; begin epilog
pop rdi
pop rsi
UNSHADOW_ARGS
pop rbp
ret
;void vp8_intra_pred_uv_dc128_mmx(
; unsigned char *dst,
; int dst_stride
; unsigned char *src,
; int src_stride,
; )
global sym(vp8_intra_pred_uv_dc128_mmx)
sym(vp8_intra_pred_uv_dc128_mmx):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 4
set up Global Offset Table in recon global values were being referenced, but the GOT was not being set up. as the GOT is only required for PIC, this issue wasn't caught in the default configuration. Change-Id: I8006e53776139362a76f2c80cf9d0f8458602b2f http://code.google.com/p/webm/issues/detail?id=328
2011-05-10 21:58:56 +02:00
GET_GOT rbx
SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}(). decoding before 10.425 10.432 10.423 =10.426 after: 10.405 10.416 10.398 =10.406, 0.2% faster encoding before 14.252 14.331 14.250 14.223 14.241 14.220 14.221 =14.248 after 14.095 14.090 14.085 14.095 14.064 14.081 14.089 =14.086, 1.1% faster Change-Id: I483d3d8f0deda8ad434cea76e16028380722aee2
2011-04-27 19:05:10 +02:00
; end prolog
; write out
movq mm1, [GLOBAL(dc_128)]
mov rax, arg(0) ;dst;
movsxd rdx, dword ptr arg(1) ;dst_stride
lea rcx, [rdx*3]
movq [rax ], mm1
movq [rax+rdx ], mm1
movq [rax+rdx*2], mm1
movq [rax+rcx ], mm1
lea rax, [rax+rdx*4]
movq [rax ], mm1
movq [rax+rdx ], mm1
movq [rax+rdx*2], mm1
movq [rax+rcx ], mm1
; begin epilog
set up Global Offset Table in recon global values were being referenced, but the GOT was not being set up. as the GOT is only required for PIC, this issue wasn't caught in the default configuration. Change-Id: I8006e53776139362a76f2c80cf9d0f8458602b2f http://code.google.com/p/webm/issues/detail?id=328
2011-05-10 21:58:56 +02:00
RESTORE_GOT
SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}(). decoding before 10.425 10.432 10.423 =10.426 after: 10.405 10.416 10.398 =10.406, 0.2% faster encoding before 14.252 14.331 14.250 14.223 14.241 14.220 14.221 =14.248 after 14.095 14.090 14.085 14.095 14.064 14.081 14.089 =14.086, 1.1% faster Change-Id: I483d3d8f0deda8ad434cea76e16028380722aee2
2011-04-27 19:05:10 +02:00
UNSHADOW_ARGS
pop rbp
ret
;void vp8_intra_pred_uv_tm_sse2(
; unsigned char *dst,
; int dst_stride
; unsigned char *src,
; int src_stride,
; )
%macro vp8_intra_pred_uv_tm 1
global sym(vp8_intra_pred_uv_tm_%1)
sym(vp8_intra_pred_uv_tm_%1):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 4
set up Global Offset Table in recon global values were being referenced, but the GOT was not being set up. as the GOT is only required for PIC, this issue wasn't caught in the default configuration. Change-Id: I8006e53776139362a76f2c80cf9d0f8458602b2f http://code.google.com/p/webm/issues/detail?id=328
2011-05-10 21:58:56 +02:00
GET_GOT rbx
SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}(). decoding before 10.425 10.432 10.423 =10.426 after: 10.405 10.416 10.398 =10.406, 0.2% faster encoding before 14.252 14.331 14.250 14.223 14.241 14.220 14.221 =14.248 after 14.095 14.090 14.085 14.095 14.064 14.081 14.089 =14.086, 1.1% faster Change-Id: I483d3d8f0deda8ad434cea76e16028380722aee2
2011-04-27 19:05:10 +02:00
push rsi
push rdi
; end prolog
; read top row
mov edx, 4
mov rsi, arg(2) ;src;
movsxd rax, dword ptr arg(3) ;src_stride;
sub rsi, rax
pxor xmm0, xmm0
%ifidn %1, ssse3
movdqa xmm2, [GLOBAL(dc_1024)]
%endif
movq xmm1, [rsi]
punpcklbw xmm1, xmm0
; set up left ptrs ans subtract topleft
movd xmm3, [rsi-1]
lea rsi, [rsi+rax-1]
%ifidn %1, sse2
punpcklbw xmm3, xmm0
pshuflw xmm3, xmm3, 0x0
punpcklqdq xmm3, xmm3
%else
pshufb xmm3, xmm2
%endif
psubw xmm1, xmm3
; set up dest ptrs
mov rdi, arg(0) ;dst;
movsxd rcx, dword ptr arg(1) ;dst_stride
Use local labels for jumps/loops in x86 assembly. Prepend . to local labels in assembly code. This allows non unique labels within a file. Also makes profiling information more informative by keeping the function name with the loop name. Change-Id: I7a983cb3a5ba2413d5dafd0a37936b268fb9e37f
2011-08-23 00:29:41 +02:00
.vp8_intra_pred_uv_tm_%1_loop:
SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}(). decoding before 10.425 10.432 10.423 =10.426 after: 10.405 10.416 10.398 =10.406, 0.2% faster encoding before 14.252 14.331 14.250 14.223 14.241 14.220 14.221 =14.248 after 14.095 14.090 14.085 14.095 14.064 14.081 14.089 =14.086, 1.1% faster Change-Id: I483d3d8f0deda8ad434cea76e16028380722aee2
2011-04-27 19:05:10 +02:00
movd xmm3, [rsi]
movd xmm5, [rsi+rax]
%ifidn %1, sse2
punpcklbw xmm3, xmm0
punpcklbw xmm5, xmm0
pshuflw xmm3, xmm3, 0x0
pshuflw xmm5, xmm5, 0x0
punpcklqdq xmm3, xmm3
punpcklqdq xmm5, xmm5
%else
pshufb xmm3, xmm2
pshufb xmm5, xmm2
%endif
paddw xmm3, xmm1
paddw xmm5, xmm1
packuswb xmm3, xmm5
movq [rdi ], xmm3
movhps[rdi+rcx], xmm3
lea rsi, [rsi+rax*2]
lea rdi, [rdi+rcx*2]
dec edx
Use local labels for jumps/loops in x86 assembly. Prepend . to local labels in assembly code. This allows non unique labels within a file. Also makes profiling information more informative by keeping the function name with the loop name. Change-Id: I7a983cb3a5ba2413d5dafd0a37936b268fb9e37f
2011-08-23 00:29:41 +02:00
jnz .vp8_intra_pred_uv_tm_%1_loop
SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}(). decoding before 10.425 10.432 10.423 =10.426 after: 10.405 10.416 10.398 =10.406, 0.2% faster encoding before 14.252 14.331 14.250 14.223 14.241 14.220 14.221 =14.248 after 14.095 14.090 14.085 14.095 14.064 14.081 14.089 =14.086, 1.1% faster Change-Id: I483d3d8f0deda8ad434cea76e16028380722aee2
2011-04-27 19:05:10 +02:00
; begin epilog
pop rdi
pop rsi
set up Global Offset Table in recon global values were being referenced, but the GOT was not being set up. as the GOT is only required for PIC, this issue wasn't caught in the default configuration. Change-Id: I8006e53776139362a76f2c80cf9d0f8458602b2f http://code.google.com/p/webm/issues/detail?id=328
2011-05-10 21:58:56 +02:00
RESTORE_GOT
SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}(). decoding before 10.425 10.432 10.423 =10.426 after: 10.405 10.416 10.398 =10.406, 0.2% faster encoding before 14.252 14.331 14.250 14.223 14.241 14.220 14.221 =14.248 after 14.095 14.090 14.085 14.095 14.064 14.081 14.089 =14.086, 1.1% faster Change-Id: I483d3d8f0deda8ad434cea76e16028380722aee2
2011-04-27 19:05:10 +02:00
UNSHADOW_ARGS
pop rbp
ret
%endmacro
vp8_intra_pred_uv_tm sse2
vp8_intra_pred_uv_tm ssse3
;void vp8_intra_pred_uv_ve_mmx(
; unsigned char *dst,
; int dst_stride
; unsigned char *src,
; int src_stride,
; )
global sym(vp8_intra_pred_uv_ve_mmx)
sym(vp8_intra_pred_uv_ve_mmx):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 4
; end prolog
; read from top
mov rax, arg(2) ;src;
movsxd rdx, dword ptr arg(3) ;src_stride;
sub rax, rdx
movq mm1, [rax]
; write out
mov rax, arg(0) ;dst;
movsxd rdx, dword ptr arg(1) ;dst_stride
lea rcx, [rdx*3]
movq [rax ], mm1
movq [rax+rdx ], mm1
movq [rax+rdx*2], mm1
movq [rax+rcx ], mm1
lea rax, [rax+rdx*4]
movq [rax ], mm1
movq [rax+rdx ], mm1
movq [rax+rdx*2], mm1
movq [rax+rcx ], mm1
; begin epilog
UNSHADOW_ARGS
pop rbp
ret
;void vp8_intra_pred_uv_ho_mmx2(
; unsigned char *dst,
; int dst_stride
; unsigned char *src,
; int src_stride,
; )
Make hor UV predict ~2x faster (73 vs 132 cycles) using SSSE3. Change-Id: I658a1df7d825f820573cb2d11ad402f9d2791035
2011-04-29 20:51:37 +02:00
%macro vp8_intra_pred_uv_ho 1
global sym(vp8_intra_pred_uv_ho_%1)
sym(vp8_intra_pred_uv_ho_%1):
SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}(). decoding before 10.425 10.432 10.423 =10.426 after: 10.405 10.416 10.398 =10.406, 0.2% faster encoding before 14.252 14.331 14.250 14.223 14.241 14.220 14.221 =14.248 after 14.095 14.090 14.085 14.095 14.064 14.081 14.089 =14.086, 1.1% faster Change-Id: I483d3d8f0deda8ad434cea76e16028380722aee2
2011-04-27 19:05:10 +02:00
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 4
push rsi
push rdi
Make hor UV predict ~2x faster (73 vs 132 cycles) using SSSE3. Change-Id: I658a1df7d825f820573cb2d11ad402f9d2791035
2011-04-29 20:51:37 +02:00
%ifidn %1, ssse3
Properly use GET_GOT/RESTORE_GOT when using GLOBAL(). This should fix binaries using PIC on x86-32. Also should fix issue 343. Change-Id: I591de3ad68c8a8bb16054bd8f987a75b4e2bad02
2011-06-30 23:04:27 +02:00
%ifndef GET_GOT_SAVE_ARG
Make hor UV predict ~2x faster (73 vs 132 cycles) using SSSE3. Change-Id: I658a1df7d825f820573cb2d11ad402f9d2791035
2011-04-29 20:51:37 +02:00
push rbx
Properly use GET_GOT/RESTORE_GOT when using GLOBAL(). This should fix binaries using PIC on x86-32. Also should fix issue 343. Change-Id: I591de3ad68c8a8bb16054bd8f987a75b4e2bad02
2011-06-30 23:04:27 +02:00
%endif
GET_GOT rbx
Make hor UV predict ~2x faster (73 vs 132 cycles) using SSSE3. Change-Id: I658a1df7d825f820573cb2d11ad402f9d2791035
2011-04-29 20:51:37 +02:00
%endif
SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}(). decoding before 10.425 10.432 10.423 =10.426 after: 10.405 10.416 10.398 =10.406, 0.2% faster encoding before 14.252 14.331 14.250 14.223 14.241 14.220 14.221 =14.248 after 14.095 14.090 14.085 14.095 14.064 14.081 14.089 =14.086, 1.1% faster Change-Id: I483d3d8f0deda8ad434cea76e16028380722aee2
2011-04-27 19:05:10 +02:00
; end prolog
; read from left and write out
Make hor UV predict ~2x faster (73 vs 132 cycles) using SSSE3. Change-Id: I658a1df7d825f820573cb2d11ad402f9d2791035
2011-04-29 20:51:37 +02:00
%ifidn %1, mmx2
SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}(). decoding before 10.425 10.432 10.423 =10.426 after: 10.405 10.416 10.398 =10.406, 0.2% faster encoding before 14.252 14.331 14.250 14.223 14.241 14.220 14.221 =14.248 after 14.095 14.090 14.085 14.095 14.064 14.081 14.089 =14.086, 1.1% faster Change-Id: I483d3d8f0deda8ad434cea76e16028380722aee2
2011-04-27 19:05:10 +02:00
mov edx, 4
Make hor UV predict ~2x faster (73 vs 132 cycles) using SSSE3. Change-Id: I658a1df7d825f820573cb2d11ad402f9d2791035
2011-04-29 20:51:37 +02:00
%endif
SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}(). decoding before 10.425 10.432 10.423 =10.426 after: 10.405 10.416 10.398 =10.406, 0.2% faster encoding before 14.252 14.331 14.250 14.223 14.241 14.220 14.221 =14.248 after 14.095 14.090 14.085 14.095 14.064 14.081 14.089 =14.086, 1.1% faster Change-Id: I483d3d8f0deda8ad434cea76e16028380722aee2
2011-04-27 19:05:10 +02:00
mov rsi, arg(2) ;src;
movsxd rax, dword ptr arg(3) ;src_stride;
mov rdi, arg(0) ;dst;
movsxd rcx, dword ptr arg(1) ;dst_stride
Make hor UV predict ~2x faster (73 vs 132 cycles) using SSSE3. Change-Id: I658a1df7d825f820573cb2d11ad402f9d2791035
2011-04-29 20:51:37 +02:00
%ifidn %1, ssse3
lea rdx, [rcx*3]
movdqa xmm2, [GLOBAL(dc_00001111)]
Properly use GET_GOT/RESTORE_GOT when using GLOBAL(). This should fix binaries using PIC on x86-32. Also should fix issue 343. Change-Id: I591de3ad68c8a8bb16054bd8f987a75b4e2bad02
2011-06-30 23:04:27 +02:00
lea rbx, [rax*3]
Make hor UV predict ~2x faster (73 vs 132 cycles) using SSSE3. Change-Id: I658a1df7d825f820573cb2d11ad402f9d2791035
2011-04-29 20:51:37 +02:00
%endif
SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}(). decoding before 10.425 10.432 10.423 =10.426 after: 10.405 10.416 10.398 =10.406, 0.2% faster encoding before 14.252 14.331 14.250 14.223 14.241 14.220 14.221 =14.248 after 14.095 14.090 14.085 14.095 14.064 14.081 14.089 =14.086, 1.1% faster Change-Id: I483d3d8f0deda8ad434cea76e16028380722aee2
2011-04-27 19:05:10 +02:00
dec rsi
Make hor UV predict ~2x faster (73 vs 132 cycles) using SSSE3. Change-Id: I658a1df7d825f820573cb2d11ad402f9d2791035
2011-04-29 20:51:37 +02:00
%ifidn %1, mmx2
Use local labels for jumps/loops in x86 assembly. Prepend . to local labels in assembly code. This allows non unique labels within a file. Also makes profiling information more informative by keeping the function name with the loop name. Change-Id: I7a983cb3a5ba2413d5dafd0a37936b268fb9e37f
2011-08-23 00:29:41 +02:00
.vp8_intra_pred_uv_ho_%1_loop:
SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}(). decoding before 10.425 10.432 10.423 =10.426 after: 10.405 10.416 10.398 =10.406, 0.2% faster encoding before 14.252 14.331 14.250 14.223 14.241 14.220 14.221 =14.248 after 14.095 14.090 14.085 14.095 14.064 14.081 14.089 =14.086, 1.1% faster Change-Id: I483d3d8f0deda8ad434cea76e16028380722aee2
2011-04-27 19:05:10 +02:00
movd mm0, [rsi]
movd mm1, [rsi+rax]
punpcklbw mm0, mm0
punpcklbw mm1, mm1
pshufw mm0, mm0, 0x0
pshufw mm1, mm1, 0x0
movq [rdi ], mm0
movq [rdi+rcx], mm1
lea rsi, [rsi+rax*2]
lea rdi, [rdi+rcx*2]
dec edx
Use local labels for jumps/loops in x86 assembly. Prepend . to local labels in assembly code. This allows non unique labels within a file. Also makes profiling information more informative by keeping the function name with the loop name. Change-Id: I7a983cb3a5ba2413d5dafd0a37936b268fb9e37f
2011-08-23 00:29:41 +02:00
jnz .vp8_intra_pred_uv_ho_%1_loop
Make hor UV predict ~2x faster (73 vs 132 cycles) using SSSE3. Change-Id: I658a1df7d825f820573cb2d11ad402f9d2791035
2011-04-29 20:51:37 +02:00
%else
movd xmm0, [rsi]
movd xmm3, [rsi+rax]
movd xmm1, [rsi+rax*2]
movd xmm4, [rsi+rbx]
punpcklbw xmm0, xmm3
punpcklbw xmm1, xmm4
pshufb xmm0, xmm2
pshufb xmm1, xmm2
movq [rdi ], xmm0
movhps [rdi+rcx], xmm0
movq [rdi+rcx*2], xmm1
movhps [rdi+rdx], xmm1
lea rsi, [rsi+rax*4]
lea rdi, [rdi+rcx*4]
movd xmm0, [rsi]
movd xmm3, [rsi+rax]
movd xmm1, [rsi+rax*2]
movd xmm4, [rsi+rbx]
punpcklbw xmm0, xmm3
punpcklbw xmm1, xmm4
pshufb xmm0, xmm2
pshufb xmm1, xmm2
movq [rdi ], xmm0
movhps [rdi+rcx], xmm0
movq [rdi+rcx*2], xmm1
movhps [rdi+rdx], xmm1
%endif
SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}(). decoding before 10.425 10.432 10.423 =10.426 after: 10.405 10.416 10.398 =10.406, 0.2% faster encoding before 14.252 14.331 14.250 14.223 14.241 14.220 14.221 =14.248 after 14.095 14.090 14.085 14.095 14.064 14.081 14.089 =14.086, 1.1% faster Change-Id: I483d3d8f0deda8ad434cea76e16028380722aee2
2011-04-27 19:05:10 +02:00
; begin epilog
Make hor UV predict ~2x faster (73 vs 132 cycles) using SSSE3. Change-Id: I658a1df7d825f820573cb2d11ad402f9d2791035
2011-04-29 20:51:37 +02:00
%ifidn %1, ssse3
Properly use GET_GOT/RESTORE_GOT when using GLOBAL(). This should fix binaries using PIC on x86-32. Also should fix issue 343. Change-Id: I591de3ad68c8a8bb16054bd8f987a75b4e2bad02
2011-06-30 23:04:27 +02:00
RESTORE_GOT
%ifndef GET_GOT_SAVE_ARG
Make hor UV predict ~2x faster (73 vs 132 cycles) using SSSE3. Change-Id: I658a1df7d825f820573cb2d11ad402f9d2791035
2011-04-29 20:51:37 +02:00
pop rbx
Properly use GET_GOT/RESTORE_GOT when using GLOBAL(). This should fix binaries using PIC on x86-32. Also should fix issue 343. Change-Id: I591de3ad68c8a8bb16054bd8f987a75b4e2bad02
2011-06-30 23:04:27 +02:00
%endif
Make hor UV predict ~2x faster (73 vs 132 cycles) using SSSE3. Change-Id: I658a1df7d825f820573cb2d11ad402f9d2791035
2011-04-29 20:51:37 +02:00
%endif
SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}(). decoding before 10.425 10.432 10.423 =10.426 after: 10.405 10.416 10.398 =10.406, 0.2% faster encoding before 14.252 14.331 14.250 14.223 14.241 14.220 14.221 =14.248 after 14.095 14.090 14.085 14.095 14.064 14.081 14.089 =14.086, 1.1% faster Change-Id: I483d3d8f0deda8ad434cea76e16028380722aee2
2011-04-27 19:05:10 +02:00
pop rdi
pop rsi
UNSHADOW_ARGS
pop rbp
ret
Make hor UV predict ~2x faster (73 vs 132 cycles) using SSSE3. Change-Id: I658a1df7d825f820573cb2d11ad402f9d2791035
2011-04-29 20:51:37 +02:00
%endmacro
vp8_intra_pred_uv_ho mmx2
vp8_intra_pred_uv_ho ssse3
SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}(). decoding before 10.425 10.432 10.423 =10.426 after: 10.405 10.416 10.398 =10.406, 0.2% faster encoding before 14.252 14.331 14.250 14.223 14.241 14.220 14.221 =14.248 after 14.095 14.090 14.085 14.095 14.064 14.081 14.089 =14.086, 1.1% faster Change-Id: I483d3d8f0deda8ad434cea76e16028380722aee2
2011-04-27 19:05:10 +02:00
SSE2 optimizations for vp8_build_intra_predictors_mby{,_s}() Ronald recently sent me this patch that he did in April. > From: Ronald S. Bultje <rbultje@google.com> > Date: Thu, 28 Apr 2011 17:30:15 -0700 > Subject: [PATCH] SSE2 optimizations for > vp8_build_intra_predictors_mby{,_s}(). HD decode tests have shown a performance boost up to 1.5%, depending on material. Patch set 3: Fixed encoder crash. Change-Id: Ie1fd1fa3dc750eec1a7a20bfa2decc079dcf48c8
2011-11-09 21:30:35 +01:00
;void vp8_intra_pred_y_dc_sse2(
; unsigned char *dst,
; int dst_stride
; unsigned char *src,
; int src_stride,
; )
global sym(vp8_intra_pred_y_dc_sse2)
sym(vp8_intra_pred_y_dc_sse2):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 4
push rsi
push rdi
; end prolog
; from top
mov rsi, arg(2) ;src;
movsxd rax, dword ptr arg(3) ;src_stride;
sub rsi, rax
pxor xmm0, xmm0
movdqa xmm1, [rsi]
psadbw xmm1, xmm0
movq xmm2, xmm1
punpckhqdq xmm1, xmm1
paddw xmm1, xmm2
; from left
dec rsi
lea rdi, [rax*3]
movzx ecx, byte [rsi+rax]
movzx edx, byte [rsi+rax*2]
add ecx, edx
movzx edx, byte [rsi+rdi]
add ecx, edx
lea rsi, [rsi+rax*4]
movzx edx, byte [rsi]
add ecx, edx
movzx edx, byte [rsi+rax]
add ecx, edx
movzx edx, byte [rsi+rax*2]
add ecx, edx
movzx edx, byte [rsi+rdi]
add ecx, edx
lea rsi, [rsi+rax*4]
movzx edx, byte [rsi]
add ecx, edx
movzx edx, byte [rsi+rax]
add ecx, edx
movzx edx, byte [rsi+rax*2]
add ecx, edx
movzx edx, byte [rsi+rdi]
add ecx, edx
lea rsi, [rsi+rax*4]
movzx edx, byte [rsi]
add ecx, edx
movzx edx, byte [rsi+rax]
add ecx, edx
movzx edx, byte [rsi+rax*2]
add ecx, edx
movzx edx, byte [rsi+rdi]
add ecx, edx
movzx edx, byte [rsi+rax*4]
add ecx, edx
; add up
pextrw edx, xmm1, 0x0
lea edx, [edx+ecx+16]
sar edx, 5
movd xmm1, edx
; FIXME use pshufb for ssse3 version
pshuflw xmm1, xmm1, 0x0
punpcklqdq xmm1, xmm1
packuswb xmm1, xmm1
; write out
mov rsi, 2
mov rdi, arg(0) ;dst;
movsxd rcx, dword ptr arg(1) ;dst_stride
lea rax, [rcx*3]
.label
movdqa [rdi ], xmm1
movdqa [rdi+rcx ], xmm1
movdqa [rdi+rcx*2], xmm1
movdqa [rdi+rax ], xmm1
lea rdi, [rdi+rcx*4]
movdqa [rdi ], xmm1
movdqa [rdi+rcx ], xmm1
movdqa [rdi+rcx*2], xmm1
movdqa [rdi+rax ], xmm1
lea rdi, [rdi+rcx*4]
dec rsi
jnz .label
; begin epilog
pop rdi
pop rsi
UNSHADOW_ARGS
pop rbp
ret
;void vp8_intra_pred_y_dctop_sse2(
; unsigned char *dst,
; int dst_stride
; unsigned char *src,
; int src_stride,
; )
global sym(vp8_intra_pred_y_dctop_sse2)
sym(vp8_intra_pred_y_dctop_sse2):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 4
push rsi
GET_GOT rbx
; end prolog
; from top
mov rcx, arg(2) ;src;
movsxd rax, dword ptr arg(3) ;src_stride;
sub rcx, rax
pxor xmm0, xmm0
movdqa xmm1, [rcx]
psadbw xmm1, xmm0
movdqa xmm2, xmm1
punpckhqdq xmm1, xmm1
paddw xmm1, xmm2
; add up
paddw xmm1, [GLOBAL(dc_8)]
psraw xmm1, 4
; FIXME use pshufb for ssse3 version
pshuflw xmm1, xmm1, 0x0
punpcklqdq xmm1, xmm1
packuswb xmm1, xmm1
; write out
mov rsi, 2
mov rdx, arg(0) ;dst;
movsxd rcx, dword ptr arg(1) ;dst_stride
lea rax, [rcx*3]
.label
movdqa [rdx ], xmm1
movdqa [rdx+rcx ], xmm1
movdqa [rdx+rcx*2], xmm1
movdqa [rdx+rax ], xmm1
lea rdx, [rdx+rcx*4]
movdqa [rdx ], xmm1
movdqa [rdx+rcx ], xmm1
movdqa [rdx+rcx*2], xmm1
movdqa [rdx+rax ], xmm1
lea rdx, [rdx+rcx*4]
dec rsi
jnz .label
; begin epilog
RESTORE_GOT
pop rsi
UNSHADOW_ARGS
pop rbp
ret
;void vp8_intra_pred_y_dcleft_sse2(
; unsigned char *dst,
; int dst_stride
; unsigned char *src,
; int src_stride,
; )
global sym(vp8_intra_pred_y_dcleft_sse2)
sym(vp8_intra_pred_y_dcleft_sse2):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 4
push rsi
push rdi
; end prolog
; from left
mov rsi, arg(2) ;src;
movsxd rax, dword ptr arg(3) ;src_stride;
dec rsi
lea rdi, [rax*3]
movzx ecx, byte [rsi]
movzx edx, byte [rsi+rax]
add ecx, edx
movzx edx, byte [rsi+rax*2]
add ecx, edx
movzx edx, byte [rsi+rdi]
add ecx, edx
lea rsi, [rsi+rax*4]
movzx edx, byte [rsi]
add ecx, edx
movzx edx, byte [rsi+rax]
add ecx, edx
movzx edx, byte [rsi+rax*2]
add ecx, edx
movzx edx, byte [rsi+rdi]
add ecx, edx
lea rsi, [rsi+rax*4]
movzx edx, byte [rsi]
add ecx, edx
movzx edx, byte [rsi+rax]
add ecx, edx
movzx edx, byte [rsi+rax*2]
add ecx, edx
movzx edx, byte [rsi+rdi]
add ecx, edx
lea rsi, [rsi+rax*4]
movzx edx, byte [rsi]
add ecx, edx
movzx edx, byte [rsi+rax]
add ecx, edx
movzx edx, byte [rsi+rax*2]
add ecx, edx
movzx edx, byte [rsi+rdi]
lea edx, [ecx+edx+8]
; add up
shr edx, 4
movd xmm1, edx
; FIXME use pshufb for ssse3 version
pshuflw xmm1, xmm1, 0x0
punpcklqdq xmm1, xmm1
packuswb xmm1, xmm1
; write out
mov rsi, 2
mov rdi, arg(0) ;dst;
movsxd rcx, dword ptr arg(1) ;dst_stride
lea rax, [rcx*3]
.label
movdqa [rdi ], xmm1
movdqa [rdi+rcx ], xmm1
movdqa [rdi+rcx*2], xmm1
movdqa [rdi+rax ], xmm1
lea rdi, [rdi+rcx*4]
movdqa [rdi ], xmm1
movdqa [rdi+rcx ], xmm1
movdqa [rdi+rcx*2], xmm1
movdqa [rdi+rax ], xmm1
lea rdi, [rdi+rcx*4]
dec rsi
jnz .label
; begin epilog
pop rdi
pop rsi
UNSHADOW_ARGS
pop rbp
ret
;void vp8_intra_pred_y_dc128_sse2(
; unsigned char *dst,
; int dst_stride
; unsigned char *src,
; int src_stride,
; )
global sym(vp8_intra_pred_y_dc128_sse2)
sym(vp8_intra_pred_y_dc128_sse2):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 4
push rsi
GET_GOT rbx
; end prolog
; write out
mov rsi, 2
movdqa xmm1, [GLOBAL(dc_128)]
mov rax, arg(0) ;dst;
movsxd rdx, dword ptr arg(1) ;dst_stride
lea rcx, [rdx*3]
.label
movdqa [rax ], xmm1
movdqa [rax+rdx ], xmm1
movdqa [rax+rdx*2], xmm1
movdqa [rax+rcx ], xmm1
lea rax, [rax+rdx*4]
movdqa [rax ], xmm1
movdqa [rax+rdx ], xmm1
movdqa [rax+rdx*2], xmm1
movdqa [rax+rcx ], xmm1
lea rax, [rax+rdx*4]
dec rsi
jnz .label
; begin epilog
RESTORE_GOT
pop rsi
UNSHADOW_ARGS
pop rbp
ret
;void vp8_intra_pred_y_tm_sse2(
; unsigned char *dst,
; int dst_stride
; unsigned char *src,
; int src_stride,
; )
%macro vp8_intra_pred_y_tm 1
global sym(vp8_intra_pred_y_tm_%1)
sym(vp8_intra_pred_y_tm_%1):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 4
push rsi
push rdi
GET_GOT rbx
; end prolog
; read top row
mov edx, 8
mov rsi, arg(2) ;src;
movsxd rax, dword ptr arg(3) ;src_stride;
sub rsi, rax
pxor xmm0, xmm0
%ifidn %1, ssse3
movdqa xmm3, [GLOBAL(dc_1024)]
%endif
movdqa xmm1, [rsi]
movdqa xmm2, xmm1
punpcklbw xmm1, xmm0
punpckhbw xmm2, xmm0
; set up left ptrs ans subtract topleft
movd xmm4, [rsi-1]
lea rsi, [rsi+rax-1]
%ifidn %1, sse2
punpcklbw xmm4, xmm0
pshuflw xmm4, xmm4, 0x0
punpcklqdq xmm4, xmm4
%else
pshufb xmm4, xmm3
%endif
psubw xmm1, xmm4
psubw xmm2, xmm4
; set up dest ptrs
mov rdi, arg(0) ;dst;
movsxd rcx, dword ptr arg(1) ;dst_stride
vp8_intra_pred_y_tm_%1_loop:
movd xmm4, [rsi]
movd xmm5, [rsi+rax]
%ifidn %1, sse2
punpcklbw xmm4, xmm0
punpcklbw xmm5, xmm0
pshuflw xmm4, xmm4, 0x0
pshuflw xmm5, xmm5, 0x0
punpcklqdq xmm4, xmm4
punpcklqdq xmm5, xmm5
%else
pshufb xmm4, xmm3
pshufb xmm5, xmm3
%endif
movdqa xmm6, xmm4
movdqa xmm7, xmm5
paddw xmm4, xmm1
paddw xmm6, xmm2
paddw xmm5, xmm1
paddw xmm7, xmm2
packuswb xmm4, xmm6
packuswb xmm5, xmm7
movdqa [rdi ], xmm4
movdqa [rdi+rcx], xmm5
lea rsi, [rsi+rax*2]
lea rdi, [rdi+rcx*2]
dec edx
jnz vp8_intra_pred_y_tm_%1_loop
; begin epilog
RESTORE_GOT
pop rdi
pop rsi
UNSHADOW_ARGS
pop rbp
ret
%endmacro
vp8_intra_pred_y_tm sse2
vp8_intra_pred_y_tm ssse3
;void vp8_intra_pred_y_ve_sse2(
; unsigned char *dst,
; int dst_stride
; unsigned char *src,
; int src_stride,
; )
global sym(vp8_intra_pred_y_ve_sse2)
sym(vp8_intra_pred_y_ve_sse2):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 4
push rsi
; end prolog
; read from top
mov rax, arg(2) ;src;
movsxd rdx, dword ptr arg(3) ;src_stride;
sub rax, rdx
movdqa xmm1, [rax]
; write out
mov rsi, 2
mov rax, arg(0) ;dst;
movsxd rdx, dword ptr arg(1) ;dst_stride
lea rcx, [rdx*3]
.label
movdqa [rax ], xmm1
movdqa [rax+rdx ], xmm1
movdqa [rax+rdx*2], xmm1
movdqa [rax+rcx ], xmm1
lea rax, [rax+rdx*4]
movdqa [rax ], xmm1
movdqa [rax+rdx ], xmm1
movdqa [rax+rdx*2], xmm1
movdqa [rax+rcx ], xmm1
lea rax, [rax+rdx*4]
dec rsi
jnz .label
; begin epilog
pop rsi
UNSHADOW_ARGS
pop rbp
ret
;void vp8_intra_pred_y_ho_sse2(
; unsigned char *dst,
; int dst_stride
; unsigned char *src,
; int src_stride,
; )
global sym(vp8_intra_pred_y_ho_sse2)
sym(vp8_intra_pred_y_ho_sse2):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 4
push rsi
push rdi
; end prolog
; read from left and write out
mov edx, 8
mov rsi, arg(2) ;src;
movsxd rax, dword ptr arg(3) ;src_stride;
mov rdi, arg(0) ;dst;
movsxd rcx, dword ptr arg(1) ;dst_stride
dec rsi
vp8_intra_pred_y_ho_sse2_loop:
movd xmm0, [rsi]
movd xmm1, [rsi+rax]
; FIXME use pshufb for ssse3 version
punpcklbw xmm0, xmm0
punpcklbw xmm1, xmm1
pshuflw xmm0, xmm0, 0x0
pshuflw xmm1, xmm1, 0x0
punpcklqdq xmm0, xmm0
punpcklqdq xmm1, xmm1
movdqa [rdi ], xmm0
movdqa [rdi+rcx], xmm1
lea rsi, [rsi+rax*2]
lea rdi, [rdi+rcx*2]
dec edx
jnz vp8_intra_pred_y_ho_sse2_loop
; begin epilog
pop rdi
pop rsi
UNSHADOW_ARGS
pop rbp
ret
SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}(). decoding before 10.425 10.432 10.423 =10.426 after: 10.405 10.416 10.398 =10.406, 0.2% faster encoding before 14.252 14.331 14.250 14.223 14.241 14.220 14.221 =14.248 after 14.095 14.090 14.085 14.095 14.064 14.081 14.089 =14.086, 1.1% faster Change-Id: I483d3d8f0deda8ad434cea76e16028380722aee2
2011-04-27 19:05:10 +02:00
SECTION_RODATA
SSE2 optimizations for vp8_build_intra_predictors_mby{,_s}() Ronald recently sent me this patch that he did in April. > From: Ronald S. Bultje <rbultje@google.com> > Date: Thu, 28 Apr 2011 17:30:15 -0700 > Subject: [PATCH] SSE2 optimizations for > vp8_build_intra_predictors_mby{,_s}(). HD decode tests have shown a performance boost up to 1.5%, depending on material. Patch set 3: Fixed encoder crash. Change-Id: Ie1fd1fa3dc750eec1a7a20bfa2decc079dcf48c8
2011-11-09 21:30:35 +01:00
align 16
SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}(). decoding before 10.425 10.432 10.423 =10.426 after: 10.405 10.416 10.398 =10.406, 0.2% faster encoding before 14.252 14.331 14.250 14.223 14.241 14.220 14.221 =14.248 after 14.095 14.090 14.085 14.095 14.064 14.081 14.089 =14.086, 1.1% faster Change-Id: I483d3d8f0deda8ad434cea76e16028380722aee2
2011-04-27 19:05:10 +02:00
dc_128:
SSE2 optimizations for vp8_build_intra_predictors_mby{,_s}() Ronald recently sent me this patch that he did in April. > From: Ronald S. Bultje <rbultje@google.com> > Date: Thu, 28 Apr 2011 17:30:15 -0700 > Subject: [PATCH] SSE2 optimizations for > vp8_build_intra_predictors_mby{,_s}(). HD decode tests have shown a performance boost up to 1.5%, depending on material. Patch set 3: Fixed encoder crash. Change-Id: Ie1fd1fa3dc750eec1a7a20bfa2decc079dcf48c8
2011-11-09 21:30:35 +01:00
times 16 db 128
SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}(). decoding before 10.425 10.432 10.423 =10.426 after: 10.405 10.416 10.398 =10.406, 0.2% faster encoding before 14.252 14.331 14.250 14.223 14.241 14.220 14.221 =14.248 after 14.095 14.090 14.085 14.095 14.064 14.081 14.089 =14.086, 1.1% faster Change-Id: I483d3d8f0deda8ad434cea76e16028380722aee2
2011-04-27 19:05:10 +02:00
dc_4:
times 4 dw 4
align 16
SSE2 optimizations for vp8_build_intra_predictors_mby{,_s}() Ronald recently sent me this patch that he did in April. > From: Ronald S. Bultje <rbultje@google.com> > Date: Thu, 28 Apr 2011 17:30:15 -0700 > Subject: [PATCH] SSE2 optimizations for > vp8_build_intra_predictors_mby{,_s}(). HD decode tests have shown a performance boost up to 1.5%, depending on material. Patch set 3: Fixed encoder crash. Change-Id: Ie1fd1fa3dc750eec1a7a20bfa2decc079dcf48c8
2011-11-09 21:30:35 +01:00
dc_8:
times 8 dw 8
align 16
SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}(). decoding before 10.425 10.432 10.423 =10.426 after: 10.405 10.416 10.398 =10.406, 0.2% faster encoding before 14.252 14.331 14.250 14.223 14.241 14.220 14.221 =14.248 after 14.095 14.090 14.085 14.095 14.064 14.081 14.089 =14.086, 1.1% faster Change-Id: I483d3d8f0deda8ad434cea76e16028380722aee2
2011-04-27 19:05:10 +02:00
dc_1024:
times 8 dw 0x400
Make hor UV predict ~2x faster (73 vs 132 cycles) using SSSE3. Change-Id: I658a1df7d825f820573cb2d11ad402f9d2791035
2011-04-29 20:51:37 +02:00
align 16
dc_00001111:
times 8 db 0
times 8 db 1
Reference in New Issue Copy Permalink