From 419f638910245f5501fcad4eede1efcab0bd22ee Mon Sep 17 00:00:00 2001 From: Yunqing Wang Date: Tue, 8 Mar 2011 16:25:06 -0500 Subject: Improve SSE2 half-pixel filter funtions Rewrote these functions to process 16 pixels once instead of 8. Change-Id: Ic67e80124467a446a3df4cfecfb76a4248602adb --- vp8/encoder/x86/variance_impl_sse2.asm | 355 +++++++++++++++++++++++++++++++-- vp8/encoder/x86/variance_sse2.c | 116 +++++------ vp8/encoder/x86/variance_ssse3.c | 34 +--- 3 files changed, 391 insertions(+), 114 deletions(-) (limited to 'vp8/encoder/x86') diff --git a/vp8/encoder/x86/variance_impl_sse2.asm b/vp8/encoder/x86/variance_impl_sse2.asm index 5d1a17d44..c2c30deb2 100644 --- a/vp8/encoder/x86/variance_impl_sse2.asm +++ b/vp8/encoder/x86/variance_impl_sse2.asm @@ -790,7 +790,7 @@ filter_block2d_bil_variance: ret -;void vp8_half_horiz_vert_variance16x_h_sse2 +;void vp8_half_horiz_vert_variance8x_h_sse2 ;( ; unsigned char *ref_ptr, ; int ref_pixels_per_line, @@ -800,8 +800,8 @@ filter_block2d_bil_variance: ; int *sum, ; unsigned int *sumsquared ;) -global sym(vp8_half_horiz_vert_variance16x_h_sse2) -sym(vp8_half_horiz_vert_variance16x_h_sse2): +global sym(vp8_half_horiz_vert_variance8x_h_sse2) +sym(vp8_half_horiz_vert_variance8x_h_sse2): push rbp mov rbp, rsp SHADOW_ARGS_TO_STACK 7 @@ -835,7 +835,7 @@ sym(vp8_half_horiz_vert_variance16x_h_sse2): add rsi, r8 %endif -vp8_half_horiz_vert_variance16x_h_1: +vp8_half_horiz_vert_variance8x_h_1: movq xmm1, QWORD PTR [rsi] ; movq xmm2, QWORD PTR [rsi+1] ; @@ -863,7 +863,7 @@ vp8_half_horiz_vert_variance16x_h_1: %endif sub rcx, 1 ; - jnz vp8_half_horiz_vert_variance16x_h_1 ; + jnz vp8_half_horiz_vert_variance8x_h_1 ; movdq2q mm6, xmm6 ; movdq2q mm7, xmm7 ; @@ -910,8 +910,123 @@ vp8_half_horiz_vert_variance16x_h_1: pop rbp ret +;void vp8_half_horiz_vert_variance16x_h_sse2 +;( +; unsigned char *ref_ptr, +; int ref_pixels_per_line, +; unsigned char *src_ptr, +; int src_pixels_per_line, +; unsigned int Height, +; int *sum, +; unsigned int *sumsquared +;) +global sym(vp8_half_horiz_vert_variance16x_h_sse2) +sym(vp8_half_horiz_vert_variance16x_h_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 7 + SAVE_XMM + GET_GOT rbx + push rsi + push rdi + ; end prolog -;void vp8_half_vert_variance16x_h_sse2 + pxor xmm6, xmm6 ; error accumulator + pxor xmm7, xmm7 ; sse eaccumulator + mov rsi, arg(0) ;ref_ptr ; + + mov rdi, arg(2) ;src_ptr ; + movsxd rcx, dword ptr arg(4) ;Height ; + movsxd rax, dword ptr arg(1) ;ref_pixels_per_line + movsxd rdx, dword ptr arg(3) ;src_pixels_per_line + + pxor xmm0, xmm0 ; + + movdqu xmm5, XMMWORD PTR [rsi] + movdqu xmm3, XMMWORD PTR [rsi+1] + pavgb xmm5, xmm3 ; xmm5 = avg(xmm1,xmm3) horizontal line 1 + + lea rsi, [rsi + rax] + +vp8_half_horiz_vert_variance16x_h_1: + movdqu xmm1, XMMWORD PTR [rsi] ; + movdqu xmm2, XMMWORD PTR [rsi+1] ; + pavgb xmm1, xmm2 ; xmm1 = avg(xmm1,xmm3) horizontal line i+1 + + pavgb xmm5, xmm1 ; xmm = vertical average of the above + + movdqa xmm4, xmm5 + punpcklbw xmm5, xmm0 ; xmm5 = words of above + punpckhbw xmm4, xmm0 + + movq xmm3, QWORD PTR [rdi] ; xmm3 = d0,d1,d2..d7 + punpcklbw xmm3, xmm0 ; xmm3 = words of above + psubw xmm5, xmm3 ; xmm5 -= xmm3 + + movq xmm3, QWORD PTR [rdi+8] + punpcklbw xmm3, xmm0 + psubw xmm4, xmm3 + + paddw xmm6, xmm5 ; xmm6 += accumulated column differences + paddw xmm6, xmm4 + pmaddwd xmm5, xmm5 ; xmm5 *= xmm5 + pmaddwd xmm4, xmm4 + paddd xmm7, xmm5 ; xmm7 += accumulated square column differences + paddd xmm7, xmm4 + + movdqa xmm5, xmm1 ; save xmm1 for use on the next row + + lea rsi, [rsi + rax] + lea rdi, [rdi + rdx] + + sub rcx, 1 ; + jnz vp8_half_horiz_vert_variance16x_h_1 ; + + pxor xmm1, xmm1 + pxor xmm5, xmm5 + + punpcklwd xmm0, xmm6 + punpckhwd xmm1, xmm6 + psrad xmm0, 16 + psrad xmm1, 16 + paddd xmm0, xmm1 + movdqa xmm1, xmm0 + + movdqa xmm6, xmm7 + punpckldq xmm6, xmm5 + punpckhdq xmm7, xmm5 + paddd xmm6, xmm7 + + punpckldq xmm0, xmm5 + punpckhdq xmm1, xmm5 + paddd xmm0, xmm1 + + movdqa xmm7, xmm6 + movdqa xmm1, xmm0 + + psrldq xmm7, 8 + psrldq xmm1, 8 + + paddd xmm6, xmm7 + paddd xmm0, xmm1 + + mov rsi, arg(5) ;[Sum] + mov rdi, arg(6) ;[SSE] + + movd [rsi], xmm0 + movd [rdi], xmm6 + + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + + +;void vp8_half_vert_variance8x_h_sse2 ;( ; unsigned char *ref_ptr, ; int ref_pixels_per_line, @@ -921,8 +1036,8 @@ vp8_half_horiz_vert_variance16x_h_1: ; int *sum, ; unsigned int *sumsquared ;) -global sym(vp8_half_vert_variance16x_h_sse2) -sym(vp8_half_vert_variance16x_h_sse2): +global sym(vp8_half_vert_variance8x_h_sse2) +sym(vp8_half_vert_variance8x_h_sse2): push rbp mov rbp, rsp SHADOW_ARGS_TO_STACK 7 @@ -945,7 +1060,7 @@ sym(vp8_half_vert_variance16x_h_sse2): movsxd rax, dword ptr arg(1) ;ref_pixels_per_line pxor xmm0, xmm0 ; -vp8_half_vert_variance16x_h_1: +vp8_half_vert_variance8x_h_1: movq xmm5, QWORD PTR [rsi] ; xmm5 = s0,s1,s2..s8 movq xmm3, QWORD PTR [rsi+rax] ; xmm3 = s1,s2,s3..s9 @@ -969,7 +1084,7 @@ vp8_half_vert_variance16x_h_1: %endif sub rcx, 1 ; - jnz vp8_half_vert_variance16x_h_1 ; + jnz vp8_half_vert_variance8x_h_1 ; movdq2q mm6, xmm6 ; movdq2q mm7, xmm7 ; @@ -1016,8 +1131,115 @@ vp8_half_vert_variance16x_h_1: pop rbp ret +;void vp8_half_vert_variance16x_h_sse2 +;( +; unsigned char *ref_ptr, +; int ref_pixels_per_line, +; unsigned char *src_ptr, +; int src_pixels_per_line, +; unsigned int Height, +; int *sum, +; unsigned int *sumsquared +;) +global sym(vp8_half_vert_variance16x_h_sse2) +sym(vp8_half_vert_variance16x_h_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 7 + SAVE_XMM + GET_GOT rbx + push rsi + push rdi + ; end prolog -;void vp8_half_horiz_variance16x_h_sse2 + pxor xmm6, xmm6 ; error accumulator + pxor xmm7, xmm7 ; sse eaccumulator + mov rsi, arg(0) ;ref_ptr + + mov rdi, arg(2) ;src_ptr + movsxd rcx, dword ptr arg(4) ;Height + movsxd rax, dword ptr arg(1) ;ref_pixels_per_line + movsxd rdx, dword ptr arg(3) ;src_pixels_per_line + + movdqu xmm5, XMMWORD PTR [rsi] + lea rsi, [rsi + rax ] + pxor xmm0, xmm0 + +vp8_half_vert_variance16x_h_1: + movdqu xmm3, XMMWORD PTR [rsi] + + pavgb xmm5, xmm3 ; xmm5 = avg(xmm1,xmm3) + movdqa xmm4, xmm5 + punpcklbw xmm5, xmm0 + punpckhbw xmm4, xmm0 + + movq xmm2, QWORD PTR [rdi] + punpcklbw xmm2, xmm0 + psubw xmm5, xmm2 + movq xmm2, QWORD PTR [rdi+8] + punpcklbw xmm2, xmm0 + psubw xmm4, xmm2 + + paddw xmm6, xmm5 ; xmm6 += accumulated column differences + paddw xmm6, xmm4 + pmaddwd xmm5, xmm5 ; xmm5 *= xmm5 + pmaddwd xmm4, xmm4 + paddd xmm7, xmm5 ; xmm7 += accumulated square column differences + paddd xmm7, xmm4 + + movdqa xmm5, xmm3 + + lea rsi, [rsi + rax] + lea rdi, [rdi + rdx] + + sub rcx, 1 + jnz vp8_half_vert_variance16x_h_1 + + pxor xmm1, xmm1 + pxor xmm5, xmm5 + + punpcklwd xmm0, xmm6 + punpckhwd xmm1, xmm6 + psrad xmm0, 16 + psrad xmm1, 16 + paddd xmm0, xmm1 + movdqa xmm1, xmm0 + + movdqa xmm6, xmm7 + punpckldq xmm6, xmm5 + punpckhdq xmm7, xmm5 + paddd xmm6, xmm7 + + punpckldq xmm0, xmm5 + punpckhdq xmm1, xmm5 + paddd xmm0, xmm1 + + movdqa xmm7, xmm6 + movdqa xmm1, xmm0 + + psrldq xmm7, 8 + psrldq xmm1, 8 + + paddd xmm6, xmm7 + paddd xmm0, xmm1 + + mov rsi, arg(5) ;[Sum] + mov rdi, arg(6) ;[SSE] + + movd [rsi], xmm0 + movd [rdi], xmm6 + + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret + + +;void vp8_half_horiz_variance8x_h_sse2 ;( ; unsigned char *ref_ptr, ; int ref_pixels_per_line, @@ -1027,8 +1249,8 @@ vp8_half_vert_variance16x_h_1: ; int *sum, ; unsigned int *sumsquared ;) -global sym(vp8_half_horiz_variance16x_h_sse2) -sym(vp8_half_horiz_variance16x_h_sse2): +global sym(vp8_half_horiz_variance8x_h_sse2) +sym(vp8_half_horiz_variance8x_h_sse2): push rbp mov rbp, rsp SHADOW_ARGS_TO_STACK 7 @@ -1050,7 +1272,7 @@ sym(vp8_half_horiz_variance16x_h_sse2): movsxd rcx, dword ptr arg(4) ;Height ; pxor xmm0, xmm0 ; -vp8_half_horiz_variance16x16_1: +vp8_half_horiz_variance8x_h_1: movq xmm5, QWORD PTR [rsi] ; xmm5 = s0,s1,s2..s8 movq xmm3, QWORD PTR [rsi+1] ; xmm3 = s1,s2,s3..s9 @@ -1073,7 +1295,7 @@ vp8_half_horiz_variance16x16_1: add rdi, r9 %endif sub rcx, 1 ; - jnz vp8_half_horiz_variance16x16_1 ; + jnz vp8_half_horiz_variance8x_h_1 ; movdq2q mm6, xmm6 ; movdq2q mm7, xmm7 ; @@ -1120,6 +1342,109 @@ vp8_half_horiz_variance16x16_1: pop rbp ret +;void vp8_half_horiz_variance16x_h_sse2 +;( +; unsigned char *ref_ptr, +; int ref_pixels_per_line, +; unsigned char *src_ptr, +; int src_pixels_per_line, +; unsigned int Height, +; int *sum, +; unsigned int *sumsquared +;) +global sym(vp8_half_horiz_variance16x_h_sse2) +sym(vp8_half_horiz_variance16x_h_sse2): + push rbp + mov rbp, rsp + SHADOW_ARGS_TO_STACK 7 + SAVE_XMM + GET_GOT rbx + push rsi + push rdi + ; end prolog + + pxor xmm6, xmm6 ; error accumulator + pxor xmm7, xmm7 ; sse eaccumulator + mov rsi, arg(0) ;ref_ptr ; + + mov rdi, arg(2) ;src_ptr ; + movsxd rcx, dword ptr arg(4) ;Height ; + movsxd rax, dword ptr arg(1) ;ref_pixels_per_line + movsxd rdx, dword ptr arg(3) ;src_pixels_per_line + + pxor xmm0, xmm0 ; + +vp8_half_horiz_variance16x_h_1: + movdqu xmm5, XMMWORD PTR [rsi] ; xmm5 = s0,s1,s2..s15 + movdqu xmm3, XMMWORD PTR [rsi+1] ; xmm3 = s1,s2,s3..s16 + + pavgb xmm5, xmm3 ; xmm5 = avg(xmm1,xmm3) + movdqa xmm1, xmm5 + punpcklbw xmm5, xmm0 ; xmm5 = words of above + punpckhbw xmm1, xmm0 + + movq xmm3, QWORD PTR [rdi] ; xmm3 = d0,d1,d2..d7 + punpcklbw xmm3, xmm0 ; xmm3 = words of above + movq xmm2, QWORD PTR [rdi+8] + punpcklbw xmm2, xmm0 + + psubw xmm5, xmm3 ; xmm5 -= xmm3 + psubw xmm1, xmm2 + paddw xmm6, xmm5 ; xmm6 += accumulated column differences + paddw xmm6, xmm1 + pmaddwd xmm5, xmm5 ; xmm5 *= xmm5 + pmaddwd xmm1, xmm1 + paddd xmm7, xmm5 ; xmm7 += accumulated square column differences + paddd xmm7, xmm1 + + lea rsi, [rsi + rax] + lea rdi, [rdi + rdx] + + sub rcx, 1 ; + jnz vp8_half_horiz_variance16x_h_1 ; + + pxor xmm1, xmm1 + pxor xmm5, xmm5 + + punpcklwd xmm0, xmm6 + punpckhwd xmm1, xmm6 + psrad xmm0, 16 + psrad xmm1, 16 + paddd xmm0, xmm1 + movdqa xmm1, xmm0 + + movdqa xmm6, xmm7 + punpckldq xmm6, xmm5 + punpckhdq xmm7, xmm5 + paddd xmm6, xmm7 + + punpckldq xmm0, xmm5 + punpckhdq xmm1, xmm5 + paddd xmm0, xmm1 + + movdqa xmm7, xmm6 + movdqa xmm1, xmm0 + + psrldq xmm7, 8 + psrldq xmm1, 8 + + paddd xmm6, xmm7 + paddd xmm0, xmm1 + + mov rsi, arg(5) ;[Sum] + mov rdi, arg(6) ;[SSE] + + movd [rsi], xmm0 + movd [rdi], xmm6 + + ; begin epilog + pop rdi + pop rsi + RESTORE_GOT + RESTORE_XMM + UNSHADOW_ARGS + pop rbp + ret SECTION_RODATA ; short xmm_bi_rd[8] = { 64, 64, 64, 64,64, 64, 64, 64}; diff --git a/vp8/encoder/x86/variance_sse2.c b/vp8/encoder/x86/variance_sse2.c index 7cf6a6308..4612a6711 100644 --- a/vp8/encoder/x86/variance_sse2.c +++ b/vp8/encoder/x86/variance_sse2.c @@ -81,6 +81,16 @@ void vp8_filter_block2d_bil_var_sse2 int *sum, unsigned int *sumsquared ); +void vp8_half_horiz_vert_variance8x_h_sse2 +( + const unsigned char *ref_ptr, + int ref_pixels_per_line, + const unsigned char *src_ptr, + int src_pixels_per_line, + unsigned int Height, + int *sum, + unsigned int *sumsquared +); void vp8_half_horiz_vert_variance16x_h_sse2 ( const unsigned char *ref_ptr, @@ -91,6 +101,16 @@ void vp8_half_horiz_vert_variance16x_h_sse2 int *sum, unsigned int *sumsquared ); +void vp8_half_horiz_variance8x_h_sse2 +( + const unsigned char *ref_ptr, + int ref_pixels_per_line, + const unsigned char *src_ptr, + int src_pixels_per_line, + unsigned int Height, + int *sum, + unsigned int *sumsquared +); void vp8_half_horiz_variance16x_h_sse2 ( const unsigned char *ref_ptr, @@ -101,6 +121,16 @@ void vp8_half_horiz_variance16x_h_sse2 int *sum, unsigned int *sumsquared ); +void vp8_half_vert_variance8x_h_sse2 +( + const unsigned char *ref_ptr, + int ref_pixels_per_line, + const unsigned char *src_ptr, + int src_pixels_per_line, + unsigned int Height, + int *sum, + unsigned int *sumsquared +); void vp8_half_vert_variance16x_h_sse2 ( const unsigned char *ref_ptr, @@ -262,21 +292,21 @@ unsigned int vp8_sub_pixel_variance8x8_wmt if (xoffset == 4 && yoffset == 0) { - vp8_half_horiz_variance16x_h_sse2( + vp8_half_horiz_variance8x_h_sse2( src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, 8, &xsum, &xxsum); } else if (xoffset == 0 && yoffset == 4) { - vp8_half_vert_variance16x_h_sse2( + vp8_half_vert_variance8x_h_sse2( src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, 8, &xsum, &xxsum); } else if (xoffset == 4 && yoffset == 4) { - vp8_half_horiz_vert_variance16x_h_sse2( + vp8_half_horiz_vert_variance8x_h_sse2( src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, 8, &xsum, &xxsum); @@ -317,11 +347,6 @@ unsigned int vp8_sub_pixel_variance16x16_wmt src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, 16, &xsum0, &xxsum0); - - vp8_half_horiz_variance16x_h_sse2( - src_ptr + 8, src_pixels_per_line, - dst_ptr + 8, dst_pixels_per_line, 16, - &xsum1, &xxsum1); } else if (xoffset == 0 && yoffset == 4) { @@ -329,11 +354,6 @@ unsigned int vp8_sub_pixel_variance16x16_wmt src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, 16, &xsum0, &xxsum0); - - vp8_half_vert_variance16x_h_sse2( - src_ptr + 8, src_pixels_per_line, - dst_ptr + 8, dst_pixels_per_line, 16, - &xsum1, &xxsum1); } else if (xoffset == 4 && yoffset == 4) { @@ -341,11 +361,6 @@ unsigned int vp8_sub_pixel_variance16x16_wmt src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, 16, &xsum0, &xxsum0); - - vp8_half_horiz_vert_variance16x_h_sse2( - src_ptr + 8, src_pixels_per_line, - dst_ptr + 8, dst_pixels_per_line, 16, - &xsum1, &xxsum1); } else { @@ -356,17 +371,16 @@ unsigned int vp8_sub_pixel_variance16x16_wmt &xsum0, &xxsum0 ); - vp8_filter_block2d_bil_var_sse2( src_ptr + 8, src_pixels_per_line, dst_ptr + 8, dst_pixels_per_line, 16, xoffset, yoffset, &xsum1, &xxsum1 ); + xsum0 += xsum1; + xxsum0 += xxsum1; } - xsum0 += xsum1; - xxsum0 += xxsum1; *sse = xxsum0; return (xxsum0 - ((xsum0 * xsum0) >> 8)); } @@ -406,11 +420,6 @@ unsigned int vp8_sub_pixel_variance16x8_wmt src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, 8, &xsum0, &xxsum0); - - vp8_half_horiz_variance16x_h_sse2( - src_ptr + 8, src_pixels_per_line, - dst_ptr + 8, dst_pixels_per_line, 8, - &xsum1, &xxsum1); } else if (xoffset == 0 && yoffset == 4) { @@ -418,11 +427,6 @@ unsigned int vp8_sub_pixel_variance16x8_wmt src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, 8, &xsum0, &xxsum0); - - vp8_half_vert_variance16x_h_sse2( - src_ptr + 8, src_pixels_per_line, - dst_ptr + 8, dst_pixels_per_line, 8, - &xsum1, &xxsum1); } else if (xoffset == 4 && yoffset == 4) { @@ -430,11 +434,6 @@ unsigned int vp8_sub_pixel_variance16x8_wmt src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, 8, &xsum0, &xxsum0); - - vp8_half_horiz_vert_variance16x_h_sse2( - src_ptr + 8, src_pixels_per_line, - dst_ptr + 8, dst_pixels_per_line, 8, - &xsum1, &xxsum1); } else { @@ -449,11 +448,10 @@ unsigned int vp8_sub_pixel_variance16x8_wmt dst_ptr + 8, dst_pixels_per_line, 8, xoffset, yoffset, &xsum1, &xxsum1); + xsum0 += xsum1; + xxsum0 += xxsum1; } - xsum0 += xsum1; - xxsum0 += xxsum1; - *sse = xxsum0; return (xxsum0 - ((xsum0 * xsum0) >> 7)); } @@ -474,21 +472,21 @@ unsigned int vp8_sub_pixel_variance8x16_wmt if (xoffset == 4 && yoffset == 0) { - vp8_half_horiz_variance16x_h_sse2( + vp8_half_horiz_variance8x_h_sse2( src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, 16, &xsum, &xxsum); } else if (xoffset == 0 && yoffset == 4) { - vp8_half_vert_variance16x_h_sse2( + vp8_half_vert_variance8x_h_sse2( src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, 16, &xsum, &xxsum); } else if (xoffset == 4 && yoffset == 4) { - vp8_half_horiz_vert_variance16x_h_sse2( + vp8_half_horiz_vert_variance8x_h_sse2( src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, 16, &xsum, &xxsum); @@ -589,21 +587,14 @@ unsigned int vp8_variance_halfpixvar16x16_h_wmt( int dst_pixels_per_line, unsigned int *sse) { - int xsum0, xsum1; - unsigned int xxsum0, xxsum1; + int xsum0; + unsigned int xxsum0; vp8_half_horiz_variance16x_h_sse2( src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, 16, &xsum0, &xxsum0); - vp8_half_horiz_variance16x_h_sse2( - src_ptr + 8, src_pixels_per_line, - dst_ptr + 8, dst_pixels_per_line, 16, - &xsum1, &xxsum1); - - xsum0 += xsum1; - xxsum0 += xxsum1; *sse = xxsum0; return (xxsum0 - ((xsum0 * xsum0) >> 8)); } @@ -616,21 +607,13 @@ unsigned int vp8_variance_halfpixvar16x16_v_wmt( int dst_pixels_per_line, unsigned int *sse) { - int xsum0, xsum1; - unsigned int xxsum0, xxsum1; - + int xsum0; + unsigned int xxsum0; vp8_half_vert_variance16x_h_sse2( src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, 16, &xsum0, &xxsum0); - vp8_half_vert_variance16x_h_sse2( - src_ptr + 8, src_pixels_per_line, - dst_ptr + 8, dst_pixels_per_line, 16, - &xsum1, &xxsum1); - - xsum0 += xsum1; - xxsum0 += xxsum1; *sse = xxsum0; return (xxsum0 - ((xsum0 * xsum0) >> 8)); } @@ -643,21 +626,14 @@ unsigned int vp8_variance_halfpixvar16x16_hv_wmt( int dst_pixels_per_line, unsigned int *sse) { - int xsum0, xsum1; - unsigned int xxsum0, xxsum1; + int xsum0; + unsigned int xxsum0; vp8_half_horiz_vert_variance16x_h_sse2( src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, 16, &xsum0, &xxsum0); - vp8_half_horiz_vert_variance16x_h_sse2( - src_ptr + 8, src_pixels_per_line, - dst_ptr + 8, dst_pixels_per_line, 16, - &xsum1, &xxsum1); - - xsum0 += xsum1; - xxsum0 += xxsum1; *sse = xxsum0; return (xxsum0 - ((xsum0 * xsum0) >> 8)); } diff --git a/vp8/encoder/x86/variance_ssse3.c b/vp8/encoder/x86/variance_ssse3.c index 750ae8b86..d50ae3ade 100644 --- a/vp8/encoder/x86/variance_ssse3.c +++ b/vp8/encoder/x86/variance_ssse3.c @@ -87,14 +87,6 @@ unsigned int vp8_sub_pixel_variance16x16_ssse3 src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, 16, &xsum0, &xxsum0); - - vp8_half_horiz_variance16x_h_sse2( - src_ptr + 8, src_pixels_per_line, - dst_ptr + 8, dst_pixels_per_line, 16, - &xsum1, &xxsum1); - - xsum0 += xsum1; - xxsum0 += xxsum1; } else if (xoffset == 0 && yoffset == 4) { @@ -102,14 +94,6 @@ unsigned int vp8_sub_pixel_variance16x16_ssse3 src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, 16, &xsum0, &xxsum0); - - vp8_half_vert_variance16x_h_sse2( - src_ptr + 8, src_pixels_per_line, - dst_ptr + 8, dst_pixels_per_line, 16, - &xsum1, &xxsum1); - - xsum0 += xsum1; - xxsum0 += xxsum1; } else if (xoffset == 4 && yoffset == 4) { @@ -117,22 +101,14 @@ unsigned int vp8_sub_pixel_variance16x16_ssse3 src_ptr, src_pixels_per_line, dst_ptr, dst_pixels_per_line, 16, &xsum0, &xxsum0); - - vp8_half_horiz_vert_variance16x_h_sse2( - src_ptr + 8, src_pixels_per_line, - dst_ptr + 8, dst_pixels_per_line, 16, - &xsum1, &xxsum1); - - xsum0 += xsum1; - xxsum0 += xxsum1; } else { - vp8_filter_block2d_bil_var_ssse3( - src_ptr, src_pixels_per_line, - dst_ptr, dst_pixels_per_line, 16, - xoffset, yoffset, - &xsum0, &xxsum0); + vp8_filter_block2d_bil_var_ssse3( + src_ptr, src_pixels_per_line, + dst_ptr, dst_pixels_per_line, 16, + xoffset, yoffset, + &xsum0, &xxsum0); } *sse = xxsum0; -- cgit v1.2.3 From 4561109a69338336d7ab1875fd2fd62f80392b14 Mon Sep 17 00:00:00 2001 From: Yunqing Wang Date: Wed, 9 Mar 2011 10:45:03 -0500 Subject: Remove unused functions Removed some unused functions Change-Id: Ifdfc27453e53cfc75997b38492901d193a16b245 --- vp8/encoder/x86/variance_mmx.c | 140 ---------------------------------------- vp8/encoder/x86/variance_sse2.c | 75 --------------------- 2 files changed, 215 deletions(-) (limited to 'vp8/encoder/x86') diff --git a/vp8/encoder/x86/variance_mmx.c b/vp8/encoder/x86/variance_mmx.c index 6eed98e07..07358c0c7 100644 --- a/vp8/encoder/x86/variance_mmx.c +++ b/vp8/encoder/x86/variance_mmx.c @@ -456,146 +456,6 @@ unsigned int vp8_sub_pixel_variance8x16_mmx return (xxsum - ((xsum * xsum) >> 7)); } -unsigned int vp8_i_variance16x16_mmx( - const unsigned char *src_ptr, - int source_stride, - const unsigned char *ref_ptr, - int recon_stride, - unsigned int *sse) -{ - unsigned int sse0, sse1, sse2, sse3, var; - int sum0, sum1, sum2, sum3, avg; - - - vp8_get8x8var_mmx(src_ptr, source_stride, ref_ptr, recon_stride, &sse0, &sum0) ; - vp8_get8x8var_mmx(src_ptr + 8, source_stride, ref_ptr + 8, recon_stride, &sse1, &sum1); - vp8_get8x8var_mmx(src_ptr + (source_stride >> 1), source_stride, ref_ptr + (recon_stride >> 1), recon_stride, &sse2, &sum2) ; - vp8_get8x8var_mmx(src_ptr + (source_stride >> 1) + 8, source_stride, ref_ptr + (recon_stride >> 1) + 8, recon_stride, &sse3, &sum3); - - var = sse0 + sse1 + sse2 + sse3; - avg = sum0 + sum1 + sum2 + sum3; - *sse = var; - return (var - ((avg * avg) >> 8)); - -} - -unsigned int vp8_i_variance8x16_mmx( - const unsigned char *src_ptr, - int source_stride, - const unsigned char *ref_ptr, - int recon_stride, - unsigned int *sse) -{ - unsigned int sse0, sse1, var; - int sum0, sum1, avg; - vp8_get8x8var_mmx(src_ptr, source_stride, ref_ptr, recon_stride, &sse0, &sum0) ; - vp8_get8x8var_mmx(src_ptr + (source_stride >> 1), source_stride, ref_ptr + (recon_stride >> 1), recon_stride, &sse1, &sum1) ; - - var = sse0 + sse1; - avg = sum0 + sum1; - - *sse = var; - return (var - ((avg * avg) >> 7)); - -} - -unsigned int vp8_i_sub_pixel_variance16x16_mmx -( - const unsigned char *src_ptr, - int src_pixels_per_line, - int xoffset, - int yoffset, - const unsigned char *dst_ptr, - int dst_pixels_per_line, - unsigned int *sse -) -{ - int xsum0, xsum1; - unsigned int xxsum0, xxsum1; - int f2soffset = (src_pixels_per_line >> 1); - int f2doffset = (dst_pixels_per_line >> 1); - - - vp8_filter_block2d_bil_var_mmx( - src_ptr, src_pixels_per_line, - dst_ptr, dst_pixels_per_line, 8, - vp8_vp7_bilinear_filters_mmx[xoffset], vp8_vp7_bilinear_filters_mmx[yoffset], - &xsum0, &xxsum0 - ); - - - vp8_filter_block2d_bil_var_mmx( - src_ptr + 8, src_pixels_per_line, - dst_ptr + 8, dst_pixels_per_line, 8, - vp8_vp7_bilinear_filters_mmx[xoffset], vp8_vp7_bilinear_filters_mmx[yoffset], - &xsum1, &xxsum1 - ); - - xsum0 += xsum1; - xxsum0 += xxsum1; - - vp8_filter_block2d_bil_var_mmx( - src_ptr + f2soffset, src_pixels_per_line, - dst_ptr + f2doffset, dst_pixels_per_line, 8, - vp8_vp7_bilinear_filters_mmx[xoffset], vp8_vp7_bilinear_filters_mmx[yoffset], - &xsum1, &xxsum1 - ); - - xsum0 += xsum1; - xxsum0 += xxsum1; - - vp8_filter_block2d_bil_var_mmx( - src_ptr + f2soffset + 8, src_pixels_per_line, - dst_ptr + f2doffset + 8, dst_pixels_per_line, 8, - vp8_vp7_bilinear_filters_mmx[xoffset], vp8_vp7_bilinear_filters_mmx[yoffset], - &xsum1, &xxsum1 - ); - - xsum0 += xsum1; - xxsum0 += xxsum1; - *sse = xxsum0; - return (xxsum0 - ((xsum0 * xsum0) >> 8)); -} - - -unsigned int vp8_i_sub_pixel_variance8x16_mmx -( - const unsigned char *src_ptr, - int src_pixels_per_line, - int xoffset, - int yoffset, - const unsigned char *dst_ptr, - int dst_pixels_per_line, - unsigned int *sse -) -{ - int xsum0, xsum1; - unsigned int xxsum0, xxsum1; - int f2soffset = (src_pixels_per_line >> 1); - int f2doffset = (dst_pixels_per_line >> 1); - - - vp8_filter_block2d_bil_var_mmx( - src_ptr, src_pixels_per_line, - dst_ptr, dst_pixels_per_line, 8, - vp8_vp7_bilinear_filters_mmx[xoffset], vp8_vp7_bilinear_filters_mmx[yoffset], - &xsum0, &xxsum0 - ); - - - vp8_filter_block2d_bil_var_mmx( - src_ptr + f2soffset, src_pixels_per_line, - dst_ptr + f2doffset, dst_pixels_per_line, 8, - vp8_vp7_bilinear_filters_mmx[xoffset], vp8_vp7_bilinear_filters_mmx[yoffset], - &xsum1, &xxsum1 - ); - - xsum0 += xsum1; - xxsum0 += xxsum1; - *sse = xxsum0; - return (xxsum0 - ((xsum0 * xsum0) >> 7)); -} - unsigned int vp8_variance_halfpixvar16x16_h_mmx( const unsigned char *src_ptr, diff --git a/vp8/encoder/x86/variance_sse2.c b/vp8/encoder/x86/variance_sse2.c index 4612a6711..0edda3062 100644 --- a/vp8/encoder/x86/variance_sse2.c +++ b/vp8/encoder/x86/variance_sse2.c @@ -504,81 +504,6 @@ unsigned int vp8_sub_pixel_variance8x16_wmt return (xxsum - ((xsum * xsum) >> 7)); } -unsigned int vp8_i_variance16x16_wmt( - const unsigned char *src_ptr, - int source_stride, - const unsigned char *ref_ptr, - int recon_stride, - unsigned int *sse) -{ - unsigned int sse0, sse1, sse2, sse3, var; - int sum0, sum1, sum2, sum3, avg; - - - vp8_get8x8var_sse2(src_ptr, source_stride, ref_ptr, recon_stride, &sse0, &sum0) ; - vp8_get8x8var_sse2(src_ptr + 8, source_stride, ref_ptr + 8, recon_stride, &sse1, &sum1); - vp8_get8x8var_sse2(src_ptr + (source_stride >> 1), source_stride, ref_ptr + (recon_stride >> 1), recon_stride, &sse2, &sum2) ; - vp8_get8x8var_sse2(src_ptr + (source_stride >> 1) + 8, source_stride, ref_ptr + (recon_stride >> 1) + 8, recon_stride, &sse3, &sum3); - - var = sse0 + sse1 + sse2 + sse3; - avg = sum0 + sum1 + sum2 + sum3; - - *sse = var; - return (var - ((avg * avg) >> 8)); - -} - -unsigned int vp8_i_variance8x16_wmt( - const unsigned char *src_ptr, - int source_stride, - const unsigned char *ref_ptr, - int recon_stride, - unsigned int *sse) -{ - unsigned int sse0, sse1, var; - int sum0, sum1, avg; - vp8_get8x8var_sse2(src_ptr, source_stride, ref_ptr, recon_stride, &sse0, &sum0) ; - vp8_get8x8var_sse2(src_ptr + (source_stride >> 1), source_stride, ref_ptr + (recon_stride >> 1), recon_stride, &sse1, &sum1) ; - - var = sse0 + sse1; - avg = sum0 + sum1; - - *sse = var; - return (var - ((avg * avg) >> 7)); - -} - - -unsigned int vp8_i_sub_pixel_variance16x16_wmt -( - const unsigned char *src_ptr, - int src_pixels_per_line, - int xoffset, - int yoffset, - const unsigned char *dst_ptr, - int dst_pixels_per_line, - unsigned int *sse -) -{ - return vp8_sub_pixel_variance16x16_wmt(src_ptr, (src_pixels_per_line >> 1), xoffset, yoffset, dst_ptr, (dst_pixels_per_line >> 1), sse); -} - - -unsigned int vp8_i_sub_pixel_variance8x16_wmt -( - const unsigned char *src_ptr, - int src_pixels_per_line, - int xoffset, - int yoffset, - const unsigned char *dst_ptr, - int dst_pixels_per_line, - unsigned int *sse -) -{ - - return vp8_sub_pixel_variance8x16_wmt(src_ptr, (src_pixels_per_line >> 1), xoffset, yoffset, dst_ptr, (dst_pixels_per_line >> 1), sse); -} - unsigned int vp8_variance_halfpixvar16x16_h_wmt( const unsigned char *src_ptr, -- cgit v1.2.3 From 7b8e7f0f3ae13ebf29200324b0c4d7fe64780a58 Mon Sep 17 00:00:00 2001 From: Yunqing Wang Date: Wed, 9 Mar 2011 11:16:30 -0500 Subject: Add vp8_sub_pixel_variance16x8_ssse3 function Added SSSE3 function Change-Id: I8c304c92458618d93fda3a2f62bd09ccb63e75ad --- vp8/encoder/x86/variance_ssse3.c | 53 ++++++++++++++++++++++++++++++++-- vp8/encoder/x86/variance_x86.h | 4 +++ vp8/encoder/x86/x86_csystemdependent.c | 1 + 3 files changed, 56 insertions(+), 2 deletions(-) (limited to 'vp8/encoder/x86') diff --git a/vp8/encoder/x86/variance_ssse3.c b/vp8/encoder/x86/variance_ssse3.c index d50ae3ade..eb5d486bf 100644 --- a/vp8/encoder/x86/variance_ssse3.c +++ b/vp8/encoder/x86/variance_ssse3.c @@ -76,8 +76,8 @@ unsigned int vp8_sub_pixel_variance16x16_ssse3 unsigned int *sse ) { - int xsum0, xsum1; - unsigned int xxsum0, xxsum1; + int xsum0; + unsigned int xxsum0; // note we could avoid these if statements if the calling function // just called the appropriate functions inside. @@ -114,3 +114,52 @@ unsigned int vp8_sub_pixel_variance16x16_ssse3 *sse = xxsum0; return (xxsum0 - ((xsum0 * xsum0) >> 8)); } + +unsigned int vp8_sub_pixel_variance16x8_ssse3 +( + const unsigned char *src_ptr, + int src_pixels_per_line, + int xoffset, + int yoffset, + const unsigned char *dst_ptr, + int dst_pixels_per_line, + unsigned int *sse + +) +{ + int xsum0; + unsigned int xxsum0; + + if (xoffset == 4 && yoffset == 0) + { + vp8_half_horiz_variance16x_h_sse2( + src_ptr, src_pixels_per_line, + dst_ptr, dst_pixels_per_line, 8, + &xsum0, &xxsum0); + } + else if (xoffset == 0 && yoffset == 4) + { + vp8_half_vert_variance16x_h_sse2( + src_ptr, src_pixels_per_line, + dst_ptr, dst_pixels_per_line, 8, + &xsum0, &xxsum0); + } + else if (xoffset == 4 && yoffset == 4) + { + vp8_half_horiz_vert_variance16x_h_sse2( + src_ptr, src_pixels_per_line, + dst_ptr, dst_pixels_per_line, 8, + &xsum0, &xxsum0); + } + else + { + vp8_filter_block2d_bil_var_ssse3( + src_ptr, src_pixels_per_line, + dst_ptr, dst_pixels_per_line, 8, + xoffset, yoffset, + &xsum0, &xxsum0); + } + + *sse = xxsum0; + return (xxsum0 - ((xsum0 * xsum0) >> 7)); +} diff --git a/vp8/encoder/x86/variance_x86.h b/vp8/encoder/x86/variance_x86.h index 1e2fb3490..3560f7413 100644 --- a/vp8/encoder/x86/variance_x86.h +++ b/vp8/encoder/x86/variance_x86.h @@ -286,6 +286,7 @@ extern prototype_sad_multi_dif_address(vp8_sad4x4x4d_sse3); #if HAVE_SSSE3 extern prototype_sad_multi_same_address(vp8_sad16x16x3_ssse3); extern prototype_sad_multi_same_address(vp8_sad16x8x3_ssse3); +extern prototype_subpixvariance(vp8_sub_pixel_variance16x8_ssse3); extern prototype_subpixvariance(vp8_sub_pixel_variance16x16_ssse3); #if !CONFIG_RUNTIME_CPU_DETECT @@ -295,6 +296,9 @@ extern prototype_subpixvariance(vp8_sub_pixel_variance16x16_ssse3); #undef vp8_variance_sad16x8x3 #define vp8_variance_sad16x8x3 vp8_sad16x8x3_ssse3 +#undef vp8_variance_subpixvar16x8 +#define vp8_variance_subpixvar16x8 vp8_sub_pixel_variance16x8_ssse3 + #undef vp8_variance_subpixvar16x16 #define vp8_variance_subpixvar16x16 vp8_sub_pixel_variance16x16_ssse3 diff --git a/vp8/encoder/x86/x86_csystemdependent.c b/vp8/encoder/x86/x86_csystemdependent.c index c7639a7e4..3158ac12b 100644 --- a/vp8/encoder/x86/x86_csystemdependent.c +++ b/vp8/encoder/x86/x86_csystemdependent.c @@ -334,6 +334,7 @@ void vp8_arch_x86_encoder_init(VP8_COMP *cpi) cpi->rtcd.variance.sad16x16x3 = vp8_sad16x16x3_ssse3; cpi->rtcd.variance.sad16x8x3 = vp8_sad16x8x3_ssse3; + cpi->rtcd.variance.subpixvar16x8 = vp8_sub_pixel_variance16x8_ssse3; cpi->rtcd.variance.subpixvar16x16 = vp8_sub_pixel_variance16x16_ssse3; cpi->rtcd.quantize.fastquantb = vp8_fast_quantize_b_ssse3; -- cgit v1.2.3