📄 pixel-sse2.asm
字号:
SBUTTERFLY dqa, wd, %3, %4, %2 SBUTTERFLY dqa, wd, %5, %6, %4 SBUTTERFLY dqa, wd, %7, %8, %6 SBUTTERFLY dqa, dq, %1, %3, %8 SBUTTERFLY dqa, dq, %9, %2, %3 SBUTTERFLY dqa, dq, %5, %7, %2 SBUTTERFLY dqa, dq, %4, %6, %7 SBUTTERFLY dqa, qdq, %1, %5, %6 SBUTTERFLY dqa, qdq, %9, %4, %5 SBUTTERFLY dqa, qdq, %8, %2, %4 SBUTTERFLY dqa, qdq, %3, %7, %2%endmacro%macro LOAD_DIFF_8P 4 ; MMP, MMT, [pix1], [pix2] movq %1, %3 movq %2, %4 punpcklbw %1, %2 punpcklbw %2, %2 psubw %1, %2%endmacro%macro LOAD_DIFF_4x8P 6 ; 4x dest, 2x temp LOAD_DIFF_8P %1, %5, [parm1q], [parm3q] LOAD_DIFF_8P %2, %6, [parm1q+parm2q], [parm3q+parm4q] LOAD_DIFF_8P %3, %5, [parm1q+2*parm2q], [parm3q+2*parm4q] LOAD_DIFF_8P %4, %6, [parm1q+r10], [parm3q+r11]%endmacro%macro SUM1x8_SSE2 3 ; 01 junk sum pxor %2, %2 psubw %2, %1 pmaxsw %1, %2 paddusw %3, %1%endmacro%macro SUM4x4_SSE2 4 ; 02 13 junk sum pxor %3, %3 psubw %3, %1 pmaxsw %1, %3 pxor %3, %3 psubw %3, %2 pmaxsw %2, %3 paddusw %4, %1 paddusw %4, %2%endmacro%macro SUM8x4_SSE2 7 ; a02 a13 junk1 b02 b13 junk2 (1=4 2=5 3=6) sum pxor %3, %3 pxor %6, %6 psubw %3, %1 psubw %6, %4 pmaxsw %1, %3 pmaxsw %4, %6 pxor %3, %3 pxor %6, %6 psubw %3, %2 psubw %6, %5 pmaxsw %2, %3 pmaxsw %5, %6 paddusw %1, %2 paddusw %4, %5 paddusw %7, %1 paddusw %7, %4%endmacro%macro SUM8x4_SSSE3 7 ; a02 a13 . b02 b13 . sum pabsw %1, %1 pabsw %2, %2 pabsw %4, %4 pabsw %5, %5 paddusw %1, %2 paddusw %4, %5 paddusw %7, %1 paddusw %7, %4%endmacro%macro SATD_TWO_SSE2 0 LOAD_DIFF_4x8P xmm0, xmm1, xmm2, xmm3, xmm4, xmm5 lea parm1q, [parm1q+4*parm2q] lea parm3q, [parm3q+4*parm4q] HADAMARD1x4 xmm0, xmm1, xmm2, xmm3 TRANSPOSE2x4x4W xmm0, xmm1, xmm2, xmm3, xmm4 HADAMARD1x4 xmm0, xmm1, xmm2, xmm3 SUM8x4 xmm0, xmm1, xmm4, xmm2, xmm3, xmm5, xmm6%endmacro%macro SATD_START 0 pxor xmm6, xmm6 lea r10, [3*parm2q] lea r11, [3*parm4q]%endmacro%macro SATD_END 0 psrlw xmm6, 1 HADDW xmm6, xmm7 movd eax, xmm6 ret%endmacro%macro SATDS 1;-----------------------------------------------------------------------------; int x264_pixel_satd_16x16_sse2 (uint8_t *, int, uint8_t *, int );-----------------------------------------------------------------------------cglobal x264_pixel_satd_16x16_%1 SATD_START mov r8, rdi mov r9, rdx SATD_TWO_SSE2 SATD_TWO_SSE2 SATD_TWO_SSE2 SATD_TWO_SSE2 lea rdi, [r8+8] lea rdx, [r9+8] SATD_TWO_SSE2 SATD_TWO_SSE2 SATD_TWO_SSE2 SATD_TWO_SSE2 SATD_END;-----------------------------------------------------------------------------; int x264_pixel_satd_8x16_sse2 (uint8_t *, int, uint8_t *, int );-----------------------------------------------------------------------------cglobal x264_pixel_satd_8x16_%1 SATD_START SATD_TWO_SSE2 SATD_TWO_SSE2 SATD_TWO_SSE2 SATD_TWO_SSE2 SATD_END;-----------------------------------------------------------------------------; int x264_pixel_satd_16x8_sse2 (uint8_t *, int, uint8_t *, int );-----------------------------------------------------------------------------cglobal x264_pixel_satd_16x8_%1 SATD_START mov r8, rdi mov r9, rdx SATD_TWO_SSE2 SATD_TWO_SSE2 lea rdi, [r8+8] lea rdx, [r9+8] SATD_TWO_SSE2 SATD_TWO_SSE2 SATD_END;-----------------------------------------------------------------------------; int x264_pixel_satd_8x8_sse2 (uint8_t *, int, uint8_t *, int );-----------------------------------------------------------------------------cglobal x264_pixel_satd_8x8_%1 SATD_START SATD_TWO_SSE2 SATD_TWO_SSE2 SATD_END;-----------------------------------------------------------------------------; int x264_pixel_satd_8x4_sse2 (uint8_t *, int, uint8_t *, int );-----------------------------------------------------------------------------cglobal x264_pixel_satd_8x4_%1 SATD_START SATD_TWO_SSE2 SATD_END;-----------------------------------------------------------------------------; int x264_pixel_sa8d_8x8_sse2( uint8_t *, int, uint8_t *, int );-----------------------------------------------------------------------------cglobal x264_pixel_sa8d_8x8_%1 lea r10, [3*parm2q] lea r11, [3*parm4q] LOAD_DIFF_4x8P xmm0, xmm1, xmm2, xmm3, xmm8, xmm8 lea parm1q, [parm1q+4*parm2q] lea parm3q, [parm3q+4*parm4q] LOAD_DIFF_4x8P xmm4, xmm5, xmm6, xmm7, xmm8, xmm8 HADAMARD1x8 xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7 TRANSPOSE8x8 xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7, xmm8 HADAMARD1x8 xmm0, xmm5, xmm7, xmm3, xmm8, xmm4, xmm2, xmm1 pxor xmm10, xmm10 SUM8x4 xmm0, xmm1, xmm6, xmm2, xmm3, xmm9, xmm10 SUM8x4 xmm4, xmm5, xmm6, xmm7, xmm8, xmm9, xmm10 psrlw xmm10, 1 HADDW xmm10, xmm0 movd eax, xmm10 add r8d, eax ; preserve rounding for 16x16 add eax, 1 shr eax, 1 ret;-----------------------------------------------------------------------------; int x264_pixel_sa8d_16x16_sse2( uint8_t *, int, uint8_t *, int );-----------------------------------------------------------------------------;; violates calling conventioncglobal x264_pixel_sa8d_16x16_%1 xor r8d, r8d call x264_pixel_sa8d_8x8_%1 ; pix[0] lea parm1q, [parm1q+4*parm2q] lea parm3q, [parm3q+4*parm4q] call x264_pixel_sa8d_8x8_%1 ; pix[8*stride] lea r10, [3*parm2q-2] lea r11, [3*parm4q-2] shl r10, 2 shl r11, 2 sub parm1q, r10 sub parm3q, r11 call x264_pixel_sa8d_8x8_%1 ; pix[8] lea parm1q, [parm1q+4*parm2q] lea parm3q, [parm3q+4*parm4q] call x264_pixel_sa8d_8x8_%1 ; pix[8*stride+8] mov eax, r8d add eax, 1 shr eax, 1 ret%endmacro ; SATDS%define SUM8x4 SUM8x4_SSE2SATDS sse2%ifdef HAVE_SSE3%define SUM8x4 SUM8x4_SSSE3SATDS ssse3%endif;-----------------------------------------------------------------------------; void x264_intra_sa8d_x3_8x8_core_sse2( uint8_t *fenc, int16_t edges[2][8], int *res );-----------------------------------------------------------------------------cglobal x264_intra_sa8d_x3_8x8_core_sse2 ; 8x8 hadamard pxor xmm4, xmm4 movq xmm0, [parm1q+0*FENC_STRIDE] movq xmm7, [parm1q+1*FENC_STRIDE] movq xmm6, [parm1q+2*FENC_STRIDE] movq xmm3, [parm1q+3*FENC_STRIDE] movq xmm5, [parm1q+4*FENC_STRIDE] movq xmm1, [parm1q+5*FENC_STRIDE] movq xmm8, [parm1q+6*FENC_STRIDE] movq xmm2, [parm1q+7*FENC_STRIDE] punpcklbw xmm0, xmm4 punpcklbw xmm7, xmm4 punpcklbw xmm6, xmm4 punpcklbw xmm3, xmm4 punpcklbw xmm5, xmm4 punpcklbw xmm1, xmm4 punpcklbw xmm8, xmm4 punpcklbw xmm2, xmm4 HADAMARD1x8 xmm0, xmm7, xmm6, xmm3, xmm5, xmm1, xmm8, xmm2 TRANSPOSE8x8 xmm0, xmm7, xmm6, xmm3, xmm5, xmm1, xmm8, xmm2, xmm4 HADAMARD1x8 xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7 ; dc movzx edi, word [parm2q+0] add di, word [parm2q+16] add edi, 8 and edi, -16 shl edi, 2 pxor xmm15, xmm15 movdqa xmm8, xmm2 movdqa xmm9, xmm3 movdqa xmm10, xmm4 movdqa xmm11, xmm5 SUM8x4_SSE2 xmm8, xmm9, xmm12, xmm10, xmm11, xmm13, xmm15 movdqa xmm8, xmm6 movdqa xmm9, xmm7 SUM4x4_SSE2 xmm8, xmm9, xmm10, xmm15 movdqa xmm8, xmm1 SUM1x8_SSE2 xmm8, xmm10, xmm15 movdqa xmm14, xmm15 ; 7x8 sum movdqa xmm8, [parm2q+0] ; left edge movd xmm9, edi psllw xmm8, 3 psubw xmm8, xmm0 psubw xmm9, xmm0 SUM1x8_SSE2 xmm8, xmm10, xmm14 SUM1x8_SSE2 xmm9, xmm11, xmm15 ; 1x8 sum punpcklwd xmm0, xmm1 punpcklwd xmm2, xmm3 punpcklwd xmm4, xmm5 punpcklwd xmm6, xmm7 punpckldq xmm0, xmm2 punpckldq xmm4, xmm6 punpcklqdq xmm0, xmm4 ; transpose movdqa xmm1, [parm2q+16] ; top edge movdqa xmm2, xmm15 psllw xmm1, 3 psrldq xmm2, 2 ; 8x7 sum psubw xmm0, xmm1 ; 8x1 sum SUM1x8_SSE2 xmm0, xmm1, xmm2 HADDW xmm14, xmm3 movd eax, xmm14 add eax, 2 shr eax, 2 mov [parm3q+4], eax ; i8x8_h sa8d HADDW xmm15, xmm4 movd eax, xmm15 add eax, 2 shr eax, 2 mov [parm3q+8], eax ; i8x8_dc sa8d HADDW xmm2, xmm5 movd eax, xmm2 add eax, 2 shr eax, 2 mov [parm3q+0], eax ; i8x8_v sa8d ret;-----------------------------------------------------------------------------; void x264_pixel_ssim_4x4x2_core_sse2( const uint8_t *pix1, int stride1,; const uint8_t *pix2, int stride2, int sums[2][4] );-----------------------------------------------------------------------------cglobal x264_pixel_ssim_4x4x2_core_sse2 pxor xmm0, xmm0 pxor xmm1, xmm1 pxor xmm2, xmm2 pxor xmm3, xmm3 pxor xmm4, xmm4 movdqa xmm8, [pw_1 GLOBAL]%rep 4 movq xmm5, [parm1q] movq xmm6, [parm3q] punpcklbw xmm5, xmm0 punpcklbw xmm6, xmm0 paddw xmm1, xmm5 paddw xmm2, xmm6 movdqa xmm7, xmm5 pmaddwd xmm5, xmm5 pmaddwd xmm7, xmm6 pmaddwd xmm6, xmm6 paddd xmm3, xmm5 paddd xmm4, xmm7 paddd xmm3, xmm6 add parm1q, parm2q add parm3q, parm4q%endrep ; PHADDW xmm1, xmm2 ; PHADDD xmm3, xmm4 pshufd xmm5, xmm3, 0xB1 pmaddwd xmm1, xmm8 pmaddwd xmm2, xmm8 pshufd xmm6, xmm4, 0xB1 packssdw xmm1, xmm2 paddd xmm3, xmm5 pmaddwd xmm1, xmm8 paddd xmm4, xmm6 pshufd xmm1, xmm1, 0xD8 movdqa xmm5, xmm3 punpckldq xmm3, xmm4 punpckhdq xmm5, xmm4 movq [parm5q+ 0], xmm1 movq [parm5q+ 8], xmm3 psrldq xmm1, 8 movq [parm5q+16], xmm1 movq [parm5q+24], xmm5 ret;-----------------------------------------------------------------------------; float x264_pixel_ssim_end_sse2( int sum0[5][4], int sum1[5][4], int width );-----------------------------------------------------------------------------cglobal x264_pixel_ssim_end4_sse2 movdqa xmm0, [parm1q+ 0] movdqa xmm1, [parm1q+16] movdqa xmm2, [parm1q+32] movdqa xmm3, [parm1q+48] movdqa xmm4, [parm1q+64] paddd xmm0, [parm2q+ 0] paddd xmm1, [parm2q+16] paddd xmm2, [parm2q+32] paddd xmm3, [parm2q+48] paddd xmm4, [parm2q+64] paddd xmm0, xmm1 paddd xmm1, xmm2 paddd xmm2, xmm3 paddd xmm3, xmm4 movdqa xmm5, [ssim_c1 GLOBAL] movdqa xmm6, [ssim_c2 GLOBAL] TRANSPOSE4x4D xmm0, xmm1, xmm2, xmm3, xmm4; s1=mm0, s2=mm3, ss=mm4, s12=mm2 movdqa xmm1, xmm3 pslld xmm3, 16 pmaddwd xmm1, xmm0 ; s1*s2 por xmm0, xmm3 pmaddwd xmm0, xmm0 ; s1*s1 + s2*s2 pslld xmm1, 1 pslld xmm2, 7 pslld xmm4, 6 psubd xmm2, xmm1 ; covar*2 psubd xmm4, xmm0 ; vars paddd xmm0, xmm5 paddd xmm1, xmm5 paddd xmm2, xmm6 paddd xmm4, xmm6 cvtdq2ps xmm0, xmm0 ; (float)(s1*s1 + s2*s2 + ssim_c1) cvtdq2ps xmm1, xmm1 ; (float)(s1*s2*2 + ssim_c1) cvtdq2ps xmm2, xmm2 ; (float)(covar*2 + ssim_c2) cvtdq2ps xmm4, xmm4 ; (float)(vars + ssim_c2) mulps xmm1, xmm2 mulps xmm0, xmm4 divps xmm1, xmm0 ; ssim neg parm3q%ifdef __PIC__ lea rax, [mask_ff + 16 GLOBAL] movdqu xmm3, [rax + parm3q*4]%else movdqu xmm3, [mask_ff + parm3q*4 + 16]%endif pand xmm1, xmm3 movhlps xmm0, xmm1 addps xmm0, xmm1 pshuflw xmm1, xmm0, 0xE addss xmm0, xmm1 ret⌨️ 快捷键说明
复制代码
Ctrl + C
搜索代码
Ctrl + F
全屏模式
F11
切换主题
Ctrl + Shift + D
显示快捷键
?
增大字号
Ctrl + =
减小字号
Ctrl + -