# HG changeset patch # User michaelni # Date 1032516181 0 # Node ID 894b6190873462aa8b3edf5a50f5721e6002993d # Parent 9abb13c21fbeb15f702424fbd49a823494d2f1c3 pix_sum16_mmx() diff -r 9abb13c21fbe -r 894b61908734 i386/dsputil_mmx.c --- a/i386/dsputil_mmx.c Thu Sep 19 22:14:53 2002 +0000 +++ b/i386/dsputil_mmx.c Fri Sep 20 10:03:01 2002 +0000 @@ -420,6 +420,44 @@ ); } +static int pix_sum16_mmx(UINT8 * pix, int line_size){ + const int h=16; + int sum; + int index= -line_size*h; + + __asm __volatile( + "pxor %%mm7, %%mm7 \n\t" + "pxor %%mm6, %%mm6 \n\t" + "1: \n\t" + "movq (%2, %1), %%mm0 \n\t" + "movq (%2, %1), %%mm1 \n\t" + "movq 8(%2, %1), %%mm2 \n\t" + "movq 8(%2, %1), %%mm3 \n\t" + "punpcklbw %%mm7, %%mm0 \n\t" + "punpckhbw %%mm7, %%mm1 \n\t" + "punpcklbw %%mm7, %%mm2 \n\t" + "punpckhbw %%mm7, %%mm3 \n\t" + "paddw %%mm0, %%mm1 \n\t" + "paddw %%mm2, %%mm3 \n\t" + "paddw %%mm1, %%mm3 \n\t" + "paddw %%mm3, %%mm6 \n\t" + "addl %3, %1 \n\t" + " js 1b \n\t" + "movq %%mm6, %%mm5 \n\t" + "psrlq $32, %%mm6 \n\t" + "paddw %%mm5, %%mm6 \n\t" + "movq %%mm6, %%mm5 \n\t" + "psrlq $16, %%mm6 \n\t" + "paddw %%mm5, %%mm6 \n\t" + "movd %%mm6, %0 \n\t" + "andl $0xFFFF, %0 \n\t" + : "=&r" (sum), "+r" (index) + : "r" (pix - index), "r" (line_size) + ); + + return sum; +} + #if 0 static void just_return() { return; } #endif @@ -448,6 +486,7 @@ put_pixels_clamped = put_pixels_clamped_mmx; add_pixels_clamped = add_pixels_clamped_mmx; clear_blocks= clear_blocks_mmx; + pix_sum= pix_sum16_mmx; pix_abs16x16 = pix_abs16x16_mmx; pix_abs16x16_x2 = pix_abs16x16_x2_mmx;