| 1 | /* |
|---|
| 2 | * Copyright (C) 2007 Michael Lewis |
|---|
| 3 | * Author: Mike Lewis <mikelikespie@gmail.com> |
|---|
| 4 | * |
|---|
| 5 | * This work is provide AS IS, and has no warranty. |
|---|
| 6 | * The author is NOT responsible for anything that happens |
|---|
| 7 | * due to use of this code, either using it or running it on |
|---|
| 8 | * your system |
|---|
| 9 | */ |
|---|
| 10 | #ifndef _HELPERS_H_INCLUDED |
|---|
| 11 | #define _HELPERS_H_INCLUDED |
|---|
| 12 | #include "fakesse.h" |
|---|
| 13 | |
|---|
| 14 | static inline __m128i shift_pack_left127( __m128i A ) { |
|---|
| 15 | return _mm_srli_si128( _mm_srli_epi32( A, 7 ), 15 ); |
|---|
| 16 | } |
|---|
| 17 | |
|---|
| 18 | static inline __m128i shift_pack_left( __m128i A ) { |
|---|
| 19 | const __m128i endmask = _mm_set_epi8( 0xff,0x00,0x00,0x00, |
|---|
| 20 | 0x00,0x00,0x00,0x00, |
|---|
| 21 | 0x00,0x00,0x00,0x00, |
|---|
| 22 | 0x00,0x00,0x00,0x00); |
|---|
| 23 | |
|---|
| 24 | __m128i temp1 = _mm_and_si128( endmask, _mm_srli_epi32( _mm_slli_si128( A, 15 ), 1 ) ); |
|---|
| 25 | __m128i temp2 = _mm_srli_si128( A, 1 ); |
|---|
| 26 | return _mm_or_si128( temp1, temp2 ); |
|---|
| 27 | } |
|---|
| 28 | |
|---|
| 29 | static inline __m128i shift_pack_right127( __m128i A ) { |
|---|
| 30 | return _mm_slli_si128( _mm_slli_epi32( A, 7 ), 15 ); |
|---|
| 31 | } |
|---|
| 32 | |
|---|
| 33 | static inline __m128i shift_pack_right( __m128i A ) { |
|---|
| 34 | const __m128i startmask = _mm_set_epi8( 0x00,0x00,0x00,0x00, |
|---|
| 35 | 0x00,0x00,0x00,0x00, |
|---|
| 36 | 0x00,0x00,0x00,0x00, |
|---|
| 37 | 0x00,0x00,0x00,0xFF); |
|---|
| 38 | |
|---|
| 39 | __m128i temp1 = _mm_and_si128( startmask, _mm_slli_epi32( _mm_srli_si128( A, 15 ), 1 ) ); |
|---|
| 40 | __m128i temp2 = _mm_slli_si128( A, 1 ); |
|---|
| 41 | return _mm_or_si128( temp1, temp2 ); |
|---|
| 42 | } |
|---|
| 43 | |
|---|
| 44 | |
|---|
| 45 | static inline __m128i unpack( __m128i A, int row ) { |
|---|
| 46 | const __m128i onemask = _mm_set1_epi8( 0x01 ); |
|---|
| 47 | __m128i shift = _mm_srli_epi32( A, row ); |
|---|
| 48 | return _mm_and_si128( shift, onemask ); |
|---|
| 49 | } |
|---|
| 50 | |
|---|
| 51 | // MAKE SURE A IS IN THE FORM (0xFF) not (0x01 |
|---|
| 52 | // ( a | mask ) | ( ~mask & b ) |
|---|
| 53 | static inline __m128i pack( __m128i A, int row ) { |
|---|
| 54 | const __m128i onemask = _mm_set1_epi8( 0x01 ); |
|---|
| 55 | A = _mm_and_si128( onemask, A ); |
|---|
| 56 | return _mm_slli_epi32(A, row); |
|---|
| 57 | } |
|---|
| 58 | |
|---|
| 59 | //The get and set ops are scalar and slow. |
|---|
| 60 | __m128i set_bit( __m128i A, int col ); |
|---|
| 61 | char get_bit( __m128i A, int col ); |
|---|
| 62 | void print_packed_row( __m128i A ); |
|---|
| 63 | |
|---|
| 64 | |
|---|
| 65 | void print_life_block( life_block *block ); |
|---|
| 66 | |
|---|
| 67 | void set_bit_in_block( life_block *block, int row, int column ); |
|---|
| 68 | |
|---|
| 69 | #endif //_HELPERS_H_INCLUDED |
|---|