src/blitter/32bpp_ssse3.hpp

   1 /* $Id$ */
   2
   3 /*
   4  * This file is part of OpenTTD.
   5  * OpenTTD is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, version 2.
   6  * OpenTTD is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
   7  * See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with OpenTTD. If not, see <http://www.gnu.org/licenses/>.
   8  */
   9
  10 /** @file 32bpp_ssse3.hpp SSSE3 32 bpp blitter. */
  11
  12 #ifndef BLITTER_32BPP_SSSE3_HPP
  13 #define BLITTER_32BPP_SSSE3_HPP
  14
  15 #ifdef WITH_SSE
  16
  17 #include "32bpp_sse2.hpp"
  18 #include "tmmintrin.h"
  19
  20 /* Alpha blend 2 pixels. */
  21 #undef ALPHA_BLEND_2
  22 #define ALPHA_BLEND_2(m_pack_mask) { \
  23         __m128i srcAB = _mm_unpacklo_epi8(srcABCD, _mm_setzero_si128()); /* PUNPCKLBW, expand each uint8 into uint16 */ \
  24         __m128i dstAB = _mm_unpacklo_epi8(dstABCD, _mm_setzero_si128()); \
  25         \
  26         __m128i alphaAB = _mm_cmpgt_epi16(srcAB, _mm_setzero_si128());   /* PCMPGTW, if (alpha > 0) a++; */ \
  27         alphaAB = _mm_srli_epi16(alphaAB, 15); \
  28         alphaAB = _mm_add_epi16(alphaAB, srcAB); \
  29         alphaAB = _mm_shuffle_epi8(alphaAB, a_cm);        /* PSHUFB, put alpha in front of each rgb */ \
  30         \
  31         srcAB = _mm_sub_epi16(srcAB, dstAB);              /* PSUBW,    (r - Cr) */ \
  32         srcAB = _mm_mullo_epi16(srcAB, alphaAB);          /* PMULLW, a*(r - Cr) */ \
  33         srcAB = _mm_srli_epi16(srcAB, 8);                 /* PSRLW,  a*(r - Cr)/256 */ \
  34         srcAB = _mm_add_epi16(srcAB, dstAB);              /* PADDW,  a*(r - Cr)/256 + Cr */ \
  35         srcABCD = _mm_shuffle_epi8(srcAB, m_pack_mask);   /* PSHUFB, pack 2 Colour (without saturation) */ \
  36 }
  37
  38 /* Adjust brightness of 2 pixels. */
  39 #define ADJUST_BRIGHTNESS_2(colourX2, brightnessX2) \
  40         /* The following dataflow differs from the one of AdjustBrightness() only for alpha.
  41          * In order to keep alpha in colAB, insert a 1 in a unused brightness byte (a*1->a).
  42          * OK, not a 1 but DEFAULT_BRIGHTNESS to compensate the div.
  43          */ \
  44         brightnessX2 &= 0xFF00FF00; \
  45         brightnessX2 += DEFAULT_BRIGHTNESS; \
  46         \
  47         __m128i zero = _mm_setzero_si128(); \
  48         __m128i colAB = _mm_unpacklo_epi8(colourX2, zero); \
  49         \
  50         __m128i briAB = _mm_cvtsi32_si128(brightnessX2); \
  51         briAB = _mm_shuffle_epi8(briAB, briAB_cm); /* DEFAULT_BRIGHTNESS in 0, 0x00 in 2. */ \
  52         colAB = _mm_mullo_epi16(colAB, briAB); \
  53         __m128i colAB_ob = _mm_srli_epi16(colAB, 8+7); \
  54         colAB = _mm_srli_epi16(colAB, 7); \
  55         \
  56         /* Sum overbright.
  57          * Maximum for each rgb is 508 => 9 bits. The highest bit tells if there is overbright.
  58          * -255 is changed in -256 so we just have to take the 8 lower bits into account.
  59          */ \
  60         colAB = _mm_and_si128(colAB, div_cleaner); \
  61         colAB_ob = _mm_and_si128(colAB_ob, ob_check); \
  62         colAB_ob = _mm_mullo_epi16(colAB_ob, ob_mask); \
  63         colAB_ob = _mm_and_si128(colAB_ob, colAB); \
  64         __m128i obAB = _mm_hadd_epi16(_mm_hadd_epi16(colAB_ob, zero), zero); \
  65         \
  66         obAB = _mm_srli_epi16(obAB, 1);       /* Reduce overbright strength. */ \
  67         obAB = _mm_shuffle_epi8(obAB, ob_cm); \
  68         __m128i retAB = ob_mask;              /* ob_mask is equal to white. */ \
  69         retAB = _mm_subs_epu16(retAB, colAB); /*    (255 - rgb) */ \
  70         retAB = _mm_mullo_epi16(retAB, obAB); /* ob*(255 - rgb) */ \
  71         retAB = _mm_srli_epi16(retAB, 8);     /* ob*(255 - rgb)/256 */ \
  72         retAB = _mm_add_epi16(retAB, colAB);  /* ob*(255 - rgb)/256 + rgb */ \
  73         \
  74         colourX2 = _mm_packus_epi16(retAB, retAB);
  75
  76 /** The SSSE3 32 bpp blitter (without palette animation). */
  77 class Blitter_32bppSSSE3 : public Blitter_32bppSSE2 {
  78 public:
  79         /* virtual */ void Draw(Blitter::BlitterParams *bp, BlitterMode mode, ZoomLevel zoom);
  80         template <BlitterMode mode, ReadMode read_mode, BlockType bt_last>
  81         void Draw(const Blitter::BlitterParams *bp, ZoomLevel zoom);
  82         /* virtual */ const char *GetName() { return "32bpp-ssse3"; }
  83 };
  84
  85 /** Factory for the SSSE3 32 bpp blitter (without palette animation). */
  86 class FBlitter_32bppSSSE3: public BlitterFactory {
  87 public:
  88         FBlitter_32bppSSSE3() : BlitterFactory("32bpp-ssse3", "32bpp SSSE3 Blitter (no palette animation)", HasCPUIDFlag(1, 2, 9)) {}
  89         /* virtual */ Blitter *CreateInstance() { return new Blitter_32bppSSSE3(); }
  90 };
  91
  92 #endif /* WITH_SSE */
  93 #endif /* BLITTER_32BPP_SSSE3_HPP */