summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
-rw-r--r--projects/openttd_vs100.vcxproj2
-rw-r--r--projects/openttd_vs100.vcxproj.filters6
-rw-r--r--projects/openttd_vs80.vcproj8
-rw-r--r--projects/openttd_vs90.vcproj8
-rw-r--r--source.list2
-rw-r--r--src/blitter/32bpp_ssse3.cpp262
-rw-r--r--src/blitter/32bpp_ssse3.hpp94
7 files changed, 382 insertions, 0 deletions
diff --git a/projects/openttd_vs100.vcxproj b/projects/openttd_vs100.vcxproj
index 4c46cdda7..95b3d0c90 100644
--- a/projects/openttd_vs100.vcxproj
+++ b/projects/openttd_vs100.vcxproj
@@ -1123,6 +1123,8 @@
<ClInclude Include="..\src\blitter\32bpp_simple.hpp" />
<ClCompile Include="..\src\blitter\32bpp_sse2.cpp" />
<ClInclude Include="..\src\blitter\32bpp_sse2.hpp" />
+ <ClCompile Include="..\src\blitter\32bpp_ssse3.cpp" />
+ <ClInclude Include="..\src\blitter\32bpp_ssse3.hpp" />
<ClCompile Include="..\src\blitter\8bpp_base.cpp" />
<ClInclude Include="..\src\blitter\8bpp_base.hpp" />
<ClCompile Include="..\src\blitter\8bpp_optimized.cpp" />
diff --git a/projects/openttd_vs100.vcxproj.filters b/projects/openttd_vs100.vcxproj.filters
index 04ad15dcb..326ba3da4 100644
--- a/projects/openttd_vs100.vcxproj.filters
+++ b/projects/openttd_vs100.vcxproj.filters
@@ -2598,6 +2598,12 @@
<ClInclude Include="..\src\blitter\32bpp_sse2.hpp">
<Filter>Blitters</Filter>
</ClInclude>
+ <ClCompile Include="..\src\blitter\32bpp_ssse3.cpp">
+ <Filter>Blitters</Filter>
+ </ClCompile>
+ <ClInclude Include="..\src\blitter\32bpp_ssse3.hpp">
+ <Filter>Blitters</Filter>
+ </ClInclude>
<ClCompile Include="..\src\blitter\8bpp_base.cpp">
<Filter>Blitters</Filter>
</ClCompile>
diff --git a/projects/openttd_vs80.vcproj b/projects/openttd_vs80.vcproj
index ad3f6b8ca..fea983c83 100644
--- a/projects/openttd_vs80.vcproj
+++ b/projects/openttd_vs80.vcproj
@@ -3835,6 +3835,14 @@
>
</File>
<File
+ RelativePath=".\..\src\blitter\32bpp_ssse3.cpp"
+ >
+ </File>
+ <File
+ RelativePath=".\..\src\blitter\32bpp_ssse3.hpp"
+ >
+ </File>
+ <File
RelativePath=".\..\src\blitter\8bpp_base.cpp"
>
</File>
diff --git a/projects/openttd_vs90.vcproj b/projects/openttd_vs90.vcproj
index b6d47b154..1333456d5 100644
--- a/projects/openttd_vs90.vcproj
+++ b/projects/openttd_vs90.vcproj
@@ -3832,6 +3832,14 @@
>
</File>
<File
+ RelativePath=".\..\src\blitter\32bpp_ssse3.cpp"
+ >
+ </File>
+ <File
+ RelativePath=".\..\src\blitter\32bpp_ssse3.hpp"
+ >
+ </File>
+ <File
RelativePath=".\..\src\blitter\8bpp_base.cpp"
>
</File>
diff --git a/source.list b/source.list
index 6d81d5a74..7cfb4f62d 100644
--- a/source.list
+++ b/source.list
@@ -910,6 +910,8 @@ blitter/32bpp_simple.hpp
#if WITH_SSE
blitter/32bpp_sse2.cpp
blitter/32bpp_sse2.hpp
+blitter/32bpp_ssse3.cpp
+blitter/32bpp_ssse3.hpp
#end
blitter/8bpp_base.cpp
blitter/8bpp_base.hpp
diff --git a/src/blitter/32bpp_ssse3.cpp b/src/blitter/32bpp_ssse3.cpp
new file mode 100644
index 000000000..eac4e8c11
--- /dev/null
+++ b/src/blitter/32bpp_ssse3.cpp
@@ -0,0 +1,262 @@
+/* $Id$ */
+
+/*
+ * This file is part of OpenTTD.
+ * 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.
+ * 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.
+ * 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/>.
+ */
+
+/** @file 32bpp_ssse3.cpp Implementation of the SSSE3 32 bpp blitter. */
+
+#ifdef WITH_SSE
+
+#include "../stdafx.h"
+#include "../zoom_func.h"
+#include "../settings_type.h"
+#include "32bpp_ssse3.hpp"
+
+/** Instantiation of the SSSE3 32bpp blitter factory. */
+static FBlitter_32bppSSSE3 iFBlitter_32bppSSSE3;
+
+#if defined(__GNUC__)
+ #pragma GCC diagnostic push
+ #pragma GCC diagnostic ignored "-Wunused-variable"
+#endif
+/**
+ * Draws a sprite to a (screen) buffer. It is templated to allow faster operation.
+ *
+ * @tparam mode blitter mode
+ * @param bp further blitting parameters
+ * @param zoom zoom level at which we are drawing
+ */
+template <BlitterMode mode, Blitter_32bppSSE2::ReadMode read_mode, Blitter_32bppSSE2::BlockType bt_last>
+inline void Blitter_32bppSSSE3::Draw(const Blitter::BlitterParams *bp, ZoomLevel zoom)
+{
+ const byte * const remap = bp->remap;
+ Colour *dst_line = (Colour *) bp->dst + bp->top * bp->pitch + bp->left;
+ int effective_width = bp->width;
+
+ /* Find where to start reading in the source sprite */
+ const SpriteData * const sd = (const SpriteData *) bp->sprite;
+ const SpriteInfo * const si = &sd->infos[zoom];
+ const MapValue *src_mv_line = (const MapValue *) &sd->data[si->mv_offset] + bp->skip_top * si->sprite_width;
+ const Colour *src_rgba_line = (const Colour *) ((const byte *) &sd->data[si->sprite_offset] + bp->skip_top * si->sprite_line_size);
+
+ if (read_mode != RM_WITH_MARGIN) {
+ src_rgba_line += bp->skip_left;
+ src_mv_line += bp->skip_left;
+ }
+
+ /* Load these variables into register before loop. */
+ const __m128i a_cm = ALPHA_CONTROL_MASK;
+ const __m128i pack_hi_cm = PACK_HIGH_CONTROL_MASK;
+ const __m128i briAB_cm = BRIGHTNESS_LOW_CONTROL_MASK;
+ const __m128i div_cleaner = BRIGHTNESS_DIV_CLEANER;
+ const __m128i ob_check = OVERBRIGHT_PRESENCE_MASK;
+ const __m128i ob_mask = OVERBRIGHT_VALUE_MASK;
+ const __m128i ob_cm = OVERBRIGHT_CONTROL_MASK;
+
+ for (int y = bp->height; y != 0; y--) {
+ Colour *dst = dst_line;
+ const Colour *src = src_rgba_line + META_LENGTH;
+ const MapValue *src_mv = src_mv_line;
+
+ switch (mode) {
+ default: {
+ switch (read_mode) {
+ case RM_WITH_MARGIN: {
+ src += src_rgba_line[0].data;
+ dst += src_rgba_line[0].data;
+ const int width_diff = si->sprite_width - bp->width;
+ effective_width = bp->width - (int) src_rgba_line[0].data;
+ const int delta_diff = (int) src_rgba_line[1].data - width_diff;
+ const int new_width = effective_width - (delta_diff & ~1);
+ effective_width = delta_diff > 0 ? new_width : effective_width;
+ if (effective_width <= 0) break;
+ /* FALLTHROUGH */
+ }
+
+ case RM_WITH_SKIP: {
+ __m128i srcABCD = _mm_loadu_si128((const __m128i*) src);
+ __m128i dstABCD = _mm_loadu_si128((__m128i*) dst);
+ for (uint x = (uint) effective_width / 2; x > 0; x--) {
+ ALPHA_BLEND_2(pack_hi_cm);
+ /* With high repack, srcABCD have its 2 blended pixels like: [S0 S1 S2 S3] -> [-- -- BS0 BS1]
+ * dstABCD shuffled: [D0 D1 D2 D3] -> [D2 D3 D0 D0]
+ * PALIGNR takes what's in (): [-- -- (BS0 BS1] [D2 D3) D0 D0]
+ */
+ dstABCD = _mm_shuffle_epi32(dstABCD, 0x0E);
+ srcABCD = _mm_alignr_epi8(dstABCD, srcABCD, 8);
+ Colour *old_dst = dst;
+ src += 2;
+ dst += 2;
+ /* It is VERY important to read next data before it gets invalidated in cpu cache. */
+ dstABCD = _mm_loadu_si128((__m128i*) dst);
+ _mm_storeu_si128((__m128i *) old_dst, srcABCD);
+ srcABCD = _mm_loadu_si128((const __m128i*) src);
+ }
+ if (bt_last == BT_ODD) {
+ ALPHA_BLEND_2(pack_hi_cm);
+ (*dst).data = EXTR32(srcABCD, 2);
+ }
+ break;
+ }
+
+ default: NOT_REACHED();
+ }
+ break;
+ }
+ case BM_COLOUR_REMAP: {
+ switch (read_mode) {
+ case RM_WITH_MARGIN: {
+ src += src_rgba_line[0].data;
+ src_mv += src_rgba_line[0].data;
+ dst += src_rgba_line[0].data;
+ const int width_diff = si->sprite_width - bp->width;
+ effective_width = bp->width - (int) src_rgba_line[0].data;
+ const int delta_diff = (int) src_rgba_line[1].data - width_diff;
+ const int nd = effective_width - delta_diff;
+ effective_width = delta_diff > 0 ? nd : effective_width;
+ if (effective_width <= 0) break;
+ /* FALLTHROUGH */
+ }
+
+ case RM_WITH_SKIP: {
+ __m128i srcABCD = _mm_loadu_si128((const __m128i*) src);
+ __m128i dstABCD = _mm_loadu_si128((__m128i*) dst);
+ uint32 mvX2 = *((uint32 *) const_cast<MapValue *>(src_mv));
+
+ for (uint x = (uint) effective_width / 2; x > 0; x--) {
+ /* Remap colours. */
+ if (mvX2 & 0x00FF00FF) {
+ /* Written so the compiler uses CMOV. */
+ const Colour src0 = src[0];
+ const uint m0 = (byte) mvX2;
+ const uint r0 = remap[m0];
+ const Colour c0map = (this->LookupColourInPalette(r0).data & 0x00FFFFFF) | (src0.data & 0xFF000000);
+ Colour c0 = 0; // Use alpha of 0 to keep dst as is.
+ c0 = r0 == 0 ? c0 : c0map;
+ c0 = m0 != 0 ? c0 : src0;
+ INSR32(c0.data, srcABCD, 0);
+
+ const Colour src1 = src[1];
+ const uint m1 = (byte) (mvX2 >> 16);
+ const uint r1 = remap[m1];
+ const Colour c1map = (this->LookupColourInPalette(r1).data & 0x00FFFFFF) | (src1.data & 0xFF000000);
+ Colour c1 = 0;
+ c1 = r1 == 0 ? c1 : c1map;
+ c1 = m1 != 0 ? c1 : src1;
+ INSR32(c1.data, srcABCD, 1);
+
+ if ((mvX2 & 0xFF00FF00) != 0x80008000) {
+ ADJUST_BRIGHTNESS_2(srcABCD, mvX2);
+ }
+ }
+
+ /* Blend colours. */
+ ALPHA_BLEND_2(pack_hi_cm);
+ dstABCD = _mm_shuffle_epi32(dstABCD, 0x0E);
+ srcABCD = _mm_alignr_epi8(dstABCD, srcABCD, 8);
+ Colour *old_dst = dst;
+ dst += 2;
+ src += 2;
+ src_mv += 2;
+ dstABCD = _mm_loadu_si128((__m128i*) dst);
+ _mm_storeu_si128((__m128i *) old_dst, srcABCD);
+ mvX2 = *((uint32 *) const_cast<MapValue *>(src_mv));
+ srcABCD = _mm_loadu_si128((const __m128i*) src);
+ }
+
+ if (effective_width & 1) {
+ /* In case the m-channel is zero, do not remap this pixel in any way */
+ if (src_mv->m == 0) {
+ if (src->a < 255) {
+ ALPHA_BLEND_2(pack_hi_cm);
+ (*dst).data = EXTR32(srcABCD, 2);
+ } else {
+ *dst = src->data;
+ }
+ } else {
+ const uint r = remap[src_mv->m];
+ if (r != 0) {
+ Colour remapped_colour = AdjustBrightness(this->LookupColourInPalette(r), src_mv->v);
+ if (src->a < 255) {
+ remapped_colour.a = src->a;
+ INSR32(remapped_colour.data, srcABCD, 0);
+ ALPHA_BLEND_2(pack_hi_cm);
+ (*dst).data = EXTR32(srcABCD, 2);
+ } else
+ *dst = remapped_colour;
+ }
+ }
+ }
+ break;
+ }
+
+ default: NOT_REACHED();
+ }
+ src_mv_line += si->sprite_width;
+ break;
+ }
+ case BM_TRANSPARENT:
+ /* Make the current colour a bit more black, so it looks like this image is transparent */
+ for (int x = bp->width; x != 0; x--) {
+ if (src->a == 255) {
+ *dst = MakeTransparent(*dst, 3, 4);
+ } else {
+ *dst = MakeTransparent(*dst, (256 * 4 - src->a), 256 * 4);
+ }
+ dst++;
+ src++;
+ }
+ break;
+ }
+
+ src_rgba_line = (const Colour*) ((const byte*) src_rgba_line + si->sprite_line_size);
+ dst_line += bp->pitch;
+ }
+}
+#if defined(__GNUC__)
+ #pragma GCC diagnostic pop
+#endif
+
+/**
+ * Draws a sprite to a (screen) buffer. Calls adequate templated function.
+ *
+ * @param bp further blitting parameters
+ * @param mode blitter mode
+ * @param zoom zoom level at which we are drawing
+ */
+void Blitter_32bppSSSE3::Draw(Blitter::BlitterParams *bp, BlitterMode mode, ZoomLevel zoom)
+{
+ switch (mode) {
+ case BM_NORMAL: {
+ const BlockType bt_last = (BlockType) (bp->width & 1);
+ if (bp->skip_left != 0 || bp->width <= MARGIN_NORMAL_THRESHOLD) {
+ switch (bt_last) {
+ case BT_EVEN: Draw<BM_NORMAL, RM_WITH_SKIP, BT_EVEN>(bp, zoom); return;
+ case BT_ODD: Draw<BM_NORMAL, RM_WITH_SKIP, BT_ODD>(bp, zoom); return;
+ default: NOT_REACHED();
+ }
+ } else {
+ switch (bt_last) {
+ case BT_EVEN: Draw<BM_NORMAL, RM_WITH_MARGIN, BT_EVEN>(bp, zoom); return;
+ case BT_ODD: Draw<BM_NORMAL, RM_WITH_MARGIN, BT_ODD>(bp, zoom); return;
+ default: NOT_REACHED();
+ }
+ }
+ break;
+ }
+ case BM_COLOUR_REMAP:
+ if (bp->skip_left != 0 || bp->width <= MARGIN_REMAP_THRESHOLD) {
+ Draw<BM_COLOUR_REMAP, RM_WITH_SKIP, BT_NONE>(bp, zoom); return;
+ } else {
+ Draw<BM_COLOUR_REMAP, RM_WITH_MARGIN, BT_NONE>(bp, zoom); return;
+ }
+ case BM_TRANSPARENT: Draw<BM_TRANSPARENT, RM_NONE, BT_NONE>(bp, zoom); return;
+ default: NOT_REACHED();
+ }
+}
+
+#endif /* WITH_SSE */
diff --git a/src/blitter/32bpp_ssse3.hpp b/src/blitter/32bpp_ssse3.hpp
new file mode 100644
index 000000000..7ea1fbd8b
--- /dev/null
+++ b/src/blitter/32bpp_ssse3.hpp
@@ -0,0 +1,94 @@
+/* $Id$ */
+
+/*
+ * This file is part of OpenTTD.
+ * 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.
+ * 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.
+ * 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/>.
+ */
+
+/** @file 32bpp_ssse3.hpp SSSE3 32 bpp blitter. */
+
+#ifndef BLITTER_32BPP_SSSE3_HPP
+#define BLITTER_32BPP_SSSE3_HPP
+
+#ifdef WITH_SSE
+
+#include "32bpp_sse2.hpp"
+#include "tmmintrin.h"
+
+/* Alpha blend 2 pixels. */
+#undef ALPHA_BLEND_2
+#define ALPHA_BLEND_2(m_pack_mask) { \
+ __m128i srcAB = _mm_unpacklo_epi8(srcABCD, _mm_setzero_si128()); /* PUNPCKLBW, expand each uint8 into uint16 */ \
+ __m128i dstAB = _mm_unpacklo_epi8(dstABCD, _mm_setzero_si128()); \
+ \
+ __m128i alphaAB = _mm_cmpgt_epi16(srcAB, _mm_setzero_si128()); /* PCMPGTW, if (alpha > 0) a++; */ \
+ alphaAB = _mm_srli_epi16(alphaAB, 15); \
+ alphaAB = _mm_add_epi16(alphaAB, srcAB); \
+ alphaAB = _mm_shuffle_epi8(alphaAB, a_cm); /* PSHUFB, put alpha in front of each rgb */ \
+ \
+ srcAB = _mm_sub_epi16(srcAB, dstAB); /* PSUBW, (r - Cr) */ \
+ srcAB = _mm_mullo_epi16(srcAB, alphaAB); /* PMULLW, a*(r - Cr) */ \
+ srcAB = _mm_srli_epi16(srcAB, 8); /* PSRLW, a*(r - Cr)/256 */ \
+ srcAB = _mm_add_epi16(srcAB, dstAB); /* PADDW, a*(r - Cr)/256 + Cr */ \
+ srcABCD = _mm_shuffle_epi8(srcAB, m_pack_mask); /* PSHUFB, pack 2 Colour (without saturation) */ \
+}
+
+/* Adjust brightness of 2 pixels. */
+#define ADJUST_BRIGHTNESS_2(colourX2, brightnessX2) \
+ /* The following dataflow differs from the one of AdjustBrightness() only for alpha.
+ * In order to keep alpha in colAB, insert a 1 in a unused brightness byte (a*1->a).
+ * OK, not a 1 but DEFAULT_BRIGHTNESS to compensate the div.
+ */ \
+ brightnessX2 &= 0xFF00FF00; \
+ brightnessX2 += DEFAULT_BRIGHTNESS; \
+ \
+ __m128i zero = _mm_setzero_si128(); \
+ __m128i colAB = _mm_unpacklo_epi8(colourX2, zero); \
+ \
+ __m128i briAB; \
+ INSR64(brightnessX2, briAB, 0); \
+ briAB = _mm_shuffle_epi8(briAB, briAB_cm); /* DEFAULT_BRIGHTNESS in 0, 0x00 in 2. */ \
+ colAB = _mm_mullo_epi16(colAB, briAB); \
+ __m128i colAB_ob = _mm_srli_epi16(colAB, 8+7); \
+ colAB = _mm_srli_epi16(colAB, 7); \
+ \
+ /* Sum overbright.
+ * Maximum for each rgb is 508 => 9 bits. The highest bit tells if there is overbright.
+ * -255 is changed in -256 so we just have to take the 8 lower bits into account.
+ */ \
+ colAB = _mm_and_si128(colAB, div_cleaner); \
+ colAB_ob = _mm_and_si128(colAB_ob, ob_check); \
+ colAB_ob = _mm_mullo_epi16(colAB_ob, ob_mask); \
+ colAB_ob = _mm_and_si128(colAB_ob, colAB); \
+ __m128i obAB = _mm_hadd_epi16(_mm_hadd_epi16(colAB_ob, zero), zero); \
+ \
+ obAB = _mm_srli_epi16(obAB, 1); /* Reduce overbright strength. */ \
+ obAB = _mm_shuffle_epi8(obAB, ob_cm); \
+ __m128i retAB = ob_mask; /* ob_mask is equal to white. */ \
+ retAB = _mm_subs_epu16(retAB, colAB); /* (255 - rgb) */ \
+ retAB = _mm_mullo_epi16(retAB, obAB); /* ob*(255 - rgb) */ \
+ retAB = _mm_srli_epi16(retAB, 8); /* ob*(255 - rgb)/256 */ \
+ retAB = _mm_add_epi16(retAB, colAB); /* ob*(255 - rgb)/256 + rgb */ \
+ \
+ colourX2 = _mm_packus_epi16(retAB, retAB);
+
+/** The SSSE3 32 bpp blitter (without palette animation). */
+class Blitter_32bppSSSE3 : public Blitter_32bppSSE2 {
+public:
+ /* virtual */ void Draw(Blitter::BlitterParams *bp, BlitterMode mode, ZoomLevel zoom);
+ template <BlitterMode mode, ReadMode read_mode, BlockType bt_last>
+ void Draw(const Blitter::BlitterParams *bp, ZoomLevel zoom);
+ /* virtual */ const char *GetName() { return "32bpp-ssse3"; }
+};
+
+/** Factory for the SSSE3 32 bpp blitter (without palette animation). */
+class FBlitter_32bppSSSE3: public BlitterFactory {
+public:
+ FBlitter_32bppSSSE3() : BlitterFactory("32bpp-ssse3", "32bpp SSSE3 Blitter (no palette animation)", HasCPUIDFlag(1, 2, 9)) {}
+ /* virtual */ Blitter *CreateInstance() { return new Blitter_32bppSSSE3(); }
+};
+
+#endif /* WITH_SSE */
+#endif /* BLITTER_32BPP_SSSE3_HPP */