1 /****************************************************************************
3 ** Copyright (C) 2011 Nokia Corporation and/or its subsidiary(-ies).
4 ** All rights reserved.
5 ** Contact: Nokia Corporation (qt-info@nokia.com)
7 ** This file is part of the QtGui module of the Qt Toolkit.
9 ** $QT_BEGIN_LICENSE:LGPL$
10 ** No Commercial Usage
11 ** This file contains pre-release code and may not be distributed.
12 ** You may use this file in accordance with the terms and conditions
13 ** contained in the Technology Preview License Agreement accompanying
16 ** GNU Lesser General Public License Usage
17 ** Alternatively, this file may be used under the terms of the GNU Lesser
18 ** General Public License version 2.1 as published by the Free Software
19 ** Foundation and appearing in the file LICENSE.LGPL included in the
20 ** packaging of this file. Please review the following information to
21 ** ensure the GNU Lesser General Public License version 2.1 requirements
22 ** will be met: http://www.gnu.org/licenses/old-licenses/lgpl-2.1.html.
24 ** In addition, as a special exception, Nokia gives you certain additional
25 ** rights. These rights are described in the Nokia Qt LGPL Exception
26 ** version 1.1, included in the file LGPL_EXCEPTION.txt in this package.
28 ** If you have questions regarding the use of this file, please contact
29 ** Nokia at qt-info@nokia.com.
40 ****************************************************************************/
42 #include <private/qdrawhelper_x86_p.h>
46 #include <private/qdrawingprimitive_sse2_p.h>
47 #include <private/qpaintengine_raster_p.h>
51 void qt_blend_argb32_on_argb32_sse2(uchar *destPixels, int dbpl,
52 const uchar *srcPixels, int sbpl,
56 const quint32 *src = (const quint32 *) srcPixels;
57 quint32 *dst = (quint32 *) destPixels;
58 if (const_alpha == 256) {
59 const __m128i alphaMask = _mm_set1_epi32(0xff000000);
60 const __m128i nullVector = _mm_set1_epi32(0);
61 const __m128i half = _mm_set1_epi16(0x80);
62 const __m128i one = _mm_set1_epi16(0xff);
63 const __m128i colorMask = _mm_set1_epi32(0x00ff00ff);
64 for (int y = 0; y < h; ++y) {
65 BLEND_SOURCE_OVER_ARGB32_SSE2(dst, src, w, nullVector, half, one, colorMask, alphaMask);
66 dst = (quint32 *)(((uchar *) dst) + dbpl);
67 src = (const quint32 *)(((const uchar *) src) + sbpl);
69 } else if (const_alpha != 0) {
70 // dest = (s + d * sia) * ca + d * cia
71 // = s * ca + d * (sia * ca + cia)
72 // = s * ca + d * (1 - sa*ca)
73 const_alpha = (const_alpha * 255) >> 8;
74 const __m128i nullVector = _mm_set1_epi32(0);
75 const __m128i half = _mm_set1_epi16(0x80);
76 const __m128i one = _mm_set1_epi16(0xff);
77 const __m128i colorMask = _mm_set1_epi32(0x00ff00ff);
78 const __m128i constAlphaVector = _mm_set1_epi16(const_alpha);
79 for (int y = 0; y < h; ++y) {
80 BLEND_SOURCE_OVER_ARGB32_WITH_CONST_ALPHA_SSE2(dst, src, w, nullVector, half, one, colorMask, constAlphaVector)
81 dst = (quint32 *)(((uchar *) dst) + dbpl);
82 src = (const quint32 *)(((const uchar *) src) + sbpl);
87 // qblendfunctions.cpp
88 void qt_blend_rgb32_on_rgb32(uchar *destPixels, int dbpl,
89 const uchar *srcPixels, int sbpl,
93 void qt_blend_rgb32_on_rgb32_sse2(uchar *destPixels, int dbpl,
94 const uchar *srcPixels, int sbpl,
98 const quint32 *src = (const quint32 *) srcPixels;
99 quint32 *dst = (quint32 *) destPixels;
100 if (const_alpha != 256) {
101 if (const_alpha != 0) {
102 const __m128i nullVector = _mm_set1_epi32(0);
103 const __m128i half = _mm_set1_epi16(0x80);
104 const __m128i colorMask = _mm_set1_epi32(0x00ff00ff);
106 const_alpha = (const_alpha * 255) >> 8;
107 int one_minus_const_alpha = 255 - const_alpha;
108 const __m128i constAlphaVector = _mm_set1_epi16(const_alpha);
109 const __m128i oneMinusConstAlpha = _mm_set1_epi16(one_minus_const_alpha);
110 for (int y = 0; y < h; ++y) {
113 // First, align dest to 16 bytes:
114 ALIGNMENT_PROLOGUE_16BYTES(dst, x, w) {
116 s = BYTE_MUL(s, const_alpha);
117 dst[x] = INTERPOLATE_PIXEL_255(src[x], const_alpha, dst[x], one_minus_const_alpha);
120 for (; x < w-3; x += 4) {
121 __m128i srcVector = _mm_loadu_si128((__m128i *)&src[x]);
122 if (_mm_movemask_epi8(_mm_cmpeq_epi32(srcVector, nullVector)) != 0xffff) {
123 const __m128i dstVector = _mm_load_si128((__m128i *)&dst[x]);
125 INTERPOLATE_PIXEL_255_SSE2(result, srcVector, dstVector, constAlphaVector, oneMinusConstAlpha, colorMask, half);
126 _mm_store_si128((__m128i *)&dst[x], result);
131 s = BYTE_MUL(s, const_alpha);
132 dst[x] = INTERPOLATE_PIXEL_255(src[x], const_alpha, dst[x], one_minus_const_alpha);
134 dst = (quint32 *)(((uchar *) dst) + dbpl);
135 src = (const quint32 *)(((const uchar *) src) + sbpl);
139 qt_blend_rgb32_on_rgb32(destPixels, dbpl, srcPixels, sbpl, w, h, const_alpha);
143 void QT_FASTCALL comp_func_SourceOver_sse2(uint *destPixels, const uint *srcPixels, int length, uint const_alpha)
145 Q_ASSERT(const_alpha < 256);
147 const quint32 *src = (const quint32 *) srcPixels;
148 quint32 *dst = (quint32 *) destPixels;
150 const __m128i nullVector = _mm_set1_epi32(0);
151 const __m128i half = _mm_set1_epi16(0x80);
152 const __m128i one = _mm_set1_epi16(0xff);
153 const __m128i colorMask = _mm_set1_epi32(0x00ff00ff);
154 if (const_alpha == 255) {
155 const __m128i alphaMask = _mm_set1_epi32(0xff000000);
156 BLEND_SOURCE_OVER_ARGB32_SSE2(dst, src, length, nullVector, half, one, colorMask, alphaMask);
158 const __m128i constAlphaVector = _mm_set1_epi16(const_alpha);
159 BLEND_SOURCE_OVER_ARGB32_WITH_CONST_ALPHA_SSE2(dst, src, length, nullVector, half, one, colorMask, constAlphaVector);
163 void QT_FASTCALL comp_func_Plus_sse2(uint *dst, const uint *src, int length, uint const_alpha)
167 if (const_alpha == 255) {
168 // 1) Prologue: align destination on 16 bytes
169 ALIGNMENT_PROLOGUE_16BYTES(dst, x, length)
170 dst[x] = comp_func_Plus_one_pixel(dst[x], src[x]);
172 // 2) composition with SSE2
173 for (; x < length - 3; x += 4) {
174 const __m128i srcVector = _mm_loadu_si128((__m128i *)&src[x]);
175 const __m128i dstVector = _mm_load_si128((__m128i *)&dst[x]);
177 const __m128i result = _mm_adds_epu8(srcVector, dstVector);
178 _mm_store_si128((__m128i *)&dst[x], result);
182 for (; x < length; ++x)
183 dst[x] = comp_func_Plus_one_pixel(dst[x], src[x]);
185 const int one_minus_const_alpha = 255 - const_alpha;
186 const __m128i constAlphaVector = _mm_set1_epi16(const_alpha);
187 const __m128i oneMinusConstAlpha = _mm_set1_epi16(one_minus_const_alpha);
189 // 1) Prologue: align destination on 16 bytes
190 ALIGNMENT_PROLOGUE_16BYTES(dst, x, length)
191 dst[x] = comp_func_Plus_one_pixel_const_alpha(dst[x], src[x], const_alpha, one_minus_const_alpha);
193 const __m128i half = _mm_set1_epi16(0x80);
194 const __m128i colorMask = _mm_set1_epi32(0x00ff00ff);
195 // 2) composition with SSE2
196 for (; x < length - 3; x += 4) {
197 const __m128i srcVector = _mm_loadu_si128((__m128i *)&src[x]);
198 const __m128i dstVector = _mm_load_si128((__m128i *)&dst[x]);
200 __m128i result = _mm_adds_epu8(srcVector, dstVector);
201 INTERPOLATE_PIXEL_255_SSE2(result, result, dstVector, constAlphaVector, oneMinusConstAlpha, colorMask, half)
202 _mm_store_si128((__m128i *)&dst[x], result);
206 for (; x < length; ++x)
207 dst[x] = comp_func_Plus_one_pixel_const_alpha(dst[x], src[x], const_alpha, one_minus_const_alpha);
211 void QT_FASTCALL comp_func_Source_sse2(uint *dst, const uint *src, int length, uint const_alpha)
213 if (const_alpha == 255) {
214 ::memcpy(dst, src, length * sizeof(uint));
216 const int ialpha = 255 - const_alpha;
220 // 1) prologue, align on 16 bytes
221 ALIGNMENT_PROLOGUE_16BYTES(dst, x, length)
222 dst[x] = INTERPOLATE_PIXEL_255(src[x], const_alpha, dst[x], ialpha);
224 // 2) interpolate pixels with SSE2
225 const __m128i half = _mm_set1_epi16(0x80);
226 const __m128i colorMask = _mm_set1_epi32(0x00ff00ff);
227 const __m128i constAlphaVector = _mm_set1_epi16(const_alpha);
228 const __m128i oneMinusConstAlpha = _mm_set1_epi16(ialpha);
229 for (; x < length - 3; x += 4) {
230 const __m128i srcVector = _mm_loadu_si128((__m128i *)&src[x]);
231 __m128i dstVector = _mm_load_si128((__m128i *)&dst[x]);
232 INTERPOLATE_PIXEL_255_SSE2(dstVector, srcVector, dstVector, constAlphaVector, oneMinusConstAlpha, colorMask, half)
233 _mm_store_si128((__m128i *)&dst[x], dstVector);
237 for (; x < length; ++x)
238 dst[x] = INTERPOLATE_PIXEL_255(src[x], const_alpha, dst[x], ialpha);
242 void qt_memfill32_sse2(quint32 *dest, quint32 value, int count)
246 case 6: *dest++ = value;
247 case 5: *dest++ = value;
248 case 4: *dest++ = value;
249 case 3: *dest++ = value;
250 case 2: *dest++ = value;
251 case 1: *dest = value;
256 const int align = (quintptr)(dest) & 0xf;
258 case 4: *dest++ = value; --count;
259 case 8: *dest++ = value; --count;
260 case 12: *dest++ = value; --count;
263 int count128 = count / 4;
264 __m128i *dst128 = reinterpret_cast<__m128i*>(dest);
265 const __m128i value128 = _mm_set_epi32(value, value, value, value);
267 int n = (count128 + 3) / 4;
268 switch (count128 & 0x3) {
269 case 0: do { _mm_stream_si128(dst128++, value128);
270 case 3: _mm_stream_si128(dst128++, value128);
271 case 2: _mm_stream_si128(dst128++, value128);
272 case 1: _mm_stream_si128(dst128++, value128);
276 const int rest = count & 0x3;
279 case 3: dest[count - 3] = value;
280 case 2: dest[count - 2] = value;
281 case 1: dest[count - 1] = value;
286 void QT_FASTCALL comp_func_solid_SourceOver_sse2(uint *destPixels, int length, uint color, uint const_alpha)
288 if ((const_alpha & qAlpha(color)) == 255) {
289 qt_memfill32_sse2(destPixels, color, length);
291 if (const_alpha != 255)
292 color = BYTE_MUL(color, const_alpha);
294 const quint32 minusAlphaOfColor = qAlpha(~color);
297 quint32 *dst = (quint32 *) destPixels;
298 const __m128i colorVector = _mm_set1_epi32(color);
299 const __m128i colorMask = _mm_set1_epi32(0x00ff00ff);
300 const __m128i half = _mm_set1_epi16(0x80);
301 const __m128i minusAlphaOfColorVector = _mm_set1_epi16(minusAlphaOfColor);
303 ALIGNMENT_PROLOGUE_16BYTES(dst, x, length)
304 destPixels[x] = color + BYTE_MUL(destPixels[x], minusAlphaOfColor);
306 for (; x < length-3; x += 4) {
307 __m128i dstVector = _mm_load_si128((__m128i *)&dst[x]);
308 BYTE_MUL_SSE2(dstVector, dstVector, minusAlphaOfColorVector, colorMask, half);
309 dstVector = _mm_add_epi8(colorVector, dstVector);
310 _mm_store_si128((__m128i *)&dst[x], dstVector);
312 for (;x < length; ++x)
313 destPixels[x] = color + BYTE_MUL(destPixels[x], minusAlphaOfColor);
317 CompositionFunctionSolid qt_functionForModeSolid_onlySSE2[numCompositionFunctions] = {
318 comp_func_solid_SourceOver_sse2,
319 comp_func_solid_DestinationOver,
320 comp_func_solid_Clear,
321 comp_func_solid_Source,
322 comp_func_solid_Destination,
323 comp_func_solid_SourceIn,
324 comp_func_solid_DestinationIn,
325 comp_func_solid_SourceOut,
326 comp_func_solid_DestinationOut,
327 comp_func_solid_SourceAtop,
328 comp_func_solid_DestinationAtop,
330 comp_func_solid_Plus,
331 comp_func_solid_Multiply,
332 comp_func_solid_Screen,
333 comp_func_solid_Overlay,
334 comp_func_solid_Darken,
335 comp_func_solid_Lighten,
336 comp_func_solid_ColorDodge,
337 comp_func_solid_ColorBurn,
338 comp_func_solid_HardLight,
339 comp_func_solid_SoftLight,
340 comp_func_solid_Difference,
341 comp_func_solid_Exclusion,
342 rasterop_solid_SourceOrDestination,
343 rasterop_solid_SourceAndDestination,
344 rasterop_solid_SourceXorDestination,
345 rasterop_solid_NotSourceAndNotDestination,
346 rasterop_solid_NotSourceOrNotDestination,
347 rasterop_solid_NotSourceXorDestination,
348 rasterop_solid_NotSource,
349 rasterop_solid_NotSourceAndDestination,
350 rasterop_solid_SourceAndNotDestination
353 CompositionFunction qt_functionForMode_onlySSE2[numCompositionFunctions] = {
354 comp_func_SourceOver_sse2,
355 comp_func_DestinationOver,
357 comp_func_Source_sse2,
358 comp_func_Destination,
360 comp_func_DestinationIn,
362 comp_func_DestinationOut,
363 comp_func_SourceAtop,
364 comp_func_DestinationAtop,
372 comp_func_ColorDodge,
376 comp_func_Difference,
378 rasterop_SourceOrDestination,
379 rasterop_SourceAndDestination,
380 rasterop_SourceXorDestination,
381 rasterop_NotSourceAndNotDestination,
382 rasterop_NotSourceOrNotDestination,
383 rasterop_NotSourceXorDestination,
385 rasterop_NotSourceAndDestination,
386 rasterop_SourceAndNotDestination
389 void qt_memfill16_sse2(quint16 *dest, quint16 value, int count)
393 case 2: *dest++ = value;
394 case 1: *dest = value;
399 const int align = (quintptr)(dest) & 0x3;
401 case 2: *dest++ = value; --count;
404 const quint32 value32 = (value << 16) | value;
405 qt_memfill32_sse2(reinterpret_cast<quint32*>(dest), value32, count / 2);
408 dest[count - 1] = value;
411 void qt_bitmapblit32_sse2(QRasterBuffer *rasterBuffer, int x, int y,
413 const uchar *src, int width, int height, int stride)
415 quint32 *dest = reinterpret_cast<quint32*>(rasterBuffer->scanLine(y)) + x;
416 const int destStride = rasterBuffer->bytesPerLine() / sizeof(quint32);
418 const __m128i c128 = _mm_set1_epi32(color);
419 const __m128i maskmask1 = _mm_set_epi32(0x10101010, 0x20202020,
420 0x40404040, 0x80808080);
421 const __m128i maskadd1 = _mm_set_epi32(0x70707070, 0x60606060,
422 0x40404040, 0x00000000);
425 const __m128i maskmask2 = _mm_set_epi32(0x01010101, 0x02020202,
426 0x04040404, 0x08080808);
427 const __m128i maskadd2 = _mm_set_epi32(0x7f7f7f7f, 0x7e7e7e7e,
428 0x7c7c7c7c, 0x78787878);
430 for (int x = 0; x < width; x += 8) {
431 const quint8 s = src[x >> 3];
434 __m128i mask1 = _mm_set1_epi8(s);
435 __m128i mask2 = mask1;
437 mask1 = _mm_and_si128(mask1, maskmask1);
438 mask1 = _mm_add_epi8(mask1, maskadd1);
439 _mm_maskmoveu_si128(c128, mask1, (char*)(dest + x));
440 mask2 = _mm_and_si128(mask2, maskmask2);
441 mask2 = _mm_add_epi8(mask2, maskadd2);
442 _mm_maskmoveu_si128(c128, mask2, (char*)(dest + x + 4));
449 const quint8 s = *src;
451 __m128i mask1 = _mm_set1_epi8(s);
452 mask1 = _mm_and_si128(mask1, maskmask1);
453 mask1 = _mm_add_epi8(mask1, maskadd1);
454 _mm_maskmoveu_si128(c128, mask1, (char*)(dest));
462 void qt_bitmapblit16_sse2(QRasterBuffer *rasterBuffer, int x, int y,
464 const uchar *src, int width, int height, int stride)
466 const quint16 c = qt_colorConvert<quint16, quint32>(color, 0);
467 quint16 *dest = reinterpret_cast<quint16*>(rasterBuffer->scanLine(y)) + x;
468 const int destStride = rasterBuffer->bytesPerLine() / sizeof(quint16);
470 const __m128i c128 = _mm_set1_epi16(c);
471 #if defined(Q_CC_MSVC)
472 # pragma warning(disable: 4309) // truncation of constant value
474 const __m128i maskmask = _mm_set_epi16(0x0101, 0x0202, 0x0404, 0x0808,
475 0x1010, 0x2020, 0x4040, 0x8080);
476 const __m128i maskadd = _mm_set_epi16(0x7f7f, 0x7e7e, 0x7c7c, 0x7878,
477 0x7070, 0x6060, 0x4040, 0x0000);
480 for (int x = 0; x < width; x += 8) {
481 const quint8 s = src[x >> 3];
484 __m128i mask = _mm_set1_epi8(s);
485 mask = _mm_and_si128(mask, maskmask);
486 mask = _mm_add_epi8(mask, maskadd);
487 _mm_maskmoveu_si128(c128, mask, (char*)(dest + x));
497 typedef __m128i Int32x4;
498 typedef __m128 Float32x4;
500 union Vect_buffer_i { Int32x4 v; int i[4]; };
501 union Vect_buffer_f { Float32x4 v; float f[4]; };
503 static inline Float32x4 v_dup(float x) { return _mm_set1_ps(x); }
504 static inline Float32x4 v_dup(double x) { return _mm_set1_ps(x); }
505 static inline Int32x4 v_dup(int x) { return _mm_set1_epi32(x); }
506 static inline Int32x4 v_dup(uint x) { return _mm_set1_epi32(x); }
508 static inline Float32x4 v_add(Float32x4 a, Float32x4 b) { return _mm_add_ps(a, b); }
509 static inline Int32x4 v_add(Int32x4 a, Int32x4 b) { return _mm_add_epi32(a, b); }
511 static inline Float32x4 v_max(Float32x4 a, Float32x4 b) { return _mm_max_ps(a, b); }
512 static inline Float32x4 v_min(Float32x4 a, Float32x4 b) { return _mm_min_ps(a, b); }
513 static inline Int32x4 v_min_16(Int32x4 a, Int32x4 b) { return _mm_min_epi16(a, b); }
515 static inline Int32x4 v_and(Int32x4 a, Int32x4 b) { return _mm_and_si128(a, b); }
517 static inline Float32x4 v_sub(Float32x4 a, Float32x4 b) { return _mm_sub_ps(a, b); }
518 static inline Int32x4 v_sub(Int32x4 a, Int32x4 b) { return _mm_sub_epi32(a, b); }
520 static inline Float32x4 v_mul(Float32x4 a, Float32x4 b) { return _mm_mul_ps(a, b); }
522 static inline Float32x4 v_sqrt(Float32x4 x) { return _mm_sqrt_ps(x); }
524 static inline Int32x4 v_toInt(Float32x4 x) { return _mm_cvttps_epi32(x); }
527 const uint * QT_FASTCALL qt_fetch_radial_gradient_sse2(uint *buffer, const Operator *op, const QSpanData *data,
528 int y, int x, int length)
530 return qt_fetch_radial_gradient_template<QRadialFetchSimd<QSimdSse2> >(buffer, op, data, y, x, length);
536 #endif // QT_HAVE_SSE2